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, NULL);
 319         if (error)
 320                 return error;
 321         ASSERT(lbp != NULL);
 322         /*
 323          * Look up the entry by hash value and name.
 324          * We know it's not there, our caller has already done a lookup.
 325          * So the index is of the entry to insert in front of.
 326          * But if there are dup hash values the index is of the first of those.
 327          */
 328         index = xfs_dir2_leaf_search_hash(args, lbp);
 329         leaf = lbp->b_addr;
 330         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 331         bestsp = xfs_dir2_leaf_bests_p(ltp);
 332         length = xfs_dir2_data_entsize(args->namelen);
 333         /*
 334          * See if there are any entries with the same hash value
 335          * and space in their block for the new entry.
 336          * This is good because it puts multiple same-hash value entries
 337          * in a data block, improving the lookup of those entries.
 338          */
 339         for (use_block = -1, lep = &leaf->ents[index];
 340              index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
 341              index++, lep++) {
 342                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 343                         continue;
 344                 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 345                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 346                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
 347                 if (be16_to_cpu(bestsp[i]) >= length) {
 348                         use_block = i;
 349                         break;
 350                 }
 351         }
 352         /*
 353          * Didn't find a block yet, linear search all the data blocks.
 354          */
 355         if (use_block == -1) {
 356                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 357                         /*
 358                          * Remember a block we see that's missing.
 359                          */
 360                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
 361                             use_block == -1)
 362                                 use_block = i;
 363                         else if (be16_to_cpu(bestsp[i]) >= length) {
 364                                 use_block = i;
 365                                 break;
 366                         }
 367                 }
 368         }
 369         /*
 370          * How many bytes do we need in the leaf block?
 371          */
 372         needbytes = 0;
 373         if (!leaf->hdr.stale)
 374                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
 375         if (use_block == -1)
 376                 needbytes += sizeof(xfs_dir2_data_off_t);
 377
 378         /*
 379          * Now kill use_block if it refers to a missing block, so we
 380          * can use it as an indication of allocation needed.
 381          */
 382         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
 383                 use_block = -1;
 384         /*
 385          * If we don't have enough free bytes but we can make enough
 386          * by compacting out stale entries, we'll do that.
 387          */
 388         if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <
 389                                 needbytes && be16_to_cpu(leaf->hdr.stale) > 1) {
 390                 compact = 1;
 391         }
 392         /*
 393          * Otherwise if we don't have enough free bytes we need to
 394          * convert to node form.
 395          */
 396         else if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(
 397                                                 leaf->hdr.count)] < needbytes) {
 398                 /*
 399                  * Just checking or no space reservation, give up.
 400                  */
 401                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 402                                                         args->total == 0) {
 403                         xfs_trans_brelse(tp, lbp);
 404                         return XFS_ERROR(ENOSPC);
 405                 }
 406                 /*
 407                  * Convert to node form.
 408                  */
 409                 error = xfs_dir2_leaf_to_node(args, lbp);
 410                 if (error)
 411                         return error;
 412                 /*
 413                  * Then add the new entry.
 414                  */
 415                 return xfs_dir2_node_addname(args);
 416         }
 417         /*
 418          * Otherwise it will fit without compaction.
 419          */
 420         else
 421                 compact = 0;
 422         /*
 423          * If just checking, then it will fit unless we needed to allocate
 424          * a new data block.
 425          */
 426         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 427                 xfs_trans_brelse(tp, lbp);
 428                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
 429         }
 430         /*
 431          * If no allocations are allowed, return now before we've
 432          * changed anything.
 433          */
 434         if (args->total == 0 && use_block == -1) {
 435                 xfs_trans_brelse(tp, lbp);
 436                 return XFS_ERROR(ENOSPC);
 437         }
 438         /*
 439          * Need to compact the leaf entries, removing stale ones.
 440          * Leave one stale entry behind - the one closest to our
 441          * insertion index - and we'll shift that one to our insertion
 442          * point later.
 443          */
 444         if (compact) {
 445                 xfs_dir2_leaf_compact_x1(lbp, &index, &lowstale, &highstale,
 446                         &lfloglow, &lfloghigh);
 447         }
 448         /*
 449          * There are stale entries, so we'll need log-low and log-high
 450          * impossibly bad values later.
 451          */
 452         else if (be16_to_cpu(leaf->hdr.stale)) {
 453                 lfloglow = be16_to_cpu(leaf->hdr.count);
 454                 lfloghigh = -1;
 455         }
 456         /*
 457          * If there was no data block space found, we need to allocate
 458          * a new one.
 459          */
 460         if (use_block == -1) {
 461                 /*
 462                  * Add the new data block.
 463                  */
 464                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 465                                 &use_block))) {
 466                         xfs_trans_brelse(tp, lbp);
 467                         return error;
 468                 }
 469                 /*
 470                  * Initialize the block.
 471                  */
 472                 if ((error = xfs_dir2_data_init(args, use_block, &dbp))) {
 473                         xfs_trans_brelse(tp, lbp);
 474                         return error;
 475                 }
 476                 /*
 477                  * If we're adding a new data block on the end we need to
 478                  * extend the bests table.  Copy it up one entry.
 479                  */
 480                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
 481                         bestsp--;
 482                         memmove(&bestsp[0], &bestsp[1],
 483                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 484                         be32_add_cpu(&ltp->bestcount, 1);
 485                         xfs_dir2_leaf_log_tail(tp, lbp);
 486                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
 487                 }
 488                 /*
 489                  * If we're filling in a previously empty block just log it.
 490                  */
 491                 else
 492                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 493                 hdr = dbp->b_addr;
 494                 bestsp[use_block] = hdr->bestfree[0].length;
 495                 grown = 1;
 496         } else {
 497                 /*
 498                  * Already had space in some data block.
 499                  * Just read that one in.
 500                  */
 501                 error = xfs_dir2_data_read(tp, dp,
 502                                            xfs_dir2_db_to_da(mp, use_block),
 503                                            -1, &dbp);
 504                 if (error) {
 505                         xfs_trans_brelse(tp, lbp);
 506                         return error;
 507                 }
 508                 hdr = dbp->b_addr;
 509                 grown = 0;
 510         }
 511         /*
 512          * Point to the biggest freespace in our data block.
 513          */
 514         dup = (xfs_dir2_data_unused_t *)
 515               ((char *)hdr + be16_to_cpu(hdr->bestfree[0].offset));
 516         ASSERT(be16_to_cpu(dup->length) >= length);
 517         needscan = needlog = 0;
 518         /*
 519          * Mark the initial part of our freespace in use for the new entry.
 520          */
 521         xfs_dir2_data_use_free(tp, dbp, dup,
 522                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
 523                 &needlog, &needscan);
 524         /*
 525          * Initialize our new entry (at last).
 526          */
 527         dep = (xfs_dir2_data_entry_t *)dup;
 528         dep->inumber = cpu_to_be64(args->inumber);
 529         dep->namelen = args->namelen;
 530         memcpy(dep->name, args->name, dep->namelen);
 531         tagp = xfs_dir2_data_entry_tag_p(dep);
 532         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 533         /*
 534          * Need to scan fix up the bestfree table.
 535          */
 536         if (needscan)
 537                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 538         /*
 539          * Need to log the data block's header.
 540          */
 541         if (needlog)
 542                 xfs_dir2_data_log_header(tp, dbp);
 543         xfs_dir2_data_log_entry(tp, dbp, dep);
 544         /*
 545          * If the bests table needs to be changed, do it.
 546          * Log the change unless we've already done that.
 547          */
 548         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(hdr->bestfree[0].length)) {
 549                 bestsp[use_block] = hdr->bestfree[0].length;
 550                 if (!grown)
 551                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 552         }
 553
 554         lep = xfs_dir2_leaf_find_entry(leaf, index, compact, lowstale,
 555                                        highstale, &lfloglow, &lfloghigh);
 556
 557         /*
 558          * Fill in the new leaf entry.
 559          */
 560         lep->hashval = cpu_to_be32(args->hashval);
 561         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
 562                                 be16_to_cpu(*tagp)));
 563         /*
 564          * Log the leaf fields and give up the buffers.
 565          */
 566         xfs_dir2_leaf_log_header(tp, lbp);
 567         xfs_dir2_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
 568         xfs_dir2_leaf_check(dp, lbp);
 569         xfs_dir2_data_check(dp, dbp);
 570         return 0;
 571 }
 572
 573 #ifdef DEBUG
 574 /*
 575  * Check the internal consistency of a leaf1 block.
 576  * Pop an assert if something is wrong.
 577  */
 578 STATIC void
 579 xfs_dir2_leaf_check(
 580         struct xfs_inode        *dp,            /* incore directory inode */
 581         struct xfs_buf          *bp)            /* leaf's buffer */
 582 {
 583         int                     i;              /* leaf index */
 584         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 585         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 586         xfs_mount_t             *mp;            /* filesystem mount point */
 587         int                     stale;          /* count of stale leaves */
 588
 589         leaf = bp->b_addr;
 590         mp = dp->i_mount;
 591         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
 592         /*
 593          * This value is not restrictive enough.
 594          * Should factor in the size of the bests table as well.
 595          * We can deduce a value for that from di_size.
 596          */
 597         ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
 598         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 599         /*
 600          * Leaves and bests don't overlap.
 601          */
 602         ASSERT((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <=
 603                (char *)xfs_dir2_leaf_bests_p(ltp));
 604         /*
 605          * Check hash value order, count stale entries.
 606          */
 607         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
 608                 if (i + 1 < be16_to_cpu(leaf->hdr.count))
 609                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
 610                                be32_to_cpu(leaf->ents[i + 1].hashval));
 611                 if (leaf->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 612                         stale++;
 613         }
 614         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
 615 }
 616 #endif  /* DEBUG */
 617
 618 /*
 619  * Compact out any stale entries in the leaf.
 620  * Log the header and changed leaf entries, if any.
 621  */
 622 void
 623 xfs_dir2_leaf_compact(
 624         xfs_da_args_t   *args,          /* operation arguments */
 625         struct xfs_buf  *bp)            /* leaf buffer */
 626 {
 627         int             from;           /* source leaf index */
 628         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 629         int             loglow;         /* first leaf entry to log */
 630         int             to;             /* target leaf index */
 631
 632         leaf = bp->b_addr;
 633         if (!leaf->hdr.stale) {
 634                 return;
 635         }
 636         /*
 637          * Compress out the stale entries in place.
 638          */
 639         for (from = to = 0, loglow = -1; from < be16_to_cpu(leaf->hdr.count); from++) {
 640                 if (leaf->ents[from].address ==
 641                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 642                         continue;
 643                 /*
 644                  * Only actually copy the entries that are different.
 645                  */
 646                 if (from > to) {
 647                         if (loglow == -1)
 648                                 loglow = to;
 649                         leaf->ents[to] = leaf->ents[from];
 650                 }
 651                 to++;
 652         }
 653         /*
 654          * Update and log the header, log the leaf entries.
 655          */
 656         ASSERT(be16_to_cpu(leaf->hdr.stale) == from - to);
 657         be16_add_cpu(&leaf->hdr.count, -(be16_to_cpu(leaf->hdr.stale)));
 658         leaf->hdr.stale = 0;
 659         xfs_dir2_leaf_log_header(args->trans, bp);
 660         if (loglow != -1)
 661                 xfs_dir2_leaf_log_ents(args->trans, bp, loglow, to - 1);
 662 }
 663
 664 /*
 665  * Compact the leaf entries, removing stale ones.
 666  * Leave one stale entry behind - the one closest to our
 667  * insertion index - and the caller will shift that one to our insertion
 668  * point later.
 669  * Return new insertion index, where the remaining stale entry is,
 670  * and leaf logging indices.
 671  */
 672 void
 673 xfs_dir2_leaf_compact_x1(
 674         struct xfs_buf  *bp,            /* leaf buffer */
 675         int             *indexp,        /* insertion index */
 676         int             *lowstalep,     /* out: stale entry before us */
 677         int             *highstalep,    /* out: stale entry after us */
 678         int             *lowlogp,       /* out: low log index */
 679         int             *highlogp)      /* out: high log index */
 680 {
 681         int             from;           /* source copy index */
 682         int             highstale;      /* stale entry at/after index */
 683         int             index;          /* insertion index */
 684         int             keepstale;      /* source index of kept stale */
 685         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 686         int             lowstale;       /* stale entry before index */
 687         int             newindex=0;     /* new insertion index */
 688         int             to;             /* destination copy index */
 689
 690         leaf = bp->b_addr;
 691         ASSERT(be16_to_cpu(leaf->hdr.stale) > 1);
 692         index = *indexp;
 693
 694         xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
 695
 696         /*
 697          * Pick the better of lowstale and highstale.
 698          */
 699         if (lowstale >= 0 &&
 700             (highstale == be16_to_cpu(leaf->hdr.count) ||
 701              index - lowstale <= highstale - index))
 702                 keepstale = lowstale;
 703         else
 704                 keepstale = highstale;
 705         /*
 706          * Copy the entries in place, removing all the stale entries
 707          * except keepstale.
 708          */
 709         for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
 710                 /*
 711                  * Notice the new value of index.
 712                  */
 713                 if (index == from)
 714                         newindex = to;
 715                 if (from != keepstale &&
 716                     leaf->ents[from].address ==
 717                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
 718                         if (from == to)
 719                                 *lowlogp = to;
 720                         continue;
 721                 }
 722                 /*
 723                  * Record the new keepstale value for the insertion.
 724                  */
 725                 if (from == keepstale)
 726                         lowstale = highstale = to;
 727                 /*
 728                  * Copy only the entries that have moved.
 729                  */
 730                 if (from > to)
 731                         leaf->ents[to] = leaf->ents[from];
 732                 to++;
 733         }
 734         ASSERT(from > to);
 735         /*
 736          * If the insertion point was past the last entry,
 737          * set the new insertion point accordingly.
 738          */
 739         if (index == from)
 740                 newindex = to;
 741         *indexp = newindex;
 742         /*
 743          * Adjust the leaf header values.
 744          */
 745         be16_add_cpu(&leaf->hdr.count, -(from - to));
 746         leaf->hdr.stale = cpu_to_be16(1);
 747         /*
 748          * Remember the low/high stale value only in the "right"
 749          * direction.
 750          */
 751         if (lowstale >= newindex)
 752                 lowstale = -1;
 753         else
 754                 highstale = be16_to_cpu(leaf->hdr.count);
 755         *highlogp = be16_to_cpu(leaf->hdr.count) - 1;
 756         *lowstalep = lowstale;
 757         *highstalep = highstale;
 758 }
 759
 760 struct xfs_dir2_leaf_map_info {
 761         xfs_extlen_t    map_blocks;     /* number of fsbs in map */
 762         xfs_dablk_t     map_off;        /* last mapped file offset */
 763         int             map_size;       /* total entries in *map */
 764         int             map_valid;      /* valid entries in *map */
 765         int             nmap;           /* mappings to ask xfs_bmapi */
 766         xfs_dir2_db_t   curdb;          /* db for current block */
 767         int             ra_current;     /* number of read-ahead blks */
 768         int             ra_index;       /* *map index for read-ahead */
 769         int             ra_offset;      /* map entry offset for ra */
 770         int             ra_want;        /* readahead count wanted */
 771         struct xfs_bmbt_irec map[];     /* map vector for blocks */
 772 };
 773
 774 STATIC int
 775 xfs_dir2_leaf_readbuf(
 776         struct xfs_inode        *dp,
 777         size_t                  bufsize,
 778         struct xfs_dir2_leaf_map_info *mip,
 779         xfs_dir2_off_t          *curoff,
 780         struct xfs_buf          **bpp)
 781 {
 782         struct xfs_mount        *mp = dp->i_mount;
 783         struct xfs_buf          *bp = *bpp;
 784         struct xfs_bmbt_irec    *map = mip->map;
 785         int                     error = 0;
 786         int                     length;
 787         int                     i;
 788         int                     j;
 789
 790         /*
 791          * If we have a buffer, we need to release it and
 792          * take it out of the mapping.
 793          */
 794
 795         if (bp) {
 796                 xfs_trans_brelse(NULL, bp);
 797                 bp = NULL;
 798                 mip->map_blocks -= mp->m_dirblkfsbs;
 799                 /*
 800                  * Loop to get rid of the extents for the
 801                  * directory block.
 802                  */
 803                 for (i = mp->m_dirblkfsbs; i > 0; ) {
 804                         j = min_t(int, map->br_blockcount, i);
 805                         map->br_blockcount -= j;
 806                         map->br_startblock += j;
 807                         map->br_startoff += j;
 808                         /*
 809                          * If mapping is done, pitch it from
 810                          * the table.
 811                          */
 812                         if (!map->br_blockcount && --mip->map_valid)
 813                                 memmove(&map[0], &map[1],
 814                                         sizeof(map[0]) * mip->map_valid);
 815                         i -= j;
 816                 }
 817         }
 818
 819         /*
 820          * Recalculate the readahead blocks wanted.
 821          */
 822         mip->ra_want = howmany(bufsize + mp->m_dirblksize,
 823                                mp->m_sb.sb_blocksize) - 1;
 824         ASSERT(mip->ra_want >= 0);
 825
 826         /*
 827          * If we don't have as many as we want, and we haven't
 828          * run out of data blocks, get some more mappings.
 829          */
 830         if (1 + mip->ra_want > mip->map_blocks &&
 831             mip->map_off < xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET)) {
 832                 /*
 833                  * Get more bmaps, fill in after the ones
 834                  * we already have in the table.
 835                  */
 836                 mip->nmap = mip->map_size - mip->map_valid;
 837                 error = xfs_bmapi_read(dp, mip->map_off,
 838                                 xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET) -
 839                                                                 mip->map_off,
 840                                 &map[mip->map_valid], &mip->nmap, 0);
 841
 842                 /*
 843                  * Don't know if we should ignore this or try to return an
 844                  * error.  The trouble with returning errors is that readdir
 845                  * will just stop without actually passing the error through.
 846                  */
 847                 if (error)
 848                         goto out;       /* XXX */
 849
 850                 /*
 851                  * If we got all the mappings we asked for, set the final map
 852                  * offset based on the last bmap value received.  Otherwise,
 853                  * we've reached the end.
 854                  */
 855                 if (mip->nmap == mip->map_size - mip->map_valid) {
 856                         i = mip->map_valid + mip->nmap - 1;
 857                         mip->map_off = map[i].br_startoff + map[i].br_blockcount;
 858                 } else
 859                         mip->map_off = xfs_dir2_byte_to_da(mp,
 860                                                         XFS_DIR2_LEAF_OFFSET);
 861
 862                 /*
 863                  * Look for holes in the mapping, and eliminate them.  Count up
 864                  * the valid blocks.
 865                  */
 866                 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
 867                         if (map[i].br_startblock == HOLESTARTBLOCK) {
 868                                 mip->nmap--;
 869                                 length = mip->map_valid + mip->nmap - i;
 870                                 if (length)
 871                                         memmove(&map[i], &map[i + 1],
 872                                                 sizeof(map[i]) * length);
 873                         } else {
 874                                 mip->map_blocks += map[i].br_blockcount;
 875                                 i++;
 876                         }
 877                 }
 878                 mip->map_valid += mip->nmap;
 879         }
 880
 881         /*
 882          * No valid mappings, so no more data blocks.
 883          */
 884         if (!mip->map_valid) {
 885                 *curoff = xfs_dir2_da_to_byte(mp, mip->map_off);
 886                 goto out;
 887         }
 888
 889         /*
 890          * Read the directory block starting at the first mapping.
 891          */
 892         mip->curdb = xfs_dir2_da_to_db(mp, map->br_startoff);
 893         error = xfs_dir2_data_read(NULL, dp, map->br_startoff,
 894                         map->br_blockcount >= mp->m_dirblkfsbs ?
 895                             XFS_FSB_TO_DADDR(mp, map->br_startblock) : -1, &bp);
 896
 897         /*
 898          * Should just skip over the data block instead of giving up.
 899          */
 900         if (error)
 901                 goto out;       /* XXX */
 902
 903         /*
 904          * Adjust the current amount of read-ahead: we just read a block that
 905          * was previously ra.
 906          */
 907         if (mip->ra_current)
 908                 mip->ra_current -= mp->m_dirblkfsbs;
 909
 910         /*
 911          * Do we need more readahead?
 912          */
 913         for (mip->ra_index = mip->ra_offset = i = 0;
 914              mip->ra_want > mip->ra_current && i < mip->map_blocks;
 915              i += mp->m_dirblkfsbs) {
 916                 ASSERT(mip->ra_index < mip->map_valid);
 917                 /*
 918                  * Read-ahead a contiguous directory block.
 919                  */
 920                 if (i > mip->ra_current &&
 921                     map[mip->ra_index].br_blockcount >= mp->m_dirblkfsbs) {
 922                         xfs_buf_readahead(mp->m_ddev_targp,
 923                                 XFS_FSB_TO_DADDR(mp,
 924                                         map[mip->ra_index].br_startblock +
 925                                                         mip->ra_offset),
 926                                 (int)BTOBB(mp->m_dirblksize), NULL);
 927                         mip->ra_current = i;
 928                 }
 929
 930                 /*
 931                  * Read-ahead a non-contiguous directory block.  This doesn't
 932                  * use our mapping, but this is a very rare case.
 933                  */
 934                 else if (i > mip->ra_current) {
 935                         xfs_da_reada_buf(NULL, dp,
 936                                         map[mip->ra_index].br_startoff +
 937                                                         mip->ra_offset,
 938                                         XFS_DATA_FORK, NULL);
 939                         mip->ra_current = i;
 940                 }
 941
 942                 /*
 943                  * Advance offset through the mapping table.
 944                  */
 945                 for (j = 0; j < mp->m_dirblkfsbs; j++) {
 946                         /*
 947                          * The rest of this extent but not more than a dir
 948                          * block.
 949                          */
 950                         length = min_t(int, mp->m_dirblkfsbs,
 951                                         map[mip->ra_index].br_blockcount -
 952                                                         mip->ra_offset);
 953                         j += length;
 954                         mip->ra_offset += length;
 955
 956                         /*
 957                          * Advance to the next mapping if this one is used up.
 958                          */
 959                         if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
 960                                 mip->ra_offset = 0;
 961                                 mip->ra_index++;
 962                         }
 963                 }
 964         }
 965
 966 out:
 967         *bpp = bp;
 968         return error;
 969 }
 970
 971 /*
 972  * Getdents (readdir) for leaf and node directories.
 973  * This reads the data blocks only, so is the same for both forms.
 974  */
 975 int                                             /* error */
 976 xfs_dir2_leaf_getdents(
 977         xfs_inode_t             *dp,            /* incore directory inode */
 978         void                    *dirent,
 979         size_t                  bufsize,
 980         xfs_off_t               *offset,
 981         filldir_t               filldir)
 982 {
 983         struct xfs_buf          *bp = NULL;     /* data block buffer */
 984         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 985         xfs_dir2_data_entry_t   *dep;           /* data entry */
 986         xfs_dir2_data_unused_t  *dup;           /* unused entry */
 987         int                     error = 0;      /* error return value */
 988         int                     length;         /* temporary length value */
 989         xfs_mount_t             *mp;            /* filesystem mount point */
 990         int                     byteoff;        /* offset in current block */
 991         xfs_dir2_off_t          curoff;         /* current overall offset */
 992         xfs_dir2_off_t          newoff;         /* new curoff after new blk */
 993         char                    *ptr = NULL;    /* pointer to current data */
 994         struct xfs_dir2_leaf_map_info *map_info;
 995
 996         /*
 997          * If the offset is at or past the largest allowed value,
 998          * give up right away.
 999          */
1000         if (*offset >= XFS_DIR2_MAX_DATAPTR)
1001                 return 0;
1002
1003         mp = dp->i_mount;
1004
1005         /*
1006          * Set up to bmap a number of blocks based on the caller's
1007          * buffer size, the directory block size, and the filesystem
1008          * block size.
1009          */
1010         length = howmany(bufsize + mp->m_dirblksize,
1011                                      mp->m_sb.sb_blocksize);
1012         map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
1013                                 (length * sizeof(struct xfs_bmbt_irec)),
1014                                KM_SLEEP);
1015         map_info->map_size = length;
1016
1017         /*
1018          * Inside the loop we keep the main offset value as a byte offset
1019          * in the directory file.
1020          */
1021         curoff = xfs_dir2_dataptr_to_byte(mp, *offset);
1022
1023         /*
1024          * Force this conversion through db so we truncate the offset
1025          * down to get the start of the data block.
1026          */
1027         map_info->map_off = xfs_dir2_db_to_da(mp,
1028                                               xfs_dir2_byte_to_db(mp, curoff));
1029
1030         /*
1031          * Loop over directory entries until we reach the end offset.
1032          * Get more blocks and readahead as necessary.
1033          */
1034         while (curoff < XFS_DIR2_LEAF_OFFSET) {
1035                 /*
1036                  * If we have no buffer, or we're off the end of the
1037                  * current buffer, need to get another one.
1038                  */
1039                 if (!bp || ptr >= (char *)bp->b_addr + mp->m_dirblksize) {
1040
1041                         error = xfs_dir2_leaf_readbuf(dp, bufsize, map_info,
1042                                                       &curoff, &bp);
1043                         if (error || !map_info->map_valid)
1044                                 break;
1045
1046                         /*
1047                          * Having done a read, we need to set a new offset.
1048                          */
1049                         newoff = xfs_dir2_db_off_to_byte(mp, map_info->curdb, 0);
1050                         /*
1051                          * Start of the current block.
1052                          */
1053                         if (curoff < newoff)
1054                                 curoff = newoff;
1055                         /*
1056                          * Make sure we're in the right block.
1057                          */
1058                         else if (curoff > newoff)
1059                                 ASSERT(xfs_dir2_byte_to_db(mp, curoff) ==
1060                                        map_info->curdb);
1061                         hdr = bp->b_addr;
1062                         xfs_dir2_data_check(dp, bp);
1063                         /*
1064                          * Find our position in the block.
1065                          */
1066                         ptr = (char *)(hdr + 1);
1067                         byteoff = xfs_dir2_byte_to_off(mp, curoff);
1068                         /*
1069                          * Skip past the header.
1070                          */
1071                         if (byteoff == 0)
1072                                 curoff += (uint)sizeof(*hdr);
1073                         /*
1074                          * Skip past entries until we reach our offset.
1075                          */
1076                         else {
1077                                 while ((char *)ptr - (char *)hdr < byteoff) {
1078                                         dup = (xfs_dir2_data_unused_t *)ptr;
1079
1080                                         if (be16_to_cpu(dup->freetag)
1081                                                   == XFS_DIR2_DATA_FREE_TAG) {
1082
1083                                                 length = be16_to_cpu(dup->length);
1084                                                 ptr += length;
1085                                                 continue;
1086                                         }
1087                                         dep = (xfs_dir2_data_entry_t *)ptr;
1088                                         length =
1089                                            xfs_dir2_data_entsize(dep->namelen);
1090                                         ptr += length;
1091                                 }
1092                                 /*
1093                                  * Now set our real offset.
1094                                  */
1095                                 curoff =
1096                                         xfs_dir2_db_off_to_byte(mp,
1097                                             xfs_dir2_byte_to_db(mp, curoff),
1098                                             (char *)ptr - (char *)hdr);
1099                                 if (ptr >= (char *)hdr + mp->m_dirblksize) {
1100                                         continue;
1101                                 }
1102                         }
1103                 }
1104                 /*
1105                  * We have a pointer to an entry.
1106                  * Is it a live one?
1107                  */
1108                 dup = (xfs_dir2_data_unused_t *)ptr;
1109                 /*
1110                  * No, it's unused, skip over it.
1111                  */
1112                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
1113                         length = be16_to_cpu(dup->length);
1114                         ptr += length;
1115                         curoff += length;
1116                         continue;
1117                 }
1118
1119                 dep = (xfs_dir2_data_entry_t *)ptr;
1120                 length = xfs_dir2_data_entsize(dep->namelen);
1121
1122                 if (filldir(dirent, (char *)dep->name, dep->namelen,
1123                             xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff,
1124                             be64_to_cpu(dep->inumber), DT_UNKNOWN))
1125                         break;
1126
1127                 /*
1128                  * Advance to next entry in the block.
1129                  */
1130                 ptr += length;
1131                 curoff += length;
1132                 /* bufsize may have just been a guess; don't go negative */
1133                 bufsize = bufsize > length ? bufsize - length : 0;
1134         }
1135
1136         /*
1137          * All done.  Set output offset value to current offset.
1138          */
1139         if (curoff > xfs_dir2_dataptr_to_byte(mp, XFS_DIR2_MAX_DATAPTR))
1140                 *offset = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
1141         else
1142                 *offset = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
1143         kmem_free(map_info);
1144         if (bp)
1145                 xfs_trans_brelse(NULL, bp);
1146         return error;
1147 }
1148
1149 /*
1150  * Initialize a new leaf block, leaf1 or leafn magic accepted.
1151  */
1152 int
1153 xfs_dir2_leaf_init(
1154         xfs_da_args_t           *args,          /* operation arguments */
1155         xfs_dir2_db_t           bno,            /* directory block number */
1156         struct xfs_buf          **bpp,          /* out: leaf buffer */
1157         int                     magic)          /* magic number for block */
1158 {
1159         struct xfs_buf          *bp;            /* leaf buffer */
1160         xfs_inode_t             *dp;            /* incore directory inode */
1161         int                     error;          /* error return code */
1162         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1163         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1164         xfs_mount_t             *mp;            /* filesystem mount point */
1165         xfs_trans_t             *tp;            /* transaction pointer */
1166
1167         dp = args->dp;
1168         ASSERT(dp != NULL);
1169         tp = args->trans;
1170         mp = dp->i_mount;
1171         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
1172                bno < XFS_DIR2_FREE_FIRSTDB(mp));
1173         /*
1174          * Get the buffer for the block.
1175          */
1176         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
1177                 XFS_DATA_FORK);
1178         if (error) {
1179                 return error;
1180         }
1181         ASSERT(bp != NULL);
1182         leaf = bp->b_addr;
1183         /*
1184          * Initialize the header.
1185          */
1186         leaf->hdr.info.magic = cpu_to_be16(magic);
1187         leaf->hdr.info.forw = 0;
1188         leaf->hdr.info.back = 0;
1189         leaf->hdr.count = 0;
1190         leaf->hdr.stale = 0;
1191         xfs_dir2_leaf_log_header(tp, bp);
1192         /*
1193          * If it's a leaf-format directory initialize the tail.
1194          * In this case our caller has the real bests table to copy into
1195          * the block.
1196          */
1197         if (magic == XFS_DIR2_LEAF1_MAGIC) {
1198                 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1199                 ltp->bestcount = 0;
1200                 xfs_dir2_leaf_log_tail(tp, bp);
1201         }
1202         *bpp = bp;
1203         return 0;
1204 }
1205
1206 /*
1207  * Log the bests entries indicated from a leaf1 block.
1208  */
1209 static void
1210 xfs_dir2_leaf_log_bests(
1211         xfs_trans_t             *tp,            /* transaction pointer */
1212         struct xfs_buf          *bp,            /* leaf buffer */
1213         int                     first,          /* first entry to log */
1214         int                     last)           /* last entry to log */
1215 {
1216         __be16                  *firstb;        /* pointer to first entry */
1217         __be16                  *lastb;         /* pointer to last entry */
1218         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1219         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1220
1221         leaf = bp->b_addr;
1222         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1223         ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1224         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1225         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1226         xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1227                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1228 }
1229
1230 /*
1231  * Log the leaf entries indicated from a leaf1 or leafn block.
1232  */
1233 void
1234 xfs_dir2_leaf_log_ents(
1235         xfs_trans_t             *tp,            /* transaction pointer */
1236         struct xfs_buf          *bp,            /* leaf buffer */
1237         int                     first,          /* first entry to log */
1238         int                     last)           /* last entry to log */
1239 {
1240         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1241         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1242         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1243
1244         leaf = bp->b_addr;
1245         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1246                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1247         firstlep = &leaf->ents[first];
1248         lastlep = &leaf->ents[last];
1249         xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1250                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1251 }
1252
1253 /*
1254  * Log the header of the leaf1 or leafn block.
1255  */
1256 void
1257 xfs_dir2_leaf_log_header(
1258         struct xfs_trans        *tp,
1259         struct xfs_buf          *bp)
1260 {
1261         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1262
1263         leaf = bp->b_addr;
1264         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1265                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1266         xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1267                 (uint)(sizeof(leaf->hdr) - 1));
1268 }
1269
1270 /*
1271  * Log the tail of the leaf1 block.
1272  */
1273 STATIC void
1274 xfs_dir2_leaf_log_tail(
1275         struct xfs_trans        *tp,
1276         struct xfs_buf          *bp)
1277 {
1278         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1279         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1280         xfs_mount_t             *mp;            /* filesystem mount point */
1281
1282         mp = tp->t_mountp;
1283         leaf = bp->b_addr;
1284         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1285         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1286         xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1287                 (uint)(mp->m_dirblksize - 1));
1288 }
1289
1290 /*
1291  * Look up the entry referred to by args in the leaf format directory.
1292  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1293  * is also used by the node-format code.
1294  */
1295 int
1296 xfs_dir2_leaf_lookup(
1297         xfs_da_args_t           *args)          /* operation arguments */
1298 {
1299         struct xfs_buf          *dbp;           /* data block buffer */
1300         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1301         xfs_inode_t             *dp;            /* incore directory inode */
1302         int                     error;          /* error return code */
1303         int                     index;          /* found entry index */
1304         struct xfs_buf          *lbp;           /* leaf buffer */
1305         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1306         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1307         xfs_trans_t             *tp;            /* transaction pointer */
1308
1309         trace_xfs_dir2_leaf_lookup(args);
1310
1311         /*
1312          * Look up name in the leaf block, returning both buffers and index.
1313          */
1314         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1315                 return error;
1316         }
1317         tp = args->trans;
1318         dp = args->dp;
1319         xfs_dir2_leaf_check(dp, lbp);
1320         leaf = lbp->b_addr;
1321         /*
1322          * Get to the leaf entry and contained data entry address.
1323          */
1324         lep = &leaf->ents[index];
1325         /*
1326          * Point to the data entry.
1327          */
1328         dep = (xfs_dir2_data_entry_t *)
1329               ((char *)dbp->b_addr +
1330                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1331         /*
1332          * Return the found inode number & CI name if appropriate
1333          */
1334         args->inumber = be64_to_cpu(dep->inumber);
1335         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1336         xfs_trans_brelse(tp, dbp);
1337         xfs_trans_brelse(tp, lbp);
1338         return XFS_ERROR(error);
1339 }
1340
1341 /*
1342  * Look up name/hash in the leaf block.
1343  * Fill in indexp with the found index, and dbpp with the data buffer.
1344  * If not found dbpp will be NULL, and ENOENT comes back.
1345  * lbpp will always be filled in with the leaf buffer unless there's an error.
1346  */
1347 static int                                      /* error */
1348 xfs_dir2_leaf_lookup_int(
1349         xfs_da_args_t           *args,          /* operation arguments */
1350         struct xfs_buf          **lbpp,         /* out: leaf buffer */
1351         int                     *indexp,        /* out: index in leaf block */
1352         struct xfs_buf          **dbpp)         /* out: data buffer */
1353 {
1354         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1355         struct xfs_buf          *dbp = NULL;    /* data buffer */
1356         xfs_dir2_data_entry_t   *dep;           /* data entry */
1357         xfs_inode_t             *dp;            /* incore directory inode */
1358         int                     error;          /* error return code */
1359         int                     index;          /* index in leaf block */
1360         struct xfs_buf          *lbp;           /* leaf buffer */
1361         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1362         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1363         xfs_mount_t             *mp;            /* filesystem mount point */
1364         xfs_dir2_db_t           newdb;          /* new data block number */
1365         xfs_trans_t             *tp;            /* transaction pointer */
1366         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1367         enum xfs_dacmp          cmp;            /* name compare result */
1368
1369         dp = args->dp;
1370         tp = args->trans;
1371         mp = dp->i_mount;
1372         /*
1373          * Read the leaf block into the buffer.
1374          */
1375         error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
1376                                                         XFS_DATA_FORK, NULL);
1377         if (error)
1378                 return error;
1379         *lbpp = lbp;
1380         leaf = lbp->b_addr;
1381         xfs_dir2_leaf_check(dp, lbp);
1382         /*
1383          * Look for the first leaf entry with our hash value.
1384          */
1385         index = xfs_dir2_leaf_search_hash(args, lbp);
1386         /*
1387          * Loop over all the entries with the right hash value
1388          * looking to match the name.
1389          */
1390         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
1391                                 be32_to_cpu(lep->hashval) == args->hashval;
1392                                 lep++, index++) {
1393                 /*
1394                  * Skip over stale leaf entries.
1395                  */
1396                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1397                         continue;
1398                 /*
1399                  * Get the new data block number.
1400                  */
1401                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1402                 /*
1403                  * If it's not the same as the old data block number,
1404                  * need to pitch the old one and read the new one.
1405                  */
1406                 if (newdb != curdb) {
1407                         if (dbp)
1408                                 xfs_trans_brelse(tp, dbp);
1409                         error = xfs_dir2_data_read(tp, dp,
1410                                                    xfs_dir2_db_to_da(mp, newdb),
1411                                                    -1, &dbp);
1412                         if (error) {
1413                                 xfs_trans_brelse(tp, lbp);
1414                                 return error;
1415                         }
1416                         curdb = newdb;
1417                 }
1418                 /*
1419                  * Point to the data entry.
1420                  */
1421                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1422                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1423                 /*
1424                  * Compare name and if it's an exact match, return the index
1425                  * and buffer. If it's the first case-insensitive match, store
1426                  * the index and buffer and continue looking for an exact match.
1427                  */
1428                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1429                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1430                         args->cmpresult = cmp;
1431                         *indexp = index;
1432                         /* case exact match: return the current buffer. */
1433                         if (cmp == XFS_CMP_EXACT) {
1434                                 *dbpp = dbp;
1435                                 return 0;
1436                         }
1437                         cidb = curdb;
1438                 }
1439         }
1440         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1441         /*
1442          * Here, we can only be doing a lookup (not a rename or remove).
1443          * If a case-insensitive match was found earlier, re-read the
1444          * appropriate data block if required and return it.
1445          */
1446         if (args->cmpresult == XFS_CMP_CASE) {
1447                 ASSERT(cidb != -1);
1448                 if (cidb != curdb) {
1449                         xfs_trans_brelse(tp, dbp);
1450                         error = xfs_dir2_data_read(tp, dp,
1451                                                    xfs_dir2_db_to_da(mp, cidb),
1452                                                    -1, &dbp);
1453                         if (error) {
1454                                 xfs_trans_brelse(tp, lbp);
1455                                 return error;
1456                         }
1457                 }
1458                 *dbpp = dbp;
1459                 return 0;
1460         }
1461         /*
1462          * No match found, return ENOENT.
1463          */
1464         ASSERT(cidb == -1);
1465         if (dbp)
1466                 xfs_trans_brelse(tp, dbp);
1467         xfs_trans_brelse(tp, lbp);
1468         return XFS_ERROR(ENOENT);
1469 }
1470
1471 /*
1472  * Remove an entry from a leaf format directory.
1473  */
1474 int                                             /* error */
1475 xfs_dir2_leaf_removename(
1476         xfs_da_args_t           *args)          /* operation arguments */
1477 {
1478         __be16                  *bestsp;        /* leaf block best freespace */
1479         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1480         xfs_dir2_db_t           db;             /* data block number */
1481         struct xfs_buf          *dbp;           /* data block buffer */
1482         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1483         xfs_inode_t             *dp;            /* incore directory inode */
1484         int                     error;          /* error return code */
1485         xfs_dir2_db_t           i;              /* temporary data block # */
1486         int                     index;          /* index into leaf entries */
1487         struct xfs_buf          *lbp;           /* leaf buffer */
1488         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1489         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1490         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1491         xfs_mount_t             *mp;            /* filesystem mount point */
1492         int                     needlog;        /* need to log data header */
1493         int                     needscan;       /* need to rescan data frees */
1494         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1495         xfs_trans_t             *tp;            /* transaction pointer */
1496
1497         trace_xfs_dir2_leaf_removename(args);
1498
1499         /*
1500          * Lookup the leaf entry, get the leaf and data blocks read in.
1501          */
1502         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1503                 return error;
1504         }
1505         dp = args->dp;
1506         tp = args->trans;
1507         mp = dp->i_mount;
1508         leaf = lbp->b_addr;
1509         hdr = dbp->b_addr;
1510         xfs_dir2_data_check(dp, dbp);
1511         /*
1512          * Point to the leaf entry, use that to point to the data entry.
1513          */
1514         lep = &leaf->ents[index];
1515         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1516         dep = (xfs_dir2_data_entry_t *)
1517               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1518         needscan = needlog = 0;
1519         oldbest = be16_to_cpu(hdr->bestfree[0].length);
1520         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1521         bestsp = xfs_dir2_leaf_bests_p(ltp);
1522         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1523         /*
1524          * Mark the former data entry unused.
1525          */
1526         xfs_dir2_data_make_free(tp, dbp,
1527                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1528                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1529         /*
1530          * We just mark the leaf entry stale by putting a null in it.
1531          */
1532         be16_add_cpu(&leaf->hdr.stale, 1);
1533         xfs_dir2_leaf_log_header(tp, lbp);
1534         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1535         xfs_dir2_leaf_log_ents(tp, lbp, index, index);
1536         /*
1537          * Scan the freespace in the data block again if necessary,
1538          * log the data block header if necessary.
1539          */
1540         if (needscan)
1541                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1542         if (needlog)
1543                 xfs_dir2_data_log_header(tp, dbp);
1544         /*
1545          * If the longest freespace in the data block has changed,
1546          * put the new value in the bests table and log that.
1547          */
1548         if (be16_to_cpu(hdr->bestfree[0].length) != oldbest) {
1549                 bestsp[db] = hdr->bestfree[0].length;
1550                 xfs_dir2_leaf_log_bests(tp, lbp, db, db);
1551         }
1552         xfs_dir2_data_check(dp, dbp);
1553         /*
1554          * If the data block is now empty then get rid of the data block.
1555          */
1556         if (be16_to_cpu(hdr->bestfree[0].length) ==
1557             mp->m_dirblksize - (uint)sizeof(*hdr)) {
1558                 ASSERT(db != mp->m_dirdatablk);
1559                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1560                         /*
1561                          * Nope, can't get rid of it because it caused
1562                          * allocation of a bmap btree block to do so.
1563                          * Just go on, returning success, leaving the
1564                          * empty block in place.
1565                          */
1566                         if (error == ENOSPC && args->total == 0)
1567                                 error = 0;
1568                         xfs_dir2_leaf_check(dp, lbp);
1569                         return error;
1570                 }
1571                 dbp = NULL;
1572                 /*
1573                  * If this is the last data block then compact the
1574                  * bests table by getting rid of entries.
1575                  */
1576                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1577                         /*
1578                          * Look for the last active entry (i).
1579                          */
1580                         for (i = db - 1; i > 0; i--) {
1581                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1582                                         break;
1583                         }
1584                         /*
1585                          * Copy the table down so inactive entries at the
1586                          * end are removed.
1587                          */
1588                         memmove(&bestsp[db - i], bestsp,
1589                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1590                         be32_add_cpu(&ltp->bestcount, -(db - i));
1591                         xfs_dir2_leaf_log_tail(tp, lbp);
1592                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1593                 } else
1594                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1595         }
1596         /*
1597          * If the data block was not the first one, drop it.
1598          */
1599         else if (db != mp->m_dirdatablk)
1600                 dbp = NULL;
1601
1602         xfs_dir2_leaf_check(dp, lbp);
1603         /*
1604          * See if we can convert to block form.
1605          */
1606         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1607 }
1608
1609 /*
1610  * Replace the inode number in a leaf format directory entry.
1611  */
1612 int                                             /* error */
1613 xfs_dir2_leaf_replace(
1614         xfs_da_args_t           *args)          /* operation arguments */
1615 {
1616         struct xfs_buf          *dbp;           /* data block buffer */
1617         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1618         xfs_inode_t             *dp;            /* incore directory inode */
1619         int                     error;          /* error return code */
1620         int                     index;          /* index of leaf entry */
1621         struct xfs_buf          *lbp;           /* leaf buffer */
1622         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1623         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1624         xfs_trans_t             *tp;            /* transaction pointer */
1625
1626         trace_xfs_dir2_leaf_replace(args);
1627
1628         /*
1629          * Look up the entry.
1630          */
1631         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1632                 return error;
1633         }
1634         dp = args->dp;
1635         leaf = lbp->b_addr;
1636         /*
1637          * Point to the leaf entry, get data address from it.
1638          */
1639         lep = &leaf->ents[index];
1640         /*
1641          * Point to the data entry.
1642          */
1643         dep = (xfs_dir2_data_entry_t *)
1644               ((char *)dbp->b_addr +
1645                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1646         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1647         /*
1648          * Put the new inode number in, log it.
1649          */
1650         dep->inumber = cpu_to_be64(args->inumber);
1651         tp = args->trans;
1652         xfs_dir2_data_log_entry(tp, dbp, dep);
1653         xfs_dir2_leaf_check(dp, lbp);
1654         xfs_trans_brelse(tp, lbp);
1655         return 0;
1656 }
1657
1658 /*
1659  * Return index in the leaf block (lbp) which is either the first
1660  * one with this hash value, or if there are none, the insert point
1661  * for that hash value.
1662  */
1663 int                                             /* index value */
1664 xfs_dir2_leaf_search_hash(
1665         xfs_da_args_t           *args,          /* operation arguments */
1666         struct xfs_buf          *lbp)           /* leaf buffer */
1667 {
1668         xfs_dahash_t            hash=0;         /* hash from this entry */
1669         xfs_dahash_t            hashwant;       /* hash value looking for */
1670         int                     high;           /* high leaf index */
1671         int                     low;            /* low leaf index */
1672         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1673         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1674         int                     mid=0;          /* current leaf index */
1675
1676         leaf = lbp->b_addr;
1677 #ifndef __KERNEL__
1678         if (!leaf->hdr.count)
1679                 return 0;
1680 #endif
1681         /*
1682          * Note, the table cannot be empty, so we have to go through the loop.
1683          * Binary search the leaf entries looking for our hash value.
1684          */
1685         for (lep = leaf->ents, low = 0, high = be16_to_cpu(leaf->hdr.count) - 1,
1686                 hashwant = args->hashval;
1687              low <= high; ) {
1688                 mid = (low + high) >> 1;
1689                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1690                         break;
1691                 if (hash < hashwant)
1692                         low = mid + 1;
1693                 else
1694                         high = mid - 1;
1695         }
1696         /*
1697          * Found one, back up through all the equal hash values.
1698          */
1699         if (hash == hashwant) {
1700                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1701                         mid--;
1702                 }
1703         }
1704         /*
1705          * Need to point to an entry higher than ours.
1706          */
1707         else if (hash < hashwant)
1708                 mid++;
1709         return mid;
1710 }
1711
1712 /*
1713  * Trim off a trailing data block.  We know it's empty since the leaf
1714  * freespace table says so.
1715  */
1716 int                                             /* error */
1717 xfs_dir2_leaf_trim_data(
1718         xfs_da_args_t           *args,          /* operation arguments */
1719         struct xfs_buf          *lbp,           /* leaf buffer */
1720         xfs_dir2_db_t           db)             /* data block number */
1721 {
1722         __be16                  *bestsp;        /* leaf bests table */
1723         struct xfs_buf          *dbp;           /* data block buffer */
1724         xfs_inode_t             *dp;            /* incore directory inode */
1725         int                     error;          /* error return value */
1726         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1727         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1728         xfs_mount_t             *mp;            /* filesystem mount point */
1729         xfs_trans_t             *tp;            /* transaction pointer */
1730
1731         dp = args->dp;
1732         mp = dp->i_mount;
1733         tp = args->trans;
1734         /*
1735          * Read the offending data block.  We need its buffer.
1736          */
1737         error = xfs_dir2_data_read(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp);
1738         if (error)
1739                 return error;
1740
1741         leaf = lbp->b_addr;
1742         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1743
1744 #ifdef DEBUG
1745 {
1746         struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1747
1748         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
1749         ASSERT(be16_to_cpu(hdr->bestfree[0].length) ==
1750                mp->m_dirblksize - (uint)sizeof(*hdr));
1751         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1752 }
1753 #endif
1754
1755         /*
1756          * Get rid of the data block.
1757          */
1758         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1759                 ASSERT(error != ENOSPC);
1760                 xfs_trans_brelse(tp, dbp);
1761                 return error;
1762         }
1763         /*
1764          * Eliminate the last bests entry from the table.
1765          */
1766         bestsp = xfs_dir2_leaf_bests_p(ltp);
1767         be32_add_cpu(&ltp->bestcount, -1);
1768         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1769         xfs_dir2_leaf_log_tail(tp, lbp);
1770         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1771         return 0;
1772 }
1773
1774 static inline size_t
1775 xfs_dir2_leaf_size(
1776         struct xfs_dir2_leaf_hdr        *hdr,
1777         int                             counts)
1778 {
1779         int                     entries;
1780
1781         entries = be16_to_cpu(hdr->count) - be16_to_cpu(hdr->stale);
1782         return sizeof(xfs_dir2_leaf_hdr_t) +
1783             entries * sizeof(xfs_dir2_leaf_entry_t) +
1784             counts * sizeof(xfs_dir2_data_off_t) +
1785             sizeof(xfs_dir2_leaf_tail_t);
1786 }
1787
1788 /*
1789  * Convert node form directory to leaf form directory.
1790  * The root of the node form dir needs to already be a LEAFN block.
1791  * Just return if we can't do anything.
1792  */
1793 int                                             /* error */
1794 xfs_dir2_node_to_leaf(
1795         xfs_da_state_t          *state)         /* directory operation state */
1796 {
1797         xfs_da_args_t           *args;          /* operation arguments */
1798         xfs_inode_t             *dp;            /* incore directory inode */
1799         int                     error;          /* error return code */
1800         struct xfs_buf          *fbp;           /* buffer for freespace block */
1801         xfs_fileoff_t           fo;             /* freespace file offset */
1802         xfs_dir2_free_t         *free;          /* freespace structure */
1803         struct xfs_buf          *lbp;           /* buffer for leaf block */
1804         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1805         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1806         xfs_mount_t             *mp;            /* filesystem mount point */
1807         int                     rval;           /* successful free trim? */
1808         xfs_trans_t             *tp;            /* transaction pointer */
1809
1810         /*
1811          * There's more than a leaf level in the btree, so there must
1812          * be multiple leafn blocks.  Give up.
1813          */
1814         if (state->path.active > 1)
1815                 return 0;
1816         args = state->args;
1817
1818         trace_xfs_dir2_node_to_leaf(args);
1819
1820         mp = state->mp;
1821         dp = args->dp;
1822         tp = args->trans;
1823         /*
1824          * Get the last offset in the file.
1825          */
1826         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1827                 return error;
1828         }
1829         fo -= mp->m_dirblkfsbs;
1830         /*
1831          * If there are freespace blocks other than the first one,
1832          * take this opportunity to remove trailing empty freespace blocks
1833          * that may have been left behind during no-space-reservation
1834          * operations.
1835          */
1836         while (fo > mp->m_dirfreeblk) {
1837                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1838                         return error;
1839                 }
1840                 if (rval)
1841                         fo -= mp->m_dirblkfsbs;
1842                 else
1843                         return 0;
1844         }
1845         /*
1846          * Now find the block just before the freespace block.
1847          */
1848         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1849                 return error;
1850         }
1851         /*
1852          * If it's not the single leaf block, give up.
1853          */
1854         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1855                 return 0;
1856         lbp = state->path.blk[0].bp;
1857         leaf = lbp->b_addr;
1858         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1859         /*
1860          * Read the freespace block.
1861          */
1862         error = xfs_dir2_free_read(tp, dp,  mp->m_dirfreeblk, &fbp);
1863         if (error)
1864                 return error;
1865         free = fbp->b_addr;
1866         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1867         ASSERT(!free->hdr.firstdb);
1868
1869         /*
1870          * Now see if the leafn and free data will fit in a leaf1.
1871          * If not, release the buffer and give up.
1872          */
1873         if (xfs_dir2_leaf_size(&leaf->hdr, be32_to_cpu(free->hdr.nvalid)) >
1874                         mp->m_dirblksize) {
1875                 xfs_trans_brelse(tp, fbp);
1876                 return 0;
1877         }
1878
1879         /*
1880          * If the leaf has any stale entries in it, compress them out.
1881          * The compact routine will log the header.
1882          */
1883         if (be16_to_cpu(leaf->hdr.stale))
1884                 xfs_dir2_leaf_compact(args, lbp);
1885         else
1886                 xfs_dir2_leaf_log_header(tp, lbp);
1887         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAF1_MAGIC);
1888         /*
1889          * Set up the leaf tail from the freespace block.
1890          */
1891         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1892         ltp->bestcount = free->hdr.nvalid;
1893         /*
1894          * Set up the leaf bests table.
1895          */
1896         memcpy(xfs_dir2_leaf_bests_p(ltp), free->bests,
1897                 be32_to_cpu(ltp->bestcount) * sizeof(xfs_dir2_data_off_t));
1898         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1899         xfs_dir2_leaf_log_tail(tp, lbp);
1900         xfs_dir2_leaf_check(dp, lbp);
1901         /*
1902          * Get rid of the freespace block.
1903          */
1904         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1905         if (error) {
1906                 /*
1907                  * This can't fail here because it can only happen when
1908                  * punching out the middle of an extent, and this is an
1909                  * isolated block.
1910                  */
1911                 ASSERT(error != ENOSPC);
1912                 return error;
1913         }
1914         fbp = NULL;
1915         /*
1916          * Now see if we can convert the single-leaf directory
1917          * down to a block form directory.
1918          * This routine always kills the dabuf for the leaf, so
1919          * eliminate it from the path.
1920          */
1921         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1922         state->path.blk[0].bp = NULL;
1923         return error;
1924 }