fs/ocfs2/suballoc.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * suballoc.c
   5  *
   6  * metadata alloc and free
   7  * Inspired by ext3 block groups.
   8  *
   9  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
  10  *
  11  * This program is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU General Public
  13  * License as published by the Free Software Foundation; either
  14  * version 2 of the License, or (at your option) any later version.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU General Public
  22  * License along with this program; if not, write to the
  23  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  24  * Boston, MA 021110-1307, USA.
  25  */
  26
  27 #include <linux/fs.h>
  28 #include <linux/types.h>
  29 #include <linux/slab.h>
  30 #include <linux/highmem.h>
  31
  32 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
  33 #include <cluster/masklog.h>
  34
  35 #include "ocfs2.h"
  36
  37 #include "alloc.h"
  38 #include "blockcheck.h"
  39 #include "dlmglue.h"
  40 #include "inode.h"
  41 #include "journal.h"
  42 #include "localalloc.h"
  43 #include "suballoc.h"
  44 #include "super.h"
  45 #include "sysfile.h"
  46 #include "uptodate.h"
  47
  48 #include "buffer_head_io.h"
  49
  50 #define NOT_ALLOC_NEW_GROUP             0
  51 #define ALLOC_NEW_GROUP                 0x1
  52 #define ALLOC_GROUPS_FROM_GLOBAL        0x2
  53
  54 #define OCFS2_MAX_TO_STEAL              1024
  55
  56 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
  57 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
  58 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
  59 static int ocfs2_block_group_fill(handle_t *handle,
  60                                   struct inode *alloc_inode,
  61                                   struct buffer_head *bg_bh,
  62                                   u64 group_blkno,
  63                                   u16 my_chain,
  64                                   struct ocfs2_chain_list *cl);
  65 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
  66                                    struct inode *alloc_inode,
  67                                    struct buffer_head *bh,
  68                                    u64 max_block,
  69                                    u64 *last_alloc_group,
  70                                    int flags);
  71
  72 static int ocfs2_cluster_group_search(struct inode *inode,
  73                                       struct buffer_head *group_bh,
  74                                       u32 bits_wanted, u32 min_bits,
  75                                       u64 max_block,
  76                                       u16 *bit_off, u16 *bits_found);
  77 static int ocfs2_block_group_search(struct inode *inode,
  78                                     struct buffer_head *group_bh,
  79                                     u32 bits_wanted, u32 min_bits,
  80                                     u64 max_block,
  81                                     u16 *bit_off, u16 *bits_found);
  82 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
  83                                      struct ocfs2_alloc_context *ac,
  84                                      handle_t *handle,
  85                                      u32 bits_wanted,
  86                                      u32 min_bits,
  87                                      u16 *bit_off,
  88                                      unsigned int *num_bits,
  89                                      u64 *bg_blkno);
  90 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
  91                                          int nr);
  92 static inline int ocfs2_block_group_set_bits(handle_t *handle,
  93                                              struct inode *alloc_inode,
  94                                              struct ocfs2_group_desc *bg,
  95                                              struct buffer_head *group_bh,
  96                                              unsigned int bit_off,
  97                                              unsigned int num_bits);
  98 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
  99                                                struct inode *alloc_inode,
 100                                                struct ocfs2_group_desc *bg,
 101                                                struct buffer_head *group_bh,
 102                                                unsigned int bit_off,
 103                                                unsigned int num_bits);
 104
 105 static int ocfs2_relink_block_group(handle_t *handle,
 106                                     struct inode *alloc_inode,
 107                                     struct buffer_head *fe_bh,
 108                                     struct buffer_head *bg_bh,
 109                                     struct buffer_head *prev_bg_bh,
 110                                     u16 chain);
 111 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
 112                                                      u32 wanted);
 113 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
 114                                                    u64 bg_blkno,
 115                                                    u16 bg_bit_off);
 116 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
 117                                                 u64 data_blkno,
 118                                                 u64 *bg_blkno,
 119                                                 u16 *bg_bit_off);
 120 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
 121                                              u32 bits_wanted, u64 max_block,
 122                                              int flags,
 123                                              struct ocfs2_alloc_context **ac);
 124
 125 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
 126 {
 127         struct inode *inode = ac->ac_inode;
 128
 129         if (inode) {
 130                 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
 131                         ocfs2_inode_unlock(inode, 1);
 132
 133                 mutex_unlock(&inode->i_mutex);
 134
 135                 iput(inode);
 136                 ac->ac_inode = NULL;
 137         }
 138         brelse(ac->ac_bh);
 139         ac->ac_bh = NULL;
 140 }
 141
 142 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
 143 {
 144         ocfs2_free_ac_resource(ac);
 145         kfree(ac);
 146 }
 147
 148 static 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, ...)                                              \
 154         do{                                                             \
 155                 if (resize)                                     \
 156                         mlog(ML_ERROR, fmt "\n", ##__VA_ARGS__);        \
 157                 else                                                    \
 158                         ocfs2_error(sb, fmt, ##__VA_ARGS__);            \
 159         } while (0)
 160
 161 static 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",
 169                          (unsigned long long)bh->b_blocknr, 7,
 170                          gd->bg_signature);
 171                 return -EINVAL;
 172         }
 173
 174         if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
 175                 do_error("Group descriptor #%llu has an invalid bg_blkno "
 176                          "of %llu",
 177                          (unsigned long long)bh->b_blocknr,
 178                          (unsigned long long)le64_to_cpu(gd->bg_blkno));
 179                 return -EINVAL;
 180         }
 181
 182         if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
 183                 do_error("Group descriptor #%llu has an invalid "
 184                          "fs_generation of #%u",
 185                          (unsigned long long)bh->b_blocknr,
 186                          le32_to_cpu(gd->bg_generation));
 187                 return -EINVAL;
 188         }
 189
 190         if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
 191                 do_error("Group descriptor #%llu has bit count %u but "
 192                          "claims that %u are free",
 193                          (unsigned long long)bh->b_blocknr,
 194                          le16_to_cpu(gd->bg_bits),
 195                          le16_to_cpu(gd->bg_free_bits_count));
 196                 return -EINVAL;
 197         }
 198
 199         if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
 200                 do_error("Group descriptor #%llu has bit count %u but "
 201                          "max bitmap bits of %u",
 202                          (unsigned long long)bh->b_blocknr,
 203                          le16_to_cpu(gd->bg_bits),
 204                          8 * le16_to_cpu(gd->bg_size));
 205                 return -EINVAL;
 206         }
 207
 208         return 0;
 209 }
 210
 211 static int ocfs2_validate_gd_parent(struct super_block *sb,
 212                                     struct ocfs2_dinode *di,
 213                                     struct buffer_head *bh,
 214                                     int resize)
 215 {
 216         unsigned int max_bits;
 217         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 218
 219         if (di->i_blkno != gd->bg_parent_dinode) {
 220                 do_error("Group descriptor #%llu has bad parent "
 221                          "pointer (%llu, expected %llu)",
 222                          (unsigned long long)bh->b_blocknr,
 223                          (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
 224                          (unsigned long long)le64_to_cpu(di->i_blkno));
 225                 return -EINVAL;
 226         }
 227
 228         max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
 229         if (le16_to_cpu(gd->bg_bits) > max_bits) {
 230                 do_error("Group descriptor #%llu has bit count of %u",
 231                          (unsigned long long)bh->b_blocknr,
 232                          le16_to_cpu(gd->bg_bits));
 233                 return -EINVAL;
 234         }
 235
 236         /* In resize, we may meet the case bg_chain == cl_next_free_rec. */
 237         if ((le16_to_cpu(gd->bg_chain) >
 238              le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
 239             ((le16_to_cpu(gd->bg_chain) ==
 240              le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
 241                 do_error("Group descriptor #%llu has bad chain %u",
 242                          (unsigned long long)bh->b_blocknr,
 243                          le16_to_cpu(gd->bg_chain));
 244                 return -EINVAL;
 245         }
 246
 247         return 0;
 248 }
 249
 250 #undef do_error
 251
 252 /*
 253  * This version only prints errors.  It does not fail the filesystem, and
 254  * exists only for resize.
 255  */
 256 int ocfs2_check_group_descriptor(struct super_block *sb,
 257                                  struct ocfs2_dinode *di,
 258                                  struct buffer_head *bh)
 259 {
 260         int rc;
 261         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 262
 263         BUG_ON(!buffer_uptodate(bh));
 264
 265         /*
 266          * If the ecc fails, we return the error but otherwise
 267          * leave the filesystem running.  We know any error is
 268          * local to this block.
 269          */
 270         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
 271         if (rc) {
 272                 mlog(ML_ERROR,
 273                      "Checksum failed for group descriptor %llu\n",
 274                      (unsigned long long)bh->b_blocknr);
 275         } else
 276                 rc = ocfs2_validate_gd_self(sb, bh, 1);
 277         if (!rc)
 278                 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
 279
 280         return rc;
 281 }
 282
 283 static int ocfs2_validate_group_descriptor(struct super_block *sb,
 284                                            struct buffer_head *bh)
 285 {
 286         int rc;
 287         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 288
 289         mlog(0, "Validating group descriptor %llu\n",
 290              (unsigned long long)bh->b_blocknr);
 291
 292         BUG_ON(!buffer_uptodate(bh));
 293
 294         /*
 295          * If the ecc fails, we return the error but otherwise
 296          * leave the filesystem running.  We know any error is
 297          * local to this block.
 298          */
 299         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
 300         if (rc)
 301                 return rc;
 302
 303         /*
 304          * Errors after here are fatal.
 305          */
 306
 307         return ocfs2_validate_gd_self(sb, bh, 0);
 308 }
 309
 310 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
 311                                 u64 gd_blkno, struct buffer_head **bh)
 312 {
 313         int rc;
 314         struct buffer_head *tmp = *bh;
 315
 316         rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
 317                               ocfs2_validate_group_descriptor);
 318         if (rc)
 319                 goto out;
 320
 321         rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
 322         if (rc) {
 323                 brelse(tmp);
 324                 goto out;
 325         }
 326
 327         /* If ocfs2_read_block() got us a new bh, pass it up. */
 328         if (!*bh)
 329                 *bh = tmp;
 330
 331 out:
 332         return rc;
 333 }
 334
 335 static int ocfs2_block_group_fill(handle_t *handle,
 336                                   struct inode *alloc_inode,
 337                                   struct buffer_head *bg_bh,
 338                                   u64 group_blkno,
 339                                   u16 my_chain,
 340                                   struct ocfs2_chain_list *cl)
 341 {
 342         int status = 0;
 343         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 344         struct super_block * sb = alloc_inode->i_sb;
 345
 346         mlog_entry_void();
 347
 348         if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
 349                 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
 350                             "b_blocknr (%llu)",
 351                             (unsigned long long)group_blkno,
 352                             (unsigned long long) bg_bh->b_blocknr);
 353                 status = -EIO;
 354                 goto bail;
 355         }
 356
 357         status = ocfs2_journal_access_gd(handle,
 358                                          INODE_CACHE(alloc_inode),
 359                                          bg_bh,
 360                                          OCFS2_JOURNAL_ACCESS_CREATE);
 361         if (status < 0) {
 362                 mlog_errno(status);
 363                 goto bail;
 364         }
 365
 366         memset(bg, 0, sb->s_blocksize);
 367         strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
 368         bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
 369         bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb));
 370         bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
 371         bg->bg_chain = cpu_to_le16(my_chain);
 372         bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
 373         bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
 374         bg->bg_blkno = cpu_to_le64(group_blkno);
 375         /* set the 1st bit in the bitmap to account for the descriptor block */
 376         ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
 377         bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
 378
 379         status = ocfs2_journal_dirty(handle, bg_bh);
 380         if (status < 0)
 381                 mlog_errno(status);
 382
 383         /* There is no need to zero out or otherwise initialize the
 384          * other blocks in a group - All valid FS metadata in a block
 385          * group stores the superblock fs_generation value at
 386          * allocation time. */
 387
 388 bail:
 389         mlog_exit(status);
 390         return status;
 391 }
 392
 393 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
 394 {
 395         u16 curr, best;
 396
 397         best = curr = 0;
 398         while (curr < le16_to_cpu(cl->cl_count)) {
 399                 if (le32_to_cpu(cl->cl_recs[best].c_total) >
 400                     le32_to_cpu(cl->cl_recs[curr].c_total))
 401                         best = curr;
 402                 curr++;
 403         }
 404         return best;
 405 }
 406
 407 /*
 408  * We expect the block group allocator to already be locked.
 409  */
 410 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
 411                                    struct inode *alloc_inode,
 412                                    struct buffer_head *bh,
 413                                    u64 max_block,
 414                                    u64 *last_alloc_group,
 415                                    int flags)
 416 {
 417         int status, credits;
 418         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
 419         struct ocfs2_chain_list *cl;
 420         struct ocfs2_alloc_context *ac = NULL;
 421         handle_t *handle = NULL;
 422         u32 bit_off, num_bits;
 423         u16 alloc_rec;
 424         u64 bg_blkno;
 425         struct buffer_head *bg_bh = NULL;
 426         struct ocfs2_group_desc *bg;
 427
 428         BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
 429
 430         mlog_entry_void();
 431
 432         cl = &fe->id2.i_chain;
 433         status = ocfs2_reserve_clusters_with_limit(osb,
 434                                                    le16_to_cpu(cl->cl_cpg),
 435                                                    max_block, flags, &ac);
 436         if (status < 0) {
 437                 if (status != -ENOSPC)
 438                         mlog_errno(status);
 439                 goto bail;
 440         }
 441
 442         credits = ocfs2_calc_group_alloc_credits(osb->sb,
 443                                                  le16_to_cpu(cl->cl_cpg));
 444         handle = ocfs2_start_trans(osb, credits);
 445         if (IS_ERR(handle)) {
 446                 status = PTR_ERR(handle);
 447                 handle = NULL;
 448                 mlog_errno(status);
 449                 goto bail;
 450         }
 451
 452         if (last_alloc_group && *last_alloc_group != 0) {
 453                 mlog(0, "use old allocation group %llu for block group alloc\n",
 454                      (unsigned long long)*last_alloc_group);
 455                 ac->ac_last_group = *last_alloc_group;
 456         }
 457         status = ocfs2_claim_clusters(osb,
 458                                       handle,
 459                                       ac,
 460                                       le16_to_cpu(cl->cl_cpg),
 461                                       &bit_off,
 462                                       &num_bits);
 463         if (status < 0) {
 464                 if (status != -ENOSPC)
 465                         mlog_errno(status);
 466                 goto bail;
 467         }
 468
 469         alloc_rec = ocfs2_find_smallest_chain(cl);
 470
 471         /* setup the group */
 472         bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
 473         mlog(0, "new descriptor, record %u, at block %llu\n",
 474              alloc_rec, (unsigned long long)bg_blkno);
 475
 476         bg_bh = sb_getblk(osb->sb, bg_blkno);
 477         if (!bg_bh) {
 478                 status = -EIO;
 479                 mlog_errno(status);
 480                 goto bail;
 481         }
 482         ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
 483
 484         status = ocfs2_block_group_fill(handle,
 485                                         alloc_inode,
 486                                         bg_bh,
 487                                         bg_blkno,
 488                                         alloc_rec,
 489                                         cl);
 490         if (status < 0) {
 491                 mlog_errno(status);
 492                 goto bail;
 493         }
 494
 495         bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 496
 497         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
 498                                          bh, OCFS2_JOURNAL_ACCESS_WRITE);
 499         if (status < 0) {
 500                 mlog_errno(status);
 501                 goto bail;
 502         }
 503
 504         le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
 505                      le16_to_cpu(bg->bg_free_bits_count));
 506         le32_add_cpu(&cl->cl_recs[alloc_rec].c_total, le16_to_cpu(bg->bg_bits));
 507         cl->cl_recs[alloc_rec].c_blkno  = cpu_to_le64(bg_blkno);
 508         if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
 509                 le16_add_cpu(&cl->cl_next_free_rec, 1);
 510
 511         le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
 512                                         le16_to_cpu(bg->bg_free_bits_count));
 513         le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
 514         le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
 515
 516         status = ocfs2_journal_dirty(handle, bh);
 517         if (status < 0) {
 518                 mlog_errno(status);
 519                 goto bail;
 520         }
 521
 522         spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
 523         OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
 524         fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
 525                                              le32_to_cpu(fe->i_clusters)));
 526         spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
 527         i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
 528         alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
 529
 530         status = 0;
 531
 532         /* save the new last alloc group so that the caller can cache it. */
 533         if (last_alloc_group)
 534                 *last_alloc_group = ac->ac_last_group;
 535
 536 bail:
 537         if (handle)
 538                 ocfs2_commit_trans(osb, handle);
 539
 540         if (ac)
 541                 ocfs2_free_alloc_context(ac);
 542
 543         brelse(bg_bh);
 544
 545         mlog_exit(status);
 546         return status;
 547 }
 548
 549 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
 550                                        struct ocfs2_alloc_context *ac,
 551                                        int type,
 552                                        u32 slot,
 553                                        u64 *last_alloc_group,
 554                                        int flags)
 555 {
 556         int status;
 557         u32 bits_wanted = ac->ac_bits_wanted;
 558         struct inode *alloc_inode;
 559         struct buffer_head *bh = NULL;
 560         struct ocfs2_dinode *fe;
 561         u32 free_bits;
 562
 563         mlog_entry_void();
 564
 565         alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
 566         if (!alloc_inode) {
 567                 mlog_errno(-EINVAL);
 568                 return -EINVAL;
 569         }
 570
 571         mutex_lock(&alloc_inode->i_mutex);
 572
 573         status = ocfs2_inode_lock(alloc_inode, &bh, 1);
 574         if (status < 0) {
 575                 mutex_unlock(&alloc_inode->i_mutex);
 576                 iput(alloc_inode);
 577
 578                 mlog_errno(status);
 579                 return status;
 580         }
 581
 582         ac->ac_inode = alloc_inode;
 583         ac->ac_alloc_slot = slot;
 584
 585         fe = (struct ocfs2_dinode *) bh->b_data;
 586
 587         /* The bh was validated by the inode read inside
 588          * ocfs2_inode_lock().  Any corruption is a code bug. */
 589         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
 590
 591         if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
 592                 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
 593                             (unsigned long long)le64_to_cpu(fe->i_blkno));
 594                 status = -EIO;
 595                 goto bail;
 596         }
 597
 598         free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
 599                 le32_to_cpu(fe->id1.bitmap1.i_used);
 600
 601         if (bits_wanted > free_bits) {
 602                 /* cluster bitmap never grows */
 603                 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
 604                         mlog(0, "Disk Full: wanted=%u, free_bits=%u\n",
 605                              bits_wanted, free_bits);
 606                         status = -ENOSPC;
 607                         goto bail;
 608                 }
 609
 610                 if (!(flags & ALLOC_NEW_GROUP)) {
 611                         mlog(0, "Alloc File %u Full: wanted=%u, free_bits=%u, "
 612                              "and we don't alloc a new group for it.\n",
 613                              slot, bits_wanted, free_bits);
 614                         status = -ENOSPC;
 615                         goto bail;
 616                 }
 617
 618                 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
 619                                                  ac->ac_max_block,
 620                                                  last_alloc_group, flags);
 621                 if (status < 0) {
 622                         if (status != -ENOSPC)
 623                                 mlog_errno(status);
 624                         goto bail;
 625                 }
 626                 atomic_inc(&osb->alloc_stats.bg_extends);
 627
 628                 /* You should never ask for this much metadata */
 629                 BUG_ON(bits_wanted >
 630                        (le32_to_cpu(fe->id1.bitmap1.i_total)
 631                         - le32_to_cpu(fe->id1.bitmap1.i_used)));
 632         }
 633
 634         get_bh(bh);
 635         ac->ac_bh = bh;
 636 bail:
 637         brelse(bh);
 638
 639         mlog_exit(status);
 640         return status;
 641 }
 642
 643 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
 644 {
 645         spin_lock(&osb->osb_lock);
 646         osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
 647         spin_unlock(&osb->osb_lock);
 648         atomic_set(&osb->s_num_inodes_stolen, 0);
 649 }
 650
 651 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
 652 {
 653         spin_lock(&osb->osb_lock);
 654         osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
 655         spin_unlock(&osb->osb_lock);
 656         atomic_set(&osb->s_num_meta_stolen, 0);
 657 }
 658
 659 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
 660 {
 661         ocfs2_init_inode_steal_slot(osb);
 662         ocfs2_init_meta_steal_slot(osb);
 663 }
 664
 665 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
 666 {
 667         spin_lock(&osb->osb_lock);
 668         if (type == INODE_ALLOC_SYSTEM_INODE)
 669                 osb->s_inode_steal_slot = slot;
 670         else if (type == EXTENT_ALLOC_SYSTEM_INODE)
 671                 osb->s_meta_steal_slot = slot;
 672         spin_unlock(&osb->osb_lock);
 673 }
 674
 675 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
 676 {
 677         int slot = OCFS2_INVALID_SLOT;
 678
 679         spin_lock(&osb->osb_lock);
 680         if (type == INODE_ALLOC_SYSTEM_INODE)
 681                 slot = osb->s_inode_steal_slot;
 682         else if (type == EXTENT_ALLOC_SYSTEM_INODE)
 683                 slot = osb->s_meta_steal_slot;
 684         spin_unlock(&osb->osb_lock);
 685
 686         return slot;
 687 }
 688
 689 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
 690 {
 691         return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
 692 }
 693
 694 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
 695 {
 696         return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
 697 }
 698
 699 static int ocfs2_steal_resource(struct ocfs2_super *osb,
 700                                 struct ocfs2_alloc_context *ac,
 701                                 int type)
 702 {
 703         int i, status = -ENOSPC;
 704         int slot = __ocfs2_get_steal_slot(osb, type);
 705
 706         /* Start to steal resource from the first slot after ours. */
 707         if (slot == OCFS2_INVALID_SLOT)
 708                 slot = osb->slot_num + 1;
 709
 710         for (i = 0; i < osb->max_slots; i++, slot++) {
 711                 if (slot == osb->max_slots)
 712                         slot = 0;
 713
 714                 if (slot == osb->slot_num)
 715                         continue;
 716
 717                 status = ocfs2_reserve_suballoc_bits(osb, ac,
 718                                                      type,
 719                                                      (u32)slot, NULL,
 720                                                      NOT_ALLOC_NEW_GROUP);
 721                 if (status >= 0) {
 722                         __ocfs2_set_steal_slot(osb, slot, type);
 723                         break;
 724                 }
 725
 726                 ocfs2_free_ac_resource(ac);
 727         }
 728
 729         return status;
 730 }
 731
 732 static int ocfs2_steal_inode(struct ocfs2_super *osb,
 733                              struct ocfs2_alloc_context *ac)
 734 {
 735         return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
 736 }
 737
 738 static int ocfs2_steal_meta(struct ocfs2_super *osb,
 739                             struct ocfs2_alloc_context *ac)
 740 {
 741         return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
 742 }
 743
 744 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
 745                                       int blocks,
 746                                       struct ocfs2_alloc_context **ac)
 747 {
 748         int status;
 749         int slot = ocfs2_get_meta_steal_slot(osb);
 750
 751         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
 752         if (!(*ac)) {
 753                 status = -ENOMEM;
 754                 mlog_errno(status);
 755                 goto bail;
 756         }
 757
 758         (*ac)->ac_bits_wanted = blocks;
 759         (*ac)->ac_which = OCFS2_AC_USE_META;
 760         (*ac)->ac_group_search = ocfs2_block_group_search;
 761
 762         if (slot != OCFS2_INVALID_SLOT &&
 763                 atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
 764                 goto extent_steal;
 765
 766         atomic_set(&osb->s_num_meta_stolen, 0);
 767         status = ocfs2_reserve_suballoc_bits(osb, (*ac),
 768                                              EXTENT_ALLOC_SYSTEM_INODE,
 769                                              (u32)osb->slot_num, NULL,
 770                                              ALLOC_NEW_GROUP);
 771
 772
 773         if (status >= 0) {
 774                 status = 0;
 775                 if (slot != OCFS2_INVALID_SLOT)
 776                         ocfs2_init_meta_steal_slot(osb);
 777                 goto bail;
 778         } else if (status < 0 && status != -ENOSPC) {
 779                 mlog_errno(status);
 780                 goto bail;
 781         }
 782
 783         ocfs2_free_ac_resource(*ac);
 784
 785 extent_steal:
 786         status = ocfs2_steal_meta(osb, *ac);
 787         atomic_inc(&osb->s_num_meta_stolen);
 788         if (status < 0) {
 789                 if (status != -ENOSPC)
 790                         mlog_errno(status);
 791                 goto bail;
 792         }
 793
 794         status = 0;
 795 bail:
 796         if ((status < 0) && *ac) {
 797                 ocfs2_free_alloc_context(*ac);
 798                 *ac = NULL;
 799         }
 800
 801         mlog_exit(status);
 802         return status;
 803 }
 804
 805 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
 806                                struct ocfs2_extent_list *root_el,
 807                                struct ocfs2_alloc_context **ac)
 808 {
 809         return ocfs2_reserve_new_metadata_blocks(osb,
 810                                         ocfs2_extend_meta_needed(root_el),
 811                                         ac);
 812 }
 813
 814 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
 815                             struct ocfs2_alloc_context **ac)
 816 {
 817         int status;
 818         int slot = ocfs2_get_inode_steal_slot(osb);
 819         u64 alloc_group;
 820
 821         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
 822         if (!(*ac)) {
 823                 status = -ENOMEM;
 824                 mlog_errno(status);
 825                 goto bail;
 826         }
 827
 828         (*ac)->ac_bits_wanted = 1;
 829         (*ac)->ac_which = OCFS2_AC_USE_INODE;
 830
 831         (*ac)->ac_group_search = ocfs2_block_group_search;
 832
 833         /*
 834          * stat(2) can't handle i_ino > 32bits, so we tell the
 835          * lower levels not to allocate us a block group past that
 836          * limit.  The 'inode64' mount option avoids this behavior.
 837          */
 838         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
 839                 (*ac)->ac_max_block = (u32)~0U;
 840
 841         /*
 842          * slot is set when we successfully steal inode from other nodes.
 843          * It is reset in 3 places:
 844          * 1. when we flush the truncate log
 845          * 2. when we complete local alloc recovery.
 846          * 3. when we successfully allocate from our own slot.
 847          * After it is set, we will go on stealing inodes until we find the
 848          * need to check our slots to see whether there is some space for us.
 849          */
 850         if (slot != OCFS2_INVALID_SLOT &&
 851             atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
 852                 goto inode_steal;
 853
 854         atomic_set(&osb->s_num_inodes_stolen, 0);
 855         alloc_group = osb->osb_inode_alloc_group;
 856         status = ocfs2_reserve_suballoc_bits(osb, *ac,
 857                                              INODE_ALLOC_SYSTEM_INODE,
 858                                              (u32)osb->slot_num,
 859                                              &alloc_group,
 860                                              ALLOC_NEW_GROUP |
 861                                              ALLOC_GROUPS_FROM_GLOBAL);
 862         if (status >= 0) {
 863                 status = 0;
 864
 865                 spin_lock(&osb->osb_lock);
 866                 osb->osb_inode_alloc_group = alloc_group;
 867                 spin_unlock(&osb->osb_lock);
 868                 mlog(0, "after reservation, new allocation group is "
 869                      "%llu\n", (unsigned long long)alloc_group);
 870
 871                 /*
 872                  * Some inodes must be freed by us, so try to allocate
 873                  * from our own next time.
 874                  */
 875                 if (slot != OCFS2_INVALID_SLOT)
 876                         ocfs2_init_inode_steal_slot(osb);
 877                 goto bail;
 878         } else if (status < 0 && status != -ENOSPC) {
 879                 mlog_errno(status);
 880                 goto bail;
 881         }
 882
 883         ocfs2_free_ac_resource(*ac);
 884
 885 inode_steal:
 886         status = ocfs2_steal_inode(osb, *ac);
 887         atomic_inc(&osb->s_num_inodes_stolen);
 888         if (status < 0) {
 889                 if (status != -ENOSPC)
 890                         mlog_errno(status);
 891                 goto bail;
 892         }
 893
 894         status = 0;
 895 bail:
 896         if ((status < 0) && *ac) {
 897                 ocfs2_free_alloc_context(*ac);
 898                 *ac = NULL;
 899         }
 900
 901         mlog_exit(status);
 902         return status;
 903 }
 904
 905 /* local alloc code has to do the same thing, so rather than do this
 906  * twice.. */
 907 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
 908                                       struct ocfs2_alloc_context *ac)
 909 {
 910         int status;
 911
 912         ac->ac_which = OCFS2_AC_USE_MAIN;
 913         ac->ac_group_search = ocfs2_cluster_group_search;
 914
 915         status = ocfs2_reserve_suballoc_bits(osb, ac,
 916                                              GLOBAL_BITMAP_SYSTEM_INODE,
 917                                              OCFS2_INVALID_SLOT, NULL,
 918                                              ALLOC_NEW_GROUP);
 919         if (status < 0 && status != -ENOSPC) {
 920                 mlog_errno(status);
 921                 goto bail;
 922         }
 923
 924 bail:
 925         return status;
 926 }
 927
 928 /* Callers don't need to care which bitmap (local alloc or main) to
 929  * use so we figure it out for them, but unfortunately this clutters
 930  * things a bit. */
 931 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
 932                                              u32 bits_wanted, u64 max_block,
 933                                              int flags,
 934                                              struct ocfs2_alloc_context **ac)
 935 {
 936         int status;
 937
 938         mlog_entry_void();
 939
 940         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
 941         if (!(*ac)) {
 942                 status = -ENOMEM;
 943                 mlog_errno(status);
 944                 goto bail;
 945         }
 946
 947         (*ac)->ac_bits_wanted = bits_wanted;
 948         (*ac)->ac_max_block = max_block;
 949
 950         status = -ENOSPC;
 951         if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
 952             ocfs2_alloc_should_use_local(osb, bits_wanted)) {
 953                 status = ocfs2_reserve_local_alloc_bits(osb,
 954                                                         bits_wanted,
 955                                                         *ac);
 956                 if (status == -EFBIG) {
 957                         /* The local alloc window is outside ac_max_block.
 958                          * use the main bitmap. */
 959                         status = -ENOSPC;
 960                 } else if ((status < 0) && (status != -ENOSPC)) {
 961                         mlog_errno(status);
 962                         goto bail;
 963                 }
 964         }
 965
 966         if (status == -ENOSPC) {
 967                 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
 968                 if (status < 0) {
 969                         if (status != -ENOSPC)
 970                                 mlog_errno(status);
 971                         goto bail;
 972                 }
 973         }
 974
 975         status = 0;
 976 bail:
 977         if ((status < 0) && *ac) {
 978                 ocfs2_free_alloc_context(*ac);
 979                 *ac = NULL;
 980         }
 981
 982         mlog_exit(status);
 983         return status;
 984 }
 985
 986 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
 987                            u32 bits_wanted,
 988                            struct ocfs2_alloc_context **ac)
 989 {
 990         return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
 991                                                  ALLOC_NEW_GROUP, ac);
 992 }
 993
 994 /*
 995  * More or less lifted from ext3. I'll leave their description below:
 996  *
 997  * "For ext3 allocations, we must not reuse any blocks which are
 998  * allocated in the bitmap buffer's "last committed data" copy.  This
 999  * prevents deletes from freeing up the page for reuse until we have
1000  * committed the delete transaction.
1001  *
1002  * If we didn't do this, then deleting something and reallocating it as
1003  * data would allow the old block to be overwritten before the
1004  * transaction committed (because we force data to disk before commit).
1005  * This would lead to corruption if we crashed between overwriting the
1006  * data and committing the delete.
1007  *
1008  * @@@ We may want to make this allocation behaviour conditional on
1009  * data-writes at some point, and disable it for metadata allocations or
1010  * sync-data inodes."
1011  *
1012  * Note: OCFS2 already does this differently for metadata vs data
1013  * allocations, as those bitmaps are separate and undo access is never
1014  * called on a metadata group descriptor.
1015  */
1016 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1017                                          int nr)
1018 {
1019         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1020         int ret;
1021
1022         if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1023                 return 0;
1024
1025         if (!buffer_jbd(bg_bh))
1026                 return 1;
1027
1028         jbd_lock_bh_state(bg_bh);
1029         bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
1030         if (bg)
1031                 ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1032         else
1033                 ret = 1;
1034         jbd_unlock_bh_state(bg_bh);
1035
1036         return ret;
1037 }
1038
1039 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1040                                              struct buffer_head *bg_bh,
1041                                              unsigned int bits_wanted,
1042                                              unsigned int total_bits,
1043                                              u16 *bit_off,
1044                                              u16 *bits_found)
1045 {
1046         void *bitmap;
1047         u16 best_offset, best_size;
1048         int offset, start, found, status = 0;
1049         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1050
1051         /* Callers got this descriptor from
1052          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1053         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1054
1055         found = start = best_offset = best_size = 0;
1056         bitmap = bg->bg_bitmap;
1057
1058         while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1059                 if (offset == total_bits)
1060                         break;
1061
1062                 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1063                         /* We found a zero, but we can't use it as it
1064                          * hasn't been put to disk yet! */
1065                         found = 0;
1066                         start = offset + 1;
1067                 } else if (offset == start) {
1068                         /* we found a zero */
1069                         found++;
1070                         /* move start to the next bit to test */
1071                         start++;
1072                 } else {
1073                         /* got a zero after some ones */
1074                         found = 1;
1075                         start = offset + 1;
1076                 }
1077                 if (found > best_size) {
1078                         best_size = found;
1079                         best_offset = start - found;
1080                 }
1081                 /* we got everything we needed */
1082                 if (found == bits_wanted) {
1083                         /* mlog(0, "Found it all!\n"); */
1084                         break;
1085                 }
1086         }
1087
1088         /* XXX: I think the first clause is equivalent to the second
1089          *      - jlbec */
1090         if (found == bits_wanted) {
1091                 *bit_off = start - found;
1092                 *bits_found = found;
1093         } else if (best_size) {
1094                 *bit_off = best_offset;
1095                 *bits_found = best_size;
1096         } else {
1097                 status = -ENOSPC;
1098                 /* No error log here -- see the comment above
1099                  * ocfs2_test_bg_bit_allocatable */
1100         }
1101
1102         return status;
1103 }
1104
1105 static inline int ocfs2_block_group_set_bits(handle_t *handle,
1106                                              struct inode *alloc_inode,
1107                                              struct ocfs2_group_desc *bg,
1108                                              struct buffer_head *group_bh,
1109                                              unsigned int bit_off,
1110                                              unsigned int num_bits)
1111 {
1112         int status;
1113         void *bitmap = bg->bg_bitmap;
1114         int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1115
1116         mlog_entry_void();
1117
1118         /* All callers get the descriptor via
1119          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1120         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1121         BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1122
1123         mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
1124              num_bits);
1125
1126         if (ocfs2_is_cluster_bitmap(alloc_inode))
1127                 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1128
1129         status = ocfs2_journal_access_gd(handle,
1130                                          INODE_CACHE(alloc_inode),
1131                                          group_bh,
1132                                          journal_type);
1133         if (status < 0) {
1134                 mlog_errno(status);
1135                 goto bail;
1136         }
1137
1138         le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1139
1140         while(num_bits--)
1141                 ocfs2_set_bit(bit_off++, bitmap);
1142
1143         status = ocfs2_journal_dirty(handle,
1144                                      group_bh);
1145         if (status < 0) {
1146                 mlog_errno(status);
1147                 goto bail;
1148         }
1149
1150 bail:
1151         mlog_exit(status);
1152         return status;
1153 }
1154
1155 /* find the one with the most empty bits */
1156 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1157 {
1158         u16 curr, best;
1159
1160         BUG_ON(!cl->cl_next_free_rec);
1161
1162         best = curr = 0;
1163         while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1164                 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1165                     le32_to_cpu(cl->cl_recs[best].c_free))
1166                         best = curr;
1167                 curr++;
1168         }
1169
1170         BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1171         return best;
1172 }
1173
1174 static int ocfs2_relink_block_group(handle_t *handle,
1175                                     struct inode *alloc_inode,
1176                                     struct buffer_head *fe_bh,
1177                                     struct buffer_head *bg_bh,
1178                                     struct buffer_head *prev_bg_bh,
1179                                     u16 chain)
1180 {
1181         int status;
1182         /* there is a really tiny chance the journal calls could fail,
1183          * but we wouldn't want inconsistent blocks in *any* case. */
1184         u64 fe_ptr, bg_ptr, prev_bg_ptr;
1185         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1186         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1187         struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1188
1189         /* The caller got these descriptors from
1190          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1191         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1192         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1193
1194         mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
1195              (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1196              (unsigned long long)le64_to_cpu(bg->bg_blkno),
1197              (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1198
1199         fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
1200         bg_ptr = le64_to_cpu(bg->bg_next_group);
1201         prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1202
1203         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1204                                          prev_bg_bh,
1205                                          OCFS2_JOURNAL_ACCESS_WRITE);
1206         if (status < 0) {
1207                 mlog_errno(status);
1208                 goto out_rollback;
1209         }
1210
1211         prev_bg->bg_next_group = bg->bg_next_group;
1212
1213         status = ocfs2_journal_dirty(handle, prev_bg_bh);
1214         if (status < 0) {
1215                 mlog_errno(status);
1216                 goto out_rollback;
1217         }
1218
1219         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1220                                          bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1221         if (status < 0) {
1222                 mlog_errno(status);
1223                 goto out_rollback;
1224         }
1225
1226         bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1227
1228         status = ocfs2_journal_dirty(handle, bg_bh);
1229         if (status < 0) {
1230                 mlog_errno(status);
1231                 goto out_rollback;
1232         }
1233
1234         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1235                                          fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1236         if (status < 0) {
1237                 mlog_errno(status);
1238                 goto out_rollback;
1239         }
1240
1241         fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1242
1243         status = ocfs2_journal_dirty(handle, fe_bh);
1244         if (status < 0) {
1245                 mlog_errno(status);
1246                 goto out_rollback;
1247         }
1248
1249         status = 0;
1250 out_rollback:
1251         if (status < 0) {
1252                 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
1253                 bg->bg_next_group = cpu_to_le64(bg_ptr);
1254                 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1255         }
1256
1257         mlog_exit(status);
1258         return status;
1259 }
1260
1261 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1262                                                      u32 wanted)
1263 {
1264         return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1265 }
1266
1267 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1268  * value on error. */
1269 static int ocfs2_cluster_group_search(struct inode *inode,
1270                                       struct buffer_head *group_bh,
1271                                       u32 bits_wanted, u32 min_bits,
1272                                       u64 max_block,
1273                                       u16 *bit_off, u16 *bits_found)
1274 {
1275         int search = -ENOSPC;
1276         int ret;
1277         u64 blkoff;
1278         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1279         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1280         u16 tmp_off, tmp_found;
1281         unsigned int max_bits, gd_cluster_off;
1282
1283         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1284
1285         if (gd->bg_free_bits_count) {
1286                 max_bits = le16_to_cpu(gd->bg_bits);
1287
1288                 /* Tail groups in cluster bitmaps which aren't cpg
1289                  * aligned are prone to partial extention by a failed
1290                  * fs resize. If the file system resize never got to
1291                  * update the dinode cluster count, then we don't want
1292                  * to trust any clusters past it, regardless of what
1293                  * the group descriptor says. */
1294                 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1295                                                           le64_to_cpu(gd->bg_blkno));
1296                 if ((gd_cluster_off + max_bits) >
1297                     OCFS2_I(inode)->ip_clusters) {
1298                         max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1299                         mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
1300                              (unsigned long long)le64_to_cpu(gd->bg_blkno),
1301                              le16_to_cpu(gd->bg_bits),
1302                              OCFS2_I(inode)->ip_clusters, max_bits);
1303                 }
1304
1305                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1306                                                         group_bh, bits_wanted,
1307                                                         max_bits,
1308                                                         &tmp_off, &tmp_found);
1309                 if (ret)
1310                         return ret;
1311
1312                 if (max_block) {
1313                         blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1314                                                           gd_cluster_off +
1315                                                           tmp_off + tmp_found);
1316                         mlog(0, "Checking %llu against %llu\n",
1317                              (unsigned long long)blkoff,
1318                              (unsigned long long)max_block);
1319                         if (blkoff > max_block)
1320                                 return -ENOSPC;
1321                 }
1322
1323                 /* ocfs2_block_group_find_clear_bits() might
1324                  * return success, but we still want to return
1325                  * -ENOSPC unless it found the minimum number
1326                  * of bits. */
1327                 if (min_bits <= tmp_found) {
1328                         *bit_off = tmp_off;
1329                         *bits_found = tmp_found;
1330                         search = 0; /* success */
1331                 } else if (tmp_found) {
1332                         /*
1333                          * Don't show bits which we'll be returning
1334                          * for allocation to the local alloc bitmap.
1335                          */
1336                         ocfs2_local_alloc_seen_free_bits(osb, tmp_found);
1337                 }
1338         }
1339
1340         return search;
1341 }
1342
1343 static int ocfs2_block_group_search(struct inode *inode,
1344                                     struct buffer_head *group_bh,
1345                                     u32 bits_wanted, u32 min_bits,
1346                                     u64 max_block,
1347                                     u16 *bit_off, u16 *bits_found)
1348 {
1349         int ret = -ENOSPC;
1350         u64 blkoff;
1351         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1352
1353         BUG_ON(min_bits != 1);
1354         BUG_ON(ocfs2_is_cluster_bitmap(inode));
1355
1356         if (bg->bg_free_bits_count) {
1357                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1358                                                         group_bh, bits_wanted,
1359                                                         le16_to_cpu(bg->bg_bits),
1360                                                         bit_off, bits_found);
1361                 if (!ret && max_block) {
1362                         blkoff = le64_to_cpu(bg->bg_blkno) + *bit_off +
1363                                 *bits_found;
1364                         mlog(0, "Checking %llu against %llu\n",
1365                              (unsigned long long)blkoff,
1366                              (unsigned long long)max_block);
1367                         if (blkoff > max_block)
1368                                 ret = -ENOSPC;
1369                 }
1370         }
1371
1372         return ret;
1373 }
1374
1375 static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1376                                        handle_t *handle,
1377                                        struct buffer_head *di_bh,
1378                                        u32 num_bits,
1379                                        u16 chain)
1380 {
1381         int ret;
1382         u32 tmp_used;
1383         struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1384         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1385
1386         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1387                                       OCFS2_JOURNAL_ACCESS_WRITE);
1388         if (ret < 0) {
1389                 mlog_errno(ret);
1390                 goto out;
1391         }
1392
1393         tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1394         di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1395         le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1396
1397         ret = ocfs2_journal_dirty(handle, di_bh);
1398         if (ret < 0)
1399                 mlog_errno(ret);
1400
1401 out:
1402         return ret;
1403 }
1404
1405 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1406                                   handle_t *handle,
1407                                   u32 bits_wanted,
1408                                   u32 min_bits,
1409                                   u16 *bit_off,
1410                                   unsigned int *num_bits,
1411                                   u64 gd_blkno,
1412                                   u16 *bits_left)
1413 {
1414         int ret;
1415         u16 found;
1416         struct buffer_head *group_bh = NULL;
1417         struct ocfs2_group_desc *gd;
1418         struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1419         struct inode *alloc_inode = ac->ac_inode;
1420
1421         ret = ocfs2_read_group_descriptor(alloc_inode, di, gd_blkno,
1422                                           &group_bh);
1423         if (ret < 0) {
1424                 mlog_errno(ret);
1425                 return ret;
1426         }
1427
1428         gd = (struct ocfs2_group_desc *) group_bh->b_data;
1429         ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1430                                   ac->ac_max_block, bit_off, &found);
1431         if (ret < 0) {
1432                 if (ret != -ENOSPC)
1433                         mlog_errno(ret);
1434                 goto out;
1435         }
1436
1437         *num_bits = found;
1438
1439         ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1440                                                *num_bits,
1441                                                le16_to_cpu(gd->bg_chain));
1442         if (ret < 0) {
1443                 mlog_errno(ret);
1444                 goto out;
1445         }
1446
1447         ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1448                                          *bit_off, *num_bits);
1449         if (ret < 0)
1450                 mlog_errno(ret);
1451
1452         *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1453
1454 out:
1455         brelse(group_bh);
1456
1457         return ret;
1458 }
1459
1460 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1461                               handle_t *handle,
1462                               u32 bits_wanted,
1463                               u32 min_bits,
1464                               u16 *bit_off,
1465                               unsigned int *num_bits,
1466                               u64 *bg_blkno,
1467                               u16 *bits_left)
1468 {
1469         int status;
1470         u16 chain, tmp_bits;
1471         u32 tmp_used;
1472         u64 next_group;
1473         struct inode *alloc_inode = ac->ac_inode;
1474         struct buffer_head *group_bh = NULL;
1475         struct buffer_head *prev_group_bh = NULL;
1476         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1477         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1478         struct ocfs2_group_desc *bg;
1479
1480         chain = ac->ac_chain;
1481         mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
1482              bits_wanted, chain,
1483              (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);
1484
1485         status = ocfs2_read_group_descriptor(alloc_inode, fe,
1486                                              le64_to_cpu(cl->cl_recs[chain].c_blkno),
1487                                              &group_bh);
1488         if (status < 0) {
1489                 mlog_errno(status);
1490                 goto bail;
1491         }
1492         bg = (struct ocfs2_group_desc *) group_bh->b_data;
1493
1494         status = -ENOSPC;
1495         /* for now, the chain search is a bit simplistic. We just use
1496          * the 1st group with any empty bits. */
1497         while ((status = ac->ac_group_search(alloc_inode, group_bh,
1498                                              bits_wanted, min_bits,
1499                                              ac->ac_max_block, bit_off,
1500                                              &tmp_bits)) == -ENOSPC) {
1501                 if (!bg->bg_next_group)
1502                         break;
1503
1504                 brelse(prev_group_bh);
1505                 prev_group_bh = NULL;
1506
1507                 next_group = le64_to_cpu(bg->bg_next_group);
1508                 prev_group_bh = group_bh;
1509                 group_bh = NULL;
1510                 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1511                                                      next_group, &group_bh);
1512                 if (status < 0) {
1513                         mlog_errno(status);
1514                         goto bail;
1515                 }
1516                 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1517         }
1518         if (status < 0) {
1519                 if (status != -ENOSPC)
1520                         mlog_errno(status);
1521                 goto bail;
1522         }
1523
1524         mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
1525              tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));
1526
1527         *num_bits = tmp_bits;
1528
1529         BUG_ON(*num_bits == 0);
1530
1531         /*
1532          * Keep track of previous block descriptor read. When
1533          * we find a target, if we have read more than X
1534          * number of descriptors, and the target is reasonably
1535          * empty, relink him to top of his chain.
1536          *
1537          * We've read 0 extra blocks and only send one more to
1538          * the transaction, yet the next guy to search has a
1539          * much easier time.
1540          *
1541          * Do this *after* figuring out how many bits we're taking out
1542          * of our target group.
1543          */
1544         if (ac->ac_allow_chain_relink &&
1545             (prev_group_bh) &&
1546             (ocfs2_block_group_reasonably_empty(bg, *num_bits))) {
1547                 status = ocfs2_relink_block_group(handle, alloc_inode,
1548                                                   ac->ac_bh, group_bh,
1549                                                   prev_group_bh, chain);
1550                 if (status < 0) {
1551                         mlog_errno(status);
1552                         goto bail;
1553                 }
1554         }
1555
1556         /* Ok, claim our bits now: set the info on dinode, chainlist
1557          * and then the group */
1558         status = ocfs2_journal_access_di(handle,
1559                                          INODE_CACHE(alloc_inode),
1560                                          ac->ac_bh,
1561                                          OCFS2_JOURNAL_ACCESS_WRITE);
1562         if (status < 0) {
1563                 mlog_errno(status);
1564                 goto bail;
1565         }
1566
1567         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1568         fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used);
1569         le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits));
1570
1571         status = ocfs2_journal_dirty(handle,
1572                                      ac->ac_bh);
1573         if (status < 0) {
1574                 mlog_errno(status);
1575                 goto bail;
1576         }
1577
1578         status = ocfs2_block_group_set_bits(handle,
1579                                             alloc_inode,
1580                                             bg,
1581                                             group_bh,
1582                                             *bit_off,
1583                                             *num_bits);
1584         if (status < 0) {
1585                 mlog_errno(status);
1586                 goto bail;
1587         }
1588
1589         mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
1590              (unsigned long long)le64_to_cpu(fe->i_blkno));
1591
1592         *bg_blkno = le64_to_cpu(bg->bg_blkno);
1593         *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1594 bail:
1595         brelse(group_bh);
1596         brelse(prev_group_bh);
1597
1598         mlog_exit(status);
1599         return status;
1600 }
1601
1602 /* will give out up to bits_wanted contiguous bits. */
1603 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
1604                                      struct ocfs2_alloc_context *ac,
1605                                      handle_t *handle,
1606                                      u32 bits_wanted,
1607                                      u32 min_bits,
1608                                      u16 *bit_off,
1609                                      unsigned int *num_bits,
1610                                      u64 *bg_blkno)
1611 {
1612         int status;
1613         u16 victim, i;
1614         u16 bits_left = 0;
1615         u64 hint_blkno = ac->ac_last_group;
1616         struct ocfs2_chain_list *cl;
1617         struct ocfs2_dinode *fe;
1618
1619         mlog_entry_void();
1620
1621         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1622         BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1623         BUG_ON(!ac->ac_bh);
1624
1625         fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1626
1627         /* The bh was validated by the inode read during
1628          * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1629         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1630
1631         if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1632             le32_to_cpu(fe->id1.bitmap1.i_total)) {
1633                 ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used "
1634                             "bits but only %u total.",
1635                             (unsigned long long)le64_to_cpu(fe->i_blkno),
1636                             le32_to_cpu(fe->id1.bitmap1.i_used),
1637                             le32_to_cpu(fe->id1.bitmap1.i_total));
1638                 status = -EIO;
1639                 goto bail;
1640         }
1641
1642         if (hint_blkno) {
1643                 /* Attempt to short-circuit the usual search mechanism
1644                  * by jumping straight to the most recently used
1645                  * allocation group. This helps us mantain some
1646                  * contiguousness across allocations. */
1647                 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1648                                                 min_bits, bit_off, num_bits,
1649                                                 hint_blkno, &bits_left);
1650                 if (!status) {
1651                         /* Be careful to update *bg_blkno here as the
1652                          * caller is expecting it to be filled in, and
1653                          * ocfs2_search_one_group() won't do that for
1654                          * us. */
1655                         *bg_blkno = hint_blkno;
1656                         goto set_hint;
1657                 }
1658                 if (status < 0 && status != -ENOSPC) {
1659                         mlog_errno(status);
1660                         goto bail;
1661                 }
1662         }
1663
1664         cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1665
1666         victim = ocfs2_find_victim_chain(cl);
1667         ac->ac_chain = victim;
1668         ac->ac_allow_chain_relink = 1;
1669
1670         status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, bit_off,
1671                                     num_bits, bg_blkno, &bits_left);
1672         if (!status)
1673                 goto set_hint;
1674         if (status < 0 && status != -ENOSPC) {
1675                 mlog_errno(status);
1676                 goto bail;
1677         }
1678
1679         mlog(0, "Search of victim chain %u came up with nothing, "
1680              "trying all chains now.\n", victim);
1681
1682         /* If we didn't pick a good victim, then just default to
1683          * searching each chain in order. Don't allow chain relinking
1684          * because we only calculate enough journal credits for one
1685          * relink per alloc. */
1686         ac->ac_allow_chain_relink = 0;
1687         for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1688                 if (i == victim)
1689                         continue;
1690                 if (!cl->cl_recs[i].c_free)
1691                         continue;
1692
1693                 ac->ac_chain = i;
1694                 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1695                                             bit_off, num_bits, bg_blkno,
1696                                             &bits_left);
1697                 if (!status)
1698                         break;
1699                 if (status < 0 && status != -ENOSPC) {
1700                         mlog_errno(status);
1701                         goto bail;
1702                 }
1703         }
1704
1705 set_hint:
1706         if (status != -ENOSPC) {
1707                 /* If the next search of this group is not likely to
1708                  * yield a suitable extent, then we reset the last
1709                  * group hint so as to not waste a disk read */
1710                 if (bits_left < min_bits)
1711                         ac->ac_last_group = 0;
1712                 else
1713                         ac->ac_last_group = *bg_blkno;
1714         }
1715
1716 bail:
1717         mlog_exit(status);
1718         return status;
1719 }
1720
1721 int ocfs2_claim_metadata(struct ocfs2_super *osb,
1722                          handle_t *handle,
1723                          struct ocfs2_alloc_context *ac,
1724                          u32 bits_wanted,
1725                          u16 *suballoc_bit_start,
1726                          unsigned int *num_bits,
1727                          u64 *blkno_start)
1728 {
1729         int status;
1730         u64 bg_blkno;
1731
1732         BUG_ON(!ac);
1733         BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1734         BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1735
1736         status = ocfs2_claim_suballoc_bits(osb,
1737                                            ac,
1738                                            handle,
1739                                            bits_wanted,
1740                                            1,
1741                                            suballoc_bit_start,
1742                                            num_bits,
1743                                            &bg_blkno);
1744         if (status < 0) {
1745                 mlog_errno(status);
1746                 goto bail;
1747         }
1748         atomic_inc(&osb->alloc_stats.bg_allocs);
1749
1750         *blkno_start = bg_blkno + (u64) *suballoc_bit_start;
1751         ac->ac_bits_given += (*num_bits);
1752         status = 0;
1753 bail:
1754         mlog_exit(status);
1755         return status;
1756 }
1757
1758 static void ocfs2_init_inode_ac_group(struct inode *dir,
1759                                       struct buffer_head *parent_fe_bh,
1760                                       struct ocfs2_alloc_context *ac)
1761 {
1762         struct ocfs2_dinode *fe = (struct ocfs2_dinode *)parent_fe_bh->b_data;
1763         /*
1764          * Try to allocate inodes from some specific group.
1765          *
1766          * If the parent dir has recorded the last group used in allocation,
1767          * cool, use it. Otherwise if we try to allocate new inode from the
1768          * same slot the parent dir belongs to, use the same chunk.
1769          *
1770          * We are very careful here to avoid the mistake of setting
1771          * ac_last_group to a group descriptor from a different (unlocked) slot.
1772          */
1773         if (OCFS2_I(dir)->ip_last_used_group &&
1774             OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
1775                 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
1776         else if (le16_to_cpu(fe->i_suballoc_slot) == ac->ac_alloc_slot)
1777                 ac->ac_last_group = ocfs2_which_suballoc_group(
1778                                         le64_to_cpu(fe->i_blkno),
1779                                         le16_to_cpu(fe->i_suballoc_bit));
1780 }
1781
1782 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
1783                                              struct ocfs2_alloc_context *ac)
1784 {
1785         OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
1786         OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
1787 }
1788
1789 int ocfs2_claim_new_inode(struct ocfs2_super *osb,
1790                           handle_t *handle,
1791                           struct inode *dir,
1792                           struct buffer_head *parent_fe_bh,
1793                           struct ocfs2_alloc_context *ac,
1794                           u16 *suballoc_bit,
1795                           u64 *fe_blkno)
1796 {
1797         int status;
1798         unsigned int num_bits;
1799         u64 bg_blkno;
1800
1801         mlog_entry_void();
1802
1803         BUG_ON(!ac);
1804         BUG_ON(ac->ac_bits_given != 0);
1805         BUG_ON(ac->ac_bits_wanted != 1);
1806         BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
1807
1808         ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
1809
1810         status = ocfs2_claim_suballoc_bits(osb,
1811                                            ac,
1812                                            handle,
1813                                            1,
1814                                            1,
1815                                            suballoc_bit,
1816                                            &num_bits,
1817                                            &bg_blkno);
1818         if (status < 0) {
1819                 mlog_errno(status);
1820                 goto bail;
1821         }
1822         atomic_inc(&osb->alloc_stats.bg_allocs);
1823
1824         BUG_ON(num_bits != 1);
1825
1826         *fe_blkno = bg_blkno + (u64) (*suballoc_bit);
1827         ac->ac_bits_given++;
1828         ocfs2_save_inode_ac_group(dir, ac);
1829         status = 0;
1830 bail:
1831         mlog_exit(status);
1832         return status;
1833 }
1834
1835 /* translate a group desc. blkno and it's bitmap offset into
1836  * disk cluster offset. */
1837 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
1838                                                    u64 bg_blkno,
1839                                                    u16 bg_bit_off)
1840 {
1841         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1842         u32 cluster = 0;
1843
1844         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1845
1846         if (bg_blkno != osb->first_cluster_group_blkno)
1847                 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
1848         cluster += (u32) bg_bit_off;
1849         return cluster;
1850 }
1851
1852 /* given a cluster offset, calculate which block group it belongs to
1853  * and return that block offset. */
1854 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
1855 {
1856         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1857         u32 group_no;
1858
1859         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1860
1861         group_no = cluster / osb->bitmap_cpg;
1862         if (!group_no)
1863                 return osb->first_cluster_group_blkno;
1864         return ocfs2_clusters_to_blocks(inode->i_sb,
1865                                         group_no * osb->bitmap_cpg);
1866 }
1867
1868 /* given the block number of a cluster start, calculate which cluster
1869  * group and descriptor bitmap offset that corresponds to. */
1870 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
1871                                                 u64 data_blkno,
1872                                                 u64 *bg_blkno,
1873                                                 u16 *bg_bit_off)
1874 {
1875         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1876         u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
1877
1878         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1879
1880         *bg_blkno = ocfs2_which_cluster_group(inode,
1881                                               data_cluster);
1882
1883         if (*bg_blkno == osb->first_cluster_group_blkno)
1884                 *bg_bit_off = (u16) data_cluster;
1885         else
1886                 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
1887                                                              data_blkno - *bg_blkno);
1888 }
1889
1890 /*
1891  * min_bits - minimum contiguous chunk from this total allocation we
1892  * can handle. set to what we asked for originally for a full
1893  * contig. allocation, set to '1' to indicate we can deal with extents
1894  * of any size.
1895  */
1896 int __ocfs2_claim_clusters(struct ocfs2_super *osb,
1897                            handle_t *handle,
1898                            struct ocfs2_alloc_context *ac,
1899                            u32 min_clusters,
1900                            u32 max_clusters,
1901                            u32 *cluster_start,
1902                            u32 *num_clusters)
1903 {
1904         int status;
1905         unsigned int bits_wanted = max_clusters;
1906         u64 bg_blkno = 0;
1907         u16 bg_bit_off;
1908
1909         mlog_entry_void();
1910
1911         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1912
1913         BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
1914                && ac->ac_which != OCFS2_AC_USE_MAIN);
1915
1916         if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
1917                 status = ocfs2_claim_local_alloc_bits(osb,
1918                                                       handle,
1919                                                       ac,
1920                                                       bits_wanted,
1921                                                       cluster_start,
1922                                                       num_clusters);
1923                 if (!status)
1924                         atomic_inc(&osb->alloc_stats.local_data);
1925         } else {
1926                 if (min_clusters > (osb->bitmap_cpg - 1)) {
1927                         /* The only paths asking for contiguousness
1928                          * should know about this already. */
1929                         mlog(ML_ERROR, "minimum allocation requested %u exceeds "
1930                              "group bitmap size %u!\n", min_clusters,
1931                              osb->bitmap_cpg);
1932                         status = -ENOSPC;
1933                         goto bail;
1934                 }
1935                 /* clamp the current request down to a realistic size. */
1936                 if (bits_wanted > (osb->bitmap_cpg - 1))
1937                         bits_wanted = osb->bitmap_cpg - 1;
1938
1939                 status = ocfs2_claim_suballoc_bits(osb,
1940                                                    ac,
1941                                                    handle,
1942                                                    bits_wanted,
1943                                                    min_clusters,
1944                                                    &bg_bit_off,
1945                                                    num_clusters,
1946                                                    &bg_blkno);
1947                 if (!status) {
1948                         *cluster_start =
1949                                 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
1950                                                                  bg_blkno,
1951                                                                  bg_bit_off);
1952                         atomic_inc(&osb->alloc_stats.bitmap_data);
1953                 }
1954         }
1955         if (status < 0) {
1956                 if (status != -ENOSPC)
1957                         mlog_errno(status);
1958                 goto bail;
1959         }
1960
1961         ac->ac_bits_given += *num_clusters;
1962
1963 bail:
1964         mlog_exit(status);
1965         return status;
1966 }
1967
1968 int ocfs2_claim_clusters(struct ocfs2_super *osb,
1969                          handle_t *handle,
1970                          struct ocfs2_alloc_context *ac,
1971                          u32 min_clusters,
1972                          u32 *cluster_start,
1973                          u32 *num_clusters)
1974 {
1975         unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
1976
1977         return __ocfs2_claim_clusters(osb, handle, ac, min_clusters,
1978                                       bits_wanted, cluster_start, num_clusters);
1979 }
1980
1981 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
1982                                                struct inode *alloc_inode,
1983                                                struct ocfs2_group_desc *bg,
1984                                                struct buffer_head *group_bh,
1985                                                unsigned int bit_off,
1986                                                unsigned int num_bits)
1987 {
1988         int status;
1989         unsigned int tmp;
1990         int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1991         struct ocfs2_group_desc *undo_bg = NULL;
1992         int cluster_bitmap = 0;
1993
1994         mlog_entry_void();
1995
1996         /* The caller got this descriptor from
1997          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1998         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1999
2000         mlog(0, "off = %u, num = %u\n", bit_off, num_bits);
2001
2002         if (ocfs2_is_cluster_bitmap(alloc_inode))
2003                 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
2004
2005         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2006                                          group_bh, journal_type);
2007         if (status < 0) {
2008                 mlog_errno(status);
2009                 goto bail;
2010         }
2011
2012         if (ocfs2_is_cluster_bitmap(alloc_inode))
2013                 cluster_bitmap = 1;
2014
2015         if (cluster_bitmap) {
2016                 jbd_lock_bh_state(group_bh);
2017                 undo_bg = (struct ocfs2_group_desc *)
2018                                         bh2jh(group_bh)->b_committed_data;
2019                 BUG_ON(!undo_bg);
2020         }
2021
2022         tmp = num_bits;
2023         while(tmp--) {
2024                 ocfs2_clear_bit((bit_off + tmp),
2025                                 (unsigned long *) bg->bg_bitmap);
2026                 if (cluster_bitmap)
2027                         ocfs2_set_bit(bit_off + tmp,
2028                                       (unsigned long *) undo_bg->bg_bitmap);
2029         }
2030         le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2031
2032         if (cluster_bitmap)
2033                 jbd_unlock_bh_state(group_bh);
2034
2035         status = ocfs2_journal_dirty(handle, group_bh);
2036         if (status < 0)
2037                 mlog_errno(status);
2038 bail:
2039         return status;
2040 }
2041
2042 /*
2043  * expects the suballoc inode to already be locked.
2044  */
2045 int ocfs2_free_suballoc_bits(handle_t *handle,
2046                              struct inode *alloc_inode,
2047                              struct buffer_head *alloc_bh,
2048                              unsigned int start_bit,
2049                              u64 bg_blkno,
2050                              unsigned int count)
2051 {
2052         int status = 0;
2053         u32 tmp_used;
2054         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2055         struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2056         struct buffer_head *group_bh = NULL;
2057         struct ocfs2_group_desc *group;
2058
2059         mlog_entry_void();
2060
2061         /* The alloc_bh comes from ocfs2_free_dinode() or
2062          * ocfs2_free_clusters().  The callers have all locked the
2063          * allocator and gotten alloc_bh from the lock call.  This
2064          * validates the dinode buffer.  Any corruption that has happended
2065          * is a code bug. */
2066         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2067         BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2068
2069         mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n",
2070              (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count,
2071              (unsigned long long)bg_blkno, start_bit);
2072
2073         status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2074                                              &group_bh);
2075         if (status < 0) {
2076                 mlog_errno(status);
2077                 goto bail;
2078         }
2079         group = (struct ocfs2_group_desc *) group_bh->b_data;
2080
2081         BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2082
2083         status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2084                                               group, group_bh,
2085                                               start_bit, count);
2086         if (status < 0) {
2087                 mlog_errno(status);
2088                 goto bail;
2089         }
2090
2091         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2092                                          alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2093         if (status < 0) {
2094                 mlog_errno(status);
2095                 goto bail;
2096         }
2097
2098         le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2099                      count);
2100         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2101         fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2102
2103         status = ocfs2_journal_dirty(handle, alloc_bh);
2104         if (status < 0) {
2105                 mlog_errno(status);
2106                 goto bail;
2107         }
2108
2109 bail:
2110         brelse(group_bh);
2111
2112         mlog_exit(status);
2113         return status;
2114 }
2115
2116 int ocfs2_free_dinode(handle_t *handle,
2117                       struct inode *inode_alloc_inode,
2118                       struct buffer_head *inode_alloc_bh,
2119                       struct ocfs2_dinode *di)
2120 {
2121         u64 blk = le64_to_cpu(di->i_blkno);
2122         u16 bit = le16_to_cpu(di->i_suballoc_bit);
2123         u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2124
2125         return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2126                                         inode_alloc_bh, bit, bg_blkno, 1);
2127 }
2128
2129 int ocfs2_free_clusters(handle_t *handle,
2130                        struct inode *bitmap_inode,
2131                        struct buffer_head *bitmap_bh,
2132                        u64 start_blk,
2133                        unsigned int num_clusters)
2134 {
2135         int status;
2136         u16 bg_start_bit;
2137         u64 bg_blkno;
2138         struct ocfs2_dinode *fe;
2139
2140         /* You can't ever have a contiguous set of clusters
2141          * bigger than a block group bitmap so we never have to worry
2142          * about looping on them. */
2143
2144         mlog_entry_void();
2145
2146         /* This is expensive. We can safely remove once this stuff has
2147          * gotten tested really well. */
2148         BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2149
2150         fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2151
2152         ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2153                                      &bg_start_bit);
2154
2155         mlog(0, "want to free %u clusters starting at block %llu\n",
2156              num_clusters, (unsigned long long)start_blk);
2157         mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n",
2158              (unsigned long long)bg_blkno, bg_start_bit);
2159
2160         status = ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2161                                           bg_start_bit, bg_blkno,
2162                                           num_clusters);
2163         if (status < 0) {
2164                 mlog_errno(status);
2165                 goto out;
2166         }
2167
2168         ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2169                                          num_clusters);
2170
2171 out:
2172         mlog_exit(status);
2173         return status;
2174 }
2175
2176 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2177 {
2178         printk("Block Group:\n");
2179         printk("bg_signature:       %s\n", bg->bg_signature);
2180         printk("bg_size:            %u\n", bg->bg_size);
2181         printk("bg_bits:            %u\n", bg->bg_bits);
2182         printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2183         printk("bg_chain:           %u\n", bg->bg_chain);
2184         printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
2185         printk("bg_next_group:      %llu\n",
2186                (unsigned long long)bg->bg_next_group);
2187         printk("bg_parent_dinode:   %llu\n",
2188                (unsigned long long)bg->bg_parent_dinode);
2189         printk("bg_blkno:           %llu\n",
2190                (unsigned long long)bg->bg_blkno);
2191 }
2192
2193 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2194 {
2195         int i;
2196
2197         printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2198         printk("i_signature:                  %s\n", fe->i_signature);
2199         printk("i_size:                       %llu\n",
2200                (unsigned long long)fe->i_size);
2201         printk("i_clusters:                   %u\n", fe->i_clusters);
2202         printk("i_generation:                 %u\n",
2203                le32_to_cpu(fe->i_generation));
2204         printk("id1.bitmap1.i_used:           %u\n",
2205                le32_to_cpu(fe->id1.bitmap1.i_used));
2206         printk("id1.bitmap1.i_total:          %u\n",
2207                le32_to_cpu(fe->id1.bitmap1.i_total));
2208         printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
2209         printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
2210         printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
2211         printk("id2.i_chain.cl_next_free_rec: %u\n",
2212                fe->id2.i_chain.cl_next_free_rec);
2213         for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2214                 printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
2215                        fe->id2.i_chain.cl_recs[i].c_free);
2216                 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2217                        fe->id2.i_chain.cl_recs[i].c_total);
2218                 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2219                        (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2220         }
2221 }
2222
2223 /*
2224  * For a given allocation, determine which allocators will need to be
2225  * accessed, and lock them, reserving the appropriate number of bits.
2226  *
2227  * Sparse file systems call this from ocfs2_write_begin_nolock()
2228  * and ocfs2_allocate_unwritten_extents().
2229  *
2230  * File systems which don't support holes call this from
2231  * ocfs2_extend_allocation().
2232  */
2233 int ocfs2_lock_allocators(struct inode *inode,
2234                           struct ocfs2_extent_tree *et,
2235                           u32 clusters_to_add, u32 extents_to_split,
2236                           struct ocfs2_alloc_context **data_ac,
2237                           struct ocfs2_alloc_context **meta_ac)
2238 {
2239         int ret = 0, num_free_extents;
2240         unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2241         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2242
2243         *meta_ac = NULL;
2244         if (data_ac)
2245                 *data_ac = NULL;
2246
2247         BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2248
2249         num_free_extents = ocfs2_num_free_extents(osb, et);
2250         if (num_free_extents < 0) {
2251                 ret = num_free_extents;
2252                 mlog_errno(ret);
2253                 goto out;
2254         }
2255
2256         /*
2257          * Sparse allocation file systems need to be more conservative
2258          * with reserving room for expansion - the actual allocation
2259          * happens while we've got a journal handle open so re-taking
2260          * a cluster lock (because we ran out of room for another
2261          * extent) will violate ordering rules.
2262          *
2263          * Most of the time we'll only be seeing this 1 cluster at a time
2264          * anyway.
2265          *
2266          * Always lock for any unwritten extents - we might want to
2267          * add blocks during a split.
2268          */
2269         if (!num_free_extents ||
2270             (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2271                 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2272                 if (ret < 0) {
2273                         if (ret != -ENOSPC)
2274                                 mlog_errno(ret);
2275                         goto out;
2276                 }
2277         }
2278
2279         if (clusters_to_add == 0)
2280                 goto out;
2281
2282         ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2283         if (ret < 0) {
2284                 if (ret != -ENOSPC)
2285                         mlog_errno(ret);
2286                 goto out;
2287         }
2288
2289 out:
2290         if (ret) {
2291                 if (*meta_ac) {
2292                         ocfs2_free_alloc_context(*meta_ac);
2293                         *meta_ac = NULL;
2294                 }
2295
2296                 /*
2297                  * We cannot have an error and a non null *data_ac.
2298                  */
2299         }
2300
2301         return ret;
2302 }
2303
2304 /*
2305  * Read the inode specified by blkno to get suballoc_slot and
2306  * suballoc_bit.
2307  */
2308 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2309                                        u16 *suballoc_slot, u16 *suballoc_bit)
2310 {
2311         int status;
2312         struct buffer_head *inode_bh = NULL;
2313         struct ocfs2_dinode *inode_fe;
2314
2315         mlog_entry("blkno: %llu\n", (unsigned long long)blkno);
2316
2317         /* dirty read disk */
2318         status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2319         if (status < 0) {
2320                 mlog(ML_ERROR, "read block %llu failed %d\n",
2321                      (unsigned long long)blkno, status);
2322                 goto bail;
2323         }
2324
2325         inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2326         if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2327                 mlog(ML_ERROR, "invalid inode %llu requested\n",
2328                      (unsigned long long)blkno);
2329                 status = -EINVAL;
2330                 goto bail;
2331         }
2332
2333         if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2334             (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2335                 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2336                      (unsigned long long)blkno,
2337                      (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2338                 status = -EINVAL;
2339                 goto bail;
2340         }
2341
2342         if (suballoc_slot)
2343                 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2344         if (suballoc_bit)
2345                 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2346
2347 bail:
2348         brelse(inode_bh);
2349
2350         mlog_exit(status);
2351         return status;
2352 }
2353
2354 /*
2355  * test whether bit is SET in allocator bitmap or not.  on success, 0
2356  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2357  * is returned and *res is meaningless.  Call this after you have
2358  * cluster locked against suballoc, or you may get a result based on
2359  * non-up2date contents
2360  */
2361 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2362                                    struct inode *suballoc,
2363                                    struct buffer_head *alloc_bh, u64 blkno,
2364                                    u16 bit, int *res)
2365 {
2366         struct ocfs2_dinode *alloc_fe;
2367         struct ocfs2_group_desc *group;
2368         struct buffer_head *group_bh = NULL;
2369         u64 bg_blkno;
2370         int status;
2371
2372         mlog_entry("blkno: %llu bit: %u\n", (unsigned long long)blkno,
2373                    (unsigned int)bit);
2374
2375         alloc_fe = (struct ocfs2_dinode *)alloc_bh->b_data;
2376         if ((bit + 1) > ocfs2_bits_per_group(&alloc_fe->id2.i_chain)) {
2377                 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2378                      (unsigned int)bit,
2379                      ocfs2_bits_per_group(&alloc_fe->id2.i_chain));
2380                 status = -EINVAL;
2381                 goto bail;
2382         }
2383
2384         bg_blkno = ocfs2_which_suballoc_group(blkno, bit);
2385         status = ocfs2_read_group_descriptor(suballoc, alloc_fe, bg_blkno,
2386                                              &group_bh);
2387         if (status < 0) {
2388                 mlog(ML_ERROR, "read group %llu failed %d\n",
2389                      (unsigned long long)bg_blkno, status);
2390                 goto bail;
2391         }
2392
2393         group = (struct ocfs2_group_desc *) group_bh->b_data;
2394         *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2395
2396 bail:
2397         brelse(group_bh);
2398
2399         mlog_exit(status);
2400         return status;
2401 }
2402
2403 /*
2404  * Test if the bit representing this inode (blkno) is set in the
2405  * suballocator.
2406  *
2407  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2408  *
2409  * In the event of failure, a negative value is returned and *res is
2410  * meaningless.
2411  *
2412  * Callers must make sure to hold nfs_sync_lock to prevent
2413  * ocfs2_delete_inode() on another node from accessing the same
2414  * suballocator concurrently.
2415  */
2416 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2417 {
2418         int status;
2419         u16 suballoc_bit = 0, suballoc_slot = 0;
2420         struct inode *inode_alloc_inode;
2421         struct buffer_head *alloc_bh = NULL;
2422
2423         mlog_entry("blkno: %llu", (unsigned long long)blkno);
2424
2425         status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2426                                              &suballoc_bit);
2427         if (status < 0) {
2428                 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2429                 goto bail;
2430         }
2431
2432         inode_alloc_inode =
2433                 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2434                                             suballoc_slot);
2435         if (!inode_alloc_inode) {
2436                 /* the error code could be inaccurate, but we are not able to
2437                  * get the correct one. */
2438                 status = -EINVAL;
2439                 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2440                      (u32)suballoc_slot);
2441                 goto bail;
2442         }
2443
2444         mutex_lock(&inode_alloc_inode->i_mutex);
2445         status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2446         if (status < 0) {
2447                 mutex_unlock(&inode_alloc_inode->i_mutex);
2448                 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2449                      (u32)suballoc_slot, status);
2450                 goto bail;
2451         }
2452
2453         status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2454                                          blkno, suballoc_bit, res);
2455         if (status < 0)
2456                 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2457
2458         ocfs2_inode_unlock(inode_alloc_inode, 0);
2459         mutex_unlock(&inode_alloc_inode->i_mutex);
2460
2461         iput(inode_alloc_inode);
2462         brelse(alloc_bh);
2463 bail:
2464         mlog_exit(status);
2465         return status;
2466 }