fs/xfs/xfs_dir2_leaf.c

   1 /*
   2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_btree.h"
  28 #include "xfs_bmap_btree.h"
  29 #include "xfs_dinode.h"
  30 #include "xfs_inode.h"
  31 #include "xfs_bmap.h"
  32 #include "xfs_dir2_format.h"
  33 #include "xfs_dir2_priv.h"
  34 #include "xfs_error.h"
  35 #include "xfs_trace.h"
  36
  37 /*
  38  * Local function declarations.
  39  */
  40 #ifdef DEBUG
  41 static void xfs_dir2_leaf_check(struct xfs_inode *dp, struct xfs_buf *bp);
  42 #else
  43 #define xfs_dir2_leaf_check(dp, bp)
  44 #endif
  45 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
  46                                     int *indexp, struct xfs_buf **dbpp);
  47 static void xfs_dir2_leaf_log_bests(struct xfs_trans *tp, struct xfs_buf *bp,
  48                                     int first, int last);
  49 static void xfs_dir2_leaf_log_tail(struct xfs_trans *tp, struct xfs_buf *bp);
  50
  51
  52 /*
  53  * Convert a block form directory to a leaf form directory.
  54  */
  55 int                                             /* error */
  56 xfs_dir2_block_to_leaf(
  57         xfs_da_args_t           *args,          /* operation arguments */
  58         struct xfs_buf          *dbp)           /* input block's buffer */
  59 {
  60         __be16                  *bestsp;        /* leaf's bestsp entries */
  61         xfs_dablk_t             blkno;          /* leaf block's bno */
  62         xfs_dir2_data_hdr_t     *hdr;           /* block header */
  63         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
  64         xfs_dir2_block_tail_t   *btp;           /* block's tail */
  65         xfs_inode_t             *dp;            /* incore directory inode */
  66         int                     error;          /* error return code */
  67         struct xfs_buf          *lbp;           /* leaf block's buffer */
  68         xfs_dir2_db_t           ldb;            /* leaf block's bno */
  69         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
  70         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
  71         xfs_mount_t             *mp;            /* filesystem mount point */
  72         int                     needlog;        /* need to log block header */
  73         int                     needscan;       /* need to rescan bestfree */
  74         xfs_trans_t             *tp;            /* transaction pointer */
  75
  76         trace_xfs_dir2_block_to_leaf(args);
  77
  78         dp = args->dp;
  79         mp = dp->i_mount;
  80         tp = args->trans;
  81         /*
  82          * Add the leaf block to the inode.
  83          * This interface will only put blocks in the leaf/node range.
  84          * Since that's empty now, we'll get the root (block 0 in range).
  85          */
  86         if ((error = xfs_da_grow_inode(args, &blkno))) {
  87                 return error;
  88         }
  89         ldb = xfs_dir2_da_to_db(mp, blkno);
  90         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
  91         /*
  92          * Initialize the leaf block, get a buffer for it.
  93          */
  94         if ((error = xfs_dir2_leaf_init(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC))) {
  95                 return error;
  96         }
  97         ASSERT(lbp != NULL);
  98         leaf = lbp->b_addr;
  99         hdr = dbp->b_addr;
 100         xfs_dir2_data_check(dp, dbp);
 101         btp = xfs_dir2_block_tail_p(mp, hdr);
 102         blp = xfs_dir2_block_leaf_p(btp);
 103         /*
 104          * Set the counts in the leaf header.
 105          */
 106         leaf->hdr.count = cpu_to_be16(be32_to_cpu(btp->count));
 107         leaf->hdr.stale = cpu_to_be16(be32_to_cpu(btp->stale));
 108         /*
 109          * Could compact these but I think we always do the conversion
 110          * after squeezing out stale entries.
 111          */
 112         memcpy(leaf->ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
 113         xfs_dir2_leaf_log_ents(tp, lbp, 0, be16_to_cpu(leaf->hdr.count) - 1);
 114         needscan = 0;
 115         needlog = 1;
 116         /*
 117          * Make the space formerly occupied by the leaf entries and block
 118          * tail be free.
 119          */
 120         xfs_dir2_data_make_free(tp, dbp,
 121                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 122                 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
 123                                        (char *)blp),
 124                 &needlog, &needscan);
 125         /*
 126          * Fix up the block header, make it a data block.
 127          */
 128         hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 129         if (needscan)
 130                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 131         /*
 132          * Set up leaf tail and bests table.
 133          */
 134         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 135         ltp->bestcount = cpu_to_be32(1);
 136         bestsp = xfs_dir2_leaf_bests_p(ltp);
 137         bestsp[0] =  hdr->bestfree[0].length;
 138         /*
 139          * Log the data header and leaf bests table.
 140          */
 141         if (needlog)
 142                 xfs_dir2_data_log_header(tp, dbp);
 143         xfs_dir2_leaf_check(dp, lbp);
 144         xfs_dir2_data_check(dp, dbp);
 145         xfs_dir2_leaf_log_bests(tp, lbp, 0, 0);
 146         return 0;
 147 }
 148
 149 STATIC void
 150 xfs_dir2_leaf_find_stale(
 151         struct xfs_dir2_leaf    *leaf,
 152         int                     index,
 153         int                     *lowstale,
 154         int                     *highstale)
 155 {
 156         /*
 157          * Find the first stale entry before our index, if any.
 158          */
 159         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
 160                 if (leaf->ents[*lowstale].address ==
 161                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 162                         break;
 163         }
 164
 165         /*
 166          * Find the first stale entry at or after our index, if any.
 167          * Stop if the result would require moving more entries than using
 168          * lowstale.
 169          */
 170         for (*highstale = index;
 171              *highstale < be16_to_cpu(leaf->hdr.count);
 172              ++*highstale) {
 173                 if (leaf->ents[*highstale].address ==
 174                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 175                         break;
 176                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
 177                         break;
 178         }
 179 }
 180
 181 struct xfs_dir2_leaf_entry *
 182 xfs_dir2_leaf_find_entry(
 183         xfs_dir2_leaf_t         *leaf,          /* leaf structure */
 184         int                     index,          /* leaf table position */
 185         int                     compact,        /* need to compact leaves */
 186         int                     lowstale,       /* index of prev stale leaf */
 187         int                     highstale,      /* index of next stale leaf */
 188         int                     *lfloglow,      /* low leaf logging index */
 189         int                     *lfloghigh)     /* high leaf logging index */
 190 {
 191         if (!leaf->hdr.stale) {
 192                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
 193
 194                 /*
 195                  * Now we need to make room to insert the leaf entry.
 196                  *
 197                  * If there are no stale entries, just insert a hole at index.
 198                  */
 199                 lep = &leaf->ents[index];
 200                 if (index < be16_to_cpu(leaf->hdr.count))
 201                         memmove(lep + 1, lep,
 202                                 (be16_to_cpu(leaf->hdr.count) - index) *
 203                                  sizeof(*lep));
 204
 205                 /*
 206                  * Record low and high logging indices for the leaf.
 207                  */
 208                 *lfloglow = index;
 209                 *lfloghigh = be16_to_cpu(leaf->hdr.count);
 210                 be16_add_cpu(&leaf->hdr.count, 1);
 211                 return lep;
 212         }
 213
 214         /*
 215          * There are stale entries.
 216          *
 217          * We will use one of them for the new entry.  It's probably not at
 218          * the right location, so we'll have to shift some up or down first.
 219          *
 220          * If we didn't compact before, we need to find the nearest stale
 221          * entries before and after our insertion point.
 222          */
 223         if (compact == 0)
 224                 xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
 225
 226         /*
 227          * If the low one is better, use it.
 228          */
 229         if (lowstale >= 0 &&
 230             (highstale == be16_to_cpu(leaf->hdr.count) ||
 231              index - lowstale - 1 < highstale - index)) {
 232                 ASSERT(index - lowstale - 1 >= 0);
 233                 ASSERT(leaf->ents[lowstale].address ==
 234                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 235
 236                 /*
 237                  * Copy entries up to cover the stale entry and make room
 238                  * for the new entry.
 239                  */
 240                 if (index - lowstale - 1 > 0) {
 241                         memmove(&leaf->ents[lowstale],
 242                                 &leaf->ents[lowstale + 1],
 243                                 (index - lowstale - 1) *
 244                                 sizeof(xfs_dir2_leaf_entry_t));
 245                 }
 246                 *lfloglow = MIN(lowstale, *lfloglow);
 247                 *lfloghigh = MAX(index - 1, *lfloghigh);
 248                 be16_add_cpu(&leaf->hdr.stale, -1);
 249                 return &leaf->ents[index - 1];
 250         }
 251
 252         /*
 253          * The high one is better, so use that one.
 254          */
 255         ASSERT(highstale - index >= 0);
 256         ASSERT(leaf->ents[highstale].address ==
 257                cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 258
 259         /*
 260          * Copy entries down to cover the stale entry and make room for the
 261          * new entry.
 262          */
 263         if (highstale - index > 0) {
 264                 memmove(&leaf->ents[index + 1],
 265                         &leaf->ents[index],
 266                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
 267         }
 268         *lfloglow = MIN(index, *lfloglow);
 269         *lfloghigh = MAX(highstale, *lfloghigh);
 270         be16_add_cpu(&leaf->hdr.stale, -1);
 271         return &leaf->ents[index];
 272 }
 273
 274 /*
 275  * Add an entry to a leaf form directory.
 276  */
 277 int                                             /* error */
 278 xfs_dir2_leaf_addname(
 279         xfs_da_args_t           *args)          /* operation arguments */
 280 {
 281         __be16                  *bestsp;        /* freespace table in leaf */
 282         int                     compact;        /* need to compact leaves */
 283         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 284         struct xfs_buf          *dbp;           /* data block buffer */
 285         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 286         xfs_inode_t             *dp;            /* incore directory inode */
 287         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
 288         int                     error;          /* error return value */
 289         int                     grown;          /* allocated new data block */
 290         int                     highstale;      /* index of next stale leaf */
 291         int                     i;              /* temporary, index */
 292         int                     index;          /* leaf table position */
 293         struct xfs_buf          *lbp;           /* leaf's buffer */
 294         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 295         int                     length;         /* length of new entry */
 296         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
 297         int                     lfloglow;       /* low leaf logging index */
 298         int                     lfloghigh;      /* high leaf logging index */
 299         int                     lowstale;       /* index of prev stale leaf */
 300         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 301         xfs_mount_t             *mp;            /* filesystem mount point */
 302         int                     needbytes;      /* leaf block bytes needed */
 303         int                     needlog;        /* need to log data header */
 304         int                     needscan;       /* need to rescan data free */
 305         __be16                  *tagp;          /* end of data entry */
 306         xfs_trans_t             *tp;            /* transaction pointer */
 307         xfs_dir2_db_t           use_block;      /* data block number */
 308
 309         trace_xfs_dir2_leaf_addname(args);
 310
 311         dp = args->dp;
 312         tp = args->trans;
 313         mp = dp->i_mount;
 314         /*
 315          * Read the leaf block.
 316          */
 317         error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
 318                 XFS_DATA_FORK);
 319         if (error) {
 320                 return error;
 321         }
 322         ASSERT(lbp != NULL);
 323         /*
 324          * Look up the entry by hash value and name.
 325          * We know it's not there, our caller has already done a lookup.
 326          * So the index is of the entry to insert in front of.
 327          * But if there are dup hash values the index is of the first of those.
 328          */
 329         index = xfs_dir2_leaf_search_hash(args, lbp);
 330         leaf = lbp->b_addr;
 331         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 332         bestsp = xfs_dir2_leaf_bests_p(ltp);
 333         length = xfs_dir2_data_entsize(args->namelen);
 334         /*
 335          * See if there are any entries with the same hash value
 336          * and space in their block for the new entry.
 337          * This is good because it puts multiple same-hash value entries
 338          * in a data block, improving the lookup of those entries.
 339          */
 340         for (use_block = -1, lep = &leaf->ents[index];
 341              index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
 342              index++, lep++) {
 343                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 344                         continue;
 345                 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 346                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 347                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
 348                 if (be16_to_cpu(bestsp[i]) >= length) {
 349                         use_block = i;
 350                         break;
 351                 }
 352         }
 353         /*
 354          * Didn't find a block yet, linear search all the data blocks.
 355          */
 356         if (use_block == -1) {
 357                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 358                         /*
 359                          * Remember a block we see that's missing.
 360                          */
 361                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
 362                             use_block == -1)
 363                                 use_block = i;
 364                         else if (be16_to_cpu(bestsp[i]) >= length) {
 365                                 use_block = i;
 366                                 break;
 367                         }
 368                 }
 369         }
 370         /*
 371          * How many bytes do we need in the leaf block?
 372          */
 373         needbytes = 0;
 374         if (!leaf->hdr.stale)
 375                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
 376         if (use_block == -1)
 377                 needbytes += sizeof(xfs_dir2_data_off_t);
 378
 379         /*
 380          * Now kill use_block if it refers to a missing block, so we
 381          * can use it as an indication of allocation needed.
 382          */
 383         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
 384                 use_block = -1;
 385         /*
 386          * If we don't have enough free bytes but we can make enough
 387          * by compacting out stale entries, we'll do that.
 388          */
 389         if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <
 390                                 needbytes && be16_to_cpu(leaf->hdr.stale) > 1) {
 391                 compact = 1;
 392         }
 393         /*
 394          * Otherwise if we don't have enough free bytes we need to
 395          * convert to node form.
 396          */
 397         else if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(
 398                                                 leaf->hdr.count)] < needbytes) {
 399                 /*
 400                  * Just checking or no space reservation, give up.
 401                  */
 402                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 403                                                         args->total == 0) {
 404                         xfs_trans_brelse(tp, lbp);
 405                         return XFS_ERROR(ENOSPC);
 406                 }
 407                 /*
 408                  * Convert to node form.
 409                  */
 410                 error = xfs_dir2_leaf_to_node(args, lbp);
 411                 if (error)
 412                         return error;
 413                 /*
 414                  * Then add the new entry.
 415                  */
 416                 return xfs_dir2_node_addname(args);
 417         }
 418         /*
 419          * Otherwise it will fit without compaction.
 420          */
 421         else
 422                 compact = 0;
 423         /*
 424          * If just checking, then it will fit unless we needed to allocate
 425          * a new data block.
 426          */
 427         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 428                 xfs_trans_brelse(tp, lbp);
 429                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
 430         }
 431         /*
 432          * If no allocations are allowed, return now before we've
 433          * changed anything.
 434          */
 435         if (args->total == 0 && use_block == -1) {
 436                 xfs_trans_brelse(tp, lbp);
 437                 return XFS_ERROR(ENOSPC);
 438         }
 439         /*
 440          * Need to compact the leaf entries, removing stale ones.
 441          * Leave one stale entry behind - the one closest to our
 442          * insertion index - and we'll shift that one to our insertion
 443          * point later.
 444          */
 445         if (compact) {
 446                 xfs_dir2_leaf_compact_x1(lbp, &index, &lowstale, &highstale,
 447                         &lfloglow, &lfloghigh);
 448         }
 449         /*
 450          * There are stale entries, so we'll need log-low and log-high
 451          * impossibly bad values later.
 452          */
 453         else if (be16_to_cpu(leaf->hdr.stale)) {
 454                 lfloglow = be16_to_cpu(leaf->hdr.count);
 455                 lfloghigh = -1;
 456         }
 457         /*
 458          * If there was no data block space found, we need to allocate
 459          * a new one.
 460          */
 461         if (use_block == -1) {
 462                 /*
 463                  * Add the new data block.
 464                  */
 465                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 466                                 &use_block))) {
 467                         xfs_trans_brelse(tp, lbp);
 468                         return error;
 469                 }
 470                 /*
 471                  * Initialize the block.
 472                  */
 473                 if ((error = xfs_dir2_data_init(args, use_block, &dbp))) {
 474                         xfs_trans_brelse(tp, lbp);
 475                         return error;
 476                 }
 477                 /*
 478                  * If we're adding a new data block on the end we need to
 479                  * extend the bests table.  Copy it up one entry.
 480                  */
 481                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
 482                         bestsp--;
 483                         memmove(&bestsp[0], &bestsp[1],
 484                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 485                         be32_add_cpu(&ltp->bestcount, 1);
 486                         xfs_dir2_leaf_log_tail(tp, lbp);
 487                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
 488                 }
 489                 /*
 490                  * If we're filling in a previously empty block just log it.
 491                  */
 492                 else
 493                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 494                 hdr = dbp->b_addr;
 495                 bestsp[use_block] = hdr->bestfree[0].length;
 496                 grown = 1;
 497         }
 498         /*
 499          * Already had space in some data block.
 500          * Just read that one in.
 501          */
 502         else {
 503                 if ((error =
 504                     xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, use_block),
 505                             -1, &dbp, XFS_DATA_FORK))) {
 506                         xfs_trans_brelse(tp, lbp);
 507                         return error;
 508                 }
 509                 hdr = dbp->b_addr;
 510                 grown = 0;
 511         }
 512         xfs_dir2_data_check(dp, dbp);
 513         /*
 514          * Point to the biggest freespace in our data block.
 515          */
 516         dup = (xfs_dir2_data_unused_t *)
 517               ((char *)hdr + be16_to_cpu(hdr->bestfree[0].offset));
 518         ASSERT(be16_to_cpu(dup->length) >= length);
 519         needscan = needlog = 0;
 520         /*
 521          * Mark the initial part of our freespace in use for the new entry.
 522          */
 523         xfs_dir2_data_use_free(tp, dbp, dup,
 524                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
 525                 &needlog, &needscan);
 526         /*
 527          * Initialize our new entry (at last).
 528          */
 529         dep = (xfs_dir2_data_entry_t *)dup;
 530         dep->inumber = cpu_to_be64(args->inumber);
 531         dep->namelen = args->namelen;
 532         memcpy(dep->name, args->name, dep->namelen);
 533         tagp = xfs_dir2_data_entry_tag_p(dep);
 534         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 535         /*
 536          * Need to scan fix up the bestfree table.
 537          */
 538         if (needscan)
 539                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 540         /*
 541          * Need to log the data block's header.
 542          */
 543         if (needlog)
 544                 xfs_dir2_data_log_header(tp, dbp);
 545         xfs_dir2_data_log_entry(tp, dbp, dep);
 546         /*
 547          * If the bests table needs to be changed, do it.
 548          * Log the change unless we've already done that.
 549          */
 550         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(hdr->bestfree[0].length)) {
 551                 bestsp[use_block] = hdr->bestfree[0].length;
 552                 if (!grown)
 553                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 554         }
 555
 556         lep = xfs_dir2_leaf_find_entry(leaf, index, compact, lowstale,
 557                                        highstale, &lfloglow, &lfloghigh);
 558
 559         /*
 560          * Fill in the new leaf entry.
 561          */
 562         lep->hashval = cpu_to_be32(args->hashval);
 563         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
 564                                 be16_to_cpu(*tagp)));
 565         /*
 566          * Log the leaf fields and give up the buffers.
 567          */
 568         xfs_dir2_leaf_log_header(tp, lbp);
 569         xfs_dir2_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
 570         xfs_dir2_leaf_check(dp, lbp);
 571         xfs_dir2_data_check(dp, dbp);
 572         return 0;
 573 }
 574
 575 #ifdef DEBUG
 576 /*
 577  * Check the internal consistency of a leaf1 block.
 578  * Pop an assert if something is wrong.
 579  */
 580 STATIC void
 581 xfs_dir2_leaf_check(
 582         struct xfs_inode        *dp,            /* incore directory inode */
 583         struct xfs_buf          *bp)            /* leaf's buffer */
 584 {
 585         int                     i;              /* leaf index */
 586         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 587         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 588         xfs_mount_t             *mp;            /* filesystem mount point */
 589         int                     stale;          /* count of stale leaves */
 590
 591         leaf = bp->b_addr;
 592         mp = dp->i_mount;
 593         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
 594         /*
 595          * This value is not restrictive enough.
 596          * Should factor in the size of the bests table as well.
 597          * We can deduce a value for that from di_size.
 598          */
 599         ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
 600         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 601         /*
 602          * Leaves and bests don't overlap.
 603          */
 604         ASSERT((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <=
 605                (char *)xfs_dir2_leaf_bests_p(ltp));
 606         /*
 607          * Check hash value order, count stale entries.
 608          */
 609         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
 610                 if (i + 1 < be16_to_cpu(leaf->hdr.count))
 611                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
 612                                be32_to_cpu(leaf->ents[i + 1].hashval));
 613                 if (leaf->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 614                         stale++;
 615         }
 616         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
 617 }
 618 #endif  /* DEBUG */
 619
 620 /*
 621  * Compact out any stale entries in the leaf.
 622  * Log the header and changed leaf entries, if any.
 623  */
 624 void
 625 xfs_dir2_leaf_compact(
 626         xfs_da_args_t   *args,          /* operation arguments */
 627         struct xfs_buf  *bp)            /* leaf buffer */
 628 {
 629         int             from;           /* source leaf index */
 630         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 631         int             loglow;         /* first leaf entry to log */
 632         int             to;             /* target leaf index */
 633
 634         leaf = bp->b_addr;
 635         if (!leaf->hdr.stale) {
 636                 return;
 637         }
 638         /*
 639          * Compress out the stale entries in place.
 640          */
 641         for (from = to = 0, loglow = -1; from < be16_to_cpu(leaf->hdr.count); from++) {
 642                 if (leaf->ents[from].address ==
 643                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 644                         continue;
 645                 /*
 646                  * Only actually copy the entries that are different.
 647                  */
 648                 if (from > to) {
 649                         if (loglow == -1)
 650                                 loglow = to;
 651                         leaf->ents[to] = leaf->ents[from];
 652                 }
 653                 to++;
 654         }
 655         /*
 656          * Update and log the header, log the leaf entries.
 657          */
 658         ASSERT(be16_to_cpu(leaf->hdr.stale) == from - to);
 659         be16_add_cpu(&leaf->hdr.count, -(be16_to_cpu(leaf->hdr.stale)));
 660         leaf->hdr.stale = 0;
 661         xfs_dir2_leaf_log_header(args->trans, bp);
 662         if (loglow != -1)
 663                 xfs_dir2_leaf_log_ents(args->trans, bp, loglow, to - 1);
 664 }
 665
 666 /*
 667  * Compact the leaf entries, removing stale ones.
 668  * Leave one stale entry behind - the one closest to our
 669  * insertion index - and the caller will shift that one to our insertion
 670  * point later.
 671  * Return new insertion index, where the remaining stale entry is,
 672  * and leaf logging indices.
 673  */
 674 void
 675 xfs_dir2_leaf_compact_x1(
 676         struct xfs_buf  *bp,            /* leaf buffer */
 677         int             *indexp,        /* insertion index */
 678         int             *lowstalep,     /* out: stale entry before us */
 679         int             *highstalep,    /* out: stale entry after us */
 680         int             *lowlogp,       /* out: low log index */
 681         int             *highlogp)      /* out: high log index */
 682 {
 683         int             from;           /* source copy index */
 684         int             highstale;      /* stale entry at/after index */
 685         int             index;          /* insertion index */
 686         int             keepstale;      /* source index of kept stale */
 687         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 688         int             lowstale;       /* stale entry before index */
 689         int             newindex=0;     /* new insertion index */
 690         int             to;             /* destination copy index */
 691
 692         leaf = bp->b_addr;
 693         ASSERT(be16_to_cpu(leaf->hdr.stale) > 1);
 694         index = *indexp;
 695
 696         xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
 697
 698         /*
 699          * Pick the better of lowstale and highstale.
 700          */
 701         if (lowstale >= 0 &&
 702             (highstale == be16_to_cpu(leaf->hdr.count) ||
 703              index - lowstale <= highstale - index))
 704                 keepstale = lowstale;
 705         else
 706                 keepstale = highstale;
 707         /*
 708          * Copy the entries in place, removing all the stale entries
 709          * except keepstale.
 710          */
 711         for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
 712                 /*
 713                  * Notice the new value of index.
 714                  */
 715                 if (index == from)
 716                         newindex = to;
 717                 if (from != keepstale &&
 718                     leaf->ents[from].address ==
 719                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
 720                         if (from == to)
 721                                 *lowlogp = to;
 722                         continue;
 723                 }
 724                 /*
 725                  * Record the new keepstale value for the insertion.
 726                  */
 727                 if (from == keepstale)
 728                         lowstale = highstale = to;
 729                 /*
 730                  * Copy only the entries that have moved.
 731                  */
 732                 if (from > to)
 733                         leaf->ents[to] = leaf->ents[from];
 734                 to++;
 735         }
 736         ASSERT(from > to);
 737         /*
 738          * If the insertion point was past the last entry,
 739          * set the new insertion point accordingly.
 740          */
 741         if (index == from)
 742                 newindex = to;
 743         *indexp = newindex;
 744         /*
 745          * Adjust the leaf header values.
 746          */
 747         be16_add_cpu(&leaf->hdr.count, -(from - to));
 748         leaf->hdr.stale = cpu_to_be16(1);
 749         /*
 750          * Remember the low/high stale value only in the "right"
 751          * direction.
 752          */
 753         if (lowstale >= newindex)
 754                 lowstale = -1;
 755         else
 756                 highstale = be16_to_cpu(leaf->hdr.count);
 757         *highlogp = be16_to_cpu(leaf->hdr.count) - 1;
 758         *lowstalep = lowstale;
 759         *highstalep = highstale;
 760 }
 761
 762 /*
 763  * Getdents (readdir) for leaf and node directories.
 764  * This reads the data blocks only, so is the same for both forms.
 765  */
 766 int                                             /* error */
 767 xfs_dir2_leaf_getdents(
 768         xfs_inode_t             *dp,            /* incore directory inode */
 769         void                    *dirent,
 770         size_t                  bufsize,
 771         xfs_off_t               *offset,
 772         filldir_t               filldir)
 773 {
 774         struct xfs_buf          *bp;            /* data block buffer */
 775         int                     byteoff;        /* offset in current block */
 776         xfs_dir2_db_t           curdb;          /* db for current block */
 777         xfs_dir2_off_t          curoff;         /* current overall offset */
 778         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 779         xfs_dir2_data_entry_t   *dep;           /* data entry */
 780         xfs_dir2_data_unused_t  *dup;           /* unused entry */
 781         int                     error = 0;      /* error return value */
 782         int                     i;              /* temporary loop index */
 783         int                     j;              /* temporary loop index */
 784         int                     length;         /* temporary length value */
 785         xfs_bmbt_irec_t         *map;           /* map vector for blocks */
 786         xfs_extlen_t            map_blocks;     /* number of fsbs in map */
 787         xfs_dablk_t             map_off;        /* last mapped file offset */
 788         int                     map_size;       /* total entries in *map */
 789         int                     map_valid;      /* valid entries in *map */
 790         xfs_mount_t             *mp;            /* filesystem mount point */
 791         xfs_dir2_off_t          newoff;         /* new curoff after new blk */
 792         int                     nmap;           /* mappings to ask xfs_bmapi */
 793         char                    *ptr = NULL;    /* pointer to current data */
 794         int                     ra_current;     /* number of read-ahead blks */
 795         int                     ra_index;       /* *map index for read-ahead */
 796         int                     ra_offset;      /* map entry offset for ra */
 797         int                     ra_want;        /* readahead count wanted */
 798
 799         /*
 800          * If the offset is at or past the largest allowed value,
 801          * give up right away.
 802          */
 803         if (*offset >= XFS_DIR2_MAX_DATAPTR)
 804                 return 0;
 805
 806         mp = dp->i_mount;
 807
 808         /*
 809          * Set up to bmap a number of blocks based on the caller's
 810          * buffer size, the directory block size, and the filesystem
 811          * block size.
 812          */
 813         map_size = howmany(bufsize + mp->m_dirblksize, mp->m_sb.sb_blocksize);
 814         map = kmem_alloc(map_size * sizeof(*map), KM_SLEEP);
 815         map_valid = ra_index = ra_offset = ra_current = map_blocks = 0;
 816         bp = NULL;
 817
 818         /*
 819          * Inside the loop we keep the main offset value as a byte offset
 820          * in the directory file.
 821          */
 822         curoff = xfs_dir2_dataptr_to_byte(mp, *offset);
 823
 824         /*
 825          * Force this conversion through db so we truncate the offset
 826          * down to get the start of the data block.
 827          */
 828         map_off = xfs_dir2_db_to_da(mp, xfs_dir2_byte_to_db(mp, curoff));
 829         /*
 830          * Loop over directory entries until we reach the end offset.
 831          * Get more blocks and readahead as necessary.
 832          */
 833         while (curoff < XFS_DIR2_LEAF_OFFSET) {
 834                 /*
 835                  * If we have no buffer, or we're off the end of the
 836                  * current buffer, need to get another one.
 837                  */
 838                 if (!bp || ptr >= (char *)bp->b_addr + mp->m_dirblksize) {
 839                         /*
 840                          * If we have a buffer, we need to release it and
 841                          * take it out of the mapping.
 842                          */
 843                         if (bp) {
 844                                 xfs_trans_brelse(NULL, bp);
 845                                 bp = NULL;
 846                                 map_blocks -= mp->m_dirblkfsbs;
 847                                 /*
 848                                  * Loop to get rid of the extents for the
 849                                  * directory block.
 850                                  */
 851                                 for (i = mp->m_dirblkfsbs; i > 0; ) {
 852                                         j = MIN((int)map->br_blockcount, i);
 853                                         map->br_blockcount -= j;
 854                                         map->br_startblock += j;
 855                                         map->br_startoff += j;
 856                                         /*
 857                                          * If mapping is done, pitch it from
 858                                          * the table.
 859                                          */
 860                                         if (!map->br_blockcount && --map_valid)
 861                                                 memmove(&map[0], &map[1],
 862                                                         sizeof(map[0]) *
 863                                                         map_valid);
 864                                         i -= j;
 865                                 }
 866                         }
 867                         /*
 868                          * Recalculate the readahead blocks wanted.
 869                          */
 870                         ra_want = howmany(bufsize + mp->m_dirblksize,
 871                                           mp->m_sb.sb_blocksize) - 1;
 872                         ASSERT(ra_want >= 0);
 873
 874                         /*
 875                          * If we don't have as many as we want, and we haven't
 876                          * run out of data blocks, get some more mappings.
 877                          */
 878                         if (1 + ra_want > map_blocks &&
 879                             map_off <
 880                             xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET)) {
 881                                 /*
 882                                  * Get more bmaps, fill in after the ones
 883                                  * we already have in the table.
 884                                  */
 885                                 nmap = map_size - map_valid;
 886                                 error = xfs_bmapi_read(dp, map_off,
 887                                         xfs_dir2_byte_to_da(mp,
 888                                                 XFS_DIR2_LEAF_OFFSET) - map_off,
 889                                         &map[map_valid], &nmap, 0);
 890                                 /*
 891                                  * Don't know if we should ignore this or
 892                                  * try to return an error.
 893                                  * The trouble with returning errors
 894                                  * is that readdir will just stop without
 895                                  * actually passing the error through.
 896                                  */
 897                                 if (error)
 898                                         break;  /* XXX */
 899                                 /*
 900                                  * If we got all the mappings we asked for,
 901                                  * set the final map offset based on the
 902                                  * last bmap value received.
 903                                  * Otherwise, we've reached the end.
 904                                  */
 905                                 if (nmap == map_size - map_valid)
 906                                         map_off =
 907                                         map[map_valid + nmap - 1].br_startoff +
 908                                         map[map_valid + nmap - 1].br_blockcount;
 909                                 else
 910                                         map_off =
 911                                                 xfs_dir2_byte_to_da(mp,
 912                                                         XFS_DIR2_LEAF_OFFSET);
 913                                 /*
 914                                  * Look for holes in the mapping, and
 915                                  * eliminate them.  Count up the valid blocks.
 916                                  */
 917                                 for (i = map_valid; i < map_valid + nmap; ) {
 918                                         if (map[i].br_startblock ==
 919                                             HOLESTARTBLOCK) {
 920                                                 nmap--;
 921                                                 length = map_valid + nmap - i;
 922                                                 if (length)
 923                                                         memmove(&map[i],
 924                                                                 &map[i + 1],
 925                                                                 sizeof(map[i]) *
 926                                                                 length);
 927                                         } else {
 928                                                 map_blocks +=
 929                                                         map[i].br_blockcount;
 930                                                 i++;
 931                                         }
 932                                 }
 933                                 map_valid += nmap;
 934                         }
 935                         /*
 936                          * No valid mappings, so no more data blocks.
 937                          */
 938                         if (!map_valid) {
 939                                 curoff = xfs_dir2_da_to_byte(mp, map_off);
 940                                 break;
 941                         }
 942                         /*
 943                          * Read the directory block starting at the first
 944                          * mapping.
 945                          */
 946                         curdb = xfs_dir2_da_to_db(mp, map->br_startoff);
 947                         error = xfs_da_read_buf(NULL, dp, map->br_startoff,
 948                                 map->br_blockcount >= mp->m_dirblkfsbs ?
 949                                     XFS_FSB_TO_DADDR(mp, map->br_startblock) :
 950                                     -1,
 951                                 &bp, XFS_DATA_FORK);
 952                         /*
 953                          * Should just skip over the data block instead
 954                          * of giving up.
 955                          */
 956                         if (error)
 957                                 break;  /* XXX */
 958                         /*
 959                          * Adjust the current amount of read-ahead: we just
 960                          * read a block that was previously ra.
 961                          */
 962                         if (ra_current)
 963                                 ra_current -= mp->m_dirblkfsbs;
 964                         /*
 965                          * Do we need more readahead?
 966                          */
 967                         for (ra_index = ra_offset = i = 0;
 968                              ra_want > ra_current && i < map_blocks;
 969                              i += mp->m_dirblkfsbs) {
 970                                 ASSERT(ra_index < map_valid);
 971                                 /*
 972                                  * Read-ahead a contiguous directory block.
 973                                  */
 974                                 if (i > ra_current &&
 975                                     map[ra_index].br_blockcount >=
 976                                     mp->m_dirblkfsbs) {
 977                                         xfs_buf_readahead(mp->m_ddev_targp,
 978                                                 XFS_FSB_TO_DADDR(mp,
 979                                                    map[ra_index].br_startblock +
 980                                                    ra_offset),
 981                                                 (int)BTOBB(mp->m_dirblksize));
 982                                         ra_current = i;
 983                                 }
 984                                 /*
 985                                  * Read-ahead a non-contiguous directory block.
 986                                  * This doesn't use our mapping, but this
 987                                  * is a very rare case.
 988                                  */
 989                                 else if (i > ra_current) {
 990                                         (void)xfs_da_reada_buf(NULL, dp,
 991                                                 map[ra_index].br_startoff +
 992                                                 ra_offset, XFS_DATA_FORK);
 993                                         ra_current = i;
 994                                 }
 995                                 /*
 996                                  * Advance offset through the mapping table.
 997                                  */
 998                                 for (j = 0; j < mp->m_dirblkfsbs; j++) {
 999                                         /*
1000                                          * The rest of this extent but not
1001                                          * more than a dir block.
1002                                          */
1003                                         length = MIN(mp->m_dirblkfsbs,
1004                                                 (int)(map[ra_index].br_blockcount -
1005                                                 ra_offset));
1006                                         j += length;
1007                                         ra_offset += length;
1008                                         /*
1009                                          * Advance to the next mapping if
1010                                          * this one is used up.
1011                                          */
1012                                         if (ra_offset ==
1013                                             map[ra_index].br_blockcount) {
1014                                                 ra_offset = 0;
1015                                                 ra_index++;
1016                                         }
1017                                 }
1018                         }
1019                         /*
1020                          * Having done a read, we need to set a new offset.
1021                          */
1022                         newoff = xfs_dir2_db_off_to_byte(mp, curdb, 0);
1023                         /*
1024                          * Start of the current block.
1025                          */
1026                         if (curoff < newoff)
1027                                 curoff = newoff;
1028                         /*
1029                          * Make sure we're in the right block.
1030                          */
1031                         else if (curoff > newoff)
1032                                 ASSERT(xfs_dir2_byte_to_db(mp, curoff) ==
1033                                        curdb);
1034                         hdr = bp->b_addr;
1035                         xfs_dir2_data_check(dp, bp);
1036                         /*
1037                          * Find our position in the block.
1038                          */
1039                         ptr = (char *)(hdr + 1);
1040                         byteoff = xfs_dir2_byte_to_off(mp, curoff);
1041                         /*
1042                          * Skip past the header.
1043                          */
1044                         if (byteoff == 0)
1045                                 curoff += (uint)sizeof(*hdr);
1046                         /*
1047                          * Skip past entries until we reach our offset.
1048                          */
1049                         else {
1050                                 while ((char *)ptr - (char *)hdr < byteoff) {
1051                                         dup = (xfs_dir2_data_unused_t *)ptr;
1052
1053                                         if (be16_to_cpu(dup->freetag)
1054                                                   == XFS_DIR2_DATA_FREE_TAG) {
1055
1056                                                 length = be16_to_cpu(dup->length);
1057                                                 ptr += length;
1058                                                 continue;
1059                                         }
1060                                         dep = (xfs_dir2_data_entry_t *)ptr;
1061                                         length =
1062                                            xfs_dir2_data_entsize(dep->namelen);
1063                                         ptr += length;
1064                                 }
1065                                 /*
1066                                  * Now set our real offset.
1067                                  */
1068                                 curoff =
1069                                         xfs_dir2_db_off_to_byte(mp,
1070                                             xfs_dir2_byte_to_db(mp, curoff),
1071                                             (char *)ptr - (char *)hdr);
1072                                 if (ptr >= (char *)hdr + mp->m_dirblksize) {
1073                                         continue;
1074                                 }
1075                         }
1076                 }
1077                 /*
1078                  * We have a pointer to an entry.
1079                  * Is it a live one?
1080                  */
1081                 dup = (xfs_dir2_data_unused_t *)ptr;
1082                 /*
1083                  * No, it's unused, skip over it.
1084                  */
1085                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
1086                         length = be16_to_cpu(dup->length);
1087                         ptr += length;
1088                         curoff += length;
1089                         continue;
1090                 }
1091
1092                 dep = (xfs_dir2_data_entry_t *)ptr;
1093                 length = xfs_dir2_data_entsize(dep->namelen);
1094
1095                 if (filldir(dirent, (char *)dep->name, dep->namelen,
1096                             xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff,
1097                             be64_to_cpu(dep->inumber), DT_UNKNOWN))
1098                         break;
1099
1100                 /*
1101                  * Advance to next entry in the block.
1102                  */
1103                 ptr += length;
1104                 curoff += length;
1105                 /* bufsize may have just been a guess; don't go negative */
1106                 bufsize = bufsize > length ? bufsize - length : 0;
1107         }
1108
1109         /*
1110          * All done.  Set output offset value to current offset.
1111          */
1112         if (curoff > xfs_dir2_dataptr_to_byte(mp, XFS_DIR2_MAX_DATAPTR))
1113                 *offset = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
1114         else
1115                 *offset = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
1116         kmem_free(map);
1117         if (bp)
1118                 xfs_trans_brelse(NULL, bp);
1119         return error;
1120 }
1121
1122 /*
1123  * Initialize a new leaf block, leaf1 or leafn magic accepted.
1124  */
1125 int
1126 xfs_dir2_leaf_init(
1127         xfs_da_args_t           *args,          /* operation arguments */
1128         xfs_dir2_db_t           bno,            /* directory block number */
1129         struct xfs_buf          **bpp,          /* out: leaf buffer */
1130         int                     magic)          /* magic number for block */
1131 {
1132         struct xfs_buf          *bp;            /* leaf buffer */
1133         xfs_inode_t             *dp;            /* incore directory inode */
1134         int                     error;          /* error return code */
1135         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1136         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1137         xfs_mount_t             *mp;            /* filesystem mount point */
1138         xfs_trans_t             *tp;            /* transaction pointer */
1139
1140         dp = args->dp;
1141         ASSERT(dp != NULL);
1142         tp = args->trans;
1143         mp = dp->i_mount;
1144         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
1145                bno < XFS_DIR2_FREE_FIRSTDB(mp));
1146         /*
1147          * Get the buffer for the block.
1148          */
1149         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
1150                 XFS_DATA_FORK);
1151         if (error) {
1152                 return error;
1153         }
1154         ASSERT(bp != NULL);
1155         leaf = bp->b_addr;
1156         /*
1157          * Initialize the header.
1158          */
1159         leaf->hdr.info.magic = cpu_to_be16(magic);
1160         leaf->hdr.info.forw = 0;
1161         leaf->hdr.info.back = 0;
1162         leaf->hdr.count = 0;
1163         leaf->hdr.stale = 0;
1164         xfs_dir2_leaf_log_header(tp, bp);
1165         /*
1166          * If it's a leaf-format directory initialize the tail.
1167          * In this case our caller has the real bests table to copy into
1168          * the block.
1169          */
1170         if (magic == XFS_DIR2_LEAF1_MAGIC) {
1171                 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1172                 ltp->bestcount = 0;
1173                 xfs_dir2_leaf_log_tail(tp, bp);
1174         }
1175         *bpp = bp;
1176         return 0;
1177 }
1178
1179 /*
1180  * Log the bests entries indicated from a leaf1 block.
1181  */
1182 static void
1183 xfs_dir2_leaf_log_bests(
1184         xfs_trans_t             *tp,            /* transaction pointer */
1185         struct xfs_buf          *bp,            /* leaf buffer */
1186         int                     first,          /* first entry to log */
1187         int                     last)           /* last entry to log */
1188 {
1189         __be16                  *firstb;        /* pointer to first entry */
1190         __be16                  *lastb;         /* pointer to last entry */
1191         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1192         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1193
1194         leaf = bp->b_addr;
1195         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1196         ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1197         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1198         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1199         xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1200                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1201 }
1202
1203 /*
1204  * Log the leaf entries indicated from a leaf1 or leafn block.
1205  */
1206 void
1207 xfs_dir2_leaf_log_ents(
1208         xfs_trans_t             *tp,            /* transaction pointer */
1209         struct xfs_buf          *bp,            /* leaf buffer */
1210         int                     first,          /* first entry to log */
1211         int                     last)           /* last entry to log */
1212 {
1213         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1214         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1215         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1216
1217         leaf = bp->b_addr;
1218         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1219                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1220         firstlep = &leaf->ents[first];
1221         lastlep = &leaf->ents[last];
1222         xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1223                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1224 }
1225
1226 /*
1227  * Log the header of the leaf1 or leafn block.
1228  */
1229 void
1230 xfs_dir2_leaf_log_header(
1231         struct xfs_trans        *tp,
1232         struct xfs_buf          *bp)
1233 {
1234         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1235
1236         leaf = bp->b_addr;
1237         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1238                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1239         xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1240                 (uint)(sizeof(leaf->hdr) - 1));
1241 }
1242
1243 /*
1244  * Log the tail of the leaf1 block.
1245  */
1246 STATIC void
1247 xfs_dir2_leaf_log_tail(
1248         struct xfs_trans        *tp,
1249         struct xfs_buf          *bp)
1250 {
1251         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1252         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1253         xfs_mount_t             *mp;            /* filesystem mount point */
1254
1255         mp = tp->t_mountp;
1256         leaf = bp->b_addr;
1257         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1258         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1259         xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1260                 (uint)(mp->m_dirblksize - 1));
1261 }
1262
1263 /*
1264  * Look up the entry referred to by args in the leaf format directory.
1265  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1266  * is also used by the node-format code.
1267  */
1268 int
1269 xfs_dir2_leaf_lookup(
1270         xfs_da_args_t           *args)          /* operation arguments */
1271 {
1272         struct xfs_buf          *dbp;           /* data block buffer */
1273         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1274         xfs_inode_t             *dp;            /* incore directory inode */
1275         int                     error;          /* error return code */
1276         int                     index;          /* found entry index */
1277         struct xfs_buf          *lbp;           /* leaf buffer */
1278         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1279         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1280         xfs_trans_t             *tp;            /* transaction pointer */
1281
1282         trace_xfs_dir2_leaf_lookup(args);
1283
1284         /*
1285          * Look up name in the leaf block, returning both buffers and index.
1286          */
1287         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1288                 return error;
1289         }
1290         tp = args->trans;
1291         dp = args->dp;
1292         xfs_dir2_leaf_check(dp, lbp);
1293         leaf = lbp->b_addr;
1294         /*
1295          * Get to the leaf entry and contained data entry address.
1296          */
1297         lep = &leaf->ents[index];
1298         /*
1299          * Point to the data entry.
1300          */
1301         dep = (xfs_dir2_data_entry_t *)
1302               ((char *)dbp->b_addr +
1303                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1304         /*
1305          * Return the found inode number & CI name if appropriate
1306          */
1307         args->inumber = be64_to_cpu(dep->inumber);
1308         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1309         xfs_trans_brelse(tp, dbp);
1310         xfs_trans_brelse(tp, lbp);
1311         return XFS_ERROR(error);
1312 }
1313
1314 /*
1315  * Look up name/hash in the leaf block.
1316  * Fill in indexp with the found index, and dbpp with the data buffer.
1317  * If not found dbpp will be NULL, and ENOENT comes back.
1318  * lbpp will always be filled in with the leaf buffer unless there's an error.
1319  */
1320 static int                                      /* error */
1321 xfs_dir2_leaf_lookup_int(
1322         xfs_da_args_t           *args,          /* operation arguments */
1323         struct xfs_buf          **lbpp,         /* out: leaf buffer */
1324         int                     *indexp,        /* out: index in leaf block */
1325         struct xfs_buf          **dbpp)         /* out: data buffer */
1326 {
1327         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1328         struct xfs_buf          *dbp = NULL;    /* data buffer */
1329         xfs_dir2_data_entry_t   *dep;           /* data entry */
1330         xfs_inode_t             *dp;            /* incore directory inode */
1331         int                     error;          /* error return code */
1332         int                     index;          /* index in leaf block */
1333         struct xfs_buf          *lbp;           /* leaf buffer */
1334         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1335         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1336         xfs_mount_t             *mp;            /* filesystem mount point */
1337         xfs_dir2_db_t           newdb;          /* new data block number */
1338         xfs_trans_t             *tp;            /* transaction pointer */
1339         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1340         enum xfs_dacmp          cmp;            /* name compare result */
1341
1342         dp = args->dp;
1343         tp = args->trans;
1344         mp = dp->i_mount;
1345         /*
1346          * Read the leaf block into the buffer.
1347          */
1348         error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
1349                                                         XFS_DATA_FORK);
1350         if (error)
1351                 return error;
1352         *lbpp = lbp;
1353         leaf = lbp->b_addr;
1354         xfs_dir2_leaf_check(dp, lbp);
1355         /*
1356          * Look for the first leaf entry with our hash value.
1357          */
1358         index = xfs_dir2_leaf_search_hash(args, lbp);
1359         /*
1360          * Loop over all the entries with the right hash value
1361          * looking to match the name.
1362          */
1363         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
1364                                 be32_to_cpu(lep->hashval) == args->hashval;
1365                                 lep++, index++) {
1366                 /*
1367                  * Skip over stale leaf entries.
1368                  */
1369                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1370                         continue;
1371                 /*
1372                  * Get the new data block number.
1373                  */
1374                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1375                 /*
1376                  * If it's not the same as the old data block number,
1377                  * need to pitch the old one and read the new one.
1378                  */
1379                 if (newdb != curdb) {
1380                         if (dbp)
1381                                 xfs_trans_brelse(tp, dbp);
1382                         error = xfs_da_read_buf(tp, dp,
1383                                                 xfs_dir2_db_to_da(mp, newdb),
1384                                                 -1, &dbp, XFS_DATA_FORK);
1385                         if (error) {
1386                                 xfs_trans_brelse(tp, lbp);
1387                                 return error;
1388                         }
1389                         xfs_dir2_data_check(dp, dbp);
1390                         curdb = newdb;
1391                 }
1392                 /*
1393                  * Point to the data entry.
1394                  */
1395                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1396                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1397                 /*
1398                  * Compare name and if it's an exact match, return the index
1399                  * and buffer. If it's the first case-insensitive match, store
1400                  * the index and buffer and continue looking for an exact match.
1401                  */
1402                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1403                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1404                         args->cmpresult = cmp;
1405                         *indexp = index;
1406                         /* case exact match: return the current buffer. */
1407                         if (cmp == XFS_CMP_EXACT) {
1408                                 *dbpp = dbp;
1409                                 return 0;
1410                         }
1411                         cidb = curdb;
1412                 }
1413         }
1414         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1415         /*
1416          * Here, we can only be doing a lookup (not a rename or remove).
1417          * If a case-insensitive match was found earlier, re-read the
1418          * appropriate data block if required and return it.
1419          */
1420         if (args->cmpresult == XFS_CMP_CASE) {
1421                 ASSERT(cidb != -1);
1422                 if (cidb != curdb) {
1423                         xfs_trans_brelse(tp, dbp);
1424                         error = xfs_da_read_buf(tp, dp,
1425                                                 xfs_dir2_db_to_da(mp, cidb),
1426                                                 -1, &dbp, XFS_DATA_FORK);
1427                         if (error) {
1428                                 xfs_trans_brelse(tp, lbp);
1429                                 return error;
1430                         }
1431                 }
1432                 *dbpp = dbp;
1433                 return 0;
1434         }
1435         /*
1436          * No match found, return ENOENT.
1437          */
1438         ASSERT(cidb == -1);
1439         if (dbp)
1440                 xfs_trans_brelse(tp, dbp);
1441         xfs_trans_brelse(tp, lbp);
1442         return XFS_ERROR(ENOENT);
1443 }
1444
1445 /*
1446  * Remove an entry from a leaf format directory.
1447  */
1448 int                                             /* error */
1449 xfs_dir2_leaf_removename(
1450         xfs_da_args_t           *args)          /* operation arguments */
1451 {
1452         __be16                  *bestsp;        /* leaf block best freespace */
1453         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1454         xfs_dir2_db_t           db;             /* data block number */
1455         struct xfs_buf          *dbp;           /* data block buffer */
1456         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1457         xfs_inode_t             *dp;            /* incore directory inode */
1458         int                     error;          /* error return code */
1459         xfs_dir2_db_t           i;              /* temporary data block # */
1460         int                     index;          /* index into leaf entries */
1461         struct xfs_buf          *lbp;           /* leaf buffer */
1462         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1463         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1464         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1465         xfs_mount_t             *mp;            /* filesystem mount point */
1466         int                     needlog;        /* need to log data header */
1467         int                     needscan;       /* need to rescan data frees */
1468         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1469         xfs_trans_t             *tp;            /* transaction pointer */
1470
1471         trace_xfs_dir2_leaf_removename(args);
1472
1473         /*
1474          * Lookup the leaf entry, get the leaf and data blocks read in.
1475          */
1476         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1477                 return error;
1478         }
1479         dp = args->dp;
1480         tp = args->trans;
1481         mp = dp->i_mount;
1482         leaf = lbp->b_addr;
1483         hdr = dbp->b_addr;
1484         xfs_dir2_data_check(dp, dbp);
1485         /*
1486          * Point to the leaf entry, use that to point to the data entry.
1487          */
1488         lep = &leaf->ents[index];
1489         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1490         dep = (xfs_dir2_data_entry_t *)
1491               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1492         needscan = needlog = 0;
1493         oldbest = be16_to_cpu(hdr->bestfree[0].length);
1494         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1495         bestsp = xfs_dir2_leaf_bests_p(ltp);
1496         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1497         /*
1498          * Mark the former data entry unused.
1499          */
1500         xfs_dir2_data_make_free(tp, dbp,
1501                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1502                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1503         /*
1504          * We just mark the leaf entry stale by putting a null in it.
1505          */
1506         be16_add_cpu(&leaf->hdr.stale, 1);
1507         xfs_dir2_leaf_log_header(tp, lbp);
1508         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1509         xfs_dir2_leaf_log_ents(tp, lbp, index, index);
1510         /*
1511          * Scan the freespace in the data block again if necessary,
1512          * log the data block header if necessary.
1513          */
1514         if (needscan)
1515                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1516         if (needlog)
1517                 xfs_dir2_data_log_header(tp, dbp);
1518         /*
1519          * If the longest freespace in the data block has changed,
1520          * put the new value in the bests table and log that.
1521          */
1522         if (be16_to_cpu(hdr->bestfree[0].length) != oldbest) {
1523                 bestsp[db] = hdr->bestfree[0].length;
1524                 xfs_dir2_leaf_log_bests(tp, lbp, db, db);
1525         }
1526         xfs_dir2_data_check(dp, dbp);
1527         /*
1528          * If the data block is now empty then get rid of the data block.
1529          */
1530         if (be16_to_cpu(hdr->bestfree[0].length) ==
1531             mp->m_dirblksize - (uint)sizeof(*hdr)) {
1532                 ASSERT(db != mp->m_dirdatablk);
1533                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1534                         /*
1535                          * Nope, can't get rid of it because it caused
1536                          * allocation of a bmap btree block to do so.
1537                          * Just go on, returning success, leaving the
1538                          * empty block in place.
1539                          */
1540                         if (error == ENOSPC && args->total == 0)
1541                                 error = 0;
1542                         xfs_dir2_leaf_check(dp, lbp);
1543                         return error;
1544                 }
1545                 dbp = NULL;
1546                 /*
1547                  * If this is the last data block then compact the
1548                  * bests table by getting rid of entries.
1549                  */
1550                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1551                         /*
1552                          * Look for the last active entry (i).
1553                          */
1554                         for (i = db - 1; i > 0; i--) {
1555                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1556                                         break;
1557                         }
1558                         /*
1559                          * Copy the table down so inactive entries at the
1560                          * end are removed.
1561                          */
1562                         memmove(&bestsp[db - i], bestsp,
1563                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1564                         be32_add_cpu(&ltp->bestcount, -(db - i));
1565                         xfs_dir2_leaf_log_tail(tp, lbp);
1566                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1567                 } else
1568                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1569         }
1570         /*
1571          * If the data block was not the first one, drop it.
1572          */
1573         else if (db != mp->m_dirdatablk)
1574                 dbp = NULL;
1575
1576         xfs_dir2_leaf_check(dp, lbp);
1577         /*
1578          * See if we can convert to block form.
1579          */
1580         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1581 }
1582
1583 /*
1584  * Replace the inode number in a leaf format directory entry.
1585  */
1586 int                                             /* error */
1587 xfs_dir2_leaf_replace(
1588         xfs_da_args_t           *args)          /* operation arguments */
1589 {
1590         struct xfs_buf          *dbp;           /* data block buffer */
1591         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1592         xfs_inode_t             *dp;            /* incore directory inode */
1593         int                     error;          /* error return code */
1594         int                     index;          /* index of leaf entry */
1595         struct xfs_buf          *lbp;           /* leaf buffer */
1596         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1597         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1598         xfs_trans_t             *tp;            /* transaction pointer */
1599
1600         trace_xfs_dir2_leaf_replace(args);
1601
1602         /*
1603          * Look up the entry.
1604          */
1605         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1606                 return error;
1607         }
1608         dp = args->dp;
1609         leaf = lbp->b_addr;
1610         /*
1611          * Point to the leaf entry, get data address from it.
1612          */
1613         lep = &leaf->ents[index];
1614         /*
1615          * Point to the data entry.
1616          */
1617         dep = (xfs_dir2_data_entry_t *)
1618               ((char *)dbp->b_addr +
1619                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1620         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1621         /*
1622          * Put the new inode number in, log it.
1623          */
1624         dep->inumber = cpu_to_be64(args->inumber);
1625         tp = args->trans;
1626         xfs_dir2_data_log_entry(tp, dbp, dep);
1627         xfs_dir2_leaf_check(dp, lbp);
1628         xfs_trans_brelse(tp, lbp);
1629         return 0;
1630 }
1631
1632 /*
1633  * Return index in the leaf block (lbp) which is either the first
1634  * one with this hash value, or if there are none, the insert point
1635  * for that hash value.
1636  */
1637 int                                             /* index value */
1638 xfs_dir2_leaf_search_hash(
1639         xfs_da_args_t           *args,          /* operation arguments */
1640         struct xfs_buf          *lbp)           /* leaf buffer */
1641 {
1642         xfs_dahash_t            hash=0;         /* hash from this entry */
1643         xfs_dahash_t            hashwant;       /* hash value looking for */
1644         int                     high;           /* high leaf index */
1645         int                     low;            /* low leaf index */
1646         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1647         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1648         int                     mid=0;          /* current leaf index */
1649
1650         leaf = lbp->b_addr;
1651 #ifndef __KERNEL__
1652         if (!leaf->hdr.count)
1653                 return 0;
1654 #endif
1655         /*
1656          * Note, the table cannot be empty, so we have to go through the loop.
1657          * Binary search the leaf entries looking for our hash value.
1658          */
1659         for (lep = leaf->ents, low = 0, high = be16_to_cpu(leaf->hdr.count) - 1,
1660                 hashwant = args->hashval;
1661              low <= high; ) {
1662                 mid = (low + high) >> 1;
1663                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1664                         break;
1665                 if (hash < hashwant)
1666                         low = mid + 1;
1667                 else
1668                         high = mid - 1;
1669         }
1670         /*
1671          * Found one, back up through all the equal hash values.
1672          */
1673         if (hash == hashwant) {
1674                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1675                         mid--;
1676                 }
1677         }
1678         /*
1679          * Need to point to an entry higher than ours.
1680          */
1681         else if (hash < hashwant)
1682                 mid++;
1683         return mid;
1684 }
1685
1686 /*
1687  * Trim off a trailing data block.  We know it's empty since the leaf
1688  * freespace table says so.
1689  */
1690 int                                             /* error */
1691 xfs_dir2_leaf_trim_data(
1692         xfs_da_args_t           *args,          /* operation arguments */
1693         struct xfs_buf          *lbp,           /* leaf buffer */
1694         xfs_dir2_db_t           db)             /* data block number */
1695 {
1696         __be16                  *bestsp;        /* leaf bests table */
1697         struct xfs_buf          *dbp;           /* data block buffer */
1698         xfs_inode_t             *dp;            /* incore directory inode */
1699         int                     error;          /* error return value */
1700         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1701         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1702         xfs_mount_t             *mp;            /* filesystem mount point */
1703         xfs_trans_t             *tp;            /* transaction pointer */
1704
1705         dp = args->dp;
1706         mp = dp->i_mount;
1707         tp = args->trans;
1708         /*
1709          * Read the offending data block.  We need its buffer.
1710          */
1711         if ((error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp,
1712                         XFS_DATA_FORK))) {
1713                 return error;
1714         }
1715
1716         leaf = lbp->b_addr;
1717         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1718
1719 #ifdef DEBUG
1720 {
1721         struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1722
1723         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
1724         ASSERT(be16_to_cpu(hdr->bestfree[0].length) ==
1725                mp->m_dirblksize - (uint)sizeof(*hdr));
1726         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1727 }
1728 #endif
1729
1730         /*
1731          * Get rid of the data block.
1732          */
1733         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1734                 ASSERT(error != ENOSPC);
1735                 xfs_trans_brelse(tp, dbp);
1736                 return error;
1737         }
1738         /*
1739          * Eliminate the last bests entry from the table.
1740          */
1741         bestsp = xfs_dir2_leaf_bests_p(ltp);
1742         be32_add_cpu(&ltp->bestcount, -1);
1743         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1744         xfs_dir2_leaf_log_tail(tp, lbp);
1745         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1746         return 0;
1747 }
1748
1749 static inline size_t
1750 xfs_dir2_leaf_size(
1751         struct xfs_dir2_leaf_hdr        *hdr,
1752         int                             counts)
1753 {
1754         int                     entries;
1755
1756         entries = be16_to_cpu(hdr->count) - be16_to_cpu(hdr->stale);
1757         return sizeof(xfs_dir2_leaf_hdr_t) +
1758             entries * sizeof(xfs_dir2_leaf_entry_t) +
1759             counts * sizeof(xfs_dir2_data_off_t) +
1760             sizeof(xfs_dir2_leaf_tail_t);
1761 }
1762
1763 /*
1764  * Convert node form directory to leaf form directory.
1765  * The root of the node form dir needs to already be a LEAFN block.
1766  * Just return if we can't do anything.
1767  */
1768 int                                             /* error */
1769 xfs_dir2_node_to_leaf(
1770         xfs_da_state_t          *state)         /* directory operation state */
1771 {
1772         xfs_da_args_t           *args;          /* operation arguments */
1773         xfs_inode_t             *dp;            /* incore directory inode */
1774         int                     error;          /* error return code */
1775         struct xfs_buf          *fbp;           /* buffer for freespace block */
1776         xfs_fileoff_t           fo;             /* freespace file offset */
1777         xfs_dir2_free_t         *free;          /* freespace structure */
1778         struct xfs_buf          *lbp;           /* buffer for leaf block */
1779         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1780         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1781         xfs_mount_t             *mp;            /* filesystem mount point */
1782         int                     rval;           /* successful free trim? */
1783         xfs_trans_t             *tp;            /* transaction pointer */
1784
1785         /*
1786          * There's more than a leaf level in the btree, so there must
1787          * be multiple leafn blocks.  Give up.
1788          */
1789         if (state->path.active > 1)
1790                 return 0;
1791         args = state->args;
1792
1793         trace_xfs_dir2_node_to_leaf(args);
1794
1795         mp = state->mp;
1796         dp = args->dp;
1797         tp = args->trans;
1798         /*
1799          * Get the last offset in the file.
1800          */
1801         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1802                 return error;
1803         }
1804         fo -= mp->m_dirblkfsbs;
1805         /*
1806          * If there are freespace blocks other than the first one,
1807          * take this opportunity to remove trailing empty freespace blocks
1808          * that may have been left behind during no-space-reservation
1809          * operations.
1810          */
1811         while (fo > mp->m_dirfreeblk) {
1812                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1813                         return error;
1814                 }
1815                 if (rval)
1816                         fo -= mp->m_dirblkfsbs;
1817                 else
1818                         return 0;
1819         }
1820         /*
1821          * Now find the block just before the freespace block.
1822          */
1823         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1824                 return error;
1825         }
1826         /*
1827          * If it's not the single leaf block, give up.
1828          */
1829         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1830                 return 0;
1831         lbp = state->path.blk[0].bp;
1832         leaf = lbp->b_addr;
1833         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1834         /*
1835          * Read the freespace block.
1836          */
1837         if ((error = xfs_da_read_buf(tp, dp, mp->m_dirfreeblk, -1, &fbp,
1838                         XFS_DATA_FORK))) {
1839                 return error;
1840         }
1841         free = fbp->b_addr;
1842         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1843         ASSERT(!free->hdr.firstdb);
1844
1845         /*
1846          * Now see if the leafn and free data will fit in a leaf1.
1847          * If not, release the buffer and give up.
1848          */
1849         if (xfs_dir2_leaf_size(&leaf->hdr, be32_to_cpu(free->hdr.nvalid)) >
1850                         mp->m_dirblksize) {
1851                 xfs_trans_brelse(tp, fbp);
1852                 return 0;
1853         }
1854
1855         /*
1856          * If the leaf has any stale entries in it, compress them out.
1857          * The compact routine will log the header.
1858          */
1859         if (be16_to_cpu(leaf->hdr.stale))
1860                 xfs_dir2_leaf_compact(args, lbp);
1861         else
1862                 xfs_dir2_leaf_log_header(tp, lbp);
1863         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAF1_MAGIC);
1864         /*
1865          * Set up the leaf tail from the freespace block.
1866          */
1867         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1868         ltp->bestcount = free->hdr.nvalid;
1869         /*
1870          * Set up the leaf bests table.
1871          */
1872         memcpy(xfs_dir2_leaf_bests_p(ltp), free->bests,
1873                 be32_to_cpu(ltp->bestcount) * sizeof(xfs_dir2_data_off_t));
1874         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1875         xfs_dir2_leaf_log_tail(tp, lbp);
1876         xfs_dir2_leaf_check(dp, lbp);
1877         /*
1878          * Get rid of the freespace block.
1879          */
1880         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1881         if (error) {
1882                 /*
1883                  * This can't fail here because it can only happen when
1884                  * punching out the middle of an extent, and this is an
1885                  * isolated block.
1886                  */
1887                 ASSERT(error != ENOSPC);
1888                 return error;
1889         }
1890         fbp = NULL;
1891         /*
1892          * Now see if we can convert the single-leaf directory
1893          * down to a block form directory.
1894          * This routine always kills the dabuf for the leaf, so
1895          * eliminate it from the path.
1896          */
1897         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1898         state->path.blk[0].bp = NULL;
1899         return error;
1900 }