alloc.c 171.4 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"

52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188
/*
 * ocfs2_extent_tree and ocfs2_extent_tree_operations are used to abstract
 * the b-tree operations in ocfs2. Now all the b-tree operations are not
 * limited to ocfs2_dinode only. Any data which need to allocate clusters
 * to store can use b-tree. And it only needs to implement its ocfs2_extent_tree
 * and operation.
 *
 * ocfs2_extent_tree contains info for the root of the b-tree, it must have a
 * root ocfs2_extent_list and a root_bh so that they can be used in the b-tree
 * functions.
 * ocfs2_extent_tree_operations abstract the normal operations we do for
 * the root of extent b-tree.
 */
struct ocfs2_extent_tree;

struct ocfs2_extent_tree_operations {
	void (*set_last_eb_blk) (struct ocfs2_extent_tree *et, u64 blkno);
	u64 (*get_last_eb_blk) (struct ocfs2_extent_tree *et);
	void (*update_clusters) (struct inode *inode,
				 struct ocfs2_extent_tree *et,
				 u32 new_clusters);
	int (*sanity_check) (struct inode *inode, struct ocfs2_extent_tree *et);
};

struct ocfs2_extent_tree {
	enum ocfs2_extent_tree_type type;
	struct ocfs2_extent_tree_operations *eops;
	struct buffer_head *root_bh;
	struct ocfs2_extent_list *root_el;
};

static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
					 u64 blkno)
{
	struct ocfs2_dinode *di = (struct ocfs2_dinode *)et->root_bh->b_data;

	BUG_ON(et->type != OCFS2_DINODE_EXTENT);
	di->i_last_eb_blk = cpu_to_le64(blkno);
}

static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et)
{
	struct ocfs2_dinode *di = (struct ocfs2_dinode *)et->root_bh->b_data;

	BUG_ON(et->type != OCFS2_DINODE_EXTENT);
	return le64_to_cpu(di->i_last_eb_blk);
}

static void ocfs2_dinode_update_clusters(struct inode *inode,
					 struct ocfs2_extent_tree *et,
					 u32 clusters)
{
	struct ocfs2_dinode *di =
			(struct ocfs2_dinode *)et->root_bh->b_data;

	le32_add_cpu(&di->i_clusters, clusters);
	spin_lock(&OCFS2_I(inode)->ip_lock);
	OCFS2_I(inode)->ip_clusters = le32_to_cpu(di->i_clusters);
	spin_unlock(&OCFS2_I(inode)->ip_lock);
}

static int ocfs2_dinode_sanity_check(struct inode *inode,
				     struct ocfs2_extent_tree *et)
{
	int ret = 0;
	struct ocfs2_dinode *di;

	BUG_ON(et->type != OCFS2_DINODE_EXTENT);

	di = (struct ocfs2_dinode *)et->root_bh->b_data;
	if (!OCFS2_IS_VALID_DINODE(di)) {
		ret = -EIO;
		ocfs2_error(inode->i_sb,
			"Inode %llu has invalid path root",
			(unsigned long long)OCFS2_I(inode)->ip_blkno);
	}

	return ret;
}

static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = {
	.set_last_eb_blk	= ocfs2_dinode_set_last_eb_blk,
	.get_last_eb_blk	= ocfs2_dinode_get_last_eb_blk,
	.update_clusters	= ocfs2_dinode_update_clusters,
	.sanity_check		= ocfs2_dinode_sanity_check,
};

static struct ocfs2_extent_tree*
	 ocfs2_new_extent_tree(struct buffer_head *bh,
			       enum ocfs2_extent_tree_type et_type)
{
	struct ocfs2_extent_tree *et;

	et = kzalloc(sizeof(*et), GFP_NOFS);
	if (!et)
		return NULL;

	et->type = et_type;
	get_bh(bh);
	et->root_bh = bh;

	/* current we only support dinode extent. */
	BUG_ON(et->type != OCFS2_DINODE_EXTENT);
	if (et_type == OCFS2_DINODE_EXTENT) {
		et->root_el = &((struct ocfs2_dinode *)bh->b_data)->id2.i_list;
		et->eops = &ocfs2_dinode_et_ops;
	}

	return et;
}

static void ocfs2_free_extent_tree(struct ocfs2_extent_tree *et)
{
	if (et) {
		brelse(et->root_bh);
		kfree(et);
	}
}

static inline void ocfs2_set_last_eb_blk(struct ocfs2_extent_tree *et,
					 u64 new_last_eb_blk)
{
	et->eops->set_last_eb_blk(et, new_last_eb_blk);
}

static inline u64 ocfs2_get_last_eb_blk(struct ocfs2_extent_tree *et)
{
	return et->eops->get_last_eb_blk(et);
}

static inline void ocfs2_update_clusters(struct inode *inode,
					 struct ocfs2_extent_tree *et,
					 u32 clusters)
{
	et->eops->update_clusters(inode, et, clusters);
}

M
Mark Fasheh 已提交
189
static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc);
190 191
static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
					 struct ocfs2_extent_block *eb);
192

M
Mark Fasheh 已提交
193 194 195 196 197 198 199 200 201 202 203
/*
 * 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;
};
204

M
Mark Fasheh 已提交
205
#define OCFS2_MAX_PATH_DEPTH	5
206

M
Mark Fasheh 已提交
207 208 209 210
struct ocfs2_path {
	int			p_tree_depth;
	struct ocfs2_path_item	p_node[OCFS2_MAX_PATH_DEPTH];
};
211

M
Mark Fasheh 已提交
212 213 214 215 216
#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)
217

M
Mark Fasheh 已提交
218 219 220 221 222 223 224 225 226
/*
 * 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;
227

M
Mark Fasheh 已提交
228 229
	if (keep_root)
		start = 1;
230

M
Mark Fasheh 已提交
231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257
	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);
	}
}

258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282
/*
 * 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);
	}
}

M
Mark Fasheh 已提交
283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329
/*
 * 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;
330

M
Mark Fasheh 已提交
331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367
	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;
}

/*
 * 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;
}

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
/*
 * 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;
}

M
Mark Fasheh 已提交
398 399 400
enum ocfs2_contig_type {
	CONTIG_NONE = 0,
	CONTIG_LEFT,
401 402
	CONTIG_RIGHT,
	CONTIG_LEFTRIGHT,
M
Mark Fasheh 已提交
403 404
};

405 406 407 408 409

/*
 * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and
 * ocfs2_extent_contig only work properly against leaf nodes!
 */
M
Mark Fasheh 已提交
410 411 412
static int ocfs2_block_extent_contig(struct super_block *sb,
				     struct ocfs2_extent_rec *ext,
				     u64 blkno)
413
{
414 415 416 417 418 419
	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;
420 421
}

M
Mark Fasheh 已提交
422 423 424
static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left,
				  struct ocfs2_extent_rec *right)
{
425 426 427 428 429 430
	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));
M
Mark Fasheh 已提交
431 432 433 434 435 436 437 438 439
}

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);

440 441 442 443 444 445 446 447
	/*
	 * 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;

M
Mark Fasheh 已提交
448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471
	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,
};

472 473 474 475 476 477
enum ocfs2_split_type {
	SPLIT_NONE = 0,
	SPLIT_LEFT,
	SPLIT_RIGHT,
};

M
Mark Fasheh 已提交
478
struct ocfs2_insert_type {
479
	enum ocfs2_split_type	ins_split;
M
Mark Fasheh 已提交
480 481 482 483 484 485
	enum ocfs2_append_type	ins_appending;
	enum ocfs2_contig_type	ins_contig;
	int			ins_contig_index;
	int			ins_tree_depth;
};

486 487 488 489 490 491
struct ocfs2_merge_ctxt {
	enum ocfs2_contig_type	c_contig_type;
	int			c_has_empty_extent;
	int			c_split_covers_rec;
};

492 493 494 495 496
/*
 * 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,
497 498
			   struct buffer_head *root_bh,
			   enum ocfs2_extent_tree_type type)
499 500
{
	int retval;
501
	struct ocfs2_extent_list *el = NULL;
502 503
	struct ocfs2_extent_block *eb;
	struct buffer_head *eb_bh = NULL;
504
	u64 last_eb_blk = 0;
505 506 507

	mlog_entry_void();

508 509 510 511 512 513 514 515 516 517 518 519
	if (type == OCFS2_DINODE_EXTENT) {
		struct ocfs2_dinode *fe =
				(struct ocfs2_dinode *)root_bh->b_data;
		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)
			last_eb_blk = le64_to_cpu(fe->i_last_eb_blk);
		el = &fe->id2.i_list;
520 521
	}

522 523
	if (last_eb_blk) {
		retval = ocfs2_read_block(osb, last_eb_blk,
524 525 526 527 528 529 530
					  &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;
531
	}
532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549

	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,
550
				     handle_t *handle,
551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632
				     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;
}

M
Mark Fasheh 已提交
633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651
/*
 * 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) +
652
		ocfs2_rec_clusters(el, &el->l_recs[i]);
M
Mark Fasheh 已提交
653 654
}

655 656 657 658 659 660 661 662 663
/*
 * 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
664
 * contain a single record with cluster count == 0.
665 666
 */
static int ocfs2_add_branch(struct ocfs2_super *osb,
667
			    handle_t *handle,
668
			    struct inode *inode,
669
			    struct ocfs2_extent_tree *et,
670
			    struct buffer_head *eb_bh,
671
			    struct buffer_head **last_eb_bh,
672 673 674 675 676 677 678 679 680
			    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_extent_block *eb;
	struct ocfs2_extent_list  *eb_el;
	struct ocfs2_extent_list  *el;
M
Mark Fasheh 已提交
681
	u32 new_cpos;
682 683 684

	mlog_entry_void();

685
	BUG_ON(!last_eb_bh || !*last_eb_bh);
686 687 688 689 690

	if (eb_bh) {
		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
		el = &eb->h_list;
	} else
691
		el = et->root_el;
692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713

	/* 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;
	}

714
	eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data;
M
Mark Fasheh 已提交
715 716
	new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list);

717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744
	/* 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);
M
Mark Fasheh 已提交
745 746 747 748 749
		/*
		 * This actually counts as an empty extent as
		 * c_clusters == 0
		 */
		eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos);
750
		eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno);
751 752 753 754 755 756
		/*
		 * 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);
757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774
		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. */
775
	status = ocfs2_journal_access(handle, inode, *last_eb_bh,
776 777 778 779 780
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
781
	status = ocfs2_journal_access(handle, inode, et->root_bh,
782 783 784 785 786 787 788 789 790 791 792 793 794 795 796
				      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
797
	 * either be on the root_bh, or the extent block passed in. */
798 799
	i = le16_to_cpu(el->l_next_free_rec);
	el->l_recs[i].e_blkno = cpu_to_le64(next_blkno);
M
Mark Fasheh 已提交
800
	el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
801
	el->l_recs[i].e_int_clusters = 0;
802 803 804 805
	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. */
806
	ocfs2_set_last_eb_blk(et, new_last_eb_blk);
807

808
	eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
809 810
	eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);

811
	status = ocfs2_journal_dirty(handle, *last_eb_bh);
812 813
	if (status < 0)
		mlog_errno(status);
814
	status = ocfs2_journal_dirty(handle, et->root_bh);
815 816 817 818 819 820 821 822
	if (status < 0)
		mlog_errno(status);
	if (eb_bh) {
		status = ocfs2_journal_dirty(handle, eb_bh);
		if (status < 0)
			mlog_errno(status);
	}

823 824 825 826 827 828 829 830
	/*
	 * 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];

831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849
	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,
850
				  handle_t *handle,
851
				  struct inode *inode,
852
				  struct ocfs2_extent_tree *et,
853 854 855 856
				  struct ocfs2_alloc_context *meta_ac,
				  struct buffer_head **ret_new_eb_bh)
{
	int status, i;
M
Mark Fasheh 已提交
857
	u32 new_clusters;
858 859
	struct buffer_head *new_eb_bh = NULL;
	struct ocfs2_extent_block *eb;
860
	struct ocfs2_extent_list  *root_el;
861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879
	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;
880
	root_el = et->root_el;
881 882 883 884 885 886 887 888

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

889 890 891 892 893
	/* copy the root extent list data into the new extent block */
	eb_el->l_tree_depth = root_el->l_tree_depth;
	eb_el->l_next_free_rec = root_el->l_next_free_rec;
	for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++)
		eb_el->l_recs[i] = root_el->l_recs[i];
894 895 896 897 898 899 900

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

901
	status = ocfs2_journal_access(handle, inode, et->root_bh,
902 903 904 905 906 907
				      OCFS2_JOURNAL_ACCESS_WRITE);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

M
Mark Fasheh 已提交
908 909
	new_clusters = ocfs2_sum_rightmost_rec(eb_el);

910 911 912 913 914 915 916 917
	/* update root_bh now */
	le16_add_cpu(&root_el->l_tree_depth, 1);
	root_el->l_recs[0].e_cpos = 0;
	root_el->l_recs[0].e_blkno = eb->h_blkno;
	root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters);
	for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
		memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
	root_el->l_next_free_rec = cpu_to_le16(1);
918 919 920

	/* If this is our 1st tree depth shift, then last_eb_blk
	 * becomes the allocated extent block */
921 922
	if (root_el->l_tree_depth == cpu_to_le16(1))
		ocfs2_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
923

924
	status = ocfs2_journal_dirty(handle, et->root_bh);
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
	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'
 *
950
 * 2) the search fails to find anything, but the root_el has room. We
951 952
 *    pass NULL back in *lowest_eb_bh, but still return '0'
 *
953
 * 3) the search fails to find anything AND the root_el is full, in
954 955 956 957 958 959
 *    which case we return > 0
 *
 * return status < 0 indicates an error.
 */
static int ocfs2_find_branch_target(struct ocfs2_super *osb,
				    struct inode *inode,
960
				    struct ocfs2_extent_tree *et,
961 962 963 964 965 966 967 968 969 970 971 972 973
				    struct buffer_head **target_bh)
{
	int status = 0, i;
	u64 blkno;
	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;

974
	el = et->root_el;
975 976 977

	while(le16_to_cpu(el->l_tree_depth) > 1) {
		if (le16_to_cpu(el->l_next_free_rec) == 0) {
978
			ocfs2_error(inode->i_sb, "Dinode %llu has empty "
979
				    "extent list (next_free_rec == 0)",
980
				    (unsigned long long)OCFS2_I(inode)->ip_blkno);
981 982 983 984 985 986
			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) {
987
			ocfs2_error(inode->i_sb, "Dinode %llu has extent "
988 989
				    "list where extent # %d has no physical "
				    "block start",
990
				    (unsigned long long)OCFS2_I(inode)->ip_blkno, i);
991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005
			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;
		}
M
Mark Fasheh 已提交
1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025

		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' */
1026 1027
	el = et->root_el;
	if (!lowest_bh && (el->l_next_free_rec == el->l_count))
M
Mark Fasheh 已提交
1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038
		status = 1;

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

	mlog_exit(status);
	return status;
}

1039 1040 1041 1042 1043 1044 1045
/*
 * 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.
1046 1047
 *
 * *last_eb_bh will be updated by ocfs2_add_branch().
1048 1049
 */
static int ocfs2_grow_tree(struct inode *inode, handle_t *handle,
1050
			   struct ocfs2_extent_tree *et, int *final_depth,
1051
			   struct buffer_head **last_eb_bh,
1052 1053 1054
			   struct ocfs2_alloc_context *meta_ac)
{
	int ret, shift;
1055 1056
	struct ocfs2_extent_list *el = et->root_el;
	int depth = le16_to_cpu(el->l_tree_depth);
1057 1058 1059 1060 1061
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
	struct buffer_head *bh = NULL;

	BUG_ON(meta_ac == NULL);

1062
	shift = ocfs2_find_branch_target(osb, inode, et, &bh);
1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078
	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). */
1079
		ret = ocfs2_shift_tree_depth(osb, handle, inode, et,
1080 1081 1082 1083 1084 1085
					     meta_ac, &bh);
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}
		depth++;
1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098
		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;
1099
			goto out;
1100
		}
1101 1102 1103 1104 1105
	}

	/* call ocfs2_add_branch to add the final part of the tree with
	 * the new data. */
	mlog(0, "add branch. bh = %p\n", bh);
1106
	ret = ocfs2_add_branch(osb, handle, inode, et, bh, last_eb_bh,
1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119
			       meta_ac);
	if (ret < 0) {
		mlog_errno(ret);
		goto out;
	}

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

M
Mark Fasheh 已提交
1120 1121 1122 1123 1124 1125 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
/*
 * 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. */
J
Julia Lawall 已提交
1151
	BUG_ON(el->l_next_free_rec == el->l_count && !has_empty);
M
Mark Fasheh 已提交
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

	/*
	 * 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;

}

1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231
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);
	}
}

M
Mark Fasheh 已提交
1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242
/*
 * 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);

1243 1244
	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);

M
Mark Fasheh 已提交
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 1344 1345 1346 1347 1348
	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) +
1349
				ocfs2_rec_clusters(el, rec);
M
Mark Fasheh 已提交
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 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 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 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462
			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.
 */
1463 1464
int ocfs2_find_leaf(struct inode *inode, struct ocfs2_extent_list *root_el,
		    u32 cpos, struct buffer_head **leaf_bh)
M
Mark Fasheh 已提交
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
{
	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);
1509 1510 1511 1512
	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);
	}
M
Mark Fasheh 已提交
1513
	left_clusters -= le32_to_cpu(left_rec->e_cpos);
1514
	left_rec->e_int_clusters = cpu_to_le32(left_clusters);
M
Mark Fasheh 已提交
1515 1516 1517

	/*
	 * Calculate the rightmost cluster count boundary before
1518
	 * moving cpos - we will need to adjust clusters after
M
Mark Fasheh 已提交
1519 1520 1521
	 * updating e_cpos to keep the same highest cluster count.
	 */
	right_end = le32_to_cpu(right_rec->e_cpos);
1522
	right_end += le32_to_cpu(right_rec->e_int_clusters);
M
Mark Fasheh 已提交
1523 1524 1525 1526 1527

	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);
1528
	right_rec->e_int_clusters = cpu_to_le32(right_end);
M
Mark Fasheh 已提交
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
}

/*
 * 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.
1571 1572
 *   - 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.
M
Mark Fasheh 已提交
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 1654 1655 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 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 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 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772
 */
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;

1773 1774
	BUG_ON(path->p_tree_depth == 0);

M
Mark Fasheh 已提交
1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808
	*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);
1809 1810 1811
				*cpos = *cpos + ocfs2_rec_clusters(el,
							   &el->l_recs[j - 1]);
				*cpos = *cpos - 1;
M
Mark Fasheh 已提交
1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834
				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;
}

1835 1836 1837 1838 1839
/*
 * Extend the transaction by enough credits to complete the rotation,
 * and still leave at least the original number of credits allocated
 * to this transaction.
 */
M
Mark Fasheh 已提交
1840
static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth,
1841
					   int op_credits,
M
Mark Fasheh 已提交
1842 1843
					   struct ocfs2_path *path)
{
1844
	int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits;
M
Mark Fasheh 已提交
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 1872 1873 1874 1875 1876 1877

	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;
}

1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900
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;
}

M
Mark Fasheh 已提交
1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918
/*
 * 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,
1919
				   enum ocfs2_split_type split,
M
Mark Fasheh 已提交
1920 1921 1922 1923
				   u32 insert_cpos,
				   struct ocfs2_path *right_path,
				   struct ocfs2_path **ret_left_path)
{
1924
	int ret, start, orig_credits = handle->h_buffer_credits;
M
Mark Fasheh 已提交
1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990
	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);

1991 1992
		if (split == SPLIT_NONE &&
		    ocfs2_rotate_requires_path_adjustment(left_path,
M
Mark Fasheh 已提交
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
							  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,
2021
						      orig_credits, right_path);
M
Mark Fasheh 已提交
2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033
		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;
		}

2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051
		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;
		}

M
Mark Fasheh 已提交
2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073
		/*
		 * 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;
}

2074 2075
static void ocfs2_update_edge_lengths(struct inode *inode, handle_t *handle,
				      struct ocfs2_path *path)
M
Mark Fasheh 已提交
2076
{
2077
	int i, idx;
M
Mark Fasheh 已提交
2078
	struct ocfs2_extent_rec *rec;
2079 2080 2081
	struct ocfs2_extent_list *el;
	struct ocfs2_extent_block *eb;
	u32 range;
M
Mark Fasheh 已提交
2082

2083 2084 2085
	/* Path should always be rightmost. */
	eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
	BUG_ON(eb->h_next_leaf_blk != 0ULL);
M
Mark Fasheh 已提交
2086

2087 2088 2089 2090 2091
	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);
M
Mark Fasheh 已提交
2092

2093 2094 2095 2096
	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];
M
Mark Fasheh 已提交
2097

2098 2099
		rec->e_int_clusters = cpu_to_le32(range);
		le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos));
M
Mark Fasheh 已提交
2100

2101
		ocfs2_journal_dirty(handle, path->p_node[i].bh);
M
Mark Fasheh 已提交
2102 2103 2104
	}
}

2105 2106 2107
static void ocfs2_unlink_path(struct inode *inode, handle_t *handle,
			      struct ocfs2_cached_dealloc_ctxt *dealloc,
			      struct ocfs2_path *path, int unlink_start)
M
Mark Fasheh 已提交
2108
{
2109 2110 2111 2112 2113 2114 2115 2116 2117 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
	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);
	}
M
Mark Fasheh 已提交
2147 2148
}

2149 2150 2151 2152 2153
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)
M
Mark Fasheh 已提交
2154
{
2155 2156 2157
	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;
M
Mark Fasheh 已提交
2158
	struct ocfs2_extent_list *el;
2159
	struct ocfs2_extent_block *eb;
M
Mark Fasheh 已提交
2160

2161
	el = path_leaf_el(left_path);
M
Mark Fasheh 已提交
2162

2163
	eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data;
2164

2165 2166 2167
	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;
M
Mark Fasheh 已提交
2168

2169
	BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec));
M
Mark Fasheh 已提交
2170

2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188
	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,
2189 2190
				     int *deleted,
				     struct ocfs2_extent_tree *et)
2191 2192
{
	int ret, i, del_right_subtree = 0, right_has_empty = 0;
2193
	struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path);
2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205
	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;
M
Mark Fasheh 已提交
2206

2207 2208
	eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data;
	if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) {
M
Mark Fasheh 已提交
2209
		/*
2210 2211 2212 2213 2214 2215 2216 2217 2218 2219
		 * 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.
M
Mark Fasheh 已提交
2220 2221
		 */

2222 2223 2224 2225 2226 2227 2228
		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);
M
Mark Fasheh 已提交
2229 2230 2231 2232 2233
			if (ret) {
				mlog_errno(ret);
				goto out;
			}

2234 2235 2236
			ocfs2_remove_empty_extent(right_leaf_el);
		} else
			right_has_empty = 1;
M
Mark Fasheh 已提交
2237 2238
	}

2239 2240 2241 2242 2243 2244
	if (eb->h_next_leaf_blk == 0ULL &&
	    le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) {
		/*
		 * We have to update i_last_eb_blk during the meta
		 * data delete.
		 */
2245
		ret = ocfs2_journal_access(handle, inode, et_root_bh,
2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319
					   OCFS2_JOURNAL_ACCESS_WRITE);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		del_right_subtree = 1;
	}

	/*
	 * Getting here with an empty extent in the right path implies
	 * that it's the rightmost path and will be deleted.
	 */
	BUG_ON(right_has_empty && !del_right_subtree);

	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;
		}
	}

	if (!right_has_empty) {
		/*
		 * Only do this if we're moving a real
		 * record. Otherwise, the action is delayed until
		 * after removal of the right path in which case we
		 * can do a simple shift to remove the empty extent.
		 */
		ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]);
		memset(&right_leaf_el->l_recs[0], 0,
		       sizeof(struct ocfs2_extent_rec));
	}
	if (eb->h_next_leaf_blk == 0ULL) {
		/*
		 * Move recs over to get rid of empty extent, decrease
		 * next_free. This is allowed to remove the last
		 * extent in our leaf (setting l_next_free_rec to
		 * zero) - the delete code below won't care.
		 */
		ocfs2_remove_empty_extent(right_leaf_el);
	}

	ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
	if (ret)
		mlog_errno(ret);
	ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
	if (ret)
		mlog_errno(ret);

	if (del_right_subtree) {
		ocfs2_unlink_subtree(inode, handle, left_path, right_path,
				     subtree_index, dealloc);
		ocfs2_update_edge_lengths(inode, handle, left_path);

		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
2320
		ocfs2_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
2321 2322 2323 2324 2325 2326 2327 2328 2329

		/*
		 * Removal of the extent in the left leaf was skipped
		 * above so we could delete the right path
		 * 1st.
		 */
		if (right_has_empty)
			ocfs2_remove_empty_extent(left_leaf_el);

2330
		ret = ocfs2_journal_dirty(handle, et_root_bh);
2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452
		if (ret)
			mlog_errno(ret);

		*deleted = 1;
	} else
		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 right of the current one.
 *
 * Will return zero if the path passed in is already the rightmost path.
 *
 * This looks similar, but is subtly different to
 * ocfs2_find_cpos_for_left_leaf().
 */
static int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
					  struct ocfs2_path *path, u32 *cpos)
{
	int i, j, ret = 0;
	u64 blkno;
	struct ocfs2_extent_list *el;

	*cpos = 0;

	if (path->p_tree_depth == 0)
		return 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) {
		int next_free;

		el = path->p_node[i].el;

		/*
		 * Find the extent record just after the one in our
		 * path.
		 */
		next_free = le16_to_cpu(el->l_next_free_rec);
		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 == (next_free - 1)) {
					if (i == 0) {
						/*
						 * We've determined that the
						 * path specified is already
						 * the rightmost one - return a
						 * cpos of zero.
						 */
						goto out;
					}
					/*
					 * The rightmost 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);
				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;
}

static int ocfs2_rotate_rightmost_leaf_left(struct inode *inode,
					    handle_t *handle,
					    struct buffer_head *bh,
					    struct ocfs2_extent_list *el)
{
	int ret;

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

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

	ocfs2_remove_empty_extent(el);

	ret = ocfs2_journal_dirty(handle, bh);
	if (ret)
		mlog_errno(ret);

out:
	return ret;
}

static int __ocfs2_rotate_tree_left(struct inode *inode,
				    handle_t *handle, int orig_credits,
				    struct ocfs2_path *path,
				    struct ocfs2_cached_dealloc_ctxt *dealloc,
2453 2454
				    struct ocfs2_path **empty_extent_path,
				    struct ocfs2_extent_tree *et)
2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512
{
	int ret, subtree_root, deleted;
	u32 right_cpos;
	struct ocfs2_path *left_path = NULL;
	struct ocfs2_path *right_path = NULL;

	BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0])));

	*empty_extent_path = NULL;

	ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, path,
					     &right_cpos);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

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

	ocfs2_cp_path(left_path, path);

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

	while (right_cpos) {
		ret = ocfs2_find_path(inode, right_path, right_cpos);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		subtree_root = ocfs2_find_subtree_root(inode, left_path,
						       right_path);

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

		ret = ocfs2_extend_rotate_transaction(handle, subtree_root,
						      orig_credits, left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524
		/*
		 * Caller might still want to make changes to the
		 * tree root, so re-add it to the journal here.
		 */
		ret = ocfs2_journal_access(handle, inode,
					   path_root_bh(left_path),
					   OCFS2_JOURNAL_ACCESS_WRITE);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2525 2526
		ret = ocfs2_rotate_subtree_left(inode, handle, left_path,
						right_path, subtree_root,
2527
						dealloc, &deleted, et);
2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569
		if (ret == -EAGAIN) {
			/*
			 * The rotation has to temporarily stop due to
			 * the right subtree having an empty
			 * extent. Pass it back to the caller for a
			 * fixup.
			 */
			*empty_extent_path = right_path;
			right_path = NULL;
			goto out;
		}
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		/*
		 * The subtree rotate might have removed records on
		 * the rightmost edge. If so, then rotation is
		 * complete.
		 */
		if (deleted)
			break;

		ocfs2_mv_path(left_path, right_path);

		ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, left_path,
						     &right_cpos);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

out:
	ocfs2_free_path(right_path);
	ocfs2_free_path(left_path);

	return ret;
}

static int ocfs2_remove_rightmost_path(struct inode *inode, handle_t *handle,
2570 2571 2572
				struct ocfs2_path *path,
				struct ocfs2_cached_dealloc_ctxt *dealloc,
				struct ocfs2_extent_tree *et)
2573 2574 2575 2576 2577 2578 2579 2580
{
	int ret, subtree_index;
	u32 cpos;
	struct ocfs2_path *left_path = NULL;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;


2581 2582 2583
	ret = et->eops->sanity_check(inode, et);
	if (ret)
		goto out;
2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639
	/*
	 * There's two ways we handle this depending on
	 * whether path is the only existing one.
	 */
	ret = ocfs2_extend_rotate_transaction(handle, 0,
					      handle->h_buffer_credits,
					      path);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	ret = ocfs2_journal_access_path(inode, handle, path);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

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

	if (cpos) {
		/*
		 * We have a path to the left of this one - it needs
		 * an update too.
		 */
		left_path = ocfs2_new_path(path_root_bh(path),
					   path_root_el(path));
		if (!left_path) {
			ret = -ENOMEM;
			mlog_errno(ret);
			goto out;
		}

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

		ret = ocfs2_journal_access_path(inode, handle, left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		subtree_index = ocfs2_find_subtree_root(inode, left_path, path);

		ocfs2_unlink_subtree(inode, handle, left_path, path,
				     subtree_index, dealloc);
		ocfs2_update_edge_lengths(inode, handle, left_path);

		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
2640
		ocfs2_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
2641 2642 2643 2644 2645 2646 2647 2648 2649 2650
	} else {
		/*
		 * 'path' is also the leftmost path which
		 * means it must be the only one. This gets
		 * handled differently because we want to
		 * revert the inode back to having extents
		 * in-line.
		 */
		ocfs2_unlink_path(inode, handle, dealloc, path, 1);

2651
		el = et->root_el;
2652 2653 2654 2655
		el->l_tree_depth = 0;
		el->l_next_free_rec = 0;
		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));

2656
		ocfs2_set_last_eb_blk(et, 0);
2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683
	}

	ocfs2_journal_dirty(handle, path_root_bh(path));

out:
	ocfs2_free_path(left_path);
	return ret;
}

/*
 * Left rotation of btree records.
 *
 * In many ways, this is (unsurprisingly) the opposite of right
 * rotation. We start at some non-rightmost path containing an empty
 * extent in the leaf block. The code works its way to the rightmost
 * path by rotating records to the left in every subtree.
 *
 * This is used by any code which reduces the number of extent records
 * in a leaf. After removal, an empty record should be placed in the
 * leftmost list position.
 *
 * This won't handle a length update of the rightmost path records if
 * the rightmost tree leaf record is removed so the caller is
 * responsible for detecting and correcting that.
 */
static int ocfs2_rotate_tree_left(struct inode *inode, handle_t *handle,
				  struct ocfs2_path *path,
2684 2685
				  struct ocfs2_cached_dealloc_ctxt *dealloc,
				  struct ocfs2_extent_tree *et)
2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698
{
	int ret, orig_credits = handle->h_buffer_credits;
	struct ocfs2_path *tmp_path = NULL, *restart_path = NULL;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;

	el = path_leaf_el(path);
	if (!ocfs2_is_empty_extent(&el->l_recs[0]))
		return 0;

	if (path->p_tree_depth == 0) {
rightmost_no_delete:
		/*
2699
		 * Inline extents. This is trivially handled, so do
2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752
		 * it up front.
		 */
		ret = ocfs2_rotate_rightmost_leaf_left(inode, handle,
						       path_leaf_bh(path),
						       path_leaf_el(path));
		if (ret)
			mlog_errno(ret);
		goto out;
	}

	/*
	 * Handle rightmost branch now. There's several cases:
	 *  1) simple rotation leaving records in there. That's trivial.
	 *  2) rotation requiring a branch delete - there's no more
	 *     records left. Two cases of this:
	 *     a) There are branches to the left.
	 *     b) This is also the leftmost (the only) branch.
	 *
	 *  1) is handled via ocfs2_rotate_rightmost_leaf_left()
	 *  2a) we need the left branch so that we can update it with the unlink
	 *  2b) we need to bring the inode back to inline extents.
	 */

	eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
	el = &eb->h_list;
	if (eb->h_next_leaf_blk == 0) {
		/*
		 * This gets a bit tricky if we're going to delete the
		 * rightmost path. Get the other cases out of the way
		 * 1st.
		 */
		if (le16_to_cpu(el->l_next_free_rec) > 1)
			goto rightmost_no_delete;

		if (le16_to_cpu(el->l_next_free_rec) == 0) {
			ret = -EIO;
			ocfs2_error(inode->i_sb,
				    "Inode %llu has empty extent block at %llu",
				    (unsigned long long)OCFS2_I(inode)->ip_blkno,
				    (unsigned long long)le64_to_cpu(eb->h_blkno));
			goto out;
		}

		/*
		 * XXX: The caller can not trust "path" any more after
		 * this as it will have been deleted. What do we do?
		 *
		 * In theory the rotate-for-merge code will never get
		 * here because it'll always ask for a rotate in a
		 * nonempty list.
		 */

		ret = ocfs2_remove_rightmost_path(inode, handle, path,
2753
						  dealloc, et);
2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764
		if (ret)
			mlog_errno(ret);
		goto out;
	}

	/*
	 * Now we can loop, remembering the path we get from -EAGAIN
	 * and restarting from there.
	 */
try_rotate:
	ret = __ocfs2_rotate_tree_left(inode, handle, orig_credits, path,
2765
				       dealloc, &restart_path, et);
2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776
	if (ret && ret != -EAGAIN) {
		mlog_errno(ret);
		goto out;
	}

	while (ret == -EAGAIN) {
		tmp_path = restart_path;
		restart_path = NULL;

		ret = __ocfs2_rotate_tree_left(inode, handle, orig_credits,
					       tmp_path, dealloc,
2777
					       &restart_path, et);
2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827
		if (ret && ret != -EAGAIN) {
			mlog_errno(ret);
			goto out;
		}

		ocfs2_free_path(tmp_path);
		tmp_path = NULL;

		if (ret == 0)
			goto try_rotate;
	}

out:
	ocfs2_free_path(tmp_path);
	ocfs2_free_path(restart_path);
	return ret;
}

static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el,
				int index)
{
	struct ocfs2_extent_rec *rec = &el->l_recs[index];
	unsigned int size;

	if (rec->e_leaf_clusters == 0) {
		/*
		 * We consumed all of the merged-from record. An empty
		 * extent cannot exist anywhere but the 1st array
		 * position, so move things over if the merged-from
		 * record doesn't occupy that position.
		 *
		 * This creates a new empty extent so the caller
		 * should be smart enough to have removed any existing
		 * ones.
		 */
		if (index > 0) {
			BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
			size = index * sizeof(struct ocfs2_extent_rec);
			memmove(&el->l_recs[1], &el->l_recs[0], size);
		}

		/*
		 * Always memset - the caller doesn't check whether it
		 * created an empty extent, so there could be junk in
		 * the other fields.
		 */
		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
	}
}

2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875
static int ocfs2_get_right_path(struct inode *inode,
				struct ocfs2_path *left_path,
				struct ocfs2_path **ret_right_path)
{
	int ret;
	u32 right_cpos;
	struct ocfs2_path *right_path = NULL;
	struct ocfs2_extent_list *left_el;

	*ret_right_path = NULL;

	/* This function shouldn't be called for non-trees. */
	BUG_ON(left_path->p_tree_depth == 0);

	left_el = path_leaf_el(left_path);
	BUG_ON(left_el->l_next_free_rec != left_el->l_count);

	ret = ocfs2_find_cpos_for_right_leaf(inode->i_sb, left_path,
					     &right_cpos);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	/* This function shouldn't be called for the rightmost leaf. */
	BUG_ON(right_cpos == 0);

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

	ret = ocfs2_find_path(inode, right_path, right_cpos);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	*ret_right_path = right_path;
out:
	if (ret)
		ocfs2_free_path(right_path);
	return ret;
}

2876 2877
/*
 * Remove split_rec clusters from the record at index and merge them
2878 2879 2880 2881
 * onto the beginning of the record "next" to it.
 * For index < l_count - 1, the next means the extent rec at index + 1.
 * For index == l_count - 1, the "next" means the 1st extent rec of the
 * next extent block.
2882
 */
2883 2884 2885 2886 2887
static int ocfs2_merge_rec_right(struct inode *inode,
				 struct ocfs2_path *left_path,
				 handle_t *handle,
				 struct ocfs2_extent_rec *split_rec,
				 int index)
2888
{
2889
	int ret, next_free, i;
2890 2891 2892
	unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
	struct ocfs2_extent_rec *left_rec;
	struct ocfs2_extent_rec *right_rec;
2893 2894 2895 2896 2897 2898
	struct ocfs2_extent_list *right_el;
	struct ocfs2_path *right_path = NULL;
	int subtree_index = 0;
	struct ocfs2_extent_list *el = path_leaf_el(left_path);
	struct buffer_head *bh = path_leaf_bh(left_path);
	struct buffer_head *root_bh = NULL;
2899 2900 2901

	BUG_ON(index >= le16_to_cpu(el->l_next_free_rec));
	left_rec = &el->l_recs[index];
2902

A
Al Viro 已提交
2903
	if (index == le16_to_cpu(el->l_next_free_rec) - 1 &&
2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916
	    le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) {
		/* we meet with a cross extent block merge. */
		ret = ocfs2_get_right_path(inode, left_path, &right_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		right_el = path_leaf_el(right_path);
		next_free = le16_to_cpu(right_el->l_next_free_rec);
		BUG_ON(next_free <= 0);
		right_rec = &right_el->l_recs[0];
		if (ocfs2_is_empty_extent(right_rec)) {
A
Al Viro 已提交
2917
			BUG_ON(next_free <= 1);
2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968
			right_rec = &right_el->l_recs[1];
		}

		BUG_ON(le32_to_cpu(left_rec->e_cpos) +
		       le16_to_cpu(left_rec->e_leaf_clusters) !=
		       le32_to_cpu(right_rec->e_cpos));

		subtree_index = ocfs2_find_subtree_root(inode,
							left_path, right_path);

		ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
						      handle->h_buffer_credits,
						      right_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		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;
			}
		}

	} else {
		BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1);
		right_rec = &el->l_recs[index + 1];
	}
2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989

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

	le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters);

	le32_add_cpu(&right_rec->e_cpos, -split_clusters);
	le64_add_cpu(&right_rec->e_blkno,
		     -ocfs2_clusters_to_blocks(inode->i_sb, split_clusters));
	le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters);

	ocfs2_cleanup_merge(el, index);

	ret = ocfs2_journal_dirty(handle, bh);
	if (ret)
		mlog_errno(ret);

2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041
	if (right_path) {
		ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
		if (ret)
			mlog_errno(ret);

		ocfs2_complete_edge_insert(inode, handle, left_path,
					   right_path, subtree_index);
	}
out:
	if (right_path)
		ocfs2_free_path(right_path);
	return ret;
}

static int ocfs2_get_left_path(struct inode *inode,
			       struct ocfs2_path *right_path,
			       struct ocfs2_path **ret_left_path)
{
	int ret;
	u32 left_cpos;
	struct ocfs2_path *left_path = NULL;

	*ret_left_path = NULL;

	/* This function shouldn't be called for non-trees. */
	BUG_ON(right_path->p_tree_depth == 0);

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

	/* This function shouldn't be called for the leftmost leaf. */
	BUG_ON(left_cpos == 0);

	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_path(inode, left_path, left_cpos);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	*ret_left_path = left_path;
3042
out:
3043 3044
	if (ret)
		ocfs2_free_path(left_path);
3045 3046 3047 3048 3049
	return ret;
}

/*
 * Remove split_rec clusters from the record at index and merge them
3050 3051 3052 3053 3054 3055 3056
 * onto the tail of the record "before" it.
 * For index > 0, the "before" means the extent rec at index - 1.
 *
 * For index == 0, the "before" means the last record of the previous
 * extent block. And there is also a situation that we may need to
 * remove the rightmost leaf extent block in the right_path and change
 * the right path to indicate the new rightmost path.
3057
 */
3058 3059
static int ocfs2_merge_rec_left(struct inode *inode,
				struct ocfs2_path *right_path,
3060 3061
				handle_t *handle,
				struct ocfs2_extent_rec *split_rec,
3062
				struct ocfs2_cached_dealloc_ctxt *dealloc,
3063
				struct ocfs2_extent_tree *et,
3064
				int index)
3065
{
3066
	int ret, i, subtree_index = 0, has_empty_extent = 0;
3067 3068 3069
	unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
	struct ocfs2_extent_rec *left_rec;
	struct ocfs2_extent_rec *right_rec;
3070 3071 3072 3073 3074
	struct ocfs2_extent_list *el = path_leaf_el(right_path);
	struct buffer_head *bh = path_leaf_bh(right_path);
	struct buffer_head *root_bh = NULL;
	struct ocfs2_path *left_path = NULL;
	struct ocfs2_extent_list *left_el;
3075

3076
	BUG_ON(index < 0);
3077 3078

	right_rec = &el->l_recs[index];
3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140
	if (index == 0) {
		/* we meet with a cross extent block merge. */
		ret = ocfs2_get_left_path(inode, right_path, &left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		left_el = path_leaf_el(left_path);
		BUG_ON(le16_to_cpu(left_el->l_next_free_rec) !=
		       le16_to_cpu(left_el->l_count));

		left_rec = &left_el->l_recs[
				le16_to_cpu(left_el->l_next_free_rec) - 1];
		BUG_ON(le32_to_cpu(left_rec->e_cpos) +
		       le16_to_cpu(left_rec->e_leaf_clusters) !=
		       le32_to_cpu(split_rec->e_cpos));

		subtree_index = ocfs2_find_subtree_root(inode,
							left_path, right_path);

		ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
						      handle->h_buffer_credits,
						      left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		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;
			}
		}
	} else {
		left_rec = &el->l_recs[index - 1];
		if (ocfs2_is_empty_extent(&el->l_recs[0]))
			has_empty_extent = 1;
	}
3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155

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

	if (has_empty_extent && index == 1) {
		/*
		 * The easy case - we can just plop the record right in.
		 */
		*left_rec = *split_rec;

		has_empty_extent = 0;
3156
	} else
3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169
		le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters);

	le32_add_cpu(&right_rec->e_cpos, split_clusters);
	le64_add_cpu(&right_rec->e_blkno,
		     ocfs2_clusters_to_blocks(inode->i_sb, split_clusters));
	le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters);

	ocfs2_cleanup_merge(el, index);

	ret = ocfs2_journal_dirty(handle, bh);
	if (ret)
		mlog_errno(ret);

3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183
	if (left_path) {
		ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
		if (ret)
			mlog_errno(ret);

		/*
		 * In the situation that the right_rec is empty and the extent
		 * block is empty also,  ocfs2_complete_edge_insert can't handle
		 * it and we need to delete the right extent block.
		 */
		if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 &&
		    le16_to_cpu(el->l_next_free_rec) == 1) {

			ret = ocfs2_remove_rightmost_path(inode, handle,
3184 3185
							  right_path,
							  dealloc, et);
3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199
			if (ret) {
				mlog_errno(ret);
				goto out;
			}

			/* Now the rightmost extent block has been deleted.
			 * So we use the new rightmost path.
			 */
			ocfs2_mv_path(right_path, left_path);
			left_path = NULL;
		} else
			ocfs2_complete_edge_insert(inode, handle, left_path,
						   right_path, subtree_index);
	}
3200
out:
3201 3202
	if (left_path)
		ocfs2_free_path(left_path);
3203 3204 3205 3206 3207
	return ret;
}

static int ocfs2_try_to_merge_extent(struct inode *inode,
				     handle_t *handle,
3208
				     struct ocfs2_path *path,
3209 3210 3211
				     int split_index,
				     struct ocfs2_extent_rec *split_rec,
				     struct ocfs2_cached_dealloc_ctxt *dealloc,
3212 3213
				     struct ocfs2_merge_ctxt *ctxt,
				     struct ocfs2_extent_tree *et)
3214 3215

{
T
Tao Mao 已提交
3216
	int ret = 0;
3217
	struct ocfs2_extent_list *el = path_leaf_el(path);
3218 3219 3220 3221
	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];

	BUG_ON(ctxt->c_contig_type == CONTIG_NONE);

T
Tao Mao 已提交
3222 3223 3224 3225 3226 3227 3228 3229
	if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) {
		/*
		 * The merge code will need to create an empty
		 * extent to take the place of the newly
		 * emptied slot. Remove any pre-existing empty
		 * extents - having more than one in a leaf is
		 * illegal.
		 */
3230
		ret = ocfs2_rotate_tree_left(inode, handle, path,
3231
					     dealloc, et);
T
Tao Mao 已提交
3232 3233 3234
		if (ret) {
			mlog_errno(ret);
			goto out;
3235
		}
T
Tao Mao 已提交
3236 3237
		split_index--;
		rec = &el->l_recs[split_index];
3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254
	}

	if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) {
		/*
		 * Left-right contig implies this.
		 */
		BUG_ON(!ctxt->c_split_covers_rec);

		/*
		 * Since the leftright insert always covers the entire
		 * extent, this call will delete the insert record
		 * entirely, resulting in an empty extent record added to
		 * the extent block.
		 *
		 * Since the adding of an empty extent shifts
		 * everything back to the right, there's no need to
		 * update split_index here.
3255 3256 3257 3258
		 *
		 * When the split_index is zero, we need to merge it to the
		 * prevoius extent block. It is more efficient and easier
		 * if we do merge_right first and merge_left later.
3259
		 */
3260 3261 3262
		ret = ocfs2_merge_rec_right(inode, path,
					    handle, split_rec,
					    split_index);
3263 3264 3265 3266 3267 3268 3269 3270 3271 3272
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		/*
		 * We can only get this from logic error above.
		 */
		BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0]));

3273
		/* The merge left us with an empty extent, remove it. */
3274 3275
		ret = ocfs2_rotate_tree_left(inode, handle, path,
					     dealloc, et);
3276 3277 3278 3279
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
3280

3281 3282 3283 3284
		rec = &el->l_recs[split_index];

		/*
		 * Note that we don't pass split_rec here on purpose -
3285
		 * we've merged it into the rec already.
3286
		 */
3287 3288
		ret = ocfs2_merge_rec_left(inode, path,
					   handle, rec,
3289
					   dealloc, et,
3290 3291
					   split_index);

3292 3293 3294 3295 3296
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

3297
		ret = ocfs2_rotate_tree_left(inode, handle, path,
3298
					     dealloc, et);
3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315
		/*
		 * Error from this last rotate is not critical, so
		 * print but don't bubble it up.
		 */
		if (ret)
			mlog_errno(ret);
		ret = 0;
	} else {
		/*
		 * Merge a record to the left or right.
		 *
		 * 'contig_type' is relative to the existing record,
		 * so for example, if we're "right contig", it's to
		 * the record on the left (hence the left merge).
		 */
		if (ctxt->c_contig_type == CONTIG_RIGHT) {
			ret = ocfs2_merge_rec_left(inode,
3316 3317
						   path,
						   handle, split_rec,
3318
						   dealloc, et,
3319 3320 3321 3322 3323 3324 3325
						   split_index);
			if (ret) {
				mlog_errno(ret);
				goto out;
			}
		} else {
			ret = ocfs2_merge_rec_right(inode,
3326 3327
						    path,
						    handle, split_rec,
3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339
						    split_index);
			if (ret) {
				mlog_errno(ret);
				goto out;
			}
		}

		if (ctxt->c_split_covers_rec) {
			/*
			 * The merge may have left an empty extent in
			 * our leaf. Try to rotate it away.
			 */
3340
			ret = ocfs2_rotate_tree_left(inode, handle, path,
3341
						     dealloc, et);
3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491
			if (ret)
				mlog_errno(ret);
			ret = 0;
		}
	}

out:
	return ret;
}

static void ocfs2_subtract_from_rec(struct super_block *sb,
				    enum ocfs2_split_type split,
				    struct ocfs2_extent_rec *rec,
				    struct ocfs2_extent_rec *split_rec)
{
	u64 len_blocks;

	len_blocks = ocfs2_clusters_to_blocks(sb,
				le16_to_cpu(split_rec->e_leaf_clusters));

	if (split == SPLIT_LEFT) {
		/*
		 * Region is on the left edge of the existing
		 * record.
		 */
		le32_add_cpu(&rec->e_cpos,
			     le16_to_cpu(split_rec->e_leaf_clusters));
		le64_add_cpu(&rec->e_blkno, len_blocks);
		le16_add_cpu(&rec->e_leaf_clusters,
			     -le16_to_cpu(split_rec->e_leaf_clusters));
	} else {
		/*
		 * Region is on the right edge of the existing
		 * record.
		 */
		le16_add_cpu(&rec->e_leaf_clusters,
			     -le16_to_cpu(split_rec->e_leaf_clusters));
	}
}

/*
 * Do the final bits of extent record insertion at the target leaf
 * list. If this leaf is part of an allocation tree, it is assumed
 * that the tree above has been prepared.
 */
static void ocfs2_insert_at_leaf(struct ocfs2_extent_rec *insert_rec,
				 struct ocfs2_extent_list *el,
				 struct ocfs2_insert_type *insert,
				 struct inode *inode)
{
	int i = insert->ins_contig_index;
	unsigned int range;
	struct ocfs2_extent_rec *rec;

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

	if (insert->ins_split != SPLIT_NONE) {
		i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos));
		BUG_ON(i == -1);
		rec = &el->l_recs[i];
		ocfs2_subtract_from_rec(inode->i_sb, insert->ins_split, rec,
					insert_rec);
		goto rotate;
	}

	/*
	 * Contiguous insert - either left or right.
	 */
	if (insert->ins_contig != CONTIG_NONE) {
		rec = &el->l_recs[i];
		if (insert->ins_contig == CONTIG_LEFT) {
			rec->e_blkno = insert_rec->e_blkno;
			rec->e_cpos = insert_rec->e_cpos;
		}
		le16_add_cpu(&rec->e_leaf_clusters,
			     le16_to_cpu(insert_rec->e_leaf_clusters));
		return;
	}

	/*
	 * Handle insert into an empty leaf.
	 */
	if (le16_to_cpu(el->l_next_free_rec) == 0 ||
	    ((le16_to_cpu(el->l_next_free_rec) == 1) &&
	     ocfs2_is_empty_extent(&el->l_recs[0]))) {
		el->l_recs[0] = *insert_rec;
		el->l_next_free_rec = cpu_to_le16(1);
		return;
	}

	/*
	 * Appending insert.
	 */
	if (insert->ins_appending == APPEND_TAIL) {
		i = le16_to_cpu(el->l_next_free_rec) - 1;
		rec = &el->l_recs[i];
		range = le32_to_cpu(rec->e_cpos)
			+ le16_to_cpu(rec->e_leaf_clusters);
		BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range);

		mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >=
				le16_to_cpu(el->l_count),
				"inode %lu, depth %u, count %u, next free %u, "
				"rec.cpos %u, rec.clusters %u, "
				"insert.cpos %u, insert.clusters %u\n",
				inode->i_ino,
				le16_to_cpu(el->l_tree_depth),
				le16_to_cpu(el->l_count),
				le16_to_cpu(el->l_next_free_rec),
				le32_to_cpu(el->l_recs[i].e_cpos),
				le16_to_cpu(el->l_recs[i].e_leaf_clusters),
				le32_to_cpu(insert_rec->e_cpos),
				le16_to_cpu(insert_rec->e_leaf_clusters));
		i++;
		el->l_recs[i] = *insert_rec;
		le16_add_cpu(&el->l_next_free_rec, 1);
		return;
	}

rotate:
	/*
	 * Ok, we have to rotate.
	 *
	 * At this point, it is safe to assume that inserting into an
	 * empty leaf and appending to a leaf have both been handled
	 * above.
	 *
	 * This leaf needs to have space, either by the empty 1st
	 * extent record, or by virtue of an l_next_rec < l_count.
	 */
	ocfs2_rotate_leaf(el, insert_rec);
}

static void ocfs2_adjust_rightmost_records(struct inode *inode,
					   handle_t *handle,
					   struct ocfs2_path *path,
					   struct ocfs2_extent_rec *insert_rec)
{
	int ret, i, next_free;
	struct buffer_head *bh;
	struct ocfs2_extent_list *el;
	struct ocfs2_extent_rec *rec;

	/*
	 * Update everything except the leaf block.
	 */
	for (i = 0; i < path->p_tree_depth; i++) {
		bh = path->p_node[i].bh;
		el = path->p_node[i].el;

M
Mark Fasheh 已提交
3492 3493 3494 3495 3496 3497
		next_free = le16_to_cpu(el->l_next_free_rec);
		if (next_free == 0) {
			ocfs2_error(inode->i_sb,
				    "Dinode %llu has a bad extent list",
				    (unsigned long long)OCFS2_I(inode)->ip_blkno);
			ret = -EIO;
3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547
			return;
		}

		rec = &el->l_recs[next_free - 1];

		rec->e_int_clusters = insert_rec->e_cpos;
		le32_add_cpu(&rec->e_int_clusters,
			     le16_to_cpu(insert_rec->e_leaf_clusters));
		le32_add_cpu(&rec->e_int_clusters,
			     -le32_to_cpu(rec->e_cpos));

		ret = ocfs2_journal_dirty(handle, bh);
		if (ret)
			mlog_errno(ret);

	}
}

static int ocfs2_append_rec_to_path(struct inode *inode, handle_t *handle,
				    struct ocfs2_extent_rec *insert_rec,
				    struct ocfs2_path *right_path,
				    struct ocfs2_path **ret_left_path)
{
	int ret, next_free;
	struct ocfs2_extent_list *el;
	struct ocfs2_path *left_path = NULL;

	*ret_left_path = NULL;

	/*
	 * This shouldn't happen for non-trees. The extent rec cluster
	 * count manipulation below only works for interior nodes.
	 */
	BUG_ON(right_path->p_tree_depth == 0);

	/*
	 * If our appending insert is at the leftmost edge of a leaf,
	 * then we might need to update the rightmost records of the
	 * neighboring path.
	 */
	el = path_leaf_el(right_path);
	next_free = le16_to_cpu(el->l_next_free_rec);
	if (next_free == 0 ||
	    (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) {
		u32 left_cpos;

		ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, right_path,
						    &left_cpos);
		if (ret) {
			mlog_errno(ret);
M
Mark Fasheh 已提交
3548 3549 3550
			goto out;
		}

3551 3552 3553
		mlog(0, "Append may need a left path update. cpos: %u, "
		     "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos),
		     left_cpos);
3554

3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566
		/*
		 * No need to worry if the append is already in the
		 * leftmost leaf.
		 */
		if (left_cpos) {
			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;
			}
M
Mark Fasheh 已提交
3567

3568 3569 3570 3571 3572
			ret = ocfs2_find_path(inode, left_path, left_cpos);
			if (ret) {
				mlog_errno(ret);
				goto out;
			}
M
Mark Fasheh 已提交
3573

3574 3575 3576 3577 3578 3579
			/*
			 * ocfs2_insert_path() will pass the left_path to the
			 * journal for us.
			 */
		}
	}
M
Mark Fasheh 已提交
3580

3581 3582 3583 3584
	ret = ocfs2_journal_access_path(inode, handle, right_path);
	if (ret) {
		mlog_errno(ret);
		goto out;
M
Mark Fasheh 已提交
3585 3586
	}

3587 3588
	ocfs2_adjust_rightmost_records(inode, handle, right_path, insert_rec);

M
Mark Fasheh 已提交
3589 3590 3591 3592 3593 3594 3595 3596 3597
	*ret_left_path = left_path;
	ret = 0;
out:
	if (ret != 0)
		ocfs2_free_path(left_path);

	return ret;
}

3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674
static void ocfs2_split_record(struct inode *inode,
			       struct ocfs2_path *left_path,
			       struct ocfs2_path *right_path,
			       struct ocfs2_extent_rec *split_rec,
			       enum ocfs2_split_type split)
{
	int index;
	u32 cpos = le32_to_cpu(split_rec->e_cpos);
	struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el;
	struct ocfs2_extent_rec *rec, *tmprec;

	right_el = path_leaf_el(right_path);;
	if (left_path)
		left_el = path_leaf_el(left_path);

	el = right_el;
	insert_el = right_el;
	index = ocfs2_search_extent_list(el, cpos);
	if (index != -1) {
		if (index == 0 && left_path) {
			BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));

			/*
			 * This typically means that the record
			 * started in the left path but moved to the
			 * right as a result of rotation. We either
			 * move the existing record to the left, or we
			 * do the later insert there.
			 *
			 * In this case, the left path should always
			 * exist as the rotate code will have passed
			 * it back for a post-insert update.
			 */

			if (split == SPLIT_LEFT) {
				/*
				 * It's a left split. Since we know
				 * that the rotate code gave us an
				 * empty extent in the left path, we
				 * can just do the insert there.
				 */
				insert_el = left_el;
			} else {
				/*
				 * Right split - we have to move the
				 * existing record over to the left
				 * leaf. The insert will be into the
				 * newly created empty extent in the
				 * right leaf.
				 */
				tmprec = &right_el->l_recs[index];
				ocfs2_rotate_leaf(left_el, tmprec);
				el = left_el;

				memset(tmprec, 0, sizeof(*tmprec));
				index = ocfs2_search_extent_list(left_el, cpos);
				BUG_ON(index == -1);
			}
		}
	} else {
		BUG_ON(!left_path);
		BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0]));
		/*
		 * Left path is easy - we can just allow the insert to
		 * happen.
		 */
		el = left_el;
		insert_el = left_el;
		index = ocfs2_search_extent_list(el, cpos);
		BUG_ON(index == -1);
	}

	rec = &el->l_recs[index];
	ocfs2_subtract_from_rec(inode->i_sb, split, rec, split_rec);
	ocfs2_rotate_leaf(insert_el, split_rec);
}

M
Mark Fasheh 已提交
3675
/*
3676 3677
 * This function only does inserts on an allocation b-tree. For tree
 * depth = 0, ocfs2_insert_at_leaf() is called directly.
M
Mark Fasheh 已提交
3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716
 *
 * right_path is the path we want to do the actual insert
 * in. left_path should only be passed in if we need to update that
 * portion of the tree after an edge insert.
 */
static int ocfs2_insert_path(struct inode *inode,
			     handle_t *handle,
			     struct ocfs2_path *left_path,
			     struct ocfs2_path *right_path,
			     struct ocfs2_extent_rec *insert_rec,
			     struct ocfs2_insert_type *insert)
{
	int ret, subtree_index;
	struct buffer_head *leaf_bh = path_leaf_bh(right_path);

	if (left_path) {
		int credits = handle->h_buffer_credits;

		/*
		 * There's a chance that left_path got passed back to
		 * us without being accounted for in the
		 * journal. Extend our transaction here to be sure we
		 * can change those blocks.
		 */
		credits += left_path->p_tree_depth;

		ret = ocfs2_extend_trans(handle, credits);
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}

		ret = ocfs2_journal_access_path(inode, handle, left_path);
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}
	}

3717 3718 3719 3720 3721 3722 3723 3724 3725 3726
	/*
	 * Pass both paths to the journal. The majority of inserts
	 * will be touching all components anyway.
	 */
	ret = ocfs2_journal_access_path(inode, handle, right_path);
	if (ret < 0) {
		mlog_errno(ret);
		goto out;
	}

3727 3728 3729
	if (insert->ins_split != SPLIT_NONE) {
		/*
		 * We could call ocfs2_insert_at_leaf() for some types
J
Joe Perches 已提交
3730
		 * of splits, but it's easier to just let one separate
3731 3732 3733 3734
		 * function sort it all out.
		 */
		ocfs2_split_record(inode, left_path, right_path,
				   insert_rec, insert->ins_split);
3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745

		/*
		 * Split might have modified either leaf and we don't
		 * have a guarantee that the later edge insert will
		 * dirty this for us.
		 */
		if (left_path)
			ret = ocfs2_journal_dirty(handle,
						  path_leaf_bh(left_path));
			if (ret)
				mlog_errno(ret);
3746 3747 3748
	} else
		ocfs2_insert_at_leaf(insert_rec, path_leaf_el(right_path),
				     insert, inode);
M
Mark Fasheh 已提交
3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773

	ret = ocfs2_journal_dirty(handle, leaf_bh);
	if (ret)
		mlog_errno(ret);

	if (left_path) {
		/*
		 * The rotate code has indicated that we need to fix
		 * up portions of the tree after the insert.
		 *
		 * XXX: Should we extend the transaction here?
		 */
		subtree_index = ocfs2_find_subtree_root(inode, left_path,
							right_path);
		ocfs2_complete_edge_insert(inode, handle, left_path,
					   right_path, subtree_index);
	}

	ret = 0;
out:
	return ret;
}

static int ocfs2_do_insert_extent(struct inode *inode,
				  handle_t *handle,
3774
				  struct ocfs2_extent_tree *et,
M
Mark Fasheh 已提交
3775 3776 3777 3778 3779 3780 3781 3782 3783
				  struct ocfs2_extent_rec *insert_rec,
				  struct ocfs2_insert_type *type)
{
	int ret, rotate = 0;
	u32 cpos;
	struct ocfs2_path *right_path = NULL;
	struct ocfs2_path *left_path = NULL;
	struct ocfs2_extent_list *el;

3784
	el = et->root_el;
M
Mark Fasheh 已提交
3785

3786
	ret = ocfs2_journal_access(handle, inode, et->root_bh,
M
Mark Fasheh 已提交
3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797
				   OCFS2_JOURNAL_ACCESS_WRITE);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	if (le16_to_cpu(el->l_tree_depth) == 0) {
		ocfs2_insert_at_leaf(insert_rec, el, type, inode);
		goto out_update_clusters;
	}

3798
	right_path = ocfs2_new_path(et->root_bh, et->root_el);
M
Mark Fasheh 已提交
3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835
	if (!right_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

	/*
	 * Determine the path to start with. Rotations need the
	 * rightmost path, everything else can go directly to the
	 * target leaf.
	 */
	cpos = le32_to_cpu(insert_rec->e_cpos);
	if (type->ins_appending == APPEND_NONE &&
	    type->ins_contig == CONTIG_NONE) {
		rotate = 1;
		cpos = UINT_MAX;
	}

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

	/*
	 * Rotations and appends need special treatment - they modify
	 * parts of the tree's above them.
	 *
	 * Both might pass back a path immediate to the left of the
	 * one being inserted to. This will be cause
	 * ocfs2_insert_path() to modify the rightmost records of
	 * left_path to account for an edge insert.
	 *
	 * XXX: When modifying this code, keep in mind that an insert
	 * can wind up skipping both of these two special cases...
	 */
	if (rotate) {
3836
		ret = ocfs2_rotate_tree_right(inode, handle, type->ins_split,
M
Mark Fasheh 已提交
3837 3838 3839 3840 3841 3842
					      le32_to_cpu(insert_rec->e_cpos),
					      right_path, &left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
3843 3844 3845 3846 3847

		/*
		 * ocfs2_rotate_tree_right() might have extended the
		 * transaction without re-journaling our tree root.
		 */
3848
		ret = ocfs2_journal_access(handle, inode, et->root_bh,
3849 3850 3851 3852 3853
					   OCFS2_JOURNAL_ACCESS_WRITE);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
M
Mark Fasheh 已提交
3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871
	} else if (type->ins_appending == APPEND_TAIL
		   && type->ins_contig != CONTIG_LEFT) {
		ret = ocfs2_append_rec_to_path(inode, handle, insert_rec,
					       right_path, &left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

	ret = ocfs2_insert_path(inode, handle, left_path, right_path,
				insert_rec, type);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

out_update_clusters:
3872
	if (type->ins_split == SPLIT_NONE)
3873 3874
		ocfs2_update_clusters(inode, et,
				      le16_to_cpu(insert_rec->e_leaf_clusters));
M
Mark Fasheh 已提交
3875

3876
	ret = ocfs2_journal_dirty(handle, et->root_bh);
M
Mark Fasheh 已提交
3877 3878 3879 3880 3881 3882 3883 3884 3885 3886
	if (ret)
		mlog_errno(ret);

out:
	ocfs2_free_path(left_path);
	ocfs2_free_path(right_path);

	return ret;
}

3887
static enum ocfs2_contig_type
T
Tao Ma 已提交
3888
ocfs2_figure_merge_contig_type(struct inode *inode, struct ocfs2_path *path,
3889 3890 3891
			       struct ocfs2_extent_list *el, int index,
			       struct ocfs2_extent_rec *split_rec)
{
T
Tao Ma 已提交
3892
	int status;
3893
	enum ocfs2_contig_type ret = CONTIG_NONE;
T
Tao Ma 已提交
3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932
	u32 left_cpos, right_cpos;
	struct ocfs2_extent_rec *rec = NULL;
	struct ocfs2_extent_list *new_el;
	struct ocfs2_path *left_path = NULL, *right_path = NULL;
	struct buffer_head *bh;
	struct ocfs2_extent_block *eb;

	if (index > 0) {
		rec = &el->l_recs[index - 1];
	} else if (path->p_tree_depth > 0) {
		status = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
						       path, &left_cpos);
		if (status)
			goto out;

		if (left_cpos != 0) {
			left_path = ocfs2_new_path(path_root_bh(path),
						   path_root_el(path));
			if (!left_path)
				goto out;

			status = ocfs2_find_path(inode, left_path, left_cpos);
			if (status)
				goto out;

			new_el = path_leaf_el(left_path);

			if (le16_to_cpu(new_el->l_next_free_rec) !=
			    le16_to_cpu(new_el->l_count)) {
				bh = path_leaf_bh(left_path);
				eb = (struct ocfs2_extent_block *)bh->b_data;
				OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb,
								 eb);
				goto out;
			}
			rec = &new_el->l_recs[
				le16_to_cpu(new_el->l_next_free_rec) - 1];
		}
	}
3933 3934 3935 3936 3937

	/*
	 * We're careful to check for an empty extent record here -
	 * the merge code will know what to do if it sees one.
	 */
T
Tao Ma 已提交
3938
	if (rec) {
3939 3940 3941 3942 3943 3944 3945 3946
		if (index == 1 && ocfs2_is_empty_extent(rec)) {
			if (split_rec->e_cpos == el->l_recs[index].e_cpos)
				ret = CONTIG_RIGHT;
		} else {
			ret = ocfs2_extent_contig(inode, rec, split_rec);
		}
	}

T
Tao Ma 已提交
3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983
	rec = NULL;
	if (index < (le16_to_cpu(el->l_next_free_rec) - 1))
		rec = &el->l_recs[index + 1];
	else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) &&
		 path->p_tree_depth > 0) {
		status = ocfs2_find_cpos_for_right_leaf(inode->i_sb,
							path, &right_cpos);
		if (status)
			goto out;

		if (right_cpos == 0)
			goto out;

		right_path = ocfs2_new_path(path_root_bh(path),
					    path_root_el(path));
		if (!right_path)
			goto out;

		status = ocfs2_find_path(inode, right_path, right_cpos);
		if (status)
			goto out;

		new_el = path_leaf_el(right_path);
		rec = &new_el->l_recs[0];
		if (ocfs2_is_empty_extent(rec)) {
			if (le16_to_cpu(new_el->l_next_free_rec) <= 1) {
				bh = path_leaf_bh(right_path);
				eb = (struct ocfs2_extent_block *)bh->b_data;
				OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb,
								 eb);
				goto out;
			}
			rec = &new_el->l_recs[1];
		}
	}

	if (rec) {
3984 3985 3986 3987 3988 3989 3990 3991 3992 3993
		enum ocfs2_contig_type contig_type;

		contig_type = ocfs2_extent_contig(inode, rec, split_rec);

		if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT)
			ret = CONTIG_LEFTRIGHT;
		else if (ret == CONTIG_NONE)
			ret = contig_type;
	}

T
Tao Ma 已提交
3994 3995 3996 3997 3998 3999
out:
	if (left_path)
		ocfs2_free_path(left_path);
	if (right_path)
		ocfs2_free_path(right_path);

4000 4001 4002
	return ret;
}

M
Mark Fasheh 已提交
4003 4004 4005 4006 4007 4008 4009 4010
static void ocfs2_figure_contig_type(struct inode *inode,
				     struct ocfs2_insert_type *insert,
				     struct ocfs2_extent_list *el,
				     struct ocfs2_extent_rec *insert_rec)
{
	int i;
	enum ocfs2_contig_type contig_type = CONTIG_NONE;

4011 4012
	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);

M
Mark Fasheh 已提交
4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029
	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
		contig_type = ocfs2_extent_contig(inode, &el->l_recs[i],
						  insert_rec);
		if (contig_type != CONTIG_NONE) {
			insert->ins_contig_index = i;
			break;
		}
	}
	insert->ins_contig = contig_type;
}

/*
 * This should only be called against the righmost leaf extent list.
 *
 * ocfs2_figure_appending_type() will figure out whether we'll have to
 * insert at the tail of the rightmost leaf.
 *
4030 4031
 * This should also work against the root extent list for tree's with 0
 * depth. If we consider the root extent list to be the rightmost leaf node
M
Mark Fasheh 已提交
4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043
 * then the logic here makes sense.
 */
static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
					struct ocfs2_extent_list *el,
					struct ocfs2_extent_rec *insert_rec)
{
	int i;
	u32 cpos = le32_to_cpu(insert_rec->e_cpos);
	struct ocfs2_extent_rec *rec;

	insert->ins_appending = APPEND_NONE;

4044
	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
M
Mark Fasheh 已提交
4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057

	if (!el->l_next_free_rec)
		goto set_tail_append;

	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
		/* Were all records empty? */
		if (le16_to_cpu(el->l_next_free_rec) == 1)
			goto set_tail_append;
	}

	i = le16_to_cpu(el->l_next_free_rec) - 1;
	rec = &el->l_recs[i];

4058 4059
	if (cpos >=
	    (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
M
Mark Fasheh 已提交
4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081
		goto set_tail_append;

	return;

set_tail_append:
	insert->ins_appending = APPEND_TAIL;
}

/*
 * Helper function called at the begining of an insert.
 *
 * This computes a few things that are commonly used in the process of
 * inserting into the btree:
 *   - Whether the new extent is contiguous with an existing one.
 *   - The current tree depth.
 *   - Whether the insert is an appending one.
 *   - The total # of free records in the tree.
 *
 * All of the information is stored on the ocfs2_insert_type
 * structure.
 */
static int ocfs2_figure_insert_type(struct inode *inode,
4082
				    struct ocfs2_extent_tree *et,
M
Mark Fasheh 已提交
4083 4084
				    struct buffer_head **last_eb_bh,
				    struct ocfs2_extent_rec *insert_rec,
4085
				    int *free_records,
M
Mark Fasheh 已提交
4086 4087 4088 4089 4090 4091 4092 4093
				    struct ocfs2_insert_type *insert)
{
	int ret;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;
	struct ocfs2_path *path = NULL;
	struct buffer_head *bh = NULL;

4094 4095
	insert->ins_split = SPLIT_NONE;

4096
	el = et->root_el;
M
Mark Fasheh 已提交
4097 4098 4099 4100 4101 4102 4103 4104 4105 4106
	insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);

	if (el->l_tree_depth) {
		/*
		 * If we have tree depth, we read in the
		 * rightmost extent block ahead of time as
		 * ocfs2_figure_insert_type() and ocfs2_add_branch()
		 * may want it later.
		 */
		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
4107
				       ocfs2_get_last_eb_blk(et), &bh,
M
Mark Fasheh 已提交
4108 4109 4110 4111 4112
				       OCFS2_BH_CACHED, inode);
		if (ret) {
			mlog_exit(ret);
			goto out;
		}
4113 4114
		eb = (struct ocfs2_extent_block *) bh->b_data;
		el = &eb->h_list;
M
Mark Fasheh 已提交
4115
	}
4116

M
Mark Fasheh 已提交
4117 4118 4119 4120 4121 4122 4123 4124
	/*
	 * Unless we have a contiguous insert, we'll need to know if
	 * there is room left in our allocation tree for another
	 * extent record.
	 *
	 * XXX: This test is simplistic, we can search for empty
	 * extent records too.
	 */
4125
	*free_records = le16_to_cpu(el->l_count) -
M
Mark Fasheh 已提交
4126 4127 4128 4129 4130 4131
		le16_to_cpu(el->l_next_free_rec);

	if (!insert->ins_tree_depth) {
		ocfs2_figure_contig_type(inode, insert, el, insert_rec);
		ocfs2_figure_appending_type(insert, el, insert_rec);
		return 0;
4132 4133
	}

4134
	path = ocfs2_new_path(et->root_bh, et->root_el);
M
Mark Fasheh 已提交
4135 4136 4137 4138 4139
	if (!path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}
4140

M
Mark Fasheh 已提交
4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151
	/*
	 * In the case that we're inserting past what the tree
	 * currently accounts for, ocfs2_find_path() will return for
	 * us the rightmost tree path. This is accounted for below in
	 * the appending code.
	 */
	ret = ocfs2_find_path(inode, path, le32_to_cpu(insert_rec->e_cpos));
	if (ret) {
		mlog_errno(ret);
		goto out;
	}
4152

M
Mark Fasheh 已提交
4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168
	el = path_leaf_el(path);

	/*
	 * Now that we have the path, there's two things we want to determine:
	 * 1) Contiguousness (also set contig_index if this is so)
	 *
	 * 2) Are we doing an append? We can trivially break this up
         *     into two types of appends: simple record append, or a
         *     rotate inside the tail leaf.
	 */
	ocfs2_figure_contig_type(inode, insert, el, insert_rec);

	/*
	 * The insert code isn't quite ready to deal with all cases of
	 * left contiguousness. Specifically, if it's an insert into
	 * the 1st record in a leaf, it will require the adjustment of
4169
	 * cluster count on the last record of the path directly to it's
M
Mark Fasheh 已提交
4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183
	 * left. For now, just catch that case and fool the layers
	 * above us. This works just fine for tree_depth == 0, which
	 * is why we allow that above.
	 */
	if (insert->ins_contig == CONTIG_LEFT &&
	    insert->ins_contig_index == 0)
		insert->ins_contig = CONTIG_NONE;

	/*
	 * Ok, so we can simply compare against last_eb to figure out
	 * whether the path doesn't exist. This will only happen in
	 * the case that we're doing a tail append, so maybe we can
	 * take advantage of that information somehow.
	 */
4184 4185
	if (ocfs2_get_last_eb_blk(et) ==
	    path_leaf_bh(path)->b_blocknr) {
M
Mark Fasheh 已提交
4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204
		/*
		 * Ok, ocfs2_find_path() returned us the rightmost
		 * tree path. This might be an appending insert. There are
		 * two cases:
		 *    1) We're doing a true append at the tail:
		 *	-This might even be off the end of the leaf
		 *    2) We're "appending" by rotating in the tail
		 */
		ocfs2_figure_appending_type(insert, el, insert_rec);
	}

out:
	ocfs2_free_path(path);

	if (ret == 0)
		*last_eb_bh = bh;
	else
		brelse(bh);
	return ret;
4205 4206
}

M
Mark Fasheh 已提交
4207 4208 4209 4210 4211
/*
 * Insert an extent into an inode btree.
 *
 * The caller needs to update fe->i_clusters
 */
4212
int ocfs2_insert_extent(struct ocfs2_super *osb,
4213
			handle_t *handle,
4214
			struct inode *inode,
4215
			struct buffer_head *root_bh,
M
Mark Fasheh 已提交
4216
			u32 cpos,
4217 4218
			u64 start_blk,
			u32 new_clusters,
4219
			u8 flags,
4220 4221
			struct ocfs2_alloc_context *meta_ac,
			enum ocfs2_extent_tree_type et_type)
4222
{
4223
	int status;
4224
	int uninitialized_var(free_records);
4225
	struct buffer_head *last_eb_bh = NULL;
M
Mark Fasheh 已提交
4226 4227
	struct ocfs2_insert_type insert = {0, };
	struct ocfs2_extent_rec rec;
4228
	struct ocfs2_extent_tree *et = NULL;
M
Mark Fasheh 已提交
4229

M
Mark Fasheh 已提交
4230 4231
	BUG_ON(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL);

4232 4233 4234 4235 4236 4237 4238
	et = ocfs2_new_extent_tree(root_bh, et_type);
	if (!et) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}

M
Mark Fasheh 已提交
4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249
	mlog(0, "add %u clusters at position %u to inode %llu\n",
	     new_clusters, cpos, (unsigned long long)OCFS2_I(inode)->ip_blkno);

	mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) &&
			(OCFS2_I(inode)->ip_clusters != cpos),
			"Device %s, asking for sparse allocation: inode %llu, "
			"cpos %u, clusters %u\n",
			osb->dev_str,
			(unsigned long long)OCFS2_I(inode)->ip_blkno, cpos,
			OCFS2_I(inode)->ip_clusters);

4250
	memset(&rec, 0, sizeof(rec));
M
Mark Fasheh 已提交
4251 4252
	rec.e_cpos = cpu_to_le32(cpos);
	rec.e_blkno = cpu_to_le64(start_blk);
4253
	rec.e_leaf_clusters = cpu_to_le16(new_clusters);
4254
	rec.e_flags = flags;
M
Mark Fasheh 已提交
4255

4256
	status = ocfs2_figure_insert_type(inode, et, &last_eb_bh, &rec,
4257
					  &free_records, &insert);
M
Mark Fasheh 已提交
4258 4259 4260
	if (status < 0) {
		mlog_errno(status);
		goto bail;
4261 4262
	}

M
Mark Fasheh 已提交
4263 4264 4265 4266
	mlog(0, "Insert.appending: %u, Insert.Contig: %u, "
	     "Insert.contig_index: %d, Insert.free_records: %d, "
	     "Insert.tree_depth: %d\n",
	     insert.ins_appending, insert.ins_contig, insert.ins_contig_index,
4267
	     free_records, insert.ins_tree_depth);
4268

4269
	if (insert.ins_contig == CONTIG_NONE && free_records == 0) {
4270
		status = ocfs2_grow_tree(inode, handle, et,
4271
					 &insert.ins_tree_depth, &last_eb_bh,
4272 4273
					 meta_ac);
		if (status) {
4274 4275 4276 4277 4278
			mlog_errno(status);
			goto bail;
		}
	}

M
Mark Fasheh 已提交
4279
	/* Finally, we can add clusters. This might rotate the tree for us. */
4280
	status = ocfs2_do_insert_extent(inode, handle, et, &rec, &insert);
4281 4282
	if (status < 0)
		mlog_errno(status);
4283
	else if (et->type == OCFS2_DINODE_EXTENT)
M
Mark Fasheh 已提交
4284
		ocfs2_extent_map_insert_rec(inode, &rec);
4285 4286 4287 4288 4289

bail:
	if (last_eb_bh)
		brelse(last_eb_bh);

4290 4291
	if (et)
		ocfs2_free_extent_tree(et);
4292 4293 4294 4295
	mlog_exit(status);
	return status;
}

4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318
static void ocfs2_make_right_split_rec(struct super_block *sb,
				       struct ocfs2_extent_rec *split_rec,
				       u32 cpos,
				       struct ocfs2_extent_rec *rec)
{
	u32 rec_cpos = le32_to_cpu(rec->e_cpos);
	u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters);

	memset(split_rec, 0, sizeof(struct ocfs2_extent_rec));

	split_rec->e_cpos = cpu_to_le32(cpos);
	split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos);

	split_rec->e_blkno = rec->e_blkno;
	le64_add_cpu(&split_rec->e_blkno,
		     ocfs2_clusters_to_blocks(sb, cpos - rec_cpos));

	split_rec->e_flags = rec->e_flags;
}

static int ocfs2_split_and_insert(struct inode *inode,
				  handle_t *handle,
				  struct ocfs2_path *path,
4319
				  struct ocfs2_extent_tree *et,
4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340
				  struct buffer_head **last_eb_bh,
				  int split_index,
				  struct ocfs2_extent_rec *orig_split_rec,
				  struct ocfs2_alloc_context *meta_ac)
{
	int ret = 0, depth;
	unsigned int insert_range, rec_range, do_leftright = 0;
	struct ocfs2_extent_rec tmprec;
	struct ocfs2_extent_list *rightmost_el;
	struct ocfs2_extent_rec rec;
	struct ocfs2_extent_rec split_rec = *orig_split_rec;
	struct ocfs2_insert_type insert;
	struct ocfs2_extent_block *eb;

leftright:
	/*
	 * Store a copy of the record on the stack - it might move
	 * around as the tree is manipulated below.
	 */
	rec = path_leaf_el(path)->l_recs[split_index];

4341
	rightmost_el = et->root_el;
4342 4343 4344 4345 4346 4347 4348 4349 4350 4351

	depth = le16_to_cpu(rightmost_el->l_tree_depth);
	if (depth) {
		BUG_ON(!(*last_eb_bh));
		eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
		rightmost_el = &eb->h_list;
	}

	if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
	    le16_to_cpu(rightmost_el->l_count)) {
4352 4353
		ret = ocfs2_grow_tree(inode, handle, et,
				      &depth, last_eb_bh, meta_ac);
4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

	memset(&insert, 0, sizeof(struct ocfs2_insert_type));
	insert.ins_appending = APPEND_NONE;
	insert.ins_contig = CONTIG_NONE;
	insert.ins_tree_depth = depth;

	insert_range = le32_to_cpu(split_rec.e_cpos) +
		le16_to_cpu(split_rec.e_leaf_clusters);
	rec_range = le32_to_cpu(rec.e_cpos) +
		le16_to_cpu(rec.e_leaf_clusters);

	if (split_rec.e_cpos == rec.e_cpos) {
		insert.ins_split = SPLIT_LEFT;
	} else if (insert_range == rec_range) {
		insert.ins_split = SPLIT_RIGHT;
	} else {
		/*
		 * Left/right split. We fake this as a right split
		 * first and then make a second pass as a left split.
		 */
		insert.ins_split = SPLIT_RIGHT;

		ocfs2_make_right_split_rec(inode->i_sb, &tmprec, insert_range,
					   &rec);

		split_rec = tmprec;

		BUG_ON(do_leftright);
		do_leftright = 1;
	}

4390
	ret = ocfs2_do_insert_extent(inode, handle, et, &split_rec, &insert);
4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	if (do_leftright == 1) {
		u32 cpos;
		struct ocfs2_extent_list *el;

		do_leftright++;
		split_rec = *orig_split_rec;

		ocfs2_reinit_path(path, 1);

		cpos = le32_to_cpu(split_rec.e_cpos);
		ret = ocfs2_find_path(inode, path, cpos);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		el = path_leaf_el(path);
		split_index = ocfs2_search_extent_list(el, cpos);
		goto leftright;
	}
out:

	return ret;
}

/*
 * Mark part or all of the extent record at split_index in the leaf
 * pointed to by path as written. This removes the unwritten
 * extent flag.
 *
 * Care is taken to handle contiguousness so as to not grow the tree.
 *
 * meta_ac is not strictly necessary - we only truly need it if growth
 * of the tree is required. All other cases will degrade into a less
 * optimal tree layout.
 *
4432 4433 4434
 * last_eb_bh should be the rightmost leaf block for any extent
 * btree. Since a split may grow the tree or a merge might shrink it,
 * the caller cannot trust the contents of that buffer after this call.
4435 4436 4437 4438 4439 4440 4441
 *
 * This code is optimized for readability - several passes might be
 * made over certain portions of the tree. All of those blocks will
 * have been brought into cache (and pinned via the journal), so the
 * extra overhead is not expressed in terms of disk reads.
 */
static int __ocfs2_mark_extent_written(struct inode *inode,
4442
				       struct ocfs2_extent_tree *et,
4443 4444 4445 4446 4447 4448 4449 4450 4451
				       handle_t *handle,
				       struct ocfs2_path *path,
				       int split_index,
				       struct ocfs2_extent_rec *split_rec,
				       struct ocfs2_alloc_context *meta_ac,
				       struct ocfs2_cached_dealloc_ctxt *dealloc)
{
	int ret = 0;
	struct ocfs2_extent_list *el = path_leaf_el(path);
4452
	struct buffer_head *last_eb_bh = NULL;
4453 4454 4455 4456
	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
	struct ocfs2_merge_ctxt ctxt;
	struct ocfs2_extent_list *rightmost_el;

4457
	if (!(rec->e_flags & OCFS2_EXT_UNWRITTEN)) {
4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470
		ret = -EIO;
		mlog_errno(ret);
		goto out;
	}

	if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) ||
	    ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) <
	     (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) {
		ret = -EIO;
		mlog_errno(ret);
		goto out;
	}

T
Tao Ma 已提交
4471
	ctxt.c_contig_type = ocfs2_figure_merge_contig_type(inode, path, el,
4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483
							    split_index,
							    split_rec);

	/*
	 * The core merge / split code wants to know how much room is
	 * left in this inodes allocation tree, so we pass the
	 * rightmost extent list.
	 */
	if (path->p_tree_depth) {
		struct ocfs2_extent_block *eb;

		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
4484
				       ocfs2_get_last_eb_blk(et),
4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509
				       &last_eb_bh, OCFS2_BH_CACHED, inode);
		if (ret) {
			mlog_exit(ret);
			goto out;
		}

		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);
			ret = -EROFS;
			goto out;
		}

		rightmost_el = &eb->h_list;
	} else
		rightmost_el = path_root_el(path);

	if (rec->e_cpos == split_rec->e_cpos &&
	    rec->e_leaf_clusters == split_rec->e_leaf_clusters)
		ctxt.c_split_covers_rec = 1;
	else
		ctxt.c_split_covers_rec = 0;

	ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]);

M
Mark Fasheh 已提交
4510 4511 4512
	mlog(0, "index: %d, contig: %u, has_empty: %u, split_covers: %u\n",
	     split_index, ctxt.c_contig_type, ctxt.c_has_empty_extent,
	     ctxt.c_split_covers_rec);
4513 4514 4515 4516 4517

	if (ctxt.c_contig_type == CONTIG_NONE) {
		if (ctxt.c_split_covers_rec)
			el->l_recs[split_index] = *split_rec;
		else
4518
			ret = ocfs2_split_and_insert(inode, handle, path, et,
4519 4520 4521 4522 4523 4524 4525
						     &last_eb_bh, split_index,
						     split_rec, meta_ac);
		if (ret)
			mlog_errno(ret);
	} else {
		ret = ocfs2_try_to_merge_extent(inode, handle, path,
						split_index, split_rec,
4526
						dealloc, &ctxt, et);
4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543
		if (ret)
			mlog_errno(ret);
	}

out:
	brelse(last_eb_bh);
	return ret;
}

/*
 * Mark the already-existing extent at cpos as written for len clusters.
 *
 * If the existing extent is larger than the request, initiate a
 * split. An attempt will be made at merging with adjacent extents.
 *
 * The caller is responsible for passing down meta_ac if we'll need it.
 */
4544
int ocfs2_mark_extent_written(struct inode *inode, struct buffer_head *root_bh,
4545 4546
			      handle_t *handle, u32 cpos, u32 len, u32 phys,
			      struct ocfs2_alloc_context *meta_ac,
4547 4548
			      struct ocfs2_cached_dealloc_ctxt *dealloc,
			      enum ocfs2_extent_tree_type et_type)
4549 4550 4551 4552 4553 4554
{
	int ret, index;
	u64 start_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys);
	struct ocfs2_extent_rec split_rec;
	struct ocfs2_path *left_path = NULL;
	struct ocfs2_extent_list *el;
4555
	struct ocfs2_extent_tree *et = NULL;
4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568

	mlog(0, "Inode %lu cpos %u, len %u, phys %u (%llu)\n",
	     inode->i_ino, cpos, len, phys, (unsigned long long)start_blkno);

	if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) {
		ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents "
			    "that are being written to, but the feature bit "
			    "is not set in the super block.",
			    (unsigned long long)OCFS2_I(inode)->ip_blkno);
		ret = -EROFS;
		goto out;
	}

4569 4570 4571 4572 4573 4574 4575
	et = ocfs2_new_extent_tree(root_bh, et_type);
	if (!et) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

4576 4577 4578 4579
	/*
	 * XXX: This should be fixed up so that we just re-insert the
	 * next extent records.
	 */
4580 4581
	if (et_type == OCFS2_DINODE_EXTENT)
		ocfs2_extent_map_trunc(inode, 0);
4582

4583
	left_path = ocfs2_new_path(et->root_bh, et->root_el);
4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613
	if (!left_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

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

	index = ocfs2_search_extent_list(el, cpos);
	if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
		ocfs2_error(inode->i_sb,
			    "Inode %llu has an extent at cpos %u which can no "
			    "longer be found.\n",
			    (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
		ret = -EROFS;
		goto out;
	}

	memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec));
	split_rec.e_cpos = cpu_to_le32(cpos);
	split_rec.e_leaf_clusters = cpu_to_le16(len);
	split_rec.e_blkno = cpu_to_le64(start_blkno);
	split_rec.e_flags = path_leaf_el(left_path)->l_recs[index].e_flags;
	split_rec.e_flags &= ~OCFS2_EXT_UNWRITTEN;

4614 4615 4616
	ret = __ocfs2_mark_extent_written(inode, et, handle, left_path,
					  index, &split_rec, meta_ac,
					  dealloc);
4617 4618 4619 4620 4621
	if (ret)
		mlog_errno(ret);

out:
	ocfs2_free_path(left_path);
4622 4623
	if (et)
		ocfs2_free_extent_tree(et);
4624 4625 4626
	return ret;
}

4627
static int ocfs2_split_tree(struct inode *inode, struct ocfs2_extent_tree *et,
4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649
			    handle_t *handle, struct ocfs2_path *path,
			    int index, u32 new_range,
			    struct ocfs2_alloc_context *meta_ac)
{
	int ret, depth, credits = handle->h_buffer_credits;
	struct buffer_head *last_eb_bh = NULL;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *rightmost_el, *el;
	struct ocfs2_extent_rec split_rec;
	struct ocfs2_extent_rec *rec;
	struct ocfs2_insert_type insert;

	/*
	 * Setup the record to split before we grow the tree.
	 */
	el = path_leaf_el(path);
	rec = &el->l_recs[index];
	ocfs2_make_right_split_rec(inode->i_sb, &split_rec, new_range, rec);

	depth = path->p_tree_depth;
	if (depth > 0) {
		ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
4650
				       ocfs2_get_last_eb_blk(et),
4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661
				       &last_eb_bh, OCFS2_BH_CACHED, inode);
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}

		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
		rightmost_el = &eb->h_list;
	} else
		rightmost_el = path_leaf_el(path);

4662
	credits += path->p_tree_depth +
4663
		   ocfs2_extend_meta_needed(et->root_el);
4664 4665 4666 4667 4668 4669 4670 4671
	ret = ocfs2_extend_trans(handle, credits);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
	    le16_to_cpu(rightmost_el->l_count)) {
4672
		ret = ocfs2_grow_tree(inode, handle, et, &depth, &last_eb_bh,
4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685
				      meta_ac);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

	memset(&insert, 0, sizeof(struct ocfs2_insert_type));
	insert.ins_appending = APPEND_NONE;
	insert.ins_contig = CONTIG_NONE;
	insert.ins_split = SPLIT_RIGHT;
	insert.ins_tree_depth = depth;

4686
	ret = ocfs2_do_insert_extent(inode, handle, et, &split_rec, &insert);
4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697
	if (ret)
		mlog_errno(ret);

out:
	brelse(last_eb_bh);
	return ret;
}

static int ocfs2_truncate_rec(struct inode *inode, handle_t *handle,
			      struct ocfs2_path *path, int index,
			      struct ocfs2_cached_dealloc_ctxt *dealloc,
4698 4699
			      u32 cpos, u32 len,
			      struct ocfs2_extent_tree *et)
4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710
{
	int ret;
	u32 left_cpos, rec_range, trunc_range;
	int wants_rotate = 0, is_rightmost_tree_rec = 0;
	struct super_block *sb = inode->i_sb;
	struct ocfs2_path *left_path = NULL;
	struct ocfs2_extent_list *el = path_leaf_el(path);
	struct ocfs2_extent_rec *rec;
	struct ocfs2_extent_block *eb;

	if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) {
4711
		ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc, et);
4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		index--;
	}

	if (index == (le16_to_cpu(el->l_next_free_rec) - 1) &&
	    path->p_tree_depth) {
		/*
		 * Check whether this is the rightmost tree record. If
		 * we remove all of this record or part of its right
		 * edge then an update of the record lengths above it
		 * will be required.
		 */
		eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
		if (eb->h_next_leaf_blk == 0)
			is_rightmost_tree_rec = 1;
	}

	rec = &el->l_recs[index];
	if (index == 0 && path->p_tree_depth &&
	    le32_to_cpu(rec->e_cpos) == cpos) {
		/*
		 * Changing the leftmost offset (via partial or whole
		 * record truncate) of an interior (or rightmost) path
		 * means we have to update the subtree that is formed
		 * by this leaf and the one to it's left.
		 *
		 * There are two cases we can skip:
		 *   1) Path is the leftmost one in our inode tree.
		 *   2) The leaf is rightmost and will be empty after
		 *      we remove the extent record - the rotate code
		 *      knows how to update the newly formed edge.
		 */

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

		if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
			left_path = ocfs2_new_path(path_root_bh(path),
						   path_root_el(path));
			if (!left_path) {
				ret = -ENOMEM;
				mlog_errno(ret);
				goto out;
			}

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

	ret = ocfs2_extend_rotate_transaction(handle, 0,
					      handle->h_buffer_credits,
					      path);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	ret = ocfs2_journal_access_path(inode, handle, path);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	ret = ocfs2_journal_access_path(inode, handle, left_path);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
	trunc_range = cpos + len;

	if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) {
		int next_free;

		memset(rec, 0, sizeof(*rec));
		ocfs2_cleanup_merge(el, index);
		wants_rotate = 1;

		next_free = le16_to_cpu(el->l_next_free_rec);
		if (is_rightmost_tree_rec && next_free > 1) {
			/*
			 * We skip the edge update if this path will
			 * be deleted by the rotate code.
			 */
			rec = &el->l_recs[next_free - 1];
			ocfs2_adjust_rightmost_records(inode, handle, path,
						       rec);
		}
	} else if (le32_to_cpu(rec->e_cpos) == cpos) {
		/* Remove leftmost portion of the record. */
		le32_add_cpu(&rec->e_cpos, len);
		le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len));
		le16_add_cpu(&rec->e_leaf_clusters, -len);
	} else if (rec_range == trunc_range) {
		/* Remove rightmost portion of the record */
		le16_add_cpu(&rec->e_leaf_clusters, -len);
		if (is_rightmost_tree_rec)
			ocfs2_adjust_rightmost_records(inode, handle, path, rec);
	} else {
		/* Caller should have trapped this. */
		mlog(ML_ERROR, "Inode %llu: Invalid record truncate: (%u, %u) "
		     "(%u, %u)\n", (unsigned long long)OCFS2_I(inode)->ip_blkno,
		     le32_to_cpu(rec->e_cpos),
		     le16_to_cpu(rec->e_leaf_clusters), cpos, len);
		BUG();
	}

	if (left_path) {
		int subtree_index;

		subtree_index = ocfs2_find_subtree_root(inode, left_path, path);
		ocfs2_complete_edge_insert(inode, handle, left_path, path,
					   subtree_index);
	}

	ocfs2_journal_dirty(handle, path_leaf_bh(path));

4842
	ret = ocfs2_rotate_tree_left(inode, handle, path, dealloc, et);
4843 4844 4845 4846 4847 4848 4849 4850 4851 4852
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

out:
	ocfs2_free_path(left_path);
	return ret;
}

4853
int ocfs2_remove_extent(struct inode *inode, struct buffer_head *root_bh,
4854 4855
			u32 cpos, u32 len, handle_t *handle,
			struct ocfs2_alloc_context *meta_ac,
4856 4857
			struct ocfs2_cached_dealloc_ctxt *dealloc,
			enum ocfs2_extent_tree_type et_type)
4858 4859 4860 4861 4862
{
	int ret, index;
	u32 rec_range, trunc_range;
	struct ocfs2_extent_rec *rec;
	struct ocfs2_extent_list *el;
4863 4864 4865 4866 4867 4868 4869 4870 4871
	struct ocfs2_path *path = NULL;
	struct ocfs2_extent_tree *et = NULL;

	et = ocfs2_new_extent_tree(root_bh, et_type);
	if (!et) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}
4872 4873 4874

	ocfs2_extent_map_trunc(inode, 0);

4875
	path = ocfs2_new_path(et->root_bh, et->root_el);
4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927
	if (!path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

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

	el = path_leaf_el(path);
	index = ocfs2_search_extent_list(el, cpos);
	if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
		ocfs2_error(inode->i_sb,
			    "Inode %llu has an extent at cpos %u which can no "
			    "longer be found.\n",
			    (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos);
		ret = -EROFS;
		goto out;
	}

	/*
	 * We have 3 cases of extent removal:
	 *   1) Range covers the entire extent rec
	 *   2) Range begins or ends on one edge of the extent rec
	 *   3) Range is in the middle of the extent rec (no shared edges)
	 *
	 * For case 1 we remove the extent rec and left rotate to
	 * fill the hole.
	 *
	 * For case 2 we just shrink the existing extent rec, with a
	 * tree update if the shrinking edge is also the edge of an
	 * extent block.
	 *
	 * For case 3 we do a right split to turn the extent rec into
	 * something case 2 can handle.
	 */
	rec = &el->l_recs[index];
	rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
	trunc_range = cpos + len;

	BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);

	mlog(0, "Inode %llu, remove (cpos %u, len %u). Existing index %d "
	     "(cpos %u, len %u)\n",
	     (unsigned long long)OCFS2_I(inode)->ip_blkno, cpos, len, index,
	     le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec));

	if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
		ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc,
4928
					 cpos, len, et);
4929 4930 4931 4932 4933
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	} else {
4934
		ret = ocfs2_split_tree(inode, et, handle, path, index,
4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982
				       trunc_range, meta_ac);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		/*
		 * The split could have manipulated the tree enough to
		 * move the record location, so we have to look for it again.
		 */
		ocfs2_reinit_path(path, 1);

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

		el = path_leaf_el(path);
		index = ocfs2_search_extent_list(el, cpos);
		if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
			ocfs2_error(inode->i_sb,
				    "Inode %llu: split at cpos %u lost record.",
				    (unsigned long long)OCFS2_I(inode)->ip_blkno,
				    cpos);
			ret = -EROFS;
			goto out;
		}

		/*
		 * Double check our values here. If anything is fishy,
		 * it's easier to catch it at the top level.
		 */
		rec = &el->l_recs[index];
		rec_range = le32_to_cpu(rec->e_cpos) +
			ocfs2_rec_clusters(el, rec);
		if (rec_range != trunc_range) {
			ocfs2_error(inode->i_sb,
				    "Inode %llu: error after split at cpos %u"
				    "trunc len %u, existing record is (%u,%u)",
				    (unsigned long long)OCFS2_I(inode)->ip_blkno,
				    cpos, len, le32_to_cpu(rec->e_cpos),
				    ocfs2_rec_clusters(el, rec));
			ret = -EROFS;
			goto out;
		}

		ret = ocfs2_truncate_rec(inode, handle, path, index, dealloc,
4983
					 cpos, len, et);
4984 4985 4986 4987 4988 4989 4990 4991
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

out:
	ocfs2_free_path(path);
4992 4993
	if (et)
		ocfs2_free_extent_tree(et);
4994 4995 4996
	return ret;
}

4997
int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029
{
	struct buffer_head *tl_bh = osb->osb_tl_bh;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;

	mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count),
			"slot %d, invalid truncate log parameters: used = "
			"%u, count = %u\n", osb->slot_num,
			le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count));
	return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count);
}

static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl,
					   unsigned int new_start)
{
	unsigned int tail_index;
	unsigned int current_tail;

	/* No records, nothing to coalesce */
	if (!le16_to_cpu(tl->tl_used))
		return 0;

	tail_index = le16_to_cpu(tl->tl_used) - 1;
	current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start);
	current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters);

	return current_tail == new_start;
}

5030 5031 5032 5033
int ocfs2_truncate_log_append(struct ocfs2_super *osb,
			      handle_t *handle,
			      u64 start_blk,
			      unsigned int num_clusters)
5034 5035 5036 5037 5038 5039 5040 5041
{
	int status, index;
	unsigned int start_cluster, tl_count;
	struct inode *tl_inode = osb->osb_tl_inode;
	struct buffer_head *tl_bh = osb->osb_tl_bh;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;

5042 5043
	mlog_entry("start_blk = %llu, num_clusters = %u\n",
		   (unsigned long long)start_blk, num_clusters);
5044

5045
	BUG_ON(mutex_trylock(&tl_inode->i_mutex));
5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059

	start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk);

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;
	if (!OCFS2_IS_VALID_DINODE(di)) {
		OCFS2_RO_ON_INVALID_DINODE(osb->sb, di);
		status = -EIO;
		goto bail;
	}

	tl_count = le16_to_cpu(tl->tl_count);
	mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
			tl_count == 0,
5060 5061 5062
			"Truncate record count on #%llu invalid "
			"wanted %u, actual %u\n",
			(unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081
			ocfs2_truncate_recs_per_inode(osb->sb),
			le16_to_cpu(tl->tl_count));

	/* Caller should have known to flush before calling us. */
	index = le16_to_cpu(tl->tl_used);
	if (index >= tl_count) {
		status = -ENOSPC;
		mlog_errno(status);
		goto bail;
	}

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

	mlog(0, "Log truncate of %u clusters starting at cluster %u to "
5082 5083
	     "%llu (index = %d)\n", num_clusters, start_cluster,
	     (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index);
5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113

	if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) {
		/*
		 * Move index back to the record we are coalescing with.
		 * ocfs2_truncate_log_can_coalesce() guarantees nonzero
		 */
		index--;

		num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters);
		mlog(0, "Coalesce with index %u (start = %u, clusters = %u)\n",
		     index, le32_to_cpu(tl->tl_recs[index].t_start),
		     num_clusters);
	} else {
		tl->tl_recs[index].t_start = cpu_to_le32(start_cluster);
		tl->tl_used = cpu_to_le16(index + 1);
	}
	tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters);

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

bail:
	mlog_exit(status);
	return status;
}

static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
5114
					 handle_t *handle,
5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187
					 struct inode *data_alloc_inode,
					 struct buffer_head *data_alloc_bh)
{
	int status = 0;
	int i;
	unsigned int num_clusters;
	u64 start_blk;
	struct ocfs2_truncate_rec rec;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;
	struct inode *tl_inode = osb->osb_tl_inode;
	struct buffer_head *tl_bh = osb->osb_tl_bh;

	mlog_entry_void();

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;
	i = le16_to_cpu(tl->tl_used) - 1;
	while (i >= 0) {
		/* Caller has given us at least enough credits to
		 * update the truncate log dinode */
		status = ocfs2_journal_access(handle, tl_inode, tl_bh,
					      OCFS2_JOURNAL_ACCESS_WRITE);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		tl->tl_used = cpu_to_le16(i);

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

		/* TODO: Perhaps we can calculate the bulk of the
		 * credits up front rather than extending like
		 * this. */
		status = ocfs2_extend_trans(handle,
					    OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		rec = tl->tl_recs[i];
		start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb,
						    le32_to_cpu(rec.t_start));
		num_clusters = le32_to_cpu(rec.t_clusters);

		/* if start_blk is not set, we ignore the record as
		 * invalid. */
		if (start_blk) {
			mlog(0, "free record %d, start = %u, clusters = %u\n",
			     i, le32_to_cpu(rec.t_start), num_clusters);

			status = ocfs2_free_clusters(handle, data_alloc_inode,
						     data_alloc_bh, start_blk,
						     num_clusters);
			if (status < 0) {
				mlog_errno(status);
				goto bail;
			}
		}
		i--;
	}

bail:
	mlog_exit(status);
	return status;
}

5188
/* Expects you to already be holding tl_inode->i_mutex */
5189
int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
5190 5191 5192
{
	int status;
	unsigned int num_to_flush;
5193
	handle_t *handle;
5194 5195 5196 5197 5198 5199 5200 5201 5202
	struct inode *tl_inode = osb->osb_tl_inode;
	struct inode *data_alloc_inode = NULL;
	struct buffer_head *tl_bh = osb->osb_tl_bh;
	struct buffer_head *data_alloc_bh = NULL;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;

	mlog_entry_void();

5203
	BUG_ON(mutex_trylock(&tl_inode->i_mutex));
5204 5205 5206 5207 5208 5209

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;
	if (!OCFS2_IS_VALID_DINODE(di)) {
		OCFS2_RO_ON_INVALID_DINODE(osb->sb, di);
		status = -EIO;
5210
		goto out;
5211 5212 5213
	}

	num_to_flush = le16_to_cpu(tl->tl_used);
5214 5215
	mlog(0, "Flush %u records from truncate log #%llu\n",
	     num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno);
5216 5217
	if (!num_to_flush) {
		status = 0;
5218
		goto out;
5219 5220 5221 5222 5223 5224 5225 5226
	}

	data_alloc_inode = ocfs2_get_system_file_inode(osb,
						       GLOBAL_BITMAP_SYSTEM_INODE,
						       OCFS2_INVALID_SLOT);
	if (!data_alloc_inode) {
		status = -EINVAL;
		mlog(ML_ERROR, "Could not get bitmap inode!\n");
5227
		goto out;
5228 5229
	}

5230 5231
	mutex_lock(&data_alloc_inode->i_mutex);

M
Mark Fasheh 已提交
5232
	status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1);
5233 5234
	if (status < 0) {
		mlog_errno(status);
5235
		goto out_mutex;
5236 5237
	}

5238
	handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
5239 5240 5241
	if (IS_ERR(handle)) {
		status = PTR_ERR(handle);
		mlog_errno(status);
5242
		goto out_unlock;
5243 5244 5245 5246
	}

	status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
					       data_alloc_bh);
5247
	if (status < 0)
5248 5249
		mlog_errno(status);

5250
	ocfs2_commit_trans(osb, handle);
5251

5252 5253
out_unlock:
	brelse(data_alloc_bh);
M
Mark Fasheh 已提交
5254
	ocfs2_inode_unlock(data_alloc_inode, 1);
5255

5256 5257 5258
out_mutex:
	mutex_unlock(&data_alloc_inode->i_mutex);
	iput(data_alloc_inode);
5259

5260
out:
5261 5262 5263 5264 5265 5266 5267 5268 5269
	mlog_exit(status);
	return status;
}

int ocfs2_flush_truncate_log(struct ocfs2_super *osb)
{
	int status;
	struct inode *tl_inode = osb->osb_tl_inode;

5270
	mutex_lock(&tl_inode->i_mutex);
5271
	status = __ocfs2_flush_truncate_log(osb);
5272
	mutex_unlock(&tl_inode->i_mutex);
5273 5274 5275 5276

	return status;
}

D
David Howells 已提交
5277
static void ocfs2_truncate_log_worker(struct work_struct *work)
5278 5279
{
	int status;
D
David Howells 已提交
5280 5281 5282
	struct ocfs2_super *osb =
		container_of(work, struct ocfs2_super,
			     osb_truncate_log_wq.work);
5283 5284 5285 5286 5287 5288

	mlog_entry_void();

	status = ocfs2_flush_truncate_log(osb);
	if (status < 0)
		mlog_errno(status);
5289 5290
	else
		ocfs2_init_inode_steal_slot(osb);
5291 5292 5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315 5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334 5335 5336 5337 5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360 5361 5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377 5378 5379 5380 5381 5382 5383 5384 5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422

	mlog_exit(status);
}

#define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ)
void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
				       int cancel)
{
	if (osb->osb_tl_inode) {
		/* We want to push off log flushes while truncates are
		 * still running. */
		if (cancel)
			cancel_delayed_work(&osb->osb_truncate_log_wq);

		queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq,
				   OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL);
	}
}

static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
				       int slot_num,
				       struct inode **tl_inode,
				       struct buffer_head **tl_bh)
{
	int status;
	struct inode *inode = NULL;
	struct buffer_head *bh = NULL;

	inode = ocfs2_get_system_file_inode(osb,
					   TRUNCATE_LOG_SYSTEM_INODE,
					   slot_num);
	if (!inode) {
		status = -EINVAL;
		mlog(ML_ERROR, "Could not get load truncate log inode!\n");
		goto bail;
	}

	status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
				  OCFS2_BH_CACHED, inode);
	if (status < 0) {
		iput(inode);
		mlog_errno(status);
		goto bail;
	}

	*tl_inode = inode;
	*tl_bh    = bh;
bail:
	mlog_exit(status);
	return status;
}

/* called during the 1st stage of node recovery. we stamp a clean
 * truncate log and pass back a copy for processing later. if the
 * truncate log does not require processing, a *tl_copy is set to
 * NULL. */
int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
				      int slot_num,
				      struct ocfs2_dinode **tl_copy)
{
	int status;
	struct inode *tl_inode = NULL;
	struct buffer_head *tl_bh = NULL;
	struct ocfs2_dinode *di;
	struct ocfs2_truncate_log *tl;

	*tl_copy = NULL;

	mlog(0, "recover truncate log from slot %d\n", slot_num);

	status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	di = (struct ocfs2_dinode *) tl_bh->b_data;
	tl = &di->id2.i_dealloc;
	if (!OCFS2_IS_VALID_DINODE(di)) {
		OCFS2_RO_ON_INVALID_DINODE(tl_inode->i_sb, di);
		status = -EIO;
		goto bail;
	}

	if (le16_to_cpu(tl->tl_used)) {
		mlog(0, "We'll have %u logs to recover\n",
		     le16_to_cpu(tl->tl_used));

		*tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL);
		if (!(*tl_copy)) {
			status = -ENOMEM;
			mlog_errno(status);
			goto bail;
		}

		/* Assuming the write-out below goes well, this copy
		 * will be passed back to recovery for processing. */
		memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size);

		/* All we need to do to clear the truncate log is set
		 * tl_used. */
		tl->tl_used = 0;

		status = ocfs2_write_block(osb, tl_bh, tl_inode);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

bail:
	if (tl_inode)
		iput(tl_inode);
	if (tl_bh)
		brelse(tl_bh);

	if (status < 0 && (*tl_copy)) {
		kfree(*tl_copy);
		*tl_copy = NULL;
	}

	mlog_exit(status);
	return status;
}

int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
					 struct ocfs2_dinode *tl_copy)
{
	int status = 0;
	int i;
	unsigned int clusters, num_recs, start_cluster;
	u64 start_blk;
5423
	handle_t *handle;
5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435
	struct inode *tl_inode = osb->osb_tl_inode;
	struct ocfs2_truncate_log *tl;

	mlog_entry_void();

	if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) {
		mlog(ML_ERROR, "Asked to recover my own truncate log!\n");
		return -EINVAL;
	}

	tl = &tl_copy->id2.i_dealloc;
	num_recs = le16_to_cpu(tl->tl_used);
5436
	mlog(0, "cleanup %u records from %llu\n", num_recs,
5437
	     (unsigned long long)le64_to_cpu(tl_copy->i_blkno));
5438

5439
	mutex_lock(&tl_inode->i_mutex);
5440 5441 5442 5443 5444 5445 5446 5447 5448
	for(i = 0; i < num_recs; i++) {
		if (ocfs2_truncate_log_needs_flush(osb)) {
			status = __ocfs2_flush_truncate_log(osb);
			if (status < 0) {
				mlog_errno(status);
				goto bail_up;
			}
		}

5449
		handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
5450 5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461
		if (IS_ERR(handle)) {
			status = PTR_ERR(handle);
			mlog_errno(status);
			goto bail_up;
		}

		clusters = le32_to_cpu(tl->tl_recs[i].t_clusters);
		start_cluster = le32_to_cpu(tl->tl_recs[i].t_start);
		start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster);

		status = ocfs2_truncate_log_append(osb, handle,
						   start_blk, clusters);
5462
		ocfs2_commit_trans(osb, handle);
5463 5464 5465 5466 5467 5468 5469
		if (status < 0) {
			mlog_errno(status);
			goto bail_up;
		}
	}

bail_up:
5470
	mutex_unlock(&tl_inode->i_mutex);
5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481 5482 5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494 5495 5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515

	mlog_exit(status);
	return status;
}

void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb)
{
	int status;
	struct inode *tl_inode = osb->osb_tl_inode;

	mlog_entry_void();

	if (tl_inode) {
		cancel_delayed_work(&osb->osb_truncate_log_wq);
		flush_workqueue(ocfs2_wq);

		status = ocfs2_flush_truncate_log(osb);
		if (status < 0)
			mlog_errno(status);

		brelse(osb->osb_tl_bh);
		iput(osb->osb_tl_inode);
	}

	mlog_exit_void();
}

int ocfs2_truncate_log_init(struct ocfs2_super *osb)
{
	int status;
	struct inode *tl_inode = NULL;
	struct buffer_head *tl_bh = NULL;

	mlog_entry_void();

	status = ocfs2_get_truncate_log_info(osb,
					     osb->slot_num,
					     &tl_inode,
					     &tl_bh);
	if (status < 0)
		mlog_errno(status);

	/* ocfs2_truncate_log_shutdown keys on the existence of
	 * osb->osb_tl_inode so we don't set any of the osb variables
	 * until we're sure all is well. */
D
David Howells 已提交
5516 5517
	INIT_DELAYED_WORK(&osb->osb_truncate_log_wq,
			  ocfs2_truncate_log_worker);
5518 5519 5520 5521 5522 5523 5524
	osb->osb_tl_bh    = tl_bh;
	osb->osb_tl_inode = tl_inode;

	mlog_exit(status);
	return status;
}

5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579
/*
 * Delayed de-allocation of suballocator blocks.
 *
 * Some sets of block de-allocations might involve multiple suballocator inodes.
 *
 * The locking for this can get extremely complicated, especially when
 * the suballocator inodes to delete from aren't known until deep
 * within an unrelated codepath.
 *
 * ocfs2_extent_block structures are a good example of this - an inode
 * btree could have been grown by any number of nodes each allocating
 * out of their own suballoc inode.
 *
 * These structures allow the delay of block de-allocation until a
 * later time, when locking of multiple cluster inodes won't cause
 * deadlock.
 */

/*
 * Describes a single block free from a suballocator
 */
struct ocfs2_cached_block_free {
	struct ocfs2_cached_block_free		*free_next;
	u64					free_blk;
	unsigned int				free_bit;
};

struct ocfs2_per_slot_free_list {
	struct ocfs2_per_slot_free_list		*f_next_suballocator;
	int					f_inode_type;
	int					f_slot;
	struct ocfs2_cached_block_free		*f_first;
};

static int ocfs2_free_cached_items(struct ocfs2_super *osb,
				   int sysfile_type,
				   int slot,
				   struct ocfs2_cached_block_free *head)
{
	int ret;
	u64 bg_blkno;
	handle_t *handle;
	struct inode *inode;
	struct buffer_head *di_bh = NULL;
	struct ocfs2_cached_block_free *tmp;

	inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot);
	if (!inode) {
		ret = -EINVAL;
		mlog_errno(ret);
		goto out;
	}

	mutex_lock(&inode->i_mutex);

M
Mark Fasheh 已提交
5580
	ret = ocfs2_inode_lock(inode, &di_bh, 1);
5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620
	if (ret) {
		mlog_errno(ret);
		goto out_mutex;
	}

	handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
	if (IS_ERR(handle)) {
		ret = PTR_ERR(handle);
		mlog_errno(ret);
		goto out_unlock;
	}

	while (head) {
		bg_blkno = ocfs2_which_suballoc_group(head->free_blk,
						      head->free_bit);
		mlog(0, "Free bit: (bit %u, blkno %llu)\n",
		     head->free_bit, (unsigned long long)head->free_blk);

		ret = ocfs2_free_suballoc_bits(handle, inode, di_bh,
					       head->free_bit, bg_blkno, 1);
		if (ret) {
			mlog_errno(ret);
			goto out_journal;
		}

		ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
		if (ret) {
			mlog_errno(ret);
			goto out_journal;
		}

		tmp = head;
		head = head->free_next;
		kfree(tmp);
	}

out_journal:
	ocfs2_commit_trans(osb, handle);

out_unlock:
M
Mark Fasheh 已提交
5621
	ocfs2_inode_unlock(inode, 1);
5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648 5649 5650 5651 5652 5653 5654 5655 5656 5657 5658 5659 5660 5661 5662 5663 5664 5665 5666 5667 5668 5669 5670 5671 5672 5673 5674 5675 5676 5677 5678 5679 5680 5681 5682 5683 5684 5685 5686 5687 5688 5689 5690 5691 5692 5693 5694 5695 5696 5697 5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710 5711 5712 5713 5714 5715 5716 5717 5718 5719 5720 5721 5722 5723 5724 5725 5726 5727 5728
	brelse(di_bh);
out_mutex:
	mutex_unlock(&inode->i_mutex);
	iput(inode);
out:
	while(head) {
		/* Premature exit may have left some dangling items. */
		tmp = head;
		head = head->free_next;
		kfree(tmp);
	}

	return ret;
}

int ocfs2_run_deallocs(struct ocfs2_super *osb,
		       struct ocfs2_cached_dealloc_ctxt *ctxt)
{
	int ret = 0, ret2;
	struct ocfs2_per_slot_free_list *fl;

	if (!ctxt)
		return 0;

	while (ctxt->c_first_suballocator) {
		fl = ctxt->c_first_suballocator;

		if (fl->f_first) {
			mlog(0, "Free items: (type %u, slot %d)\n",
			     fl->f_inode_type, fl->f_slot);
			ret2 = ocfs2_free_cached_items(osb, fl->f_inode_type,
						       fl->f_slot, fl->f_first);
			if (ret2)
				mlog_errno(ret2);
			if (!ret)
				ret = ret2;
		}

		ctxt->c_first_suballocator = fl->f_next_suballocator;
		kfree(fl);
	}

	return ret;
}

static struct ocfs2_per_slot_free_list *
ocfs2_find_per_slot_free_list(int type,
			      int slot,
			      struct ocfs2_cached_dealloc_ctxt *ctxt)
{
	struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator;

	while (fl) {
		if (fl->f_inode_type == type && fl->f_slot == slot)
			return fl;

		fl = fl->f_next_suballocator;
	}

	fl = kmalloc(sizeof(*fl), GFP_NOFS);
	if (fl) {
		fl->f_inode_type = type;
		fl->f_slot = slot;
		fl->f_first = NULL;
		fl->f_next_suballocator = ctxt->c_first_suballocator;

		ctxt->c_first_suballocator = fl;
	}
	return fl;
}

static int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
				     int type, int slot, u64 blkno,
				     unsigned int bit)
{
	int ret;
	struct ocfs2_per_slot_free_list *fl;
	struct ocfs2_cached_block_free *item;

	fl = ocfs2_find_per_slot_free_list(type, slot, ctxt);
	if (fl == NULL) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

	item = kmalloc(sizeof(*item), GFP_NOFS);
	if (item == NULL) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

	mlog(0, "Insert: (type %d, slot %u, bit %u, blk %llu)\n",
	     type, slot, bit, (unsigned long long)blkno);

	item->free_blk = blkno;
	item->free_bit = bit;
	item->free_next = fl->f_first;

	fl->f_first = item;

	ret = 0;
out:
	return ret;
}

5729 5730 5731 5732 5733 5734 5735 5736 5737
static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
					 struct ocfs2_extent_block *eb)
{
	return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE,
					 le16_to_cpu(eb->h_suballoc_slot),
					 le64_to_cpu(eb->h_blkno),
					 le16_to_cpu(eb->h_suballoc_bit));
}

5738 5739 5740 5741
/* This function will figure out whether the currently last extent
 * block will be deleted, and if it will, what the new last extent
 * block will be so we can update his h_next_leaf_blk field, as well
 * as the dinodes i_last_eb_blk */
M
Mark Fasheh 已提交
5742
static int ocfs2_find_new_last_ext_blk(struct inode *inode,
5743
				       unsigned int clusters_to_del,
M
Mark Fasheh 已提交
5744
				       struct ocfs2_path *path,
5745 5746
				       struct buffer_head **new_last_eb)
{
5747
	int next_free, ret = 0;
M
Mark Fasheh 已提交
5748
	u32 cpos;
5749
	struct ocfs2_extent_rec *rec;
5750 5751 5752 5753 5754 5755 5756
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;
	struct buffer_head *bh = NULL;

	*new_last_eb = NULL;

	/* we have no tree, so of course, no last_eb. */
M
Mark Fasheh 已提交
5757 5758
	if (!path->p_tree_depth)
		goto out;
5759 5760 5761

	/* trunc to zero special case - this makes tree_depth = 0
	 * regardless of what it is.  */
5762
	if (OCFS2_I(inode)->ip_clusters == clusters_to_del)
M
Mark Fasheh 已提交
5763
		goto out;
5764

M
Mark Fasheh 已提交
5765
	el = path_leaf_el(path);
5766 5767
	BUG_ON(!el->l_next_free_rec);

5768 5769 5770 5771 5772 5773 5774 5775 5776
	/*
	 * Make sure that this extent list will actually be empty
	 * after we clear away the data. We can shortcut out if
	 * there's more than one non-empty extent in the
	 * list. Otherwise, a check of the remaining extent is
	 * necessary.
	 */
	next_free = le16_to_cpu(el->l_next_free_rec);
	rec = NULL;
M
Mark Fasheh 已提交
5777
	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
5778
		if (next_free > 2)
M
Mark Fasheh 已提交
5779
			goto out;
5780 5781 5782 5783 5784 5785 5786 5787 5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799 5800

		/* We may have a valid extent in index 1, check it. */
		if (next_free == 2)
			rec = &el->l_recs[1];

		/*
		 * Fall through - no more nonempty extents, so we want
		 * to delete this leaf.
		 */
	} else {
		if (next_free > 1)
			goto out;

		rec = &el->l_recs[0];
	}

	if (rec) {
		/*
		 * Check it we'll only be trimming off the end of this
		 * cluster.
		 */
5801
		if (le16_to_cpu(rec->e_leaf_clusters) > clusters_to_del)
5802 5803
			goto out;
	}
5804

M
Mark Fasheh 已提交
5805 5806 5807 5808 5809
	ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}
5810

M
Mark Fasheh 已提交
5811 5812 5813 5814 5815
	ret = ocfs2_find_leaf(inode, path_root_el(path), cpos, &bh);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}
5816

M
Mark Fasheh 已提交
5817 5818 5819 5820 5821 5822 5823
	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 = -EROFS;
		goto out;
	}
5824 5825 5826

	*new_last_eb = bh;
	get_bh(*new_last_eb);
M
Mark Fasheh 已提交
5827 5828 5829 5830
	mlog(0, "returning block %llu, (cpos: %u)\n",
	     (unsigned long long)le64_to_cpu(eb->h_blkno), cpos);
out:
	brelse(bh);
5831

M
Mark Fasheh 已提交
5832
	return ret;
5833 5834
}

5835 5836 5837 5838 5839 5840 5841 5842 5843 5844 5845 5846 5847 5848 5849 5850 5851 5852 5853 5854 5855 5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871 5872 5873 5874 5875 5876 5877 5878 5879 5880
/*
 * Trim some clusters off the rightmost edge of a tree. Only called
 * during truncate.
 *
 * The caller needs to:
 *   - start journaling of each path component.
 *   - compute and fully set up any new last ext block
 */
static int ocfs2_trim_tree(struct inode *inode, struct ocfs2_path *path,
			   handle_t *handle, struct ocfs2_truncate_context *tc,
			   u32 clusters_to_del, u64 *delete_start)
{
	int ret, i, index = path->p_tree_depth;
	u32 new_edge = 0;
	u64 deleted_eb = 0;
	struct buffer_head *bh;
	struct ocfs2_extent_list *el;
	struct ocfs2_extent_rec *rec;

	*delete_start = 0;

	while (index >= 0) {
		bh = path->p_node[index].bh;
		el = path->p_node[index].el;

		mlog(0, "traveling tree (index = %d, block = %llu)\n",
		     index,  (unsigned long long)bh->b_blocknr);

		BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);

		if (index !=
		    (path->p_tree_depth - le16_to_cpu(el->l_tree_depth))) {
			ocfs2_error(inode->i_sb,
				    "Inode %lu has invalid ext. block %llu",
				    inode->i_ino,
				    (unsigned long long)bh->b_blocknr);
			ret = -EROFS;
			goto out;
		}

find_tail_record:
		i = le16_to_cpu(el->l_next_free_rec) - 1;
		rec = &el->l_recs[i];

		mlog(0, "Extent list before: record %d: (%u, %u, %llu), "
		     "next = %u\n", i, le32_to_cpu(rec->e_cpos),
5881
		     ocfs2_rec_clusters(el, rec),
5882 5883 5884
		     (unsigned long long)le64_to_cpu(rec->e_blkno),
		     le16_to_cpu(el->l_next_free_rec));

5885
		BUG_ON(ocfs2_rec_clusters(el, rec) < clusters_to_del);
5886 5887 5888 5889 5890 5891 5892 5893 5894 5895 5896 5897 5898 5899 5900 5901 5902 5903 5904 5905 5906 5907 5908 5909 5910 5911 5912 5913 5914 5915 5916 5917 5918 5919 5920 5921 5922 5923 5924 5925 5926

		if (le16_to_cpu(el->l_tree_depth) == 0) {
			/*
			 * If the leaf block contains a single empty
			 * extent and no records, we can just remove
			 * the block.
			 */
			if (i == 0 && ocfs2_is_empty_extent(rec)) {
				memset(rec, 0,
				       sizeof(struct ocfs2_extent_rec));
				el->l_next_free_rec = cpu_to_le16(0);

				goto delete;
			}

			/*
			 * Remove any empty extents by shifting things
			 * left. That should make life much easier on
			 * the code below. This condition is rare
			 * enough that we shouldn't see a performance
			 * hit.
			 */
			if (ocfs2_is_empty_extent(&el->l_recs[0])) {
				le16_add_cpu(&el->l_next_free_rec, -1);

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

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

				/*
				 * We've modified our extent list. The
				 * simplest way to handle this change
				 * is to being the search from the
				 * start again.
				 */
				goto find_tail_record;
			}

5927
			le16_add_cpu(&rec->e_leaf_clusters, -clusters_to_del);
5928 5929 5930 5931 5932

			/*
			 * We'll use "new_edge" on our way back up the
			 * tree to know what our rightmost cpos is.
			 */
5933
			new_edge = le16_to_cpu(rec->e_leaf_clusters);
5934 5935 5936 5937 5938 5939 5940
			new_edge += le32_to_cpu(rec->e_cpos);

			/*
			 * The caller will use this to delete data blocks.
			 */
			*delete_start = le64_to_cpu(rec->e_blkno)
				+ ocfs2_clusters_to_blocks(inode->i_sb,
5941
					le16_to_cpu(rec->e_leaf_clusters));
5942 5943 5944 5945

			/*
			 * If it's now empty, remove this record.
			 */
5946
			if (le16_to_cpu(rec->e_leaf_clusters) == 0) {
5947 5948 5949 5950 5951 5952 5953 5954 5955 5956 5957 5958 5959 5960 5961 5962 5963 5964 5965 5966 5967 5968 5969 5970 5971
				memset(rec, 0,
				       sizeof(struct ocfs2_extent_rec));
				le16_add_cpu(&el->l_next_free_rec, -1);
			}
		} else {
			if (le64_to_cpu(rec->e_blkno) == deleted_eb) {
				memset(rec, 0,
				       sizeof(struct ocfs2_extent_rec));
				le16_add_cpu(&el->l_next_free_rec, -1);

				goto delete;
			}

			/* Can this actually happen? */
			if (le16_to_cpu(el->l_next_free_rec) == 0)
				goto delete;

			/*
			 * We never actually deleted any clusters
			 * because our leaf was empty. There's no
			 * reason to adjust the rightmost edge then.
			 */
			if (new_edge == 0)
				goto delete;

5972 5973
			rec->e_int_clusters = cpu_to_le32(new_edge);
			le32_add_cpu(&rec->e_int_clusters,
5974 5975 5976 5977 5978 5979
				     -le32_to_cpu(rec->e_cpos));

			 /*
			  * A deleted child record should have been
			  * caught above.
			  */
5980
			 BUG_ON(le32_to_cpu(rec->e_int_clusters) == 0);
5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992
		}

delete:
		ret = ocfs2_journal_dirty(handle, bh);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		mlog(0, "extent list container %llu, after: record %d: "
		     "(%u, %u, %llu), next = %u.\n",
		     (unsigned long long)bh->b_blocknr, i,
5993
		     le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec),
5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011 6012 6013 6014
		     (unsigned long long)le64_to_cpu(rec->e_blkno),
		     le16_to_cpu(el->l_next_free_rec));

		/*
		 * We must be careful to only attempt delete of an
		 * extent block (and not the root inode block).
		 */
		if (index > 0 && le16_to_cpu(el->l_next_free_rec) == 0) {
			struct ocfs2_extent_block *eb =
				(struct ocfs2_extent_block *)bh->b_data;

			/*
			 * Save this for use when processing the
			 * parent block.
			 */
			deleted_eb = le64_to_cpu(eb->h_blkno);

			mlog(0, "deleting this extent block.\n");

			ocfs2_remove_from_cache(inode, bh);

6015
			BUG_ON(ocfs2_rec_clusters(el, &el->l_recs[0]));
6016 6017 6018
			BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos));
			BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno));

6019 6020 6021 6022
			ret = ocfs2_cache_extent_block_free(&tc->tc_dealloc, eb);
			/* An error here is not fatal. */
			if (ret < 0)
				mlog_errno(ret);
6023 6024 6025 6026 6027 6028 6029 6030 6031 6032 6033 6034
		} else {
			deleted_eb = 0;
		}

		index--;
	}

	ret = 0;
out:
	return ret;
}

6035 6036 6037 6038
static int ocfs2_do_truncate(struct ocfs2_super *osb,
			     unsigned int clusters_to_del,
			     struct inode *inode,
			     struct buffer_head *fe_bh,
6039
			     handle_t *handle,
M
Mark Fasheh 已提交
6040 6041
			     struct ocfs2_truncate_context *tc,
			     struct ocfs2_path *path)
6042
{
6043
	int status;
6044 6045 6046 6047 6048 6049 6050 6051
	struct ocfs2_dinode *fe;
	struct ocfs2_extent_block *last_eb = NULL;
	struct ocfs2_extent_list *el;
	struct buffer_head *last_eb_bh = NULL;
	u64 delete_blk = 0;

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

6052
	status = ocfs2_find_new_last_ext_blk(inode, clusters_to_del,
M
Mark Fasheh 已提交
6053
					     path, &last_eb_bh);
6054 6055 6056 6057
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
M
Mark Fasheh 已提交
6058 6059 6060 6061 6062

	/*
	 * Each component will be touched, so we might as well journal
	 * here to avoid having to handle errors later.
	 */
6063 6064 6065 6066
	status = ocfs2_journal_access_path(inode, handle, path);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
M
Mark Fasheh 已提交
6067 6068 6069 6070 6071 6072 6073 6074 6075 6076
	}

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

6077
		last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
M
Mark Fasheh 已提交
6078
	}
6079

M
Mark Fasheh 已提交
6080 6081 6082 6083 6084 6085
	el = &(fe->id2.i_list);

	/*
	 * Lower levels depend on this never happening, but it's best
	 * to check it up here before changing the tree.
	 */
6086
	if (el->l_tree_depth && el->l_recs[0].e_int_clusters == 0) {
M
Mark Fasheh 已提交
6087 6088 6089
		ocfs2_error(inode->i_sb,
			    "Inode %lu has an empty extent record, depth %u\n",
			    inode->i_ino, le16_to_cpu(el->l_tree_depth));
6090
		status = -EROFS;
6091 6092 6093 6094 6095 6096 6097 6098
		goto bail;
	}

	spin_lock(&OCFS2_I(inode)->ip_lock);
	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters) -
				      clusters_to_del;
	spin_unlock(&OCFS2_I(inode)->ip_lock);
	le32_add_cpu(&fe->i_clusters, -clusters_to_del);
6099
	inode->i_blocks = ocfs2_inode_sector_count(inode);
6100

6101 6102 6103 6104 6105
	status = ocfs2_trim_tree(inode, path, handle, tc,
				 clusters_to_del, &delete_blk);
	if (status) {
		mlog_errno(status);
		goto bail;
6106 6107
	}

M
Mark Fasheh 已提交
6108
	if (le32_to_cpu(fe->i_clusters) == 0) {
6109 6110 6111 6112 6113 6114 6115 6116 6117 6118 6119 6120 6121 6122 6123 6124 6125 6126 6127 6128 6129 6130 6131 6132
		/* trunc to zero is a special case. */
		el->l_tree_depth = 0;
		fe->i_last_eb_blk = 0;
	} else if (last_eb)
		fe->i_last_eb_blk = last_eb->h_blkno;

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

	if (last_eb) {
		/* If there will be a new last extent block, then by
		 * definition, there cannot be any leaves to the right of
		 * him. */
		last_eb->h_next_leaf_blk = 0;
		status = ocfs2_journal_dirty(handle, last_eb_bh);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

6133 6134 6135
	if (delete_blk) {
		status = ocfs2_truncate_log_append(osb, handle, delete_blk,
						   clusters_to_del);
6136 6137 6138 6139 6140 6141 6142
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}
	status = 0;
bail:
M
Mark Fasheh 已提交
6143

6144 6145 6146 6147
	mlog_exit(status);
	return status;
}

6148 6149 6150 6151 6152 6153 6154 6155 6156 6157 6158 6159 6160 6161
static int ocfs2_writeback_zero_func(handle_t *handle, struct buffer_head *bh)
{
	set_buffer_uptodate(bh);
	mark_buffer_dirty(bh);
	return 0;
}

static int ocfs2_ordered_zero_func(handle_t *handle, struct buffer_head *bh)
{
	set_buffer_uptodate(bh);
	mark_buffer_dirty(bh);
	return ocfs2_journal_dirty_data(handle, bh);
}

6162 6163 6164 6165 6166 6167 6168 6169 6170 6171 6172
static void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
				     unsigned int from, unsigned int to,
				     struct page *page, int zero, u64 *phys)
{
	int ret, partial = 0;

	ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0);
	if (ret)
		mlog_errno(ret);

	if (zero)
6173
		zero_user_segment(page, from, to);
6174 6175 6176 6177 6178 6179 6180 6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198 6199 6200

	/*
	 * Need to set the buffers we zero'd into uptodate
	 * here if they aren't - ocfs2_map_page_blocks()
	 * might've skipped some
	 */
	if (ocfs2_should_order_data(inode)) {
		ret = walk_page_buffers(handle,
					page_buffers(page),
					from, to, &partial,
					ocfs2_ordered_zero_func);
		if (ret < 0)
			mlog_errno(ret);
	} else {
		ret = walk_page_buffers(handle, page_buffers(page),
					from, to, &partial,
					ocfs2_writeback_zero_func);
		if (ret < 0)
			mlog_errno(ret);
	}

	if (!partial)
		SetPageUptodate(page);

	flush_dcache_page(page);
}

6201 6202 6203
static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start,
				     loff_t end, struct page **pages,
				     int numpages, u64 phys, handle_t *handle)
6204
{
6205
	int i;
6206 6207 6208 6209 6210 6211 6212 6213 6214
	struct page *page;
	unsigned int from, to = PAGE_CACHE_SIZE;
	struct super_block *sb = inode->i_sb;

	BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));

	if (numpages == 0)
		goto out;

6215
	to = PAGE_CACHE_SIZE;
6216 6217 6218
	for(i = 0; i < numpages; i++) {
		page = pages[i];

6219 6220 6221 6222
		from = start & (PAGE_CACHE_SIZE - 1);
		if ((end >> PAGE_CACHE_SHIFT) == page->index)
			to = end & (PAGE_CACHE_SIZE - 1);

6223 6224 6225
		BUG_ON(from > PAGE_CACHE_SIZE);
		BUG_ON(to > PAGE_CACHE_SIZE);

6226 6227
		ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1,
					 &phys);
6228

6229
		start = (page->index + 1) << PAGE_CACHE_SHIFT;
6230 6231
	}
out:
6232 6233
	if (pages)
		ocfs2_unlock_and_free_pages(pages, numpages);
6234 6235
}

6236
static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end,
6237
				struct page **pages, int *num)
6238
{
6239
	int numpages, ret = 0;
6240 6241 6242
	struct super_block *sb = inode->i_sb;
	struct address_space *mapping = inode->i_mapping;
	unsigned long index;
6243
	loff_t last_page_bytes;
6244

6245
	BUG_ON(start > end);
6246

6247 6248 6249
	BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits !=
	       (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits);

6250
	numpages = 0;
6251 6252
	last_page_bytes = PAGE_ALIGN(end);
	index = start >> PAGE_CACHE_SHIFT;
6253 6254 6255 6256 6257 6258 6259 6260 6261 6262
	do {
		pages[numpages] = grab_cache_page(mapping, index);
		if (!pages[numpages]) {
			ret = -ENOMEM;
			mlog_errno(ret);
			goto out;
		}

		numpages++;
		index++;
6263
	} while (index < (last_page_bytes >> PAGE_CACHE_SHIFT));
6264 6265 6266

out:
	if (ret != 0) {
6267 6268
		if (pages)
			ocfs2_unlock_and_free_pages(pages, numpages);
6269 6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283 6284 6285
		numpages = 0;
	}

	*num = numpages;

	return ret;
}

/*
 * Zero the area past i_size but still within an allocated
 * cluster. This avoids exposing nonzero data on subsequent file
 * extends.
 *
 * We need to call this before i_size is updated on the inode because
 * otherwise block_write_full_page() will skip writeout of pages past
 * i_size. The new_i_size parameter is passed for this reason.
 */
6286 6287
int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
				  u64 range_start, u64 range_end)
6288
{
6289
	int ret = 0, numpages;
6290 6291
	struct page **pages = NULL;
	u64 phys;
6292 6293
	unsigned int ext_flags;
	struct super_block *sb = inode->i_sb;
6294 6295 6296 6297 6298

	/*
	 * File systems which don't support sparse files zero on every
	 * extend.
	 */
6299
	if (!ocfs2_sparse_alloc(OCFS2_SB(sb)))
6300 6301
		return 0;

6302
	pages = kcalloc(ocfs2_pages_per_cluster(sb),
6303 6304 6305 6306 6307 6308 6309
			sizeof(struct page *), GFP_NOFS);
	if (pages == NULL) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

6310 6311 6312 6313 6314 6315
	if (range_start == range_end)
		goto out;

	ret = ocfs2_extent_map_get_blocks(inode,
					  range_start >> sb->s_blocksize_bits,
					  &phys, NULL, &ext_flags);
6316 6317 6318 6319 6320
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

6321 6322 6323 6324 6325
	/*
	 * Tail is a hole, or is marked unwritten. In either case, we
	 * can count on read and write to return/push zero's.
	 */
	if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN)
6326 6327
		goto out;

6328 6329 6330 6331 6332 6333 6334
	ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages,
				   &numpages);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

6335 6336
	ocfs2_zero_cluster_pages(inode, range_start, range_end, pages,
				 numpages, phys, handle);
6337 6338 6339 6340 6341 6342

	/*
	 * Initiate writeout of the pages we zero'd here. We don't
	 * wait on them - the truncate_inode_pages() call later will
	 * do that for us.
	 */
6343 6344
	ret = do_sync_mapping_range(inode->i_mapping, range_start,
				    range_end - 1, SYNC_FILE_RANGE_WRITE);
6345 6346 6347 6348 6349 6350 6351 6352 6353 6354
	if (ret)
		mlog_errno(ret);

out:
	if (pages)
		kfree(pages);

	return ret;
}

M
Mark Fasheh 已提交
6355 6356 6357 6358 6359 6360 6361
static void ocfs2_zero_dinode_id2(struct inode *inode, struct ocfs2_dinode *di)
{
	unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits;

	memset(&di->id2, 0, blocksize - offsetof(struct ocfs2_dinode, id2));
}

6362 6363 6364 6365 6366 6367 6368 6369 6370
void ocfs2_dinode_new_extent_list(struct inode *inode,
				  struct ocfs2_dinode *di)
{
	ocfs2_zero_dinode_id2(inode, di);
	di->id2.i_list.l_tree_depth = 0;
	di->id2.i_list.l_next_free_rec = 0;
	di->id2.i_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(inode->i_sb));
}

M
Mark Fasheh 已提交
6371 6372 6373 6374 6375 6376 6377 6378 6379 6380 6381 6382 6383 6384 6385 6386 6387 6388 6389 6390 6391 6392 6393 6394 6395 6396 6397 6398 6399 6400 6401 6402 6403 6404 6405 6406 6407 6408 6409 6410 6411 6412 6413 6414 6415 6416 6417 6418 6419 6420 6421 6422 6423 6424 6425 6426 6427 6428 6429 6430 6431 6432 6433 6434 6435 6436 6437 6438 6439 6440 6441 6442 6443 6444 6445 6446 6447 6448 6449 6450 6451 6452 6453 6454 6455 6456 6457 6458 6459 6460 6461 6462 6463 6464 6465 6466 6467 6468 6469 6470 6471 6472 6473 6474 6475 6476 6477 6478 6479 6480 6481 6482 6483 6484 6485 6486 6487 6488 6489 6490
void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di)
{
	struct ocfs2_inode_info *oi = OCFS2_I(inode);
	struct ocfs2_inline_data *idata = &di->id2.i_data;

	spin_lock(&oi->ip_lock);
	oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
	spin_unlock(&oi->ip_lock);

	/*
	 * We clear the entire i_data structure here so that all
	 * fields can be properly initialized.
	 */
	ocfs2_zero_dinode_id2(inode, di);

	idata->id_count = cpu_to_le16(ocfs2_max_inline_data(inode->i_sb));
}

int ocfs2_convert_inline_data_to_extents(struct inode *inode,
					 struct buffer_head *di_bh)
{
	int ret, i, has_data, num_pages = 0;
	handle_t *handle;
	u64 uninitialized_var(block);
	struct ocfs2_inode_info *oi = OCFS2_I(inode);
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
	struct ocfs2_alloc_context *data_ac = NULL;
	struct page **pages = NULL;
	loff_t end = osb->s_clustersize;

	has_data = i_size_read(inode) ? 1 : 0;

	if (has_data) {
		pages = kcalloc(ocfs2_pages_per_cluster(osb->sb),
				sizeof(struct page *), GFP_NOFS);
		if (pages == NULL) {
			ret = -ENOMEM;
			mlog_errno(ret);
			goto out;
		}

		ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

	handle = ocfs2_start_trans(osb, OCFS2_INLINE_TO_EXTENTS_CREDITS);
	if (IS_ERR(handle)) {
		ret = PTR_ERR(handle);
		mlog_errno(ret);
		goto out_unlock;
	}

	ret = ocfs2_journal_access(handle, inode, di_bh,
				   OCFS2_JOURNAL_ACCESS_WRITE);
	if (ret) {
		mlog_errno(ret);
		goto out_commit;
	}

	if (has_data) {
		u32 bit_off, num;
		unsigned int page_end;
		u64 phys;

		ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off,
					   &num);
		if (ret) {
			mlog_errno(ret);
			goto out_commit;
		}

		/*
		 * Save two copies, one for insert, and one that can
		 * be changed by ocfs2_map_and_dirty_page() below.
		 */
		block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);

		/*
		 * Non sparse file systems zero on extend, so no need
		 * to do that now.
		 */
		if (!ocfs2_sparse_alloc(osb) &&
		    PAGE_CACHE_SIZE < osb->s_clustersize)
			end = PAGE_CACHE_SIZE;

		ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages);
		if (ret) {
			mlog_errno(ret);
			goto out_commit;
		}

		/*
		 * This should populate the 1st page for us and mark
		 * it up to date.
		 */
		ret = ocfs2_read_inline_data(inode, pages[0], di_bh);
		if (ret) {
			mlog_errno(ret);
			goto out_commit;
		}

		page_end = PAGE_CACHE_SIZE;
		if (PAGE_CACHE_SIZE > osb->s_clustersize)
			page_end = osb->s_clustersize;

		for (i = 0; i < num_pages; i++)
			ocfs2_map_and_dirty_page(inode, handle, 0, page_end,
						 pages[i], i > 0, &phys);
	}

	spin_lock(&oi->ip_lock);
	oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
	spin_unlock(&oi->ip_lock);

6491
	ocfs2_dinode_new_extent_list(inode, di);
M
Mark Fasheh 已提交
6492 6493 6494 6495 6496 6497 6498 6499 6500 6501

	ocfs2_journal_dirty(handle, di_bh);

	if (has_data) {
		/*
		 * An error at this point should be extremely rare. If
		 * this proves to be false, we could always re-build
		 * the in-inode data from our pages.
		 */
		ret = ocfs2_insert_extent(osb, handle, inode, di_bh,
6502 6503
					  0, block, 1, 0,
					  NULL, OCFS2_DINODE_EXTENT);
M
Mark Fasheh 已提交
6504 6505 6506 6507 6508 6509 6510 6511 6512 6513 6514 6515 6516 6517 6518 6519 6520 6521 6522 6523 6524 6525 6526 6527
		if (ret) {
			mlog_errno(ret);
			goto out_commit;
		}

		inode->i_blocks = ocfs2_inode_sector_count(inode);
	}

out_commit:
	ocfs2_commit_trans(osb, handle);

out_unlock:
	if (data_ac)
		ocfs2_free_alloc_context(data_ac);

out:
	if (pages) {
		ocfs2_unlock_and_free_pages(pages, num_pages);
		kfree(pages);
	}

	return ret;
}

6528 6529 6530 6531 6532 6533 6534 6535 6536 6537 6538 6539
/*
 * It is expected, that by the time you call this function,
 * inode->i_size and fe->i_size have been adjusted.
 *
 * WARNING: This will kfree the truncate context
 */
int ocfs2_commit_truncate(struct ocfs2_super *osb,
			  struct inode *inode,
			  struct buffer_head *fe_bh,
			  struct ocfs2_truncate_context *tc)
{
	int status, i, credits, tl_sem = 0;
M
Mark Fasheh 已提交
6540
	u32 clusters_to_del, new_highest_cpos, range;
6541
	struct ocfs2_extent_list *el;
6542
	handle_t *handle = NULL;
6543
	struct inode *tl_inode = osb->osb_tl_inode;
M
Mark Fasheh 已提交
6544
	struct ocfs2_path *path = NULL;
6545
	struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6546 6547 6548

	mlog_entry_void();

M
Mark Fasheh 已提交
6549
	new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
6550 6551
						     i_size_read(inode));

6552
	path = ocfs2_new_path(fe_bh, &di->id2.i_list);
M
Mark Fasheh 已提交
6553 6554 6555 6556 6557
	if (!path) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}
M
Mark Fasheh 已提交
6558 6559 6560

	ocfs2_extent_map_trunc(inode, new_highest_cpos);

6561
start:
6562 6563 6564 6565 6566 6567 6568 6569
	/*
	 * Check that we still have allocation to delete.
	 */
	if (OCFS2_I(inode)->ip_clusters == 0) {
		status = 0;
		goto bail;
	}

M
Mark Fasheh 已提交
6570 6571 6572 6573 6574 6575 6576
	/*
	 * Truncate always works against the rightmost tree branch.
	 */
	status = ocfs2_find_path(inode, path, UINT_MAX);
	if (status) {
		mlog_errno(status);
		goto bail;
6577 6578
	}

M
Mark Fasheh 已提交
6579 6580 6581 6582 6583 6584 6585 6586 6587 6588 6589 6590 6591 6592 6593
	mlog(0, "inode->ip_clusters = %u, tree_depth = %u\n",
	     OCFS2_I(inode)->ip_clusters, path->p_tree_depth);

	/*
	 * By now, el will point to the extent list on the bottom most
	 * portion of this tree. Only the tail record is considered in
	 * each pass.
	 *
	 * We handle the following cases, in order:
	 * - empty extent: delete the remaining branch
	 * - remove the entire record
	 * - remove a partial record
	 * - no record needs to be removed (truncate has completed)
	 */
	el = path_leaf_el(path);
6594 6595 6596 6597 6598 6599 6600 6601 6602
	if (le16_to_cpu(el->l_next_free_rec) == 0) {
		ocfs2_error(inode->i_sb,
			    "Inode %llu has empty extent block at %llu\n",
			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
			    (unsigned long long)path_leaf_bh(path)->b_blocknr);
		status = -EROFS;
		goto bail;
	}

6603
	i = le16_to_cpu(el->l_next_free_rec) - 1;
M
Mark Fasheh 已提交
6604
	range = le32_to_cpu(el->l_recs[i].e_cpos) +
6605
		ocfs2_rec_clusters(el, &el->l_recs[i]);
M
Mark Fasheh 已提交
6606 6607 6608
	if (i == 0 && ocfs2_is_empty_extent(&el->l_recs[i])) {
		clusters_to_del = 0;
	} else if (le32_to_cpu(el->l_recs[i].e_cpos) >= new_highest_cpos) {
6609
		clusters_to_del = ocfs2_rec_clusters(el, &el->l_recs[i]);
M
Mark Fasheh 已提交
6610
	} else if (range > new_highest_cpos) {
6611
		clusters_to_del = (ocfs2_rec_clusters(el, &el->l_recs[i]) +
6612
				   le32_to_cpu(el->l_recs[i].e_cpos)) -
M
Mark Fasheh 已提交
6613 6614 6615 6616 6617
				  new_highest_cpos;
	} else {
		status = 0;
		goto bail;
	}
6618

M
Mark Fasheh 已提交
6619 6620 6621
	mlog(0, "clusters_to_del = %u in this pass, tail blk=%llu\n",
	     clusters_to_del, (unsigned long long)path_leaf_bh(path)->b_blocknr);

6622
	mutex_lock(&tl_inode->i_mutex);
6623 6624 6625 6626 6627 6628 6629 6630 6631 6632 6633 6634 6635
	tl_sem = 1;
	/* ocfs2_truncate_log_needs_flush guarantees us at least one
	 * record is free for use. If there isn't any, we flush to get
	 * an empty truncate log.  */
	if (ocfs2_truncate_log_needs_flush(osb)) {
		status = __ocfs2_flush_truncate_log(osb);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

	credits = ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del,
M
Mark Fasheh 已提交
6636 6637
						(struct ocfs2_dinode *)fe_bh->b_data,
						el);
6638
	handle = ocfs2_start_trans(osb, credits);
6639 6640 6641 6642 6643 6644 6645
	if (IS_ERR(handle)) {
		status = PTR_ERR(handle);
		handle = NULL;
		mlog_errno(status);
		goto bail;
	}

M
Mark Fasheh 已提交
6646 6647
	status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh, handle,
				   tc, path);
6648 6649 6650 6651 6652
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

6653
	mutex_unlock(&tl_inode->i_mutex);
6654 6655
	tl_sem = 0;

6656
	ocfs2_commit_trans(osb, handle);
6657 6658
	handle = NULL;

M
Mark Fasheh 已提交
6659 6660 6661
	ocfs2_reinit_path(path, 1);

	/*
6662 6663
	 * The check above will catch the case where we've truncated
	 * away all allocation.
M
Mark Fasheh 已提交
6664
	 */
6665 6666
	goto start;

6667 6668 6669 6670 6671
bail:

	ocfs2_schedule_truncate_log_flush(osb, 1);

	if (tl_sem)
6672
		mutex_unlock(&tl_inode->i_mutex);
6673 6674

	if (handle)
6675
		ocfs2_commit_trans(osb, handle);
6676

6677 6678
	ocfs2_run_deallocs(osb, &tc->tc_dealloc);

M
Mark Fasheh 已提交
6679
	ocfs2_free_path(path);
6680 6681 6682 6683 6684 6685 6686 6687 6688

	/* This will drop the ext_alloc cluster lock for us */
	ocfs2_free_truncate_context(tc);

	mlog_exit(status);
	return status;
}

/*
6689
 * Expects the inode to already be locked.
6690 6691 6692 6693 6694 6695
 */
int ocfs2_prepare_truncate(struct ocfs2_super *osb,
			   struct inode *inode,
			   struct buffer_head *fe_bh,
			   struct ocfs2_truncate_context **tc)
{
6696
	int status;
6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710
	unsigned int new_i_clusters;
	struct ocfs2_dinode *fe;
	struct ocfs2_extent_block *eb;
	struct buffer_head *last_eb_bh = NULL;

	mlog_entry_void();

	*tc = NULL;

	new_i_clusters = ocfs2_clusters_for_bytes(osb->sb,
						  i_size_read(inode));
	fe = (struct ocfs2_dinode *) fe_bh->b_data;

	mlog(0, "fe->i_clusters = %u, new_i_clusters = %u, fe->i_size ="
6711 6712
	     "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters,
	     (unsigned long long)le64_to_cpu(fe->i_size));
6713

6714
	*tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL);
6715 6716 6717 6718 6719
	if (!(*tc)) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}
6720
	ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc);
6721 6722 6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736 6737 6738 6739 6740 6741 6742 6743 6744 6745 6746 6747 6748 6749 6750 6751

	if (fe->id2.i_list.l_tree_depth) {
		status = ocfs2_read_block(osb, le64_to_cpu(fe->i_last_eb_blk),
					  &last_eb_bh, OCFS2_BH_CACHED, inode);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
		if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
			OCFS2_RO_ON_INVALID_EXTENT_BLOCK(inode->i_sb, eb);

			brelse(last_eb_bh);
			status = -EIO;
			goto bail;
		}
	}

	(*tc)->tc_last_eb_bh = last_eb_bh;

	status = 0;
bail:
	if (status < 0) {
		if (*tc)
			ocfs2_free_truncate_context(*tc);
		*tc = NULL;
	}
	mlog_exit_void();
	return status;
}

M
Mark Fasheh 已提交
6752 6753 6754 6755 6756 6757 6758 6759 6760 6761 6762 6763 6764 6765 6766 6767 6768 6769 6770 6771 6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783 6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796 6797 6798 6799 6800 6801 6802 6803 6804 6805 6806 6807 6808 6809 6810 6811 6812 6813 6814 6815 6816 6817 6818 6819 6820 6821 6822 6823 6824 6825 6826
/*
 * 'start' is inclusive, 'end' is not.
 */
int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh,
			  unsigned int start, unsigned int end, int trunc)
{
	int ret;
	unsigned int numbytes;
	handle_t *handle;
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
	struct ocfs2_inline_data *idata = &di->id2.i_data;

	if (end > i_size_read(inode))
		end = i_size_read(inode);

	BUG_ON(start >= end);

	if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
	    !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) ||
	    !ocfs2_supports_inline_data(osb)) {
		ocfs2_error(inode->i_sb,
			    "Inline data flags for inode %llu don't agree! "
			    "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n",
			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
			    le16_to_cpu(di->i_dyn_features),
			    OCFS2_I(inode)->ip_dyn_features,
			    osb->s_feature_incompat);
		ret = -EROFS;
		goto out;
	}

	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
	if (IS_ERR(handle)) {
		ret = PTR_ERR(handle);
		mlog_errno(ret);
		goto out;
	}

	ret = ocfs2_journal_access(handle, inode, di_bh,
				   OCFS2_JOURNAL_ACCESS_WRITE);
	if (ret) {
		mlog_errno(ret);
		goto out_commit;
	}

	numbytes = end - start;
	memset(idata->id_data + start, 0, numbytes);

	/*
	 * No need to worry about the data page here - it's been
	 * truncated already and inline data doesn't need it for
	 * pushing zero's to disk, so we'll let readpage pick it up
	 * later.
	 */
	if (trunc) {
		i_size_write(inode, start);
		di->i_size = cpu_to_le64(start);
	}

	inode->i_blocks = ocfs2_inode_sector_count(inode);
	inode->i_ctime = inode->i_mtime = CURRENT_TIME;

	di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
	di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);

	ocfs2_journal_dirty(handle, di_bh);

out_commit:
	ocfs2_commit_trans(osb, handle);

out:
	return ret;
}

6827 6828
static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc)
{
6829 6830 6831 6832 6833 6834 6835
	/*
	 * The caller is responsible for completing deallocation
	 * before freeing the context.
	 */
	if (tc->tc_dealloc.c_first_suballocator != NULL)
		mlog(ML_NOTICE,
		     "Truncate completion has non-empty dealloc context\n");
6836 6837 6838 6839 6840 6841

	if (tc->tc_last_eb_bh)
		brelse(tc->tc_last_eb_bh);

	kfree(tc);
}