xref: /openbmc/linux/fs/ext4/extents.c (revision 901181b7)
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 <linux/iomap.h>
32 #include "ext4_jbd2.h"
33 #include "ext4_extents.h"
34 #include "xattr.h"
35 
36 #include <trace/events/ext4.h>
37 
38 /*
39  * used by extent splitting.
40  */
41 #define EXT4_EXT_MAY_ZEROOUT	0x1  /* safe to zeroout if split fails \
42 					due to ENOSPC */
43 #define EXT4_EXT_MARK_UNWRIT1	0x2  /* mark first half unwritten */
44 #define EXT4_EXT_MARK_UNWRIT2	0x4  /* mark second half unwritten */
45 
46 #define EXT4_EXT_DATA_VALID1	0x8  /* first half contains valid data */
47 #define EXT4_EXT_DATA_VALID2	0x10 /* second half contains valid data */
48 
49 static __le32 ext4_extent_block_csum(struct inode *inode,
50 				     struct ext4_extent_header *eh)
51 {
52 	struct ext4_inode_info *ei = EXT4_I(inode);
53 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
54 	__u32 csum;
55 
56 	csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)eh,
57 			   EXT4_EXTENT_TAIL_OFFSET(eh));
58 	return cpu_to_le32(csum);
59 }
60 
61 static int ext4_extent_block_csum_verify(struct inode *inode,
62 					 struct ext4_extent_header *eh)
63 {
64 	struct ext4_extent_tail *et;
65 
66 	if (!ext4_has_metadata_csum(inode->i_sb))
67 		return 1;
68 
69 	et = find_ext4_extent_tail(eh);
70 	if (et->et_checksum != ext4_extent_block_csum(inode, eh))
71 		return 0;
72 	return 1;
73 }
74 
75 static void ext4_extent_block_csum_set(struct inode *inode,
76 				       struct ext4_extent_header *eh)
77 {
78 	struct ext4_extent_tail *et;
79 
80 	if (!ext4_has_metadata_csum(inode->i_sb))
81 		return;
82 
83 	et = find_ext4_extent_tail(eh);
84 	et->et_checksum = ext4_extent_block_csum(inode, eh);
85 }
86 
87 static int ext4_split_extent_at(handle_t *handle,
88 			     struct inode *inode,
89 			     struct ext4_ext_path **ppath,
90 			     ext4_lblk_t split,
91 			     int split_flag,
92 			     int flags);
93 
94 static int ext4_ext_trunc_restart_fn(struct inode *inode, int *dropped)
95 {
96 	/*
97 	 * Drop i_data_sem to avoid deadlock with ext4_map_blocks.  At this
98 	 * moment, get_block can be called only for blocks inside i_size since
99 	 * page cache has been already dropped and writes are blocked by
100 	 * i_mutex. So we can safely drop the i_data_sem here.
101 	 */
102 	BUG_ON(EXT4_JOURNAL(inode) == NULL);
103 	ext4_discard_preallocations(inode, 0);
104 	up_write(&EXT4_I(inode)->i_data_sem);
105 	*dropped = 1;
106 	return 0;
107 }
108 
109 /*
110  * Make sure 'handle' has at least 'check_cred' credits. If not, restart
111  * transaction with 'restart_cred' credits. The function drops i_data_sem
112  * when restarting transaction and gets it after transaction is restarted.
113  *
114  * The function returns 0 on success, 1 if transaction had to be restarted,
115  * and < 0 in case of fatal error.
116  */
117 int ext4_datasem_ensure_credits(handle_t *handle, struct inode *inode,
118 				int check_cred, int restart_cred,
119 				int revoke_cred)
120 {
121 	int ret;
122 	int dropped = 0;
123 
124 	ret = ext4_journal_ensure_credits_fn(handle, check_cred, restart_cred,
125 		revoke_cred, ext4_ext_trunc_restart_fn(inode, &dropped));
126 	if (dropped)
127 		down_write(&EXT4_I(inode)->i_data_sem);
128 	return ret;
129 }
130 
131 /*
132  * could return:
133  *  - EROFS
134  *  - ENOMEM
135  */
136 static int ext4_ext_get_access(handle_t *handle, struct inode *inode,
137 				struct ext4_ext_path *path)
138 {
139 	int err = 0;
140 
141 	if (path->p_bh) {
142 		/* path points to block */
143 		BUFFER_TRACE(path->p_bh, "get_write_access");
144 		err = ext4_journal_get_write_access(handle, inode->i_sb,
145 						    path->p_bh, EXT4_JTR_NONE);
146 		/*
147 		 * The extent buffer's verified bit will be set again in
148 		 * __ext4_ext_dirty(). We could leave an inconsistent
149 		 * buffer if the extents updating procudure break off du
150 		 * to some error happens, force to check it again.
151 		 */
152 		if (!err)
153 			clear_buffer_verified(path->p_bh);
154 	}
155 	/* path points to leaf/index in inode body */
156 	/* we use in-core data, no need to protect them */
157 	return err;
158 }
159 
160 /*
161  * could return:
162  *  - EROFS
163  *  - ENOMEM
164  *  - EIO
165  */
166 static int __ext4_ext_dirty(const char *where, unsigned int line,
167 			    handle_t *handle, struct inode *inode,
168 			    struct ext4_ext_path *path)
169 {
170 	int err;
171 
172 	WARN_ON(!rwsem_is_locked(&EXT4_I(inode)->i_data_sem));
173 	if (path->p_bh) {
174 		ext4_extent_block_csum_set(inode, ext_block_hdr(path->p_bh));
175 		/* path points to block */
176 		err = __ext4_handle_dirty_metadata(where, line, handle,
177 						   inode, path->p_bh);
178 		/* Extents updating done, re-set verified flag */
179 		if (!err)
180 			set_buffer_verified(path->p_bh);
181 	} else {
182 		/* path points to leaf/index in inode body */
183 		err = ext4_mark_inode_dirty(handle, inode);
184 	}
185 	return err;
186 }
187 
188 #define ext4_ext_dirty(handle, inode, path) \
189 		__ext4_ext_dirty(__func__, __LINE__, (handle), (inode), (path))
190 
191 static ext4_fsblk_t ext4_ext_find_goal(struct inode *inode,
192 			      struct ext4_ext_path *path,
193 			      ext4_lblk_t block)
194 {
195 	if (path) {
196 		int depth = path->p_depth;
197 		struct ext4_extent *ex;
198 
199 		/*
200 		 * Try to predict block placement assuming that we are
201 		 * filling in a file which will eventually be
202 		 * non-sparse --- i.e., in the case of libbfd writing
203 		 * an ELF object sections out-of-order but in a way
204 		 * the eventually results in a contiguous object or
205 		 * executable file, or some database extending a table
206 		 * space file.  However, this is actually somewhat
207 		 * non-ideal if we are writing a sparse file such as
208 		 * qemu or KVM writing a raw image file that is going
209 		 * to stay fairly sparse, since it will end up
210 		 * fragmenting the file system's free space.  Maybe we
211 		 * should have some hueristics or some way to allow
212 		 * userspace to pass a hint to file system,
213 		 * especially if the latter case turns out to be
214 		 * common.
215 		 */
216 		ex = path[depth].p_ext;
217 		if (ex) {
218 			ext4_fsblk_t ext_pblk = ext4_ext_pblock(ex);
219 			ext4_lblk_t ext_block = le32_to_cpu(ex->ee_block);
220 
221 			if (block > ext_block)
222 				return ext_pblk + (block - ext_block);
223 			else
224 				return ext_pblk - (ext_block - block);
225 		}
226 
227 		/* it looks like index is empty;
228 		 * try to find starting block from index itself */
229 		if (path[depth].p_bh)
230 			return path[depth].p_bh->b_blocknr;
231 	}
232 
233 	/* OK. use inode's group */
234 	return ext4_inode_to_goal_block(inode);
235 }
236 
237 /*
238  * Allocation for a meta data block
239  */
240 static ext4_fsblk_t
241 ext4_ext_new_meta_block(handle_t *handle, struct inode *inode,
242 			struct ext4_ext_path *path,
243 			struct ext4_extent *ex, int *err, unsigned int flags)
244 {
245 	ext4_fsblk_t goal, newblock;
246 
247 	goal = ext4_ext_find_goal(inode, path, le32_to_cpu(ex->ee_block));
248 	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
249 					NULL, err);
250 	return newblock;
251 }
252 
253 static inline int ext4_ext_space_block(struct inode *inode, int check)
254 {
255 	int size;
256 
257 	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
258 			/ sizeof(struct ext4_extent);
259 #ifdef AGGRESSIVE_TEST
260 	if (!check && size > 6)
261 		size = 6;
262 #endif
263 	return size;
264 }
265 
266 static inline int ext4_ext_space_block_idx(struct inode *inode, int check)
267 {
268 	int size;
269 
270 	size = (inode->i_sb->s_blocksize - sizeof(struct ext4_extent_header))
271 			/ sizeof(struct ext4_extent_idx);
272 #ifdef AGGRESSIVE_TEST
273 	if (!check && size > 5)
274 		size = 5;
275 #endif
276 	return size;
277 }
278 
279 static inline int ext4_ext_space_root(struct inode *inode, int check)
280 {
281 	int size;
282 
283 	size = sizeof(EXT4_I(inode)->i_data);
284 	size -= sizeof(struct ext4_extent_header);
285 	size /= sizeof(struct ext4_extent);
286 #ifdef AGGRESSIVE_TEST
287 	if (!check && size > 3)
288 		size = 3;
289 #endif
290 	return size;
291 }
292 
293 static inline int ext4_ext_space_root_idx(struct inode *inode, int check)
294 {
295 	int size;
296 
297 	size = sizeof(EXT4_I(inode)->i_data);
298 	size -= sizeof(struct ext4_extent_header);
299 	size /= sizeof(struct ext4_extent_idx);
300 #ifdef AGGRESSIVE_TEST
301 	if (!check && size > 4)
302 		size = 4;
303 #endif
304 	return size;
305 }
306 
307 static inline int
308 ext4_force_split_extent_at(handle_t *handle, struct inode *inode,
309 			   struct ext4_ext_path **ppath, ext4_lblk_t lblk,
310 			   int nofail)
311 {
312 	struct ext4_ext_path *path = *ppath;
313 	int unwritten = ext4_ext_is_unwritten(path[path->p_depth].p_ext);
314 	int flags = EXT4_EX_NOCACHE | EXT4_GET_BLOCKS_PRE_IO;
315 
316 	if (nofail)
317 		flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL | EXT4_EX_NOFAIL;
318 
319 	return ext4_split_extent_at(handle, inode, ppath, lblk, unwritten ?
320 			EXT4_EXT_MARK_UNWRIT1|EXT4_EXT_MARK_UNWRIT2 : 0,
321 			flags);
322 }
323 
324 static int
325 ext4_ext_max_entries(struct inode *inode, int depth)
326 {
327 	int max;
328 
329 	if (depth == ext_depth(inode)) {
330 		if (depth == 0)
331 			max = ext4_ext_space_root(inode, 1);
332 		else
333 			max = ext4_ext_space_root_idx(inode, 1);
334 	} else {
335 		if (depth == 0)
336 			max = ext4_ext_space_block(inode, 1);
337 		else
338 			max = ext4_ext_space_block_idx(inode, 1);
339 	}
340 
341 	return max;
342 }
343 
344 static int ext4_valid_extent(struct inode *inode, struct ext4_extent *ext)
345 {
346 	ext4_fsblk_t block = ext4_ext_pblock(ext);
347 	int len = ext4_ext_get_actual_len(ext);
348 	ext4_lblk_t lblock = le32_to_cpu(ext->ee_block);
349 
350 	/*
351 	 * We allow neither:
352 	 *  - zero length
353 	 *  - overflow/wrap-around
354 	 */
355 	if (lblock + len <= lblock)
356 		return 0;
357 	return ext4_inode_block_valid(inode, block, len);
358 }
359 
360 static int ext4_valid_extent_idx(struct inode *inode,
361 				struct ext4_extent_idx *ext_idx)
362 {
363 	ext4_fsblk_t block = ext4_idx_pblock(ext_idx);
364 
365 	return ext4_inode_block_valid(inode, block, 1);
366 }
367 
368 static int ext4_valid_extent_entries(struct inode *inode,
369 				     struct ext4_extent_header *eh,
370 				     ext4_lblk_t lblk, ext4_fsblk_t *pblk,
371 				     int depth)
372 {
373 	unsigned short entries;
374 	ext4_lblk_t lblock = 0;
375 	ext4_lblk_t prev = 0;
376 
377 	if (eh->eh_entries == 0)
378 		return 1;
379 
380 	entries = le16_to_cpu(eh->eh_entries);
381 
382 	if (depth == 0) {
383 		/* leaf entries */
384 		struct ext4_extent *ext = EXT_FIRST_EXTENT(eh);
385 
386 		/*
387 		 * The logical block in the first entry should equal to
388 		 * the number in the index block.
389 		 */
390 		if (depth != ext_depth(inode) &&
391 		    lblk != le32_to_cpu(ext->ee_block))
392 			return 0;
393 		while (entries) {
394 			if (!ext4_valid_extent(inode, ext))
395 				return 0;
396 
397 			/* Check for overlapping extents */
398 			lblock = le32_to_cpu(ext->ee_block);
399 			if ((lblock <= prev) && prev) {
400 				*pblk = ext4_ext_pblock(ext);
401 				return 0;
402 			}
403 			prev = lblock + ext4_ext_get_actual_len(ext) - 1;
404 			ext++;
405 			entries--;
406 		}
407 	} else {
408 		struct ext4_extent_idx *ext_idx = EXT_FIRST_INDEX(eh);
409 
410 		/*
411 		 * The logical block in the first entry should equal to
412 		 * the number in the parent index block.
413 		 */
414 		if (depth != ext_depth(inode) &&
415 		    lblk != le32_to_cpu(ext_idx->ei_block))
416 			return 0;
417 		while (entries) {
418 			if (!ext4_valid_extent_idx(inode, ext_idx))
419 				return 0;
420 
421 			/* Check for overlapping index extents */
422 			lblock = le32_to_cpu(ext_idx->ei_block);
423 			if ((lblock <= prev) && prev) {
424 				*pblk = ext4_idx_pblock(ext_idx);
425 				return 0;
426 			}
427 			ext_idx++;
428 			entries--;
429 			prev = lblock;
430 		}
431 	}
432 	return 1;
433 }
434 
435 static int __ext4_ext_check(const char *function, unsigned int line,
436 			    struct inode *inode, struct ext4_extent_header *eh,
437 			    int depth, ext4_fsblk_t pblk, ext4_lblk_t lblk)
438 {
439 	const char *error_msg;
440 	int max = 0, err = -EFSCORRUPTED;
441 
442 	if (unlikely(eh->eh_magic != EXT4_EXT_MAGIC)) {
443 		error_msg = "invalid magic";
444 		goto corrupted;
445 	}
446 	if (unlikely(le16_to_cpu(eh->eh_depth) != depth)) {
447 		error_msg = "unexpected eh_depth";
448 		goto corrupted;
449 	}
450 	if (unlikely(eh->eh_max == 0)) {
451 		error_msg = "invalid eh_max";
452 		goto corrupted;
453 	}
454 	max = ext4_ext_max_entries(inode, depth);
455 	if (unlikely(le16_to_cpu(eh->eh_max) > max)) {
456 		error_msg = "too large eh_max";
457 		goto corrupted;
458 	}
459 	if (unlikely(le16_to_cpu(eh->eh_entries) > le16_to_cpu(eh->eh_max))) {
460 		error_msg = "invalid eh_entries";
461 		goto corrupted;
462 	}
463 	if (!ext4_valid_extent_entries(inode, eh, lblk, &pblk, depth)) {
464 		error_msg = "invalid extent entries";
465 		goto corrupted;
466 	}
467 	if (unlikely(depth > 32)) {
468 		error_msg = "too large eh_depth";
469 		goto corrupted;
470 	}
471 	/* Verify checksum on non-root extent tree nodes */
472 	if (ext_depth(inode) != depth &&
473 	    !ext4_extent_block_csum_verify(inode, eh)) {
474 		error_msg = "extent tree corrupted";
475 		err = -EFSBADCRC;
476 		goto corrupted;
477 	}
478 	return 0;
479 
480 corrupted:
481 	ext4_error_inode_err(inode, function, line, 0, -err,
482 			     "pblk %llu bad header/extent: %s - magic %x, "
483 			     "entries %u, max %u(%u), depth %u(%u)",
484 			     (unsigned long long) pblk, error_msg,
485 			     le16_to_cpu(eh->eh_magic),
486 			     le16_to_cpu(eh->eh_entries),
487 			     le16_to_cpu(eh->eh_max),
488 			     max, le16_to_cpu(eh->eh_depth), depth);
489 	return err;
490 }
491 
492 #define ext4_ext_check(inode, eh, depth, pblk)			\
493 	__ext4_ext_check(__func__, __LINE__, (inode), (eh), (depth), (pblk), 0)
494 
495 int ext4_ext_check_inode(struct inode *inode)
496 {
497 	return ext4_ext_check(inode, ext_inode_hdr(inode), ext_depth(inode), 0);
498 }
499 
500 static void ext4_cache_extents(struct inode *inode,
501 			       struct ext4_extent_header *eh)
502 {
503 	struct ext4_extent *ex = EXT_FIRST_EXTENT(eh);
504 	ext4_lblk_t prev = 0;
505 	int i;
506 
507 	for (i = le16_to_cpu(eh->eh_entries); i > 0; i--, ex++) {
508 		unsigned int status = EXTENT_STATUS_WRITTEN;
509 		ext4_lblk_t lblk = le32_to_cpu(ex->ee_block);
510 		int len = ext4_ext_get_actual_len(ex);
511 
512 		if (prev && (prev != lblk))
513 			ext4_es_cache_extent(inode, prev, lblk - prev, ~0,
514 					     EXTENT_STATUS_HOLE);
515 
516 		if (ext4_ext_is_unwritten(ex))
517 			status = EXTENT_STATUS_UNWRITTEN;
518 		ext4_es_cache_extent(inode, lblk, len,
519 				     ext4_ext_pblock(ex), status);
520 		prev = lblk + len;
521 	}
522 }
523 
524 static struct buffer_head *
525 __read_extent_tree_block(const char *function, unsigned int line,
526 			 struct inode *inode, struct ext4_extent_idx *idx,
527 			 int depth, int flags)
528 {
529 	struct buffer_head		*bh;
530 	int				err;
531 	gfp_t				gfp_flags = __GFP_MOVABLE | GFP_NOFS;
532 	ext4_fsblk_t			pblk;
533 
534 	if (flags & EXT4_EX_NOFAIL)
535 		gfp_flags |= __GFP_NOFAIL;
536 
537 	pblk = ext4_idx_pblock(idx);
538 	bh = sb_getblk_gfp(inode->i_sb, pblk, gfp_flags);
539 	if (unlikely(!bh))
540 		return ERR_PTR(-ENOMEM);
541 
542 	if (!bh_uptodate_or_lock(bh)) {
543 		trace_ext4_ext_load_extent(inode, pblk, _RET_IP_);
544 		err = ext4_read_bh(bh, 0, NULL);
545 		if (err < 0)
546 			goto errout;
547 	}
548 	if (buffer_verified(bh) && !(flags & EXT4_EX_FORCE_CACHE))
549 		return bh;
550 	err = __ext4_ext_check(function, line, inode, ext_block_hdr(bh),
551 			       depth, pblk, le32_to_cpu(idx->ei_block));
552 	if (err)
553 		goto errout;
554 	set_buffer_verified(bh);
555 	/*
556 	 * If this is a leaf block, cache all of its entries
557 	 */
558 	if (!(flags & EXT4_EX_NOCACHE) && depth == 0) {
559 		struct ext4_extent_header *eh = ext_block_hdr(bh);
560 		ext4_cache_extents(inode, eh);
561 	}
562 	return bh;
563 errout:
564 	put_bh(bh);
565 	return ERR_PTR(err);
566 
567 }
568 
569 #define read_extent_tree_block(inode, idx, depth, flags)		\
570 	__read_extent_tree_block(__func__, __LINE__, (inode), (idx),	\
571 				 (depth), (flags))
572 
573 /*
574  * This function is called to cache a file's extent information in the
575  * extent status tree
576  */
577 int ext4_ext_precache(struct inode *inode)
578 {
579 	struct ext4_inode_info *ei = EXT4_I(inode);
580 	struct ext4_ext_path *path = NULL;
581 	struct buffer_head *bh;
582 	int i = 0, depth, ret = 0;
583 
584 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
585 		return 0;	/* not an extent-mapped inode */
586 
587 	down_read(&ei->i_data_sem);
588 	depth = ext_depth(inode);
589 
590 	/* Don't cache anything if there are no external extent blocks */
591 	if (!depth) {
592 		up_read(&ei->i_data_sem);
593 		return ret;
594 	}
595 
596 	path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
597 		       GFP_NOFS);
598 	if (path == NULL) {
599 		up_read(&ei->i_data_sem);
600 		return -ENOMEM;
601 	}
602 
603 	path[0].p_hdr = ext_inode_hdr(inode);
604 	ret = ext4_ext_check(inode, path[0].p_hdr, depth, 0);
605 	if (ret)
606 		goto out;
607 	path[0].p_idx = EXT_FIRST_INDEX(path[0].p_hdr);
608 	while (i >= 0) {
609 		/*
610 		 * If this is a leaf block or we've reached the end of
611 		 * the index block, go up
612 		 */
613 		if ((i == depth) ||
614 		    path[i].p_idx > EXT_LAST_INDEX(path[i].p_hdr)) {
615 			brelse(path[i].p_bh);
616 			path[i].p_bh = NULL;
617 			i--;
618 			continue;
619 		}
620 		bh = read_extent_tree_block(inode, path[i].p_idx++,
621 					    depth - i - 1,
622 					    EXT4_EX_FORCE_CACHE);
623 		if (IS_ERR(bh)) {
624 			ret = PTR_ERR(bh);
625 			break;
626 		}
627 		i++;
628 		path[i].p_bh = bh;
629 		path[i].p_hdr = ext_block_hdr(bh);
630 		path[i].p_idx = EXT_FIRST_INDEX(path[i].p_hdr);
631 	}
632 	ext4_set_inode_state(inode, EXT4_STATE_EXT_PRECACHED);
633 out:
634 	up_read(&ei->i_data_sem);
635 	ext4_ext_drop_refs(path);
636 	kfree(path);
637 	return ret;
638 }
639 
640 #ifdef EXT_DEBUG
641 static void ext4_ext_show_path(struct inode *inode, struct ext4_ext_path *path)
642 {
643 	int k, l = path->p_depth;
644 
645 	ext_debug(inode, "path:");
646 	for (k = 0; k <= l; k++, path++) {
647 		if (path->p_idx) {
648 			ext_debug(inode, "  %d->%llu",
649 				  le32_to_cpu(path->p_idx->ei_block),
650 				  ext4_idx_pblock(path->p_idx));
651 		} else if (path->p_ext) {
652 			ext_debug(inode, "  %d:[%d]%d:%llu ",
653 				  le32_to_cpu(path->p_ext->ee_block),
654 				  ext4_ext_is_unwritten(path->p_ext),
655 				  ext4_ext_get_actual_len(path->p_ext),
656 				  ext4_ext_pblock(path->p_ext));
657 		} else
658 			ext_debug(inode, "  []");
659 	}
660 	ext_debug(inode, "\n");
661 }
662 
663 static void ext4_ext_show_leaf(struct inode *inode, struct ext4_ext_path *path)
664 {
665 	int depth = ext_depth(inode);
666 	struct ext4_extent_header *eh;
667 	struct ext4_extent *ex;
668 	int i;
669 
670 	if (!path)
671 		return;
672 
673 	eh = path[depth].p_hdr;
674 	ex = EXT_FIRST_EXTENT(eh);
675 
676 	ext_debug(inode, "Displaying leaf extents\n");
677 
678 	for (i = 0; i < le16_to_cpu(eh->eh_entries); i++, ex++) {
679 		ext_debug(inode, "%d:[%d]%d:%llu ", le32_to_cpu(ex->ee_block),
680 			  ext4_ext_is_unwritten(ex),
681 			  ext4_ext_get_actual_len(ex), ext4_ext_pblock(ex));
682 	}
683 	ext_debug(inode, "\n");
684 }
685 
686 static void ext4_ext_show_move(struct inode *inode, struct ext4_ext_path *path,
687 			ext4_fsblk_t newblock, int level)
688 {
689 	int depth = ext_depth(inode);
690 	struct ext4_extent *ex;
691 
692 	if (depth != level) {
693 		struct ext4_extent_idx *idx;
694 		idx = path[level].p_idx;
695 		while (idx <= EXT_MAX_INDEX(path[level].p_hdr)) {
696 			ext_debug(inode, "%d: move %d:%llu in new index %llu\n",
697 				  level, le32_to_cpu(idx->ei_block),
698 				  ext4_idx_pblock(idx), newblock);
699 			idx++;
700 		}
701 
702 		return;
703 	}
704 
705 	ex = path[depth].p_ext;
706 	while (ex <= EXT_MAX_EXTENT(path[depth].p_hdr)) {
707 		ext_debug(inode, "move %d:%llu:[%d]%d in new leaf %llu\n",
708 				le32_to_cpu(ex->ee_block),
709 				ext4_ext_pblock(ex),
710 				ext4_ext_is_unwritten(ex),
711 				ext4_ext_get_actual_len(ex),
712 				newblock);
713 		ex++;
714 	}
715 }
716 
717 #else
718 #define ext4_ext_show_path(inode, path)
719 #define ext4_ext_show_leaf(inode, path)
720 #define ext4_ext_show_move(inode, path, newblock, level)
721 #endif
722 
723 void ext4_ext_drop_refs(struct ext4_ext_path *path)
724 {
725 	int depth, i;
726 
727 	if (!path)
728 		return;
729 	depth = path->p_depth;
730 	for (i = 0; i <= depth; i++, path++) {
731 		brelse(path->p_bh);
732 		path->p_bh = NULL;
733 	}
734 }
735 
736 /*
737  * ext4_ext_binsearch_idx:
738  * binary search for the closest index of the given block
739  * the header must be checked before calling this
740  */
741 static void
742 ext4_ext_binsearch_idx(struct inode *inode,
743 			struct ext4_ext_path *path, ext4_lblk_t block)
744 {
745 	struct ext4_extent_header *eh = path->p_hdr;
746 	struct ext4_extent_idx *r, *l, *m;
747 
748 
749 	ext_debug(inode, "binsearch for %u(idx):  ", block);
750 
751 	l = EXT_FIRST_INDEX(eh) + 1;
752 	r = EXT_LAST_INDEX(eh);
753 	while (l <= r) {
754 		m = l + (r - l) / 2;
755 		ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
756 			  le32_to_cpu(l->ei_block), m, le32_to_cpu(m->ei_block),
757 			  r, le32_to_cpu(r->ei_block));
758 
759 		if (block < le32_to_cpu(m->ei_block))
760 			r = m - 1;
761 		else
762 			l = m + 1;
763 	}
764 
765 	path->p_idx = l - 1;
766 	ext_debug(inode, "  -> %u->%lld ", le32_to_cpu(path->p_idx->ei_block),
767 		  ext4_idx_pblock(path->p_idx));
768 
769 #ifdef CHECK_BINSEARCH
770 	{
771 		struct ext4_extent_idx *chix, *ix;
772 		int k;
773 
774 		chix = ix = EXT_FIRST_INDEX(eh);
775 		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ix++) {
776 			if (k != 0 && le32_to_cpu(ix->ei_block) <=
777 			    le32_to_cpu(ix[-1].ei_block)) {
778 				printk(KERN_DEBUG "k=%d, ix=0x%p, "
779 				       "first=0x%p\n", k,
780 				       ix, EXT_FIRST_INDEX(eh));
781 				printk(KERN_DEBUG "%u <= %u\n",
782 				       le32_to_cpu(ix->ei_block),
783 				       le32_to_cpu(ix[-1].ei_block));
784 			}
785 			BUG_ON(k && le32_to_cpu(ix->ei_block)
786 					   <= le32_to_cpu(ix[-1].ei_block));
787 			if (block < le32_to_cpu(ix->ei_block))
788 				break;
789 			chix = ix;
790 		}
791 		BUG_ON(chix != path->p_idx);
792 	}
793 #endif
794 
795 }
796 
797 /*
798  * ext4_ext_binsearch:
799  * binary search for closest extent of the given block
800  * the header must be checked before calling this
801  */
802 static void
803 ext4_ext_binsearch(struct inode *inode,
804 		struct ext4_ext_path *path, ext4_lblk_t block)
805 {
806 	struct ext4_extent_header *eh = path->p_hdr;
807 	struct ext4_extent *r, *l, *m;
808 
809 	if (eh->eh_entries == 0) {
810 		/*
811 		 * this leaf is empty:
812 		 * we get such a leaf in split/add case
813 		 */
814 		return;
815 	}
816 
817 	ext_debug(inode, "binsearch for %u:  ", block);
818 
819 	l = EXT_FIRST_EXTENT(eh) + 1;
820 	r = EXT_LAST_EXTENT(eh);
821 
822 	while (l <= r) {
823 		m = l + (r - l) / 2;
824 		ext_debug(inode, "%p(%u):%p(%u):%p(%u) ", l,
825 			  le32_to_cpu(l->ee_block), m, le32_to_cpu(m->ee_block),
826 			  r, le32_to_cpu(r->ee_block));
827 
828 		if (block < le32_to_cpu(m->ee_block))
829 			r = m - 1;
830 		else
831 			l = m + 1;
832 	}
833 
834 	path->p_ext = l - 1;
835 	ext_debug(inode, "  -> %d:%llu:[%d]%d ",
836 			le32_to_cpu(path->p_ext->ee_block),
837 			ext4_ext_pblock(path->p_ext),
838 			ext4_ext_is_unwritten(path->p_ext),
839 			ext4_ext_get_actual_len(path->p_ext));
840 
841 #ifdef CHECK_BINSEARCH
842 	{
843 		struct ext4_extent *chex, *ex;
844 		int k;
845 
846 		chex = ex = EXT_FIRST_EXTENT(eh);
847 		for (k = 0; k < le16_to_cpu(eh->eh_entries); k++, ex++) {
848 			BUG_ON(k && le32_to_cpu(ex->ee_block)
849 					  <= le32_to_cpu(ex[-1].ee_block));
850 			if (block < le32_to_cpu(ex->ee_block))
851 				break;
852 			chex = ex;
853 		}
854 		BUG_ON(chex != path->p_ext);
855 	}
856 #endif
857 
858 }
859 
860 void ext4_ext_tree_init(handle_t *handle, struct inode *inode)
861 {
862 	struct ext4_extent_header *eh;
863 
864 	eh = ext_inode_hdr(inode);
865 	eh->eh_depth = 0;
866 	eh->eh_entries = 0;
867 	eh->eh_magic = EXT4_EXT_MAGIC;
868 	eh->eh_max = cpu_to_le16(ext4_ext_space_root(inode, 0));
869 	eh->eh_generation = 0;
870 	ext4_mark_inode_dirty(handle, inode);
871 }
872 
873 struct ext4_ext_path *
874 ext4_find_extent(struct inode *inode, ext4_lblk_t block,
875 		 struct ext4_ext_path **orig_path, int flags)
876 {
877 	struct ext4_extent_header *eh;
878 	struct buffer_head *bh;
879 	struct ext4_ext_path *path = orig_path ? *orig_path : NULL;
880 	short int depth, i, ppos = 0;
881 	int ret;
882 	gfp_t gfp_flags = GFP_NOFS;
883 
884 	if (flags & EXT4_EX_NOFAIL)
885 		gfp_flags |= __GFP_NOFAIL;
886 
887 	eh = ext_inode_hdr(inode);
888 	depth = ext_depth(inode);
889 	if (depth < 0 || depth > EXT4_MAX_EXTENT_DEPTH) {
890 		EXT4_ERROR_INODE(inode, "inode has invalid extent depth: %d",
891 				 depth);
892 		ret = -EFSCORRUPTED;
893 		goto err;
894 	}
895 
896 	if (path) {
897 		ext4_ext_drop_refs(path);
898 		if (depth > path[0].p_maxdepth) {
899 			kfree(path);
900 			*orig_path = path = NULL;
901 		}
902 	}
903 	if (!path) {
904 		/* account possible depth increase */
905 		path = kcalloc(depth + 2, sizeof(struct ext4_ext_path),
906 				gfp_flags);
907 		if (unlikely(!path))
908 			return ERR_PTR(-ENOMEM);
909 		path[0].p_maxdepth = depth + 1;
910 	}
911 	path[0].p_hdr = eh;
912 	path[0].p_bh = NULL;
913 
914 	i = depth;
915 	if (!(flags & EXT4_EX_NOCACHE) && depth == 0)
916 		ext4_cache_extents(inode, eh);
917 	/* walk through the tree */
918 	while (i) {
919 		ext_debug(inode, "depth %d: num %d, max %d\n",
920 			  ppos, le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
921 
922 		ext4_ext_binsearch_idx(inode, path + ppos, block);
923 		path[ppos].p_block = ext4_idx_pblock(path[ppos].p_idx);
924 		path[ppos].p_depth = i;
925 		path[ppos].p_ext = NULL;
926 
927 		bh = read_extent_tree_block(inode, path[ppos].p_idx, --i, flags);
928 		if (IS_ERR(bh)) {
929 			ret = PTR_ERR(bh);
930 			goto err;
931 		}
932 
933 		eh = ext_block_hdr(bh);
934 		ppos++;
935 		path[ppos].p_bh = bh;
936 		path[ppos].p_hdr = eh;
937 	}
938 
939 	path[ppos].p_depth = i;
940 	path[ppos].p_ext = NULL;
941 	path[ppos].p_idx = NULL;
942 
943 	/* find extent */
944 	ext4_ext_binsearch(inode, path + ppos, block);
945 	/* if not an empty leaf */
946 	if (path[ppos].p_ext)
947 		path[ppos].p_block = ext4_ext_pblock(path[ppos].p_ext);
948 
949 	ext4_ext_show_path(inode, path);
950 
951 	return path;
952 
953 err:
954 	ext4_ext_drop_refs(path);
955 	kfree(path);
956 	if (orig_path)
957 		*orig_path = NULL;
958 	return ERR_PTR(ret);
959 }
960 
961 /*
962  * ext4_ext_insert_index:
963  * insert new index [@logical;@ptr] into the block at @curp;
964  * check where to insert: before @curp or after @curp
965  */
966 static int ext4_ext_insert_index(handle_t *handle, struct inode *inode,
967 				 struct ext4_ext_path *curp,
968 				 int logical, ext4_fsblk_t ptr)
969 {
970 	struct ext4_extent_idx *ix;
971 	int len, err;
972 
973 	err = ext4_ext_get_access(handle, inode, curp);
974 	if (err)
975 		return err;
976 
977 	if (unlikely(logical == le32_to_cpu(curp->p_idx->ei_block))) {
978 		EXT4_ERROR_INODE(inode,
979 				 "logical %d == ei_block %d!",
980 				 logical, le32_to_cpu(curp->p_idx->ei_block));
981 		return -EFSCORRUPTED;
982 	}
983 
984 	if (unlikely(le16_to_cpu(curp->p_hdr->eh_entries)
985 			     >= le16_to_cpu(curp->p_hdr->eh_max))) {
986 		EXT4_ERROR_INODE(inode,
987 				 "eh_entries %d >= eh_max %d!",
988 				 le16_to_cpu(curp->p_hdr->eh_entries),
989 				 le16_to_cpu(curp->p_hdr->eh_max));
990 		return -EFSCORRUPTED;
991 	}
992 
993 	if (logical > le32_to_cpu(curp->p_idx->ei_block)) {
994 		/* insert after */
995 		ext_debug(inode, "insert new index %d after: %llu\n",
996 			  logical, ptr);
997 		ix = curp->p_idx + 1;
998 	} else {
999 		/* insert before */
1000 		ext_debug(inode, "insert new index %d before: %llu\n",
1001 			  logical, ptr);
1002 		ix = curp->p_idx;
1003 	}
1004 
1005 	len = EXT_LAST_INDEX(curp->p_hdr) - ix + 1;
1006 	BUG_ON(len < 0);
1007 	if (len > 0) {
1008 		ext_debug(inode, "insert new index %d: "
1009 				"move %d indices from 0x%p to 0x%p\n",
1010 				logical, len, ix, ix + 1);
1011 		memmove(ix + 1, ix, len * sizeof(struct ext4_extent_idx));
1012 	}
1013 
1014 	if (unlikely(ix > EXT_MAX_INDEX(curp->p_hdr))) {
1015 		EXT4_ERROR_INODE(inode, "ix > EXT_MAX_INDEX!");
1016 		return -EFSCORRUPTED;
1017 	}
1018 
1019 	ix->ei_block = cpu_to_le32(logical);
1020 	ext4_idx_store_pblock(ix, ptr);
1021 	le16_add_cpu(&curp->p_hdr->eh_entries, 1);
1022 
1023 	if (unlikely(ix > EXT_LAST_INDEX(curp->p_hdr))) {
1024 		EXT4_ERROR_INODE(inode, "ix > EXT_LAST_INDEX!");
1025 		return -EFSCORRUPTED;
1026 	}
1027 
1028 	err = ext4_ext_dirty(handle, inode, curp);
1029 	ext4_std_error(inode->i_sb, err);
1030 
1031 	return err;
1032 }
1033 
1034 /*
1035  * ext4_ext_split:
1036  * inserts new subtree into the path, using free index entry
1037  * at depth @at:
1038  * - allocates all needed blocks (new leaf and all intermediate index blocks)
1039  * - makes decision where to split
1040  * - moves remaining extents and index entries (right to the split point)
1041  *   into the newly allocated blocks
1042  * - initializes subtree
1043  */
1044 static int ext4_ext_split(handle_t *handle, struct inode *inode,
1045 			  unsigned int flags,
1046 			  struct ext4_ext_path *path,
1047 			  struct ext4_extent *newext, int at)
1048 {
1049 	struct buffer_head *bh = NULL;
1050 	int depth = ext_depth(inode);
1051 	struct ext4_extent_header *neh;
1052 	struct ext4_extent_idx *fidx;
1053 	int i = at, k, m, a;
1054 	ext4_fsblk_t newblock, oldblock;
1055 	__le32 border;
1056 	ext4_fsblk_t *ablocks = NULL; /* array of allocated blocks */
1057 	gfp_t gfp_flags = GFP_NOFS;
1058 	int err = 0;
1059 	size_t ext_size = 0;
1060 
1061 	if (flags & EXT4_EX_NOFAIL)
1062 		gfp_flags |= __GFP_NOFAIL;
1063 
1064 	/* make decision: where to split? */
1065 	/* FIXME: now decision is simplest: at current extent */
1066 
1067 	/* if current leaf will be split, then we should use
1068 	 * border from split point */
1069 	if (unlikely(path[depth].p_ext > EXT_MAX_EXTENT(path[depth].p_hdr))) {
1070 		EXT4_ERROR_INODE(inode, "p_ext > EXT_MAX_EXTENT!");
1071 		return -EFSCORRUPTED;
1072 	}
1073 	if (path[depth].p_ext != EXT_MAX_EXTENT(path[depth].p_hdr)) {
1074 		border = path[depth].p_ext[1].ee_block;
1075 		ext_debug(inode, "leaf will be split."
1076 				" next leaf starts at %d\n",
1077 				  le32_to_cpu(border));
1078 	} else {
1079 		border = newext->ee_block;
1080 		ext_debug(inode, "leaf will be added."
1081 				" next leaf starts at %d\n",
1082 				le32_to_cpu(border));
1083 	}
1084 
1085 	/*
1086 	 * If error occurs, then we break processing
1087 	 * and mark filesystem read-only. index won't
1088 	 * be inserted and tree will be in consistent
1089 	 * state. Next mount will repair buffers too.
1090 	 */
1091 
1092 	/*
1093 	 * Get array to track all allocated blocks.
1094 	 * We need this to handle errors and free blocks
1095 	 * upon them.
1096 	 */
1097 	ablocks = kcalloc(depth, sizeof(ext4_fsblk_t), gfp_flags);
1098 	if (!ablocks)
1099 		return -ENOMEM;
1100 
1101 	/* allocate all needed blocks */
1102 	ext_debug(inode, "allocate %d blocks for indexes/leaf\n", depth - at);
1103 	for (a = 0; a < depth - at; a++) {
1104 		newblock = ext4_ext_new_meta_block(handle, inode, path,
1105 						   newext, &err, flags);
1106 		if (newblock == 0)
1107 			goto cleanup;
1108 		ablocks[a] = newblock;
1109 	}
1110 
1111 	/* initialize new leaf */
1112 	newblock = ablocks[--a];
1113 	if (unlikely(newblock == 0)) {
1114 		EXT4_ERROR_INODE(inode, "newblock == 0!");
1115 		err = -EFSCORRUPTED;
1116 		goto cleanup;
1117 	}
1118 	bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1119 	if (unlikely(!bh)) {
1120 		err = -ENOMEM;
1121 		goto cleanup;
1122 	}
1123 	lock_buffer(bh);
1124 
1125 	err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1126 					     EXT4_JTR_NONE);
1127 	if (err)
1128 		goto cleanup;
1129 
1130 	neh = ext_block_hdr(bh);
1131 	neh->eh_entries = 0;
1132 	neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1133 	neh->eh_magic = EXT4_EXT_MAGIC;
1134 	neh->eh_depth = 0;
1135 	neh->eh_generation = 0;
1136 
1137 	/* move remainder of path[depth] to the new leaf */
1138 	if (unlikely(path[depth].p_hdr->eh_entries !=
1139 		     path[depth].p_hdr->eh_max)) {
1140 		EXT4_ERROR_INODE(inode, "eh_entries %d != eh_max %d!",
1141 				 path[depth].p_hdr->eh_entries,
1142 				 path[depth].p_hdr->eh_max);
1143 		err = -EFSCORRUPTED;
1144 		goto cleanup;
1145 	}
1146 	/* start copy from next extent */
1147 	m = EXT_MAX_EXTENT(path[depth].p_hdr) - path[depth].p_ext++;
1148 	ext4_ext_show_move(inode, path, newblock, depth);
1149 	if (m) {
1150 		struct ext4_extent *ex;
1151 		ex = EXT_FIRST_EXTENT(neh);
1152 		memmove(ex, path[depth].p_ext, sizeof(struct ext4_extent) * m);
1153 		le16_add_cpu(&neh->eh_entries, m);
1154 	}
1155 
1156 	/* zero out unused area in the extent block */
1157 	ext_size = sizeof(struct ext4_extent_header) +
1158 		sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries);
1159 	memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1160 	ext4_extent_block_csum_set(inode, neh);
1161 	set_buffer_uptodate(bh);
1162 	unlock_buffer(bh);
1163 
1164 	err = ext4_handle_dirty_metadata(handle, inode, bh);
1165 	if (err)
1166 		goto cleanup;
1167 	brelse(bh);
1168 	bh = NULL;
1169 
1170 	/* correct old leaf */
1171 	if (m) {
1172 		err = ext4_ext_get_access(handle, inode, path + depth);
1173 		if (err)
1174 			goto cleanup;
1175 		le16_add_cpu(&path[depth].p_hdr->eh_entries, -m);
1176 		err = ext4_ext_dirty(handle, inode, path + depth);
1177 		if (err)
1178 			goto cleanup;
1179 
1180 	}
1181 
1182 	/* create intermediate indexes */
1183 	k = depth - at - 1;
1184 	if (unlikely(k < 0)) {
1185 		EXT4_ERROR_INODE(inode, "k %d < 0!", k);
1186 		err = -EFSCORRUPTED;
1187 		goto cleanup;
1188 	}
1189 	if (k)
1190 		ext_debug(inode, "create %d intermediate indices\n", k);
1191 	/* insert new index into current index block */
1192 	/* current depth stored in i var */
1193 	i = depth - 1;
1194 	while (k--) {
1195 		oldblock = newblock;
1196 		newblock = ablocks[--a];
1197 		bh = sb_getblk(inode->i_sb, newblock);
1198 		if (unlikely(!bh)) {
1199 			err = -ENOMEM;
1200 			goto cleanup;
1201 		}
1202 		lock_buffer(bh);
1203 
1204 		err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1205 						     EXT4_JTR_NONE);
1206 		if (err)
1207 			goto cleanup;
1208 
1209 		neh = ext_block_hdr(bh);
1210 		neh->eh_entries = cpu_to_le16(1);
1211 		neh->eh_magic = EXT4_EXT_MAGIC;
1212 		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1213 		neh->eh_depth = cpu_to_le16(depth - i);
1214 		neh->eh_generation = 0;
1215 		fidx = EXT_FIRST_INDEX(neh);
1216 		fidx->ei_block = border;
1217 		ext4_idx_store_pblock(fidx, oldblock);
1218 
1219 		ext_debug(inode, "int.index at %d (block %llu): %u -> %llu\n",
1220 				i, newblock, le32_to_cpu(border), oldblock);
1221 
1222 		/* move remainder of path[i] to the new index block */
1223 		if (unlikely(EXT_MAX_INDEX(path[i].p_hdr) !=
1224 					EXT_LAST_INDEX(path[i].p_hdr))) {
1225 			EXT4_ERROR_INODE(inode,
1226 					 "EXT_MAX_INDEX != EXT_LAST_INDEX ee_block %d!",
1227 					 le32_to_cpu(path[i].p_ext->ee_block));
1228 			err = -EFSCORRUPTED;
1229 			goto cleanup;
1230 		}
1231 		/* start copy indexes */
1232 		m = EXT_MAX_INDEX(path[i].p_hdr) - path[i].p_idx++;
1233 		ext_debug(inode, "cur 0x%p, last 0x%p\n", path[i].p_idx,
1234 				EXT_MAX_INDEX(path[i].p_hdr));
1235 		ext4_ext_show_move(inode, path, newblock, i);
1236 		if (m) {
1237 			memmove(++fidx, path[i].p_idx,
1238 				sizeof(struct ext4_extent_idx) * m);
1239 			le16_add_cpu(&neh->eh_entries, m);
1240 		}
1241 		/* zero out unused area in the extent block */
1242 		ext_size = sizeof(struct ext4_extent_header) +
1243 		   (sizeof(struct ext4_extent) * le16_to_cpu(neh->eh_entries));
1244 		memset(bh->b_data + ext_size, 0,
1245 			inode->i_sb->s_blocksize - ext_size);
1246 		ext4_extent_block_csum_set(inode, neh);
1247 		set_buffer_uptodate(bh);
1248 		unlock_buffer(bh);
1249 
1250 		err = ext4_handle_dirty_metadata(handle, inode, bh);
1251 		if (err)
1252 			goto cleanup;
1253 		brelse(bh);
1254 		bh = NULL;
1255 
1256 		/* correct old index */
1257 		if (m) {
1258 			err = ext4_ext_get_access(handle, inode, path + i);
1259 			if (err)
1260 				goto cleanup;
1261 			le16_add_cpu(&path[i].p_hdr->eh_entries, -m);
1262 			err = ext4_ext_dirty(handle, inode, path + i);
1263 			if (err)
1264 				goto cleanup;
1265 		}
1266 
1267 		i--;
1268 	}
1269 
1270 	/* insert new index */
1271 	err = ext4_ext_insert_index(handle, inode, path + at,
1272 				    le32_to_cpu(border), newblock);
1273 
1274 cleanup:
1275 	if (bh) {
1276 		if (buffer_locked(bh))
1277 			unlock_buffer(bh);
1278 		brelse(bh);
1279 	}
1280 
1281 	if (err) {
1282 		/* free all allocated blocks in error case */
1283 		for (i = 0; i < depth; i++) {
1284 			if (!ablocks[i])
1285 				continue;
1286 			ext4_free_blocks(handle, inode, NULL, ablocks[i], 1,
1287 					 EXT4_FREE_BLOCKS_METADATA);
1288 		}
1289 	}
1290 	kfree(ablocks);
1291 
1292 	return err;
1293 }
1294 
1295 /*
1296  * ext4_ext_grow_indepth:
1297  * implements tree growing procedure:
1298  * - allocates new block
1299  * - moves top-level data (index block or leaf) into the new block
1300  * - initializes new top-level, creating index that points to the
1301  *   just created block
1302  */
1303 static int ext4_ext_grow_indepth(handle_t *handle, struct inode *inode,
1304 				 unsigned int flags)
1305 {
1306 	struct ext4_extent_header *neh;
1307 	struct buffer_head *bh;
1308 	ext4_fsblk_t newblock, goal = 0;
1309 	struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
1310 	int err = 0;
1311 	size_t ext_size = 0;
1312 
1313 	/* Try to prepend new index to old one */
1314 	if (ext_depth(inode))
1315 		goal = ext4_idx_pblock(EXT_FIRST_INDEX(ext_inode_hdr(inode)));
1316 	if (goal > le32_to_cpu(es->s_first_data_block)) {
1317 		flags |= EXT4_MB_HINT_TRY_GOAL;
1318 		goal--;
1319 	} else
1320 		goal = ext4_inode_to_goal_block(inode);
1321 	newblock = ext4_new_meta_blocks(handle, inode, goal, flags,
1322 					NULL, &err);
1323 	if (newblock == 0)
1324 		return err;
1325 
1326 	bh = sb_getblk_gfp(inode->i_sb, newblock, __GFP_MOVABLE | GFP_NOFS);
1327 	if (unlikely(!bh))
1328 		return -ENOMEM;
1329 	lock_buffer(bh);
1330 
1331 	err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
1332 					     EXT4_JTR_NONE);
1333 	if (err) {
1334 		unlock_buffer(bh);
1335 		goto out;
1336 	}
1337 
1338 	ext_size = sizeof(EXT4_I(inode)->i_data);
1339 	/* move top-level index/leaf into new block */
1340 	memmove(bh->b_data, EXT4_I(inode)->i_data, ext_size);
1341 	/* zero out unused area in the extent block */
1342 	memset(bh->b_data + ext_size, 0, inode->i_sb->s_blocksize - ext_size);
1343 
1344 	/* set size of new block */
1345 	neh = ext_block_hdr(bh);
1346 	/* old root could have indexes or leaves
1347 	 * so calculate e_max right way */
1348 	if (ext_depth(inode))
1349 		neh->eh_max = cpu_to_le16(ext4_ext_space_block_idx(inode, 0));
1350 	else
1351 		neh->eh_max = cpu_to_le16(ext4_ext_space_block(inode, 0));
1352 	neh->eh_magic = EXT4_EXT_MAGIC;
1353 	ext4_extent_block_csum_set(inode, neh);
1354 	set_buffer_uptodate(bh);
1355 	set_buffer_verified(bh);
1356 	unlock_buffer(bh);
1357 
1358 	err = ext4_handle_dirty_metadata(handle, inode, bh);
1359 	if (err)
1360 		goto out;
1361 
1362 	/* Update top-level index: num,max,pointer */
1363 	neh = ext_inode_hdr(inode);
1364 	neh->eh_entries = cpu_to_le16(1);
1365 	ext4_idx_store_pblock(EXT_FIRST_INDEX(neh), newblock);
1366 	if (neh->eh_depth == 0) {
1367 		/* Root extent block becomes index block */
1368 		neh->eh_max = cpu_to_le16(ext4_ext_space_root_idx(inode, 0));
1369 		EXT_FIRST_INDEX(neh)->ei_block =
1370 			EXT_FIRST_EXTENT(neh)->ee_block;
1371 	}
1372 	ext_debug(inode, "new root: num %d(%d), lblock %d, ptr %llu\n",
1373 		  le16_to_cpu(neh->eh_entries), le16_to_cpu(neh->eh_max),
1374 		  le32_to_cpu(EXT_FIRST_INDEX(neh)->ei_block),
1375 		  ext4_idx_pblock(EXT_FIRST_INDEX(neh)));
1376 
1377 	le16_add_cpu(&neh->eh_depth, 1);
1378 	err = ext4_mark_inode_dirty(handle, inode);
1379 out:
1380 	brelse(bh);
1381 
1382 	return err;
1383 }
1384 
1385 /*
1386  * ext4_ext_create_new_leaf:
1387  * finds empty index and adds new leaf.
1388  * if no free index is found, then it requests in-depth growing.
1389  */
1390 static int ext4_ext_create_new_leaf(handle_t *handle, struct inode *inode,
1391 				    unsigned int mb_flags,
1392 				    unsigned int gb_flags,
1393 				    struct ext4_ext_path **ppath,
1394 				    struct ext4_extent *newext)
1395 {
1396 	struct ext4_ext_path *path = *ppath;
1397 	struct ext4_ext_path *curp;
1398 	int depth, i, err = 0;
1399 
1400 repeat:
1401 	i = depth = ext_depth(inode);
1402 
1403 	/* walk up to the tree and look for free index entry */
1404 	curp = path + depth;
1405 	while (i > 0 && !EXT_HAS_FREE_INDEX(curp)) {
1406 		i--;
1407 		curp--;
1408 	}
1409 
1410 	/* we use already allocated block for index block,
1411 	 * so subsequent data blocks should be contiguous */
1412 	if (EXT_HAS_FREE_INDEX(curp)) {
1413 		/* if we found index with free entry, then use that
1414 		 * entry: create all needed subtree and add new leaf */
1415 		err = ext4_ext_split(handle, inode, mb_flags, path, newext, i);
1416 		if (err)
1417 			goto out;
1418 
1419 		/* refill path */
1420 		path = ext4_find_extent(inode,
1421 				    (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1422 				    ppath, gb_flags);
1423 		if (IS_ERR(path))
1424 			err = PTR_ERR(path);
1425 	} else {
1426 		/* tree is full, time to grow in depth */
1427 		err = ext4_ext_grow_indepth(handle, inode, mb_flags);
1428 		if (err)
1429 			goto out;
1430 
1431 		/* refill path */
1432 		path = ext4_find_extent(inode,
1433 				   (ext4_lblk_t)le32_to_cpu(newext->ee_block),
1434 				    ppath, gb_flags);
1435 		if (IS_ERR(path)) {
1436 			err = PTR_ERR(path);
1437 			goto out;
1438 		}
1439 
1440 		/*
1441 		 * only first (depth 0 -> 1) produces free space;
1442 		 * in all other cases we have to split the grown tree
1443 		 */
1444 		depth = ext_depth(inode);
1445 		if (path[depth].p_hdr->eh_entries == path[depth].p_hdr->eh_max) {
1446 			/* now we need to split */
1447 			goto repeat;
1448 		}
1449 	}
1450 
1451 out:
1452 	return err;
1453 }
1454 
1455 /*
1456  * search the closest allocated block to the left for *logical
1457  * and returns it at @logical + it's physical address at @phys
1458  * if *logical is the smallest allocated block, the function
1459  * returns 0 at @phys
1460  * return value contains 0 (success) or error code
1461  */
1462 static int ext4_ext_search_left(struct inode *inode,
1463 				struct ext4_ext_path *path,
1464 				ext4_lblk_t *logical, ext4_fsblk_t *phys)
1465 {
1466 	struct ext4_extent_idx *ix;
1467 	struct ext4_extent *ex;
1468 	int depth, ee_len;
1469 
1470 	if (unlikely(path == NULL)) {
1471 		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1472 		return -EFSCORRUPTED;
1473 	}
1474 	depth = path->p_depth;
1475 	*phys = 0;
1476 
1477 	if (depth == 0 && path->p_ext == NULL)
1478 		return 0;
1479 
1480 	/* usually extent in the path covers blocks smaller
1481 	 * then *logical, but it can be that extent is the
1482 	 * first one in the file */
1483 
1484 	ex = path[depth].p_ext;
1485 	ee_len = ext4_ext_get_actual_len(ex);
1486 	if (*logical < le32_to_cpu(ex->ee_block)) {
1487 		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1488 			EXT4_ERROR_INODE(inode,
1489 					 "EXT_FIRST_EXTENT != ex *logical %d ee_block %d!",
1490 					 *logical, le32_to_cpu(ex->ee_block));
1491 			return -EFSCORRUPTED;
1492 		}
1493 		while (--depth >= 0) {
1494 			ix = path[depth].p_idx;
1495 			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1496 				EXT4_ERROR_INODE(inode,
1497 				  "ix (%d) != EXT_FIRST_INDEX (%d) (depth %d)!",
1498 				  ix != NULL ? le32_to_cpu(ix->ei_block) : 0,
1499 				  EXT_FIRST_INDEX(path[depth].p_hdr) != NULL ?
1500 		le32_to_cpu(EXT_FIRST_INDEX(path[depth].p_hdr)->ei_block) : 0,
1501 				  depth);
1502 				return -EFSCORRUPTED;
1503 			}
1504 		}
1505 		return 0;
1506 	}
1507 
1508 	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1509 		EXT4_ERROR_INODE(inode,
1510 				 "logical %d < ee_block %d + ee_len %d!",
1511 				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1512 		return -EFSCORRUPTED;
1513 	}
1514 
1515 	*logical = le32_to_cpu(ex->ee_block) + ee_len - 1;
1516 	*phys = ext4_ext_pblock(ex) + ee_len - 1;
1517 	return 0;
1518 }
1519 
1520 /*
1521  * Search the closest allocated block to the right for *logical
1522  * and returns it at @logical + it's physical address at @phys.
1523  * If not exists, return 0 and @phys is set to 0. We will return
1524  * 1 which means we found an allocated block and ret_ex is valid.
1525  * Or return a (< 0) error code.
1526  */
1527 static int ext4_ext_search_right(struct inode *inode,
1528 				 struct ext4_ext_path *path,
1529 				 ext4_lblk_t *logical, ext4_fsblk_t *phys,
1530 				 struct ext4_extent *ret_ex)
1531 {
1532 	struct buffer_head *bh = NULL;
1533 	struct ext4_extent_header *eh;
1534 	struct ext4_extent_idx *ix;
1535 	struct ext4_extent *ex;
1536 	int depth;	/* Note, NOT eh_depth; depth from top of tree */
1537 	int ee_len;
1538 
1539 	if (unlikely(path == NULL)) {
1540 		EXT4_ERROR_INODE(inode, "path == NULL *logical %d!", *logical);
1541 		return -EFSCORRUPTED;
1542 	}
1543 	depth = path->p_depth;
1544 	*phys = 0;
1545 
1546 	if (depth == 0 && path->p_ext == NULL)
1547 		return 0;
1548 
1549 	/* usually extent in the path covers blocks smaller
1550 	 * then *logical, but it can be that extent is the
1551 	 * first one in the file */
1552 
1553 	ex = path[depth].p_ext;
1554 	ee_len = ext4_ext_get_actual_len(ex);
1555 	if (*logical < le32_to_cpu(ex->ee_block)) {
1556 		if (unlikely(EXT_FIRST_EXTENT(path[depth].p_hdr) != ex)) {
1557 			EXT4_ERROR_INODE(inode,
1558 					 "first_extent(path[%d].p_hdr) != ex",
1559 					 depth);
1560 			return -EFSCORRUPTED;
1561 		}
1562 		while (--depth >= 0) {
1563 			ix = path[depth].p_idx;
1564 			if (unlikely(ix != EXT_FIRST_INDEX(path[depth].p_hdr))) {
1565 				EXT4_ERROR_INODE(inode,
1566 						 "ix != EXT_FIRST_INDEX *logical %d!",
1567 						 *logical);
1568 				return -EFSCORRUPTED;
1569 			}
1570 		}
1571 		goto found_extent;
1572 	}
1573 
1574 	if (unlikely(*logical < (le32_to_cpu(ex->ee_block) + ee_len))) {
1575 		EXT4_ERROR_INODE(inode,
1576 				 "logical %d < ee_block %d + ee_len %d!",
1577 				 *logical, le32_to_cpu(ex->ee_block), ee_len);
1578 		return -EFSCORRUPTED;
1579 	}
1580 
1581 	if (ex != EXT_LAST_EXTENT(path[depth].p_hdr)) {
1582 		/* next allocated block in this leaf */
1583 		ex++;
1584 		goto found_extent;
1585 	}
1586 
1587 	/* go up and search for index to the right */
1588 	while (--depth >= 0) {
1589 		ix = path[depth].p_idx;
1590 		if (ix != EXT_LAST_INDEX(path[depth].p_hdr))
1591 			goto got_index;
1592 	}
1593 
1594 	/* we've gone up to the root and found no index to the right */
1595 	return 0;
1596 
1597 got_index:
1598 	/* we've found index to the right, let's
1599 	 * follow it and find the closest allocated
1600 	 * block to the right */
1601 	ix++;
1602 	while (++depth < path->p_depth) {
1603 		/* subtract from p_depth to get proper eh_depth */
1604 		bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1605 		if (IS_ERR(bh))
1606 			return PTR_ERR(bh);
1607 		eh = ext_block_hdr(bh);
1608 		ix = EXT_FIRST_INDEX(eh);
1609 		put_bh(bh);
1610 	}
1611 
1612 	bh = read_extent_tree_block(inode, ix, path->p_depth - depth, 0);
1613 	if (IS_ERR(bh))
1614 		return PTR_ERR(bh);
1615 	eh = ext_block_hdr(bh);
1616 	ex = EXT_FIRST_EXTENT(eh);
1617 found_extent:
1618 	*logical = le32_to_cpu(ex->ee_block);
1619 	*phys = ext4_ext_pblock(ex);
1620 	if (ret_ex)
1621 		*ret_ex = *ex;
1622 	if (bh)
1623 		put_bh(bh);
1624 	return 1;
1625 }
1626 
1627 /*
1628  * ext4_ext_next_allocated_block:
1629  * returns allocated block in subsequent extent or EXT_MAX_BLOCKS.
1630  * NOTE: it considers block number from index entry as
1631  * allocated block. Thus, index entries have to be consistent
1632  * with leaves.
1633  */
1634 ext4_lblk_t
1635 ext4_ext_next_allocated_block(struct ext4_ext_path *path)
1636 {
1637 	int depth;
1638 
1639 	BUG_ON(path == NULL);
1640 	depth = path->p_depth;
1641 
1642 	if (depth == 0 && path->p_ext == NULL)
1643 		return EXT_MAX_BLOCKS;
1644 
1645 	while (depth >= 0) {
1646 		struct ext4_ext_path *p = &path[depth];
1647 
1648 		if (depth == path->p_depth) {
1649 			/* leaf */
1650 			if (p->p_ext && p->p_ext != EXT_LAST_EXTENT(p->p_hdr))
1651 				return le32_to_cpu(p->p_ext[1].ee_block);
1652 		} else {
1653 			/* index */
1654 			if (p->p_idx != EXT_LAST_INDEX(p->p_hdr))
1655 				return le32_to_cpu(p->p_idx[1].ei_block);
1656 		}
1657 		depth--;
1658 	}
1659 
1660 	return EXT_MAX_BLOCKS;
1661 }
1662 
1663 /*
1664  * ext4_ext_next_leaf_block:
1665  * returns first allocated block from next leaf or EXT_MAX_BLOCKS
1666  */
1667 static ext4_lblk_t ext4_ext_next_leaf_block(struct ext4_ext_path *path)
1668 {
1669 	int depth;
1670 
1671 	BUG_ON(path == NULL);
1672 	depth = path->p_depth;
1673 
1674 	/* zero-tree has no leaf blocks at all */
1675 	if (depth == 0)
1676 		return EXT_MAX_BLOCKS;
1677 
1678 	/* go to index block */
1679 	depth--;
1680 
1681 	while (depth >= 0) {
1682 		if (path[depth].p_idx !=
1683 				EXT_LAST_INDEX(path[depth].p_hdr))
1684 			return (ext4_lblk_t)
1685 				le32_to_cpu(path[depth].p_idx[1].ei_block);
1686 		depth--;
1687 	}
1688 
1689 	return EXT_MAX_BLOCKS;
1690 }
1691 
1692 /*
1693  * ext4_ext_correct_indexes:
1694  * if leaf gets modified and modified extent is first in the leaf,
1695  * then we have to correct all indexes above.
1696  * TODO: do we need to correct tree in all cases?
1697  */
1698 static int ext4_ext_correct_indexes(handle_t *handle, struct inode *inode,
1699 				struct ext4_ext_path *path)
1700 {
1701 	struct ext4_extent_header *eh;
1702 	int depth = ext_depth(inode);
1703 	struct ext4_extent *ex;
1704 	__le32 border;
1705 	int k, err = 0;
1706 
1707 	eh = path[depth].p_hdr;
1708 	ex = path[depth].p_ext;
1709 
1710 	if (unlikely(ex == NULL || eh == NULL)) {
1711 		EXT4_ERROR_INODE(inode,
1712 				 "ex %p == NULL or eh %p == NULL", ex, eh);
1713 		return -EFSCORRUPTED;
1714 	}
1715 
1716 	if (depth == 0) {
1717 		/* there is no tree at all */
1718 		return 0;
1719 	}
1720 
1721 	if (ex != EXT_FIRST_EXTENT(eh)) {
1722 		/* we correct tree if first leaf got modified only */
1723 		return 0;
1724 	}
1725 
1726 	/*
1727 	 * TODO: we need correction if border is smaller than current one
1728 	 */
1729 	k = depth - 1;
1730 	border = path[depth].p_ext->ee_block;
1731 	err = ext4_ext_get_access(handle, inode, path + k);
1732 	if (err)
1733 		return err;
1734 	path[k].p_idx->ei_block = border;
1735 	err = ext4_ext_dirty(handle, inode, path + k);
1736 	if (err)
1737 		return err;
1738 
1739 	while (k--) {
1740 		/* change all left-side indexes */
1741 		if (path[k+1].p_idx != EXT_FIRST_INDEX(path[k+1].p_hdr))
1742 			break;
1743 		err = ext4_ext_get_access(handle, inode, path + k);
1744 		if (err)
1745 			break;
1746 		path[k].p_idx->ei_block = border;
1747 		err = ext4_ext_dirty(handle, inode, path + k);
1748 		if (err)
1749 			break;
1750 	}
1751 
1752 	return err;
1753 }
1754 
1755 static int ext4_can_extents_be_merged(struct inode *inode,
1756 				      struct ext4_extent *ex1,
1757 				      struct ext4_extent *ex2)
1758 {
1759 	unsigned short ext1_ee_len, ext2_ee_len;
1760 
1761 	if (ext4_ext_is_unwritten(ex1) != ext4_ext_is_unwritten(ex2))
1762 		return 0;
1763 
1764 	ext1_ee_len = ext4_ext_get_actual_len(ex1);
1765 	ext2_ee_len = ext4_ext_get_actual_len(ex2);
1766 
1767 	if (le32_to_cpu(ex1->ee_block) + ext1_ee_len !=
1768 			le32_to_cpu(ex2->ee_block))
1769 		return 0;
1770 
1771 	if (ext1_ee_len + ext2_ee_len > EXT_INIT_MAX_LEN)
1772 		return 0;
1773 
1774 	if (ext4_ext_is_unwritten(ex1) &&
1775 	    ext1_ee_len + ext2_ee_len > EXT_UNWRITTEN_MAX_LEN)
1776 		return 0;
1777 #ifdef AGGRESSIVE_TEST
1778 	if (ext1_ee_len >= 4)
1779 		return 0;
1780 #endif
1781 
1782 	if (ext4_ext_pblock(ex1) + ext1_ee_len == ext4_ext_pblock(ex2))
1783 		return 1;
1784 	return 0;
1785 }
1786 
1787 /*
1788  * This function tries to merge the "ex" extent to the next extent in the tree.
1789  * It always tries to merge towards right. If you want to merge towards
1790  * left, pass "ex - 1" as argument instead of "ex".
1791  * Returns 0 if the extents (ex and ex+1) were _not_ merged and returns
1792  * 1 if they got merged.
1793  */
1794 static int ext4_ext_try_to_merge_right(struct inode *inode,
1795 				 struct ext4_ext_path *path,
1796 				 struct ext4_extent *ex)
1797 {
1798 	struct ext4_extent_header *eh;
1799 	unsigned int depth, len;
1800 	int merge_done = 0, unwritten;
1801 
1802 	depth = ext_depth(inode);
1803 	BUG_ON(path[depth].p_hdr == NULL);
1804 	eh = path[depth].p_hdr;
1805 
1806 	while (ex < EXT_LAST_EXTENT(eh)) {
1807 		if (!ext4_can_extents_be_merged(inode, ex, ex + 1))
1808 			break;
1809 		/* merge with next extent! */
1810 		unwritten = ext4_ext_is_unwritten(ex);
1811 		ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
1812 				+ ext4_ext_get_actual_len(ex + 1));
1813 		if (unwritten)
1814 			ext4_ext_mark_unwritten(ex);
1815 
1816 		if (ex + 1 < EXT_LAST_EXTENT(eh)) {
1817 			len = (EXT_LAST_EXTENT(eh) - ex - 1)
1818 				* sizeof(struct ext4_extent);
1819 			memmove(ex + 1, ex + 2, len);
1820 		}
1821 		le16_add_cpu(&eh->eh_entries, -1);
1822 		merge_done = 1;
1823 		WARN_ON(eh->eh_entries == 0);
1824 		if (!eh->eh_entries)
1825 			EXT4_ERROR_INODE(inode, "eh->eh_entries = 0!");
1826 	}
1827 
1828 	return merge_done;
1829 }
1830 
1831 /*
1832  * This function does a very simple check to see if we can collapse
1833  * an extent tree with a single extent tree leaf block into the inode.
1834  */
1835 static void ext4_ext_try_to_merge_up(handle_t *handle,
1836 				     struct inode *inode,
1837 				     struct ext4_ext_path *path)
1838 {
1839 	size_t s;
1840 	unsigned max_root = ext4_ext_space_root(inode, 0);
1841 	ext4_fsblk_t blk;
1842 
1843 	if ((path[0].p_depth != 1) ||
1844 	    (le16_to_cpu(path[0].p_hdr->eh_entries) != 1) ||
1845 	    (le16_to_cpu(path[1].p_hdr->eh_entries) > max_root))
1846 		return;
1847 
1848 	/*
1849 	 * We need to modify the block allocation bitmap and the block
1850 	 * group descriptor to release the extent tree block.  If we
1851 	 * can't get the journal credits, give up.
1852 	 */
1853 	if (ext4_journal_extend(handle, 2,
1854 			ext4_free_metadata_revoke_credits(inode->i_sb, 1)))
1855 		return;
1856 
1857 	/*
1858 	 * Copy the extent data up to the inode
1859 	 */
1860 	blk = ext4_idx_pblock(path[0].p_idx);
1861 	s = le16_to_cpu(path[1].p_hdr->eh_entries) *
1862 		sizeof(struct ext4_extent_idx);
1863 	s += sizeof(struct ext4_extent_header);
1864 
1865 	path[1].p_maxdepth = path[0].p_maxdepth;
1866 	memcpy(path[0].p_hdr, path[1].p_hdr, s);
1867 	path[0].p_depth = 0;
1868 	path[0].p_ext = EXT_FIRST_EXTENT(path[0].p_hdr) +
1869 		(path[1].p_ext - EXT_FIRST_EXTENT(path[1].p_hdr));
1870 	path[0].p_hdr->eh_max = cpu_to_le16(max_root);
1871 
1872 	brelse(path[1].p_bh);
1873 	ext4_free_blocks(handle, inode, NULL, blk, 1,
1874 			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
1875 }
1876 
1877 /*
1878  * This function tries to merge the @ex extent to neighbours in the tree, then
1879  * tries to collapse the extent tree into the inode.
1880  */
1881 static void ext4_ext_try_to_merge(handle_t *handle,
1882 				  struct inode *inode,
1883 				  struct ext4_ext_path *path,
1884 				  struct ext4_extent *ex)
1885 {
1886 	struct ext4_extent_header *eh;
1887 	unsigned int depth;
1888 	int merge_done = 0;
1889 
1890 	depth = ext_depth(inode);
1891 	BUG_ON(path[depth].p_hdr == NULL);
1892 	eh = path[depth].p_hdr;
1893 
1894 	if (ex > EXT_FIRST_EXTENT(eh))
1895 		merge_done = ext4_ext_try_to_merge_right(inode, path, ex - 1);
1896 
1897 	if (!merge_done)
1898 		(void) ext4_ext_try_to_merge_right(inode, path, ex);
1899 
1900 	ext4_ext_try_to_merge_up(handle, inode, path);
1901 }
1902 
1903 /*
1904  * check if a portion of the "newext" extent overlaps with an
1905  * existing extent.
1906  *
1907  * If there is an overlap discovered, it updates the length of the newext
1908  * such that there will be no overlap, and then returns 1.
1909  * If there is no overlap found, it returns 0.
1910  */
1911 static unsigned int ext4_ext_check_overlap(struct ext4_sb_info *sbi,
1912 					   struct inode *inode,
1913 					   struct ext4_extent *newext,
1914 					   struct ext4_ext_path *path)
1915 {
1916 	ext4_lblk_t b1, b2;
1917 	unsigned int depth, len1;
1918 	unsigned int ret = 0;
1919 
1920 	b1 = le32_to_cpu(newext->ee_block);
1921 	len1 = ext4_ext_get_actual_len(newext);
1922 	depth = ext_depth(inode);
1923 	if (!path[depth].p_ext)
1924 		goto out;
1925 	b2 = EXT4_LBLK_CMASK(sbi, le32_to_cpu(path[depth].p_ext->ee_block));
1926 
1927 	/*
1928 	 * get the next allocated block if the extent in the path
1929 	 * is before the requested block(s)
1930 	 */
1931 	if (b2 < b1) {
1932 		b2 = ext4_ext_next_allocated_block(path);
1933 		if (b2 == EXT_MAX_BLOCKS)
1934 			goto out;
1935 		b2 = EXT4_LBLK_CMASK(sbi, b2);
1936 	}
1937 
1938 	/* check for wrap through zero on extent logical start block*/
1939 	if (b1 + len1 < b1) {
1940 		len1 = EXT_MAX_BLOCKS - b1;
1941 		newext->ee_len = cpu_to_le16(len1);
1942 		ret = 1;
1943 	}
1944 
1945 	/* check for overlap */
1946 	if (b1 + len1 > b2) {
1947 		newext->ee_len = cpu_to_le16(b2 - b1);
1948 		ret = 1;
1949 	}
1950 out:
1951 	return ret;
1952 }
1953 
1954 /*
1955  * ext4_ext_insert_extent:
1956  * tries to merge requested extent into the existing extent or
1957  * inserts requested extent as new one into the tree,
1958  * creating new leaf in the no-space case.
1959  */
1960 int ext4_ext_insert_extent(handle_t *handle, struct inode *inode,
1961 				struct ext4_ext_path **ppath,
1962 				struct ext4_extent *newext, int gb_flags)
1963 {
1964 	struct ext4_ext_path *path = *ppath;
1965 	struct ext4_extent_header *eh;
1966 	struct ext4_extent *ex, *fex;
1967 	struct ext4_extent *nearex; /* nearest extent */
1968 	struct ext4_ext_path *npath = NULL;
1969 	int depth, len, err;
1970 	ext4_lblk_t next;
1971 	int mb_flags = 0, unwritten;
1972 
1973 	if (gb_flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
1974 		mb_flags |= EXT4_MB_DELALLOC_RESERVED;
1975 	if (unlikely(ext4_ext_get_actual_len(newext) == 0)) {
1976 		EXT4_ERROR_INODE(inode, "ext4_ext_get_actual_len(newext) == 0");
1977 		return -EFSCORRUPTED;
1978 	}
1979 	depth = ext_depth(inode);
1980 	ex = path[depth].p_ext;
1981 	eh = path[depth].p_hdr;
1982 	if (unlikely(path[depth].p_hdr == NULL)) {
1983 		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
1984 		return -EFSCORRUPTED;
1985 	}
1986 
1987 	/* try to insert block into found extent and return */
1988 	if (ex && !(gb_flags & EXT4_GET_BLOCKS_PRE_IO)) {
1989 
1990 		/*
1991 		 * Try to see whether we should rather test the extent on
1992 		 * right from ex, or from the left of ex. This is because
1993 		 * ext4_find_extent() can return either extent on the
1994 		 * left, or on the right from the searched position. This
1995 		 * will make merging more effective.
1996 		 */
1997 		if (ex < EXT_LAST_EXTENT(eh) &&
1998 		    (le32_to_cpu(ex->ee_block) +
1999 		    ext4_ext_get_actual_len(ex) <
2000 		    le32_to_cpu(newext->ee_block))) {
2001 			ex += 1;
2002 			goto prepend;
2003 		} else if ((ex > EXT_FIRST_EXTENT(eh)) &&
2004 			   (le32_to_cpu(newext->ee_block) +
2005 			   ext4_ext_get_actual_len(newext) <
2006 			   le32_to_cpu(ex->ee_block)))
2007 			ex -= 1;
2008 
2009 		/* Try to append newex to the ex */
2010 		if (ext4_can_extents_be_merged(inode, ex, newext)) {
2011 			ext_debug(inode, "append [%d]%d block to %u:[%d]%d"
2012 				  "(from %llu)\n",
2013 				  ext4_ext_is_unwritten(newext),
2014 				  ext4_ext_get_actual_len(newext),
2015 				  le32_to_cpu(ex->ee_block),
2016 				  ext4_ext_is_unwritten(ex),
2017 				  ext4_ext_get_actual_len(ex),
2018 				  ext4_ext_pblock(ex));
2019 			err = ext4_ext_get_access(handle, inode,
2020 						  path + depth);
2021 			if (err)
2022 				return err;
2023 			unwritten = ext4_ext_is_unwritten(ex);
2024 			ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2025 					+ ext4_ext_get_actual_len(newext));
2026 			if (unwritten)
2027 				ext4_ext_mark_unwritten(ex);
2028 			eh = path[depth].p_hdr;
2029 			nearex = ex;
2030 			goto merge;
2031 		}
2032 
2033 prepend:
2034 		/* Try to prepend newex to the ex */
2035 		if (ext4_can_extents_be_merged(inode, newext, ex)) {
2036 			ext_debug(inode, "prepend %u[%d]%d block to %u:[%d]%d"
2037 				  "(from %llu)\n",
2038 				  le32_to_cpu(newext->ee_block),
2039 				  ext4_ext_is_unwritten(newext),
2040 				  ext4_ext_get_actual_len(newext),
2041 				  le32_to_cpu(ex->ee_block),
2042 				  ext4_ext_is_unwritten(ex),
2043 				  ext4_ext_get_actual_len(ex),
2044 				  ext4_ext_pblock(ex));
2045 			err = ext4_ext_get_access(handle, inode,
2046 						  path + depth);
2047 			if (err)
2048 				return err;
2049 
2050 			unwritten = ext4_ext_is_unwritten(ex);
2051 			ex->ee_block = newext->ee_block;
2052 			ext4_ext_store_pblock(ex, ext4_ext_pblock(newext));
2053 			ex->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ex)
2054 					+ ext4_ext_get_actual_len(newext));
2055 			if (unwritten)
2056 				ext4_ext_mark_unwritten(ex);
2057 			eh = path[depth].p_hdr;
2058 			nearex = ex;
2059 			goto merge;
2060 		}
2061 	}
2062 
2063 	depth = ext_depth(inode);
2064 	eh = path[depth].p_hdr;
2065 	if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max))
2066 		goto has_space;
2067 
2068 	/* probably next leaf has space for us? */
2069 	fex = EXT_LAST_EXTENT(eh);
2070 	next = EXT_MAX_BLOCKS;
2071 	if (le32_to_cpu(newext->ee_block) > le32_to_cpu(fex->ee_block))
2072 		next = ext4_ext_next_leaf_block(path);
2073 	if (next != EXT_MAX_BLOCKS) {
2074 		ext_debug(inode, "next leaf block - %u\n", next);
2075 		BUG_ON(npath != NULL);
2076 		npath = ext4_find_extent(inode, next, NULL, gb_flags);
2077 		if (IS_ERR(npath))
2078 			return PTR_ERR(npath);
2079 		BUG_ON(npath->p_depth != path->p_depth);
2080 		eh = npath[depth].p_hdr;
2081 		if (le16_to_cpu(eh->eh_entries) < le16_to_cpu(eh->eh_max)) {
2082 			ext_debug(inode, "next leaf isn't full(%d)\n",
2083 				  le16_to_cpu(eh->eh_entries));
2084 			path = npath;
2085 			goto has_space;
2086 		}
2087 		ext_debug(inode, "next leaf has no free space(%d,%d)\n",
2088 			  le16_to_cpu(eh->eh_entries), le16_to_cpu(eh->eh_max));
2089 	}
2090 
2091 	/*
2092 	 * There is no free space in the found leaf.
2093 	 * We're gonna add a new leaf in the tree.
2094 	 */
2095 	if (gb_flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
2096 		mb_flags |= EXT4_MB_USE_RESERVED;
2097 	err = ext4_ext_create_new_leaf(handle, inode, mb_flags, gb_flags,
2098 				       ppath, newext);
2099 	if (err)
2100 		goto cleanup;
2101 	depth = ext_depth(inode);
2102 	eh = path[depth].p_hdr;
2103 
2104 has_space:
2105 	nearex = path[depth].p_ext;
2106 
2107 	err = ext4_ext_get_access(handle, inode, path + depth);
2108 	if (err)
2109 		goto cleanup;
2110 
2111 	if (!nearex) {
2112 		/* there is no extent in this leaf, create first one */
2113 		ext_debug(inode, "first extent in the leaf: %u:%llu:[%d]%d\n",
2114 				le32_to_cpu(newext->ee_block),
2115 				ext4_ext_pblock(newext),
2116 				ext4_ext_is_unwritten(newext),
2117 				ext4_ext_get_actual_len(newext));
2118 		nearex = EXT_FIRST_EXTENT(eh);
2119 	} else {
2120 		if (le32_to_cpu(newext->ee_block)
2121 			   > le32_to_cpu(nearex->ee_block)) {
2122 			/* Insert after */
2123 			ext_debug(inode, "insert %u:%llu:[%d]%d before: "
2124 					"nearest %p\n",
2125 					le32_to_cpu(newext->ee_block),
2126 					ext4_ext_pblock(newext),
2127 					ext4_ext_is_unwritten(newext),
2128 					ext4_ext_get_actual_len(newext),
2129 					nearex);
2130 			nearex++;
2131 		} else {
2132 			/* Insert before */
2133 			BUG_ON(newext->ee_block == nearex->ee_block);
2134 			ext_debug(inode, "insert %u:%llu:[%d]%d after: "
2135 					"nearest %p\n",
2136 					le32_to_cpu(newext->ee_block),
2137 					ext4_ext_pblock(newext),
2138 					ext4_ext_is_unwritten(newext),
2139 					ext4_ext_get_actual_len(newext),
2140 					nearex);
2141 		}
2142 		len = EXT_LAST_EXTENT(eh) - nearex + 1;
2143 		if (len > 0) {
2144 			ext_debug(inode, "insert %u:%llu:[%d]%d: "
2145 					"move %d extents from 0x%p to 0x%p\n",
2146 					le32_to_cpu(newext->ee_block),
2147 					ext4_ext_pblock(newext),
2148 					ext4_ext_is_unwritten(newext),
2149 					ext4_ext_get_actual_len(newext),
2150 					len, nearex, nearex + 1);
2151 			memmove(nearex + 1, nearex,
2152 				len * sizeof(struct ext4_extent));
2153 		}
2154 	}
2155 
2156 	le16_add_cpu(&eh->eh_entries, 1);
2157 	path[depth].p_ext = nearex;
2158 	nearex->ee_block = newext->ee_block;
2159 	ext4_ext_store_pblock(nearex, ext4_ext_pblock(newext));
2160 	nearex->ee_len = newext->ee_len;
2161 
2162 merge:
2163 	/* try to merge extents */
2164 	if (!(gb_flags & EXT4_GET_BLOCKS_PRE_IO))
2165 		ext4_ext_try_to_merge(handle, inode, path, nearex);
2166 
2167 
2168 	/* time to correct all indexes above */
2169 	err = ext4_ext_correct_indexes(handle, inode, path);
2170 	if (err)
2171 		goto cleanup;
2172 
2173 	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
2174 
2175 cleanup:
2176 	ext4_ext_drop_refs(npath);
2177 	kfree(npath);
2178 	return err;
2179 }
2180 
2181 static int ext4_fill_es_cache_info(struct inode *inode,
2182 				   ext4_lblk_t block, ext4_lblk_t num,
2183 				   struct fiemap_extent_info *fieinfo)
2184 {
2185 	ext4_lblk_t next, end = block + num - 1;
2186 	struct extent_status es;
2187 	unsigned char blksize_bits = inode->i_sb->s_blocksize_bits;
2188 	unsigned int flags;
2189 	int err;
2190 
2191 	while (block <= end) {
2192 		next = 0;
2193 		flags = 0;
2194 		if (!ext4_es_lookup_extent(inode, block, &next, &es))
2195 			break;
2196 		if (ext4_es_is_unwritten(&es))
2197 			flags |= FIEMAP_EXTENT_UNWRITTEN;
2198 		if (ext4_es_is_delayed(&es))
2199 			flags |= (FIEMAP_EXTENT_DELALLOC |
2200 				  FIEMAP_EXTENT_UNKNOWN);
2201 		if (ext4_es_is_hole(&es))
2202 			flags |= EXT4_FIEMAP_EXTENT_HOLE;
2203 		if (next == 0)
2204 			flags |= FIEMAP_EXTENT_LAST;
2205 		if (flags & (FIEMAP_EXTENT_DELALLOC|
2206 			     EXT4_FIEMAP_EXTENT_HOLE))
2207 			es.es_pblk = 0;
2208 		else
2209 			es.es_pblk = ext4_es_pblock(&es);
2210 		err = fiemap_fill_next_extent(fieinfo,
2211 				(__u64)es.es_lblk << blksize_bits,
2212 				(__u64)es.es_pblk << blksize_bits,
2213 				(__u64)es.es_len << blksize_bits,
2214 				flags);
2215 		if (next == 0)
2216 			break;
2217 		block = next;
2218 		if (err < 0)
2219 			return err;
2220 		if (err == 1)
2221 			return 0;
2222 	}
2223 	return 0;
2224 }
2225 
2226 
2227 /*
2228  * ext4_ext_determine_hole - determine hole around given block
2229  * @inode:	inode we lookup in
2230  * @path:	path in extent tree to @lblk
2231  * @lblk:	pointer to logical block around which we want to determine hole
2232  *
2233  * Determine hole length (and start if easily possible) around given logical
2234  * block. We don't try too hard to find the beginning of the hole but @path
2235  * actually points to extent before @lblk, we provide it.
2236  *
2237  * The function returns the length of a hole starting at @lblk. We update @lblk
2238  * to the beginning of the hole if we managed to find it.
2239  */
2240 static ext4_lblk_t ext4_ext_determine_hole(struct inode *inode,
2241 					   struct ext4_ext_path *path,
2242 					   ext4_lblk_t *lblk)
2243 {
2244 	int depth = ext_depth(inode);
2245 	struct ext4_extent *ex;
2246 	ext4_lblk_t len;
2247 
2248 	ex = path[depth].p_ext;
2249 	if (ex == NULL) {
2250 		/* there is no extent yet, so gap is [0;-] */
2251 		*lblk = 0;
2252 		len = EXT_MAX_BLOCKS;
2253 	} else if (*lblk < le32_to_cpu(ex->ee_block)) {
2254 		len = le32_to_cpu(ex->ee_block) - *lblk;
2255 	} else if (*lblk >= le32_to_cpu(ex->ee_block)
2256 			+ ext4_ext_get_actual_len(ex)) {
2257 		ext4_lblk_t next;
2258 
2259 		*lblk = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
2260 		next = ext4_ext_next_allocated_block(path);
2261 		BUG_ON(next == *lblk);
2262 		len = next - *lblk;
2263 	} else {
2264 		BUG();
2265 	}
2266 	return len;
2267 }
2268 
2269 /*
2270  * ext4_ext_put_gap_in_cache:
2271  * calculate boundaries of the gap that the requested block fits into
2272  * and cache this gap
2273  */
2274 static void
2275 ext4_ext_put_gap_in_cache(struct inode *inode, ext4_lblk_t hole_start,
2276 			  ext4_lblk_t hole_len)
2277 {
2278 	struct extent_status es;
2279 
2280 	ext4_es_find_extent_range(inode, &ext4_es_is_delayed, hole_start,
2281 				  hole_start + hole_len - 1, &es);
2282 	if (es.es_len) {
2283 		/* There's delayed extent containing lblock? */
2284 		if (es.es_lblk <= hole_start)
2285 			return;
2286 		hole_len = min(es.es_lblk - hole_start, hole_len);
2287 	}
2288 	ext_debug(inode, " -> %u:%u\n", hole_start, hole_len);
2289 	ext4_es_insert_extent(inode, hole_start, hole_len, ~0,
2290 			      EXTENT_STATUS_HOLE);
2291 }
2292 
2293 /*
2294  * ext4_ext_rm_idx:
2295  * removes index from the index block.
2296  */
2297 static int ext4_ext_rm_idx(handle_t *handle, struct inode *inode,
2298 			struct ext4_ext_path *path, int depth)
2299 {
2300 	int err;
2301 	ext4_fsblk_t leaf;
2302 
2303 	/* free index block */
2304 	depth--;
2305 	path = path + depth;
2306 	leaf = ext4_idx_pblock(path->p_idx);
2307 	if (unlikely(path->p_hdr->eh_entries == 0)) {
2308 		EXT4_ERROR_INODE(inode, "path->p_hdr->eh_entries == 0");
2309 		return -EFSCORRUPTED;
2310 	}
2311 	err = ext4_ext_get_access(handle, inode, path);
2312 	if (err)
2313 		return err;
2314 
2315 	if (path->p_idx != EXT_LAST_INDEX(path->p_hdr)) {
2316 		int len = EXT_LAST_INDEX(path->p_hdr) - path->p_idx;
2317 		len *= sizeof(struct ext4_extent_idx);
2318 		memmove(path->p_idx, path->p_idx + 1, len);
2319 	}
2320 
2321 	le16_add_cpu(&path->p_hdr->eh_entries, -1);
2322 	err = ext4_ext_dirty(handle, inode, path);
2323 	if (err)
2324 		return err;
2325 	ext_debug(inode, "index is empty, remove it, free block %llu\n", leaf);
2326 	trace_ext4_ext_rm_idx(inode, leaf);
2327 
2328 	ext4_free_blocks(handle, inode, NULL, leaf, 1,
2329 			 EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET);
2330 
2331 	while (--depth >= 0) {
2332 		if (path->p_idx != EXT_FIRST_INDEX(path->p_hdr))
2333 			break;
2334 		path--;
2335 		err = ext4_ext_get_access(handle, inode, path);
2336 		if (err)
2337 			break;
2338 		path->p_idx->ei_block = (path+1)->p_idx->ei_block;
2339 		err = ext4_ext_dirty(handle, inode, path);
2340 		if (err)
2341 			break;
2342 	}
2343 	return err;
2344 }
2345 
2346 /*
2347  * ext4_ext_calc_credits_for_single_extent:
2348  * This routine returns max. credits that needed to insert an extent
2349  * to the extent tree.
2350  * When pass the actual path, the caller should calculate credits
2351  * under i_data_sem.
2352  */
2353 int ext4_ext_calc_credits_for_single_extent(struct inode *inode, int nrblocks,
2354 						struct ext4_ext_path *path)
2355 {
2356 	if (path) {
2357 		int depth = ext_depth(inode);
2358 		int ret = 0;
2359 
2360 		/* probably there is space in leaf? */
2361 		if (le16_to_cpu(path[depth].p_hdr->eh_entries)
2362 				< le16_to_cpu(path[depth].p_hdr->eh_max)) {
2363 
2364 			/*
2365 			 *  There are some space in the leaf tree, no
2366 			 *  need to account for leaf block credit
2367 			 *
2368 			 *  bitmaps and block group descriptor blocks
2369 			 *  and other metadata blocks still need to be
2370 			 *  accounted.
2371 			 */
2372 			/* 1 bitmap, 1 block group descriptor */
2373 			ret = 2 + EXT4_META_TRANS_BLOCKS(inode->i_sb);
2374 			return ret;
2375 		}
2376 	}
2377 
2378 	return ext4_chunk_trans_blocks(inode, nrblocks);
2379 }
2380 
2381 /*
2382  * How many index/leaf blocks need to change/allocate to add @extents extents?
2383  *
2384  * If we add a single extent, then in the worse case, each tree level
2385  * index/leaf need to be changed in case of the tree split.
2386  *
2387  * If more extents are inserted, they could cause the whole tree split more
2388  * than once, but this is really rare.
2389  */
2390 int ext4_ext_index_trans_blocks(struct inode *inode, int extents)
2391 {
2392 	int index;
2393 	int depth;
2394 
2395 	/* If we are converting the inline data, only one is needed here. */
2396 	if (ext4_has_inline_data(inode))
2397 		return 1;
2398 
2399 	depth = ext_depth(inode);
2400 
2401 	if (extents <= 1)
2402 		index = depth * 2;
2403 	else
2404 		index = depth * 3;
2405 
2406 	return index;
2407 }
2408 
2409 static inline int get_default_free_blocks_flags(struct inode *inode)
2410 {
2411 	if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode) ||
2412 	    ext4_test_inode_flag(inode, EXT4_INODE_EA_INODE))
2413 		return EXT4_FREE_BLOCKS_METADATA | EXT4_FREE_BLOCKS_FORGET;
2414 	else if (ext4_should_journal_data(inode))
2415 		return EXT4_FREE_BLOCKS_FORGET;
2416 	return 0;
2417 }
2418 
2419 /*
2420  * ext4_rereserve_cluster - increment the reserved cluster count when
2421  *                          freeing a cluster with a pending reservation
2422  *
2423  * @inode - file containing the cluster
2424  * @lblk - logical block in cluster to be reserved
2425  *
2426  * Increments the reserved cluster count and adjusts quota in a bigalloc
2427  * file system when freeing a partial cluster containing at least one
2428  * delayed and unwritten block.  A partial cluster meeting that
2429  * requirement will have a pending reservation.  If so, the
2430  * RERESERVE_CLUSTER flag is used when calling ext4_free_blocks() to
2431  * defer reserved and allocated space accounting to a subsequent call
2432  * to this function.
2433  */
2434 static void ext4_rereserve_cluster(struct inode *inode, ext4_lblk_t lblk)
2435 {
2436 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2437 	struct ext4_inode_info *ei = EXT4_I(inode);
2438 
2439 	dquot_reclaim_block(inode, EXT4_C2B(sbi, 1));
2440 
2441 	spin_lock(&ei->i_block_reservation_lock);
2442 	ei->i_reserved_data_blocks++;
2443 	percpu_counter_add(&sbi->s_dirtyclusters_counter, 1);
2444 	spin_unlock(&ei->i_block_reservation_lock);
2445 
2446 	percpu_counter_add(&sbi->s_freeclusters_counter, 1);
2447 	ext4_remove_pending(inode, lblk);
2448 }
2449 
2450 static int ext4_remove_blocks(handle_t *handle, struct inode *inode,
2451 			      struct ext4_extent *ex,
2452 			      struct partial_cluster *partial,
2453 			      ext4_lblk_t from, ext4_lblk_t to)
2454 {
2455 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2456 	unsigned short ee_len = ext4_ext_get_actual_len(ex);
2457 	ext4_fsblk_t last_pblk, pblk;
2458 	ext4_lblk_t num;
2459 	int flags;
2460 
2461 	/* only extent tail removal is allowed */
2462 	if (from < le32_to_cpu(ex->ee_block) ||
2463 	    to != le32_to_cpu(ex->ee_block) + ee_len - 1) {
2464 		ext4_error(sbi->s_sb,
2465 			   "strange request: removal(2) %u-%u from %u:%u",
2466 			   from, to, le32_to_cpu(ex->ee_block), ee_len);
2467 		return 0;
2468 	}
2469 
2470 #ifdef EXTENTS_STATS
2471 	spin_lock(&sbi->s_ext_stats_lock);
2472 	sbi->s_ext_blocks += ee_len;
2473 	sbi->s_ext_extents++;
2474 	if (ee_len < sbi->s_ext_min)
2475 		sbi->s_ext_min = ee_len;
2476 	if (ee_len > sbi->s_ext_max)
2477 		sbi->s_ext_max = ee_len;
2478 	if (ext_depth(inode) > sbi->s_depth_max)
2479 		sbi->s_depth_max = ext_depth(inode);
2480 	spin_unlock(&sbi->s_ext_stats_lock);
2481 #endif
2482 
2483 	trace_ext4_remove_blocks(inode, ex, from, to, partial);
2484 
2485 	/*
2486 	 * if we have a partial cluster, and it's different from the
2487 	 * cluster of the last block in the extent, we free it
2488 	 */
2489 	last_pblk = ext4_ext_pblock(ex) + ee_len - 1;
2490 
2491 	if (partial->state != initial &&
2492 	    partial->pclu != EXT4_B2C(sbi, last_pblk)) {
2493 		if (partial->state == tofree) {
2494 			flags = get_default_free_blocks_flags(inode);
2495 			if (ext4_is_pending(inode, partial->lblk))
2496 				flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2497 			ext4_free_blocks(handle, inode, NULL,
2498 					 EXT4_C2B(sbi, partial->pclu),
2499 					 sbi->s_cluster_ratio, flags);
2500 			if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2501 				ext4_rereserve_cluster(inode, partial->lblk);
2502 		}
2503 		partial->state = initial;
2504 	}
2505 
2506 	num = le32_to_cpu(ex->ee_block) + ee_len - from;
2507 	pblk = ext4_ext_pblock(ex) + ee_len - num;
2508 
2509 	/*
2510 	 * We free the partial cluster at the end of the extent (if any),
2511 	 * unless the cluster is used by another extent (partial_cluster
2512 	 * state is nofree).  If a partial cluster exists here, it must be
2513 	 * shared with the last block in the extent.
2514 	 */
2515 	flags = get_default_free_blocks_flags(inode);
2516 
2517 	/* partial, left end cluster aligned, right end unaligned */
2518 	if ((EXT4_LBLK_COFF(sbi, to) != sbi->s_cluster_ratio - 1) &&
2519 	    (EXT4_LBLK_CMASK(sbi, to) >= from) &&
2520 	    (partial->state != nofree)) {
2521 		if (ext4_is_pending(inode, to))
2522 			flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2523 		ext4_free_blocks(handle, inode, NULL,
2524 				 EXT4_PBLK_CMASK(sbi, last_pblk),
2525 				 sbi->s_cluster_ratio, flags);
2526 		if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2527 			ext4_rereserve_cluster(inode, to);
2528 		partial->state = initial;
2529 		flags = get_default_free_blocks_flags(inode);
2530 	}
2531 
2532 	flags |= EXT4_FREE_BLOCKS_NOFREE_LAST_CLUSTER;
2533 
2534 	/*
2535 	 * For bigalloc file systems, we never free a partial cluster
2536 	 * at the beginning of the extent.  Instead, we check to see if we
2537 	 * need to free it on a subsequent call to ext4_remove_blocks,
2538 	 * or at the end of ext4_ext_rm_leaf or ext4_ext_remove_space.
2539 	 */
2540 	flags |= EXT4_FREE_BLOCKS_NOFREE_FIRST_CLUSTER;
2541 	ext4_free_blocks(handle, inode, NULL, pblk, num, flags);
2542 
2543 	/* reset the partial cluster if we've freed past it */
2544 	if (partial->state != initial && partial->pclu != EXT4_B2C(sbi, pblk))
2545 		partial->state = initial;
2546 
2547 	/*
2548 	 * If we've freed the entire extent but the beginning is not left
2549 	 * cluster aligned and is not marked as ineligible for freeing we
2550 	 * record the partial cluster at the beginning of the extent.  It
2551 	 * wasn't freed by the preceding ext4_free_blocks() call, and we
2552 	 * need to look farther to the left to determine if it's to be freed
2553 	 * (not shared with another extent). Else, reset the partial
2554 	 * cluster - we're either  done freeing or the beginning of the
2555 	 * extent is left cluster aligned.
2556 	 */
2557 	if (EXT4_LBLK_COFF(sbi, from) && num == ee_len) {
2558 		if (partial->state == initial) {
2559 			partial->pclu = EXT4_B2C(sbi, pblk);
2560 			partial->lblk = from;
2561 			partial->state = tofree;
2562 		}
2563 	} else {
2564 		partial->state = initial;
2565 	}
2566 
2567 	return 0;
2568 }
2569 
2570 /*
2571  * ext4_ext_rm_leaf() Removes the extents associated with the
2572  * blocks appearing between "start" and "end".  Both "start"
2573  * and "end" must appear in the same extent or EIO is returned.
2574  *
2575  * @handle: The journal handle
2576  * @inode:  The files inode
2577  * @path:   The path to the leaf
2578  * @partial_cluster: The cluster which we'll have to free if all extents
2579  *                   has been released from it.  However, if this value is
2580  *                   negative, it's a cluster just to the right of the
2581  *                   punched region and it must not be freed.
2582  * @start:  The first block to remove
2583  * @end:   The last block to remove
2584  */
2585 static int
2586 ext4_ext_rm_leaf(handle_t *handle, struct inode *inode,
2587 		 struct ext4_ext_path *path,
2588 		 struct partial_cluster *partial,
2589 		 ext4_lblk_t start, ext4_lblk_t end)
2590 {
2591 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2592 	int err = 0, correct_index = 0;
2593 	int depth = ext_depth(inode), credits, revoke_credits;
2594 	struct ext4_extent_header *eh;
2595 	ext4_lblk_t a, b;
2596 	unsigned num;
2597 	ext4_lblk_t ex_ee_block;
2598 	unsigned short ex_ee_len;
2599 	unsigned unwritten = 0;
2600 	struct ext4_extent *ex;
2601 	ext4_fsblk_t pblk;
2602 
2603 	/* the header must be checked already in ext4_ext_remove_space() */
2604 	ext_debug(inode, "truncate since %u in leaf to %u\n", start, end);
2605 	if (!path[depth].p_hdr)
2606 		path[depth].p_hdr = ext_block_hdr(path[depth].p_bh);
2607 	eh = path[depth].p_hdr;
2608 	if (unlikely(path[depth].p_hdr == NULL)) {
2609 		EXT4_ERROR_INODE(inode, "path[%d].p_hdr == NULL", depth);
2610 		return -EFSCORRUPTED;
2611 	}
2612 	/* find where to start removing */
2613 	ex = path[depth].p_ext;
2614 	if (!ex)
2615 		ex = EXT_LAST_EXTENT(eh);
2616 
2617 	ex_ee_block = le32_to_cpu(ex->ee_block);
2618 	ex_ee_len = ext4_ext_get_actual_len(ex);
2619 
2620 	trace_ext4_ext_rm_leaf(inode, start, ex, partial);
2621 
2622 	while (ex >= EXT_FIRST_EXTENT(eh) &&
2623 			ex_ee_block + ex_ee_len > start) {
2624 
2625 		if (ext4_ext_is_unwritten(ex))
2626 			unwritten = 1;
2627 		else
2628 			unwritten = 0;
2629 
2630 		ext_debug(inode, "remove ext %u:[%d]%d\n", ex_ee_block,
2631 			  unwritten, ex_ee_len);
2632 		path[depth].p_ext = ex;
2633 
2634 		a = ex_ee_block > start ? ex_ee_block : start;
2635 		b = ex_ee_block+ex_ee_len - 1 < end ?
2636 			ex_ee_block+ex_ee_len - 1 : end;
2637 
2638 		ext_debug(inode, "  border %u:%u\n", a, b);
2639 
2640 		/* If this extent is beyond the end of the hole, skip it */
2641 		if (end < ex_ee_block) {
2642 			/*
2643 			 * We're going to skip this extent and move to another,
2644 			 * so note that its first cluster is in use to avoid
2645 			 * freeing it when removing blocks.  Eventually, the
2646 			 * right edge of the truncated/punched region will
2647 			 * be just to the left.
2648 			 */
2649 			if (sbi->s_cluster_ratio > 1) {
2650 				pblk = ext4_ext_pblock(ex);
2651 				partial->pclu = EXT4_B2C(sbi, pblk);
2652 				partial->state = nofree;
2653 			}
2654 			ex--;
2655 			ex_ee_block = le32_to_cpu(ex->ee_block);
2656 			ex_ee_len = ext4_ext_get_actual_len(ex);
2657 			continue;
2658 		} else if (b != ex_ee_block + ex_ee_len - 1) {
2659 			EXT4_ERROR_INODE(inode,
2660 					 "can not handle truncate %u:%u "
2661 					 "on extent %u:%u",
2662 					 start, end, ex_ee_block,
2663 					 ex_ee_block + ex_ee_len - 1);
2664 			err = -EFSCORRUPTED;
2665 			goto out;
2666 		} else if (a != ex_ee_block) {
2667 			/* remove tail of the extent */
2668 			num = a - ex_ee_block;
2669 		} else {
2670 			/* remove whole extent: excellent! */
2671 			num = 0;
2672 		}
2673 		/*
2674 		 * 3 for leaf, sb, and inode plus 2 (bmap and group
2675 		 * descriptor) for each block group; assume two block
2676 		 * groups plus ex_ee_len/blocks_per_block_group for
2677 		 * the worst case
2678 		 */
2679 		credits = 7 + 2*(ex_ee_len/EXT4_BLOCKS_PER_GROUP(inode->i_sb));
2680 		if (ex == EXT_FIRST_EXTENT(eh)) {
2681 			correct_index = 1;
2682 			credits += (ext_depth(inode)) + 1;
2683 		}
2684 		credits += EXT4_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb);
2685 		/*
2686 		 * We may end up freeing some index blocks and data from the
2687 		 * punched range. Note that partial clusters are accounted for
2688 		 * by ext4_free_data_revoke_credits().
2689 		 */
2690 		revoke_credits =
2691 			ext4_free_metadata_revoke_credits(inode->i_sb,
2692 							  ext_depth(inode)) +
2693 			ext4_free_data_revoke_credits(inode, b - a + 1);
2694 
2695 		err = ext4_datasem_ensure_credits(handle, inode, credits,
2696 						  credits, revoke_credits);
2697 		if (err) {
2698 			if (err > 0)
2699 				err = -EAGAIN;
2700 			goto out;
2701 		}
2702 
2703 		err = ext4_ext_get_access(handle, inode, path + depth);
2704 		if (err)
2705 			goto out;
2706 
2707 		err = ext4_remove_blocks(handle, inode, ex, partial, a, b);
2708 		if (err)
2709 			goto out;
2710 
2711 		if (num == 0)
2712 			/* this extent is removed; mark slot entirely unused */
2713 			ext4_ext_store_pblock(ex, 0);
2714 
2715 		ex->ee_len = cpu_to_le16(num);
2716 		/*
2717 		 * Do not mark unwritten if all the blocks in the
2718 		 * extent have been removed.
2719 		 */
2720 		if (unwritten && num)
2721 			ext4_ext_mark_unwritten(ex);
2722 		/*
2723 		 * If the extent was completely released,
2724 		 * we need to remove it from the leaf
2725 		 */
2726 		if (num == 0) {
2727 			if (end != EXT_MAX_BLOCKS - 1) {
2728 				/*
2729 				 * For hole punching, we need to scoot all the
2730 				 * extents up when an extent is removed so that
2731 				 * we dont have blank extents in the middle
2732 				 */
2733 				memmove(ex, ex+1, (EXT_LAST_EXTENT(eh) - ex) *
2734 					sizeof(struct ext4_extent));
2735 
2736 				/* Now get rid of the one at the end */
2737 				memset(EXT_LAST_EXTENT(eh), 0,
2738 					sizeof(struct ext4_extent));
2739 			}
2740 			le16_add_cpu(&eh->eh_entries, -1);
2741 		}
2742 
2743 		err = ext4_ext_dirty(handle, inode, path + depth);
2744 		if (err)
2745 			goto out;
2746 
2747 		ext_debug(inode, "new extent: %u:%u:%llu\n", ex_ee_block, num,
2748 				ext4_ext_pblock(ex));
2749 		ex--;
2750 		ex_ee_block = le32_to_cpu(ex->ee_block);
2751 		ex_ee_len = ext4_ext_get_actual_len(ex);
2752 	}
2753 
2754 	if (correct_index && eh->eh_entries)
2755 		err = ext4_ext_correct_indexes(handle, inode, path);
2756 
2757 	/*
2758 	 * If there's a partial cluster and at least one extent remains in
2759 	 * the leaf, free the partial cluster if it isn't shared with the
2760 	 * current extent.  If it is shared with the current extent
2761 	 * we reset the partial cluster because we've reached the start of the
2762 	 * truncated/punched region and we're done removing blocks.
2763 	 */
2764 	if (partial->state == tofree && ex >= EXT_FIRST_EXTENT(eh)) {
2765 		pblk = ext4_ext_pblock(ex) + ex_ee_len - 1;
2766 		if (partial->pclu != EXT4_B2C(sbi, pblk)) {
2767 			int flags = get_default_free_blocks_flags(inode);
2768 
2769 			if (ext4_is_pending(inode, partial->lblk))
2770 				flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
2771 			ext4_free_blocks(handle, inode, NULL,
2772 					 EXT4_C2B(sbi, partial->pclu),
2773 					 sbi->s_cluster_ratio, flags);
2774 			if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
2775 				ext4_rereserve_cluster(inode, partial->lblk);
2776 		}
2777 		partial->state = initial;
2778 	}
2779 
2780 	/* if this leaf is free, then we should
2781 	 * remove it from index block above */
2782 	if (err == 0 && eh->eh_entries == 0 && path[depth].p_bh != NULL)
2783 		err = ext4_ext_rm_idx(handle, inode, path, depth);
2784 
2785 out:
2786 	return err;
2787 }
2788 
2789 /*
2790  * ext4_ext_more_to_rm:
2791  * returns 1 if current index has to be freed (even partial)
2792  */
2793 static int
2794 ext4_ext_more_to_rm(struct ext4_ext_path *path)
2795 {
2796 	BUG_ON(path->p_idx == NULL);
2797 
2798 	if (path->p_idx < EXT_FIRST_INDEX(path->p_hdr))
2799 		return 0;
2800 
2801 	/*
2802 	 * if truncate on deeper level happened, it wasn't partial,
2803 	 * so we have to consider current index for truncation
2804 	 */
2805 	if (le16_to_cpu(path->p_hdr->eh_entries) == path->p_block)
2806 		return 0;
2807 	return 1;
2808 }
2809 
2810 int ext4_ext_remove_space(struct inode *inode, ext4_lblk_t start,
2811 			  ext4_lblk_t end)
2812 {
2813 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
2814 	int depth = ext_depth(inode);
2815 	struct ext4_ext_path *path = NULL;
2816 	struct partial_cluster partial;
2817 	handle_t *handle;
2818 	int i = 0, err = 0;
2819 
2820 	partial.pclu = 0;
2821 	partial.lblk = 0;
2822 	partial.state = initial;
2823 
2824 	ext_debug(inode, "truncate since %u to %u\n", start, end);
2825 
2826 	/* probably first extent we're gonna free will be last in block */
2827 	handle = ext4_journal_start_with_revoke(inode, EXT4_HT_TRUNCATE,
2828 			depth + 1,
2829 			ext4_free_metadata_revoke_credits(inode->i_sb, depth));
2830 	if (IS_ERR(handle))
2831 		return PTR_ERR(handle);
2832 
2833 again:
2834 	trace_ext4_ext_remove_space(inode, start, end, depth);
2835 
2836 	/*
2837 	 * Check if we are removing extents inside the extent tree. If that
2838 	 * is the case, we are going to punch a hole inside the extent tree
2839 	 * so we have to check whether we need to split the extent covering
2840 	 * the last block to remove so we can easily remove the part of it
2841 	 * in ext4_ext_rm_leaf().
2842 	 */
2843 	if (end < EXT_MAX_BLOCKS - 1) {
2844 		struct ext4_extent *ex;
2845 		ext4_lblk_t ee_block, ex_end, lblk;
2846 		ext4_fsblk_t pblk;
2847 
2848 		/* find extent for or closest extent to this block */
2849 		path = ext4_find_extent(inode, end, NULL,
2850 					EXT4_EX_NOCACHE | EXT4_EX_NOFAIL);
2851 		if (IS_ERR(path)) {
2852 			ext4_journal_stop(handle);
2853 			return PTR_ERR(path);
2854 		}
2855 		depth = ext_depth(inode);
2856 		/* Leaf not may not exist only if inode has no blocks at all */
2857 		ex = path[depth].p_ext;
2858 		if (!ex) {
2859 			if (depth) {
2860 				EXT4_ERROR_INODE(inode,
2861 						 "path[%d].p_hdr == NULL",
2862 						 depth);
2863 				err = -EFSCORRUPTED;
2864 			}
2865 			goto out;
2866 		}
2867 
2868 		ee_block = le32_to_cpu(ex->ee_block);
2869 		ex_end = ee_block + ext4_ext_get_actual_len(ex) - 1;
2870 
2871 		/*
2872 		 * See if the last block is inside the extent, if so split
2873 		 * the extent at 'end' block so we can easily remove the
2874 		 * tail of the first part of the split extent in
2875 		 * ext4_ext_rm_leaf().
2876 		 */
2877 		if (end >= ee_block && end < ex_end) {
2878 
2879 			/*
2880 			 * If we're going to split the extent, note that
2881 			 * the cluster containing the block after 'end' is
2882 			 * in use to avoid freeing it when removing blocks.
2883 			 */
2884 			if (sbi->s_cluster_ratio > 1) {
2885 				pblk = ext4_ext_pblock(ex) + end - ee_block + 1;
2886 				partial.pclu = EXT4_B2C(sbi, pblk);
2887 				partial.state = nofree;
2888 			}
2889 
2890 			/*
2891 			 * Split the extent in two so that 'end' is the last
2892 			 * block in the first new extent. Also we should not
2893 			 * fail removing space due to ENOSPC so try to use
2894 			 * reserved block if that happens.
2895 			 */
2896 			err = ext4_force_split_extent_at(handle, inode, &path,
2897 							 end + 1, 1);
2898 			if (err < 0)
2899 				goto out;
2900 
2901 		} else if (sbi->s_cluster_ratio > 1 && end >= ex_end &&
2902 			   partial.state == initial) {
2903 			/*
2904 			 * If we're punching, there's an extent to the right.
2905 			 * If the partial cluster hasn't been set, set it to
2906 			 * that extent's first cluster and its state to nofree
2907 			 * so it won't be freed should it contain blocks to be
2908 			 * removed. If it's already set (tofree/nofree), we're
2909 			 * retrying and keep the original partial cluster info
2910 			 * so a cluster marked tofree as a result of earlier
2911 			 * extent removal is not lost.
2912 			 */
2913 			lblk = ex_end + 1;
2914 			err = ext4_ext_search_right(inode, path, &lblk, &pblk,
2915 						    NULL);
2916 			if (err < 0)
2917 				goto out;
2918 			if (pblk) {
2919 				partial.pclu = EXT4_B2C(sbi, pblk);
2920 				partial.state = nofree;
2921 			}
2922 		}
2923 	}
2924 	/*
2925 	 * We start scanning from right side, freeing all the blocks
2926 	 * after i_size and walking into the tree depth-wise.
2927 	 */
2928 	depth = ext_depth(inode);
2929 	if (path) {
2930 		int k = i = depth;
2931 		while (--k > 0)
2932 			path[k].p_block =
2933 				le16_to_cpu(path[k].p_hdr->eh_entries)+1;
2934 	} else {
2935 		path = kcalloc(depth + 1, sizeof(struct ext4_ext_path),
2936 			       GFP_NOFS | __GFP_NOFAIL);
2937 		if (path == NULL) {
2938 			ext4_journal_stop(handle);
2939 			return -ENOMEM;
2940 		}
2941 		path[0].p_maxdepth = path[0].p_depth = depth;
2942 		path[0].p_hdr = ext_inode_hdr(inode);
2943 		i = 0;
2944 
2945 		if (ext4_ext_check(inode, path[0].p_hdr, depth, 0)) {
2946 			err = -EFSCORRUPTED;
2947 			goto out;
2948 		}
2949 	}
2950 	err = 0;
2951 
2952 	while (i >= 0 && err == 0) {
2953 		if (i == depth) {
2954 			/* this is leaf block */
2955 			err = ext4_ext_rm_leaf(handle, inode, path,
2956 					       &partial, start, end);
2957 			/* root level has p_bh == NULL, brelse() eats this */
2958 			brelse(path[i].p_bh);
2959 			path[i].p_bh = NULL;
2960 			i--;
2961 			continue;
2962 		}
2963 
2964 		/* this is index block */
2965 		if (!path[i].p_hdr) {
2966 			ext_debug(inode, "initialize header\n");
2967 			path[i].p_hdr = ext_block_hdr(path[i].p_bh);
2968 		}
2969 
2970 		if (!path[i].p_idx) {
2971 			/* this level hasn't been touched yet */
2972 			path[i].p_idx = EXT_LAST_INDEX(path[i].p_hdr);
2973 			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries)+1;
2974 			ext_debug(inode, "init index ptr: hdr 0x%p, num %d\n",
2975 				  path[i].p_hdr,
2976 				  le16_to_cpu(path[i].p_hdr->eh_entries));
2977 		} else {
2978 			/* we were already here, see at next index */
2979 			path[i].p_idx--;
2980 		}
2981 
2982 		ext_debug(inode, "level %d - index, first 0x%p, cur 0x%p\n",
2983 				i, EXT_FIRST_INDEX(path[i].p_hdr),
2984 				path[i].p_idx);
2985 		if (ext4_ext_more_to_rm(path + i)) {
2986 			struct buffer_head *bh;
2987 			/* go to the next level */
2988 			ext_debug(inode, "move to level %d (block %llu)\n",
2989 				  i + 1, ext4_idx_pblock(path[i].p_idx));
2990 			memset(path + i + 1, 0, sizeof(*path));
2991 			bh = read_extent_tree_block(inode, path[i].p_idx,
2992 						    depth - i - 1,
2993 						    EXT4_EX_NOCACHE);
2994 			if (IS_ERR(bh)) {
2995 				/* should we reset i_size? */
2996 				err = PTR_ERR(bh);
2997 				break;
2998 			}
2999 			/* Yield here to deal with large extent trees.
3000 			 * Should be a no-op if we did IO above. */
3001 			cond_resched();
3002 			if (WARN_ON(i + 1 > depth)) {
3003 				err = -EFSCORRUPTED;
3004 				break;
3005 			}
3006 			path[i + 1].p_bh = bh;
3007 
3008 			/* save actual number of indexes since this
3009 			 * number is changed at the next iteration */
3010 			path[i].p_block = le16_to_cpu(path[i].p_hdr->eh_entries);
3011 			i++;
3012 		} else {
3013 			/* we finished processing this index, go up */
3014 			if (path[i].p_hdr->eh_entries == 0 && i > 0) {
3015 				/* index is empty, remove it;
3016 				 * handle must be already prepared by the
3017 				 * truncatei_leaf() */
3018 				err = ext4_ext_rm_idx(handle, inode, path, i);
3019 			}
3020 			/* root level has p_bh == NULL, brelse() eats this */
3021 			brelse(path[i].p_bh);
3022 			path[i].p_bh = NULL;
3023 			i--;
3024 			ext_debug(inode, "return to level %d\n", i);
3025 		}
3026 	}
3027 
3028 	trace_ext4_ext_remove_space_done(inode, start, end, depth, &partial,
3029 					 path->p_hdr->eh_entries);
3030 
3031 	/*
3032 	 * if there's a partial cluster and we have removed the first extent
3033 	 * in the file, then we also free the partial cluster, if any
3034 	 */
3035 	if (partial.state == tofree && err == 0) {
3036 		int flags = get_default_free_blocks_flags(inode);
3037 
3038 		if (ext4_is_pending(inode, partial.lblk))
3039 			flags |= EXT4_FREE_BLOCKS_RERESERVE_CLUSTER;
3040 		ext4_free_blocks(handle, inode, NULL,
3041 				 EXT4_C2B(sbi, partial.pclu),
3042 				 sbi->s_cluster_ratio, flags);
3043 		if (flags & EXT4_FREE_BLOCKS_RERESERVE_CLUSTER)
3044 			ext4_rereserve_cluster(inode, partial.lblk);
3045 		partial.state = initial;
3046 	}
3047 
3048 	/* TODO: flexible tree reduction should be here */
3049 	if (path->p_hdr->eh_entries == 0) {
3050 		/*
3051 		 * truncate to zero freed all the tree,
3052 		 * so we need to correct eh_depth
3053 		 */
3054 		err = ext4_ext_get_access(handle, inode, path);
3055 		if (err == 0) {
3056 			ext_inode_hdr(inode)->eh_depth = 0;
3057 			ext_inode_hdr(inode)->eh_max =
3058 				cpu_to_le16(ext4_ext_space_root(inode, 0));
3059 			err = ext4_ext_dirty(handle, inode, path);
3060 		}
3061 	}
3062 out:
3063 	ext4_ext_drop_refs(path);
3064 	kfree(path);
3065 	path = NULL;
3066 	if (err == -EAGAIN)
3067 		goto again;
3068 	ext4_journal_stop(handle);
3069 
3070 	return err;
3071 }
3072 
3073 /*
3074  * called at mount time
3075  */
3076 void ext4_ext_init(struct super_block *sb)
3077 {
3078 	/*
3079 	 * possible initialization would be here
3080 	 */
3081 
3082 	if (ext4_has_feature_extents(sb)) {
3083 #if defined(AGGRESSIVE_TEST) || defined(CHECK_BINSEARCH) || defined(EXTENTS_STATS)
3084 		printk(KERN_INFO "EXT4-fs: file extents enabled"
3085 #ifdef AGGRESSIVE_TEST
3086 		       ", aggressive tests"
3087 #endif
3088 #ifdef CHECK_BINSEARCH
3089 		       ", check binsearch"
3090 #endif
3091 #ifdef EXTENTS_STATS
3092 		       ", stats"
3093 #endif
3094 		       "\n");
3095 #endif
3096 #ifdef EXTENTS_STATS
3097 		spin_lock_init(&EXT4_SB(sb)->s_ext_stats_lock);
3098 		EXT4_SB(sb)->s_ext_min = 1 << 30;
3099 		EXT4_SB(sb)->s_ext_max = 0;
3100 #endif
3101 	}
3102 }
3103 
3104 /*
3105  * called at umount time
3106  */
3107 void ext4_ext_release(struct super_block *sb)
3108 {
3109 	if (!ext4_has_feature_extents(sb))
3110 		return;
3111 
3112 #ifdef EXTENTS_STATS
3113 	if (EXT4_SB(sb)->s_ext_blocks && EXT4_SB(sb)->s_ext_extents) {
3114 		struct ext4_sb_info *sbi = EXT4_SB(sb);
3115 		printk(KERN_ERR "EXT4-fs: %lu blocks in %lu extents (%lu ave)\n",
3116 			sbi->s_ext_blocks, sbi->s_ext_extents,
3117 			sbi->s_ext_blocks / sbi->s_ext_extents);
3118 		printk(KERN_ERR "EXT4-fs: extents: %lu min, %lu max, max depth %lu\n",
3119 			sbi->s_ext_min, sbi->s_ext_max, sbi->s_depth_max);
3120 	}
3121 #endif
3122 }
3123 
3124 static int ext4_zeroout_es(struct inode *inode, struct ext4_extent *ex)
3125 {
3126 	ext4_lblk_t  ee_block;
3127 	ext4_fsblk_t ee_pblock;
3128 	unsigned int ee_len;
3129 
3130 	ee_block  = le32_to_cpu(ex->ee_block);
3131 	ee_len    = ext4_ext_get_actual_len(ex);
3132 	ee_pblock = ext4_ext_pblock(ex);
3133 
3134 	if (ee_len == 0)
3135 		return 0;
3136 
3137 	return ext4_es_insert_extent(inode, ee_block, ee_len, ee_pblock,
3138 				     EXTENT_STATUS_WRITTEN);
3139 }
3140 
3141 /* FIXME!! we need to try to merge to left or right after zero-out  */
3142 static int ext4_ext_zeroout(struct inode *inode, struct ext4_extent *ex)
3143 {
3144 	ext4_fsblk_t ee_pblock;
3145 	unsigned int ee_len;
3146 
3147 	ee_len    = ext4_ext_get_actual_len(ex);
3148 	ee_pblock = ext4_ext_pblock(ex);
3149 	return ext4_issue_zeroout(inode, le32_to_cpu(ex->ee_block), ee_pblock,
3150 				  ee_len);
3151 }
3152 
3153 /*
3154  * ext4_split_extent_at() splits an extent at given block.
3155  *
3156  * @handle: the journal handle
3157  * @inode: the file inode
3158  * @path: the path to the extent
3159  * @split: the logical block where the extent is splitted.
3160  * @split_flags: indicates if the extent could be zeroout if split fails, and
3161  *		 the states(init or unwritten) of new extents.
3162  * @flags: flags used to insert new extent to extent tree.
3163  *
3164  *
3165  * Splits extent [a, b] into two extents [a, @split) and [@split, b], states
3166  * of which are determined by split_flag.
3167  *
3168  * There are two cases:
3169  *  a> the extent are splitted into two extent.
3170  *  b> split is not needed, and just mark the extent.
3171  *
3172  * return 0 on success.
3173  */
3174 static int ext4_split_extent_at(handle_t *handle,
3175 			     struct inode *inode,
3176 			     struct ext4_ext_path **ppath,
3177 			     ext4_lblk_t split,
3178 			     int split_flag,
3179 			     int flags)
3180 {
3181 	struct ext4_ext_path *path = *ppath;
3182 	ext4_fsblk_t newblock;
3183 	ext4_lblk_t ee_block;
3184 	struct ext4_extent *ex, newex, orig_ex, zero_ex;
3185 	struct ext4_extent *ex2 = NULL;
3186 	unsigned int ee_len, depth;
3187 	int err = 0;
3188 
3189 	BUG_ON((split_flag & (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2)) ==
3190 	       (EXT4_EXT_DATA_VALID1 | EXT4_EXT_DATA_VALID2));
3191 
3192 	ext_debug(inode, "logical block %llu\n", (unsigned long long)split);
3193 
3194 	ext4_ext_show_leaf(inode, path);
3195 
3196 	depth = ext_depth(inode);
3197 	ex = path[depth].p_ext;
3198 	ee_block = le32_to_cpu(ex->ee_block);
3199 	ee_len = ext4_ext_get_actual_len(ex);
3200 	newblock = split - ee_block + ext4_ext_pblock(ex);
3201 
3202 	BUG_ON(split < ee_block || split >= (ee_block + ee_len));
3203 	BUG_ON(!ext4_ext_is_unwritten(ex) &&
3204 	       split_flag & (EXT4_EXT_MAY_ZEROOUT |
3205 			     EXT4_EXT_MARK_UNWRIT1 |
3206 			     EXT4_EXT_MARK_UNWRIT2));
3207 
3208 	err = ext4_ext_get_access(handle, inode, path + depth);
3209 	if (err)
3210 		goto out;
3211 
3212 	if (split == ee_block) {
3213 		/*
3214 		 * case b: block @split is the block that the extent begins with
3215 		 * then we just change the state of the extent, and splitting
3216 		 * is not needed.
3217 		 */
3218 		if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3219 			ext4_ext_mark_unwritten(ex);
3220 		else
3221 			ext4_ext_mark_initialized(ex);
3222 
3223 		if (!(flags & EXT4_GET_BLOCKS_PRE_IO))
3224 			ext4_ext_try_to_merge(handle, inode, path, ex);
3225 
3226 		err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3227 		goto out;
3228 	}
3229 
3230 	/* case a */
3231 	memcpy(&orig_ex, ex, sizeof(orig_ex));
3232 	ex->ee_len = cpu_to_le16(split - ee_block);
3233 	if (split_flag & EXT4_EXT_MARK_UNWRIT1)
3234 		ext4_ext_mark_unwritten(ex);
3235 
3236 	/*
3237 	 * path may lead to new leaf, not to original leaf any more
3238 	 * after ext4_ext_insert_extent() returns,
3239 	 */
3240 	err = ext4_ext_dirty(handle, inode, path + depth);
3241 	if (err)
3242 		goto fix_extent_len;
3243 
3244 	ex2 = &newex;
3245 	ex2->ee_block = cpu_to_le32(split);
3246 	ex2->ee_len   = cpu_to_le16(ee_len - (split - ee_block));
3247 	ext4_ext_store_pblock(ex2, newblock);
3248 	if (split_flag & EXT4_EXT_MARK_UNWRIT2)
3249 		ext4_ext_mark_unwritten(ex2);
3250 
3251 	err = ext4_ext_insert_extent(handle, inode, ppath, &newex, flags);
3252 	if (err != -ENOSPC && err != -EDQUOT)
3253 		goto out;
3254 
3255 	if (EXT4_EXT_MAY_ZEROOUT & split_flag) {
3256 		if (split_flag & (EXT4_EXT_DATA_VALID1|EXT4_EXT_DATA_VALID2)) {
3257 			if (split_flag & EXT4_EXT_DATA_VALID1) {
3258 				err = ext4_ext_zeroout(inode, ex2);
3259 				zero_ex.ee_block = ex2->ee_block;
3260 				zero_ex.ee_len = cpu_to_le16(
3261 						ext4_ext_get_actual_len(ex2));
3262 				ext4_ext_store_pblock(&zero_ex,
3263 						      ext4_ext_pblock(ex2));
3264 			} else {
3265 				err = ext4_ext_zeroout(inode, ex);
3266 				zero_ex.ee_block = ex->ee_block;
3267 				zero_ex.ee_len = cpu_to_le16(
3268 						ext4_ext_get_actual_len(ex));
3269 				ext4_ext_store_pblock(&zero_ex,
3270 						      ext4_ext_pblock(ex));
3271 			}
3272 		} else {
3273 			err = ext4_ext_zeroout(inode, &orig_ex);
3274 			zero_ex.ee_block = orig_ex.ee_block;
3275 			zero_ex.ee_len = cpu_to_le16(
3276 						ext4_ext_get_actual_len(&orig_ex));
3277 			ext4_ext_store_pblock(&zero_ex,
3278 					      ext4_ext_pblock(&orig_ex));
3279 		}
3280 
3281 		if (!err) {
3282 			/* update the extent length and mark as initialized */
3283 			ex->ee_len = cpu_to_le16(ee_len);
3284 			ext4_ext_try_to_merge(handle, inode, path, ex);
3285 			err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3286 			if (!err)
3287 				/* update extent status tree */
3288 				err = ext4_zeroout_es(inode, &zero_ex);
3289 			/* If we failed at this point, we don't know in which
3290 			 * state the extent tree exactly is so don't try to fix
3291 			 * length of the original extent as it may do even more
3292 			 * damage.
3293 			 */
3294 			goto out;
3295 		}
3296 	}
3297 
3298 fix_extent_len:
3299 	ex->ee_len = orig_ex.ee_len;
3300 	/*
3301 	 * Ignore ext4_ext_dirty return value since we are already in error path
3302 	 * and err is a non-zero error code.
3303 	 */
3304 	ext4_ext_dirty(handle, inode, path + path->p_depth);
3305 	return err;
3306 out:
3307 	ext4_ext_show_leaf(inode, path);
3308 	return err;
3309 }
3310 
3311 /*
3312  * ext4_split_extents() splits an extent and mark extent which is covered
3313  * by @map as split_flags indicates
3314  *
3315  * It may result in splitting the extent into multiple extents (up to three)
3316  * There are three possibilities:
3317  *   a> There is no split required
3318  *   b> Splits in two extents: Split is happening at either end of the extent
3319  *   c> Splits in three extents: Somone is splitting in middle of the extent
3320  *
3321  */
3322 static int ext4_split_extent(handle_t *handle,
3323 			      struct inode *inode,
3324 			      struct ext4_ext_path **ppath,
3325 			      struct ext4_map_blocks *map,
3326 			      int split_flag,
3327 			      int flags)
3328 {
3329 	struct ext4_ext_path *path = *ppath;
3330 	ext4_lblk_t ee_block;
3331 	struct ext4_extent *ex;
3332 	unsigned int ee_len, depth;
3333 	int err = 0;
3334 	int unwritten;
3335 	int split_flag1, flags1;
3336 	int allocated = map->m_len;
3337 
3338 	depth = ext_depth(inode);
3339 	ex = path[depth].p_ext;
3340 	ee_block = le32_to_cpu(ex->ee_block);
3341 	ee_len = ext4_ext_get_actual_len(ex);
3342 	unwritten = ext4_ext_is_unwritten(ex);
3343 
3344 	if (map->m_lblk + map->m_len < ee_block + ee_len) {
3345 		split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT;
3346 		flags1 = flags | EXT4_GET_BLOCKS_PRE_IO;
3347 		if (unwritten)
3348 			split_flag1 |= EXT4_EXT_MARK_UNWRIT1 |
3349 				       EXT4_EXT_MARK_UNWRIT2;
3350 		if (split_flag & EXT4_EXT_DATA_VALID2)
3351 			split_flag1 |= EXT4_EXT_DATA_VALID1;
3352 		err = ext4_split_extent_at(handle, inode, ppath,
3353 				map->m_lblk + map->m_len, split_flag1, flags1);
3354 		if (err)
3355 			goto out;
3356 	} else {
3357 		allocated = ee_len - (map->m_lblk - ee_block);
3358 	}
3359 	/*
3360 	 * Update path is required because previous ext4_split_extent_at() may
3361 	 * result in split of original leaf or extent zeroout.
3362 	 */
3363 	path = ext4_find_extent(inode, map->m_lblk, ppath, flags);
3364 	if (IS_ERR(path))
3365 		return PTR_ERR(path);
3366 	depth = ext_depth(inode);
3367 	ex = path[depth].p_ext;
3368 	if (!ex) {
3369 		EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3370 				 (unsigned long) map->m_lblk);
3371 		return -EFSCORRUPTED;
3372 	}
3373 	unwritten = ext4_ext_is_unwritten(ex);
3374 	split_flag1 = 0;
3375 
3376 	if (map->m_lblk >= ee_block) {
3377 		split_flag1 = split_flag & EXT4_EXT_DATA_VALID2;
3378 		if (unwritten) {
3379 			split_flag1 |= EXT4_EXT_MARK_UNWRIT1;
3380 			split_flag1 |= split_flag & (EXT4_EXT_MAY_ZEROOUT |
3381 						     EXT4_EXT_MARK_UNWRIT2);
3382 		}
3383 		err = ext4_split_extent_at(handle, inode, ppath,
3384 				map->m_lblk, split_flag1, flags);
3385 		if (err)
3386 			goto out;
3387 	}
3388 
3389 	ext4_ext_show_leaf(inode, path);
3390 out:
3391 	return err ? err : allocated;
3392 }
3393 
3394 /*
3395  * This function is called by ext4_ext_map_blocks() if someone tries to write
3396  * to an unwritten extent. It may result in splitting the unwritten
3397  * extent into multiple extents (up to three - one initialized and two
3398  * unwritten).
3399  * There are three possibilities:
3400  *   a> There is no split required: Entire extent should be initialized
3401  *   b> Splits in two extents: Write is happening at either end of the extent
3402  *   c> Splits in three extents: Somone is writing in middle of the extent
3403  *
3404  * Pre-conditions:
3405  *  - The extent pointed to by 'path' is unwritten.
3406  *  - The extent pointed to by 'path' contains a superset
3407  *    of the logical span [map->m_lblk, map->m_lblk + map->m_len).
3408  *
3409  * Post-conditions on success:
3410  *  - the returned value is the number of blocks beyond map->l_lblk
3411  *    that are allocated and initialized.
3412  *    It is guaranteed to be >= map->m_len.
3413  */
3414 static int ext4_ext_convert_to_initialized(handle_t *handle,
3415 					   struct inode *inode,
3416 					   struct ext4_map_blocks *map,
3417 					   struct ext4_ext_path **ppath,
3418 					   int flags)
3419 {
3420 	struct ext4_ext_path *path = *ppath;
3421 	struct ext4_sb_info *sbi;
3422 	struct ext4_extent_header *eh;
3423 	struct ext4_map_blocks split_map;
3424 	struct ext4_extent zero_ex1, zero_ex2;
3425 	struct ext4_extent *ex, *abut_ex;
3426 	ext4_lblk_t ee_block, eof_block;
3427 	unsigned int ee_len, depth, map_len = map->m_len;
3428 	int allocated = 0, max_zeroout = 0;
3429 	int err = 0;
3430 	int split_flag = EXT4_EXT_DATA_VALID2;
3431 
3432 	ext_debug(inode, "logical block %llu, max_blocks %u\n",
3433 		  (unsigned long long)map->m_lblk, map_len);
3434 
3435 	sbi = EXT4_SB(inode->i_sb);
3436 	eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3437 			>> inode->i_sb->s_blocksize_bits;
3438 	if (eof_block < map->m_lblk + map_len)
3439 		eof_block = map->m_lblk + map_len;
3440 
3441 	depth = ext_depth(inode);
3442 	eh = path[depth].p_hdr;
3443 	ex = path[depth].p_ext;
3444 	ee_block = le32_to_cpu(ex->ee_block);
3445 	ee_len = ext4_ext_get_actual_len(ex);
3446 	zero_ex1.ee_len = 0;
3447 	zero_ex2.ee_len = 0;
3448 
3449 	trace_ext4_ext_convert_to_initialized_enter(inode, map, ex);
3450 
3451 	/* Pre-conditions */
3452 	BUG_ON(!ext4_ext_is_unwritten(ex));
3453 	BUG_ON(!in_range(map->m_lblk, ee_block, ee_len));
3454 
3455 	/*
3456 	 * Attempt to transfer newly initialized blocks from the currently
3457 	 * unwritten extent to its neighbor. This is much cheaper
3458 	 * than an insertion followed by a merge as those involve costly
3459 	 * memmove() calls. Transferring to the left is the common case in
3460 	 * steady state for workloads doing fallocate(FALLOC_FL_KEEP_SIZE)
3461 	 * followed by append writes.
3462 	 *
3463 	 * Limitations of the current logic:
3464 	 *  - L1: we do not deal with writes covering the whole extent.
3465 	 *    This would require removing the extent if the transfer
3466 	 *    is possible.
3467 	 *  - L2: we only attempt to merge with an extent stored in the
3468 	 *    same extent tree node.
3469 	 */
3470 	if ((map->m_lblk == ee_block) &&
3471 		/* See if we can merge left */
3472 		(map_len < ee_len) &&		/*L1*/
3473 		(ex > EXT_FIRST_EXTENT(eh))) {	/*L2*/
3474 		ext4_lblk_t prev_lblk;
3475 		ext4_fsblk_t prev_pblk, ee_pblk;
3476 		unsigned int prev_len;
3477 
3478 		abut_ex = ex - 1;
3479 		prev_lblk = le32_to_cpu(abut_ex->ee_block);
3480 		prev_len = ext4_ext_get_actual_len(abut_ex);
3481 		prev_pblk = ext4_ext_pblock(abut_ex);
3482 		ee_pblk = ext4_ext_pblock(ex);
3483 
3484 		/*
3485 		 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3486 		 * upon those conditions:
3487 		 * - C1: abut_ex is initialized,
3488 		 * - C2: abut_ex is logically abutting ex,
3489 		 * - C3: abut_ex is physically abutting ex,
3490 		 * - C4: abut_ex can receive the additional blocks without
3491 		 *   overflowing the (initialized) length limit.
3492 		 */
3493 		if ((!ext4_ext_is_unwritten(abut_ex)) &&		/*C1*/
3494 			((prev_lblk + prev_len) == ee_block) &&		/*C2*/
3495 			((prev_pblk + prev_len) == ee_pblk) &&		/*C3*/
3496 			(prev_len < (EXT_INIT_MAX_LEN - map_len))) {	/*C4*/
3497 			err = ext4_ext_get_access(handle, inode, path + depth);
3498 			if (err)
3499 				goto out;
3500 
3501 			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3502 				map, ex, abut_ex);
3503 
3504 			/* Shift the start of ex by 'map_len' blocks */
3505 			ex->ee_block = cpu_to_le32(ee_block + map_len);
3506 			ext4_ext_store_pblock(ex, ee_pblk + map_len);
3507 			ex->ee_len = cpu_to_le16(ee_len - map_len);
3508 			ext4_ext_mark_unwritten(ex); /* Restore the flag */
3509 
3510 			/* Extend abut_ex by 'map_len' blocks */
3511 			abut_ex->ee_len = cpu_to_le16(prev_len + map_len);
3512 
3513 			/* Result: number of initialized blocks past m_lblk */
3514 			allocated = map_len;
3515 		}
3516 	} else if (((map->m_lblk + map_len) == (ee_block + ee_len)) &&
3517 		   (map_len < ee_len) &&	/*L1*/
3518 		   ex < EXT_LAST_EXTENT(eh)) {	/*L2*/
3519 		/* See if we can merge right */
3520 		ext4_lblk_t next_lblk;
3521 		ext4_fsblk_t next_pblk, ee_pblk;
3522 		unsigned int next_len;
3523 
3524 		abut_ex = ex + 1;
3525 		next_lblk = le32_to_cpu(abut_ex->ee_block);
3526 		next_len = ext4_ext_get_actual_len(abut_ex);
3527 		next_pblk = ext4_ext_pblock(abut_ex);
3528 		ee_pblk = ext4_ext_pblock(ex);
3529 
3530 		/*
3531 		 * A transfer of blocks from 'ex' to 'abut_ex' is allowed
3532 		 * upon those conditions:
3533 		 * - C1: abut_ex is initialized,
3534 		 * - C2: abut_ex is logically abutting ex,
3535 		 * - C3: abut_ex is physically abutting ex,
3536 		 * - C4: abut_ex can receive the additional blocks without
3537 		 *   overflowing the (initialized) length limit.
3538 		 */
3539 		if ((!ext4_ext_is_unwritten(abut_ex)) &&		/*C1*/
3540 		    ((map->m_lblk + map_len) == next_lblk) &&		/*C2*/
3541 		    ((ee_pblk + ee_len) == next_pblk) &&		/*C3*/
3542 		    (next_len < (EXT_INIT_MAX_LEN - map_len))) {	/*C4*/
3543 			err = ext4_ext_get_access(handle, inode, path + depth);
3544 			if (err)
3545 				goto out;
3546 
3547 			trace_ext4_ext_convert_to_initialized_fastpath(inode,
3548 				map, ex, abut_ex);
3549 
3550 			/* Shift the start of abut_ex by 'map_len' blocks */
3551 			abut_ex->ee_block = cpu_to_le32(next_lblk - map_len);
3552 			ext4_ext_store_pblock(abut_ex, next_pblk - map_len);
3553 			ex->ee_len = cpu_to_le16(ee_len - map_len);
3554 			ext4_ext_mark_unwritten(ex); /* Restore the flag */
3555 
3556 			/* Extend abut_ex by 'map_len' blocks */
3557 			abut_ex->ee_len = cpu_to_le16(next_len + map_len);
3558 
3559 			/* Result: number of initialized blocks past m_lblk */
3560 			allocated = map_len;
3561 		}
3562 	}
3563 	if (allocated) {
3564 		/* Mark the block containing both extents as dirty */
3565 		err = ext4_ext_dirty(handle, inode, path + depth);
3566 
3567 		/* Update path to point to the right extent */
3568 		path[depth].p_ext = abut_ex;
3569 		goto out;
3570 	} else
3571 		allocated = ee_len - (map->m_lblk - ee_block);
3572 
3573 	WARN_ON(map->m_lblk < ee_block);
3574 	/*
3575 	 * It is safe to convert extent to initialized via explicit
3576 	 * zeroout only if extent is fully inside i_size or new_size.
3577 	 */
3578 	split_flag |= ee_block + ee_len <= eof_block ? EXT4_EXT_MAY_ZEROOUT : 0;
3579 
3580 	if (EXT4_EXT_MAY_ZEROOUT & split_flag)
3581 		max_zeroout = sbi->s_extent_max_zeroout_kb >>
3582 			(inode->i_sb->s_blocksize_bits - 10);
3583 
3584 	/*
3585 	 * five cases:
3586 	 * 1. split the extent into three extents.
3587 	 * 2. split the extent into two extents, zeroout the head of the first
3588 	 *    extent.
3589 	 * 3. split the extent into two extents, zeroout the tail of the second
3590 	 *    extent.
3591 	 * 4. split the extent into two extents with out zeroout.
3592 	 * 5. no splitting needed, just possibly zeroout the head and / or the
3593 	 *    tail of the extent.
3594 	 */
3595 	split_map.m_lblk = map->m_lblk;
3596 	split_map.m_len = map->m_len;
3597 
3598 	if (max_zeroout && (allocated > split_map.m_len)) {
3599 		if (allocated <= max_zeroout) {
3600 			/* case 3 or 5 */
3601 			zero_ex1.ee_block =
3602 				 cpu_to_le32(split_map.m_lblk +
3603 					     split_map.m_len);
3604 			zero_ex1.ee_len =
3605 				cpu_to_le16(allocated - split_map.m_len);
3606 			ext4_ext_store_pblock(&zero_ex1,
3607 				ext4_ext_pblock(ex) + split_map.m_lblk +
3608 				split_map.m_len - ee_block);
3609 			err = ext4_ext_zeroout(inode, &zero_ex1);
3610 			if (err)
3611 				goto fallback;
3612 			split_map.m_len = allocated;
3613 		}
3614 		if (split_map.m_lblk - ee_block + split_map.m_len <
3615 								max_zeroout) {
3616 			/* case 2 or 5 */
3617 			if (split_map.m_lblk != ee_block) {
3618 				zero_ex2.ee_block = ex->ee_block;
3619 				zero_ex2.ee_len = cpu_to_le16(split_map.m_lblk -
3620 							ee_block);
3621 				ext4_ext_store_pblock(&zero_ex2,
3622 						      ext4_ext_pblock(ex));
3623 				err = ext4_ext_zeroout(inode, &zero_ex2);
3624 				if (err)
3625 					goto fallback;
3626 			}
3627 
3628 			split_map.m_len += split_map.m_lblk - ee_block;
3629 			split_map.m_lblk = ee_block;
3630 			allocated = map->m_len;
3631 		}
3632 	}
3633 
3634 fallback:
3635 	err = ext4_split_extent(handle, inode, ppath, &split_map, split_flag,
3636 				flags);
3637 	if (err > 0)
3638 		err = 0;
3639 out:
3640 	/* If we have gotten a failure, don't zero out status tree */
3641 	if (!err) {
3642 		err = ext4_zeroout_es(inode, &zero_ex1);
3643 		if (!err)
3644 			err = ext4_zeroout_es(inode, &zero_ex2);
3645 	}
3646 	return err ? err : allocated;
3647 }
3648 
3649 /*
3650  * This function is called by ext4_ext_map_blocks() from
3651  * ext4_get_blocks_dio_write() when DIO to write
3652  * to an unwritten extent.
3653  *
3654  * Writing to an unwritten extent may result in splitting the unwritten
3655  * extent into multiple initialized/unwritten extents (up to three)
3656  * There are three possibilities:
3657  *   a> There is no split required: Entire extent should be unwritten
3658  *   b> Splits in two extents: Write is happening at either end of the extent
3659  *   c> Splits in three extents: Somone is writing in middle of the extent
3660  *
3661  * This works the same way in the case of initialized -> unwritten conversion.
3662  *
3663  * One of more index blocks maybe needed if the extent tree grow after
3664  * the unwritten extent split. To prevent ENOSPC occur at the IO
3665  * complete, we need to split the unwritten extent before DIO submit
3666  * the IO. The unwritten extent called at this time will be split
3667  * into three unwritten extent(at most). After IO complete, the part
3668  * being filled will be convert to initialized by the end_io callback function
3669  * via ext4_convert_unwritten_extents().
3670  *
3671  * Returns the size of unwritten extent to be written on success.
3672  */
3673 static int ext4_split_convert_extents(handle_t *handle,
3674 					struct inode *inode,
3675 					struct ext4_map_blocks *map,
3676 					struct ext4_ext_path **ppath,
3677 					int flags)
3678 {
3679 	struct ext4_ext_path *path = *ppath;
3680 	ext4_lblk_t eof_block;
3681 	ext4_lblk_t ee_block;
3682 	struct ext4_extent *ex;
3683 	unsigned int ee_len;
3684 	int split_flag = 0, depth;
3685 
3686 	ext_debug(inode, "logical block %llu, max_blocks %u\n",
3687 		  (unsigned long long)map->m_lblk, map->m_len);
3688 
3689 	eof_block = (EXT4_I(inode)->i_disksize + inode->i_sb->s_blocksize - 1)
3690 			>> inode->i_sb->s_blocksize_bits;
3691 	if (eof_block < map->m_lblk + map->m_len)
3692 		eof_block = map->m_lblk + map->m_len;
3693 	/*
3694 	 * It is safe to convert extent to initialized via explicit
3695 	 * zeroout only if extent is fully inside i_size or new_size.
3696 	 */
3697 	depth = ext_depth(inode);
3698 	ex = path[depth].p_ext;
3699 	ee_block = le32_to_cpu(ex->ee_block);
3700 	ee_len = ext4_ext_get_actual_len(ex);
3701 
3702 	/* Convert to unwritten */
3703 	if (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN) {
3704 		split_flag |= EXT4_EXT_DATA_VALID1;
3705 	/* Convert to initialized */
3706 	} else if (flags & EXT4_GET_BLOCKS_CONVERT) {
3707 		split_flag |= ee_block + ee_len <= eof_block ?
3708 			      EXT4_EXT_MAY_ZEROOUT : 0;
3709 		split_flag |= (EXT4_EXT_MARK_UNWRIT2 | EXT4_EXT_DATA_VALID2);
3710 	}
3711 	flags |= EXT4_GET_BLOCKS_PRE_IO;
3712 	return ext4_split_extent(handle, inode, ppath, map, split_flag, flags);
3713 }
3714 
3715 static int ext4_convert_unwritten_extents_endio(handle_t *handle,
3716 						struct inode *inode,
3717 						struct ext4_map_blocks *map,
3718 						struct ext4_ext_path **ppath)
3719 {
3720 	struct ext4_ext_path *path = *ppath;
3721 	struct ext4_extent *ex;
3722 	ext4_lblk_t ee_block;
3723 	unsigned int ee_len;
3724 	int depth;
3725 	int err = 0;
3726 
3727 	depth = ext_depth(inode);
3728 	ex = path[depth].p_ext;
3729 	ee_block = le32_to_cpu(ex->ee_block);
3730 	ee_len = ext4_ext_get_actual_len(ex);
3731 
3732 	ext_debug(inode, "logical block %llu, max_blocks %u\n",
3733 		  (unsigned long long)ee_block, ee_len);
3734 
3735 	/* If extent is larger than requested it is a clear sign that we still
3736 	 * have some extent state machine issues left. So extent_split is still
3737 	 * required.
3738 	 * TODO: Once all related issues will be fixed this situation should be
3739 	 * illegal.
3740 	 */
3741 	if (ee_block != map->m_lblk || ee_len > map->m_len) {
3742 #ifdef CONFIG_EXT4_DEBUG
3743 		ext4_warning(inode->i_sb, "Inode (%ld) finished: extent logical block %llu,"
3744 			     " len %u; IO logical block %llu, len %u",
3745 			     inode->i_ino, (unsigned long long)ee_block, ee_len,
3746 			     (unsigned long long)map->m_lblk, map->m_len);
3747 #endif
3748 		err = ext4_split_convert_extents(handle, inode, map, ppath,
3749 						 EXT4_GET_BLOCKS_CONVERT);
3750 		if (err < 0)
3751 			return err;
3752 		path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3753 		if (IS_ERR(path))
3754 			return PTR_ERR(path);
3755 		depth = ext_depth(inode);
3756 		ex = path[depth].p_ext;
3757 	}
3758 
3759 	err = ext4_ext_get_access(handle, inode, path + depth);
3760 	if (err)
3761 		goto out;
3762 	/* first mark the extent as initialized */
3763 	ext4_ext_mark_initialized(ex);
3764 
3765 	/* note: ext4_ext_correct_indexes() isn't needed here because
3766 	 * borders are not changed
3767 	 */
3768 	ext4_ext_try_to_merge(handle, inode, path, ex);
3769 
3770 	/* Mark modified extent as dirty */
3771 	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3772 out:
3773 	ext4_ext_show_leaf(inode, path);
3774 	return err;
3775 }
3776 
3777 static int
3778 convert_initialized_extent(handle_t *handle, struct inode *inode,
3779 			   struct ext4_map_blocks *map,
3780 			   struct ext4_ext_path **ppath,
3781 			   unsigned int *allocated)
3782 {
3783 	struct ext4_ext_path *path = *ppath;
3784 	struct ext4_extent *ex;
3785 	ext4_lblk_t ee_block;
3786 	unsigned int ee_len;
3787 	int depth;
3788 	int err = 0;
3789 
3790 	/*
3791 	 * Make sure that the extent is no bigger than we support with
3792 	 * unwritten extent
3793 	 */
3794 	if (map->m_len > EXT_UNWRITTEN_MAX_LEN)
3795 		map->m_len = EXT_UNWRITTEN_MAX_LEN / 2;
3796 
3797 	depth = ext_depth(inode);
3798 	ex = path[depth].p_ext;
3799 	ee_block = le32_to_cpu(ex->ee_block);
3800 	ee_len = ext4_ext_get_actual_len(ex);
3801 
3802 	ext_debug(inode, "logical block %llu, max_blocks %u\n",
3803 		  (unsigned long long)ee_block, ee_len);
3804 
3805 	if (ee_block != map->m_lblk || ee_len > map->m_len) {
3806 		err = ext4_split_convert_extents(handle, inode, map, ppath,
3807 				EXT4_GET_BLOCKS_CONVERT_UNWRITTEN);
3808 		if (err < 0)
3809 			return err;
3810 		path = ext4_find_extent(inode, map->m_lblk, ppath, 0);
3811 		if (IS_ERR(path))
3812 			return PTR_ERR(path);
3813 		depth = ext_depth(inode);
3814 		ex = path[depth].p_ext;
3815 		if (!ex) {
3816 			EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
3817 					 (unsigned long) map->m_lblk);
3818 			return -EFSCORRUPTED;
3819 		}
3820 	}
3821 
3822 	err = ext4_ext_get_access(handle, inode, path + depth);
3823 	if (err)
3824 		return err;
3825 	/* first mark the extent as unwritten */
3826 	ext4_ext_mark_unwritten(ex);
3827 
3828 	/* note: ext4_ext_correct_indexes() isn't needed here because
3829 	 * borders are not changed
3830 	 */
3831 	ext4_ext_try_to_merge(handle, inode, path, ex);
3832 
3833 	/* Mark modified extent as dirty */
3834 	err = ext4_ext_dirty(handle, inode, path + path->p_depth);
3835 	if (err)
3836 		return err;
3837 	ext4_ext_show_leaf(inode, path);
3838 
3839 	ext4_update_inode_fsync_trans(handle, inode, 1);
3840 
3841 	map->m_flags |= EXT4_MAP_UNWRITTEN;
3842 	if (*allocated > map->m_len)
3843 		*allocated = map->m_len;
3844 	map->m_len = *allocated;
3845 	return 0;
3846 }
3847 
3848 static int
3849 ext4_ext_handle_unwritten_extents(handle_t *handle, struct inode *inode,
3850 			struct ext4_map_blocks *map,
3851 			struct ext4_ext_path **ppath, int flags,
3852 			unsigned int allocated, ext4_fsblk_t newblock)
3853 {
3854 	struct ext4_ext_path __maybe_unused *path = *ppath;
3855 	int ret = 0;
3856 	int err = 0;
3857 
3858 	ext_debug(inode, "logical block %llu, max_blocks %u, flags 0x%x, allocated %u\n",
3859 		  (unsigned long long)map->m_lblk, map->m_len, flags,
3860 		  allocated);
3861 	ext4_ext_show_leaf(inode, path);
3862 
3863 	/*
3864 	 * When writing into unwritten space, we should not fail to
3865 	 * allocate metadata blocks for the new extent block if needed.
3866 	 */
3867 	flags |= EXT4_GET_BLOCKS_METADATA_NOFAIL;
3868 
3869 	trace_ext4_ext_handle_unwritten_extents(inode, map, flags,
3870 						    allocated, newblock);
3871 
3872 	/* get_block() before submitting IO, split the extent */
3873 	if (flags & EXT4_GET_BLOCKS_PRE_IO) {
3874 		ret = ext4_split_convert_extents(handle, inode, map, ppath,
3875 					 flags | EXT4_GET_BLOCKS_CONVERT);
3876 		if (ret < 0) {
3877 			err = ret;
3878 			goto out2;
3879 		}
3880 		/*
3881 		 * shouldn't get a 0 return when splitting an extent unless
3882 		 * m_len is 0 (bug) or extent has been corrupted
3883 		 */
3884 		if (unlikely(ret == 0)) {
3885 			EXT4_ERROR_INODE(inode,
3886 					 "unexpected ret == 0, m_len = %u",
3887 					 map->m_len);
3888 			err = -EFSCORRUPTED;
3889 			goto out2;
3890 		}
3891 		map->m_flags |= EXT4_MAP_UNWRITTEN;
3892 		goto out;
3893 	}
3894 	/* IO end_io complete, convert the filled extent to written */
3895 	if (flags & EXT4_GET_BLOCKS_CONVERT) {
3896 		err = ext4_convert_unwritten_extents_endio(handle, inode, map,
3897 							   ppath);
3898 		if (err < 0)
3899 			goto out2;
3900 		ext4_update_inode_fsync_trans(handle, inode, 1);
3901 		goto map_out;
3902 	}
3903 	/* buffered IO cases */
3904 	/*
3905 	 * repeat fallocate creation request
3906 	 * we already have an unwritten extent
3907 	 */
3908 	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
3909 		map->m_flags |= EXT4_MAP_UNWRITTEN;
3910 		goto map_out;
3911 	}
3912 
3913 	/* buffered READ or buffered write_begin() lookup */
3914 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
3915 		/*
3916 		 * We have blocks reserved already.  We
3917 		 * return allocated blocks so that delalloc
3918 		 * won't do block reservation for us.  But
3919 		 * the buffer head will be unmapped so that
3920 		 * a read from the block returns 0s.
3921 		 */
3922 		map->m_flags |= EXT4_MAP_UNWRITTEN;
3923 		goto out1;
3924 	}
3925 
3926 	/*
3927 	 * Default case when (flags & EXT4_GET_BLOCKS_CREATE) == 1.
3928 	 * For buffered writes, at writepage time, etc.  Convert a
3929 	 * discovered unwritten extent to written.
3930 	 */
3931 	ret = ext4_ext_convert_to_initialized(handle, inode, map, ppath, flags);
3932 	if (ret < 0) {
3933 		err = ret;
3934 		goto out2;
3935 	}
3936 	ext4_update_inode_fsync_trans(handle, inode, 1);
3937 	/*
3938 	 * shouldn't get a 0 return when converting an unwritten extent
3939 	 * unless m_len is 0 (bug) or extent has been corrupted
3940 	 */
3941 	if (unlikely(ret == 0)) {
3942 		EXT4_ERROR_INODE(inode, "unexpected ret == 0, m_len = %u",
3943 				 map->m_len);
3944 		err = -EFSCORRUPTED;
3945 		goto out2;
3946 	}
3947 
3948 out:
3949 	allocated = ret;
3950 	map->m_flags |= EXT4_MAP_NEW;
3951 map_out:
3952 	map->m_flags |= EXT4_MAP_MAPPED;
3953 out1:
3954 	map->m_pblk = newblock;
3955 	if (allocated > map->m_len)
3956 		allocated = map->m_len;
3957 	map->m_len = allocated;
3958 	ext4_ext_show_leaf(inode, path);
3959 out2:
3960 	return err ? err : allocated;
3961 }
3962 
3963 /*
3964  * get_implied_cluster_alloc - check to see if the requested
3965  * allocation (in the map structure) overlaps with a cluster already
3966  * allocated in an extent.
3967  *	@sb	The filesystem superblock structure
3968  *	@map	The requested lblk->pblk mapping
3969  *	@ex	The extent structure which might contain an implied
3970  *			cluster allocation
3971  *
3972  * This function is called by ext4_ext_map_blocks() after we failed to
3973  * find blocks that were already in the inode's extent tree.  Hence,
3974  * we know that the beginning of the requested region cannot overlap
3975  * the extent from the inode's extent tree.  There are three cases we
3976  * want to catch.  The first is this case:
3977  *
3978  *		 |--- cluster # N--|
3979  *    |--- extent ---|	|---- requested region ---|
3980  *			|==========|
3981  *
3982  * The second case that we need to test for is this one:
3983  *
3984  *   |--------- cluster # N ----------------|
3985  *	   |--- requested region --|   |------- extent ----|
3986  *	   |=======================|
3987  *
3988  * The third case is when the requested region lies between two extents
3989  * within the same cluster:
3990  *          |------------- cluster # N-------------|
3991  * |----- ex -----|                  |---- ex_right ----|
3992  *                  |------ requested region ------|
3993  *                  |================|
3994  *
3995  * In each of the above cases, we need to set the map->m_pblk and
3996  * map->m_len so it corresponds to the return the extent labelled as
3997  * "|====|" from cluster #N, since it is already in use for data in
3998  * cluster EXT4_B2C(sbi, map->m_lblk).	We will then return 1 to
3999  * signal to ext4_ext_map_blocks() that map->m_pblk should be treated
4000  * as a new "allocated" block region.  Otherwise, we will return 0 and
4001  * ext4_ext_map_blocks() will then allocate one or more new clusters
4002  * by calling ext4_mb_new_blocks().
4003  */
4004 static int get_implied_cluster_alloc(struct super_block *sb,
4005 				     struct ext4_map_blocks *map,
4006 				     struct ext4_extent *ex,
4007 				     struct ext4_ext_path *path)
4008 {
4009 	struct ext4_sb_info *sbi = EXT4_SB(sb);
4010 	ext4_lblk_t c_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4011 	ext4_lblk_t ex_cluster_start, ex_cluster_end;
4012 	ext4_lblk_t rr_cluster_start;
4013 	ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4014 	ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4015 	unsigned short ee_len = ext4_ext_get_actual_len(ex);
4016 
4017 	/* The extent passed in that we are trying to match */
4018 	ex_cluster_start = EXT4_B2C(sbi, ee_block);
4019 	ex_cluster_end = EXT4_B2C(sbi, ee_block + ee_len - 1);
4020 
4021 	/* The requested region passed into ext4_map_blocks() */
4022 	rr_cluster_start = EXT4_B2C(sbi, map->m_lblk);
4023 
4024 	if ((rr_cluster_start == ex_cluster_end) ||
4025 	    (rr_cluster_start == ex_cluster_start)) {
4026 		if (rr_cluster_start == ex_cluster_end)
4027 			ee_start += ee_len - 1;
4028 		map->m_pblk = EXT4_PBLK_CMASK(sbi, ee_start) + c_offset;
4029 		map->m_len = min(map->m_len,
4030 				 (unsigned) sbi->s_cluster_ratio - c_offset);
4031 		/*
4032 		 * Check for and handle this case:
4033 		 *
4034 		 *   |--------- cluster # N-------------|
4035 		 *		       |------- extent ----|
4036 		 *	   |--- requested region ---|
4037 		 *	   |===========|
4038 		 */
4039 
4040 		if (map->m_lblk < ee_block)
4041 			map->m_len = min(map->m_len, ee_block - map->m_lblk);
4042 
4043 		/*
4044 		 * Check for the case where there is already another allocated
4045 		 * block to the right of 'ex' but before the end of the cluster.
4046 		 *
4047 		 *          |------------- cluster # N-------------|
4048 		 * |----- ex -----|                  |---- ex_right ----|
4049 		 *                  |------ requested region ------|
4050 		 *                  |================|
4051 		 */
4052 		if (map->m_lblk > ee_block) {
4053 			ext4_lblk_t next = ext4_ext_next_allocated_block(path);
4054 			map->m_len = min(map->m_len, next - map->m_lblk);
4055 		}
4056 
4057 		trace_ext4_get_implied_cluster_alloc_exit(sb, map, 1);
4058 		return 1;
4059 	}
4060 
4061 	trace_ext4_get_implied_cluster_alloc_exit(sb, map, 0);
4062 	return 0;
4063 }
4064 
4065 
4066 /*
4067  * Block allocation/map/preallocation routine for extents based files
4068  *
4069  *
4070  * Need to be called with
4071  * down_read(&EXT4_I(inode)->i_data_sem) if not allocating file system block
4072  * (ie, create is zero). Otherwise down_write(&EXT4_I(inode)->i_data_sem)
4073  *
4074  * return > 0, number of blocks already mapped/allocated
4075  *          if create == 0 and these are pre-allocated blocks
4076  *          	buffer head is unmapped
4077  *          otherwise blocks are mapped
4078  *
4079  * return = 0, if plain look up failed (blocks have not been allocated)
4080  *          buffer head is unmapped
4081  *
4082  * return < 0, error case.
4083  */
4084 int ext4_ext_map_blocks(handle_t *handle, struct inode *inode,
4085 			struct ext4_map_blocks *map, int flags)
4086 {
4087 	struct ext4_ext_path *path = NULL;
4088 	struct ext4_extent newex, *ex, ex2;
4089 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4090 	ext4_fsblk_t newblock = 0, pblk;
4091 	int err = 0, depth, ret;
4092 	unsigned int allocated = 0, offset = 0;
4093 	unsigned int allocated_clusters = 0;
4094 	struct ext4_allocation_request ar;
4095 	ext4_lblk_t cluster_offset;
4096 
4097 	ext_debug(inode, "blocks %u/%u requested\n", map->m_lblk, map->m_len);
4098 	trace_ext4_ext_map_blocks_enter(inode, map->m_lblk, map->m_len, flags);
4099 
4100 	/* find extent for this block */
4101 	path = ext4_find_extent(inode, map->m_lblk, NULL, 0);
4102 	if (IS_ERR(path)) {
4103 		err = PTR_ERR(path);
4104 		path = NULL;
4105 		goto out;
4106 	}
4107 
4108 	depth = ext_depth(inode);
4109 
4110 	/*
4111 	 * consistent leaf must not be empty;
4112 	 * this situation is possible, though, _during_ tree modification;
4113 	 * this is why assert can't be put in ext4_find_extent()
4114 	 */
4115 	if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
4116 		EXT4_ERROR_INODE(inode, "bad extent address "
4117 				 "lblock: %lu, depth: %d pblock %lld",
4118 				 (unsigned long) map->m_lblk, depth,
4119 				 path[depth].p_block);
4120 		err = -EFSCORRUPTED;
4121 		goto out;
4122 	}
4123 
4124 	ex = path[depth].p_ext;
4125 	if (ex) {
4126 		ext4_lblk_t ee_block = le32_to_cpu(ex->ee_block);
4127 		ext4_fsblk_t ee_start = ext4_ext_pblock(ex);
4128 		unsigned short ee_len;
4129 
4130 
4131 		/*
4132 		 * unwritten extents are treated as holes, except that
4133 		 * we split out initialized portions during a write.
4134 		 */
4135 		ee_len = ext4_ext_get_actual_len(ex);
4136 
4137 		trace_ext4_ext_show_extent(inode, ee_block, ee_start, ee_len);
4138 
4139 		/* if found extent covers block, simply return it */
4140 		if (in_range(map->m_lblk, ee_block, ee_len)) {
4141 			newblock = map->m_lblk - ee_block + ee_start;
4142 			/* number of remaining blocks in the extent */
4143 			allocated = ee_len - (map->m_lblk - ee_block);
4144 			ext_debug(inode, "%u fit into %u:%d -> %llu\n",
4145 				  map->m_lblk, ee_block, ee_len, newblock);
4146 
4147 			/*
4148 			 * If the extent is initialized check whether the
4149 			 * caller wants to convert it to unwritten.
4150 			 */
4151 			if ((!ext4_ext_is_unwritten(ex)) &&
4152 			    (flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) {
4153 				err = convert_initialized_extent(handle,
4154 					inode, map, &path, &allocated);
4155 				goto out;
4156 			} else if (!ext4_ext_is_unwritten(ex)) {
4157 				map->m_flags |= EXT4_MAP_MAPPED;
4158 				map->m_pblk = newblock;
4159 				if (allocated > map->m_len)
4160 					allocated = map->m_len;
4161 				map->m_len = allocated;
4162 				ext4_ext_show_leaf(inode, path);
4163 				goto out;
4164 			}
4165 
4166 			ret = ext4_ext_handle_unwritten_extents(
4167 				handle, inode, map, &path, flags,
4168 				allocated, newblock);
4169 			if (ret < 0)
4170 				err = ret;
4171 			else
4172 				allocated = ret;
4173 			goto out;
4174 		}
4175 	}
4176 
4177 	/*
4178 	 * requested block isn't allocated yet;
4179 	 * we couldn't try to create block if create flag is zero
4180 	 */
4181 	if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) {
4182 		ext4_lblk_t hole_start, hole_len;
4183 
4184 		hole_start = map->m_lblk;
4185 		hole_len = ext4_ext_determine_hole(inode, path, &hole_start);
4186 		/*
4187 		 * put just found gap into cache to speed up
4188 		 * subsequent requests
4189 		 */
4190 		ext4_ext_put_gap_in_cache(inode, hole_start, hole_len);
4191 
4192 		/* Update hole_len to reflect hole size after map->m_lblk */
4193 		if (hole_start != map->m_lblk)
4194 			hole_len -= map->m_lblk - hole_start;
4195 		map->m_pblk = 0;
4196 		map->m_len = min_t(unsigned int, map->m_len, hole_len);
4197 
4198 		goto out;
4199 	}
4200 
4201 	/*
4202 	 * Okay, we need to do block allocation.
4203 	 */
4204 	newex.ee_block = cpu_to_le32(map->m_lblk);
4205 	cluster_offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4206 
4207 	/*
4208 	 * If we are doing bigalloc, check to see if the extent returned
4209 	 * by ext4_find_extent() implies a cluster we can use.
4210 	 */
4211 	if (cluster_offset && ex &&
4212 	    get_implied_cluster_alloc(inode->i_sb, map, ex, path)) {
4213 		ar.len = allocated = map->m_len;
4214 		newblock = map->m_pblk;
4215 		goto got_allocated_blocks;
4216 	}
4217 
4218 	/* find neighbour allocated blocks */
4219 	ar.lleft = map->m_lblk;
4220 	err = ext4_ext_search_left(inode, path, &ar.lleft, &ar.pleft);
4221 	if (err)
4222 		goto out;
4223 	ar.lright = map->m_lblk;
4224 	err = ext4_ext_search_right(inode, path, &ar.lright, &ar.pright, &ex2);
4225 	if (err < 0)
4226 		goto out;
4227 
4228 	/* Check if the extent after searching to the right implies a
4229 	 * cluster we can use. */
4230 	if ((sbi->s_cluster_ratio > 1) && err &&
4231 	    get_implied_cluster_alloc(inode->i_sb, map, &ex2, path)) {
4232 		ar.len = allocated = map->m_len;
4233 		newblock = map->m_pblk;
4234 		goto got_allocated_blocks;
4235 	}
4236 
4237 	/*
4238 	 * See if request is beyond maximum number of blocks we can have in
4239 	 * a single extent. For an initialized extent this limit is
4240 	 * EXT_INIT_MAX_LEN and for an unwritten extent this limit is
4241 	 * EXT_UNWRITTEN_MAX_LEN.
4242 	 */
4243 	if (map->m_len > EXT_INIT_MAX_LEN &&
4244 	    !(flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4245 		map->m_len = EXT_INIT_MAX_LEN;
4246 	else if (map->m_len > EXT_UNWRITTEN_MAX_LEN &&
4247 		 (flags & EXT4_GET_BLOCKS_UNWRIT_EXT))
4248 		map->m_len = EXT_UNWRITTEN_MAX_LEN;
4249 
4250 	/* Check if we can really insert (m_lblk)::(m_lblk + m_len) extent */
4251 	newex.ee_len = cpu_to_le16(map->m_len);
4252 	err = ext4_ext_check_overlap(sbi, inode, &newex, path);
4253 	if (err)
4254 		allocated = ext4_ext_get_actual_len(&newex);
4255 	else
4256 		allocated = map->m_len;
4257 
4258 	/* allocate new block */
4259 	ar.inode = inode;
4260 	ar.goal = ext4_ext_find_goal(inode, path, map->m_lblk);
4261 	ar.logical = map->m_lblk;
4262 	/*
4263 	 * We calculate the offset from the beginning of the cluster
4264 	 * for the logical block number, since when we allocate a
4265 	 * physical cluster, the physical block should start at the
4266 	 * same offset from the beginning of the cluster.  This is
4267 	 * needed so that future calls to get_implied_cluster_alloc()
4268 	 * work correctly.
4269 	 */
4270 	offset = EXT4_LBLK_COFF(sbi, map->m_lblk);
4271 	ar.len = EXT4_NUM_B2C(sbi, offset+allocated);
4272 	ar.goal -= offset;
4273 	ar.logical -= offset;
4274 	if (S_ISREG(inode->i_mode))
4275 		ar.flags = EXT4_MB_HINT_DATA;
4276 	else
4277 		/* disable in-core preallocation for non-regular files */
4278 		ar.flags = 0;
4279 	if (flags & EXT4_GET_BLOCKS_NO_NORMALIZE)
4280 		ar.flags |= EXT4_MB_HINT_NOPREALLOC;
4281 	if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4282 		ar.flags |= EXT4_MB_DELALLOC_RESERVED;
4283 	if (flags & EXT4_GET_BLOCKS_METADATA_NOFAIL)
4284 		ar.flags |= EXT4_MB_USE_RESERVED;
4285 	newblock = ext4_mb_new_blocks(handle, &ar, &err);
4286 	if (!newblock)
4287 		goto out;
4288 	allocated_clusters = ar.len;
4289 	ar.len = EXT4_C2B(sbi, ar.len) - offset;
4290 	ext_debug(inode, "allocate new block: goal %llu, found %llu/%u, requested %u\n",
4291 		  ar.goal, newblock, ar.len, allocated);
4292 	if (ar.len > allocated)
4293 		ar.len = allocated;
4294 
4295 got_allocated_blocks:
4296 	/* try to insert new extent into found leaf and return */
4297 	pblk = newblock + offset;
4298 	ext4_ext_store_pblock(&newex, pblk);
4299 	newex.ee_len = cpu_to_le16(ar.len);
4300 	/* Mark unwritten */
4301 	if (flags & EXT4_GET_BLOCKS_UNWRIT_EXT) {
4302 		ext4_ext_mark_unwritten(&newex);
4303 		map->m_flags |= EXT4_MAP_UNWRITTEN;
4304 	}
4305 
4306 	err = ext4_ext_insert_extent(handle, inode, &path, &newex, flags);
4307 	if (err) {
4308 		if (allocated_clusters) {
4309 			int fb_flags = 0;
4310 
4311 			/*
4312 			 * free data blocks we just allocated.
4313 			 * not a good idea to call discard here directly,
4314 			 * but otherwise we'd need to call it every free().
4315 			 */
4316 			ext4_discard_preallocations(inode, 0);
4317 			if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
4318 				fb_flags = EXT4_FREE_BLOCKS_NO_QUOT_UPDATE;
4319 			ext4_free_blocks(handle, inode, NULL, newblock,
4320 					 EXT4_C2B(sbi, allocated_clusters),
4321 					 fb_flags);
4322 		}
4323 		goto out;
4324 	}
4325 
4326 	/*
4327 	 * Reduce the reserved cluster count to reflect successful deferred
4328 	 * allocation of delayed allocated clusters or direct allocation of
4329 	 * clusters discovered to be delayed allocated.  Once allocated, a
4330 	 * cluster is not included in the reserved count.
4331 	 */
4332 	if (test_opt(inode->i_sb, DELALLOC) && allocated_clusters) {
4333 		if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) {
4334 			/*
4335 			 * When allocating delayed allocated clusters, simply
4336 			 * reduce the reserved cluster count and claim quota
4337 			 */
4338 			ext4_da_update_reserve_space(inode, allocated_clusters,
4339 							1);
4340 		} else {
4341 			ext4_lblk_t lblk, len;
4342 			unsigned int n;
4343 
4344 			/*
4345 			 * When allocating non-delayed allocated clusters
4346 			 * (from fallocate, filemap, DIO, or clusters
4347 			 * allocated when delalloc has been disabled by
4348 			 * ext4_nonda_switch), reduce the reserved cluster
4349 			 * count by the number of allocated clusters that
4350 			 * have previously been delayed allocated.  Quota
4351 			 * has been claimed by ext4_mb_new_blocks() above,
4352 			 * so release the quota reservations made for any
4353 			 * previously delayed allocated clusters.
4354 			 */
4355 			lblk = EXT4_LBLK_CMASK(sbi, map->m_lblk);
4356 			len = allocated_clusters << sbi->s_cluster_bits;
4357 			n = ext4_es_delayed_clu(inode, lblk, len);
4358 			if (n > 0)
4359 				ext4_da_update_reserve_space(inode, (int) n, 0);
4360 		}
4361 	}
4362 
4363 	/*
4364 	 * Cache the extent and update transaction to commit on fdatasync only
4365 	 * when it is _not_ an unwritten extent.
4366 	 */
4367 	if ((flags & EXT4_GET_BLOCKS_UNWRIT_EXT) == 0)
4368 		ext4_update_inode_fsync_trans(handle, inode, 1);
4369 	else
4370 		ext4_update_inode_fsync_trans(handle, inode, 0);
4371 
4372 	map->m_flags |= (EXT4_MAP_NEW | EXT4_MAP_MAPPED);
4373 	map->m_pblk = pblk;
4374 	map->m_len = ar.len;
4375 	allocated = map->m_len;
4376 	ext4_ext_show_leaf(inode, path);
4377 out:
4378 	ext4_ext_drop_refs(path);
4379 	kfree(path);
4380 
4381 	trace_ext4_ext_map_blocks_exit(inode, flags, map,
4382 				       err ? err : allocated);
4383 	return err ? err : allocated;
4384 }
4385 
4386 int ext4_ext_truncate(handle_t *handle, struct inode *inode)
4387 {
4388 	struct super_block *sb = inode->i_sb;
4389 	ext4_lblk_t last_block;
4390 	int err = 0;
4391 
4392 	/*
4393 	 * TODO: optimization is possible here.
4394 	 * Probably we need not scan at all,
4395 	 * because page truncation is enough.
4396 	 */
4397 
4398 	/* we have to know where to truncate from in crash case */
4399 	EXT4_I(inode)->i_disksize = inode->i_size;
4400 	err = ext4_mark_inode_dirty(handle, inode);
4401 	if (err)
4402 		return err;
4403 
4404 	last_block = (inode->i_size + sb->s_blocksize - 1)
4405 			>> EXT4_BLOCK_SIZE_BITS(sb);
4406 retry:
4407 	err = ext4_es_remove_extent(inode, last_block,
4408 				    EXT_MAX_BLOCKS - last_block);
4409 	if (err == -ENOMEM) {
4410 		cond_resched();
4411 		congestion_wait(BLK_RW_ASYNC, HZ/50);
4412 		goto retry;
4413 	}
4414 	if (err)
4415 		return err;
4416 retry_remove_space:
4417 	err = ext4_ext_remove_space(inode, last_block, EXT_MAX_BLOCKS - 1);
4418 	if (err == -ENOMEM) {
4419 		cond_resched();
4420 		congestion_wait(BLK_RW_ASYNC, HZ/50);
4421 		goto retry_remove_space;
4422 	}
4423 	return err;
4424 }
4425 
4426 static int ext4_alloc_file_blocks(struct file *file, ext4_lblk_t offset,
4427 				  ext4_lblk_t len, loff_t new_size,
4428 				  int flags)
4429 {
4430 	struct inode *inode = file_inode(file);
4431 	handle_t *handle;
4432 	int ret = 0, ret2 = 0, ret3 = 0;
4433 	int retries = 0;
4434 	int depth = 0;
4435 	struct ext4_map_blocks map;
4436 	unsigned int credits;
4437 	loff_t epos;
4438 
4439 	BUG_ON(!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS));
4440 	map.m_lblk = offset;
4441 	map.m_len = len;
4442 	/*
4443 	 * Don't normalize the request if it can fit in one extent so
4444 	 * that it doesn't get unnecessarily split into multiple
4445 	 * extents.
4446 	 */
4447 	if (len <= EXT_UNWRITTEN_MAX_LEN)
4448 		flags |= EXT4_GET_BLOCKS_NO_NORMALIZE;
4449 
4450 	/*
4451 	 * credits to insert 1 extent into extent tree
4452 	 */
4453 	credits = ext4_chunk_trans_blocks(inode, len);
4454 	depth = ext_depth(inode);
4455 
4456 retry:
4457 	while (len) {
4458 		/*
4459 		 * Recalculate credits when extent tree depth changes.
4460 		 */
4461 		if (depth != ext_depth(inode)) {
4462 			credits = ext4_chunk_trans_blocks(inode, len);
4463 			depth = ext_depth(inode);
4464 		}
4465 
4466 		handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4467 					    credits);
4468 		if (IS_ERR(handle)) {
4469 			ret = PTR_ERR(handle);
4470 			break;
4471 		}
4472 		ret = ext4_map_blocks(handle, inode, &map, flags);
4473 		if (ret <= 0) {
4474 			ext4_debug("inode #%lu: block %u: len %u: "
4475 				   "ext4_ext_map_blocks returned %d",
4476 				   inode->i_ino, map.m_lblk,
4477 				   map.m_len, ret);
4478 			ext4_mark_inode_dirty(handle, inode);
4479 			ext4_journal_stop(handle);
4480 			break;
4481 		}
4482 		/*
4483 		 * allow a full retry cycle for any remaining allocations
4484 		 */
4485 		retries = 0;
4486 		map.m_lblk += ret;
4487 		map.m_len = len = len - ret;
4488 		epos = (loff_t)map.m_lblk << inode->i_blkbits;
4489 		inode->i_ctime = current_time(inode);
4490 		if (new_size) {
4491 			if (epos > new_size)
4492 				epos = new_size;
4493 			if (ext4_update_inode_size(inode, epos) & 0x1)
4494 				inode->i_mtime = inode->i_ctime;
4495 		}
4496 		ret2 = ext4_mark_inode_dirty(handle, inode);
4497 		ext4_update_inode_fsync_trans(handle, inode, 1);
4498 		ret3 = ext4_journal_stop(handle);
4499 		ret2 = ret3 ? ret3 : ret2;
4500 		if (unlikely(ret2))
4501 			break;
4502 	}
4503 	if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
4504 		goto retry;
4505 
4506 	return ret > 0 ? ret2 : ret;
4507 }
4508 
4509 static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len);
4510 
4511 static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len);
4512 
4513 static long ext4_zero_range(struct file *file, loff_t offset,
4514 			    loff_t len, int mode)
4515 {
4516 	struct inode *inode = file_inode(file);
4517 	struct address_space *mapping = file->f_mapping;
4518 	handle_t *handle = NULL;
4519 	unsigned int max_blocks;
4520 	loff_t new_size = 0;
4521 	int ret = 0;
4522 	int flags;
4523 	int credits;
4524 	int partial_begin, partial_end;
4525 	loff_t start, end;
4526 	ext4_lblk_t lblk;
4527 	unsigned int blkbits = inode->i_blkbits;
4528 
4529 	trace_ext4_zero_range(inode, offset, len, mode);
4530 
4531 	/* Call ext4_force_commit to flush all data in case of data=journal. */
4532 	if (ext4_should_journal_data(inode)) {
4533 		ret = ext4_force_commit(inode->i_sb);
4534 		if (ret)
4535 			return ret;
4536 	}
4537 
4538 	/*
4539 	 * Round up offset. This is not fallocate, we need to zero out
4540 	 * blocks, so convert interior block aligned part of the range to
4541 	 * unwritten and possibly manually zero out unaligned parts of the
4542 	 * range.
4543 	 */
4544 	start = round_up(offset, 1 << blkbits);
4545 	end = round_down((offset + len), 1 << blkbits);
4546 
4547 	if (start < offset || end > offset + len)
4548 		return -EINVAL;
4549 	partial_begin = offset & ((1 << blkbits) - 1);
4550 	partial_end = (offset + len) & ((1 << blkbits) - 1);
4551 
4552 	lblk = start >> blkbits;
4553 	max_blocks = (end >> blkbits);
4554 	if (max_blocks < lblk)
4555 		max_blocks = 0;
4556 	else
4557 		max_blocks -= lblk;
4558 
4559 	inode_lock(inode);
4560 
4561 	/*
4562 	 * Indirect files do not support unwritten extents
4563 	 */
4564 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4565 		ret = -EOPNOTSUPP;
4566 		goto out_mutex;
4567 	}
4568 
4569 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4570 	    (offset + len > inode->i_size ||
4571 	     offset + len > EXT4_I(inode)->i_disksize)) {
4572 		new_size = offset + len;
4573 		ret = inode_newsize_ok(inode, new_size);
4574 		if (ret)
4575 			goto out_mutex;
4576 	}
4577 
4578 	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4579 
4580 	/* Wait all existing dio workers, newcomers will block on i_mutex */
4581 	inode_dio_wait(inode);
4582 
4583 	/* Preallocate the range including the unaligned edges */
4584 	if (partial_begin || partial_end) {
4585 		ret = ext4_alloc_file_blocks(file,
4586 				round_down(offset, 1 << blkbits) >> blkbits,
4587 				(round_up((offset + len), 1 << blkbits) -
4588 				 round_down(offset, 1 << blkbits)) >> blkbits,
4589 				new_size, flags);
4590 		if (ret)
4591 			goto out_mutex;
4592 
4593 	}
4594 
4595 	/* Zero range excluding the unaligned edges */
4596 	if (max_blocks > 0) {
4597 		flags |= (EXT4_GET_BLOCKS_CONVERT_UNWRITTEN |
4598 			  EXT4_EX_NOCACHE);
4599 
4600 		/*
4601 		 * Prevent page faults from reinstantiating pages we have
4602 		 * released from page cache.
4603 		 */
4604 		filemap_invalidate_lock(mapping);
4605 
4606 		ret = ext4_break_layouts(inode);
4607 		if (ret) {
4608 			filemap_invalidate_unlock(mapping);
4609 			goto out_mutex;
4610 		}
4611 
4612 		ret = ext4_update_disksize_before_punch(inode, offset, len);
4613 		if (ret) {
4614 			filemap_invalidate_unlock(mapping);
4615 			goto out_mutex;
4616 		}
4617 		/* Now release the pages and zero block aligned part of pages */
4618 		truncate_pagecache_range(inode, start, end - 1);
4619 		inode->i_mtime = inode->i_ctime = current_time(inode);
4620 
4621 		ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size,
4622 					     flags);
4623 		filemap_invalidate_unlock(mapping);
4624 		if (ret)
4625 			goto out_mutex;
4626 	}
4627 	if (!partial_begin && !partial_end)
4628 		goto out_mutex;
4629 
4630 	/*
4631 	 * In worst case we have to writeout two nonadjacent unwritten
4632 	 * blocks and update the inode
4633 	 */
4634 	credits = (2 * ext4_ext_index_trans_blocks(inode, 2)) + 1;
4635 	if (ext4_should_journal_data(inode))
4636 		credits += 2;
4637 	handle = ext4_journal_start(inode, EXT4_HT_MISC, credits);
4638 	if (IS_ERR(handle)) {
4639 		ret = PTR_ERR(handle);
4640 		ext4_std_error(inode->i_sb, ret);
4641 		goto out_mutex;
4642 	}
4643 
4644 	inode->i_mtime = inode->i_ctime = current_time(inode);
4645 	if (new_size)
4646 		ext4_update_inode_size(inode, new_size);
4647 	ret = ext4_mark_inode_dirty(handle, inode);
4648 	if (unlikely(ret))
4649 		goto out_handle;
4650 	ext4_fc_track_range(handle, inode, offset >> inode->i_sb->s_blocksize_bits,
4651 			(offset + len - 1) >> inode->i_sb->s_blocksize_bits);
4652 	/* Zero out partial block at the edges of the range */
4653 	ret = ext4_zero_partial_blocks(handle, inode, offset, len);
4654 	if (ret >= 0)
4655 		ext4_update_inode_fsync_trans(handle, inode, 1);
4656 
4657 	if (file->f_flags & O_SYNC)
4658 		ext4_handle_sync(handle);
4659 
4660 out_handle:
4661 	ext4_journal_stop(handle);
4662 out_mutex:
4663 	inode_unlock(inode);
4664 	return ret;
4665 }
4666 
4667 /*
4668  * preallocate space for a file. This implements ext4's fallocate file
4669  * operation, which gets called from sys_fallocate system call.
4670  * For block-mapped files, posix_fallocate should fall back to the method
4671  * of writing zeroes to the required new blocks (the same behavior which is
4672  * expected for file systems which do not support fallocate() system call).
4673  */
4674 long ext4_fallocate(struct file *file, int mode, loff_t offset, loff_t len)
4675 {
4676 	struct inode *inode = file_inode(file);
4677 	loff_t new_size = 0;
4678 	unsigned int max_blocks;
4679 	int ret = 0;
4680 	int flags;
4681 	ext4_lblk_t lblk;
4682 	unsigned int blkbits = inode->i_blkbits;
4683 
4684 	/*
4685 	 * Encrypted inodes can't handle collapse range or insert
4686 	 * range since we would need to re-encrypt blocks with a
4687 	 * different IV or XTS tweak (which are based on the logical
4688 	 * block number).
4689 	 */
4690 	if (IS_ENCRYPTED(inode) &&
4691 	    (mode & (FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_INSERT_RANGE)))
4692 		return -EOPNOTSUPP;
4693 
4694 	/* Return error if mode is not supported */
4695 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |
4696 		     FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE |
4697 		     FALLOC_FL_INSERT_RANGE))
4698 		return -EOPNOTSUPP;
4699 
4700 	ext4_fc_start_update(inode);
4701 
4702 	if (mode & FALLOC_FL_PUNCH_HOLE) {
4703 		ret = ext4_punch_hole(inode, offset, len);
4704 		goto exit;
4705 	}
4706 
4707 	ret = ext4_convert_inline_data(inode);
4708 	if (ret)
4709 		goto exit;
4710 
4711 	if (mode & FALLOC_FL_COLLAPSE_RANGE) {
4712 		ret = ext4_collapse_range(inode, offset, len);
4713 		goto exit;
4714 	}
4715 
4716 	if (mode & FALLOC_FL_INSERT_RANGE) {
4717 		ret = ext4_insert_range(inode, offset, len);
4718 		goto exit;
4719 	}
4720 
4721 	if (mode & FALLOC_FL_ZERO_RANGE) {
4722 		ret = ext4_zero_range(file, offset, len, mode);
4723 		goto exit;
4724 	}
4725 	trace_ext4_fallocate_enter(inode, offset, len, mode);
4726 	lblk = offset >> blkbits;
4727 
4728 	max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4729 	flags = EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT;
4730 
4731 	inode_lock(inode);
4732 
4733 	/*
4734 	 * We only support preallocation for extent-based files only
4735 	 */
4736 	if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
4737 		ret = -EOPNOTSUPP;
4738 		goto out;
4739 	}
4740 
4741 	if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4742 	    (offset + len > inode->i_size ||
4743 	     offset + len > EXT4_I(inode)->i_disksize)) {
4744 		new_size = offset + len;
4745 		ret = inode_newsize_ok(inode, new_size);
4746 		if (ret)
4747 			goto out;
4748 	}
4749 
4750 	/* Wait all existing dio workers, newcomers will block on i_mutex */
4751 	inode_dio_wait(inode);
4752 
4753 	ret = ext4_alloc_file_blocks(file, lblk, max_blocks, new_size, flags);
4754 	if (ret)
4755 		goto out;
4756 
4757 	if (file->f_flags & O_SYNC && EXT4_SB(inode->i_sb)->s_journal) {
4758 		ret = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
4759 					EXT4_I(inode)->i_sync_tid);
4760 	}
4761 out:
4762 	inode_unlock(inode);
4763 	trace_ext4_fallocate_exit(inode, offset, max_blocks, ret);
4764 exit:
4765 	ext4_fc_stop_update(inode);
4766 	return ret;
4767 }
4768 
4769 /*
4770  * This function convert a range of blocks to written extents
4771  * The caller of this function will pass the start offset and the size.
4772  * all unwritten extents within this range will be converted to
4773  * written extents.
4774  *
4775  * This function is called from the direct IO end io call back
4776  * function, to convert the fallocated extents after IO is completed.
4777  * Returns 0 on success.
4778  */
4779 int ext4_convert_unwritten_extents(handle_t *handle, struct inode *inode,
4780 				   loff_t offset, ssize_t len)
4781 {
4782 	unsigned int max_blocks;
4783 	int ret = 0, ret2 = 0, ret3 = 0;
4784 	struct ext4_map_blocks map;
4785 	unsigned int blkbits = inode->i_blkbits;
4786 	unsigned int credits = 0;
4787 
4788 	map.m_lblk = offset >> blkbits;
4789 	max_blocks = EXT4_MAX_BLOCKS(len, offset, blkbits);
4790 
4791 	if (!handle) {
4792 		/*
4793 		 * credits to insert 1 extent into extent tree
4794 		 */
4795 		credits = ext4_chunk_trans_blocks(inode, max_blocks);
4796 	}
4797 	while (ret >= 0 && ret < max_blocks) {
4798 		map.m_lblk += ret;
4799 		map.m_len = (max_blocks -= ret);
4800 		if (credits) {
4801 			handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
4802 						    credits);
4803 			if (IS_ERR(handle)) {
4804 				ret = PTR_ERR(handle);
4805 				break;
4806 			}
4807 		}
4808 		ret = ext4_map_blocks(handle, inode, &map,
4809 				      EXT4_GET_BLOCKS_IO_CONVERT_EXT);
4810 		if (ret <= 0)
4811 			ext4_warning(inode->i_sb,
4812 				     "inode #%lu: block %u: len %u: "
4813 				     "ext4_ext_map_blocks returned %d",
4814 				     inode->i_ino, map.m_lblk,
4815 				     map.m_len, ret);
4816 		ret2 = ext4_mark_inode_dirty(handle, inode);
4817 		if (credits) {
4818 			ret3 = ext4_journal_stop(handle);
4819 			if (unlikely(ret3))
4820 				ret2 = ret3;
4821 		}
4822 
4823 		if (ret <= 0 || ret2)
4824 			break;
4825 	}
4826 	return ret > 0 ? ret2 : ret;
4827 }
4828 
4829 int ext4_convert_unwritten_io_end_vec(handle_t *handle, ext4_io_end_t *io_end)
4830 {
4831 	int ret = 0, err = 0;
4832 	struct ext4_io_end_vec *io_end_vec;
4833 
4834 	/*
4835 	 * This is somewhat ugly but the idea is clear: When transaction is
4836 	 * reserved, everything goes into it. Otherwise we rather start several
4837 	 * smaller transactions for conversion of each extent separately.
4838 	 */
4839 	if (handle) {
4840 		handle = ext4_journal_start_reserved(handle,
4841 						     EXT4_HT_EXT_CONVERT);
4842 		if (IS_ERR(handle))
4843 			return PTR_ERR(handle);
4844 	}
4845 
4846 	list_for_each_entry(io_end_vec, &io_end->list_vec, list) {
4847 		ret = ext4_convert_unwritten_extents(handle, io_end->inode,
4848 						     io_end_vec->offset,
4849 						     io_end_vec->size);
4850 		if (ret)
4851 			break;
4852 	}
4853 
4854 	if (handle)
4855 		err = ext4_journal_stop(handle);
4856 
4857 	return ret < 0 ? ret : err;
4858 }
4859 
4860 static int ext4_iomap_xattr_fiemap(struct inode *inode, struct iomap *iomap)
4861 {
4862 	__u64 physical = 0;
4863 	__u64 length = 0;
4864 	int blockbits = inode->i_sb->s_blocksize_bits;
4865 	int error = 0;
4866 	u16 iomap_type;
4867 
4868 	/* in-inode? */
4869 	if (ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4870 		struct ext4_iloc iloc;
4871 		int offset;	/* offset of xattr in inode */
4872 
4873 		error = ext4_get_inode_loc(inode, &iloc);
4874 		if (error)
4875 			return error;
4876 		physical = (__u64)iloc.bh->b_blocknr << blockbits;
4877 		offset = EXT4_GOOD_OLD_INODE_SIZE +
4878 				EXT4_I(inode)->i_extra_isize;
4879 		physical += offset;
4880 		length = EXT4_SB(inode->i_sb)->s_inode_size - offset;
4881 		brelse(iloc.bh);
4882 		iomap_type = IOMAP_INLINE;
4883 	} else if (EXT4_I(inode)->i_file_acl) { /* external block */
4884 		physical = (__u64)EXT4_I(inode)->i_file_acl << blockbits;
4885 		length = inode->i_sb->s_blocksize;
4886 		iomap_type = IOMAP_MAPPED;
4887 	} else {
4888 		/* no in-inode or external block for xattr, so return -ENOENT */
4889 		error = -ENOENT;
4890 		goto out;
4891 	}
4892 
4893 	iomap->addr = physical;
4894 	iomap->offset = 0;
4895 	iomap->length = length;
4896 	iomap->type = iomap_type;
4897 	iomap->flags = 0;
4898 out:
4899 	return error;
4900 }
4901 
4902 static int ext4_iomap_xattr_begin(struct inode *inode, loff_t offset,
4903 				  loff_t length, unsigned flags,
4904 				  struct iomap *iomap, struct iomap *srcmap)
4905 {
4906 	int error;
4907 
4908 	error = ext4_iomap_xattr_fiemap(inode, iomap);
4909 	if (error == 0 && (offset >= iomap->length))
4910 		error = -ENOENT;
4911 	return error;
4912 }
4913 
4914 static const struct iomap_ops ext4_iomap_xattr_ops = {
4915 	.iomap_begin		= ext4_iomap_xattr_begin,
4916 };
4917 
4918 static int ext4_fiemap_check_ranges(struct inode *inode, u64 start, u64 *len)
4919 {
4920 	u64 maxbytes;
4921 
4922 	if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4923 		maxbytes = inode->i_sb->s_maxbytes;
4924 	else
4925 		maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
4926 
4927 	if (*len == 0)
4928 		return -EINVAL;
4929 	if (start > maxbytes)
4930 		return -EFBIG;
4931 
4932 	/*
4933 	 * Shrink request scope to what the fs can actually handle.
4934 	 */
4935 	if (*len > maxbytes || (maxbytes - *len) < start)
4936 		*len = maxbytes - start;
4937 	return 0;
4938 }
4939 
4940 int ext4_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4941 		u64 start, u64 len)
4942 {
4943 	int error = 0;
4944 
4945 	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4946 		error = ext4_ext_precache(inode);
4947 		if (error)
4948 			return error;
4949 		fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4950 	}
4951 
4952 	/*
4953 	 * For bitmap files the maximum size limit could be smaller than
4954 	 * s_maxbytes, so check len here manually instead of just relying on the
4955 	 * generic check.
4956 	 */
4957 	error = ext4_fiemap_check_ranges(inode, start, &len);
4958 	if (error)
4959 		return error;
4960 
4961 	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
4962 		fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR;
4963 		return iomap_fiemap(inode, fieinfo, start, len,
4964 				    &ext4_iomap_xattr_ops);
4965 	}
4966 
4967 	return iomap_fiemap(inode, fieinfo, start, len, &ext4_iomap_report_ops);
4968 }
4969 
4970 int ext4_get_es_cache(struct inode *inode, struct fiemap_extent_info *fieinfo,
4971 		      __u64 start, __u64 len)
4972 {
4973 	ext4_lblk_t start_blk, len_blks;
4974 	__u64 last_blk;
4975 	int error = 0;
4976 
4977 	if (ext4_has_inline_data(inode)) {
4978 		int has_inline;
4979 
4980 		down_read(&EXT4_I(inode)->xattr_sem);
4981 		has_inline = ext4_has_inline_data(inode);
4982 		up_read(&EXT4_I(inode)->xattr_sem);
4983 		if (has_inline)
4984 			return 0;
4985 	}
4986 
4987 	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
4988 		error = ext4_ext_precache(inode);
4989 		if (error)
4990 			return error;
4991 		fieinfo->fi_flags &= ~FIEMAP_FLAG_CACHE;
4992 	}
4993 
4994 	error = fiemap_prep(inode, fieinfo, start, &len, 0);
4995 	if (error)
4996 		return error;
4997 
4998 	error = ext4_fiemap_check_ranges(inode, start, &len);
4999 	if (error)
5000 		return error;
5001 
5002 	start_blk = start >> inode->i_sb->s_blocksize_bits;
5003 	last_blk = (start + len - 1) >> inode->i_sb->s_blocksize_bits;
5004 	if (last_blk >= EXT_MAX_BLOCKS)
5005 		last_blk = EXT_MAX_BLOCKS-1;
5006 	len_blks = ((ext4_lblk_t) last_blk) - start_blk + 1;
5007 
5008 	/*
5009 	 * Walk the extent tree gathering extent information
5010 	 * and pushing extents back to the user.
5011 	 */
5012 	return ext4_fill_es_cache_info(inode, start_blk, len_blks, fieinfo);
5013 }
5014 
5015 /*
5016  * ext4_ext_shift_path_extents:
5017  * Shift the extents of a path structure lying between path[depth].p_ext
5018  * and EXT_LAST_EXTENT(path[depth].p_hdr), by @shift blocks. @SHIFT tells
5019  * if it is right shift or left shift operation.
5020  */
5021 static int
5022 ext4_ext_shift_path_extents(struct ext4_ext_path *path, ext4_lblk_t shift,
5023 			    struct inode *inode, handle_t *handle,
5024 			    enum SHIFT_DIRECTION SHIFT)
5025 {
5026 	int depth, err = 0;
5027 	struct ext4_extent *ex_start, *ex_last;
5028 	bool update = false;
5029 	int credits, restart_credits;
5030 	depth = path->p_depth;
5031 
5032 	while (depth >= 0) {
5033 		if (depth == path->p_depth) {
5034 			ex_start = path[depth].p_ext;
5035 			if (!ex_start)
5036 				return -EFSCORRUPTED;
5037 
5038 			ex_last = EXT_LAST_EXTENT(path[depth].p_hdr);
5039 			/* leaf + sb + inode */
5040 			credits = 3;
5041 			if (ex_start == EXT_FIRST_EXTENT(path[depth].p_hdr)) {
5042 				update = true;
5043 				/* extent tree + sb + inode */
5044 				credits = depth + 2;
5045 			}
5046 
5047 			restart_credits = ext4_writepage_trans_blocks(inode);
5048 			err = ext4_datasem_ensure_credits(handle, inode, credits,
5049 					restart_credits, 0);
5050 			if (err) {
5051 				if (err > 0)
5052 					err = -EAGAIN;
5053 				goto out;
5054 			}
5055 
5056 			err = ext4_ext_get_access(handle, inode, path + depth);
5057 			if (err)
5058 				goto out;
5059 
5060 			while (ex_start <= ex_last) {
5061 				if (SHIFT == SHIFT_LEFT) {
5062 					le32_add_cpu(&ex_start->ee_block,
5063 						-shift);
5064 					/* Try to merge to the left. */
5065 					if ((ex_start >
5066 					    EXT_FIRST_EXTENT(path[depth].p_hdr))
5067 					    &&
5068 					    ext4_ext_try_to_merge_right(inode,
5069 					    path, ex_start - 1))
5070 						ex_last--;
5071 					else
5072 						ex_start++;
5073 				} else {
5074 					le32_add_cpu(&ex_last->ee_block, shift);
5075 					ext4_ext_try_to_merge_right(inode, path,
5076 						ex_last);
5077 					ex_last--;
5078 				}
5079 			}
5080 			err = ext4_ext_dirty(handle, inode, path + depth);
5081 			if (err)
5082 				goto out;
5083 
5084 			if (--depth < 0 || !update)
5085 				break;
5086 		}
5087 
5088 		/* Update index too */
5089 		err = ext4_ext_get_access(handle, inode, path + depth);
5090 		if (err)
5091 			goto out;
5092 
5093 		if (SHIFT == SHIFT_LEFT)
5094 			le32_add_cpu(&path[depth].p_idx->ei_block, -shift);
5095 		else
5096 			le32_add_cpu(&path[depth].p_idx->ei_block, shift);
5097 		err = ext4_ext_dirty(handle, inode, path + depth);
5098 		if (err)
5099 			goto out;
5100 
5101 		/* we are done if current index is not a starting index */
5102 		if (path[depth].p_idx != EXT_FIRST_INDEX(path[depth].p_hdr))
5103 			break;
5104 
5105 		depth--;
5106 	}
5107 
5108 out:
5109 	return err;
5110 }
5111 
5112 /*
5113  * ext4_ext_shift_extents:
5114  * All the extents which lies in the range from @start to the last allocated
5115  * block for the @inode are shifted either towards left or right (depending
5116  * upon @SHIFT) by @shift blocks.
5117  * On success, 0 is returned, error otherwise.
5118  */
5119 static int
5120 ext4_ext_shift_extents(struct inode *inode, handle_t *handle,
5121 		       ext4_lblk_t start, ext4_lblk_t shift,
5122 		       enum SHIFT_DIRECTION SHIFT)
5123 {
5124 	struct ext4_ext_path *path;
5125 	int ret = 0, depth;
5126 	struct ext4_extent *extent;
5127 	ext4_lblk_t stop, *iterator, ex_start, ex_end;
5128 	ext4_lblk_t tmp = EXT_MAX_BLOCKS;
5129 
5130 	/* Let path point to the last extent */
5131 	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5132 				EXT4_EX_NOCACHE);
5133 	if (IS_ERR(path))
5134 		return PTR_ERR(path);
5135 
5136 	depth = path->p_depth;
5137 	extent = path[depth].p_ext;
5138 	if (!extent)
5139 		goto out;
5140 
5141 	stop = le32_to_cpu(extent->ee_block);
5142 
5143        /*
5144 	* For left shifts, make sure the hole on the left is big enough to
5145 	* accommodate the shift.  For right shifts, make sure the last extent
5146 	* won't be shifted beyond EXT_MAX_BLOCKS.
5147 	*/
5148 	if (SHIFT == SHIFT_LEFT) {
5149 		path = ext4_find_extent(inode, start - 1, &path,
5150 					EXT4_EX_NOCACHE);
5151 		if (IS_ERR(path))
5152 			return PTR_ERR(path);
5153 		depth = path->p_depth;
5154 		extent =  path[depth].p_ext;
5155 		if (extent) {
5156 			ex_start = le32_to_cpu(extent->ee_block);
5157 			ex_end = le32_to_cpu(extent->ee_block) +
5158 				ext4_ext_get_actual_len(extent);
5159 		} else {
5160 			ex_start = 0;
5161 			ex_end = 0;
5162 		}
5163 
5164 		if ((start == ex_start && shift > ex_start) ||
5165 		    (shift > start - ex_end)) {
5166 			ret = -EINVAL;
5167 			goto out;
5168 		}
5169 	} else {
5170 		if (shift > EXT_MAX_BLOCKS -
5171 		    (stop + ext4_ext_get_actual_len(extent))) {
5172 			ret = -EINVAL;
5173 			goto out;
5174 		}
5175 	}
5176 
5177 	/*
5178 	 * In case of left shift, iterator points to start and it is increased
5179 	 * till we reach stop. In case of right shift, iterator points to stop
5180 	 * and it is decreased till we reach start.
5181 	 */
5182 again:
5183 	if (SHIFT == SHIFT_LEFT)
5184 		iterator = &start;
5185 	else
5186 		iterator = &stop;
5187 
5188 	if (tmp != EXT_MAX_BLOCKS)
5189 		*iterator = tmp;
5190 
5191 	/*
5192 	 * Its safe to start updating extents.  Start and stop are unsigned, so
5193 	 * in case of right shift if extent with 0 block is reached, iterator
5194 	 * becomes NULL to indicate the end of the loop.
5195 	 */
5196 	while (iterator && start <= stop) {
5197 		path = ext4_find_extent(inode, *iterator, &path,
5198 					EXT4_EX_NOCACHE);
5199 		if (IS_ERR(path))
5200 			return PTR_ERR(path);
5201 		depth = path->p_depth;
5202 		extent = path[depth].p_ext;
5203 		if (!extent) {
5204 			EXT4_ERROR_INODE(inode, "unexpected hole at %lu",
5205 					 (unsigned long) *iterator);
5206 			return -EFSCORRUPTED;
5207 		}
5208 		if (SHIFT == SHIFT_LEFT && *iterator >
5209 		    le32_to_cpu(extent->ee_block)) {
5210 			/* Hole, move to the next extent */
5211 			if (extent < EXT_LAST_EXTENT(path[depth].p_hdr)) {
5212 				path[depth].p_ext++;
5213 			} else {
5214 				*iterator = ext4_ext_next_allocated_block(path);
5215 				continue;
5216 			}
5217 		}
5218 
5219 		tmp = *iterator;
5220 		if (SHIFT == SHIFT_LEFT) {
5221 			extent = EXT_LAST_EXTENT(path[depth].p_hdr);
5222 			*iterator = le32_to_cpu(extent->ee_block) +
5223 					ext4_ext_get_actual_len(extent);
5224 		} else {
5225 			extent = EXT_FIRST_EXTENT(path[depth].p_hdr);
5226 			if (le32_to_cpu(extent->ee_block) > 0)
5227 				*iterator = le32_to_cpu(extent->ee_block) - 1;
5228 			else
5229 				/* Beginning is reached, end of the loop */
5230 				iterator = NULL;
5231 			/* Update path extent in case we need to stop */
5232 			while (le32_to_cpu(extent->ee_block) < start)
5233 				extent++;
5234 			path[depth].p_ext = extent;
5235 		}
5236 		ret = ext4_ext_shift_path_extents(path, shift, inode,
5237 				handle, SHIFT);
5238 		/* iterator can be NULL which means we should break */
5239 		if (ret == -EAGAIN)
5240 			goto again;
5241 		if (ret)
5242 			break;
5243 	}
5244 out:
5245 	ext4_ext_drop_refs(path);
5246 	kfree(path);
5247 	return ret;
5248 }
5249 
5250 /*
5251  * ext4_collapse_range:
5252  * This implements the fallocate's collapse range functionality for ext4
5253  * Returns: 0 and non-zero on error.
5254  */
5255 static int ext4_collapse_range(struct inode *inode, loff_t offset, loff_t len)
5256 {
5257 	struct super_block *sb = inode->i_sb;
5258 	struct address_space *mapping = inode->i_mapping;
5259 	ext4_lblk_t punch_start, punch_stop;
5260 	handle_t *handle;
5261 	unsigned int credits;
5262 	loff_t new_size, ioffset;
5263 	int ret;
5264 
5265 	/*
5266 	 * We need to test this early because xfstests assumes that a
5267 	 * collapse range of (0, 1) will return EOPNOTSUPP if the file
5268 	 * system does not support collapse range.
5269 	 */
5270 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5271 		return -EOPNOTSUPP;
5272 
5273 	/* Collapse range works only on fs cluster size aligned regions. */
5274 	if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5275 		return -EINVAL;
5276 
5277 	trace_ext4_collapse_range(inode, offset, len);
5278 
5279 	punch_start = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5280 	punch_stop = (offset + len) >> EXT4_BLOCK_SIZE_BITS(sb);
5281 
5282 	/* Call ext4_force_commit to flush all data in case of data=journal. */
5283 	if (ext4_should_journal_data(inode)) {
5284 		ret = ext4_force_commit(inode->i_sb);
5285 		if (ret)
5286 			return ret;
5287 	}
5288 
5289 	inode_lock(inode);
5290 	/*
5291 	 * There is no need to overlap collapse range with EOF, in which case
5292 	 * it is effectively a truncate operation
5293 	 */
5294 	if (offset + len >= inode->i_size) {
5295 		ret = -EINVAL;
5296 		goto out_mutex;
5297 	}
5298 
5299 	/* Currently just for extent based files */
5300 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5301 		ret = -EOPNOTSUPP;
5302 		goto out_mutex;
5303 	}
5304 
5305 	/* Wait for existing dio to complete */
5306 	inode_dio_wait(inode);
5307 
5308 	/*
5309 	 * Prevent page faults from reinstantiating pages we have released from
5310 	 * page cache.
5311 	 */
5312 	filemap_invalidate_lock(mapping);
5313 
5314 	ret = ext4_break_layouts(inode);
5315 	if (ret)
5316 		goto out_mmap;
5317 
5318 	/*
5319 	 * Need to round down offset to be aligned with page size boundary
5320 	 * for page size > block size.
5321 	 */
5322 	ioffset = round_down(offset, PAGE_SIZE);
5323 	/*
5324 	 * Write tail of the last page before removed range since it will get
5325 	 * removed from the page cache below.
5326 	 */
5327 	ret = filemap_write_and_wait_range(mapping, ioffset, offset);
5328 	if (ret)
5329 		goto out_mmap;
5330 	/*
5331 	 * Write data that will be shifted to preserve them when discarding
5332 	 * page cache below. We are also protected from pages becoming dirty
5333 	 * by i_rwsem and invalidate_lock.
5334 	 */
5335 	ret = filemap_write_and_wait_range(mapping, offset + len,
5336 					   LLONG_MAX);
5337 	if (ret)
5338 		goto out_mmap;
5339 	truncate_pagecache(inode, ioffset);
5340 
5341 	credits = ext4_writepage_trans_blocks(inode);
5342 	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5343 	if (IS_ERR(handle)) {
5344 		ret = PTR_ERR(handle);
5345 		goto out_mmap;
5346 	}
5347 	ext4_fc_start_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE);
5348 
5349 	down_write(&EXT4_I(inode)->i_data_sem);
5350 	ext4_discard_preallocations(inode, 0);
5351 
5352 	ret = ext4_es_remove_extent(inode, punch_start,
5353 				    EXT_MAX_BLOCKS - punch_start);
5354 	if (ret) {
5355 		up_write(&EXT4_I(inode)->i_data_sem);
5356 		goto out_stop;
5357 	}
5358 
5359 	ret = ext4_ext_remove_space(inode, punch_start, punch_stop - 1);
5360 	if (ret) {
5361 		up_write(&EXT4_I(inode)->i_data_sem);
5362 		goto out_stop;
5363 	}
5364 	ext4_discard_preallocations(inode, 0);
5365 
5366 	ret = ext4_ext_shift_extents(inode, handle, punch_stop,
5367 				     punch_stop - punch_start, SHIFT_LEFT);
5368 	if (ret) {
5369 		up_write(&EXT4_I(inode)->i_data_sem);
5370 		goto out_stop;
5371 	}
5372 
5373 	new_size = inode->i_size - len;
5374 	i_size_write(inode, new_size);
5375 	EXT4_I(inode)->i_disksize = new_size;
5376 
5377 	up_write(&EXT4_I(inode)->i_data_sem);
5378 	if (IS_SYNC(inode))
5379 		ext4_handle_sync(handle);
5380 	inode->i_mtime = inode->i_ctime = current_time(inode);
5381 	ret = ext4_mark_inode_dirty(handle, inode);
5382 	ext4_update_inode_fsync_trans(handle, inode, 1);
5383 
5384 out_stop:
5385 	ext4_journal_stop(handle);
5386 	ext4_fc_stop_ineligible(sb);
5387 out_mmap:
5388 	filemap_invalidate_unlock(mapping);
5389 out_mutex:
5390 	inode_unlock(inode);
5391 	return ret;
5392 }
5393 
5394 /*
5395  * ext4_insert_range:
5396  * This function implements the FALLOC_FL_INSERT_RANGE flag of fallocate.
5397  * The data blocks starting from @offset to the EOF are shifted by @len
5398  * towards right to create a hole in the @inode. Inode size is increased
5399  * by len bytes.
5400  * Returns 0 on success, error otherwise.
5401  */
5402 static int ext4_insert_range(struct inode *inode, loff_t offset, loff_t len)
5403 {
5404 	struct super_block *sb = inode->i_sb;
5405 	struct address_space *mapping = inode->i_mapping;
5406 	handle_t *handle;
5407 	struct ext4_ext_path *path;
5408 	struct ext4_extent *extent;
5409 	ext4_lblk_t offset_lblk, len_lblk, ee_start_lblk = 0;
5410 	unsigned int credits, ee_len;
5411 	int ret = 0, depth, split_flag = 0;
5412 	loff_t ioffset;
5413 
5414 	/*
5415 	 * We need to test this early because xfstests assumes that an
5416 	 * insert range of (0, 1) will return EOPNOTSUPP if the file
5417 	 * system does not support insert range.
5418 	 */
5419 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5420 		return -EOPNOTSUPP;
5421 
5422 	/* Insert range works only on fs cluster size aligned regions. */
5423 	if (!IS_ALIGNED(offset | len, EXT4_CLUSTER_SIZE(sb)))
5424 		return -EINVAL;
5425 
5426 	trace_ext4_insert_range(inode, offset, len);
5427 
5428 	offset_lblk = offset >> EXT4_BLOCK_SIZE_BITS(sb);
5429 	len_lblk = len >> EXT4_BLOCK_SIZE_BITS(sb);
5430 
5431 	/* Call ext4_force_commit to flush all data in case of data=journal */
5432 	if (ext4_should_journal_data(inode)) {
5433 		ret = ext4_force_commit(inode->i_sb);
5434 		if (ret)
5435 			return ret;
5436 	}
5437 
5438 	inode_lock(inode);
5439 	/* Currently just for extent based files */
5440 	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
5441 		ret = -EOPNOTSUPP;
5442 		goto out_mutex;
5443 	}
5444 
5445 	/* Check whether the maximum file size would be exceeded */
5446 	if (len > inode->i_sb->s_maxbytes - inode->i_size) {
5447 		ret = -EFBIG;
5448 		goto out_mutex;
5449 	}
5450 
5451 	/* Offset must be less than i_size */
5452 	if (offset >= inode->i_size) {
5453 		ret = -EINVAL;
5454 		goto out_mutex;
5455 	}
5456 
5457 	/* Wait for existing dio to complete */
5458 	inode_dio_wait(inode);
5459 
5460 	/*
5461 	 * Prevent page faults from reinstantiating pages we have released from
5462 	 * page cache.
5463 	 */
5464 	filemap_invalidate_lock(mapping);
5465 
5466 	ret = ext4_break_layouts(inode);
5467 	if (ret)
5468 		goto out_mmap;
5469 
5470 	/*
5471 	 * Need to round down to align start offset to page size boundary
5472 	 * for page size > block size.
5473 	 */
5474 	ioffset = round_down(offset, PAGE_SIZE);
5475 	/* Write out all dirty pages */
5476 	ret = filemap_write_and_wait_range(inode->i_mapping, ioffset,
5477 			LLONG_MAX);
5478 	if (ret)
5479 		goto out_mmap;
5480 	truncate_pagecache(inode, ioffset);
5481 
5482 	credits = ext4_writepage_trans_blocks(inode);
5483 	handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
5484 	if (IS_ERR(handle)) {
5485 		ret = PTR_ERR(handle);
5486 		goto out_mmap;
5487 	}
5488 	ext4_fc_start_ineligible(sb, EXT4_FC_REASON_FALLOC_RANGE);
5489 
5490 	/* Expand file to avoid data loss if there is error while shifting */
5491 	inode->i_size += len;
5492 	EXT4_I(inode)->i_disksize += len;
5493 	inode->i_mtime = inode->i_ctime = current_time(inode);
5494 	ret = ext4_mark_inode_dirty(handle, inode);
5495 	if (ret)
5496 		goto out_stop;
5497 
5498 	down_write(&EXT4_I(inode)->i_data_sem);
5499 	ext4_discard_preallocations(inode, 0);
5500 
5501 	path = ext4_find_extent(inode, offset_lblk, NULL, 0);
5502 	if (IS_ERR(path)) {
5503 		up_write(&EXT4_I(inode)->i_data_sem);
5504 		goto out_stop;
5505 	}
5506 
5507 	depth = ext_depth(inode);
5508 	extent = path[depth].p_ext;
5509 	if (extent) {
5510 		ee_start_lblk = le32_to_cpu(extent->ee_block);
5511 		ee_len = ext4_ext_get_actual_len(extent);
5512 
5513 		/*
5514 		 * If offset_lblk is not the starting block of extent, split
5515 		 * the extent @offset_lblk
5516 		 */
5517 		if ((offset_lblk > ee_start_lblk) &&
5518 				(offset_lblk < (ee_start_lblk + ee_len))) {
5519 			if (ext4_ext_is_unwritten(extent))
5520 				split_flag = EXT4_EXT_MARK_UNWRIT1 |
5521 					EXT4_EXT_MARK_UNWRIT2;
5522 			ret = ext4_split_extent_at(handle, inode, &path,
5523 					offset_lblk, split_flag,
5524 					EXT4_EX_NOCACHE |
5525 					EXT4_GET_BLOCKS_PRE_IO |
5526 					EXT4_GET_BLOCKS_METADATA_NOFAIL);
5527 		}
5528 
5529 		ext4_ext_drop_refs(path);
5530 		kfree(path);
5531 		if (ret < 0) {
5532 			up_write(&EXT4_I(inode)->i_data_sem);
5533 			goto out_stop;
5534 		}
5535 	} else {
5536 		ext4_ext_drop_refs(path);
5537 		kfree(path);
5538 	}
5539 
5540 	ret = ext4_es_remove_extent(inode, offset_lblk,
5541 			EXT_MAX_BLOCKS - offset_lblk);
5542 	if (ret) {
5543 		up_write(&EXT4_I(inode)->i_data_sem);
5544 		goto out_stop;
5545 	}
5546 
5547 	/*
5548 	 * if offset_lblk lies in a hole which is at start of file, use
5549 	 * ee_start_lblk to shift extents
5550 	 */
5551 	ret = ext4_ext_shift_extents(inode, handle,
5552 		ee_start_lblk > offset_lblk ? ee_start_lblk : offset_lblk,
5553 		len_lblk, SHIFT_RIGHT);
5554 
5555 	up_write(&EXT4_I(inode)->i_data_sem);
5556 	if (IS_SYNC(inode))
5557 		ext4_handle_sync(handle);
5558 	if (ret >= 0)
5559 		ext4_update_inode_fsync_trans(handle, inode, 1);
5560 
5561 out_stop:
5562 	ext4_journal_stop(handle);
5563 	ext4_fc_stop_ineligible(sb);
5564 out_mmap:
5565 	filemap_invalidate_unlock(mapping);
5566 out_mutex:
5567 	inode_unlock(inode);
5568 	return ret;
5569 }
5570 
5571 /**
5572  * ext4_swap_extents() - Swap extents between two inodes
5573  * @handle: handle for this transaction
5574  * @inode1:	First inode
5575  * @inode2:	Second inode
5576  * @lblk1:	Start block for first inode
5577  * @lblk2:	Start block for second inode
5578  * @count:	Number of blocks to swap
5579  * @unwritten: Mark second inode's extents as unwritten after swap
5580  * @erp:	Pointer to save error value
5581  *
5582  * This helper routine does exactly what is promise "swap extents". All other
5583  * stuff such as page-cache locking consistency, bh mapping consistency or
5584  * extent's data copying must be performed by caller.
5585  * Locking:
5586  * 		i_mutex is held for both inodes
5587  * 		i_data_sem is locked for write for both inodes
5588  * Assumptions:
5589  *		All pages from requested range are locked for both inodes
5590  */
5591 int
5592 ext4_swap_extents(handle_t *handle, struct inode *inode1,
5593 		  struct inode *inode2, ext4_lblk_t lblk1, ext4_lblk_t lblk2,
5594 		  ext4_lblk_t count, int unwritten, int *erp)
5595 {
5596 	struct ext4_ext_path *path1 = NULL;
5597 	struct ext4_ext_path *path2 = NULL;
5598 	int replaced_count = 0;
5599 
5600 	BUG_ON(!rwsem_is_locked(&EXT4_I(inode1)->i_data_sem));
5601 	BUG_ON(!rwsem_is_locked(&EXT4_I(inode2)->i_data_sem));
5602 	BUG_ON(!inode_is_locked(inode1));
5603 	BUG_ON(!inode_is_locked(inode2));
5604 
5605 	*erp = ext4_es_remove_extent(inode1, lblk1, count);
5606 	if (unlikely(*erp))
5607 		return 0;
5608 	*erp = ext4_es_remove_extent(inode2, lblk2, count);
5609 	if (unlikely(*erp))
5610 		return 0;
5611 
5612 	while (count) {
5613 		struct ext4_extent *ex1, *ex2, tmp_ex;
5614 		ext4_lblk_t e1_blk, e2_blk;
5615 		int e1_len, e2_len, len;
5616 		int split = 0;
5617 
5618 		path1 = ext4_find_extent(inode1, lblk1, NULL, EXT4_EX_NOCACHE);
5619 		if (IS_ERR(path1)) {
5620 			*erp = PTR_ERR(path1);
5621 			path1 = NULL;
5622 		finish:
5623 			count = 0;
5624 			goto repeat;
5625 		}
5626 		path2 = ext4_find_extent(inode2, lblk2, NULL, EXT4_EX_NOCACHE);
5627 		if (IS_ERR(path2)) {
5628 			*erp = PTR_ERR(path2);
5629 			path2 = NULL;
5630 			goto finish;
5631 		}
5632 		ex1 = path1[path1->p_depth].p_ext;
5633 		ex2 = path2[path2->p_depth].p_ext;
5634 		/* Do we have something to swap ? */
5635 		if (unlikely(!ex2 || !ex1))
5636 			goto finish;
5637 
5638 		e1_blk = le32_to_cpu(ex1->ee_block);
5639 		e2_blk = le32_to_cpu(ex2->ee_block);
5640 		e1_len = ext4_ext_get_actual_len(ex1);
5641 		e2_len = ext4_ext_get_actual_len(ex2);
5642 
5643 		/* Hole handling */
5644 		if (!in_range(lblk1, e1_blk, e1_len) ||
5645 		    !in_range(lblk2, e2_blk, e2_len)) {
5646 			ext4_lblk_t next1, next2;
5647 
5648 			/* if hole after extent, then go to next extent */
5649 			next1 = ext4_ext_next_allocated_block(path1);
5650 			next2 = ext4_ext_next_allocated_block(path2);
5651 			/* If hole before extent, then shift to that extent */
5652 			if (e1_blk > lblk1)
5653 				next1 = e1_blk;
5654 			if (e2_blk > lblk2)
5655 				next2 = e2_blk;
5656 			/* Do we have something to swap */
5657 			if (next1 == EXT_MAX_BLOCKS || next2 == EXT_MAX_BLOCKS)
5658 				goto finish;
5659 			/* Move to the rightest boundary */
5660 			len = next1 - lblk1;
5661 			if (len < next2 - lblk2)
5662 				len = next2 - lblk2;
5663 			if (len > count)
5664 				len = count;
5665 			lblk1 += len;
5666 			lblk2 += len;
5667 			count -= len;
5668 			goto repeat;
5669 		}
5670 
5671 		/* Prepare left boundary */
5672 		if (e1_blk < lblk1) {
5673 			split = 1;
5674 			*erp = ext4_force_split_extent_at(handle, inode1,
5675 						&path1, lblk1, 0);
5676 			if (unlikely(*erp))
5677 				goto finish;
5678 		}
5679 		if (e2_blk < lblk2) {
5680 			split = 1;
5681 			*erp = ext4_force_split_extent_at(handle, inode2,
5682 						&path2,  lblk2, 0);
5683 			if (unlikely(*erp))
5684 				goto finish;
5685 		}
5686 		/* ext4_split_extent_at() may result in leaf extent split,
5687 		 * path must to be revalidated. */
5688 		if (split)
5689 			goto repeat;
5690 
5691 		/* Prepare right boundary */
5692 		len = count;
5693 		if (len > e1_blk + e1_len - lblk1)
5694 			len = e1_blk + e1_len - lblk1;
5695 		if (len > e2_blk + e2_len - lblk2)
5696 			len = e2_blk + e2_len - lblk2;
5697 
5698 		if (len != e1_len) {
5699 			split = 1;
5700 			*erp = ext4_force_split_extent_at(handle, inode1,
5701 						&path1, lblk1 + len, 0);
5702 			if (unlikely(*erp))
5703 				goto finish;
5704 		}
5705 		if (len != e2_len) {
5706 			split = 1;
5707 			*erp = ext4_force_split_extent_at(handle, inode2,
5708 						&path2, lblk2 + len, 0);
5709 			if (*erp)
5710 				goto finish;
5711 		}
5712 		/* ext4_split_extent_at() may result in leaf extent split,
5713 		 * path must to be revalidated. */
5714 		if (split)
5715 			goto repeat;
5716 
5717 		BUG_ON(e2_len != e1_len);
5718 		*erp = ext4_ext_get_access(handle, inode1, path1 + path1->p_depth);
5719 		if (unlikely(*erp))
5720 			goto finish;
5721 		*erp = ext4_ext_get_access(handle, inode2, path2 + path2->p_depth);
5722 		if (unlikely(*erp))
5723 			goto finish;
5724 
5725 		/* Both extents are fully inside boundaries. Swap it now */
5726 		tmp_ex = *ex1;
5727 		ext4_ext_store_pblock(ex1, ext4_ext_pblock(ex2));
5728 		ext4_ext_store_pblock(ex2, ext4_ext_pblock(&tmp_ex));
5729 		ex1->ee_len = cpu_to_le16(e2_len);
5730 		ex2->ee_len = cpu_to_le16(e1_len);
5731 		if (unwritten)
5732 			ext4_ext_mark_unwritten(ex2);
5733 		if (ext4_ext_is_unwritten(&tmp_ex))
5734 			ext4_ext_mark_unwritten(ex1);
5735 
5736 		ext4_ext_try_to_merge(handle, inode2, path2, ex2);
5737 		ext4_ext_try_to_merge(handle, inode1, path1, ex1);
5738 		*erp = ext4_ext_dirty(handle, inode2, path2 +
5739 				      path2->p_depth);
5740 		if (unlikely(*erp))
5741 			goto finish;
5742 		*erp = ext4_ext_dirty(handle, inode1, path1 +
5743 				      path1->p_depth);
5744 		/*
5745 		 * Looks scarry ah..? second inode already points to new blocks,
5746 		 * and it was successfully dirtied. But luckily error may happen
5747 		 * only due to journal error, so full transaction will be
5748 		 * aborted anyway.
5749 		 */
5750 		if (unlikely(*erp))
5751 			goto finish;
5752 		lblk1 += len;
5753 		lblk2 += len;
5754 		replaced_count += len;
5755 		count -= len;
5756 
5757 	repeat:
5758 		ext4_ext_drop_refs(path1);
5759 		kfree(path1);
5760 		ext4_ext_drop_refs(path2);
5761 		kfree(path2);
5762 		path1 = path2 = NULL;
5763 	}
5764 	return replaced_count;
5765 }
5766 
5767 /*
5768  * ext4_clu_mapped - determine whether any block in a logical cluster has
5769  *                   been mapped to a physical cluster
5770  *
5771  * @inode - file containing the logical cluster
5772  * @lclu - logical cluster of interest
5773  *
5774  * Returns 1 if any block in the logical cluster is mapped, signifying
5775  * that a physical cluster has been allocated for it.  Otherwise,
5776  * returns 0.  Can also return negative error codes.  Derived from
5777  * ext4_ext_map_blocks().
5778  */
5779 int ext4_clu_mapped(struct inode *inode, ext4_lblk_t lclu)
5780 {
5781 	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5782 	struct ext4_ext_path *path;
5783 	int depth, mapped = 0, err = 0;
5784 	struct ext4_extent *extent;
5785 	ext4_lblk_t first_lblk, first_lclu, last_lclu;
5786 
5787 	/* search for the extent closest to the first block in the cluster */
5788 	path = ext4_find_extent(inode, EXT4_C2B(sbi, lclu), NULL, 0);
5789 	if (IS_ERR(path)) {
5790 		err = PTR_ERR(path);
5791 		path = NULL;
5792 		goto out;
5793 	}
5794 
5795 	depth = ext_depth(inode);
5796 
5797 	/*
5798 	 * A consistent leaf must not be empty.  This situation is possible,
5799 	 * though, _during_ tree modification, and it's why an assert can't
5800 	 * be put in ext4_find_extent().
5801 	 */
5802 	if (unlikely(path[depth].p_ext == NULL && depth != 0)) {
5803 		EXT4_ERROR_INODE(inode,
5804 		    "bad extent address - lblock: %lu, depth: %d, pblock: %lld",
5805 				 (unsigned long) EXT4_C2B(sbi, lclu),
5806 				 depth, path[depth].p_block);
5807 		err = -EFSCORRUPTED;
5808 		goto out;
5809 	}
5810 
5811 	extent = path[depth].p_ext;
5812 
5813 	/* can't be mapped if the extent tree is empty */
5814 	if (extent == NULL)
5815 		goto out;
5816 
5817 	first_lblk = le32_to_cpu(extent->ee_block);
5818 	first_lclu = EXT4_B2C(sbi, first_lblk);
5819 
5820 	/*
5821 	 * Three possible outcomes at this point - found extent spanning
5822 	 * the target cluster, to the left of the target cluster, or to the
5823 	 * right of the target cluster.  The first two cases are handled here.
5824 	 * The last case indicates the target cluster is not mapped.
5825 	 */
5826 	if (lclu >= first_lclu) {
5827 		last_lclu = EXT4_B2C(sbi, first_lblk +
5828 				     ext4_ext_get_actual_len(extent) - 1);
5829 		if (lclu <= last_lclu) {
5830 			mapped = 1;
5831 		} else {
5832 			first_lblk = ext4_ext_next_allocated_block(path);
5833 			first_lclu = EXT4_B2C(sbi, first_lblk);
5834 			if (lclu == first_lclu)
5835 				mapped = 1;
5836 		}
5837 	}
5838 
5839 out:
5840 	ext4_ext_drop_refs(path);
5841 	kfree(path);
5842 
5843 	return err ? err : mapped;
5844 }
5845 
5846 /*
5847  * Updates physical block address and unwritten status of extent
5848  * starting at lblk start and of len. If such an extent doesn't exist,
5849  * this function splits the extent tree appropriately to create an
5850  * extent like this.  This function is called in the fast commit
5851  * replay path.  Returns 0 on success and error on failure.
5852  */
5853 int ext4_ext_replay_update_ex(struct inode *inode, ext4_lblk_t start,
5854 			      int len, int unwritten, ext4_fsblk_t pblk)
5855 {
5856 	struct ext4_ext_path *path = NULL, *ppath;
5857 	struct ext4_extent *ex;
5858 	int ret;
5859 
5860 	path = ext4_find_extent(inode, start, NULL, 0);
5861 	if (IS_ERR(path))
5862 		return PTR_ERR(path);
5863 	ex = path[path->p_depth].p_ext;
5864 	if (!ex) {
5865 		ret = -EFSCORRUPTED;
5866 		goto out;
5867 	}
5868 
5869 	if (le32_to_cpu(ex->ee_block) != start ||
5870 		ext4_ext_get_actual_len(ex) != len) {
5871 		/* We need to split this extent to match our extent first */
5872 		ppath = path;
5873 		down_write(&EXT4_I(inode)->i_data_sem);
5874 		ret = ext4_force_split_extent_at(NULL, inode, &ppath, start, 1);
5875 		up_write(&EXT4_I(inode)->i_data_sem);
5876 		if (ret)
5877 			goto out;
5878 		kfree(path);
5879 		path = ext4_find_extent(inode, start, NULL, 0);
5880 		if (IS_ERR(path))
5881 			return -1;
5882 		ppath = path;
5883 		ex = path[path->p_depth].p_ext;
5884 		WARN_ON(le32_to_cpu(ex->ee_block) != start);
5885 		if (ext4_ext_get_actual_len(ex) != len) {
5886 			down_write(&EXT4_I(inode)->i_data_sem);
5887 			ret = ext4_force_split_extent_at(NULL, inode, &ppath,
5888 							 start + len, 1);
5889 			up_write(&EXT4_I(inode)->i_data_sem);
5890 			if (ret)
5891 				goto out;
5892 			kfree(path);
5893 			path = ext4_find_extent(inode, start, NULL, 0);
5894 			if (IS_ERR(path))
5895 				return -EINVAL;
5896 			ex = path[path->p_depth].p_ext;
5897 		}
5898 	}
5899 	if (unwritten)
5900 		ext4_ext_mark_unwritten(ex);
5901 	else
5902 		ext4_ext_mark_initialized(ex);
5903 	ext4_ext_store_pblock(ex, pblk);
5904 	down_write(&EXT4_I(inode)->i_data_sem);
5905 	ret = ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5906 	up_write(&EXT4_I(inode)->i_data_sem);
5907 out:
5908 	ext4_ext_drop_refs(path);
5909 	kfree(path);
5910 	ext4_mark_inode_dirty(NULL, inode);
5911 	return ret;
5912 }
5913 
5914 /* Try to shrink the extent tree */
5915 void ext4_ext_replay_shrink_inode(struct inode *inode, ext4_lblk_t end)
5916 {
5917 	struct ext4_ext_path *path = NULL;
5918 	struct ext4_extent *ex;
5919 	ext4_lblk_t old_cur, cur = 0;
5920 
5921 	while (cur < end) {
5922 		path = ext4_find_extent(inode, cur, NULL, 0);
5923 		if (IS_ERR(path))
5924 			return;
5925 		ex = path[path->p_depth].p_ext;
5926 		if (!ex) {
5927 			ext4_ext_drop_refs(path);
5928 			kfree(path);
5929 			ext4_mark_inode_dirty(NULL, inode);
5930 			return;
5931 		}
5932 		old_cur = cur;
5933 		cur = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5934 		if (cur <= old_cur)
5935 			cur = old_cur + 1;
5936 		ext4_ext_try_to_merge(NULL, inode, path, ex);
5937 		down_write(&EXT4_I(inode)->i_data_sem);
5938 		ext4_ext_dirty(NULL, inode, &path[path->p_depth]);
5939 		up_write(&EXT4_I(inode)->i_data_sem);
5940 		ext4_mark_inode_dirty(NULL, inode);
5941 		ext4_ext_drop_refs(path);
5942 		kfree(path);
5943 	}
5944 }
5945 
5946 /* Check if *cur is a hole and if it is, skip it */
5947 static int skip_hole(struct inode *inode, ext4_lblk_t *cur)
5948 {
5949 	int ret;
5950 	struct ext4_map_blocks map;
5951 
5952 	map.m_lblk = *cur;
5953 	map.m_len = ((inode->i_size) >> inode->i_sb->s_blocksize_bits) - *cur;
5954 
5955 	ret = ext4_map_blocks(NULL, inode, &map, 0);
5956 	if (ret < 0)
5957 		return ret;
5958 	if (ret != 0)
5959 		return 0;
5960 	*cur = *cur + map.m_len;
5961 	return 0;
5962 }
5963 
5964 /* Count number of blocks used by this inode and update i_blocks */
5965 int ext4_ext_replay_set_iblocks(struct inode *inode)
5966 {
5967 	struct ext4_ext_path *path = NULL, *path2 = NULL;
5968 	struct ext4_extent *ex;
5969 	ext4_lblk_t cur = 0, end;
5970 	int numblks = 0, i, ret = 0;
5971 	ext4_fsblk_t cmp1, cmp2;
5972 	struct ext4_map_blocks map;
5973 
5974 	/* Determin the size of the file first */
5975 	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
5976 					EXT4_EX_NOCACHE);
5977 	if (IS_ERR(path))
5978 		return PTR_ERR(path);
5979 	ex = path[path->p_depth].p_ext;
5980 	if (!ex) {
5981 		ext4_ext_drop_refs(path);
5982 		kfree(path);
5983 		goto out;
5984 	}
5985 	end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
5986 	ext4_ext_drop_refs(path);
5987 	kfree(path);
5988 
5989 	/* Count the number of data blocks */
5990 	cur = 0;
5991 	while (cur < end) {
5992 		map.m_lblk = cur;
5993 		map.m_len = end - cur;
5994 		ret = ext4_map_blocks(NULL, inode, &map, 0);
5995 		if (ret < 0)
5996 			break;
5997 		if (ret > 0)
5998 			numblks += ret;
5999 		cur = cur + map.m_len;
6000 	}
6001 
6002 	/*
6003 	 * Count the number of extent tree blocks. We do it by looking up
6004 	 * two successive extents and determining the difference between
6005 	 * their paths. When path is different for 2 successive extents
6006 	 * we compare the blocks in the path at each level and increment
6007 	 * iblocks by total number of differences found.
6008 	 */
6009 	cur = 0;
6010 	ret = skip_hole(inode, &cur);
6011 	if (ret < 0)
6012 		goto out;
6013 	path = ext4_find_extent(inode, cur, NULL, 0);
6014 	if (IS_ERR(path))
6015 		goto out;
6016 	numblks += path->p_depth;
6017 	ext4_ext_drop_refs(path);
6018 	kfree(path);
6019 	while (cur < end) {
6020 		path = ext4_find_extent(inode, cur, NULL, 0);
6021 		if (IS_ERR(path))
6022 			break;
6023 		ex = path[path->p_depth].p_ext;
6024 		if (!ex) {
6025 			ext4_ext_drop_refs(path);
6026 			kfree(path);
6027 			return 0;
6028 		}
6029 		cur = max(cur + 1, le32_to_cpu(ex->ee_block) +
6030 					ext4_ext_get_actual_len(ex));
6031 		ret = skip_hole(inode, &cur);
6032 		if (ret < 0) {
6033 			ext4_ext_drop_refs(path);
6034 			kfree(path);
6035 			break;
6036 		}
6037 		path2 = ext4_find_extent(inode, cur, NULL, 0);
6038 		if (IS_ERR(path2)) {
6039 			ext4_ext_drop_refs(path);
6040 			kfree(path);
6041 			break;
6042 		}
6043 		for (i = 0; i <= max(path->p_depth, path2->p_depth); i++) {
6044 			cmp1 = cmp2 = 0;
6045 			if (i <= path->p_depth)
6046 				cmp1 = path[i].p_bh ?
6047 					path[i].p_bh->b_blocknr : 0;
6048 			if (i <= path2->p_depth)
6049 				cmp2 = path2[i].p_bh ?
6050 					path2[i].p_bh->b_blocknr : 0;
6051 			if (cmp1 != cmp2 && cmp2 != 0)
6052 				numblks++;
6053 		}
6054 		ext4_ext_drop_refs(path);
6055 		ext4_ext_drop_refs(path2);
6056 		kfree(path);
6057 		kfree(path2);
6058 	}
6059 
6060 out:
6061 	inode->i_blocks = numblks << (inode->i_sb->s_blocksize_bits - 9);
6062 	ext4_mark_inode_dirty(NULL, inode);
6063 	return 0;
6064 }
6065 
6066 int ext4_ext_clear_bb(struct inode *inode)
6067 {
6068 	struct ext4_ext_path *path = NULL;
6069 	struct ext4_extent *ex;
6070 	ext4_lblk_t cur = 0, end;
6071 	int j, ret = 0;
6072 	struct ext4_map_blocks map;
6073 
6074 	if (ext4_test_inode_flag(inode, EXT4_INODE_INLINE_DATA))
6075 		return 0;
6076 
6077 	/* Determin the size of the file first */
6078 	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL,
6079 					EXT4_EX_NOCACHE);
6080 	if (IS_ERR(path))
6081 		return PTR_ERR(path);
6082 	ex = path[path->p_depth].p_ext;
6083 	if (!ex) {
6084 		ext4_ext_drop_refs(path);
6085 		kfree(path);
6086 		return 0;
6087 	}
6088 	end = le32_to_cpu(ex->ee_block) + ext4_ext_get_actual_len(ex);
6089 	ext4_ext_drop_refs(path);
6090 	kfree(path);
6091 
6092 	cur = 0;
6093 	while (cur < end) {
6094 		map.m_lblk = cur;
6095 		map.m_len = end - cur;
6096 		ret = ext4_map_blocks(NULL, inode, &map, 0);
6097 		if (ret < 0)
6098 			break;
6099 		if (ret > 0) {
6100 			path = ext4_find_extent(inode, map.m_lblk, NULL, 0);
6101 			if (!IS_ERR_OR_NULL(path)) {
6102 				for (j = 0; j < path->p_depth; j++) {
6103 
6104 					ext4_mb_mark_bb(inode->i_sb,
6105 							path[j].p_block, 1, 0);
6106 				}
6107 				ext4_ext_drop_refs(path);
6108 				kfree(path);
6109 			}
6110 			ext4_mb_mark_bb(inode->i_sb, map.m_pblk, map.m_len, 0);
6111 		}
6112 		cur = cur + map.m_len;
6113 	}
6114 
6115 	return 0;
6116 }
6117