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_INODES_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 (clean_error)                                        \
 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 clean_error)
 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 clean_error)
 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         if (le16_to_cpu(gd->bg_chain) >=
 237             le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) {
 238                 do_error("Group descriptor #%llu has bad chain %u",
 239                          (unsigned long long)bh->b_blocknr,
 240                          le16_to_cpu(gd->bg_chain));
 241                 return -EINVAL;
 242         }
 243
 244         return 0;
 245 }
 246
 247 #undef do_error
 248
 249 /*
 250  * This version only prints errors.  It does not fail the filesystem, and
 251  * exists only for resize.
 252  */
 253 int ocfs2_check_group_descriptor(struct super_block *sb,
 254                                  struct ocfs2_dinode *di,
 255                                  struct buffer_head *bh)
 256 {
 257         int rc;
 258         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 259
 260         BUG_ON(!buffer_uptodate(bh));
 261
 262         /*
 263          * If the ecc fails, we return the error but otherwise
 264          * leave the filesystem running.  We know any error is
 265          * local to this block.
 266          */
 267         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
 268         if (rc) {
 269                 mlog(ML_ERROR,
 270                      "Checksum failed for group descriptor %llu\n",
 271                      (unsigned long long)bh->b_blocknr);
 272         } else
 273                 rc = ocfs2_validate_gd_self(sb, bh, 1);
 274         if (!rc)
 275                 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
 276
 277         return rc;
 278 }
 279
 280 static int ocfs2_validate_group_descriptor(struct super_block *sb,
 281                                            struct buffer_head *bh)
 282 {
 283         int rc;
 284         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
 285
 286         mlog(0, "Validating group descriptor %llu\n",
 287              (unsigned long long)bh->b_blocknr);
 288
 289         BUG_ON(!buffer_uptodate(bh));
 290
 291         /*
 292          * If the ecc fails, we return the error but otherwise
 293          * leave the filesystem running.  We know any error is
 294          * local to this block.
 295          */
 296         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
 297         if (rc)
 298                 return rc;
 299
 300         /*
 301          * Errors after here are fatal.
 302          */
 303
 304         return ocfs2_validate_gd_self(sb, bh, 0);
 305 }
 306
 307 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
 308                                 u64 gd_blkno, struct buffer_head **bh)
 309 {
 310         int rc;
 311         struct buffer_head *tmp = *bh;
 312
 313         rc = ocfs2_read_block(inode, gd_blkno, &tmp,
 314                               ocfs2_validate_group_descriptor);
 315         if (rc)
 316                 goto out;
 317
 318         rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
 319         if (rc) {
 320                 brelse(tmp);
 321                 goto out;
 322         }
 323
 324         /* If ocfs2_read_block() got us a new bh, pass it up. */
 325         if (!*bh)
 326                 *bh = tmp;
 327
 328 out:
 329         return rc;
 330 }
 331
 332 static int ocfs2_block_group_fill(handle_t *handle,
 333                                   struct inode *alloc_inode,
 334                                   struct buffer_head *bg_bh,
 335                                   u64 group_blkno,
 336                                   u16 my_chain,
 337                                   struct ocfs2_chain_list *cl)
 338 {
 339         int status = 0;
 340         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 341         struct super_block * sb = alloc_inode->i_sb;
 342
 343         mlog_entry_void();
 344
 345         if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
 346                 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
 347                             "b_blocknr (%llu)",
 348                             (unsigned long long)group_blkno,
 349                             (unsigned long long) bg_bh->b_blocknr);
 350                 status = -EIO;
 351                 goto bail;
 352         }
 353
 354         status = ocfs2_journal_access_gd(handle,
 355                                          alloc_inode,
 356                                          bg_bh,
 357                                          OCFS2_JOURNAL_ACCESS_CREATE);
 358         if (status < 0) {
 359                 mlog_errno(status);
 360                 goto bail;
 361         }
 362
 363         memset(bg, 0, sb->s_blocksize);
 364         strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
 365         bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
 366         bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb));
 367         bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
 368         bg->bg_chain = cpu_to_le16(my_chain);
 369         bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
 370         bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
 371         bg->bg_blkno = cpu_to_le64(group_blkno);
 372         /* set the 1st bit in the bitmap to account for the descriptor block */
 373         ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
 374         bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
 375
 376         status = ocfs2_journal_dirty(handle, bg_bh);
 377         if (status < 0)
 378                 mlog_errno(status);
 379
 380         /* There is no need to zero out or otherwise initialize the
 381          * other blocks in a group - All valid FS metadata in a block
 382          * group stores the superblock fs_generation value at
 383          * allocation time. */
 384
 385 bail:
 386         mlog_exit(status);
 387         return status;
 388 }
 389
 390 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
 391 {
 392         u16 curr, best;
 393
 394         best = curr = 0;
 395         while (curr < le16_to_cpu(cl->cl_count)) {
 396                 if (le32_to_cpu(cl->cl_recs[best].c_total) >
 397                     le32_to_cpu(cl->cl_recs[curr].c_total))
 398                         best = curr;
 399                 curr++;
 400         }
 401         return best;
 402 }
 403
 404 /*
 405  * We expect the block group allocator to already be locked.
 406  */
 407 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
 408                                    struct inode *alloc_inode,
 409                                    struct buffer_head *bh,
 410                                    u64 max_block,
 411                                    u64 *last_alloc_group,
 412                                    int flags)
 413 {
 414         int status, credits;
 415         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
 416         struct ocfs2_chain_list *cl;
 417         struct ocfs2_alloc_context *ac = NULL;
 418         handle_t *handle = NULL;
 419         u32 bit_off, num_bits;
 420         u16 alloc_rec;
 421         u64 bg_blkno;
 422         struct buffer_head *bg_bh = NULL;
 423         struct ocfs2_group_desc *bg;
 424
 425         BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
 426
 427         mlog_entry_void();
 428
 429         cl = &fe->id2.i_chain;
 430         status = ocfs2_reserve_clusters_with_limit(osb,
 431                                                    le16_to_cpu(cl->cl_cpg),
 432                                                    max_block, flags, &ac);
 433         if (status < 0) {
 434                 if (status != -ENOSPC)
 435                         mlog_errno(status);
 436                 goto bail;
 437         }
 438
 439         credits = ocfs2_calc_group_alloc_credits(osb->sb,
 440                                                  le16_to_cpu(cl->cl_cpg));
 441         handle = ocfs2_start_trans(osb, credits);
 442         if (IS_ERR(handle)) {
 443                 status = PTR_ERR(handle);
 444                 handle = NULL;
 445                 mlog_errno(status);
 446                 goto bail;
 447         }
 448
 449         if (last_alloc_group && *last_alloc_group != 0) {
 450                 mlog(0, "use old allocation group %llu for block group alloc\n",
 451                      (unsigned long long)*last_alloc_group);
 452                 ac->ac_last_group = *last_alloc_group;
 453         }
 454         status = ocfs2_claim_clusters(osb,
 455                                       handle,
 456                                       ac,
 457                                       le16_to_cpu(cl->cl_cpg),
 458                                       &bit_off,
 459                                       &num_bits);
 460         if (status < 0) {
 461                 if (status != -ENOSPC)
 462                         mlog_errno(status);
 463                 goto bail;
 464         }
 465
 466         alloc_rec = ocfs2_find_smallest_chain(cl);
 467
 468         /* setup the group */
 469         bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
 470         mlog(0, "new descriptor, record %u, at block %llu\n",
 471              alloc_rec, (unsigned long long)bg_blkno);
 472
 473         bg_bh = sb_getblk(osb->sb, bg_blkno);
 474         if (!bg_bh) {
 475                 status = -EIO;
 476                 mlog_errno(status);
 477                 goto bail;
 478         }
 479         ocfs2_set_new_buffer_uptodate(alloc_inode, bg_bh);
 480
 481         status = ocfs2_block_group_fill(handle,
 482                                         alloc_inode,
 483                                         bg_bh,
 484                                         bg_blkno,
 485                                         alloc_rec,
 486                                         cl);
 487         if (status < 0) {
 488                 mlog_errno(status);
 489                 goto bail;
 490         }
 491
 492         bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 493
 494         status = ocfs2_journal_access_di(handle, alloc_inode,
 495                                          bh, OCFS2_JOURNAL_ACCESS_WRITE);
 496         if (status < 0) {
 497                 mlog_errno(status);
 498                 goto bail;
 499         }
 500
 501         le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
 502                      le16_to_cpu(bg->bg_free_bits_count));
 503         le32_add_cpu(&cl->cl_recs[alloc_rec].c_total, le16_to_cpu(bg->bg_bits));
 504         cl->cl_recs[alloc_rec].c_blkno  = cpu_to_le64(bg_blkno);
 505         if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
 506                 le16_add_cpu(&cl->cl_next_free_rec, 1);
 507
 508         le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
 509                                         le16_to_cpu(bg->bg_free_bits_count));
 510         le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
 511         le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
 512
 513         status = ocfs2_journal_dirty(handle, bh);
 514         if (status < 0) {
 515                 mlog_errno(status);
 516                 goto bail;
 517         }
 518
 519         spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
 520         OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
 521         fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
 522                                              le32_to_cpu(fe->i_clusters)));
 523         spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
 524         i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
 525         alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
 526
 527         status = 0;
 528
 529         /* save the new last alloc group so that the caller can cache it. */
 530         if (last_alloc_group)
 531                 *last_alloc_group = ac->ac_last_group;
 532
 533 bail:
 534         if (handle)
 535                 ocfs2_commit_trans(osb, handle);
 536
 537         if (ac)
 538                 ocfs2_free_alloc_context(ac);
 539
 540         brelse(bg_bh);
 541
 542         mlog_exit(status);
 543         return status;
 544 }
 545
 546 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
 547                                        struct ocfs2_alloc_context *ac,
 548                                        int type,
 549                                        u32 slot,
 550                                        u64 *last_alloc_group,
 551                                        int flags)
 552 {
 553         int status;
 554         u32 bits_wanted = ac->ac_bits_wanted;
 555         struct inode *alloc_inode;
 556         struct buffer_head *bh = NULL;
 557         struct ocfs2_dinode *fe;
 558         u32 free_bits;
 559
 560         mlog_entry_void();
 561
 562         alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
 563         if (!alloc_inode) {
 564                 mlog_errno(-EINVAL);
 565                 return -EINVAL;
 566         }
 567
 568         mutex_lock(&alloc_inode->i_mutex);
 569
 570         status = ocfs2_inode_lock(alloc_inode, &bh, 1);
 571         if (status < 0) {
 572                 mutex_unlock(&alloc_inode->i_mutex);
 573                 iput(alloc_inode);
 574
 575                 mlog_errno(status);
 576                 return status;
 577         }
 578
 579         ac->ac_inode = alloc_inode;
 580         ac->ac_alloc_slot = slot;
 581
 582         fe = (struct ocfs2_dinode *) bh->b_data;
 583
 584         /* The bh was validated by the inode read inside
 585          * ocfs2_inode_lock().  Any corruption is a code bug. */
 586         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
 587
 588         if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
 589                 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
 590                             (unsigned long long)le64_to_cpu(fe->i_blkno));
 591                 status = -EIO;
 592                 goto bail;
 593         }
 594
 595         free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
 596                 le32_to_cpu(fe->id1.bitmap1.i_used);
 597
 598         if (bits_wanted > free_bits) {
 599                 /* cluster bitmap never grows */
 600                 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
 601                         mlog(0, "Disk Full: wanted=%u, free_bits=%u\n",
 602                              bits_wanted, free_bits);
 603                         status = -ENOSPC;
 604                         goto bail;
 605                 }
 606
 607                 if (!(flags & ALLOC_NEW_GROUP)) {
 608                         mlog(0, "Alloc File %u Full: wanted=%u, free_bits=%u, "
 609                              "and we don't alloc a new group for it.\n",
 610                              slot, bits_wanted, free_bits);
 611                         status = -ENOSPC;
 612                         goto bail;
 613                 }
 614
 615                 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
 616                                                  ac->ac_max_block,
 617                                                  last_alloc_group, flags);
 618                 if (status < 0) {
 619                         if (status != -ENOSPC)
 620                                 mlog_errno(status);
 621                         goto bail;
 622                 }
 623                 atomic_inc(&osb->alloc_stats.bg_extends);
 624
 625                 /* You should never ask for this much metadata */
 626                 BUG_ON(bits_wanted >
 627                        (le32_to_cpu(fe->id1.bitmap1.i_total)
 628                         - le32_to_cpu(fe->id1.bitmap1.i_used)));
 629         }
 630
 631         get_bh(bh);
 632         ac->ac_bh = bh;
 633 bail:
 634         brelse(bh);
 635
 636         mlog_exit(status);
 637         return status;
 638 }
 639
 640 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
 641                                       int blocks,
 642                                       struct ocfs2_alloc_context **ac)
 643 {
 644         int status;
 645         u32 slot;
 646
 647         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
 648         if (!(*ac)) {
 649                 status = -ENOMEM;
 650                 mlog_errno(status);
 651                 goto bail;
 652         }
 653
 654         (*ac)->ac_bits_wanted = blocks;
 655         (*ac)->ac_which = OCFS2_AC_USE_META;
 656         slot = osb->slot_num;
 657         (*ac)->ac_group_search = ocfs2_block_group_search;
 658
 659         status = ocfs2_reserve_suballoc_bits(osb, (*ac),
 660                                              EXTENT_ALLOC_SYSTEM_INODE,
 661                                              slot, NULL, ALLOC_NEW_GROUP);
 662         if (status < 0) {
 663                 if (status != -ENOSPC)
 664                         mlog_errno(status);
 665                 goto bail;
 666         }
 667
 668         status = 0;
 669 bail:
 670         if ((status < 0) && *ac) {
 671                 ocfs2_free_alloc_context(*ac);
 672                 *ac = NULL;
 673         }
 674
 675         mlog_exit(status);
 676         return status;
 677 }
 678
 679 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
 680                                struct ocfs2_extent_list *root_el,
 681                                struct ocfs2_alloc_context **ac)
 682 {
 683         return ocfs2_reserve_new_metadata_blocks(osb,
 684                                         ocfs2_extend_meta_needed(root_el),
 685                                         ac);
 686 }
 687
 688 static int ocfs2_steal_inode_from_other_nodes(struct ocfs2_super *osb,
 689                                               struct ocfs2_alloc_context *ac)
 690 {
 691         int i, status = -ENOSPC;
 692         s16 slot = ocfs2_get_inode_steal_slot(osb);
 693
 694         /* Start to steal inodes from the first slot after ours. */
 695         if (slot == OCFS2_INVALID_SLOT)
 696                 slot = osb->slot_num + 1;
 697
 698         for (i = 0; i < osb->max_slots; i++, slot++) {
 699                 if (slot == osb->max_slots)
 700                         slot = 0;
 701
 702                 if (slot == osb->slot_num)
 703                         continue;
 704
 705                 status = ocfs2_reserve_suballoc_bits(osb, ac,
 706                                                      INODE_ALLOC_SYSTEM_INODE,
 707                                                      slot, NULL,
 708                                                      NOT_ALLOC_NEW_GROUP);
 709                 if (status >= 0) {
 710                         ocfs2_set_inode_steal_slot(osb, slot);
 711                         break;
 712                 }
 713
 714                 ocfs2_free_ac_resource(ac);
 715         }
 716
 717         return status;
 718 }
 719
 720 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
 721                             struct ocfs2_alloc_context **ac)
 722 {
 723         int status;
 724         s16 slot = ocfs2_get_inode_steal_slot(osb);
 725         u64 alloc_group;
 726
 727         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
 728         if (!(*ac)) {
 729                 status = -ENOMEM;
 730                 mlog_errno(status);
 731                 goto bail;
 732         }
 733
 734         (*ac)->ac_bits_wanted = 1;
 735         (*ac)->ac_which = OCFS2_AC_USE_INODE;
 736
 737         (*ac)->ac_group_search = ocfs2_block_group_search;
 738
 739         /*
 740          * stat(2) can't handle i_ino > 32bits, so we tell the
 741          * lower levels not to allocate us a block group past that
 742          * limit.  The 'inode64' mount option avoids this behavior.
 743          */
 744         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
 745                 (*ac)->ac_max_block = (u32)~0U;
 746
 747         /*
 748          * slot is set when we successfully steal inode from other nodes.
 749          * It is reset in 3 places:
 750          * 1. when we flush the truncate log
 751          * 2. when we complete local alloc recovery.
 752          * 3. when we successfully allocate from our own slot.
 753          * After it is set, we will go on stealing inodes until we find the
 754          * need to check our slots to see whether there is some space for us.
 755          */
 756         if (slot != OCFS2_INVALID_SLOT &&
 757             atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_INODES_TO_STEAL)
 758                 goto inode_steal;
 759
 760         atomic_set(&osb->s_num_inodes_stolen, 0);
 761         alloc_group = osb->osb_inode_alloc_group;
 762         status = ocfs2_reserve_suballoc_bits(osb, *ac,
 763                                              INODE_ALLOC_SYSTEM_INODE,
 764                                              osb->slot_num,
 765                                              &alloc_group,
 766                                              ALLOC_NEW_GROUP |
 767                                              ALLOC_GROUPS_FROM_GLOBAL);
 768         if (status >= 0) {
 769                 status = 0;
 770
 771                 spin_lock(&osb->osb_lock);
 772                 osb->osb_inode_alloc_group = alloc_group;
 773                 spin_unlock(&osb->osb_lock);
 774                 mlog(0, "after reservation, new allocation group is "
 775                      "%llu\n", (unsigned long long)alloc_group);
 776
 777                 /*
 778                  * Some inodes must be freed by us, so try to allocate
 779                  * from our own next time.
 780                  */
 781                 if (slot != OCFS2_INVALID_SLOT)
 782                         ocfs2_init_inode_steal_slot(osb);
 783                 goto bail;
 784         } else if (status < 0 && status != -ENOSPC) {
 785                 mlog_errno(status);
 786                 goto bail;
 787         }
 788
 789         ocfs2_free_ac_resource(*ac);
 790
 791 inode_steal:
 792         status = ocfs2_steal_inode_from_other_nodes(osb, *ac);
 793         atomic_inc(&osb->s_num_inodes_stolen);
 794         if (status < 0) {
 795                 if (status != -ENOSPC)
 796                         mlog_errno(status);
 797                 goto bail;
 798         }
 799
 800         status = 0;
 801 bail:
 802         if ((status < 0) && *ac) {
 803                 ocfs2_free_alloc_context(*ac);
 804                 *ac = NULL;
 805         }
 806
 807         mlog_exit(status);
 808         return status;
 809 }
 810
 811 /* local alloc code has to do the same thing, so rather than do this
 812  * twice.. */
 813 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
 814                                       struct ocfs2_alloc_context *ac)
 815 {
 816         int status;
 817
 818         ac->ac_which = OCFS2_AC_USE_MAIN;
 819         ac->ac_group_search = ocfs2_cluster_group_search;
 820
 821         status = ocfs2_reserve_suballoc_bits(osb, ac,
 822                                              GLOBAL_BITMAP_SYSTEM_INODE,
 823                                              OCFS2_INVALID_SLOT, NULL,
 824                                              ALLOC_NEW_GROUP);
 825         if (status < 0 && status != -ENOSPC) {
 826                 mlog_errno(status);
 827                 goto bail;
 828         }
 829
 830 bail:
 831         return status;
 832 }
 833
 834 /* Callers don't need to care which bitmap (local alloc or main) to
 835  * use so we figure it out for them, but unfortunately this clutters
 836  * things a bit. */
 837 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
 838                                              u32 bits_wanted, u64 max_block,
 839                                              int flags,
 840                                              struct ocfs2_alloc_context **ac)
 841 {
 842         int status;
 843
 844         mlog_entry_void();
 845
 846         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
 847         if (!(*ac)) {
 848                 status = -ENOMEM;
 849                 mlog_errno(status);
 850                 goto bail;
 851         }
 852
 853         (*ac)->ac_bits_wanted = bits_wanted;
 854         (*ac)->ac_max_block = max_block;
 855
 856         status = -ENOSPC;
 857         if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
 858             ocfs2_alloc_should_use_local(osb, bits_wanted)) {
 859                 status = ocfs2_reserve_local_alloc_bits(osb,
 860                                                         bits_wanted,
 861                                                         *ac);
 862                 if (status == -EFBIG) {
 863                         /* The local alloc window is outside ac_max_block.
 864                          * use the main bitmap. */
 865                         status = -ENOSPC;
 866                 } else if ((status < 0) && (status != -ENOSPC)) {
 867                         mlog_errno(status);
 868                         goto bail;
 869                 }
 870         }
 871
 872         if (status == -ENOSPC) {
 873                 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
 874                 if (status < 0) {
 875                         if (status != -ENOSPC)
 876                                 mlog_errno(status);
 877                         goto bail;
 878                 }
 879         }
 880
 881         status = 0;
 882 bail:
 883         if ((status < 0) && *ac) {
 884                 ocfs2_free_alloc_context(*ac);
 885                 *ac = NULL;
 886         }
 887
 888         mlog_exit(status);
 889         return status;
 890 }
 891
 892 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
 893                            u32 bits_wanted,
 894                            struct ocfs2_alloc_context **ac)
 895 {
 896         return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
 897                                                  ALLOC_NEW_GROUP, ac);
 898 }
 899
 900 /*
 901  * More or less lifted from ext3. I'll leave their description below:
 902  *
 903  * "For ext3 allocations, we must not reuse any blocks which are
 904  * allocated in the bitmap buffer's "last committed data" copy.  This
 905  * prevents deletes from freeing up the page for reuse until we have
 906  * committed the delete transaction.
 907  *
 908  * If we didn't do this, then deleting something and reallocating it as
 909  * data would allow the old block to be overwritten before the
 910  * transaction committed (because we force data to disk before commit).
 911  * This would lead to corruption if we crashed between overwriting the
 912  * data and committing the delete.
 913  *
 914  * @@@ We may want to make this allocation behaviour conditional on
 915  * data-writes at some point, and disable it for metadata allocations or
 916  * sync-data inodes."
 917  *
 918  * Note: OCFS2 already does this differently for metadata vs data
 919  * allocations, as those bitmaps are separate and undo access is never
 920  * called on a metadata group descriptor.
 921  */
 922 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
 923                                          int nr)
 924 {
 925         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 926
 927         if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
 928                 return 0;
 929         if (!buffer_jbd(bg_bh) || !bh2jh(bg_bh)->b_committed_data)
 930                 return 1;
 931
 932         bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
 933         return !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
 934 }
 935
 936 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
 937                                              struct buffer_head *bg_bh,
 938                                              unsigned int bits_wanted,
 939                                              unsigned int total_bits,
 940                                              u16 *bit_off,
 941                                              u16 *bits_found)
 942 {
 943         void *bitmap;
 944         u16 best_offset, best_size;
 945         int offset, start, found, status = 0;
 946         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
 947
 948         /* Callers got this descriptor from
 949          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
 950         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
 951
 952         found = start = best_offset = best_size = 0;
 953         bitmap = bg->bg_bitmap;
 954
 955         while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
 956                 if (offset == total_bits)
 957                         break;
 958
 959                 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
 960                         /* We found a zero, but we can't use it as it
 961                          * hasn't been put to disk yet! */
 962                         found = 0;
 963                         start = offset + 1;
 964                 } else if (offset == start) {
 965                         /* we found a zero */
 966                         found++;
 967                         /* move start to the next bit to test */
 968                         start++;
 969                 } else {
 970                         /* got a zero after some ones */
 971                         found = 1;
 972                         start = offset + 1;
 973                 }
 974                 if (found > best_size) {
 975                         best_size = found;
 976                         best_offset = start - found;
 977                 }
 978                 /* we got everything we needed */
 979                 if (found == bits_wanted) {
 980                         /* mlog(0, "Found it all!\n"); */
 981                         break;
 982                 }
 983         }
 984
 985         /* XXX: I think the first clause is equivalent to the second
 986          *      - jlbec */
 987         if (found == bits_wanted) {
 988                 *bit_off = start - found;
 989                 *bits_found = found;
 990         } else if (best_size) {
 991                 *bit_off = best_offset;
 992                 *bits_found = best_size;
 993         } else {
 994                 status = -ENOSPC;
 995                 /* No error log here -- see the comment above
 996                  * ocfs2_test_bg_bit_allocatable */
 997         }
 998
 999         return status;
1000 }
1001
1002 static inline int ocfs2_block_group_set_bits(handle_t *handle,
1003                                              struct inode *alloc_inode,
1004                                              struct ocfs2_group_desc *bg,
1005                                              struct buffer_head *group_bh,
1006                                              unsigned int bit_off,
1007                                              unsigned int num_bits)
1008 {
1009         int status;
1010         void *bitmap = bg->bg_bitmap;
1011         int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1012
1013         mlog_entry_void();
1014
1015         /* All callers get the descriptor via
1016          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1017         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1018         BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1019
1020         mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
1021              num_bits);
1022
1023         if (ocfs2_is_cluster_bitmap(alloc_inode))
1024                 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1025
1026         status = ocfs2_journal_access_gd(handle,
1027                                          alloc_inode,
1028                                          group_bh,
1029                                          journal_type);
1030         if (status < 0) {
1031                 mlog_errno(status);
1032                 goto bail;
1033         }
1034
1035         le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1036
1037         while(num_bits--)
1038                 ocfs2_set_bit(bit_off++, bitmap);
1039
1040         status = ocfs2_journal_dirty(handle,
1041                                      group_bh);
1042         if (status < 0) {
1043                 mlog_errno(status);
1044                 goto bail;
1045         }
1046
1047 bail:
1048         mlog_exit(status);
1049         return status;
1050 }
1051
1052 /* find the one with the most empty bits */
1053 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1054 {
1055         u16 curr, best;
1056
1057         BUG_ON(!cl->cl_next_free_rec);
1058
1059         best = curr = 0;
1060         while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1061                 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1062                     le32_to_cpu(cl->cl_recs[best].c_free))
1063                         best = curr;
1064                 curr++;
1065         }
1066
1067         BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1068         return best;
1069 }
1070
1071 static int ocfs2_relink_block_group(handle_t *handle,
1072                                     struct inode *alloc_inode,
1073                                     struct buffer_head *fe_bh,
1074                                     struct buffer_head *bg_bh,
1075                                     struct buffer_head *prev_bg_bh,
1076                                     u16 chain)
1077 {
1078         int status;
1079         /* there is a really tiny chance the journal calls could fail,
1080          * but we wouldn't want inconsistent blocks in *any* case. */
1081         u64 fe_ptr, bg_ptr, prev_bg_ptr;
1082         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1083         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1084         struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1085
1086         /* The caller got these descriptors from
1087          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1088         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1089         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1090
1091         mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
1092              (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1093              (unsigned long long)le64_to_cpu(bg->bg_blkno),
1094              (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1095
1096         fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
1097         bg_ptr = le64_to_cpu(bg->bg_next_group);
1098         prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1099
1100         status = ocfs2_journal_access_gd(handle, alloc_inode, prev_bg_bh,
1101                                          OCFS2_JOURNAL_ACCESS_WRITE);
1102         if (status < 0) {
1103                 mlog_errno(status);
1104                 goto out_rollback;
1105         }
1106
1107         prev_bg->bg_next_group = bg->bg_next_group;
1108
1109         status = ocfs2_journal_dirty(handle, prev_bg_bh);
1110         if (status < 0) {
1111                 mlog_errno(status);
1112                 goto out_rollback;
1113         }
1114
1115         status = ocfs2_journal_access_gd(handle, alloc_inode, bg_bh,
1116                                          OCFS2_JOURNAL_ACCESS_WRITE);
1117         if (status < 0) {
1118                 mlog_errno(status);
1119                 goto out_rollback;
1120         }
1121
1122         bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1123
1124         status = ocfs2_journal_dirty(handle, bg_bh);
1125         if (status < 0) {
1126                 mlog_errno(status);
1127                 goto out_rollback;
1128         }
1129
1130         status = ocfs2_journal_access_di(handle, alloc_inode, fe_bh,
1131                                          OCFS2_JOURNAL_ACCESS_WRITE);
1132         if (status < 0) {
1133                 mlog_errno(status);
1134                 goto out_rollback;
1135         }
1136
1137         fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1138
1139         status = ocfs2_journal_dirty(handle, fe_bh);
1140         if (status < 0) {
1141                 mlog_errno(status);
1142                 goto out_rollback;
1143         }
1144
1145         status = 0;
1146 out_rollback:
1147         if (status < 0) {
1148                 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
1149                 bg->bg_next_group = cpu_to_le64(bg_ptr);
1150                 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1151         }
1152
1153         mlog_exit(status);
1154         return status;
1155 }
1156
1157 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1158                                                      u32 wanted)
1159 {
1160         return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1161 }
1162
1163 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1164  * value on error. */
1165 static int ocfs2_cluster_group_search(struct inode *inode,
1166                                       struct buffer_head *group_bh,
1167                                       u32 bits_wanted, u32 min_bits,
1168                                       u64 max_block,
1169                                       u16 *bit_off, u16 *bits_found)
1170 {
1171         int search = -ENOSPC;
1172         int ret;
1173         u64 blkoff;
1174         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1175         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1176         u16 tmp_off, tmp_found;
1177         unsigned int max_bits, gd_cluster_off;
1178
1179         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1180
1181         if (gd->bg_free_bits_count) {
1182                 max_bits = le16_to_cpu(gd->bg_bits);
1183
1184                 /* Tail groups in cluster bitmaps which aren't cpg
1185                  * aligned are prone to partial extention by a failed
1186                  * fs resize. If the file system resize never got to
1187                  * update the dinode cluster count, then we don't want
1188                  * to trust any clusters past it, regardless of what
1189                  * the group descriptor says. */
1190                 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1191                                                           le64_to_cpu(gd->bg_blkno));
1192                 if ((gd_cluster_off + max_bits) >
1193                     OCFS2_I(inode)->ip_clusters) {
1194                         max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1195                         mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
1196                              (unsigned long long)le64_to_cpu(gd->bg_blkno),
1197                              le16_to_cpu(gd->bg_bits),
1198                              OCFS2_I(inode)->ip_clusters, max_bits);
1199                 }
1200
1201                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1202                                                         group_bh, bits_wanted,
1203                                                         max_bits,
1204                                                         &tmp_off, &tmp_found);
1205                 if (ret)
1206                         return ret;
1207
1208                 if (max_block) {
1209                         blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1210                                                           gd_cluster_off +
1211                                                           tmp_off + tmp_found);
1212                         mlog(0, "Checking %llu against %llu\n",
1213                              (unsigned long long)blkoff,
1214                              (unsigned long long)max_block);
1215                         if (blkoff > max_block)
1216                                 return -ENOSPC;
1217                 }
1218
1219                 /* ocfs2_block_group_find_clear_bits() might
1220                  * return success, but we still want to return
1221                  * -ENOSPC unless it found the minimum number
1222                  * of bits. */
1223                 if (min_bits <= tmp_found) {
1224                         *bit_off = tmp_off;
1225                         *bits_found = tmp_found;
1226                         search = 0; /* success */
1227                 } else if (tmp_found) {
1228                         /*
1229                          * Don't show bits which we'll be returning
1230                          * for allocation to the local alloc bitmap.
1231                          */
1232                         ocfs2_local_alloc_seen_free_bits(osb, tmp_found);
1233                 }
1234         }
1235
1236         return search;
1237 }
1238
1239 static int ocfs2_block_group_search(struct inode *inode,
1240                                     struct buffer_head *group_bh,
1241                                     u32 bits_wanted, u32 min_bits,
1242                                     u64 max_block,
1243                                     u16 *bit_off, u16 *bits_found)
1244 {
1245         int ret = -ENOSPC;
1246         u64 blkoff;
1247         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1248
1249         BUG_ON(min_bits != 1);
1250         BUG_ON(ocfs2_is_cluster_bitmap(inode));
1251
1252         if (bg->bg_free_bits_count) {
1253                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1254                                                         group_bh, bits_wanted,
1255                                                         le16_to_cpu(bg->bg_bits),
1256                                                         bit_off, bits_found);
1257                 if (!ret && max_block) {
1258                         blkoff = le64_to_cpu(bg->bg_blkno) + *bit_off +
1259                                 *bits_found;
1260                         mlog(0, "Checking %llu against %llu\n",
1261                              (unsigned long long)blkoff,
1262                              (unsigned long long)max_block);
1263                         if (blkoff > max_block)
1264                                 ret = -ENOSPC;
1265                 }
1266         }
1267
1268         return ret;
1269 }
1270
1271 static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1272                                        handle_t *handle,
1273                                        struct buffer_head *di_bh,
1274                                        u32 num_bits,
1275                                        u16 chain)
1276 {
1277         int ret;
1278         u32 tmp_used;
1279         struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1280         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1281
1282         ret = ocfs2_journal_access_di(handle, inode, di_bh,
1283                                       OCFS2_JOURNAL_ACCESS_WRITE);
1284         if (ret < 0) {
1285                 mlog_errno(ret);
1286                 goto out;
1287         }
1288
1289         tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1290         di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1291         le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1292
1293         ret = ocfs2_journal_dirty(handle, di_bh);
1294         if (ret < 0)
1295                 mlog_errno(ret);
1296
1297 out:
1298         return ret;
1299 }
1300
1301 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1302                                   handle_t *handle,
1303                                   u32 bits_wanted,
1304                                   u32 min_bits,
1305                                   u16 *bit_off,
1306                                   unsigned int *num_bits,
1307                                   u64 gd_blkno,
1308                                   u16 *bits_left)
1309 {
1310         int ret;
1311         u16 found;
1312         struct buffer_head *group_bh = NULL;
1313         struct ocfs2_group_desc *gd;
1314         struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1315         struct inode *alloc_inode = ac->ac_inode;
1316
1317         ret = ocfs2_read_group_descriptor(alloc_inode, di, gd_blkno,
1318                                           &group_bh);
1319         if (ret < 0) {
1320                 mlog_errno(ret);
1321                 return ret;
1322         }
1323
1324         gd = (struct ocfs2_group_desc *) group_bh->b_data;
1325         ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1326                                   ac->ac_max_block, bit_off, &found);
1327         if (ret < 0) {
1328                 if (ret != -ENOSPC)
1329                         mlog_errno(ret);
1330                 goto out;
1331         }
1332
1333         *num_bits = found;
1334
1335         ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1336                                                *num_bits,
1337                                                le16_to_cpu(gd->bg_chain));
1338         if (ret < 0) {
1339                 mlog_errno(ret);
1340                 goto out;
1341         }
1342
1343         ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1344                                          *bit_off, *num_bits);
1345         if (ret < 0)
1346                 mlog_errno(ret);
1347
1348         *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1349
1350 out:
1351         brelse(group_bh);
1352
1353         return ret;
1354 }
1355
1356 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1357                               handle_t *handle,
1358                               u32 bits_wanted,
1359                               u32 min_bits,
1360                               u16 *bit_off,
1361                               unsigned int *num_bits,
1362                               u64 *bg_blkno,
1363                               u16 *bits_left)
1364 {
1365         int status;
1366         u16 chain, tmp_bits;
1367         u32 tmp_used;
1368         u64 next_group;
1369         struct inode *alloc_inode = ac->ac_inode;
1370         struct buffer_head *group_bh = NULL;
1371         struct buffer_head *prev_group_bh = NULL;
1372         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1373         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1374         struct ocfs2_group_desc *bg;
1375
1376         chain = ac->ac_chain;
1377         mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
1378              bits_wanted, chain,
1379              (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);
1380
1381         status = ocfs2_read_group_descriptor(alloc_inode, fe,
1382                                              le64_to_cpu(cl->cl_recs[chain].c_blkno),
1383                                              &group_bh);
1384         if (status < 0) {
1385                 mlog_errno(status);
1386                 goto bail;
1387         }
1388         bg = (struct ocfs2_group_desc *) group_bh->b_data;
1389
1390         status = -ENOSPC;
1391         /* for now, the chain search is a bit simplistic. We just use
1392          * the 1st group with any empty bits. */
1393         while ((status = ac->ac_group_search(alloc_inode, group_bh,
1394                                              bits_wanted, min_bits,
1395                                              ac->ac_max_block, bit_off,
1396                                              &tmp_bits)) == -ENOSPC) {
1397                 if (!bg->bg_next_group)
1398                         break;
1399
1400                 brelse(prev_group_bh);
1401                 prev_group_bh = NULL;
1402
1403                 next_group = le64_to_cpu(bg->bg_next_group);
1404                 prev_group_bh = group_bh;
1405                 group_bh = NULL;
1406                 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1407                                                      next_group, &group_bh);
1408                 if (status < 0) {
1409                         mlog_errno(status);
1410                         goto bail;
1411                 }
1412                 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1413         }
1414         if (status < 0) {
1415                 if (status != -ENOSPC)
1416                         mlog_errno(status);
1417                 goto bail;
1418         }
1419
1420         mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
1421              tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));
1422
1423         *num_bits = tmp_bits;
1424
1425         BUG_ON(*num_bits == 0);
1426
1427         /*
1428          * Keep track of previous block descriptor read. When
1429          * we find a target, if we have read more than X
1430          * number of descriptors, and the target is reasonably
1431          * empty, relink him to top of his chain.
1432          *
1433          * We've read 0 extra blocks and only send one more to
1434          * the transaction, yet the next guy to search has a
1435          * much easier time.
1436          *
1437          * Do this *after* figuring out how many bits we're taking out
1438          * of our target group.
1439          */
1440         if (ac->ac_allow_chain_relink &&
1441             (prev_group_bh) &&
1442             (ocfs2_block_group_reasonably_empty(bg, *num_bits))) {
1443                 status = ocfs2_relink_block_group(handle, alloc_inode,
1444                                                   ac->ac_bh, group_bh,
1445                                                   prev_group_bh, chain);
1446                 if (status < 0) {
1447                         mlog_errno(status);
1448                         goto bail;
1449                 }
1450         }
1451
1452         /* Ok, claim our bits now: set the info on dinode, chainlist
1453          * and then the group */
1454         status = ocfs2_journal_access_di(handle,
1455                                          alloc_inode,
1456                                          ac->ac_bh,
1457                                          OCFS2_JOURNAL_ACCESS_WRITE);
1458         if (status < 0) {
1459                 mlog_errno(status);
1460                 goto bail;
1461         }
1462
1463         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1464         fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used);
1465         le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits));
1466
1467         status = ocfs2_journal_dirty(handle,
1468                                      ac->ac_bh);
1469         if (status < 0) {
1470                 mlog_errno(status);
1471                 goto bail;
1472         }
1473
1474         status = ocfs2_block_group_set_bits(handle,
1475                                             alloc_inode,
1476                                             bg,
1477                                             group_bh,
1478                                             *bit_off,
1479                                             *num_bits);
1480         if (status < 0) {
1481                 mlog_errno(status);
1482                 goto bail;
1483         }
1484
1485         mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
1486              (unsigned long long)le64_to_cpu(fe->i_blkno));
1487
1488         *bg_blkno = le64_to_cpu(bg->bg_blkno);
1489         *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1490 bail:
1491         brelse(group_bh);
1492         brelse(prev_group_bh);
1493
1494         mlog_exit(status);
1495         return status;
1496 }
1497
1498 /* will give out up to bits_wanted contiguous bits. */
1499 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
1500                                      struct ocfs2_alloc_context *ac,
1501                                      handle_t *handle,
1502                                      u32 bits_wanted,
1503                                      u32 min_bits,
1504                                      u16 *bit_off,
1505                                      unsigned int *num_bits,
1506                                      u64 *bg_blkno)
1507 {
1508         int status;
1509         u16 victim, i;
1510         u16 bits_left = 0;
1511         u64 hint_blkno = ac->ac_last_group;
1512         struct ocfs2_chain_list *cl;
1513         struct ocfs2_dinode *fe;
1514
1515         mlog_entry_void();
1516
1517         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1518         BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1519         BUG_ON(!ac->ac_bh);
1520
1521         fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1522
1523         /* The bh was validated by the inode read during
1524          * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1525         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1526
1527         if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1528             le32_to_cpu(fe->id1.bitmap1.i_total)) {
1529                 ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used "
1530                             "bits but only %u total.",
1531                             (unsigned long long)le64_to_cpu(fe->i_blkno),
1532                             le32_to_cpu(fe->id1.bitmap1.i_used),
1533                             le32_to_cpu(fe->id1.bitmap1.i_total));
1534                 status = -EIO;
1535                 goto bail;
1536         }
1537
1538         if (hint_blkno) {
1539                 /* Attempt to short-circuit the usual search mechanism
1540                  * by jumping straight to the most recently used
1541                  * allocation group. This helps us mantain some
1542                  * contiguousness across allocations. */
1543                 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1544                                                 min_bits, bit_off, num_bits,
1545                                                 hint_blkno, &bits_left);
1546                 if (!status) {
1547                         /* Be careful to update *bg_blkno here as the
1548                          * caller is expecting it to be filled in, and
1549                          * ocfs2_search_one_group() won't do that for
1550                          * us. */
1551                         *bg_blkno = hint_blkno;
1552                         goto set_hint;
1553                 }
1554                 if (status < 0 && status != -ENOSPC) {
1555                         mlog_errno(status);
1556                         goto bail;
1557                 }
1558         }
1559
1560         cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1561
1562         victim = ocfs2_find_victim_chain(cl);
1563         ac->ac_chain = victim;
1564         ac->ac_allow_chain_relink = 1;
1565
1566         status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, bit_off,
1567                                     num_bits, bg_blkno, &bits_left);
1568         if (!status)
1569                 goto set_hint;
1570         if (status < 0 && status != -ENOSPC) {
1571                 mlog_errno(status);
1572                 goto bail;
1573         }
1574
1575         mlog(0, "Search of victim chain %u came up with nothing, "
1576              "trying all chains now.\n", victim);
1577
1578         /* If we didn't pick a good victim, then just default to
1579          * searching each chain in order. Don't allow chain relinking
1580          * because we only calculate enough journal credits for one
1581          * relink per alloc. */
1582         ac->ac_allow_chain_relink = 0;
1583         for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1584                 if (i == victim)
1585                         continue;
1586                 if (!cl->cl_recs[i].c_free)
1587                         continue;
1588
1589                 ac->ac_chain = i;
1590                 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1591                                             bit_off, num_bits, bg_blkno,
1592                                             &bits_left);
1593                 if (!status)
1594                         break;
1595                 if (status < 0 && status != -ENOSPC) {
1596                         mlog_errno(status);
1597                         goto bail;
1598                 }
1599         }
1600
1601 set_hint:
1602         if (status != -ENOSPC) {
1603                 /* If the next search of this group is not likely to
1604                  * yield a suitable extent, then we reset the last
1605                  * group hint so as to not waste a disk read */
1606                 if (bits_left < min_bits)
1607                         ac->ac_last_group = 0;
1608                 else
1609                         ac->ac_last_group = *bg_blkno;
1610         }
1611
1612 bail:
1613         mlog_exit(status);
1614         return status;
1615 }
1616
1617 int ocfs2_claim_metadata(struct ocfs2_super *osb,
1618                          handle_t *handle,
1619                          struct ocfs2_alloc_context *ac,
1620                          u32 bits_wanted,
1621                          u16 *suballoc_bit_start,
1622                          unsigned int *num_bits,
1623                          u64 *blkno_start)
1624 {
1625         int status;
1626         u64 bg_blkno;
1627
1628         BUG_ON(!ac);
1629         BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1630         BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1631
1632         status = ocfs2_claim_suballoc_bits(osb,
1633                                            ac,
1634                                            handle,
1635                                            bits_wanted,
1636                                            1,
1637                                            suballoc_bit_start,
1638                                            num_bits,
1639                                            &bg_blkno);
1640         if (status < 0) {
1641                 mlog_errno(status);
1642                 goto bail;
1643         }
1644         atomic_inc(&osb->alloc_stats.bg_allocs);
1645
1646         *blkno_start = bg_blkno + (u64) *suballoc_bit_start;
1647         ac->ac_bits_given += (*num_bits);
1648         status = 0;
1649 bail:
1650         mlog_exit(status);
1651         return status;
1652 }
1653
1654 static void ocfs2_init_inode_ac_group(struct inode *dir,
1655                                       struct buffer_head *parent_fe_bh,
1656                                       struct ocfs2_alloc_context *ac)
1657 {
1658         struct ocfs2_dinode *fe = (struct ocfs2_dinode *)parent_fe_bh->b_data;
1659         /*
1660          * Try to allocate inodes from some specific group.
1661          *
1662          * If the parent dir has recorded the last group used in allocation,
1663          * cool, use it. Otherwise if we try to allocate new inode from the
1664          * same slot the parent dir belongs to, use the same chunk.
1665          *
1666          * We are very careful here to avoid the mistake of setting
1667          * ac_last_group to a group descriptor from a different (unlocked) slot.
1668          */
1669         if (OCFS2_I(dir)->ip_last_used_group &&
1670             OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
1671                 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
1672         else if (le16_to_cpu(fe->i_suballoc_slot) == ac->ac_alloc_slot)
1673                 ac->ac_last_group = ocfs2_which_suballoc_group(
1674                                         le64_to_cpu(fe->i_blkno),
1675                                         le16_to_cpu(fe->i_suballoc_bit));
1676 }
1677
1678 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
1679                                              struct ocfs2_alloc_context *ac)
1680 {
1681         OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
1682         OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
1683 }
1684
1685 int ocfs2_claim_new_inode(struct ocfs2_super *osb,
1686                           handle_t *handle,
1687                           struct inode *dir,
1688                           struct buffer_head *parent_fe_bh,
1689                           struct ocfs2_alloc_context *ac,
1690                           u16 *suballoc_bit,
1691                           u64 *fe_blkno)
1692 {
1693         int status;
1694         unsigned int num_bits;
1695         u64 bg_blkno;
1696
1697         mlog_entry_void();
1698
1699         BUG_ON(!ac);
1700         BUG_ON(ac->ac_bits_given != 0);
1701         BUG_ON(ac->ac_bits_wanted != 1);
1702         BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
1703
1704         ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
1705
1706         status = ocfs2_claim_suballoc_bits(osb,
1707                                            ac,
1708                                            handle,
1709                                            1,
1710                                            1,
1711                                            suballoc_bit,
1712                                            &num_bits,
1713                                            &bg_blkno);
1714         if (status < 0) {
1715                 mlog_errno(status);
1716                 goto bail;
1717         }
1718         atomic_inc(&osb->alloc_stats.bg_allocs);
1719
1720         BUG_ON(num_bits != 1);
1721
1722         *fe_blkno = bg_blkno + (u64) (*suballoc_bit);
1723         ac->ac_bits_given++;
1724         ocfs2_save_inode_ac_group(dir, ac);
1725         status = 0;
1726 bail:
1727         mlog_exit(status);
1728         return status;
1729 }
1730
1731 /* translate a group desc. blkno and it's bitmap offset into
1732  * disk cluster offset. */
1733 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
1734                                                    u64 bg_blkno,
1735                                                    u16 bg_bit_off)
1736 {
1737         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1738         u32 cluster = 0;
1739
1740         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1741
1742         if (bg_blkno != osb->first_cluster_group_blkno)
1743                 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
1744         cluster += (u32) bg_bit_off;
1745         return cluster;
1746 }
1747
1748 /* given a cluster offset, calculate which block group it belongs to
1749  * and return that block offset. */
1750 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
1751 {
1752         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1753         u32 group_no;
1754
1755         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1756
1757         group_no = cluster / osb->bitmap_cpg;
1758         if (!group_no)
1759                 return osb->first_cluster_group_blkno;
1760         return ocfs2_clusters_to_blocks(inode->i_sb,
1761                                         group_no * osb->bitmap_cpg);
1762 }
1763
1764 /* given the block number of a cluster start, calculate which cluster
1765  * group and descriptor bitmap offset that corresponds to. */
1766 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
1767                                                 u64 data_blkno,
1768                                                 u64 *bg_blkno,
1769                                                 u16 *bg_bit_off)
1770 {
1771         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1772         u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
1773
1774         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1775
1776         *bg_blkno = ocfs2_which_cluster_group(inode,
1777                                               data_cluster);
1778
1779         if (*bg_blkno == osb->first_cluster_group_blkno)
1780                 *bg_bit_off = (u16) data_cluster;
1781         else
1782                 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
1783                                                              data_blkno - *bg_blkno);
1784 }
1785
1786 /*
1787  * min_bits - minimum contiguous chunk from this total allocation we
1788  * can handle. set to what we asked for originally for a full
1789  * contig. allocation, set to '1' to indicate we can deal with extents
1790  * of any size.
1791  */
1792 int __ocfs2_claim_clusters(struct ocfs2_super *osb,
1793                            handle_t *handle,
1794                            struct ocfs2_alloc_context *ac,
1795                            u32 min_clusters,
1796                            u32 max_clusters,
1797                            u32 *cluster_start,
1798                            u32 *num_clusters)
1799 {
1800         int status;
1801         unsigned int bits_wanted = max_clusters;
1802         u64 bg_blkno = 0;
1803         u16 bg_bit_off;
1804
1805         mlog_entry_void();
1806
1807         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1808
1809         BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
1810                && ac->ac_which != OCFS2_AC_USE_MAIN);
1811
1812         if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
1813                 status = ocfs2_claim_local_alloc_bits(osb,
1814                                                       handle,
1815                                                       ac,
1816                                                       bits_wanted,
1817                                                       cluster_start,
1818                                                       num_clusters);
1819                 if (!status)
1820                         atomic_inc(&osb->alloc_stats.local_data);
1821         } else {
1822                 if (min_clusters > (osb->bitmap_cpg - 1)) {
1823                         /* The only paths asking for contiguousness
1824                          * should know about this already. */
1825                         mlog(ML_ERROR, "minimum allocation requested %u exceeds "
1826                              "group bitmap size %u!\n", min_clusters,
1827                              osb->bitmap_cpg);
1828                         status = -ENOSPC;
1829                         goto bail;
1830                 }
1831                 /* clamp the current request down to a realistic size. */
1832                 if (bits_wanted > (osb->bitmap_cpg - 1))
1833                         bits_wanted = osb->bitmap_cpg - 1;
1834
1835                 status = ocfs2_claim_suballoc_bits(osb,
1836                                                    ac,
1837                                                    handle,
1838                                                    bits_wanted,
1839                                                    min_clusters,
1840                                                    &bg_bit_off,
1841                                                    num_clusters,
1842                                                    &bg_blkno);
1843                 if (!status) {
1844                         *cluster_start =
1845                                 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
1846                                                                  bg_blkno,
1847                                                                  bg_bit_off);
1848                         atomic_inc(&osb->alloc_stats.bitmap_data);
1849                 }
1850         }
1851         if (status < 0) {
1852                 if (status != -ENOSPC)
1853                         mlog_errno(status);
1854                 goto bail;
1855         }
1856
1857         ac->ac_bits_given += *num_clusters;
1858
1859 bail:
1860         mlog_exit(status);
1861         return status;
1862 }
1863
1864 int ocfs2_claim_clusters(struct ocfs2_super *osb,
1865                          handle_t *handle,
1866                          struct ocfs2_alloc_context *ac,
1867                          u32 min_clusters,
1868                          u32 *cluster_start,
1869                          u32 *num_clusters)
1870 {
1871         unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
1872
1873         return __ocfs2_claim_clusters(osb, handle, ac, min_clusters,
1874                                       bits_wanted, cluster_start, num_clusters);
1875 }
1876
1877 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
1878                                                struct inode *alloc_inode,
1879                                                struct ocfs2_group_desc *bg,
1880                                                struct buffer_head *group_bh,
1881                                                unsigned int bit_off,
1882                                                unsigned int num_bits)
1883 {
1884         int status;
1885         unsigned int tmp;
1886         int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1887         struct ocfs2_group_desc *undo_bg = NULL;
1888
1889         mlog_entry_void();
1890
1891         /* The caller got this descriptor from
1892          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1893         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1894
1895         mlog(0, "off = %u, num = %u\n", bit_off, num_bits);
1896
1897         if (ocfs2_is_cluster_bitmap(alloc_inode))
1898                 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1899
1900         status = ocfs2_journal_access_gd(handle, alloc_inode, group_bh,
1901                                          journal_type);
1902         if (status < 0) {
1903                 mlog_errno(status);
1904                 goto bail;
1905         }
1906
1907         if (ocfs2_is_cluster_bitmap(alloc_inode))
1908                 undo_bg = (struct ocfs2_group_desc *) bh2jh(group_bh)->b_committed_data;
1909
1910         tmp = num_bits;
1911         while(tmp--) {
1912                 ocfs2_clear_bit((bit_off + tmp),
1913                                 (unsigned long *) bg->bg_bitmap);
1914                 if (ocfs2_is_cluster_bitmap(alloc_inode))
1915                         ocfs2_set_bit(bit_off + tmp,
1916                                       (unsigned long *) undo_bg->bg_bitmap);
1917         }
1918         le16_add_cpu(&bg->bg_free_bits_count, num_bits);
1919
1920         status = ocfs2_journal_dirty(handle, group_bh);
1921         if (status < 0)
1922                 mlog_errno(status);
1923 bail:
1924         return status;
1925 }
1926
1927 /*
1928  * expects the suballoc inode to already be locked.
1929  */
1930 int ocfs2_free_suballoc_bits(handle_t *handle,
1931                              struct inode *alloc_inode,
1932                              struct buffer_head *alloc_bh,
1933                              unsigned int start_bit,
1934                              u64 bg_blkno,
1935                              unsigned int count)
1936 {
1937         int status = 0;
1938         u32 tmp_used;
1939         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
1940         struct ocfs2_chain_list *cl = &fe->id2.i_chain;
1941         struct buffer_head *group_bh = NULL;
1942         struct ocfs2_group_desc *group;
1943
1944         mlog_entry_void();
1945
1946         /* The alloc_bh comes from ocfs2_free_dinode() or
1947          * ocfs2_free_clusters().  The callers have all locked the
1948          * allocator and gotten alloc_bh from the lock call.  This
1949          * validates the dinode buffer.  Any corruption that has happended
1950          * is a code bug. */
1951         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1952         BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
1953
1954         mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n",
1955              (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count,
1956              (unsigned long long)bg_blkno, start_bit);
1957
1958         status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
1959                                              &group_bh);
1960         if (status < 0) {
1961                 mlog_errno(status);
1962                 goto bail;
1963         }
1964         group = (struct ocfs2_group_desc *) group_bh->b_data;
1965
1966         BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
1967
1968         status = ocfs2_block_group_clear_bits(handle, alloc_inode,
1969                                               group, group_bh,
1970                                               start_bit, count);
1971         if (status < 0) {
1972                 mlog_errno(status);
1973                 goto bail;
1974         }
1975
1976         status = ocfs2_journal_access_di(handle, alloc_inode, alloc_bh,
1977                                          OCFS2_JOURNAL_ACCESS_WRITE);
1978         if (status < 0) {
1979                 mlog_errno(status);
1980                 goto bail;
1981         }
1982
1983         le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
1984                      count);
1985         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1986         fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
1987
1988         status = ocfs2_journal_dirty(handle, alloc_bh);
1989         if (status < 0) {
1990                 mlog_errno(status);
1991                 goto bail;
1992         }
1993
1994 bail:
1995         brelse(group_bh);
1996
1997         mlog_exit(status);
1998         return status;
1999 }
2000
2001 int ocfs2_free_dinode(handle_t *handle,
2002                       struct inode *inode_alloc_inode,
2003                       struct buffer_head *inode_alloc_bh,
2004                       struct ocfs2_dinode *di)
2005 {
2006         u64 blk = le64_to_cpu(di->i_blkno);
2007         u16 bit = le16_to_cpu(di->i_suballoc_bit);
2008         u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2009
2010         return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2011                                         inode_alloc_bh, bit, bg_blkno, 1);
2012 }
2013
2014 int ocfs2_free_clusters(handle_t *handle,
2015                        struct inode *bitmap_inode,
2016                        struct buffer_head *bitmap_bh,
2017                        u64 start_blk,
2018                        unsigned int num_clusters)
2019 {
2020         int status;
2021         u16 bg_start_bit;
2022         u64 bg_blkno;
2023         struct ocfs2_dinode *fe;
2024
2025         /* You can't ever have a contiguous set of clusters
2026          * bigger than a block group bitmap so we never have to worry
2027          * about looping on them. */
2028
2029         mlog_entry_void();
2030
2031         /* This is expensive. We can safely remove once this stuff has
2032          * gotten tested really well. */
2033         BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2034
2035         fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2036
2037         ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2038                                      &bg_start_bit);
2039
2040         mlog(0, "want to free %u clusters starting at block %llu\n",
2041              num_clusters, (unsigned long long)start_blk);
2042         mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n",
2043              (unsigned long long)bg_blkno, bg_start_bit);
2044
2045         status = ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2046                                           bg_start_bit, bg_blkno,
2047                                           num_clusters);
2048         if (status < 0) {
2049                 mlog_errno(status);
2050                 goto out;
2051         }
2052
2053         ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2054                                          num_clusters);
2055
2056 out:
2057         mlog_exit(status);
2058         return status;
2059 }
2060
2061 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2062 {
2063         printk("Block Group:\n");
2064         printk("bg_signature:       %s\n", bg->bg_signature);
2065         printk("bg_size:            %u\n", bg->bg_size);
2066         printk("bg_bits:            %u\n", bg->bg_bits);
2067         printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2068         printk("bg_chain:           %u\n", bg->bg_chain);
2069         printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
2070         printk("bg_next_group:      %llu\n",
2071                (unsigned long long)bg->bg_next_group);
2072         printk("bg_parent_dinode:   %llu\n",
2073                (unsigned long long)bg->bg_parent_dinode);
2074         printk("bg_blkno:           %llu\n",
2075                (unsigned long long)bg->bg_blkno);
2076 }
2077
2078 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2079 {
2080         int i;
2081
2082         printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2083         printk("i_signature:                  %s\n", fe->i_signature);
2084         printk("i_size:                       %llu\n",
2085                (unsigned long long)fe->i_size);
2086         printk("i_clusters:                   %u\n", fe->i_clusters);
2087         printk("i_generation:                 %u\n",
2088                le32_to_cpu(fe->i_generation));
2089         printk("id1.bitmap1.i_used:           %u\n",
2090                le32_to_cpu(fe->id1.bitmap1.i_used));
2091         printk("id1.bitmap1.i_total:          %u\n",
2092                le32_to_cpu(fe->id1.bitmap1.i_total));
2093         printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
2094         printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
2095         printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
2096         printk("id2.i_chain.cl_next_free_rec: %u\n",
2097                fe->id2.i_chain.cl_next_free_rec);
2098         for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2099                 printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
2100                        fe->id2.i_chain.cl_recs[i].c_free);
2101                 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2102                        fe->id2.i_chain.cl_recs[i].c_total);
2103                 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2104                        (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2105         }
2106 }
2107
2108 /*
2109  * For a given allocation, determine which allocators will need to be
2110  * accessed, and lock them, reserving the appropriate number of bits.
2111  *
2112  * Sparse file systems call this from ocfs2_write_begin_nolock()
2113  * and ocfs2_allocate_unwritten_extents().
2114  *
2115  * File systems which don't support holes call this from
2116  * ocfs2_extend_allocation().
2117  */
2118 int ocfs2_lock_allocators(struct inode *inode,
2119                           struct ocfs2_extent_tree *et,
2120                           u32 clusters_to_add, u32 extents_to_split,
2121                           struct ocfs2_alloc_context **data_ac,
2122                           struct ocfs2_alloc_context **meta_ac)
2123 {
2124         int ret = 0, num_free_extents;
2125         unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2126         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2127
2128         *meta_ac = NULL;
2129         if (data_ac)
2130                 *data_ac = NULL;
2131
2132         BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2133
2134         num_free_extents = ocfs2_num_free_extents(osb, inode, et);
2135         if (num_free_extents < 0) {
2136                 ret = num_free_extents;
2137                 mlog_errno(ret);
2138                 goto out;
2139         }
2140
2141         /*
2142          * Sparse allocation file systems need to be more conservative
2143          * with reserving room for expansion - the actual allocation
2144          * happens while we've got a journal handle open so re-taking
2145          * a cluster lock (because we ran out of room for another
2146          * extent) will violate ordering rules.
2147          *
2148          * Most of the time we'll only be seeing this 1 cluster at a time
2149          * anyway.
2150          *
2151          * Always lock for any unwritten extents - we might want to
2152          * add blocks during a split.
2153          */
2154         if (!num_free_extents ||
2155             (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2156                 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2157                 if (ret < 0) {
2158                         if (ret != -ENOSPC)
2159                                 mlog_errno(ret);
2160                         goto out;
2161                 }
2162         }
2163
2164         if (clusters_to_add == 0)
2165                 goto out;
2166
2167         ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2168         if (ret < 0) {
2169                 if (ret != -ENOSPC)
2170                         mlog_errno(ret);
2171                 goto out;
2172         }
2173
2174 out:
2175         if (ret) {
2176                 if (*meta_ac) {
2177                         ocfs2_free_alloc_context(*meta_ac);
2178                         *meta_ac = NULL;
2179                 }
2180
2181                 /*
2182                  * We cannot have an error and a non null *data_ac.
2183                  */
2184         }
2185
2186         return ret;
2187 }
2188
2189 /*
2190  * Read the inode specified by blkno to get suballoc_slot and
2191  * suballoc_bit.
2192  */
2193 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2194                                        u16 *suballoc_slot, u16 *suballoc_bit)
2195 {
2196         int status;
2197         struct buffer_head *inode_bh = NULL;
2198         struct ocfs2_dinode *inode_fe;
2199
2200         mlog_entry("blkno: %llu\n", (unsigned long long)blkno);
2201
2202         /* dirty read disk */
2203         status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2204         if (status < 0) {
2205                 mlog(ML_ERROR, "read block %llu failed %d\n",
2206                      (unsigned long long)blkno, status);
2207                 goto bail;
2208         }
2209
2210         inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2211         if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2212                 mlog(ML_ERROR, "invalid inode %llu requested\n",
2213                      (unsigned long long)blkno);
2214                 status = -EINVAL;
2215                 goto bail;
2216         }
2217
2218         if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2219             (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2220                 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2221                      (unsigned long long)blkno,
2222                      (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2223                 status = -EINVAL;
2224                 goto bail;
2225         }
2226
2227         if (suballoc_slot)
2228                 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2229         if (suballoc_bit)
2230                 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2231
2232 bail:
2233         brelse(inode_bh);
2234
2235         mlog_exit(status);
2236         return status;
2237 }
2238
2239 /*
2240  * test whether bit is SET in allocator bitmap or not.  on success, 0
2241  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2242  * is returned and *res is meaningless.  Call this after you have
2243  * cluster locked against suballoc, or you may get a result based on
2244  * non-up2date contents
2245  */
2246 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2247                                    struct inode *suballoc,
2248                                    struct buffer_head *alloc_bh, u64 blkno,
2249                                    u16 bit, int *res)
2250 {
2251         struct ocfs2_dinode *alloc_fe;
2252         struct ocfs2_group_desc *group;
2253         struct buffer_head *group_bh = NULL;
2254         u64 bg_blkno;
2255         int status;
2256
2257         mlog_entry("blkno: %llu bit: %u\n", (unsigned long long)blkno,
2258                    (unsigned int)bit);
2259
2260         alloc_fe = (struct ocfs2_dinode *)alloc_bh->b_data;
2261         if ((bit + 1) > ocfs2_bits_per_group(&alloc_fe->id2.i_chain)) {
2262                 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2263                      (unsigned int)bit,
2264                      ocfs2_bits_per_group(&alloc_fe->id2.i_chain));
2265                 status = -EINVAL;
2266                 goto bail;
2267         }
2268
2269         bg_blkno = ocfs2_which_suballoc_group(blkno, bit);
2270         status = ocfs2_read_group_descriptor(suballoc, alloc_fe, bg_blkno,
2271                                              &group_bh);
2272         if (status < 0) {
2273                 mlog(ML_ERROR, "read group %llu failed %d\n",
2274                      (unsigned long long)bg_blkno, status);
2275                 goto bail;
2276         }
2277
2278         group = (struct ocfs2_group_desc *) group_bh->b_data;
2279         *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2280
2281 bail:
2282         brelse(group_bh);
2283
2284         mlog_exit(status);
2285         return status;
2286 }
2287
2288 /*
2289  * Test if the bit representing this inode (blkno) is set in the
2290  * suballocator.
2291  *
2292  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2293  *
2294  * In the event of failure, a negative value is returned and *res is
2295  * meaningless.
2296  *
2297  * Callers must make sure to hold nfs_sync_lock to prevent
2298  * ocfs2_delete_inode() on another node from accessing the same
2299  * suballocator concurrently.
2300  */
2301 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2302 {
2303         int status;
2304         u16 suballoc_bit = 0, suballoc_slot = 0;
2305         struct inode *inode_alloc_inode;
2306         struct buffer_head *alloc_bh = NULL;
2307
2308         mlog_entry("blkno: %llu", (unsigned long long)blkno);
2309
2310         status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2311                                              &suballoc_bit);
2312         if (status < 0) {
2313                 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2314                 goto bail;
2315         }
2316
2317         inode_alloc_inode =
2318                 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2319                                             suballoc_slot);
2320         if (!inode_alloc_inode) {
2321                 /* the error code could be inaccurate, but we are not able to
2322                  * get the correct one. */
2323                 status = -EINVAL;
2324                 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2325                      (u32)suballoc_slot);
2326                 goto bail;
2327         }
2328
2329         mutex_lock(&inode_alloc_inode->i_mutex);
2330         status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2331         if (status < 0) {
2332                 mutex_unlock(&inode_alloc_inode->i_mutex);
2333                 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2334                      (u32)suballoc_slot, status);
2335                 goto bail;
2336         }
2337
2338         status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2339                                          blkno, suballoc_bit, res);
2340         if (status < 0)
2341                 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2342
2343         ocfs2_inode_unlock(inode_alloc_inode, 0);
2344         mutex_unlock(&inode_alloc_inode->i_mutex);
2345
2346         iput(inode_alloc_inode);
2347         brelse(alloc_bh);
2348 bail:
2349         mlog_exit(status);
2350         return status;
2351 }