alloc.c 168 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
/* -*- mode: c; c-basic-offset: 8; -*-
 * vim: noexpandtab sw=8 ts=8 sts=0:
 *
 * alloc.c
 *
 * Extent allocs and frees
 *
 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public
 * License as published by the Free Software Foundation; either
 * version 2 of the License, or (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public
 * License along with this program; if not, write to the
 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 * Boston, MA 021110-1307, USA.
 */

#include <linux/fs.h>
#include <linux/types.h>
#include <linux/slab.h>
#include <linux/highmem.h>
30
#include <linux/swap.h>
31
32
33
34
35
36
37

#define MLOG_MASK_PREFIX ML_DISK_ALLOC
#include <cluster/masklog.h>

#include "ocfs2.h"

#include "alloc.h"
38
#include "aops.h"
39
40
41
42
43
44
45
46
47
48
49
50
51
#include "dlmglue.h"
#include "extent_map.h"
#include "inode.h"
#include "journal.h"
#include "localalloc.h"
#include "suballoc.h"
#include "sysfile.h"
#include "file.h"
#include "super.h"
#include "uptodate.h"

#include "buffer_head_io.h"

Mark Fasheh's avatar
Mark Fasheh committed
52
static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc);
53
54
static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
					 struct ocfs2_extent_block *eb);
55

Mark Fasheh's avatar
Mark Fasheh committed
56
57
58
59
60
61
62
63
64
65
66
/*
 * Structures which describe a path through a btree, and functions to
 * manipulate them.
 *
 * The idea here is to be as generic as possible with the tree
 * manipulation code.
 */
struct ocfs2_path_item {
	struct buffer_head		*bh;
	struct ocfs2_extent_list	*el;
};
67

Mark Fasheh's avatar
Mark Fasheh committed
68
#define OCFS2_MAX_PATH_DEPTH	5
69

Mark Fasheh's avatar
Mark Fasheh committed
70
71
72
73
struct ocfs2_path {
	int			p_tree_depth;
	struct ocfs2_path_item	p_node[OCFS2_MAX_PATH_DEPTH];
};
74

Mark Fasheh's avatar
Mark Fasheh committed
75
76
77
78
79
#define path_root_bh(_path) ((_path)->p_node[0].bh)
#define path_root_el(_path) ((_path)->p_node[0].el)
#define path_leaf_bh(_path) ((_path)->p_node[(_path)->p_tree_depth].bh)
#define path_leaf_el(_path) ((_path)->p_node[(_path)->p_tree_depth].el)
#define path_num_items(_path) ((_path)->p_tree_depth + 1)
80

Mark Fasheh's avatar
Mark Fasheh committed
81
82
83
84
85
86
87
88
89
/*
 * Reset the actual path elements so that we can re-use the structure
 * to build another path. Generally, this involves freeing the buffer
 * heads.
 */
static void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root)
{
	int i, start = 0, depth = 0;
	struct ocfs2_path_item *node;
90

Mark Fasheh's avatar
Mark Fasheh committed
91
92
	if (keep_root)
		start = 1;
93

Mark Fasheh's avatar
Mark Fasheh committed
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
	for(i = start; i < path_num_items(path); i++) {
		node = &path->p_node[i];

		brelse(node->bh);
		node->bh = NULL;
		node->el = NULL;
	}

	/*
	 * Tree depth may change during truncate, or insert. If we're
	 * keeping the root extent list, then make sure that our path
	 * structure reflects the proper depth.
	 */
	if (keep_root)
		depth = le16_to_cpu(path_root_el(path)->l_tree_depth);

	path->p_tree_depth = depth;
}

static void ocfs2_free_path(struct ocfs2_path *path)
{
	if (path) {
		ocfs2_reinit_path(path, 0);
		kfree(path);
	}
}

121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
/*
 * All the elements of src into dest. After this call, src could be freed
 * without affecting dest.
 *
 * Both paths should have the same root. Any non-root elements of dest
 * will be freed.
 */
static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src)
{
	int i;

	BUG_ON(path_root_bh(dest) != path_root_bh(src));
	BUG_ON(path_root_el(dest) != path_root_el(src));

	ocfs2_reinit_path(dest, 1);

	for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
		dest->p_node[i].bh = src->p_node[i].bh;
		dest->p_node[i].el = src->p_node[i].el;

		if (dest->p_node[i].bh)
			get_bh(dest->p_node[i].bh);
	}
}

Mark Fasheh's avatar
Mark Fasheh committed
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
/*
 * Make the *dest path the same as src and re-initialize src path to
 * have a root only.
 */
static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src)
{
	int i;

	BUG_ON(path_root_bh(dest) != path_root_bh(src));

	for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
		brelse(dest->p_node[i].bh);

		dest->p_node[i].bh = src->p_node[i].bh;
		dest->p_node[i].el = src->p_node[i].el;

		src->p_node[i].bh = NULL;
		src->p_node[i].el = NULL;
	}
}

/*
 * Insert an extent block at given index.
 *
 * This will not take an additional reference on eb_bh.
 */
static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index,
					struct buffer_head *eb_bh)
{
	struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data;

	/*
	 * Right now, no root bh is an extent block, so this helps
	 * catch code errors with dinode trees. The assertion can be
	 * safely removed if we ever need to insert extent block
	 * structures at the root.
	 */
	BUG_ON(index == 0);

	path->p_node[index].bh = eb_bh;
	path->p_node[index].el = &eb->h_list;
}

static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh,
					 struct ocfs2_extent_list *root_el)
{
	struct ocfs2_path *path;
193

Mark Fasheh's avatar
Mark Fasheh committed
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
	BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH);

	path = kzalloc(sizeof(*path), GFP_NOFS);
	if (path) {
		path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth);
		get_bh(root_bh);
		path_root_bh(path) = root_bh;
		path_root_el(path) = root_el;
	}

	return path;
}

/*
 * Allocate and initialize a new path based on a disk inode tree.
 */
static struct ocfs2_path *ocfs2_new_inode_path(struct buffer_head *di_bh)
{
	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
	struct ocfs2_extent_list *el = &di->id2.i_list;

	return ocfs2_new_path(di_bh, el);
}

/*
 * Convenience function to journal all components in a path.
 */
static int ocfs2_journal_access_path(struct inode *inode, handle_t *handle,
				     struct ocfs2_path *path)
{
	int i, ret = 0;

	if (!path)
		goto out;

	for(i = 0; i < path_num_items(path); i++) {
		ret = ocfs2_journal_access(handle, inode, path->p_node[i].bh,
					   OCFS2_JOURNAL_ACCESS_WRITE);
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}
	}

out:
	return ret;
}

242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
/*
 * Return the index of the extent record which contains cluster #v_cluster.
 * -1 is returned if it was not found.
 *
 * Should work fine on interior and exterior nodes.
 */
int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster)
{
	int ret = -1;
	int i;
	struct ocfs2_extent_rec *rec;
	u32 rec_end, rec_start, clusters;

	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
		rec = &el->l_recs[i];

		rec_start = le32_to_cpu(rec->e_cpos);
		clusters = ocfs2_rec_clusters(el, rec);

		rec_end = rec_start + clusters;

		if (v_cluster >= rec_start && v_cluster < rec_end) {
			ret = i;
			break;
		}
	}

	return ret;
}

Mark Fasheh's avatar
Mark Fasheh committed
272
273
274
enum ocfs2_contig_type {
	CONTIG_NONE = 0,
	CONTIG_LEFT,
275
276
	CONTIG_RIGHT,
	CONTIG_LEFTRIGHT,
Mark Fasheh's avatar
Mark Fasheh committed
277
278
};

279
280
281
282
283

/*
 * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and
 * ocfs2_extent_contig only work properly against leaf nodes!
 */
Mark Fasheh's avatar
Mark Fasheh committed
284
285
286
static int ocfs2_block_extent_contig(struct super_block *sb,
				     struct ocfs2_extent_rec *ext,
				     u64 blkno)
287
{
288
289
290
291
292
293
	u64 blk_end = le64_to_cpu(ext->e_blkno);

	blk_end += ocfs2_clusters_to_blocks(sb,
				    le16_to_cpu(ext->e_leaf_clusters));

	return blkno == blk_end;
294
295
}

Mark Fasheh's avatar
Mark Fasheh committed
296
297
298
static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left,
				  struct ocfs2_extent_rec *right)
{
299
300
301
302
303
304
	u32 left_range;

	left_range = le32_to_cpu(left->e_cpos) +
		le16_to_cpu(left->e_leaf_clusters);

	return (left_range == le32_to_cpu(right->e_cpos));
Mark Fasheh's avatar
Mark Fasheh committed
305
306
307
308
309
310
311
312
313
}

static enum ocfs2_contig_type
	ocfs2_extent_contig(struct inode *inode,
			    struct ocfs2_extent_rec *ext,
			    struct ocfs2_extent_rec *insert_rec)
{
	u64 blkno = le64_to_cpu(insert_rec->e_blkno);

314
315
316
317
318
319
320
321
	/*
	 * Refuse to coalesce extent records with different flag
	 * fields - we don't want to mix unwritten extents with user
	 * data.
	 */
	if (ext->e_flags != insert_rec->e_flags)
		return CONTIG_NONE;

Mark Fasheh's avatar
Mark Fasheh committed
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
	if (ocfs2_extents_adjacent(ext, insert_rec) &&
	    ocfs2_block_extent_contig(inode->i_sb, ext, blkno))
			return CONTIG_RIGHT;

	blkno = le64_to_cpu(ext->e_blkno);
	if (ocfs2_extents_adjacent(insert_rec, ext) &&
	    ocfs2_block_extent_contig(inode->i_sb, insert_rec, blkno))
		return CONTIG_LEFT;

	return CONTIG_NONE;
}

/*
 * NOTE: We can have pretty much any combination of contiguousness and
 * appending.
 *
 * The usefulness of APPEND_TAIL is more in that it lets us know that
 * we'll have to update the path to that leaf.
 */
enum ocfs2_append_type {
	APPEND_NONE = 0,
	APPEND_TAIL,
};

346
347
348
349
350
351
enum ocfs2_split_type {
	SPLIT_NONE = 0,
	SPLIT_LEFT,
	SPLIT_RIGHT,
};

Mark Fasheh's avatar
Mark Fasheh committed
352
struct ocfs2_insert_type {
353
	enum ocfs2_split_type	ins_split;
Mark Fasheh's avatar
Mark Fasheh committed
354
355
356
357
358
359
	enum ocfs2_append_type	ins_appending;
	enum ocfs2_contig_type	ins_contig;
	int			ins_contig_index;
	int			ins_tree_depth;
};

360
361
362
363
364
365
struct ocfs2_merge_ctxt {
	enum ocfs2_contig_type	c_contig_type;
	int			c_has_empty_extent;
	int			c_split_covers_rec;
};

366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
/*
 * How many free extents have we got before we need more meta data?
 */
int ocfs2_num_free_extents(struct ocfs2_super *osb,
			   struct inode *inode,
			   struct ocfs2_dinode *fe)
{
	int retval;
	struct ocfs2_extent_list *el;
	struct ocfs2_extent_block *eb;
	struct buffer_head *eb_bh = NULL;

	mlog_entry_void();

	if (!OCFS2_IS_VALID_DINODE(fe)) {
		OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
		retval = -EIO;
		goto bail;
	}

	if (fe->i_last_eb_blk) {
		retval = ocfs2_read_block(osb, le64_to_cpu(fe->i_last_eb_blk),
					  &eb_bh, OCFS2_BH_CACHED, inode);
		if (retval < 0) {
			mlog_errno(retval);
			goto bail;
		}
		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
		el = &eb->h_list;
	} else
		el = &fe->id2.i_list;

	BUG_ON(el->l_tree_depth != 0);

	retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec);
bail:
	if (eb_bh)
		brelse(eb_bh);

	mlog_exit(retval);
	return retval;
}

/* expects array to already be allocated
 *
 * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and
 * l_count for you
 */
static int ocfs2_create_new_meta_bhs(struct ocfs2_super *osb,
415
				     handle_t *handle,
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
				     struct inode *inode,
				     int wanted,
				     struct ocfs2_alloc_context *meta_ac,
				     struct buffer_head *bhs[])
{
	int count, status, i;
	u16 suballoc_bit_start;
	u32 num_got;
	u64 first_blkno;
	struct ocfs2_extent_block *eb;

	mlog_entry_void();

	count = 0;
	while (count < wanted) {
		status = ocfs2_claim_metadata(osb,
					      handle,
					      meta_ac,
					      wanted - count,
					      &suballoc_bit_start,
					      &num_got,
					      &first_blkno);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		for(i = count;  i < (num_got + count); i++) {
			bhs[i] = sb_getblk(osb->sb, first_blkno);
			if (bhs[i] == NULL) {
				status = -EIO;
				mlog_errno(status);
				goto bail;
			}
			ocfs2_set_new_buffer_uptodate(inode, bhs[i]);

			status = ocfs2_journal_access(handle, inode, bhs[i],
						      OCFS2_JOURNAL_ACCESS_CREATE);
			if (status < 0) {
				mlog_errno(status);
				goto bail;
			}

			memset(bhs[i]->b_data, 0, osb->sb->s_blocksize);
			eb = (struct ocfs2_extent_block *) bhs[i]->b_data;
			/* Ok, setup the minimal stuff here. */
			strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE);
			eb->h_blkno = cpu_to_le64(first_blkno);
			eb->h_fs_generation = cpu_to_le32(osb->fs_generation);
			eb->h_suballoc_slot = cpu_to_le16(osb->slot_num);
			eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start);
			eb->h_list.l_count =
				cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb));

			suballoc_bit_start++;
			first_blkno++;

			/* We'll also be dirtied by the caller, so
			 * this isn't absolutely necessary. */
			status = ocfs2_journal_dirty(handle, bhs[i]);
			if (status < 0) {
				mlog_errno(status);
				goto bail;
			}
		}

		count += num_got;
	}

	status = 0;
bail:
	if (status < 0) {
		for(i = 0; i < wanted; i++) {
			if (bhs[i])
				brelse(bhs[i]);
			bhs[i] = NULL;
		}
	}
	mlog_exit(status);
	return status;
}

Mark Fasheh's avatar
Mark Fasheh committed
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
/*
 * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth().
 *
 * Returns the sum of the rightmost extent rec logical offset and
 * cluster count.
 *
 * ocfs2_add_branch() uses this to determine what logical cluster
 * value should be populated into the leftmost new branch records.
 *
 * ocfs2_shift_tree_depth() uses this to determine the # clusters
 * value for the new topmost tree record.
 */
static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list  *el)
{
	int i;

	i = le16_to_cpu(el->l_next_free_rec) - 1;

	return le32_to_cpu(el->l_recs[i].e_cpos) +
517
		ocfs2_rec_clusters(el, &el->l_recs[i]);
Mark Fasheh's avatar
Mark Fasheh committed
518
519
}

520
521
522
523
524
525
526
527
528
/*
 * Add an entire tree branch to our inode. eb_bh is the extent block
 * to start at, if we don't want to start the branch at the dinode
 * structure.
 *
 * last_eb_bh is required as we have to update it's next_leaf pointer
 * for the new last extent block.
 *
 * the new branch will be 'empty' in the sense that every block will
529
 * contain a single record with cluster count == 0.
530
531
 */
static int ocfs2_add_branch(struct ocfs2_super *osb,
532
			    handle_t *handle,
533
534
535
			    struct inode *inode,
			    struct buffer_head *fe_bh,
			    struct buffer_head *eb_bh,
536
			    struct buffer_head **last_eb_bh,
537
538
539
540
541
542
543
544
545
546
			    struct ocfs2_alloc_context *meta_ac)
{
	int status, new_blocks, i;
	u64 next_blkno, new_last_eb_blk;
	struct buffer_head *bh;
	struct buffer_head **new_eb_bhs = NULL;
	struct ocfs2_dinode *fe;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list  *eb_el;
	struct ocfs2_extent_list  *el;
Mark Fasheh's avatar
Mark Fasheh committed
547
	u32 new_cpos;
548
549
550

	mlog_entry_void();

551
	BUG_ON(!last_eb_bh || !*last_eb_bh);
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581

	fe = (struct ocfs2_dinode *) fe_bh->b_data;

	if (eb_bh) {
		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
		el = &eb->h_list;
	} else
		el = &fe->id2.i_list;

	/* we never add a branch to a leaf. */
	BUG_ON(!el->l_tree_depth);

	new_blocks = le16_to_cpu(el->l_tree_depth);

	/* allocate the number of new eb blocks we need */
	new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *),
			     GFP_KERNEL);
	if (!new_eb_bhs) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}

	status = ocfs2_create_new_meta_bhs(osb, handle, inode, new_blocks,
					   meta_ac, new_eb_bhs);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

582
	eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data;
Mark Fasheh's avatar
Mark Fasheh committed
583
584
	new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list);

585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
	/* Note: new_eb_bhs[new_blocks - 1] is the guy which will be
	 * linked with the rest of the tree.
	 * conversly, new_eb_bhs[0] is the new bottommost leaf.
	 *
	 * when we leave the loop, new_last_eb_blk will point to the
	 * newest leaf, and next_blkno will point to the topmost extent
	 * block. */
	next_blkno = new_last_eb_blk = 0;
	for(i = 0; i < new_blocks; i++) {
		bh = new_eb_bhs[i];
		eb = (struct ocfs2_extent_block *) bh->b_data;
		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
			status = -EIO;
			goto bail;
		}
		eb_el = &eb->h_list;

		status = ocfs2_journal_access(handle, inode, bh,
					      OCFS2_JOURNAL_ACCESS_CREATE);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		eb->h_next_leaf_blk = 0;
		eb_el->l_tree_depth = cpu_to_le16(i);
		eb_el->l_next_free_rec = cpu_to_le16(1);
Mark Fasheh's avatar
Mark Fasheh committed
613
614
615
616
617
		/*
		 * This actually counts as an empty extent as
		 * c_clusters == 0
		 */
		eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos);
618
		eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno);
619
620
621
622
623
624
		/*
		 * eb_el isn't always an interior node, but even leaf
		 * nodes want a zero'd flags and reserved field so
		 * this gets the whole 32 bits regardless of use.
		 */
		eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0);
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
		if (!eb_el->l_tree_depth)
			new_last_eb_blk = le64_to_cpu(eb->h_blkno);

		status = ocfs2_journal_dirty(handle, bh);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		next_blkno = le64_to_cpu(eb->h_blkno);
	}

	/* This is a bit hairy. We want to update up to three blocks
	 * here without leaving any of them in an inconsistent state
	 * in case of error. We don't have to worry about
	 * journal_dirty erroring as it won't unless we've aborted the
	 * handle (in which case we would never be here) so reserving
	 * the write with journal_access is all we need to do. */
643
	status = ocfs2_journal_access(handle, inode, *last_eb_bh,
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
	status = ocfs2_journal_access(handle, inode, fe_bh,
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
	if (eb_bh) {
		status = ocfs2_journal_access(handle, inode, eb_bh,
					      OCFS2_JOURNAL_ACCESS_WRITE);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

	/* Link the new branch into the rest of the tree (el will
	 * either be on the fe, or the extent block passed in. */
	i = le16_to_cpu(el->l_next_free_rec);
	el->l_recs[i].e_blkno = cpu_to_le64(next_blkno);
Mark Fasheh's avatar
Mark Fasheh committed
668
	el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
669
	el->l_recs[i].e_int_clusters = 0;
670
671
672
673
674
675
	le16_add_cpu(&el->l_next_free_rec, 1);

	/* fe needs a new last extent block pointer, as does the
	 * next_leaf on the previously last-extent-block. */
	fe->i_last_eb_blk = cpu_to_le64(new_last_eb_blk);

676
	eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
677
678
	eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);

679
	status = ocfs2_journal_dirty(handle, *last_eb_bh);
680
681
682
683
684
685
686
687
688
689
690
	if (status < 0)
		mlog_errno(status);
	status = ocfs2_journal_dirty(handle, fe_bh);
	if (status < 0)
		mlog_errno(status);
	if (eb_bh) {
		status = ocfs2_journal_dirty(handle, eb_bh);
		if (status < 0)
			mlog_errno(status);
	}

691
692
693
694
695
696
697
698
	/*
	 * Some callers want to track the rightmost leaf so pass it
	 * back here.
	 */
	brelse(*last_eb_bh);
	get_bh(new_eb_bhs[0]);
	*last_eb_bh = new_eb_bhs[0];

699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
	status = 0;
bail:
	if (new_eb_bhs) {
		for (i = 0; i < new_blocks; i++)
			if (new_eb_bhs[i])
				brelse(new_eb_bhs[i]);
		kfree(new_eb_bhs);
	}

	mlog_exit(status);
	return status;
}

/*
 * adds another level to the allocation tree.
 * returns back the new extent block so you can add a branch to it
 * after this call.
 */
static int ocfs2_shift_tree_depth(struct ocfs2_super *osb,
718
				  handle_t *handle,
719
720
721
722
723
724
				  struct inode *inode,
				  struct buffer_head *fe_bh,
				  struct ocfs2_alloc_context *meta_ac,
				  struct buffer_head **ret_new_eb_bh)
{
	int status, i;
Mark Fasheh's avatar
Mark Fasheh committed
725
	u32 new_clusters;
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
	struct buffer_head *new_eb_bh = NULL;
	struct ocfs2_dinode *fe;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list  *fe_el;
	struct ocfs2_extent_list  *eb_el;

	mlog_entry_void();

	status = ocfs2_create_new_meta_bhs(osb, handle, inode, 1, meta_ac,
					   &new_eb_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	eb = (struct ocfs2_extent_block *) new_eb_bh->b_data;
	if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
		OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
		status = -EIO;
		goto bail;
	}

	eb_el = &eb->h_list;
	fe = (struct ocfs2_dinode *) fe_bh->b_data;
	fe_el = &fe->id2.i_list;

	status = ocfs2_journal_access(handle, inode, new_eb_bh,
				      OCFS2_JOURNAL_ACCESS_CREATE);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	/* copy the fe data into the new extent block */
	eb_el->l_tree_depth = fe_el->l_tree_depth;
	eb_el->l_next_free_rec = fe_el->l_next_free_rec;
762
763
	for(i = 0; i < le16_to_cpu(fe_el->l_next_free_rec); i++)
		eb_el->l_recs[i] = fe_el->l_recs[i];
764
765
766
767
768
769
770
771
772
773
774
775
776
777

	status = ocfs2_journal_dirty(handle, new_eb_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	status = ocfs2_journal_access(handle, inode, fe_bh,
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

Mark Fasheh's avatar
Mark Fasheh committed
778
779
	new_clusters = ocfs2_sum_rightmost_rec(eb_el);

780
781
782
783
	/* update fe now */
	le16_add_cpu(&fe_el->l_tree_depth, 1);
	fe_el->l_recs[0].e_cpos = 0;
	fe_el->l_recs[0].e_blkno = eb->h_blkno;
784
785
786
	fe_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters);
	for(i = 1; i < le16_to_cpu(fe_el->l_next_free_rec); i++)
		memset(&fe_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
	fe_el->l_next_free_rec = cpu_to_le16(1);

	/* If this is our 1st tree depth shift, then last_eb_blk
	 * becomes the allocated extent block */
	if (fe_el->l_tree_depth == cpu_to_le16(1))
		fe->i_last_eb_blk = eb->h_blkno;

	status = ocfs2_journal_dirty(handle, fe_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	*ret_new_eb_bh = new_eb_bh;
	new_eb_bh = NULL;
	status = 0;
bail:
	if (new_eb_bh)
		brelse(new_eb_bh);

	mlog_exit(status);
	return status;
}

/*
 * Should only be called when there is no space left in any of the
 * leaf nodes. What we want to do is find the lowest tree depth
 * non-leaf extent block with room for new records. There are three
 * valid results of this search:
 *
 * 1) a lowest extent block is found, then we pass it back in
 *    *lowest_eb_bh and return '0'
 *
 * 2) the search fails to find anything, but the dinode has room. We
 *    pass NULL back in *lowest_eb_bh, but still return '0'
 *
 * 3) the search fails to find anything AND the dinode is full, in
 *    which case we return > 0
 *
 * return status < 0 indicates an error.
 */
static int ocfs2_find_branch_target(struct ocfs2_super *osb,
				    struct inode *inode,
				    struct buffer_head *fe_bh,
				    struct buffer_head **target_bh)
{
	int status = 0, i;
	u64 blkno;
	struct ocfs2_dinode *fe;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list  *el;
	struct buffer_head *bh = NULL;
	struct buffer_head *lowest_bh = NULL;

	mlog_entry_void();

	*target_bh = NULL;

	fe = (struct ocfs2_dinode *) fe_bh->b_data;
	el = &fe->id2.i_list;

	while(le16_to_cpu(el->l_tree_depth) > 1) {
		if (le16_to_cpu(el->l_next_free_rec) == 0) {
850
			ocfs2_error(inode->i_sb, "Dinode %llu has empty "
851
				    "extent list (next_free_rec == 0)",
852
				    (unsigned long long)OCFS2_I(inode)->ip_blkno);
853
854
855
856
857
858
			status = -EIO;
			goto bail;
		}
		i = le16_to_cpu(el->l_next_free_rec) - 1;
		blkno = le64_to_cpu(el->l_recs[i].e_blkno);
		if (!blkno) {
859
			ocfs2_error(inode->i_sb, "Dinode %llu has extent "
860
861
				    "list where extent # %d has no physical "
				    "block start",
862
				    (unsigned long long)OCFS2_I(inode)->ip_blkno, i);
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
			status = -EIO;
			goto bail;
		}

		if (bh) {
			brelse(bh);
			bh = NULL;
		}

		status = ocfs2_read_block(osb, blkno, &bh, OCFS2_BH_CACHED,
					  inode);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
Mark Fasheh's avatar
Mark Fasheh committed
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910

		eb = (struct ocfs2_extent_block *) bh->b_data;
		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
			status = -EIO;
			goto bail;
		}
		el = &eb->h_list;

		if (le16_to_cpu(el->l_next_free_rec) <
		    le16_to_cpu(el->l_count)) {
			if (lowest_bh)
				brelse(lowest_bh);
			lowest_bh = bh;
			get_bh(lowest_bh);
		}
	}

	/* If we didn't find one and the fe doesn't have any room,
	 * then return '1' */
	if (!lowest_bh
	    && (fe->id2.i_list.l_next_free_rec == fe->id2.i_list.l_count))
		status = 1;

	*target_bh = lowest_bh;
bail:
	if (bh)
		brelse(bh);

	mlog_exit(status);
	return status;
}

911
912
913
914
915
916
917
/*
 * Grow a b-tree so that it has more records.
 *
 * We might shift the tree depth in which case existing paths should
 * be considered invalid.
 *
 * Tree depth after the grow is returned via *final_depth.
918
919
 *
 * *last_eb_bh will be updated by ocfs2_add_branch().
920
921
922
 */
static int ocfs2_grow_tree(struct inode *inode, handle_t *handle,
			   struct buffer_head *di_bh, int *final_depth,
923
			   struct buffer_head **last_eb_bh,
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
			   struct ocfs2_alloc_context *meta_ac)
{
	int ret, shift;
	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
	int depth = le16_to_cpu(di->id2.i_list.l_tree_depth);
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
	struct buffer_head *bh = NULL;

	BUG_ON(meta_ac == NULL);

	shift = ocfs2_find_branch_target(osb, inode, di_bh, &bh);
	if (shift < 0) {
		ret = shift;
		mlog_errno(ret);
		goto out;
	}

	/* We traveled all the way to the bottom of the allocation tree
	 * and didn't find room for any more extents - we need to add
	 * another tree level */
	if (shift) {
		BUG_ON(bh);
		mlog(0, "need to shift tree depth (current = %d)\n", depth);

		/* ocfs2_shift_tree_depth will return us a buffer with
		 * the new extent block (so we can pass that to
		 * ocfs2_add_branch). */
		ret = ocfs2_shift_tree_depth(osb, handle, inode, di_bh,
					     meta_ac, &bh);
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}
		depth++;
958
959
960
961
962
963
964
965
966
967
968
969
970
		if (depth == 1) {
			/*
			 * Special case: we have room now if we shifted from
			 * tree_depth 0, so no more work needs to be done.
			 *
			 * We won't be calling add_branch, so pass
			 * back *last_eb_bh as the new leaf. At depth
			 * zero, it should always be null so there's
			 * no reason to brelse.
			 */
			BUG_ON(*last_eb_bh);
			get_bh(bh);
			*last_eb_bh = bh;
971
			goto out;
972
		}
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
	}

	/* call ocfs2_add_branch to add the final part of the tree with
	 * the new data. */
	mlog(0, "add branch. bh = %p\n", bh);
	ret = ocfs2_add_branch(osb, handle, inode, di_bh, bh, last_eb_bh,
			       meta_ac);
	if (ret < 0) {
		mlog_errno(ret);
		goto out;
	}

out:
	if (final_depth)
		*final_depth = depth;
	brelse(bh);
	return ret;
}

992
993
994
995
/*
 * This is only valid for leaf nodes, which are the only ones that can
 * have empty extents anyway.
 */
Mark Fasheh's avatar
Mark Fasheh committed
996
997
static inline int ocfs2_is_empty_extent(struct ocfs2_extent_rec *rec)
{
998
	return !rec->e_leaf_clusters;
Mark Fasheh's avatar
Mark Fasheh committed
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
}

/*
 * This function will discard the rightmost extent record.
 */
static void ocfs2_shift_records_right(struct ocfs2_extent_list *el)
{
	int next_free = le16_to_cpu(el->l_next_free_rec);
	int count = le16_to_cpu(el->l_count);
	unsigned int num_bytes;

	BUG_ON(!next_free);
	/* This will cause us to go off the end of our extent list. */
	BUG_ON(next_free >= count);

	num_bytes = sizeof(struct ocfs2_extent_rec) * next_free;

	memmove(&el->l_recs[1], &el->l_recs[0], num_bytes);
}

static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el,
			      struct ocfs2_extent_rec *insert_rec)
{
	int i, insert_index, next_free, has_empty, num_bytes;
	u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos);
	struct ocfs2_extent_rec *rec;

	next_free = le16_to_cpu(el->l_next_free_rec);
	has_empty = ocfs2_is_empty_extent(&el->l_recs[0]);

	BUG_ON(!next_free);

	/* The tree code before us didn't allow enough room in the leaf. */
Julia Lawall's avatar
Julia Lawall committed
1032
	BUG_ON(el->l_next_free_rec == el->l_count && !has_empty);
Mark Fasheh's avatar
Mark Fasheh committed
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096

	/*
	 * The easiest way to approach this is to just remove the
	 * empty extent and temporarily decrement next_free.
	 */
	if (has_empty) {
		/*
		 * If next_free was 1 (only an empty extent), this
		 * loop won't execute, which is fine. We still want
		 * the decrement above to happen.
		 */
		for(i = 0; i < (next_free - 1); i++)
			el->l_recs[i] = el->l_recs[i+1];

		next_free--;
	}

	/*
	 * Figure out what the new record index should be.
	 */
	for(i = 0; i < next_free; i++) {
		rec = &el->l_recs[i];

		if (insert_cpos < le32_to_cpu(rec->e_cpos))
			break;
	}
	insert_index = i;

	mlog(0, "ins %u: index %d, has_empty %d, next_free %d, count %d\n",
	     insert_cpos, insert_index, has_empty, next_free, le16_to_cpu(el->l_count));

	BUG_ON(insert_index < 0);
	BUG_ON(insert_index >= le16_to_cpu(el->l_count));
	BUG_ON(insert_index > next_free);

	/*
	 * No need to memmove if we're just adding to the tail.
	 */
	if (insert_index != next_free) {
		BUG_ON(next_free >= le16_to_cpu(el->l_count));

		num_bytes = next_free - insert_index;
		num_bytes *= sizeof(struct ocfs2_extent_rec);
		memmove(&el->l_recs[insert_index + 1],
			&el->l_recs[insert_index],
			num_bytes);
	}

	/*
	 * Either we had an empty extent, and need to re-increment or
	 * there was no empty extent on a non full rightmost leaf node,
	 * in which case we still need to increment.
	 */
	next_free++;
	el->l_next_free_rec = cpu_to_le16(next_free);
	/*
	 * Make sure none of the math above just messed up our tree.
	 */
	BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count));

	el->l_recs[insert_index] = *insert_rec;

}

1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el)
{
	int size, num_recs = le16_to_cpu(el->l_next_free_rec);

	BUG_ON(num_recs == 0);

	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
		num_recs--;
		size = num_recs * sizeof(struct ocfs2_extent_rec);
		memmove(&el->l_recs[0], &el->l_recs[1], size);
		memset(&el->l_recs[num_recs], 0,
		       sizeof(struct ocfs2_extent_rec));
		el->l_next_free_rec = cpu_to_le16(num_recs);
	}
}

Mark Fasheh's avatar
Mark Fasheh committed
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
/*
 * Create an empty extent record .
 *
 * l_next_free_rec may be updated.
 *
 * If an empty extent already exists do nothing.
 */
static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el)
{
	int next_free = le16_to_cpu(el->l_next_free_rec);

1124
1125
	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);

Mark Fasheh's avatar
Mark Fasheh committed
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
	if (next_free == 0)
		goto set_and_inc;

	if (ocfs2_is_empty_extent(&el->l_recs[0]))
		return;

	mlog_bug_on_msg(el->l_count == el->l_next_free_rec,
			"Asked to create an empty extent in a full list:\n"
			"count = %u, tree depth = %u",
			le16_to_cpu(el->l_count),
			le16_to_cpu(el->l_tree_depth));

	ocfs2_shift_records_right(el);

set_and_inc:
	le16_add_cpu(&el->l_next_free_rec, 1);
	memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
}

/*
 * For a rotation which involves two leaf nodes, the "root node" is
 * the lowest level tree node which contains a path to both leafs. This
 * resulting set of information can be used to form a complete "subtree"
 *
 * This function is passed two full paths from the dinode down to a
 * pair of adjacent leaves. It's task is to figure out which path
 * index contains the subtree root - this can be the root index itself
 * in a worst-case rotation.
 *
 * The array index of the subtree root is passed back.
 */
static int ocfs2_find_subtree_root(struct inode *inode,
				   struct ocfs2_path *left,
				   struct ocfs2_path *right)
{
	int i = 0;

	/*
	 * Check that the caller passed in two paths from the same tree.
	 */
	BUG_ON(path_root_bh(left) != path_root_bh(right));

	do {
		i++;

		/*
		 * The caller didn't pass two adjacent paths.
		 */
		mlog_bug_on_msg(i > left->p_tree_depth,
				"Inode %lu, left depth %u, right depth %u\n"
				"left leaf blk %llu, right leaf blk %llu\n",
				inode->i_ino, left->p_tree_depth,
				right->p_tree_depth,
				(unsigned long long)path_leaf_bh(left)->b_blocknr,
				(unsigned long long)path_leaf_bh(right)->b_blocknr);
	} while (left->p_node[i].bh->b_blocknr ==
		 right->p_node[i].bh->b_blocknr);

	return i - 1;
}

typedef void (path_insert_t)(void *, struct buffer_head *);

/*
 * Traverse a btree path in search of cpos, starting at root_el.
 *
 * This code can be called with a cpos larger than the tree, in which
 * case it will return the rightmost path.
 */
static int __ocfs2_find_path(struct inode *inode,
			     struct ocfs2_extent_list *root_el, u32 cpos,
			     path_insert_t *func, void *data)
{
	int i, ret = 0;
	u32 range;
	u64 blkno;
	struct buffer_head *bh = NULL;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;
	struct ocfs2_extent_rec *rec;
	struct ocfs2_inode_info *oi = OCFS2_I(inode);

	el = root_el;
	while (el->l_tree_depth) {
		if (le16_to_cpu(el->l_next_free_rec) == 0) {
			ocfs2_error(inode->i_sb,
				    "Inode %llu has empty extent list at "
				    "depth %u\n",
				    (unsigned long long)oi->ip_blkno,
				    le16_to_cpu(el->l_tree_depth));
			ret = -EROFS;
			goto out;

		}

		for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) {
			rec = &el->l_recs[i];

			/*
			 * In the case that cpos is off the allocation
			 * tree, this should just wind up returning the
			 * rightmost record.
			 */
			range = le32_to_cpu(rec->e_cpos) +
1230
				ocfs2_rec_clusters(el, rec);
Mark Fasheh's avatar
Mark Fasheh committed
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
			if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
			    break;
		}

		blkno = le64_to_cpu(el->l_recs[i].e_blkno);
		if (blkno == 0) {
			ocfs2_error(inode->i_sb,
				    "Inode %llu has bad blkno in extent list "
				    "at depth %u (index %d)\n",
				    (unsigned long long)oi->ip_blkno,
				    le16_to_cpu(el->l_tree_depth), i);
			ret = -EROFS;
			goto out;
		}

		brelse(bh);
		bh = NULL;
		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
				       &bh, OCFS2_BH_CACHED, inode);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		eb = (struct ocfs2_extent_block *) bh->b_data;
		el = &eb->h_list;
		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
			ret = -EIO;
			goto out;
		}

		if (le16_to_cpu(el->l_next_free_rec) >
		    le16_to_cpu(el->l_count)) {
			ocfs2_error(inode->i_sb,
				    "Inode %llu has bad count in extent list "
				    "at block %llu (next free=%u, count=%u)\n",
				    (unsigned long long)oi->ip_blkno,
				    (unsigned long long)bh->b_blocknr,
				    le16_to_cpu(el->l_next_free_rec),
				    le16_to_cpu(el->l_count));
			ret = -EROFS;
			goto out;
		}

		if (func)
			func(data, bh);
	}

out:
	/*
	 * Catch any trailing bh that the loop didn't handle.
	 */
	brelse(bh);

	return ret;
}

/*
 * Given an initialized path (that is, it has a valid root extent
 * list), this function will traverse the btree in search of the path
 * which would contain cpos.
 *
 * The path traveled is recorded in the path structure.
 *
 * Note that this will not do any comparisons on leaf node extent
 * records, so it will work fine in the case that we just added a tree
 * branch.
 */
struct find_path_data {
	int index;
	struct ocfs2_path *path;
};
static void find_path_ins(void *data, struct buffer_head *bh)
{
	struct find_path_data *fp = data;

	get_bh(bh);
	ocfs2_path_insert_eb(fp->path, fp->index, bh);
	fp->index++;
}
static int ocfs2_find_path(struct inode *inode, struct ocfs2_path *path,
			   u32 cpos)
{
	struct find_path_data data;

	data.index = 1;
	data.path = path;
	return __ocfs2_find_path(inode, path_root_el(path), cpos,
				 find_path_ins, &data);
}

static void find_leaf_ins(void *data, struct buffer_head *bh)
{
	struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data;
	struct ocfs2_extent_list *el = &eb->h_list;
	struct buffer_head **ret = data;

	/* We want to retain only the leaf block. */
	if (le16_to_cpu(el->l_tree_depth) == 0) {
		get_bh(bh);
		*ret = bh;
	}
}
/*
 * Find the leaf block in the tree which would contain cpos. No
 * checking of the actual leaf is done.
 *
 * Some paths want to call this instead of allocating a path structure
 * and calling ocfs2_find_path().
 *
 * This function doesn't handle non btree extent lists.
 */
1344
1345
int ocfs2_find_leaf(struct inode *inode, struct ocfs2_extent_list *root_el,
		    u32 cpos, struct buffer_head **leaf_bh)
Mark Fasheh's avatar
Mark Fasheh committed
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
{
	int ret;
	struct buffer_head *bh = NULL;

	ret = __ocfs2_find_path(inode, root_el, cpos, find_leaf_ins, &bh);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	*leaf_bh = bh;
out:
	return ret;
}

/*
 * Adjust the adjacent records (left_rec, right_rec) involved in a rotation.
 *
 * Basically, we've moved stuff around at the bottom of the tree and
 * we need to fix up the extent records above the changes to reflect
 * the new changes.
 *
 * left_rec: the record on the left.
 * left_child_el: is the child list pointed to by left_rec
 * right_rec: the record to the right of left_rec
 * right_child_el: is the child list pointed to by right_rec
 *
 * By definition, this only works on interior nodes.
 */
static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec,
				  struct ocfs2_extent_list *left_child_el,
				  struct ocfs2_extent_rec *right_rec,
				  struct ocfs2_extent_list *right_child_el)
{
	u32 left_clusters, right_end;

	/*
	 * Interior nodes never have holes. Their cpos is the cpos of
	 * the leftmost record in their child list. Their cluster
	 * count covers the full theoretical range of their child list
	 * - the range between their cpos and the cpos of the record
	 * immediately to their right.
	 */
	left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos);
1390
1391
1392
1393
	if (ocfs2_is_empty_extent(&right_child_el->l_recs[0])) {
		BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1);
		left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos);
	}
Mark Fasheh's avatar
Mark Fasheh committed
1394
	left_clusters -= le32_to_cpu(left_rec->e_cpos);
1395
	left_rec->e_int_clusters = cpu_to_le32(left_clusters);
Mark Fasheh's avatar
Mark Fasheh committed
1396
1397
1398

	/*
	 * Calculate the rightmost cluster count boundary before
1399
	 * moving cpos - we will need to adjust clusters after
Mark Fasheh's avatar
Mark Fasheh committed
1400
1401
1402
	 * updating e_cpos to keep the same highest cluster count.
	 */
	right_end = le32_to_cpu(right_rec->e_cpos);
1403
	right_end += le32_to_cpu(right_rec->e_int_clusters);
Mark Fasheh's avatar
Mark Fasheh committed
1404
1405
1406
1407
1408

	right_rec->e_cpos = left_rec->e_cpos;
	le32_add_cpu(&right_rec->e_cpos, left_clusters);

	right_end -= le32_to_cpu(right_rec->e_cpos);
1409
	right_rec->e_int_clusters = cpu_to_le32(right_end);
Mark Fasheh's avatar
Mark Fasheh committed
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
}

/*
 * Adjust the adjacent root node records involved in a
 * rotation. left_el_blkno is passed in as a key so that we can easily
 * find it's index in the root list.
 */
static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el,
				      struct ocfs2_extent_list *left_el,
				      struct ocfs2_extent_list *right_el,
				      u64 left_el_blkno)
{
	int i;

	BUG_ON(le16_to_cpu(root_el->l_tree_depth) <=
	       le16_to_cpu(left_el->l_tree_depth));

	for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) {
		if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno)
			break;
	}

	/*
	 * The path walking code should have never returned a root and
	 * two paths which are not adjacent.
	 */
	BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1));

	ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el,
				      &root_el->l_recs[i + 1], right_el);
}

/*
 * We've changed a leaf block (in right_path) and need to reflect that
 * change back up the subtree.
 *
 * This happens in multiple places:
 *   - When we've moved an extent record from the left path leaf to the right
 *     path leaf to make room for an empty extent in the left path leaf.
 *   - When our insert into the right path leaf is at the leftmost edge
 *     and requires an update of the path immediately to it's left. This
 *     can occur at the end of some types of rotation and appending inserts.
1452
1453
 *   - When we've adjusted the last extent record in the left path leaf and the
 *     1st extent record in the right path leaf during cross extent block merge.
Mark Fasheh's avatar
Mark Fasheh committed
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
 */
static void ocfs2_complete_edge_insert(struct inode *inode, handle_t *handle,
				       struct ocfs2_path *left_path,
				       struct ocfs2_path *right_path,
				       int subtree_index)
{
	int ret, i, idx;
	struct ocfs2_extent_list *el, *left_el, *right_el;
	struct ocfs2_extent_rec *left_rec, *right_rec;
	struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;

	/*
	 * Update the counts and position values within all the
	 * interior nodes to reflect the leaf rotation we just did.
	 *
	 * The root node is handled below the loop.
	 *
	 * We begin the loop with right_el and left_el pointing to the
	 * leaf lists and work our way up.
	 *
	 * NOTE: within this loop, left_el and right_el always refer
	 * to the *child* lists.
	 */
	left_el = path_leaf_el(left_path);
	right_el = path_leaf_el(right_path);
	for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) {
		mlog(0, "Adjust records at index %u\n", i);

		/*
		 * One nice property of knowing that all of these
		 * nodes are below the root is that we only deal with
		 * the leftmost right node record and the rightmost
		 * left node record.
		 */
		el = left_path->p_node[i].el;
		idx = le16_to_cpu(left_el->l_next_free_rec) - 1;
		left_rec = &el->l_recs[idx];

		el = right_path->p_node[i].el;
		right_rec = &el->l_recs[0];

		ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec,
					      right_el);

		ret = ocfs2_journal_dirty(handle, left_path->p_node[i].bh);
		if (ret)
			mlog_errno(ret);

		ret = ocfs2_journal_dirty(handle, right_path->p_node[i].bh);
		if (ret)
			mlog_errno(ret);

		/*
		 * Setup our list pointers now so that the current
		 * parents become children in the next iteration.
		 */
		left_el = left_path->p_node[i].el;
		right_el = right_path->p_node[i].el;
	}

	/*
	 * At the root node, adjust the two adjacent records which
	 * begin our path to the leaves.
	 */

	el = left_path->p_node[subtree_index].el;
	left_el = left_path->p_node[subtree_index + 1].el;
	right_el = right_path->p_node[subtree_index + 1].el;

	ocfs2_adjust_root_records(el, left_el, right_el,
				  left_path->p_node[subtree_index + 1].bh->b_blocknr);

	root_bh = left_path->p_node[subtree_index].bh;

	ret = ocfs2_journal_dirty(handle, root_bh);
	if (ret)
		mlog_errno(ret);
}

static int ocfs2_rotate_subtree_right(struct inode *inode,
				      handle_t *handle,
				      struct ocfs2_path *left_path,
				      struct ocfs2_path *right_path,
				      int subtree_index)
{
	int ret, i;
	struct buffer_head *right_leaf_bh;
	struct buffer_head *left_leaf_bh = NULL;
	struct buffer_head *root_bh;
	struct ocfs2_extent_list *right_el, *left_el;
	struct ocfs2_extent_rec move_rec;

	left_leaf_bh = path_leaf_bh(left_path);
	left_el = path_leaf_el(left_path);

	if (left_el->l_next_free_rec != left_el->l_count) {
		ocfs2_error(inode->i_sb,
			    "Inode %llu has non-full interior leaf node %llu"
			    "(next free = %u)",
			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
			    (unsigned long long)left_leaf_bh->b_blocknr,
			    le16_to_cpu(left_el->l_next_free_rec));
		return -EROFS;
	}

	/*
	 * This extent block may already have an empty record, so we
	 * return early if so.
	 */
	if (ocfs2_is_empty_extent(&left_el->l_recs[0]))
		return 0;

	root_bh = left_path->p_node[subtree_index].bh;
	BUG_ON(root_bh != right_path->p_node[subtree_index].bh);

	ret = ocfs2_journal_access(handle, inode, root_bh,
				   OCFS2_JOURNAL_ACCESS_WRITE);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
		ret = ocfs2_journal_access(handle, inode,
					   right_path->p_node[i].bh,
					   OCFS2_JOURNAL_ACCESS_WRITE);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		ret = ocfs2_journal_access(handle, inode,
					   left_path->p_node[i].bh,
					   OCFS2_JOURNAL_ACCESS_WRITE);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

	right_leaf_bh = path_leaf_bh(right_path);
	right_el = path_leaf_el(right_path);

	/* This is a code error, not a disk corruption. */
	mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails "
			"because rightmost leaf block %llu is empty\n",
			(unsigned long long)OCFS2_I(inode)->ip_blkno,
			(unsigned long long)right_leaf_bh->b_blocknr);

	ocfs2_create_empty_extent(right_el);

	ret = ocfs2_journal_dirty(handle, right_leaf_bh);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	/* Do the copy now. */
	i = le16_to_cpu(left_el->l_next_free_rec) - 1;
	move_rec = left_el->l_recs[i];
	right_el->l_recs[0] = move_rec;

	/*
	 * Clear out the record we just copied and shift everything
	 * over, leaving an empty extent in the left leaf.
	 *
	 * We temporarily subtract from next_free_rec so that the
	 * shift will lose the tail record (which is now defunct).
	 */
	le16_add_cpu(&left_el->l_next_free_rec, -1);
	ocfs2_shift_records_right(left_el);
	memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
	le16_add_cpu(&left_el->l_next_free_rec, 1);

	ret = ocfs2_journal_dirty(handle, left_leaf_bh);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	ocfs2_complete_edge_insert(inode, handle, left_path, right_path,
				subtree_index);

out:
	return ret;
}

/*
 * Given a full path, determine what cpos value would return us a path
 * containing the leaf immediately to the left of the current one.
 *
 * Will return zero if the path passed in is already the leftmost path.
 */
static int ocfs2_find_cpos_for_left_leaf(struct super_block *sb,
					 struct ocfs2_path *path, u32 *cpos)
{
	int i, j, ret = 0;
	u64 blkno;
	struct ocfs2_extent_list *el;

1654
1655
	BUG_ON(path->p_tree_depth == 0);

Mark Fasheh's avatar
Mark Fasheh committed
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
	*cpos = 0;

	blkno = path_leaf_bh(path)->b_blocknr;

	/* Start at the tree node just above the leaf and work our way up. */
	i = path->p_tree_depth - 1;
	while (i >= 0) {
		el = path->p_node[i].el;

		/*
		 * Find the extent record just before the one in our
		 * path.
		 */
		for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
			if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
				if (j == 0) {
					if (i == 0) {
						/*
						 * We've determined that the
						 * path specified is already
						 * the leftmost one - return a
						 * cpos of zero.
						 */
						goto out;
					}
					/*
					 * The leftmost record points to our
					 * leaf - we need to travel up the
					 * tree one level.
					 */
					goto next_node;
				}

				*cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos);
1690
1691
1692
				*cpos = *cpos + ocfs2_rec_clusters(el,
							   &el->l_recs[j - 1]);
				*cpos = *cpos - 1;
Mark Fasheh's avatar
Mark Fasheh committed
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
				goto out;
			}
		}

		/*
		 * If we got here, we never found a valid node where
		 * the tree indicated one should be.
		 */
		ocfs2_error(sb,
			    "Invalid extent tree at extent block %llu\n",
			    (unsigned long long)blkno);
		ret = -EROFS;
		goto out;

next_node:
		blkno = path->p_node[i].bh->b_blocknr;
		i--;
	}

out:
	return ret;
}

1716
1717
1718
1719
1720
/*
 * Extend the transaction by enough credits to complete the rotation,
 * and still leave at least the original number of credits allocated
 * to this transaction.
 */
Mark Fasheh's avatar
Mark Fasheh committed
1721
static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth,
1722
					   int op_credits,
Mark Fasheh's avatar
Mark Fasheh committed
1723
1724
					   struct ocfs2_path *path)
{
1725
	int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits;
Mark Fasheh's avatar
Mark Fasheh committed
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758

	if (handle->h_buffer_credits < credits)
		return ocfs2_extend_trans(handle, credits);

	return 0;
}

/*
 * Trap the case where we're inserting into the theoretical range past
 * the _actual_ left leaf range. Otherwise, we'll rotate a record
 * whose cpos is less than ours into the right leaf.
 *
 * It's only necessary to look at the rightmost record of the left
 * leaf because the logic that calls us should ensure that the
 * theoretical ranges in the path components above the leaves are
 * correct.
 */
static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path,
						 u32 insert_cpos)
{
	struct ocfs2_extent_list *left_el;
	struct ocfs2_extent_rec *rec;
	int next_free;

	left_el = path_leaf_el(left_path);
	next_free = le16_to_cpu(left_el->l_next_free_rec);
	rec = &left_el->l_recs[next_free - 1];

	if (insert_cpos > le32_to_cpu(rec->e_cpos))
		return 1;
	return 0;
}

1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos)
{
	int next_free = le16_to_cpu(el->l_next_free_rec);
	unsigned int range;
	struct ocfs2_extent_rec *rec;

	if (next_free == 0)
		return 0;

	rec = &el->l_recs[0];
	if (ocfs2_is_empty_extent(rec)) {
		/* Empty list. */
		if (next_free == 1)
			return 0;
		rec = &el->l_recs[1];
	}

	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
	if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
		return 1;
	return 0;
}

Mark Fasheh's avatar
Mark Fasheh committed
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
/*
 * Rotate all the records in a btree right one record, starting at insert_cpos.
 *
 * The path to the rightmost leaf should be passed in.
 *
 * The array is assumed to be large enough to hold an entire path (tree depth).
 *
 * Upon succesful return from this function:
 *
 * - The 'right_path' array will contain a path to the leaf block
 *   whose range contains e_cpos.
 * - That leaf block will have a single empty extent in list index 0.
 * - In the case that the rotation requires a post-insert update,
 *   *ret_left_path will contain a valid path which can be passed to
 *   ocfs2_insert_path().
 */
static int ocfs2_rotate_tree_right(struct inode *inode,
				   handle_t *handle,
1800
				   enum ocfs2_split_type split,
Mark Fasheh's avatar
Mark Fasheh committed
1801
1802
1803
1804
				   u32 insert_cpos,
				   struct ocfs2_path *right_path,
				   struct ocfs2_path **ret_left_path)
{
1805
	int ret, start, orig_credits = handle->h_buffer_credits;
Mark Fasheh's avatar
Mark Fasheh committed
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
	u32 cpos;
	struct ocfs2_path *left_path = NULL;

	*ret_left_path = NULL;

	left_path = ocfs2_new_path(path_root_bh(right_path),
				   path_root_el(right_path));
	if (!left_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

	ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path, &cpos);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	mlog(0, "Insert: %u, first left path cpos: %u\n", insert_cpos, cpos);

	/*
	 * What we want to do here is:
	 *
	 * 1) Start with the rightmost path.
	 *
	 * 2) Determine a path to the leaf block directly to the left
	 *    of that leaf.
	 *
	 * 3) Determine the 'subtree root' - the lowest level tree node
	 *    which contains a path to both leaves.
	 *
	 * 4) Rotate the subtree.
	 *
	 * 5) Find the next subtree by considering the left path to be
	 *    the new right path.
	 *
	 * The check at the top of this while loop also accepts
	 * insert_cpos == cpos because cpos is only a _theoretical_
	 * value to get us the left path - insert_cpos might very well
	 * be filling that hole.
	 *
	 * Stop at a cpos of '0' because we either started at the
	 * leftmost branch (i.e., a tree with one branch and a
	 * rotation inside of it), or we've gone as far as we can in
	 * rotating subtrees.
	 */
	while (cpos && insert_cpos <= cpos) {
		mlog(0, "Rotating a tree: ins. cpos: %u, left path cpos: %u\n",
		     insert_cpos, cpos);

		ret = ocfs2_find_path(inode, left_path, cpos);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		mlog_bug_on_msg(path_leaf_bh(left_path) ==
				path_leaf_bh(right_path),
				"Inode %lu: error during insert of %u "
				"(left path cpos %u) results in two identical "
				"paths ending at %llu\n",
				inode->i_ino, insert_cpos, cpos,
				(unsigned long long)
				path_leaf_bh(left_path)->b_blocknr);

1872
1873
		if (split == SPLIT_NONE &&
		    ocfs2_rotate_requires_path_adjustment(left_path,
Mark Fasheh's avatar
Mark Fasheh committed
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
							  insert_cpos)) {

			/*
			 * We've rotated the tree as much as we
			 * should. The rest is up to
			 * ocfs2_insert_path() to complete, after the
			 * record insertion. We indicate this
			 * situation by returning the left path.
			 *
			 * The reason we don't adjust the records here
			 * before the record insert is that an error
			 * later might break the rule where a parent
			 * record e_cpos will reflect the actual
			 * e_cpos of the 1st nonempty record of the
			 * child list.
			 */
			*ret_left_path = left_path;
			goto out_ret_path;
		}

		start = ocfs2_find_subtree_root(inode, left_path, right_path);

		mlog(0, "Subtree root at index %d (blk %llu, depth %d)\n",
		     start,
		     (unsigned long long) right_path->p_node[start].bh->b_blocknr,
		     right_path->p_tree_depth);

		ret = ocfs2_extend_rotate_transaction(handle, start,
1902
						      orig_credits, right_path);
Mark Fasheh's avatar
Mark Fasheh committed
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		ret = ocfs2_rotate_subtree_right(inode, handle, left_path,
						 right_path, start);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
		if (split != SPLIT_NONE &&
		    ocfs2_leftmost_rec_contains(path_leaf_el(right_path),
						insert_cpos)) {
			/*
			 * A rotate moves the rightmost left leaf
			 * record over to the leftmost right leaf
			 * slot. If we're doing an extent split
			 * instead of a real insert, then we have to
			 * check that the extent to be split wasn't
			 * just moved over. If it was, then we can
			 * exit here, passing left_path back -
			 * ocfs2_split_extent() is smart enough to
			 * search both leaves.
			 */
			*ret_left_path = left_path;
			goto out_ret_path;
		}

Mark Fasheh's avatar
Mark Fasheh committed
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
		/*
		 * There is no need to re-read the next right path
		 * as we know that it'll be our current left
		 * path. Optimize by copying values instead.
		 */
		ocfs2_mv_path(right_path, left_path);

		ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path,
						    &cpos);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

out:
	ocfs2_free_path(left_path);

out_ret_path:
	return ret;
}

1955
1956
static void ocfs2_update_edge_lengths(struct inode *inode, handle_t *handle,
				      struct ocfs2_path *path)
Mark Fasheh's avatar
Mark Fasheh committed
1957
{
1958
	int i, idx;
Mark Fasheh's avatar
Mark Fasheh committed
1959
	struct ocfs2_extent_rec *rec;
1960
1961
1962
	struct ocfs2_extent_list *el;
	struct ocfs2_extent_block *eb;
	u32 range;
Mark Fasheh's avatar
Mark Fasheh committed
1963

1964
1965
1966
	/* Path should always be rightmost. */
	eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
	BUG_ON(eb->h_next_leaf_blk != 0ULL);
Mark Fasheh's avatar
Mark Fasheh committed
1967

1968
1969
1970
1971
1972
	el = &eb->h_list;
	BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
	idx = le16_to_cpu(el->l_next_free_rec) - 1;
	rec = &el->l_recs[idx];
	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
Mark Fasheh's avatar
Mark Fasheh committed
1973

1974
1975
1976
1977
	for (i = 0; i < path->p_tree_depth; i++) {
		el = path->p_node[i].el;
		idx = le16_to_cpu(el->l_next_free_rec) - 1;
		rec = &el->l_recs[idx];
Mark Fasheh's avatar
Mark Fasheh committed
1978

1979
1980
		rec->e_int_clusters = cpu_to_le32(range);
		le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos));
Mark Fasheh's avatar
Mark Fasheh committed
1981

1982
		ocfs2_journal_dirty(handle, path->p_node[i].bh);
Mark Fasheh's avatar
Mark Fasheh committed
1983
1984
1985
	}
}

1986
1987
1988
static void ocfs2_unlink_path(struct inode *inode, handle_t *handle,
			      struct ocfs2_cached_dealloc_ctxt *dealloc,
			      struct ocfs2_path *path, int unlink_start)
Mark Fasheh's avatar
Mark Fasheh committed
1989
{
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
	int ret, i;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;
	struct buffer_head *bh;

	for(i = unlink_start; i < path_num_items(path); i++) {
		bh = path->p_node[i].bh;

		eb = (struct ocfs2_extent_block *)bh->b_data;
		/*
		 * Not all nodes might have had their final count
		 * decremented by the caller - handle this here.
		 */
		el = &eb->h_list;
		if (le16_to_cpu(el->l_next_free_rec) > 1) {
			mlog(ML_ERROR,
			     "Inode %llu, attempted to remove extent block "
			     "%llu with %u records\n",
			     (unsigned long long)OCFS2_I(inode)->ip_blkno,
			     (unsigned long long)le64_to_cpu(eb->h_blkno),
			     le16_to_cpu(el->l_next_free_rec));

			ocfs2_journal_dirty(handle, bh);
			ocfs2_remove_from_cache(inode, bh);
			continue;
		}

		el->l_next_free_rec = 0;
		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));

		ocfs2_journal_dirty(handle, bh);

		ret = ocfs2_cache_extent_block_free(dealloc, eb);
		if (ret)
			mlog_errno(ret);

		ocfs2_remove_from_cache(inode, bh);
	}
Mark Fasheh's avatar
Mark Fasheh committed
2028
2029
}

2030
2031
2032
2033
2034
static void ocfs2_unlink_subtree(struct inode *inode, handle_t *handle,
				 struct ocfs2_path *left_path,
				 struct ocfs2_path *right_path,
				 int subtree_index,
				 struct ocfs2_cached_dealloc_ctxt *dealloc)
Mark Fasheh's avatar
Mark Fasheh committed
2035
{
2036
2037
2038
	int i;
	struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
	struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el;
Mark Fasheh's avatar
Mark Fasheh committed
2039
	struct ocfs2_extent_list *el;
2040
	struct ocfs2_extent_block *eb;
Mark Fasheh's avatar
Mark Fasheh committed
2041

2042
	el = path_leaf_el(left_path);
Mark Fasheh's avatar
Mark Fasheh committed
2043

2044
	eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data;
2045

2046
2047
2048
	for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
		if (root_el->l_recs[i].e_blkno == eb->h_blkno)
			break;
Mark Fasheh's avatar
Mark Fasheh committed
2049

2050
	BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec));
Mark Fasheh's avatar
Mark Fasheh committed
2051

2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
	memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
	le16_add_cpu(&root_el->l_next_free_rec, -1);

	eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
	eb->h_next_leaf_blk = 0;

	ocfs2_journal_dirty(handle, root_bh);
	ocfs2_journal_dirty(handle, path_leaf_bh(left_path));

	ocfs2_unlink_path(inode, handle, dealloc, right_path,
			  subtree_index + 1);
}

static int ocfs2_rotate_subtree_left(struct inode *inode, handle_t *handle,
				     struct ocfs2_path *left_path,
				     struct ocfs2_path *right_path,
				     int subtree_index,
				     struct ocfs2_cached_dealloc_ctxt *dealloc,
				     int *deleted)
{
	int ret, i, del_right_subtree = 0, right_has_empty = 0;
	struct buffer_head *root_bh, *di_bh = path_root_bh(right_path);
	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
	struct ocfs2_extent_list *right_leaf_el, *left_leaf_el;
	struct ocfs2_extent_block *eb;

	*deleted = 0;

	right_leaf_el = path_leaf_el(right_path);
	left_leaf_el = path_leaf_el(left_path);
	root_bh = left_path->p_node[subtree_index].bh;
	BUG_ON(root_bh != right_path->p_node[subtree_index].bh);

	if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0]))
		return 0;
Mark Fasheh's avatar
Mark Fasheh committed
2087

2088
2089
	eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data;
	if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) {
Mark Fasheh's avatar
Mark Fasheh committed
2090
		/*
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
		 * It's legal for us to proceed if the right leaf is
		 * the rightmost one and it has an empty extent. There
		 * are two cases to handle - whether the leaf will be
		 * empty after removal or not. If the leaf isn't empty
		 * then just remove the empty extent up front. The
		 * next block will handle empty leaves by flagging
		 * them for unlink.
		 *
		 * Non rightmost leaves will throw -EAGAIN and the
		 * caller can manually move the subtree and retry.
Mark Fasheh's avatar
Mark Fasheh committed
2101
2102
		 */

2103
2104
2105
2106
2107
2108
2109
		if (eb->h_next_leaf_blk != 0ULL)
			return -EAGAIN;

		if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) {
			ret = ocfs2_journal_access(handle, inode,
						   path_leaf_bh(right_path),
						   OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh's avatar
Mark Fasheh committed
2110
2111
2112
2113
2114
			if (ret) {
				mlog_errno(ret);
				goto out;
			}

2115
2116
2117
			ocfs2_remove_empty_extent(right_leaf_el);
		} else
			right_has_empty = 1;
Mark Fasheh's avatar
Mark Fasheh committed
2118
2119
	}

2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210