alloc.c 194.9 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
#include <linux/quotaops.h>
32 33 34 35 36 37 38

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

#include "ocfs2.h"

#include "alloc.h"
39
#include "aops.h"
J
Joel Becker 已提交
40
#include "blockcheck.h"
41 42 43 44 45 46 47 48 49 50
#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"
51
#include "xattr.h"
52
#include "refcounttree.h"
53 54 55

#include "buffer_head_io.h"

56 57 58 59 60 61
enum ocfs2_contig_type {
	CONTIG_NONE = 0,
	CONTIG_LEFT,
	CONTIG_RIGHT,
	CONTIG_LEFTRIGHT,
};
62

63 64 65 66
static enum ocfs2_contig_type
	ocfs2_extent_rec_contig(struct super_block *sb,
				struct ocfs2_extent_rec *ext,
				struct ocfs2_extent_rec *insert_rec);
67 68 69 70 71
/*
 * Operations for a specific extent tree type.
 *
 * To implement an on-disk btree (extent tree) type in ocfs2, add
 * an ocfs2_extent_tree_operations structure and the matching
72
 * ocfs2_init_<thingy>_extent_tree() function.  That's pretty much it
73 74
 * for the allocation portion of the extent tree.
 */
75
struct ocfs2_extent_tree_operations {
76 77 78 79 80 81 82
	/*
	 * last_eb_blk is the block number of the right most leaf extent
	 * block.  Most on-disk structures containing an extent tree store
	 * this value for fast access.  The ->eo_set_last_eb_blk() and
	 * ->eo_get_last_eb_blk() operations access this value.  They are
	 *  both required.
	 */
83 84 85
	void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et,
				   u64 blkno);
	u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et);
86 87 88 89 90 91 92

	/*
	 * The on-disk structure usually keeps track of how many total
	 * clusters are stored in this extent tree.  This function updates
	 * that value.  new_clusters is the delta, and must be
	 * added to the total.  Required.
	 */
93
	void (*eo_update_clusters)(struct ocfs2_extent_tree *et,
94
				   u32 new_clusters);
95

96 97 98 99 100 101 102
	/*
	 * If this extent tree is supported by an extent map, insert
	 * a record into the map.
	 */
	void (*eo_extent_map_insert)(struct ocfs2_extent_tree *et,
				     struct ocfs2_extent_rec *rec);

103 104 105 106 107 108 109
	/*
	 * If this extent tree is supported by an extent map, truncate the
	 * map to clusters,
	 */
	void (*eo_extent_map_truncate)(struct ocfs2_extent_tree *et,
				       u32 clusters);

110 111 112 113
	/*
	 * If ->eo_insert_check() exists, it is called before rec is
	 * inserted into the extent tree.  It is optional.
	 */
114
	int (*eo_insert_check)(struct ocfs2_extent_tree *et,
115
			       struct ocfs2_extent_rec *rec);
116
	int (*eo_sanity_check)(struct ocfs2_extent_tree *et);
117

118 119 120 121 122 123 124 125 126 127
	/*
	 * --------------------------------------------------------------
	 * The remaining are internal to ocfs2_extent_tree and don't have
	 * accessor functions
	 */

	/*
	 * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el.
	 * It is required.
	 */
128
	void (*eo_fill_root_el)(struct ocfs2_extent_tree *et);
129 130 131 132 133 134

	/*
	 * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if
	 * it exists.  If it does not, et->et_max_leaf_clusters is set
	 * to 0 (unlimited).  Optional.
	 */
135
	void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et);
136 137 138 139 140 141 142 143 144 145

	/*
	 * ->eo_extent_contig test whether the 2 ocfs2_extent_rec
	 * are contiguous or not. Optional. Don't need to set it if use
	 * ocfs2_extent_rec as the tree leaf.
	 */
	enum ocfs2_contig_type
		(*eo_extent_contig)(struct ocfs2_extent_tree *et,
				    struct ocfs2_extent_rec *ext,
				    struct ocfs2_extent_rec *insert_rec);
146 147 148
};


149 150 151 152 153 154 155
/*
 * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check
 * in the methods.
 */
static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et);
static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
					 u64 blkno);
156
static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
157
					 u32 clusters);
158 159
static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et,
					   struct ocfs2_extent_rec *rec);
160 161
static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
					     u32 clusters);
162
static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
163
				     struct ocfs2_extent_rec *rec);
164
static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et);
165 166 167 168 169
static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et);
static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = {
	.eo_set_last_eb_blk	= ocfs2_dinode_set_last_eb_blk,
	.eo_get_last_eb_blk	= ocfs2_dinode_get_last_eb_blk,
	.eo_update_clusters	= ocfs2_dinode_update_clusters,
170
	.eo_extent_map_insert	= ocfs2_dinode_extent_map_insert,
171
	.eo_extent_map_truncate	= ocfs2_dinode_extent_map_truncate,
172 173 174 175
	.eo_insert_check	= ocfs2_dinode_insert_check,
	.eo_sanity_check	= ocfs2_dinode_sanity_check,
	.eo_fill_root_el	= ocfs2_dinode_fill_root_el,
};
176

177 178 179
static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
					 u64 blkno)
{
180
	struct ocfs2_dinode *di = et->et_object;
181

182
	BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
183 184 185 186 187
	di->i_last_eb_blk = cpu_to_le64(blkno);
}

static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et)
{
188
	struct ocfs2_dinode *di = et->et_object;
189

190
	BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
191 192 193
	return le64_to_cpu(di->i_last_eb_blk);
}

194
static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
195 196
					 u32 clusters)
{
197
	struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
198
	struct ocfs2_dinode *di = et->et_object;
199 200

	le32_add_cpu(&di->i_clusters, clusters);
201 202 203
	spin_lock(&oi->ip_lock);
	oi->ip_clusters = le32_to_cpu(di->i_clusters);
	spin_unlock(&oi->ip_lock);
204 205
}

206 207 208 209 210 211 212 213
static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et,
					   struct ocfs2_extent_rec *rec)
{
	struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode;

	ocfs2_extent_map_insert_rec(inode, rec);
}

214 215 216 217 218 219 220 221
static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
					     u32 clusters)
{
	struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode;

	ocfs2_extent_map_trunc(inode, clusters);
}

222
static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
223 224
				     struct ocfs2_extent_rec *rec)
{
225 226
	struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
	struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb);
227

228
	BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL);
229
	mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) &&
230
			(oi->ip_clusters != le32_to_cpu(rec->e_cpos)),
231 232 233
			"Device %s, asking for sparse allocation: inode %llu, "
			"cpos %u, clusters %u\n",
			osb->dev_str,
234 235
			(unsigned long long)oi->ip_blkno,
			rec->e_cpos, oi->ip_clusters);
236 237 238 239

	return 0;
}

240
static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et)
241
{
242
	struct ocfs2_dinode *di = et->et_object;
243

244
	BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
245
	BUG_ON(!OCFS2_IS_VALID_DINODE(di));
246

247
	return 0;
248 249
}

250 251 252 253 254 255 256
static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et)
{
	struct ocfs2_dinode *di = et->et_object;

	et->et_root_el = &di->id2.i_list;
}

257

258 259
static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et)
{
260
	struct ocfs2_xattr_value_buf *vb = et->et_object;
261

262
	et->et_root_el = &vb->vb_xv->xr_list;
263 264
}

265 266 267
static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et,
					      u64 blkno)
{
268
	struct ocfs2_xattr_value_buf *vb = et->et_object;
269

270
	vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno);
271 272 273 274
}

static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et)
{
275
	struct ocfs2_xattr_value_buf *vb = et->et_object;
276

277
	return le64_to_cpu(vb->vb_xv->xr_last_eb_blk);
278 279
}

280
static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et,
281 282
					      u32 clusters)
{
283
	struct ocfs2_xattr_value_buf *vb = et->et_object;
284

285
	le32_add_cpu(&vb->vb_xv->xr_clusters, clusters);
286 287
}

288
static struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = {
289 290 291
	.eo_set_last_eb_blk	= ocfs2_xattr_value_set_last_eb_blk,
	.eo_get_last_eb_blk	= ocfs2_xattr_value_get_last_eb_blk,
	.eo_update_clusters	= ocfs2_xattr_value_update_clusters,
292
	.eo_fill_root_el	= ocfs2_xattr_value_fill_root_el,
293 294
};

295 296 297 298 299 300 301
static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et)
{
	struct ocfs2_xattr_block *xb = et->et_object;

	et->et_root_el = &xb->xb_attrs.xb_root.xt_list;
}

302
static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et)
303
{
304
	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
305
	et->et_max_leaf_clusters =
306
		ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE);
307 308
}

T
Tao Ma 已提交
309 310 311
static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et,
					     u64 blkno)
{
312
	struct ocfs2_xattr_block *xb = et->et_object;
T
Tao Ma 已提交
313 314 315 316 317 318 319
	struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;

	xt->xt_last_eb_blk = cpu_to_le64(blkno);
}

static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et)
{
320
	struct ocfs2_xattr_block *xb = et->et_object;
T
Tao Ma 已提交
321 322 323 324 325
	struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;

	return le64_to_cpu(xt->xt_last_eb_blk);
}

326
static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et,
T
Tao Ma 已提交
327 328
					     u32 clusters)
{
329
	struct ocfs2_xattr_block *xb = et->et_object;
T
Tao Ma 已提交
330 331 332 333 334

	le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters);
}

static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = {
335 336 337
	.eo_set_last_eb_blk	= ocfs2_xattr_tree_set_last_eb_blk,
	.eo_get_last_eb_blk	= ocfs2_xattr_tree_get_last_eb_blk,
	.eo_update_clusters	= ocfs2_xattr_tree_update_clusters,
338
	.eo_fill_root_el	= ocfs2_xattr_tree_fill_root_el,
339
	.eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters,
T
Tao Ma 已提交
340 341
};

342 343 344 345 346 347 348 349 350 351 352 353 354 355 356
static void ocfs2_dx_root_set_last_eb_blk(struct ocfs2_extent_tree *et,
					  u64 blkno)
{
	struct ocfs2_dx_root_block *dx_root = et->et_object;

	dx_root->dr_last_eb_blk = cpu_to_le64(blkno);
}

static u64 ocfs2_dx_root_get_last_eb_blk(struct ocfs2_extent_tree *et)
{
	struct ocfs2_dx_root_block *dx_root = et->et_object;

	return le64_to_cpu(dx_root->dr_last_eb_blk);
}

357
static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et,
358 359 360 361 362 363 364
					  u32 clusters)
{
	struct ocfs2_dx_root_block *dx_root = et->et_object;

	le32_add_cpu(&dx_root->dr_clusters, clusters);
}

365
static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et)
366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388
{
	struct ocfs2_dx_root_block *dx_root = et->et_object;

	BUG_ON(!OCFS2_IS_VALID_DX_ROOT(dx_root));

	return 0;
}

static void ocfs2_dx_root_fill_root_el(struct ocfs2_extent_tree *et)
{
	struct ocfs2_dx_root_block *dx_root = et->et_object;

	et->et_root_el = &dx_root->dr_list;
}

static struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = {
	.eo_set_last_eb_blk	= ocfs2_dx_root_set_last_eb_blk,
	.eo_get_last_eb_blk	= ocfs2_dx_root_get_last_eb_blk,
	.eo_update_clusters	= ocfs2_dx_root_update_clusters,
	.eo_sanity_check	= ocfs2_dx_root_sanity_check,
	.eo_fill_root_el	= ocfs2_dx_root_fill_root_el,
};

389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434
static void ocfs2_refcount_tree_fill_root_el(struct ocfs2_extent_tree *et)
{
	struct ocfs2_refcount_block *rb = et->et_object;

	et->et_root_el = &rb->rf_list;
}

static void ocfs2_refcount_tree_set_last_eb_blk(struct ocfs2_extent_tree *et,
						u64 blkno)
{
	struct ocfs2_refcount_block *rb = et->et_object;

	rb->rf_last_eb_blk = cpu_to_le64(blkno);
}

static u64 ocfs2_refcount_tree_get_last_eb_blk(struct ocfs2_extent_tree *et)
{
	struct ocfs2_refcount_block *rb = et->et_object;

	return le64_to_cpu(rb->rf_last_eb_blk);
}

static void ocfs2_refcount_tree_update_clusters(struct ocfs2_extent_tree *et,
						u32 clusters)
{
	struct ocfs2_refcount_block *rb = et->et_object;

	le32_add_cpu(&rb->rf_clusters, clusters);
}

static enum ocfs2_contig_type
ocfs2_refcount_tree_extent_contig(struct ocfs2_extent_tree *et,
				  struct ocfs2_extent_rec *ext,
				  struct ocfs2_extent_rec *insert_rec)
{
	return CONTIG_NONE;
}

static struct ocfs2_extent_tree_operations ocfs2_refcount_tree_et_ops = {
	.eo_set_last_eb_blk	= ocfs2_refcount_tree_set_last_eb_blk,
	.eo_get_last_eb_blk	= ocfs2_refcount_tree_get_last_eb_blk,
	.eo_update_clusters	= ocfs2_refcount_tree_update_clusters,
	.eo_fill_root_el	= ocfs2_refcount_tree_fill_root_el,
	.eo_extent_contig	= ocfs2_refcount_tree_extent_contig,
};

435
static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et,
436
				     struct ocfs2_caching_info *ci,
437
				     struct buffer_head *bh,
438
				     ocfs2_journal_access_func access,
439 440
				     void *obj,
				     struct ocfs2_extent_tree_operations *ops)
441
{
442
	et->et_ops = ops;
443
	et->et_root_bh = bh;
444
	et->et_ci = ci;
445
	et->et_root_journal_access = access;
446 447 448
	if (!obj)
		obj = (void *)bh->b_data;
	et->et_object = obj;
449

450
	et->et_ops->eo_fill_root_el(et);
451 452 453
	if (!et->et_ops->eo_fill_max_leaf_clusters)
		et->et_max_leaf_clusters = 0;
	else
454
		et->et_ops->eo_fill_max_leaf_clusters(et);
455 456
}

457
void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et,
458
				   struct ocfs2_caching_info *ci,
459
				   struct buffer_head *bh)
460
{
461
	__ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_di,
462
				 NULL, &ocfs2_dinode_et_ops);
463 464
}

465
void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et,
466
				       struct ocfs2_caching_info *ci,
467
				       struct buffer_head *bh)
468
{
469
	__ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_xb,
470
				 NULL, &ocfs2_xattr_tree_et_ops);
471 472
}

473
void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et,
474
					struct ocfs2_caching_info *ci,
475
					struct ocfs2_xattr_value_buf *vb)
476
{
477
	__ocfs2_init_extent_tree(et, ci, vb->vb_bh, vb->vb_access, vb,
478
				 &ocfs2_xattr_value_et_ops);
479 480
}

481
void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et,
482
				    struct ocfs2_caching_info *ci,
483 484
				    struct buffer_head *bh)
{
485
	__ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_dr,
486 487 488
				 NULL, &ocfs2_dx_root_et_ops);
}

489 490 491 492 493 494 495 496
void ocfs2_init_refcount_extent_tree(struct ocfs2_extent_tree *et,
				     struct ocfs2_caching_info *ci,
				     struct buffer_head *bh)
{
	__ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_rb,
				 NULL, &ocfs2_refcount_tree_et_ops);
}

497 498
static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et,
					    u64 new_last_eb_blk)
499
{
500
	et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk);
501 502
}

503
static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et)
504
{
505
	return et->et_ops->eo_get_last_eb_blk(et);
506 507
}

508
static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et,
509 510
					    u32 clusters)
{
511
	et->et_ops->eo_update_clusters(et, clusters);
512 513
}

514 515 516 517 518 519 520
static inline void ocfs2_et_extent_map_insert(struct ocfs2_extent_tree *et,
					      struct ocfs2_extent_rec *rec)
{
	if (et->et_ops->eo_extent_map_insert)
		et->et_ops->eo_extent_map_insert(et, rec);
}

521 522 523 524 525 526 527
static inline void ocfs2_et_extent_map_truncate(struct ocfs2_extent_tree *et,
						u32 clusters)
{
	if (et->et_ops->eo_extent_map_truncate)
		et->et_ops->eo_extent_map_truncate(et, clusters);
}

528 529 530 531
static inline int ocfs2_et_root_journal_access(handle_t *handle,
					       struct ocfs2_extent_tree *et,
					       int type)
{
532
	return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh,
533 534 535
					  type);
}

536 537 538 539 540 541 542 543 544 545 546 547 548
static inline enum ocfs2_contig_type
	ocfs2_et_extent_contig(struct ocfs2_extent_tree *et,
			       struct ocfs2_extent_rec *rec,
			       struct ocfs2_extent_rec *insert_rec)
{
	if (et->et_ops->eo_extent_contig)
		return et->et_ops->eo_extent_contig(et, rec, insert_rec);

	return ocfs2_extent_rec_contig(
				ocfs2_metadata_cache_get_super(et->et_ci),
				rec, insert_rec);
}

549
static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et,
550 551 552 553 554
					struct ocfs2_extent_rec *rec)
{
	int ret = 0;

	if (et->et_ops->eo_insert_check)
555
		ret = et->et_ops->eo_insert_check(et, rec);
556 557 558
	return ret;
}

559
static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et)
560
{
561 562 563
	int ret = 0;

	if (et->et_ops->eo_sanity_check)
564
		ret = et->et_ops->eo_sanity_check(et);
565
	return ret;
566 567
}

M
Mark Fasheh 已提交
568
static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc);
569 570
static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
					 struct ocfs2_extent_block *eb);
571 572
static void ocfs2_adjust_rightmost_records(handle_t *handle,
					   struct ocfs2_extent_tree *et,
573 574
					   struct ocfs2_path *path,
					   struct ocfs2_extent_rec *insert_rec);
M
Mark Fasheh 已提交
575 576 577 578 579
/*
 * Reset the actual path elements so that we can re-use the structure
 * to build another path. Generally, this involves freeing the buffer
 * heads.
 */
580
void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root)
M
Mark Fasheh 已提交
581 582 583
{
	int i, start = 0, depth = 0;
	struct ocfs2_path_item *node;
584

M
Mark Fasheh 已提交
585 586
	if (keep_root)
		start = 1;
587

M
Mark Fasheh 已提交
588 589 590 591 592 593 594 595 596 597 598 599 600 601 602
	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);
603 604
	else
		path_root_access(path) = NULL;
M
Mark Fasheh 已提交
605 606 607 608

	path->p_tree_depth = depth;
}

609
void ocfs2_free_path(struct ocfs2_path *path)
M
Mark Fasheh 已提交
610 611 612 613 614 615 616
{
	if (path) {
		ocfs2_reinit_path(path, 0);
		kfree(path);
	}
}

617 618 619 620 621 622 623 624 625 626 627 628 629
/*
 * 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));
630
	BUG_ON(path_root_access(dest) != path_root_access(src));
631 632 633 634 635 636 637 638 639 640 641 642

	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
643 644 645 646 647 648 649 650 651
/*
 * 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));
652
	BUG_ON(path_root_access(dest) != path_root_access(src));
M
Mark Fasheh 已提交
653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687

	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,
688 689
					 struct ocfs2_extent_list *root_el,
					 ocfs2_journal_access_func access)
M
Mark Fasheh 已提交
690 691
{
	struct ocfs2_path *path;
692

M
Mark Fasheh 已提交
693 694 695 696 697 698 699 700
	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;
701
		path_root_access(path) = access;
M
Mark Fasheh 已提交
702 703 704 705 706
	}

	return path;
}

707
struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path)
708
{
709 710
	return ocfs2_new_path(path_root_bh(path), path_root_el(path),
			      path_root_access(path));
711 712
}

713
struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et)
714
{
715 716 717 718 719 720 721 722 723 724 725
	return ocfs2_new_path(et->et_root_bh, et->et_root_el,
			      et->et_root_journal_access);
}

/*
 * Journal the buffer at depth idx.  All idx>0 are extent_blocks,
 * otherwise it's the root_access function.
 *
 * I don't like the way this function's name looks next to
 * ocfs2_journal_access_path(), but I don't have a better one.
 */
726 727 728 729
int ocfs2_path_bh_journal_access(handle_t *handle,
				 struct ocfs2_caching_info *ci,
				 struct ocfs2_path *path,
				 int idx)
730 731 732 733 734 735 736 737 738
{
	ocfs2_journal_access_func access = path_root_access(path);

	if (!access)
		access = ocfs2_journal_access;

	if (idx)
		access = ocfs2_journal_access_eb;

739
	return access(handle, ci, path->p_node[idx].bh,
740
		      OCFS2_JOURNAL_ACCESS_WRITE);
741 742
}

M
Mark Fasheh 已提交
743 744 745
/*
 * Convenience function to journal all components in a path.
 */
746 747 748
int ocfs2_journal_access_path(struct ocfs2_caching_info *ci,
			      handle_t *handle,
			      struct ocfs2_path *path)
M
Mark Fasheh 已提交
749 750 751 752 753 754 755
{
	int i, ret = 0;

	if (!path)
		goto out;

	for(i = 0; i < path_num_items(path); i++) {
756
		ret = ocfs2_path_bh_journal_access(handle, ci, path, i);
M
Mark Fasheh 已提交
757 758 759 760 761 762 763 764 765 766
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}
	}

out:
	return ret;
}

767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796
/*
 * 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;
}

797 798
/*
 * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and
799
 * ocfs2_extent_rec_contig only work properly against leaf nodes!
800
 */
M
Mark Fasheh 已提交
801 802 803
static int ocfs2_block_extent_contig(struct super_block *sb,
				     struct ocfs2_extent_rec *ext,
				     u64 blkno)
804
{
805 806 807 808 809 810
	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;
811 812
}

M
Mark Fasheh 已提交
813 814 815
static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left,
				  struct ocfs2_extent_rec *right)
{
816 817 818 819 820 821
	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
822 823 824
}

static enum ocfs2_contig_type
825 826 827
	ocfs2_extent_rec_contig(struct super_block *sb,
				struct ocfs2_extent_rec *ext,
				struct ocfs2_extent_rec *insert_rec)
M
Mark Fasheh 已提交
828 829 830
{
	u64 blkno = le64_to_cpu(insert_rec->e_blkno);

831 832 833 834 835 836 837 838
	/*
	 * 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
839
	if (ocfs2_extents_adjacent(ext, insert_rec) &&
840
	    ocfs2_block_extent_contig(sb, ext, blkno))
M
Mark Fasheh 已提交
841 842 843 844
			return CONTIG_RIGHT;

	blkno = le64_to_cpu(ext->e_blkno);
	if (ocfs2_extents_adjacent(insert_rec, ext) &&
845
	    ocfs2_block_extent_contig(sb, insert_rec, blkno))
M
Mark Fasheh 已提交
846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862
		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,
};

863 864 865 866 867 868
enum ocfs2_split_type {
	SPLIT_NONE = 0,
	SPLIT_LEFT,
	SPLIT_RIGHT,
};

M
Mark Fasheh 已提交
869
struct ocfs2_insert_type {
870
	enum ocfs2_split_type	ins_split;
M
Mark Fasheh 已提交
871 872 873 874 875 876
	enum ocfs2_append_type	ins_appending;
	enum ocfs2_contig_type	ins_contig;
	int			ins_contig_index;
	int			ins_tree_depth;
};

877 878 879 880 881 882
struct ocfs2_merge_ctxt {
	enum ocfs2_contig_type	c_contig_type;
	int			c_has_empty_extent;
	int			c_split_covers_rec;
};

883 884 885
static int ocfs2_validate_extent_block(struct super_block *sb,
				       struct buffer_head *bh)
{
J
Joel Becker 已提交
886
	int rc;
887 888 889
	struct ocfs2_extent_block *eb =
		(struct ocfs2_extent_block *)bh->b_data;

890 891 892
	mlog(0, "Validating extent block %llu\n",
	     (unsigned long long)bh->b_blocknr);

J
Joel Becker 已提交
893 894 895 896 897 898 899 900
	BUG_ON(!buffer_uptodate(bh));

	/*
	 * If the ecc fails, we return the error but otherwise
	 * leave the filesystem running.  We know any error is
	 * local to this block.
	 */
	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &eb->h_check);
901 902 903
	if (rc) {
		mlog(ML_ERROR, "Checksum failed for extent block %llu\n",
		     (unsigned long long)bh->b_blocknr);
J
Joel Becker 已提交
904
		return rc;
905
	}
J
Joel Becker 已提交
906 907 908 909 910

	/*
	 * Errors after here are fatal.
	 */

911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939
	if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
		ocfs2_error(sb,
			    "Extent block #%llu has bad signature %.*s",
			    (unsigned long long)bh->b_blocknr, 7,
			    eb->h_signature);
		return -EINVAL;
	}

	if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) {
		ocfs2_error(sb,
			    "Extent block #%llu has an invalid h_blkno "
			    "of %llu",
			    (unsigned long long)bh->b_blocknr,
			    (unsigned long long)le64_to_cpu(eb->h_blkno));
		return -EINVAL;
	}

	if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation) {
		ocfs2_error(sb,
			    "Extent block #%llu has an invalid "
			    "h_fs_generation of #%u",
			    (unsigned long long)bh->b_blocknr,
			    le32_to_cpu(eb->h_fs_generation));
		return -EINVAL;
	}

	return 0;
}

940
int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno,
941 942 943 944 945
			    struct buffer_head **bh)
{
	int rc;
	struct buffer_head *tmp = *bh;

946
	rc = ocfs2_read_block(ci, eb_blkno, &tmp,
947
			      ocfs2_validate_extent_block);
948 949

	/* If ocfs2_read_block() got us a new bh, pass it up. */
950
	if (!rc && !*bh)
951 952 953 954 955 956
		*bh = tmp;

	return rc;
}


957 958 959 960
/*
 * How many free extents have we got before we need more meta data?
 */
int ocfs2_num_free_extents(struct ocfs2_super *osb,
961
			   struct ocfs2_extent_tree *et)
962 963
{
	int retval;
964
	struct ocfs2_extent_list *el = NULL;
965 966
	struct ocfs2_extent_block *eb;
	struct buffer_head *eb_bh = NULL;
967
	u64 last_eb_blk = 0;
968 969 970

	mlog_entry_void();

971 972
	el = et->et_root_el;
	last_eb_blk = ocfs2_et_get_last_eb_blk(et);
973

974
	if (last_eb_blk) {
975 976
		retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk,
						 &eb_bh);
977 978 979 980 981 982
		if (retval < 0) {
			mlog_errno(retval);
			goto bail;
		}
		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
		el = &eb->h_list;
983
	}
984 985 986 987 988

	BUG_ON(el->l_tree_depth != 0);

	retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec);
bail:
989
	brelse(eb_bh);
990 991 992 993 994 995 996 997 998 999

	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
 */
1000 1001
static int ocfs2_create_new_meta_bhs(handle_t *handle,
				     struct ocfs2_extent_tree *et,
1002 1003 1004 1005 1006 1007 1008
				     int wanted,
				     struct ocfs2_alloc_context *meta_ac,
				     struct buffer_head *bhs[])
{
	int count, status, i;
	u16 suballoc_bit_start;
	u32 num_got;
1009
	u64 suballoc_loc, first_blkno;
1010 1011
	struct ocfs2_super *osb =
		OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
1012 1013 1014 1015 1016 1017
	struct ocfs2_extent_block *eb;

	mlog_entry_void();

	count = 0;
	while (count < wanted) {
1018
		status = ocfs2_claim_metadata(handle,
1019 1020
					      meta_ac,
					      wanted - count,
1021
					      &suballoc_loc,
1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036
					      &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;
			}
1037
			ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]);
1038

1039 1040
			status = ocfs2_journal_access_eb(handle, et->et_ci,
							 bhs[i],
1041
							 OCFS2_JOURNAL_ACCESS_CREATE);
1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052
			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);
1053 1054
			eb->h_suballoc_slot =
				cpu_to_le16(meta_ac->ac_alloc_slot);
1055
			eb->h_suballoc_loc = cpu_to_le64(suballoc_loc);
1056 1057 1058 1059 1060 1061 1062 1063 1064
			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. */
1065
			ocfs2_journal_dirty(handle, bhs[i]);
1066 1067 1068 1069 1070 1071 1072 1073 1074
		}

		count += num_got;
	}

	status = 0;
bail:
	if (status < 0) {
		for(i = 0; i < wanted; i++) {
1075
			brelse(bhs[i]);
1076 1077 1078 1079 1080 1081 1082
			bhs[i] = NULL;
		}
	}
	mlog_exit(status);
	return status;
}

M
Mark Fasheh 已提交
1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101
/*
 * 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) +
1102
		ocfs2_rec_clusters(el, &el->l_recs[i]);
M
Mark Fasheh 已提交
1103 1104
}

1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122
/*
 * Change range of the branches in the right most path according to the leaf
 * extent block's rightmost record.
 */
static int ocfs2_adjust_rightmost_branch(handle_t *handle,
					 struct ocfs2_extent_tree *et)
{
	int status;
	struct ocfs2_path *path = NULL;
	struct ocfs2_extent_list *el;
	struct ocfs2_extent_rec *rec;

	path = ocfs2_new_path_from_et(et);
	if (!path) {
		status = -ENOMEM;
		return status;
	}

1123
	status = ocfs2_find_path(et->et_ci, path, UINT_MAX);
1124 1125 1126 1127 1128
	if (status < 0) {
		mlog_errno(status);
		goto out;
	}

1129
	status = ocfs2_extend_trans(handle, path_num_items(path));
1130 1131 1132 1133 1134
	if (status < 0) {
		mlog_errno(status);
		goto out;
	}

1135
	status = ocfs2_journal_access_path(et->et_ci, handle, path);
1136 1137 1138 1139 1140 1141 1142 1143
	if (status < 0) {
		mlog_errno(status);
		goto out;
	}

	el = path_leaf_el(path);
	rec = &el->l_recs[le32_to_cpu(el->l_next_free_rec) - 1];

1144
	ocfs2_adjust_rightmost_records(handle, et, path, rec);
1145 1146 1147 1148 1149 1150

out:
	ocfs2_free_path(path);
	return status;
}

1151 1152
/*
 * Add an entire tree branch to our inode. eb_bh is the extent block
1153
 * to start at, if we don't want to start the branch at the root
1154 1155 1156 1157 1158 1159
 * 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
1160
 * contain a single record with cluster count == 0.
1161
 */
1162
static int ocfs2_add_branch(handle_t *handle,
1163
			    struct ocfs2_extent_tree *et,
1164
			    struct buffer_head *eb_bh,
1165
			    struct buffer_head **last_eb_bh,
1166 1167 1168 1169 1170 1171 1172 1173 1174
			    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;
1175
	u32 new_cpos, root_end;
1176 1177 1178

	mlog_entry_void();

1179
	BUG_ON(!last_eb_bh || !*last_eb_bh);
1180 1181 1182 1183 1184

	if (eb_bh) {
		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
		el = &eb->h_list;
	} else
1185
		el = et->et_root_el;
1186 1187 1188 1189 1190 1191

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

	new_blocks = le16_to_cpu(el->l_tree_depth);

1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205
	eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data;
	new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list);
	root_end = ocfs2_sum_rightmost_rec(et->et_root_el);

	/*
	 * If there is a gap before the root end and the real end
	 * of the righmost leaf block, we need to remove the gap
	 * between new_cpos and root_end first so that the tree
	 * is consistent after we add a new branch(it will start
	 * from new_cpos).
	 */
	if (root_end > new_cpos) {
		mlog(0, "adjust the cluster end from %u to %u\n",
		     root_end, new_cpos);
1206
		status = ocfs2_adjust_rightmost_branch(handle, et);
1207 1208 1209 1210 1211 1212
		if (status) {
			mlog_errno(status);
			goto bail;
		}
	}

1213 1214 1215 1216 1217 1218 1219 1220 1221
	/* 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;
	}

1222
	status = ocfs2_create_new_meta_bhs(handle, et, new_blocks,
1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239
					   meta_ac, new_eb_bhs);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	/* 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;
1240 1241
		/* ocfs2_create_new_meta_bhs() should create it right! */
		BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
1242 1243
		eb_el = &eb->h_list;

1244
		status = ocfs2_journal_access_eb(handle, et->et_ci, bh,
1245
						 OCFS2_JOURNAL_ACCESS_CREATE);
1246 1247 1248 1249 1250 1251 1252 1253
		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
1254 1255 1256 1257 1258
		/*
		 * This actually counts as an empty extent as
		 * c_clusters == 0
		 */
		eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos);
1259
		eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno);
1260 1261 1262 1263 1264 1265
		/*
		 * 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);
1266 1267 1268
		if (!eb_el->l_tree_depth)
			new_last_eb_blk = le64_to_cpu(eb->h_blkno);

1269
		ocfs2_journal_dirty(handle, bh);
1270 1271 1272 1273 1274 1275 1276 1277 1278
		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. */
1279
	status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh,
1280
					 OCFS2_JOURNAL_ACCESS_WRITE);
1281 1282 1283 1284
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
1285
	status = ocfs2_et_root_journal_access(handle, et,
1286
					      OCFS2_JOURNAL_ACCESS_WRITE);
1287 1288 1289 1290 1291
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
	if (eb_bh) {
1292
		status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh,
1293
						 OCFS2_JOURNAL_ACCESS_WRITE);
1294 1295 1296 1297 1298 1299 1300
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

	/* Link the new branch into the rest of the tree (el will
1301
	 * either be on the root_bh, or the extent block passed in. */
1302 1303
	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
1304
	el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1305
	el->l_recs[i].e_int_clusters = 0;
1306 1307 1308 1309
	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. */
1310
	ocfs2_et_set_last_eb_blk(et, new_last_eb_blk);
1311

1312
	eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
1313 1314
	eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);

1315 1316 1317 1318
	ocfs2_journal_dirty(handle, *last_eb_bh);
	ocfs2_journal_dirty(handle, et->et_root_bh);
	if (eb_bh)
		ocfs2_journal_dirty(handle, eb_bh);
1319

1320 1321 1322 1323 1324 1325 1326 1327
	/*
	 * 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];

1328 1329 1330 1331
	status = 0;
bail:
	if (new_eb_bhs) {
		for (i = 0; i < new_blocks; i++)
1332
			brelse(new_eb_bhs[i]);
1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344
		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.
 */
1345
static int ocfs2_shift_tree_depth(handle_t *handle,
1346
				  struct ocfs2_extent_tree *et,
1347 1348 1349 1350
				  struct ocfs2_alloc_context *meta_ac,
				  struct buffer_head **ret_new_eb_bh)
{
	int status, i;
M
Mark Fasheh 已提交
1351
	u32 new_clusters;
1352 1353
	struct buffer_head *new_eb_bh = NULL;
	struct ocfs2_extent_block *eb;
1354
	struct ocfs2_extent_list  *root_el;
1355 1356 1357 1358
	struct ocfs2_extent_list  *eb_el;

	mlog_entry_void();

1359
	status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac,
1360 1361 1362 1363 1364 1365 1366
					   &new_eb_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	eb = (struct ocfs2_extent_block *) new_eb_bh->b_data;
1367 1368
	/* ocfs2_create_new_meta_bhs() should create it right! */
	BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
1369 1370

	eb_el = &eb->h_list;
1371
	root_el = et->et_root_el;
1372

1373
	status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh,
1374
					 OCFS2_JOURNAL_ACCESS_CREATE);
1375 1376 1377 1378 1379
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

1380 1381 1382 1383 1384
	/* 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];
1385

1386
	ocfs2_journal_dirty(handle, new_eb_bh);
1387

1388
	status = ocfs2_et_root_journal_access(handle, et,
1389
					      OCFS2_JOURNAL_ACCESS_WRITE);
1390 1391 1392 1393 1394
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

M
Mark Fasheh 已提交
1395 1396
	new_clusters = ocfs2_sum_rightmost_rec(eb_el);

1397 1398 1399 1400 1401 1402 1403 1404
	/* 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);
1405 1406 1407

	/* If this is our 1st tree depth shift, then last_eb_blk
	 * becomes the allocated extent block */
1408
	if (root_el->l_tree_depth == cpu_to_le16(1))
1409
		ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
1410

1411
	ocfs2_journal_dirty(handle, et->et_root_bh);
1412 1413 1414 1415 1416

	*ret_new_eb_bh = new_eb_bh;
	new_eb_bh = NULL;
	status = 0;
bail:
1417
	brelse(new_eb_bh);
1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431

	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'
 *
1432
 * 2) the search fails to find anything, but the root_el has room. We
1433 1434
 *    pass NULL back in *lowest_eb_bh, but still return '0'
 *
1435
 * 3) the search fails to find anything AND the root_el is full, in
1436 1437 1438 1439
 *    which case we return > 0
 *
 * return status < 0 indicates an error.
 */
1440
static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et,
1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453
				    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;

1454
	el = et->et_root_el;
1455 1456 1457

	while(le16_to_cpu(el->l_tree_depth) > 1) {
		if (le16_to_cpu(el->l_next_free_rec) == 0) {
1458 1459
			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
				    "Owner %llu has empty "
1460
				    "extent list (next_free_rec == 0)",
1461
				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
1462 1463 1464 1465 1466 1467
			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) {
1468 1469
			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
				    "Owner %llu has extent "
1470 1471
				    "list where extent # %d has no physical "
				    "block start",
1472
				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i);
1473 1474 1475 1476
			status = -EIO;
			goto bail;
		}

1477 1478
		brelse(bh);
		bh = NULL;
1479

1480
		status = ocfs2_read_extent_block(et->et_ci, blkno, &bh);
1481 1482 1483 1484
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
M
Mark Fasheh 已提交
1485 1486 1487 1488 1489 1490

		eb = (struct ocfs2_extent_block *) bh->b_data;
		el = &eb->h_list;

		if (le16_to_cpu(el->l_next_free_rec) <
		    le16_to_cpu(el->l_count)) {
1491
			brelse(lowest_bh);
M
Mark Fasheh 已提交
1492 1493 1494 1495 1496 1497 1498
			lowest_bh = bh;
			get_bh(lowest_bh);
		}
	}

	/* If we didn't find one and the fe doesn't have any room,
	 * then return '1' */
1499
	el = et->et_root_el;
1500
	if (!lowest_bh && (el->l_next_free_rec == el->l_count))
M
Mark Fasheh 已提交
1501 1502 1503 1504
		status = 1;

	*target_bh = lowest_bh;
bail:
1505
	brelse(bh);
M
Mark Fasheh 已提交
1506 1507 1508 1509 1510

	mlog_exit(status);
	return status;
}

1511 1512 1513 1514 1515 1516 1517
/*
 * 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.
1518 1519
 *
 * *last_eb_bh will be updated by ocfs2_add_branch().
1520
 */
1521 1522
static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et,
			   int *final_depth, struct buffer_head **last_eb_bh,
1523 1524 1525
			   struct ocfs2_alloc_context *meta_ac)
{
	int ret, shift;
1526
	struct ocfs2_extent_list *el = et->et_root_el;
1527
	int depth = le16_to_cpu(el->l_tree_depth);
1528 1529 1530 1531
	struct buffer_head *bh = NULL;

	BUG_ON(meta_ac == NULL);

1532
	shift = ocfs2_find_branch_target(et, &bh);
1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548
	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). */
1549
		ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh);
1550 1551 1552 1553 1554
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}
		depth++;
1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567
		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;
1568
			goto out;
1569
		}
1570 1571 1572 1573 1574
	}

	/* call ocfs2_add_branch to add the final part of the tree with
	 * the new data. */
	mlog(0, "add branch. bh = %p\n", bh);
1575
	ret = ocfs2_add_branch(handle, et, bh, last_eb_bh,
1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588
			       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
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
/*
 * 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
1620
	BUG_ON(el->l_next_free_rec == el->l_count && !has_empty);
M
Mark Fasheh 已提交
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

	/*
	 * 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;

}

1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700
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
1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711
/*
 * 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);

1712 1713
	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);

M
Mark Fasheh 已提交
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
	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.
 */
1745 1746 1747
int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et,
			    struct ocfs2_path *left,
			    struct ocfs2_path *right)
M
Mark Fasheh 已提交
1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762
{
	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,
1763
				"Owner %llu, left depth %u, right depth %u\n"
M
Mark Fasheh 已提交
1764
				"left leaf blk %llu, right leaf blk %llu\n",
1765 1766
				(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
				left->p_tree_depth, right->p_tree_depth,
M
Mark Fasheh 已提交
1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782
				(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.
 */
1783
static int __ocfs2_find_path(struct ocfs2_caching_info *ci,
M
Mark Fasheh 已提交
1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797
			     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;

	el = root_el;
	while (el->l_tree_depth) {
		if (le16_to_cpu(el->l_next_free_rec) == 0) {
1798 1799
			ocfs2_error(ocfs2_metadata_cache_get_super(ci),
				    "Owner %llu has empty extent list at "
M
Mark Fasheh 已提交
1800
				    "depth %u\n",
1801
				    (unsigned long long)ocfs2_metadata_cache_owner(ci),
M
Mark Fasheh 已提交
1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816
				    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) +
1817
				ocfs2_rec_clusters(el, rec);
M
Mark Fasheh 已提交
1818 1819 1820 1821 1822 1823
			if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
			    break;
		}

		blkno = le64_to_cpu(el->l_recs[i].e_blkno);
		if (blkno == 0) {
1824 1825
			ocfs2_error(ocfs2_metadata_cache_get_super(ci),
				    "Owner %llu has bad blkno in extent list "
M
Mark Fasheh 已提交
1826
				    "at depth %u (index %d)\n",
1827
				    (unsigned long long)ocfs2_metadata_cache_owner(ci),
M
Mark Fasheh 已提交
1828 1829 1830 1831 1832 1833 1834
				    le16_to_cpu(el->l_tree_depth), i);
			ret = -EROFS;
			goto out;
		}

		brelse(bh);
		bh = NULL;
1835
		ret = ocfs2_read_extent_block(ci, blkno, &bh);
M
Mark Fasheh 已提交
1836 1837 1838 1839 1840 1841 1842 1843 1844 1845
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		eb = (struct ocfs2_extent_block *) bh->b_data;
		el = &eb->h_list;

		if (le16_to_cpu(el->l_next_free_rec) >
		    le16_to_cpu(el->l_count)) {
1846 1847
			ocfs2_error(ocfs2_metadata_cache_get_super(ci),
				    "Owner %llu has bad count in extent list "
M
Mark Fasheh 已提交
1848
				    "at block %llu (next free=%u, count=%u)\n",
1849
				    (unsigned long long)ocfs2_metadata_cache_owner(ci),
M
Mark Fasheh 已提交
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 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892
				    (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++;
}
1893 1894
int ocfs2_find_path(struct ocfs2_caching_info *ci,
		    struct ocfs2_path *path, u32 cpos)
M
Mark Fasheh 已提交
1895 1896 1897 1898 1899
{
	struct find_path_data data;

	data.index = 1;
	data.path = path;
1900
	return __ocfs2_find_path(ci, path_root_el(path), cpos,
M
Mark Fasheh 已提交
1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924
				 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.
 */
1925 1926 1927
int ocfs2_find_leaf(struct ocfs2_caching_info *ci,
		    struct ocfs2_extent_list *root_el, u32 cpos,
		    struct buffer_head **leaf_bh)
M
Mark Fasheh 已提交
1928 1929 1930 1931
{
	int ret;
	struct buffer_head *bh = NULL;

1932
	ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh);
M
Mark Fasheh 已提交
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
	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);
1972 1973
	if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) {
		BUG_ON(right_child_el->l_tree_depth);
1974 1975 1976
		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
1977
	left_clusters -= le32_to_cpu(left_rec->e_cpos);
1978
	left_rec->e_int_clusters = cpu_to_le32(left_clusters);
M
Mark Fasheh 已提交
1979 1980 1981

	/*
	 * Calculate the rightmost cluster count boundary before
1982
	 * moving cpos - we will need to adjust clusters after
M
Mark Fasheh 已提交
1983 1984 1985
	 * updating e_cpos to keep the same highest cluster count.
	 */
	right_end = le32_to_cpu(right_rec->e_cpos);
1986
	right_end += le32_to_cpu(right_rec->e_int_clusters);
M
Mark Fasheh 已提交
1987 1988 1989 1990 1991

	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);
1992
	right_rec->e_int_clusters = cpu_to_le32(right_end);
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 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034
}

/*
 * 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.
2035 2036
 *   - 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
2037
 */
2038
static void ocfs2_complete_edge_insert(handle_t *handle,
M
Mark Fasheh 已提交
2039 2040 2041 2042
				       struct ocfs2_path *left_path,
				       struct ocfs2_path *right_path,
				       int subtree_index)
{
2043
	int i, idx;
M
Mark Fasheh 已提交
2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080
	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);

2081 2082
		ocfs2_journal_dirty(handle, left_path->p_node[i].bh);
		ocfs2_journal_dirty(handle, right_path->p_node[i].bh);
M
Mark Fasheh 已提交
2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105

		/*
		 * 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;

2106
	ocfs2_journal_dirty(handle, root_bh);
M
Mark Fasheh 已提交
2107 2108
}

2109 2110
static int ocfs2_rotate_subtree_right(handle_t *handle,
				      struct ocfs2_extent_tree *et,
M
Mark Fasheh 已提交
2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125
				      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) {
2126
		ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
M
Mark Fasheh 已提交
2127 2128
			    "Inode %llu has non-full interior leaf node %llu"
			    "(next free = %u)",
2129
			    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
M
Mark Fasheh 已提交
2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144
			    (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);

2145
	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
2146
					   subtree_index);
M
Mark Fasheh 已提交
2147 2148 2149 2150 2151 2152
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
2153
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2154
						   right_path, i);
M
Mark Fasheh 已提交
2155 2156 2157 2158 2159
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2160
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2161
						   left_path, i);
M
Mark Fasheh 已提交
2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173
		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",
2174
			(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
M
Mark Fasheh 已提交
2175 2176 2177 2178
			(unsigned long long)right_leaf_bh->b_blocknr);

	ocfs2_create_empty_extent(right_el);

2179
	ocfs2_journal_dirty(handle, right_leaf_bh);
M
Mark Fasheh 已提交
2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197

	/* 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);

2198
	ocfs2_journal_dirty(handle, left_leaf_bh);
M
Mark Fasheh 已提交
2199

2200 2201
	ocfs2_complete_edge_insert(handle, left_path, right_path,
				   subtree_index);
M
Mark Fasheh 已提交
2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219

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;

2220 2221
	BUG_ON(path->p_tree_depth == 0);

M
Mark Fasheh 已提交
2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255
	*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);
2256 2257 2258
				*cpos = *cpos + ocfs2_rec_clusters(el,
							   &el->l_recs[j - 1]);
				*cpos = *cpos - 1;
M
Mark Fasheh 已提交
2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281
				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;
}

2282 2283 2284 2285 2286
/*
 * 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
2287
static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth,
2288
					   int op_credits,
M
Mark Fasheh 已提交
2289 2290
					   struct ocfs2_path *path)
{
2291
	int ret = 0;
2292
	int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits;
M
Mark Fasheh 已提交
2293

2294
	if (handle->h_buffer_credits < credits)
2295 2296
		ret = ocfs2_extend_trans(handle,
					 credits - handle->h_buffer_credits);
M
Mark Fasheh 已提交
2297

2298
	return ret;
M
Mark Fasheh 已提交
2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326
}

/*
 * 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;
}

2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349
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
2350 2351 2352 2353 2354 2355 2356
/*
 * 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).
 *
2357
 * Upon successful return from this function:
M
Mark Fasheh 已提交
2358 2359 2360 2361 2362 2363 2364 2365
 *
 * - 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().
 */
2366
static int ocfs2_rotate_tree_right(handle_t *handle,
2367
				   struct ocfs2_extent_tree *et,
2368
				   enum ocfs2_split_type split,
M
Mark Fasheh 已提交
2369 2370 2371 2372
				   u32 insert_cpos,
				   struct ocfs2_path *right_path,
				   struct ocfs2_path **ret_left_path)
{
2373
	int ret, start, orig_credits = handle->h_buffer_credits;
M
Mark Fasheh 已提交
2374 2375
	u32 cpos;
	struct ocfs2_path *left_path = NULL;
2376
	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
M
Mark Fasheh 已提交
2377 2378 2379

	*ret_left_path = NULL;

2380
	left_path = ocfs2_new_path_from_path(right_path);
M
Mark Fasheh 已提交
2381 2382 2383 2384 2385 2386
	if (!left_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

2387
	ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
M
Mark Fasheh 已提交
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
	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);

2425
		ret = ocfs2_find_path(et->et_ci, left_path, cpos);
M
Mark Fasheh 已提交
2426 2427 2428 2429 2430 2431 2432
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		mlog_bug_on_msg(path_leaf_bh(left_path) ==
				path_leaf_bh(right_path),
2433
				"Owner %llu: error during insert of %u "
M
Mark Fasheh 已提交
2434 2435
				"(left path cpos %u) results in two identical "
				"paths ending at %llu\n",
2436 2437
				(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
				insert_cpos, cpos,
M
Mark Fasheh 已提交
2438 2439 2440
				(unsigned long long)
				path_leaf_bh(left_path)->b_blocknr);

2441 2442
		if (split == SPLIT_NONE &&
		    ocfs2_rotate_requires_path_adjustment(left_path,
M
Mark Fasheh 已提交
2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462
							  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;
		}

2463
		start = ocfs2_find_subtree_root(et, left_path, right_path);
M
Mark Fasheh 已提交
2464 2465 2466 2467 2468 2469 2470

		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,
2471
						      orig_credits, right_path);
M
Mark Fasheh 已提交
2472 2473 2474 2475 2476
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2477
		ret = ocfs2_rotate_subtree_right(handle, et, left_path,
M
Mark Fasheh 已提交
2478 2479 2480 2481 2482 2483
						 right_path, start);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501
		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
2502 2503 2504 2505 2506 2507 2508
		/*
		 * 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);

2509
		ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
M
Mark Fasheh 已提交
2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

out:
	ocfs2_free_path(left_path);

out_ret_path:
	return ret;
}

2523 2524
static int ocfs2_update_edge_lengths(handle_t *handle,
				     struct ocfs2_extent_tree *et,
2525
				     int subtree_index, struct ocfs2_path *path)
M
Mark Fasheh 已提交
2526
{
2527
	int i, idx, ret;
M
Mark Fasheh 已提交
2528
	struct ocfs2_extent_rec *rec;
2529 2530 2531
	struct ocfs2_extent_list *el;
	struct ocfs2_extent_block *eb;
	u32 range;
M
Mark Fasheh 已提交
2532

2533 2534 2535 2536 2537 2538 2539 2540 2541
	/*
	 * In normal tree rotation process, we will never touch the
	 * tree branch above subtree_index and ocfs2_extend_rotate_transaction
	 * doesn't reserve the credits for them either.
	 *
	 * But we do have a special case here which will update the rightmost
	 * records for all the bh in the path.
	 * So we have to allocate extra credits and access them.
	 */
2542
	ret = ocfs2_extend_trans(handle, subtree_index);
2543 2544 2545 2546 2547
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

2548
	ret = ocfs2_journal_access_path(et->et_ci, handle, path);
2549 2550 2551 2552 2553
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

2554 2555 2556
	/* 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
2557

2558 2559 2560 2561 2562
	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
2563

2564 2565 2566 2567
	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
2568

2569 2570
		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
2571

2572
		ocfs2_journal_dirty(handle, path->p_node[i].bh);
M
Mark Fasheh 已提交
2573
	}
2574 2575
out:
	return ret;
M
Mark Fasheh 已提交
2576 2577
}

2578 2579
static void ocfs2_unlink_path(handle_t *handle,
			      struct ocfs2_extent_tree *et,
2580 2581
			      struct ocfs2_cached_dealloc_ctxt *dealloc,
			      struct ocfs2_path *path, int unlink_start)
M
Mark Fasheh 已提交
2582
{
2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600
	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",
2601
			     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2602 2603 2604 2605
			     (unsigned long long)le64_to_cpu(eb->h_blkno),
			     le16_to_cpu(el->l_next_free_rec));

			ocfs2_journal_dirty(handle, bh);
2606
			ocfs2_remove_from_cache(et->et_ci, bh);
2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618
			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);

2619
		ocfs2_remove_from_cache(et->et_ci, bh);
2620
	}
M
Mark Fasheh 已提交
2621 2622
}

2623 2624
static void ocfs2_unlink_subtree(handle_t *handle,
				 struct ocfs2_extent_tree *et,
2625 2626 2627 2628
				 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
2629
{
2630 2631 2632
	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
2633
	struct ocfs2_extent_list *el;
2634
	struct ocfs2_extent_block *eb;
M
Mark Fasheh 已提交
2635

2636
	el = path_leaf_el(left_path);
M
Mark Fasheh 已提交
2637

2638
	eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data;
2639

2640 2641 2642
	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
2643

2644
	BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec));
M
Mark Fasheh 已提交
2645

2646 2647 2648 2649 2650 2651 2652 2653 2654
	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));

2655
	ocfs2_unlink_path(handle, et, dealloc, right_path,
2656 2657 2658
			  subtree_index + 1);
}

2659 2660
static int ocfs2_rotate_subtree_left(handle_t *handle,
				     struct ocfs2_extent_tree *et,
2661 2662 2663 2664
				     struct ocfs2_path *left_path,
				     struct ocfs2_path *right_path,
				     int subtree_index,
				     struct ocfs2_cached_dealloc_ctxt *dealloc,
2665
				     int *deleted)
2666 2667
{
	int ret, i, del_right_subtree = 0, right_has_empty = 0;
2668
	struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path);
2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680
	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
2681

2682 2683
	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
2684
		/*
2685 2686 2687 2688 2689 2690 2691 2692 2693 2694
		 * 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
2695 2696
		 */

2697 2698 2699 2700
		if (eb->h_next_leaf_blk != 0ULL)
			return -EAGAIN;

		if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) {
2701
			ret = ocfs2_journal_access_eb(handle, et->et_ci,
2702 2703
						      path_leaf_bh(right_path),
						      OCFS2_JOURNAL_ACCESS_WRITE);
M
Mark Fasheh 已提交
2704 2705 2706 2707 2708
			if (ret) {
				mlog_errno(ret);
				goto out;
			}

2709 2710 2711
			ocfs2_remove_empty_extent(right_leaf_el);
		} else
			right_has_empty = 1;
M
Mark Fasheh 已提交
2712 2713
	}

2714 2715 2716 2717 2718 2719
	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.
		 */
2720
		ret = ocfs2_et_root_journal_access(handle, et,
2721
						   OCFS2_JOURNAL_ACCESS_WRITE);
2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735
		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);

2736
	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
2737
					   subtree_index);
2738 2739 2740 2741 2742 2743
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
2744
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2745
						   right_path, i);
2746 2747 2748 2749 2750
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2751
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2752
						   left_path, i);
2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779
		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);
	}

2780 2781
	ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
	ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
2782 2783

	if (del_right_subtree) {
2784
		ocfs2_unlink_subtree(handle, et, left_path, right_path,
2785
				     subtree_index, dealloc);
2786
		ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
2787 2788 2789 2790 2791
						left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
2792 2793

		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
2794
		ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
2795 2796 2797 2798 2799 2800 2801 2802 2803

		/*
		 * 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);

2804
		ocfs2_journal_dirty(handle, et_root_bh);
2805 2806 2807

		*deleted = 1;
	} else
2808
		ocfs2_complete_edge_insert(handle, left_path, right_path,
2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823
					   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().
 */
2824 2825
int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
				   struct ocfs2_path *path, u32 *cpos)
2826 2827 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 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893
{
	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;
}

2894 2895
static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle,
					    struct ocfs2_extent_tree *et,
2896
					    struct ocfs2_path *path)
2897 2898
{
	int ret;
2899 2900
	struct buffer_head *bh = path_leaf_bh(path);
	struct ocfs2_extent_list *el = path_leaf_el(path);
2901 2902 2903 2904

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

2905
	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
2906
					   path_num_items(path) - 1);
2907 2908 2909 2910 2911 2912
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	ocfs2_remove_empty_extent(el);
2913
	ocfs2_journal_dirty(handle, bh);
2914 2915 2916 2917 2918

out:
	return ret;
}

2919 2920 2921
static int __ocfs2_rotate_tree_left(handle_t *handle,
				    struct ocfs2_extent_tree *et,
				    int orig_credits,
2922 2923
				    struct ocfs2_path *path,
				    struct ocfs2_cached_dealloc_ctxt *dealloc,
2924
				    struct ocfs2_path **empty_extent_path)
2925 2926 2927 2928 2929
{
	int ret, subtree_root, deleted;
	u32 right_cpos;
	struct ocfs2_path *left_path = NULL;
	struct ocfs2_path *right_path = NULL;
2930
	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
2931 2932 2933 2934 2935

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

	*empty_extent_path = NULL;

2936
	ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
2937 2938 2939 2940 2941
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

2942
	left_path = ocfs2_new_path_from_path(path);
2943 2944 2945 2946 2947 2948 2949 2950
	if (!left_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

	ocfs2_cp_path(left_path, path);

2951
	right_path = ocfs2_new_path_from_path(path);
2952 2953 2954 2955 2956 2957 2958
	if (!right_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

	while (right_cpos) {
2959
		ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
2960 2961 2962 2963 2964
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2965
		subtree_root = ocfs2_find_subtree_root(et, left_path,
2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980
						       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;
		}

2981 2982 2983 2984
		/*
		 * Caller might still want to make changes to the
		 * tree root, so re-add it to the journal here.
		 */
2985
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2986
						   left_path, 0);
2987 2988 2989 2990 2991
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2992
		ret = ocfs2_rotate_subtree_left(handle, et, left_path,
2993
						right_path, subtree_root,
2994
						dealloc, &deleted);
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
		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);

3021
		ret = ocfs2_find_cpos_for_right_leaf(sb, left_path,
3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035
						     &right_cpos);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

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

	return ret;
}

3036 3037
static int ocfs2_remove_rightmost_path(handle_t *handle,
				struct ocfs2_extent_tree *et,
3038
				struct ocfs2_path *path,
3039
				struct ocfs2_cached_dealloc_ctxt *dealloc)
3040 3041 3042 3043 3044 3045 3046 3047
{
	int ret, subtree_index;
	u32 cpos;
	struct ocfs2_path *left_path = NULL;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;


3048
	ret = ocfs2_et_sanity_check(et);
3049 3050
	if (ret)
		goto out;
3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062
	/*
	 * 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;
	}

3063
	ret = ocfs2_journal_access_path(et->et_ci, handle, path);
3064 3065 3066 3067 3068
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

3069 3070
	ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
					    path, &cpos);
3071 3072 3073 3074 3075 3076 3077 3078 3079 3080
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	if (cpos) {
		/*
		 * We have a path to the left of this one - it needs
		 * an update too.
		 */
3081
		left_path = ocfs2_new_path_from_path(path);
3082 3083 3084 3085 3086 3087
		if (!left_path) {
			ret = -ENOMEM;
			mlog_errno(ret);
			goto out;
		}

3088
		ret = ocfs2_find_path(et->et_ci, left_path, cpos);
3089 3090 3091 3092 3093
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

3094
		ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
3095 3096 3097 3098 3099
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

3100
		subtree_index = ocfs2_find_subtree_root(et, left_path, path);
3101

3102
		ocfs2_unlink_subtree(handle, et, left_path, path,
3103
				     subtree_index, dealloc);
3104
		ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
3105 3106 3107 3108 3109
						left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
3110 3111

		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
3112
		ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
3113 3114 3115 3116 3117
	} else {
		/*
		 * 'path' is also the leftmost path which
		 * means it must be the only one. This gets
		 * handled differently because we want to
3118
		 * revert the root back to having extents
3119 3120
		 * in-line.
		 */
3121
		ocfs2_unlink_path(handle, et, dealloc, path, 1);
3122

3123
		el = et->et_root_el;
3124 3125 3126 3127
		el->l_tree_depth = 0;
		el->l_next_free_rec = 0;
		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));

3128
		ocfs2_et_set_last_eb_blk(et, 0);
3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153
	}

	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.
 */
3154 3155
static int ocfs2_rotate_tree_left(handle_t *handle,
				  struct ocfs2_extent_tree *et,
3156
				  struct ocfs2_path *path,
3157
				  struct ocfs2_cached_dealloc_ctxt *dealloc)
3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170
{
	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:
		/*
3171
		 * Inline extents. This is trivially handled, so do
3172 3173
		 * it up front.
		 */
3174
		ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path);
3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189
		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
3190
	 *  2b) we need to bring the root back to inline extents.
3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205
	 */

	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;
3206 3207 3208
			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
				    "Owner %llu has empty extent block at %llu",
				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221
				    (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.
		 */

3222 3223
		ret = ocfs2_remove_rightmost_path(handle, et, path,
						  dealloc);
3224 3225 3226 3227 3228 3229 3230 3231 3232 3233
		if (ret)
			mlog_errno(ret);
		goto out;
	}

	/*
	 * Now we can loop, remembering the path we get from -EAGAIN
	 * and restarting from there.
	 */
try_rotate:
3234 3235
	ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path,
				       dealloc, &restart_path);
3236 3237 3238 3239 3240 3241 3242 3243 3244
	if (ret && ret != -EAGAIN) {
		mlog_errno(ret);
		goto out;
	}

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

3245
		ret = __ocfs2_rotate_tree_left(handle, et, orig_credits,
3246
					       tmp_path, dealloc,
3247
					       &restart_path);
3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297
		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));
	}
}

3298
static int ocfs2_get_right_path(struct ocfs2_extent_tree *et,
3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314
				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);

3315 3316
	ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
					     left_path, &right_cpos);
3317 3318 3319 3320 3321 3322 3323 3324
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

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

3325
	right_path = ocfs2_new_path_from_path(left_path);
3326 3327 3328 3329 3330 3331
	if (!right_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

3332
	ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

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

3345 3346
/*
 * Remove split_rec clusters from the record at index and merge them
3347 3348 3349 3350
 * 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.
3351
 */
3352
static int ocfs2_merge_rec_right(struct ocfs2_path *left_path,
3353
				 handle_t *handle,
3354
				 struct ocfs2_extent_tree *et,
3355 3356
				 struct ocfs2_extent_rec *split_rec,
				 int index)
3357
{
3358
	int ret, next_free, i;
3359 3360 3361
	unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
	struct ocfs2_extent_rec *left_rec;
	struct ocfs2_extent_rec *right_rec;
3362 3363 3364 3365 3366 3367
	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;
3368 3369 3370

	BUG_ON(index >= le16_to_cpu(el->l_next_free_rec));
	left_rec = &el->l_recs[index];
3371

A
Al Viro 已提交
3372
	if (index == le16_to_cpu(el->l_next_free_rec) - 1 &&
3373 3374
	    le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) {
		/* we meet with a cross extent block merge. */
3375
		ret = ocfs2_get_right_path(et, left_path, &right_path);
3376 3377 3378 3379 3380 3381 3382 3383 3384 3385
		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
3386
			BUG_ON(next_free <= 1);
3387 3388 3389 3390 3391 3392 3393
			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));

3394 3395
		subtree_index = ocfs2_find_subtree_root(et, left_path,
							right_path);
3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407

		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);

3408
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
3409
						   subtree_index);
3410 3411 3412 3413 3414 3415 3416
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		for (i = subtree_index + 1;
		     i < path_num_items(right_path); i++) {
3417
			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
3418
							   right_path, i);
3419 3420 3421 3422 3423
			if (ret) {
				mlog_errno(ret);
				goto out;
			}

3424
			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
3425
							   left_path, i);
3426 3427 3428 3429 3430 3431 3432 3433 3434 3435
			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];
	}
3436

3437
	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path,
3438
					   path_num_items(left_path) - 1);
3439 3440 3441 3442 3443 3444 3445 3446 3447
	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,
3448 3449
		     -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
					       split_clusters));
3450 3451 3452 3453
	le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters);

	ocfs2_cleanup_merge(el, index);

3454
	ocfs2_journal_dirty(handle, bh);
3455
	if (right_path) {
3456
		ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
3457 3458
		ocfs2_complete_edge_insert(handle, left_path, right_path,
					   subtree_index);
3459 3460 3461 3462 3463 3464 3465
	}
out:
	if (right_path)
		ocfs2_free_path(right_path);
	return ret;
}

3466
static int ocfs2_get_left_path(struct ocfs2_extent_tree *et,
3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478
			       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);

3479
	ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3480 3481 3482 3483 3484 3485 3486 3487 3488
					    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);

3489
	left_path = ocfs2_new_path_from_path(right_path);
3490 3491 3492 3493 3494 3495
	if (!left_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

3496
	ret = ocfs2_find_path(et->et_ci, left_path, left_cpos);
3497 3498 3499 3500 3501 3502
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	*ret_left_path = left_path;
3503
out:
3504 3505
	if (ret)
		ocfs2_free_path(left_path);
3506 3507 3508 3509 3510
	return ret;
}

/*
 * Remove split_rec clusters from the record at index and merge them
3511 3512 3513 3514 3515 3516 3517
 * 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.
3518
 */
3519
static int ocfs2_merge_rec_left(struct ocfs2_path *right_path,
3520
				handle_t *handle,
3521
				struct ocfs2_extent_tree *et,
3522
				struct ocfs2_extent_rec *split_rec,
3523 3524
				struct ocfs2_cached_dealloc_ctxt *dealloc,
				int index)
3525
{
3526
	int ret, i, subtree_index = 0, has_empty_extent = 0;
3527 3528 3529
	unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
	struct ocfs2_extent_rec *left_rec;
	struct ocfs2_extent_rec *right_rec;
3530 3531 3532 3533 3534
	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;
3535

3536
	BUG_ON(index < 0);
3537 3538

	right_rec = &el->l_recs[index];
3539 3540
	if (index == 0) {
		/* we meet with a cross extent block merge. */
3541
		ret = ocfs2_get_left_path(et, right_path, &left_path);
3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556
		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));

3557 3558
		subtree_index = ocfs2_find_subtree_root(et, left_path,
							right_path);
3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570

		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);

3571
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
3572
						   subtree_index);
3573 3574 3575 3576 3577 3578 3579
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		for (i = subtree_index + 1;
		     i < path_num_items(right_path); i++) {
3580
			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
3581
							   right_path, i);
3582 3583 3584 3585 3586
			if (ret) {
				mlog_errno(ret);
				goto out;
			}

3587
			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
3588
							   left_path, i);
3589 3590 3591 3592 3593 3594 3595 3596 3597 3598
			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;
	}
3599

3600
	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
3601
					   path_num_items(right_path) - 1);
3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613
	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;
3614
	} else
3615 3616 3617 3618
		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,
3619 3620
		     ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
					      split_clusters));
3621 3622 3623 3624
	le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters);

	ocfs2_cleanup_merge(el, index);

3625
	ocfs2_journal_dirty(handle, bh);
3626
	if (left_path) {
3627
		ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
3628 3629 3630 3631 3632 3633 3634 3635 3636

		/*
		 * 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) {

3637
			ret = ocfs2_remove_rightmost_path(handle, et,
3638
							  right_path,
3639
							  dealloc);
3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650
			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
3651
			ocfs2_complete_edge_insert(handle, left_path,
3652 3653
						   right_path, subtree_index);
	}
3654
out:
3655 3656
	if (left_path)
		ocfs2_free_path(left_path);
3657 3658 3659
	return ret;
}

3660 3661
static int ocfs2_try_to_merge_extent(handle_t *handle,
				     struct ocfs2_extent_tree *et,
3662
				     struct ocfs2_path *path,
3663 3664 3665
				     int split_index,
				     struct ocfs2_extent_rec *split_rec,
				     struct ocfs2_cached_dealloc_ctxt *dealloc,
3666
				     struct ocfs2_merge_ctxt *ctxt)
3667
{
T
Tao Mao 已提交
3668
	int ret = 0;
3669
	struct ocfs2_extent_list *el = path_leaf_el(path);
3670 3671 3672 3673
	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];

	BUG_ON(ctxt->c_contig_type == CONTIG_NONE);

T
Tao Mao 已提交
3674 3675 3676 3677 3678 3679 3680 3681
	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.
		 */
3682
		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
T
Tao Mao 已提交
3683 3684 3685
		if (ret) {
			mlog_errno(ret);
			goto out;
3686
		}
T
Tao Mao 已提交
3687 3688
		split_index--;
		rec = &el->l_recs[split_index];
3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705
	}

	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.
3706 3707 3708 3709
		 *
		 * 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.
3710
		 */
3711
		ret = ocfs2_merge_rec_right(path, handle, et, split_rec,
3712
					    split_index);
3713 3714 3715 3716 3717 3718 3719 3720 3721 3722
		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]));

3723
		/* The merge left us with an empty extent, remove it. */
3724
		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
3725 3726 3727 3728
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
3729

3730 3731 3732 3733
		rec = &el->l_recs[split_index];

		/*
		 * Note that we don't pass split_rec here on purpose -
3734
		 * we've merged it into the rec already.
3735
		 */
3736 3737
		ret = ocfs2_merge_rec_left(path, handle, et, rec,
					   dealloc, split_index);
3738

3739 3740 3741 3742 3743
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

3744
		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760
		/*
		 * 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) {
3761 3762
			ret = ocfs2_merge_rec_left(path, handle, et,
						   split_rec, dealloc,
3763 3764 3765 3766 3767 3768
						   split_index);
			if (ret) {
				mlog_errno(ret);
				goto out;
			}
		} else {
3769
			ret = ocfs2_merge_rec_right(path, handle,
3770
						    et, split_rec,
3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782
						    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.
			 */
3783 3784
			ret = ocfs2_rotate_tree_left(handle, et, path,
						     dealloc);
3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 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
			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.
 */
3830 3831
static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et,
				 struct ocfs2_extent_rec *insert_rec,
3832
				 struct ocfs2_extent_list *el,
3833
				 struct ocfs2_insert_type *insert)
3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844
{
	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];
3845 3846
		ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
					insert->ins_split, rec,
3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887
					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),
3888
				"owner %llu, depth %u, count %u, next free %u, "
3889 3890
				"rec.cpos %u, rec.clusters %u, "
				"insert.cpos %u, insert.clusters %u\n",
3891
				ocfs2_metadata_cache_owner(et->et_ci),
3892 3893 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
				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);
}

3919 3920
static void ocfs2_adjust_rightmost_records(handle_t *handle,
					   struct ocfs2_extent_tree *et,
3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935
					   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
3936 3937
		next_free = le16_to_cpu(el->l_next_free_rec);
		if (next_free == 0) {
3938 3939 3940
			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
				    "Owner %llu has a bad extent list",
				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
M
Mark Fasheh 已提交
3941
			ret = -EIO;
3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952
			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));

3953
		ocfs2_journal_dirty(handle, bh);
3954 3955 3956
	}
}

3957 3958
static int ocfs2_append_rec_to_path(handle_t *handle,
				    struct ocfs2_extent_tree *et,
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 3984 3985
				    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;

3986 3987
		ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
						    right_path, &left_cpos);
3988 3989
		if (ret) {
			mlog_errno(ret);
M
Mark Fasheh 已提交
3990 3991 3992
			goto out;
		}

3993 3994 3995
		mlog(0, "Append may need a left path update. cpos: %u, "
		     "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos),
		     left_cpos);
3996

3997 3998 3999 4000 4001
		/*
		 * No need to worry if the append is already in the
		 * leftmost leaf.
		 */
		if (left_cpos) {
4002
			left_path = ocfs2_new_path_from_path(right_path);
4003 4004 4005 4006 4007
			if (!left_path) {
				ret = -ENOMEM;
				mlog_errno(ret);
				goto out;
			}
M
Mark Fasheh 已提交
4008

4009
			ret = ocfs2_find_path(et->et_ci, left_path,
4010
					      left_cpos);
4011 4012 4013 4014
			if (ret) {
				mlog_errno(ret);
				goto out;
			}
M
Mark Fasheh 已提交
4015

4016 4017 4018 4019 4020 4021
			/*
			 * ocfs2_insert_path() will pass the left_path to the
			 * journal for us.
			 */
		}
	}
M
Mark Fasheh 已提交
4022

4023
	ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
4024 4025 4026
	if (ret) {
		mlog_errno(ret);
		goto out;
M
Mark Fasheh 已提交
4027 4028
	}

4029
	ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec);
4030

M
Mark Fasheh 已提交
4031 4032 4033 4034 4035 4036 4037 4038 4039
	*ret_left_path = left_path;
	ret = 0;
out:
	if (ret != 0)
		ocfs2_free_path(left_path);

	return ret;
}

4040
static void ocfs2_split_record(struct ocfs2_extent_tree *et,
4041 4042 4043 4044 4045 4046 4047 4048 4049 4050
			       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;

4051
	right_el = path_leaf_el(right_path);
4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112
	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];
4113 4114
	ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
				split, rec, split_rec);
4115 4116 4117
	ocfs2_rotate_leaf(insert_el, split_rec);
}

M
Mark Fasheh 已提交
4118
/*
4119 4120
 * 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
4121 4122 4123 4124 4125
 *
 * 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.
 */
4126
static int ocfs2_insert_path(handle_t *handle,
4127
			     struct ocfs2_extent_tree *et,
M
Mark Fasheh 已提交
4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142
			     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) {
		/*
		 * 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.
		 */
4143
		ret = ocfs2_extend_trans(handle, left_path->p_tree_depth);
M
Mark Fasheh 已提交
4144 4145 4146 4147 4148
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}

4149
		ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
M
Mark Fasheh 已提交
4150 4151 4152 4153 4154 4155
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}
	}

4156 4157 4158 4159
	/*
	 * Pass both paths to the journal. The majority of inserts
	 * will be touching all components anyway.
	 */
4160
	ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
4161 4162 4163 4164 4165
	if (ret < 0) {
		mlog_errno(ret);
		goto out;
	}

4166 4167 4168
	if (insert->ins_split != SPLIT_NONE) {
		/*
		 * We could call ocfs2_insert_at_leaf() for some types
J
Joe Perches 已提交
4169
		 * of splits, but it's easier to just let one separate
4170 4171
		 * function sort it all out.
		 */
4172
		ocfs2_split_record(et, left_path, right_path,
4173
				   insert_rec, insert->ins_split);
4174 4175 4176 4177 4178 4179 4180

		/*
		 * 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)
4181 4182
			ocfs2_journal_dirty(handle,
					    path_leaf_bh(left_path));
4183
	} else
4184 4185
		ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path),
				     insert);
M
Mark Fasheh 已提交
4186

4187
	ocfs2_journal_dirty(handle, leaf_bh);
M
Mark Fasheh 已提交
4188 4189 4190 4191 4192 4193 4194 4195

	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?
		 */
4196
		subtree_index = ocfs2_find_subtree_root(et, left_path,
M
Mark Fasheh 已提交
4197
							right_path);
4198 4199
		ocfs2_complete_edge_insert(handle, left_path, right_path,
					   subtree_index);
M
Mark Fasheh 已提交
4200 4201 4202 4203 4204 4205 4206
	}

	ret = 0;
out:
	return ret;
}

4207
static int ocfs2_do_insert_extent(handle_t *handle,
4208
				  struct ocfs2_extent_tree *et,
M
Mark Fasheh 已提交
4209 4210 4211 4212 4213 4214 4215 4216 4217
				  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;

4218
	el = et->et_root_el;
M
Mark Fasheh 已提交
4219

4220
	ret = ocfs2_et_root_journal_access(handle, et,
4221
					   OCFS2_JOURNAL_ACCESS_WRITE);
M
Mark Fasheh 已提交
4222 4223 4224 4225 4226 4227
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	if (le16_to_cpu(el->l_tree_depth) == 0) {
4228
		ocfs2_insert_at_leaf(et, insert_rec, el, type);
M
Mark Fasheh 已提交
4229 4230 4231
		goto out_update_clusters;
	}

4232
	right_path = ocfs2_new_path_from_et(et);
M
Mark Fasheh 已提交
4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250
	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;
	}

4251
	ret = ocfs2_find_path(et->et_ci, right_path, cpos);
M
Mark Fasheh 已提交
4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269
	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) {
4270
		ret = ocfs2_rotate_tree_right(handle, et, type->ins_split,
M
Mark Fasheh 已提交
4271 4272 4273 4274 4275 4276
					      le32_to_cpu(insert_rec->e_cpos),
					      right_path, &left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
4277 4278 4279 4280 4281

		/*
		 * ocfs2_rotate_tree_right() might have extended the
		 * transaction without re-journaling our tree root.
		 */
4282
		ret = ocfs2_et_root_journal_access(handle, et,
4283
						   OCFS2_JOURNAL_ACCESS_WRITE);
4284 4285 4286 4287
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
M
Mark Fasheh 已提交
4288 4289
	} else if (type->ins_appending == APPEND_TAIL
		   && type->ins_contig != CONTIG_LEFT) {
4290
		ret = ocfs2_append_rec_to_path(handle, et, insert_rec,
M
Mark Fasheh 已提交
4291 4292 4293 4294 4295 4296 4297
					       right_path, &left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

4298
	ret = ocfs2_insert_path(handle, et, left_path, right_path,
M
Mark Fasheh 已提交
4299 4300 4301 4302 4303 4304 4305
				insert_rec, type);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

out_update_clusters:
4306
	if (type->ins_split == SPLIT_NONE)
4307
		ocfs2_et_update_clusters(et,
4308
					 le16_to_cpu(insert_rec->e_leaf_clusters));
M
Mark Fasheh 已提交
4309

4310
	ocfs2_journal_dirty(handle, et->et_root_bh);
M
Mark Fasheh 已提交
4311 4312 4313 4314 4315 4316 4317 4318

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

	return ret;
}

4319
static enum ocfs2_contig_type
4320 4321
ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et,
			       struct ocfs2_path *path,
4322 4323 4324
			       struct ocfs2_extent_list *el, int index,
			       struct ocfs2_extent_rec *split_rec)
{
T
Tao Ma 已提交
4325
	int status;
4326
	enum ocfs2_contig_type ret = CONTIG_NONE;
T
Tao Ma 已提交
4327 4328 4329 4330 4331 4332
	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;
4333
	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
T
Tao Ma 已提交
4334 4335 4336 4337

	if (index > 0) {
		rec = &el->l_recs[index - 1];
	} else if (path->p_tree_depth > 0) {
4338
		status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
T
Tao Ma 已提交
4339 4340 4341 4342
		if (status)
			goto out;

		if (left_cpos != 0) {
4343
			left_path = ocfs2_new_path_from_path(path);
T
Tao Ma 已提交
4344 4345 4346
			if (!left_path)
				goto out;

4347 4348
			status = ocfs2_find_path(et->et_ci, left_path,
						 left_cpos);
T
Tao Ma 已提交
4349 4350 4351 4352 4353 4354 4355 4356 4357
			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;
4358
				ocfs2_error(sb,
4359 4360 4361 4362 4363 4364 4365 4366
					    "Extent block #%llu has an "
					    "invalid l_next_free_rec of "
					    "%d.  It should have "
					    "matched the l_count of %d",
					    (unsigned long long)le64_to_cpu(eb->h_blkno),
					    le16_to_cpu(new_el->l_next_free_rec),
					    le16_to_cpu(new_el->l_count));
				status = -EINVAL;
T
Tao Ma 已提交
4367 4368 4369 4370 4371 4372
				goto out;
			}
			rec = &new_el->l_recs[
				le16_to_cpu(new_el->l_next_free_rec) - 1];
		}
	}
4373 4374 4375 4376 4377

	/*
	 * 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
4378
	if (rec) {
4379 4380 4381 4382
		if (index == 1 && ocfs2_is_empty_extent(rec)) {
			if (split_rec->e_cpos == el->l_recs[index].e_cpos)
				ret = CONTIG_RIGHT;
		} else {
4383
			ret = ocfs2_et_extent_contig(et, rec, split_rec);
4384 4385 4386
		}
	}

T
Tao Ma 已提交
4387 4388 4389 4390 4391
	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) {
4392
		status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
T
Tao Ma 已提交
4393 4394 4395 4396 4397 4398
		if (status)
			goto out;

		if (right_cpos == 0)
			goto out;

4399
		right_path = ocfs2_new_path_from_path(path);
T
Tao Ma 已提交
4400 4401 4402
		if (!right_path)
			goto out;

4403
		status = ocfs2_find_path(et->et_ci, right_path, right_cpos);
T
Tao Ma 已提交
4404 4405 4406 4407 4408 4409 4410 4411 4412
		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;
4413
				ocfs2_error(sb,
4414 4415 4416 4417 4418
					    "Extent block #%llu has an "
					    "invalid l_next_free_rec of %d",
					    (unsigned long long)le64_to_cpu(eb->h_blkno),
					    le16_to_cpu(new_el->l_next_free_rec));
				status = -EINVAL;
T
Tao Ma 已提交
4419 4420 4421 4422 4423 4424 4425
				goto out;
			}
			rec = &new_el->l_recs[1];
		}
	}

	if (rec) {
4426 4427
		enum ocfs2_contig_type contig_type;

4428
		contig_type = ocfs2_et_extent_contig(et, rec, split_rec);
4429 4430 4431 4432 4433 4434 4435

		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
4436 4437 4438 4439 4440 4441
out:
	if (left_path)
		ocfs2_free_path(left_path);
	if (right_path)
		ocfs2_free_path(right_path);

4442 4443 4444
	return ret;
}

4445
static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et,
M
Mark Fasheh 已提交
4446 4447
				     struct ocfs2_insert_type *insert,
				     struct ocfs2_extent_list *el,
4448
				     struct ocfs2_extent_rec *insert_rec)
M
Mark Fasheh 已提交
4449 4450 4451 4452
{
	int i;
	enum ocfs2_contig_type contig_type = CONTIG_NONE;

4453 4454
	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);

M
Mark Fasheh 已提交
4455
	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
4456 4457
		contig_type = ocfs2_et_extent_contig(et, &el->l_recs[i],
						     insert_rec);
M
Mark Fasheh 已提交
4458 4459 4460 4461 4462 4463
		if (contig_type != CONTIG_NONE) {
			insert->ins_contig_index = i;
			break;
		}
	}
	insert->ins_contig = contig_type;
4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474

	if (insert->ins_contig != CONTIG_NONE) {
		struct ocfs2_extent_rec *rec =
				&el->l_recs[insert->ins_contig_index];
		unsigned int len = le16_to_cpu(rec->e_leaf_clusters) +
				   le16_to_cpu(insert_rec->e_leaf_clusters);

		/*
		 * Caller might want us to limit the size of extents, don't
		 * calculate contiguousness if we might exceed that limit.
		 */
4475 4476
		if (et->et_max_leaf_clusters &&
		    (len > et->et_max_leaf_clusters))
4477 4478
			insert->ins_contig = CONTIG_NONE;
	}
M
Mark Fasheh 已提交
4479 4480 4481 4482 4483 4484 4485 4486
}

/*
 * 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.
 *
4487 4488
 * 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
4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500
 * 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;

4501
	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
M
Mark Fasheh 已提交
4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514

	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];

4515 4516
	if (cpos >=
	    (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
M
Mark Fasheh 已提交
4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537
		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.
 */
4538
static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et,
M
Mark Fasheh 已提交
4539 4540
				    struct buffer_head **last_eb_bh,
				    struct ocfs2_extent_rec *insert_rec,
4541
				    int *free_records,
M
Mark Fasheh 已提交
4542 4543 4544 4545 4546 4547 4548 4549
				    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;

4550 4551
	insert->ins_split = SPLIT_NONE;

4552
	el = et->et_root_el;
M
Mark Fasheh 已提交
4553 4554 4555 4556 4557 4558 4559 4560 4561
	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.
		 */
4562
		ret = ocfs2_read_extent_block(et->et_ci,
4563 4564
					      ocfs2_et_get_last_eb_blk(et),
					      &bh);
M
Mark Fasheh 已提交
4565 4566 4567 4568
		if (ret) {
			mlog_exit(ret);
			goto out;
		}
4569 4570
		eb = (struct ocfs2_extent_block *) bh->b_data;
		el = &eb->h_list;
M
Mark Fasheh 已提交
4571
	}
4572

M
Mark Fasheh 已提交
4573 4574 4575 4576 4577 4578 4579 4580
	/*
	 * 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.
	 */
4581
	*free_records = le16_to_cpu(el->l_count) -
M
Mark Fasheh 已提交
4582 4583 4584
		le16_to_cpu(el->l_next_free_rec);

	if (!insert->ins_tree_depth) {
4585
		ocfs2_figure_contig_type(et, insert, el, insert_rec);
M
Mark Fasheh 已提交
4586 4587
		ocfs2_figure_appending_type(insert, el, insert_rec);
		return 0;
4588 4589
	}

4590
	path = ocfs2_new_path_from_et(et);
M
Mark Fasheh 已提交
4591 4592 4593 4594 4595
	if (!path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}
4596

M
Mark Fasheh 已提交
4597 4598 4599 4600 4601 4602
	/*
	 * 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.
	 */
4603
	ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos));
M
Mark Fasheh 已提交
4604 4605 4606 4607
	if (ret) {
		mlog_errno(ret);
		goto out;
	}
4608

M
Mark Fasheh 已提交
4609 4610 4611 4612 4613 4614 4615 4616 4617 4618
	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.
	 */
4619
	ocfs2_figure_contig_type(et, insert, el, insert_rec);
M
Mark Fasheh 已提交
4620 4621 4622 4623 4624

	/*
	 * 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
4625
	 * cluster count on the last record of the path directly to it's
M
Mark Fasheh 已提交
4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639
	 * 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.
	 */
4640
	if (ocfs2_et_get_last_eb_blk(et) ==
4641
	    path_leaf_bh(path)->b_blocknr) {
M
Mark Fasheh 已提交
4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660
		/*
		 * 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;
4661 4662
}

M
Mark Fasheh 已提交
4663
/*
4664
 * Insert an extent into a btree.
M
Mark Fasheh 已提交
4665
 *
4666
 * The caller needs to update the owning btree's cluster count.
M
Mark Fasheh 已提交
4667
 */
4668
int ocfs2_insert_extent(handle_t *handle,
4669 4670 4671 4672 4673 4674
			struct ocfs2_extent_tree *et,
			u32 cpos,
			u64 start_blk,
			u32 new_clusters,
			u8 flags,
			struct ocfs2_alloc_context *meta_ac)
4675
{
4676
	int status;
4677
	int uninitialized_var(free_records);
4678
	struct buffer_head *last_eb_bh = NULL;
M
Mark Fasheh 已提交
4679 4680 4681
	struct ocfs2_insert_type insert = {0, };
	struct ocfs2_extent_rec rec;

4682 4683 4684
	mlog(0, "add %u clusters at position %u to owner %llu\n",
	     new_clusters, cpos,
	     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
M
Mark Fasheh 已提交
4685

4686
	memset(&rec, 0, sizeof(rec));
M
Mark Fasheh 已提交
4687 4688
	rec.e_cpos = cpu_to_le32(cpos);
	rec.e_blkno = cpu_to_le64(start_blk);
4689
	rec.e_leaf_clusters = cpu_to_le16(new_clusters);
4690
	rec.e_flags = flags;
4691
	status = ocfs2_et_insert_check(et, &rec);
4692 4693 4694 4695
	if (status) {
		mlog_errno(status);
		goto bail;
	}
M
Mark Fasheh 已提交
4696

4697
	status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec,
4698
					  &free_records, &insert);
M
Mark Fasheh 已提交
4699 4700 4701
	if (status < 0) {
		mlog_errno(status);
		goto bail;
4702 4703
	}

M
Mark Fasheh 已提交
4704 4705 4706 4707
	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,
4708
	     free_records, insert.ins_tree_depth);
4709

4710
	if (insert.ins_contig == CONTIG_NONE && free_records == 0) {
4711
		status = ocfs2_grow_tree(handle, et,
4712
					 &insert.ins_tree_depth, &last_eb_bh,
4713 4714
					 meta_ac);
		if (status) {
4715 4716 4717 4718 4719
			mlog_errno(status);
			goto bail;
		}
	}

M
Mark Fasheh 已提交
4720
	/* Finally, we can add clusters. This might rotate the tree for us. */
4721
	status = ocfs2_do_insert_extent(handle, et, &rec, &insert);
4722 4723
	if (status < 0)
		mlog_errno(status);
4724 4725
	else
		ocfs2_et_extent_map_insert(et, &rec);
4726 4727

bail:
4728
	brelse(last_eb_bh);
4729

4730 4731 4732 4733
	mlog_exit(status);
	return status;
}

4734 4735 4736
/*
 * Allcate and add clusters into the extent b-tree.
 * The new clusters(clusters_to_add) will be inserted at logical_offset.
4737
 * The extent b-tree's root is specified by et, and
4738 4739 4740
 * it is not limited to the file storage. Any extent tree can use this
 * function if it implements the proper ocfs2_extent_tree.
 */
4741 4742
int ocfs2_add_clusters_in_btree(handle_t *handle,
				struct ocfs2_extent_tree *et,
4743 4744 4745 4746 4747
				u32 *logical_offset,
				u32 clusters_to_add,
				int mark_unwritten,
				struct ocfs2_alloc_context *data_ac,
				struct ocfs2_alloc_context *meta_ac,
4748
				enum ocfs2_alloc_restarted *reason_ret)
4749 4750 4751 4752 4753 4754 4755
{
	int status = 0;
	int free_extents;
	enum ocfs2_alloc_restarted reason = RESTART_NONE;
	u32 bit_off, num_bits;
	u64 block;
	u8 flags = 0;
4756 4757
	struct ocfs2_super *osb =
		OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
4758 4759 4760 4761 4762 4763

	BUG_ON(!clusters_to_add);

	if (mark_unwritten)
		flags = OCFS2_EXT_UNWRITTEN;

4764
	free_extents = ocfs2_num_free_extents(osb, et);
4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782
	if (free_extents < 0) {
		status = free_extents;
		mlog_errno(status);
		goto leave;
	}

	/* there are two cases which could cause us to EAGAIN in the
	 * we-need-more-metadata case:
	 * 1) we haven't reserved *any*
	 * 2) we are so fragmented, we've needed to add metadata too
	 *    many times. */
	if (!free_extents && !meta_ac) {
		mlog(0, "we haven't reserved any metadata!\n");
		status = -EAGAIN;
		reason = RESTART_META;
		goto leave;
	} else if ((!free_extents)
		   && (ocfs2_alloc_context_bits_left(meta_ac)
4783
		       < ocfs2_extend_meta_needed(et->et_root_el))) {
4784 4785 4786 4787 4788 4789
		mlog(0, "filesystem is really fragmented...\n");
		status = -EAGAIN;
		reason = RESTART_META;
		goto leave;
	}

4790
	status = __ocfs2_claim_clusters(handle, data_ac, 1,
4791 4792 4793 4794 4795 4796 4797 4798 4799
					clusters_to_add, &bit_off, &num_bits);
	if (status < 0) {
		if (status != -ENOSPC)
			mlog_errno(status);
		goto leave;
	}

	BUG_ON(num_bits > clusters_to_add);

4800
	/* reserve our write early -- insert_extent may update the tree root */
4801
	status = ocfs2_et_root_journal_access(handle, et,
4802
					      OCFS2_JOURNAL_ACCESS_WRITE);
4803 4804 4805 4806 4807 4808
	if (status < 0) {
		mlog_errno(status);
		goto leave;
	}

	block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4809 4810 4811
	mlog(0, "Allocating %u clusters at block %u for owner %llu\n",
	     num_bits, bit_off,
	     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
4812
	status = ocfs2_insert_extent(handle, et, *logical_offset, block,
4813
				     num_bits, flags, meta_ac);
4814 4815 4816 4817 4818
	if (status < 0) {
		mlog_errno(status);
		goto leave;
	}

4819
	ocfs2_journal_dirty(handle, et->et_root_bh);
4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837

	clusters_to_add -= num_bits;
	*logical_offset += num_bits;

	if (clusters_to_add) {
		mlog(0, "need to alloc once more, wanted = %u\n",
		     clusters_to_add);
		status = -EAGAIN;
		reason = RESTART_TRANS;
	}

leave:
	mlog_exit(status);
	if (reason_ret)
		*reason_ret = reason;
	return status;
}

4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857
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;
}

4858
static int ocfs2_split_and_insert(handle_t *handle,
4859
				  struct ocfs2_extent_tree *et,
4860
				  struct ocfs2_path *path,
4861 4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881
				  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];

4882
	rightmost_el = et->et_root_el;
4883 4884 4885 4886 4887 4888 4889 4890 4891 4892

	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)) {
4893
		ret = ocfs2_grow_tree(handle, et,
4894
				      &depth, last_eb_bh, meta_ac);
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
		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;

4922 4923
		ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
					   &tmprec, insert_range, &rec);
4924 4925 4926 4927 4928 4929 4930

		split_rec = tmprec;

		BUG_ON(do_leftright);
		do_leftright = 1;
	}

4931
	ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946
	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);
4947
		ret = ocfs2_find_path(et->et_ci, path, cpos);
4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961
		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;
}

4962 4963
static int ocfs2_replace_extent_rec(handle_t *handle,
				    struct ocfs2_extent_tree *et,
4964 4965 4966 4967 4968 4969 4970
				    struct ocfs2_path *path,
				    struct ocfs2_extent_list *el,
				    int split_index,
				    struct ocfs2_extent_rec *split_rec)
{
	int ret;

4971
	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984
					   path_num_items(path) - 1);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	el->l_recs[split_index] = *split_rec;

	ocfs2_journal_dirty(handle, path_leaf_bh(path));
out:
	return ret;
}

4985
/*
T
Tao Ma 已提交
4986 4987
 * Split part or all of the extent record at split_index in the leaf
 * pointed to by path. Merge with the contiguous extent record if needed.
4988 4989 4990 4991 4992 4993 4994
 *
 * 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.
 *
4995 4996 4997
 * 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.
4998 4999 5000 5001 5002 5003
 *
 * 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.
 */
5004 5005 5006 5007 5008 5009 5010
int ocfs2_split_extent(handle_t *handle,
		       struct ocfs2_extent_tree *et,
		       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)
5011 5012 5013
{
	int ret = 0;
	struct ocfs2_extent_list *el = path_leaf_el(path);
5014
	struct buffer_head *last_eb_bh = NULL;
5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026
	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
	struct ocfs2_merge_ctxt ctxt;
	struct ocfs2_extent_list *rightmost_el;

	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;
	}

5027
	ctxt.c_contig_type = ocfs2_figure_merge_contig_type(et, path, el,
5028 5029 5030 5031 5032
							    split_index,
							    split_rec);

	/*
	 * The core merge / split code wants to know how much room is
5033
	 * left in this allocation tree, so we pass the
5034 5035 5036 5037 5038
	 * rightmost extent list.
	 */
	if (path->p_tree_depth) {
		struct ocfs2_extent_block *eb;

5039
		ret = ocfs2_read_extent_block(et->et_ci,
5040 5041
					      ocfs2_et_get_last_eb_blk(et),
					      &last_eb_bh);
5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059
		if (ret) {
			mlog_exit(ret);
			goto out;
		}

		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
		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
5060 5061 5062
	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);
5063 5064 5065

	if (ctxt.c_contig_type == CONTIG_NONE) {
		if (ctxt.c_split_covers_rec)
5066
			ret = ocfs2_replace_extent_rec(handle, et, path, el,
5067
						       split_index, split_rec);
5068
		else
5069
			ret = ocfs2_split_and_insert(handle, et, path,
5070 5071 5072 5073 5074
						     &last_eb_bh, split_index,
						     split_rec, meta_ac);
		if (ret)
			mlog_errno(ret);
	} else {
5075
		ret = ocfs2_try_to_merge_extent(handle, et, path,
5076
						split_index, split_rec,
5077
						dealloc, &ctxt);
5078 5079 5080 5081 5082 5083 5084 5085 5086 5087
		if (ret)
			mlog_errno(ret);
	}

out:
	brelse(last_eb_bh);
	return ret;
}

/*
T
Tao Ma 已提交
5088 5089 5090 5091 5092
 * Change the flags of the already-existing extent at cpos for len clusters.
 *
 * new_flags: the flags we want to set.
 * clear_flags: the flags we want to clear.
 * phys: the new physical offset we want this new extent starts from.
5093 5094 5095 5096 5097 5098
 *
 * 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.
 */
5099 5100 5101 5102 5103 5104
int ocfs2_change_extent_flag(handle_t *handle,
			     struct ocfs2_extent_tree *et,
			     u32 cpos, u32 len, u32 phys,
			     struct ocfs2_alloc_context *meta_ac,
			     struct ocfs2_cached_dealloc_ctxt *dealloc,
			     int new_flags, int clear_flags)
5105 5106
{
	int ret, index;
T
Tao Ma 已提交
5107 5108
	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
	u64 start_blkno = ocfs2_clusters_to_blocks(sb, phys);
5109 5110 5111
	struct ocfs2_extent_rec split_rec;
	struct ocfs2_path *left_path = NULL;
	struct ocfs2_extent_list *el;
T
Tao Ma 已提交
5112
	struct ocfs2_extent_rec *rec;
5113

5114
	left_path = ocfs2_new_path_from_et(et);
5115 5116 5117 5118 5119 5120
	if (!left_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

5121
	ret = ocfs2_find_path(et->et_ci, left_path, cpos);
5122 5123 5124 5125 5126 5127 5128 5129
	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)) {
T
Tao Ma 已提交
5130 5131
		ocfs2_error(sb,
			    "Owner %llu has an extent at cpos %u which can no "
5132
			    "longer be found.\n",
T
Tao Ma 已提交
5133 5134
			     (unsigned long long)
			     ocfs2_metadata_cache_owner(et->et_ci), cpos);
5135 5136 5137 5138
		ret = -EROFS;
		goto out;
	}

T
Tao Ma 已提交
5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156
	ret = -EIO;
	rec = &el->l_recs[index];
	if (new_flags && (rec->e_flags & new_flags)) {
		mlog(ML_ERROR, "Owner %llu tried to set %d flags on an "
		     "extent that already had them",
		     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
		     new_flags);
		goto out;
	}

	if (clear_flags && !(rec->e_flags & clear_flags)) {
		mlog(ML_ERROR, "Owner %llu tried to clear %d flags on an "
		     "extent that didn't have them",
		     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
		     clear_flags);
		goto out;
	}

5157 5158 5159 5160
	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);
T
Tao Ma 已提交
5161 5162 5163 5164 5165 5166
	split_rec.e_flags = rec->e_flags;
	if (new_flags)
		split_rec.e_flags |= new_flags;
	if (clear_flags)
		split_rec.e_flags &= ~clear_flags;

5167 5168 5169
	ret = ocfs2_split_extent(handle, et, left_path,
				 index, &split_rec, meta_ac,
				 dealloc);
5170 5171 5172 5173 5174 5175
	if (ret)
		mlog_errno(ret);

out:
	ocfs2_free_path(left_path);
	return ret;
T
Tao Ma 已提交
5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221

}

/*
 * Mark the already-existing extent at cpos as written for len clusters.
 * This removes the unwritten extent flag.
 *
 * 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.
 */
int ocfs2_mark_extent_written(struct inode *inode,
			      struct ocfs2_extent_tree *et,
			      handle_t *handle, u32 cpos, u32 len, u32 phys,
			      struct ocfs2_alloc_context *meta_ac,
			      struct ocfs2_cached_dealloc_ctxt *dealloc)
{
	int ret;

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

	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;
	}

	/*
	 * XXX: This should be fixed up so that we just re-insert the
	 * next extent records.
	 */
	ocfs2_et_extent_map_truncate(et, 0);

	ret = ocfs2_change_extent_flag(handle, et, cpos,
				       len, phys, meta_ac, dealloc,
				       0, OCFS2_EXT_UNWRITTEN);
	if (ret)
		mlog_errno(ret);

out:
	return ret;
5222 5223
}

5224 5225
static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et,
			    struct ocfs2_path *path,
5226 5227 5228
			    int index, u32 new_range,
			    struct ocfs2_alloc_context *meta_ac)
{
5229
	int ret, depth, credits;
5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241
	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];
5242 5243
	ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
				   &split_rec, new_range, rec);
5244 5245 5246

	depth = path->p_tree_depth;
	if (depth > 0) {
5247
		ret = ocfs2_read_extent_block(et->et_ci,
5248 5249
					      ocfs2_et_get_last_eb_blk(et),
					      &last_eb_bh);
5250 5251 5252 5253 5254 5255 5256 5257 5258 5259
		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);

5260 5261
	credits = path->p_tree_depth +
		  ocfs2_extend_meta_needed(et->et_root_el);
5262 5263 5264 5265 5266 5267 5268 5269
	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)) {
5270
		ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh,
5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283
				      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;

5284
	ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
5285 5286 5287 5288 5289 5290 5291 5292
	if (ret)
		mlog_errno(ret);

out:
	brelse(last_eb_bh);
	return ret;
}

5293 5294
static int ocfs2_truncate_rec(handle_t *handle,
			      struct ocfs2_extent_tree *et,
5295 5296
			      struct ocfs2_path *path, int index,
			      struct ocfs2_cached_dealloc_ctxt *dealloc,
5297
			      u32 cpos, u32 len)
5298 5299 5300 5301
{
	int ret;
	u32 left_cpos, rec_range, trunc_range;
	int wants_rotate = 0, is_rightmost_tree_rec = 0;
5302
	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
5303 5304 5305 5306 5307 5308
	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) {
5309
		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
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
		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:
5341
		 *   1) Path is the leftmost one in our btree.
5342 5343 5344 5345 5346
		 *   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.
		 */

5347
		ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
5348 5349 5350 5351 5352 5353
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
5354
			left_path = ocfs2_new_path_from_path(path);
5355 5356 5357 5358 5359 5360
			if (!left_path) {
				ret = -ENOMEM;
				mlog_errno(ret);
				goto out;
			}

5361 5362
			ret = ocfs2_find_path(et->et_ci, left_path,
					      left_cpos);
5363 5364 5365 5366 5367 5368 5369 5370 5371 5372 5373 5374 5375 5376 5377
			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;
	}

5378
	ret = ocfs2_journal_access_path(et->et_ci, handle, path);
5379 5380 5381 5382 5383
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

5384
	ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
5385 5386 5387 5388 5389 5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406
	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];
5407
			ocfs2_adjust_rightmost_records(handle, et, path,
5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418
						       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)
5419
			ocfs2_adjust_rightmost_records(handle, et, path, rec);
5420 5421
	} else {
		/* Caller should have trapped this. */
5422 5423 5424
		mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) "
		     "(%u, %u)\n",
		     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5425 5426 5427 5428 5429 5430 5431 5432
		     le32_to_cpu(rec->e_cpos),
		     le16_to_cpu(rec->e_leaf_clusters), cpos, len);
		BUG();
	}

	if (left_path) {
		int subtree_index;

5433
		subtree_index = ocfs2_find_subtree_root(et, left_path, path);
5434
		ocfs2_complete_edge_insert(handle, left_path, path,
5435 5436 5437 5438 5439
					   subtree_index);
	}

	ocfs2_journal_dirty(handle, path_leaf_bh(path));

5440
	ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
5441 5442 5443 5444 5445 5446 5447 5448 5449 5450
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

out:
	ocfs2_free_path(left_path);
	return ret;
}

5451
int ocfs2_remove_extent(handle_t *handle,
5452
			struct ocfs2_extent_tree *et,
5453
			u32 cpos, u32 len,
5454
			struct ocfs2_alloc_context *meta_ac,
5455
			struct ocfs2_cached_dealloc_ctxt *dealloc)
5456 5457 5458 5459 5460
{
	int ret, index;
	u32 rec_range, trunc_range;
	struct ocfs2_extent_rec *rec;
	struct ocfs2_extent_list *el;
5461
	struct ocfs2_path *path = NULL;
5462

5463 5464 5465 5466 5467
	/*
	 * XXX: Why are we truncating to 0 instead of wherever this
	 * affects us?
	 */
	ocfs2_et_extent_map_truncate(et, 0);
5468

5469
	path = ocfs2_new_path_from_et(et);
5470 5471 5472 5473 5474 5475
	if (!path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

5476
	ret = ocfs2_find_path(et->et_ci, path, cpos);
5477 5478 5479 5480 5481 5482 5483 5484
	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)) {
5485 5486
		ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
			    "Owner %llu has an extent at cpos %u which can no "
5487
			    "longer be found.\n",
5488 5489
			    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
			    cpos);
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
		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);

5516
	mlog(0, "Owner %llu, remove (cpos %u, len %u). Existing index %d "
5517
	     "(cpos %u, len %u)\n",
5518 5519
	     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
	     cpos, len, index,
5520 5521 5522
	     le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec));

	if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
5523 5524
		ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
					 cpos, len);
5525 5526 5527 5528 5529
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	} else {
5530
		ret = ocfs2_split_tree(handle, et, path, index,
5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542
				       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);

5543
		ret = ocfs2_find_path(et->et_ci, path, cpos);
5544 5545 5546 5547 5548 5549 5550 5551
		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)) {
5552 5553 5554
			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
				    "Owner %llu: split at cpos %u lost record.",
				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567
				    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) {
5568 5569
			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
				    "Owner %llu: error after split at cpos %u"
5570
				    "trunc len %u, existing record is (%u,%u)",
5571
				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5572 5573 5574 5575 5576 5577
				    cpos, len, le32_to_cpu(rec->e_cpos),
				    ocfs2_rec_clusters(el, rec));
			ret = -EROFS;
			goto out;
		}

5578 5579
		ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
					 cpos, len);
5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

out:
	ocfs2_free_path(path);
	return ret;
}

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
int ocfs2_remove_btree_range(struct inode *inode,
			     struct ocfs2_extent_tree *et,
			     u32 cpos, u32 phys_cpos, u32 len,
			     struct ocfs2_cached_dealloc_ctxt *dealloc)
{
	int ret;
	u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
	struct inode *tl_inode = osb->osb_tl_inode;
	handle_t *handle;
	struct ocfs2_alloc_context *meta_ac = NULL;

	ret = ocfs2_lock_allocators(inode, et, 0, 1, NULL, &meta_ac);
	if (ret) {
		mlog_errno(ret);
		return ret;
	}

	mutex_lock(&tl_inode->i_mutex);

	if (ocfs2_truncate_log_needs_flush(osb)) {
		ret = __ocfs2_flush_truncate_log(osb);
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}
	}

5619
	handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5620 5621 5622 5623 5624 5625
	if (IS_ERR(handle)) {
		ret = PTR_ERR(handle);
		mlog_errno(ret);
		goto out;
	}

5626
	ret = ocfs2_et_root_journal_access(handle, et,
5627
					   OCFS2_JOURNAL_ACCESS_WRITE);
5628 5629 5630 5631 5632
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

5633
	dquot_free_space_nodirty(inode,
5634 5635
				  ocfs2_clusters_to_bytes(inode->i_sb, len));

5636
	ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc);
5637 5638 5639 5640 5641
	if (ret) {
		mlog_errno(ret);
		goto out_commit;
	}

5642
	ocfs2_et_update_clusters(et, -len);
5643

5644
	ocfs2_journal_dirty(handle, et->et_root_bh);
5645 5646 5647 5648 5649 5650 5651 5652 5653 5654 5655 5656 5657 5658 5659 5660

	ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
	if (ret)
		mlog_errno(ret);

out_commit:
	ocfs2_commit_trans(osb, handle);
out:
	mutex_unlock(&tl_inode->i_mutex);

	if (meta_ac)
		ocfs2_free_alloc_context(meta_ac);

	return ret;
}

5661
int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
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
{
	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;
}

5694 5695 5696 5697
int ocfs2_truncate_log_append(struct ocfs2_super *osb,
			      handle_t *handle,
			      u64 start_blk,
			      unsigned int num_clusters)
5698 5699 5700 5701 5702 5703 5704 5705
{
	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;

5706 5707
	mlog_entry("start_blk = %llu, num_clusters = %u\n",
		   (unsigned long long)start_blk, num_clusters);
5708

5709
	BUG_ON(mutex_trylock(&tl_inode->i_mutex));
5710 5711 5712 5713 5714

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

	di = (struct ocfs2_dinode *) tl_bh->b_data;

5715 5716 5717 5718 5719 5720
	/* tl_bh is loaded from ocfs2_truncate_log_init().  It's validated
	 * by the underlying call to ocfs2_read_inode_block(), so any
	 * corruption is a code bug */
	BUG_ON(!OCFS2_IS_VALID_DINODE(di));

	tl = &di->id2.i_dealloc;
5721 5722 5723
	tl_count = le16_to_cpu(tl->tl_count);
	mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
			tl_count == 0,
5724 5725 5726
			"Truncate record count on #%llu invalid "
			"wanted %u, actual %u\n",
			(unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5727 5728 5729 5730 5731 5732 5733 5734 5735 5736 5737
			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;
	}

5738
	status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
5739
					 OCFS2_JOURNAL_ACCESS_WRITE);
5740 5741 5742 5743 5744 5745
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	mlog(0, "Log truncate of %u clusters starting at cluster %u to "
5746 5747
	     "%llu (index = %d)\n", num_clusters, start_cluster,
	     (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index);
5748 5749 5750 5751 5752 5753 5754 5755 5756 5757 5758 5759 5760 5761 5762 5763 5764 5765

	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);

5766
	ocfs2_journal_dirty(handle, tl_bh);
5767 5768 5769 5770 5771 5772 5773

bail:
	mlog_exit(status);
	return status;
}

static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
5774
					 handle_t *handle,
5775 5776 5777 5778 5779 5780 5781 5782 5783 5784 5785 5786 5787 5788 5789 5790 5791 5792 5793 5794 5795
					 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 */
5796
		status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
5797
						 OCFS2_JOURNAL_ACCESS_WRITE);
5798 5799 5800 5801 5802 5803 5804
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		tl->tl_used = cpu_to_le16(i);

5805
		ocfs2_journal_dirty(handle, tl_bh);
5806 5807 5808 5809 5810 5811 5812 5813 5814 5815 5816 5817 5818 5819 5820 5821 5822 5823 5824 5825 5826 5827 5828 5829 5830 5831 5832 5833 5834 5835 5836 5837 5838 5839 5840 5841 5842 5843

		/* 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;
}

5844
/* Expects you to already be holding tl_inode->i_mutex */
5845
int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
5846 5847 5848
{
	int status;
	unsigned int num_to_flush;
5849
	handle_t *handle;
5850 5851 5852 5853 5854 5855 5856 5857 5858
	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();

5859
	BUG_ON(mutex_trylock(&tl_inode->i_mutex));
5860 5861 5862

	di = (struct ocfs2_dinode *) tl_bh->b_data;

5863 5864 5865 5866 5867 5868
	/* tl_bh is loaded from ocfs2_truncate_log_init().  It's validated
	 * by the underlying call to ocfs2_read_inode_block(), so any
	 * corruption is a code bug */
	BUG_ON(!OCFS2_IS_VALID_DINODE(di));

	tl = &di->id2.i_dealloc;
5869
	num_to_flush = le16_to_cpu(tl->tl_used);
5870 5871
	mlog(0, "Flush %u records from truncate log #%llu\n",
	     num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno);
5872 5873
	if (!num_to_flush) {
		status = 0;
5874
		goto out;
5875 5876 5877 5878 5879 5880 5881 5882
	}

	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");
5883
		goto out;
5884 5885
	}

5886 5887
	mutex_lock(&data_alloc_inode->i_mutex);

M
Mark Fasheh 已提交
5888
	status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1);
5889 5890
	if (status < 0) {
		mlog_errno(status);
5891
		goto out_mutex;
5892 5893
	}

5894
	handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
5895 5896 5897
	if (IS_ERR(handle)) {
		status = PTR_ERR(handle);
		mlog_errno(status);
5898
		goto out_unlock;
5899 5900 5901 5902
	}

	status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
					       data_alloc_bh);
5903
	if (status < 0)
5904 5905
		mlog_errno(status);

5906
	ocfs2_commit_trans(osb, handle);
5907

5908 5909
out_unlock:
	brelse(data_alloc_bh);
M
Mark Fasheh 已提交
5910
	ocfs2_inode_unlock(data_alloc_inode, 1);
5911

5912 5913 5914
out_mutex:
	mutex_unlock(&data_alloc_inode->i_mutex);
	iput(data_alloc_inode);
5915

5916
out:
5917 5918 5919 5920 5921 5922 5923 5924 5925
	mlog_exit(status);
	return status;
}

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

5926
	mutex_lock(&tl_inode->i_mutex);
5927
	status = __ocfs2_flush_truncate_log(osb);
5928
	mutex_unlock(&tl_inode->i_mutex);
5929 5930 5931 5932

	return status;
}

D
David Howells 已提交
5933
static void ocfs2_truncate_log_worker(struct work_struct *work)
5934 5935
{
	int status;
D
David Howells 已提交
5936 5937 5938
	struct ocfs2_super *osb =
		container_of(work, struct ocfs2_super,
			     osb_truncate_log_wq.work);
5939 5940 5941 5942 5943 5944

	mlog_entry_void();

	status = ocfs2_flush_truncate_log(osb);
	if (status < 0)
		mlog_errno(status);
5945
	else
5946
		ocfs2_init_steal_slots(osb);
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 5972 5973 5974 5975 5976 5977 5978 5979 5980 5981 5982 5983

	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;
	}

5984
	status = ocfs2_read_inode_block(inode, &bh);
5985 5986 5987 5988 5989 5990 5991 5992 5993 5994 5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021 6022 6023
	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;

6024 6025 6026 6027 6028 6029
	/* tl_bh is loaded from ocfs2_get_truncate_log_info().  It's
	 * validated by the underlying call to ocfs2_read_inode_block(),
	 * so any corruption is a code bug */
	BUG_ON(!OCFS2_IS_VALID_DINODE(di));

	tl = &di->id2.i_dealloc;
6030 6031 6032 6033 6034 6035 6036 6037 6038 6039 6040 6041 6042 6043 6044 6045 6046 6047 6048
	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;

6049
		ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check);
6050
		status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode));
6051 6052 6053 6054 6055 6056 6057 6058 6059
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

bail:
	if (tl_inode)
		iput(tl_inode);
6060
	brelse(tl_bh);
6061 6062 6063 6064 6065 6066 6067 6068 6069 6070 6071 6072 6073 6074 6075 6076 6077

	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;
6078
	handle_t *handle;
6079 6080 6081 6082 6083 6084 6085 6086 6087 6088 6089 6090
	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);
6091
	mlog(0, "cleanup %u records from %llu\n", num_recs,
6092
	     (unsigned long long)le64_to_cpu(tl_copy->i_blkno));
6093

6094
	mutex_lock(&tl_inode->i_mutex);
6095 6096 6097 6098 6099 6100 6101 6102 6103
	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;
			}
		}

6104
		handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
6105 6106 6107 6108 6109 6110 6111 6112 6113 6114 6115 6116
		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);
6117
		ocfs2_commit_trans(osb, handle);
6118 6119 6120 6121 6122 6123 6124
		if (status < 0) {
			mlog_errno(status);
			goto bail_up;
		}
	}

bail_up:
6125
	mutex_unlock(&tl_inode->i_mutex);
6126 6127 6128 6129 6130 6131 6132 6133 6134 6135 6136 6137 6138 6139 6140 6141 6142 6143 6144 6145 6146 6147 6148 6149 6150 6151 6152 6153 6154 6155 6156 6157 6158 6159 6160 6161 6162 6163 6164 6165 6166 6167 6168 6169 6170

	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
6171 6172
	INIT_DELAYED_WORK(&osb->osb_truncate_log_wq,
			  ocfs2_truncate_log_worker);
6173 6174 6175 6176 6177 6178 6179
	osb->osb_tl_bh    = tl_bh;
	osb->osb_tl_inode = tl_inode;

	mlog_exit(status);
	return status;
}

6180 6181 6182 6183 6184 6185 6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198
/*
 * 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.
 */

/*
6199 6200 6201 6202
 * Describe a single bit freed from a suballocator.  For the block
 * suballocators, it represents one block.  For the global cluster
 * allocator, it represents some clusters and free_bit indicates
 * clusters number.
6203 6204 6205 6206 6207 6208 6209 6210 6211 6212 6213 6214 6215 6216
 */
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;
};

6217 6218 6219 6220
static int ocfs2_free_cached_blocks(struct ocfs2_super *osb,
				    int sysfile_type,
				    int slot,
				    struct ocfs2_cached_block_free *head)
6221 6222 6223 6224 6225 6226 6227 6228 6229 6230 6231 6232 6233 6234 6235 6236 6237
{
	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
6238
	ret = ocfs2_inode_lock(inode, &di_bh, 1);
6239 6240 6241 6242 6243 6244 6245 6246 6247 6248 6249 6250 6251 6252 6253 6254 6255 6256 6257 6258 6259 6260 6261 6262 6263 6264 6265 6266 6267 6268 6269 6270 6271 6272 6273 6274 6275 6276 6277 6278
	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
6279
	ocfs2_inode_unlock(inode, 1);
6280 6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292 6293 6294
	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;
}

6295 6296 6297 6298 6299 6300 6301 6302 6303 6304 6305 6306 6307 6308 6309 6310 6311 6312 6313 6314 6315 6316 6317 6318 6319 6320 6321 6322 6323 6324 6325 6326 6327 6328 6329 6330 6331 6332 6333 6334 6335 6336 6337 6338 6339 6340 6341 6342 6343 6344 6345 6346 6347 6348 6349 6350 6351 6352 6353 6354 6355 6356 6357 6358 6359 6360 6361 6362 6363 6364 6365 6366 6367 6368 6369 6370
int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
				u64 blkno, unsigned int bit)
{
	int ret = 0;
	struct ocfs2_cached_block_free *item;

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

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

	item->free_blk = blkno;
	item->free_bit = bit;
	item->free_next = ctxt->c_global_allocator;

	ctxt->c_global_allocator = item;
	return ret;
}

static int ocfs2_free_cached_clusters(struct ocfs2_super *osb,
				      struct ocfs2_cached_block_free *head)
{
	struct ocfs2_cached_block_free *tmp;
	struct inode *tl_inode = osb->osb_tl_inode;
	handle_t *handle;
	int ret = 0;

	mutex_lock(&tl_inode->i_mutex);

	while (head) {
		if (ocfs2_truncate_log_needs_flush(osb)) {
			ret = __ocfs2_flush_truncate_log(osb);
			if (ret < 0) {
				mlog_errno(ret);
				break;
			}
		}

		handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
		if (IS_ERR(handle)) {
			ret = PTR_ERR(handle);
			mlog_errno(ret);
			break;
		}

		ret = ocfs2_truncate_log_append(osb, handle, head->free_blk,
						head->free_bit);

		ocfs2_commit_trans(osb, handle);
		tmp = head;
		head = head->free_next;
		kfree(tmp);

		if (ret < 0) {
			mlog_errno(ret);
			break;
		}
	}

	mutex_unlock(&tl_inode->i_mutex);

	while (head) {
		/* Premature exit may have left some dangling items. */
		tmp = head;
		head = head->free_next;
		kfree(tmp);
	}

	return ret;
}

6371 6372 6373 6374 6375 6376 6377 6378 6379 6380 6381 6382 6383 6384 6385
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);
6386 6387 6388 6389
			ret2 = ocfs2_free_cached_blocks(osb,
							fl->f_inode_type,
							fl->f_slot,
							fl->f_first);
6390 6391 6392 6393 6394 6395 6396 6397 6398 6399
			if (ret2)
				mlog_errno(ret2);
			if (!ret)
				ret = ret2;
		}

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

6400 6401 6402 6403 6404 6405 6406 6407 6408 6409 6410
	if (ctxt->c_global_allocator) {
		ret2 = ocfs2_free_cached_clusters(osb,
						  ctxt->c_global_allocator);
		if (ret2)
			mlog_errno(ret2);
		if (!ret)
			ret = ret2;

		ctxt->c_global_allocator = NULL;
	}

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
	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;
}

6440 6441 6442
int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
			      int type, int slot, u64 blkno,
			      unsigned int bit)
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
{
	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;
}

6476 6477 6478 6479 6480 6481 6482 6483 6484
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));
}

6485 6486 6487 6488
/* 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
6489
static int ocfs2_find_new_last_ext_blk(struct inode *inode,
6490
				       unsigned int clusters_to_del,
M
Mark Fasheh 已提交
6491
				       struct ocfs2_path *path,
6492 6493
				       struct buffer_head **new_last_eb)
{
6494
	int next_free, ret = 0;
M
Mark Fasheh 已提交
6495
	u32 cpos;
6496
	struct ocfs2_extent_rec *rec;
6497 6498 6499 6500 6501 6502 6503
	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
6504 6505
	if (!path->p_tree_depth)
		goto out;
6506 6507 6508

	/* trunc to zero special case - this makes tree_depth = 0
	 * regardless of what it is.  */
6509
	if (OCFS2_I(inode)->ip_clusters == clusters_to_del)
M
Mark Fasheh 已提交
6510
		goto out;
6511

M
Mark Fasheh 已提交
6512
	el = path_leaf_el(path);
6513 6514
	BUG_ON(!el->l_next_free_rec);

6515 6516 6517 6518 6519 6520 6521 6522 6523
	/*
	 * 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
6524
	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
6525
		if (next_free > 2)
M
Mark Fasheh 已提交
6526
			goto out;
6527 6528 6529 6530 6531 6532 6533 6534 6535 6536 6537 6538 6539 6540 6541 6542 6543 6544 6545 6546 6547

		/* 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.
		 */
6548
		if (le16_to_cpu(rec->e_leaf_clusters) > clusters_to_del)
6549 6550
			goto out;
	}
6551

M
Mark Fasheh 已提交
6552 6553 6554 6555 6556
	ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}
6557

6558
	ret = ocfs2_find_leaf(INODE_CACHE(inode), path_root_el(path), cpos, &bh);
M
Mark Fasheh 已提交
6559 6560 6561 6562
	if (ret) {
		mlog_errno(ret);
		goto out;
	}
6563

M
Mark Fasheh 已提交
6564 6565
	eb = (struct ocfs2_extent_block *) bh->b_data;
	el = &eb->h_list;
6566 6567 6568 6569

	/* ocfs2_find_leaf() gets the eb from ocfs2_read_extent_block().
	 * Any corruption is a code bug. */
	BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
6570 6571 6572

	*new_last_eb = bh;
	get_bh(*new_last_eb);
M
Mark Fasheh 已提交
6573 6574 6575 6576
	mlog(0, "returning block %llu, (cpos: %u)\n",
	     (unsigned long long)le64_to_cpu(eb->h_blkno), cpos);
out:
	brelse(bh);
6577

M
Mark Fasheh 已提交
6578
	return ret;
6579 6580
}

6581 6582 6583 6584 6585 6586 6587 6588 6589 6590
/*
 * 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,
6591
			   u32 clusters_to_del, u64 *delete_start, u8 *flags)
6592 6593 6594 6595 6596 6597 6598 6599 6600
{
	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;
6601
	*flags = 0;
6602 6603 6604 6605 6606 6607 6608 6609 6610 6611 6612 6613 6614 6615 6616 6617 6618 6619 6620 6621 6622 6623 6624 6625 6626 6627

	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),
6628
		     ocfs2_rec_clusters(el, rec),
6629 6630 6631
		     (unsigned long long)le64_to_cpu(rec->e_blkno),
		     le16_to_cpu(el->l_next_free_rec));

6632
		BUG_ON(ocfs2_rec_clusters(el, rec) < clusters_to_del);
6633 6634 6635 6636 6637 6638 6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650 6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673

		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;
			}

6674
			le16_add_cpu(&rec->e_leaf_clusters, -clusters_to_del);
6675 6676 6677 6678 6679

			/*
			 * We'll use "new_edge" on our way back up the
			 * tree to know what our rightmost cpos is.
			 */
6680
			new_edge = le16_to_cpu(rec->e_leaf_clusters);
6681 6682 6683 6684 6685 6686 6687
			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,
6688
					le16_to_cpu(rec->e_leaf_clusters));
6689
			*flags = rec->e_flags;
6690 6691 6692 6693

			/*
			 * If it's now empty, remove this record.
			 */
6694
			if (le16_to_cpu(rec->e_leaf_clusters) == 0) {
6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717 6718 6719
				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;

6720 6721
			rec->e_int_clusters = cpu_to_le32(new_edge);
			le32_add_cpu(&rec->e_int_clusters,
6722 6723 6724 6725 6726 6727
				     -le32_to_cpu(rec->e_cpos));

			 /*
			  * A deleted child record should have been
			  * caught above.
			  */
6728
			 BUG_ON(le32_to_cpu(rec->e_int_clusters) == 0);
6729 6730 6731
		}

delete:
6732
		ocfs2_journal_dirty(handle, bh);
6733 6734 6735 6736

		mlog(0, "extent list container %llu, after: record %d: "
		     "(%u, %u, %llu), next = %u.\n",
		     (unsigned long long)bh->b_blocknr, i,
6737
		     le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec),
6738 6739 6740 6741 6742 6743 6744 6745 6746 6747 6748 6749 6750 6751 6752 6753 6754 6755 6756
		     (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");

6757
			ocfs2_remove_from_cache(INODE_CACHE(inode), bh);
6758

6759
			BUG_ON(ocfs2_rec_clusters(el, &el->l_recs[0]));
6760 6761 6762
			BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos));
			BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno));

6763 6764 6765 6766
			ret = ocfs2_cache_extent_block_free(&tc->tc_dealloc, eb);
			/* An error here is not fatal. */
			if (ret < 0)
				mlog_errno(ret);
6767 6768 6769 6770 6771 6772 6773 6774 6775 6776 6777 6778
		} else {
			deleted_eb = 0;
		}

		index--;
	}

	ret = 0;
out:
	return ret;
}

6779 6780 6781 6782
static int ocfs2_do_truncate(struct ocfs2_super *osb,
			     unsigned int clusters_to_del,
			     struct inode *inode,
			     struct buffer_head *fe_bh,
6783
			     handle_t *handle,
M
Mark Fasheh 已提交
6784
			     struct ocfs2_truncate_context *tc,
6785 6786
			     struct ocfs2_path *path,
			     struct ocfs2_alloc_context *meta_ac)
6787
{
6788
	int status;
6789 6790 6791 6792 6793
	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;
6794
	u8 rec_flags;
6795 6796 6797

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

6798
	status = ocfs2_find_new_last_ext_blk(inode, clusters_to_del,
M
Mark Fasheh 已提交
6799
					     path, &last_eb_bh);
6800 6801 6802 6803
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
M
Mark Fasheh 已提交
6804 6805 6806 6807 6808

	/*
	 * Each component will be touched, so we might as well journal
	 * here to avoid having to handle errors later.
	 */
6809
	status = ocfs2_journal_access_path(INODE_CACHE(inode), handle, path);
6810 6811 6812
	if (status < 0) {
		mlog_errno(status);
		goto bail;
M
Mark Fasheh 已提交
6813 6814 6815
	}

	if (last_eb_bh) {
6816
		status = ocfs2_journal_access_eb(handle, INODE_CACHE(inode), last_eb_bh,
6817
						 OCFS2_JOURNAL_ACCESS_WRITE);
M
Mark Fasheh 已提交
6818 6819 6820 6821 6822
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

6823
		last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
M
Mark Fasheh 已提交
6824
	}
6825

M
Mark Fasheh 已提交
6826 6827 6828 6829 6830 6831
	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.
	 */
6832
	if (el->l_tree_depth && el->l_recs[0].e_int_clusters == 0) {
M
Mark Fasheh 已提交
6833 6834 6835
		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));
6836
		status = -EROFS;
6837 6838 6839
		goto bail;
	}

6840
	dquot_free_space_nodirty(inode,
6841
			ocfs2_clusters_to_bytes(osb->sb, clusters_to_del));
6842 6843 6844 6845 6846
	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);
6847
	inode->i_blocks = ocfs2_inode_sector_count(inode);
6848

6849
	status = ocfs2_trim_tree(inode, path, handle, tc,
6850
				 clusters_to_del, &delete_blk, &rec_flags);
6851 6852 6853
	if (status) {
		mlog_errno(status);
		goto bail;
6854 6855
	}

M
Mark Fasheh 已提交
6856
	if (le32_to_cpu(fe->i_clusters) == 0) {
6857 6858 6859 6860 6861 6862
		/* 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;

6863
	ocfs2_journal_dirty(handle, fe_bh);
6864 6865 6866 6867 6868 6869

	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;
6870
		ocfs2_journal_dirty(handle, last_eb_bh);
6871 6872
	}

6873
	if (delete_blk) {
6874 6875 6876 6877 6878
		if (rec_flags & OCFS2_EXT_REFCOUNTED)
			status = ocfs2_decrease_refcount(inode, handle,
					ocfs2_blocks_to_clusters(osb->sb,
								 delete_blk),
					clusters_to_del, meta_ac,
T
Tao Ma 已提交
6879
					&tc->tc_dealloc, 1);
6880 6881 6882 6883
		else
			status = ocfs2_truncate_log_append(osb, handle,
							   delete_blk,
							   clusters_to_del);
6884 6885 6886 6887 6888 6889 6890
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}
	status = 0;
bail:
6891
	brelse(last_eb_bh);
6892 6893 6894 6895
	mlog_exit(status);
	return status;
}

J
Joel Becker 已提交
6896
static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh)
6897 6898 6899 6900 6901 6902
{
	set_buffer_uptodate(bh);
	mark_buffer_dirty(bh);
	return 0;
}

T
Tao Ma 已提交
6903 6904 6905
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)
6906 6907 6908 6909 6910 6911 6912 6913
{
	int ret, partial = 0;

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

	if (zero)
6914
		zero_user_segment(page, from, to);
6915 6916 6917 6918 6919 6920

	/*
	 * Need to set the buffers we zero'd into uptodate
	 * here if they aren't - ocfs2_map_page_blocks()
	 * might've skipped some
	 */
J
Joel Becker 已提交
6921 6922 6923 6924 6925 6926 6927
	ret = walk_page_buffers(handle, page_buffers(page),
				from, to, &partial,
				ocfs2_zero_func);
	if (ret < 0)
		mlog_errno(ret);
	else if (ocfs2_should_order_data(inode)) {
		ret = ocfs2_jbd2_file_inode(handle, inode);
6928 6929 6930 6931 6932 6933 6934 6935 6936 6937
		if (ret < 0)
			mlog_errno(ret);
	}

	if (!partial)
		SetPageUptodate(page);

	flush_dcache_page(page);
}

6938 6939 6940
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)
6941
{
6942
	int i;
6943 6944 6945 6946 6947 6948 6949 6950 6951
	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;

6952
	to = PAGE_CACHE_SIZE;
6953 6954 6955
	for(i = 0; i < numpages; i++) {
		page = pages[i];

6956 6957 6958 6959
		from = start & (PAGE_CACHE_SIZE - 1);
		if ((end >> PAGE_CACHE_SHIFT) == page->index)
			to = end & (PAGE_CACHE_SIZE - 1);

6960 6961 6962
		BUG_ON(from > PAGE_CACHE_SIZE);
		BUG_ON(to > PAGE_CACHE_SIZE);

6963 6964
		ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1,
					 &phys);
6965

6966
		start = (page->index + 1) << PAGE_CACHE_SHIFT;
6967 6968
	}
out:
6969 6970
	if (pages)
		ocfs2_unlock_and_free_pages(pages, numpages);
6971 6972
}

T
Tao Ma 已提交
6973 6974
int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end,
		     struct page **pages, int *num)
6975
{
6976
	int numpages, ret = 0;
6977 6978
	struct address_space *mapping = inode->i_mapping;
	unsigned long index;
6979
	loff_t last_page_bytes;
6980

6981
	BUG_ON(start > end);
6982

6983
	numpages = 0;
6984 6985
	last_page_bytes = PAGE_ALIGN(end);
	index = start >> PAGE_CACHE_SHIFT;
6986 6987 6988 6989 6990 6991 6992 6993 6994 6995
	do {
		pages[numpages] = grab_cache_page(mapping, index);
		if (!pages[numpages]) {
			ret = -ENOMEM;
			mlog_errno(ret);
			goto out;
		}

		numpages++;
		index++;
6996
	} while (index < (last_page_bytes >> PAGE_CACHE_SHIFT));
6997 6998 6999

out:
	if (ret != 0) {
7000 7001
		if (pages)
			ocfs2_unlock_and_free_pages(pages, numpages);
7002 7003 7004 7005 7006 7007 7008 7009
		numpages = 0;
	}

	*num = numpages;

	return ret;
}

T
Tao Ma 已提交
7010 7011 7012 7013 7014 7015 7016 7017 7018 7019 7020
static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end,
				struct page **pages, int *num)
{
	struct super_block *sb = inode->i_sb;

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

	return ocfs2_grab_pages(inode, start, end, pages, num);
}

7021 7022 7023 7024 7025 7026 7027 7028 7029
/*
 * 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.
 */
7030 7031
int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
				  u64 range_start, u64 range_end)
7032
{
7033
	int ret = 0, numpages;
7034 7035
	struct page **pages = NULL;
	u64 phys;
7036 7037
	unsigned int ext_flags;
	struct super_block *sb = inode->i_sb;
7038 7039 7040 7041 7042

	/*
	 * File systems which don't support sparse files zero on every
	 * extend.
	 */
7043
	if (!ocfs2_sparse_alloc(OCFS2_SB(sb)))
7044 7045
		return 0;

7046
	pages = kcalloc(ocfs2_pages_per_cluster(sb),
7047 7048 7049 7050 7051 7052 7053
			sizeof(struct page *), GFP_NOFS);
	if (pages == NULL) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

7054 7055 7056 7057 7058 7059
	if (range_start == range_end)
		goto out;

	ret = ocfs2_extent_map_get_blocks(inode,
					  range_start >> sb->s_blocksize_bits,
					  &phys, NULL, &ext_flags);
7060 7061 7062 7063 7064
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

7065 7066 7067 7068 7069
	/*
	 * 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)
7070 7071
		goto out;

7072 7073 7074 7075 7076 7077 7078
	ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages,
				   &numpages);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

7079 7080
	ocfs2_zero_cluster_pages(inode, range_start, range_end, pages,
				 numpages, phys, handle);
7081 7082 7083 7084 7085 7086

	/*
	 * 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.
	 */
7087 7088
	ret = filemap_fdatawrite_range(inode->i_mapping, range_start,
				       range_end - 1);
7089 7090 7091 7092 7093 7094 7095 7096 7097 7098
	if (ret)
		mlog_errno(ret);

out:
	if (pages)
		kfree(pages);

	return ret;
}

7099 7100
static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode,
					     struct ocfs2_dinode *di)
M
Mark Fasheh 已提交
7101 7102
{
	unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits;
7103
	unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size);
M
Mark Fasheh 已提交
7104

7105 7106 7107 7108 7109 7110 7111
	if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL)
		memset(&di->id2, 0, blocksize -
				    offsetof(struct ocfs2_dinode, id2) -
				    xattrsize);
	else
		memset(&di->id2, 0, blocksize -
				    offsetof(struct ocfs2_dinode, id2));
M
Mark Fasheh 已提交
7112 7113
}

7114 7115 7116
void ocfs2_dinode_new_extent_list(struct inode *inode,
				  struct ocfs2_dinode *di)
{
7117
	ocfs2_zero_dinode_id2_with_xattr(inode, di);
7118 7119
	di->id2.i_list.l_tree_depth = 0;
	di->id2.i_list.l_next_free_rec = 0;
7120 7121
	di->id2.i_list.l_count = cpu_to_le16(
		ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di));
7122 7123
}

M
Mark Fasheh 已提交
7124 7125 7126 7127 7128 7129 7130 7131 7132 7133 7134 7135 7136 7137
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.
	 */
7138
	ocfs2_zero_dinode_id2_with_xattr(inode, di);
M
Mark Fasheh 已提交
7139

7140 7141
	idata->id_count = cpu_to_le16(
			ocfs2_max_inline_data_with_xattr(inode->i_sb, di));
M
Mark Fasheh 已提交
7142 7143 7144 7145 7146 7147 7148 7149 7150 7151 7152 7153 7154 7155
}

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;
7156
	struct ocfs2_extent_tree et;
7157
	int did_quota = 0;
M
Mark Fasheh 已提交
7158 7159 7160 7161 7162 7163 7164 7165 7166 7167 7168 7169 7170 7171 7172 7173 7174 7175 7176

	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;
		}
	}

7177 7178
	handle = ocfs2_start_trans(osb,
				   ocfs2_inline_to_extents_credits(osb->sb));
M
Mark Fasheh 已提交
7179 7180 7181 7182 7183 7184
	if (IS_ERR(handle)) {
		ret = PTR_ERR(handle);
		mlog_errno(ret);
		goto out_unlock;
	}

7185
	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
7186
				      OCFS2_JOURNAL_ACCESS_WRITE);
M
Mark Fasheh 已提交
7187 7188 7189 7190 7191 7192 7193 7194 7195 7196
	if (ret) {
		mlog_errno(ret);
		goto out_commit;
	}

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

7197 7198 7199
		ret = dquot_alloc_space_nodirty(inode,
				       ocfs2_clusters_to_bytes(osb->sb, 1));
		if (ret)
7200 7201 7202
			goto out_commit;
		did_quota = 1;

7203 7204
		data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;

7205
		ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
M
Mark Fasheh 已提交
7206 7207 7208 7209 7210 7211 7212 7213 7214 7215 7216 7217 7218 7219 7220 7221 7222 7223 7224 7225 7226 7227 7228 7229 7230 7231 7232 7233 7234 7235 7236 7237 7238 7239 7240 7241 7242 7243 7244 7245 7246 7247 7248 7249 7250 7251 7252 7253 7254 7255
					   &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);

7256
	ocfs2_dinode_new_extent_list(inode, di);
M
Mark Fasheh 已提交
7257 7258 7259 7260 7261 7262 7263 7264 7265

	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.
		 */
7266
		ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
7267
		ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL);
M
Mark Fasheh 已提交
7268 7269 7270 7271 7272 7273 7274 7275 7276
		if (ret) {
			mlog_errno(ret);
			goto out_commit;
		}

		inode->i_blocks = ocfs2_inode_sector_count(inode);
	}

out_commit:
7277
	if (ret < 0 && did_quota)
7278
		dquot_free_space_nodirty(inode,
7279 7280
					  ocfs2_clusters_to_bytes(osb->sb, 1));

M
Mark Fasheh 已提交
7281 7282 7283 7284 7285 7286 7287 7288 7289 7290 7291 7292 7293 7294 7295
	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;
}

7296 7297 7298 7299 7300 7301 7302 7303 7304 7305 7306 7307
/*
 * 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
7308
	u32 clusters_to_del, new_highest_cpos, range;
7309
	u64 blkno = 0;
7310
	struct ocfs2_extent_list *el;
7311
	handle_t *handle = NULL;
7312
	struct inode *tl_inode = osb->osb_tl_inode;
M
Mark Fasheh 已提交
7313
	struct ocfs2_path *path = NULL;
7314
	struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
7315 7316
	struct ocfs2_alloc_context *meta_ac = NULL;
	struct ocfs2_refcount_tree *ref_tree = NULL;
7317 7318 7319

	mlog_entry_void();

M
Mark Fasheh 已提交
7320
	new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
7321 7322
						     i_size_read(inode));

7323 7324
	path = ocfs2_new_path(fe_bh, &di->id2.i_list,
			      ocfs2_journal_access_di);
M
Mark Fasheh 已提交
7325 7326 7327 7328 7329
	if (!path) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}
M
Mark Fasheh 已提交
7330 7331 7332

	ocfs2_extent_map_trunc(inode, new_highest_cpos);

7333
start:
7334 7335 7336 7337 7338 7339 7340 7341
	/*
	 * Check that we still have allocation to delete.
	 */
	if (OCFS2_I(inode)->ip_clusters == 0) {
		status = 0;
		goto bail;
	}

7342 7343
	credits = 0;

M
Mark Fasheh 已提交
7344 7345 7346
	/*
	 * Truncate always works against the rightmost tree branch.
	 */
7347
	status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX);
M
Mark Fasheh 已提交
7348 7349 7350
	if (status) {
		mlog_errno(status);
		goto bail;
7351 7352
	}

M
Mark Fasheh 已提交
7353 7354 7355 7356 7357 7358 7359 7360 7361 7362 7363 7364 7365 7366 7367
	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);
7368 7369 7370 7371 7372 7373 7374 7375 7376
	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;
	}

7377
	i = le16_to_cpu(el->l_next_free_rec) - 1;
M
Mark Fasheh 已提交
7378
	range = le32_to_cpu(el->l_recs[i].e_cpos) +
7379
		ocfs2_rec_clusters(el, &el->l_recs[i]);
M
Mark Fasheh 已提交
7380 7381 7382
	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) {
7383
		clusters_to_del = ocfs2_rec_clusters(el, &el->l_recs[i]);
7384
		blkno = le64_to_cpu(el->l_recs[i].e_blkno);
M
Mark Fasheh 已提交
7385
	} else if (range > new_highest_cpos) {
7386
		clusters_to_del = (ocfs2_rec_clusters(el, &el->l_recs[i]) +
7387
				   le32_to_cpu(el->l_recs[i].e_cpos)) -
M
Mark Fasheh 已提交
7388
				  new_highest_cpos;
7389 7390 7391 7392
		blkno = le64_to_cpu(el->l_recs[i].e_blkno) +
			ocfs2_clusters_to_blocks(inode->i_sb,
				ocfs2_rec_clusters(el, &el->l_recs[i]) -
				clusters_to_del);
M
Mark Fasheh 已提交
7393 7394 7395 7396
	} else {
		status = 0;
		goto bail;
	}
7397

M
Mark Fasheh 已提交
7398 7399 7400
	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);

7401 7402 7403 7404 7405 7406 7407 7408 7409 7410 7411 7412 7413 7414 7415 7416 7417 7418 7419 7420 7421 7422 7423
	if (el->l_recs[i].e_flags & OCFS2_EXT_REFCOUNTED && clusters_to_del) {
		BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
			 OCFS2_HAS_REFCOUNT_FL));

		status = ocfs2_lock_refcount_tree(osb,
						le64_to_cpu(di->i_refcount_loc),
						1, &ref_tree, NULL);
		if (status) {
			mlog_errno(status);
			goto bail;
		}

		status = ocfs2_prepare_refcount_change_for_del(inode, fe_bh,
							       blkno,
							       clusters_to_del,
							       &credits,
							       &meta_ac);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

7424
	mutex_lock(&tl_inode->i_mutex);
7425 7426 7427 7428 7429 7430 7431 7432 7433 7434 7435 7436
	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;
		}
	}

7437
	credits += ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del,
M
Mark Fasheh 已提交
7438 7439
						(struct ocfs2_dinode *)fe_bh->b_data,
						el);
7440
	handle = ocfs2_start_trans(osb, credits);
7441 7442 7443 7444 7445 7446 7447
	if (IS_ERR(handle)) {
		status = PTR_ERR(handle);
		handle = NULL;
		mlog_errno(status);
		goto bail;
	}

M
Mark Fasheh 已提交
7448
	status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh, handle,
7449
				   tc, path, meta_ac);
7450 7451 7452 7453 7454
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

7455
	mutex_unlock(&tl_inode->i_mutex);
7456 7457
	tl_sem = 0;

7458
	ocfs2_commit_trans(osb, handle);
7459 7460
	handle = NULL;

M
Mark Fasheh 已提交
7461 7462
	ocfs2_reinit_path(path, 1);

7463 7464 7465 7466 7467 7468 7469 7470 7471 7472
	if (meta_ac) {
		ocfs2_free_alloc_context(meta_ac);
		meta_ac = NULL;
	}

	if (ref_tree) {
		ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
		ref_tree = NULL;
	}

M
Mark Fasheh 已提交
7473
	/*
7474 7475
	 * The check above will catch the case where we've truncated
	 * away all allocation.
M
Mark Fasheh 已提交
7476
	 */
7477 7478
	goto start;

7479 7480 7481 7482 7483
bail:

	ocfs2_schedule_truncate_log_flush(osb, 1);

	if (tl_sem)
7484
		mutex_unlock(&tl_inode->i_mutex);
7485 7486

	if (handle)
7487
		ocfs2_commit_trans(osb, handle);
7488

7489 7490 7491 7492 7493 7494
	if (meta_ac)
		ocfs2_free_alloc_context(meta_ac);

	if (ref_tree)
		ocfs2_unlock_refcount_tree(osb, ref_tree, 1);

7495 7496
	ocfs2_run_deallocs(osb, &tc->tc_dealloc);

M
Mark Fasheh 已提交
7497
	ocfs2_free_path(path);
7498 7499 7500 7501 7502 7503 7504 7505 7506

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

	mlog_exit(status);
	return status;
}

/*
7507
 * Expects the inode to already be locked.
7508 7509 7510 7511 7512 7513
 */
int ocfs2_prepare_truncate(struct ocfs2_super *osb,
			   struct inode *inode,
			   struct buffer_head *fe_bh,
			   struct ocfs2_truncate_context **tc)
{
7514
	int status;
7515 7516 7517 7518 7519 7520 7521 7522 7523 7524 7525 7526 7527 7528
	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 ="
7529 7530
	     "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters,
	     (unsigned long long)le64_to_cpu(fe->i_size));
7531

7532
	*tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL);
7533 7534 7535 7536 7537
	if (!(*tc)) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}
7538
	ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc);
7539 7540

	if (fe->id2.i_list.l_tree_depth) {
7541
		status = ocfs2_read_extent_block(INODE_CACHE(inode),
7542 7543
						 le64_to_cpu(fe->i_last_eb_blk),
						 &last_eb_bh);
7544 7545 7546 7547 7548 7549 7550 7551 7552 7553 7554 7555 7556 7557 7558 7559 7560 7561 7562 7563
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
	}

	(*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
7564 7565 7566 7567 7568 7569 7570 7571 7572 7573 7574 7575 7576 7577 7578 7579 7580 7581 7582 7583 7584 7585 7586 7587 7588 7589 7590 7591 7592 7593 7594 7595 7596 7597 7598 7599 7600 7601 7602
/*
 * '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;
	}

7603
	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
7604
				      OCFS2_JOURNAL_ACCESS_WRITE);
M
Mark Fasheh 已提交
7605 7606 7607 7608 7609 7610 7611 7612 7613 7614 7615 7616 7617 7618 7619 7620 7621 7622 7623 7624 7625 7626 7627 7628 7629 7630 7631 7632 7633 7634 7635 7636 7637 7638
	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;
}

7639 7640
static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc)
{
7641 7642 7643 7644 7645 7646 7647
	/*
	 * 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");
7648

7649
	brelse(tc->tc_last_eb_bh);
7650 7651 7652

	kfree(tc);
}