xref: /openbmc/linux/fs/ext4/extents.c (revision 78560d41)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
4  * Written by Alex Tomas <alex@clusterfs.com>
5  *
6  * Architecture independence:
7  *   Copyright (c) 2005, Bull S.A.
8  *   Written by Pierre Peiffer <pierre.peiffer@bull.net>
9  */
10 
11 /*
12  * Extents support for EXT4
13  *
14  * TODO:
15  *   - ext4*_error() should be used in some situations
16  *   - analyze all BUG()/BUG_ON(), use -EIO where appropriate
17  *   - smart tree reduction
18  */
19 
20 #include <linux/fs.h>
21 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/highuid.h>
24 #include <linux/pagemap.h>
25 #include <linux/quotaops.h>
26 #include <linux/string.h>
27 #include <linux/slab.h>
28 #include <linux/uaccess.h>
29 #include <linux/fiemap.h>
30 #include <linux/backing-dev.h>
31 #include "ext4_jbd2.h"
32 #include "ext4_extents.h"
33 #include "xattr.h"
34 
35 #include <trace/events/ext4.h>
36 
37 /*
38  * used by extent splitting.
39  */
40 #define EXT4_EXT_MAY_ZEROOUT	0x1  /* safe to zeroout if split fails \
41 					due to ENOSPC */
42 #define EXT4_EXT_MARK_UNWRIT1	0x2  /* mark first half unwritten */
43 #define EXT4_EXT_MARK_UNWRIT2	0x4  /* mark second half unwritten */
44 
45 #define EXT4_EXT_DATA_VALID1	0x8  /* first half contains valid data */
46 #define EXT4_EXT_DATA_VALID2	0x10 /* second half contains valid data */
47 
48 static __le32 ext4_extent_block_csum(struct inode *inode,
49 				     struct ext4_extent_header *eh)
50 {
51 	struct ext4_inode_info *ei = EXT4_I(inode);
52 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
53 	__u32 csum;
54 
55 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
56 			   EXT4_EXTENT_TAIL_OFFSET(eh));
57 	return cpu_to_le32(csum);
58 }
59 
60 static int ext4_extent_block_csum_verify(struct inode *inode,
61 					 struct ext4_extent_header *eh)
62 {
63 	struct ext4_extent_tail *et;
64 
65 	if (!ext4_has_metadata_csum(inode->i_sb))
66 		return 1;
67 
68 	et = find_ext4_extent_tail(eh);
69 	if (et->et_checksum != ext4_extent_block_csum(inode, eh))
70 		return 0;
71 	return 1;
72 }
73 
74 static void ext4_extent_block_csum_set(struct inode *inode,
75 				       struct ext4_extent_header *eh)
76 {
77 	struct ext4_extent_tail *et;
78 
79 	if (!ext4_has_metadata_csum(inode->i_sb))
80 		return;
81 
82 	et = find_ext4_extent_tail(eh);
83 	et->et_checksum = ext4_extent_block_csum(inode, eh);
84 }
85 
86 static int ext4_split_extent(handle_t *handle,
87 				struct inode *inode,
88 				struct ext4_ext_path **ppath,
89 				struct ext4_map_blocks *map,
90 				int split_flag,
91 				int flags);
92 
93 static int ext4_split_extent_at(handle_t *handle,
94 			     struct inode *inode,
95 			     struct ext4_ext_path **ppath,
96 			     ext4_lblk_t split,
97 			     int split_flag,
98 			     int flags);
99 
100 static int ext4_find_delayed_extent(struct inode *inode,
101 				    struct extent_status *newes);
102 
103 static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
104 {
105 	/*
106 	 * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
107 	 * moment, get_block can be called only for blocks inside i_size since
108 	 * page cache has been already dropped and writes are blocked by
109 	 * i_mutex. So we can safely drop the i_data_sem here.
110 	 */
111 	BUG_ON(EXT4_JOURNAL(inode) == NULL);
112 	ext4_discard_preallocations(inode);
113 	up_write(&EXT4_I(inode)->i_data_sem);
114 	*dropped = 1;
115 	return 0;
116 }
117 
118 /*
119  * Make sure 'handle' has at least 'check_cred' credits. If not, restart
120  * transaction with 'restart_cred' credits. The function drops i_data_sem
121  * when restarting transaction and gets it after transaction is restarted.
122  *
123  * The function returns 0 on success, 1 if transaction had to be restarted,
124  * and < 0 in case of fatal error.
125  */
126 int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
127 				int check_cred, int restart_cred,
128 				int revoke_cred)
129 {
130 	int ret;
131 	int dropped = 0;
132 
133 	ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
134 		revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
135 	if (dropped)
136 		down_write(&EXT4_I(inode)->i_data_sem);
137 	return ret;
138 }
139 
140 /*
141  * could return:
142  *  - EROFS
143  *  - ENOMEM
144  */
145 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
146 				struct ext4_ext_path *path)
147 {
148 	if (path->p_bh) {
149 		/* path points to block */
150 		BUFFER_TRACE(path->p_bh, "get_write_access");
151 		return ext4_journal_get_write_access(handle, path->p_bh);
152 	}
153 	/* path points to leaf/index in inode body */
154 	/* we use in-core data, no need to protect them */
155 	return 0;
156 }
157 
158 /*
159  * could return:
160  *  - EROFS
161  *  - ENOMEM
162  *  - EIO
163  */
164 static int __ext4_ext_dirty(const char *where, unsigned int line,
165 			    handle_t *handle, struct inode *inode,
166 			    struct ext4_ext_path *path)
167 {
168 	int err;
169 
170 	WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
171 	if (path->p_bh) {
172 		ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
173 		/* path points to block */
174 		err = __ext4_handle_dirty_metadata(where, line, handle,
175 						   inode, path->p_bh);
176 	} else {
177 		/* path points to leaf/index in inode body */
178 		err = ext4_mark_inode_dirty(handle, inode);
179 	}
180 	return err;
181 }
182 
183 #define ext4_ext_dirty(handle, inode, path) \
184 		__ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
185 
186 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
187 			      struct ext4_ext_path *path,
188 			      ext4_lblk_t block)
189 {
190 	if (path) {
191 		int depth = path->p_depth;
192 		struct ext4_extent *ex;
193 
194 		/*
195 		 * Try to predict block placement assuming that we are
196 		 * filling in a file which will eventually be
197 		 * non-sparse --- i.e., in the case of libbfd writing
198 		 * an ELF object sections out-of-order but in a way
199 		 * the eventually results in a contiguous object or
200 		 * executable file, or some database extending a table
201 		 * space file.  However, this is actually somewhat
202 		 * non-ideal if we are writing a sparse file such as
203 		 * qemu or KVM writing a raw image file that is going
204 		 * to stay fairly sparse, since it will end up
205 		 * fragmenting the file system's free space.  Maybe we
206 		 * should have some hueristics or some way to allow
207 		 * userspace to pass a hint to file system,
208 		 * especially if the latter case turns out to be
209 		 * common.
210 		 */
211 		ex = path[depth].p_ext;
212 		if (ex) {
213 			ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
214 			ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
215 
216 			if (block > ext_block)
217 				return ext_pblk + (block - ext_block);
218 			else
219 				return ext_pblk - (ext_block - block);
220 		}
221 
222 		/* it looks like index is empty;
223 		 * try to find starting block from index itself */
224 		if (path[depth].p_bh)
225 			return path[depth].p_bh->b_blocknr;
226 	}
227 
228 	/* OK. use inode's group */
229 	return ext4_inode_to_goal_block(inode);
230 }
231 
232 /*
233  * Allocation for a meta data block
234  */
235 static ext4_fsblk_t
236 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
237 			struct ext4_ext_path *path,
238 			struct ext4_extent *ex, int *err, unsigned int flags)
239 {
240 	ext4_fsblk_t goal, newblock;
241 
242 	goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
243 	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
244 					NULL, err);
245 	return newblock;
246 }
247 
248 static inline int ext4_ext_space_block(struct inode *inode, int check)
249 {
250 	int size;
251 
252 	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
253 			/ sizeof(struct ext4_extent);
254 #ifdef AGGRESSIVE_TEST
255 	if (!check && size > 6)
256 		size = 6;
257 #endif
258 	return size;
259 }
260 
261 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
262 {
263 	int size;
264 
265 	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
266 			/ sizeof(struct ext4_extent_idx);
267 #ifdef AGGRESSIVE_TEST
268 	if (!check && size > 5)
269 		size = 5;
270 #endif
271 	return size;
272 }
273 
274 static inline int ext4_ext_space_root(struct inode *inode, int check)
275 {
276 	int size;
277 
278 	size = sizeof(EXT4_I(inode)->i_data);
279 	size -= sizeof(struct ext4_extent_header);
280 	size /= sizeof(struct ext4_extent);
281 #ifdef AGGRESSIVE_TEST
282 	if (!check && size > 3)
283 		size = 3;
284 #endif
285 	return size;
286 }
287 
288 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
289 {
290 	int size;
291 
292 	size = sizeof(EXT4_I(inode)->i_data);
293 	size -= sizeof(struct ext4_extent_header);
294 	size /= sizeof(struct ext4_extent_idx);
295 #ifdef AGGRESSIVE_TEST
296 	if (!check && size > 4)
297 		size = 4;
298 #endif
299 	return size;
300 }
301 
302 static inline int
303 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
304 			   struct ext4_ext_path **ppath, ext4_lblk_t lblk,
305 			   int nofail)
306 {
307 	struct ext4_ext_path *path = *ppath;
308 	int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
309 
310 	return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
311 			EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
312 			EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO |
313 			(nofail ? EXT4_GET_BLOCKS_METADATA_NOFAIL:0));
314 }
315 
316 static int
317 ext4_ext_max_entries(struct inode *inode, int depth)
318 {
319 	int max;
320 
321 	if (depth == ext_depth(inode)) {
322 		if (depth == 0)
323 			max = ext4_ext_space_root(inode, 1);
324 		else
325 			max = ext4_ext_space_root_idx(inode, 1);
326 	} else {
327 		if (depth == 0)
328 			max = ext4_ext_space_block(inode, 1);
329 		else
330 			max = ext4_ext_space_block_idx(inode, 1);
331 	}
332 
333 	return max;
334 }
335 
336 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
337 {
338 	ext4_fsblk_t block = ext4_ext_pblock(ext);
339 	int len = ext4_ext_get_actual_len(ext);
340 	ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
341 
342 	/*
343 	 * We allow neither:
344 	 *  - zero length
345 	 *  - overflow/wrap-around
346 	 */
347 	if (lblock + len <= lblock)
348 		return 0;
349 	return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, len);
350 }
351 
352 static int ext4_valid_extent_idx(struct inode *inode,
353 				struct ext4_extent_idx *ext_idx)
354 {
355 	ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
356 
357 	return ext4_data_block_valid(EXT4_SB(inode->i_sb), block, 1);
358 }
359 
360 static int ext4_valid_extent_entries(struct inode *inode,
361 				struct ext4_extent_header *eh,
362 				int depth)
363 {
364 	unsigned short entries;
365 	if (eh->eh_entries == 0)
366 		return 1;
367 
368 	entries = le16_to_cpu(eh->eh_entries);
369 
370 	if (depth == 0) {
371 		/* leaf entries */
372 		struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
373 		struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
374 		ext4_fsblk_t pblock = 0;
375 		ext4_lblk_t lblock = 0;
376 		ext4_lblk_t prev = 0;
377 		int len = 0;
378 		while (entries) {
379 			if (!ext4_valid_extent(inode, ext))
380 				return 0;
381 
382 			/* Check for overlapping extents */
383 			lblock = le32_to_cpu(ext->ee_block);
384 			len = ext4_ext_get_actual_len(ext);
385 			if ((lblock <= prev) && prev) {
386 				pblock = ext4_ext_pblock(ext);
387 				es->s_last_error_block = cpu_to_le64(pblock);
388 				return 0;
389 			}
390 			ext++;
391 			entries--;
392 			prev = lblock + len - 1;
393 		}
394 	} else {
395 		struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
396 		while (entries) {
397 			if (!ext4_valid_extent_idx(inode, ext_idx))
398 				return 0;
399 			ext_idx++;
400 			entries--;
401 		}
402 	}
403 	return 1;
404 }
405 
406 static int __ext4_ext_check(const char *function, unsigned int line,
407 			    struct inode *inode, struct ext4_extent_header *eh,
408 			    int depth, ext4_fsblk_t pblk)
409 {
410 	const char *error_msg;
411 	int max = 0, err = -EFSCORRUPTED;
412 
413 	if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
414 		error_msg = "invalid magic";
415 		goto corrupted;
416 	}
417 	if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
418 		error_msg = "unexpected eh_depth";
419 		goto corrupted;
420 	}
421 	if (unlikely(eh->eh_max == 0)) {
422 		error_msg = "invalid eh_max";
423 		goto corrupted;
424 	}
425 	max = ext4_ext_max_entries(inode, depth);
426 	if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
427 		error_msg = "too large eh_max";
428 		goto corrupted;
429 	}
430 	if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
431 		error_msg = "invalid eh_entries";
432 		goto corrupted;
433 	}
434 	if (!ext4_valid_extent_entries(inode, eh, depth)) {
435 		error_msg = "invalid extent entries";
436 		goto corrupted;
437 	}
438 	if (unlikely(depth > 32)) {
439 		error_msg = "too large eh_depth";
440 		goto corrupted;
441 	}
442 	/* Verify checksum on non-root extent tree nodes */
443 	if (ext_depth(inode) != depth &&
444 	    !ext4_extent_block_csum_verify(inode, eh)) {
445 		error_msg = "extent tree corrupted";
446 		err = -EFSBADCRC;
447 		goto corrupted;
448 	}
449 	return 0;
450 
451 corrupted:
452 	ext4_set_errno(inode->i_sb, -err);
453 	ext4_error_inode(inode, function, line, 0,
454 			 "pblk %llu bad header/extent: %s - magic %x, "
455 			 "entries %u, max %u(%u), depth %u(%u)",
456 			 (unsigned long long) pblk, error_msg,
457 			 le16_to_cpu(eh->eh_magic),
458 			 le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max),
459 			 max, le16_to_cpu(eh->eh_depth), depth);
460 	return err;
461 }
462 
463 #define ext4_ext_check(inode, eh, depth, pblk)			\
464 	__ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk))
465 
466 int ext4_ext_check_inode(struct inode *inode)
467 {
468 	return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
469 }
470 
471 static void ext4_cache_extents(struct inode *inode,
472 			       struct ext4_extent_header *eh)
473 {
474 	struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
475 	ext4_lblk_t prev = 0;
476 	int i;
477 
478 	for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
479 		unsigned int status = EXTENT_STATUS_WRITTEN;
480 		ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
481 		int len = ext4_ext_get_actual_len(ex);
482 
483 		if (prev && (prev != lblk))
484 			ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
485 					     EXTENT_STATUS_HOLE);
486 
487 		if (ext4_ext_is_unwritten(ex))
488 			status = EXTENT_STATUS_UNWRITTEN;
489 		ext4_es_cache_extent(inode, lblk, len,
490 				     ext4_ext_pblock(ex), status);
491 		prev = lblk + len;
492 	}
493 }
494 
495 static struct buffer_head *
496 __read_extent_tree_block(const char *function, unsigned int line,
497 			 struct inode *inode, ext4_fsblk_t pblk, int depth,
498 			 int flags)
499 {
500 	struct buffer_head		*bh;
501 	int				err;
502 
503 	bh = sb_getblk_gfp(inode->i_sb, pblk, __GFP_MOVABLE | GFP_NOFS);
504 	if (unlikely(!bh))
505 		return ERR_PTR(-ENOMEM);
506 
507 	if (!bh_uptodate_or_lock(bh)) {
508 		trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
509 		err = bh_submit_read(bh);
510 		if (err < 0)
511 			goto errout;
512 	}
513 	if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
514 		return bh;
515 	if (!ext4_has_feature_journal(inode->i_sb) ||
516 	    (inode->i_ino !=
517 	     le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) {
518 		err = __ext4_ext_check(function, line, inode,
519 				       ext_block_hdr(bh), depth, pblk);
520 		if (err)
521 			goto errout;
522 	}
523 	set_buffer_verified(bh);
524 	/*
525 	 * If this is a leaf block, cache all of its entries
526 	 */
527 	if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
528 		struct ext4_extent_header *eh = ext_block_hdr(bh);
529 		ext4_cache_extents(inode, eh);
530 	}
531 	return bh;
532 errout:
533 	put_bh(bh);
534 	return ERR_PTR(err);
535 
536 }
537 
538 #define read_extent_tree_block(inode, pblk, depth, flags)		\
539 	__read_extent_tree_block(__func__, __LINE__, (inode), (pblk),   \
540 				 (depth), (flags))
541 
542 /*
543  * This function is called to cache a file's extent information in the
544  * extent status tree
545  */
546 int ext4_ext_precache(struct inode *inode)
547 {
548 	struct ext4_inode_info *ei = EXT4_I(inode);
549 	struct ext4_ext_path *path = NULL;
550 	struct buffer_head *bh;
551 	int i = 0, depth, ret = 0;
552 
553 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
554 		return 0;	/* not an extent-mapped inode */
555 
556 	down_read(&ei->i_data_sem);
557 	depth = ext_depth(inode);
558 
559 	path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
560 		       GFP_NOFS);
561 	if (path == NULL) {
562 		up_read(&ei->i_data_sem);
563 		return -ENOMEM;
564 	}
565 
566 	/* Don't cache anything if there are no external extent blocks */
567 	if (depth == 0)
568 		goto out;
569 	path[0].p_hdr = ext_inode_hdr(inode);
570 	ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
571 	if (ret)
572 		goto out;
573 	path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
574 	while (i >= 0) {
575 		/*
576 		 * If this is a leaf block or we've reached the end of
577 		 * the index block, go up
578 		 */
579 		if ((i == depth) ||
580 		    path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
581 			brelse(path[i].p_bh);
582 			path[i].p_bh = NULL;
583 			i--;
584 			continue;
585 		}
586 		bh = read_extent_tree_block(inode,
587 					    ext4_idx_pblock(path[i].p_idx++),
588 					    depth - i - 1,
589 					    EXT4_EX_FORCE_CACHE);
590 		if (IS_ERR(bh)) {
591 			ret = PTR_ERR(bh);
592 			break;
593 		}
594 		i++;
595 		path[i].p_bh = bh;
596 		path[i].p_hdr = ext_block_hdr(bh);
597 		path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
598 	}
599 	ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
600 out:
601 	up_read(&ei->i_data_sem);
602 	ext4_ext_drop_refs(path);
603 	kfree(path);
604 	return ret;
605 }
606 
607 #ifdef EXT_DEBUG
608 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
609 {
610 	int k, l = path->p_depth;
611 
612 	ext_debug("path:");
613 	for (k = 0; k <= l; k++, path++) {
614 		if (path->p_idx) {
615 			ext_debug("  %d->%llu",
616 				  le32_to_cpu(path->p_idx->ei_block),
617 				  ext4_idx_pblock(path->p_idx));
618 		} else if (path->p_ext) {
619 			ext_debug("  %d:[%d]%d:%llu ",
620 				  le32_to_cpu(path->p_ext->ee_block),
621 				  ext4_ext_is_unwritten(path->p_ext),
622 				  ext4_ext_get_actual_len(path->p_ext),
623 				  ext4_ext_pblock(path->p_ext));
624 		} else
625 			ext_debug("  []");
626 	}
627 	ext_debug("\n");
628 }
629 
630 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
631 {
632 	int depth = ext_depth(inode);
633 	struct ext4_extent_header *eh;
634 	struct ext4_extent *ex;
635 	int i;
636 
637 	if (!path)
638 		return;
639 
640 	eh = path[depth].p_hdr;
641 	ex = EXT_FIRST_EXTENT(eh);
642 
643 	ext_debug("Displaying leaf extents for inode %lu\n", inode->i_ino);
644 
645 	for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
646 		ext_debug("%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
647 			  ext4_ext_is_unwritten(ex),
648 			  ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
649 	}
650 	ext_debug("\n");
651 }
652 
653 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
654 			ext4_fsblk_t newblock, int level)
655 {
656 	int depth = ext_depth(inode);
657 	struct ext4_extent *ex;
658 
659 	if (depth != level) {
660 		struct ext4_extent_idx *idx;
661 		idx = path[level].p_idx;
662 		while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
663 			ext_debug("%d: move %d:%llu in new index %llu\n", level,
664 					le32_to_cpu(idx->ei_block),
665 					ext4_idx_pblock(idx),
666 					newblock);
667 			idx++;
668 		}
669 
670 		return;
671 	}
672 
673 	ex = path[depth].p_ext;
674 	while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
675 		ext_debug("move %d:%llu:[%d]%d in new leaf %llu\n",
676 				le32_to_cpu(ex->ee_block),
677 				ext4_ext_pblock(ex),
678 				ext4_ext_is_unwritten(ex),
679 				ext4_ext_get_actual_len(ex),
680 				newblock);
681 		ex++;
682 	}
683 }
684 
685 #else
686 #define ext4_ext_show_path(inode, path)
687 #define ext4_ext_show_leaf(inode, path)
688 #define ext4_ext_show_move(inode, path, newblock, level)
689 #endif
690 
691 void ext4_ext_drop_refs(struct ext4_ext_path *path)
692 {
693 	int depth, i;
694 
695 	if (!path)
696 		return;
697 	depth = path->p_depth;
698 	for (i = 0; i <= depth; i++, path++) {
699 		if (path->p_bh) {
700 			brelse(path->p_bh);
701 			path->p_bh = NULL;
702 		}
703 	}
704 }
705 
706 /*
707  * ext4_ext_binsearch_idx:
708  * binary search for the closest index of the given block
709  * the header must be checked before calling this
710  */
711 static void
712 ext4_ext_binsearch_idx(struct inode *inode,
713 			struct ext4_ext_path *path, ext4_lblk_t block)
714 {
715 	struct ext4_extent_header *eh = path->p_hdr;
716 	struct ext4_extent_idx *r, *l, *m;
717 
718 
719 	ext_debug("binsearch for %u(idx):  ", block);
720 
721 	l = EXT_FIRST_INDEX(eh) + 1;
722 	r = EXT_LAST_INDEX(eh);
723 	while (l <= r) {
724 		m = l + (r - l) / 2;
725 		if (block < le32_to_cpu(m->ei_block))
726 			r = m - 1;
727 		else
728 			l = m + 1;
729 		ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ei_block),
730 				m, le32_to_cpu(m->ei_block),
731 				r, le32_to_cpu(r->ei_block));
732 	}
733 
734 	path->p_idx = l - 1;
735 	ext_debug("  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
736 		  ext4_idx_pblock(path->p_idx));
737 
738 #ifdef CHECK_BINSEARCH
739 	{
740 		struct ext4_extent_idx *chix, *ix;
741 		int k;
742 
743 		chix = ix = EXT_FIRST_INDEX(eh);
744 		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
745 			if (k != 0 && le32_to_cpu(ix->ei_block) <=
746 			    le32_to_cpu(ix[-1].ei_block)) {
747 				printk(KERN_DEBUG "k=%d, ix=0x%p, "
748 				       "first=0x%p\n", k,
749 				       ix, EXT_FIRST_INDEX(eh));
750 				printk(KERN_DEBUG "%u <= %u\n",
751 				       le32_to_cpu(ix->ei_block),
752 				       le32_to_cpu(ix[-1].ei_block));
753 			}
754 			BUG_ON(k && le32_to_cpu(ix->ei_block)
755 					   <= le32_to_cpu(ix[-1].ei_block));
756 			if (block < le32_to_cpu(ix->ei_block))
757 				break;
758 			chix = ix;
759 		}
760 		BUG_ON(chix != path->p_idx);
761 	}
762 #endif
763 
764 }
765 
766 /*
767  * ext4_ext_binsearch:
768  * binary search for closest extent of the given block
769  * the header must be checked before calling this
770  */
771 static void
772 ext4_ext_binsearch(struct inode *inode,
773 		struct ext4_ext_path *path, ext4_lblk_t block)
774 {
775 	struct ext4_extent_header *eh = path->p_hdr;
776 	struct ext4_extent *r, *l, *m;
777 
778 	if (eh->eh_entries == 0) {
779 		/*
780 		 * this leaf is empty:
781 		 * we get such a leaf in split/add case
782 		 */
783 		return;
784 	}
785 
786 	ext_debug("binsearch for %u:  ", block);
787 
788 	l = EXT_FIRST_EXTENT(eh) + 1;
789 	r = EXT_LAST_EXTENT(eh);
790 
791 	while (l <= r) {
792 		m = l + (r - l) / 2;
793 		if (block < le32_to_cpu(m->ee_block))
794 			r = m - 1;
795 		else
796 			l = m + 1;
797 		ext_debug("%p(%u):%p(%u):%p(%u) ", l, le32_to_cpu(l->ee_block),
798 				m, le32_to_cpu(m->ee_block),
799 				r, le32_to_cpu(r->ee_block));
800 	}
801 
802 	path->p_ext = l - 1;
803 	ext_debug("  -> %d:%llu:[%d]%d ",
804 			le32_to_cpu(path->p_ext->ee_block),
805 			ext4_ext_pblock(path->p_ext),
806 			ext4_ext_is_unwritten(path->p_ext),
807 			ext4_ext_get_actual_len(path->p_ext));
808 
809 #ifdef CHECK_BINSEARCH
810 	{
811 		struct ext4_extent *chex, *ex;
812 		int k;
813 
814 		chex = ex = EXT_FIRST_EXTENT(eh);
815 		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
816 			BUG_ON(k && le32_to_cpu(ex->ee_block)
817 					  <= le32_to_cpu(ex[-1].ee_block));
818 			if (block < le32_to_cpu(ex->ee_block))
819 				break;
820 			chex = ex;
821 		}
822 		BUG_ON(chex != path->p_ext);
823 	}
824 #endif
825 
826 }
827 
828 int ext4_ext_tree_init(handle_t *handle, struct inode *inode)
829 {
830 	struct ext4_extent_header *eh;
831 
832 	eh = ext_inode_hdr(inode);
833 	eh->eh_depth = 0;
834 	eh->eh_entries = 0;
835 	eh->eh_magic = EXT4_EXT_MAGIC;
836 	eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
837 	ext4_mark_inode_dirty(handle, inode);
838 	return 0;
839 }
840 
841 struct ext4_ext_path *
842 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
843 		 struct ext4_ext_path **orig_path, int flags)
844 {
845 	struct ext4_extent_header *eh;
846 	struct buffer_head *bh;
847 	struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
848 	short int depth, i, ppos = 0;
849 	int ret;
850 
851 	eh = ext_inode_hdr(inode);
852 	depth = ext_depth(inode);
853 	if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
854 		EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
855 				 depth);
856 		ret = -EFSCORRUPTED;
857 		goto err;
858 	}
859 
860 	if (path) {
861 		ext4_ext_drop_refs(path);
862 		if (depth > path[0].p_maxdepth) {
863 			kfree(path);
864 			*orig_path = path = NULL;
865 		}
866 	}
867 	if (!path) {
868 		/* account possible depth increase */
869 		path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
870 				GFP_NOFS);
871 		if (unlikely(!path))
872 			return ERR_PTR(-ENOMEM);
873 		path[0].p_maxdepth = depth + 1;
874 	}
875 	path[0].p_hdr = eh;
876 	path[0].p_bh = NULL;
877 
878 	i = depth;
879 	if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
880 		ext4_cache_extents(inode, eh);
881 	/* walk through the tree */
882 	while (i) {
883 		ext_debug("depth %d: num %d, max %d\n",
884 			  ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
885 
886 		ext4_ext_binsearch_idx(inode, path + ppos, block);
887 		path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
888 		path[ppos].p_depth = i;
889 		path[ppos].p_ext = NULL;
890 
891 		bh = read_extent_tree_block(inode, path[ppos].p_block, --i,
892 					    flags);
893 		if (IS_ERR(bh)) {
894 			ret = PTR_ERR(bh);
895 			goto err;
896 		}
897 
898 		eh = ext_block_hdr(bh);
899 		ppos++;
900 		path[ppos].p_bh = bh;
901 		path[ppos].p_hdr = eh;
902 	}
903 
904 	path[ppos].p_depth = i;
905 	path[ppos].p_ext = NULL;
906 	path[ppos].p_idx = NULL;
907 
908 	/* find extent */
909 	ext4_ext_binsearch(inode, path + ppos, block);
910 	/* if not an empty leaf */
911 	if (path[ppos].p_ext)
912 		path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
913 
914 	ext4_ext_show_path(inode, path);
915 
916 	return path;
917 
918 err:
919 	ext4_ext_drop_refs(path);
920 	kfree(path);
921 	if (orig_path)
922 		*orig_path = NULL;
923 	return ERR_PTR(ret);
924 }
925 
926 /*
927  * ext4_ext_insert_index:
928  * insert new index [@logical;@ptr] into the block at @curp;
929  * check where to insert: before @curp or after @curp
930  */
931 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
932 				 struct ext4_ext_path *curp,
933 				 int logical, ext4_fsblk_t ptr)
934 {
935 	struct ext4_extent_idx *ix;
936 	int len, err;
937 
938 	err = ext4_ext_get_access(handle, inode, curp);
939 	if (err)
940 		return err;
941 
942 	if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
943 		EXT4_ERROR_INODE(inode,
944 				 "logical %d == ei_block %d!",
945 				 logical, le32_to_cpu(curp->p_idx->ei_block));
946 		return -EFSCORRUPTED;
947 	}
948 
949 	if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
950 			     >= le16_to_cpu(curp->p_hdr->eh_max))) {
951 		EXT4_ERROR_INODE(inode,
952 				 "eh_entries %d >= eh_max %d!",
953 				 le16_to_cpu(curp->p_hdr->eh_entries),
954 				 le16_to_cpu(curp->p_hdr->eh_max));
955 		return -EFSCORRUPTED;
956 	}
957 
958 	if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
959 		/* insert after */
960 		ext_debug("insert new index %d after: %llu\n", logical, ptr);
961 		ix = curp->p_idx + 1;
962 	} else {
963 		/* insert before */
964 		ext_debug("insert new index %d before: %llu\n", logical, ptr);
965 		ix = curp->p_idx;
966 	}
967 
968 	len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
969 	BUG_ON(len < 0);
970 	if (len > 0) {
971 		ext_debug("insert new index %d: "
972 				"move %d indices from 0x%p to 0x%p\n",
973 				logical, len, ix, ix + 1);
974 		memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
975 	}
976 
977 	if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
978 		EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
979 		return -EFSCORRUPTED;
980 	}
981 
982 	ix->ei_block = cpu_to_le32(logical);
983 	ext4_idx_store_pblock(ix, ptr);
984 	le16_add_cpu(&curp->p_hdr->eh_entries, 1);
985 
986 	if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
987 		EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
988 		return -EFSCORRUPTED;
989 	}
990 
991 	err = ext4_ext_dirty(handle, inode, curp);
992 	ext4_std_error(inode->i_sb, err);
993 
994 	return err;
995 }
996 
997 /*
998  * ext4_ext_split:
999  * inserts new subtree into the path, using free index entry
1000  * at depth @at:
1001  * - allocates all needed blocks (new leaf and all intermediate index blocks)
1002  * - makes decision where to split
1003  * - moves remaining extents and index entries (right to the split point)
1004  *   into the newly allocated blocks
1005  * - initializes subtree
1006  */
1007 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1008 			  unsigned int flags,
1009 			  struct ext4_ext_path *path,
1010 			  struct ext4_extent *newext, int at)
1011 {
1012 	struct buffer_head *bh = NULL;
1013 	int depth = ext_depth(inode);
1014 	struct ext4_extent_header *neh;
1015 	struct ext4_extent_idx *fidx;
1016 	int i = at, k, m, a;
1017 	ext4_fsblk_t newblock, oldblock;
1018 	__le32 border;
1019 	ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1020 	int err = 0;
1021 	size_t ext_size = 0;
1022 
1023 	/* make decision: where to split? */
1024 	/* FIXME: now decision is simplest: at current extent */
1025 
1026 	/* if current leaf will be split, then we should use
1027 	 * border from split point */
1028 	if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1029 		EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1030 		return -EFSCORRUPTED;
1031 	}
1032 	if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1033 		border = path[depth].p_ext[1].ee_block;
1034 		ext_debug("leaf will be split."
1035 				" next leaf starts at %d\n",
1036 				  le32_to_cpu(border));
1037 	} else {
1038 		border = newext->ee_block;
1039 		ext_debug("leaf will be added."
1040 				" next leaf starts at %d\n",
1041 				le32_to_cpu(border));
1042 	}
1043 
1044 	/*
1045 	 * If error occurs, then we break processing
1046 	 * and mark filesystem read-only. index won't
1047 	 * be inserted and tree will be in consistent
1048 	 * state. Next mount will repair buffers too.
1049 	 */
1050 
1051 	/*
1052 	 * Get array to track all allocated blocks.
1053 	 * We need this to handle errors and free blocks
1054 	 * upon them.
1055 	 */
1056 	ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), GFP_NOFS);
1057 	if (!ablocks)
1058 		return -ENOMEM;
1059 
1060 	/* allocate all needed blocks */
1061 	ext_debug("allocate %d blocks for indexes/leaf\n", depth - at);
1062 	for (a = 0; a < depth - at; a++) {
1063 		newblock = ext4_ext_new_meta_block(handle, inode, path,
1064 						   newext, &err, flags);
1065 		if (newblock == 0)
1066 			goto cleanup;
1067 		ablocks[a] = newblock;
1068 	}
1069 
1070 	/* initialize new leaf */
1071 	newblock = ablocks[--a];
1072 	if (unlikely(newblock == 0)) {
1073 		EXT4_ERROR_INODE(inode, "newblock == 0!");
1074 		err = -EFSCORRUPTED;
1075 		goto cleanup;
1076 	}
1077 	bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1078 	if (unlikely(!bh)) {
1079 		err = -ENOMEM;
1080 		goto cleanup;
1081 	}
1082 	lock_buffer(bh);
1083 
1084 	err = ext4_journal_get_create_access(handle, bh);
1085 	if (err)
1086 		goto cleanup;
1087 
1088 	neh = ext_block_hdr(bh);
1089 	neh->eh_entries = 0;
1090 	neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1091 	neh->eh_magic = EXT4_EXT_MAGIC;
1092 	neh->eh_depth = 0;
1093 
1094 	/* move remainder of path[depth] to the new leaf */
1095 	if (unlikely(path[depth].p_hdr->eh_entries !=
1096 		     path[depth].p_hdr->eh_max)) {
1097 		EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1098 				 path[depth].p_hdr->eh_entries,
1099 				 path[depth].p_hdr->eh_max);
1100 		err = -EFSCORRUPTED;
1101 		goto cleanup;
1102 	}
1103 	/* start copy from next extent */
1104 	m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1105 	ext4_ext_show_move(inode, path, newblock, depth);
1106 	if (m) {
1107 		struct ext4_extent *ex;
1108 		ex = EXT_FIRST_EXTENT(neh);
1109 		memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1110 		le16_add_cpu(&neh->eh_entries, m);
1111 	}
1112 
1113 	/* zero out unused area in the extent block */
1114 	ext_size = sizeof(struct ext4_extent_header) +
1115 		sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1116 	memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1117 	ext4_extent_block_csum_set(inode, neh);
1118 	set_buffer_uptodate(bh);
1119 	unlock_buffer(bh);
1120 
1121 	err = ext4_handle_dirty_metadata(handle, inode, bh);
1122 	if (err)
1123 		goto cleanup;
1124 	brelse(bh);
1125 	bh = NULL;
1126 
1127 	/* correct old leaf */
1128 	if (m) {
1129 		err = ext4_ext_get_access(handle, inode, path + depth);
1130 		if (err)
1131 			goto cleanup;
1132 		le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1133 		err = ext4_ext_dirty(handle, inode, path + depth);
1134 		if (err)
1135 			goto cleanup;
1136 
1137 	}
1138 
1139 	/* create intermediate indexes */
1140 	k = depth - at - 1;
1141 	if (unlikely(k < 0)) {
1142 		EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1143 		err = -EFSCORRUPTED;
1144 		goto cleanup;
1145 	}
1146 	if (k)
1147 		ext_debug("create %d intermediate indices\n", k);
1148 	/* insert new index into current index block */
1149 	/* current depth stored in i var */
1150 	i = depth - 1;
1151 	while (k--) {
1152 		oldblock = newblock;
1153 		newblock = ablocks[--a];
1154 		bh = sb_getblk(inode->i_sb, newblock);
1155 		if (unlikely(!bh)) {
1156 			err = -ENOMEM;
1157 			goto cleanup;
1158 		}
1159 		lock_buffer(bh);
1160 
1161 		err = ext4_journal_get_create_access(handle, bh);
1162 		if (err)
1163 			goto cleanup;
1164 
1165 		neh = ext_block_hdr(bh);
1166 		neh->eh_entries = cpu_to_le16(1);
1167 		neh->eh_magic = EXT4_EXT_MAGIC;
1168 		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1169 		neh->eh_depth = cpu_to_le16(depth - i);
1170 		fidx = EXT_FIRST_INDEX(neh);
1171 		fidx->ei_block = border;
1172 		ext4_idx_store_pblock(fidx, oldblock);
1173 
1174 		ext_debug("int.index at %d (block %llu): %u -> %llu\n",
1175 				i, newblock, le32_to_cpu(border), oldblock);
1176 
1177 		/* move remainder of path[i] to the new index block */
1178 		if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1179 					EXT_LAST_INDEX(path[i].p_hdr))) {
1180 			EXT4_ERROR_INODE(inode,
1181 					 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1182 					 le32_to_cpu(path[i].p_ext->ee_block));
1183 			err = -EFSCORRUPTED;
1184 			goto cleanup;
1185 		}
1186 		/* start copy indexes */
1187 		m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1188 		ext_debug("cur 0x%p, last 0x%p\n", path[i].p_idx,
1189 				EXT_MAX_INDEX(path[i].p_hdr));
1190 		ext4_ext_show_move(inode, path, newblock, i);
1191 		if (m) {
1192 			memmove(++fidx, path[i].p_idx,
1193 				sizeof(struct ext4_extent_idx) * m);
1194 			le16_add_cpu(&neh->eh_entries, m);
1195 		}
1196 		/* zero out unused area in the extent block */
1197 		ext_size = sizeof(struct ext4_extent_header) +
1198 		   (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1199 		memset(bh->b_data + ext_size, 0,
1200 			inode->i_sb->s_blocksize - ext_size);
1201 		ext4_extent_block_csum_set(inode, neh);
1202 		set_buffer_uptodate(bh);
1203 		unlock_buffer(bh);
1204 
1205 		err = ext4_handle_dirty_metadata(handle, inode, bh);
1206 		if (err)
1207 			goto cleanup;
1208 		brelse(bh);
1209 		bh = NULL;
1210 
1211 		/* correct old index */
1212 		if (m) {
1213 			err = ext4_ext_get_access(handle, inode, path + i);
1214 			if (err)
1215 				goto cleanup;
1216 			le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1217 			err = ext4_ext_dirty(handle, inode, path + i);
1218 			if (err)
1219 				goto cleanup;
1220 		}
1221 
1222 		i--;
1223 	}
1224 
1225 	/* insert new index */
1226 	err = ext4_ext_insert_index(handle, inode, path + at,
1227 				    le32_to_cpu(border), newblock);
1228 
1229 cleanup:
1230 	if (bh) {
1231 		if (buffer_locked(bh))
1232 			unlock_buffer(bh);
1233 		brelse(bh);
1234 	}
1235 
1236 	if (err) {
1237 		/* free all allocated blocks in error case */
1238 		for (i = 0; i < depth; i++) {
1239 			if (!ablocks[i])
1240 				continue;
1241 			ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1242 					 EXT4_FREE_BLOCKS_METADATA);
1243 		}
1244 	}
1245 	kfree(ablocks);
1246 
1247 	return err;
1248 }
1249 
1250 /*
1251  * ext4_ext_grow_indepth:
1252  * implements tree growing procedure:
1253  * - allocates new block
1254  * - moves top-level data (index block or leaf) into the new block
1255  * - initializes new top-level, creating index that points to the
1256  *   just created block
1257  */
1258 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1259 				 unsigned int flags)
1260 {
1261 	struct ext4_extent_header *neh;
1262 	struct buffer_head *bh;
1263 	ext4_fsblk_t newblock, goal = 0;
1264 	struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1265 	int err = 0;
1266 	size_t ext_size = 0;
1267 
1268 	/* Try to prepend new index to old one */
1269 	if (ext_depth(inode))
1270 		goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1271 	if (goal > le32_to_cpu(es->s_first_data_block)) {
1272 		flags |= EXT4_MB_HINT_TRY_GOAL;
1273 		goal--;
1274 	} else
1275 		goal = ext4_inode_to_goal_block(inode);
1276 	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1277 					NULL, &err);
1278 	if (newblock == 0)
1279 		return err;
1280 
1281 	bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1282 	if (unlikely(!bh))
1283 		return -ENOMEM;
1284 	lock_buffer(bh);
1285 
1286 	err = ext4_journal_get_create_access(handle, bh);
1287 	if (err) {
1288 		unlock_buffer(bh);
1289 		goto out;
1290 	}
1291 
1292 	ext_size = sizeof(EXT4_I(inode)->i_data);
1293 	/* move top-level index/leaf into new block */
1294 	memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1295 	/* zero out unused area in the extent block */
1296 	memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1297 
1298 	/* set size of new block */
1299 	neh = ext_block_hdr(bh);
1300 	/* old root could have indexes or leaves
1301 	 * so calculate e_max right way */
1302 	if (ext_depth(inode))
1303 		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1304 	else
1305 		neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1306 	neh->eh_magic = EXT4_EXT_MAGIC;
1307 	ext4_extent_block_csum_set(inode, neh);
1308 	set_buffer_uptodate(bh);
1309 	unlock_buffer(bh);
1310 
1311 	err = ext4_handle_dirty_metadata(handle, inode, bh);
1312 	if (err)
1313 		goto out;
1314 
1315 	/* Update top-level index: num,max,pointer */
1316 	neh = ext_inode_hdr(inode);
1317 	neh->eh_entries = cpu_to_le16(1);
1318 	ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1319 	if (neh->eh_depth == 0) {
1320 		/* Root extent block becomes index block */
1321 		neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1322 		EXT_FIRST_INDEX(neh)->ei_block =
1323 			EXT_FIRST_EXTENT(neh)->ee_block;
1324 	}
1325 	ext_debug("new root: num %d(%d), lblock %d, ptr %llu\n",
1326 		  le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1327 		  le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1328 		  ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1329 
1330 	le16_add_cpu(&neh->eh_depth, 1);
1331 	ext4_mark_inode_dirty(handle, inode);
1332 out:
1333 	brelse(bh);
1334 
1335 	return err;
1336 }
1337 
1338 /*
1339  * ext4_ext_create_new_leaf:
1340  * finds empty index and adds new leaf.
1341  * if no free index is found, then it requests in-depth growing.
1342  */
1343 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1344 				    unsigned int mb_flags,
1345 				    unsigned int gb_flags,
1346 				    struct ext4_ext_path **ppath,
1347 				    struct ext4_extent *newext)
1348 {
1349 	struct ext4_ext_path *path = *ppath;
1350 	struct ext4_ext_path *curp;
1351 	int depth, i, err = 0;
1352 
1353 repeat:
1354 	i = depth = ext_depth(inode);
1355 
1356 	/* walk up to the tree and look for free index entry */
1357 	curp = path + depth;
1358 	while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1359 		i--;
1360 		curp--;
1361 	}
1362 
1363 	/* we use already allocated block for index block,
1364 	 * so subsequent data blocks should be contiguous */
1365 	if (EXT_HAS_FREE_INDEX(curp)) {
1366 		/* if we found index with free entry, then use that
1367 		 * entry: create all needed subtree and add new leaf */
1368 		err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1369 		if (err)
1370 			goto out;
1371 
1372 		/* refill path */
1373 		path = ext4_find_extent(inode,
1374 				    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1375 				    ppath, gb_flags);
1376 		if (IS_ERR(path))
1377 			err = PTR_ERR(path);
1378 	} else {
1379 		/* tree is full, time to grow in depth */
1380 		err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1381 		if (err)
1382 			goto out;
1383 
1384 		/* refill path */
1385 		path = ext4_find_extent(inode,
1386 				   (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1387 				    ppath, gb_flags);
1388 		if (IS_ERR(path)) {
1389 			err = PTR_ERR(path);
1390 			goto out;
1391 		}
1392 
1393 		/*
1394 		 * only first (depth 0 -> 1) produces free space;
1395 		 * in all other cases we have to split the grown tree
1396 		 */
1397 		depth = ext_depth(inode);
1398 		if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1399 			/* now we need to split */
1400 			goto repeat;
1401 		}
1402 	}
1403 
1404 out:
1405 	return err;
1406 }
1407 
1408 /*
1409  * search the closest allocated block to the left for *logical
1410  * and returns it at @logical + it's physical address at @phys
1411  * if *logical is the smallest allocated block, the function
1412  * returns 0 at @phys
1413  * return value contains 0 (success) or error code
1414  */
1415 static int ext4_ext_search_left(struct inode *inode,
1416 				struct ext4_ext_path *path,
1417 				ext4_lblk_t *logical, ext4_fsblk_t *phys)
1418 {
1419 	struct ext4_extent_idx *ix;
1420 	struct ext4_extent *ex;
1421 	int depth, ee_len;
1422 
1423 	if (unlikely(path == NULL)) {
1424 		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1425 		return -EFSCORRUPTED;
1426 	}
1427 	depth = path->p_depth;
1428 	*phys = 0;
1429 
1430 	if (depth == 0 && path->p_ext == NULL)
1431 		return 0;
1432 
1433 	/* usually extent in the path covers blocks smaller
1434 	 * then *logical, but it can be that extent is the
1435 	 * first one in the file */
1436 
1437 	ex = path[depth].p_ext;
1438 	ee_len = ext4_ext_get_actual_len(ex);
1439 	if (*logical < le32_to_cpu(ex->ee_block)) {
1440 		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1441 			EXT4_ERROR_INODE(inode,
1442 					 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1443 					 *logical, le32_to_cpu(ex->ee_block));
1444 			return -EFSCORRUPTED;
1445 		}
1446 		while (--depth >= 0) {
1447 			ix = path[depth].p_idx;
1448 			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1449 				EXT4_ERROR_INODE(inode,
1450 				  "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1451 				  ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1452 				  EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1453 		le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1454 				  depth);
1455 				return -EFSCORRUPTED;
1456 			}
1457 		}
1458 		return 0;
1459 	}
1460 
1461 	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1462 		EXT4_ERROR_INODE(inode,
1463 				 "logical %d < ee_block %d + ee_len %d!",
1464 				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1465 		return -EFSCORRUPTED;
1466 	}
1467 
1468 	*logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1469 	*phys = ext4_ext_pblock(ex) + ee_len - 1;
1470 	return 0;
1471 }
1472 
1473 /*
1474  * search the closest allocated block to the right for *logical
1475  * and returns it at @logical + it's physical address at @phys
1476  * if *logical is the largest allocated block, the function
1477  * returns 0 at @phys
1478  * return value contains 0 (success) or error code
1479  */
1480 static int ext4_ext_search_right(struct inode *inode,
1481 				 struct ext4_ext_path *path,
1482 				 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1483 				 struct ext4_extent **ret_ex)
1484 {
1485 	struct buffer_head *bh = NULL;
1486 	struct ext4_extent_header *eh;
1487 	struct ext4_extent_idx *ix;
1488 	struct ext4_extent *ex;
1489 	ext4_fsblk_t block;
1490 	int depth;	/* Note, NOT eh_depth; depth from top of tree */
1491 	int ee_len;
1492 
1493 	if (unlikely(path == NULL)) {
1494 		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1495 		return -EFSCORRUPTED;
1496 	}
1497 	depth = path->p_depth;
1498 	*phys = 0;
1499 
1500 	if (depth == 0 && path->p_ext == NULL)
1501 		return 0;
1502 
1503 	/* usually extent in the path covers blocks smaller
1504 	 * then *logical, but it can be that extent is the
1505 	 * first one in the file */
1506 
1507 	ex = path[depth].p_ext;
1508 	ee_len = ext4_ext_get_actual_len(ex);
1509 	if (*logical < le32_to_cpu(ex->ee_block)) {
1510 		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1511 			EXT4_ERROR_INODE(inode,
1512 					 "first_extent(path[%d].p_hdr) != ex",
1513 					 depth);
1514 			return -EFSCORRUPTED;
1515 		}
1516 		while (--depth >= 0) {
1517 			ix = path[depth].p_idx;
1518 			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1519 				EXT4_ERROR_INODE(inode,
1520 						 "ix != EXT_FIRST_INDEX *logical %d!",
1521 						 *logical);
1522 				return -EFSCORRUPTED;
1523 			}
1524 		}
1525 		goto found_extent;
1526 	}
1527 
1528 	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1529 		EXT4_ERROR_INODE(inode,
1530 				 "logical %d < ee_block %d + ee_len %d!",
1531 				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1532 		return -EFSCORRUPTED;
1533 	}
1534 
1535 	if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1536 		/* next allocated block in this leaf */
1537 		ex++;
1538 		goto found_extent;
1539 	}
1540 
1541 	/* go up and search for index to the right */
1542 	while (--depth >= 0) {
1543 		ix = path[depth].p_idx;
1544 		if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1545 			goto got_index;
1546 	}
1547 
1548 	/* we've gone up to the root and found no index to the right */
1549 	return 0;
1550 
1551 got_index:
1552 	/* we've found index to the right, let's
1553 	 * follow it and find the closest allocated
1554 	 * block to the right */
1555 	ix++;
1556 	block = ext4_idx_pblock(ix);
1557 	while (++depth < path->p_depth) {
1558 		/* subtract from p_depth to get proper eh_depth */
1559 		bh = read_extent_tree_block(inode, block,
1560 					    path->p_depth - depth, 0);
1561 		if (IS_ERR(bh))
1562 			return PTR_ERR(bh);
1563 		eh = ext_block_hdr(bh);
1564 		ix = EXT_FIRST_INDEX(eh);
1565 		block = ext4_idx_pblock(ix);
1566 		put_bh(bh);
1567 	}
1568 
1569 	bh = read_extent_tree_block(inode, block, path->p_depth - depth, 0);
1570 	if (IS_ERR(bh))
1571 		return PTR_ERR(bh);
1572 	eh = ext_block_hdr(bh);
1573 	ex = EXT_FIRST_EXTENT(eh);
1574 found_extent:
1575 	*logical = le32_to_cpu(ex->ee_block);
1576 	*phys = ext4_ext_pblock(ex);
1577 	*ret_ex = ex;
1578 	if (bh)
1579 		put_bh(bh);
1580 	return 0;
1581 }
1582 
1583 /*
1584  * ext4_ext_next_allocated_block:
1585  * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1586  * NOTE: it considers block number from index entry as
1587  * allocated block. Thus, index entries have to be consistent
1588  * with leaves.
1589  */
1590 ext4_lblk_t
1591 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1592 {
1593 	int depth;
1594 
1595 	BUG_ON(path == NULL);
1596 	depth = path->p_depth;
1597 
1598 	if (depth == 0 && path->p_ext == NULL)
1599 		return EXT_MAX_BLOCKS;
1600 
1601 	while (depth >= 0) {
1602 		struct ext4_ext_path *p = &path[depth];
1603 
1604 		if (depth == path->p_depth) {
1605 			/* leaf */
1606 			if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1607 				return le32_to_cpu(p->p_ext[1].ee_block);
1608 		} else {
1609 			/* index */
1610 			if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1611 				return le32_to_cpu(p->p_idx[1].ei_block);
1612 		}
1613 		depth--;
1614 	}
1615 
1616 	return EXT_MAX_BLOCKS;
1617 }
1618 
1619 /*
1620  * ext4_ext_next_leaf_block:
1621  * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1622  */
1623 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1624 {
1625 	int depth;
1626 
1627 	BUG_ON(path == NULL);
1628 	depth = path->p_depth;
1629 
1630 	/* zero-tree has no leaf blocks at all */
1631 	if (depth == 0)
1632 		return EXT_MAX_BLOCKS;
1633 
1634 	/* go to index block */
1635 	depth--;
1636 
1637 	while (depth >= 0) {
1638 		if (path[depth].p_idx !=
1639 				EXT_LAST_INDEX(path[depth].p_hdr))
1640 			return (ext4_lblk_t)
1641 				le32_to_cpu(path[depth].p_idx[1].ei_block);
1642 		depth--;
1643 	}
1644 
1645 	return EXT_MAX_BLOCKS;
1646 }
1647 
1648 /*
1649  * ext4_ext_correct_indexes:
1650  * if leaf gets modified and modified extent is first in the leaf,
1651  * then we have to correct all indexes above.
1652  * TODO: do we need to correct tree in all cases?
1653  */
1654 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1655 				struct ext4_ext_path *path)
1656 {
1657 	struct ext4_extent_header *eh;
1658 	int depth = ext_depth(inode);
1659 	struct ext4_extent *ex;
1660 	__le32 border;
1661 	int k, err = 0;
1662 
1663 	eh = path[depth].p_hdr;
1664 	ex = path[depth].p_ext;
1665 
1666 	if (unlikely(ex == NULL || eh == NULL)) {
1667 		EXT4_ERROR_INODE(inode,
1668 				 "ex %p == NULL or eh %p == NULL", ex, eh);
1669 		return -EFSCORRUPTED;
1670 	}
1671 
1672 	if (depth == 0) {
1673 		/* there is no tree at all */
1674 		return 0;
1675 	}
1676 
1677 	if (ex != EXT_FIRST_EXTENT(eh)) {
1678 		/* we correct tree if first leaf got modified only */
1679 		return 0;
1680 	}
1681 
1682 	/*
1683 	 * TODO: we need correction if border is smaller than current one
1684 	 */
1685 	k = depth - 1;
1686 	border = path[depth].p_ext->ee_block;
1687 	err = ext4_ext_get_access(handle, inode, path + k);
1688 	if (err)
1689 		return err;
1690 	path[k].p_idx->ei_block = border;
1691 	err = ext4_ext_dirty(handle, inode, path + k);
1692 	if (err)
1693 		return err;
1694 
1695 	while (k--) {
1696 		/* change all left-side indexes */
1697 		if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1698 			break;
1699 		err = ext4_ext_get_access(handle, inode, path + k);
1700 		if (err)
1701 			break;
1702 		path[k].p_idx->ei_block = border;
1703 		err = ext4_ext_dirty(handle, inode, path + k);
1704 		if (err)
1705 			break;
1706 	}
1707 
1708 	return err;
1709 }
1710 
1711 static int ext4_can_extents_be_merged(struct inode *inode,
1712 				      struct ext4_extent *ex1,
1713 				      struct ext4_extent *ex2)
1714 {
1715 	unsigned short ext1_ee_len, ext2_ee_len;
1716 
1717 	if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1718 		return 0;
1719 
1720 	ext1_ee_len = ext4_ext_get_actual_len(ex1);
1721 	ext2_ee_len = ext4_ext_get_actual_len(ex2);
1722 
1723 	if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1724 			le32_to_cpu(ex2->ee_block))
1725 		return 0;
1726 
1727 	if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1728 		return 0;
1729 
1730 	if (ext4_ext_is_unwritten(ex1) &&
1731 	    ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1732 		return 0;
1733 #ifdef AGGRESSIVE_TEST
1734 	if (ext1_ee_len >= 4)
1735 		return 0;
1736 #endif
1737 
1738 	if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1739 		return 1;
1740 	return 0;
1741 }
1742 
1743 /*
1744  * This function tries to merge the "ex" extent to the next extent in the tree.
1745  * It always tries to merge towards right. If you want to merge towards
1746  * left, pass "ex - 1" as argument instead of "ex".
1747  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1748  * 1 if they got merged.
1749  */
1750 static int ext4_ext_try_to_merge_right(struct inode *inode,
1751 				 struct ext4_ext_path *path,
1752 				 struct ext4_extent *ex)
1753 {
1754 	struct ext4_extent_header *eh;
1755 	unsigned int depth, len;
1756 	int merge_done = 0, unwritten;
1757 
1758 	depth = ext_depth(inode);
1759 	BUG_ON(path[depth].p_hdr == NULL);
1760 	eh = path[depth].p_hdr;
1761 
1762 	while (ex < EXT_LAST_EXTENT(eh)) {
1763 		if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1764 			break;
1765 		/* merge with next extent! */
1766 		unwritten = ext4_ext_is_unwritten(ex);
1767 		ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1768 				+ ext4_ext_get_actual_len(ex + 1));
1769 		if (unwritten)
1770 			ext4_ext_mark_unwritten(ex);
1771 
1772 		if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1773 			len = (EXT_LAST_EXTENT(eh) - ex - 1)
1774 				* sizeof(struct ext4_extent);
1775 			memmove(ex + 1, ex + 2, len);
1776 		}
1777 		le16_add_cpu(&eh->eh_entries, -1);
1778 		merge_done = 1;
1779 		WARN_ON(eh->eh_entries == 0);
1780 		if (!eh->eh_entries)
1781 			EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1782 	}
1783 
1784 	return merge_done;
1785 }
1786 
1787 /*
1788  * This function does a very simple check to see if we can collapse
1789  * an extent tree with a single extent tree leaf block into the inode.
1790  */
1791 static void ext4_ext_try_to_merge_up(handle_t *handle,
1792 				     struct inode *inode,
1793 				     struct ext4_ext_path *path)
1794 {
1795 	size_t s;
1796 	unsigned max_root = ext4_ext_space_root(inode, 0);
1797 	ext4_fsblk_t blk;
1798 
1799 	if ((path[0].p_depth != 1) ||
1800 	    (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1801 	    (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1802 		return;
1803 
1804 	/*
1805 	 * We need to modify the block allocation bitmap and the block
1806 	 * group descriptor to release the extent tree block.  If we
1807 	 * can't get the journal credits, give up.
1808 	 */
1809 	if (ext4_journal_extend(handle, 2,
1810 			ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1811 		return;
1812 
1813 	/*
1814 	 * Copy the extent data up to the inode
1815 	 */
1816 	blk = ext4_idx_pblock(path[0].p_idx);
1817 	s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1818 		sizeof(struct ext4_extent_idx);
1819 	s += sizeof(struct ext4_extent_header);
1820 
1821 	path[1].p_maxdepth = path[0].p_maxdepth;
1822 	memcpy(path[0].p_hdr, path[1].p_hdr, s);
1823 	path[0].p_depth = 0;
1824 	path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1825 		(path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1826 	path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1827 
1828 	brelse(path[1].p_bh);
1829 	ext4_free_blocks(handle, inode, NULL, blk, 1,
1830 			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1831 }
1832 
1833 /*
1834  * This function tries to merge the @ex extent to neighbours in the tree, then
1835  * tries to collapse the extent tree into the inode.
1836  */
1837 static void ext4_ext_try_to_merge(handle_t *handle,
1838 				  struct inode *inode,
1839 				  struct ext4_ext_path *path,
1840 				  struct ext4_extent *ex)
1841 {
1842 	struct ext4_extent_header *eh;
1843 	unsigned int depth;
1844 	int merge_done = 0;
1845 
1846 	depth = ext_depth(inode);
1847 	BUG_ON(path[depth].p_hdr == NULL);
1848 	eh = path[depth].p_hdr;
1849 
1850 	if (ex > EXT_FIRST_EXTENT(eh))
1851 		merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1852 
1853 	if (!merge_done)
1854 		(void) ext4_ext_try_to_merge_right(inode, path, ex);
1855 
1856 	ext4_ext_try_to_merge_up(handle, inode, path);
1857 }
1858 
1859 /*
1860  * check if a portion of the "newext" extent overlaps with an
1861  * existing extent.
1862  *
1863  * If there is an overlap discovered, it updates the length of the newext
1864  * such that there will be no overlap, and then returns 1.
1865  * If there is no overlap found, it returns 0.
1866  */
1867 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1868 					   struct inode *inode,
1869 					   struct ext4_extent *newext,
1870 					   struct ext4_ext_path *path)
1871 {
1872 	ext4_lblk_t b1, b2;
1873 	unsigned int depth, len1;
1874 	unsigned int ret = 0;
1875 
1876 	b1 = le32_to_cpu(newext->ee_block);
1877 	len1 = ext4_ext_get_actual_len(newext);
1878 	depth = ext_depth(inode);
1879 	if (!path[depth].p_ext)
1880 		goto out;
1881 	b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1882 
1883 	/*
1884 	 * get the next allocated block if the extent in the path
1885 	 * is before the requested block(s)
1886 	 */
1887 	if (b2 < b1) {
1888 		b2 = ext4_ext_next_allocated_block(path);
1889 		if (b2 == EXT_MAX_BLOCKS)
1890 			goto out;
1891 		b2 = EXT4_LBLK_CMASK(sbi, b2);
1892 	}
1893 
1894 	/* check for wrap through zero on extent logical start block*/
1895 	if (b1 + len1 < b1) {
1896 		len1 = EXT_MAX_BLOCKS - b1;
1897 		newext->ee_len = cpu_to_le16(len1);
1898 		ret = 1;
1899 	}
1900 
1901 	/* check for overlap */
1902 	if (b1 + len1 > b2) {
1903 		newext->ee_len = cpu_to_le16(b2 - b1);
1904 		ret = 1;
1905 	}
1906 out:
1907 	return ret;
1908 }
1909 
1910 /*
1911  * ext4_ext_insert_extent:
1912  * tries to merge requsted extent into the existing extent or
1913  * inserts requested extent as new one into the tree,
1914  * creating new leaf in the no-space case.
1915  */
1916 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1917 				struct ext4_ext_path **ppath,
1918 				struct ext4_extent *newext, int gb_flags)
1919 {
1920 	struct ext4_ext_path *path = *ppath;
1921 	struct ext4_extent_header *eh;
1922 	struct ext4_extent *ex, *fex;
1923 	struct ext4_extent *nearex; /* nearest extent */
1924 	struct ext4_ext_path *npath = NULL;
1925 	int depth, len, err;
1926 	ext4_lblk_t next;
1927 	int mb_flags = 0, unwritten;
1928 
1929 	if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1930 		mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1931 	if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1932 		EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1933 		return -EFSCORRUPTED;
1934 	}
1935 	depth = ext_depth(inode);
1936 	ex = path[depth].p_ext;
1937 	eh = path[depth].p_hdr;
1938 	if (unlikely(path[depth].p_hdr == NULL)) {
1939 		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1940 		return -EFSCORRUPTED;
1941 	}
1942 
1943 	/* try to insert block into found extent and return */
1944 	if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1945 
1946 		/*
1947 		 * Try to see whether we should rather test the extent on
1948 		 * right from ex, or from the left of ex. This is because
1949 		 * ext4_find_extent() can return either extent on the
1950 		 * left, or on the right from the searched position. This
1951 		 * will make merging more effective.
1952 		 */
1953 		if (ex < EXT_LAST_EXTENT(eh) &&
1954 		    (le32_to_cpu(ex->ee_block) +
1955 		    ext4_ext_get_actual_len(ex) <
1956 		    le32_to_cpu(newext->ee_block))) {
1957 			ex += 1;
1958 			goto prepend;
1959 		} else if ((ex > EXT_FIRST_EXTENT(eh)) &&
1960 			   (le32_to_cpu(newext->ee_block) +
1961 			   ext4_ext_get_actual_len(newext) <
1962 			   le32_to_cpu(ex->ee_block)))
1963 			ex -= 1;
1964 
1965 		/* Try to append newex to the ex */
1966 		if (ext4_can_extents_be_merged(inode, ex, newext)) {
1967 			ext_debug("append [%d]%d block to %u:[%d]%d"
1968 				  "(from %llu)\n",
1969 				  ext4_ext_is_unwritten(newext),
1970 				  ext4_ext_get_actual_len(newext),
1971 				  le32_to_cpu(ex->ee_block),
1972 				  ext4_ext_is_unwritten(ex),
1973 				  ext4_ext_get_actual_len(ex),
1974 				  ext4_ext_pblock(ex));
1975 			err = ext4_ext_get_access(handle, inode,
1976 						  path + depth);
1977 			if (err)
1978 				return err;
1979 			unwritten = ext4_ext_is_unwritten(ex);
1980 			ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1981 					+ ext4_ext_get_actual_len(newext));
1982 			if (unwritten)
1983 				ext4_ext_mark_unwritten(ex);
1984 			eh = path[depth].p_hdr;
1985 			nearex = ex;
1986 			goto merge;
1987 		}
1988 
1989 prepend:
1990 		/* Try to prepend newex to the ex */
1991 		if (ext4_can_extents_be_merged(inode, newext, ex)) {
1992 			ext_debug("prepend %u[%d]%d block to %u:[%d]%d"
1993 				  "(from %llu)\n",
1994 				  le32_to_cpu(newext->ee_block),
1995 				  ext4_ext_is_unwritten(newext),
1996 				  ext4_ext_get_actual_len(newext),
1997 				  le32_to_cpu(ex->ee_block),
1998 				  ext4_ext_is_unwritten(ex),
1999 				  ext4_ext_get_actual_len(ex),
2000 				  ext4_ext_pblock(ex));
2001 			err = ext4_ext_get_access(handle, inode,
2002 						  path + depth);
2003 			if (err)
2004 				return err;
2005 
2006 			unwritten = ext4_ext_is_unwritten(ex);
2007 			ex->ee_block = newext->ee_block;
2008 			ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2009 			ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2010 					+ ext4_ext_get_actual_len(newext));
2011 			if (unwritten)
2012 				ext4_ext_mark_unwritten(ex);
2013 			eh = path[depth].p_hdr;
2014 			nearex = ex;
2015 			goto merge;
2016 		}
2017 	}
2018 
2019 	depth = ext_depth(inode);
2020 	eh = path[depth].p_hdr;
2021 	if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2022 		goto has_space;
2023 
2024 	/* probably next leaf has space for us? */
2025 	fex = EXT_LAST_EXTENT(eh);
2026 	next = EXT_MAX_BLOCKS;
2027 	if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2028 		next = ext4_ext_next_leaf_block(path);
2029 	if (next != EXT_MAX_BLOCKS) {
2030 		ext_debug("next leaf block - %u\n", next);
2031 		BUG_ON(npath != NULL);
2032 		npath = ext4_find_extent(inode, next, NULL, 0);
2033 		if (IS_ERR(npath))
2034 			return PTR_ERR(npath);
2035 		BUG_ON(npath->p_depth != path->p_depth);
2036 		eh = npath[depth].p_hdr;
2037 		if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2038 			ext_debug("next leaf isn't full(%d)\n",
2039 				  le16_to_cpu(eh->eh_entries));
2040 			path = npath;
2041 			goto has_space;
2042 		}
2043 		ext_debug("next leaf has no free space(%d,%d)\n",
2044 			  le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2045 	}
2046 
2047 	/*
2048 	 * There is no free space in the found leaf.
2049 	 * We're gonna add a new leaf in the tree.
2050 	 */
2051 	if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2052 		mb_flags |= EXT4_MB_USE_RESERVED;
2053 	err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2054 				       ppath, newext);
2055 	if (err)
2056 		goto cleanup;
2057 	depth = ext_depth(inode);
2058 	eh = path[depth].p_hdr;
2059 
2060 has_space:
2061 	nearex = path[depth].p_ext;
2062 
2063 	err = ext4_ext_get_access(handle, inode, path + depth);
2064 	if (err)
2065 		goto cleanup;
2066 
2067 	if (!nearex) {
2068 		/* there is no extent in this leaf, create first one */
2069 		ext_debug("first extent in the leaf: %u:%llu:[%d]%d\n",
2070 				le32_to_cpu(newext->ee_block),
2071 				ext4_ext_pblock(newext),
2072 				ext4_ext_is_unwritten(newext),
2073 				ext4_ext_get_actual_len(newext));
2074 		nearex = EXT_FIRST_EXTENT(eh);
2075 	} else {
2076 		if (le32_to_cpu(newext->ee_block)
2077 			   > le32_to_cpu(nearex->ee_block)) {
2078 			/* Insert after */
2079 			ext_debug("insert %u:%llu:[%d]%d before: "
2080 					"nearest %p\n",
2081 					le32_to_cpu(newext->ee_block),
2082 					ext4_ext_pblock(newext),
2083 					ext4_ext_is_unwritten(newext),
2084 					ext4_ext_get_actual_len(newext),
2085 					nearex);
2086 			nearex++;
2087 		} else {
2088 			/* Insert before */
2089 			BUG_ON(newext->ee_block == nearex->ee_block);
2090 			ext_debug("insert %u:%llu:[%d]%d after: "
2091 					"nearest %p\n",
2092 					le32_to_cpu(newext->ee_block),
2093 					ext4_ext_pblock(newext),
2094 					ext4_ext_is_unwritten(newext),
2095 					ext4_ext_get_actual_len(newext),
2096 					nearex);
2097 		}
2098 		len = EXT_LAST_EXTENT(eh) - nearex + 1;
2099 		if (len > 0) {
2100 			ext_debug("insert %u:%llu:[%d]%d: "
2101 					"move %d extents from 0x%p to 0x%p\n",
2102 					le32_to_cpu(newext->ee_block),
2103 					ext4_ext_pblock(newext),
2104 					ext4_ext_is_unwritten(newext),
2105 					ext4_ext_get_actual_len(newext),
2106 					len, nearex, nearex + 1);
2107 			memmove(nearex + 1, nearex,
2108 				len * sizeof(struct ext4_extent));
2109 		}
2110 	}
2111 
2112 	le16_add_cpu(&eh->eh_entries, 1);
2113 	path[depth].p_ext = nearex;
2114 	nearex->ee_block = newext->ee_block;
2115 	ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2116 	nearex->ee_len = newext->ee_len;
2117 
2118 merge:
2119 	/* try to merge extents */
2120 	if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2121 		ext4_ext_try_to_merge(handle, inode, path, nearex);
2122 
2123 
2124 	/* time to correct all indexes above */
2125 	err = ext4_ext_correct_indexes(handle, inode, path);
2126 	if (err)
2127 		goto cleanup;
2128 
2129 	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2130 
2131 cleanup:
2132 	ext4_ext_drop_refs(npath);
2133 	kfree(npath);
2134 	return err;
2135 }
2136 
2137 static int ext4_fill_fiemap_extents(struct inode *inode,
2138 				    ext4_lblk_t block, ext4_lblk_t num,
2139 				    struct fiemap_extent_info *fieinfo)
2140 {
2141 	struct ext4_ext_path *path = NULL;
2142 	struct ext4_extent *ex;
2143 	struct extent_status es;
2144 	ext4_lblk_t next, next_del, start = 0, end = 0;
2145 	ext4_lblk_t last = block + num;
2146 	int exists, depth = 0, err = 0;
2147 	unsigned int flags = 0;
2148 	unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2149 
2150 	while (block < last && block != EXT_MAX_BLOCKS) {
2151 		num = last - block;
2152 		/* find extent for this block */
2153 		down_read(&EXT4_I(inode)->i_data_sem);
2154 
2155 		path = ext4_find_extent(inode, block, &path, 0);
2156 		if (IS_ERR(path)) {
2157 			up_read(&EXT4_I(inode)->i_data_sem);
2158 			err = PTR_ERR(path);
2159 			path = NULL;
2160 			break;
2161 		}
2162 
2163 		depth = ext_depth(inode);
2164 		if (unlikely(path[depth].p_hdr == NULL)) {
2165 			up_read(&EXT4_I(inode)->i_data_sem);
2166 			EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2167 			err = -EFSCORRUPTED;
2168 			break;
2169 		}
2170 		ex = path[depth].p_ext;
2171 		next = ext4_ext_next_allocated_block(path);
2172 
2173 		flags = 0;
2174 		exists = 0;
2175 		if (!ex) {
2176 			/* there is no extent yet, so try to allocate
2177 			 * all requested space */
2178 			start = block;
2179 			end = block + num;
2180 		} else if (le32_to_cpu(ex->ee_block) > block) {
2181 			/* need to allocate space before found extent */
2182 			start = block;
2183 			end = le32_to_cpu(ex->ee_block);
2184 			if (block + num < end)
2185 				end = block + num;
2186 		} else if (block >= le32_to_cpu(ex->ee_block)
2187 					+ ext4_ext_get_actual_len(ex)) {
2188 			/* need to allocate space after found extent */
2189 			start = block;
2190 			end = block + num;
2191 			if (end >= next)
2192 				end = next;
2193 		} else if (block >= le32_to_cpu(ex->ee_block)) {
2194 			/*
2195 			 * some part of requested space is covered
2196 			 * by found extent
2197 			 */
2198 			start = block;
2199 			end = le32_to_cpu(ex->ee_block)
2200 				+ ext4_ext_get_actual_len(ex);
2201 			if (block + num < end)
2202 				end = block + num;
2203 			exists = 1;
2204 		} else {
2205 			BUG();
2206 		}
2207 		BUG_ON(end <= start);
2208 
2209 		if (!exists) {
2210 			es.es_lblk = start;
2211 			es.es_len = end - start;
2212 			es.es_pblk = 0;
2213 		} else {
2214 			es.es_lblk = le32_to_cpu(ex->ee_block);
2215 			es.es_len = ext4_ext_get_actual_len(ex);
2216 			es.es_pblk = ext4_ext_pblock(ex);
2217 			if (ext4_ext_is_unwritten(ex))
2218 				flags |= FIEMAP_EXTENT_UNWRITTEN;
2219 		}
2220 
2221 		/*
2222 		 * Find delayed extent and update es accordingly. We call
2223 		 * it even in !exists case to find out whether es is the
2224 		 * last existing extent or not.
2225 		 */
2226 		next_del = ext4_find_delayed_extent(inode, &es);
2227 		if (!exists && next_del) {
2228 			exists = 1;
2229 			flags |= (FIEMAP_EXTENT_DELALLOC |
2230 				  FIEMAP_EXTENT_UNKNOWN);
2231 		}
2232 		up_read(&EXT4_I(inode)->i_data_sem);
2233 
2234 		if (unlikely(es.es_len == 0)) {
2235 			EXT4_ERROR_INODE(inode, "es.es_len == 0");
2236 			err = -EFSCORRUPTED;
2237 			break;
2238 		}
2239 
2240 		/*
2241 		 * This is possible iff next == next_del == EXT_MAX_BLOCKS.
2242 		 * we need to check next == EXT_MAX_BLOCKS because it is
2243 		 * possible that an extent is with unwritten and delayed
2244 		 * status due to when an extent is delayed allocated and
2245 		 * is allocated by fallocate status tree will track both of
2246 		 * them in a extent.
2247 		 *
2248 		 * So we could return a unwritten and delayed extent, and
2249 		 * its block is equal to 'next'.
2250 		 */
2251 		if (next == next_del && next == EXT_MAX_BLOCKS) {
2252 			flags |= FIEMAP_EXTENT_LAST;
2253 			if (unlikely(next_del != EXT_MAX_BLOCKS ||
2254 				     next != EXT_MAX_BLOCKS)) {
2255 				EXT4_ERROR_INODE(inode,
2256 						 "next extent == %u, next "
2257 						 "delalloc extent = %u",
2258 						 next, next_del);
2259 				err = -EFSCORRUPTED;
2260 				break;
2261 			}
2262 		}
2263 
2264 		if (exists) {
2265 			err = fiemap_fill_next_extent(fieinfo,
2266 				(__u64)es.es_lblk << blksize_bits,
2267 				(__u64)es.es_pblk << blksize_bits,
2268 				(__u64)es.es_len << blksize_bits,
2269 				flags);
2270 			if (err < 0)
2271 				break;
2272 			if (err == 1) {
2273 				err = 0;
2274 				break;
2275 			}
2276 		}
2277 
2278 		block = es.es_lblk + es.es_len;
2279 	}
2280 
2281 	ext4_ext_drop_refs(path);
2282 	kfree(path);
2283 	return err;
2284 }
2285 
2286 static int ext4_fill_es_cache_info(struct inode *inode,
2287 				   ext4_lblk_t block, ext4_lblk_t num,
2288 				   struct fiemap_extent_info *fieinfo)
2289 {
2290 	ext4_lblk_t next, end = block + num - 1;
2291 	struct extent_status es;
2292 	unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2293 	unsigned int flags;
2294 	int err;
2295 
2296 	while (block <= end) {
2297 		next = 0;
2298 		flags = 0;
2299 		if (!ext4_es_lookup_extent(inode, block, &next, &es))
2300 			break;
2301 		if (ext4_es_is_unwritten(&es))
2302 			flags |= FIEMAP_EXTENT_UNWRITTEN;
2303 		if (ext4_es_is_delayed(&es))
2304 			flags |= (FIEMAP_EXTENT_DELALLOC |
2305 				  FIEMAP_EXTENT_UNKNOWN);
2306 		if (ext4_es_is_hole(&es))
2307 			flags |= EXT4_FIEMAP_EXTENT_HOLE;
2308 		if (next == 0)
2309 			flags |= FIEMAP_EXTENT_LAST;
2310 		if (flags & (FIEMAP_EXTENT_DELALLOC|
2311 			     EXT4_FIEMAP_EXTENT_HOLE))
2312 			es.es_pblk = 0;
2313 		else
2314 			es.es_pblk = ext4_es_pblock(&es);
2315 		err = fiemap_fill_next_extent(fieinfo,
2316 				(__u64)es.es_lblk << blksize_bits,
2317 				(__u64)es.es_pblk << blksize_bits,
2318 				(__u64)es.es_len << blksize_bits,
2319 				flags);
2320 		if (next == 0)
2321 			break;
2322 		block = next;
2323 		if (err < 0)
2324 			return err;
2325 		if (err == 1)
2326 			return 0;
2327 	}
2328 	return 0;
2329 }
2330 
2331 
2332 /*
2333  * ext4_ext_determine_hole - determine hole around given block
2334  * @inode:	inode we lookup in
2335  * @path:	path in extent tree to @lblk
2336  * @lblk:	pointer to logical block around which we want to determine hole
2337  *
2338  * Determine hole length (and start if easily possible) around given logical
2339  * block. We don't try too hard to find the beginning of the hole but @path
2340  * actually points to extent before @lblk, we provide it.
2341  *
2342  * The function returns the length of a hole starting at @lblk. We update @lblk
2343  * to the beginning of the hole if we managed to find it.
2344  */
2345 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2346 					   struct ext4_ext_path *path,
2347 					   ext4_lblk_t *lblk)
2348 {
2349 	int depth = ext_depth(inode);
2350 	struct ext4_extent *ex;
2351 	ext4_lblk_t len;
2352 
2353 	ex = path[depth].p_ext;
2354 	if (ex == NULL) {
2355 		/* there is no extent yet, so gap is [0;-] */
2356 		*lblk = 0;
2357 		len = EXT_MAX_BLOCKS;
2358 	} else if (*lblk < le32_to_cpu(ex->ee_block)) {
2359 		len = le32_to_cpu(ex->ee_block) - *lblk;
2360 	} else if (*lblk >= le32_to_cpu(ex->ee_block)
2361 			+ ext4_ext_get_actual_len(ex)) {
2362 		ext4_lblk_t next;
2363 
2364 		*lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2365 		next = ext4_ext_next_allocated_block(path);
2366 		BUG_ON(next == *lblk);
2367 		len = next - *lblk;
2368 	} else {
2369 		BUG();
2370 	}
2371 	return len;
2372 }
2373 
2374 /*
2375  * ext4_ext_put_gap_in_cache:
2376  * calculate boundaries of the gap that the requested block fits into
2377  * and cache this gap
2378  */
2379 static void
2380 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2381 			  ext4_lblk_t hole_len)
2382 {
2383 	struct extent_status es;
2384 
2385 	ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2386 				  hole_start + hole_len - 1, &es);
2387 	if (es.es_len) {
2388 		/* There's delayed extent containing lblock? */
2389 		if (es.es_lblk <= hole_start)
2390 			return;
2391 		hole_len = min(es.es_lblk - hole_start, hole_len);
2392 	}
2393 	ext_debug(" -> %u:%u\n", hole_start, hole_len);
2394 	ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2395 			      EXTENT_STATUS_HOLE);
2396 }
2397 
2398 /*
2399  * ext4_ext_rm_idx:
2400  * removes index from the index block.
2401  */
2402 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2403 			struct ext4_ext_path *path, int depth)
2404 {
2405 	int err;
2406 	ext4_fsblk_t leaf;
2407 
2408 	/* free index block */
2409 	depth--;
2410 	path = path + depth;
2411 	leaf = ext4_idx_pblock(path->p_idx);
2412 	if (unlikely(path->p_hdr->eh_entries == 0)) {
2413 		EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2414 		return -EFSCORRUPTED;
2415 	}
2416 	err = ext4_ext_get_access(handle, inode, path);
2417 	if (err)
2418 		return err;
2419 
2420 	if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2421 		int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2422 		len *= sizeof(struct ext4_extent_idx);
2423 		memmove(path->p_idx, path->p_idx + 1, len);
2424 	}
2425 
2426 	le16_add_cpu(&path->p_hdr->eh_entries, -1);
2427 	err = ext4_ext_dirty(handle, inode, path);
2428 	if (err)
2429 		return err;
2430 	ext_debug("index is empty, remove it, free block %llu\n", leaf);
2431 	trace_ext4_ext_rm_idx(inode, leaf);
2432 
2433 	ext4_free_blocks(handle, inode, NULL, leaf, 1,
2434 			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2435 
2436 	while (--depth >= 0) {
2437 		if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2438 			break;
2439 		path--;
2440 		err = ext4_ext_get_access(handle, inode, path);
2441 		if (err)
2442 			break;
2443 		path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2444 		err = ext4_ext_dirty(handle, inode, path);
2445 		if (err)
2446 			break;
2447 	}
2448 	return err;
2449 }
2450 
2451 /*
2452  * ext4_ext_calc_credits_for_single_extent:
2453  * This routine returns max. credits that needed to insert an extent
2454  * to the extent tree.
2455  * When pass the actual path, the caller should calculate credits
2456  * under i_data_sem.
2457  */
2458 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2459 						struct ext4_ext_path *path)
2460 {
2461 	if (path) {
2462 		int depth = ext_depth(inode);
2463 		int ret = 0;
2464 
2465 		/* probably there is space in leaf? */
2466 		if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2467 				< le16_to_cpu(path[depth].p_hdr->eh_max)) {
2468 
2469 			/*
2470 			 *  There are some space in the leaf tree, no
2471 			 *  need to account for leaf block credit
2472 			 *
2473 			 *  bitmaps and block group descriptor blocks
2474 			 *  and other metadata blocks still need to be
2475 			 *  accounted.
2476 			 */
2477 			/* 1 bitmap, 1 block group descriptor */
2478 			ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2479 			return ret;
2480 		}
2481 	}
2482 
2483 	return ext4_chunk_trans_blocks(inode, nrblocks);
2484 }
2485 
2486 /*
2487  * How many index/leaf blocks need to change/allocate to add @extents extents?
2488  *
2489  * If we add a single extent, then in the worse case, each tree level
2490  * index/leaf need to be changed in case of the tree split.
2491  *
2492  * If more extents are inserted, they could cause the whole tree split more
2493  * than once, but this is really rare.
2494  */
2495 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2496 {
2497 	int index;
2498 	int depth;
2499 
2500 	/* If we are converting the inline data, only one is needed here. */
2501 	if (ext4_has_inline_data(inode))
2502 		return 1;
2503 
2504 	depth = ext_depth(inode);
2505 
2506 	if (extents <= 1)
2507 		index = depth * 2;
2508 	else
2509 		index = depth * 3;
2510 
2511 	return index;
2512 }
2513 
2514 static inline int get_default_free_blocks_flags(struct inode *inode)
2515 {
2516 	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2517 	    ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2518 		return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2519 	else if (ext4_should_journal_data(inode))
2520 		return EXT4_FREE_BLOCKS_FORGET;
2521 	return 0;
2522 }
2523 
2524 /*
2525  * ext4_rereserve_cluster - increment the reserved cluster count when
2526  *                          freeing a cluster with a pending reservation
2527  *
2528  * @inode - file containing the cluster
2529  * @lblk - logical block in cluster to be reserved
2530  *
2531  * Increments the reserved cluster count and adjusts quota in a bigalloc
2532  * file system when freeing a partial cluster containing at least one
2533  * delayed and unwritten block.  A partial cluster meeting that
2534  * requirement will have a pending reservation.  If so, the
2535  * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2536  * defer reserved and allocated space accounting to a subsequent call
2537  * to this function.
2538  */
2539 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2540 {
2541 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2542 	struct ext4_inode_info *ei = EXT4_I(inode);
2543 
2544 	dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2545 
2546 	spin_lock(&ei->i_block_reservation_lock);
2547 	ei->i_reserved_data_blocks++;
2548 	percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2549 	spin_unlock(&ei->i_block_reservation_lock);
2550 
2551 	percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2552 	ext4_remove_pending(inode, lblk);
2553 }
2554 
2555 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2556 			      struct ext4_extent *ex,
2557 			      struct partial_cluster *partial,
2558 			      ext4_lblk_t from, ext4_lblk_t to)
2559 {
2560 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2561 	unsigned short ee_len = ext4_ext_get_actual_len(ex);
2562 	ext4_fsblk_t last_pblk, pblk;
2563 	ext4_lblk_t num;
2564 	int flags;
2565 
2566 	/* only extent tail removal is allowed */
2567 	if (from < le32_to_cpu(ex->ee_block) ||
2568 	    to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2569 		ext4_error(sbi->s_sb,
2570 			   "strange request: removal(2) %u-%u from %u:%u",
2571 			   from, to, le32_to_cpu(ex->ee_block), ee_len);
2572 		return 0;
2573 	}
2574 
2575 #ifdef EXTENTS_STATS
2576 	spin_lock(&sbi->s_ext_stats_lock);
2577 	sbi->s_ext_blocks += ee_len;
2578 	sbi->s_ext_extents++;
2579 	if (ee_len < sbi->s_ext_min)
2580 		sbi->s_ext_min = ee_len;
2581 	if (ee_len > sbi->s_ext_max)
2582 		sbi->s_ext_max = ee_len;
2583 	if (ext_depth(inode) > sbi->s_depth_max)
2584 		sbi->s_depth_max = ext_depth(inode);
2585 	spin_unlock(&sbi->s_ext_stats_lock);
2586 #endif
2587 
2588 	trace_ext4_remove_blocks(inode, ex, from, to, partial);
2589 
2590 	/*
2591 	 * if we have a partial cluster, and it's different from the
2592 	 * cluster of the last block in the extent, we free it
2593 	 */
2594 	last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2595 
2596 	if (partial->state != initial &&
2597 	    partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2598 		if (partial->state == tofree) {
2599 			flags = get_default_free_blocks_flags(inode);
2600 			if (ext4_is_pending(inode, partial->lblk))
2601 				flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2602 			ext4_free_blocks(handle, inode, NULL,
2603 					 EXT4_C2B(sbi, partial->pclu),
2604 					 sbi->s_cluster_ratio, flags);
2605 			if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2606 				ext4_rereserve_cluster(inode, partial->lblk);
2607 		}
2608 		partial->state = initial;
2609 	}
2610 
2611 	num = le32_to_cpu(ex->ee_block) + ee_len - from;
2612 	pblk = ext4_ext_pblock(ex) + ee_len - num;
2613 
2614 	/*
2615 	 * We free the partial cluster at the end of the extent (if any),
2616 	 * unless the cluster is used by another extent (partial_cluster
2617 	 * state is nofree).  If a partial cluster exists here, it must be
2618 	 * shared with the last block in the extent.
2619 	 */
2620 	flags = get_default_free_blocks_flags(inode);
2621 
2622 	/* partial, left end cluster aligned, right end unaligned */
2623 	if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2624 	    (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2625 	    (partial->state != nofree)) {
2626 		if (ext4_is_pending(inode, to))
2627 			flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2628 		ext4_free_blocks(handle, inode, NULL,
2629 				 EXT4_PBLK_CMASK(sbi, last_pblk),
2630 				 sbi->s_cluster_ratio, flags);
2631 		if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2632 			ext4_rereserve_cluster(inode, to);
2633 		partial->state = initial;
2634 		flags = get_default_free_blocks_flags(inode);
2635 	}
2636 
2637 	flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2638 
2639 	/*
2640 	 * For bigalloc file systems, we never free a partial cluster
2641 	 * at the beginning of the extent.  Instead, we check to see if we
2642 	 * need to free it on a subsequent call to ext4_remove_blocks,
2643 	 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2644 	 */
2645 	flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2646 	ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2647 
2648 	/* reset the partial cluster if we've freed past it */
2649 	if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2650 		partial->state = initial;
2651 
2652 	/*
2653 	 * If we've freed the entire extent but the beginning is not left
2654 	 * cluster aligned and is not marked as ineligible for freeing we
2655 	 * record the partial cluster at the beginning of the extent.  It
2656 	 * wasn't freed by the preceding ext4_free_blocks() call, and we
2657 	 * need to look farther to the left to determine if it's to be freed
2658 	 * (not shared with another extent). Else, reset the partial
2659 	 * cluster - we're either  done freeing or the beginning of the
2660 	 * extent is left cluster aligned.
2661 	 */
2662 	if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2663 		if (partial->state == initial) {
2664 			partial->pclu = EXT4_B2C(sbi, pblk);
2665 			partial->lblk = from;
2666 			partial->state = tofree;
2667 		}
2668 	} else {
2669 		partial->state = initial;
2670 	}
2671 
2672 	return 0;
2673 }
2674 
2675 /*
2676  * ext4_ext_rm_leaf() Removes the extents associated with the
2677  * blocks appearing between "start" and "end".  Both "start"
2678  * and "end" must appear in the same extent or EIO is returned.
2679  *
2680  * @handle: The journal handle
2681  * @inode:  The files inode
2682  * @path:   The path to the leaf
2683  * @partial_cluster: The cluster which we'll have to free if all extents
2684  *                   has been released from it.  However, if this value is
2685  *                   negative, it's a cluster just to the right of the
2686  *                   punched region and it must not be freed.
2687  * @start:  The first block to remove
2688  * @end:   The last block to remove
2689  */
2690 static int
2691 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2692 		 struct ext4_ext_path *path,
2693 		 struct partial_cluster *partial,
2694 		 ext4_lblk_t start, ext4_lblk_t end)
2695 {
2696 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2697 	int err = 0, correct_index = 0;
2698 	int depth = ext_depth(inode), credits, revoke_credits;
2699 	struct ext4_extent_header *eh;
2700 	ext4_lblk_t a, b;
2701 	unsigned num;
2702 	ext4_lblk_t ex_ee_block;
2703 	unsigned short ex_ee_len;
2704 	unsigned unwritten = 0;
2705 	struct ext4_extent *ex;
2706 	ext4_fsblk_t pblk;
2707 
2708 	/* the header must be checked already in ext4_ext_remove_space() */
2709 	ext_debug("truncate since %u in leaf to %u\n", start, end);
2710 	if (!path[depth].p_hdr)
2711 		path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2712 	eh = path[depth].p_hdr;
2713 	if (unlikely(path[depth].p_hdr == NULL)) {
2714 		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2715 		return -EFSCORRUPTED;
2716 	}
2717 	/* find where to start removing */
2718 	ex = path[depth].p_ext;
2719 	if (!ex)
2720 		ex = EXT_LAST_EXTENT(eh);
2721 
2722 	ex_ee_block = le32_to_cpu(ex->ee_block);
2723 	ex_ee_len = ext4_ext_get_actual_len(ex);
2724 
2725 	trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2726 
2727 	while (ex >= EXT_FIRST_EXTENT(eh) &&
2728 			ex_ee_block + ex_ee_len > start) {
2729 
2730 		if (ext4_ext_is_unwritten(ex))
2731 			unwritten = 1;
2732 		else
2733 			unwritten = 0;
2734 
2735 		ext_debug("remove ext %u:[%d]%d\n", ex_ee_block,
2736 			  unwritten, ex_ee_len);
2737 		path[depth].p_ext = ex;
2738 
2739 		a = ex_ee_block > start ? ex_ee_block : start;
2740 		b = ex_ee_block+ex_ee_len - 1 < end ?
2741 			ex_ee_block+ex_ee_len - 1 : end;
2742 
2743 		ext_debug("  border %u:%u\n", a, b);
2744 
2745 		/* If this extent is beyond the end of the hole, skip it */
2746 		if (end < ex_ee_block) {
2747 			/*
2748 			 * We're going to skip this extent and move to another,
2749 			 * so note that its first cluster is in use to avoid
2750 			 * freeing it when removing blocks.  Eventually, the
2751 			 * right edge of the truncated/punched region will
2752 			 * be just to the left.
2753 			 */
2754 			if (sbi->s_cluster_ratio > 1) {
2755 				pblk = ext4_ext_pblock(ex);
2756 				partial->pclu = EXT4_B2C(sbi, pblk);
2757 				partial->state = nofree;
2758 			}
2759 			ex--;
2760 			ex_ee_block = le32_to_cpu(ex->ee_block);
2761 			ex_ee_len = ext4_ext_get_actual_len(ex);
2762 			continue;
2763 		} else if (b != ex_ee_block + ex_ee_len - 1) {
2764 			EXT4_ERROR_INODE(inode,
2765 					 "can not handle truncate %u:%u "
2766 					 "on extent %u:%u",
2767 					 start, end, ex_ee_block,
2768 					 ex_ee_block + ex_ee_len - 1);
2769 			err = -EFSCORRUPTED;
2770 			goto out;
2771 		} else if (a != ex_ee_block) {
2772 			/* remove tail of the extent */
2773 			num = a - ex_ee_block;
2774 		} else {
2775 			/* remove whole extent: excellent! */
2776 			num = 0;
2777 		}
2778 		/*
2779 		 * 3 for leaf, sb, and inode plus 2 (bmap and group
2780 		 * descriptor) for each block group; assume two block
2781 		 * groups plus ex_ee_len/blocks_per_block_group for
2782 		 * the worst case
2783 		 */
2784 		credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2785 		if (ex == EXT_FIRST_EXTENT(eh)) {
2786 			correct_index = 1;
2787 			credits += (ext_depth(inode)) + 1;
2788 		}
2789 		credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2790 		/*
2791 		 * We may end up freeing some index blocks and data from the
2792 		 * punched range. Note that partial clusters are accounted for
2793 		 * by ext4_free_data_revoke_credits().
2794 		 */
2795 		revoke_credits =
2796 			ext4_free_metadata_revoke_credits(inode->i_sb,
2797 							  ext_depth(inode)) +
2798 			ext4_free_data_revoke_credits(inode, b - a + 1);
2799 
2800 		err = ext4_datasem_ensure_credits(handle, inode, credits,
2801 						  credits, revoke_credits);
2802 		if (err) {
2803 			if (err > 0)
2804 				err = -EAGAIN;
2805 			goto out;
2806 		}
2807 
2808 		err = ext4_ext_get_access(handle, inode, path + depth);
2809 		if (err)
2810 			goto out;
2811 
2812 		err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2813 		if (err)
2814 			goto out;
2815 
2816 		if (num == 0)
2817 			/* this extent is removed; mark slot entirely unused */
2818 			ext4_ext_store_pblock(ex, 0);
2819 
2820 		ex->ee_len = cpu_to_le16(num);
2821 		/*
2822 		 * Do not mark unwritten if all the blocks in the
2823 		 * extent have been removed.
2824 		 */
2825 		if (unwritten && num)
2826 			ext4_ext_mark_unwritten(ex);
2827 		/*
2828 		 * If the extent was completely released,
2829 		 * we need to remove it from the leaf
2830 		 */
2831 		if (num == 0) {
2832 			if (end != EXT_MAX_BLOCKS - 1) {
2833 				/*
2834 				 * For hole punching, we need to scoot all the
2835 				 * extents up when an extent is removed so that
2836 				 * we dont have blank extents in the middle
2837 				 */
2838 				memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2839 					sizeof(struct ext4_extent));
2840 
2841 				/* Now get rid of the one at the end */
2842 				memset(EXT_LAST_EXTENT(eh), 0,
2843 					sizeof(struct ext4_extent));
2844 			}
2845 			le16_add_cpu(&eh->eh_entries, -1);
2846 		}
2847 
2848 		err = ext4_ext_dirty(handle, inode, path + depth);
2849 		if (err)
2850 			goto out;
2851 
2852 		ext_debug("new extent: %u:%u:%llu\n", ex_ee_block, num,
2853 				ext4_ext_pblock(ex));
2854 		ex--;
2855 		ex_ee_block = le32_to_cpu(ex->ee_block);
2856 		ex_ee_len = ext4_ext_get_actual_len(ex);
2857 	}
2858 
2859 	if (correct_index && eh->eh_entries)
2860 		err = ext4_ext_correct_indexes(handle, inode, path);
2861 
2862 	/*
2863 	 * If there's a partial cluster and at least one extent remains in
2864 	 * the leaf, free the partial cluster if it isn't shared with the
2865 	 * current extent.  If it is shared with the current extent
2866 	 * we reset the partial cluster because we've reached the start of the
2867 	 * truncated/punched region and we're done removing blocks.
2868 	 */
2869 	if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2870 		pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2871 		if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2872 			int flags = get_default_free_blocks_flags(inode);
2873 
2874 			if (ext4_is_pending(inode, partial->lblk))
2875 				flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2876 			ext4_free_blocks(handle, inode, NULL,
2877 					 EXT4_C2B(sbi, partial->pclu),
2878 					 sbi->s_cluster_ratio, flags);
2879 			if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2880 				ext4_rereserve_cluster(inode, partial->lblk);
2881 		}
2882 		partial->state = initial;
2883 	}
2884 
2885 	/* if this leaf is free, then we should
2886 	 * remove it from index block above */
2887 	if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2888 		err = ext4_ext_rm_idx(handle, inode, path, depth);
2889 
2890 out:
2891 	return err;
2892 }
2893 
2894 /*
2895  * ext4_ext_more_to_rm:
2896  * returns 1 if current index has to be freed (even partial)
2897  */
2898 static int
2899 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2900 {
2901 	BUG_ON(path->p_idx == NULL);
2902 
2903 	if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2904 		return 0;
2905 
2906 	/*
2907 	 * if truncate on deeper level happened, it wasn't partial,
2908 	 * so we have to consider current index for truncation
2909 	 */
2910 	if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2911 		return 0;
2912 	return 1;
2913 }
2914 
2915 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2916 			  ext4_lblk_t end)
2917 {
2918 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2919 	int depth = ext_depth(inode);
2920 	struct ext4_ext_path *path = NULL;
2921 	struct partial_cluster partial;
2922 	handle_t *handle;
2923 	int i = 0, err = 0;
2924 
2925 	partial.pclu = 0;
2926 	partial.lblk = 0;
2927 	partial.state = initial;
2928 
2929 	ext_debug("truncate since %u to %u\n", start, end);
2930 
2931 	/* probably first extent we're gonna free will be last in block */
2932 	handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2933 			depth + 1,
2934 			ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2935 	if (IS_ERR(handle))
2936 		return PTR_ERR(handle);
2937 
2938 again:
2939 	trace_ext4_ext_remove_space(inode, start, end, depth);
2940 
2941 	/*
2942 	 * Check if we are removing extents inside the extent tree. If that
2943 	 * is the case, we are going to punch a hole inside the extent tree
2944 	 * so we have to check whether we need to split the extent covering
2945 	 * the last block to remove so we can easily remove the part of it
2946 	 * in ext4_ext_rm_leaf().
2947 	 */
2948 	if (end < EXT_MAX_BLOCKS - 1) {
2949 		struct ext4_extent *ex;
2950 		ext4_lblk_t ee_block, ex_end, lblk;
2951 		ext4_fsblk_t pblk;
2952 
2953 		/* find extent for or closest extent to this block */
2954 		path = ext4_find_extent(inode, end, NULL, EXT4_EX_NOCACHE);
2955 		if (IS_ERR(path)) {
2956 			ext4_journal_stop(handle);
2957 			return PTR_ERR(path);
2958 		}
2959 		depth = ext_depth(inode);
2960 		/* Leaf not may not exist only if inode has no blocks at all */
2961 		ex = path[depth].p_ext;
2962 		if (!ex) {
2963 			if (depth) {
2964 				EXT4_ERROR_INODE(inode,
2965 						 "path[%d].p_hdr == NULL",
2966 						 depth);
2967 				err = -EFSCORRUPTED;
2968 			}
2969 			goto out;
2970 		}
2971 
2972 		ee_block = le32_to_cpu(ex->ee_block);
2973 		ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2974 
2975 		/*
2976 		 * See if the last block is inside the extent, if so split
2977 		 * the extent at 'end' block so we can easily remove the
2978 		 * tail of the first part of the split extent in
2979 		 * ext4_ext_rm_leaf().
2980 		 */
2981 		if (end >= ee_block && end < ex_end) {
2982 
2983 			/*
2984 			 * If we're going to split the extent, note that
2985 			 * the cluster containing the block after 'end' is
2986 			 * in use to avoid freeing it when removing blocks.
2987 			 */
2988 			if (sbi->s_cluster_ratio > 1) {
2989 				pblk = ext4_ext_pblock(ex) + end - ee_block + 2;
2990 				partial.pclu = EXT4_B2C(sbi, pblk);
2991 				partial.state = nofree;
2992 			}
2993 
2994 			/*
2995 			 * Split the extent in two so that 'end' is the last
2996 			 * block in the first new extent. Also we should not
2997 			 * fail removing space due to ENOSPC so try to use
2998 			 * reserved block if that happens.
2999 			 */
3000 			err = ext4_force_split_extent_at(handle, inode, &path,
3001 							 end + 1, 1);
3002 			if (err < 0)
3003 				goto out;
3004 
3005 		} else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
3006 			   partial.state == initial) {
3007 			/*
3008 			 * If we're punching, there's an extent to the right.
3009 			 * If the partial cluster hasn't been set, set it to
3010 			 * that extent's first cluster and its state to nofree
3011 			 * so it won't be freed should it contain blocks to be
3012 			 * removed. If it's already set (tofree/nofree), we're
3013 			 * retrying and keep the original partial cluster info
3014 			 * so a cluster marked tofree as a result of earlier
3015 			 * extent removal is not lost.
3016 			 */
3017 			lblk = ex_end + 1;
3018 			err = ext4_ext_search_right(inode, path, &lblk, &pblk,
3019 						    &ex);
3020 			if (err)
3021 				goto out;
3022 			if (pblk) {
3023 				partial.pclu = EXT4_B2C(sbi, pblk);
3024 				partial.state = nofree;
3025 			}
3026 		}
3027 	}
3028 	/*
3029 	 * We start scanning from right side, freeing all the blocks
3030 	 * after i_size and walking into the tree depth-wise.
3031 	 */
3032 	depth = ext_depth(inode);
3033 	if (path) {
3034 		int k = i = depth;
3035 		while (--k > 0)
3036 			path[k].p_block =
3037 				le16_to_cpu(path[k].p_hdr->eh_entries)+1;
3038 	} else {
3039 		path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
3040 			       GFP_NOFS);
3041 		if (path == NULL) {
3042 			ext4_journal_stop(handle);
3043 			return -ENOMEM;
3044 		}
3045 		path[0].p_maxdepth = path[0].p_depth = depth;
3046 		path[0].p_hdr = ext_inode_hdr(inode);
3047 		i = 0;
3048 
3049 		if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
3050 			err = -EFSCORRUPTED;
3051 			goto out;
3052 		}
3053 	}
3054 	err = 0;
3055 
3056 	while (i >= 0 && err == 0) {
3057 		if (i == depth) {
3058 			/* this is leaf block */
3059 			err = ext4_ext_rm_leaf(handle, inode, path,
3060 					       &partial, start, end);
3061 			/* root level has p_bh == NULL, brelse() eats this */
3062 			brelse(path[i].p_bh);
3063 			path[i].p_bh = NULL;
3064 			i--;
3065 			continue;
3066 		}
3067 
3068 		/* this is index block */
3069 		if (!path[i].p_hdr) {
3070 			ext_debug("initialize header\n");
3071 			path[i].p_hdr = ext_block_hdr(path[i].p_bh);
3072 		}
3073 
3074 		if (!path[i].p_idx) {
3075 			/* this level hasn't been touched yet */
3076 			path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
3077 			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
3078 			ext_debug("init index ptr: hdr 0x%p, num %d\n",
3079 				  path[i].p_hdr,
3080 				  le16_to_cpu(path[i].p_hdr->eh_entries));
3081 		} else {
3082 			/* we were already here, see at next index */
3083 			path[i].p_idx--;
3084 		}
3085 
3086 		ext_debug("level %d - index, first 0x%p, cur 0x%p\n",
3087 				i, EXT_FIRST_INDEX(path[i].p_hdr),
3088 				path[i].p_idx);
3089 		if (ext4_ext_more_to_rm(path + i)) {
3090 			struct buffer_head *bh;
3091 			/* go to the next level */
3092 			ext_debug("move to level %d (block %llu)\n",
3093 				  i + 1, ext4_idx_pblock(path[i].p_idx));
3094 			memset(path + i + 1, 0, sizeof(*path));
3095 			bh = read_extent_tree_block(inode,
3096 				ext4_idx_pblock(path[i].p_idx), depth - i - 1,
3097 				EXT4_EX_NOCACHE);
3098 			if (IS_ERR(bh)) {
3099 				/* should we reset i_size? */
3100 				err = PTR_ERR(bh);
3101 				break;
3102 			}
3103 			/* Yield here to deal with large extent trees.
3104 			 * Should be a no-op if we did IO above. */
3105 			cond_resched();
3106 			if (WARN_ON(i + 1 > depth)) {
3107 				err = -EFSCORRUPTED;
3108 				break;
3109 			}
3110 			path[i + 1].p_bh = bh;
3111 
3112 			/* save actual number of indexes since this
3113 			 * number is changed at the next iteration */
3114 			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3115 			i++;
3116 		} else {
3117 			/* we finished processing this index, go up */
3118 			if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3119 				/* index is empty, remove it;
3120 				 * handle must be already prepared by the
3121 				 * truncatei_leaf() */
3122 				err = ext4_ext_rm_idx(handle, inode, path, i);
3123 			}
3124 			/* root level has p_bh == NULL, brelse() eats this */
3125 			brelse(path[i].p_bh);
3126 			path[i].p_bh = NULL;
3127 			i--;
3128 			ext_debug("return to level %d\n", i);
3129 		}
3130 	}
3131 
3132 	trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3133 					 path->p_hdr->eh_entries);
3134 
3135 	/*
3136 	 * if there's a partial cluster and we have removed the first extent
3137 	 * in the file, then we also free the partial cluster, if any
3138 	 */
3139 	if (partial.state == tofree && err == 0) {
3140 		int flags = get_default_free_blocks_flags(inode);
3141 
3142 		if (ext4_is_pending(inode, partial.lblk))
3143 			flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3144 		ext4_free_blocks(handle, inode, NULL,
3145 				 EXT4_C2B(sbi, partial.pclu),
3146 				 sbi->s_cluster_ratio, flags);
3147 		if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3148 			ext4_rereserve_cluster(inode, partial.lblk);
3149 		partial.state = initial;
3150 	}
3151 
3152 	/* TODO: flexible tree reduction should be here */
3153 	if (path->p_hdr->eh_entries == 0) {
3154 		/*
3155 		 * truncate to zero freed all the tree,
3156 		 * so we need to correct eh_depth
3157 		 */
3158 		err = ext4_ext_get_access(handle, inode, path);
3159 		if (err == 0) {
3160 			ext_inode_hdr(inode)->eh_depth = 0;
3161 			ext_inode_hdr(inode)->eh_max =
3162 				cpu_to_le16(ext4_ext_space_root(inode, 0));
3163 			err = ext4_ext_dirty(handle, inode, path);
3164 		}
3165 	}
3166 out:
3167 	ext4_ext_drop_refs(path);
3168 	kfree(path);
3169 	path = NULL;
3170 	if (err == -EAGAIN)
3171 		goto again;
3172 	ext4_journal_stop(handle);
3173 
3174 	return err;
3175 }
3176 
3177 /*
3178  * called at mount time
3179  */
3180 void ext4_ext_init(struct super_block *sb)
3181 {
3182 	/*
3183 	 * possible initialization would be here
3184 	 */
3185 
3186 	if (ext4_has_feature_extents(sb)) {
3187 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3188 		printk(KERN_INFO "EXT4-fs: file extents enabled"
3189 #ifdef AGGRESSIVE_TEST
3190 		       ", aggressive tests"
3191 #endif
3192 #ifdef CHECK_BINSEARCH
3193 		       ", check binsearch"
3194 #endif
3195 #ifdef EXTENTS_STATS
3196 		       ", stats"
3197 #endif
3198 		       "\n");
3199 #endif
3200 #ifdef EXTENTS_STATS
3201 		spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3202 		EXT4_SB(sb)->s_ext_min = 1 << 30;
3203 		EXT4_SB(sb)->s_ext_max = 0;
3204 #endif
3205 	}
3206 }
3207 
3208 /*
3209  * called at umount time
3210  */
3211 void ext4_ext_release(struct super_block *sb)
3212 {
3213 	if (!ext4_has_feature_extents(sb))
3214 		return;
3215 
3216 #ifdef EXTENTS_STATS
3217 	if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3218 		struct ext4_sb_info *sbi = EXT4_SB(sb);
3219 		printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3220 			sbi->s_ext_blocks, sbi->s_ext_extents,
3221 			sbi->s_ext_blocks / sbi->s_ext_extents);
3222 		printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3223 			sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3224 	}
3225 #endif
3226 }
3227 
3228 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3229 {
3230 	ext4_lblk_t  ee_block;
3231 	ext4_fsblk_t ee_pblock;
3232 	unsigned int ee_len;
3233 
3234 	ee_block  = le32_to_cpu(ex->ee_block);
3235 	ee_len    = ext4_ext_get_actual_len(ex);
3236 	ee_pblock = ext4_ext_pblock(ex);
3237 
3238 	if (ee_len == 0)
3239 		return 0;
3240 
3241 	return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3242 				     EXTENT_STATUS_WRITTEN);
3243 }
3244 
3245 /* FIXME!! we need to try to merge to left or right after zero-out  */
3246 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3247 {
3248 	ext4_fsblk_t ee_pblock;
3249 	unsigned int ee_len;
3250 
3251 	ee_len    = ext4_ext_get_actual_len(ex);
3252 	ee_pblock = ext4_ext_pblock(ex);
3253 	return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3254 				  ee_len);
3255 }
3256 
3257 /*
3258  * ext4_split_extent_at() splits an extent at given block.
3259  *
3260  * @handle: the journal handle
3261  * @inode: the file inode
3262  * @path: the path to the extent
3263  * @split: the logical block where the extent is splitted.
3264  * @split_flags: indicates if the extent could be zeroout if split fails, and
3265  *		 the states(init or unwritten) of new extents.
3266  * @flags: flags used to insert new extent to extent tree.
3267  *
3268  *
3269  * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3270  * of which are deterimined by split_flag.
3271  *
3272  * There are two cases:
3273  *  a> the extent are splitted into two extent.
3274  *  b> split is not needed, and just mark the extent.
3275  *
3276  * return 0 on success.
3277  */
3278 static int ext4_split_extent_at(handle_t *handle,
3279 			     struct inode *inode,
3280 			     struct ext4_ext_path **ppath,
3281 			     ext4_lblk_t split,
3282 			     int split_flag,
3283 			     int flags)
3284 {
3285 	struct ext4_ext_path *path = *ppath;
3286 	ext4_fsblk_t newblock;
3287 	ext4_lblk_t ee_block;
3288 	struct ext4_extent *ex, newex, orig_ex, zero_ex;
3289 	struct ext4_extent *ex2 = NULL;
3290 	unsigned int ee_len, depth;
3291 	int err = 0;
3292 
3293 	BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3294 	       (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3295 
3296 	ext_debug("ext4_split_extents_at: inode %lu, logical"
3297 		"block %llu\n", inode->i_ino, (unsigned long long)split);
3298 
3299 	ext4_ext_show_leaf(inode, path);
3300 
3301 	depth = ext_depth(inode);
3302 	ex = path[depth].p_ext;
3303 	ee_block = le32_to_cpu(ex->ee_block);
3304 	ee_len = ext4_ext_get_actual_len(ex);
3305 	newblock = split - ee_block + ext4_ext_pblock(ex);
3306 
3307 	BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3308 	BUG_ON(!ext4_ext_is_unwritten(ex) &&
3309 	       split_flag & (EXT4_EXT_MAY_ZEROOUT |
3310 			     EXT4_EXT_MARK_UNWRIT1 |
3311 			     EXT4_EXT_MARK_UNWRIT2));
3312 
3313 	err = ext4_ext_get_access(handle, inode, path + depth);
3314 	if (err)
3315 		goto out;
3316 
3317 	if (split == ee_block) {
3318 		/*
3319 		 * case b: block @split is the block that the extent begins with
3320 		 * then we just change the state of the extent, and splitting
3321 		 * is not needed.
3322 		 */
3323 		if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3324 			ext4_ext_mark_unwritten(ex);
3325 		else
3326 			ext4_ext_mark_initialized(ex);
3327 
3328 		if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3329 			ext4_ext_try_to_merge(handle, inode, path, ex);
3330 
3331 		err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3332 		goto out;
3333 	}
3334 
3335 	/* case a */
3336 	memcpy(&orig_ex, ex, sizeof(orig_ex));
3337 	ex->ee_len = cpu_to_le16(split - ee_block);
3338 	if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3339 		ext4_ext_mark_unwritten(ex);
3340 
3341 	/*
3342 	 * path may lead to new leaf, not to original leaf any more
3343 	 * after ext4_ext_insert_extent() returns,
3344 	 */
3345 	err = ext4_ext_dirty(handle, inode, path + depth);
3346 	if (err)
3347 		goto fix_extent_len;
3348 
3349 	ex2 = &newex;
3350 	ex2->ee_block = cpu_to_le32(split);
3351 	ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
3352 	ext4_ext_store_pblock(ex2, newblock);
3353 	if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3354 		ext4_ext_mark_unwritten(ex2);
3355 
3356 	err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3357 	if (err == -ENOSPC && (EXT4_EXT_MAY_ZEROOUT & split_flag)) {
3358 		if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3359 			if (split_flag & EXT4_EXT_DATA_VALID1) {
3360 				err = ext4_ext_zeroout(inode, ex2);
3361 				zero_ex.ee_block = ex2->ee_block;
3362 				zero_ex.ee_len = cpu_to_le16(
3363 						ext4_ext_get_actual_len(ex2));
3364 				ext4_ext_store_pblock(&zero_ex,
3365 						      ext4_ext_pblock(ex2));
3366 			} else {
3367 				err = ext4_ext_zeroout(inode, ex);
3368 				zero_ex.ee_block = ex->ee_block;
3369 				zero_ex.ee_len = cpu_to_le16(
3370 						ext4_ext_get_actual_len(ex));
3371 				ext4_ext_store_pblock(&zero_ex,
3372 						      ext4_ext_pblock(ex));
3373 			}
3374 		} else {
3375 			err = ext4_ext_zeroout(inode, &orig_ex);
3376 			zero_ex.ee_block = orig_ex.ee_block;
3377 			zero_ex.ee_len = cpu_to_le16(
3378 						ext4_ext_get_actual_len(&orig_ex));
3379 			ext4_ext_store_pblock(&zero_ex,
3380 					      ext4_ext_pblock(&orig_ex));
3381 		}
3382 
3383 		if (err)
3384 			goto fix_extent_len;
3385 		/* update the extent length and mark as initialized */
3386 		ex->ee_len = cpu_to_le16(ee_len);
3387 		ext4_ext_try_to_merge(handle, inode, path, ex);
3388 		err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3389 		if (err)
3390 			goto fix_extent_len;
3391 
3392 		/* update extent status tree */
3393 		err = ext4_zeroout_es(inode, &zero_ex);
3394 
3395 		goto out;
3396 	} else if (err)
3397 		goto fix_extent_len;
3398 
3399 out:
3400 	ext4_ext_show_leaf(inode, path);
3401 	return err;
3402 
3403 fix_extent_len:
3404 	ex->ee_len = orig_ex.ee_len;
3405 	ext4_ext_dirty(handle, inode, path + path->p_depth);
3406 	return err;
3407 }
3408 
3409 /*
3410  * ext4_split_extents() splits an extent and mark extent which is covered
3411  * by @map as split_flags indicates
3412  *
3413  * It may result in splitting the extent into multiple extents (up to three)
3414  * There are three possibilities:
3415  *   a> There is no split required
3416  *   b> Splits in two extents: Split is happening at either end of the extent
3417  *   c> Splits in three extents: Somone is splitting in middle of the extent
3418  *
3419  */
3420 static int ext4_split_extent(handle_t *handle,
3421 			      struct inode *inode,
3422 			      struct ext4_ext_path **ppath,
3423 			      struct ext4_map_blocks *map,
3424 			      int split_flag,
3425 			      int flags)
3426 {
3427 	struct ext4_ext_path *path = *ppath;
3428 	ext4_lblk_t ee_block;
3429 	struct ext4_extent *ex;
3430 	unsigned int ee_len, depth;
3431 	int err = 0;
3432 	int unwritten;
3433 	int split_flag1, flags1;
3434 	int allocated = map->m_len;
3435 
3436 	depth = ext_depth(inode);
3437 	ex = path[depth].p_ext;
3438 	ee_block = le32_to_cpu(ex->ee_block);
3439 	ee_len = ext4_ext_get_actual_len(ex);
3440 	unwritten = ext4_ext_is_unwritten(ex);
3441 
3442 	if (map->m_lblk + map->m_len < ee_block + ee_len) {
3443 		split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3444 		flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3445 		if (unwritten)
3446 			split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3447 				       EXT4_EXT_MARK_UNWRIT2;
3448 		if (split_flag & EXT4_EXT_DATA_VALID2)
3449 			split_flag1 |= EXT4_EXT_DATA_VALID1;
3450 		err = ext4_split_extent_at(handle, inode, ppath,
3451 				map->m_lblk + map->m_len, split_flag1, flags1);
3452 		if (err)
3453 			goto out;
3454 	} else {
3455 		allocated = ee_len - (map->m_lblk - ee_block);
3456 	}
3457 	/*
3458 	 * Update path is required because previous ext4_split_extent_at() may
3459 	 * result in split of original leaf or extent zeroout.
3460 	 */
3461 	path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3462 	if (IS_ERR(path))
3463 		return PTR_ERR(path);
3464 	depth = ext_depth(inode);
3465 	ex = path[depth].p_ext;
3466 	if (!ex) {
3467 		EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3468 				 (unsigned long) map->m_lblk);
3469 		return -EFSCORRUPTED;
3470 	}
3471 	unwritten = ext4_ext_is_unwritten(ex);
3472 	split_flag1 = 0;
3473 
3474 	if (map->m_lblk >= ee_block) {
3475 		split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3476 		if (unwritten) {
3477 			split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3478 			split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3479 						     EXT4_EXT_MARK_UNWRIT2);
3480 		}
3481 		err = ext4_split_extent_at(handle, inode, ppath,
3482 				map->m_lblk, split_flag1, flags);
3483 		if (err)
3484 			goto out;
3485 	}
3486 
3487 	ext4_ext_show_leaf(inode, path);
3488 out:
3489 	return err ? err : allocated;
3490 }
3491 
3492 /*
3493  * This function is called by ext4_ext_map_blocks() if someone tries to write
3494  * to an unwritten extent. It may result in splitting the unwritten
3495  * extent into multiple extents (up to three - one initialized and two
3496  * unwritten).
3497  * There are three possibilities:
3498  *   a> There is no split required: Entire extent should be initialized
3499  *   b> Splits in two extents: Write is happening at either end of the extent
3500  *   c> Splits in three extents: Somone is writing in middle of the extent
3501  *
3502  * Pre-conditions:
3503  *  - The extent pointed to by 'path' is unwritten.
3504  *  - The extent pointed to by 'path' contains a superset
3505  *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3506  *
3507  * Post-conditions on success:
3508  *  - the returned value is the number of blocks beyond map->l_lblk
3509  *    that are allocated and initialized.
3510  *    It is guaranteed to be >= map->m_len.
3511  */
3512 static int ext4_ext_convert_to_initialized(handle_t *handle,
3513 					   struct inode *inode,
3514 					   struct ext4_map_blocks *map,
3515 					   struct ext4_ext_path **ppath,
3516 					   int flags)
3517 {
3518 	struct ext4_ext_path *path = *ppath;
3519 	struct ext4_sb_info *sbi;
3520 	struct ext4_extent_header *eh;
3521 	struct ext4_map_blocks split_map;
3522 	struct ext4_extent zero_ex1, zero_ex2;
3523 	struct ext4_extent *ex, *abut_ex;
3524 	ext4_lblk_t ee_block, eof_block;
3525 	unsigned int ee_len, depth, map_len = map->m_len;
3526 	int allocated = 0, max_zeroout = 0;
3527 	int err = 0;
3528 	int split_flag = EXT4_EXT_DATA_VALID2;
3529 
3530 	ext_debug("ext4_ext_convert_to_initialized: inode %lu, logical"
3531 		"block %llu, max_blocks %u\n", inode->i_ino,
3532 		(unsigned long long)map->m_lblk, map_len);
3533 
3534 	sbi = EXT4_SB(inode->i_sb);
3535 	eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3536 		inode->i_sb->s_blocksize_bits;
3537 	if (eof_block < map->m_lblk + map_len)
3538 		eof_block = map->m_lblk + map_len;
3539 
3540 	depth = ext_depth(inode);
3541 	eh = path[depth].p_hdr;
3542 	ex = path[depth].p_ext;
3543 	ee_block = le32_to_cpu(ex->ee_block);
3544 	ee_len = ext4_ext_get_actual_len(ex);
3545 	zero_ex1.ee_len = 0;
3546 	zero_ex2.ee_len = 0;
3547 
3548 	trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3549 
3550 	/* Pre-conditions */
3551 	BUG_ON(!ext4_ext_is_unwritten(ex));
3552 	BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3553 
3554 	/*
3555 	 * Attempt to transfer newly initialized blocks from the currently
3556 	 * unwritten extent to its neighbor. This is much cheaper
3557 	 * than an insertion followed by a merge as those involve costly
3558 	 * memmove() calls. Transferring to the left is the common case in
3559 	 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3560 	 * followed by append writes.
3561 	 *
3562 	 * Limitations of the current logic:
3563 	 *  - L1: we do not deal with writes covering the whole extent.
3564 	 *    This would require removing the extent if the transfer
3565 	 *    is possible.
3566 	 *  - L2: we only attempt to merge with an extent stored in the
3567 	 *    same extent tree node.
3568 	 */
3569 	if ((map->m_lblk == ee_block) &&
3570 		/* See if we can merge left */
3571 		(map_len < ee_len) &&		/*L1*/
3572 		(ex > EXT_FIRST_EXTENT(eh))) {	/*L2*/
3573 		ext4_lblk_t prev_lblk;
3574 		ext4_fsblk_t prev_pblk, ee_pblk;
3575 		unsigned int prev_len;
3576 
3577 		abut_ex = ex - 1;
3578 		prev_lblk = le32_to_cpu(abut_ex->ee_block);
3579 		prev_len = ext4_ext_get_actual_len(abut_ex);
3580 		prev_pblk = ext4_ext_pblock(abut_ex);
3581 		ee_pblk = ext4_ext_pblock(ex);
3582 
3583 		/*
3584 		 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3585 		 * upon those conditions:
3586 		 * - C1: abut_ex is initialized,
3587 		 * - C2: abut_ex is logically abutting ex,
3588 		 * - C3: abut_ex is physically abutting ex,
3589 		 * - C4: abut_ex can receive the additional blocks without
3590 		 *   overflowing the (initialized) length limit.
3591 		 */
3592 		if ((!ext4_ext_is_unwritten(abut_ex)) &&		/*C1*/
3593 			((prev_lblk + prev_len) == ee_block) &&		/*C2*/
3594 			((prev_pblk + prev_len) == ee_pblk) &&		/*C3*/
3595 			(prev_len < (EXT_INIT_MAX_LEN - map_len))) {	/*C4*/
3596 			err = ext4_ext_get_access(handle, inode, path + depth);
3597 			if (err)
3598 				goto out;
3599 
3600 			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3601 				map, ex, abut_ex);
3602 
3603 			/* Shift the start of ex by 'map_len' blocks */
3604 			ex->ee_block = cpu_to_le32(ee_block + map_len);
3605 			ext4_ext_store_pblock(ex, ee_pblk + map_len);
3606 			ex->ee_len = cpu_to_le16(ee_len - map_len);
3607 			ext4_ext_mark_unwritten(ex); /* Restore the flag */
3608 
3609 			/* Extend abut_ex by 'map_len' blocks */
3610 			abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3611 
3612 			/* Result: number of initialized blocks past m_lblk */
3613 			allocated = map_len;
3614 		}
3615 	} else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3616 		   (map_len < ee_len) &&	/*L1*/
3617 		   ex < EXT_LAST_EXTENT(eh)) {	/*L2*/
3618 		/* See if we can merge right */
3619 		ext4_lblk_t next_lblk;
3620 		ext4_fsblk_t next_pblk, ee_pblk;
3621 		unsigned int next_len;
3622 
3623 		abut_ex = ex + 1;
3624 		next_lblk = le32_to_cpu(abut_ex->ee_block);
3625 		next_len = ext4_ext_get_actual_len(abut_ex);
3626 		next_pblk = ext4_ext_pblock(abut_ex);
3627 		ee_pblk = ext4_ext_pblock(ex);
3628 
3629 		/*
3630 		 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3631 		 * upon those conditions:
3632 		 * - C1: abut_ex is initialized,
3633 		 * - C2: abut_ex is logically abutting ex,
3634 		 * - C3: abut_ex is physically abutting ex,
3635 		 * - C4: abut_ex can receive the additional blocks without
3636 		 *   overflowing the (initialized) length limit.
3637 		 */
3638 		if ((!ext4_ext_is_unwritten(abut_ex)) &&		/*C1*/
3639 		    ((map->m_lblk + map_len) == next_lblk) &&		/*C2*/
3640 		    ((ee_pblk + ee_len) == next_pblk) &&		/*C3*/
3641 		    (next_len < (EXT_INIT_MAX_LEN - map_len))) {	/*C4*/
3642 			err = ext4_ext_get_access(handle, inode, path + depth);
3643 			if (err)
3644 				goto out;
3645 
3646 			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3647 				map, ex, abut_ex);
3648 
3649 			/* Shift the start of abut_ex by 'map_len' blocks */
3650 			abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3651 			ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3652 			ex->ee_len = cpu_to_le16(ee_len - map_len);
3653 			ext4_ext_mark_unwritten(ex); /* Restore the flag */
3654 
3655 			/* Extend abut_ex by 'map_len' blocks */
3656 			abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3657 
3658 			/* Result: number of initialized blocks past m_lblk */
3659 			allocated = map_len;
3660 		}
3661 	}
3662 	if (allocated) {
3663 		/* Mark the block containing both extents as dirty */
3664 		ext4_ext_dirty(handle, inode, path + depth);
3665 
3666 		/* Update path to point to the right extent */
3667 		path[depth].p_ext = abut_ex;
3668 		goto out;
3669 	} else
3670 		allocated = ee_len - (map->m_lblk - ee_block);
3671 
3672 	WARN_ON(map->m_lblk < ee_block);
3673 	/*
3674 	 * It is safe to convert extent to initialized via explicit
3675 	 * zeroout only if extent is fully inside i_size or new_size.
3676 	 */
3677 	split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3678 
3679 	if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3680 		max_zeroout = sbi->s_extent_max_zeroout_kb >>
3681 			(inode->i_sb->s_blocksize_bits - 10);
3682 
3683 	/*
3684 	 * five cases:
3685 	 * 1. split the extent into three extents.
3686 	 * 2. split the extent into two extents, zeroout the head of the first
3687 	 *    extent.
3688 	 * 3. split the extent into two extents, zeroout the tail of the second
3689 	 *    extent.
3690 	 * 4. split the extent into two extents with out zeroout.
3691 	 * 5. no splitting needed, just possibly zeroout the head and / or the
3692 	 *    tail of the extent.
3693 	 */
3694 	split_map.m_lblk = map->m_lblk;
3695 	split_map.m_len = map->m_len;
3696 
3697 	if (max_zeroout && (allocated > split_map.m_len)) {
3698 		if (allocated <= max_zeroout) {
3699 			/* case 3 or 5 */
3700 			zero_ex1.ee_block =
3701 				 cpu_to_le32(split_map.m_lblk +
3702 					     split_map.m_len);
3703 			zero_ex1.ee_len =
3704 				cpu_to_le16(allocated - split_map.m_len);
3705 			ext4_ext_store_pblock(&zero_ex1,
3706 				ext4_ext_pblock(ex) + split_map.m_lblk +
3707 				split_map.m_len - ee_block);
3708 			err = ext4_ext_zeroout(inode, &zero_ex1);
3709 			if (err)
3710 				goto out;
3711 			split_map.m_len = allocated;
3712 		}
3713 		if (split_map.m_lblk - ee_block + split_map.m_len <
3714 								max_zeroout) {
3715 			/* case 2 or 5 */
3716 			if (split_map.m_lblk != ee_block) {
3717 				zero_ex2.ee_block = ex->ee_block;
3718 				zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3719 							ee_block);
3720 				ext4_ext_store_pblock(&zero_ex2,
3721 						      ext4_ext_pblock(ex));
3722 				err = ext4_ext_zeroout(inode, &zero_ex2);
3723 				if (err)
3724 					goto out;
3725 			}
3726 
3727 			split_map.m_len += split_map.m_lblk - ee_block;
3728 			split_map.m_lblk = ee_block;
3729 			allocated = map->m_len;
3730 		}
3731 	}
3732 
3733 	err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3734 				flags);
3735 	if (err > 0)
3736 		err = 0;
3737 out:
3738 	/* If we have gotten a failure, don't zero out status tree */
3739 	if (!err) {
3740 		err = ext4_zeroout_es(inode, &zero_ex1);
3741 		if (!err)
3742 			err = ext4_zeroout_es(inode, &zero_ex2);
3743 	}
3744 	return err ? err : allocated;
3745 }
3746 
3747 /*
3748  * This function is called by ext4_ext_map_blocks() from
3749  * ext4_get_blocks_dio_write() when DIO to write
3750  * to an unwritten extent.
3751  *
3752  * Writing to an unwritten extent may result in splitting the unwritten
3753  * extent into multiple initialized/unwritten extents (up to three)
3754  * There are three possibilities:
3755  *   a> There is no split required: Entire extent should be unwritten
3756  *   b> Splits in two extents: Write is happening at either end of the extent
3757  *   c> Splits in three extents: Somone is writing in middle of the extent
3758  *
3759  * This works the same way in the case of initialized -> unwritten conversion.
3760  *
3761  * One of more index blocks maybe needed if the extent tree grow after
3762  * the unwritten extent split. To prevent ENOSPC occur at the IO
3763  * complete, we need to split the unwritten extent before DIO submit
3764  * the IO. The unwritten extent called at this time will be split
3765  * into three unwritten extent(at most). After IO complete, the part
3766  * being filled will be convert to initialized by the end_io callback function
3767  * via ext4_convert_unwritten_extents().
3768  *
3769  * Returns the size of unwritten extent to be written on success.
3770  */
3771 static int ext4_split_convert_extents(handle_t *handle,
3772 					struct inode *inode,
3773 					struct ext4_map_blocks *map,
3774 					struct ext4_ext_path **ppath,
3775 					int flags)
3776 {
3777 	struct ext4_ext_path *path = *ppath;
3778 	ext4_lblk_t eof_block;
3779 	ext4_lblk_t ee_block;
3780 	struct ext4_extent *ex;
3781 	unsigned int ee_len;
3782 	int split_flag = 0, depth;
3783 
3784 	ext_debug("%s: inode %lu, logical block %llu, max_blocks %u\n",
3785 		  __func__, inode->i_ino,
3786 		  (unsigned long long)map->m_lblk, map->m_len);
3787 
3788 	eof_block = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
3789 		inode->i_sb->s_blocksize_bits;
3790 	if (eof_block < map->m_lblk + map->m_len)
3791 		eof_block = map->m_lblk + map->m_len;
3792 	/*
3793 	 * It is safe to convert extent to initialized via explicit
3794 	 * zeroout only if extent is fully insde i_size or new_size.
3795 	 */
3796 	depth = ext_depth(inode);
3797 	ex = path[depth].p_ext;
3798 	ee_block = le32_to_cpu(ex->ee_block);
3799 	ee_len = ext4_ext_get_actual_len(ex);
3800 
3801 	/* Convert to unwritten */
3802 	if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3803 		split_flag |= EXT4_EXT_DATA_VALID1;
3804 	/* Convert to initialized */
3805 	} else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3806 		split_flag |= ee_block + ee_len <= eof_block ?
3807 			      EXT4_EXT_MAY_ZEROOUT : 0;
3808 		split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3809 	}
3810 	flags |= EXT4_GET_BLOCKS_PRE_IO;
3811 	return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3812 }
3813 
3814 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3815 						struct inode *inode,
3816 						struct ext4_map_blocks *map,
3817 						struct ext4_ext_path **ppath)
3818 {
3819 	struct ext4_ext_path *path = *ppath;
3820 	struct ext4_extent *ex;
3821 	ext4_lblk_t ee_block;
3822 	unsigned int ee_len;
3823 	int depth;
3824 	int err = 0;
3825 
3826 	depth = ext_depth(inode);
3827 	ex = path[depth].p_ext;
3828 	ee_block = le32_to_cpu(ex->ee_block);
3829 	ee_len = ext4_ext_get_actual_len(ex);
3830 
3831 	ext_debug("ext4_convert_unwritten_extents_endio: inode %lu, logical"
3832 		"block %llu, max_blocks %u\n", inode->i_ino,
3833 		  (unsigned long long)ee_block, ee_len);
3834 
3835 	/* If extent is larger than requested it is a clear sign that we still
3836 	 * have some extent state machine issues left. So extent_split is still
3837 	 * required.
3838 	 * TODO: Once all related issues will be fixed this situation should be
3839 	 * illegal.
3840 	 */
3841 	if (ee_block != map->m_lblk || ee_len > map->m_len) {
3842 #ifdef CONFIG_EXT4_DEBUG
3843 		ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3844 			     " len %u; IO logical block %llu, len %u",
3845 			     inode->i_ino, (unsigned long long)ee_block, ee_len,
3846 			     (unsigned long long)map->m_lblk, map->m_len);
3847 #endif
3848 		err = ext4_split_convert_extents(handle, inode, map, ppath,
3849 						 EXT4_GET_BLOCKS_CONVERT);
3850 		if (err < 0)
3851 			return err;
3852 		path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3853 		if (IS_ERR(path))
3854 			return PTR_ERR(path);
3855 		depth = ext_depth(inode);
3856 		ex = path[depth].p_ext;
3857 	}
3858 
3859 	err = ext4_ext_get_access(handle, inode, path + depth);
3860 	if (err)
3861 		goto out;
3862 	/* first mark the extent as initialized */
3863 	ext4_ext_mark_initialized(ex);
3864 
3865 	/* note: ext4_ext_correct_indexes() isn't needed here because
3866 	 * borders are not changed
3867 	 */
3868 	ext4_ext_try_to_merge(handle, inode, path, ex);
3869 
3870 	/* Mark modified extent as dirty */
3871 	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3872 out:
3873 	ext4_ext_show_leaf(inode, path);
3874 	return err;
3875 }
3876 
3877 /*
3878  * Handle EOFBLOCKS_FL flag, clearing it if necessary
3879  */
3880 static int check_eofblocks_fl(handle_t *handle, struct inode *inode,
3881 			      ext4_lblk_t lblk,
3882 			      struct ext4_ext_path *path,
3883 			      unsigned int len)
3884 {
3885 	int i, depth;
3886 	struct ext4_extent_header *eh;
3887 	struct ext4_extent *last_ex;
3888 
3889 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EOFBLOCKS))
3890 		return 0;
3891 
3892 	depth = ext_depth(inode);
3893 	eh = path[depth].p_hdr;
3894 
3895 	/*
3896 	 * We're going to remove EOFBLOCKS_FL entirely in future so we
3897 	 * do not care for this case anymore. Simply remove the flag
3898 	 * if there are no extents.
3899 	 */
3900 	if (unlikely(!eh->eh_entries))
3901 		goto out;
3902 	last_ex = EXT_LAST_EXTENT(eh);
3903 	/*
3904 	 * We should clear the EOFBLOCKS_FL flag if we are writing the
3905 	 * last block in the last extent in the file.  We test this by
3906 	 * first checking to see if the caller to
3907 	 * ext4_ext_get_blocks() was interested in the last block (or
3908 	 * a block beyond the last block) in the current extent.  If
3909 	 * this turns out to be false, we can bail out from this
3910 	 * function immediately.
3911 	 */
3912 	if (lblk + len < le32_to_cpu(last_ex->ee_block) +
3913 	    ext4_ext_get_actual_len(last_ex))
3914 		return 0;
3915 	/*
3916 	 * If the caller does appear to be planning to write at or
3917 	 * beyond the end of the current extent, we then test to see
3918 	 * if the current extent is the last extent in the file, by
3919 	 * checking to make sure it was reached via the rightmost node
3920 	 * at each level of the tree.
3921 	 */
3922 	for (i = depth-1; i >= 0; i--)
3923 		if (path[i].p_idx != EXT_LAST_INDEX(path[i].p_hdr))
3924 			return 0;
3925 out:
3926 	ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
3927 	return ext4_mark_inode_dirty(handle, inode);
3928 }
3929 
3930 static int
3931 convert_initialized_extent(handle_t *handle, struct inode *inode,
3932 			   struct ext4_map_blocks *map,
3933 			   struct ext4_ext_path **ppath,
3934 			   unsigned int allocated)
3935 {
3936 	struct ext4_ext_path *path = *ppath;
3937 	struct ext4_extent *ex;
3938 	ext4_lblk_t ee_block;
3939 	unsigned int ee_len;
3940 	int depth;
3941 	int err = 0;
3942 
3943 	/*
3944 	 * Make sure that the extent is no bigger than we support with
3945 	 * unwritten extent
3946 	 */
3947 	if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3948 		map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3949 
3950 	depth = ext_depth(inode);
3951 	ex = path[depth].p_ext;
3952 	ee_block = le32_to_cpu(ex->ee_block);
3953 	ee_len = ext4_ext_get_actual_len(ex);
3954 
3955 	ext_debug("%s: inode %lu, logical"
3956 		"block %llu, max_blocks %u\n", __func__, inode->i_ino,
3957 		  (unsigned long long)ee_block, ee_len);
3958 
3959 	if (ee_block != map->m_lblk || ee_len > map->m_len) {
3960 		err = ext4_split_convert_extents(handle, inode, map, ppath,
3961 				EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3962 		if (err < 0)
3963 			return err;
3964 		path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3965 		if (IS_ERR(path))
3966 			return PTR_ERR(path);
3967 		depth = ext_depth(inode);
3968 		ex = path[depth].p_ext;
3969 		if (!ex) {
3970 			EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3971 					 (unsigned long) map->m_lblk);
3972 			return -EFSCORRUPTED;
3973 		}
3974 	}
3975 
3976 	err = ext4_ext_get_access(handle, inode, path + depth);
3977 	if (err)
3978 		return err;
3979 	/* first mark the extent as unwritten */
3980 	ext4_ext_mark_unwritten(ex);
3981 
3982 	/* note: ext4_ext_correct_indexes() isn't needed here because
3983 	 * borders are not changed
3984 	 */
3985 	ext4_ext_try_to_merge(handle, inode, path, ex);
3986 
3987 	/* Mark modified extent as dirty */
3988 	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3989 	if (err)
3990 		return err;
3991 	ext4_ext_show_leaf(inode, path);
3992 
3993 	ext4_update_inode_fsync_trans(handle, inode, 1);
3994 	err = check_eofblocks_fl(handle, inode, map->m_lblk, path, map->m_len);
3995 	if (err)
3996 		return err;
3997 	map->m_flags |= EXT4_MAP_UNWRITTEN;
3998 	if (allocated > map->m_len)
3999 		allocated = map->m_len;
4000 	map->m_len = allocated;
4001 	return allocated;
4002 }
4003 
4004 static int
4005 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
4006 			struct ext4_map_blocks *map,
4007 			struct ext4_ext_path **ppath, int flags,
4008 			unsigned int allocated, ext4_fsblk_t newblock)
4009 {
4010 	struct ext4_ext_path *path = *ppath;
4011 	int ret = 0;
4012 	int err = 0;
4013 
4014 	ext_debug("ext4_ext_handle_unwritten_extents: inode %lu, logical "
4015 		  "block %llu, max_blocks %u, flags %x, allocated %u\n",
4016 		  inode->i_ino, (unsigned long long)map->m_lblk, map->m_len,
4017 		  flags, allocated);
4018 	ext4_ext_show_leaf(inode, path);
4019 
4020 	/*
4021 	 * When writing into unwritten space, we should not fail to
4022 	 * allocate metadata blocks for the new extent block if needed.
4023 	 */
4024 	flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
4025 
4026 	trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
4027 						    allocated, newblock);
4028 
4029 	/* get_block() before submit the IO, split the extent */
4030 	if (flags & EXT4_GET_BLOCKS_PRE_IO) {
4031 		ret = ext4_split_convert_extents(handle, inode, map, ppath,
4032 					 flags | EXT4_GET_BLOCKS_CONVERT);
4033 		if (ret <= 0)
4034 			goto out;
4035 		map->m_flags |= EXT4_MAP_UNWRITTEN;
4036 		goto out;
4037 	}
4038 	/* IO end_io complete, convert the filled extent to written */
4039 	if (flags & EXT4_GET_BLOCKS_CONVERT) {
4040 		if (flags & EXT4_GET_BLOCKS_ZERO) {
4041 			if (allocated > map->m_len)
4042 				allocated = map->m_len;
4043 			err = ext4_issue_zeroout(inode, map->m_lblk, newblock,
4044 						 allocated);
4045 			if (err < 0)
4046 				goto out2;
4047 		}
4048 		ret = ext4_convert_unwritten_extents_endio(handle, inode, map,
4049 							   ppath);
4050 		if (ret >= 0) {
4051 			ext4_update_inode_fsync_trans(handle, inode, 1);
4052 			err = check_eofblocks_fl(handle, inode, map->m_lblk,
4053 						 path, map->m_len);
4054 		} else
4055 			err = ret;
4056 		map->m_flags |= EXT4_MAP_MAPPED;
4057 		map->m_pblk = newblock;
4058 		if (allocated > map->m_len)
4059 			allocated = map->m_len;
4060 		map->m_len = allocated;
4061 		goto out2;
4062 	}
4063 	/* buffered IO case */
4064 	/*
4065 	 * repeat fallocate creation request
4066 	 * we already have an unwritten extent
4067 	 */
4068 	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4069 		map->m_flags |= EXT4_MAP_UNWRITTEN;
4070 		goto map_out;
4071 	}
4072 
4073 	/* buffered READ or buffered write_begin() lookup */
4074 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4075 		/*
4076 		 * We have blocks reserved already.  We
4077 		 * return allocated blocks so that delalloc
4078 		 * won't do block reservation for us.  But
4079 		 * the buffer head will be unmapped so that
4080 		 * a read from the block returns 0s.
4081 		 */
4082 		map->m_flags |= EXT4_MAP_UNWRITTEN;
4083 		goto out1;
4084 	}
4085 
4086 	/* buffered write, writepage time, convert*/
4087 	ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
4088 	if (ret >= 0)
4089 		ext4_update_inode_fsync_trans(handle, inode, 1);
4090 out:
4091 	if (ret <= 0) {
4092 		err = ret;
4093 		goto out2;
4094 	} else
4095 		allocated = ret;
4096 	map->m_flags |= EXT4_MAP_NEW;
4097 	if (allocated > map->m_len)
4098 		allocated = map->m_len;
4099 	map->m_len = allocated;
4100 
4101 map_out:
4102 	map->m_flags |= EXT4_MAP_MAPPED;
4103 	if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0) {
4104 		err = check_eofblocks_fl(handle, inode, map->m_lblk, path,
4105 					 map->m_len);
4106 		if (err < 0)
4107 			goto out2;
4108 	}
4109 out1:
4110 	if (allocated > map->m_len)
4111 		allocated = map->m_len;
4112 	ext4_ext_show_leaf(inode, path);
4113 	map->m_pblk = newblock;
4114 	map->m_len = allocated;
4115 out2:
4116 	return err ? err : allocated;
4117 }
4118 
4119 /*
4120  * get_implied_cluster_alloc - check to see if the requested
4121  * allocation (in the map structure) overlaps with a cluster already
4122  * allocated in an extent.
4123  *	@sb	The filesystem superblock structure
4124  *	@map	The requested lblk->pblk mapping
4125  *	@ex	The extent structure which might contain an implied
4126  *			cluster allocation
4127  *
4128  * This function is called by ext4_ext_map_blocks() after we failed to
4129  * find blocks that were already in the inode's extent tree.  Hence,
4130  * we know that the beginning of the requested region cannot overlap
4131  * the extent from the inode's extent tree.  There are three cases we
4132  * want to catch.  The first is this case:
4133  *
4134  *		 |--- cluster # N--|
4135  *    |--- extent ---|	|---- requested region ---|
4136  *			|==========|
4137  *
4138  * The second case that we need to test for is this one:
4139  *
4140  *   |--------- cluster # N ----------------|
4141  *	   |--- requested region --|   |------- extent ----|
4142  *	   |=======================|
4143  *
4144  * The third case is when the requested region lies between two extents
4145  * within the same cluster:
4146  *          |------------- cluster # N-------------|
4147  * |----- ex -----|                  |---- ex_right ----|
4148  *                  |------ requested region ------|
4149  *                  |================|
4150  *
4151  * In each of the above cases, we need to set the map->m_pblk and
4152  * map->m_len so it corresponds to the return the extent labelled as
4153  * "|====|" from cluster #N, since it is already in use for data in
4154  * cluster EXT4_B2C(sbi, map->m_lblk).	We will then return 1 to
4155  * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4156  * as a new "allocated" block region.  Otherwise, we will return 0 and
4157  * ext4_ext_map_blocks() will then allocate one or more new clusters
4158  * by calling ext4_mb_new_blocks().
4159  */
4160 static int get_implied_cluster_alloc(struct super_block *sb,
4161 				     struct ext4_map_blocks *map,
4162 				     struct ext4_extent *ex,
4163 				     struct ext4_ext_path *path)
4164 {
4165 	struct ext4_sb_info *sbi = EXT4_SB(sb);
4166 	ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4167 	ext4_lblk_t ex_cluster_start, ex_cluster_end;
4168 	ext4_lblk_t rr_cluster_start;
4169 	ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4170 	ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4171 	unsigned short ee_len = ext4_ext_get_actual_len(ex);
4172 
4173 	/* The extent passed in that we are trying to match */
4174 	ex_cluster_start = EXT4_B2C(sbi, ee_block);
4175 	ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4176 
4177 	/* The requested region passed into ext4_map_blocks() */
4178 	rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4179 
4180 	if ((rr_cluster_start == ex_cluster_end) ||
4181 	    (rr_cluster_start == ex_cluster_start)) {
4182 		if (rr_cluster_start == ex_cluster_end)
4183 			ee_start += ee_len - 1;
4184 		map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4185 		map->m_len = min(map->m_len,
4186 				 (unsigned) sbi->s_cluster_ratio - c_offset);
4187 		/*
4188 		 * Check for and handle this case:
4189 		 *
4190 		 *   |--------- cluster # N-------------|
4191 		 *		       |------- extent ----|
4192 		 *	   |--- requested region ---|
4193 		 *	   |===========|
4194 		 */
4195 
4196 		if (map->m_lblk < ee_block)
4197 			map->m_len = min(map->m_len, ee_block - map->m_lblk);
4198 
4199 		/*
4200 		 * Check for the case where there is already another allocated
4201 		 * block to the right of 'ex' but before the end of the cluster.
4202 		 *
4203 		 *          |------------- cluster # N-------------|
4204 		 * |----- ex -----|                  |---- ex_right ----|
4205 		 *                  |------ requested region ------|
4206 		 *                  |================|
4207 		 */
4208 		if (map->m_lblk > ee_block) {
4209 			ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4210 			map->m_len = min(map->m_len, next - map->m_lblk);
4211 		}
4212 
4213 		trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4214 		return 1;
4215 	}
4216 
4217 	trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4218 	return 0;
4219 }
4220 
4221 
4222 /*
4223  * Block allocation/map/preallocation routine for extents based files
4224  *
4225  *
4226  * Need to be called with
4227  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4228  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4229  *
4230  * return > 0, number of of blocks already mapped/allocated
4231  *          if create == 0 and these are pre-allocated blocks
4232  *          	buffer head is unmapped
4233  *          otherwise blocks are mapped
4234  *
4235  * return = 0, if plain look up failed (blocks have not been allocated)
4236  *          buffer head is unmapped
4237  *
4238  * return < 0, error case.
4239  */
4240 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4241 			struct ext4_map_blocks *map, int flags)
4242 {
4243 	struct ext4_ext_path *path = NULL;
4244 	struct ext4_extent newex, *ex, *ex2;
4245 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4246 	ext4_fsblk_t newblock = 0;
4247 	int free_on_err = 0, err = 0, depth, ret;
4248 	unsigned int allocated = 0, offset = 0;
4249 	unsigned int allocated_clusters = 0;
4250 	struct ext4_allocation_request ar;
4251 	ext4_lblk_t cluster_offset;
4252 	bool map_from_cluster = false;
4253 
4254 	ext_debug("blocks %u/%u requested for inode %lu\n",
4255 		  map->m_lblk, map->m_len, inode->i_ino);
4256 	trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4257 
4258 	/* find extent for this block */
4259 	path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4260 	if (IS_ERR(path)) {
4261 		err = PTR_ERR(path);
4262 		path = NULL;
4263 		goto out2;
4264 	}
4265 
4266 	depth = ext_depth(inode);
4267 
4268 	/*
4269 	 * consistent leaf must not be empty;
4270 	 * this situation is possible, though, _during_ tree modification;
4271 	 * this is why assert can't be put in ext4_find_extent()
4272 	 */
4273 	if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4274 		EXT4_ERROR_INODE(inode, "bad extent address "
4275 				 "lblock: %lu, depth: %d pblock %lld",
4276 				 (unsigned long) map->m_lblk, depth,
4277 				 path[depth].p_block);
4278 		err = -EFSCORRUPTED;
4279 		goto out2;
4280 	}
4281 
4282 	ex = path[depth].p_ext;
4283 	if (ex) {
4284 		ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4285 		ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4286 		unsigned short ee_len;
4287 
4288 
4289 		/*
4290 		 * unwritten extents are treated as holes, except that
4291 		 * we split out initialized portions during a write.
4292 		 */
4293 		ee_len = ext4_ext_get_actual_len(ex);
4294 
4295 		trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4296 
4297 		/* if found extent covers block, simply return it */
4298 		if (in_range(map->m_lblk, ee_block, ee_len)) {
4299 			newblock = map->m_lblk - ee_block + ee_start;
4300 			/* number of remaining blocks in the extent */
4301 			allocated = ee_len - (map->m_lblk - ee_block);
4302 			ext_debug("%u fit into %u:%d -> %llu\n", map->m_lblk,
4303 				  ee_block, ee_len, newblock);
4304 
4305 			/*
4306 			 * If the extent is initialized check whether the
4307 			 * caller wants to convert it to unwritten.
4308 			 */
4309 			if ((!ext4_ext_is_unwritten(ex)) &&
4310 			    (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4311 				allocated = convert_initialized_extent(
4312 						handle, inode, map, &path,
4313 						allocated);
4314 				goto out2;
4315 			} else if (!ext4_ext_is_unwritten(ex))
4316 				goto out;
4317 
4318 			ret = ext4_ext_handle_unwritten_extents(
4319 				handle, inode, map, &path, flags,
4320 				allocated, newblock);
4321 			if (ret < 0)
4322 				err = ret;
4323 			else
4324 				allocated = ret;
4325 			goto out2;
4326 		}
4327 	}
4328 
4329 	/*
4330 	 * requested block isn't allocated yet;
4331 	 * we couldn't try to create block if create flag is zero
4332 	 */
4333 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4334 		ext4_lblk_t hole_start, hole_len;
4335 
4336 		hole_start = map->m_lblk;
4337 		hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4338 		/*
4339 		 * put just found gap into cache to speed up
4340 		 * subsequent requests
4341 		 */
4342 		ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4343 
4344 		/* Update hole_len to reflect hole size after map->m_lblk */
4345 		if (hole_start != map->m_lblk)
4346 			hole_len -= map->m_lblk - hole_start;
4347 		map->m_pblk = 0;
4348 		map->m_len = min_t(unsigned int, map->m_len, hole_len);
4349 
4350 		goto out2;
4351 	}
4352 
4353 	/*
4354 	 * Okay, we need to do block allocation.
4355 	 */
4356 	newex.ee_block = cpu_to_le32(map->m_lblk);
4357 	cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4358 
4359 	/*
4360 	 * If we are doing bigalloc, check to see if the extent returned
4361 	 * by ext4_find_extent() implies a cluster we can use.
4362 	 */
4363 	if (cluster_offset && ex &&
4364 	    get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4365 		ar.len = allocated = map->m_len;
4366 		newblock = map->m_pblk;
4367 		map_from_cluster = true;
4368 		goto got_allocated_blocks;
4369 	}
4370 
4371 	/* find neighbour allocated blocks */
4372 	ar.lleft = map->m_lblk;
4373 	err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4374 	if (err)
4375 		goto out2;
4376 	ar.lright = map->m_lblk;
4377 	ex2 = NULL;
4378 	err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4379 	if (err)
4380 		goto out2;
4381 
4382 	/* Check if the extent after searching to the right implies a
4383 	 * cluster we can use. */
4384 	if ((sbi->s_cluster_ratio > 1) && ex2 &&
4385 	    get_implied_cluster_alloc(inode->i_sb, map, ex2, path)) {
4386 		ar.len = allocated = map->m_len;
4387 		newblock = map->m_pblk;
4388 		map_from_cluster = true;
4389 		goto got_allocated_blocks;
4390 	}
4391 
4392 	/*
4393 	 * See if request is beyond maximum number of blocks we can have in
4394 	 * a single extent. For an initialized extent this limit is
4395 	 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4396 	 * EXT_UNWRITTEN_MAX_LEN.
4397 	 */
4398 	if (map->m_len > EXT_INIT_MAX_LEN &&
4399 	    !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4400 		map->m_len = EXT_INIT_MAX_LEN;
4401 	else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4402 		 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4403 		map->m_len = EXT_UNWRITTEN_MAX_LEN;
4404 
4405 	/* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4406 	newex.ee_len = cpu_to_le16(map->m_len);
4407 	err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4408 	if (err)
4409 		allocated = ext4_ext_get_actual_len(&newex);
4410 	else
4411 		allocated = map->m_len;
4412 
4413 	/* allocate new block */
4414 	ar.inode = inode;
4415 	ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4416 	ar.logical = map->m_lblk;
4417 	/*
4418 	 * We calculate the offset from the beginning of the cluster
4419 	 * for the logical block number, since when we allocate a
4420 	 * physical cluster, the physical block should start at the
4421 	 * same offset from the beginning of the cluster.  This is
4422 	 * needed so that future calls to get_implied_cluster_alloc()
4423 	 * work correctly.
4424 	 */
4425 	offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4426 	ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4427 	ar.goal -= offset;
4428 	ar.logical -= offset;
4429 	if (S_ISREG(inode->i_mode))
4430 		ar.flags = EXT4_MB_HINT_DATA;
4431 	else
4432 		/* disable in-core preallocation for non-regular files */
4433 		ar.flags = 0;
4434 	if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4435 		ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4436 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4437 		ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4438 	if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4439 		ar.flags |= EXT4_MB_USE_RESERVED;
4440 	newblock = ext4_mb_new_blocks(handle, &ar, &err);
4441 	if (!newblock)
4442 		goto out2;
4443 	ext_debug("allocate new block: goal %llu, found %llu/%u\n",
4444 		  ar.goal, newblock, allocated);
4445 	free_on_err = 1;
4446 	allocated_clusters = ar.len;
4447 	ar.len = EXT4_C2B(sbi, ar.len) - offset;
4448 	if (ar.len > allocated)
4449 		ar.len = allocated;
4450 
4451 got_allocated_blocks:
4452 	/* try to insert new extent into found leaf and return */
4453 	ext4_ext_store_pblock(&newex, newblock + offset);
4454 	newex.ee_len = cpu_to_le16(ar.len);
4455 	/* Mark unwritten */
4456 	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT){
4457 		ext4_ext_mark_unwritten(&newex);
4458 		map->m_flags |= EXT4_MAP_UNWRITTEN;
4459 	}
4460 
4461 	err = 0;
4462 	if ((flags & EXT4_GET_BLOCKS_KEEP_SIZE) == 0)
4463 		err = check_eofblocks_fl(handle, inode, map->m_lblk,
4464 					 path, ar.len);
4465 	if (!err)
4466 		err = ext4_ext_insert_extent(handle, inode, &path,
4467 					     &newex, flags);
4468 
4469 	if (err && free_on_err) {
4470 		int fb_flags = flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE ?
4471 			EXT4_FREE_BLOCKS_NO_QUOT_UPDATE : 0;
4472 		/* free data blocks we just allocated */
4473 		/* not a good idea to call discard here directly,
4474 		 * but otherwise we'd need to call it every free() */
4475 		ext4_discard_preallocations(inode);
4476 		ext4_free_blocks(handle, inode, NULL, newblock,
4477 				 EXT4_C2B(sbi, allocated_clusters), fb_flags);
4478 		goto out2;
4479 	}
4480 
4481 	/* previous routine could use block we allocated */
4482 	newblock = ext4_ext_pblock(&newex);
4483 	allocated = ext4_ext_get_actual_len(&newex);
4484 	if (allocated > map->m_len)
4485 		allocated = map->m_len;
4486 	map->m_flags |= EXT4_MAP_NEW;
4487 
4488 	/*
4489 	 * Reduce the reserved cluster count to reflect successful deferred
4490 	 * allocation of delayed allocated clusters or direct allocation of
4491 	 * clusters discovered to be delayed allocated.  Once allocated, a
4492 	 * cluster is not included in the reserved count.
4493 	 */
4494 	if (test_opt(inode->i_sb, DELALLOC) && !map_from_cluster) {
4495 		if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4496 			/*
4497 			 * When allocating delayed allocated clusters, simply
4498 			 * reduce the reserved cluster count and claim quota
4499 			 */
4500 			ext4_da_update_reserve_space(inode, allocated_clusters,
4501 							1);
4502 		} else {
4503 			ext4_lblk_t lblk, len;
4504 			unsigned int n;
4505 
4506 			/*
4507 			 * When allocating non-delayed allocated clusters
4508 			 * (from fallocate, filemap, DIO, or clusters
4509 			 * allocated when delalloc has been disabled by
4510 			 * ext4_nonda_switch), reduce the reserved cluster
4511 			 * count by the number of allocated clusters that
4512 			 * have previously been delayed allocated.  Quota
4513 			 * has been claimed by ext4_mb_new_blocks() above,
4514 			 * so release the quota reservations made for any
4515 			 * previously delayed allocated clusters.
4516 			 */
4517 			lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4518 			len = allocated_clusters << sbi->s_cluster_bits;
4519 			n = ext4_es_delayed_clu(inode, lblk, len);
4520 			if (n > 0)
4521 				ext4_da_update_reserve_space(inode, (int) n, 0);
4522 		}
4523 	}
4524 
4525 	/*
4526 	 * Cache the extent and update transaction to commit on fdatasync only
4527 	 * when it is _not_ an unwritten extent.
4528 	 */
4529 	if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4530 		ext4_update_inode_fsync_trans(handle, inode, 1);
4531 	else
4532 		ext4_update_inode_fsync_trans(handle, inode, 0);
4533 out:
4534 	if (allocated > map->m_len)
4535 		allocated = map->m_len;
4536 	ext4_ext_show_leaf(inode, path);
4537 	map->m_flags |= EXT4_MAP_MAPPED;
4538 	map->m_pblk = newblock;
4539 	map->m_len = allocated;
4540 out2:
4541 	ext4_ext_drop_refs(path);
4542 	kfree(path);
4543 
4544 	trace_ext4_ext_map_blocks_exit(inode, flags, map,
4545 				       err ? err : allocated);
4546 	return err ? err : allocated;
4547 }
4548 
4549 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4550 {
4551 	struct super_block *sb = inode->i_sb;
4552 	ext4_lblk_t last_block;
4553 	int err = 0;
4554 
4555 	/*
4556 	 * TODO: optimization is possible here.
4557 	 * Probably we need not scan at all,
4558 	 * because page truncation is enough.
4559 	 */
4560 
4561 	/* we have to know where to truncate from in crash case */
4562 	EXT4_I(inode)->i_disksize = inode->i_size;
4563 	err = ext4_mark_inode_dirty(handle, inode);
4564 	if (err)
4565 		return err;
4566 
4567 	last_block = (inode->i_size + sb->s_blocksize - 1)
4568 			>> EXT4_BLOCK_SIZE_BITS(sb);
4569 retry:
4570 	err = ext4_es_remove_extent(inode, last_block,
4571 				    EXT_MAX_BLOCKS - last_block);
4572 	if (err == -ENOMEM) {
4573 		cond_resched();
4574 		congestion_wait(BLK_RW_ASYNC, HZ/50);
4575 		goto retry;
4576 	}
4577 	if (err)
4578 		return err;
4579 	return ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4580 }
4581 
4582 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4583 				  ext4_lblk_t len, loff_t new_size,
4584 				  int flags)
4585 {
4586 	struct inode *inode = file_inode(file);
4587 	handle_t *handle;
4588 	int ret = 0;
4589 	int ret2 = 0;
4590 	int retries = 0;
4591 	int depth = 0;
4592 	struct ext4_map_blocks map;
4593 	unsigned int credits;
4594 	loff_t epos;
4595 
4596 	BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4597 	map.m_lblk = offset;
4598 	map.m_len = len;
4599 	/*
4600 	 * Don't normalize the request if it can fit in one extent so
4601 	 * that it doesn't get unnecessarily split into multiple
4602 	 * extents.
4603 	 */
4604 	if (len <= EXT_UNWRITTEN_MAX_LEN)
4605 		flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4606 
4607 	/*
4608 	 * credits to insert 1 extent into extent tree
4609 	 */
4610 	credits = ext4_chunk_trans_blocks(inode, len);
4611 	depth = ext_depth(inode);
4612 
4613 retry:
4614 	while (ret >= 0 && len) {
4615 		/*
4616 		 * Recalculate credits when extent tree depth changes.
4617 		 */
4618 		if (depth != ext_depth(inode)) {
4619 			credits = ext4_chunk_trans_blocks(inode, len);
4620 			depth = ext_depth(inode);
4621 		}
4622 
4623 		handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4624 					    credits);
4625 		if (IS_ERR(handle)) {
4626 			ret = PTR_ERR(handle);
4627 			break;
4628 		}
4629 		ret = ext4_map_blocks(handle, inode, &map, flags);
4630 		if (ret <= 0) {
4631 			ext4_debug("inode #%lu: block %u: len %u: "
4632 				   "ext4_ext_map_blocks returned %d",
4633 				   inode->i_ino, map.m_lblk,
4634 				   map.m_len, ret);
4635 			ext4_mark_inode_dirty(handle, inode);
4636 			ret2 = ext4_journal_stop(handle);
4637 			break;
4638 		}
4639 		map.m_lblk += ret;
4640 		map.m_len = len = len - ret;
4641 		epos = (loff_t)map.m_lblk << inode->i_blkbits;
4642 		inode->i_ctime = current_time(inode);
4643 		if (new_size) {
4644 			if (epos > new_size)
4645 				epos = new_size;
4646 			if (ext4_update_inode_size(inode, epos) & 0x1)
4647 				inode->i_mtime = inode->i_ctime;
4648 		} else {
4649 			if (epos > inode->i_size)
4650 				ext4_set_inode_flag(inode,
4651 						    EXT4_INODE_EOFBLOCKS);
4652 		}
4653 		ext4_mark_inode_dirty(handle, inode);
4654 		ext4_update_inode_fsync_trans(handle, inode, 1);
4655 		ret2 = ext4_journal_stop(handle);
4656 		if (ret2)
4657 			break;
4658 	}
4659 	if (ret == -ENOSPC &&
4660 			ext4_should_retry_alloc(inode->i_sb, &retries)) {
4661 		ret = 0;
4662 		goto retry;
4663 	}
4664 
4665 	return ret > 0 ? ret2 : ret;
4666 }
4667 
4668 static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len);
4669 
4670 static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len);
4671 
4672 static long ext4_zero_range(struct file *file, loff_t offset,
4673 			    loff_t len, int mode)
4674 {
4675 	struct inode *inode = file_inode(file);
4676 	handle_t *handle = NULL;
4677 	unsigned int max_blocks;
4678 	loff_t new_size = 0;
4679 	int ret = 0;
4680 	int flags;
4681 	int credits;
4682 	int partial_begin, partial_end;
4683 	loff_t start, end;
4684 	ext4_lblk_t lblk;
4685 	unsigned int blkbits = inode->i_blkbits;
4686 
4687 	trace_ext4_zero_range(inode, offset, len, mode);
4688 
4689 	/* Call ext4_force_commit to flush all data in case of data=journal. */
4690 	if (ext4_should_journal_data(inode)) {
4691 		ret = ext4_force_commit(inode->i_sb);
4692 		if (ret)
4693 			return ret;
4694 	}
4695 
4696 	/*
4697 	 * Round up offset. This is not fallocate, we neet to zero out
4698 	 * blocks, so convert interior block aligned part of the range to
4699 	 * unwritten and possibly manually zero out unaligned parts of the
4700 	 * range.
4701 	 */
4702 	start = round_up(offset, 1 << blkbits);
4703 	end = round_down((offset + len), 1 << blkbits);
4704 
4705 	if (start < offset || end > offset + len)
4706 		return -EINVAL;
4707 	partial_begin = offset & ((1 << blkbits) - 1);
4708 	partial_end = (offset + len) & ((1 << blkbits) - 1);
4709 
4710 	lblk = start >> blkbits;
4711 	max_blocks = (end >> blkbits);
4712 	if (max_blocks < lblk)
4713 		max_blocks = 0;
4714 	else
4715 		max_blocks -= lblk;
4716 
4717 	inode_lock(inode);
4718 
4719 	/*
4720 	 * Indirect files do not support unwritten extnets
4721 	 */
4722 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4723 		ret = -EOPNOTSUPP;
4724 		goto out_mutex;
4725 	}
4726 
4727 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4728 	    (offset + len > inode->i_size ||
4729 	     offset + len > EXT4_I(inode)->i_disksize)) {
4730 		new_size = offset + len;
4731 		ret = inode_newsize_ok(inode, new_size);
4732 		if (ret)
4733 			goto out_mutex;
4734 	}
4735 
4736 	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4737 	if (mode & FALLOC_FL_KEEP_SIZE)
4738 		flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4739 
4740 	/* Wait all existing dio workers, newcomers will block on i_mutex */
4741 	inode_dio_wait(inode);
4742 
4743 	/* Preallocate the range including the unaligned edges */
4744 	if (partial_begin || partial_end) {
4745 		ret = ext4_alloc_file_blocks(file,
4746 				round_down(offset, 1 << blkbits) >> blkbits,
4747 				(round_up((offset + len), 1 << blkbits) -
4748 				 round_down(offset, 1 << blkbits)) >> blkbits,
4749 				new_size, flags);
4750 		if (ret)
4751 			goto out_mutex;
4752 
4753 	}
4754 
4755 	/* Zero range excluding the unaligned edges */
4756 	if (max_blocks > 0) {
4757 		flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4758 			  EXT4_EX_NOCACHE);
4759 
4760 		/*
4761 		 * Prevent page faults from reinstantiating pages we have
4762 		 * released from page cache.
4763 		 */
4764 		down_write(&EXT4_I(inode)->i_mmap_sem);
4765 
4766 		ret = ext4_break_layouts(inode);
4767 		if (ret) {
4768 			up_write(&EXT4_I(inode)->i_mmap_sem);
4769 			goto out_mutex;
4770 		}
4771 
4772 		ret = ext4_update_disksize_before_punch(inode, offset, len);
4773 		if (ret) {
4774 			up_write(&EXT4_I(inode)->i_mmap_sem);
4775 			goto out_mutex;
4776 		}
4777 		/* Now release the pages and zero block aligned part of pages */
4778 		truncate_pagecache_range(inode, start, end - 1);
4779 		inode->i_mtime = inode->i_ctime = current_time(inode);
4780 
4781 		ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4782 					     flags);
4783 		up_write(&EXT4_I(inode)->i_mmap_sem);
4784 		if (ret)
4785 			goto out_mutex;
4786 	}
4787 	if (!partial_begin && !partial_end)
4788 		goto out_mutex;
4789 
4790 	/*
4791 	 * In worst case we have to writeout two nonadjacent unwritten
4792 	 * blocks and update the inode
4793 	 */
4794 	credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4795 	if (ext4_should_journal_data(inode))
4796 		credits += 2;
4797 	handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4798 	if (IS_ERR(handle)) {
4799 		ret = PTR_ERR(handle);
4800 		ext4_std_error(inode->i_sb, ret);
4801 		goto out_mutex;
4802 	}
4803 
4804 	inode->i_mtime = inode->i_ctime = current_time(inode);
4805 	if (new_size) {
4806 		ext4_update_inode_size(inode, new_size);
4807 	} else {
4808 		/*
4809 		* Mark that we allocate beyond EOF so the subsequent truncate
4810 		* can proceed even if the new size is the same as i_size.
4811 		*/
4812 		if (offset + len > inode->i_size)
4813 			ext4_set_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4814 	}
4815 	ext4_mark_inode_dirty(handle, inode);
4816 
4817 	/* Zero out partial block at the edges of the range */
4818 	ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4819 	if (ret >= 0)
4820 		ext4_update_inode_fsync_trans(handle, inode, 1);
4821 
4822 	if (file->f_flags & O_SYNC)
4823 		ext4_handle_sync(handle);
4824 
4825 	ext4_journal_stop(handle);
4826 out_mutex:
4827 	inode_unlock(inode);
4828 	return ret;
4829 }
4830 
4831 /*
4832  * preallocate space for a file. This implements ext4's fallocate file
4833  * operation, which gets called from sys_fallocate system call.
4834  * For block-mapped files, posix_fallocate should fall back to the method
4835  * of writing zeroes to the required new blocks (the same behavior which is
4836  * expected for file systems which do not support fallocate() system call).
4837  */
4838 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4839 {
4840 	struct inode *inode = file_inode(file);
4841 	loff_t new_size = 0;
4842 	unsigned int max_blocks;
4843 	int ret = 0;
4844 	int flags;
4845 	ext4_lblk_t lblk;
4846 	unsigned int blkbits = inode->i_blkbits;
4847 
4848 	/*
4849 	 * Encrypted inodes can't handle collapse range or insert
4850 	 * range since we would need to re-encrypt blocks with a
4851 	 * different IV or XTS tweak (which are based on the logical
4852 	 * block number).
4853 	 */
4854 	if (IS_ENCRYPTED(inode) &&
4855 	    (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4856 		return -EOPNOTSUPP;
4857 
4858 	/* Return error if mode is not supported */
4859 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4860 		     FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4861 		     FALLOC_FL_INSERT_RANGE))
4862 		return -EOPNOTSUPP;
4863 
4864 	if (mode & FALLOC_FL_PUNCH_HOLE)
4865 		return ext4_punch_hole(inode, offset, len);
4866 
4867 	ret = ext4_convert_inline_data(inode);
4868 	if (ret)
4869 		return ret;
4870 
4871 	if (mode & FALLOC_FL_COLLAPSE_RANGE)
4872 		return ext4_collapse_range(inode, offset, len);
4873 
4874 	if (mode & FALLOC_FL_INSERT_RANGE)
4875 		return ext4_insert_range(inode, offset, len);
4876 
4877 	if (mode & FALLOC_FL_ZERO_RANGE)
4878 		return ext4_zero_range(file, offset, len, mode);
4879 
4880 	trace_ext4_fallocate_enter(inode, offset, len, mode);
4881 	lblk = offset >> blkbits;
4882 
4883 	max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4884 	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4885 	if (mode & FALLOC_FL_KEEP_SIZE)
4886 		flags |= EXT4_GET_BLOCKS_KEEP_SIZE;
4887 
4888 	inode_lock(inode);
4889 
4890 	/*
4891 	 * We only support preallocation for extent-based files only
4892 	 */
4893 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4894 		ret = -EOPNOTSUPP;
4895 		goto out;
4896 	}
4897 
4898 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4899 	    (offset + len > inode->i_size ||
4900 	     offset + len > EXT4_I(inode)->i_disksize)) {
4901 		new_size = offset + len;
4902 		ret = inode_newsize_ok(inode, new_size);
4903 		if (ret)
4904 			goto out;
4905 	}
4906 
4907 	/* Wait all existing dio workers, newcomers will block on i_mutex */
4908 	inode_dio_wait(inode);
4909 
4910 	ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4911 	if (ret)
4912 		goto out;
4913 
4914 	if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4915 		ret = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
4916 						EXT4_I(inode)->i_sync_tid);
4917 	}
4918 out:
4919 	inode_unlock(inode);
4920 	trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4921 	return ret;
4922 }
4923 
4924 /*
4925  * This function convert a range of blocks to written extents
4926  * The caller of this function will pass the start offset and the size.
4927  * all unwritten extents within this range will be converted to
4928  * written extents.
4929  *
4930  * This function is called from the direct IO end io call back
4931  * function, to convert the fallocated extents after IO is completed.
4932  * Returns 0 on success.
4933  */
4934 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4935 				   loff_t offset, ssize_t len)
4936 {
4937 	unsigned int max_blocks;
4938 	int ret = 0;
4939 	int ret2 = 0;
4940 	struct ext4_map_blocks map;
4941 	unsigned int blkbits = inode->i_blkbits;
4942 	unsigned int credits = 0;
4943 
4944 	map.m_lblk = offset >> blkbits;
4945 	max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4946 
4947 	if (!handle) {
4948 		/*
4949 		 * credits to insert 1 extent into extent tree
4950 		 */
4951 		credits = ext4_chunk_trans_blocks(inode, max_blocks);
4952 	}
4953 	while (ret >= 0 && ret < max_blocks) {
4954 		map.m_lblk += ret;
4955 		map.m_len = (max_blocks -= ret);
4956 		if (credits) {
4957 			handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4958 						    credits);
4959 			if (IS_ERR(handle)) {
4960 				ret = PTR_ERR(handle);
4961 				break;
4962 			}
4963 		}
4964 		ret = ext4_map_blocks(handle, inode, &map,
4965 				      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4966 		if (ret <= 0)
4967 			ext4_warning(inode->i_sb,
4968 				     "inode #%lu: block %u: len %u: "
4969 				     "ext4_ext_map_blocks returned %d",
4970 				     inode->i_ino, map.m_lblk,
4971 				     map.m_len, ret);
4972 		ext4_mark_inode_dirty(handle, inode);
4973 		if (credits)
4974 			ret2 = ext4_journal_stop(handle);
4975 		if (ret <= 0 || ret2)
4976 			break;
4977 	}
4978 	return ret > 0 ? ret2 : ret;
4979 }
4980 
4981 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4982 {
4983 	int ret, err = 0;
4984 	struct ext4_io_end_vec *io_end_vec;
4985 
4986 	/*
4987 	 * This is somewhat ugly but the idea is clear: When transaction is
4988 	 * reserved, everything goes into it. Otherwise we rather start several
4989 	 * smaller transactions for conversion of each extent separately.
4990 	 */
4991 	if (handle) {
4992 		handle = ext4_journal_start_reserved(handle,
4993 						     EXT4_HT_EXT_CONVERT);
4994 		if (IS_ERR(handle))
4995 			return PTR_ERR(handle);
4996 	}
4997 
4998 	list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4999 		ret = ext4_convert_unwritten_extents(handle, io_end->inode,
5000 						     io_end_vec->offset,
5001 						     io_end_vec->size);
5002 		if (ret)
5003 			break;
5004 	}
5005 
5006 	if (handle)
5007 		err = ext4_journal_stop(handle);
5008 
5009 	return ret < 0 ? ret : err;
5010 }
5011 
5012 /*
5013  * If newes is not existing extent (newes->ec_pblk equals zero) find
5014  * delayed extent at start of newes and update newes accordingly and
5015  * return start of the next delayed extent.
5016  *
5017  * If newes is existing extent (newes->ec_pblk is not equal zero)
5018  * return start of next delayed extent or EXT_MAX_BLOCKS if no delayed
5019  * extent found. Leave newes unmodified.
5020  */
5021 static int ext4_find_delayed_extent(struct inode *inode,
5022 				    struct extent_status *newes)
5023 {
5024 	struct extent_status es;
5025 	ext4_lblk_t block, next_del;
5026 
5027 	if (newes->es_pblk == 0) {
5028 		ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
5029 					  newes->es_lblk,
5030 					  newes->es_lblk + newes->es_len - 1,
5031 					  &es);
5032 
5033 		/*
5034 		 * No extent in extent-tree contains block @newes->es_pblk,
5035 		 * then the block may stay in 1)a hole or 2)delayed-extent.
5036 		 */
5037 		if (es.es_len == 0)
5038 			/* A hole found. */
5039 			return 0;
5040 
5041 		if (es.es_lblk > newes->es_lblk) {
5042 			/* A hole found. */
5043 			newes->es_len = min(es.es_lblk - newes->es_lblk,
5044 					    newes->es_len);
5045 			return 0;
5046 		}
5047 
5048 		newes->es_len = es.es_lblk + es.es_len - newes->es_lblk;
5049 	}
5050 
5051 	block = newes->es_lblk + newes->es_len;
5052 	ext4_es_find_extent_range(inode, &ext4_es_is_delayed, block,
5053 				  EXT_MAX_BLOCKS, &es);
5054 	if (es.es_len == 0)
5055 		next_del = EXT_MAX_BLOCKS;
5056 	else
5057 		next_del = es.es_lblk;
5058 
5059 	return next_del;
5060 }
5061 
5062 static int ext4_xattr_fiemap(struct inode *inode,
5063 				struct fiemap_extent_info *fieinfo)
5064 {
5065 	__u64 physical = 0;
5066 	__u64 length;
5067 	__u32 flags = FIEMAP_EXTENT_LAST;
5068 	int blockbits = inode->i_sb->s_blocksize_bits;
5069 	int error = 0;
5070 
5071 	/* in-inode? */
5072 	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
5073 		struct ext4_iloc iloc;
5074 		int offset;	/* offset of xattr in inode */
5075 
5076 		error = ext4_get_inode_loc(inode, &iloc);
5077 		if (error)
5078 			return error;
5079 		physical = (__u64)iloc.bh->b_blocknr << blockbits;
5080 		offset = EXT4_GOOD_OLD_INODE_SIZE +
5081 				EXT4_I(inode)->i_extra_isize;
5082 		physical += offset;
5083 		length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
5084 		flags |= FIEMAP_EXTENT_DATA_INLINE;
5085 		brelse(iloc.bh);
5086 	} else { /* external block */
5087 		physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
5088 		length = inode->i_sb->s_blocksize;
5089 	}
5090 
5091 	if (physical)
5092 		error = fiemap_fill_next_extent(fieinfo, 0, physical,
5093 						length, flags);
5094 	return (error < 0 ? error : 0);
5095 }
5096 
5097 static int _ext4_fiemap(struct inode *inode,
5098 			struct fiemap_extent_info *fieinfo,
5099 			__u64 start, __u64 len,
5100 			int (*fill)(struct inode *, ext4_lblk_t,
5101 				    ext4_lblk_t,
5102 				    struct fiemap_extent_info *))
5103 {
5104 	ext4_lblk_t start_blk;
5105 	u32 ext4_fiemap_flags = FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR;
5106 
5107 	int error = 0;
5108 
5109 	if (ext4_has_inline_data(inode)) {
5110 		int has_inline = 1;
5111 
5112 		error = ext4_inline_data_fiemap(inode, fieinfo, &has_inline,
5113 						start, len);
5114 
5115 		if (has_inline)
5116 			return error;
5117 	}
5118 
5119 	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
5120 		error = ext4_ext_precache(inode);
5121 		if (error)
5122 			return error;
5123 		fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
5124 	}
5125 
5126 	/* fallback to generic here if not in extents fmt */
5127 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) &&
5128 	    fill == ext4_fill_fiemap_extents)
5129 		return generic_block_fiemap(inode, fieinfo, start, len,
5130 			ext4_get_block);
5131 
5132 	if (fill == ext4_fill_es_cache_info)
5133 		ext4_fiemap_flags &= FIEMAP_FLAG_XATTR;
5134 	if (fiemap_check_flags(fieinfo, ext4_fiemap_flags))
5135 		return -EBADR;
5136 
5137 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
5138 		error = ext4_xattr_fiemap(inode, fieinfo);
5139 	} else {
5140 		ext4_lblk_t len_blks;
5141 		__u64 last_blk;
5142 
5143 		start_blk = start >> inode->i_sb->s_blocksize_bits;
5144 		last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5145 		if (last_blk >= EXT_MAX_BLOCKS)
5146 			last_blk = EXT_MAX_BLOCKS-1;
5147 		len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5148 
5149 		/*
5150 		 * Walk the extent tree gathering extent information
5151 		 * and pushing extents back to the user.
5152 		 */
5153 		error = fill(inode, start_blk, len_blks, fieinfo);
5154 	}
5155 	return error;
5156 }
5157 
5158 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5159 		__u64 start, __u64 len)
5160 {
5161 	return _ext4_fiemap(inode, fieinfo, start, len,
5162 			    ext4_fill_fiemap_extents);
5163 }
5164 
5165 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
5166 		      __u64 start, __u64 len)
5167 {
5168 	if (ext4_has_inline_data(inode)) {
5169 		int has_inline;
5170 
5171 		down_read(&EXT4_I(inode)->xattr_sem);
5172 		has_inline = ext4_has_inline_data(inode);
5173 		up_read(&EXT4_I(inode)->xattr_sem);
5174 		if (has_inline)
5175 			return 0;
5176 	}
5177 
5178 	return _ext4_fiemap(inode, fieinfo, start, len,
5179 			    ext4_fill_es_cache_info);
5180 }
5181 
5182 
5183 /*
5184  * ext4_access_path:
5185  * Function to access the path buffer for marking it dirty.
5186  * It also checks if there are sufficient credits left in the journal handle
5187  * to update path.
5188  */
5189 static int
5190 ext4_access_path(handle_t *handle, struct inode *inode,
5191 		struct ext4_ext_path *path)
5192 {
5193 	int credits, err;
5194 
5195 	if (!ext4_handle_valid(handle))
5196 		return 0;
5197 
5198 	/*
5199 	 * Check if need to extend journal credits
5200 	 * 3 for leaf, sb, and inode plus 2 (bmap and group
5201 	 * descriptor) for each block group; assume two block
5202 	 * groups
5203 	 */
5204 	credits = ext4_writepage_trans_blocks(inode);
5205 	err = ext4_datasem_ensure_credits(handle, inode, 7, credits, 0);
5206 	if (err < 0)
5207 		return err;
5208 
5209 	err = ext4_ext_get_access(handle, inode, path);
5210 	return err;
5211 }
5212 
5213 /*
5214  * ext4_ext_shift_path_extents:
5215  * Shift the extents of a path structure lying between path[depth].p_ext
5216  * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5217  * if it is right shift or left shift operation.
5218  */
5219 static int
5220 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5221 			    struct inode *inode, handle_t *handle,
5222 			    enum SHIFT_DIRECTION SHIFT)
5223 {
5224 	int depth, err = 0;
5225 	struct ext4_extent *ex_start, *ex_last;
5226 	bool update = false;
5227 	depth = path->p_depth;
5228 
5229 	while (depth >= 0) {
5230 		if (depth == path->p_depth) {
5231 			ex_start = path[depth].p_ext;
5232 			if (!ex_start)
5233 				return -EFSCORRUPTED;
5234 
5235 			ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5236 
5237 			err = ext4_access_path(handle, inode, path + depth);
5238 			if (err)
5239 				goto out;
5240 
5241 			if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr))
5242 				update = true;
5243 
5244 			while (ex_start <= ex_last) {
5245 				if (SHIFT == SHIFT_LEFT) {
5246 					le32_add_cpu(&ex_start->ee_block,
5247 						-shift);
5248 					/* Try to merge to the left. */
5249 					if ((ex_start >
5250 					    EXT_FIRST_EXTENT(path[depth].p_hdr))
5251 					    &&
5252 					    ext4_ext_try_to_merge_right(inode,
5253 					    path, ex_start - 1))
5254 						ex_last--;
5255 					else
5256 						ex_start++;
5257 				} else {
5258 					le32_add_cpu(&ex_last->ee_block, shift);
5259 					ext4_ext_try_to_merge_right(inode, path,
5260 						ex_last);
5261 					ex_last--;
5262 				}
5263 			}
5264 			err = ext4_ext_dirty(handle, inode, path + depth);
5265 			if (err)
5266 				goto out;
5267 
5268 			if (--depth < 0 || !update)
5269 				break;
5270 		}
5271 
5272 		/* Update index too */
5273 		err = ext4_access_path(handle, inode, path + depth);
5274 		if (err)
5275 			goto out;
5276 
5277 		if (SHIFT == SHIFT_LEFT)
5278 			le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5279 		else
5280 			le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5281 		err = ext4_ext_dirty(handle, inode, path + depth);
5282 		if (err)
5283 			goto out;
5284 
5285 		/* we are done if current index is not a starting index */
5286 		if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5287 			break;
5288 
5289 		depth--;
5290 	}
5291 
5292 out:
5293 	return err;
5294 }
5295 
5296 /*
5297  * ext4_ext_shift_extents:
5298  * All the extents which lies in the range from @start to the last allocated
5299  * block for the @inode are shifted either towards left or right (depending
5300  * upon @SHIFT) by @shift blocks.
5301  * On success, 0 is returned, error otherwise.
5302  */
5303 static int
5304 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5305 		       ext4_lblk_t start, ext4_lblk_t shift,
5306 		       enum SHIFT_DIRECTION SHIFT)
5307 {
5308 	struct ext4_ext_path *path;
5309 	int ret = 0, depth;
5310 	struct ext4_extent *extent;
5311 	ext4_lblk_t stop, *iterator, ex_start, ex_end;
5312 
5313 	/* Let path point to the last extent */
5314 	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5315 				EXT4_EX_NOCACHE);
5316 	if (IS_ERR(path))
5317 		return PTR_ERR(path);
5318 
5319 	depth = path->p_depth;
5320 	extent = path[depth].p_ext;
5321 	if (!extent)
5322 		goto out;
5323 
5324 	stop = le32_to_cpu(extent->ee_block);
5325 
5326        /*
5327 	* For left shifts, make sure the hole on the left is big enough to
5328 	* accommodate the shift.  For right shifts, make sure the last extent
5329 	* won't be shifted beyond EXT_MAX_BLOCKS.
5330 	*/
5331 	if (SHIFT == SHIFT_LEFT) {
5332 		path = ext4_find_extent(inode, start - 1, &path,
5333 					EXT4_EX_NOCACHE);
5334 		if (IS_ERR(path))
5335 			return PTR_ERR(path);
5336 		depth = path->p_depth;
5337 		extent =  path[depth].p_ext;
5338 		if (extent) {
5339 			ex_start = le32_to_cpu(extent->ee_block);
5340 			ex_end = le32_to_cpu(extent->ee_block) +
5341 				ext4_ext_get_actual_len(extent);
5342 		} else {
5343 			ex_start = 0;
5344 			ex_end = 0;
5345 		}
5346 
5347 		if ((start == ex_start && shift > ex_start) ||
5348 		    (shift > start - ex_end)) {
5349 			ret = -EINVAL;
5350 			goto out;
5351 		}
5352 	} else {
5353 		if (shift > EXT_MAX_BLOCKS -
5354 		    (stop + ext4_ext_get_actual_len(extent))) {
5355 			ret = -EINVAL;
5356 			goto out;
5357 		}
5358 	}
5359 
5360 	/*
5361 	 * In case of left shift, iterator points to start and it is increased
5362 	 * till we reach stop. In case of right shift, iterator points to stop
5363 	 * and it is decreased till we reach start.
5364 	 */
5365 	if (SHIFT == SHIFT_LEFT)
5366 		iterator = &start;
5367 	else
5368 		iterator = &stop;
5369 
5370 	/*
5371 	 * Its safe to start updating extents.  Start and stop are unsigned, so
5372 	 * in case of right shift if extent with 0 block is reached, iterator
5373 	 * becomes NULL to indicate the end of the loop.
5374 	 */
5375 	while (iterator && start <= stop) {
5376 		path = ext4_find_extent(inode, *iterator, &path,
5377 					EXT4_EX_NOCACHE);
5378 		if (IS_ERR(path))
5379 			return PTR_ERR(path);
5380 		depth = path->p_depth;
5381 		extent = path[depth].p_ext;
5382 		if (!extent) {
5383 			EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5384 					 (unsigned long) *iterator);
5385 			return -EFSCORRUPTED;
5386 		}
5387 		if (SHIFT == SHIFT_LEFT && *iterator >
5388 		    le32_to_cpu(extent->ee_block)) {
5389 			/* Hole, move to the next extent */
5390 			if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5391 				path[depth].p_ext++;
5392 			} else {
5393 				*iterator = ext4_ext_next_allocated_block(path);
5394 				continue;
5395 			}
5396 		}
5397 
5398 		if (SHIFT == SHIFT_LEFT) {
5399 			extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5400 			*iterator = le32_to_cpu(extent->ee_block) +
5401 					ext4_ext_get_actual_len(extent);
5402 		} else {
5403 			extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5404 			if (le32_to_cpu(extent->ee_block) > 0)
5405 				*iterator = le32_to_cpu(extent->ee_block) - 1;
5406 			else
5407 				/* Beginning is reached, end of the loop */
5408 				iterator = NULL;
5409 			/* Update path extent in case we need to stop */
5410 			while (le32_to_cpu(extent->ee_block) < start)
5411 				extent++;
5412 			path[depth].p_ext = extent;
5413 		}
5414 		ret = ext4_ext_shift_path_extents(path, shift, inode,
5415 				handle, SHIFT);
5416 		if (ret)
5417 			break;
5418 	}
5419 out:
5420 	ext4_ext_drop_refs(path);
5421 	kfree(path);
5422 	return ret;
5423 }
5424 
5425 /*
5426  * ext4_collapse_range:
5427  * This implements the fallocate's collapse range functionality for ext4
5428  * Returns: 0 and non-zero on error.
5429  */
5430 static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5431 {
5432 	struct super_block *sb = inode->i_sb;
5433 	ext4_lblk_t punch_start, punch_stop;
5434 	handle_t *handle;
5435 	unsigned int credits;
5436 	loff_t new_size, ioffset;
5437 	int ret;
5438 
5439 	/*
5440 	 * We need to test this early because xfstests assumes that a
5441 	 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5442 	 * system does not support collapse range.
5443 	 */
5444 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5445 		return -EOPNOTSUPP;
5446 
5447 	/* Collapse range works only on fs cluster size aligned regions. */
5448 	if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5449 		return -EINVAL;
5450 
5451 	trace_ext4_collapse_range(inode, offset, len);
5452 
5453 	punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5454 	punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5455 
5456 	/* Call ext4_force_commit to flush all data in case of data=journal. */
5457 	if (ext4_should_journal_data(inode)) {
5458 		ret = ext4_force_commit(inode->i_sb);
5459 		if (ret)
5460 			return ret;
5461 	}
5462 
5463 	inode_lock(inode);
5464 	/*
5465 	 * There is no need to overlap collapse range with EOF, in which case
5466 	 * it is effectively a truncate operation
5467 	 */
5468 	if (offset + len >= inode->i_size) {
5469 		ret = -EINVAL;
5470 		goto out_mutex;
5471 	}
5472 
5473 	/* Currently just for extent based files */
5474 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5475 		ret = -EOPNOTSUPP;
5476 		goto out_mutex;
5477 	}
5478 
5479 	/* Wait for existing dio to complete */
5480 	inode_dio_wait(inode);
5481 
5482 	/*
5483 	 * Prevent page faults from reinstantiating pages we have released from
5484 	 * page cache.
5485 	 */
5486 	down_write(&EXT4_I(inode)->i_mmap_sem);
5487 
5488 	ret = ext4_break_layouts(inode);
5489 	if (ret)
5490 		goto out_mmap;
5491 
5492 	/*
5493 	 * Need to round down offset to be aligned with page size boundary
5494 	 * for page size > block size.
5495 	 */
5496 	ioffset = round_down(offset, PAGE_SIZE);
5497 	/*
5498 	 * Write tail of the last page before removed range since it will get
5499 	 * removed from the page cache below.
5500 	 */
5501 	ret = filemap_write_and_wait_range(inode->i_mapping, ioffset, offset);
5502 	if (ret)
5503 		goto out_mmap;
5504 	/*
5505 	 * Write data that will be shifted to preserve them when discarding
5506 	 * page cache below. We are also protected from pages becoming dirty
5507 	 * by i_mmap_sem.
5508 	 */
5509 	ret = filemap_write_and_wait_range(inode->i_mapping, offset + len,
5510 					   LLONG_MAX);
5511 	if (ret)
5512 		goto out_mmap;
5513 	truncate_pagecache(inode, ioffset);
5514 
5515 	credits = ext4_writepage_trans_blocks(inode);
5516 	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5517 	if (IS_ERR(handle)) {
5518 		ret = PTR_ERR(handle);
5519 		goto out_mmap;
5520 	}
5521 
5522 	down_write(&EXT4_I(inode)->i_data_sem);
5523 	ext4_discard_preallocations(inode);
5524 
5525 	ret = ext4_es_remove_extent(inode, punch_start,
5526 				    EXT_MAX_BLOCKS - punch_start);
5527 	if (ret) {
5528 		up_write(&EXT4_I(inode)->i_data_sem);
5529 		goto out_stop;
5530 	}
5531 
5532 	ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5533 	if (ret) {
5534 		up_write(&EXT4_I(inode)->i_data_sem);
5535 		goto out_stop;
5536 	}
5537 	ext4_discard_preallocations(inode);
5538 
5539 	ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5540 				     punch_stop - punch_start, SHIFT_LEFT);
5541 	if (ret) {
5542 		up_write(&EXT4_I(inode)->i_data_sem);
5543 		goto out_stop;
5544 	}
5545 
5546 	new_size = inode->i_size - len;
5547 	i_size_write(inode, new_size);
5548 	EXT4_I(inode)->i_disksize = new_size;
5549 
5550 	up_write(&EXT4_I(inode)->i_data_sem);
5551 	if (IS_SYNC(inode))
5552 		ext4_handle_sync(handle);
5553 	inode->i_mtime = inode->i_ctime = current_time(inode);
5554 	ext4_mark_inode_dirty(handle, inode);
5555 	ext4_update_inode_fsync_trans(handle, inode, 1);
5556 
5557 out_stop:
5558 	ext4_journal_stop(handle);
5559 out_mmap:
5560 	up_write(&EXT4_I(inode)->i_mmap_sem);
5561 out_mutex:
5562 	inode_unlock(inode);
5563 	return ret;
5564 }
5565 
5566 /*
5567  * ext4_insert_range:
5568  * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5569  * The data blocks starting from @offset to the EOF are shifted by @len
5570  * towards right to create a hole in the @inode. Inode size is increased
5571  * by len bytes.
5572  * Returns 0 on success, error otherwise.
5573  */
5574 static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5575 {
5576 	struct super_block *sb = inode->i_sb;
5577 	handle_t *handle;
5578 	struct ext4_ext_path *path;
5579 	struct ext4_extent *extent;
5580 	ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5581 	unsigned int credits, ee_len;
5582 	int ret = 0, depth, split_flag = 0;
5583 	loff_t ioffset;
5584 
5585 	/*
5586 	 * We need to test this early because xfstests assumes that an
5587 	 * insert range of (0, 1) will return EOPNOTSUPP if the file
5588 	 * system does not support insert range.
5589 	 */
5590 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5591 		return -EOPNOTSUPP;
5592 
5593 	/* Insert range works only on fs cluster size aligned regions. */
5594 	if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5595 		return -EINVAL;
5596 
5597 	trace_ext4_insert_range(inode, offset, len);
5598 
5599 	offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5600 	len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5601 
5602 	/* Call ext4_force_commit to flush all data in case of data=journal */
5603 	if (ext4_should_journal_data(inode)) {
5604 		ret = ext4_force_commit(inode->i_sb);
5605 		if (ret)
5606 			return ret;
5607 	}
5608 
5609 	inode_lock(inode);
5610 	/* Currently just for extent based files */
5611 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5612 		ret = -EOPNOTSUPP;
5613 		goto out_mutex;
5614 	}
5615 
5616 	/* Check whether the maximum file size would be exceeded */
5617 	if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5618 		ret = -EFBIG;
5619 		goto out_mutex;
5620 	}
5621 
5622 	/* Offset must be less than i_size */
5623 	if (offset >= inode->i_size) {
5624 		ret = -EINVAL;
5625 		goto out_mutex;
5626 	}
5627 
5628 	/* Wait for existing dio to complete */
5629 	inode_dio_wait(inode);
5630 
5631 	/*
5632 	 * Prevent page faults from reinstantiating pages we have released from
5633 	 * page cache.
5634 	 */
5635 	down_write(&EXT4_I(inode)->i_mmap_sem);
5636 
5637 	ret = ext4_break_layouts(inode);
5638 	if (ret)
5639 		goto out_mmap;
5640 
5641 	/*
5642 	 * Need to round down to align start offset to page size boundary
5643 	 * for page size > block size.
5644 	 */
5645 	ioffset = round_down(offset, PAGE_SIZE);
5646 	/* Write out all dirty pages */
5647 	ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5648 			LLONG_MAX);
5649 	if (ret)
5650 		goto out_mmap;
5651 	truncate_pagecache(inode, ioffset);
5652 
5653 	credits = ext4_writepage_trans_blocks(inode);
5654 	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5655 	if (IS_ERR(handle)) {
5656 		ret = PTR_ERR(handle);
5657 		goto out_mmap;
5658 	}
5659 
5660 	/* Expand file to avoid data loss if there is error while shifting */
5661 	inode->i_size += len;
5662 	EXT4_I(inode)->i_disksize += len;
5663 	inode->i_mtime = inode->i_ctime = current_time(inode);
5664 	ret = ext4_mark_inode_dirty(handle, inode);
5665 	if (ret)
5666 		goto out_stop;
5667 
5668 	down_write(&EXT4_I(inode)->i_data_sem);
5669 	ext4_discard_preallocations(inode);
5670 
5671 	path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5672 	if (IS_ERR(path)) {
5673 		up_write(&EXT4_I(inode)->i_data_sem);
5674 		goto out_stop;
5675 	}
5676 
5677 	depth = ext_depth(inode);
5678 	extent = path[depth].p_ext;
5679 	if (extent) {
5680 		ee_start_lblk = le32_to_cpu(extent->ee_block);
5681 		ee_len = ext4_ext_get_actual_len(extent);
5682 
5683 		/*
5684 		 * If offset_lblk is not the starting block of extent, split
5685 		 * the extent @offset_lblk
5686 		 */
5687 		if ((offset_lblk > ee_start_lblk) &&
5688 				(offset_lblk < (ee_start_lblk + ee_len))) {
5689 			if (ext4_ext_is_unwritten(extent))
5690 				split_flag = EXT4_EXT_MARK_UNWRIT1 |
5691 					EXT4_EXT_MARK_UNWRIT2;
5692 			ret = ext4_split_extent_at(handle, inode, &path,
5693 					offset_lblk, split_flag,
5694 					EXT4_EX_NOCACHE |
5695 					EXT4_GET_BLOCKS_PRE_IO |
5696 					EXT4_GET_BLOCKS_METADATA_NOFAIL);
5697 		}
5698 
5699 		ext4_ext_drop_refs(path);
5700 		kfree(path);
5701 		if (ret < 0) {
5702 			up_write(&EXT4_I(inode)->i_data_sem);
5703 			goto out_stop;
5704 		}
5705 	} else {
5706 		ext4_ext_drop_refs(path);
5707 		kfree(path);
5708 	}
5709 
5710 	ret = ext4_es_remove_extent(inode, offset_lblk,
5711 			EXT_MAX_BLOCKS - offset_lblk);
5712 	if (ret) {
5713 		up_write(&EXT4_I(inode)->i_data_sem);
5714 		goto out_stop;
5715 	}
5716 
5717 	/*
5718 	 * if offset_lblk lies in a hole which is at start of file, use
5719 	 * ee_start_lblk to shift extents
5720 	 */
5721 	ret = ext4_ext_shift_extents(inode, handle,
5722 		ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5723 		len_lblk, SHIFT_RIGHT);
5724 
5725 	up_write(&EXT4_I(inode)->i_data_sem);
5726 	if (IS_SYNC(inode))
5727 		ext4_handle_sync(handle);
5728 	if (ret >= 0)
5729 		ext4_update_inode_fsync_trans(handle, inode, 1);
5730 
5731 out_stop:
5732 	ext4_journal_stop(handle);
5733 out_mmap:
5734 	up_write(&EXT4_I(inode)->i_mmap_sem);
5735 out_mutex:
5736 	inode_unlock(inode);
5737 	return ret;
5738 }
5739 
5740 /**
5741  * ext4_swap_extents() - Swap extents between two inodes
5742  * @handle: handle for this transaction
5743  * @inode1:	First inode
5744  * @inode2:	Second inode
5745  * @lblk1:	Start block for first inode
5746  * @lblk2:	Start block for second inode
5747  * @count:	Number of blocks to swap
5748  * @unwritten: Mark second inode's extents as unwritten after swap
5749  * @erp:	Pointer to save error value
5750  *
5751  * This helper routine does exactly what is promise "swap extents". All other
5752  * stuff such as page-cache locking consistency, bh mapping consistency or
5753  * extent's data copying must be performed by caller.
5754  * Locking:
5755  * 		i_mutex is held for both inodes
5756  * 		i_data_sem is locked for write for both inodes
5757  * Assumptions:
5758  *		All pages from requested range are locked for both inodes
5759  */
5760 int
5761 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5762 		  struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5763 		  ext4_lblk_t count, int unwritten, int *erp)
5764 {
5765 	struct ext4_ext_path *path1 = NULL;
5766 	struct ext4_ext_path *path2 = NULL;
5767 	int replaced_count = 0;
5768 
5769 	BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5770 	BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5771 	BUG_ON(!inode_is_locked(inode1));
5772 	BUG_ON(!inode_is_locked(inode2));
5773 
5774 	*erp = ext4_es_remove_extent(inode1, lblk1, count);
5775 	if (unlikely(*erp))
5776 		return 0;
5777 	*erp = ext4_es_remove_extent(inode2, lblk2, count);
5778 	if (unlikely(*erp))
5779 		return 0;
5780 
5781 	while (count) {
5782 		struct ext4_extent *ex1, *ex2, tmp_ex;
5783 		ext4_lblk_t e1_blk, e2_blk;
5784 		int e1_len, e2_len, len;
5785 		int split = 0;
5786 
5787 		path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5788 		if (IS_ERR(path1)) {
5789 			*erp = PTR_ERR(path1);
5790 			path1 = NULL;
5791 		finish:
5792 			count = 0;
5793 			goto repeat;
5794 		}
5795 		path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5796 		if (IS_ERR(path2)) {
5797 			*erp = PTR_ERR(path2);
5798 			path2 = NULL;
5799 			goto finish;
5800 		}
5801 		ex1 = path1[path1->p_depth].p_ext;
5802 		ex2 = path2[path2->p_depth].p_ext;
5803 		/* Do we have somthing to swap ? */
5804 		if (unlikely(!ex2 || !ex1))
5805 			goto finish;
5806 
5807 		e1_blk = le32_to_cpu(ex1->ee_block);
5808 		e2_blk = le32_to_cpu(ex2->ee_block);
5809 		e1_len = ext4_ext_get_actual_len(ex1);
5810 		e2_len = ext4_ext_get_actual_len(ex2);
5811 
5812 		/* Hole handling */
5813 		if (!in_range(lblk1, e1_blk, e1_len) ||
5814 		    !in_range(lblk2, e2_blk, e2_len)) {
5815 			ext4_lblk_t next1, next2;
5816 
5817 			/* if hole after extent, then go to next extent */
5818 			next1 = ext4_ext_next_allocated_block(path1);
5819 			next2 = ext4_ext_next_allocated_block(path2);
5820 			/* If hole before extent, then shift to that extent */
5821 			if (e1_blk > lblk1)
5822 				next1 = e1_blk;
5823 			if (e2_blk > lblk2)
5824 				next2 = e2_blk;
5825 			/* Do we have something to swap */
5826 			if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5827 				goto finish;
5828 			/* Move to the rightest boundary */
5829 			len = next1 - lblk1;
5830 			if (len < next2 - lblk2)
5831 				len = next2 - lblk2;
5832 			if (len > count)
5833 				len = count;
5834 			lblk1 += len;
5835 			lblk2 += len;
5836 			count -= len;
5837 			goto repeat;
5838 		}
5839 
5840 		/* Prepare left boundary */
5841 		if (e1_blk < lblk1) {
5842 			split = 1;
5843 			*erp = ext4_force_split_extent_at(handle, inode1,
5844 						&path1, lblk1, 0);
5845 			if (unlikely(*erp))
5846 				goto finish;
5847 		}
5848 		if (e2_blk < lblk2) {
5849 			split = 1;
5850 			*erp = ext4_force_split_extent_at(handle, inode2,
5851 						&path2,  lblk2, 0);
5852 			if (unlikely(*erp))
5853 				goto finish;
5854 		}
5855 		/* ext4_split_extent_at() may result in leaf extent split,
5856 		 * path must to be revalidated. */
5857 		if (split)
5858 			goto repeat;
5859 
5860 		/* Prepare right boundary */
5861 		len = count;
5862 		if (len > e1_blk + e1_len - lblk1)
5863 			len = e1_blk + e1_len - lblk1;
5864 		if (len > e2_blk + e2_len - lblk2)
5865 			len = e2_blk + e2_len - lblk2;
5866 
5867 		if (len != e1_len) {
5868 			split = 1;
5869 			*erp = ext4_force_split_extent_at(handle, inode1,
5870 						&path1, lblk1 + len, 0);
5871 			if (unlikely(*erp))
5872 				goto finish;
5873 		}
5874 		if (len != e2_len) {
5875 			split = 1;
5876 			*erp = ext4_force_split_extent_at(handle, inode2,
5877 						&path2, lblk2 + len, 0);
5878 			if (*erp)
5879 				goto finish;
5880 		}
5881 		/* ext4_split_extent_at() may result in leaf extent split,
5882 		 * path must to be revalidated. */
5883 		if (split)
5884 			goto repeat;
5885 
5886 		BUG_ON(e2_len != e1_len);
5887 		*erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5888 		if (unlikely(*erp))
5889 			goto finish;
5890 		*erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5891 		if (unlikely(*erp))
5892 			goto finish;
5893 
5894 		/* Both extents are fully inside boundaries. Swap it now */
5895 		tmp_ex = *ex1;
5896 		ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5897 		ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5898 		ex1->ee_len = cpu_to_le16(e2_len);
5899 		ex2->ee_len = cpu_to_le16(e1_len);
5900 		if (unwritten)
5901 			ext4_ext_mark_unwritten(ex2);
5902 		if (ext4_ext_is_unwritten(&tmp_ex))
5903 			ext4_ext_mark_unwritten(ex1);
5904 
5905 		ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5906 		ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5907 		*erp = ext4_ext_dirty(handle, inode2, path2 +
5908 				      path2->p_depth);
5909 		if (unlikely(*erp))
5910 			goto finish;
5911 		*erp = ext4_ext_dirty(handle, inode1, path1 +
5912 				      path1->p_depth);
5913 		/*
5914 		 * Looks scarry ah..? second inode already points to new blocks,
5915 		 * and it was successfully dirtied. But luckily error may happen
5916 		 * only due to journal error, so full transaction will be
5917 		 * aborted anyway.
5918 		 */
5919 		if (unlikely(*erp))
5920 			goto finish;
5921 		lblk1 += len;
5922 		lblk2 += len;
5923 		replaced_count += len;
5924 		count -= len;
5925 
5926 	repeat:
5927 		ext4_ext_drop_refs(path1);
5928 		kfree(path1);
5929 		ext4_ext_drop_refs(path2);
5930 		kfree(path2);
5931 		path1 = path2 = NULL;
5932 	}
5933 	return replaced_count;
5934 }
5935 
5936 /*
5937  * ext4_clu_mapped - determine whether any block in a logical cluster has
5938  *                   been mapped to a physical cluster
5939  *
5940  * @inode - file containing the logical cluster
5941  * @lclu - logical cluster of interest
5942  *
5943  * Returns 1 if any block in the logical cluster is mapped, signifying
5944  * that a physical cluster has been allocated for it.  Otherwise,
5945  * returns 0.  Can also return negative error codes.  Derived from
5946  * ext4_ext_map_blocks().
5947  */
5948 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5949 {
5950 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5951 	struct ext4_ext_path *path;
5952 	int depth, mapped = 0, err = 0;
5953 	struct ext4_extent *extent;
5954 	ext4_lblk_t first_lblk, first_lclu, last_lclu;
5955 
5956 	/* search for the extent closest to the first block in the cluster */
5957 	path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5958 	if (IS_ERR(path)) {
5959 		err = PTR_ERR(path);
5960 		path = NULL;
5961 		goto out;
5962 	}
5963 
5964 	depth = ext_depth(inode);
5965 
5966 	/*
5967 	 * A consistent leaf must not be empty.  This situation is possible,
5968 	 * though, _during_ tree modification, and it's why an assert can't
5969 	 * be put in ext4_find_extent().
5970 	 */
5971 	if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5972 		EXT4_ERROR_INODE(inode,
5973 		    "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5974 				 (unsigned long) EXT4_C2B(sbi, lclu),
5975 				 depth, path[depth].p_block);
5976 		err = -EFSCORRUPTED;
5977 		goto out;
5978 	}
5979 
5980 	extent = path[depth].p_ext;
5981 
5982 	/* can't be mapped if the extent tree is empty */
5983 	if (extent == NULL)
5984 		goto out;
5985 
5986 	first_lblk = le32_to_cpu(extent->ee_block);
5987 	first_lclu = EXT4_B2C(sbi, first_lblk);
5988 
5989 	/*
5990 	 * Three possible outcomes at this point - found extent spanning
5991 	 * the target cluster, to the left of the target cluster, or to the
5992 	 * right of the target cluster.  The first two cases are handled here.
5993 	 * The last case indicates the target cluster is not mapped.
5994 	 */
5995 	if (lclu >= first_lclu) {
5996 		last_lclu = EXT4_B2C(sbi, first_lblk +
5997 				     ext4_ext_get_actual_len(extent) - 1);
5998 		if (lclu <= last_lclu) {
5999 			mapped = 1;
6000 		} else {
6001 			first_lblk = ext4_ext_next_allocated_block(path);
6002 			first_lclu = EXT4_B2C(sbi, first_lblk);
6003 			if (lclu == first_lclu)
6004 				mapped = 1;
6005 		}
6006 	}
6007 
6008 out:
6009 	ext4_ext_drop_refs(path);
6010 	kfree(path);
6011 
6012 	return err ? err : mapped;
6013 }
6014