alloc.c 191.2 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 53 54

#include "buffer_head_io.h"

55

56 57 58 59 60
/*
 * 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
61
 * ocfs2_init_<thingy>_extent_tree() function.  That's pretty much it
62 63
 * for the allocation portion of the extent tree.
 */
64
struct ocfs2_extent_tree_operations {
65 66 67 68 69 70 71
	/*
	 * 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.
	 */
72 73 74
	void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et,
				   u64 blkno);
	u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et);
75 76 77 78 79 80 81

	/*
	 * 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.
	 */
82
	void (*eo_update_clusters)(struct ocfs2_extent_tree *et,
83
				   u32 new_clusters);
84

85 86 87 88 89 90 91
	/*
	 * 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);

92 93 94 95
	/*
	 * If ->eo_insert_check() exists, it is called before rec is
	 * inserted into the extent tree.  It is optional.
	 */
96
	int (*eo_insert_check)(struct ocfs2_extent_tree *et,
97
			       struct ocfs2_extent_rec *rec);
98
	int (*eo_sanity_check)(struct ocfs2_extent_tree *et);
99

100 101 102 103 104 105 106 107 108 109
	/*
	 * --------------------------------------------------------------
	 * 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.
	 */
110
	void (*eo_fill_root_el)(struct ocfs2_extent_tree *et);
111 112 113 114 115 116

	/*
	 * ->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.
	 */
117
	void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et);
118 119 120
};


121 122 123 124 125 126 127
/*
 * 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);
128
static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
129
					 u32 clusters);
130 131
static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
					     u32 clusters);
132
static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
133
				     struct ocfs2_extent_rec *rec);
134
static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et);
135 136 137 138 139
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,
140
	.eo_extent_map_truncate	= ocfs2_dinode_extent_map_truncate,
141 142 143 144
	.eo_insert_check	= ocfs2_dinode_insert_check,
	.eo_sanity_check	= ocfs2_dinode_sanity_check,
	.eo_fill_root_el	= ocfs2_dinode_fill_root_el,
};
145

146 147 148
static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
					 u64 blkno)
{
149
	struct ocfs2_dinode *di = et->et_object;
150

151
	BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
152 153 154 155 156
	di->i_last_eb_blk = cpu_to_le64(blkno);
}

static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et)
{
157
	struct ocfs2_dinode *di = et->et_object;
158

159
	BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
160 161 162
	return le64_to_cpu(di->i_last_eb_blk);
}

163
static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
164 165
					 u32 clusters)
{
166
	struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
167
	struct ocfs2_dinode *di = et->et_object;
168 169

	le32_add_cpu(&di->i_clusters, clusters);
170 171 172
	spin_lock(&oi->ip_lock);
	oi->ip_clusters = le32_to_cpu(di->i_clusters);
	spin_unlock(&oi->ip_lock);
173 174
}

175 176 177 178 179 180 181 182
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);
}

183
static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
184 185
				     struct ocfs2_extent_rec *rec)
{
186 187
	struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
	struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb);
188

189
	BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL);
190
	mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) &&
191
			(oi->ip_clusters != le32_to_cpu(rec->e_cpos)),
192 193 194
			"Device %s, asking for sparse allocation: inode %llu, "
			"cpos %u, clusters %u\n",
			osb->dev_str,
195 196
			(unsigned long long)oi->ip_blkno,
			rec->e_cpos, oi->ip_clusters);
197 198 199 200

	return 0;
}

201
static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et)
202
{
203
	struct ocfs2_dinode *di = et->et_object;
204

205
	BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
206
	BUG_ON(!OCFS2_IS_VALID_DINODE(di));
207

208
	return 0;
209 210
}

211 212 213 214 215 216 217
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;
}

218

219 220
static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et)
{
221
	struct ocfs2_xattr_value_buf *vb = et->et_object;
222

223
	et->et_root_el = &vb->vb_xv->xr_list;
224 225
}

226 227 228
static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et,
					      u64 blkno)
{
229
	struct ocfs2_xattr_value_buf *vb = et->et_object;
230

231
	vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno);
232 233 234 235
}

static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et)
{
236
	struct ocfs2_xattr_value_buf *vb = et->et_object;
237

238
	return le64_to_cpu(vb->vb_xv->xr_last_eb_blk);
239 240
}

241
static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et,
242 243
					      u32 clusters)
{
244
	struct ocfs2_xattr_value_buf *vb = et->et_object;
245

246
	le32_add_cpu(&vb->vb_xv->xr_clusters, clusters);
247 248
}

249
static struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = {
250 251 252
	.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,
253
	.eo_fill_root_el	= ocfs2_xattr_value_fill_root_el,
254 255
};

256 257 258 259 260 261 262
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;
}

263
static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et)
264
{
265
	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
266
	et->et_max_leaf_clusters =
267
		ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE);
268 269
}

T
Tao Ma 已提交
270 271 272
static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et,
					     u64 blkno)
{
273
	struct ocfs2_xattr_block *xb = et->et_object;
T
Tao Ma 已提交
274 275 276 277 278 279 280
	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)
{
281
	struct ocfs2_xattr_block *xb = et->et_object;
T
Tao Ma 已提交
282 283 284 285 286
	struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;

	return le64_to_cpu(xt->xt_last_eb_blk);
}

287
static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et,
T
Tao Ma 已提交
288 289
					     u32 clusters)
{
290
	struct ocfs2_xattr_block *xb = et->et_object;
T
Tao Ma 已提交
291 292 293 294 295

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

static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = {
296 297 298
	.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,
299
	.eo_fill_root_el	= ocfs2_xattr_tree_fill_root_el,
300
	.eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters,
T
Tao Ma 已提交
301 302
};

303 304 305 306 307 308 309 310 311 312 313 314 315 316 317
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);
}

318
static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et,
319 320 321 322 323 324 325
					  u32 clusters)
{
	struct ocfs2_dx_root_block *dx_root = et->et_object;

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

326
static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et)
327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349
{
	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,
};

350 351 352
static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et,
				     struct inode *inode,
				     struct buffer_head *bh,
353
				     ocfs2_journal_access_func access,
354 355
				     void *obj,
				     struct ocfs2_extent_tree_operations *ops)
356
{
357
	et->et_ops = ops;
358
	et->et_root_bh = bh;
359
	et->et_ci = INODE_CACHE(inode);
360
	et->et_root_journal_access = access;
361 362 363
	if (!obj)
		obj = (void *)bh->b_data;
	et->et_object = obj;
364

365
	et->et_ops->eo_fill_root_el(et);
366 367 368
	if (!et->et_ops->eo_fill_max_leaf_clusters)
		et->et_max_leaf_clusters = 0;
	else
369
		et->et_ops->eo_fill_max_leaf_clusters(et);
370 371
}

372 373 374
void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et,
				   struct inode *inode,
				   struct buffer_head *bh)
375
{
376 377
	__ocfs2_init_extent_tree(et, inode, bh, ocfs2_journal_access_di,
				 NULL, &ocfs2_dinode_et_ops);
378 379
}

380
void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et,
381
				       struct inode *inode,
382
				       struct buffer_head *bh)
383
{
384 385
	__ocfs2_init_extent_tree(et, inode, bh, ocfs2_journal_access_xb,
				 NULL, &ocfs2_xattr_tree_et_ops);
386 387
}

388 389
void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et,
					struct inode *inode,
390
					struct ocfs2_xattr_value_buf *vb)
391
{
392
	__ocfs2_init_extent_tree(et, inode, vb->vb_bh, vb->vb_access, vb,
393
				 &ocfs2_xattr_value_et_ops);
394 395
}

396 397 398 399 400 401 402 403
void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et,
				    struct inode *inode,
				    struct buffer_head *bh)
{
	__ocfs2_init_extent_tree(et, inode, bh, ocfs2_journal_access_dr,
				 NULL, &ocfs2_dx_root_et_ops);
}

404 405
static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et,
					    u64 new_last_eb_blk)
406
{
407
	et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk);
408 409
}

410
static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et)
411
{
412
	return et->et_ops->eo_get_last_eb_blk(et);
413 414
}

415
static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et,
416 417
					    u32 clusters)
{
418
	et->et_ops->eo_update_clusters(et, clusters);
419 420
}

421 422 423 424 425 426 427
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);
}

428 429 430 431
static inline int ocfs2_et_root_journal_access(handle_t *handle,
					       struct ocfs2_extent_tree *et,
					       int type)
{
432
	return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh,
433 434 435
					  type);
}

436
static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et,
437 438 439 440 441
					struct ocfs2_extent_rec *rec)
{
	int ret = 0;

	if (et->et_ops->eo_insert_check)
442
		ret = et->et_ops->eo_insert_check(et, rec);
443 444 445
	return ret;
}

446
static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et)
447
{
448 449 450
	int ret = 0;

	if (et->et_ops->eo_sanity_check)
451
		ret = et->et_ops->eo_sanity_check(et);
452
	return ret;
453 454
}

M
Mark Fasheh 已提交
455
static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc);
456 457
static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
					 struct ocfs2_extent_block *eb);
458

M
Mark Fasheh 已提交
459 460 461 462 463 464 465 466 467 468 469
/*
 * Structures which describe a path through a btree, and functions to
 * manipulate them.
 *
 * The idea here is to be as generic as possible with the tree
 * manipulation code.
 */
struct ocfs2_path_item {
	struct buffer_head		*bh;
	struct ocfs2_extent_list	*el;
};
470

M
Mark Fasheh 已提交
471
#define OCFS2_MAX_PATH_DEPTH	5
472

M
Mark Fasheh 已提交
473
struct ocfs2_path {
474 475 476
	int				p_tree_depth;
	ocfs2_journal_access_func	p_root_access;
	struct ocfs2_path_item		p_node[OCFS2_MAX_PATH_DEPTH];
M
Mark Fasheh 已提交
477
};
478

M
Mark Fasheh 已提交
479 480
#define path_root_bh(_path) ((_path)->p_node[0].bh)
#define path_root_el(_path) ((_path)->p_node[0].el)
481
#define path_root_access(_path)((_path)->p_root_access)
M
Mark Fasheh 已提交
482 483 484
#define path_leaf_bh(_path) ((_path)->p_node[(_path)->p_tree_depth].bh)
#define path_leaf_el(_path) ((_path)->p_node[(_path)->p_tree_depth].el)
#define path_num_items(_path) ((_path)->p_tree_depth + 1)
485

486 487
static int ocfs2_find_path(struct ocfs2_caching_info *ci,
			   struct ocfs2_path *path, u32 cpos);
488 489
static void ocfs2_adjust_rightmost_records(handle_t *handle,
					   struct ocfs2_extent_tree *et,
490 491
					   struct ocfs2_path *path,
					   struct ocfs2_extent_rec *insert_rec);
M
Mark Fasheh 已提交
492 493 494 495 496 497 498 499 500
/*
 * Reset the actual path elements so that we can re-use the structure
 * to build another path. Generally, this involves freeing the buffer
 * heads.
 */
static void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root)
{
	int i, start = 0, depth = 0;
	struct ocfs2_path_item *node;
501

M
Mark Fasheh 已提交
502 503
	if (keep_root)
		start = 1;
504

M
Mark Fasheh 已提交
505 506 507 508 509 510 511 512 513 514 515 516 517 518 519
	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);
520 521
	else
		path_root_access(path) = NULL;
M
Mark Fasheh 已提交
522 523 524 525 526 527 528 529 530 531 532 533

	path->p_tree_depth = depth;
}

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

534 535 536 537 538 539 540 541 542 543 544 545 546
/*
 * 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));
547
	BUG_ON(path_root_access(dest) != path_root_access(src));
548 549 550 551 552 553 554 555 556 557 558 559

	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
560 561 562 563 564 565 566 567 568
/*
 * 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));
569
	BUG_ON(path_root_access(dest) != path_root_access(src));
M
Mark Fasheh 已提交
570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604

	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,
605 606
					 struct ocfs2_extent_list *root_el,
					 ocfs2_journal_access_func access)
M
Mark Fasheh 已提交
607 608
{
	struct ocfs2_path *path;
609

M
Mark Fasheh 已提交
610 611 612 613 614 615 616 617
	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;
618
		path_root_access(path) = access;
M
Mark Fasheh 已提交
619 620 621 622 623
	}

	return path;
}

624 625
static struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path)
{
626 627
	return ocfs2_new_path(path_root_bh(path), path_root_el(path),
			      path_root_access(path));
628 629 630 631
}

static struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et)
{
632 633 634 635 636 637 638 639 640 641 642 643
	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.
 */
static int ocfs2_path_bh_journal_access(handle_t *handle,
644
					struct ocfs2_caching_info *ci,
645 646 647 648 649 650 651 652 653 654 655
					struct ocfs2_path *path,
					int idx)
{
	ocfs2_journal_access_func access = path_root_access(path);

	if (!access)
		access = ocfs2_journal_access;

	if (idx)
		access = ocfs2_journal_access_eb;

656
	return access(handle, ci, path->p_node[idx].bh,
657
		      OCFS2_JOURNAL_ACCESS_WRITE);
658 659
}

M
Mark Fasheh 已提交
660 661 662
/*
 * Convenience function to journal all components in a path.
 */
663 664
static int ocfs2_journal_access_path(struct ocfs2_caching_info *ci,
				     handle_t *handle,
M
Mark Fasheh 已提交
665 666 667 668 669 670 671 672
				     struct ocfs2_path *path)
{
	int i, ret = 0;

	if (!path)
		goto out;

	for(i = 0; i < path_num_items(path); i++) {
673
		ret = ocfs2_path_bh_journal_access(handle, ci, path, i);
M
Mark Fasheh 已提交
674 675 676 677 678 679 680 681 682 683
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}
	}

out:
	return ret;
}

684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713
/*
 * Return the index of the extent record which contains cluster #v_cluster.
 * -1 is returned if it was not found.
 *
 * Should work fine on interior and exterior nodes.
 */
int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster)
{
	int ret = -1;
	int i;
	struct ocfs2_extent_rec *rec;
	u32 rec_end, rec_start, clusters;

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

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

		rec_end = rec_start + clusters;

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

	return ret;
}

M
Mark Fasheh 已提交
714 715 716
enum ocfs2_contig_type {
	CONTIG_NONE = 0,
	CONTIG_LEFT,
717 718
	CONTIG_RIGHT,
	CONTIG_LEFTRIGHT,
M
Mark Fasheh 已提交
719 720
};

721 722 723 724 725

/*
 * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and
 * ocfs2_extent_contig only work properly against leaf nodes!
 */
M
Mark Fasheh 已提交
726 727 728
static int ocfs2_block_extent_contig(struct super_block *sb,
				     struct ocfs2_extent_rec *ext,
				     u64 blkno)
729
{
730 731 732 733 734 735
	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;
736 737
}

M
Mark Fasheh 已提交
738 739 740
static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left,
				  struct ocfs2_extent_rec *right)
{
741 742 743 744 745 746
	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
747 748 749
}

static enum ocfs2_contig_type
750
	ocfs2_extent_contig(struct super_block *sb,
M
Mark Fasheh 已提交
751 752 753 754 755
			    struct ocfs2_extent_rec *ext,
			    struct ocfs2_extent_rec *insert_rec)
{
	u64 blkno = le64_to_cpu(insert_rec->e_blkno);

756 757 758 759 760 761 762 763
	/*
	 * 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
764
	if (ocfs2_extents_adjacent(ext, insert_rec) &&
765
	    ocfs2_block_extent_contig(sb, ext, blkno))
M
Mark Fasheh 已提交
766 767 768 769
			return CONTIG_RIGHT;

	blkno = le64_to_cpu(ext->e_blkno);
	if (ocfs2_extents_adjacent(insert_rec, ext) &&
770
	    ocfs2_block_extent_contig(sb, insert_rec, blkno))
M
Mark Fasheh 已提交
771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787
		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,
};

788 789 790 791 792 793
enum ocfs2_split_type {
	SPLIT_NONE = 0,
	SPLIT_LEFT,
	SPLIT_RIGHT,
};

M
Mark Fasheh 已提交
794
struct ocfs2_insert_type {
795
	enum ocfs2_split_type	ins_split;
M
Mark Fasheh 已提交
796 797 798 799 800 801
	enum ocfs2_append_type	ins_appending;
	enum ocfs2_contig_type	ins_contig;
	int			ins_contig_index;
	int			ins_tree_depth;
};

802 803 804 805 806 807
struct ocfs2_merge_ctxt {
	enum ocfs2_contig_type	c_contig_type;
	int			c_has_empty_extent;
	int			c_split_covers_rec;
};

808 809 810
static int ocfs2_validate_extent_block(struct super_block *sb,
				       struct buffer_head *bh)
{
J
Joel Becker 已提交
811
	int rc;
812 813 814
	struct ocfs2_extent_block *eb =
		(struct ocfs2_extent_block *)bh->b_data;

815 816 817
	mlog(0, "Validating extent block %llu\n",
	     (unsigned long long)bh->b_blocknr);

J
Joel Becker 已提交
818 819 820 821 822 823 824 825
	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);
826 827 828
	if (rc) {
		mlog(ML_ERROR, "Checksum failed for extent block %llu\n",
		     (unsigned long long)bh->b_blocknr);
J
Joel Becker 已提交
829
		return rc;
830
	}
J
Joel Becker 已提交
831 832 833 834 835

	/*
	 * Errors after here are fatal.
	 */

836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864
	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;
}

865
int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno,
866 867 868 869 870
			    struct buffer_head **bh)
{
	int rc;
	struct buffer_head *tmp = *bh;

871
	rc = ocfs2_read_block(ci, eb_blkno, &tmp,
872
			      ocfs2_validate_extent_block);
873 874

	/* If ocfs2_read_block() got us a new bh, pass it up. */
875
	if (!rc && !*bh)
876 877 878 879 880 881
		*bh = tmp;

	return rc;
}


882 883 884 885
/*
 * How many free extents have we got before we need more meta data?
 */
int ocfs2_num_free_extents(struct ocfs2_super *osb,
886
			   struct ocfs2_extent_tree *et)
887 888
{
	int retval;
889
	struct ocfs2_extent_list *el = NULL;
890 891
	struct ocfs2_extent_block *eb;
	struct buffer_head *eb_bh = NULL;
892
	u64 last_eb_blk = 0;
893 894 895

	mlog_entry_void();

896 897
	el = et->et_root_el;
	last_eb_blk = ocfs2_et_get_last_eb_blk(et);
898

899
	if (last_eb_blk) {
900 901
		retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk,
						 &eb_bh);
902 903 904 905 906 907
		if (retval < 0) {
			mlog_errno(retval);
			goto bail;
		}
		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
		el = &eb->h_list;
908
	}
909 910 911 912 913

	BUG_ON(el->l_tree_depth != 0);

	retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec);
bail:
914
	brelse(eb_bh);
915 916 917 918 919 920 921 922 923 924

	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
 */
925 926
static int ocfs2_create_new_meta_bhs(handle_t *handle,
				     struct ocfs2_extent_tree *et,
927 928 929 930 931 932 933 934
				     int wanted,
				     struct ocfs2_alloc_context *meta_ac,
				     struct buffer_head *bhs[])
{
	int count, status, i;
	u16 suballoc_bit_start;
	u32 num_got;
	u64 first_blkno;
935 936
	struct ocfs2_super *osb =
		OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961
	struct ocfs2_extent_block *eb;

	mlog_entry_void();

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

		for(i = count;  i < (num_got + count); i++) {
			bhs[i] = sb_getblk(osb->sb, first_blkno);
			if (bhs[i] == NULL) {
				status = -EIO;
				mlog_errno(status);
				goto bail;
			}
962
			ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]);
963

964 965
			status = ocfs2_journal_access_eb(handle, et->et_ci,
							 bhs[i],
966
							 OCFS2_JOURNAL_ACCESS_CREATE);
967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001
			if (status < 0) {
				mlog_errno(status);
				goto bail;
			}

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

			suballoc_bit_start++;
			first_blkno++;

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

		count += num_got;
	}

	status = 0;
bail:
	if (status < 0) {
		for(i = 0; i < wanted; i++) {
1002
			brelse(bhs[i]);
1003 1004 1005 1006 1007 1008 1009
			bhs[i] = NULL;
		}
	}
	mlog_exit(status);
	return status;
}

M
Mark Fasheh 已提交
1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028
/*
 * 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) +
1029
		ocfs2_rec_clusters(el, &el->l_recs[i]);
M
Mark Fasheh 已提交
1030 1031
}

1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049
/*
 * 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;
	}

1050
	status = ocfs2_find_path(et->et_ci, path, UINT_MAX);
1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062
	if (status < 0) {
		mlog_errno(status);
		goto out;
	}

	status = ocfs2_extend_trans(handle, path_num_items(path) +
				    handle->h_buffer_credits);
	if (status < 0) {
		mlog_errno(status);
		goto out;
	}

1063
	status = ocfs2_journal_access_path(et->et_ci, handle, path);
1064 1065 1066 1067 1068 1069 1070 1071
	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];

1072
	ocfs2_adjust_rightmost_records(handle, et, path, rec);
1073 1074 1075 1076 1077 1078

out:
	ocfs2_free_path(path);
	return status;
}

1079 1080
/*
 * Add an entire tree branch to our inode. eb_bh is the extent block
1081
 * to start at, if we don't want to start the branch at the root
1082 1083 1084 1085 1086 1087
 * 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
1088
 * contain a single record with cluster count == 0.
1089
 */
1090
static int ocfs2_add_branch(handle_t *handle,
1091
			    struct ocfs2_extent_tree *et,
1092
			    struct buffer_head *eb_bh,
1093
			    struct buffer_head **last_eb_bh,
1094 1095 1096 1097 1098 1099 1100 1101 1102
			    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;
1103
	u32 new_cpos, root_end;
1104 1105 1106

	mlog_entry_void();

1107
	BUG_ON(!last_eb_bh || !*last_eb_bh);
1108 1109 1110 1111 1112

	if (eb_bh) {
		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
		el = &eb->h_list;
	} else
1113
		el = et->et_root_el;
1114 1115 1116 1117 1118 1119

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

	new_blocks = le16_to_cpu(el->l_tree_depth);

1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133
	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);
1134
		status = ocfs2_adjust_rightmost_branch(handle, et);
1135 1136 1137 1138 1139 1140
		if (status) {
			mlog_errno(status);
			goto bail;
		}
	}

1141 1142 1143 1144 1145 1146 1147 1148 1149
	/* 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;
	}

1150
	status = ocfs2_create_new_meta_bhs(handle, et, new_blocks,
1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167
					   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;
1168 1169
		/* ocfs2_create_new_meta_bhs() should create it right! */
		BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
1170 1171
		eb_el = &eb->h_list;

1172
		status = ocfs2_journal_access_eb(handle, et->et_ci, bh,
1173
						 OCFS2_JOURNAL_ACCESS_CREATE);
1174 1175 1176 1177 1178 1179 1180 1181
		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
1182 1183 1184 1185 1186
		/*
		 * This actually counts as an empty extent as
		 * c_clusters == 0
		 */
		eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos);
1187
		eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno);
1188 1189 1190 1191 1192 1193
		/*
		 * 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);
1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211
		if (!eb_el->l_tree_depth)
			new_last_eb_blk = le64_to_cpu(eb->h_blkno);

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

		next_blkno = le64_to_cpu(eb->h_blkno);
	}

	/* This is a bit hairy. We want to update up to three blocks
	 * here without leaving any of them in an inconsistent state
	 * in case of error. We don't have to worry about
	 * journal_dirty erroring as it won't unless we've aborted the
	 * handle (in which case we would never be here) so reserving
	 * the write with journal_access is all we need to do. */
1212
	status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh,
1213
					 OCFS2_JOURNAL_ACCESS_WRITE);
1214 1215 1216 1217
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
1218
	status = ocfs2_et_root_journal_access(handle, et,
1219
					      OCFS2_JOURNAL_ACCESS_WRITE);
1220 1221 1222 1223 1224
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
	if (eb_bh) {
1225
		status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh,
1226
						 OCFS2_JOURNAL_ACCESS_WRITE);
1227 1228 1229 1230 1231 1232 1233
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

	/* Link the new branch into the rest of the tree (el will
1234
	 * either be on the root_bh, or the extent block passed in. */
1235 1236
	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
1237
	el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1238
	el->l_recs[i].e_int_clusters = 0;
1239 1240 1241 1242
	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. */
1243
	ocfs2_et_set_last_eb_blk(et, new_last_eb_blk);
1244

1245
	eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
1246 1247
	eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);

1248
	status = ocfs2_journal_dirty(handle, *last_eb_bh);
1249 1250
	if (status < 0)
		mlog_errno(status);
1251
	status = ocfs2_journal_dirty(handle, et->et_root_bh);
1252 1253 1254 1255 1256 1257 1258 1259
	if (status < 0)
		mlog_errno(status);
	if (eb_bh) {
		status = ocfs2_journal_dirty(handle, eb_bh);
		if (status < 0)
			mlog_errno(status);
	}

1260 1261 1262 1263 1264 1265 1266 1267
	/*
	 * 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];

1268 1269 1270 1271
	status = 0;
bail:
	if (new_eb_bhs) {
		for (i = 0; i < new_blocks; i++)
1272
			brelse(new_eb_bhs[i]);
1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284
		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.
 */
1285
static int ocfs2_shift_tree_depth(handle_t *handle,
1286
				  struct ocfs2_extent_tree *et,
1287 1288 1289 1290
				  struct ocfs2_alloc_context *meta_ac,
				  struct buffer_head **ret_new_eb_bh)
{
	int status, i;
M
Mark Fasheh 已提交
1291
	u32 new_clusters;
1292 1293
	struct buffer_head *new_eb_bh = NULL;
	struct ocfs2_extent_block *eb;
1294
	struct ocfs2_extent_list  *root_el;
1295 1296 1297 1298
	struct ocfs2_extent_list  *eb_el;

	mlog_entry_void();

1299
	status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac,
1300 1301 1302 1303 1304 1305 1306
					   &new_eb_bh);
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	eb = (struct ocfs2_extent_block *) new_eb_bh->b_data;
1307 1308
	/* ocfs2_create_new_meta_bhs() should create it right! */
	BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
1309 1310

	eb_el = &eb->h_list;
1311
	root_el = et->et_root_el;
1312

1313
	status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh,
1314
					 OCFS2_JOURNAL_ACCESS_CREATE);
1315 1316 1317 1318 1319
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

1320 1321 1322 1323 1324
	/* 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];
1325 1326 1327 1328 1329 1330 1331

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

1332
	status = ocfs2_et_root_journal_access(handle, et,
1333
					      OCFS2_JOURNAL_ACCESS_WRITE);
1334 1335 1336 1337 1338
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

M
Mark Fasheh 已提交
1339 1340
	new_clusters = ocfs2_sum_rightmost_rec(eb_el);

1341 1342 1343 1344 1345 1346 1347 1348
	/* 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);
1349 1350 1351

	/* If this is our 1st tree depth shift, then last_eb_blk
	 * becomes the allocated extent block */
1352
	if (root_el->l_tree_depth == cpu_to_le16(1))
1353
		ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
1354

1355
	status = ocfs2_journal_dirty(handle, et->et_root_bh);
1356 1357 1358 1359 1360 1361 1362 1363 1364
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	*ret_new_eb_bh = new_eb_bh;
	new_eb_bh = NULL;
	status = 0;
bail:
1365
	brelse(new_eb_bh);
1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379

	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'
 *
1380
 * 2) the search fails to find anything, but the root_el has room. We
1381 1382
 *    pass NULL back in *lowest_eb_bh, but still return '0'
 *
1383
 * 3) the search fails to find anything AND the root_el is full, in
1384 1385 1386 1387
 *    which case we return > 0
 *
 * return status < 0 indicates an error.
 */
1388
static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et,
1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401
				    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;

1402
	el = et->et_root_el;
1403 1404 1405

	while(le16_to_cpu(el->l_tree_depth) > 1) {
		if (le16_to_cpu(el->l_next_free_rec) == 0) {
1406 1407
			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
				    "Owner %llu has empty "
1408
				    "extent list (next_free_rec == 0)",
1409
				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
1410 1411 1412 1413 1414 1415
			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) {
1416 1417
			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
				    "Owner %llu has extent "
1418 1419
				    "list where extent # %d has no physical "
				    "block start",
1420
				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i);
1421 1422 1423 1424
			status = -EIO;
			goto bail;
		}

1425 1426
		brelse(bh);
		bh = NULL;
1427

1428
		status = ocfs2_read_extent_block(et->et_ci, blkno, &bh);
1429 1430 1431 1432
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
M
Mark Fasheh 已提交
1433 1434 1435 1436 1437 1438

		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)) {
1439
			brelse(lowest_bh);
M
Mark Fasheh 已提交
1440 1441 1442 1443 1444 1445 1446
			lowest_bh = bh;
			get_bh(lowest_bh);
		}
	}

	/* If we didn't find one and the fe doesn't have any room,
	 * then return '1' */
1447
	el = et->et_root_el;
1448
	if (!lowest_bh && (el->l_next_free_rec == el->l_count))
M
Mark Fasheh 已提交
1449 1450 1451 1452
		status = 1;

	*target_bh = lowest_bh;
bail:
1453
	brelse(bh);
M
Mark Fasheh 已提交
1454 1455 1456 1457 1458

	mlog_exit(status);
	return status;
}

1459 1460 1461 1462 1463 1464 1465
/*
 * 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.
1466 1467
 *
 * *last_eb_bh will be updated by ocfs2_add_branch().
1468
 */
1469 1470
static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et,
			   int *final_depth, struct buffer_head **last_eb_bh,
1471 1472 1473
			   struct ocfs2_alloc_context *meta_ac)
{
	int ret, shift;
1474
	struct ocfs2_extent_list *el = et->et_root_el;
1475
	int depth = le16_to_cpu(el->l_tree_depth);
1476 1477 1478 1479
	struct buffer_head *bh = NULL;

	BUG_ON(meta_ac == NULL);

1480
	shift = ocfs2_find_branch_target(et, &bh);
1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496
	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). */
1497
		ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh);
1498 1499 1500 1501 1502
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}
		depth++;
1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515
		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;
1516
			goto out;
1517
		}
1518 1519 1520 1521 1522
	}

	/* call ocfs2_add_branch to add the final part of the tree with
	 * the new data. */
	mlog(0, "add branch. bh = %p\n", bh);
1523
	ret = ocfs2_add_branch(handle, et, bh, last_eb_bh,
1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536
			       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
1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567
/*
 * 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
1568
	BUG_ON(el->l_next_free_rec == el->l_count && !has_empty);
M
Mark Fasheh 已提交
1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632

	/*
	 * 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;

}

1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648
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
1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659
/*
 * 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);

1660 1661
	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);

M
Mark Fasheh 已提交
1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692
	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.
 */
1693
static int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et,
M
Mark Fasheh 已提交
1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710
				   struct ocfs2_path *left,
				   struct ocfs2_path *right)
{
	int i = 0;

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

	do {
		i++;

		/*
		 * The caller didn't pass two adjacent paths.
		 */
		mlog_bug_on_msg(i > left->p_tree_depth,
1711
				"Owner %llu, left depth %u, right depth %u\n"
M
Mark Fasheh 已提交
1712
				"left leaf blk %llu, right leaf blk %llu\n",
1713 1714
				(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
				left->p_tree_depth, right->p_tree_depth,
M
Mark Fasheh 已提交
1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730
				(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.
 */
1731
static int __ocfs2_find_path(struct ocfs2_caching_info *ci,
M
Mark Fasheh 已提交
1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745
			     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) {
1746 1747
			ocfs2_error(ocfs2_metadata_cache_get_super(ci),
				    "Owner %llu has empty extent list at "
M
Mark Fasheh 已提交
1748
				    "depth %u\n",
1749
				    (unsigned long long)ocfs2_metadata_cache_owner(ci),
M
Mark Fasheh 已提交
1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764
				    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) +
1765
				ocfs2_rec_clusters(el, rec);
M
Mark Fasheh 已提交
1766 1767 1768 1769 1770 1771
			if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
			    break;
		}

		blkno = le64_to_cpu(el->l_recs[i].e_blkno);
		if (blkno == 0) {
1772 1773
			ocfs2_error(ocfs2_metadata_cache_get_super(ci),
				    "Owner %llu has bad blkno in extent list "
M
Mark Fasheh 已提交
1774
				    "at depth %u (index %d)\n",
1775
				    (unsigned long long)ocfs2_metadata_cache_owner(ci),
M
Mark Fasheh 已提交
1776 1777 1778 1779 1780 1781 1782
				    le16_to_cpu(el->l_tree_depth), i);
			ret = -EROFS;
			goto out;
		}

		brelse(bh);
		bh = NULL;
1783
		ret = ocfs2_read_extent_block(ci, blkno, &bh);
M
Mark Fasheh 已提交
1784 1785 1786 1787 1788 1789 1790 1791 1792 1793
		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)) {
1794 1795
			ocfs2_error(ocfs2_metadata_cache_get_super(ci),
				    "Owner %llu has bad count in extent list "
M
Mark Fasheh 已提交
1796
				    "at block %llu (next free=%u, count=%u)\n",
1797
				    (unsigned long long)ocfs2_metadata_cache_owner(ci),
M
Mark Fasheh 已提交
1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840
				    (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++;
}
1841 1842
static int ocfs2_find_path(struct ocfs2_caching_info *ci,
			   struct ocfs2_path *path, u32 cpos)
M
Mark Fasheh 已提交
1843 1844 1845 1846 1847
{
	struct find_path_data data;

	data.index = 1;
	data.path = path;
1848
	return __ocfs2_find_path(ci, path_root_el(path), cpos,
M
Mark Fasheh 已提交
1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872
				 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.
 */
1873 1874 1875
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
1876 1877 1878 1879
{
	int ret;
	struct buffer_head *bh = NULL;

1880
	ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh);
M
Mark Fasheh 已提交
1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919
	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);
1920 1921
	if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) {
		BUG_ON(right_child_el->l_tree_depth);
1922 1923 1924
		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
1925
	left_clusters -= le32_to_cpu(left_rec->e_cpos);
1926
	left_rec->e_int_clusters = cpu_to_le32(left_clusters);
M
Mark Fasheh 已提交
1927 1928 1929

	/*
	 * Calculate the rightmost cluster count boundary before
1930
	 * moving cpos - we will need to adjust clusters after
M
Mark Fasheh 已提交
1931 1932 1933
	 * updating e_cpos to keep the same highest cluster count.
	 */
	right_end = le32_to_cpu(right_rec->e_cpos);
1934
	right_end += le32_to_cpu(right_rec->e_int_clusters);
M
Mark Fasheh 已提交
1935 1936 1937 1938 1939

	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);
1940
	right_rec->e_int_clusters = cpu_to_le32(right_end);
M
Mark Fasheh 已提交
1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982
}

/*
 * 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.
1983 1984
 *   - 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
1985
 */
1986
static void ocfs2_complete_edge_insert(handle_t *handle,
M
Mark Fasheh 已提交
1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063
				       struct ocfs2_path *left_path,
				       struct ocfs2_path *right_path,
				       int subtree_index)
{
	int ret, i, idx;
	struct ocfs2_extent_list *el, *left_el, *right_el;
	struct ocfs2_extent_rec *left_rec, *right_rec;
	struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;

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

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

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

		ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec,
					      right_el);

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

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

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

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

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

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

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

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

2064 2065
static int ocfs2_rotate_subtree_right(handle_t *handle,
				      struct ocfs2_extent_tree *et,
M
Mark Fasheh 已提交
2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080
				      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) {
2081
		ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
M
Mark Fasheh 已提交
2082 2083
			    "Inode %llu has non-full interior leaf node %llu"
			    "(next free = %u)",
2084
			    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
M
Mark Fasheh 已提交
2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099
			    (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);

2100
	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
2101
					   subtree_index);
M
Mark Fasheh 已提交
2102 2103 2104 2105 2106 2107
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
2108
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2109
						   right_path, i);
M
Mark Fasheh 已提交
2110 2111 2112 2113 2114
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2115
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2116
						   left_path, i);
M
Mark Fasheh 已提交
2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128
		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",
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 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162
			(unsigned long long)right_leaf_bh->b_blocknr);

	ocfs2_create_empty_extent(right_el);

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

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

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

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

2163 2164
	ocfs2_complete_edge_insert(handle, left_path, right_path,
				   subtree_index);
M
Mark Fasheh 已提交
2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182

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;

2183 2184
	BUG_ON(path->p_tree_depth == 0);

M
Mark Fasheh 已提交
2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218
	*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);
2219 2220 2221
				*cpos = *cpos + ocfs2_rec_clusters(el,
							   &el->l_recs[j - 1]);
				*cpos = *cpos - 1;
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
				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;
}

2245 2246 2247 2248 2249
/*
 * 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
2250
static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth,
2251
					   int op_credits,
M
Mark Fasheh 已提交
2252 2253
					   struct ocfs2_path *path)
{
2254
	int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits;
M
Mark Fasheh 已提交
2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287

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

	return 0;
}

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

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

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

2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310
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
2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326
/*
 * Rotate all the records in a btree right one record, starting at insert_cpos.
 *
 * The path to the rightmost leaf should be passed in.
 *
 * The array is assumed to be large enough to hold an entire path (tree depth).
 *
 * Upon succesful return from this function:
 *
 * - The 'right_path' array will contain a path to the leaf block
 *   whose range contains e_cpos.
 * - That leaf block will have a single empty extent in list index 0.
 * - In the case that the rotation requires a post-insert update,
 *   *ret_left_path will contain a valid path which can be passed to
 *   ocfs2_insert_path().
 */
2327
static int ocfs2_rotate_tree_right(handle_t *handle,
2328
				   struct ocfs2_extent_tree *et,
2329
				   enum ocfs2_split_type split,
M
Mark Fasheh 已提交
2330 2331 2332 2333
				   u32 insert_cpos,
				   struct ocfs2_path *right_path,
				   struct ocfs2_path **ret_left_path)
{
2334
	int ret, start, orig_credits = handle->h_buffer_credits;
M
Mark Fasheh 已提交
2335 2336
	u32 cpos;
	struct ocfs2_path *left_path = NULL;
2337
	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
M
Mark Fasheh 已提交
2338 2339 2340

	*ret_left_path = NULL;

2341
	left_path = ocfs2_new_path_from_path(right_path);
M
Mark Fasheh 已提交
2342 2343 2344 2345 2346 2347
	if (!left_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

2348
	ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
M
Mark Fasheh 已提交
2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385
	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);

2386
		ret = ocfs2_find_path(et->et_ci, left_path, cpos);
M
Mark Fasheh 已提交
2387 2388 2389 2390 2391 2392 2393
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		mlog_bug_on_msg(path_leaf_bh(left_path) ==
				path_leaf_bh(right_path),
2394
				"Owner %llu: error during insert of %u "
M
Mark Fasheh 已提交
2395 2396
				"(left path cpos %u) results in two identical "
				"paths ending at %llu\n",
2397 2398
				(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
				insert_cpos, cpos,
M
Mark Fasheh 已提交
2399 2400 2401
				(unsigned long long)
				path_leaf_bh(left_path)->b_blocknr);

2402 2403
		if (split == SPLIT_NONE &&
		    ocfs2_rotate_requires_path_adjustment(left_path,
M
Mark Fasheh 已提交
2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423
							  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;
		}

2424
		start = ocfs2_find_subtree_root(et, left_path, right_path);
M
Mark Fasheh 已提交
2425 2426 2427 2428 2429 2430 2431

		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,
2432
						      orig_credits, right_path);
M
Mark Fasheh 已提交
2433 2434 2435 2436 2437
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2438
		ret = ocfs2_rotate_subtree_right(handle, et, left_path,
M
Mark Fasheh 已提交
2439 2440 2441 2442 2443 2444
						 right_path, start);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462
		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
2463 2464 2465 2466 2467 2468 2469
		/*
		 * 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);

2470
		ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
M
Mark Fasheh 已提交
2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

out:
	ocfs2_free_path(left_path);

out_ret_path:
	return ret;
}

2484 2485
static int ocfs2_update_edge_lengths(handle_t *handle,
				     struct ocfs2_extent_tree *et,
2486
				     int subtree_index, struct ocfs2_path *path)
M
Mark Fasheh 已提交
2487
{
2488
	int i, idx, ret;
M
Mark Fasheh 已提交
2489
	struct ocfs2_extent_rec *rec;
2490 2491 2492
	struct ocfs2_extent_list *el;
	struct ocfs2_extent_block *eb;
	u32 range;
M
Mark Fasheh 已提交
2493

2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509
	/*
	 * 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.
	 */
	ret = ocfs2_extend_trans(handle,
				 handle->h_buffer_credits + subtree_index);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

2510
	ret = ocfs2_journal_access_path(et->et_ci, handle, path);
2511 2512 2513 2514 2515
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

2516 2517 2518
	/* 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
2519

2520 2521 2522 2523 2524
	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
2525

2526 2527 2528 2529
	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
2530

2531 2532
		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
2533

2534
		ocfs2_journal_dirty(handle, path->p_node[i].bh);
M
Mark Fasheh 已提交
2535
	}
2536 2537
out:
	return ret;
M
Mark Fasheh 已提交
2538 2539
}

2540 2541
static void ocfs2_unlink_path(handle_t *handle,
			      struct ocfs2_extent_tree *et,
2542 2543
			      struct ocfs2_cached_dealloc_ctxt *dealloc,
			      struct ocfs2_path *path, int unlink_start)
M
Mark Fasheh 已提交
2544
{
2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562
	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",
2563
			     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2564 2565 2566 2567
			     (unsigned long long)le64_to_cpu(eb->h_blkno),
			     le16_to_cpu(el->l_next_free_rec));

			ocfs2_journal_dirty(handle, bh);
2568
			ocfs2_remove_from_cache(et->et_ci, bh);
2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580
			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);

2581
		ocfs2_remove_from_cache(et->et_ci, bh);
2582
	}
M
Mark Fasheh 已提交
2583 2584
}

2585 2586
static void ocfs2_unlink_subtree(handle_t *handle,
				 struct ocfs2_extent_tree *et,
2587 2588 2589 2590
				 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
2591
{
2592 2593 2594
	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
2595
	struct ocfs2_extent_list *el;
2596
	struct ocfs2_extent_block *eb;
M
Mark Fasheh 已提交
2597

2598
	el = path_leaf_el(left_path);
M
Mark Fasheh 已提交
2599

2600
	eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data;
2601

2602 2603 2604
	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
2605

2606
	BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec));
M
Mark Fasheh 已提交
2607

2608 2609 2610 2611 2612 2613 2614 2615 2616
	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));

2617
	ocfs2_unlink_path(handle, et, dealloc, right_path,
2618 2619 2620
			  subtree_index + 1);
}

2621 2622
static int ocfs2_rotate_subtree_left(handle_t *handle,
				     struct ocfs2_extent_tree *et,
2623 2624 2625 2626
				     struct ocfs2_path *left_path,
				     struct ocfs2_path *right_path,
				     int subtree_index,
				     struct ocfs2_cached_dealloc_ctxt *dealloc,
2627
				     int *deleted)
2628 2629
{
	int ret, i, del_right_subtree = 0, right_has_empty = 0;
2630
	struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path);
2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642
	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
2643

2644 2645
	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
2646
		/*
2647 2648 2649 2650 2651 2652 2653 2654 2655 2656
		 * 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
2657 2658
		 */

2659 2660 2661 2662
		if (eb->h_next_leaf_blk != 0ULL)
			return -EAGAIN;

		if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) {
2663
			ret = ocfs2_journal_access_eb(handle, et->et_ci,
2664 2665
						      path_leaf_bh(right_path),
						      OCFS2_JOURNAL_ACCESS_WRITE);
M
Mark Fasheh 已提交
2666 2667 2668 2669 2670
			if (ret) {
				mlog_errno(ret);
				goto out;
			}

2671 2672 2673
			ocfs2_remove_empty_extent(right_leaf_el);
		} else
			right_has_empty = 1;
M
Mark Fasheh 已提交
2674 2675
	}

2676 2677 2678 2679 2680 2681
	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.
		 */
2682
		ret = ocfs2_et_root_journal_access(handle, et,
2683
						   OCFS2_JOURNAL_ACCESS_WRITE);
2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697
		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);

2698
	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
2699
					   subtree_index);
2700 2701 2702 2703 2704 2705
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
2706
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2707
						   right_path, i);
2708 2709 2710 2711 2712
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2713
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2714
						   left_path, i);
2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

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

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

	if (del_right_subtree) {
2750
		ocfs2_unlink_subtree(handle, et, left_path, right_path,
2751
				     subtree_index, dealloc);
2752
		ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
2753 2754 2755 2756 2757
						left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
2758 2759

		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
2760
		ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
2761 2762 2763 2764 2765 2766 2767 2768 2769

		/*
		 * 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);

2770
		ret = ocfs2_journal_dirty(handle, et_root_bh);
2771 2772 2773 2774 2775
		if (ret)
			mlog_errno(ret);

		*deleted = 1;
	} else
2776
		ocfs2_complete_edge_insert(handle, left_path, right_path,
2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 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
					   subtree_index);

out:
	return ret;
}

/*
 * Given a full path, determine what cpos value would return us a path
 * containing the leaf immediately to the right of the current one.
 *
 * Will return zero if the path passed in is already the rightmost path.
 *
 * This looks similar, but is subtly different to
 * ocfs2_find_cpos_for_left_leaf().
 */
static int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
					  struct ocfs2_path *path, u32 *cpos)
{
	int i, j, ret = 0;
	u64 blkno;
	struct ocfs2_extent_list *el;

	*cpos = 0;

	if (path->p_tree_depth == 0)
		return 0;

	blkno = path_leaf_bh(path)->b_blocknr;

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

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

		/*
		 * Find the extent record just after the one in our
		 * path.
		 */
		next_free = le16_to_cpu(el->l_next_free_rec);
		for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
			if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
				if (j == (next_free - 1)) {
					if (i == 0) {
						/*
						 * We've determined that the
						 * path specified is already
						 * the rightmost one - return a
						 * cpos of zero.
						 */
						goto out;
					}
					/*
					 * The rightmost record points to our
					 * leaf - we need to travel up the
					 * tree one level.
					 */
					goto next_node;
				}

				*cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos);
				goto out;
			}
		}

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

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

out:
	return ret;
}

2862 2863
static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle,
					    struct ocfs2_extent_tree *et,
2864
					    struct ocfs2_path *path)
2865 2866
{
	int ret;
2867 2868
	struct buffer_head *bh = path_leaf_bh(path);
	struct ocfs2_extent_list *el = path_leaf_el(path);
2869 2870 2871 2872

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

2873
	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
2874
					   path_num_items(path) - 1);
2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	ocfs2_remove_empty_extent(el);

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

out:
	return ret;
}

2890 2891 2892
static int __ocfs2_rotate_tree_left(handle_t *handle,
				    struct ocfs2_extent_tree *et,
				    int orig_credits,
2893 2894
				    struct ocfs2_path *path,
				    struct ocfs2_cached_dealloc_ctxt *dealloc,
2895
				    struct ocfs2_path **empty_extent_path)
2896 2897 2898 2899 2900
{
	int ret, subtree_root, deleted;
	u32 right_cpos;
	struct ocfs2_path *left_path = NULL;
	struct ocfs2_path *right_path = NULL;
2901
	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
2902 2903 2904 2905 2906

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

	*empty_extent_path = NULL;

2907
	ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
2908 2909 2910 2911 2912
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

2913
	left_path = ocfs2_new_path_from_path(path);
2914 2915 2916 2917 2918 2919 2920 2921
	if (!left_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

	ocfs2_cp_path(left_path, path);

2922
	right_path = ocfs2_new_path_from_path(path);
2923 2924 2925 2926 2927 2928 2929
	if (!right_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

	while (right_cpos) {
2930
		ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
2931 2932 2933 2934 2935
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2936
		subtree_root = ocfs2_find_subtree_root(et, left_path,
2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951
						       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;
		}

2952 2953 2954 2955
		/*
		 * Caller might still want to make changes to the
		 * tree root, so re-add it to the journal here.
		 */
2956
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2957
						   left_path, 0);
2958 2959 2960 2961 2962
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

2963
		ret = ocfs2_rotate_subtree_left(handle, et, left_path,
2964
						right_path, subtree_root,
2965
						dealloc, &deleted);
2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991
		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);

2992
		ret = ocfs2_find_cpos_for_right_leaf(sb, left_path,
2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006
						     &right_cpos);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

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

	return ret;
}

3007 3008
static int ocfs2_remove_rightmost_path(handle_t *handle,
				struct ocfs2_extent_tree *et,
3009
				struct ocfs2_path *path,
3010
				struct ocfs2_cached_dealloc_ctxt *dealloc)
3011 3012 3013 3014 3015 3016 3017 3018
{
	int ret, subtree_index;
	u32 cpos;
	struct ocfs2_path *left_path = NULL;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *el;


3019
	ret = ocfs2_et_sanity_check(et);
3020 3021
	if (ret)
		goto out;
3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033
	/*
	 * 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;
	}

3034
	ret = ocfs2_journal_access_path(et->et_ci, handle, path);
3035 3036 3037 3038 3039
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

3040 3041
	ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
					    path, &cpos);
3042 3043 3044 3045 3046 3047 3048 3049 3050 3051
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	if (cpos) {
		/*
		 * We have a path to the left of this one - it needs
		 * an update too.
		 */
3052
		left_path = ocfs2_new_path_from_path(path);
3053 3054 3055 3056 3057 3058
		if (!left_path) {
			ret = -ENOMEM;
			mlog_errno(ret);
			goto out;
		}

3059
		ret = ocfs2_find_path(et->et_ci, left_path, cpos);
3060 3061 3062 3063 3064
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

3065
		ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
3066 3067 3068 3069 3070
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

3071
		subtree_index = ocfs2_find_subtree_root(et, left_path, path);
3072

3073
		ocfs2_unlink_subtree(handle, et, left_path, path,
3074
				     subtree_index, dealloc);
3075
		ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
3076 3077 3078 3079 3080
						left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
3081 3082

		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
3083
		ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
3084 3085 3086 3087 3088
	} else {
		/*
		 * 'path' is also the leftmost path which
		 * means it must be the only one. This gets
		 * handled differently because we want to
3089
		 * revert the root back to having extents
3090 3091
		 * in-line.
		 */
3092
		ocfs2_unlink_path(handle, et, dealloc, path, 1);
3093

3094
		el = et->et_root_el;
3095 3096 3097 3098
		el->l_tree_depth = 0;
		el->l_next_free_rec = 0;
		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));

3099
		ocfs2_et_set_last_eb_blk(et, 0);
3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124
	}

	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.
 */
3125 3126
static int ocfs2_rotate_tree_left(handle_t *handle,
				  struct ocfs2_extent_tree *et,
3127
				  struct ocfs2_path *path,
3128
				  struct ocfs2_cached_dealloc_ctxt *dealloc)
3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141
{
	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:
		/*
3142
		 * Inline extents. This is trivially handled, so do
3143 3144
		 * it up front.
		 */
3145
		ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path);
3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160
		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
3161
	 *  2b) we need to bring the root back to inline extents.
3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176
	 */

	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;
3177 3178 3179
			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),
3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192
				    (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.
		 */

3193 3194
		ret = ocfs2_remove_rightmost_path(handle, et, path,
						  dealloc);
3195 3196 3197 3198 3199 3200 3201 3202 3203 3204
		if (ret)
			mlog_errno(ret);
		goto out;
	}

	/*
	 * Now we can loop, remembering the path we get from -EAGAIN
	 * and restarting from there.
	 */
try_rotate:
3205 3206
	ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path,
				       dealloc, &restart_path);
3207 3208 3209 3210 3211 3212 3213 3214 3215
	if (ret && ret != -EAGAIN) {
		mlog_errno(ret);
		goto out;
	}

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

3216
		ret = __ocfs2_rotate_tree_left(handle, et, orig_credits,
3217
					       tmp_path, dealloc,
3218
					       &restart_path);
3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268
		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));
	}
}

3269
static int ocfs2_get_right_path(struct ocfs2_extent_tree *et,
3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285
				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);

3286 3287
	ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
					     left_path, &right_cpos);
3288 3289 3290 3291 3292 3293 3294 3295
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

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

3296
	right_path = ocfs2_new_path_from_path(left_path);
3297 3298 3299 3300 3301 3302
	if (!right_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

3303
	ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

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

3316 3317
/*
 * Remove split_rec clusters from the record at index and merge them
3318 3319 3320 3321
 * 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.
3322
 */
3323
static int ocfs2_merge_rec_right(struct ocfs2_path *left_path,
3324
				 handle_t *handle,
3325
				 struct ocfs2_extent_tree *et,
3326 3327
				 struct ocfs2_extent_rec *split_rec,
				 int index)
3328
{
3329
	int ret, next_free, i;
3330 3331 3332
	unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
	struct ocfs2_extent_rec *left_rec;
	struct ocfs2_extent_rec *right_rec;
3333 3334 3335 3336 3337 3338
	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;
3339 3340 3341

	BUG_ON(index >= le16_to_cpu(el->l_next_free_rec));
	left_rec = &el->l_recs[index];
3342

A
Al Viro 已提交
3343
	if (index == le16_to_cpu(el->l_next_free_rec) - 1 &&
3344 3345
	    le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) {
		/* we meet with a cross extent block merge. */
3346
		ret = ocfs2_get_right_path(et, left_path, &right_path);
3347 3348 3349 3350 3351 3352 3353 3354 3355 3356
		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
3357
			BUG_ON(next_free <= 1);
3358 3359 3360 3361 3362 3363 3364
			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));

3365 3366
		subtree_index = ocfs2_find_subtree_root(et, left_path,
							right_path);
3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378

		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);

3379
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
3380
						   subtree_index);
3381 3382 3383 3384 3385 3386 3387
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		for (i = subtree_index + 1;
		     i < path_num_items(right_path); i++) {
3388
			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
3389
							   right_path, i);
3390 3391 3392 3393 3394
			if (ret) {
				mlog_errno(ret);
				goto out;
			}

3395
			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
3396
							   left_path, i);
3397 3398 3399 3400 3401 3402 3403 3404 3405 3406
			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];
	}
3407

3408
	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path,
3409
					   path_num_items(left_path) - 1);
3410 3411 3412 3413 3414 3415 3416 3417 3418
	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,
3419 3420
		     -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
					       split_clusters));
3421 3422 3423 3424 3425 3426 3427 3428
	le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters);

	ocfs2_cleanup_merge(el, index);

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

3429 3430 3431 3432 3433
	if (right_path) {
		ret = ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
		if (ret)
			mlog_errno(ret);

3434 3435
		ocfs2_complete_edge_insert(handle, left_path, right_path,
					   subtree_index);
3436 3437 3438 3439 3440 3441 3442
	}
out:
	if (right_path)
		ocfs2_free_path(right_path);
	return ret;
}

3443
static int ocfs2_get_left_path(struct ocfs2_extent_tree *et,
3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455
			       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);

3456
	ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3457 3458 3459 3460 3461 3462 3463 3464 3465
					    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);

3466
	left_path = ocfs2_new_path_from_path(right_path);
3467 3468 3469 3470 3471 3472
	if (!left_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

3473
	ret = ocfs2_find_path(et->et_ci, left_path, left_cpos);
3474 3475 3476 3477 3478 3479
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	*ret_left_path = left_path;
3480
out:
3481 3482
	if (ret)
		ocfs2_free_path(left_path);
3483 3484 3485 3486 3487
	return ret;
}

/*
 * Remove split_rec clusters from the record at index and merge them
3488 3489 3490 3491 3492 3493 3494
 * 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.
3495
 */
3496
static int ocfs2_merge_rec_left(struct ocfs2_path *right_path,
3497
				handle_t *handle,
3498
				struct ocfs2_extent_tree *et,
3499
				struct ocfs2_extent_rec *split_rec,
3500 3501
				struct ocfs2_cached_dealloc_ctxt *dealloc,
				int index)
3502
{
3503
	int ret, i, subtree_index = 0, has_empty_extent = 0;
3504 3505 3506
	unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
	struct ocfs2_extent_rec *left_rec;
	struct ocfs2_extent_rec *right_rec;
3507 3508 3509 3510 3511
	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;
3512

3513
	BUG_ON(index < 0);
3514 3515

	right_rec = &el->l_recs[index];
3516 3517
	if (index == 0) {
		/* we meet with a cross extent block merge. */
3518
		ret = ocfs2_get_left_path(et, right_path, &left_path);
3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533
		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));

3534 3535
		subtree_index = ocfs2_find_subtree_root(et, left_path,
							right_path);
3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547

		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);

3548
		ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
3549
						   subtree_index);
3550 3551 3552 3553 3554 3555 3556
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		for (i = subtree_index + 1;
		     i < path_num_items(right_path); i++) {
3557
			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
3558
							   right_path, i);
3559 3560 3561 3562 3563
			if (ret) {
				mlog_errno(ret);
				goto out;
			}

3564
			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
3565
							   left_path, i);
3566 3567 3568 3569 3570 3571 3572 3573 3574 3575
			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;
	}
3576

3577
	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
3578
					   path_num_items(right_path) - 1);
3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590
	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;
3591
	} else
3592 3593 3594 3595
		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,
3596 3597
		     ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
					      split_clusters));
3598 3599 3600 3601 3602 3603 3604 3605
	le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters);

	ocfs2_cleanup_merge(el, index);

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

3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618
	if (left_path) {
		ret = ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
		if (ret)
			mlog_errno(ret);

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

3619
			ret = ocfs2_remove_rightmost_path(handle, et,
3620
							  right_path,
3621
							  dealloc);
3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632
			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
3633
			ocfs2_complete_edge_insert(handle, left_path,
3634 3635
						   right_path, subtree_index);
	}
3636
out:
3637 3638
	if (left_path)
		ocfs2_free_path(left_path);
3639 3640 3641
	return ret;
}

3642 3643
static int ocfs2_try_to_merge_extent(handle_t *handle,
				     struct ocfs2_extent_tree *et,
3644
				     struct ocfs2_path *path,
3645 3646 3647
				     int split_index,
				     struct ocfs2_extent_rec *split_rec,
				     struct ocfs2_cached_dealloc_ctxt *dealloc,
3648
				     struct ocfs2_merge_ctxt *ctxt)
3649
{
T
Tao Mao 已提交
3650
	int ret = 0;
3651
	struct ocfs2_extent_list *el = path_leaf_el(path);
3652 3653 3654 3655
	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];

	BUG_ON(ctxt->c_contig_type == CONTIG_NONE);

T
Tao Mao 已提交
3656 3657 3658 3659 3660 3661 3662 3663
	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.
		 */
3664
		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
T
Tao Mao 已提交
3665 3666 3667
		if (ret) {
			mlog_errno(ret);
			goto out;
3668
		}
T
Tao Mao 已提交
3669 3670
		split_index--;
		rec = &el->l_recs[split_index];
3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687
	}

	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.
3688 3689 3690 3691
		 *
		 * 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.
3692
		 */
3693
		ret = ocfs2_merge_rec_right(path, handle, et, split_rec,
3694
					    split_index);
3695 3696 3697 3698 3699 3700 3701 3702 3703 3704
		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]));

3705
		/* The merge left us with an empty extent, remove it. */
3706
		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
3707 3708 3709 3710
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
3711

3712 3713 3714 3715
		rec = &el->l_recs[split_index];

		/*
		 * Note that we don't pass split_rec here on purpose -
3716
		 * we've merged it into the rec already.
3717
		 */
3718 3719
		ret = ocfs2_merge_rec_left(path, handle, et, rec,
					   dealloc, split_index);
3720

3721 3722 3723 3724 3725
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

3726
		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742
		/*
		 * 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) {
3743 3744
			ret = ocfs2_merge_rec_left(path, handle, et,
						   split_rec, dealloc,
3745 3746 3747 3748 3749 3750
						   split_index);
			if (ret) {
				mlog_errno(ret);
				goto out;
			}
		} else {
3751
			ret = ocfs2_merge_rec_right(path, handle,
3752
						    et, split_rec,
3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764
						    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.
			 */
3765 3766
			ret = ocfs2_rotate_tree_left(handle, et, path,
						     dealloc);
3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 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
			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.
 */
3812 3813
static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et,
				 struct ocfs2_extent_rec *insert_rec,
3814
				 struct ocfs2_extent_list *el,
3815
				 struct ocfs2_insert_type *insert)
3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826
{
	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];
3827 3828
		ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
					insert->ins_split, rec,
3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869
					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),
3870
				"owner %llu, depth %u, count %u, next free %u, "
3871 3872
				"rec.cpos %u, rec.clusters %u, "
				"insert.cpos %u, insert.clusters %u\n",
3873
				ocfs2_metadata_cache_owner(et->et_ci),
3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900
				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);
}

3901 3902
static void ocfs2_adjust_rightmost_records(handle_t *handle,
					   struct ocfs2_extent_tree *et,
3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917
					   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
3918 3919
		next_free = le16_to_cpu(el->l_next_free_rec);
		if (next_free == 0) {
3920 3921 3922
			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
3923
			ret = -EIO;
3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941
			return;
		}

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

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

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

	}
}

3942 3943
static int ocfs2_append_rec_to_path(handle_t *handle,
				    struct ocfs2_extent_tree *et,
3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970
				    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;

3971 3972
		ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
						    right_path, &left_cpos);
3973 3974
		if (ret) {
			mlog_errno(ret);
M
Mark Fasheh 已提交
3975 3976 3977
			goto out;
		}

3978 3979 3980
		mlog(0, "Append may need a left path update. cpos: %u, "
		     "left_cpos: %u\n", le32_to_cpu(insert_rec->e_cpos),
		     left_cpos);
3981

3982 3983 3984 3985 3986
		/*
		 * No need to worry if the append is already in the
		 * leftmost leaf.
		 */
		if (left_cpos) {
3987
			left_path = ocfs2_new_path_from_path(right_path);
3988 3989 3990 3991 3992
			if (!left_path) {
				ret = -ENOMEM;
				mlog_errno(ret);
				goto out;
			}
M
Mark Fasheh 已提交
3993

3994
			ret = ocfs2_find_path(et->et_ci, left_path,
3995
					      left_cpos);
3996 3997 3998 3999
			if (ret) {
				mlog_errno(ret);
				goto out;
			}
M
Mark Fasheh 已提交
4000

4001 4002 4003 4004 4005 4006
			/*
			 * ocfs2_insert_path() will pass the left_path to the
			 * journal for us.
			 */
		}
	}
M
Mark Fasheh 已提交
4007

4008
	ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
4009 4010 4011
	if (ret) {
		mlog_errno(ret);
		goto out;
M
Mark Fasheh 已提交
4012 4013
	}

4014
	ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec);
4015

M
Mark Fasheh 已提交
4016 4017 4018 4019 4020 4021 4022 4023 4024
	*ret_left_path = left_path;
	ret = 0;
out:
	if (ret != 0)
		ocfs2_free_path(left_path);

	return ret;
}

4025
static void ocfs2_split_record(struct ocfs2_extent_tree *et,
4026 4027 4028 4029 4030 4031 4032 4033 4034 4035
			       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;

4036
	right_el = path_leaf_el(right_path);
4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 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
	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];
4098 4099
	ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
				split, rec, split_rec);
4100 4101 4102
	ocfs2_rotate_leaf(insert_el, split_rec);
}

M
Mark Fasheh 已提交
4103
/*
4104 4105
 * 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
4106 4107 4108 4109 4110
 *
 * 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.
 */
4111
static int ocfs2_insert_path(handle_t *handle,
4112
			     struct ocfs2_extent_tree *et,
M
Mark Fasheh 已提交
4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137
			     struct ocfs2_path *left_path,
			     struct ocfs2_path *right_path,
			     struct ocfs2_extent_rec *insert_rec,
			     struct ocfs2_insert_type *insert)
{
	int ret, subtree_index;
	struct buffer_head *leaf_bh = path_leaf_bh(right_path);

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

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

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

4138
		ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
M
Mark Fasheh 已提交
4139 4140 4141 4142 4143 4144
		if (ret < 0) {
			mlog_errno(ret);
			goto out;
		}
	}

4145 4146 4147 4148
	/*
	 * Pass both paths to the journal. The majority of inserts
	 * will be touching all components anyway.
	 */
4149
	ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
4150 4151 4152 4153 4154
	if (ret < 0) {
		mlog_errno(ret);
		goto out;
	}

4155 4156 4157
	if (insert->ins_split != SPLIT_NONE) {
		/*
		 * We could call ocfs2_insert_at_leaf() for some types
J
Joe Perches 已提交
4158
		 * of splits, but it's easier to just let one separate
4159 4160
		 * function sort it all out.
		 */
4161
		ocfs2_split_record(et, left_path, right_path,
4162
				   insert_rec, insert->ins_split);
4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173

		/*
		 * Split might have modified either leaf and we don't
		 * have a guarantee that the later edge insert will
		 * dirty this for us.
		 */
		if (left_path)
			ret = ocfs2_journal_dirty(handle,
						  path_leaf_bh(left_path));
			if (ret)
				mlog_errno(ret);
4174
	} else
4175 4176
		ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path),
				     insert);
M
Mark Fasheh 已提交
4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188

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

	if (left_path) {
		/*
		 * The rotate code has indicated that we need to fix
		 * up portions of the tree after the insert.
		 *
		 * XXX: Should we extend the transaction here?
		 */
4189
		subtree_index = ocfs2_find_subtree_root(et, left_path,
M
Mark Fasheh 已提交
4190
							right_path);
4191 4192
		ocfs2_complete_edge_insert(handle, left_path, right_path,
					   subtree_index);
M
Mark Fasheh 已提交
4193 4194 4195 4196 4197 4198 4199
	}

	ret = 0;
out:
	return ret;
}

4200
static int ocfs2_do_insert_extent(handle_t *handle,
4201
				  struct ocfs2_extent_tree *et,
M
Mark Fasheh 已提交
4202 4203 4204 4205 4206 4207 4208 4209 4210
				  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;

4211
	el = et->et_root_el;
M
Mark Fasheh 已提交
4212

4213
	ret = ocfs2_et_root_journal_access(handle, et,
4214
					   OCFS2_JOURNAL_ACCESS_WRITE);
M
Mark Fasheh 已提交
4215 4216 4217 4218 4219 4220
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

	if (le16_to_cpu(el->l_tree_depth) == 0) {
4221
		ocfs2_insert_at_leaf(et, insert_rec, el, type);
M
Mark Fasheh 已提交
4222 4223 4224
		goto out_update_clusters;
	}

4225
	right_path = ocfs2_new_path_from_et(et);
M
Mark Fasheh 已提交
4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243
	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;
	}

4244
	ret = ocfs2_find_path(et->et_ci, right_path, cpos);
M
Mark Fasheh 已提交
4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262
	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) {
4263
		ret = ocfs2_rotate_tree_right(handle, et, type->ins_split,
M
Mark Fasheh 已提交
4264 4265 4266 4267 4268 4269
					      le32_to_cpu(insert_rec->e_cpos),
					      right_path, &left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
4270 4271 4272 4273 4274

		/*
		 * ocfs2_rotate_tree_right() might have extended the
		 * transaction without re-journaling our tree root.
		 */
4275
		ret = ocfs2_et_root_journal_access(handle, et,
4276
						   OCFS2_JOURNAL_ACCESS_WRITE);
4277 4278 4279 4280
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
M
Mark Fasheh 已提交
4281 4282
	} else if (type->ins_appending == APPEND_TAIL
		   && type->ins_contig != CONTIG_LEFT) {
4283
		ret = ocfs2_append_rec_to_path(handle, et, insert_rec,
M
Mark Fasheh 已提交
4284 4285 4286 4287 4288 4289 4290
					       right_path, &left_path);
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

4291
	ret = ocfs2_insert_path(handle, et, left_path, right_path,
M
Mark Fasheh 已提交
4292 4293 4294 4295 4296 4297 4298
				insert_rec, type);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

out_update_clusters:
4299
	if (type->ins_split == SPLIT_NONE)
4300
		ocfs2_et_update_clusters(et,
4301
					 le16_to_cpu(insert_rec->e_leaf_clusters));
M
Mark Fasheh 已提交
4302

4303
	ret = ocfs2_journal_dirty(handle, et->et_root_bh);
M
Mark Fasheh 已提交
4304 4305 4306 4307 4308 4309 4310 4311 4312 4313
	if (ret)
		mlog_errno(ret);

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

	return ret;
}

4314
static enum ocfs2_contig_type
4315 4316
ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et,
			       struct ocfs2_path *path,
4317 4318 4319
			       struct ocfs2_extent_list *el, int index,
			       struct ocfs2_extent_rec *split_rec)
{
T
Tao Ma 已提交
4320
	int status;
4321
	enum ocfs2_contig_type ret = CONTIG_NONE;
T
Tao Ma 已提交
4322 4323 4324 4325 4326 4327
	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;
4328
	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
T
Tao Ma 已提交
4329 4330 4331 4332

	if (index > 0) {
		rec = &el->l_recs[index - 1];
	} else if (path->p_tree_depth > 0) {
4333
		status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
T
Tao Ma 已提交
4334 4335 4336 4337
		if (status)
			goto out;

		if (left_cpos != 0) {
4338
			left_path = ocfs2_new_path_from_path(path);
T
Tao Ma 已提交
4339 4340 4341
			if (!left_path)
				goto out;

4342 4343
			status = ocfs2_find_path(et->et_ci, left_path,
						 left_cpos);
T
Tao Ma 已提交
4344 4345 4346 4347 4348 4349 4350 4351 4352
			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;
4353
				ocfs2_error(sb,
4354 4355 4356 4357 4358 4359 4360 4361
					    "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
4362 4363 4364 4365 4366 4367
				goto out;
			}
			rec = &new_el->l_recs[
				le16_to_cpu(new_el->l_next_free_rec) - 1];
		}
	}
4368 4369 4370 4371 4372

	/*
	 * 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
4373
	if (rec) {
4374 4375 4376 4377
		if (index == 1 && ocfs2_is_empty_extent(rec)) {
			if (split_rec->e_cpos == el->l_recs[index].e_cpos)
				ret = CONTIG_RIGHT;
		} else {
4378
			ret = ocfs2_extent_contig(sb, rec, split_rec);
4379 4380 4381
		}
	}

T
Tao Ma 已提交
4382 4383 4384 4385 4386
	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) {
4387
		status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
T
Tao Ma 已提交
4388 4389 4390 4391 4392 4393
		if (status)
			goto out;

		if (right_cpos == 0)
			goto out;

4394
		right_path = ocfs2_new_path_from_path(path);
T
Tao Ma 已提交
4395 4396 4397
		if (!right_path)
			goto out;

4398
		status = ocfs2_find_path(et->et_ci, right_path, right_cpos);
T
Tao Ma 已提交
4399 4400 4401 4402 4403 4404 4405 4406 4407
		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;
4408
				ocfs2_error(sb,
4409 4410 4411 4412 4413
					    "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
4414 4415 4416 4417 4418 4419 4420
				goto out;
			}
			rec = &new_el->l_recs[1];
		}
	}

	if (rec) {
4421 4422
		enum ocfs2_contig_type contig_type;

4423
		contig_type = ocfs2_extent_contig(sb, rec, split_rec);
4424 4425 4426 4427 4428 4429 4430

		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
4431 4432 4433 4434 4435 4436
out:
	if (left_path)
		ocfs2_free_path(left_path);
	if (right_path)
		ocfs2_free_path(right_path);

4437 4438 4439
	return ret;
}

M
Mark Fasheh 已提交
4440 4441 4442
static void ocfs2_figure_contig_type(struct inode *inode,
				     struct ocfs2_insert_type *insert,
				     struct ocfs2_extent_list *el,
4443 4444
				     struct ocfs2_extent_rec *insert_rec,
				     struct ocfs2_extent_tree *et)
M
Mark Fasheh 已提交
4445 4446 4447 4448
{
	int i;
	enum ocfs2_contig_type contig_type = CONTIG_NONE;

4449 4450
	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);

M
Mark Fasheh 已提交
4451
	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
4452
		contig_type = ocfs2_extent_contig(inode->i_sb, &el->l_recs[i],
M
Mark Fasheh 已提交
4453 4454 4455 4456 4457 4458 4459
						  insert_rec);
		if (contig_type != CONTIG_NONE) {
			insert->ins_contig_index = i;
			break;
		}
	}
	insert->ins_contig = contig_type;
4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470

	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.
		 */
4471 4472
		if (et->et_max_leaf_clusters &&
		    (len > et->et_max_leaf_clusters))
4473 4474
			insert->ins_contig = CONTIG_NONE;
	}
M
Mark Fasheh 已提交
4475 4476 4477 4478 4479 4480 4481 4482
}

/*
 * 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.
 *
4483 4484
 * 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
4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496
 * 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;

4497
	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
M
Mark Fasheh 已提交
4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510

	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];

4511 4512
	if (cpos >=
	    (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
M
Mark Fasheh 已提交
4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534
		goto set_tail_append;

	return;

set_tail_append:
	insert->ins_appending = APPEND_TAIL;
}

/*
 * Helper function called at the begining of an insert.
 *
 * This computes a few things that are commonly used in the process of
 * inserting into the btree:
 *   - Whether the new extent is contiguous with an existing one.
 *   - The current tree depth.
 *   - Whether the insert is an appending one.
 *   - The total # of free records in the tree.
 *
 * All of the information is stored on the ocfs2_insert_type
 * structure.
 */
static int ocfs2_figure_insert_type(struct inode *inode,
4535
				    struct ocfs2_extent_tree *et,
M
Mark Fasheh 已提交
4536 4537
				    struct buffer_head **last_eb_bh,
				    struct ocfs2_extent_rec *insert_rec,
4538
				    int *free_records,
M
Mark Fasheh 已提交
4539 4540 4541 4542 4543 4544 4545 4546
				    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;

4547 4548
	insert->ins_split = SPLIT_NONE;

4549
	el = et->et_root_el;
M
Mark Fasheh 已提交
4550 4551 4552 4553 4554 4555 4556 4557 4558
	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.
		 */
4559
		ret = ocfs2_read_extent_block(et->et_ci,
4560 4561
					      ocfs2_et_get_last_eb_blk(et),
					      &bh);
M
Mark Fasheh 已提交
4562 4563 4564 4565
		if (ret) {
			mlog_exit(ret);
			goto out;
		}
4566 4567
		eb = (struct ocfs2_extent_block *) bh->b_data;
		el = &eb->h_list;
M
Mark Fasheh 已提交
4568
	}
4569

M
Mark Fasheh 已提交
4570 4571 4572 4573 4574 4575 4576 4577
	/*
	 * 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.
	 */
4578
	*free_records = le16_to_cpu(el->l_count) -
M
Mark Fasheh 已提交
4579 4580 4581
		le16_to_cpu(el->l_next_free_rec);

	if (!insert->ins_tree_depth) {
4582
		ocfs2_figure_contig_type(inode, insert, el, insert_rec, et);
M
Mark Fasheh 已提交
4583 4584
		ocfs2_figure_appending_type(insert, el, insert_rec);
		return 0;
4585 4586
	}

4587
	path = ocfs2_new_path_from_et(et);
M
Mark Fasheh 已提交
4588 4589 4590 4591 4592
	if (!path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}
4593

M
Mark Fasheh 已提交
4594 4595 4596 4597 4598 4599
	/*
	 * 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.
	 */
4600
	ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos));
M
Mark Fasheh 已提交
4601 4602 4603 4604
	if (ret) {
		mlog_errno(ret);
		goto out;
	}
4605

M
Mark Fasheh 已提交
4606 4607 4608 4609 4610 4611 4612 4613 4614 4615
	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.
	 */
4616
	ocfs2_figure_contig_type(inode, insert, el, insert_rec, et);
M
Mark Fasheh 已提交
4617 4618 4619 4620 4621

	/*
	 * 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
4622
	 * cluster count on the last record of the path directly to it's
M
Mark Fasheh 已提交
4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636
	 * 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.
	 */
4637
	if (ocfs2_et_get_last_eb_blk(et) ==
4638
	    path_leaf_bh(path)->b_blocknr) {
M
Mark Fasheh 已提交
4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657
		/*
		 * 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;
4658 4659
}

M
Mark Fasheh 已提交
4660 4661 4662 4663 4664
/*
 * Insert an extent into an inode btree.
 *
 * The caller needs to update fe->i_clusters
 */
4665 4666 4667 4668 4669 4670 4671 4672 4673
int ocfs2_insert_extent(struct ocfs2_super *osb,
			handle_t *handle,
			struct inode *inode,
			struct ocfs2_extent_tree *et,
			u32 cpos,
			u64 start_blk,
			u32 new_clusters,
			u8 flags,
			struct ocfs2_alloc_context *meta_ac)
4674
{
4675
	int status;
4676
	int uninitialized_var(free_records);
4677
	struct buffer_head *last_eb_bh = NULL;
M
Mark Fasheh 已提交
4678 4679 4680 4681 4682 4683
	struct ocfs2_insert_type insert = {0, };
	struct ocfs2_extent_rec rec;

	mlog(0, "add %u clusters at position %u to inode %llu\n",
	     new_clusters, cpos, (unsigned long long)OCFS2_I(inode)->ip_blkno);

4684
	memset(&rec, 0, sizeof(rec));
M
Mark Fasheh 已提交
4685 4686
	rec.e_cpos = cpu_to_le32(cpos);
	rec.e_blkno = cpu_to_le64(start_blk);
4687
	rec.e_leaf_clusters = cpu_to_le16(new_clusters);
4688
	rec.e_flags = flags;
4689
	status = ocfs2_et_insert_check(et, &rec);
4690 4691 4692 4693
	if (status) {
		mlog_errno(status);
		goto bail;
	}
M
Mark Fasheh 已提交
4694

4695
	status = ocfs2_figure_insert_type(inode, et, &last_eb_bh, &rec,
4696
					  &free_records, &insert);
M
Mark Fasheh 已提交
4697 4698 4699
	if (status < 0) {
		mlog_errno(status);
		goto bail;
4700 4701
	}

M
Mark Fasheh 已提交
4702 4703 4704 4705
	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,
4706
	     free_records, insert.ins_tree_depth);
4707

4708
	if (insert.ins_contig == CONTIG_NONE && free_records == 0) {
4709
		status = ocfs2_grow_tree(handle, et,
4710
					 &insert.ins_tree_depth, &last_eb_bh,
4711 4712
					 meta_ac);
		if (status) {
4713 4714 4715 4716 4717
			mlog_errno(status);
			goto bail;
		}
	}

M
Mark Fasheh 已提交
4718
	/* Finally, we can add clusters. This might rotate the tree for us. */
4719
	status = ocfs2_do_insert_extent(handle, et, &rec, &insert);
4720 4721
	if (status < 0)
		mlog_errno(status);
4722
	else if (et->et_ops == &ocfs2_dinode_et_ops)
M
Mark Fasheh 已提交
4723
		ocfs2_extent_map_insert_rec(inode, &rec);
4724 4725

bail:
4726
	brelse(last_eb_bh);
4727

4728 4729 4730 4731
	mlog_exit(status);
	return status;
}

4732 4733 4734
/*
 * Allcate and add clusters into the extent b-tree.
 * The new clusters(clusters_to_add) will be inserted at logical_offset.
4735
 * The extent b-tree's root is specified by et, and
4736 4737 4738 4739 4740 4741 4742 4743
 * it is not limited to the file storage. Any extent tree can use this
 * function if it implements the proper ocfs2_extent_tree.
 */
int ocfs2_add_clusters_in_btree(struct ocfs2_super *osb,
				struct inode *inode,
				u32 *logical_offset,
				u32 clusters_to_add,
				int mark_unwritten,
4744
				struct ocfs2_extent_tree *et,
4745 4746 4747
				handle_t *handle,
				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 4756 4757 4758 4759 4760 4761
{
	int status = 0;
	int free_extents;
	enum ocfs2_alloc_restarted reason = RESTART_NONE;
	u32 bit_off, num_bits;
	u64 block;
	u8 flags = 0;

	BUG_ON(!clusters_to_add);

	if (mark_unwritten)
		flags = OCFS2_EXT_UNWRITTEN;

4762
	free_extents = ocfs2_num_free_extents(osb, et);
4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780
	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)
4781
		       < ocfs2_extend_meta_needed(et->et_root_el))) {
4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797
		mlog(0, "filesystem is really fragmented...\n");
		status = -EAGAIN;
		reason = RESTART_META;
		goto leave;
	}

	status = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
					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);

4798
	/* reserve our write early -- insert_extent may update the tree root */
4799
	status = ocfs2_et_root_journal_access(handle, et,
4800
					      OCFS2_JOURNAL_ACCESS_WRITE);
4801 4802 4803 4804 4805 4806 4807 4808
	if (status < 0) {
		mlog_errno(status);
		goto leave;
	}

	block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
	mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
	     num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4809 4810 4811
	status = ocfs2_insert_extent(osb, handle, inode, et,
				     *logical_offset, block,
				     num_bits, flags, meta_ac);
4812 4813 4814 4815 4816
	if (status < 0) {
		mlog_errno(status);
		goto leave;
	}

4817
	status = ocfs2_journal_dirty(handle, et->et_root_bh);
4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839
	if (status < 0) {
		mlog_errno(status);
		goto leave;
	}

	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;
}

4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862
static void ocfs2_make_right_split_rec(struct super_block *sb,
				       struct ocfs2_extent_rec *split_rec,
				       u32 cpos,
				       struct ocfs2_extent_rec *rec)
{
	u32 rec_cpos = le32_to_cpu(rec->e_cpos);
	u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters);

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

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

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

	split_rec->e_flags = rec->e_flags;
}

static int ocfs2_split_and_insert(struct inode *inode,
				  handle_t *handle,
				  struct ocfs2_path *path,
4863
				  struct ocfs2_extent_tree *et,
4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884
				  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];

4885
	rightmost_el = et->et_root_el;
4886 4887 4888 4889 4890 4891 4892 4893 4894 4895

	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)) {
4896
		ret = ocfs2_grow_tree(handle, et,
4897
				      &depth, last_eb_bh, meta_ac);
4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

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

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

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

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

		split_rec = tmprec;

		BUG_ON(do_leftright);
		do_leftright = 1;
	}

4934
	ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949
	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);
4950
		ret = ocfs2_find_path(et->et_ci, path, cpos);
4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964
		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;
}

4965 4966 4967 4968 4969 4970 4971 4972 4973
static int ocfs2_replace_extent_rec(struct inode *inode,
				    handle_t *handle,
				    struct ocfs2_path *path,
				    struct ocfs2_extent_list *el,
				    int split_index,
				    struct ocfs2_extent_rec *split_rec)
{
	int ret;

4974
	ret = ocfs2_path_bh_journal_access(handle, INODE_CACHE(inode), path,
4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987
					   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;
}

4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998
/*
 * Mark part or all of the extent record at split_index in the leaf
 * pointed to by path as written. This removes the unwritten
 * extent flag.
 *
 * Care is taken to handle contiguousness so as to not grow the tree.
 *
 * meta_ac is not strictly necessary - we only truly need it if growth
 * of the tree is required. All other cases will degrade into a less
 * optimal tree layout.
 *
4999 5000 5001
 * 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.
5002 5003 5004 5005 5006 5007 5008
 *
 * This code is optimized for readability - several passes might be
 * made over certain portions of the tree. All of those blocks will
 * have been brought into cache (and pinned via the journal), so the
 * extra overhead is not expressed in terms of disk reads.
 */
static int __ocfs2_mark_extent_written(struct inode *inode,
5009
				       struct ocfs2_extent_tree *et,
5010 5011 5012 5013 5014 5015 5016 5017 5018
				       handle_t *handle,
				       struct ocfs2_path *path,
				       int split_index,
				       struct ocfs2_extent_rec *split_rec,
				       struct ocfs2_alloc_context *meta_ac,
				       struct ocfs2_cached_dealloc_ctxt *dealloc)
{
	int ret = 0;
	struct ocfs2_extent_list *el = path_leaf_el(path);
5019
	struct buffer_head *last_eb_bh = NULL;
5020 5021 5022 5023
	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
	struct ocfs2_merge_ctxt ctxt;
	struct ocfs2_extent_list *rightmost_el;

5024
	if (!(rec->e_flags & OCFS2_EXT_UNWRITTEN)) {
5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037
		ret = -EIO;
		mlog_errno(ret);
		goto out;
	}

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

5038
	ctxt.c_contig_type = ocfs2_figure_merge_contig_type(et, path, el,
5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049
							    split_index,
							    split_rec);

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

5050
		ret = ocfs2_read_extent_block(et->et_ci,
5051 5052
					      ocfs2_et_get_last_eb_blk(et),
					      &last_eb_bh);
5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070
		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
5071 5072 5073
	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);
5074 5075 5076

	if (ctxt.c_contig_type == CONTIG_NONE) {
		if (ctxt.c_split_covers_rec)
5077 5078 5079
			ret = ocfs2_replace_extent_rec(inode, handle,
						       path, el,
						       split_index, split_rec);
5080
		else
5081
			ret = ocfs2_split_and_insert(inode, handle, path, et,
5082 5083 5084 5085 5086
						     &last_eb_bh, split_index,
						     split_rec, meta_ac);
		if (ret)
			mlog_errno(ret);
	} else {
5087
		ret = ocfs2_try_to_merge_extent(handle, et, path,
5088
						split_index, split_rec,
5089
						dealloc, &ctxt);
5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106
		if (ret)
			mlog_errno(ret);
	}

out:
	brelse(last_eb_bh);
	return ret;
}

/*
 * Mark the already-existing extent at cpos as written for len clusters.
 *
 * If the existing extent is larger than the request, initiate a
 * split. An attempt will be made at merging with adjacent extents.
 *
 * The caller is responsible for passing down meta_ac if we'll need it.
 */
5107 5108
int ocfs2_mark_extent_written(struct inode *inode,
			      struct ocfs2_extent_tree *et,
5109 5110
			      handle_t *handle, u32 cpos, u32 len, u32 phys,
			      struct ocfs2_alloc_context *meta_ac,
5111
			      struct ocfs2_cached_dealloc_ctxt *dealloc)
5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134
{
	int ret, index;
	u64 start_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys);
	struct ocfs2_extent_rec split_rec;
	struct ocfs2_path *left_path = NULL;
	struct ocfs2_extent_list *el;

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

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

	/*
	 * XXX: This should be fixed up so that we just re-insert the
	 * next extent records.
	 */
5135
	ocfs2_et_extent_map_truncate(et, 0);
5136

5137
	left_path = ocfs2_new_path_from_et(et);
5138 5139 5140 5141 5142 5143
	if (!left_path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

5144
	ret = ocfs2_find_path(et->et_ci, left_path, cpos);
5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167
	if (ret) {
		mlog_errno(ret);
		goto out;
	}
	el = path_leaf_el(left_path);

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

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

5168
	ret = __ocfs2_mark_extent_written(inode, et, handle, left_path,
5169 5170
					  index, &split_rec, meta_ac,
					  dealloc);
5171 5172 5173 5174 5175 5176 5177 5178
	if (ret)
		mlog_errno(ret);

out:
	ocfs2_free_path(left_path);
	return ret;
}

5179
static int ocfs2_split_tree(struct inode *inode, struct ocfs2_extent_tree *et,
5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200
			    handle_t *handle, struct ocfs2_path *path,
			    int index, u32 new_range,
			    struct ocfs2_alloc_context *meta_ac)
{
	int ret, depth, credits = handle->h_buffer_credits;
	struct buffer_head *last_eb_bh = NULL;
	struct ocfs2_extent_block *eb;
	struct ocfs2_extent_list *rightmost_el, *el;
	struct ocfs2_extent_rec split_rec;
	struct ocfs2_extent_rec *rec;
	struct ocfs2_insert_type insert;

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

	depth = path->p_tree_depth;
	if (depth > 0) {
5201
		ret = ocfs2_read_extent_block(et->et_ci,
5202 5203
					      ocfs2_et_get_last_eb_blk(et),
					      &last_eb_bh);
5204 5205 5206 5207 5208 5209 5210 5211 5212 5213
		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);

5214
	credits += path->p_tree_depth +
5215
		   ocfs2_extend_meta_needed(et->et_root_el);
5216 5217 5218 5219 5220 5221 5222 5223
	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)) {
5224
		ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh,
5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237
				      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;

5238
	ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
5239 5240 5241 5242 5243 5244 5245 5246
	if (ret)
		mlog_errno(ret);

out:
	brelse(last_eb_bh);
	return ret;
}

5247 5248
static int ocfs2_truncate_rec(handle_t *handle,
			      struct ocfs2_extent_tree *et,
5249 5250
			      struct ocfs2_path *path, int index,
			      struct ocfs2_cached_dealloc_ctxt *dealloc,
5251
			      u32 cpos, u32 len)
5252 5253 5254 5255
{
	int ret;
	u32 left_cpos, rec_range, trunc_range;
	int wants_rotate = 0, is_rightmost_tree_rec = 0;
5256
	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
5257 5258 5259 5260 5261 5262
	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) {
5263
		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
5264 5265 5266 5267 5268 5269 5270 5271 5272 5273 5274 5275 5276 5277 5278 5279 5280 5281 5282 5283 5284 5285 5286 5287 5288 5289 5290 5291 5292 5293 5294
		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:
5295
		 *   1) Path is the leftmost one in our btree.
5296 5297 5298 5299 5300
		 *   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.
		 */

5301
		ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
5302 5303 5304 5305 5306 5307
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

		if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
5308
			left_path = ocfs2_new_path_from_path(path);
5309 5310 5311 5312 5313 5314
			if (!left_path) {
				ret = -ENOMEM;
				mlog_errno(ret);
				goto out;
			}

5315 5316
			ret = ocfs2_find_path(et->et_ci, left_path,
					      left_cpos);
5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331
			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;
	}

5332
	ret = ocfs2_journal_access_path(et->et_ci, handle, path);
5333 5334 5335 5336 5337
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

5338
	ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
5339 5340 5341 5342 5343 5344 5345 5346 5347 5348 5349 5350 5351 5352 5353 5354 5355 5356 5357 5358 5359 5360
	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];
5361
			ocfs2_adjust_rightmost_records(handle, et, path,
5362 5363 5364 5365 5366 5367 5368 5369 5370 5371 5372
						       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)
5373
			ocfs2_adjust_rightmost_records(handle, et, path, rec);
5374 5375
	} else {
		/* Caller should have trapped this. */
5376 5377 5378
		mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) "
		     "(%u, %u)\n",
		     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5379 5380 5381 5382 5383 5384 5385 5386
		     le32_to_cpu(rec->e_cpos),
		     le16_to_cpu(rec->e_leaf_clusters), cpos, len);
		BUG();
	}

	if (left_path) {
		int subtree_index;

5387
		subtree_index = ocfs2_find_subtree_root(et, left_path, path);
5388
		ocfs2_complete_edge_insert(handle, left_path, path,
5389 5390 5391 5392 5393
					   subtree_index);
	}

	ocfs2_journal_dirty(handle, path_leaf_bh(path));

5394
	ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
5395 5396 5397 5398 5399 5400 5401 5402 5403 5404
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

out:
	ocfs2_free_path(left_path);
	return ret;
}

5405 5406
int ocfs2_remove_extent(struct inode *inode,
			struct ocfs2_extent_tree *et,
5407 5408
			u32 cpos, u32 len, handle_t *handle,
			struct ocfs2_alloc_context *meta_ac,
5409
			struct ocfs2_cached_dealloc_ctxt *dealloc)
5410 5411 5412 5413 5414
{
	int ret, index;
	u32 rec_range, trunc_range;
	struct ocfs2_extent_rec *rec;
	struct ocfs2_extent_list *el;
5415
	struct ocfs2_path *path = NULL;
5416

5417 5418 5419 5420 5421
	/*
	 * XXX: Why are we truncating to 0 instead of wherever this
	 * affects us?
	 */
	ocfs2_et_extent_map_truncate(et, 0);
5422

5423
	path = ocfs2_new_path_from_et(et);
5424 5425 5426 5427 5428 5429
	if (!path) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

5430
	ret = ocfs2_find_path(et->et_ci, path, cpos);
5431 5432 5433 5434 5435 5436 5437 5438 5439 5440 5441 5442 5443 5444 5445 5446 5447 5448 5449 5450 5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461 5462 5463 5464 5465 5466 5467 5468 5469 5470 5471 5472 5473 5474
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

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

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

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

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

	if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
5475 5476
		ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
					 cpos, len);
5477 5478 5479 5480 5481
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	} else {
5482
		ret = ocfs2_split_tree(inode, et, handle, path, index,
5483 5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494
				       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);

5495
		ret = ocfs2_find_path(et->et_ci, path, cpos);
5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529
		if (ret) {
			mlog_errno(ret);
			goto out;
		}

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

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

5530 5531
		ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
					 cpos, len);
5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542
		if (ret) {
			mlog_errno(ret);
			goto out;
		}
	}

out:
	ocfs2_free_path(path);
	return ret;
}

5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557 5558 5559 5560 5561 5562 5563 5564 5565 5566 5567 5568 5569 5570
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;
		}
	}

5571
	handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5572 5573 5574 5575 5576 5577
	if (IS_ERR(handle)) {
		ret = PTR_ERR(handle);
		mlog_errno(ret);
		goto out;
	}

5578
	ret = ocfs2_et_root_journal_access(handle, et,
5579
					   OCFS2_JOURNAL_ACCESS_WRITE);
5580 5581 5582 5583 5584
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

5585 5586 5587
	vfs_dq_free_space_nodirty(inode,
				  ocfs2_clusters_to_bytes(inode->i_sb, len));

5588 5589 5590 5591 5592 5593 5594
	ret = ocfs2_remove_extent(inode, et, cpos, len, handle, meta_ac,
				  dealloc);
	if (ret) {
		mlog_errno(ret);
		goto out_commit;
	}

5595
	ocfs2_et_update_clusters(et, -len);
5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617

	ret = ocfs2_journal_dirty(handle, et->et_root_bh);
	if (ret) {
		mlog_errno(ret);
		goto out_commit;
	}

	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;
}

5618
int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
5619 5620 5621 5622 5623 5624 5625 5626 5627 5628 5629 5630 5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648 5649 5650
{
	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;
}

5651 5652 5653 5654
int ocfs2_truncate_log_append(struct ocfs2_super *osb,
			      handle_t *handle,
			      u64 start_blk,
			      unsigned int num_clusters)
5655 5656 5657 5658 5659 5660 5661 5662
{
	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;

5663 5664
	mlog_entry("start_blk = %llu, num_clusters = %u\n",
		   (unsigned long long)start_blk, num_clusters);
5665

5666
	BUG_ON(mutex_trylock(&tl_inode->i_mutex));
5667 5668 5669 5670 5671

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

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

5672 5673 5674 5675 5676 5677
	/* 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;
5678 5679 5680
	tl_count = le16_to_cpu(tl->tl_count);
	mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
			tl_count == 0,
5681 5682 5683
			"Truncate record count on #%llu invalid "
			"wanted %u, actual %u\n",
			(unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5684 5685 5686 5687 5688 5689 5690 5691 5692 5693 5694
			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;
	}

5695
	status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
5696
					 OCFS2_JOURNAL_ACCESS_WRITE);
5697 5698 5699 5700 5701 5702
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

	mlog(0, "Log truncate of %u clusters starting at cluster %u to "
5703 5704
	     "%llu (index = %d)\n", num_clusters, start_cluster,
	     (unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index);
5705 5706 5707 5708 5709 5710 5711 5712 5713 5714 5715 5716 5717 5718 5719 5720 5721 5722 5723 5724 5725 5726 5727 5728 5729 5730 5731 5732 5733 5734

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

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

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

bail:
	mlog_exit(status);
	return status;
}

static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
5735
					 handle_t *handle,
5736 5737 5738 5739 5740 5741 5742 5743 5744 5745 5746 5747 5748 5749 5750 5751 5752 5753 5754 5755 5756
					 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 */
5757
		status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
5758
						 OCFS2_JOURNAL_ACCESS_WRITE);
5759 5760 5761 5762 5763 5764 5765 5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777 5778 5779 5780 5781 5782 5783 5784 5785 5786 5787 5788 5789 5790 5791 5792 5793 5794 5795 5796 5797 5798 5799 5800 5801 5802 5803 5804 5805 5806 5807 5808
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

		tl->tl_used = cpu_to_le16(i);

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

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

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

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

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

bail:
	mlog_exit(status);
	return status;
}

5809
/* Expects you to already be holding tl_inode->i_mutex */
5810
int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
5811 5812 5813
{
	int status;
	unsigned int num_to_flush;
5814
	handle_t *handle;
5815 5816 5817 5818 5819 5820 5821 5822 5823
	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();

5824
	BUG_ON(mutex_trylock(&tl_inode->i_mutex));
5825 5826 5827

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

5828 5829 5830 5831 5832 5833
	/* 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;
5834
	num_to_flush = le16_to_cpu(tl->tl_used);
5835 5836
	mlog(0, "Flush %u records from truncate log #%llu\n",
	     num_to_flush, (unsigned long long)OCFS2_I(tl_inode)->ip_blkno);
5837 5838
	if (!num_to_flush) {
		status = 0;
5839
		goto out;
5840 5841 5842 5843 5844 5845 5846 5847
	}

	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");
5848
		goto out;
5849 5850
	}

5851 5852
	mutex_lock(&data_alloc_inode->i_mutex);

M
Mark Fasheh 已提交
5853
	status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1);
5854 5855
	if (status < 0) {
		mlog_errno(status);
5856
		goto out_mutex;
5857 5858
	}

5859
	handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
5860 5861 5862
	if (IS_ERR(handle)) {
		status = PTR_ERR(handle);
		mlog_errno(status);
5863
		goto out_unlock;
5864 5865 5866 5867
	}

	status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
					       data_alloc_bh);
5868
	if (status < 0)
5869 5870
		mlog_errno(status);

5871
	ocfs2_commit_trans(osb, handle);
5872

5873 5874
out_unlock:
	brelse(data_alloc_bh);
M
Mark Fasheh 已提交
5875
	ocfs2_inode_unlock(data_alloc_inode, 1);
5876

5877 5878 5879
out_mutex:
	mutex_unlock(&data_alloc_inode->i_mutex);
	iput(data_alloc_inode);
5880

5881
out:
5882 5883 5884 5885 5886 5887 5888 5889 5890
	mlog_exit(status);
	return status;
}

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

5891
	mutex_lock(&tl_inode->i_mutex);
5892
	status = __ocfs2_flush_truncate_log(osb);
5893
	mutex_unlock(&tl_inode->i_mutex);
5894 5895 5896 5897

	return status;
}

D
David Howells 已提交
5898
static void ocfs2_truncate_log_worker(struct work_struct *work)
5899 5900
{
	int status;
D
David Howells 已提交
5901 5902 5903
	struct ocfs2_super *osb =
		container_of(work, struct ocfs2_super,
			     osb_truncate_log_wq.work);
5904 5905 5906 5907 5908 5909

	mlog_entry_void();

	status = ocfs2_flush_truncate_log(osb);
	if (status < 0)
		mlog_errno(status);
5910 5911
	else
		ocfs2_init_inode_steal_slot(osb);
5912 5913 5914 5915 5916 5917 5918 5919 5920 5921 5922 5923 5924 5925 5926 5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948

	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;
	}

5949
	status = ocfs2_read_inode_block(inode, &bh);
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 5984 5985 5986 5987 5988
	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;

5989 5990 5991 5992 5993 5994
	/* 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;
5995 5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010 6011 6012 6013
	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;

6014
		ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check);
6015
		status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode));
6016 6017 6018 6019 6020 6021 6022 6023 6024
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

bail:
	if (tl_inode)
		iput(tl_inode);
6025
	brelse(tl_bh);
6026 6027 6028 6029 6030 6031 6032 6033 6034 6035 6036 6037 6038 6039 6040 6041 6042

	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;
6043
	handle_t *handle;
6044 6045 6046 6047 6048 6049 6050 6051 6052 6053 6054 6055
	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);
6056
	mlog(0, "cleanup %u records from %llu\n", num_recs,
6057
	     (unsigned long long)le64_to_cpu(tl_copy->i_blkno));
6058

6059
	mutex_lock(&tl_inode->i_mutex);
6060 6061 6062 6063 6064 6065 6066 6067 6068
	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;
			}
		}

6069
		handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
6070 6071 6072 6073 6074 6075 6076 6077 6078 6079 6080 6081
		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);
6082
		ocfs2_commit_trans(osb, handle);
6083 6084 6085 6086 6087 6088 6089
		if (status < 0) {
			mlog_errno(status);
			goto bail_up;
		}
	}

bail_up:
6090
	mutex_unlock(&tl_inode->i_mutex);
6091 6092 6093 6094 6095 6096 6097 6098 6099 6100 6101 6102 6103 6104 6105 6106 6107 6108 6109 6110 6111 6112 6113 6114 6115 6116 6117 6118 6119 6120 6121 6122 6123 6124 6125 6126 6127 6128 6129 6130 6131 6132 6133 6134 6135

	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
6136 6137
	INIT_DELAYED_WORK(&osb->osb_truncate_log_wq,
			  ocfs2_truncate_log_worker);
6138 6139 6140 6141 6142 6143 6144
	osb->osb_tl_bh    = tl_bh;
	osb->osb_tl_inode = tl_inode;

	mlog_exit(status);
	return status;
}

6145 6146 6147 6148 6149 6150 6151 6152 6153 6154 6155 6156 6157 6158 6159 6160 6161 6162 6163
/*
 * 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.
 */

/*
6164 6165 6166 6167
 * 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.
6168 6169 6170 6171 6172 6173 6174 6175 6176 6177 6178 6179 6180 6181
 */
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;
};

6182 6183 6184 6185
static int ocfs2_free_cached_blocks(struct ocfs2_super *osb,
				    int sysfile_type,
				    int slot,
				    struct ocfs2_cached_block_free *head)
6186 6187 6188 6189 6190 6191 6192 6193 6194 6195 6196 6197 6198 6199 6200 6201 6202
{
	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
6203
	ret = ocfs2_inode_lock(inode, &di_bh, 1);
6204 6205 6206 6207 6208 6209 6210 6211 6212 6213 6214 6215 6216 6217 6218 6219 6220 6221 6222 6223 6224 6225 6226 6227 6228 6229 6230 6231 6232 6233 6234 6235 6236 6237 6238 6239 6240 6241 6242 6243
	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
6244
	ocfs2_inode_unlock(inode, 1);
6245 6246 6247 6248 6249 6250 6251 6252 6253 6254 6255 6256 6257 6258 6259
	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;
}

6260 6261 6262 6263 6264 6265 6266 6267 6268 6269 6270 6271 6272 6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292 6293 6294 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
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;
}

6336 6337 6338 6339 6340 6341 6342 6343 6344 6345 6346 6347 6348 6349 6350
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);
6351 6352 6353 6354
			ret2 = ocfs2_free_cached_blocks(osb,
							fl->f_inode_type,
							fl->f_slot,
							fl->f_first);
6355 6356 6357 6358 6359 6360 6361 6362 6363 6364
			if (ret2)
				mlog_errno(ret2);
			if (!ret)
				ret = ret2;
		}

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

6365 6366 6367 6368 6369 6370 6371 6372 6373 6374 6375
	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;
	}

6376 6377 6378 6379 6380 6381 6382 6383 6384 6385 6386 6387 6388 6389 6390 6391 6392 6393 6394 6395 6396 6397 6398 6399 6400 6401 6402 6403 6404 6405 6406 6407 6408 6409 6410 6411 6412 6413 6414 6415 6416 6417 6418 6419 6420 6421 6422 6423 6424 6425 6426 6427 6428 6429 6430 6431 6432 6433 6434 6435 6436 6437 6438 6439 6440
	return ret;
}

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

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

		fl = fl->f_next_suballocator;
	}

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

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

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

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

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

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

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

	fl->f_first = item;

	ret = 0;
out:
	return ret;
}

6441 6442 6443 6444 6445 6446 6447 6448 6449
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));
}

6450 6451 6452 6453
/* 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
6454
static int ocfs2_find_new_last_ext_blk(struct inode *inode,
6455
				       unsigned int clusters_to_del,
M
Mark Fasheh 已提交
6456
				       struct ocfs2_path *path,
6457 6458
				       struct buffer_head **new_last_eb)
{
6459
	int next_free, ret = 0;
M
Mark Fasheh 已提交
6460
	u32 cpos;
6461
	struct ocfs2_extent_rec *rec;
6462 6463 6464 6465 6466 6467 6468
	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
6469 6470
	if (!path->p_tree_depth)
		goto out;
6471 6472 6473

	/* trunc to zero special case - this makes tree_depth = 0
	 * regardless of what it is.  */
6474
	if (OCFS2_I(inode)->ip_clusters == clusters_to_del)
M
Mark Fasheh 已提交
6475
		goto out;
6476

M
Mark Fasheh 已提交
6477
	el = path_leaf_el(path);
6478 6479
	BUG_ON(!el->l_next_free_rec);

6480 6481 6482 6483 6484 6485 6486 6487 6488
	/*
	 * 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
6489
	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
6490
		if (next_free > 2)
M
Mark Fasheh 已提交
6491
			goto out;
6492 6493 6494 6495 6496 6497 6498 6499 6500 6501 6502 6503 6504 6505 6506 6507 6508 6509 6510 6511 6512

		/* 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.
		 */
6513
		if (le16_to_cpu(rec->e_leaf_clusters) > clusters_to_del)
6514 6515
			goto out;
	}
6516

M
Mark Fasheh 已提交
6517 6518 6519 6520 6521
	ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, path, &cpos);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}
6522

6523
	ret = ocfs2_find_leaf(INODE_CACHE(inode), path_root_el(path), cpos, &bh);
M
Mark Fasheh 已提交
6524 6525 6526 6527
	if (ret) {
		mlog_errno(ret);
		goto out;
	}
6528

M
Mark Fasheh 已提交
6529 6530
	eb = (struct ocfs2_extent_block *) bh->b_data;
	el = &eb->h_list;
6531 6532 6533 6534

	/* 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));
6535 6536 6537

	*new_last_eb = bh;
	get_bh(*new_last_eb);
M
Mark Fasheh 已提交
6538 6539 6540 6541
	mlog(0, "returning block %llu, (cpos: %u)\n",
	     (unsigned long long)le64_to_cpu(eb->h_blkno), cpos);
out:
	brelse(bh);
6542

M
Mark Fasheh 已提交
6543
	return ret;
6544 6545
}

6546 6547 6548 6549 6550 6551 6552 6553 6554 6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570 6571 6572 6573 6574 6575 6576 6577 6578 6579 6580 6581 6582 6583 6584 6585 6586 6587 6588 6589 6590 6591
/*
 * Trim some clusters off the rightmost edge of a tree. Only called
 * during truncate.
 *
 * The caller needs to:
 *   - start journaling of each path component.
 *   - compute and fully set up any new last ext block
 */
static int ocfs2_trim_tree(struct inode *inode, struct ocfs2_path *path,
			   handle_t *handle, struct ocfs2_truncate_context *tc,
			   u32 clusters_to_del, u64 *delete_start)
{
	int ret, i, index = path->p_tree_depth;
	u32 new_edge = 0;
	u64 deleted_eb = 0;
	struct buffer_head *bh;
	struct ocfs2_extent_list *el;
	struct ocfs2_extent_rec *rec;

	*delete_start = 0;

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

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

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

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

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

		mlog(0, "Extent list before: record %d: (%u, %u, %llu), "
		     "next = %u\n", i, le32_to_cpu(rec->e_cpos),
6592
		     ocfs2_rec_clusters(el, rec),
6593 6594 6595
		     (unsigned long long)le64_to_cpu(rec->e_blkno),
		     le16_to_cpu(el->l_next_free_rec));

6596
		BUG_ON(ocfs2_rec_clusters(el, rec) < clusters_to_del);
6597 6598 6599 6600 6601 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 6628 6629 6630 6631 6632 6633 6634 6635 6636 6637

		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;
			}

6638
			le16_add_cpu(&rec->e_leaf_clusters, -clusters_to_del);
6639 6640 6641 6642 6643

			/*
			 * We'll use "new_edge" on our way back up the
			 * tree to know what our rightmost cpos is.
			 */
6644
			new_edge = le16_to_cpu(rec->e_leaf_clusters);
6645 6646 6647 6648 6649 6650 6651
			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,
6652
					le16_to_cpu(rec->e_leaf_clusters));
6653 6654 6655 6656

			/*
			 * If it's now empty, remove this record.
			 */
6657
			if (le16_to_cpu(rec->e_leaf_clusters) == 0) {
6658 6659 6660 6661 6662 6663 6664 6665 6666 6667 6668 6669 6670 6671 6672 6673 6674 6675 6676 6677 6678 6679 6680 6681 6682
				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;

6683 6684
			rec->e_int_clusters = cpu_to_le32(new_edge);
			le32_add_cpu(&rec->e_int_clusters,
6685 6686 6687 6688 6689 6690
				     -le32_to_cpu(rec->e_cpos));

			 /*
			  * A deleted child record should have been
			  * caught above.
			  */
6691
			 BUG_ON(le32_to_cpu(rec->e_int_clusters) == 0);
6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703
		}

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

		mlog(0, "extent list container %llu, after: record %d: "
		     "(%u, %u, %llu), next = %u.\n",
		     (unsigned long long)bh->b_blocknr, i,
6704
		     le32_to_cpu(rec->e_cpos), ocfs2_rec_clusters(el, rec),
6705 6706 6707 6708 6709 6710 6711 6712 6713 6714 6715 6716 6717 6718 6719 6720 6721 6722 6723
		     (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");

6724
			ocfs2_remove_from_cache(INODE_CACHE(inode), bh);
6725

6726
			BUG_ON(ocfs2_rec_clusters(el, &el->l_recs[0]));
6727 6728 6729
			BUG_ON(le32_to_cpu(el->l_recs[0].e_cpos));
			BUG_ON(le64_to_cpu(el->l_recs[0].e_blkno));

6730 6731 6732 6733
			ret = ocfs2_cache_extent_block_free(&tc->tc_dealloc, eb);
			/* An error here is not fatal. */
			if (ret < 0)
				mlog_errno(ret);
6734 6735 6736 6737 6738 6739 6740 6741 6742 6743 6744 6745
		} else {
			deleted_eb = 0;
		}

		index--;
	}

	ret = 0;
out:
	return ret;
}

6746 6747 6748 6749
static int ocfs2_do_truncate(struct ocfs2_super *osb,
			     unsigned int clusters_to_del,
			     struct inode *inode,
			     struct buffer_head *fe_bh,
6750
			     handle_t *handle,
M
Mark Fasheh 已提交
6751 6752
			     struct ocfs2_truncate_context *tc,
			     struct ocfs2_path *path)
6753
{
6754
	int status;
6755 6756 6757 6758 6759 6760 6761 6762
	struct ocfs2_dinode *fe;
	struct ocfs2_extent_block *last_eb = NULL;
	struct ocfs2_extent_list *el;
	struct buffer_head *last_eb_bh = NULL;
	u64 delete_blk = 0;

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

6763
	status = ocfs2_find_new_last_ext_blk(inode, clusters_to_del,
M
Mark Fasheh 已提交
6764
					     path, &last_eb_bh);
6765 6766 6767 6768
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}
M
Mark Fasheh 已提交
6769 6770 6771 6772 6773

	/*
	 * Each component will be touched, so we might as well journal
	 * here to avoid having to handle errors later.
	 */
6774
	status = ocfs2_journal_access_path(INODE_CACHE(inode), handle, path);
6775 6776 6777
	if (status < 0) {
		mlog_errno(status);
		goto bail;
M
Mark Fasheh 已提交
6778 6779 6780
	}

	if (last_eb_bh) {
6781
		status = ocfs2_journal_access_eb(handle, INODE_CACHE(inode), last_eb_bh,
6782
						 OCFS2_JOURNAL_ACCESS_WRITE);
M
Mark Fasheh 已提交
6783 6784 6785 6786 6787
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}

6788
		last_eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
M
Mark Fasheh 已提交
6789
	}
6790

M
Mark Fasheh 已提交
6791 6792 6793 6794 6795 6796
	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.
	 */
6797
	if (el->l_tree_depth && el->l_recs[0].e_int_clusters == 0) {
M
Mark Fasheh 已提交
6798 6799 6800
		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));
6801
		status = -EROFS;
6802 6803 6804
		goto bail;
	}

6805 6806
	vfs_dq_free_space_nodirty(inode,
			ocfs2_clusters_to_bytes(osb->sb, clusters_to_del));
6807 6808 6809 6810 6811
	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);
6812
	inode->i_blocks = ocfs2_inode_sector_count(inode);
6813

6814 6815 6816 6817 6818
	status = ocfs2_trim_tree(inode, path, handle, tc,
				 clusters_to_del, &delete_blk);
	if (status) {
		mlog_errno(status);
		goto bail;
6819 6820
	}

M
Mark Fasheh 已提交
6821
	if (le32_to_cpu(fe->i_clusters) == 0) {
6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841 6842 6843 6844 6845
		/* trunc to zero is a special case. */
		el->l_tree_depth = 0;
		fe->i_last_eb_blk = 0;
	} else if (last_eb)
		fe->i_last_eb_blk = last_eb->h_blkno;

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

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

6846 6847 6848
	if (delete_blk) {
		status = ocfs2_truncate_log_append(osb, handle, delete_blk,
						   clusters_to_del);
6849 6850 6851 6852 6853 6854 6855
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}
	status = 0;
bail:
6856
	brelse(last_eb_bh);
6857 6858 6859 6860
	mlog_exit(status);
	return status;
}

J
Joel Becker 已提交
6861
static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh)
6862 6863 6864 6865 6866 6867
{
	set_buffer_uptodate(bh);
	mark_buffer_dirty(bh);
	return 0;
}

6868 6869 6870 6871 6872 6873 6874 6875 6876 6877 6878
static void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
				     unsigned int from, unsigned int to,
				     struct page *page, int zero, u64 *phys)
{
	int ret, partial = 0;

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

	if (zero)
6879
		zero_user_segment(page, from, to);
6880 6881 6882 6883 6884 6885

	/*
	 * 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
6886 6887 6888 6889 6890 6891 6892
	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);
6893 6894 6895 6896 6897 6898 6899 6900 6901 6902
		if (ret < 0)
			mlog_errno(ret);
	}

	if (!partial)
		SetPageUptodate(page);

	flush_dcache_page(page);
}

6903 6904 6905
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)
6906
{
6907
	int i;
6908 6909 6910 6911 6912 6913 6914 6915 6916
	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;

6917
	to = PAGE_CACHE_SIZE;
6918 6919 6920
	for(i = 0; i < numpages; i++) {
		page = pages[i];

6921 6922 6923 6924
		from = start & (PAGE_CACHE_SIZE - 1);
		if ((end >> PAGE_CACHE_SHIFT) == page->index)
			to = end & (PAGE_CACHE_SIZE - 1);

6925 6926 6927
		BUG_ON(from > PAGE_CACHE_SIZE);
		BUG_ON(to > PAGE_CACHE_SIZE);

6928 6929
		ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1,
					 &phys);
6930

6931
		start = (page->index + 1) << PAGE_CACHE_SHIFT;
6932 6933
	}
out:
6934 6935
	if (pages)
		ocfs2_unlock_and_free_pages(pages, numpages);
6936 6937
}

6938
static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end,
6939
				struct page **pages, int *num)
6940
{
6941
	int numpages, ret = 0;
6942 6943 6944
	struct super_block *sb = inode->i_sb;
	struct address_space *mapping = inode->i_mapping;
	unsigned long index;
6945
	loff_t last_page_bytes;
6946

6947
	BUG_ON(start > end);
6948

6949 6950 6951
	BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits !=
	       (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits);

6952
	numpages = 0;
6953 6954
	last_page_bytes = PAGE_ALIGN(end);
	index = start >> PAGE_CACHE_SHIFT;
6955 6956 6957 6958 6959 6960 6961 6962 6963 6964
	do {
		pages[numpages] = grab_cache_page(mapping, index);
		if (!pages[numpages]) {
			ret = -ENOMEM;
			mlog_errno(ret);
			goto out;
		}

		numpages++;
		index++;
6965
	} while (index < (last_page_bytes >> PAGE_CACHE_SHIFT));
6966 6967 6968

out:
	if (ret != 0) {
6969 6970
		if (pages)
			ocfs2_unlock_and_free_pages(pages, numpages);
6971 6972 6973 6974 6975 6976 6977 6978 6979 6980 6981 6982 6983 6984 6985 6986 6987
		numpages = 0;
	}

	*num = numpages;

	return ret;
}

/*
 * Zero the area past i_size but still within an allocated
 * cluster. This avoids exposing nonzero data on subsequent file
 * extends.
 *
 * We need to call this before i_size is updated on the inode because
 * otherwise block_write_full_page() will skip writeout of pages past
 * i_size. The new_i_size parameter is passed for this reason.
 */
6988 6989
int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
				  u64 range_start, u64 range_end)
6990
{
6991
	int ret = 0, numpages;
6992 6993
	struct page **pages = NULL;
	u64 phys;
6994 6995
	unsigned int ext_flags;
	struct super_block *sb = inode->i_sb;
6996 6997 6998 6999 7000

	/*
	 * File systems which don't support sparse files zero on every
	 * extend.
	 */
7001
	if (!ocfs2_sparse_alloc(OCFS2_SB(sb)))
7002 7003
		return 0;

7004
	pages = kcalloc(ocfs2_pages_per_cluster(sb),
7005 7006 7007 7008 7009 7010 7011
			sizeof(struct page *), GFP_NOFS);
	if (pages == NULL) {
		ret = -ENOMEM;
		mlog_errno(ret);
		goto out;
	}

7012 7013 7014 7015 7016 7017
	if (range_start == range_end)
		goto out;

	ret = ocfs2_extent_map_get_blocks(inode,
					  range_start >> sb->s_blocksize_bits,
					  &phys, NULL, &ext_flags);
7018 7019 7020 7021 7022
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

7023 7024 7025 7026 7027
	/*
	 * 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)
7028 7029
		goto out;

7030 7031 7032 7033 7034 7035 7036
	ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages,
				   &numpages);
	if (ret) {
		mlog_errno(ret);
		goto out;
	}

7037 7038
	ocfs2_zero_cluster_pages(inode, range_start, range_end, pages,
				 numpages, phys, handle);
7039 7040 7041 7042 7043 7044

	/*
	 * 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.
	 */
7045 7046
	ret = do_sync_mapping_range(inode->i_mapping, range_start,
				    range_end - 1, SYNC_FILE_RANGE_WRITE);
7047 7048 7049 7050 7051 7052 7053 7054 7055 7056
	if (ret)
		mlog_errno(ret);

out:
	if (pages)
		kfree(pages);

	return ret;
}

7057 7058
static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode,
					     struct ocfs2_dinode *di)
M
Mark Fasheh 已提交
7059 7060
{
	unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits;
7061
	unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size);
M
Mark Fasheh 已提交
7062

7063 7064 7065 7066 7067 7068 7069
	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
7070 7071
}

7072 7073 7074
void ocfs2_dinode_new_extent_list(struct inode *inode,
				  struct ocfs2_dinode *di)
{
7075
	ocfs2_zero_dinode_id2_with_xattr(inode, di);
7076 7077
	di->id2.i_list.l_tree_depth = 0;
	di->id2.i_list.l_next_free_rec = 0;
7078 7079
	di->id2.i_list.l_count = cpu_to_le16(
		ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di));
7080 7081
}

M
Mark Fasheh 已提交
7082 7083 7084 7085 7086 7087 7088 7089 7090 7091 7092 7093 7094 7095
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.
	 */
7096
	ocfs2_zero_dinode_id2_with_xattr(inode, di);
M
Mark Fasheh 已提交
7097

7098 7099
	idata->id_count = cpu_to_le16(
			ocfs2_max_inline_data_with_xattr(inode->i_sb, di));
M
Mark Fasheh 已提交
7100 7101 7102 7103 7104 7105 7106 7107 7108 7109 7110 7111 7112 7113
}

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;
7114
	struct ocfs2_extent_tree et;
7115
	int did_quota = 0;
M
Mark Fasheh 已提交
7116 7117 7118 7119 7120 7121 7122 7123 7124 7125 7126 7127 7128 7129 7130 7131 7132 7133 7134

	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;
		}
	}

7135 7136
	handle = ocfs2_start_trans(osb,
				   ocfs2_inline_to_extents_credits(osb->sb));
M
Mark Fasheh 已提交
7137 7138 7139 7140 7141 7142
	if (IS_ERR(handle)) {
		ret = PTR_ERR(handle);
		mlog_errno(ret);
		goto out_unlock;
	}

7143
	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
7144
				      OCFS2_JOURNAL_ACCESS_WRITE);
M
Mark Fasheh 已提交
7145 7146 7147 7148 7149 7150 7151 7152 7153 7154
	if (ret) {
		mlog_errno(ret);
		goto out_commit;
	}

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

7155 7156 7157 7158 7159 7160 7161
		if (vfs_dq_alloc_space_nodirty(inode,
				       ocfs2_clusters_to_bytes(osb->sb, 1))) {
			ret = -EDQUOT;
			goto out_commit;
		}
		did_quota = 1;

M
Mark Fasheh 已提交
7162 7163 7164 7165 7166 7167 7168 7169 7170 7171 7172 7173 7174 7175 7176 7177 7178 7179 7180 7181 7182 7183 7184 7185 7186 7187 7188 7189 7190 7191 7192 7193 7194 7195 7196 7197 7198 7199 7200 7201 7202 7203 7204 7205 7206 7207 7208 7209 7210 7211 7212
		ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off,
					   &num);
		if (ret) {
			mlog_errno(ret);
			goto out_commit;
		}

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

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

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

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

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

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

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

7213
	ocfs2_dinode_new_extent_list(inode, di);
M
Mark Fasheh 已提交
7214 7215 7216 7217 7218 7219 7220 7221 7222

	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.
		 */
7223
		ocfs2_init_dinode_extent_tree(&et, inode, di_bh);
7224 7225
		ret = ocfs2_insert_extent(osb, handle, inode, &et,
					  0, block, 1, 0, NULL);
M
Mark Fasheh 已提交
7226 7227 7228 7229 7230 7231 7232 7233 7234
		if (ret) {
			mlog_errno(ret);
			goto out_commit;
		}

		inode->i_blocks = ocfs2_inode_sector_count(inode);
	}

out_commit:
7235 7236 7237 7238
	if (ret < 0 && did_quota)
		vfs_dq_free_space_nodirty(inode,
					  ocfs2_clusters_to_bytes(osb->sb, 1));

M
Mark Fasheh 已提交
7239 7240 7241 7242 7243 7244 7245 7246 7247 7248 7249 7250 7251 7252 7253
	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;
}

7254 7255 7256 7257 7258 7259 7260 7261 7262 7263 7264 7265
/*
 * 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
7266
	u32 clusters_to_del, new_highest_cpos, range;
7267
	struct ocfs2_extent_list *el;
7268
	handle_t *handle = NULL;
7269
	struct inode *tl_inode = osb->osb_tl_inode;
M
Mark Fasheh 已提交
7270
	struct ocfs2_path *path = NULL;
7271
	struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
7272 7273 7274

	mlog_entry_void();

M
Mark Fasheh 已提交
7275
	new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
7276 7277
						     i_size_read(inode));

7278 7279
	path = ocfs2_new_path(fe_bh, &di->id2.i_list,
			      ocfs2_journal_access_di);
M
Mark Fasheh 已提交
7280 7281 7282 7283 7284
	if (!path) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}
M
Mark Fasheh 已提交
7285 7286 7287

	ocfs2_extent_map_trunc(inode, new_highest_cpos);

7288
start:
7289 7290 7291 7292 7293 7294 7295 7296
	/*
	 * Check that we still have allocation to delete.
	 */
	if (OCFS2_I(inode)->ip_clusters == 0) {
		status = 0;
		goto bail;
	}

M
Mark Fasheh 已提交
7297 7298 7299
	/*
	 * Truncate always works against the rightmost tree branch.
	 */
7300
	status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX);
M
Mark Fasheh 已提交
7301 7302 7303
	if (status) {
		mlog_errno(status);
		goto bail;
7304 7305
	}

M
Mark Fasheh 已提交
7306 7307 7308 7309 7310 7311 7312 7313 7314 7315 7316 7317 7318 7319 7320
	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);
7321 7322 7323 7324 7325 7326 7327 7328 7329
	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;
	}

7330
	i = le16_to_cpu(el->l_next_free_rec) - 1;
M
Mark Fasheh 已提交
7331
	range = le32_to_cpu(el->l_recs[i].e_cpos) +
7332
		ocfs2_rec_clusters(el, &el->l_recs[i]);
M
Mark Fasheh 已提交
7333 7334 7335
	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) {
7336
		clusters_to_del = ocfs2_rec_clusters(el, &el->l_recs[i]);
M
Mark Fasheh 已提交
7337
	} else if (range > new_highest_cpos) {
7338
		clusters_to_del = (ocfs2_rec_clusters(el, &el->l_recs[i]) +
7339
				   le32_to_cpu(el->l_recs[i].e_cpos)) -
M
Mark Fasheh 已提交
7340 7341 7342 7343 7344
				  new_highest_cpos;
	} else {
		status = 0;
		goto bail;
	}
7345

M
Mark Fasheh 已提交
7346 7347 7348
	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);

7349
	mutex_lock(&tl_inode->i_mutex);
7350 7351 7352 7353 7354 7355 7356 7357 7358 7359 7360 7361 7362
	tl_sem = 1;
	/* ocfs2_truncate_log_needs_flush guarantees us at least one
	 * record is free for use. If there isn't any, we flush to get
	 * an empty truncate log.  */
	if (ocfs2_truncate_log_needs_flush(osb)) {
		status = __ocfs2_flush_truncate_log(osb);
		if (status < 0) {
			mlog_errno(status);
			goto bail;
		}
	}

	credits = ocfs2_calc_tree_trunc_credits(osb->sb, clusters_to_del,
M
Mark Fasheh 已提交
7363 7364
						(struct ocfs2_dinode *)fe_bh->b_data,
						el);
7365
	handle = ocfs2_start_trans(osb, credits);
7366 7367 7368 7369 7370 7371 7372
	if (IS_ERR(handle)) {
		status = PTR_ERR(handle);
		handle = NULL;
		mlog_errno(status);
		goto bail;
	}

M
Mark Fasheh 已提交
7373 7374
	status = ocfs2_do_truncate(osb, clusters_to_del, inode, fe_bh, handle,
				   tc, path);
7375 7376 7377 7378 7379
	if (status < 0) {
		mlog_errno(status);
		goto bail;
	}

7380
	mutex_unlock(&tl_inode->i_mutex);
7381 7382
	tl_sem = 0;

7383
	ocfs2_commit_trans(osb, handle);
7384 7385
	handle = NULL;

M
Mark Fasheh 已提交
7386 7387 7388
	ocfs2_reinit_path(path, 1);

	/*
7389 7390
	 * The check above will catch the case where we've truncated
	 * away all allocation.
M
Mark Fasheh 已提交
7391
	 */
7392 7393
	goto start;

7394 7395 7396 7397 7398
bail:

	ocfs2_schedule_truncate_log_flush(osb, 1);

	if (tl_sem)
7399
		mutex_unlock(&tl_inode->i_mutex);
7400 7401

	if (handle)
7402
		ocfs2_commit_trans(osb, handle);
7403

7404 7405
	ocfs2_run_deallocs(osb, &tc->tc_dealloc);

M
Mark Fasheh 已提交
7406
	ocfs2_free_path(path);
7407 7408 7409 7410 7411 7412 7413 7414 7415

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

	mlog_exit(status);
	return status;
}

/*
7416
 * Expects the inode to already be locked.
7417 7418 7419 7420 7421 7422
 */
int ocfs2_prepare_truncate(struct ocfs2_super *osb,
			   struct inode *inode,
			   struct buffer_head *fe_bh,
			   struct ocfs2_truncate_context **tc)
{
7423
	int status;
7424 7425 7426 7427 7428 7429 7430 7431 7432 7433 7434 7435 7436 7437
	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 ="
7438 7439
	     "%llu\n", le32_to_cpu(fe->i_clusters), new_i_clusters,
	     (unsigned long long)le64_to_cpu(fe->i_size));
7440

7441
	*tc = kzalloc(sizeof(struct ocfs2_truncate_context), GFP_KERNEL);
7442 7443 7444 7445 7446
	if (!(*tc)) {
		status = -ENOMEM;
		mlog_errno(status);
		goto bail;
	}
7447
	ocfs2_init_dealloc_ctxt(&(*tc)->tc_dealloc);
7448 7449

	if (fe->id2.i_list.l_tree_depth) {
7450
		status = ocfs2_read_extent_block(INODE_CACHE(inode),
7451 7452
						 le64_to_cpu(fe->i_last_eb_blk),
						 &last_eb_bh);
7453 7454 7455 7456 7457 7458 7459 7460 7461 7462 7463 7464 7465 7466 7467 7468 7469 7470 7471 7472
		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
7473 7474 7475 7476 7477 7478 7479 7480 7481 7482 7483 7484 7485 7486 7487 7488 7489 7490 7491 7492 7493 7494 7495 7496 7497 7498 7499 7500 7501 7502 7503 7504 7505 7506 7507 7508 7509 7510 7511
/*
 * '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;
	}

7512
	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
7513
				      OCFS2_JOURNAL_ACCESS_WRITE);
M
Mark Fasheh 已提交
7514 7515 7516 7517 7518 7519 7520 7521 7522 7523 7524 7525 7526 7527 7528 7529 7530 7531 7532 7533 7534 7535 7536 7537 7538 7539 7540 7541 7542 7543 7544 7545 7546 7547
	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;
}

7548 7549
static void ocfs2_free_truncate_context(struct ocfs2_truncate_context *tc)
{
7550 7551 7552 7553 7554 7555 7556
	/*
	 * 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");
7557

7558
	brelse(tc->tc_last_eb_bh);
7559 7560 7561

	kfree(tc);
}