Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/* -*- mode: c; c-basic-offset: 8; -*-
   3 * vim: noexpandtab sw=8 ts=8 sts=0:
   4 *
   5 * suballoc.c
   6 *
   7 * metadata alloc and free
   8 * Inspired by ext3 block groups.
   9 *
  10 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
  11 */
  12
  13#include <linux/fs.h>
  14#include <linux/types.h>
  15#include <linux/slab.h>
  16#include <linux/highmem.h>
  17
  18#include <cluster/masklog.h>
  19
  20#include "ocfs2.h"
  21
  22#include "alloc.h"
  23#include "blockcheck.h"
  24#include "dlmglue.h"
  25#include "inode.h"
  26#include "journal.h"
  27#include "localalloc.h"
  28#include "suballoc.h"
  29#include "super.h"
  30#include "sysfile.h"
  31#include "uptodate.h"
  32#include "ocfs2_trace.h"
  33
  34#include "buffer_head_io.h"
  35
  36#define NOT_ALLOC_NEW_GROUP		0
  37#define ALLOC_NEW_GROUP			0x1
  38#define ALLOC_GROUPS_FROM_GLOBAL	0x2
  39
  40#define OCFS2_MAX_TO_STEAL		1024
  41
  42struct ocfs2_suballoc_result {
  43	u64		sr_bg_blkno;	/* The bg we allocated from.  Set
  44					   to 0 when a block group is
  45					   contiguous. */
  46	u64		sr_bg_stable_blkno; /*
  47					     * Doesn't change, always
  48					     * set to target block
  49					     * group descriptor
  50					     * block.
  51					     */
  52	u64		sr_blkno;	/* The first allocated block */
  53	unsigned int	sr_bit_offset;	/* The bit in the bg */
  54	unsigned int	sr_bits;	/* How many bits we claimed */
 
 
 
 
  55};
  56
  57static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
  58{
  59	if (res->sr_blkno == 0)
  60		return 0;
  61
  62	if (res->sr_bg_blkno)
  63		return res->sr_bg_blkno;
  64
  65	return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
  66}
  67
  68static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
  69static int ocfs2_block_group_fill(handle_t *handle,
  70				  struct inode *alloc_inode,
  71				  struct buffer_head *bg_bh,
  72				  u64 group_blkno,
  73				  unsigned int group_clusters,
  74				  u16 my_chain,
  75				  struct ocfs2_chain_list *cl);
  76static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
  77				   struct inode *alloc_inode,
  78				   struct buffer_head *bh,
  79				   u64 max_block,
  80				   u64 *last_alloc_group,
  81				   int flags);
  82
  83static int ocfs2_cluster_group_search(struct inode *inode,
  84				      struct buffer_head *group_bh,
  85				      u32 bits_wanted, u32 min_bits,
  86				      u64 max_block,
  87				      struct ocfs2_suballoc_result *res);
  88static int ocfs2_block_group_search(struct inode *inode,
  89				    struct buffer_head *group_bh,
  90				    u32 bits_wanted, u32 min_bits,
  91				    u64 max_block,
  92				    struct ocfs2_suballoc_result *res);
  93static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
  94				     handle_t *handle,
  95				     u32 bits_wanted,
  96				     u32 min_bits,
  97				     struct ocfs2_suballoc_result *res);
  98static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
  99					 int nr);
 100static int ocfs2_relink_block_group(handle_t *handle,
 101				    struct inode *alloc_inode,
 102				    struct buffer_head *fe_bh,
 103				    struct buffer_head *bg_bh,
 104				    struct buffer_head *prev_bg_bh,
 105				    u16 chain);
 106static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
 107						     u32 wanted);
 108static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
 109						   u64 bg_blkno,
 110						   u16 bg_bit_off);
 111static inline void ocfs2_block_to_cluster_group(struct inode *inode,
 112						u64 data_blkno,
 113						u64 *bg_blkno,
 114						u16 *bg_bit_off);
 115static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
 116					     u32 bits_wanted, u64 max_block,
 117					     int flags,
 118					     struct ocfs2_alloc_context **ac);
 119
 120void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
 121{
 122	struct inode *inode = ac->ac_inode;
 123
 124	if (inode) {
 125		if (ac->ac_which != OCFS2_AC_USE_LOCAL)
 126			ocfs2_inode_unlock(inode, 1);
 127
 128		inode_unlock(inode);
 129
 130		iput(inode);
 131		ac->ac_inode = NULL;
 132	}
 133	brelse(ac->ac_bh);
 134	ac->ac_bh = NULL;
 135	ac->ac_resv = NULL;
 136	kfree(ac->ac_find_loc_priv);
 137	ac->ac_find_loc_priv = NULL;
 138}
 139
 140void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
 141{
 142	ocfs2_free_ac_resource(ac);
 143	kfree(ac);
 144}
 145
 146static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
 147{
 148	return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
 149}
 150
 151#define do_error(fmt, ...)						\
 152do {									\
 153	if (resize)							\
 154		mlog(ML_ERROR, fmt, ##__VA_ARGS__);			\
 155	else								\
 156		return ocfs2_error(sb, fmt, ##__VA_ARGS__);		\
 157} while (0)
 158
 159static int ocfs2_validate_gd_self(struct super_block *sb,
 160				  struct buffer_head *bh,
 161				  int resize)
 162{
 163	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 164
 165	if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
 166		do_error("Group descriptor #%llu has bad signature %.*s\n",
 167			 (unsigned long long)bh->b_blocknr, 7,
 168			 gd->bg_signature);
 169	}
 170
 171	if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
 172		do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
 173			 (unsigned long long)bh->b_blocknr,
 174			 (unsigned long long)le64_to_cpu(gd->bg_blkno));
 175	}
 176
 177	if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
 178		do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
 179			 (unsigned long long)bh->b_blocknr,
 180			 le32_to_cpu(gd->bg_generation));
 181	}
 182
 183	if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
 184		do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
 185			 (unsigned long long)bh->b_blocknr,
 186			 le16_to_cpu(gd->bg_bits),
 187			 le16_to_cpu(gd->bg_free_bits_count));
 188	}
 189
 190	if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
 191		do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
 192			 (unsigned long long)bh->b_blocknr,
 193			 le16_to_cpu(gd->bg_bits),
 194			 8 * le16_to_cpu(gd->bg_size));
 195	}
 196
 197	return 0;
 198}
 199
 200static int ocfs2_validate_gd_parent(struct super_block *sb,
 201				    struct ocfs2_dinode *di,
 202				    struct buffer_head *bh,
 203				    int resize)
 204{
 205	unsigned int max_bits;
 206	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 207
 208	if (di->i_blkno != gd->bg_parent_dinode) {
 209		do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
 210			 (unsigned long long)bh->b_blocknr,
 211			 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
 212			 (unsigned long long)le64_to_cpu(di->i_blkno));
 213	}
 214
 215	max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
 216	if (le16_to_cpu(gd->bg_bits) > max_bits) {
 217		do_error("Group descriptor #%llu has bit count of %u\n",
 218			 (unsigned long long)bh->b_blocknr,
 219			 le16_to_cpu(gd->bg_bits));
 220	}
 221
 222	/* In resize, we may meet the case bg_chain == cl_next_free_rec. */
 223	if ((le16_to_cpu(gd->bg_chain) >
 224	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
 225	    ((le16_to_cpu(gd->bg_chain) ==
 226	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
 227		do_error("Group descriptor #%llu has bad chain %u\n",
 228			 (unsigned long long)bh->b_blocknr,
 229			 le16_to_cpu(gd->bg_chain));
 230	}
 231
 232	return 0;
 233}
 234
 235#undef do_error
 236
 237/*
 238 * This version only prints errors.  It does not fail the filesystem, and
 239 * exists only for resize.
 240 */
 241int ocfs2_check_group_descriptor(struct super_block *sb,
 242				 struct ocfs2_dinode *di,
 243				 struct buffer_head *bh)
 244{
 245	int rc;
 246	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 247
 248	BUG_ON(!buffer_uptodate(bh));
 249
 250	/*
 251	 * If the ecc fails, we return the error but otherwise
 252	 * leave the filesystem running.  We know any error is
 253	 * local to this block.
 254	 */
 255	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
 256	if (rc) {
 257		mlog(ML_ERROR,
 258		     "Checksum failed for group descriptor %llu\n",
 259		     (unsigned long long)bh->b_blocknr);
 260	} else
 261		rc = ocfs2_validate_gd_self(sb, bh, 1);
 262	if (!rc)
 263		rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
 264
 265	return rc;
 266}
 267
 268static int ocfs2_validate_group_descriptor(struct super_block *sb,
 269					   struct buffer_head *bh)
 270{
 271	int rc;
 272	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 273
 274	trace_ocfs2_validate_group_descriptor(
 275					(unsigned long long)bh->b_blocknr);
 276
 277	BUG_ON(!buffer_uptodate(bh));
 278
 279	/*
 280	 * If the ecc fails, we return the error but otherwise
 281	 * leave the filesystem running.  We know any error is
 282	 * local to this block.
 283	 */
 284	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
 285	if (rc)
 286		return rc;
 287
 288	/*
 289	 * Errors after here are fatal.
 290	 */
 291
 292	return ocfs2_validate_gd_self(sb, bh, 0);
 293}
 294
 295int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
 296				u64 gd_blkno, struct buffer_head **bh)
 297{
 298	int rc;
 299	struct buffer_head *tmp = *bh;
 300
 301	rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
 302			      ocfs2_validate_group_descriptor);
 303	if (rc)
 304		goto out;
 305
 306	rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
 307	if (rc) {
 308		brelse(tmp);
 309		goto out;
 310	}
 311
 312	/* If ocfs2_read_block() got us a new bh, pass it up. */
 313	if (!*bh)
 314		*bh = tmp;
 315
 316out:
 317	return rc;
 318}
 319
 320static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
 321					  struct ocfs2_group_desc *bg,
 322					  struct ocfs2_chain_list *cl,
 323					  u64 p_blkno, unsigned int clusters)
 324{
 325	struct ocfs2_extent_list *el = &bg->bg_list;
 326	struct ocfs2_extent_rec *rec;
 327
 328	BUG_ON(!ocfs2_supports_discontig_bg(osb));
 329	if (!el->l_next_free_rec)
 330		el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
 331	rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
 332	rec->e_blkno = cpu_to_le64(p_blkno);
 333	rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
 334				  le16_to_cpu(cl->cl_bpc));
 335	rec->e_leaf_clusters = cpu_to_le16(clusters);
 336	le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
 337	le16_add_cpu(&bg->bg_free_bits_count,
 338		     clusters * le16_to_cpu(cl->cl_bpc));
 339	le16_add_cpu(&el->l_next_free_rec, 1);
 340}
 341
 342static int ocfs2_block_group_fill(handle_t *handle,
 343				  struct inode *alloc_inode,
 344				  struct buffer_head *bg_bh,
 345				  u64 group_blkno,
 346				  unsigned int group_clusters,
 347				  u16 my_chain,
 348				  struct ocfs2_chain_list *cl)
 349{
 350	int status = 0;
 351	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
 352	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 353	struct super_block * sb = alloc_inode->i_sb;
 354
 355	if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
 356		status = ocfs2_error(alloc_inode->i_sb,
 357				     "group block (%llu) != b_blocknr (%llu)\n",
 358				     (unsigned long long)group_blkno,
 359				     (unsigned long long) bg_bh->b_blocknr);
 360		goto bail;
 361	}
 362
 363	status = ocfs2_journal_access_gd(handle,
 364					 INODE_CACHE(alloc_inode),
 365					 bg_bh,
 366					 OCFS2_JOURNAL_ACCESS_CREATE);
 367	if (status < 0) {
 368		mlog_errno(status);
 369		goto bail;
 370	}
 371
 372	memset(bg, 0, sb->s_blocksize);
 373	strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
 374	bg->bg_generation = cpu_to_le32(osb->fs_generation);
 375	bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
 376						osb->s_feature_incompat));
 377	bg->bg_chain = cpu_to_le16(my_chain);
 378	bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
 379	bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
 380	bg->bg_blkno = cpu_to_le64(group_blkno);
 381	if (group_clusters == le16_to_cpu(cl->cl_cpg))
 382		bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
 383	else
 384		ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
 385					      group_clusters);
 386
 387	/* set the 1st bit in the bitmap to account for the descriptor block */
 388	ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
 389	bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
 390
 391	ocfs2_journal_dirty(handle, bg_bh);
 392
 393	/* There is no need to zero out or otherwise initialize the
 394	 * other blocks in a group - All valid FS metadata in a block
 395	 * group stores the superblock fs_generation value at
 396	 * allocation time. */
 397
 398bail:
 399	if (status)
 400		mlog_errno(status);
 401	return status;
 402}
 403
 404static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
 405{
 406	u16 curr, best;
 407
 408	best = curr = 0;
 409	while (curr < le16_to_cpu(cl->cl_count)) {
 410		if (le32_to_cpu(cl->cl_recs[best].c_total) >
 411		    le32_to_cpu(cl->cl_recs[curr].c_total))
 412			best = curr;
 413		curr++;
 414	}
 415	return best;
 416}
 417
 418static struct buffer_head *
 419ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
 420			       struct inode *alloc_inode,
 421			       struct ocfs2_alloc_context *ac,
 422			       struct ocfs2_chain_list *cl)
 423{
 424	int status;
 425	u32 bit_off, num_bits;
 426	u64 bg_blkno;
 427	struct buffer_head *bg_bh;
 428	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
 429
 430	status = ocfs2_claim_clusters(handle, ac,
 431				      le16_to_cpu(cl->cl_cpg), &bit_off,
 432				      &num_bits);
 433	if (status < 0) {
 434		if (status != -ENOSPC)
 435			mlog_errno(status);
 436		goto bail;
 437	}
 438
 439	/* setup the group */
 440	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
 441	trace_ocfs2_block_group_alloc_contig(
 442	     (unsigned long long)bg_blkno, alloc_rec);
 443
 444	bg_bh = sb_getblk(osb->sb, bg_blkno);
 445	if (!bg_bh) {
 446		status = -ENOMEM;
 447		mlog_errno(status);
 448		goto bail;
 449	}
 450	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
 451
 452	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
 453					bg_blkno, num_bits, alloc_rec, cl);
 454	if (status < 0) {
 455		brelse(bg_bh);
 456		mlog_errno(status);
 457	}
 458
 459bail:
 460	return status ? ERR_PTR(status) : bg_bh;
 461}
 462
 463static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
 464					handle_t *handle,
 465					struct ocfs2_alloc_context *ac,
 466					unsigned int min_bits,
 467					u32 *bit_off, u32 *num_bits)
 468{
 469	int status = 0;
 470
 471	while (min_bits) {
 472		status = ocfs2_claim_clusters(handle, ac, min_bits,
 473					      bit_off, num_bits);
 474		if (status != -ENOSPC)
 475			break;
 476
 477		min_bits >>= 1;
 478	}
 479
 480	return status;
 481}
 482
 483static int ocfs2_block_group_grow_discontig(handle_t *handle,
 484					    struct inode *alloc_inode,
 485					    struct buffer_head *bg_bh,
 486					    struct ocfs2_alloc_context *ac,
 487					    struct ocfs2_chain_list *cl,
 488					    unsigned int min_bits)
 489{
 490	int status;
 491	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
 492	struct ocfs2_group_desc *bg =
 493		(struct ocfs2_group_desc *)bg_bh->b_data;
 494	unsigned int needed = le16_to_cpu(cl->cl_cpg) -
 495			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
 496	u32 p_cpos, clusters;
 497	u64 p_blkno;
 498	struct ocfs2_extent_list *el = &bg->bg_list;
 499
 500	status = ocfs2_journal_access_gd(handle,
 501					 INODE_CACHE(alloc_inode),
 502					 bg_bh,
 503					 OCFS2_JOURNAL_ACCESS_CREATE);
 504	if (status < 0) {
 505		mlog_errno(status);
 506		goto bail;
 507	}
 508
 509	while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
 510				le16_to_cpu(el->l_count))) {
 511		if (min_bits > needed)
 512			min_bits = needed;
 513		status = ocfs2_block_group_claim_bits(osb, handle, ac,
 514						      min_bits, &p_cpos,
 515						      &clusters);
 516		if (status < 0) {
 517			if (status != -ENOSPC)
 518				mlog_errno(status);
 519			goto bail;
 520		}
 521		p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
 522		ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
 523					      clusters);
 524
 525		min_bits = clusters;
 526		needed = le16_to_cpu(cl->cl_cpg) -
 527			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
 528	}
 529
 530	if (needed > 0) {
 531		/*
 532		 * We have used up all the extent rec but can't fill up
 533		 * the cpg. So bail out.
 534		 */
 535		status = -ENOSPC;
 536		goto bail;
 537	}
 538
 539	ocfs2_journal_dirty(handle, bg_bh);
 540
 541bail:
 542	return status;
 543}
 544
 545static void ocfs2_bg_alloc_cleanup(handle_t *handle,
 546				   struct ocfs2_alloc_context *cluster_ac,
 547				   struct inode *alloc_inode,
 548				   struct buffer_head *bg_bh)
 549{
 550	int i, ret;
 551	struct ocfs2_group_desc *bg;
 552	struct ocfs2_extent_list *el;
 553	struct ocfs2_extent_rec *rec;
 554
 555	if (!bg_bh)
 556		return;
 557
 558	bg = (struct ocfs2_group_desc *)bg_bh->b_data;
 559	el = &bg->bg_list;
 560	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
 561		rec = &el->l_recs[i];
 562		ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
 563					  cluster_ac->ac_bh,
 564					  le64_to_cpu(rec->e_blkno),
 565					  le16_to_cpu(rec->e_leaf_clusters));
 566		if (ret)
 567			mlog_errno(ret);
 568		/* Try all the clusters to free */
 569	}
 570
 571	ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
 572	brelse(bg_bh);
 573}
 574
 575static struct buffer_head *
 576ocfs2_block_group_alloc_discontig(handle_t *handle,
 577				  struct inode *alloc_inode,
 578				  struct ocfs2_alloc_context *ac,
 579				  struct ocfs2_chain_list *cl)
 580{
 581	int status;
 582	u32 bit_off, num_bits;
 583	u64 bg_blkno;
 584	unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
 585	struct buffer_head *bg_bh = NULL;
 586	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
 587	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
 588
 589	if (!ocfs2_supports_discontig_bg(osb)) {
 590		status = -ENOSPC;
 591		goto bail;
 592	}
 593
 594	status = ocfs2_extend_trans(handle,
 595				    ocfs2_calc_bg_discontig_credits(osb->sb));
 596	if (status) {
 597		mlog_errno(status);
 598		goto bail;
 599	}
 600
 601	/*
 602	 * We're going to be grabbing from multiple cluster groups.
 603	 * We don't have enough credits to relink them all, and the
 604	 * cluster groups will be staying in cache for the duration of
 605	 * this operation.
 606	 */
 607	ac->ac_disable_chain_relink = 1;
 608
 609	/* Claim the first region */
 610	status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
 611					      &bit_off, &num_bits);
 612	if (status < 0) {
 613		if (status != -ENOSPC)
 614			mlog_errno(status);
 615		goto bail;
 616	}
 617	min_bits = num_bits;
 618
 619	/* setup the group */
 620	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
 621	trace_ocfs2_block_group_alloc_discontig(
 622				(unsigned long long)bg_blkno, alloc_rec);
 623
 624	bg_bh = sb_getblk(osb->sb, bg_blkno);
 625	if (!bg_bh) {
 626		status = -ENOMEM;
 627		mlog_errno(status);
 628		goto bail;
 629	}
 630	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
 631
 632	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
 633					bg_blkno, num_bits, alloc_rec, cl);
 634	if (status < 0) {
 635		mlog_errno(status);
 636		goto bail;
 637	}
 638
 639	status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
 640						  bg_bh, ac, cl, min_bits);
 641	if (status)
 642		mlog_errno(status);
 643
 644bail:
 645	if (status)
 646		ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
 647	return status ? ERR_PTR(status) : bg_bh;
 648}
 649
 650/*
 651 * We expect the block group allocator to already be locked.
 652 */
 653static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
 654				   struct inode *alloc_inode,
 655				   struct buffer_head *bh,
 656				   u64 max_block,
 657				   u64 *last_alloc_group,
 658				   int flags)
 659{
 660	int status, credits;
 661	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
 662	struct ocfs2_chain_list *cl;
 663	struct ocfs2_alloc_context *ac = NULL;
 664	handle_t *handle = NULL;
 665	u16 alloc_rec;
 666	struct buffer_head *bg_bh = NULL;
 667	struct ocfs2_group_desc *bg;
 668
 669	BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
 670
 671	cl = &fe->id2.i_chain;
 672	status = ocfs2_reserve_clusters_with_limit(osb,
 673						   le16_to_cpu(cl->cl_cpg),
 674						   max_block, flags, &ac);
 675	if (status < 0) {
 676		if (status != -ENOSPC)
 677			mlog_errno(status);
 678		goto bail;
 679	}
 680
 681	credits = ocfs2_calc_group_alloc_credits(osb->sb,
 682						 le16_to_cpu(cl->cl_cpg));
 683	handle = ocfs2_start_trans(osb, credits);
 684	if (IS_ERR(handle)) {
 685		status = PTR_ERR(handle);
 686		handle = NULL;
 687		mlog_errno(status);
 688		goto bail;
 689	}
 690
 691	if (last_alloc_group && *last_alloc_group != 0) {
 692		trace_ocfs2_block_group_alloc(
 693				(unsigned long long)*last_alloc_group);
 694		ac->ac_last_group = *last_alloc_group;
 695	}
 696
 697	bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
 698					       ac, cl);
 699	if (IS_ERR(bg_bh) && (PTR_ERR(bg_bh) == -ENOSPC))
 700		bg_bh = ocfs2_block_group_alloc_discontig(handle,
 701							  alloc_inode,
 702							  ac, cl);
 703	if (IS_ERR(bg_bh)) {
 704		status = PTR_ERR(bg_bh);
 705		bg_bh = NULL;
 706		if (status != -ENOSPC)
 707			mlog_errno(status);
 708		goto bail;
 709	}
 710	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 711
 712	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
 713					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
 714	if (status < 0) {
 715		mlog_errno(status);
 716		goto bail;
 717	}
 718
 719	alloc_rec = le16_to_cpu(bg->bg_chain);
 720	le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
 721		     le16_to_cpu(bg->bg_free_bits_count));
 722	le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
 723		     le16_to_cpu(bg->bg_bits));
 724	cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
 725	if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
 726		le16_add_cpu(&cl->cl_next_free_rec, 1);
 727
 728	le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
 729					le16_to_cpu(bg->bg_free_bits_count));
 730	le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
 731	le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
 732
 733	ocfs2_journal_dirty(handle, bh);
 734
 735	spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
 736	OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
 737	fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
 738					     le32_to_cpu(fe->i_clusters)));
 739	spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
 740	i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
 741	alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
 742	ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
 743
 744	status = 0;
 745
 746	/* save the new last alloc group so that the caller can cache it. */
 747	if (last_alloc_group)
 748		*last_alloc_group = ac->ac_last_group;
 749
 750bail:
 751	if (handle)
 752		ocfs2_commit_trans(osb, handle);
 753
 754	if (ac)
 755		ocfs2_free_alloc_context(ac);
 756
 757	brelse(bg_bh);
 758
 759	if (status)
 760		mlog_errno(status);
 761	return status;
 762}
 763
 764static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
 765				       struct ocfs2_alloc_context *ac,
 766				       int type,
 767				       u32 slot,
 768				       u64 *last_alloc_group,
 769				       int flags)
 770{
 771	int status;
 772	u32 bits_wanted = ac->ac_bits_wanted;
 773	struct inode *alloc_inode;
 774	struct buffer_head *bh = NULL;
 775	struct ocfs2_dinode *fe;
 776	u32 free_bits;
 777
 778	alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
 779	if (!alloc_inode) {
 780		mlog_errno(-EINVAL);
 781		return -EINVAL;
 782	}
 783
 784	inode_lock(alloc_inode);
 785
 786	status = ocfs2_inode_lock(alloc_inode, &bh, 1);
 787	if (status < 0) {
 788		inode_unlock(alloc_inode);
 789		iput(alloc_inode);
 790
 791		mlog_errno(status);
 792		return status;
 793	}
 794
 795	ac->ac_inode = alloc_inode;
 796	ac->ac_alloc_slot = slot;
 797
 798	fe = (struct ocfs2_dinode *) bh->b_data;
 799
 800	/* The bh was validated by the inode read inside
 801	 * ocfs2_inode_lock().  Any corruption is a code bug. */
 802	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
 803
 804	if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
 805		status = ocfs2_error(alloc_inode->i_sb,
 806				     "Invalid chain allocator %llu\n",
 807				     (unsigned long long)le64_to_cpu(fe->i_blkno));
 808		goto bail;
 809	}
 810
 811	free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
 812		le32_to_cpu(fe->id1.bitmap1.i_used);
 813
 814	if (bits_wanted > free_bits) {
 815		/* cluster bitmap never grows */
 816		if (ocfs2_is_cluster_bitmap(alloc_inode)) {
 817			trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
 818								free_bits);
 819			status = -ENOSPC;
 820			goto bail;
 821		}
 822
 823		if (!(flags & ALLOC_NEW_GROUP)) {
 824			trace_ocfs2_reserve_suballoc_bits_no_new_group(
 825						slot, bits_wanted, free_bits);
 826			status = -ENOSPC;
 827			goto bail;
 828		}
 829
 830		status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
 831						 ac->ac_max_block,
 832						 last_alloc_group, flags);
 833		if (status < 0) {
 834			if (status != -ENOSPC)
 835				mlog_errno(status);
 836			goto bail;
 837		}
 838		atomic_inc(&osb->alloc_stats.bg_extends);
 839
 840		/* You should never ask for this much metadata */
 841		BUG_ON(bits_wanted >
 842		       (le32_to_cpu(fe->id1.bitmap1.i_total)
 843			- le32_to_cpu(fe->id1.bitmap1.i_used)));
 844	}
 845
 846	get_bh(bh);
 847	ac->ac_bh = bh;
 848bail:
 849	brelse(bh);
 850
 851	if (status)
 852		mlog_errno(status);
 853	return status;
 854}
 855
 856static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
 857{
 858	spin_lock(&osb->osb_lock);
 859	osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
 860	spin_unlock(&osb->osb_lock);
 861	atomic_set(&osb->s_num_inodes_stolen, 0);
 862}
 863
 864static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
 865{
 866	spin_lock(&osb->osb_lock);
 867	osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
 868	spin_unlock(&osb->osb_lock);
 869	atomic_set(&osb->s_num_meta_stolen, 0);
 870}
 871
 872void ocfs2_init_steal_slots(struct ocfs2_super *osb)
 873{
 874	ocfs2_init_inode_steal_slot(osb);
 875	ocfs2_init_meta_steal_slot(osb);
 876}
 877
 878static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
 879{
 880	spin_lock(&osb->osb_lock);
 881	if (type == INODE_ALLOC_SYSTEM_INODE)
 882		osb->s_inode_steal_slot = slot;
 883	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
 884		osb->s_meta_steal_slot = slot;
 885	spin_unlock(&osb->osb_lock);
 886}
 887
 888static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
 889{
 890	int slot = OCFS2_INVALID_SLOT;
 891
 892	spin_lock(&osb->osb_lock);
 893	if (type == INODE_ALLOC_SYSTEM_INODE)
 894		slot = osb->s_inode_steal_slot;
 895	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
 896		slot = osb->s_meta_steal_slot;
 897	spin_unlock(&osb->osb_lock);
 898
 899	return slot;
 900}
 901
 902static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
 903{
 904	return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
 905}
 906
 907static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
 908{
 909	return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
 910}
 911
 912static int ocfs2_steal_resource(struct ocfs2_super *osb,
 913				struct ocfs2_alloc_context *ac,
 914				int type)
 915{
 916	int i, status = -ENOSPC;
 917	int slot = __ocfs2_get_steal_slot(osb, type);
 918
 919	/* Start to steal resource from the first slot after ours. */
 920	if (slot == OCFS2_INVALID_SLOT)
 921		slot = osb->slot_num + 1;
 922
 923	for (i = 0; i < osb->max_slots; i++, slot++) {
 924		if (slot == osb->max_slots)
 925			slot = 0;
 926
 927		if (slot == osb->slot_num)
 928			continue;
 929
 930		status = ocfs2_reserve_suballoc_bits(osb, ac,
 931						     type,
 932						     (u32)slot, NULL,
 933						     NOT_ALLOC_NEW_GROUP);
 934		if (status >= 0) {
 935			__ocfs2_set_steal_slot(osb, slot, type);
 936			break;
 937		}
 938
 939		ocfs2_free_ac_resource(ac);
 940	}
 941
 942	return status;
 943}
 944
 945static int ocfs2_steal_inode(struct ocfs2_super *osb,
 946			     struct ocfs2_alloc_context *ac)
 947{
 948	return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
 949}
 950
 951static int ocfs2_steal_meta(struct ocfs2_super *osb,
 952			    struct ocfs2_alloc_context *ac)
 953{
 954	return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
 955}
 956
 957int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
 958				      int blocks,
 959				      struct ocfs2_alloc_context **ac)
 960{
 961	int status;
 962	int slot = ocfs2_get_meta_steal_slot(osb);
 963
 964	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
 965	if (!(*ac)) {
 966		status = -ENOMEM;
 967		mlog_errno(status);
 968		goto bail;
 969	}
 970
 971	(*ac)->ac_bits_wanted = blocks;
 972	(*ac)->ac_which = OCFS2_AC_USE_META;
 973	(*ac)->ac_group_search = ocfs2_block_group_search;
 974
 975	if (slot != OCFS2_INVALID_SLOT &&
 976		atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
 977		goto extent_steal;
 978
 979	atomic_set(&osb->s_num_meta_stolen, 0);
 980	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
 981					     EXTENT_ALLOC_SYSTEM_INODE,
 982					     (u32)osb->slot_num, NULL,
 983					     ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
 984
 985
 986	if (status >= 0) {
 987		status = 0;
 988		if (slot != OCFS2_INVALID_SLOT)
 989			ocfs2_init_meta_steal_slot(osb);
 990		goto bail;
 991	} else if (status < 0 && status != -ENOSPC) {
 992		mlog_errno(status);
 993		goto bail;
 994	}
 995
 996	ocfs2_free_ac_resource(*ac);
 997
 998extent_steal:
 999	status = ocfs2_steal_meta(osb, *ac);
1000	atomic_inc(&osb->s_num_meta_stolen);
1001	if (status < 0) {
1002		if (status != -ENOSPC)
1003			mlog_errno(status);
1004		goto bail;
1005	}
1006
1007	status = 0;
1008bail:
1009	if ((status < 0) && *ac) {
1010		ocfs2_free_alloc_context(*ac);
1011		*ac = NULL;
1012	}
1013
1014	if (status)
1015		mlog_errno(status);
1016	return status;
1017}
1018
1019int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1020			       struct ocfs2_extent_list *root_el,
1021			       struct ocfs2_alloc_context **ac)
1022{
1023	return ocfs2_reserve_new_metadata_blocks(osb,
1024					ocfs2_extend_meta_needed(root_el),
1025					ac);
1026}
1027
1028int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1029			    struct ocfs2_alloc_context **ac)
1030{
1031	int status;
1032	int slot = ocfs2_get_inode_steal_slot(osb);
1033	u64 alloc_group;
1034
1035	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1036	if (!(*ac)) {
1037		status = -ENOMEM;
1038		mlog_errno(status);
1039		goto bail;
1040	}
1041
1042	(*ac)->ac_bits_wanted = 1;
1043	(*ac)->ac_which = OCFS2_AC_USE_INODE;
1044
1045	(*ac)->ac_group_search = ocfs2_block_group_search;
1046
1047	/*
1048	 * stat(2) can't handle i_ino > 32bits, so we tell the
1049	 * lower levels not to allocate us a block group past that
1050	 * limit.  The 'inode64' mount option avoids this behavior.
1051	 */
1052	if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1053		(*ac)->ac_max_block = (u32)~0U;
1054
1055	/*
1056	 * slot is set when we successfully steal inode from other nodes.
1057	 * It is reset in 3 places:
1058	 * 1. when we flush the truncate log
1059	 * 2. when we complete local alloc recovery.
1060	 * 3. when we successfully allocate from our own slot.
1061	 * After it is set, we will go on stealing inodes until we find the
1062	 * need to check our slots to see whether there is some space for us.
1063	 */
1064	if (slot != OCFS2_INVALID_SLOT &&
1065	    atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1066		goto inode_steal;
1067
1068	atomic_set(&osb->s_num_inodes_stolen, 0);
1069	alloc_group = osb->osb_inode_alloc_group;
1070	status = ocfs2_reserve_suballoc_bits(osb, *ac,
1071					     INODE_ALLOC_SYSTEM_INODE,
1072					     (u32)osb->slot_num,
1073					     &alloc_group,
1074					     ALLOC_NEW_GROUP |
1075					     ALLOC_GROUPS_FROM_GLOBAL);
1076	if (status >= 0) {
1077		status = 0;
1078
1079		spin_lock(&osb->osb_lock);
1080		osb->osb_inode_alloc_group = alloc_group;
1081		spin_unlock(&osb->osb_lock);
1082		trace_ocfs2_reserve_new_inode_new_group(
1083			(unsigned long long)alloc_group);
1084
1085		/*
1086		 * Some inodes must be freed by us, so try to allocate
1087		 * from our own next time.
1088		 */
1089		if (slot != OCFS2_INVALID_SLOT)
1090			ocfs2_init_inode_steal_slot(osb);
1091		goto bail;
1092	} else if (status < 0 && status != -ENOSPC) {
1093		mlog_errno(status);
1094		goto bail;
1095	}
1096
1097	ocfs2_free_ac_resource(*ac);
1098
1099inode_steal:
1100	status = ocfs2_steal_inode(osb, *ac);
1101	atomic_inc(&osb->s_num_inodes_stolen);
1102	if (status < 0) {
1103		if (status != -ENOSPC)
1104			mlog_errno(status);
1105		goto bail;
1106	}
1107
1108	status = 0;
1109bail:
1110	if ((status < 0) && *ac) {
1111		ocfs2_free_alloc_context(*ac);
1112		*ac = NULL;
1113	}
1114
1115	if (status)
1116		mlog_errno(status);
1117	return status;
1118}
1119
1120/* local alloc code has to do the same thing, so rather than do this
1121 * twice.. */
1122int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1123				      struct ocfs2_alloc_context *ac)
1124{
1125	int status;
1126
1127	ac->ac_which = OCFS2_AC_USE_MAIN;
1128	ac->ac_group_search = ocfs2_cluster_group_search;
1129
1130	status = ocfs2_reserve_suballoc_bits(osb, ac,
1131					     GLOBAL_BITMAP_SYSTEM_INODE,
1132					     OCFS2_INVALID_SLOT, NULL,
1133					     ALLOC_NEW_GROUP);
1134	if (status < 0 && status != -ENOSPC)
1135		mlog_errno(status);
1136
1137	return status;
1138}
1139
1140/* Callers don't need to care which bitmap (local alloc or main) to
1141 * use so we figure it out for them, but unfortunately this clutters
1142 * things a bit. */
1143static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1144					     u32 bits_wanted, u64 max_block,
1145					     int flags,
1146					     struct ocfs2_alloc_context **ac)
1147{
1148	int status, ret = 0;
1149	int retried = 0;
1150
1151	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1152	if (!(*ac)) {
1153		status = -ENOMEM;
1154		mlog_errno(status);
1155		goto bail;
1156	}
1157
1158	(*ac)->ac_bits_wanted = bits_wanted;
1159	(*ac)->ac_max_block = max_block;
1160
1161	status = -ENOSPC;
1162	if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1163	    ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1164		status = ocfs2_reserve_local_alloc_bits(osb,
1165							bits_wanted,
1166							*ac);
1167		if ((status < 0) && (status != -ENOSPC)) {
1168			mlog_errno(status);
1169			goto bail;
1170		}
1171	}
1172
1173	if (status == -ENOSPC) {
1174retry:
1175		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1176		/* Retry if there is sufficient space cached in truncate log */
1177		if (status == -ENOSPC && !retried) {
1178			retried = 1;
1179			ocfs2_inode_unlock((*ac)->ac_inode, 1);
1180			inode_unlock((*ac)->ac_inode);
1181
1182			ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
1183			if (ret == 1) {
1184				iput((*ac)->ac_inode);
1185				(*ac)->ac_inode = NULL;
1186				goto retry;
1187			}
1188
1189			if (ret < 0)
1190				mlog_errno(ret);
1191
1192			inode_lock((*ac)->ac_inode);
1193			ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
1194			if (ret < 0) {
1195				mlog_errno(ret);
1196				inode_unlock((*ac)->ac_inode);
1197				iput((*ac)->ac_inode);
1198				(*ac)->ac_inode = NULL;
1199				goto bail;
1200			}
1201		}
1202		if (status < 0) {
1203			if (status != -ENOSPC)
1204				mlog_errno(status);
1205			goto bail;
1206		}
1207	}
1208
1209	status = 0;
1210bail:
1211	if ((status < 0) && *ac) {
1212		ocfs2_free_alloc_context(*ac);
1213		*ac = NULL;
1214	}
1215
1216	if (status)
1217		mlog_errno(status);
1218	return status;
1219}
1220
1221int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1222			   u32 bits_wanted,
1223			   struct ocfs2_alloc_context **ac)
1224{
1225	return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1226						 ALLOC_NEW_GROUP, ac);
1227}
1228
1229/*
1230 * More or less lifted from ext3. I'll leave their description below:
1231 *
1232 * "For ext3 allocations, we must not reuse any blocks which are
1233 * allocated in the bitmap buffer's "last committed data" copy.  This
1234 * prevents deletes from freeing up the page for reuse until we have
1235 * committed the delete transaction.
1236 *
1237 * If we didn't do this, then deleting something and reallocating it as
1238 * data would allow the old block to be overwritten before the
1239 * transaction committed (because we force data to disk before commit).
1240 * This would lead to corruption if we crashed between overwriting the
1241 * data and committing the delete.
1242 *
1243 * @@@ We may want to make this allocation behaviour conditional on
1244 * data-writes at some point, and disable it for metadata allocations or
1245 * sync-data inodes."
1246 *
1247 * Note: OCFS2 already does this differently for metadata vs data
1248 * allocations, as those bitmaps are separate and undo access is never
1249 * called on a metadata group descriptor.
1250 */
1251static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1252					 int nr)
1253{
1254	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 
1255	int ret;
1256
1257	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1258		return 0;
1259
1260	if (!buffer_jbd(bg_bh))
 
1261		return 1;
1262
1263	jbd_lock_bh_state(bg_bh);
1264	bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
1265	if (bg)
1266		ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1267	else
1268		ret = 1;
1269	jbd_unlock_bh_state(bg_bh);
 
1270
1271	return ret;
1272}
1273
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1274static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1275					     struct buffer_head *bg_bh,
1276					     unsigned int bits_wanted,
1277					     unsigned int total_bits,
1278					     struct ocfs2_suballoc_result *res)
1279{
1280	void *bitmap;
1281	u16 best_offset, best_size;
 
1282	int offset, start, found, status = 0;
1283	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1284
1285	/* Callers got this descriptor from
1286	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1287	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1288
1289	found = start = best_offset = best_size = 0;
1290	bitmap = bg->bg_bitmap;
1291
1292	while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1293		if (offset == total_bits)
1294			break;
1295
1296		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1297			/* We found a zero, but we can't use it as it
1298			 * hasn't been put to disk yet! */
1299			found = 0;
1300			start = offset + 1;
1301		} else if (offset == start) {
1302			/* we found a zero */
1303			found++;
1304			/* move start to the next bit to test */
1305			start++;
1306		} else {
1307			/* got a zero after some ones */
1308			found = 1;
1309			start = offset + 1;
 
1310		}
1311		if (found > best_size) {
1312			best_size = found;
1313			best_offset = start - found;
1314		}
1315		/* we got everything we needed */
1316		if (found == bits_wanted) {
1317			/* mlog(0, "Found it all!\n"); */
1318			break;
1319		}
1320	}
1321
 
 
1322	if (best_size) {
1323		res->sr_bit_offset = best_offset;
1324		res->sr_bits = best_size;
1325	} else {
1326		status = -ENOSPC;
1327		/* No error log here -- see the comment above
1328		 * ocfs2_test_bg_bit_allocatable */
1329	}
1330
1331	return status;
1332}
1333
1334int ocfs2_block_group_set_bits(handle_t *handle,
1335					     struct inode *alloc_inode,
1336					     struct ocfs2_group_desc *bg,
1337					     struct buffer_head *group_bh,
1338					     unsigned int bit_off,
1339					     unsigned int num_bits)
 
 
1340{
1341	int status;
1342	void *bitmap = bg->bg_bitmap;
1343	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
 
 
 
1344
1345	/* All callers get the descriptor via
1346	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1347	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1348	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1349
1350	trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1351
1352	if (ocfs2_is_cluster_bitmap(alloc_inode))
1353		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1354
1355	status = ocfs2_journal_access_gd(handle,
1356					 INODE_CACHE(alloc_inode),
1357					 group_bh,
1358					 journal_type);
1359	if (status < 0) {
1360		mlog_errno(status);
1361		goto bail;
1362	}
1363
1364	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1365	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1366		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1367				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
1368				   le16_to_cpu(bg->bg_bits),
1369				   le16_to_cpu(bg->bg_free_bits_count),
1370				   num_bits);
1371	}
1372	while(num_bits--)
1373		ocfs2_set_bit(bit_off++, bitmap);
1374
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1375	ocfs2_journal_dirty(handle, group_bh);
1376
1377bail:
1378	return status;
1379}
1380
1381/* find the one with the most empty bits */
1382static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1383{
1384	u16 curr, best;
1385
1386	BUG_ON(!cl->cl_next_free_rec);
1387
1388	best = curr = 0;
1389	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1390		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1391		    le32_to_cpu(cl->cl_recs[best].c_free))
1392			best = curr;
1393		curr++;
1394	}
1395
1396	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1397	return best;
1398}
1399
1400static int ocfs2_relink_block_group(handle_t *handle,
1401				    struct inode *alloc_inode,
1402				    struct buffer_head *fe_bh,
1403				    struct buffer_head *bg_bh,
1404				    struct buffer_head *prev_bg_bh,
1405				    u16 chain)
1406{
1407	int status;
1408	/* there is a really tiny chance the journal calls could fail,
1409	 * but we wouldn't want inconsistent blocks in *any* case. */
1410	u64 bg_ptr, prev_bg_ptr;
1411	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1412	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1413	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1414
1415	/* The caller got these descriptors from
1416	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1417	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1418	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1419
1420	trace_ocfs2_relink_block_group(
1421		(unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1422		(unsigned long long)le64_to_cpu(bg->bg_blkno),
1423		(unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1424
1425	bg_ptr = le64_to_cpu(bg->bg_next_group);
1426	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1427
1428	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1429					 prev_bg_bh,
1430					 OCFS2_JOURNAL_ACCESS_WRITE);
1431	if (status < 0)
1432		goto out;
1433
1434	prev_bg->bg_next_group = bg->bg_next_group;
1435	ocfs2_journal_dirty(handle, prev_bg_bh);
1436
1437	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1438					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1439	if (status < 0)
1440		goto out_rollback_prev_bg;
1441
1442	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1443	ocfs2_journal_dirty(handle, bg_bh);
1444
1445	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1446					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1447	if (status < 0)
1448		goto out_rollback_bg;
1449
1450	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1451	ocfs2_journal_dirty(handle, fe_bh);
1452
1453out:
1454	if (status < 0)
1455		mlog_errno(status);
1456	return status;
1457
1458out_rollback_bg:
1459	bg->bg_next_group = cpu_to_le64(bg_ptr);
1460out_rollback_prev_bg:
1461	prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1462	goto out;
1463}
1464
1465static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1466						     u32 wanted)
1467{
1468	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1469}
1470
1471/* return 0 on success, -ENOSPC to keep searching and any other < 0
1472 * value on error. */
1473static int ocfs2_cluster_group_search(struct inode *inode,
1474				      struct buffer_head *group_bh,
1475				      u32 bits_wanted, u32 min_bits,
1476				      u64 max_block,
1477				      struct ocfs2_suballoc_result *res)
1478{
1479	int search = -ENOSPC;
1480	int ret;
1481	u64 blkoff;
1482	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1483	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1484	unsigned int max_bits, gd_cluster_off;
1485
1486	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1487
1488	if (gd->bg_free_bits_count) {
 
 
 
 
 
1489		max_bits = le16_to_cpu(gd->bg_bits);
1490
1491		/* Tail groups in cluster bitmaps which aren't cpg
1492		 * aligned are prone to partial extension by a failed
1493		 * fs resize. If the file system resize never got to
1494		 * update the dinode cluster count, then we don't want
1495		 * to trust any clusters past it, regardless of what
1496		 * the group descriptor says. */
1497		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1498							  le64_to_cpu(gd->bg_blkno));
1499		if ((gd_cluster_off + max_bits) >
1500		    OCFS2_I(inode)->ip_clusters) {
1501			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1502			trace_ocfs2_cluster_group_search_wrong_max_bits(
1503				(unsigned long long)le64_to_cpu(gd->bg_blkno),
1504				le16_to_cpu(gd->bg_bits),
1505				OCFS2_I(inode)->ip_clusters, max_bits);
1506		}
1507
1508		ret = ocfs2_block_group_find_clear_bits(osb,
1509							group_bh, bits_wanted,
1510							max_bits, res);
1511		if (ret)
1512			return ret;
1513
1514		if (max_block) {
1515			blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1516							  gd_cluster_off +
1517							  res->sr_bit_offset +
1518							  res->sr_bits);
1519			trace_ocfs2_cluster_group_search_max_block(
1520				(unsigned long long)blkoff,
1521				(unsigned long long)max_block);
1522			if (blkoff > max_block)
1523				return -ENOSPC;
1524		}
1525
1526		/* ocfs2_block_group_find_clear_bits() might
1527		 * return success, but we still want to return
1528		 * -ENOSPC unless it found the minimum number
1529		 * of bits. */
1530		if (min_bits <= res->sr_bits)
1531			search = 0; /* success */
1532		else if (res->sr_bits) {
1533			/*
1534			 * Don't show bits which we'll be returning
1535			 * for allocation to the local alloc bitmap.
1536			 */
1537			ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1538		}
1539	}
1540
1541	return search;
1542}
1543
1544static int ocfs2_block_group_search(struct inode *inode,
1545				    struct buffer_head *group_bh,
1546				    u32 bits_wanted, u32 min_bits,
1547				    u64 max_block,
1548				    struct ocfs2_suballoc_result *res)
1549{
1550	int ret = -ENOSPC;
1551	u64 blkoff;
1552	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1553
1554	BUG_ON(min_bits != 1);
1555	BUG_ON(ocfs2_is_cluster_bitmap(inode));
1556
1557	if (bg->bg_free_bits_count) {
1558		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1559							group_bh, bits_wanted,
1560							le16_to_cpu(bg->bg_bits),
1561							res);
1562		if (!ret && max_block) {
1563			blkoff = le64_to_cpu(bg->bg_blkno) +
1564				res->sr_bit_offset + res->sr_bits;
1565			trace_ocfs2_block_group_search_max_block(
1566				(unsigned long long)blkoff,
1567				(unsigned long long)max_block);
1568			if (blkoff > max_block)
1569				ret = -ENOSPC;
1570		}
1571	}
1572
1573	return ret;
1574}
1575
1576int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1577				       handle_t *handle,
1578				       struct buffer_head *di_bh,
1579				       u32 num_bits,
1580				       u16 chain)
1581{
1582	int ret;
1583	u32 tmp_used;
1584	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1585	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1586
1587	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1588				      OCFS2_JOURNAL_ACCESS_WRITE);
1589	if (ret < 0) {
1590		mlog_errno(ret);
1591		goto out;
1592	}
1593
1594	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1595	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1596	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1597	ocfs2_journal_dirty(handle, di_bh);
1598
1599out:
1600	return ret;
1601}
1602
1603void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1604				       struct buffer_head *di_bh,
1605				       u32 num_bits,
1606				       u16 chain)
1607{
1608	u32 tmp_used;
1609	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1610	struct ocfs2_chain_list *cl;
1611
1612	cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1613	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1614	di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1615	le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1616}
1617
1618static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1619					 struct ocfs2_extent_rec *rec,
1620					 struct ocfs2_chain_list *cl)
1621{
1622	unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1623	unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1624	unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1625
1626	if (res->sr_bit_offset < bitoff)
1627		return 0;
1628	if (res->sr_bit_offset >= (bitoff + bitcount))
1629		return 0;
1630	res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1631		(res->sr_bit_offset - bitoff);
1632	if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1633		res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1634	return 1;
1635}
1636
1637static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1638					  struct ocfs2_group_desc *bg,
1639					  struct ocfs2_suballoc_result *res)
1640{
1641	int i;
1642	u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1643	struct ocfs2_extent_rec *rec;
1644	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1645	struct ocfs2_chain_list *cl = &di->id2.i_chain;
1646
1647	if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1648		res->sr_blkno = 0;
1649		return;
1650	}
1651
1652	res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1653	res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1654	if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1655	    !bg->bg_list.l_next_free_rec)
1656		return;
1657
1658	for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1659		rec = &bg->bg_list.l_recs[i];
1660		if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1661			res->sr_bg_blkno = bg_blkno;  /* Restore */
1662			break;
1663		}
1664	}
1665}
1666
1667static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1668				  handle_t *handle,
1669				  u32 bits_wanted,
1670				  u32 min_bits,
1671				  struct ocfs2_suballoc_result *res,
1672				  u16 *bits_left)
1673{
1674	int ret;
1675	struct buffer_head *group_bh = NULL;
1676	struct ocfs2_group_desc *gd;
1677	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1678	struct inode *alloc_inode = ac->ac_inode;
1679
1680	ret = ocfs2_read_group_descriptor(alloc_inode, di,
1681					  res->sr_bg_blkno, &group_bh);
1682	if (ret < 0) {
1683		mlog_errno(ret);
1684		return ret;
1685	}
1686
1687	gd = (struct ocfs2_group_desc *) group_bh->b_data;
1688	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1689				  ac->ac_max_block, res);
1690	if (ret < 0) {
1691		if (ret != -ENOSPC)
1692			mlog_errno(ret);
1693		goto out;
1694	}
1695
1696	if (!ret)
1697		ocfs2_bg_discontig_fix_result(ac, gd, res);
1698
1699	/*
1700	 * sr_bg_blkno might have been changed by
1701	 * ocfs2_bg_discontig_fix_result
1702	 */
1703	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1704
1705	if (ac->ac_find_loc_only)
1706		goto out_loc_only;
1707
1708	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1709					       res->sr_bits,
1710					       le16_to_cpu(gd->bg_chain));
1711	if (ret < 0) {
1712		mlog_errno(ret);
1713		goto out;
1714	}
1715
1716	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1717					 res->sr_bit_offset, res->sr_bits);
 
1718	if (ret < 0) {
1719		ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1720					       res->sr_bits,
1721					       le16_to_cpu(gd->bg_chain));
1722		mlog_errno(ret);
1723	}
1724
1725out_loc_only:
1726	*bits_left = le16_to_cpu(gd->bg_free_bits_count);
1727
1728out:
1729	brelse(group_bh);
1730
1731	return ret;
1732}
1733
1734static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1735			      handle_t *handle,
1736			      u32 bits_wanted,
1737			      u32 min_bits,
1738			      struct ocfs2_suballoc_result *res,
1739			      u16 *bits_left)
1740{
1741	int status;
1742	u16 chain;
1743	u64 next_group;
1744	struct inode *alloc_inode = ac->ac_inode;
1745	struct buffer_head *group_bh = NULL;
1746	struct buffer_head *prev_group_bh = NULL;
1747	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1748	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1749	struct ocfs2_group_desc *bg;
1750
1751	chain = ac->ac_chain;
1752	trace_ocfs2_search_chain_begin(
1753		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1754		bits_wanted, chain);
1755
1756	status = ocfs2_read_group_descriptor(alloc_inode, fe,
1757					     le64_to_cpu(cl->cl_recs[chain].c_blkno),
1758					     &group_bh);
1759	if (status < 0) {
1760		mlog_errno(status);
1761		goto bail;
1762	}
1763	bg = (struct ocfs2_group_desc *) group_bh->b_data;
1764
1765	status = -ENOSPC;
1766	/* for now, the chain search is a bit simplistic. We just use
1767	 * the 1st group with any empty bits. */
1768	while ((status = ac->ac_group_search(alloc_inode, group_bh,
1769					     bits_wanted, min_bits,
1770					     ac->ac_max_block,
1771					     res)) == -ENOSPC) {
1772		if (!bg->bg_next_group)
1773			break;
1774
1775		brelse(prev_group_bh);
1776		prev_group_bh = NULL;
1777
1778		next_group = le64_to_cpu(bg->bg_next_group);
1779		prev_group_bh = group_bh;
1780		group_bh = NULL;
1781		status = ocfs2_read_group_descriptor(alloc_inode, fe,
1782						     next_group, &group_bh);
1783		if (status < 0) {
1784			mlog_errno(status);
1785			goto bail;
1786		}
1787		bg = (struct ocfs2_group_desc *) group_bh->b_data;
1788	}
1789	if (status < 0) {
1790		if (status != -ENOSPC)
1791			mlog_errno(status);
1792		goto bail;
1793	}
1794
1795	trace_ocfs2_search_chain_succ(
1796		(unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1797
1798	res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1799
1800	BUG_ON(res->sr_bits == 0);
1801	if (!status)
1802		ocfs2_bg_discontig_fix_result(ac, bg, res);
1803
1804	/*
1805	 * sr_bg_blkno might have been changed by
1806	 * ocfs2_bg_discontig_fix_result
1807	 */
1808	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1809
1810	/*
1811	 * Keep track of previous block descriptor read. When
1812	 * we find a target, if we have read more than X
1813	 * number of descriptors, and the target is reasonably
1814	 * empty, relink him to top of his chain.
1815	 *
1816	 * We've read 0 extra blocks and only send one more to
1817	 * the transaction, yet the next guy to search has a
1818	 * much easier time.
1819	 *
1820	 * Do this *after* figuring out how many bits we're taking out
1821	 * of our target group.
1822	 */
1823	if (!ac->ac_disable_chain_relink &&
1824	    (prev_group_bh) &&
1825	    (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1826		status = ocfs2_relink_block_group(handle, alloc_inode,
1827						  ac->ac_bh, group_bh,
1828						  prev_group_bh, chain);
1829		if (status < 0) {
1830			mlog_errno(status);
1831			goto bail;
1832		}
1833	}
1834
1835	if (ac->ac_find_loc_only)
1836		goto out_loc_only;
1837
1838	status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1839						  ac->ac_bh, res->sr_bits,
1840						  chain);
1841	if (status) {
1842		mlog_errno(status);
1843		goto bail;
1844	}
1845
1846	status = ocfs2_block_group_set_bits(handle,
1847					    alloc_inode,
1848					    bg,
1849					    group_bh,
1850					    res->sr_bit_offset,
1851					    res->sr_bits);
 
 
1852	if (status < 0) {
1853		ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1854					ac->ac_bh, res->sr_bits, chain);
1855		mlog_errno(status);
1856		goto bail;
1857	}
1858
1859	trace_ocfs2_search_chain_end(
1860			(unsigned long long)le64_to_cpu(fe->i_blkno),
1861			res->sr_bits);
1862
1863out_loc_only:
1864	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
1865bail:
1866	brelse(group_bh);
1867	brelse(prev_group_bh);
1868
1869	if (status)
1870		mlog_errno(status);
1871	return status;
1872}
1873
1874/* will give out up to bits_wanted contiguous bits. */
1875static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1876				     handle_t *handle,
1877				     u32 bits_wanted,
1878				     u32 min_bits,
1879				     struct ocfs2_suballoc_result *res)
1880{
1881	int status;
1882	u16 victim, i;
1883	u16 bits_left = 0;
1884	u64 hint = ac->ac_last_group;
1885	struct ocfs2_chain_list *cl;
1886	struct ocfs2_dinode *fe;
1887
1888	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1889	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1890	BUG_ON(!ac->ac_bh);
1891
1892	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1893
1894	/* The bh was validated by the inode read during
1895	 * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1896	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1897
1898	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1899	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
1900		status = ocfs2_error(ac->ac_inode->i_sb,
1901				     "Chain allocator dinode %llu has %u used bits but only %u total\n",
1902				     (unsigned long long)le64_to_cpu(fe->i_blkno),
1903				     le32_to_cpu(fe->id1.bitmap1.i_used),
1904				     le32_to_cpu(fe->id1.bitmap1.i_total));
1905		goto bail;
1906	}
1907
1908	res->sr_bg_blkno = hint;
1909	if (res->sr_bg_blkno) {
1910		/* Attempt to short-circuit the usual search mechanism
1911		 * by jumping straight to the most recently used
1912		 * allocation group. This helps us maintain some
1913		 * contiguousness across allocations. */
1914		status = ocfs2_search_one_group(ac, handle, bits_wanted,
1915						min_bits, res, &bits_left);
1916		if (!status)
1917			goto set_hint;
1918		if (status < 0 && status != -ENOSPC) {
1919			mlog_errno(status);
1920			goto bail;
1921		}
1922	}
1923
1924	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1925
1926	victim = ocfs2_find_victim_chain(cl);
1927	ac->ac_chain = victim;
1928
1929	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1930				    res, &bits_left);
1931	if (!status) {
1932		if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1933			hint = res->sr_bg_blkno;
1934		else
1935			hint = ocfs2_group_from_res(res);
1936		goto set_hint;
1937	}
1938	if (status < 0 && status != -ENOSPC) {
1939		mlog_errno(status);
1940		goto bail;
1941	}
1942
1943	trace_ocfs2_claim_suballoc_bits(victim);
1944
1945	/* If we didn't pick a good victim, then just default to
1946	 * searching each chain in order. Don't allow chain relinking
1947	 * because we only calculate enough journal credits for one
1948	 * relink per alloc. */
1949	ac->ac_disable_chain_relink = 1;
1950	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1951		if (i == victim)
1952			continue;
1953		if (!cl->cl_recs[i].c_free)
1954			continue;
1955
1956		ac->ac_chain = i;
1957		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1958					    res, &bits_left);
1959		if (!status) {
1960			hint = ocfs2_group_from_res(res);
1961			break;
1962		}
1963		if (status < 0 && status != -ENOSPC) {
1964			mlog_errno(status);
1965			goto bail;
1966		}
1967	}
1968
1969set_hint:
1970	if (status != -ENOSPC) {
1971		/* If the next search of this group is not likely to
1972		 * yield a suitable extent, then we reset the last
1973		 * group hint so as to not waste a disk read */
1974		if (bits_left < min_bits)
1975			ac->ac_last_group = 0;
1976		else
1977			ac->ac_last_group = hint;
1978	}
1979
1980bail:
1981	if (status)
1982		mlog_errno(status);
1983	return status;
1984}
1985
1986int ocfs2_claim_metadata(handle_t *handle,
1987			 struct ocfs2_alloc_context *ac,
1988			 u32 bits_wanted,
1989			 u64 *suballoc_loc,
1990			 u16 *suballoc_bit_start,
1991			 unsigned int *num_bits,
1992			 u64 *blkno_start)
1993{
1994	int status;
1995	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
1996
1997	BUG_ON(!ac);
1998	BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1999	BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2000
2001	status = ocfs2_claim_suballoc_bits(ac,
2002					   handle,
2003					   bits_wanted,
2004					   1,
2005					   &res);
2006	if (status < 0) {
2007		mlog_errno(status);
2008		goto bail;
2009	}
2010	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2011
2012	*suballoc_loc = res.sr_bg_blkno;
2013	*suballoc_bit_start = res.sr_bit_offset;
2014	*blkno_start = res.sr_blkno;
2015	ac->ac_bits_given += res.sr_bits;
2016	*num_bits = res.sr_bits;
2017	status = 0;
2018bail:
2019	if (status)
2020		mlog_errno(status);
2021	return status;
2022}
2023
2024static void ocfs2_init_inode_ac_group(struct inode *dir,
2025				      struct buffer_head *parent_di_bh,
2026				      struct ocfs2_alloc_context *ac)
2027{
2028	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2029	/*
2030	 * Try to allocate inodes from some specific group.
2031	 *
2032	 * If the parent dir has recorded the last group used in allocation,
2033	 * cool, use it. Otherwise if we try to allocate new inode from the
2034	 * same slot the parent dir belongs to, use the same chunk.
2035	 *
2036	 * We are very careful here to avoid the mistake of setting
2037	 * ac_last_group to a group descriptor from a different (unlocked) slot.
2038	 */
2039	if (OCFS2_I(dir)->ip_last_used_group &&
2040	    OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2041		ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2042	else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2043		if (di->i_suballoc_loc)
2044			ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2045		else
2046			ac->ac_last_group = ocfs2_which_suballoc_group(
2047					le64_to_cpu(di->i_blkno),
2048					le16_to_cpu(di->i_suballoc_bit));
2049	}
2050}
2051
2052static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2053					     struct ocfs2_alloc_context *ac)
2054{
2055	OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2056	OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2057}
2058
2059int ocfs2_find_new_inode_loc(struct inode *dir,
2060			     struct buffer_head *parent_fe_bh,
2061			     struct ocfs2_alloc_context *ac,
2062			     u64 *fe_blkno)
2063{
2064	int ret;
2065	handle_t *handle = NULL;
2066	struct ocfs2_suballoc_result *res;
2067
2068	BUG_ON(!ac);
2069	BUG_ON(ac->ac_bits_given != 0);
2070	BUG_ON(ac->ac_bits_wanted != 1);
2071	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2072
2073	res = kzalloc(sizeof(*res), GFP_NOFS);
2074	if (res == NULL) {
2075		ret = -ENOMEM;
2076		mlog_errno(ret);
2077		goto out;
2078	}
2079
2080	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2081
2082	/*
2083	 * The handle started here is for chain relink. Alternatively,
2084	 * we could just disable relink for these calls.
2085	 */
2086	handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2087	if (IS_ERR(handle)) {
2088		ret = PTR_ERR(handle);
2089		handle = NULL;
2090		mlog_errno(ret);
2091		goto out;
2092	}
2093
2094	/*
2095	 * This will instruct ocfs2_claim_suballoc_bits and
2096	 * ocfs2_search_one_group to search but save actual allocation
2097	 * for later.
2098	 */
2099	ac->ac_find_loc_only = 1;
2100
2101	ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2102	if (ret < 0) {
2103		mlog_errno(ret);
2104		goto out;
2105	}
2106
2107	ac->ac_find_loc_priv = res;
2108	*fe_blkno = res->sr_blkno;
2109	ocfs2_update_inode_fsync_trans(handle, dir, 0);
2110out:
2111	if (handle)
2112		ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2113
2114	if (ret)
2115		kfree(res);
2116
2117	return ret;
2118}
2119
2120int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2121				 struct inode *dir,
2122				 struct ocfs2_alloc_context *ac,
2123				 u64 *suballoc_loc,
2124				 u16 *suballoc_bit,
2125				 u64 di_blkno)
2126{
2127	int ret;
2128	u16 chain;
2129	struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2130	struct buffer_head *bg_bh = NULL;
2131	struct ocfs2_group_desc *bg;
2132	struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2133
2134	/*
2135	 * Since di_blkno is being passed back in, we check for any
2136	 * inconsistencies which may have happened between
2137	 * calls. These are code bugs as di_blkno is not expected to
2138	 * change once returned from ocfs2_find_new_inode_loc()
2139	 */
2140	BUG_ON(res->sr_blkno != di_blkno);
2141
2142	ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2143					  res->sr_bg_stable_blkno, &bg_bh);
2144	if (ret) {
2145		mlog_errno(ret);
2146		goto out;
2147	}
2148
2149	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2150	chain = le16_to_cpu(bg->bg_chain);
2151
2152	ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2153					       ac->ac_bh, res->sr_bits,
2154					       chain);
2155	if (ret) {
2156		mlog_errno(ret);
2157		goto out;
2158	}
2159
2160	ret = ocfs2_block_group_set_bits(handle,
2161					 ac->ac_inode,
2162					 bg,
2163					 bg_bh,
2164					 res->sr_bit_offset,
2165					 res->sr_bits);
 
 
2166	if (ret < 0) {
2167		ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2168					       ac->ac_bh, res->sr_bits, chain);
2169		mlog_errno(ret);
2170		goto out;
2171	}
2172
2173	trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2174					   res->sr_bits);
2175
2176	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2177
2178	BUG_ON(res->sr_bits != 1);
2179
2180	*suballoc_loc = res->sr_bg_blkno;
2181	*suballoc_bit = res->sr_bit_offset;
2182	ac->ac_bits_given++;
2183	ocfs2_save_inode_ac_group(dir, ac);
2184
2185out:
2186	brelse(bg_bh);
2187
2188	return ret;
2189}
2190
2191int ocfs2_claim_new_inode(handle_t *handle,
2192			  struct inode *dir,
2193			  struct buffer_head *parent_fe_bh,
2194			  struct ocfs2_alloc_context *ac,
2195			  u64 *suballoc_loc,
2196			  u16 *suballoc_bit,
2197			  u64 *fe_blkno)
2198{
2199	int status;
2200	struct ocfs2_suballoc_result res;
2201
2202	BUG_ON(!ac);
2203	BUG_ON(ac->ac_bits_given != 0);
2204	BUG_ON(ac->ac_bits_wanted != 1);
2205	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2206
2207	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2208
2209	status = ocfs2_claim_suballoc_bits(ac,
2210					   handle,
2211					   1,
2212					   1,
2213					   &res);
2214	if (status < 0) {
2215		mlog_errno(status);
2216		goto bail;
2217	}
2218	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2219
2220	BUG_ON(res.sr_bits != 1);
2221
2222	*suballoc_loc = res.sr_bg_blkno;
2223	*suballoc_bit = res.sr_bit_offset;
2224	*fe_blkno = res.sr_blkno;
2225	ac->ac_bits_given++;
2226	ocfs2_save_inode_ac_group(dir, ac);
2227	status = 0;
2228bail:
2229	if (status)
2230		mlog_errno(status);
2231	return status;
2232}
2233
2234/* translate a group desc. blkno and it's bitmap offset into
2235 * disk cluster offset. */
2236static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2237						   u64 bg_blkno,
2238						   u16 bg_bit_off)
2239{
2240	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2241	u32 cluster = 0;
2242
2243	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2244
2245	if (bg_blkno != osb->first_cluster_group_blkno)
2246		cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2247	cluster += (u32) bg_bit_off;
2248	return cluster;
2249}
2250
2251/* given a cluster offset, calculate which block group it belongs to
2252 * and return that block offset. */
2253u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2254{
2255	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2256	u32 group_no;
2257
2258	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2259
2260	group_no = cluster / osb->bitmap_cpg;
2261	if (!group_no)
2262		return osb->first_cluster_group_blkno;
2263	return ocfs2_clusters_to_blocks(inode->i_sb,
2264					group_no * osb->bitmap_cpg);
2265}
2266
2267/* given the block number of a cluster start, calculate which cluster
2268 * group and descriptor bitmap offset that corresponds to. */
2269static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2270						u64 data_blkno,
2271						u64 *bg_blkno,
2272						u16 *bg_bit_off)
2273{
2274	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2275	u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2276
2277	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2278
2279	*bg_blkno = ocfs2_which_cluster_group(inode,
2280					      data_cluster);
2281
2282	if (*bg_blkno == osb->first_cluster_group_blkno)
2283		*bg_bit_off = (u16) data_cluster;
2284	else
2285		*bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2286							     data_blkno - *bg_blkno);
2287}
2288
2289/*
2290 * min_bits - minimum contiguous chunk from this total allocation we
2291 * can handle. set to what we asked for originally for a full
2292 * contig. allocation, set to '1' to indicate we can deal with extents
2293 * of any size.
2294 */
2295int __ocfs2_claim_clusters(handle_t *handle,
2296			   struct ocfs2_alloc_context *ac,
2297			   u32 min_clusters,
2298			   u32 max_clusters,
2299			   u32 *cluster_start,
2300			   u32 *num_clusters)
2301{
2302	int status;
2303	unsigned int bits_wanted = max_clusters;
2304	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2305	struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2306
2307	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2308
2309	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2310	       && ac->ac_which != OCFS2_AC_USE_MAIN);
2311
2312	if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2313		WARN_ON(min_clusters > 1);
2314
2315		status = ocfs2_claim_local_alloc_bits(osb,
2316						      handle,
2317						      ac,
2318						      bits_wanted,
2319						      cluster_start,
2320						      num_clusters);
2321		if (!status)
2322			atomic_inc(&osb->alloc_stats.local_data);
2323	} else {
2324		if (min_clusters > (osb->bitmap_cpg - 1)) {
2325			/* The only paths asking for contiguousness
2326			 * should know about this already. */
2327			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2328			     "group bitmap size %u!\n", min_clusters,
2329			     osb->bitmap_cpg);
2330			status = -ENOSPC;
2331			goto bail;
2332		}
2333		/* clamp the current request down to a realistic size. */
2334		if (bits_wanted > (osb->bitmap_cpg - 1))
2335			bits_wanted = osb->bitmap_cpg - 1;
2336
2337		status = ocfs2_claim_suballoc_bits(ac,
2338						   handle,
2339						   bits_wanted,
2340						   min_clusters,
2341						   &res);
2342		if (!status) {
2343			BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2344			*cluster_start =
2345				ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2346								 res.sr_bg_blkno,
2347								 res.sr_bit_offset);
2348			atomic_inc(&osb->alloc_stats.bitmap_data);
2349			*num_clusters = res.sr_bits;
2350		}
2351	}
2352	if (status < 0) {
2353		if (status != -ENOSPC)
2354			mlog_errno(status);
2355		goto bail;
2356	}
2357
2358	ac->ac_bits_given += *num_clusters;
2359
2360bail:
2361	if (status)
2362		mlog_errno(status);
2363	return status;
2364}
2365
2366int ocfs2_claim_clusters(handle_t *handle,
2367			 struct ocfs2_alloc_context *ac,
2368			 u32 min_clusters,
2369			 u32 *cluster_start,
2370			 u32 *num_clusters)
2371{
2372	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2373
2374	return __ocfs2_claim_clusters(handle, ac, min_clusters,
2375				      bits_wanted, cluster_start, num_clusters);
2376}
2377
2378static int ocfs2_block_group_clear_bits(handle_t *handle,
2379					struct inode *alloc_inode,
2380					struct ocfs2_group_desc *bg,
2381					struct buffer_head *group_bh,
2382					unsigned int bit_off,
2383					unsigned int num_bits,
 
2384					void (*undo_fn)(unsigned int bit,
2385							unsigned long *bmap))
2386{
2387	int status;
2388	unsigned int tmp;
 
2389	struct ocfs2_group_desc *undo_bg = NULL;
 
2390
2391	/* The caller got this descriptor from
2392	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2393	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2394
2395	trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2396
2397	BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2398	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2399					 group_bh,
2400					 undo_fn ?
2401					 OCFS2_JOURNAL_ACCESS_UNDO :
2402					 OCFS2_JOURNAL_ACCESS_WRITE);
2403	if (status < 0) {
2404		mlog_errno(status);
2405		goto bail;
2406	}
2407
 
2408	if (undo_fn) {
2409		jbd_lock_bh_state(group_bh);
2410		undo_bg = (struct ocfs2_group_desc *)
2411					bh2jh(group_bh)->b_committed_data;
2412		BUG_ON(!undo_bg);
2413	}
2414
2415	tmp = num_bits;
2416	while(tmp--) {
2417		ocfs2_clear_bit((bit_off + tmp),
2418				(unsigned long *) bg->bg_bitmap);
2419		if (undo_fn)
2420			undo_fn(bit_off + tmp,
2421				(unsigned long *) undo_bg->bg_bitmap);
2422	}
2423	le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2424	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2425		if (undo_fn)
2426			jbd_unlock_bh_state(group_bh);
2427		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2428				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
2429				   le16_to_cpu(bg->bg_bits),
2430				   le16_to_cpu(bg->bg_free_bits_count),
2431				   num_bits);
2432	}
2433
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2434	if (undo_fn)
2435		jbd_unlock_bh_state(group_bh);
2436
2437	ocfs2_journal_dirty(handle, group_bh);
2438bail:
2439	return status;
2440}
2441
2442/*
2443 * expects the suballoc inode to already be locked.
2444 */
2445static int _ocfs2_free_suballoc_bits(handle_t *handle,
2446				     struct inode *alloc_inode,
2447				     struct buffer_head *alloc_bh,
2448				     unsigned int start_bit,
2449				     u64 bg_blkno,
2450				     unsigned int count,
2451				     void (*undo_fn)(unsigned int bit,
2452						     unsigned long *bitmap))
2453{
2454	int status = 0;
2455	u32 tmp_used;
2456	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2457	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2458	struct buffer_head *group_bh = NULL;
2459	struct ocfs2_group_desc *group;
 
2460
2461	/* The alloc_bh comes from ocfs2_free_dinode() or
2462	 * ocfs2_free_clusters().  The callers have all locked the
2463	 * allocator and gotten alloc_bh from the lock call.  This
2464	 * validates the dinode buffer.  Any corruption that has happened
2465	 * is a code bug. */
2466	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2467	BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2468
2469	trace_ocfs2_free_suballoc_bits(
2470		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2471		(unsigned long long)bg_blkno,
2472		start_bit, count);
2473
2474	status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2475					     &group_bh);
2476	if (status < 0) {
2477		mlog_errno(status);
2478		goto bail;
2479	}
2480	group = (struct ocfs2_group_desc *) group_bh->b_data;
2481
2482	BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2483
 
 
2484	status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2485					      group, group_bh,
2486					      start_bit, count, undo_fn);
2487	if (status < 0) {
2488		mlog_errno(status);
2489		goto bail;
2490	}
2491
2492	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2493					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2494	if (status < 0) {
2495		mlog_errno(status);
2496		ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2497				start_bit, count);
 
2498		goto bail;
2499	}
2500
2501	le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2502		     count);
2503	tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2504	fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2505	ocfs2_journal_dirty(handle, alloc_bh);
2506
2507bail:
2508	brelse(group_bh);
2509
2510	if (status)
2511		mlog_errno(status);
2512	return status;
2513}
2514
2515int ocfs2_free_suballoc_bits(handle_t *handle,
2516			     struct inode *alloc_inode,
2517			     struct buffer_head *alloc_bh,
2518			     unsigned int start_bit,
2519			     u64 bg_blkno,
2520			     unsigned int count)
2521{
2522	return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2523					 start_bit, bg_blkno, count, NULL);
2524}
2525
2526int ocfs2_free_dinode(handle_t *handle,
2527		      struct inode *inode_alloc_inode,
2528		      struct buffer_head *inode_alloc_bh,
2529		      struct ocfs2_dinode *di)
2530{
2531	u64 blk = le64_to_cpu(di->i_blkno);
2532	u16 bit = le16_to_cpu(di->i_suballoc_bit);
2533	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2534
2535	if (di->i_suballoc_loc)
2536		bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2537	return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2538					inode_alloc_bh, bit, bg_blkno, 1);
2539}
2540
2541static int _ocfs2_free_clusters(handle_t *handle,
2542				struct inode *bitmap_inode,
2543				struct buffer_head *bitmap_bh,
2544				u64 start_blk,
2545				unsigned int num_clusters,
2546				void (*undo_fn)(unsigned int bit,
2547						unsigned long *bitmap))
2548{
2549	int status;
2550	u16 bg_start_bit;
2551	u64 bg_blkno;
2552
2553	/* You can't ever have a contiguous set of clusters
2554	 * bigger than a block group bitmap so we never have to worry
2555	 * about looping on them.
2556	 * This is expensive. We can safely remove once this stuff has
2557	 * gotten tested really well. */
2558	BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb,
2559				ocfs2_blocks_to_clusters(bitmap_inode->i_sb,
2560							 start_blk)));
2561
2562
2563	ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2564				     &bg_start_bit);
2565
2566	trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2567			(unsigned long long)start_blk,
2568			bg_start_bit, num_clusters);
2569
2570	status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2571					   bg_start_bit, bg_blkno,
2572					   num_clusters, undo_fn);
2573	if (status < 0) {
2574		mlog_errno(status);
2575		goto out;
2576	}
2577
2578	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2579					 num_clusters);
2580
2581out:
2582	if (status)
2583		mlog_errno(status);
2584	return status;
2585}
2586
2587int ocfs2_free_clusters(handle_t *handle,
2588			struct inode *bitmap_inode,
2589			struct buffer_head *bitmap_bh,
2590			u64 start_blk,
2591			unsigned int num_clusters)
2592{
2593	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2594				    start_blk, num_clusters,
2595				    _ocfs2_set_bit);
2596}
2597
2598/*
2599 * Give never-used clusters back to the global bitmap.  We don't need
2600 * to protect these bits in the undo buffer.
2601 */
2602int ocfs2_release_clusters(handle_t *handle,
2603			   struct inode *bitmap_inode,
2604			   struct buffer_head *bitmap_bh,
2605			   u64 start_blk,
2606			   unsigned int num_clusters)
2607{
2608	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2609				    start_blk, num_clusters,
2610				    _ocfs2_clear_bit);
2611}
2612
2613/*
2614 * For a given allocation, determine which allocators will need to be
2615 * accessed, and lock them, reserving the appropriate number of bits.
2616 *
2617 * Sparse file systems call this from ocfs2_write_begin_nolock()
2618 * and ocfs2_allocate_unwritten_extents().
2619 *
2620 * File systems which don't support holes call this from
2621 * ocfs2_extend_allocation().
2622 */
2623int ocfs2_lock_allocators(struct inode *inode,
2624			  struct ocfs2_extent_tree *et,
2625			  u32 clusters_to_add, u32 extents_to_split,
2626			  struct ocfs2_alloc_context **data_ac,
2627			  struct ocfs2_alloc_context **meta_ac)
2628{
2629	int ret = 0, num_free_extents;
2630	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2631	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2632
2633	*meta_ac = NULL;
2634	if (data_ac)
2635		*data_ac = NULL;
2636
2637	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2638
2639	num_free_extents = ocfs2_num_free_extents(et);
2640	if (num_free_extents < 0) {
2641		ret = num_free_extents;
2642		mlog_errno(ret);
2643		goto out;
2644	}
2645
2646	/*
2647	 * Sparse allocation file systems need to be more conservative
2648	 * with reserving room for expansion - the actual allocation
2649	 * happens while we've got a journal handle open so re-taking
2650	 * a cluster lock (because we ran out of room for another
2651	 * extent) will violate ordering rules.
2652	 *
2653	 * Most of the time we'll only be seeing this 1 cluster at a time
2654	 * anyway.
2655	 *
2656	 * Always lock for any unwritten extents - we might want to
2657	 * add blocks during a split.
2658	 */
2659	if (!num_free_extents ||
2660	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2661		ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2662		if (ret < 0) {
2663			if (ret != -ENOSPC)
2664				mlog_errno(ret);
2665			goto out;
2666		}
2667	}
2668
2669	if (clusters_to_add == 0)
2670		goto out;
2671
2672	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2673	if (ret < 0) {
2674		if (ret != -ENOSPC)
2675			mlog_errno(ret);
2676		goto out;
2677	}
2678
2679out:
2680	if (ret) {
2681		if (*meta_ac) {
2682			ocfs2_free_alloc_context(*meta_ac);
2683			*meta_ac = NULL;
2684		}
2685
2686		/*
2687		 * We cannot have an error and a non null *data_ac.
2688		 */
2689	}
2690
2691	return ret;
2692}
2693
2694/*
2695 * Read the inode specified by blkno to get suballoc_slot and
2696 * suballoc_bit.
2697 */
2698static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2699				       u16 *suballoc_slot, u64 *group_blkno,
2700				       u16 *suballoc_bit)
2701{
2702	int status;
2703	struct buffer_head *inode_bh = NULL;
2704	struct ocfs2_dinode *inode_fe;
2705
2706	trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2707
2708	/* dirty read disk */
2709	status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2710	if (status < 0) {
2711		mlog(ML_ERROR, "read block %llu failed %d\n",
2712		     (unsigned long long)blkno, status);
2713		goto bail;
2714	}
2715
2716	inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2717	if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2718		mlog(ML_ERROR, "invalid inode %llu requested\n",
2719		     (unsigned long long)blkno);
2720		status = -EINVAL;
2721		goto bail;
2722	}
2723
2724	if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2725	    (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2726		mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2727		     (unsigned long long)blkno,
2728		     (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2729		status = -EINVAL;
2730		goto bail;
2731	}
2732
2733	if (suballoc_slot)
2734		*suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2735	if (suballoc_bit)
2736		*suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2737	if (group_blkno)
2738		*group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2739
2740bail:
2741	brelse(inode_bh);
2742
2743	if (status)
2744		mlog_errno(status);
2745	return status;
2746}
2747
2748/*
2749 * test whether bit is SET in allocator bitmap or not.  on success, 0
2750 * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2751 * is returned and *res is meaningless.  Call this after you have
2752 * cluster locked against suballoc, or you may get a result based on
2753 * non-up2date contents
2754 */
2755static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2756				   struct inode *suballoc,
2757				   struct buffer_head *alloc_bh,
2758				   u64 group_blkno, u64 blkno,
2759				   u16 bit, int *res)
2760{
2761	struct ocfs2_dinode *alloc_di;
2762	struct ocfs2_group_desc *group;
2763	struct buffer_head *group_bh = NULL;
2764	u64 bg_blkno;
2765	int status;
2766
2767	trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2768				      (unsigned int)bit);
2769
2770	alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2771	if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2772		mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2773		     (unsigned int)bit,
2774		     ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2775		status = -EINVAL;
2776		goto bail;
2777	}
2778
2779	bg_blkno = group_blkno ? group_blkno :
2780		   ocfs2_which_suballoc_group(blkno, bit);
2781	status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2782					     &group_bh);
2783	if (status < 0) {
2784		mlog(ML_ERROR, "read group %llu failed %d\n",
2785		     (unsigned long long)bg_blkno, status);
2786		goto bail;
2787	}
2788
2789	group = (struct ocfs2_group_desc *) group_bh->b_data;
2790	*res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2791
2792bail:
2793	brelse(group_bh);
2794
2795	if (status)
2796		mlog_errno(status);
2797	return status;
2798}
2799
2800/*
2801 * Test if the bit representing this inode (blkno) is set in the
2802 * suballocator.
2803 *
2804 * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2805 *
2806 * In the event of failure, a negative value is returned and *res is
2807 * meaningless.
2808 *
2809 * Callers must make sure to hold nfs_sync_lock to prevent
2810 * ocfs2_delete_inode() on another node from accessing the same
2811 * suballocator concurrently.
2812 */
2813int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2814{
2815	int status;
2816	u64 group_blkno = 0;
2817	u16 suballoc_bit = 0, suballoc_slot = 0;
2818	struct inode *inode_alloc_inode;
2819	struct buffer_head *alloc_bh = NULL;
2820
2821	trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2822
2823	status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2824					     &group_blkno, &suballoc_bit);
2825	if (status < 0) {
2826		mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2827		goto bail;
2828	}
2829
2830	inode_alloc_inode =
2831		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2832					    suballoc_slot);
 
 
 
2833	if (!inode_alloc_inode) {
2834		/* the error code could be inaccurate, but we are not able to
2835		 * get the correct one. */
2836		status = -EINVAL;
2837		mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2838		     (u32)suballoc_slot);
2839		goto bail;
2840	}
2841
2842	inode_lock(inode_alloc_inode);
2843	status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2844	if (status < 0) {
2845		inode_unlock(inode_alloc_inode);
2846		iput(inode_alloc_inode);
2847		mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2848		     (u32)suballoc_slot, status);
2849		goto bail;
2850	}
2851
2852	status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2853					 group_blkno, blkno, suballoc_bit, res);
2854	if (status < 0)
2855		mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2856
2857	ocfs2_inode_unlock(inode_alloc_inode, 0);
2858	inode_unlock(inode_alloc_inode);
2859
2860	iput(inode_alloc_inode);
2861	brelse(alloc_bh);
2862bail:
2863	if (status)
2864		mlog_errno(status);
2865	return status;
2866}
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
 
 
   3 * suballoc.c
   4 *
   5 * metadata alloc and free
   6 * Inspired by ext3 block groups.
   7 *
   8 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9 */
  10
  11#include <linux/fs.h>
  12#include <linux/types.h>
  13#include <linux/slab.h>
  14#include <linux/highmem.h>
  15
  16#include <cluster/masklog.h>
  17
  18#include "ocfs2.h"
  19
  20#include "alloc.h"
  21#include "blockcheck.h"
  22#include "dlmglue.h"
  23#include "inode.h"
  24#include "journal.h"
  25#include "localalloc.h"
  26#include "suballoc.h"
  27#include "super.h"
  28#include "sysfile.h"
  29#include "uptodate.h"
  30#include "ocfs2_trace.h"
  31
  32#include "buffer_head_io.h"
  33
  34#define NOT_ALLOC_NEW_GROUP		0
  35#define ALLOC_NEW_GROUP			0x1
  36#define ALLOC_GROUPS_FROM_GLOBAL	0x2
  37
  38#define OCFS2_MAX_TO_STEAL		1024
  39
  40struct ocfs2_suballoc_result {
  41	u64		sr_bg_blkno;	/* The bg we allocated from.  Set
  42					   to 0 when a block group is
  43					   contiguous. */
  44	u64		sr_bg_stable_blkno; /*
  45					     * Doesn't change, always
  46					     * set to target block
  47					     * group descriptor
  48					     * block.
  49					     */
  50	u64		sr_blkno;	/* The first allocated block */
  51	unsigned int	sr_bit_offset;	/* The bit in the bg */
  52	unsigned int	sr_bits;	/* How many bits we claimed */
  53	unsigned int	sr_max_contig_bits; /* The length for contiguous
  54					     * free bits, only available
  55					     * for cluster group
  56					     */
  57};
  58
  59static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
  60{
  61	if (res->sr_blkno == 0)
  62		return 0;
  63
  64	if (res->sr_bg_blkno)
  65		return res->sr_bg_blkno;
  66
  67	return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
  68}
  69
  70static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
  71static int ocfs2_block_group_fill(handle_t *handle,
  72				  struct inode *alloc_inode,
  73				  struct buffer_head *bg_bh,
  74				  u64 group_blkno,
  75				  unsigned int group_clusters,
  76				  u16 my_chain,
  77				  struct ocfs2_chain_list *cl);
  78static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
  79				   struct inode *alloc_inode,
  80				   struct buffer_head *bh,
  81				   u64 max_block,
  82				   u64 *last_alloc_group,
  83				   int flags);
  84
  85static int ocfs2_cluster_group_search(struct inode *inode,
  86				      struct buffer_head *group_bh,
  87				      u32 bits_wanted, u32 min_bits,
  88				      u64 max_block,
  89				      struct ocfs2_suballoc_result *res);
  90static int ocfs2_block_group_search(struct inode *inode,
  91				    struct buffer_head *group_bh,
  92				    u32 bits_wanted, u32 min_bits,
  93				    u64 max_block,
  94				    struct ocfs2_suballoc_result *res);
  95static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
  96				     handle_t *handle,
  97				     u32 bits_wanted,
  98				     u32 min_bits,
  99				     struct ocfs2_suballoc_result *res);
 100static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
 101					 int nr);
 102static int ocfs2_relink_block_group(handle_t *handle,
 103				    struct inode *alloc_inode,
 104				    struct buffer_head *fe_bh,
 105				    struct buffer_head *bg_bh,
 106				    struct buffer_head *prev_bg_bh,
 107				    u16 chain);
 108static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
 109						     u32 wanted);
 110static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
 111						   u64 bg_blkno,
 112						   u16 bg_bit_off);
 113static inline void ocfs2_block_to_cluster_group(struct inode *inode,
 114						u64 data_blkno,
 115						u64 *bg_blkno,
 116						u16 *bg_bit_off);
 117static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
 118					     u32 bits_wanted, u64 max_block,
 119					     int flags,
 120					     struct ocfs2_alloc_context **ac);
 121
 122void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
 123{
 124	struct inode *inode = ac->ac_inode;
 125
 126	if (inode) {
 127		if (ac->ac_which != OCFS2_AC_USE_LOCAL)
 128			ocfs2_inode_unlock(inode, 1);
 129
 130		inode_unlock(inode);
 131
 132		iput(inode);
 133		ac->ac_inode = NULL;
 134	}
 135	brelse(ac->ac_bh);
 136	ac->ac_bh = NULL;
 137	ac->ac_resv = NULL;
 138	kfree(ac->ac_find_loc_priv);
 139	ac->ac_find_loc_priv = NULL;
 140}
 141
 142void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
 143{
 144	ocfs2_free_ac_resource(ac);
 145	kfree(ac);
 146}
 147
 148static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
 149{
 150	return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
 151}
 152
 153#define do_error(fmt, ...)						\
 154do {									\
 155	if (resize)							\
 156		mlog(ML_ERROR, fmt, ##__VA_ARGS__);			\
 157	else								\
 158		return ocfs2_error(sb, fmt, ##__VA_ARGS__);		\
 159} while (0)
 160
 161static int ocfs2_validate_gd_self(struct super_block *sb,
 162				  struct buffer_head *bh,
 163				  int resize)
 164{
 165	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 166
 167	if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
 168		do_error("Group descriptor #%llu has bad signature %.*s\n",
 169			 (unsigned long long)bh->b_blocknr, 7,
 170			 gd->bg_signature);
 171	}
 172
 173	if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
 174		do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
 175			 (unsigned long long)bh->b_blocknr,
 176			 (unsigned long long)le64_to_cpu(gd->bg_blkno));
 177	}
 178
 179	if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
 180		do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
 181			 (unsigned long long)bh->b_blocknr,
 182			 le32_to_cpu(gd->bg_generation));
 183	}
 184
 185	if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
 186		do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
 187			 (unsigned long long)bh->b_blocknr,
 188			 le16_to_cpu(gd->bg_bits),
 189			 le16_to_cpu(gd->bg_free_bits_count));
 190	}
 191
 192	if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
 193		do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
 194			 (unsigned long long)bh->b_blocknr,
 195			 le16_to_cpu(gd->bg_bits),
 196			 8 * le16_to_cpu(gd->bg_size));
 197	}
 198
 199	return 0;
 200}
 201
 202static int ocfs2_validate_gd_parent(struct super_block *sb,
 203				    struct ocfs2_dinode *di,
 204				    struct buffer_head *bh,
 205				    int resize)
 206{
 207	unsigned int max_bits;
 208	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 209
 210	if (di->i_blkno != gd->bg_parent_dinode) {
 211		do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
 212			 (unsigned long long)bh->b_blocknr,
 213			 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
 214			 (unsigned long long)le64_to_cpu(di->i_blkno));
 215	}
 216
 217	max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
 218	if (le16_to_cpu(gd->bg_bits) > max_bits) {
 219		do_error("Group descriptor #%llu has bit count of %u\n",
 220			 (unsigned long long)bh->b_blocknr,
 221			 le16_to_cpu(gd->bg_bits));
 222	}
 223
 224	/* In resize, we may meet the case bg_chain == cl_next_free_rec. */
 225	if ((le16_to_cpu(gd->bg_chain) >
 226	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
 227	    ((le16_to_cpu(gd->bg_chain) ==
 228	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
 229		do_error("Group descriptor #%llu has bad chain %u\n",
 230			 (unsigned long long)bh->b_blocknr,
 231			 le16_to_cpu(gd->bg_chain));
 232	}
 233
 234	return 0;
 235}
 236
 237#undef do_error
 238
 239/*
 240 * This version only prints errors.  It does not fail the filesystem, and
 241 * exists only for resize.
 242 */
 243int ocfs2_check_group_descriptor(struct super_block *sb,
 244				 struct ocfs2_dinode *di,
 245				 struct buffer_head *bh)
 246{
 247	int rc;
 248	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 249
 250	BUG_ON(!buffer_uptodate(bh));
 251
 252	/*
 253	 * If the ecc fails, we return the error but otherwise
 254	 * leave the filesystem running.  We know any error is
 255	 * local to this block.
 256	 */
 257	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
 258	if (rc) {
 259		mlog(ML_ERROR,
 260		     "Checksum failed for group descriptor %llu\n",
 261		     (unsigned long long)bh->b_blocknr);
 262	} else
 263		rc = ocfs2_validate_gd_self(sb, bh, 1);
 264	if (!rc)
 265		rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
 266
 267	return rc;
 268}
 269
 270static int ocfs2_validate_group_descriptor(struct super_block *sb,
 271					   struct buffer_head *bh)
 272{
 273	int rc;
 274	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 275
 276	trace_ocfs2_validate_group_descriptor(
 277					(unsigned long long)bh->b_blocknr);
 278
 279	BUG_ON(!buffer_uptodate(bh));
 280
 281	/*
 282	 * If the ecc fails, we return the error but otherwise
 283	 * leave the filesystem running.  We know any error is
 284	 * local to this block.
 285	 */
 286	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
 287	if (rc)
 288		return rc;
 289
 290	/*
 291	 * Errors after here are fatal.
 292	 */
 293
 294	return ocfs2_validate_gd_self(sb, bh, 0);
 295}
 296
 297int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
 298				u64 gd_blkno, struct buffer_head **bh)
 299{
 300	int rc;
 301	struct buffer_head *tmp = *bh;
 302
 303	rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
 304			      ocfs2_validate_group_descriptor);
 305	if (rc)
 306		goto out;
 307
 308	rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
 309	if (rc) {
 310		brelse(tmp);
 311		goto out;
 312	}
 313
 314	/* If ocfs2_read_block() got us a new bh, pass it up. */
 315	if (!*bh)
 316		*bh = tmp;
 317
 318out:
 319	return rc;
 320}
 321
 322static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
 323					  struct ocfs2_group_desc *bg,
 324					  struct ocfs2_chain_list *cl,
 325					  u64 p_blkno, unsigned int clusters)
 326{
 327	struct ocfs2_extent_list *el = &bg->bg_list;
 328	struct ocfs2_extent_rec *rec;
 329
 330	BUG_ON(!ocfs2_supports_discontig_bg(osb));
 331	if (!el->l_next_free_rec)
 332		el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
 333	rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
 334	rec->e_blkno = cpu_to_le64(p_blkno);
 335	rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
 336				  le16_to_cpu(cl->cl_bpc));
 337	rec->e_leaf_clusters = cpu_to_le16(clusters);
 338	le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
 339	le16_add_cpu(&bg->bg_free_bits_count,
 340		     clusters * le16_to_cpu(cl->cl_bpc));
 341	le16_add_cpu(&el->l_next_free_rec, 1);
 342}
 343
 344static int ocfs2_block_group_fill(handle_t *handle,
 345				  struct inode *alloc_inode,
 346				  struct buffer_head *bg_bh,
 347				  u64 group_blkno,
 348				  unsigned int group_clusters,
 349				  u16 my_chain,
 350				  struct ocfs2_chain_list *cl)
 351{
 352	int status = 0;
 353	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
 354	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 355	struct super_block * sb = alloc_inode->i_sb;
 356
 357	if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
 358		status = ocfs2_error(alloc_inode->i_sb,
 359				     "group block (%llu) != b_blocknr (%llu)\n",
 360				     (unsigned long long)group_blkno,
 361				     (unsigned long long) bg_bh->b_blocknr);
 362		goto bail;
 363	}
 364
 365	status = ocfs2_journal_access_gd(handle,
 366					 INODE_CACHE(alloc_inode),
 367					 bg_bh,
 368					 OCFS2_JOURNAL_ACCESS_CREATE);
 369	if (status < 0) {
 370		mlog_errno(status);
 371		goto bail;
 372	}
 373
 374	memset(bg, 0, sb->s_blocksize);
 375	strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
 376	bg->bg_generation = cpu_to_le32(osb->fs_generation);
 377	bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
 378						osb->s_feature_incompat));
 379	bg->bg_chain = cpu_to_le16(my_chain);
 380	bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
 381	bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
 382	bg->bg_blkno = cpu_to_le64(group_blkno);
 383	if (group_clusters == le16_to_cpu(cl->cl_cpg))
 384		bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
 385	else
 386		ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
 387					      group_clusters);
 388
 389	/* set the 1st bit in the bitmap to account for the descriptor block */
 390	ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
 391	bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
 392
 393	ocfs2_journal_dirty(handle, bg_bh);
 394
 395	/* There is no need to zero out or otherwise initialize the
 396	 * other blocks in a group - All valid FS metadata in a block
 397	 * group stores the superblock fs_generation value at
 398	 * allocation time. */
 399
 400bail:
 401	if (status)
 402		mlog_errno(status);
 403	return status;
 404}
 405
 406static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
 407{
 408	u16 curr, best;
 409
 410	best = curr = 0;
 411	while (curr < le16_to_cpu(cl->cl_count)) {
 412		if (le32_to_cpu(cl->cl_recs[best].c_total) >
 413		    le32_to_cpu(cl->cl_recs[curr].c_total))
 414			best = curr;
 415		curr++;
 416	}
 417	return best;
 418}
 419
 420static struct buffer_head *
 421ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
 422			       struct inode *alloc_inode,
 423			       struct ocfs2_alloc_context *ac,
 424			       struct ocfs2_chain_list *cl)
 425{
 426	int status;
 427	u32 bit_off, num_bits;
 428	u64 bg_blkno;
 429	struct buffer_head *bg_bh;
 430	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
 431
 432	status = ocfs2_claim_clusters(handle, ac,
 433				      le16_to_cpu(cl->cl_cpg), &bit_off,
 434				      &num_bits);
 435	if (status < 0) {
 436		if (status != -ENOSPC)
 437			mlog_errno(status);
 438		goto bail;
 439	}
 440
 441	/* setup the group */
 442	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
 443	trace_ocfs2_block_group_alloc_contig(
 444	     (unsigned long long)bg_blkno, alloc_rec);
 445
 446	bg_bh = sb_getblk(osb->sb, bg_blkno);
 447	if (!bg_bh) {
 448		status = -ENOMEM;
 449		mlog_errno(status);
 450		goto bail;
 451	}
 452	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
 453
 454	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
 455					bg_blkno, num_bits, alloc_rec, cl);
 456	if (status < 0) {
 457		brelse(bg_bh);
 458		mlog_errno(status);
 459	}
 460
 461bail:
 462	return status ? ERR_PTR(status) : bg_bh;
 463}
 464
 465static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
 466					handle_t *handle,
 467					struct ocfs2_alloc_context *ac,
 468					unsigned int min_bits,
 469					u32 *bit_off, u32 *num_bits)
 470{
 471	int status = 0;
 472
 473	while (min_bits) {
 474		status = ocfs2_claim_clusters(handle, ac, min_bits,
 475					      bit_off, num_bits);
 476		if (status != -ENOSPC)
 477			break;
 478
 479		min_bits >>= 1;
 480	}
 481
 482	return status;
 483}
 484
 485static int ocfs2_block_group_grow_discontig(handle_t *handle,
 486					    struct inode *alloc_inode,
 487					    struct buffer_head *bg_bh,
 488					    struct ocfs2_alloc_context *ac,
 489					    struct ocfs2_chain_list *cl,
 490					    unsigned int min_bits)
 491{
 492	int status;
 493	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
 494	struct ocfs2_group_desc *bg =
 495		(struct ocfs2_group_desc *)bg_bh->b_data;
 496	unsigned int needed = le16_to_cpu(cl->cl_cpg) -
 497			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
 498	u32 p_cpos, clusters;
 499	u64 p_blkno;
 500	struct ocfs2_extent_list *el = &bg->bg_list;
 501
 502	status = ocfs2_journal_access_gd(handle,
 503					 INODE_CACHE(alloc_inode),
 504					 bg_bh,
 505					 OCFS2_JOURNAL_ACCESS_CREATE);
 506	if (status < 0) {
 507		mlog_errno(status);
 508		goto bail;
 509	}
 510
 511	while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
 512				le16_to_cpu(el->l_count))) {
 513		if (min_bits > needed)
 514			min_bits = needed;
 515		status = ocfs2_block_group_claim_bits(osb, handle, ac,
 516						      min_bits, &p_cpos,
 517						      &clusters);
 518		if (status < 0) {
 519			if (status != -ENOSPC)
 520				mlog_errno(status);
 521			goto bail;
 522		}
 523		p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
 524		ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
 525					      clusters);
 526
 527		min_bits = clusters;
 528		needed = le16_to_cpu(cl->cl_cpg) -
 529			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
 530	}
 531
 532	if (needed > 0) {
 533		/*
 534		 * We have used up all the extent rec but can't fill up
 535		 * the cpg. So bail out.
 536		 */
 537		status = -ENOSPC;
 538		goto bail;
 539	}
 540
 541	ocfs2_journal_dirty(handle, bg_bh);
 542
 543bail:
 544	return status;
 545}
 546
 547static void ocfs2_bg_alloc_cleanup(handle_t *handle,
 548				   struct ocfs2_alloc_context *cluster_ac,
 549				   struct inode *alloc_inode,
 550				   struct buffer_head *bg_bh)
 551{
 552	int i, ret;
 553	struct ocfs2_group_desc *bg;
 554	struct ocfs2_extent_list *el;
 555	struct ocfs2_extent_rec *rec;
 556
 557	if (!bg_bh)
 558		return;
 559
 560	bg = (struct ocfs2_group_desc *)bg_bh->b_data;
 561	el = &bg->bg_list;
 562	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
 563		rec = &el->l_recs[i];
 564		ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
 565					  cluster_ac->ac_bh,
 566					  le64_to_cpu(rec->e_blkno),
 567					  le16_to_cpu(rec->e_leaf_clusters));
 568		if (ret)
 569			mlog_errno(ret);
 570		/* Try all the clusters to free */
 571	}
 572
 573	ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
 574	brelse(bg_bh);
 575}
 576
 577static struct buffer_head *
 578ocfs2_block_group_alloc_discontig(handle_t *handle,
 579				  struct inode *alloc_inode,
 580				  struct ocfs2_alloc_context *ac,
 581				  struct ocfs2_chain_list *cl)
 582{
 583	int status;
 584	u32 bit_off, num_bits;
 585	u64 bg_blkno;
 586	unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
 587	struct buffer_head *bg_bh = NULL;
 588	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
 589	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
 590
 591	if (!ocfs2_supports_discontig_bg(osb)) {
 592		status = -ENOSPC;
 593		goto bail;
 594	}
 595
 596	status = ocfs2_extend_trans(handle,
 597				    ocfs2_calc_bg_discontig_credits(osb->sb));
 598	if (status) {
 599		mlog_errno(status);
 600		goto bail;
 601	}
 602
 603	/*
 604	 * We're going to be grabbing from multiple cluster groups.
 605	 * We don't have enough credits to relink them all, and the
 606	 * cluster groups will be staying in cache for the duration of
 607	 * this operation.
 608	 */
 609	ac->ac_disable_chain_relink = 1;
 610
 611	/* Claim the first region */
 612	status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
 613					      &bit_off, &num_bits);
 614	if (status < 0) {
 615		if (status != -ENOSPC)
 616			mlog_errno(status);
 617		goto bail;
 618	}
 619	min_bits = num_bits;
 620
 621	/* setup the group */
 622	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
 623	trace_ocfs2_block_group_alloc_discontig(
 624				(unsigned long long)bg_blkno, alloc_rec);
 625
 626	bg_bh = sb_getblk(osb->sb, bg_blkno);
 627	if (!bg_bh) {
 628		status = -ENOMEM;
 629		mlog_errno(status);
 630		goto bail;
 631	}
 632	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
 633
 634	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
 635					bg_blkno, num_bits, alloc_rec, cl);
 636	if (status < 0) {
 637		mlog_errno(status);
 638		goto bail;
 639	}
 640
 641	status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
 642						  bg_bh, ac, cl, min_bits);
 643	if (status)
 644		mlog_errno(status);
 645
 646bail:
 647	if (status)
 648		ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
 649	return status ? ERR_PTR(status) : bg_bh;
 650}
 651
 652/*
 653 * We expect the block group allocator to already be locked.
 654 */
 655static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
 656				   struct inode *alloc_inode,
 657				   struct buffer_head *bh,
 658				   u64 max_block,
 659				   u64 *last_alloc_group,
 660				   int flags)
 661{
 662	int status, credits;
 663	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
 664	struct ocfs2_chain_list *cl;
 665	struct ocfs2_alloc_context *ac = NULL;
 666	handle_t *handle = NULL;
 667	u16 alloc_rec;
 668	struct buffer_head *bg_bh = NULL;
 669	struct ocfs2_group_desc *bg;
 670
 671	BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
 672
 673	cl = &fe->id2.i_chain;
 674	status = ocfs2_reserve_clusters_with_limit(osb,
 675						   le16_to_cpu(cl->cl_cpg),
 676						   max_block, flags, &ac);
 677	if (status < 0) {
 678		if (status != -ENOSPC)
 679			mlog_errno(status);
 680		goto bail;
 681	}
 682
 683	credits = ocfs2_calc_group_alloc_credits(osb->sb,
 684						 le16_to_cpu(cl->cl_cpg));
 685	handle = ocfs2_start_trans(osb, credits);
 686	if (IS_ERR(handle)) {
 687		status = PTR_ERR(handle);
 688		handle = NULL;
 689		mlog_errno(status);
 690		goto bail;
 691	}
 692
 693	if (last_alloc_group && *last_alloc_group != 0) {
 694		trace_ocfs2_block_group_alloc(
 695				(unsigned long long)*last_alloc_group);
 696		ac->ac_last_group = *last_alloc_group;
 697	}
 698
 699	bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
 700					       ac, cl);
 701	if (PTR_ERR(bg_bh) == -ENOSPC)
 702		bg_bh = ocfs2_block_group_alloc_discontig(handle,
 703							  alloc_inode,
 704							  ac, cl);
 705	if (IS_ERR(bg_bh)) {
 706		status = PTR_ERR(bg_bh);
 707		bg_bh = NULL;
 708		if (status != -ENOSPC)
 709			mlog_errno(status);
 710		goto bail;
 711	}
 712	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 713
 714	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
 715					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
 716	if (status < 0) {
 717		mlog_errno(status);
 718		goto bail;
 719	}
 720
 721	alloc_rec = le16_to_cpu(bg->bg_chain);
 722	le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
 723		     le16_to_cpu(bg->bg_free_bits_count));
 724	le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
 725		     le16_to_cpu(bg->bg_bits));
 726	cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
 727	if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
 728		le16_add_cpu(&cl->cl_next_free_rec, 1);
 729
 730	le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
 731					le16_to_cpu(bg->bg_free_bits_count));
 732	le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
 733	le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
 734
 735	ocfs2_journal_dirty(handle, bh);
 736
 737	spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
 738	OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
 739	fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
 740					     le32_to_cpu(fe->i_clusters)));
 741	spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
 742	i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
 743	alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
 744	ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
 745
 746	status = 0;
 747
 748	/* save the new last alloc group so that the caller can cache it. */
 749	if (last_alloc_group)
 750		*last_alloc_group = ac->ac_last_group;
 751
 752bail:
 753	if (handle)
 754		ocfs2_commit_trans(osb, handle);
 755
 756	if (ac)
 757		ocfs2_free_alloc_context(ac);
 758
 759	brelse(bg_bh);
 760
 761	if (status)
 762		mlog_errno(status);
 763	return status;
 764}
 765
 766static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
 767				       struct ocfs2_alloc_context *ac,
 768				       int type,
 769				       u32 slot,
 770				       u64 *last_alloc_group,
 771				       int flags)
 772{
 773	int status;
 774	u32 bits_wanted = ac->ac_bits_wanted;
 775	struct inode *alloc_inode;
 776	struct buffer_head *bh = NULL;
 777	struct ocfs2_dinode *fe;
 778	u32 free_bits;
 779
 780	alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
 781	if (!alloc_inode) {
 782		mlog_errno(-EINVAL);
 783		return -EINVAL;
 784	}
 785
 786	inode_lock(alloc_inode);
 787
 788	status = ocfs2_inode_lock(alloc_inode, &bh, 1);
 789	if (status < 0) {
 790		inode_unlock(alloc_inode);
 791		iput(alloc_inode);
 792
 793		mlog_errno(status);
 794		return status;
 795	}
 796
 797	ac->ac_inode = alloc_inode;
 798	ac->ac_alloc_slot = slot;
 799
 800	fe = (struct ocfs2_dinode *) bh->b_data;
 801
 802	/* The bh was validated by the inode read inside
 803	 * ocfs2_inode_lock().  Any corruption is a code bug. */
 804	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
 805
 806	if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
 807		status = ocfs2_error(alloc_inode->i_sb,
 808				     "Invalid chain allocator %llu\n",
 809				     (unsigned long long)le64_to_cpu(fe->i_blkno));
 810		goto bail;
 811	}
 812
 813	free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
 814		le32_to_cpu(fe->id1.bitmap1.i_used);
 815
 816	if (bits_wanted > free_bits) {
 817		/* cluster bitmap never grows */
 818		if (ocfs2_is_cluster_bitmap(alloc_inode)) {
 819			trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
 820								free_bits);
 821			status = -ENOSPC;
 822			goto bail;
 823		}
 824
 825		if (!(flags & ALLOC_NEW_GROUP)) {
 826			trace_ocfs2_reserve_suballoc_bits_no_new_group(
 827						slot, bits_wanted, free_bits);
 828			status = -ENOSPC;
 829			goto bail;
 830		}
 831
 832		status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
 833						 ac->ac_max_block,
 834						 last_alloc_group, flags);
 835		if (status < 0) {
 836			if (status != -ENOSPC)
 837				mlog_errno(status);
 838			goto bail;
 839		}
 840		atomic_inc(&osb->alloc_stats.bg_extends);
 841
 842		/* You should never ask for this much metadata */
 843		BUG_ON(bits_wanted >
 844		       (le32_to_cpu(fe->id1.bitmap1.i_total)
 845			- le32_to_cpu(fe->id1.bitmap1.i_used)));
 846	}
 847
 848	get_bh(bh);
 849	ac->ac_bh = bh;
 850bail:
 851	brelse(bh);
 852
 853	if (status)
 854		mlog_errno(status);
 855	return status;
 856}
 857
 858static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
 859{
 860	spin_lock(&osb->osb_lock);
 861	osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
 862	spin_unlock(&osb->osb_lock);
 863	atomic_set(&osb->s_num_inodes_stolen, 0);
 864}
 865
 866static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
 867{
 868	spin_lock(&osb->osb_lock);
 869	osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
 870	spin_unlock(&osb->osb_lock);
 871	atomic_set(&osb->s_num_meta_stolen, 0);
 872}
 873
 874void ocfs2_init_steal_slots(struct ocfs2_super *osb)
 875{
 876	ocfs2_init_inode_steal_slot(osb);
 877	ocfs2_init_meta_steal_slot(osb);
 878}
 879
 880static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
 881{
 882	spin_lock(&osb->osb_lock);
 883	if (type == INODE_ALLOC_SYSTEM_INODE)
 884		osb->s_inode_steal_slot = (u16)slot;
 885	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
 886		osb->s_meta_steal_slot = (u16)slot;
 887	spin_unlock(&osb->osb_lock);
 888}
 889
 890static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
 891{
 892	int slot = OCFS2_INVALID_SLOT;
 893
 894	spin_lock(&osb->osb_lock);
 895	if (type == INODE_ALLOC_SYSTEM_INODE)
 896		slot = osb->s_inode_steal_slot;
 897	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
 898		slot = osb->s_meta_steal_slot;
 899	spin_unlock(&osb->osb_lock);
 900
 901	return slot;
 902}
 903
 904static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
 905{
 906	return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
 907}
 908
 909static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
 910{
 911	return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
 912}
 913
 914static int ocfs2_steal_resource(struct ocfs2_super *osb,
 915				struct ocfs2_alloc_context *ac,
 916				int type)
 917{
 918	int i, status = -ENOSPC;
 919	int slot = __ocfs2_get_steal_slot(osb, type);
 920
 921	/* Start to steal resource from the first slot after ours. */
 922	if (slot == OCFS2_INVALID_SLOT)
 923		slot = osb->slot_num + 1;
 924
 925	for (i = 0; i < osb->max_slots; i++, slot++) {
 926		if (slot == osb->max_slots)
 927			slot = 0;
 928
 929		if (slot == osb->slot_num)
 930			continue;
 931
 932		status = ocfs2_reserve_suballoc_bits(osb, ac,
 933						     type,
 934						     (u32)slot, NULL,
 935						     NOT_ALLOC_NEW_GROUP);
 936		if (status >= 0) {
 937			__ocfs2_set_steal_slot(osb, slot, type);
 938			break;
 939		}
 940
 941		ocfs2_free_ac_resource(ac);
 942	}
 943
 944	return status;
 945}
 946
 947static int ocfs2_steal_inode(struct ocfs2_super *osb,
 948			     struct ocfs2_alloc_context *ac)
 949{
 950	return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
 951}
 952
 953static int ocfs2_steal_meta(struct ocfs2_super *osb,
 954			    struct ocfs2_alloc_context *ac)
 955{
 956	return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
 957}
 958
 959int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
 960				      int blocks,
 961				      struct ocfs2_alloc_context **ac)
 962{
 963	int status;
 964	int slot = ocfs2_get_meta_steal_slot(osb);
 965
 966	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
 967	if (!(*ac)) {
 968		status = -ENOMEM;
 969		mlog_errno(status);
 970		goto bail;
 971	}
 972
 973	(*ac)->ac_bits_wanted = blocks;
 974	(*ac)->ac_which = OCFS2_AC_USE_META;
 975	(*ac)->ac_group_search = ocfs2_block_group_search;
 976
 977	if (slot != OCFS2_INVALID_SLOT &&
 978		atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
 979		goto extent_steal;
 980
 981	atomic_set(&osb->s_num_meta_stolen, 0);
 982	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
 983					     EXTENT_ALLOC_SYSTEM_INODE,
 984					     (u32)osb->slot_num, NULL,
 985					     ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
 986
 987
 988	if (status >= 0) {
 989		status = 0;
 990		if (slot != OCFS2_INVALID_SLOT)
 991			ocfs2_init_meta_steal_slot(osb);
 992		goto bail;
 993	} else if (status < 0 && status != -ENOSPC) {
 994		mlog_errno(status);
 995		goto bail;
 996	}
 997
 998	ocfs2_free_ac_resource(*ac);
 999
1000extent_steal:
1001	status = ocfs2_steal_meta(osb, *ac);
1002	atomic_inc(&osb->s_num_meta_stolen);
1003	if (status < 0) {
1004		if (status != -ENOSPC)
1005			mlog_errno(status);
1006		goto bail;
1007	}
1008
1009	status = 0;
1010bail:
1011	if ((status < 0) && *ac) {
1012		ocfs2_free_alloc_context(*ac);
1013		*ac = NULL;
1014	}
1015
1016	if (status)
1017		mlog_errno(status);
1018	return status;
1019}
1020
1021int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1022			       struct ocfs2_extent_list *root_el,
1023			       struct ocfs2_alloc_context **ac)
1024{
1025	return ocfs2_reserve_new_metadata_blocks(osb,
1026					ocfs2_extend_meta_needed(root_el),
1027					ac);
1028}
1029
1030int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1031			    struct ocfs2_alloc_context **ac)
1032{
1033	int status;
1034	int slot = ocfs2_get_inode_steal_slot(osb);
1035	u64 alloc_group;
1036
1037	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1038	if (!(*ac)) {
1039		status = -ENOMEM;
1040		mlog_errno(status);
1041		goto bail;
1042	}
1043
1044	(*ac)->ac_bits_wanted = 1;
1045	(*ac)->ac_which = OCFS2_AC_USE_INODE;
1046
1047	(*ac)->ac_group_search = ocfs2_block_group_search;
1048
1049	/*
1050	 * stat(2) can't handle i_ino > 32bits, so we tell the
1051	 * lower levels not to allocate us a block group past that
1052	 * limit.  The 'inode64' mount option avoids this behavior.
1053	 */
1054	if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1055		(*ac)->ac_max_block = (u32)~0U;
1056
1057	/*
1058	 * slot is set when we successfully steal inode from other nodes.
1059	 * It is reset in 3 places:
1060	 * 1. when we flush the truncate log
1061	 * 2. when we complete local alloc recovery.
1062	 * 3. when we successfully allocate from our own slot.
1063	 * After it is set, we will go on stealing inodes until we find the
1064	 * need to check our slots to see whether there is some space for us.
1065	 */
1066	if (slot != OCFS2_INVALID_SLOT &&
1067	    atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1068		goto inode_steal;
1069
1070	atomic_set(&osb->s_num_inodes_stolen, 0);
1071	alloc_group = osb->osb_inode_alloc_group;
1072	status = ocfs2_reserve_suballoc_bits(osb, *ac,
1073					     INODE_ALLOC_SYSTEM_INODE,
1074					     (u32)osb->slot_num,
1075					     &alloc_group,
1076					     ALLOC_NEW_GROUP |
1077					     ALLOC_GROUPS_FROM_GLOBAL);
1078	if (status >= 0) {
1079		status = 0;
1080
1081		spin_lock(&osb->osb_lock);
1082		osb->osb_inode_alloc_group = alloc_group;
1083		spin_unlock(&osb->osb_lock);
1084		trace_ocfs2_reserve_new_inode_new_group(
1085			(unsigned long long)alloc_group);
1086
1087		/*
1088		 * Some inodes must be freed by us, so try to allocate
1089		 * from our own next time.
1090		 */
1091		if (slot != OCFS2_INVALID_SLOT)
1092			ocfs2_init_inode_steal_slot(osb);
1093		goto bail;
1094	} else if (status < 0 && status != -ENOSPC) {
1095		mlog_errno(status);
1096		goto bail;
1097	}
1098
1099	ocfs2_free_ac_resource(*ac);
1100
1101inode_steal:
1102	status = ocfs2_steal_inode(osb, *ac);
1103	atomic_inc(&osb->s_num_inodes_stolen);
1104	if (status < 0) {
1105		if (status != -ENOSPC)
1106			mlog_errno(status);
1107		goto bail;
1108	}
1109
1110	status = 0;
1111bail:
1112	if ((status < 0) && *ac) {
1113		ocfs2_free_alloc_context(*ac);
1114		*ac = NULL;
1115	}
1116
1117	if (status)
1118		mlog_errno(status);
1119	return status;
1120}
1121
1122/* local alloc code has to do the same thing, so rather than do this
1123 * twice.. */
1124int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1125				      struct ocfs2_alloc_context *ac)
1126{
1127	int status;
1128
1129	ac->ac_which = OCFS2_AC_USE_MAIN;
1130	ac->ac_group_search = ocfs2_cluster_group_search;
1131
1132	status = ocfs2_reserve_suballoc_bits(osb, ac,
1133					     GLOBAL_BITMAP_SYSTEM_INODE,
1134					     OCFS2_INVALID_SLOT, NULL,
1135					     ALLOC_NEW_GROUP);
1136	if (status < 0 && status != -ENOSPC)
1137		mlog_errno(status);
1138
1139	return status;
1140}
1141
1142/* Callers don't need to care which bitmap (local alloc or main) to
1143 * use so we figure it out for them, but unfortunately this clutters
1144 * things a bit. */
1145static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1146					     u32 bits_wanted, u64 max_block,
1147					     int flags,
1148					     struct ocfs2_alloc_context **ac)
1149{
1150	int status, ret = 0;
1151	int retried = 0;
1152
1153	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1154	if (!(*ac)) {
1155		status = -ENOMEM;
1156		mlog_errno(status);
1157		goto bail;
1158	}
1159
1160	(*ac)->ac_bits_wanted = bits_wanted;
1161	(*ac)->ac_max_block = max_block;
1162
1163	status = -ENOSPC;
1164	if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1165	    ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1166		status = ocfs2_reserve_local_alloc_bits(osb,
1167							bits_wanted,
1168							*ac);
1169		if ((status < 0) && (status != -ENOSPC)) {
1170			mlog_errno(status);
1171			goto bail;
1172		}
1173	}
1174
1175	if (status == -ENOSPC) {
1176retry:
1177		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1178		/* Retry if there is sufficient space cached in truncate log */
1179		if (status == -ENOSPC && !retried) {
1180			retried = 1;
1181			ocfs2_inode_unlock((*ac)->ac_inode, 1);
1182			inode_unlock((*ac)->ac_inode);
1183
1184			ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
1185			if (ret == 1) {
1186				iput((*ac)->ac_inode);
1187				(*ac)->ac_inode = NULL;
1188				goto retry;
1189			}
1190
1191			if (ret < 0)
1192				mlog_errno(ret);
1193
1194			inode_lock((*ac)->ac_inode);
1195			ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
1196			if (ret < 0) {
1197				mlog_errno(ret);
1198				inode_unlock((*ac)->ac_inode);
1199				iput((*ac)->ac_inode);
1200				(*ac)->ac_inode = NULL;
1201				goto bail;
1202			}
1203		}
1204		if (status < 0) {
1205			if (status != -ENOSPC)
1206				mlog_errno(status);
1207			goto bail;
1208		}
1209	}
1210
1211	status = 0;
1212bail:
1213	if ((status < 0) && *ac) {
1214		ocfs2_free_alloc_context(*ac);
1215		*ac = NULL;
1216	}
1217
1218	if (status)
1219		mlog_errno(status);
1220	return status;
1221}
1222
1223int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1224			   u32 bits_wanted,
1225			   struct ocfs2_alloc_context **ac)
1226{
1227	return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1228						 ALLOC_NEW_GROUP, ac);
1229}
1230
1231/*
1232 * More or less lifted from ext3. I'll leave their description below:
1233 *
1234 * "For ext3 allocations, we must not reuse any blocks which are
1235 * allocated in the bitmap buffer's "last committed data" copy.  This
1236 * prevents deletes from freeing up the page for reuse until we have
1237 * committed the delete transaction.
1238 *
1239 * If we didn't do this, then deleting something and reallocating it as
1240 * data would allow the old block to be overwritten before the
1241 * transaction committed (because we force data to disk before commit).
1242 * This would lead to corruption if we crashed between overwriting the
1243 * data and committing the delete.
1244 *
1245 * @@@ We may want to make this allocation behaviour conditional on
1246 * data-writes at some point, and disable it for metadata allocations or
1247 * sync-data inodes."
1248 *
1249 * Note: OCFS2 already does this differently for metadata vs data
1250 * allocations, as those bitmaps are separate and undo access is never
1251 * called on a metadata group descriptor.
1252 */
1253static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1254					 int nr)
1255{
1256	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1257	struct journal_head *jh;
1258	int ret;
1259
1260	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1261		return 0;
1262
1263	jh = jbd2_journal_grab_journal_head(bg_bh);
1264	if (!jh)
1265		return 1;
1266
1267	spin_lock(&jh->b_state_lock);
1268	bg = (struct ocfs2_group_desc *) jh->b_committed_data;
1269	if (bg)
1270		ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1271	else
1272		ret = 1;
1273	spin_unlock(&jh->b_state_lock);
1274	jbd2_journal_put_journal_head(jh);
1275
1276	return ret;
1277}
1278
1279u16 ocfs2_find_max_contig_free_bits(void *bitmap,
1280			 u16 total_bits, u16 start)
1281{
1282	u16 offset, free_bits;
1283	u16 contig_bits = 0;
1284
1285	while (start < total_bits) {
1286		offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start);
1287		if (offset == total_bits)
1288			break;
1289
1290		start = ocfs2_find_next_bit(bitmap, total_bits, offset);
1291		free_bits = start - offset;
1292		if (contig_bits < free_bits)
1293			contig_bits = free_bits;
1294	}
1295
1296	return contig_bits;
1297}
1298
1299static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1300					     struct buffer_head *bg_bh,
1301					     unsigned int bits_wanted,
1302					     unsigned int total_bits,
1303					     struct ocfs2_suballoc_result *res)
1304{
1305	void *bitmap;
1306	u16 best_offset, best_size;
1307	u16 prev_best_size = 0;
1308	int offset, start, found, status = 0;
1309	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1310
1311	/* Callers got this descriptor from
1312	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1313	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1314
1315	found = start = best_offset = best_size = 0;
1316	bitmap = bg->bg_bitmap;
1317
1318	while ((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) <
1319	       total_bits) {
 
 
1320		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1321			/* We found a zero, but we can't use it as it
1322			 * hasn't been put to disk yet! */
1323			found = 0;
1324			start = offset + 1;
1325		} else if (offset == start) {
1326			/* we found a zero */
1327			found++;
1328			/* move start to the next bit to test */
1329			start++;
1330		} else {
1331			/* got a zero after some ones */
1332			found = 1;
1333			start = offset + 1;
1334			prev_best_size = best_size;
1335		}
1336		if (found > best_size) {
1337			best_size = found;
1338			best_offset = start - found;
1339		}
1340		/* we got everything we needed */
1341		if (found == bits_wanted) {
1342			/* mlog(0, "Found it all!\n"); */
1343			break;
1344		}
1345	}
1346
1347	/* best_size will be allocated, we save prev_best_size */
1348	res->sr_max_contig_bits = prev_best_size;
1349	if (best_size) {
1350		res->sr_bit_offset = best_offset;
1351		res->sr_bits = best_size;
1352	} else {
1353		status = -ENOSPC;
1354		/* No error log here -- see the comment above
1355		 * ocfs2_test_bg_bit_allocatable */
1356	}
1357
1358	return status;
1359}
1360
1361int ocfs2_block_group_set_bits(handle_t *handle,
1362					     struct inode *alloc_inode,
1363					     struct ocfs2_group_desc *bg,
1364					     struct buffer_head *group_bh,
1365					     unsigned int bit_off,
1366					     unsigned int num_bits,
1367					     unsigned int max_contig_bits,
1368					     int fastpath)
1369{
1370	int status;
1371	void *bitmap = bg->bg_bitmap;
1372	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1373	unsigned int start = bit_off + num_bits;
1374	u16 contig_bits;
1375	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
1376
1377	/* All callers get the descriptor via
1378	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1379	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1380	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1381
1382	trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1383
1384	if (ocfs2_is_cluster_bitmap(alloc_inode))
1385		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1386
1387	status = ocfs2_journal_access_gd(handle,
1388					 INODE_CACHE(alloc_inode),
1389					 group_bh,
1390					 journal_type);
1391	if (status < 0) {
1392		mlog_errno(status);
1393		goto bail;
1394	}
1395
1396	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1397	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1398		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1399				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
1400				   le16_to_cpu(bg->bg_bits),
1401				   le16_to_cpu(bg->bg_free_bits_count),
1402				   num_bits);
1403	}
1404	while(num_bits--)
1405		ocfs2_set_bit(bit_off++, bitmap);
1406
1407	/*
1408	 * this is optimize path, caller set old contig value
1409	 * in max_contig_bits to bypass finding action.
1410	 */
1411	if (fastpath) {
1412		bg->bg_contig_free_bits = cpu_to_le16(max_contig_bits);
1413	} else if (ocfs2_is_cluster_bitmap(alloc_inode)) {
1414		/*
1415		 * Usually, the block group bitmap allocates only 1 bit
1416		 * at a time, while the cluster group allocates n bits
1417		 * each time. Therefore, we only save the contig bits for
1418		 * the cluster group.
1419		 */
1420		contig_bits = ocfs2_find_max_contig_free_bits(bitmap,
1421				    le16_to_cpu(bg->bg_bits), start);
1422		if (contig_bits > max_contig_bits)
1423			max_contig_bits = contig_bits;
1424		bg->bg_contig_free_bits = cpu_to_le16(max_contig_bits);
1425		ocfs2_local_alloc_seen_free_bits(osb, max_contig_bits);
1426	} else {
1427		bg->bg_contig_free_bits = 0;
1428	}
1429
1430	ocfs2_journal_dirty(handle, group_bh);
1431
1432bail:
1433	return status;
1434}
1435
1436/* find the one with the most empty bits */
1437static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1438{
1439	u16 curr, best;
1440
1441	BUG_ON(!cl->cl_next_free_rec);
1442
1443	best = curr = 0;
1444	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1445		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1446		    le32_to_cpu(cl->cl_recs[best].c_free))
1447			best = curr;
1448		curr++;
1449	}
1450
1451	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1452	return best;
1453}
1454
1455static int ocfs2_relink_block_group(handle_t *handle,
1456				    struct inode *alloc_inode,
1457				    struct buffer_head *fe_bh,
1458				    struct buffer_head *bg_bh,
1459				    struct buffer_head *prev_bg_bh,
1460				    u16 chain)
1461{
1462	int status;
1463	/* there is a really tiny chance the journal calls could fail,
1464	 * but we wouldn't want inconsistent blocks in *any* case. */
1465	u64 bg_ptr, prev_bg_ptr;
1466	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1467	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1468	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1469
1470	/* The caller got these descriptors from
1471	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1472	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1473	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1474
1475	trace_ocfs2_relink_block_group(
1476		(unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1477		(unsigned long long)le64_to_cpu(bg->bg_blkno),
1478		(unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1479
1480	bg_ptr = le64_to_cpu(bg->bg_next_group);
1481	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1482
1483	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1484					 prev_bg_bh,
1485					 OCFS2_JOURNAL_ACCESS_WRITE);
1486	if (status < 0)
1487		goto out;
1488
1489	prev_bg->bg_next_group = bg->bg_next_group;
1490	ocfs2_journal_dirty(handle, prev_bg_bh);
1491
1492	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1493					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1494	if (status < 0)
1495		goto out_rollback_prev_bg;
1496
1497	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1498	ocfs2_journal_dirty(handle, bg_bh);
1499
1500	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1501					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1502	if (status < 0)
1503		goto out_rollback_bg;
1504
1505	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1506	ocfs2_journal_dirty(handle, fe_bh);
1507
1508out:
1509	if (status < 0)
1510		mlog_errno(status);
1511	return status;
1512
1513out_rollback_bg:
1514	bg->bg_next_group = cpu_to_le64(bg_ptr);
1515out_rollback_prev_bg:
1516	prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1517	goto out;
1518}
1519
1520static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1521						     u32 wanted)
1522{
1523	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1524}
1525
1526/* return 0 on success, -ENOSPC to keep searching and any other < 0
1527 * value on error. */
1528static int ocfs2_cluster_group_search(struct inode *inode,
1529				      struct buffer_head *group_bh,
1530				      u32 bits_wanted, u32 min_bits,
1531				      u64 max_block,
1532				      struct ocfs2_suballoc_result *res)
1533{
1534	int search = -ENOSPC;
1535	int ret;
1536	u64 blkoff;
1537	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1538	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1539	unsigned int max_bits, gd_cluster_off;
1540
1541	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1542
1543	if (le16_to_cpu(gd->bg_contig_free_bits) &&
1544	    le16_to_cpu(gd->bg_contig_free_bits) < bits_wanted)
1545		return -ENOSPC;
1546
1547	/* ->bg_contig_free_bits may un-initialized, so compare again */
1548	if (le16_to_cpu(gd->bg_free_bits_count) >= bits_wanted) {
1549		max_bits = le16_to_cpu(gd->bg_bits);
1550
1551		/* Tail groups in cluster bitmaps which aren't cpg
1552		 * aligned are prone to partial extension by a failed
1553		 * fs resize. If the file system resize never got to
1554		 * update the dinode cluster count, then we don't want
1555		 * to trust any clusters past it, regardless of what
1556		 * the group descriptor says. */
1557		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1558							  le64_to_cpu(gd->bg_blkno));
1559		if ((gd_cluster_off + max_bits) >
1560		    OCFS2_I(inode)->ip_clusters) {
1561			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1562			trace_ocfs2_cluster_group_search_wrong_max_bits(
1563				(unsigned long long)le64_to_cpu(gd->bg_blkno),
1564				le16_to_cpu(gd->bg_bits),
1565				OCFS2_I(inode)->ip_clusters, max_bits);
1566		}
1567
1568		ret = ocfs2_block_group_find_clear_bits(osb,
1569							group_bh, bits_wanted,
1570							max_bits, res);
1571		if (ret)
1572			return ret;
1573
1574		if (max_block) {
1575			blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1576							  gd_cluster_off +
1577							  res->sr_bit_offset +
1578							  res->sr_bits);
1579			trace_ocfs2_cluster_group_search_max_block(
1580				(unsigned long long)blkoff,
1581				(unsigned long long)max_block);
1582			if (blkoff > max_block)
1583				return -ENOSPC;
1584		}
1585
1586		/* ocfs2_block_group_find_clear_bits() might
1587		 * return success, but we still want to return
1588		 * -ENOSPC unless it found the minimum number
1589		 * of bits. */
1590		if (min_bits <= res->sr_bits)
1591			search = 0; /* success */
 
 
 
 
 
 
 
1592	}
1593
1594	return search;
1595}
1596
1597static int ocfs2_block_group_search(struct inode *inode,
1598				    struct buffer_head *group_bh,
1599				    u32 bits_wanted, u32 min_bits,
1600				    u64 max_block,
1601				    struct ocfs2_suballoc_result *res)
1602{
1603	int ret = -ENOSPC;
1604	u64 blkoff;
1605	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1606
1607	BUG_ON(min_bits != 1);
1608	BUG_ON(ocfs2_is_cluster_bitmap(inode));
1609
1610	if (le16_to_cpu(bg->bg_free_bits_count) >= bits_wanted) {
1611		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1612							group_bh, bits_wanted,
1613							le16_to_cpu(bg->bg_bits),
1614							res);
1615		if (!ret && max_block) {
1616			blkoff = le64_to_cpu(bg->bg_blkno) +
1617				res->sr_bit_offset + res->sr_bits;
1618			trace_ocfs2_block_group_search_max_block(
1619				(unsigned long long)blkoff,
1620				(unsigned long long)max_block);
1621			if (blkoff > max_block)
1622				ret = -ENOSPC;
1623		}
1624	}
1625
1626	return ret;
1627}
1628
1629int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1630				       handle_t *handle,
1631				       struct buffer_head *di_bh,
1632				       u32 num_bits,
1633				       u16 chain)
1634{
1635	int ret;
1636	u32 tmp_used;
1637	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1638	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1639
1640	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1641				      OCFS2_JOURNAL_ACCESS_WRITE);
1642	if (ret < 0) {
1643		mlog_errno(ret);
1644		goto out;
1645	}
1646
1647	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1648	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1649	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1650	ocfs2_journal_dirty(handle, di_bh);
1651
1652out:
1653	return ret;
1654}
1655
1656void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1657				       struct buffer_head *di_bh,
1658				       u32 num_bits,
1659				       u16 chain)
1660{
1661	u32 tmp_used;
1662	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1663	struct ocfs2_chain_list *cl;
1664
1665	cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1666	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1667	di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1668	le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1669}
1670
1671static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1672					 struct ocfs2_extent_rec *rec,
1673					 struct ocfs2_chain_list *cl)
1674{
1675	unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1676	unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1677	unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1678
1679	if (res->sr_bit_offset < bitoff)
1680		return 0;
1681	if (res->sr_bit_offset >= (bitoff + bitcount))
1682		return 0;
1683	res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1684		(res->sr_bit_offset - bitoff);
1685	if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1686		res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1687	return 1;
1688}
1689
1690static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1691					  struct ocfs2_group_desc *bg,
1692					  struct ocfs2_suballoc_result *res)
1693{
1694	int i;
1695	u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1696	struct ocfs2_extent_rec *rec;
1697	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1698	struct ocfs2_chain_list *cl = &di->id2.i_chain;
1699
1700	if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1701		res->sr_blkno = 0;
1702		return;
1703	}
1704
1705	res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1706	res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1707	if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1708	    !bg->bg_list.l_next_free_rec)
1709		return;
1710
1711	for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1712		rec = &bg->bg_list.l_recs[i];
1713		if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1714			res->sr_bg_blkno = bg_blkno;  /* Restore */
1715			break;
1716		}
1717	}
1718}
1719
1720static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1721				  handle_t *handle,
1722				  u32 bits_wanted,
1723				  u32 min_bits,
1724				  struct ocfs2_suballoc_result *res,
1725				  u16 *bits_left)
1726{
1727	int ret;
1728	struct buffer_head *group_bh = NULL;
1729	struct ocfs2_group_desc *gd;
1730	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1731	struct inode *alloc_inode = ac->ac_inode;
1732
1733	ret = ocfs2_read_group_descriptor(alloc_inode, di,
1734					  res->sr_bg_blkno, &group_bh);
1735	if (ret < 0) {
1736		mlog_errno(ret);
1737		return ret;
1738	}
1739
1740	gd = (struct ocfs2_group_desc *) group_bh->b_data;
1741	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1742				  ac->ac_max_block, res);
1743	if (ret < 0) {
1744		if (ret != -ENOSPC)
1745			mlog_errno(ret);
1746		goto out;
1747	}
1748
1749	if (!ret)
1750		ocfs2_bg_discontig_fix_result(ac, gd, res);
1751
1752	/*
1753	 * sr_bg_blkno might have been changed by
1754	 * ocfs2_bg_discontig_fix_result
1755	 */
1756	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1757
1758	if (ac->ac_find_loc_only)
1759		goto out_loc_only;
1760
1761	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1762					       res->sr_bits,
1763					       le16_to_cpu(gd->bg_chain));
1764	if (ret < 0) {
1765		mlog_errno(ret);
1766		goto out;
1767	}
1768
1769	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1770					 res->sr_bit_offset, res->sr_bits,
1771					 res->sr_max_contig_bits, 0);
1772	if (ret < 0) {
1773		ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1774					       res->sr_bits,
1775					       le16_to_cpu(gd->bg_chain));
1776		mlog_errno(ret);
1777	}
1778
1779out_loc_only:
1780	*bits_left = le16_to_cpu(gd->bg_free_bits_count);
1781
1782out:
1783	brelse(group_bh);
1784
1785	return ret;
1786}
1787
1788static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1789			      handle_t *handle,
1790			      u32 bits_wanted,
1791			      u32 min_bits,
1792			      struct ocfs2_suballoc_result *res,
1793			      u16 *bits_left)
1794{
1795	int status;
1796	u16 chain;
1797	u64 next_group;
1798	struct inode *alloc_inode = ac->ac_inode;
1799	struct buffer_head *group_bh = NULL;
1800	struct buffer_head *prev_group_bh = NULL;
1801	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1802	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1803	struct ocfs2_group_desc *bg;
1804
1805	chain = ac->ac_chain;
1806	trace_ocfs2_search_chain_begin(
1807		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1808		bits_wanted, chain);
1809
1810	status = ocfs2_read_group_descriptor(alloc_inode, fe,
1811					     le64_to_cpu(cl->cl_recs[chain].c_blkno),
1812					     &group_bh);
1813	if (status < 0) {
1814		mlog_errno(status);
1815		goto bail;
1816	}
1817	bg = (struct ocfs2_group_desc *) group_bh->b_data;
1818
1819	status = -ENOSPC;
1820	/* for now, the chain search is a bit simplistic. We just use
1821	 * the 1st group with any empty bits. */
1822	while ((status = ac->ac_group_search(alloc_inode, group_bh,
1823					     bits_wanted, min_bits,
1824					     ac->ac_max_block,
1825					     res)) == -ENOSPC) {
1826		if (!bg->bg_next_group)
1827			break;
1828
1829		brelse(prev_group_bh);
1830		prev_group_bh = NULL;
1831
1832		next_group = le64_to_cpu(bg->bg_next_group);
1833		prev_group_bh = group_bh;
1834		group_bh = NULL;
1835		status = ocfs2_read_group_descriptor(alloc_inode, fe,
1836						     next_group, &group_bh);
1837		if (status < 0) {
1838			mlog_errno(status);
1839			goto bail;
1840		}
1841		bg = (struct ocfs2_group_desc *) group_bh->b_data;
1842	}
1843	if (status < 0) {
1844		if (status != -ENOSPC)
1845			mlog_errno(status);
1846		goto bail;
1847	}
1848
1849	trace_ocfs2_search_chain_succ(
1850		(unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1851
1852	res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1853
1854	BUG_ON(res->sr_bits == 0);
1855	if (!status)
1856		ocfs2_bg_discontig_fix_result(ac, bg, res);
1857
1858	/*
1859	 * sr_bg_blkno might have been changed by
1860	 * ocfs2_bg_discontig_fix_result
1861	 */
1862	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1863
1864	/*
1865	 * Keep track of previous block descriptor read. When
1866	 * we find a target, if we have read more than X
1867	 * number of descriptors, and the target is reasonably
1868	 * empty, relink him to top of his chain.
1869	 *
1870	 * We've read 0 extra blocks and only send one more to
1871	 * the transaction, yet the next guy to search has a
1872	 * much easier time.
1873	 *
1874	 * Do this *after* figuring out how many bits we're taking out
1875	 * of our target group.
1876	 */
1877	if (!ac->ac_disable_chain_relink &&
1878	    (prev_group_bh) &&
1879	    (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1880		status = ocfs2_relink_block_group(handle, alloc_inode,
1881						  ac->ac_bh, group_bh,
1882						  prev_group_bh, chain);
1883		if (status < 0) {
1884			mlog_errno(status);
1885			goto bail;
1886		}
1887	}
1888
1889	if (ac->ac_find_loc_only)
1890		goto out_loc_only;
1891
1892	status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1893						  ac->ac_bh, res->sr_bits,
1894						  chain);
1895	if (status) {
1896		mlog_errno(status);
1897		goto bail;
1898	}
1899
1900	status = ocfs2_block_group_set_bits(handle,
1901					    alloc_inode,
1902					    bg,
1903					    group_bh,
1904					    res->sr_bit_offset,
1905					    res->sr_bits,
1906					    res->sr_max_contig_bits,
1907					    0);
1908	if (status < 0) {
1909		ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1910					ac->ac_bh, res->sr_bits, chain);
1911		mlog_errno(status);
1912		goto bail;
1913	}
1914
1915	trace_ocfs2_search_chain_end(
1916			(unsigned long long)le64_to_cpu(fe->i_blkno),
1917			res->sr_bits);
1918
1919out_loc_only:
1920	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
1921bail:
1922	brelse(group_bh);
1923	brelse(prev_group_bh);
1924
1925	if (status)
1926		mlog_errno(status);
1927	return status;
1928}
1929
1930/* will give out up to bits_wanted contiguous bits. */
1931static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1932				     handle_t *handle,
1933				     u32 bits_wanted,
1934				     u32 min_bits,
1935				     struct ocfs2_suballoc_result *res)
1936{
1937	int status;
1938	u16 victim, i;
1939	u16 bits_left = 0;
1940	u64 hint = ac->ac_last_group;
1941	struct ocfs2_chain_list *cl;
1942	struct ocfs2_dinode *fe;
1943
1944	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1945	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1946	BUG_ON(!ac->ac_bh);
1947
1948	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1949
1950	/* The bh was validated by the inode read during
1951	 * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1952	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1953
1954	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1955	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
1956		status = ocfs2_error(ac->ac_inode->i_sb,
1957				     "Chain allocator dinode %llu has %u used bits but only %u total\n",
1958				     (unsigned long long)le64_to_cpu(fe->i_blkno),
1959				     le32_to_cpu(fe->id1.bitmap1.i_used),
1960				     le32_to_cpu(fe->id1.bitmap1.i_total));
1961		goto bail;
1962	}
1963
1964	res->sr_bg_blkno = hint;
1965	if (res->sr_bg_blkno) {
1966		/* Attempt to short-circuit the usual search mechanism
1967		 * by jumping straight to the most recently used
1968		 * allocation group. This helps us maintain some
1969		 * contiguousness across allocations. */
1970		status = ocfs2_search_one_group(ac, handle, bits_wanted,
1971						min_bits, res, &bits_left);
1972		if (!status)
1973			goto set_hint;
1974		if (status < 0 && status != -ENOSPC) {
1975			mlog_errno(status);
1976			goto bail;
1977		}
1978	}
1979
1980	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1981
1982	victim = ocfs2_find_victim_chain(cl);
1983	ac->ac_chain = victim;
1984
1985	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1986				    res, &bits_left);
1987	if (!status) {
1988		if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1989			hint = res->sr_bg_blkno;
1990		else
1991			hint = ocfs2_group_from_res(res);
1992		goto set_hint;
1993	}
1994	if (status < 0 && status != -ENOSPC) {
1995		mlog_errno(status);
1996		goto bail;
1997	}
1998
1999	trace_ocfs2_claim_suballoc_bits(victim);
2000
2001	/* If we didn't pick a good victim, then just default to
2002	 * searching each chain in order. Don't allow chain relinking
2003	 * because we only calculate enough journal credits for one
2004	 * relink per alloc. */
2005	ac->ac_disable_chain_relink = 1;
2006	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
2007		if (i == victim)
2008			continue;
2009		if (le32_to_cpu(cl->cl_recs[i].c_free) < bits_wanted)
2010			continue;
2011
2012		ac->ac_chain = i;
2013		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
2014					    res, &bits_left);
2015		if (!status) {
2016			hint = ocfs2_group_from_res(res);
2017			break;
2018		}
2019		if (status < 0 && status != -ENOSPC) {
2020			mlog_errno(status);
2021			goto bail;
2022		}
2023	}
2024
2025set_hint:
2026	if (status != -ENOSPC) {
2027		/* If the next search of this group is not likely to
2028		 * yield a suitable extent, then we reset the last
2029		 * group hint so as to not waste a disk read */
2030		if (bits_left < min_bits)
2031			ac->ac_last_group = 0;
2032		else
2033			ac->ac_last_group = hint;
2034	}
2035
2036bail:
2037	if (status)
2038		mlog_errno(status);
2039	return status;
2040}
2041
2042int ocfs2_claim_metadata(handle_t *handle,
2043			 struct ocfs2_alloc_context *ac,
2044			 u32 bits_wanted,
2045			 u64 *suballoc_loc,
2046			 u16 *suballoc_bit_start,
2047			 unsigned int *num_bits,
2048			 u64 *blkno_start)
2049{
2050	int status;
2051	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2052
2053	BUG_ON(!ac);
2054	BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2055	BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2056
2057	status = ocfs2_claim_suballoc_bits(ac,
2058					   handle,
2059					   bits_wanted,
2060					   1,
2061					   &res);
2062	if (status < 0) {
2063		mlog_errno(status);
2064		goto bail;
2065	}
2066	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2067
2068	*suballoc_loc = res.sr_bg_blkno;
2069	*suballoc_bit_start = res.sr_bit_offset;
2070	*blkno_start = res.sr_blkno;
2071	ac->ac_bits_given += res.sr_bits;
2072	*num_bits = res.sr_bits;
2073	status = 0;
2074bail:
2075	if (status)
2076		mlog_errno(status);
2077	return status;
2078}
2079
2080static void ocfs2_init_inode_ac_group(struct inode *dir,
2081				      struct buffer_head *parent_di_bh,
2082				      struct ocfs2_alloc_context *ac)
2083{
2084	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2085	/*
2086	 * Try to allocate inodes from some specific group.
2087	 *
2088	 * If the parent dir has recorded the last group used in allocation,
2089	 * cool, use it. Otherwise if we try to allocate new inode from the
2090	 * same slot the parent dir belongs to, use the same chunk.
2091	 *
2092	 * We are very careful here to avoid the mistake of setting
2093	 * ac_last_group to a group descriptor from a different (unlocked) slot.
2094	 */
2095	if (OCFS2_I(dir)->ip_last_used_group &&
2096	    OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2097		ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2098	else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2099		if (di->i_suballoc_loc)
2100			ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2101		else
2102			ac->ac_last_group = ocfs2_which_suballoc_group(
2103					le64_to_cpu(di->i_blkno),
2104					le16_to_cpu(di->i_suballoc_bit));
2105	}
2106}
2107
2108static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2109					     struct ocfs2_alloc_context *ac)
2110{
2111	OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2112	OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2113}
2114
2115int ocfs2_find_new_inode_loc(struct inode *dir,
2116			     struct buffer_head *parent_fe_bh,
2117			     struct ocfs2_alloc_context *ac,
2118			     u64 *fe_blkno)
2119{
2120	int ret;
2121	handle_t *handle = NULL;
2122	struct ocfs2_suballoc_result *res;
2123
2124	BUG_ON(!ac);
2125	BUG_ON(ac->ac_bits_given != 0);
2126	BUG_ON(ac->ac_bits_wanted != 1);
2127	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2128
2129	res = kzalloc(sizeof(*res), GFP_NOFS);
2130	if (res == NULL) {
2131		ret = -ENOMEM;
2132		mlog_errno(ret);
2133		goto out;
2134	}
2135
2136	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2137
2138	/*
2139	 * The handle started here is for chain relink. Alternatively,
2140	 * we could just disable relink for these calls.
2141	 */
2142	handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2143	if (IS_ERR(handle)) {
2144		ret = PTR_ERR(handle);
2145		handle = NULL;
2146		mlog_errno(ret);
2147		goto out;
2148	}
2149
2150	/*
2151	 * This will instruct ocfs2_claim_suballoc_bits and
2152	 * ocfs2_search_one_group to search but save actual allocation
2153	 * for later.
2154	 */
2155	ac->ac_find_loc_only = 1;
2156
2157	ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2158	if (ret < 0) {
2159		mlog_errno(ret);
2160		goto out;
2161	}
2162
2163	ac->ac_find_loc_priv = res;
2164	*fe_blkno = res->sr_blkno;
2165	ocfs2_update_inode_fsync_trans(handle, dir, 0);
2166out:
2167	if (handle)
2168		ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2169
2170	if (ret)
2171		kfree(res);
2172
2173	return ret;
2174}
2175
2176int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2177				 struct inode *dir,
2178				 struct ocfs2_alloc_context *ac,
2179				 u64 *suballoc_loc,
2180				 u16 *suballoc_bit,
2181				 u64 di_blkno)
2182{
2183	int ret;
2184	u16 chain;
2185	struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2186	struct buffer_head *bg_bh = NULL;
2187	struct ocfs2_group_desc *bg;
2188	struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2189
2190	/*
2191	 * Since di_blkno is being passed back in, we check for any
2192	 * inconsistencies which may have happened between
2193	 * calls. These are code bugs as di_blkno is not expected to
2194	 * change once returned from ocfs2_find_new_inode_loc()
2195	 */
2196	BUG_ON(res->sr_blkno != di_blkno);
2197
2198	ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2199					  res->sr_bg_stable_blkno, &bg_bh);
2200	if (ret) {
2201		mlog_errno(ret);
2202		goto out;
2203	}
2204
2205	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2206	chain = le16_to_cpu(bg->bg_chain);
2207
2208	ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2209					       ac->ac_bh, res->sr_bits,
2210					       chain);
2211	if (ret) {
2212		mlog_errno(ret);
2213		goto out;
2214	}
2215
2216	ret = ocfs2_block_group_set_bits(handle,
2217					 ac->ac_inode,
2218					 bg,
2219					 bg_bh,
2220					 res->sr_bit_offset,
2221					 res->sr_bits,
2222					 res->sr_max_contig_bits,
2223					 0);
2224	if (ret < 0) {
2225		ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2226					       ac->ac_bh, res->sr_bits, chain);
2227		mlog_errno(ret);
2228		goto out;
2229	}
2230
2231	trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2232					   res->sr_bits);
2233
2234	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2235
2236	BUG_ON(res->sr_bits != 1);
2237
2238	*suballoc_loc = res->sr_bg_blkno;
2239	*suballoc_bit = res->sr_bit_offset;
2240	ac->ac_bits_given++;
2241	ocfs2_save_inode_ac_group(dir, ac);
2242
2243out:
2244	brelse(bg_bh);
2245
2246	return ret;
2247}
2248
2249int ocfs2_claim_new_inode(handle_t *handle,
2250			  struct inode *dir,
2251			  struct buffer_head *parent_fe_bh,
2252			  struct ocfs2_alloc_context *ac,
2253			  u64 *suballoc_loc,
2254			  u16 *suballoc_bit,
2255			  u64 *fe_blkno)
2256{
2257	int status;
2258	struct ocfs2_suballoc_result res;
2259
2260	BUG_ON(!ac);
2261	BUG_ON(ac->ac_bits_given != 0);
2262	BUG_ON(ac->ac_bits_wanted != 1);
2263	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2264
2265	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2266
2267	status = ocfs2_claim_suballoc_bits(ac,
2268					   handle,
2269					   1,
2270					   1,
2271					   &res);
2272	if (status < 0) {
2273		mlog_errno(status);
2274		goto bail;
2275	}
2276	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2277
2278	BUG_ON(res.sr_bits != 1);
2279
2280	*suballoc_loc = res.sr_bg_blkno;
2281	*suballoc_bit = res.sr_bit_offset;
2282	*fe_blkno = res.sr_blkno;
2283	ac->ac_bits_given++;
2284	ocfs2_save_inode_ac_group(dir, ac);
2285	status = 0;
2286bail:
2287	if (status)
2288		mlog_errno(status);
2289	return status;
2290}
2291
2292/* translate a group desc. blkno and it's bitmap offset into
2293 * disk cluster offset. */
2294static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2295						   u64 bg_blkno,
2296						   u16 bg_bit_off)
2297{
2298	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2299	u32 cluster = 0;
2300
2301	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2302
2303	if (bg_blkno != osb->first_cluster_group_blkno)
2304		cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2305	cluster += (u32) bg_bit_off;
2306	return cluster;
2307}
2308
2309/* given a cluster offset, calculate which block group it belongs to
2310 * and return that block offset. */
2311u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2312{
2313	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2314	u32 group_no;
2315
2316	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2317
2318	group_no = cluster / osb->bitmap_cpg;
2319	if (!group_no)
2320		return osb->first_cluster_group_blkno;
2321	return ocfs2_clusters_to_blocks(inode->i_sb,
2322					group_no * osb->bitmap_cpg);
2323}
2324
2325/* given the block number of a cluster start, calculate which cluster
2326 * group and descriptor bitmap offset that corresponds to. */
2327static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2328						u64 data_blkno,
2329						u64 *bg_blkno,
2330						u16 *bg_bit_off)
2331{
2332	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2333	u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2334
2335	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2336
2337	*bg_blkno = ocfs2_which_cluster_group(inode,
2338					      data_cluster);
2339
2340	if (*bg_blkno == osb->first_cluster_group_blkno)
2341		*bg_bit_off = (u16) data_cluster;
2342	else
2343		*bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2344							     data_blkno - *bg_blkno);
2345}
2346
2347/*
2348 * min_bits - minimum contiguous chunk from this total allocation we
2349 * can handle. set to what we asked for originally for a full
2350 * contig. allocation, set to '1' to indicate we can deal with extents
2351 * of any size.
2352 */
2353int __ocfs2_claim_clusters(handle_t *handle,
2354			   struct ocfs2_alloc_context *ac,
2355			   u32 min_clusters,
2356			   u32 max_clusters,
2357			   u32 *cluster_start,
2358			   u32 *num_clusters)
2359{
2360	int status;
2361	unsigned int bits_wanted = max_clusters;
2362	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2363	struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2364
2365	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2366
2367	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2368	       && ac->ac_which != OCFS2_AC_USE_MAIN);
2369
2370	if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2371		WARN_ON(min_clusters > 1);
2372
2373		status = ocfs2_claim_local_alloc_bits(osb,
2374						      handle,
2375						      ac,
2376						      bits_wanted,
2377						      cluster_start,
2378						      num_clusters);
2379		if (!status)
2380			atomic_inc(&osb->alloc_stats.local_data);
2381	} else {
2382		if (min_clusters > (osb->bitmap_cpg - 1)) {
2383			/* The only paths asking for contiguousness
2384			 * should know about this already. */
2385			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2386			     "group bitmap size %u!\n", min_clusters,
2387			     osb->bitmap_cpg);
2388			status = -ENOSPC;
2389			goto bail;
2390		}
2391		/* clamp the current request down to a realistic size. */
2392		if (bits_wanted > (osb->bitmap_cpg - 1))
2393			bits_wanted = osb->bitmap_cpg - 1;
2394
2395		status = ocfs2_claim_suballoc_bits(ac,
2396						   handle,
2397						   bits_wanted,
2398						   min_clusters,
2399						   &res);
2400		if (!status) {
2401			BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2402			*cluster_start =
2403				ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2404								 res.sr_bg_blkno,
2405								 res.sr_bit_offset);
2406			atomic_inc(&osb->alloc_stats.bitmap_data);
2407			*num_clusters = res.sr_bits;
2408		}
2409	}
2410	if (status < 0) {
2411		if (status != -ENOSPC)
2412			mlog_errno(status);
2413		goto bail;
2414	}
2415
2416	ac->ac_bits_given += *num_clusters;
2417
2418bail:
2419	if (status)
2420		mlog_errno(status);
2421	return status;
2422}
2423
2424int ocfs2_claim_clusters(handle_t *handle,
2425			 struct ocfs2_alloc_context *ac,
2426			 u32 min_clusters,
2427			 u32 *cluster_start,
2428			 u32 *num_clusters)
2429{
2430	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2431
2432	return __ocfs2_claim_clusters(handle, ac, min_clusters,
2433				      bits_wanted, cluster_start, num_clusters);
2434}
2435
2436static int ocfs2_block_group_clear_bits(handle_t *handle,
2437					struct inode *alloc_inode,
2438					struct ocfs2_group_desc *bg,
2439					struct buffer_head *group_bh,
2440					unsigned int bit_off,
2441					unsigned int num_bits,
2442					unsigned int max_contig_bits,
2443					void (*undo_fn)(unsigned int bit,
2444							unsigned long *bmap))
2445{
2446	int status;
2447	unsigned int tmp;
2448	u16 contig_bits;
2449	struct ocfs2_group_desc *undo_bg = NULL;
2450	struct journal_head *jh;
2451
2452	/* The caller got this descriptor from
2453	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2454	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2455
2456	trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2457
2458	BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2459	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2460					 group_bh,
2461					 undo_fn ?
2462					 OCFS2_JOURNAL_ACCESS_UNDO :
2463					 OCFS2_JOURNAL_ACCESS_WRITE);
2464	if (status < 0) {
2465		mlog_errno(status);
2466		goto bail;
2467	}
2468
2469	jh = bh2jh(group_bh);
2470	if (undo_fn) {
2471		spin_lock(&jh->b_state_lock);
2472		undo_bg = (struct ocfs2_group_desc *) jh->b_committed_data;
 
2473		BUG_ON(!undo_bg);
2474	}
2475
2476	tmp = num_bits;
2477	while(tmp--) {
2478		ocfs2_clear_bit((bit_off + tmp),
2479				(unsigned long *) bg->bg_bitmap);
2480		if (undo_fn)
2481			undo_fn(bit_off + tmp,
2482				(unsigned long *) undo_bg->bg_bitmap);
2483	}
2484	le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2485	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2486		if (undo_fn)
2487			spin_unlock(&jh->b_state_lock);
2488		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2489				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
2490				   le16_to_cpu(bg->bg_bits),
2491				   le16_to_cpu(bg->bg_free_bits_count),
2492				   num_bits);
2493	}
2494
2495	/*
2496	 * TODO: even 'num_bits == 1' (the worst case, release 1 cluster),
2497	 * we still need to rescan whole bitmap.
2498	 */
2499	if (ocfs2_is_cluster_bitmap(alloc_inode)) {
2500		contig_bits = ocfs2_find_max_contig_free_bits(bg->bg_bitmap,
2501				    le16_to_cpu(bg->bg_bits), 0);
2502		if (contig_bits > max_contig_bits)
2503			max_contig_bits = contig_bits;
2504		bg->bg_contig_free_bits = cpu_to_le16(max_contig_bits);
2505	} else {
2506		bg->bg_contig_free_bits = 0;
2507	}
2508
2509	if (undo_fn)
2510		spin_unlock(&jh->b_state_lock);
2511
2512	ocfs2_journal_dirty(handle, group_bh);
2513bail:
2514	return status;
2515}
2516
2517/*
2518 * expects the suballoc inode to already be locked.
2519 */
2520static int _ocfs2_free_suballoc_bits(handle_t *handle,
2521				     struct inode *alloc_inode,
2522				     struct buffer_head *alloc_bh,
2523				     unsigned int start_bit,
2524				     u64 bg_blkno,
2525				     unsigned int count,
2526				     void (*undo_fn)(unsigned int bit,
2527						     unsigned long *bitmap))
2528{
2529	int status = 0;
2530	u32 tmp_used;
2531	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2532	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2533	struct buffer_head *group_bh = NULL;
2534	struct ocfs2_group_desc *group;
2535	__le16 old_bg_contig_free_bits = 0;
2536
2537	/* The alloc_bh comes from ocfs2_free_dinode() or
2538	 * ocfs2_free_clusters().  The callers have all locked the
2539	 * allocator and gotten alloc_bh from the lock call.  This
2540	 * validates the dinode buffer.  Any corruption that has happened
2541	 * is a code bug. */
2542	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2543	BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2544
2545	trace_ocfs2_free_suballoc_bits(
2546		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2547		(unsigned long long)bg_blkno,
2548		start_bit, count);
2549
2550	status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2551					     &group_bh);
2552	if (status < 0) {
2553		mlog_errno(status);
2554		goto bail;
2555	}
2556	group = (struct ocfs2_group_desc *) group_bh->b_data;
2557
2558	BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2559
2560	if (ocfs2_is_cluster_bitmap(alloc_inode))
2561		old_bg_contig_free_bits = group->bg_contig_free_bits;
2562	status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2563					      group, group_bh,
2564					      start_bit, count, 0, undo_fn);
2565	if (status < 0) {
2566		mlog_errno(status);
2567		goto bail;
2568	}
2569
2570	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2571					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2572	if (status < 0) {
2573		mlog_errno(status);
2574		ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2575				start_bit, count,
2576				le16_to_cpu(old_bg_contig_free_bits), 1);
2577		goto bail;
2578	}
2579
2580	le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2581		     count);
2582	tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2583	fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2584	ocfs2_journal_dirty(handle, alloc_bh);
2585
2586bail:
2587	brelse(group_bh);
 
 
 
2588	return status;
2589}
2590
2591int ocfs2_free_suballoc_bits(handle_t *handle,
2592			     struct inode *alloc_inode,
2593			     struct buffer_head *alloc_bh,
2594			     unsigned int start_bit,
2595			     u64 bg_blkno,
2596			     unsigned int count)
2597{
2598	return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2599					 start_bit, bg_blkno, count, NULL);
2600}
2601
2602int ocfs2_free_dinode(handle_t *handle,
2603		      struct inode *inode_alloc_inode,
2604		      struct buffer_head *inode_alloc_bh,
2605		      struct ocfs2_dinode *di)
2606{
2607	u64 blk = le64_to_cpu(di->i_blkno);
2608	u16 bit = le16_to_cpu(di->i_suballoc_bit);
2609	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2610
2611	if (di->i_suballoc_loc)
2612		bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2613	return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2614					inode_alloc_bh, bit, bg_blkno, 1);
2615}
2616
2617static int _ocfs2_free_clusters(handle_t *handle,
2618				struct inode *bitmap_inode,
2619				struct buffer_head *bitmap_bh,
2620				u64 start_blk,
2621				unsigned int num_clusters,
2622				void (*undo_fn)(unsigned int bit,
2623						unsigned long *bitmap))
2624{
2625	int status;
2626	u16 bg_start_bit;
2627	u64 bg_blkno;
2628
2629	/* You can't ever have a contiguous set of clusters
2630	 * bigger than a block group bitmap so we never have to worry
2631	 * about looping on them.
2632	 * This is expensive. We can safely remove once this stuff has
2633	 * gotten tested really well. */
2634	BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb,
2635				ocfs2_blocks_to_clusters(bitmap_inode->i_sb,
2636							 start_blk)));
2637
2638
2639	ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2640				     &bg_start_bit);
2641
2642	trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2643			(unsigned long long)start_blk,
2644			bg_start_bit, num_clusters);
2645
2646	status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2647					   bg_start_bit, bg_blkno,
2648					   num_clusters, undo_fn);
2649	if (status < 0) {
2650		mlog_errno(status);
2651		goto out;
2652	}
2653
2654	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2655					 num_clusters);
2656
2657out:
 
 
2658	return status;
2659}
2660
2661int ocfs2_free_clusters(handle_t *handle,
2662			struct inode *bitmap_inode,
2663			struct buffer_head *bitmap_bh,
2664			u64 start_blk,
2665			unsigned int num_clusters)
2666{
2667	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2668				    start_blk, num_clusters,
2669				    _ocfs2_set_bit);
2670}
2671
2672/*
2673 * Give never-used clusters back to the global bitmap.  We don't need
2674 * to protect these bits in the undo buffer.
2675 */
2676int ocfs2_release_clusters(handle_t *handle,
2677			   struct inode *bitmap_inode,
2678			   struct buffer_head *bitmap_bh,
2679			   u64 start_blk,
2680			   unsigned int num_clusters)
2681{
2682	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2683				    start_blk, num_clusters,
2684				    _ocfs2_clear_bit);
2685}
2686
2687/*
2688 * For a given allocation, determine which allocators will need to be
2689 * accessed, and lock them, reserving the appropriate number of bits.
2690 *
2691 * Sparse file systems call this from ocfs2_write_begin_nolock()
2692 * and ocfs2_allocate_unwritten_extents().
2693 *
2694 * File systems which don't support holes call this from
2695 * ocfs2_extend_allocation().
2696 */
2697int ocfs2_lock_allocators(struct inode *inode,
2698			  struct ocfs2_extent_tree *et,
2699			  u32 clusters_to_add, u32 extents_to_split,
2700			  struct ocfs2_alloc_context **data_ac,
2701			  struct ocfs2_alloc_context **meta_ac)
2702{
2703	int ret = 0, num_free_extents;
2704	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2705	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2706
2707	*meta_ac = NULL;
2708	if (data_ac)
2709		*data_ac = NULL;
2710
2711	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2712
2713	num_free_extents = ocfs2_num_free_extents(et);
2714	if (num_free_extents < 0) {
2715		ret = num_free_extents;
2716		mlog_errno(ret);
2717		goto out;
2718	}
2719
2720	/*
2721	 * Sparse allocation file systems need to be more conservative
2722	 * with reserving room for expansion - the actual allocation
2723	 * happens while we've got a journal handle open so re-taking
2724	 * a cluster lock (because we ran out of room for another
2725	 * extent) will violate ordering rules.
2726	 *
2727	 * Most of the time we'll only be seeing this 1 cluster at a time
2728	 * anyway.
2729	 *
2730	 * Always lock for any unwritten extents - we might want to
2731	 * add blocks during a split.
2732	 */
2733	if (!num_free_extents ||
2734	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2735		ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2736		if (ret < 0) {
2737			if (ret != -ENOSPC)
2738				mlog_errno(ret);
2739			goto out;
2740		}
2741	}
2742
2743	if (clusters_to_add == 0)
2744		goto out;
2745
2746	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2747	if (ret < 0) {
2748		if (ret != -ENOSPC)
2749			mlog_errno(ret);
2750		goto out;
2751	}
2752
2753out:
2754	if (ret) {
2755		if (*meta_ac) {
2756			ocfs2_free_alloc_context(*meta_ac);
2757			*meta_ac = NULL;
2758		}
2759
2760		/*
2761		 * We cannot have an error and a non null *data_ac.
2762		 */
2763	}
2764
2765	return ret;
2766}
2767
2768/*
2769 * Read the inode specified by blkno to get suballoc_slot and
2770 * suballoc_bit.
2771 */
2772static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2773				       u16 *suballoc_slot, u64 *group_blkno,
2774				       u16 *suballoc_bit)
2775{
2776	int status;
2777	struct buffer_head *inode_bh = NULL;
2778	struct ocfs2_dinode *inode_fe;
2779
2780	trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2781
2782	/* dirty read disk */
2783	status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2784	if (status < 0) {
2785		mlog(ML_ERROR, "read block %llu failed %d\n",
2786		     (unsigned long long)blkno, status);
2787		goto bail;
2788	}
2789
2790	inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2791	if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2792		mlog(ML_ERROR, "invalid inode %llu requested\n",
2793		     (unsigned long long)blkno);
2794		status = -EINVAL;
2795		goto bail;
2796	}
2797
2798	if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2799	    (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2800		mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2801		     (unsigned long long)blkno,
2802		     (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2803		status = -EINVAL;
2804		goto bail;
2805	}
2806
2807	if (suballoc_slot)
2808		*suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2809	if (suballoc_bit)
2810		*suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2811	if (group_blkno)
2812		*group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2813
2814bail:
2815	brelse(inode_bh);
2816
2817	if (status)
2818		mlog_errno(status);
2819	return status;
2820}
2821
2822/*
2823 * test whether bit is SET in allocator bitmap or not.  on success, 0
2824 * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2825 * is returned and *res is meaningless.  Call this after you have
2826 * cluster locked against suballoc, or you may get a result based on
2827 * non-up2date contents
2828 */
2829static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2830				   struct inode *suballoc,
2831				   struct buffer_head *alloc_bh,
2832				   u64 group_blkno, u64 blkno,
2833				   u16 bit, int *res)
2834{
2835	struct ocfs2_dinode *alloc_di;
2836	struct ocfs2_group_desc *group;
2837	struct buffer_head *group_bh = NULL;
2838	u64 bg_blkno;
2839	int status;
2840
2841	trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2842				      (unsigned int)bit);
2843
2844	alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2845	if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2846		mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2847		     (unsigned int)bit,
2848		     ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2849		status = -EINVAL;
2850		goto bail;
2851	}
2852
2853	bg_blkno = group_blkno ? group_blkno :
2854		   ocfs2_which_suballoc_group(blkno, bit);
2855	status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2856					     &group_bh);
2857	if (status < 0) {
2858		mlog(ML_ERROR, "read group %llu failed %d\n",
2859		     (unsigned long long)bg_blkno, status);
2860		goto bail;
2861	}
2862
2863	group = (struct ocfs2_group_desc *) group_bh->b_data;
2864	*res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2865
2866bail:
2867	brelse(group_bh);
2868
2869	if (status)
2870		mlog_errno(status);
2871	return status;
2872}
2873
2874/*
2875 * Test if the bit representing this inode (blkno) is set in the
2876 * suballocator.
2877 *
2878 * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2879 *
2880 * In the event of failure, a negative value is returned and *res is
2881 * meaningless.
2882 *
2883 * Callers must make sure to hold nfs_sync_lock to prevent
2884 * ocfs2_delete_inode() on another node from accessing the same
2885 * suballocator concurrently.
2886 */
2887int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2888{
2889	int status;
2890	u64 group_blkno = 0;
2891	u16 suballoc_bit = 0, suballoc_slot = 0;
2892	struct inode *inode_alloc_inode;
2893	struct buffer_head *alloc_bh = NULL;
2894
2895	trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2896
2897	status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2898					     &group_blkno, &suballoc_bit);
2899	if (status < 0) {
2900		mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2901		goto bail;
2902	}
2903
2904	if (suballoc_slot == (u16)OCFS2_INVALID_SLOT)
2905		inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2906			GLOBAL_INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2907	else
2908		inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2909			INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2910	if (!inode_alloc_inode) {
2911		/* the error code could be inaccurate, but we are not able to
2912		 * get the correct one. */
2913		status = -EINVAL;
2914		mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2915		     (u32)suballoc_slot);
2916		goto bail;
2917	}
2918
2919	inode_lock(inode_alloc_inode);
2920	status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2921	if (status < 0) {
2922		inode_unlock(inode_alloc_inode);
2923		iput(inode_alloc_inode);
2924		mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2925		     (u32)suballoc_slot, status);
2926		goto bail;
2927	}
2928
2929	status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2930					 group_blkno, blkno, suballoc_bit, res);
2931	if (status < 0)
2932		mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2933
2934	ocfs2_inode_unlock(inode_alloc_inode, 0);
2935	inode_unlock(inode_alloc_inode);
2936
2937	iput(inode_alloc_inode);
2938	brelse(alloc_bh);
2939bail:
2940	if (status)
2941		mlog_errno(status);
2942	return status;
2943}