git.oblomov.eu Git - linux-2.6/blob - 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                 1
  52
  53 #define OCFS2_MAX_INODES_TO_STEAL       1024
  54
  55 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
  56 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
  57 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
  58 static int ocfs2_block_group_fill(handle_t *handle,
  59                                   struct inode *alloc_inode,
  60                                   struct buffer_head *bg_bh,
  61                                   u64 group_blkno,
  62                                   u16 my_chain,
  63                                   struct ocfs2_chain_list *cl);
  64 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
  65                                    struct inode *alloc_inode,
  66                                    struct buffer_head *bh,
  67                                    u64 max_block);
  68
  69 static int ocfs2_cluster_group_search(struct inode *inode,
  70                                       struct buffer_head *group_bh,
  71                                       u32 bits_wanted, u32 min_bits,
  72                                       u64 max_block,
  73                                       u16 *bit_off, u16 *bits_found);
  74 static int ocfs2_block_group_search(struct inode *inode,
  75                                     struct buffer_head *group_bh,
  76                                     u32 bits_wanted, u32 min_bits,
  77                                     u64 max_block,
  78                                     u16 *bit_off, u16 *bits_found);
  79 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
  80                                      struct ocfs2_alloc_context *ac,
  81                                      handle_t *handle,
  82                                      u32 bits_wanted,
  83                                      u32 min_bits,
  84                                      u16 *bit_off,
  85                                      unsigned int *num_bits,
  86                                      u64 *bg_blkno);
  87 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
  88                                          int nr);
  89 static inline int ocfs2_block_group_set_bits(handle_t *handle,
  90                                              struct inode *alloc_inode,
  91                                              struct ocfs2_group_desc *bg,
  92                                              struct buffer_head *group_bh,
  93                                              unsigned int bit_off,
  94                                              unsigned int num_bits);
  95 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
  96                                                struct inode *alloc_inode,
  97                                                struct ocfs2_group_desc *bg,
  98                                                struct buffer_head *group_bh,
  99                                                unsigned int bit_off,
 100                                                unsigned int num_bits);
 101
 102 static int ocfs2_relink_block_group(handle_t *handle,
 103                                     struct inode *alloc_inode,
 104                                     struct buffer_head *fe_bh,
 105                                     struct buffer_head *bg_bh,
 106                                     struct buffer_head *prev_bg_bh,
 107                                     u16 chain);
 108 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
 109                                                      u32 wanted);
 110 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
 111                                                    u64 bg_blkno,
 112                                                    u16 bg_bit_off);
 113 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
 114                                                 u64 data_blkno,
 115                                                 u64 *bg_blkno,
 116                                                 u16 *bg_bit_off);
 117 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
 118                                              u32 bits_wanted, u64 max_block,
 119                                              struct ocfs2_alloc_context **ac);
 120
 121 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
 122 {
 123         struct inode *inode = ac->ac_inode;
 124
 125         if (inode) {
 126                 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
 127                         ocfs2_inode_unlock(inode, 1);
 128
 129                 mutex_unlock(&inode->i_mutex);
 130
 131                 iput(inode);
 132                 ac->ac_inode = NULL;
 133         }
 134         brelse(ac->ac_bh);
 135         ac->ac_bh = NULL;
 136 }
 137
 138 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
 139 {
 140         ocfs2_free_ac_resource(ac);
 141         kfree(ac);
 142 }
 143
 144 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
 145 {
 146         return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
 147 }
 148
 149 #define do_error(fmt, ...)                                              \
 150         do{                                                             \
 151                 if (clean_error)                                        \
 152                         mlog(ML_ERROR, fmt "\n", ##__VA_ARGS__);        \
 153                 else                                                    \
 154                         ocfs2_error(sb, fmt, ##__VA_ARGS__);            \
 155         } while (0)
 156
 157 static int ocfs2_validate_gd_self(struct super_block *sb,
 158                                   struct buffer_head *bh,
 159                                   int clean_error)
 160 {
 161         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 162
 163         if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
 164                 do_error("Group descriptor #%llu has bad signature %.*s",
 165                          (unsigned long long)bh->b_blocknr, 7,
 166                          gd->bg_signature);
 167                 return -EINVAL;
 168         }
 169
 170         if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
 171                 do_error("Group descriptor #%llu has an invalid bg_blkno "
 172                          "of %llu",
 173                          (unsigned long long)bh->b_blocknr,
 174                          (unsigned long long)le64_to_cpu(gd->bg_blkno));
 175                 return -EINVAL;
 176         }
 177
 178         if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
 179                 do_error("Group descriptor #%llu has an invalid "
 180                          "fs_generation of #%u",
 181                          (unsigned long long)bh->b_blocknr,
 182                          le32_to_cpu(gd->bg_generation));
 183                 return -EINVAL;
 184         }
 185
 186         if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
 187                 do_error("Group descriptor #%llu has bit count %u but "
 188                          "claims that %u are free",
 189                          (unsigned long long)bh->b_blocknr,
 190                          le16_to_cpu(gd->bg_bits),
 191                          le16_to_cpu(gd->bg_free_bits_count));
 192                 return -EINVAL;
 193         }
 194
 195         if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
 196                 do_error("Group descriptor #%llu has bit count %u but "
 197                          "max bitmap bits of %u",
 198                          (unsigned long long)bh->b_blocknr,
 199                          le16_to_cpu(gd->bg_bits),
 200                          8 * le16_to_cpu(gd->bg_size));
 201                 return -EINVAL;
 202         }
 203
 204         return 0;
 205 }
 206
 207 static int ocfs2_validate_gd_parent(struct super_block *sb,
 208                                     struct ocfs2_dinode *di,
 209                                     struct buffer_head *bh,
 210                                     int clean_error)
 211 {
 212         unsigned int max_bits;
 213         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 214
 215         if (di->i_blkno != gd->bg_parent_dinode) {
 216                 do_error("Group descriptor #%llu has bad parent "
 217                          "pointer (%llu, expected %llu)",
 218                          (unsigned long long)bh->b_blocknr,
 219                          (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
 220                          (unsigned long long)le64_to_cpu(di->i_blkno));
 221                 return -EINVAL;
 222         }
 223
 224         max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
 225         if (le16_to_cpu(gd->bg_bits) > max_bits) {
 226                 do_error("Group descriptor #%llu has bit count of %u",
 227                          (unsigned long long)bh->b_blocknr,
 228                          le16_to_cpu(gd->bg_bits));
 229                 return -EINVAL;
 230         }
 231
 232         if (le16_to_cpu(gd->bg_chain) >=
 233             le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) {
 234                 do_error("Group descriptor #%llu has bad chain %u",
 235                          (unsigned long long)bh->b_blocknr,
 236                          le16_to_cpu(gd->bg_chain));
 237                 return -EINVAL;
 238         }
 239
 240         return 0;
 241 }
 242
 243 #undef do_error
 244
 245 /*
 246  * This version only prints errors.  It does not fail the filesystem, and
 247  * exists only for resize.
 248  */
 249 int ocfs2_check_group_descriptor(struct super_block *sb,
 250                                  struct ocfs2_dinode *di,
 251                                  struct buffer_head *bh)
 252 {
 253         int rc;
 254         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 255
 256         BUG_ON(!buffer_uptodate(bh));
 257
 258         /*
 259          * If the ecc fails, we return the error but otherwise
 260          * leave the filesystem running.  We know any error is
 261          * local to this block.
 262          */
 263         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
 264         if (rc) {
 265                 mlog(ML_ERROR,
 266                      "Checksum failed for group descriptor %llu\n",
 267                      (unsigned long long)bh->b_blocknr);
 268         } else
 269                 rc = ocfs2_validate_gd_self(sb, bh, 1);
 270         if (!rc)
 271                 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
 272
 273         return rc;
 274 }
 275
 276 static int ocfs2_validate_group_descriptor(struct super_block *sb,
 277                                            struct buffer_head *bh)
 278 {
 279         int rc;
 280         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 281
 282         mlog(0, "Validating group descriptor %llu\n",
 283              (unsigned long long)bh->b_blocknr);
 284
 285         BUG_ON(!buffer_uptodate(bh));
 286
 287         /*
 288          * If the ecc fails, we return the error but otherwise
 289          * leave the filesystem running.  We know any error is
 290          * local to this block.
 291          */
 292         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
 293         if (rc)
 294                 return rc;
 295
 296         /*
 297          * Errors after here are fatal.
 298          */
 299
 300         return ocfs2_validate_gd_self(sb, bh, 0);
 301 }
 302
 303 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
 304                                 u64 gd_blkno, struct buffer_head **bh)
 305 {
 306         int rc;
 307         struct buffer_head *tmp = *bh;
 308
 309         rc = ocfs2_read_block(inode, gd_blkno, &tmp,
 310                               ocfs2_validate_group_descriptor);
 311         if (rc)
 312                 goto out;
 313
 314         rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
 315         if (rc) {
 316                 brelse(tmp);
 317                 goto out;
 318         }
 319
 320         /* If ocfs2_read_block() got us a new bh, pass it up. */
 321         if (!*bh)
 322                 *bh = tmp;
 323
 324 out:
 325         return rc;
 326 }
 327
 328 static int ocfs2_block_group_fill(handle_t *handle,
 329                                   struct inode *alloc_inode,
 330                                   struct buffer_head *bg_bh,
 331                                   u64 group_blkno,
 332                                   u16 my_chain,
 333                                   struct ocfs2_chain_list *cl)
 334 {
 335         int status = 0;
 336         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 337         struct super_block * sb = alloc_inode->i_sb;
 338
 339         mlog_entry_void();
 340
 341         if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
 342                 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
 343                             "b_blocknr (%llu)",
 344                             (unsigned long long)group_blkno,
 345                             (unsigned long long) bg_bh->b_blocknr);
 346                 status = -EIO;
 347                 goto bail;
 348         }
 349
 350         status = ocfs2_journal_access_gd(handle,
 351                                          alloc_inode,
 352                                          bg_bh,
 353                                          OCFS2_JOURNAL_ACCESS_CREATE);
 354         if (status < 0) {
 355                 mlog_errno(status);
 356                 goto bail;
 357         }
 358
 359         memset(bg, 0, sb->s_blocksize);
 360         strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
 361         bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
 362         bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb));
 363         bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
 364         bg->bg_chain = cpu_to_le16(my_chain);
 365         bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
 366         bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
 367         bg->bg_blkno = cpu_to_le64(group_blkno);
 368         /* set the 1st bit in the bitmap to account for the descriptor block */
 369         ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
 370         bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
 371
 372         status = ocfs2_journal_dirty(handle, bg_bh);
 373         if (status < 0)
 374                 mlog_errno(status);
 375
 376         /* There is no need to zero out or otherwise initialize the
 377          * other blocks in a group - All valid FS metadata in a block
 378          * group stores the superblock fs_generation value at
 379          * allocation time. */
 380
 381 bail:
 382         mlog_exit(status);
 383         return status;
 384 }
 385
 386 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
 387 {
 388         u16 curr, best;
 389
 390         best = curr = 0;
 391         while (curr < le16_to_cpu(cl->cl_count)) {
 392                 if (le32_to_cpu(cl->cl_recs[best].c_total) >
 393                     le32_to_cpu(cl->cl_recs[curr].c_total))
 394                         best = curr;
 395                 curr++;
 396         }
 397         return best;
 398 }
 399
 400 /*
 401  * We expect the block group allocator to already be locked.
 402  */
 403 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
 404                                    struct inode *alloc_inode,
 405                                    struct buffer_head *bh,
 406                                    u64 max_block)
 407 {
 408         int status, credits;
 409         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
 410         struct ocfs2_chain_list *cl;
 411         struct ocfs2_alloc_context *ac = NULL;
 412         handle_t *handle = NULL;
 413         u32 bit_off, num_bits;
 414         u16 alloc_rec;
 415         u64 bg_blkno;
 416         struct buffer_head *bg_bh = NULL;
 417         struct ocfs2_group_desc *bg;
 418
 419         BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
 420
 421         mlog_entry_void();
 422
 423         cl = &fe->id2.i_chain;
 424         status = ocfs2_reserve_clusters_with_limit(osb,
 425                                                    le16_to_cpu(cl->cl_cpg),
 426                                                    max_block, &ac);
 427         if (status < 0) {
 428                 if (status != -ENOSPC)
 429                         mlog_errno(status);
 430                 goto bail;
 431         }
 432
 433         credits = ocfs2_calc_group_alloc_credits(osb->sb,
 434                                                  le16_to_cpu(cl->cl_cpg));
 435         handle = ocfs2_start_trans(osb, credits);
 436         if (IS_ERR(handle)) {
 437                 status = PTR_ERR(handle);
 438                 handle = NULL;
 439                 mlog_errno(status);
 440                 goto bail;
 441         }
 442
 443         status = ocfs2_claim_clusters(osb,
 444                                       handle,
 445                                       ac,
 446                                       le16_to_cpu(cl->cl_cpg),
 447                                       &bit_off,
 448                                       &num_bits);
 449         if (status < 0) {
 450                 if (status != -ENOSPC)
 451                         mlog_errno(status);
 452                 goto bail;
 453         }
 454
 455         alloc_rec = ocfs2_find_smallest_chain(cl);
 456
 457         /* setup the group */
 458         bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
 459         mlog(0, "new descriptor, record %u, at block %llu\n",
 460              alloc_rec, (unsigned long long)bg_blkno);
 461
 462         bg_bh = sb_getblk(osb->sb, bg_blkno);
 463         if (!bg_bh) {
 464                 status = -EIO;
 465                 mlog_errno(status);
 466                 goto bail;
 467         }
 468         ocfs2_set_new_buffer_uptodate(alloc_inode, bg_bh);
 469
 470         status = ocfs2_block_group_fill(handle,
 471                                         alloc_inode,
 472                                         bg_bh,
 473                                         bg_blkno,
 474                                         alloc_rec,
 475                                         cl);
 476         if (status < 0) {
 477                 mlog_errno(status);
 478                 goto bail;
 479         }
 480
 481         bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 482
 483         status = ocfs2_journal_access_di(handle, alloc_inode,
 484                                          bh, OCFS2_JOURNAL_ACCESS_WRITE);
 485         if (status < 0) {
 486                 mlog_errno(status);
 487                 goto bail;
 488         }
 489
 490         le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
 491                      le16_to_cpu(bg->bg_free_bits_count));
 492         le32_add_cpu(&cl->cl_recs[alloc_rec].c_total, le16_to_cpu(bg->bg_bits));
 493         cl->cl_recs[alloc_rec].c_blkno  = cpu_to_le64(bg_blkno);
 494         if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
 495                 le16_add_cpu(&cl->cl_next_free_rec, 1);
 496
 497         le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
 498                                         le16_to_cpu(bg->bg_free_bits_count));
 499         le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
 500         le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
 501
 502         status = ocfs2_journal_dirty(handle, bh);
 503         if (status < 0) {
 504                 mlog_errno(status);
 505                 goto bail;
 506         }
 507
 508         spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
 509         OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
 510         fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
 511                                              le32_to_cpu(fe->i_clusters)));
 512         spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
 513         i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
 514         alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
 515
 516         status = 0;
 517 bail:
 518         if (handle)
 519                 ocfs2_commit_trans(osb, handle);
 520
 521         if (ac)
 522                 ocfs2_free_alloc_context(ac);
 523
 524         brelse(bg_bh);
 525
 526         mlog_exit(status);
 527         return status;
 528 }
 529
 530 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
 531                                        struct ocfs2_alloc_context *ac,
 532                                        int type,
 533                                        u32 slot,
 534                                        int alloc_new_group)
 535 {
 536         int status;
 537         u32 bits_wanted = ac->ac_bits_wanted;
 538         struct inode *alloc_inode;
 539         struct buffer_head *bh = NULL;
 540         struct ocfs2_dinode *fe;
 541         u32 free_bits;
 542
 543         mlog_entry_void();
 544
 545         alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
 546         if (!alloc_inode) {
 547                 mlog_errno(-EINVAL);
 548                 return -EINVAL;
 549         }
 550
 551         mutex_lock(&alloc_inode->i_mutex);
 552
 553         status = ocfs2_inode_lock(alloc_inode, &bh, 1);
 554         if (status < 0) {
 555                 mutex_unlock(&alloc_inode->i_mutex);
 556                 iput(alloc_inode);
 557
 558                 mlog_errno(status);
 559                 return status;
 560         }
 561
 562         ac->ac_inode = alloc_inode;
 563         ac->ac_alloc_slot = slot;
 564
 565         fe = (struct ocfs2_dinode *) bh->b_data;
 566
 567         /* The bh was validated by the inode read inside
 568          * ocfs2_inode_lock().  Any corruption is a code bug. */
 569         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
 570
 571         if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
 572                 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
 573                             (unsigned long long)le64_to_cpu(fe->i_blkno));
 574                 status = -EIO;
 575                 goto bail;
 576         }
 577
 578         free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
 579                 le32_to_cpu(fe->id1.bitmap1.i_used);
 580
 581         if (bits_wanted > free_bits) {
 582                 /* cluster bitmap never grows */
 583                 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
 584                         mlog(0, "Disk Full: wanted=%u, free_bits=%u\n",
 585                              bits_wanted, free_bits);
 586                         status = -ENOSPC;
 587                         goto bail;
 588                 }
 589
 590                 if (alloc_new_group != ALLOC_NEW_GROUP) {
 591                         mlog(0, "Alloc File %u Full: wanted=%u, free_bits=%u, "
 592                              "and we don't alloc a new group for it.\n",
 593                              slot, bits_wanted, free_bits);
 594                         status = -ENOSPC;
 595                         goto bail;
 596                 }
 597
 598                 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
 599                                                  ac->ac_max_block);
 600                 if (status < 0) {
 601                         if (status != -ENOSPC)
 602                                 mlog_errno(status);
 603                         goto bail;
 604                 }
 605                 atomic_inc(&osb->alloc_stats.bg_extends);
 606
 607                 /* You should never ask for this much metadata */
 608                 BUG_ON(bits_wanted >
 609                        (le32_to_cpu(fe->id1.bitmap1.i_total)
 610                         - le32_to_cpu(fe->id1.bitmap1.i_used)));
 611         }
 612
 613         get_bh(bh);
 614         ac->ac_bh = bh;
 615 bail:
 616         brelse(bh);
 617
 618         mlog_exit(status);
 619         return status;
 620 }
 621
 622 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
 623                                       int blocks,
 624                                       struct ocfs2_alloc_context **ac)
 625 {
 626         int status;
 627         u32 slot;
 628
 629         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
 630         if (!(*ac)) {
 631                 status = -ENOMEM;
 632                 mlog_errno(status);
 633                 goto bail;
 634         }
 635
 636         (*ac)->ac_bits_wanted = blocks;
 637         (*ac)->ac_which = OCFS2_AC_USE_META;
 638         slot = osb->slot_num;
 639         (*ac)->ac_group_search = ocfs2_block_group_search;
 640
 641         status = ocfs2_reserve_suballoc_bits(osb, (*ac),
 642                                              EXTENT_ALLOC_SYSTEM_INODE,
 643                                              slot, ALLOC_NEW_GROUP);
 644         if (status < 0) {
 645                 if (status != -ENOSPC)
 646                         mlog_errno(status);
 647                 goto bail;
 648         }
 649
 650         status = 0;
 651 bail:
 652         if ((status < 0) && *ac) {
 653                 ocfs2_free_alloc_context(*ac);
 654                 *ac = NULL;
 655         }
 656
 657         mlog_exit(status);
 658         return status;
 659 }
 660
 661 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
 662                                struct ocfs2_extent_list *root_el,
 663                                struct ocfs2_alloc_context **ac)
 664 {
 665         return ocfs2_reserve_new_metadata_blocks(osb,
 666                                         ocfs2_extend_meta_needed(root_el),
 667                                         ac);
 668 }
 669
 670 static int ocfs2_steal_inode_from_other_nodes(struct ocfs2_super *osb,
 671                                               struct ocfs2_alloc_context *ac)
 672 {
 673         int i, status = -ENOSPC;
 674         s16 slot = ocfs2_get_inode_steal_slot(osb);
 675
 676         /* Start to steal inodes from the first slot after ours. */
 677         if (slot == OCFS2_INVALID_SLOT)
 678                 slot = osb->slot_num + 1;
 679
 680         for (i = 0; i < osb->max_slots; i++, slot++) {
 681                 if (slot == osb->max_slots)
 682                         slot = 0;
 683
 684                 if (slot == osb->slot_num)
 685                         continue;
 686
 687                 status = ocfs2_reserve_suballoc_bits(osb, ac,
 688                                                      INODE_ALLOC_SYSTEM_INODE,
 689                                                      slot, NOT_ALLOC_NEW_GROUP);
 690                 if (status >= 0) {
 691                         ocfs2_set_inode_steal_slot(osb, slot);
 692                         break;
 693                 }
 694
 695                 ocfs2_free_ac_resource(ac);
 696         }
 697
 698         return status;
 699 }
 700
 701 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
 702                             struct ocfs2_alloc_context **ac)
 703 {
 704         int status;
 705         s16 slot = ocfs2_get_inode_steal_slot(osb);
 706
 707         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
 708         if (!(*ac)) {
 709                 status = -ENOMEM;
 710                 mlog_errno(status);
 711                 goto bail;
 712         }
 713
 714         (*ac)->ac_bits_wanted = 1;
 715         (*ac)->ac_which = OCFS2_AC_USE_INODE;
 716
 717         (*ac)->ac_group_search = ocfs2_block_group_search;
 718
 719         /*
 720          * stat(2) can't handle i_ino > 32bits, so we tell the
 721          * lower levels not to allocate us a block group past that
 722          * limit.  The 'inode64' mount option avoids this behavior.
 723          */
 724         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
 725                 (*ac)->ac_max_block = (u32)~0U;
 726
 727         /*
 728          * slot is set when we successfully steal inode from other nodes.
 729          * It is reset in 3 places:
 730          * 1. when we flush the truncate log
 731          * 2. when we complete local alloc recovery.
 732          * 3. when we successfully allocate from our own slot.
 733          * After it is set, we will go on stealing inodes until we find the
 734          * need to check our slots to see whether there is some space for us.
 735          */
 736         if (slot != OCFS2_INVALID_SLOT &&
 737             atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_INODES_TO_STEAL)
 738                 goto inode_steal;
 739
 740         atomic_set(&osb->s_num_inodes_stolen, 0);
 741         status = ocfs2_reserve_suballoc_bits(osb, *ac,
 742                                              INODE_ALLOC_SYSTEM_INODE,
 743                                              osb->slot_num, ALLOC_NEW_GROUP);
 744         if (status >= 0) {
 745                 status = 0;
 746
 747                 /*
 748                  * Some inodes must be freed by us, so try to allocate
 749                  * from our own next time.
 750                  */
 751                 if (slot != OCFS2_INVALID_SLOT)
 752                         ocfs2_init_inode_steal_slot(osb);
 753                 goto bail;
 754         } else if (status < 0 && status != -ENOSPC) {
 755                 mlog_errno(status);
 756                 goto bail;
 757         }
 758
 759         ocfs2_free_ac_resource(*ac);
 760
 761 inode_steal:
 762         status = ocfs2_steal_inode_from_other_nodes(osb, *ac);
 763         atomic_inc(&osb->s_num_inodes_stolen);
 764         if (status < 0) {
 765                 if (status != -ENOSPC)
 766                         mlog_errno(status);
 767                 goto bail;
 768         }
 769
 770         status = 0;
 771 bail:
 772         if ((status < 0) && *ac) {
 773                 ocfs2_free_alloc_context(*ac);
 774                 *ac = NULL;
 775         }
 776
 777         mlog_exit(status);
 778         return status;
 779 }
 780
 781 /* local alloc code has to do the same thing, so rather than do this
 782  * twice.. */
 783 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
 784                                       struct ocfs2_alloc_context *ac)
 785 {
 786         int status;
 787
 788         ac->ac_which = OCFS2_AC_USE_MAIN;
 789         ac->ac_group_search = ocfs2_cluster_group_search;
 790
 791         status = ocfs2_reserve_suballoc_bits(osb, ac,
 792                                              GLOBAL_BITMAP_SYSTEM_INODE,
 793                                              OCFS2_INVALID_SLOT,
 794                                              ALLOC_NEW_GROUP);
 795         if (status < 0 && status != -ENOSPC) {
 796                 mlog_errno(status);
 797                 goto bail;
 798         }
 799
 800 bail:
 801         return status;
 802 }
 803
 804 /* Callers don't need to care which bitmap (local alloc or main) to
 805  * use so we figure it out for them, but unfortunately this clutters
 806  * things a bit. */
 807 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
 808                                              u32 bits_wanted, u64 max_block,
 809                                              struct ocfs2_alloc_context **ac)
 810 {
 811         int status;
 812
 813         mlog_entry_void();
 814
 815         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
 816         if (!(*ac)) {
 817                 status = -ENOMEM;
 818                 mlog_errno(status);
 819                 goto bail;
 820         }
 821
 822         (*ac)->ac_bits_wanted = bits_wanted;
 823         (*ac)->ac_max_block = max_block;
 824
 825         status = -ENOSPC;
 826         if (ocfs2_alloc_should_use_local(osb, bits_wanted)) {
 827                 status = ocfs2_reserve_local_alloc_bits(osb,
 828                                                         bits_wanted,
 829                                                         *ac);
 830                 if (status == -EFBIG) {
 831                         /* The local alloc window is outside ac_max_block.
 832                          * use the main bitmap. */
 833                         status = -ENOSPC;
 834                 } else if ((status < 0) && (status != -ENOSPC)) {
 835                         mlog_errno(status);
 836                         goto bail;
 837                 }
 838         }
 839
 840         if (status == -ENOSPC) {
 841                 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
 842                 if (status < 0) {
 843                         if (status != -ENOSPC)
 844                                 mlog_errno(status);
 845                         goto bail;
 846                 }
 847         }
 848
 849         status = 0;
 850 bail:
 851         if ((status < 0) && *ac) {
 852                 ocfs2_free_alloc_context(*ac);
 853                 *ac = NULL;
 854         }
 855
 856         mlog_exit(status);
 857         return status;
 858 }
 859
 860 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
 861                            u32 bits_wanted,
 862                            struct ocfs2_alloc_context **ac)
 863 {
 864         return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0, ac);
 865 }
 866
 867 /*
 868  * More or less lifted from ext3. I'll leave their description below:
 869  *
 870  * "For ext3 allocations, we must not reuse any blocks which are
 871  * allocated in the bitmap buffer's "last committed data" copy.  This
 872  * prevents deletes from freeing up the page for reuse until we have
 873  * committed the delete transaction.
 874  *
 875  * If we didn't do this, then deleting something and reallocating it as
 876  * data would allow the old block to be overwritten before the
 877  * transaction committed (because we force data to disk before commit).
 878  * This would lead to corruption if we crashed between overwriting the
 879  * data and committing the delete.
 880  *
 881  * @@@ We may want to make this allocation behaviour conditional on
 882  * data-writes at some point, and disable it for metadata allocations or
 883  * sync-data inodes."
 884  *
 885  * Note: OCFS2 already does this differently for metadata vs data
 886  * allocations, as those bitmaps are separate and undo access is never
 887  * called on a metadata group descriptor.
 888  */
 889 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
 890                                          int nr)
 891 {
 892         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 893
 894         if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
 895                 return 0;
 896         if (!buffer_jbd(bg_bh) || !bh2jh(bg_bh)->b_committed_data)
 897                 return 1;
 898
 899         bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
 900         return !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
 901 }
 902
 903 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
 904                                              struct buffer_head *bg_bh,
 905                                              unsigned int bits_wanted,
 906                                              unsigned int total_bits,
 907                                              u16 *bit_off,
 908                                              u16 *bits_found)
 909 {
 910         void *bitmap;
 911         u16 best_offset, best_size;
 912         int offset, start, found, status = 0;
 913         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 914
 915         /* Callers got this descriptor from
 916          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
 917         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
 918
 919         found = start = best_offset = best_size = 0;
 920         bitmap = bg->bg_bitmap;
 921
 922         while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
 923                 if (offset == total_bits)
 924                         break;
 925
 926                 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
 927                         /* We found a zero, but we can't use it as it
 928                          * hasn't been put to disk yet! */
 929                         found = 0;
 930                         start = offset + 1;
 931                 } else if (offset == start) {
 932                         /* we found a zero */
 933                         found++;
 934                         /* move start to the next bit to test */
 935                         start++;
 936                 } else {
 937                         /* got a zero after some ones */
 938                         found = 1;
 939                         start = offset + 1;
 940                 }
 941                 if (found > best_size) {
 942                         best_size = found;
 943                         best_offset = start - found;
 944                 }
 945                 /* we got everything we needed */
 946                 if (found == bits_wanted) {
 947                         /* mlog(0, "Found it all!\n"); */
 948                         break;
 949                 }
 950         }
 951
 952         /* XXX: I think the first clause is equivalent to the second
 953          *      - jlbec */
 954         if (found == bits_wanted) {
 955                 *bit_off = start - found;
 956                 *bits_found = found;
 957         } else if (best_size) {
 958                 *bit_off = best_offset;
 959                 *bits_found = best_size;
 960         } else {
 961                 status = -ENOSPC;
 962                 /* No error log here -- see the comment above
 963                  * ocfs2_test_bg_bit_allocatable */
 964         }
 965
 966         return status;
 967 }
 968
 969 static inline int ocfs2_block_group_set_bits(handle_t *handle,
 970                                              struct inode *alloc_inode,
 971                                              struct ocfs2_group_desc *bg,
 972                                              struct buffer_head *group_bh,
 973                                              unsigned int bit_off,
 974                                              unsigned int num_bits)
 975 {
 976         int status;
 977         void *bitmap = bg->bg_bitmap;
 978         int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
 979
 980         mlog_entry_void();
 981
 982         /* All callers get the descriptor via
 983          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
 984         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
 985         BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
 986
 987         mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
 988              num_bits);
 989
 990         if (ocfs2_is_cluster_bitmap(alloc_inode))
 991                 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
 992
 993         status = ocfs2_journal_access_gd(handle,
 994                                          alloc_inode,
 995                                          group_bh,
 996                                          journal_type);
 997         if (status < 0) {
 998                 mlog_errno(status);
 999                 goto bail;
1000         }
1001
1002         le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1003
1004         while(num_bits--)
1005                 ocfs2_set_bit(bit_off++, bitmap);
1006
1007         status = ocfs2_journal_dirty(handle,
1008                                      group_bh);
1009         if (status < 0) {
1010                 mlog_errno(status);
1011                 goto bail;
1012         }
1013
1014 bail:
1015         mlog_exit(status);
1016         return status;
1017 }
1018
1019 /* find the one with the most empty bits */
1020 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1021 {
1022         u16 curr, best;
1023
1024         BUG_ON(!cl->cl_next_free_rec);
1025
1026         best = curr = 0;
1027         while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1028                 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1029                     le32_to_cpu(cl->cl_recs[best].c_free))
1030                         best = curr;
1031                 curr++;
1032         }
1033
1034         BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1035         return best;
1036 }
1037
1038 static int ocfs2_relink_block_group(handle_t *handle,
1039                                     struct inode *alloc_inode,
1040                                     struct buffer_head *fe_bh,
1041                                     struct buffer_head *bg_bh,
1042                                     struct buffer_head *prev_bg_bh,
1043                                     u16 chain)
1044 {
1045         int status;
1046         /* there is a really tiny chance the journal calls could fail,
1047          * but we wouldn't want inconsistent blocks in *any* case. */
1048         u64 fe_ptr, bg_ptr, prev_bg_ptr;
1049         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1050         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1051         struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1052
1053         /* The caller got these descriptors from
1054          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1055         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1056         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1057
1058         mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
1059              (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1060              (unsigned long long)le64_to_cpu(bg->bg_blkno),
1061              (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1062
1063         fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
1064         bg_ptr = le64_to_cpu(bg->bg_next_group);
1065         prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1066
1067         status = ocfs2_journal_access_gd(handle, alloc_inode, prev_bg_bh,
1068                                          OCFS2_JOURNAL_ACCESS_WRITE);
1069         if (status < 0) {
1070                 mlog_errno(status);
1071                 goto out_rollback;
1072         }
1073
1074         prev_bg->bg_next_group = bg->bg_next_group;
1075
1076         status = ocfs2_journal_dirty(handle, prev_bg_bh);
1077         if (status < 0) {
1078                 mlog_errno(status);
1079                 goto out_rollback;
1080         }
1081
1082         status = ocfs2_journal_access_gd(handle, alloc_inode, bg_bh,
1083                                          OCFS2_JOURNAL_ACCESS_WRITE);
1084         if (status < 0) {
1085                 mlog_errno(status);
1086                 goto out_rollback;
1087         }
1088
1089         bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1090
1091         status = ocfs2_journal_dirty(handle, bg_bh);
1092         if (status < 0) {
1093                 mlog_errno(status);
1094                 goto out_rollback;
1095         }
1096
1097         status = ocfs2_journal_access_di(handle, alloc_inode, fe_bh,
1098                                          OCFS2_JOURNAL_ACCESS_WRITE);
1099         if (status < 0) {
1100                 mlog_errno(status);
1101                 goto out_rollback;
1102         }
1103
1104         fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1105
1106         status = ocfs2_journal_dirty(handle, fe_bh);
1107         if (status < 0) {
1108                 mlog_errno(status);
1109                 goto out_rollback;
1110         }
1111
1112         status = 0;
1113 out_rollback:
1114         if (status < 0) {
1115                 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
1116                 bg->bg_next_group = cpu_to_le64(bg_ptr);
1117                 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1118         }
1119
1120         mlog_exit(status);
1121         return status;
1122 }
1123
1124 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1125                                                      u32 wanted)
1126 {
1127         return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1128 }
1129
1130 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1131  * value on error. */
1132 static int ocfs2_cluster_group_search(struct inode *inode,
1133                                       struct buffer_head *group_bh,
1134                                       u32 bits_wanted, u32 min_bits,
1135                                       u64 max_block,
1136                                       u16 *bit_off, u16 *bits_found)
1137 {
1138         int search = -ENOSPC;
1139         int ret;
1140         u64 blkoff;
1141         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1142         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1143         u16 tmp_off, tmp_found;
1144         unsigned int max_bits, gd_cluster_off;
1145
1146         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1147
1148         if (gd->bg_free_bits_count) {
1149                 max_bits = le16_to_cpu(gd->bg_bits);
1150
1151                 /* Tail groups in cluster bitmaps which aren't cpg
1152                  * aligned are prone to partial extention by a failed
1153                  * fs resize. If the file system resize never got to
1154                  * update the dinode cluster count, then we don't want
1155                  * to trust any clusters past it, regardless of what
1156                  * the group descriptor says. */
1157                 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1158                                                           le64_to_cpu(gd->bg_blkno));
1159                 if ((gd_cluster_off + max_bits) >
1160                     OCFS2_I(inode)->ip_clusters) {
1161                         max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1162                         mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
1163                              (unsigned long long)le64_to_cpu(gd->bg_blkno),
1164                              le16_to_cpu(gd->bg_bits),
1165                              OCFS2_I(inode)->ip_clusters, max_bits);
1166                 }
1167
1168                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1169                                                         group_bh, bits_wanted,
1170                                                         max_bits,
1171                                                         &tmp_off, &tmp_found);
1172                 if (ret)
1173                         return ret;
1174
1175                 if (max_block) {
1176                         blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1177                                                           gd_cluster_off +
1178                                                           tmp_off + tmp_found);
1179                         mlog(0, "Checking %llu against %llu\n",
1180                              (unsigned long long)blkoff,
1181                              (unsigned long long)max_block);
1182                         if (blkoff > max_block)
1183                                 return -ENOSPC;
1184                 }
1185
1186                 /* ocfs2_block_group_find_clear_bits() might
1187                  * return success, but we still want to return
1188                  * -ENOSPC unless it found the minimum number
1189                  * of bits. */
1190                 if (min_bits <= tmp_found) {
1191                         *bit_off = tmp_off;
1192                         *bits_found = tmp_found;
1193                         search = 0; /* success */
1194                 } else if (tmp_found) {
1195                         /*
1196                          * Don't show bits which we'll be returning
1197                          * for allocation to the local alloc bitmap.
1198                          */
1199                         ocfs2_local_alloc_seen_free_bits(osb, tmp_found);
1200                 }
1201         }
1202
1203         return search;
1204 }
1205
1206 static int ocfs2_block_group_search(struct inode *inode,
1207                                     struct buffer_head *group_bh,
1208                                     u32 bits_wanted, u32 min_bits,
1209                                     u64 max_block,
1210                                     u16 *bit_off, u16 *bits_found)
1211 {
1212         int ret = -ENOSPC;
1213         u64 blkoff;
1214         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1215
1216         BUG_ON(min_bits != 1);
1217         BUG_ON(ocfs2_is_cluster_bitmap(inode));
1218
1219         if (bg->bg_free_bits_count) {
1220                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1221                                                         group_bh, bits_wanted,
1222                                                         le16_to_cpu(bg->bg_bits),
1223                                                         bit_off, bits_found);
1224                 if (!ret && max_block) {
1225                         blkoff = le64_to_cpu(bg->bg_blkno) + *bit_off +
1226                                 *bits_found;
1227                         mlog(0, "Checking %llu against %llu\n",
1228                              (unsigned long long)blkoff,
1229                              (unsigned long long)max_block);
1230                         if (blkoff > max_block)
1231                                 ret = -ENOSPC;
1232                 }
1233         }
1234
1235         return ret;
1236 }
1237
1238 static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1239                                        handle_t *handle,
1240                                        struct buffer_head *di_bh,
1241                                        u32 num_bits,
1242                                        u16 chain)
1243 {
1244         int ret;
1245         u32 tmp_used;
1246         struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1247         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1248
1249         ret = ocfs2_journal_access_di(handle, inode, di_bh,
1250                                       OCFS2_JOURNAL_ACCESS_WRITE);
1251         if (ret < 0) {
1252                 mlog_errno(ret);
1253                 goto out;
1254         }
1255
1256         tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1257         di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1258         le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1259
1260         ret = ocfs2_journal_dirty(handle, di_bh);
1261         if (ret < 0)
1262                 mlog_errno(ret);
1263
1264 out:
1265         return ret;
1266 }
1267
1268 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1269                                   handle_t *handle,
1270                                   u32 bits_wanted,
1271                                   u32 min_bits,
1272                                   u16 *bit_off,
1273                                   unsigned int *num_bits,
1274                                   u64 gd_blkno,
1275                                   u16 *bits_left)
1276 {
1277         int ret;
1278         u16 found;
1279         struct buffer_head *group_bh = NULL;
1280         struct ocfs2_group_desc *gd;
1281         struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1282         struct inode *alloc_inode = ac->ac_inode;
1283
1284         ret = ocfs2_read_group_descriptor(alloc_inode, di, gd_blkno,
1285                                           &group_bh);
1286         if (ret < 0) {
1287                 mlog_errno(ret);
1288                 return ret;
1289         }
1290
1291         gd = (struct ocfs2_group_desc *) group_bh->b_data;
1292         ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1293                                   ac->ac_max_block, bit_off, &found);
1294         if (ret < 0) {
1295                 if (ret != -ENOSPC)
1296                         mlog_errno(ret);
1297                 goto out;
1298         }
1299
1300         *num_bits = found;
1301
1302         ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1303                                                *num_bits,
1304                                                le16_to_cpu(gd->bg_chain));
1305         if (ret < 0) {
1306                 mlog_errno(ret);
1307                 goto out;
1308         }
1309
1310         ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1311                                          *bit_off, *num_bits);
1312         if (ret < 0)
1313                 mlog_errno(ret);
1314
1315         *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1316
1317 out:
1318         brelse(group_bh);
1319
1320         return ret;
1321 }
1322
1323 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1324                               handle_t *handle,
1325                               u32 bits_wanted,
1326                               u32 min_bits,
1327                               u16 *bit_off,
1328                               unsigned int *num_bits,
1329                               u64 *bg_blkno,
1330                               u16 *bits_left)
1331 {
1332         int status;
1333         u16 chain, tmp_bits;
1334         u32 tmp_used;
1335         u64 next_group;
1336         struct inode *alloc_inode = ac->ac_inode;
1337         struct buffer_head *group_bh = NULL;
1338         struct buffer_head *prev_group_bh = NULL;
1339         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1340         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1341         struct ocfs2_group_desc *bg;
1342
1343         chain = ac->ac_chain;
1344         mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
1345              bits_wanted, chain,
1346              (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);
1347
1348         status = ocfs2_read_group_descriptor(alloc_inode, fe,
1349                                              le64_to_cpu(cl->cl_recs[chain].c_blkno),
1350                                              &group_bh);
1351         if (status < 0) {
1352                 mlog_errno(status);
1353                 goto bail;
1354         }
1355         bg = (struct ocfs2_group_desc *) group_bh->b_data;
1356
1357         status = -ENOSPC;
1358         /* for now, the chain search is a bit simplistic. We just use
1359          * the 1st group with any empty bits. */
1360         while ((status = ac->ac_group_search(alloc_inode, group_bh,
1361                                              bits_wanted, min_bits,
1362                                              ac->ac_max_block, bit_off,
1363                                              &tmp_bits)) == -ENOSPC) {
1364                 if (!bg->bg_next_group)
1365                         break;
1366
1367                 brelse(prev_group_bh);
1368                 prev_group_bh = NULL;
1369
1370                 next_group = le64_to_cpu(bg->bg_next_group);
1371                 prev_group_bh = group_bh;
1372                 group_bh = NULL;
1373                 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1374                                                      next_group, &group_bh);
1375                 if (status < 0) {
1376                         mlog_errno(status);
1377                         goto bail;
1378                 }
1379                 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1380         }
1381         if (status < 0) {
1382                 if (status != -ENOSPC)
1383                         mlog_errno(status);
1384                 goto bail;
1385         }
1386
1387         mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
1388              tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));
1389
1390         *num_bits = tmp_bits;
1391
1392         BUG_ON(*num_bits == 0);
1393
1394         /*
1395          * Keep track of previous block descriptor read. When
1396          * we find a target, if we have read more than X
1397          * number of descriptors, and the target is reasonably
1398          * empty, relink him to top of his chain.
1399          *
1400          * We've read 0 extra blocks and only send one more to
1401          * the transaction, yet the next guy to search has a
1402          * much easier time.
1403          *
1404          * Do this *after* figuring out how many bits we're taking out
1405          * of our target group.
1406          */
1407         if (ac->ac_allow_chain_relink &&
1408             (prev_group_bh) &&
1409             (ocfs2_block_group_reasonably_empty(bg, *num_bits))) {
1410                 status = ocfs2_relink_block_group(handle, alloc_inode,
1411                                                   ac->ac_bh, group_bh,
1412                                                   prev_group_bh, chain);
1413                 if (status < 0) {
1414                         mlog_errno(status);
1415                         goto bail;
1416                 }
1417         }
1418
1419         /* Ok, claim our bits now: set the info on dinode, chainlist
1420          * and then the group */
1421         status = ocfs2_journal_access_di(handle,
1422                                          alloc_inode,
1423                                          ac->ac_bh,
1424                                          OCFS2_JOURNAL_ACCESS_WRITE);
1425         if (status < 0) {
1426                 mlog_errno(status);
1427                 goto bail;
1428         }
1429
1430         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1431         fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used);
1432         le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits));
1433
1434         status = ocfs2_journal_dirty(handle,
1435                                      ac->ac_bh);
1436         if (status < 0) {
1437                 mlog_errno(status);
1438                 goto bail;
1439         }
1440
1441         status = ocfs2_block_group_set_bits(handle,
1442                                             alloc_inode,
1443                                             bg,
1444                                             group_bh,
1445                                             *bit_off,
1446                                             *num_bits);
1447         if (status < 0) {
1448                 mlog_errno(status);
1449                 goto bail;
1450         }
1451
1452         mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
1453              (unsigned long long)le64_to_cpu(fe->i_blkno));
1454
1455         *bg_blkno = le64_to_cpu(bg->bg_blkno);
1456         *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1457 bail:
1458         brelse(group_bh);
1459         brelse(prev_group_bh);
1460
1461         mlog_exit(status);
1462         return status;
1463 }
1464
1465 /* will give out up to bits_wanted contiguous bits. */
1466 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
1467                                      struct ocfs2_alloc_context *ac,
1468                                      handle_t *handle,
1469                                      u32 bits_wanted,
1470                                      u32 min_bits,
1471                                      u16 *bit_off,
1472                                      unsigned int *num_bits,
1473                                      u64 *bg_blkno)
1474 {
1475         int status;
1476         u16 victim, i;
1477         u16 bits_left = 0;
1478         u64 hint_blkno = ac->ac_last_group;
1479         struct ocfs2_chain_list *cl;
1480         struct ocfs2_dinode *fe;
1481
1482         mlog_entry_void();
1483
1484         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1485         BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1486         BUG_ON(!ac->ac_bh);
1487
1488         fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1489
1490         /* The bh was validated by the inode read during
1491          * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1492         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1493
1494         if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1495             le32_to_cpu(fe->id1.bitmap1.i_total)) {
1496                 ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used "
1497                             "bits but only %u total.",
1498                             (unsigned long long)le64_to_cpu(fe->i_blkno),
1499                             le32_to_cpu(fe->id1.bitmap1.i_used),
1500                             le32_to_cpu(fe->id1.bitmap1.i_total));
1501                 status = -EIO;
1502                 goto bail;
1503         }
1504
1505         if (hint_blkno) {
1506                 /* Attempt to short-circuit the usual search mechanism
1507                  * by jumping straight to the most recently used
1508                  * allocation group. This helps us mantain some
1509                  * contiguousness across allocations. */
1510                 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1511                                                 min_bits, bit_off, num_bits,
1512                                                 hint_blkno, &bits_left);
1513                 if (!status) {
1514                         /* Be careful to update *bg_blkno here as the
1515                          * caller is expecting it to be filled in, and
1516                          * ocfs2_search_one_group() won't do that for
1517                          * us. */
1518                         *bg_blkno = hint_blkno;
1519                         goto set_hint;
1520                 }
1521                 if (status < 0 && status != -ENOSPC) {
1522                         mlog_errno(status);
1523                         goto bail;
1524                 }
1525         }
1526
1527         cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1528
1529         victim = ocfs2_find_victim_chain(cl);
1530         ac->ac_chain = victim;
1531         ac->ac_allow_chain_relink = 1;
1532
1533         status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, bit_off,
1534                                     num_bits, bg_blkno, &bits_left);
1535         if (!status)
1536                 goto set_hint;
1537         if (status < 0 && status != -ENOSPC) {
1538                 mlog_errno(status);
1539                 goto bail;
1540         }
1541
1542         mlog(0, "Search of victim chain %u came up with nothing, "
1543              "trying all chains now.\n", victim);
1544
1545         /* If we didn't pick a good victim, then just default to
1546          * searching each chain in order. Don't allow chain relinking
1547          * because we only calculate enough journal credits for one
1548          * relink per alloc. */
1549         ac->ac_allow_chain_relink = 0;
1550         for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1551                 if (i == victim)
1552                         continue;
1553                 if (!cl->cl_recs[i].c_free)
1554                         continue;
1555
1556                 ac->ac_chain = i;
1557                 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1558                                             bit_off, num_bits, bg_blkno,
1559                                             &bits_left);
1560                 if (!status)
1561                         break;
1562                 if (status < 0 && status != -ENOSPC) {
1563                         mlog_errno(status);
1564                         goto bail;
1565                 }
1566         }
1567
1568 set_hint:
1569         if (status != -ENOSPC) {
1570                 /* If the next search of this group is not likely to
1571                  * yield a suitable extent, then we reset the last
1572                  * group hint so as to not waste a disk read */
1573                 if (bits_left < min_bits)
1574                         ac->ac_last_group = 0;
1575                 else
1576                         ac->ac_last_group = *bg_blkno;
1577         }
1578
1579 bail:
1580         mlog_exit(status);
1581         return status;
1582 }
1583
1584 int ocfs2_claim_metadata(struct ocfs2_super *osb,
1585                          handle_t *handle,
1586                          struct ocfs2_alloc_context *ac,
1587                          u32 bits_wanted,
1588                          u16 *suballoc_bit_start,
1589                          unsigned int *num_bits,
1590                          u64 *blkno_start)
1591 {
1592         int status;
1593         u64 bg_blkno;
1594
1595         BUG_ON(!ac);
1596         BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1597         BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1598
1599         status = ocfs2_claim_suballoc_bits(osb,
1600                                            ac,
1601                                            handle,
1602                                            bits_wanted,
1603                                            1,
1604                                            suballoc_bit_start,
1605                                            num_bits,
1606                                            &bg_blkno);
1607         if (status < 0) {
1608                 mlog_errno(status);
1609                 goto bail;
1610         }
1611         atomic_inc(&osb->alloc_stats.bg_allocs);
1612
1613         *blkno_start = bg_blkno + (u64) *suballoc_bit_start;
1614         ac->ac_bits_given += (*num_bits);
1615         status = 0;
1616 bail:
1617         mlog_exit(status);
1618         return status;
1619 }
1620
1621 int ocfs2_claim_new_inode(struct ocfs2_super *osb,
1622                           handle_t *handle,
1623                           struct ocfs2_alloc_context *ac,
1624                           u16 *suballoc_bit,
1625                           u64 *fe_blkno)
1626 {
1627         int status;
1628         unsigned int num_bits;
1629         u64 bg_blkno;
1630
1631         mlog_entry_void();
1632
1633         BUG_ON(!ac);
1634         BUG_ON(ac->ac_bits_given != 0);
1635         BUG_ON(ac->ac_bits_wanted != 1);
1636         BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
1637
1638         status = ocfs2_claim_suballoc_bits(osb,
1639                                            ac,
1640                                            handle,
1641                                            1,
1642                                            1,
1643                                            suballoc_bit,
1644                                            &num_bits,
1645                                            &bg_blkno);
1646         if (status < 0) {
1647                 mlog_errno(status);
1648                 goto bail;
1649         }
1650         atomic_inc(&osb->alloc_stats.bg_allocs);
1651
1652         BUG_ON(num_bits != 1);
1653
1654         *fe_blkno = bg_blkno + (u64) (*suballoc_bit);
1655         ac->ac_bits_given++;
1656         status = 0;
1657 bail:
1658         mlog_exit(status);
1659         return status;
1660 }
1661
1662 /* translate a group desc. blkno and it's bitmap offset into
1663  * disk cluster offset. */
1664 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
1665                                                    u64 bg_blkno,
1666                                                    u16 bg_bit_off)
1667 {
1668         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1669         u32 cluster = 0;
1670
1671         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1672
1673         if (bg_blkno != osb->first_cluster_group_blkno)
1674                 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
1675         cluster += (u32) bg_bit_off;
1676         return cluster;
1677 }
1678
1679 /* given a cluster offset, calculate which block group it belongs to
1680  * and return that block offset. */
1681 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
1682 {
1683         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1684         u32 group_no;
1685
1686         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1687
1688         group_no = cluster / osb->bitmap_cpg;
1689         if (!group_no)
1690                 return osb->first_cluster_group_blkno;
1691         return ocfs2_clusters_to_blocks(inode->i_sb,
1692                                         group_no * osb->bitmap_cpg);
1693 }
1694
1695 /* given the block number of a cluster start, calculate which cluster
1696  * group and descriptor bitmap offset that corresponds to. */
1697 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
1698                                                 u64 data_blkno,
1699                                                 u64 *bg_blkno,
1700                                                 u16 *bg_bit_off)
1701 {
1702         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1703         u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
1704
1705         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1706
1707         *bg_blkno = ocfs2_which_cluster_group(inode,
1708                                               data_cluster);
1709
1710         if (*bg_blkno == osb->first_cluster_group_blkno)
1711                 *bg_bit_off = (u16) data_cluster;
1712         else
1713                 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
1714                                                              data_blkno - *bg_blkno);
1715 }
1716
1717 /*
1718  * min_bits - minimum contiguous chunk from this total allocation we
1719  * can handle. set to what we asked for originally for a full
1720  * contig. allocation, set to '1' to indicate we can deal with extents
1721  * of any size.
1722  */
1723 int __ocfs2_claim_clusters(struct ocfs2_super *osb,
1724                            handle_t *handle,
1725                            struct ocfs2_alloc_context *ac,
1726                            u32 min_clusters,
1727                            u32 max_clusters,
1728                            u32 *cluster_start,
1729                            u32 *num_clusters)
1730 {
1731         int status;
1732         unsigned int bits_wanted = max_clusters;
1733         u64 bg_blkno = 0;
1734         u16 bg_bit_off;
1735
1736         mlog_entry_void();
1737
1738         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1739
1740         BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
1741                && ac->ac_which != OCFS2_AC_USE_MAIN);
1742
1743         if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
1744                 status = ocfs2_claim_local_alloc_bits(osb,
1745                                                       handle,
1746                                                       ac,
1747                                                       bits_wanted,
1748                                                       cluster_start,
1749                                                       num_clusters);
1750                 if (!status)
1751                         atomic_inc(&osb->alloc_stats.local_data);
1752         } else {
1753                 if (min_clusters > (osb->bitmap_cpg - 1)) {
1754                         /* The only paths asking for contiguousness
1755                          * should know about this already. */
1756                         mlog(ML_ERROR, "minimum allocation requested %u exceeds "
1757                              "group bitmap size %u!\n", min_clusters,
1758                              osb->bitmap_cpg);
1759                         status = -ENOSPC;
1760                         goto bail;
1761                 }
1762                 /* clamp the current request down to a realistic size. */
1763                 if (bits_wanted > (osb->bitmap_cpg - 1))
1764                         bits_wanted = osb->bitmap_cpg - 1;
1765
1766                 status = ocfs2_claim_suballoc_bits(osb,
1767                                                    ac,
1768                                                    handle,
1769                                                    bits_wanted,
1770                                                    min_clusters,
1771                                                    &bg_bit_off,
1772                                                    num_clusters,
1773                                                    &bg_blkno);
1774                 if (!status) {
1775                         *cluster_start =
1776                                 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
1777                                                                  bg_blkno,
1778                                                                  bg_bit_off);
1779                         atomic_inc(&osb->alloc_stats.bitmap_data);
1780                 }
1781         }
1782         if (status < 0) {
1783                 if (status != -ENOSPC)
1784                         mlog_errno(status);
1785                 goto bail;
1786         }
1787
1788         ac->ac_bits_given += *num_clusters;
1789
1790 bail:
1791         mlog_exit(status);
1792         return status;
1793 }
1794
1795 int ocfs2_claim_clusters(struct ocfs2_super *osb,
1796                          handle_t *handle,
1797                          struct ocfs2_alloc_context *ac,
1798                          u32 min_clusters,
1799                          u32 *cluster_start,
1800                          u32 *num_clusters)
1801 {
1802         unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
1803
1804         return __ocfs2_claim_clusters(osb, handle, ac, min_clusters,
1805                                       bits_wanted, cluster_start, num_clusters);
1806 }
1807
1808 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
1809                                                struct inode *alloc_inode,
1810                                                struct ocfs2_group_desc *bg,
1811                                                struct buffer_head *group_bh,
1812                                                unsigned int bit_off,
1813                                                unsigned int num_bits)
1814 {
1815         int status;
1816         unsigned int tmp;
1817         int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1818         struct ocfs2_group_desc *undo_bg = NULL;
1819
1820         mlog_entry_void();
1821
1822         /* The caller got this descriptor from
1823          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1824         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1825
1826         mlog(0, "off = %u, num = %u\n", bit_off, num_bits);
1827
1828         if (ocfs2_is_cluster_bitmap(alloc_inode))
1829                 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1830
1831         status = ocfs2_journal_access_gd(handle, alloc_inode, group_bh,
1832                                          journal_type);
1833         if (status < 0) {
1834                 mlog_errno(status);
1835                 goto bail;
1836         }
1837
1838         if (ocfs2_is_cluster_bitmap(alloc_inode))
1839                 undo_bg = (struct ocfs2_group_desc *) bh2jh(group_bh)->b_committed_data;
1840
1841         tmp = num_bits;
1842         while(tmp--) {
1843                 ocfs2_clear_bit((bit_off + tmp),
1844                                 (unsigned long *) bg->bg_bitmap);
1845                 if (ocfs2_is_cluster_bitmap(alloc_inode))
1846                         ocfs2_set_bit(bit_off + tmp,
1847                                       (unsigned long *) undo_bg->bg_bitmap);
1848         }
1849         le16_add_cpu(&bg->bg_free_bits_count, num_bits);
1850
1851         status = ocfs2_journal_dirty(handle, group_bh);
1852         if (status < 0)
1853                 mlog_errno(status);
1854 bail:
1855         return status;
1856 }
1857
1858 /*
1859  * expects the suballoc inode to already be locked.
1860  */
1861 int ocfs2_free_suballoc_bits(handle_t *handle,
1862                              struct inode *alloc_inode,
1863                              struct buffer_head *alloc_bh,
1864                              unsigned int start_bit,
1865                              u64 bg_blkno,
1866                              unsigned int count)
1867 {
1868         int status = 0;
1869         u32 tmp_used;
1870         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
1871         struct ocfs2_chain_list *cl = &fe->id2.i_chain;
1872         struct buffer_head *group_bh = NULL;
1873         struct ocfs2_group_desc *group;
1874
1875         mlog_entry_void();
1876
1877         /* The alloc_bh comes from ocfs2_free_dinode() or
1878          * ocfs2_free_clusters().  The callers have all locked the
1879          * allocator and gotten alloc_bh from the lock call.  This
1880          * validates the dinode buffer.  Any corruption that has happended
1881          * is a code bug. */
1882         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1883         BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
1884
1885         mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n",
1886              (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count,
1887              (unsigned long long)bg_blkno, start_bit);
1888
1889         status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
1890                                              &group_bh);
1891         if (status < 0) {
1892                 mlog_errno(status);
1893                 goto bail;
1894         }
1895         group = (struct ocfs2_group_desc *) group_bh->b_data;
1896
1897         BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
1898
1899         status = ocfs2_block_group_clear_bits(handle, alloc_inode,
1900                                               group, group_bh,
1901                                               start_bit, count);
1902         if (status < 0) {
1903                 mlog_errno(status);
1904                 goto bail;
1905         }
1906
1907         status = ocfs2_journal_access_di(handle, alloc_inode, alloc_bh,
1908                                          OCFS2_JOURNAL_ACCESS_WRITE);
1909         if (status < 0) {
1910                 mlog_errno(status);
1911                 goto bail;
1912         }
1913
1914         le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
1915                      count);
1916         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1917         fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
1918
1919         status = ocfs2_journal_dirty(handle, alloc_bh);
1920         if (status < 0) {
1921                 mlog_errno(status);
1922                 goto bail;
1923         }
1924
1925 bail:
1926         brelse(group_bh);
1927
1928         mlog_exit(status);
1929         return status;
1930 }
1931
1932 int ocfs2_free_dinode(handle_t *handle,
1933                       struct inode *inode_alloc_inode,
1934                       struct buffer_head *inode_alloc_bh,
1935                       struct ocfs2_dinode *di)
1936 {
1937         u64 blk = le64_to_cpu(di->i_blkno);
1938         u16 bit = le16_to_cpu(di->i_suballoc_bit);
1939         u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1940
1941         return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
1942                                         inode_alloc_bh, bit, bg_blkno, 1);
1943 }
1944
1945 int ocfs2_free_clusters(handle_t *handle,
1946                        struct inode *bitmap_inode,
1947                        struct buffer_head *bitmap_bh,
1948                        u64 start_blk,
1949                        unsigned int num_clusters)
1950 {
1951         int status;
1952         u16 bg_start_bit;
1953         u64 bg_blkno;
1954         struct ocfs2_dinode *fe;
1955
1956         /* You can't ever have a contiguous set of clusters
1957          * bigger than a block group bitmap so we never have to worry
1958          * about looping on them. */
1959
1960         mlog_entry_void();
1961
1962         /* This is expensive. We can safely remove once this stuff has
1963          * gotten tested really well. */
1964         BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
1965
1966         fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
1967
1968         ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
1969                                      &bg_start_bit);
1970
1971         mlog(0, "want to free %u clusters starting at block %llu\n",
1972              num_clusters, (unsigned long long)start_blk);
1973         mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n",
1974              (unsigned long long)bg_blkno, bg_start_bit);
1975
1976         status = ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
1977                                           bg_start_bit, bg_blkno,
1978                                           num_clusters);
1979         if (status < 0) {
1980                 mlog_errno(status);
1981                 goto out;
1982         }
1983
1984         ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
1985                                          num_clusters);
1986
1987 out:
1988         mlog_exit(status);
1989         return status;
1990 }
1991
1992 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
1993 {
1994         printk("Block Group:\n");
1995         printk("bg_signature:       %s\n", bg->bg_signature);
1996         printk("bg_size:            %u\n", bg->bg_size);
1997         printk("bg_bits:            %u\n", bg->bg_bits);
1998         printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
1999         printk("bg_chain:           %u\n", bg->bg_chain);
2000         printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
2001         printk("bg_next_group:      %llu\n",
2002                (unsigned long long)bg->bg_next_group);
2003         printk("bg_parent_dinode:   %llu\n",
2004                (unsigned long long)bg->bg_parent_dinode);
2005         printk("bg_blkno:           %llu\n",
2006                (unsigned long long)bg->bg_blkno);
2007 }
2008
2009 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2010 {
2011         int i;
2012
2013         printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2014         printk("i_signature:                  %s\n", fe->i_signature);
2015         printk("i_size:                       %llu\n",
2016                (unsigned long long)fe->i_size);
2017         printk("i_clusters:                   %u\n", fe->i_clusters);
2018         printk("i_generation:                 %u\n",
2019                le32_to_cpu(fe->i_generation));
2020         printk("id1.bitmap1.i_used:           %u\n",
2021                le32_to_cpu(fe->id1.bitmap1.i_used));
2022         printk("id1.bitmap1.i_total:          %u\n",
2023                le32_to_cpu(fe->id1.bitmap1.i_total));
2024         printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
2025         printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
2026         printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
2027         printk("id2.i_chain.cl_next_free_rec: %u\n",
2028                fe->id2.i_chain.cl_next_free_rec);
2029         for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2030                 printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
2031                        fe->id2.i_chain.cl_recs[i].c_free);
2032                 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2033                        fe->id2.i_chain.cl_recs[i].c_total);
2034                 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2035                        (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2036         }
2037 }
2038
2039 /*
2040  * For a given allocation, determine which allocators will need to be
2041  * accessed, and lock them, reserving the appropriate number of bits.
2042  *
2043  * Sparse file systems call this from ocfs2_write_begin_nolock()
2044  * and ocfs2_allocate_unwritten_extents().
2045  *
2046  * File systems which don't support holes call this from
2047  * ocfs2_extend_allocation().
2048  */
2049 int ocfs2_lock_allocators(struct inode *inode,
2050                           struct ocfs2_extent_tree *et,
2051                           u32 clusters_to_add, u32 extents_to_split,
2052                           struct ocfs2_alloc_context **data_ac,
2053                           struct ocfs2_alloc_context **meta_ac)
2054 {
2055         int ret = 0, num_free_extents;
2056         unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2057         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2058
2059         *meta_ac = NULL;
2060         if (data_ac)
2061                 *data_ac = NULL;
2062
2063         BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2064
2065         num_free_extents = ocfs2_num_free_extents(osb, inode, et);
2066         if (num_free_extents < 0) {
2067                 ret = num_free_extents;
2068                 mlog_errno(ret);
2069                 goto out;
2070         }
2071
2072         /*
2073          * Sparse allocation file systems need to be more conservative
2074          * with reserving room for expansion - the actual allocation
2075          * happens while we've got a journal handle open so re-taking
2076          * a cluster lock (because we ran out of room for another
2077          * extent) will violate ordering rules.
2078          *
2079          * Most of the time we'll only be seeing this 1 cluster at a time
2080          * anyway.
2081          *
2082          * Always lock for any unwritten extents - we might want to
2083          * add blocks during a split.
2084          */
2085         if (!num_free_extents ||
2086             (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2087                 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2088                 if (ret < 0) {
2089                         if (ret != -ENOSPC)
2090                                 mlog_errno(ret);
2091                         goto out;
2092                 }
2093         }
2094
2095         if (clusters_to_add == 0)
2096                 goto out;
2097
2098         ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2099         if (ret < 0) {
2100                 if (ret != -ENOSPC)
2101                         mlog_errno(ret);
2102                 goto out;
2103         }
2104
2105 out:
2106         if (ret) {
2107                 if (*meta_ac) {
2108                         ocfs2_free_alloc_context(*meta_ac);
2109                         *meta_ac = NULL;
2110                 }
2111
2112                 /*
2113                  * We cannot have an error and a non null *data_ac.
2114                  */
2115         }
2116
2117         return ret;
2118 }