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 static void
  52 xfs_dir2_leaf_verify(
  53         struct xfs_buf          *bp,
  54         __be16                  magic)
  55 {
  56         struct xfs_mount        *mp = bp->b_target->bt_mount;
  57         struct xfs_dir2_leaf_hdr *hdr = bp->b_addr;
  58         int                     block_ok = 0;
  59
  60         block_ok = hdr->info.magic == magic;
  61         if (!block_ok) {
  62                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, hdr);
  63                 xfs_buf_ioerror(bp, EFSCORRUPTED);
  64         }
  65 }
  66
  67 static void
  68 xfs_dir2_leaf1_read_verify(
  69         struct xfs_buf  *bp)
  70 {
  71         xfs_dir2_leaf_verify(bp, cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
  72 }
  73
  74 static void
  75 xfs_dir2_leaf1_write_verify(
  76         struct xfs_buf  *bp)
  77 {
  78         xfs_dir2_leaf_verify(bp, cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
  79 }
  80
  81 void
  82 xfs_dir2_leafn_read_verify(
  83         struct xfs_buf  *bp)
  84 {
  85         xfs_dir2_leaf_verify(bp, cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
  86 }
  87
  88 void
  89 xfs_dir2_leafn_write_verify(
  90         struct xfs_buf  *bp)
  91 {
  92         xfs_dir2_leaf_verify(bp, cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
  93 }
  94
  95 static const struct xfs_buf_ops xfs_dir2_leaf1_buf_ops = {
  96         .verify_read = xfs_dir2_leaf1_read_verify,
  97         .verify_write = xfs_dir2_leaf1_write_verify,
  98 };
  99
 100 const struct xfs_buf_ops xfs_dir2_leafn_buf_ops = {
 101         .verify_read = xfs_dir2_leafn_read_verify,
 102         .verify_write = xfs_dir2_leafn_write_verify,
 103 };
 104
 105 static int
 106 xfs_dir2_leaf_read(
 107         struct xfs_trans        *tp,
 108         struct xfs_inode        *dp,
 109         xfs_dablk_t             fbno,
 110         xfs_daddr_t             mappedbno,
 111         struct xfs_buf          **bpp)
 112 {
 113         return xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 114                                 XFS_DATA_FORK, &xfs_dir2_leaf1_buf_ops);
 115 }
 116
 117 int
 118 xfs_dir2_leafn_read(
 119         struct xfs_trans        *tp,
 120         struct xfs_inode        *dp,
 121         xfs_dablk_t             fbno,
 122         xfs_daddr_t             mappedbno,
 123         struct xfs_buf          **bpp)
 124 {
 125         return xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 126                                 XFS_DATA_FORK, &xfs_dir2_leafn_buf_ops);
 127 }
 128
 129 /*
 130  * Convert a block form directory to a leaf form directory.
 131  */
 132 int                                             /* error */
 133 xfs_dir2_block_to_leaf(
 134         xfs_da_args_t           *args,          /* operation arguments */
 135         struct xfs_buf          *dbp)           /* input block's buffer */
 136 {
 137         __be16                  *bestsp;        /* leaf's bestsp entries */
 138         xfs_dablk_t             blkno;          /* leaf block's bno */
 139         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 140         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
 141         xfs_dir2_block_tail_t   *btp;           /* block's tail */
 142         xfs_inode_t             *dp;            /* incore directory inode */
 143         int                     error;          /* error return code */
 144         struct xfs_buf          *lbp;           /* leaf block's buffer */
 145         xfs_dir2_db_t           ldb;            /* leaf block's bno */
 146         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 147         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
 148         xfs_mount_t             *mp;            /* filesystem mount point */
 149         int                     needlog;        /* need to log block header */
 150         int                     needscan;       /* need to rescan bestfree */
 151         xfs_trans_t             *tp;            /* transaction pointer */
 152         struct xfs_dir2_data_free       *bf;
 153
 154         trace_xfs_dir2_block_to_leaf(args);
 155
 156         dp = args->dp;
 157         mp = dp->i_mount;
 158         tp = args->trans;
 159         /*
 160          * Add the leaf block to the inode.
 161          * This interface will only put blocks in the leaf/node range.
 162          * Since that's empty now, we'll get the root (block 0 in range).
 163          */
 164         if ((error = xfs_da_grow_inode(args, &blkno))) {
 165                 return error;
 166         }
 167         ldb = xfs_dir2_da_to_db(mp, blkno);
 168         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
 169         /*
 170          * Initialize the leaf block, get a buffer for it.
 171          */
 172         if ((error = xfs_dir2_leaf_init(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC))) {
 173                 return error;
 174         }
 175         ASSERT(lbp != NULL);
 176         leaf = lbp->b_addr;
 177         hdr = dbp->b_addr;
 178         xfs_dir2_data_check(dp, dbp);
 179         btp = xfs_dir2_block_tail_p(mp, hdr);
 180         blp = xfs_dir2_block_leaf_p(btp);
 181         bf = xfs_dir3_data_bestfree_p(hdr);
 182         /*
 183          * Set the counts in the leaf header.
 184          */
 185         leaf->hdr.count = cpu_to_be16(be32_to_cpu(btp->count));
 186         leaf->hdr.stale = cpu_to_be16(be32_to_cpu(btp->stale));
 187         /*
 188          * Could compact these but I think we always do the conversion
 189          * after squeezing out stale entries.
 190          */
 191         memcpy(leaf->ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
 192         xfs_dir2_leaf_log_ents(tp, lbp, 0, be16_to_cpu(leaf->hdr.count) - 1);
 193         needscan = 0;
 194         needlog = 1;
 195         /*
 196          * Make the space formerly occupied by the leaf entries and block
 197          * tail be free.
 198          */
 199         xfs_dir2_data_make_free(tp, dbp,
 200                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 201                 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
 202                                        (char *)blp),
 203                 &needlog, &needscan);
 204         /*
 205          * Fix up the block header, make it a data block.
 206          */
 207         dbp->b_ops = &xfs_dir2_data_buf_ops;
 208         hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 209         if (needscan)
 210                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 211         /*
 212          * Set up leaf tail and bests table.
 213          */
 214         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 215         ltp->bestcount = cpu_to_be32(1);
 216         bestsp = xfs_dir2_leaf_bests_p(ltp);
 217         bestsp[0] =  bf[0].length;
 218         /*
 219          * Log the data header and leaf bests table.
 220          */
 221         if (needlog)
 222                 xfs_dir2_data_log_header(tp, dbp);
 223         xfs_dir2_leaf_check(dp, lbp);
 224         xfs_dir2_data_check(dp, dbp);
 225         xfs_dir2_leaf_log_bests(tp, lbp, 0, 0);
 226         return 0;
 227 }
 228
 229 STATIC void
 230 xfs_dir2_leaf_find_stale(
 231         struct xfs_dir2_leaf    *leaf,
 232         int                     index,
 233         int                     *lowstale,
 234         int                     *highstale)
 235 {
 236         /*
 237          * Find the first stale entry before our index, if any.
 238          */
 239         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
 240                 if (leaf->ents[*lowstale].address ==
 241                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 242                         break;
 243         }
 244
 245         /*
 246          * Find the first stale entry at or after our index, if any.
 247          * Stop if the result would require moving more entries than using
 248          * lowstale.
 249          */
 250         for (*highstale = index;
 251              *highstale < be16_to_cpu(leaf->hdr.count);
 252              ++*highstale) {
 253                 if (leaf->ents[*highstale].address ==
 254                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 255                         break;
 256                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
 257                         break;
 258         }
 259 }
 260
 261 struct xfs_dir2_leaf_entry *
 262 xfs_dir2_leaf_find_entry(
 263         xfs_dir2_leaf_t         *leaf,          /* leaf structure */
 264         int                     index,          /* leaf table position */
 265         int                     compact,        /* need to compact leaves */
 266         int                     lowstale,       /* index of prev stale leaf */
 267         int                     highstale,      /* index of next stale leaf */
 268         int                     *lfloglow,      /* low leaf logging index */
 269         int                     *lfloghigh)     /* high leaf logging index */
 270 {
 271         if (!leaf->hdr.stale) {
 272                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
 273
 274                 /*
 275                  * Now we need to make room to insert the leaf entry.
 276                  *
 277                  * If there are no stale entries, just insert a hole at index.
 278                  */
 279                 lep = &leaf->ents[index];
 280                 if (index < be16_to_cpu(leaf->hdr.count))
 281                         memmove(lep + 1, lep,
 282                                 (be16_to_cpu(leaf->hdr.count) - index) *
 283                                  sizeof(*lep));
 284
 285                 /*
 286                  * Record low and high logging indices for the leaf.
 287                  */
 288                 *lfloglow = index;
 289                 *lfloghigh = be16_to_cpu(leaf->hdr.count);
 290                 be16_add_cpu(&leaf->hdr.count, 1);
 291                 return lep;
 292         }
 293
 294         /*
 295          * There are stale entries.
 296          *
 297          * We will use one of them for the new entry.  It's probably not at
 298          * the right location, so we'll have to shift some up or down first.
 299          *
 300          * If we didn't compact before, we need to find the nearest stale
 301          * entries before and after our insertion point.
 302          */
 303         if (compact == 0)
 304                 xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
 305
 306         /*
 307          * If the low one is better, use it.
 308          */
 309         if (lowstale >= 0 &&
 310             (highstale == be16_to_cpu(leaf->hdr.count) ||
 311              index - lowstale - 1 < highstale - index)) {
 312                 ASSERT(index - lowstale - 1 >= 0);
 313                 ASSERT(leaf->ents[lowstale].address ==
 314                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 315
 316                 /*
 317                  * Copy entries up to cover the stale entry and make room
 318                  * for the new entry.
 319                  */
 320                 if (index - lowstale - 1 > 0) {
 321                         memmove(&leaf->ents[lowstale],
 322                                 &leaf->ents[lowstale + 1],
 323                                 (index - lowstale - 1) *
 324                                 sizeof(xfs_dir2_leaf_entry_t));
 325                 }
 326                 *lfloglow = MIN(lowstale, *lfloglow);
 327                 *lfloghigh = MAX(index - 1, *lfloghigh);
 328                 be16_add_cpu(&leaf->hdr.stale, -1);
 329                 return &leaf->ents[index - 1];
 330         }
 331
 332         /*
 333          * The high one is better, so use that one.
 334          */
 335         ASSERT(highstale - index >= 0);
 336         ASSERT(leaf->ents[highstale].address ==
 337                cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 338
 339         /*
 340          * Copy entries down to cover the stale entry and make room for the
 341          * new entry.
 342          */
 343         if (highstale - index > 0) {
 344                 memmove(&leaf->ents[index + 1],
 345                         &leaf->ents[index],
 346                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
 347         }
 348         *lfloglow = MIN(index, *lfloglow);
 349         *lfloghigh = MAX(highstale, *lfloghigh);
 350         be16_add_cpu(&leaf->hdr.stale, -1);
 351         return &leaf->ents[index];
 352 }
 353
 354 /*
 355  * Add an entry to a leaf form directory.
 356  */
 357 int                                             /* error */
 358 xfs_dir2_leaf_addname(
 359         xfs_da_args_t           *args)          /* operation arguments */
 360 {
 361         __be16                  *bestsp;        /* freespace table in leaf */
 362         int                     compact;        /* need to compact leaves */
 363         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 364         struct xfs_buf          *dbp;           /* data block buffer */
 365         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 366         xfs_inode_t             *dp;            /* incore directory inode */
 367         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
 368         int                     error;          /* error return value */
 369         int                     grown;          /* allocated new data block */
 370         int                     highstale;      /* index of next stale leaf */
 371         int                     i;              /* temporary, index */
 372         int                     index;          /* leaf table position */
 373         struct xfs_buf          *lbp;           /* leaf's buffer */
 374         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 375         int                     length;         /* length of new entry */
 376         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
 377         int                     lfloglow;       /* low leaf logging index */
 378         int                     lfloghigh;      /* high leaf logging index */
 379         int                     lowstale;       /* index of prev stale leaf */
 380         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 381         xfs_mount_t             *mp;            /* filesystem mount point */
 382         int                     needbytes;      /* leaf block bytes needed */
 383         int                     needlog;        /* need to log data header */
 384         int                     needscan;       /* need to rescan data free */
 385         __be16                  *tagp;          /* end of data entry */
 386         xfs_trans_t             *tp;            /* transaction pointer */
 387         xfs_dir2_db_t           use_block;      /* data block number */
 388
 389         trace_xfs_dir2_leaf_addname(args);
 390
 391         dp = args->dp;
 392         tp = args->trans;
 393         mp = dp->i_mount;
 394
 395         error = xfs_dir2_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
 396         if (error)
 397                 return error;
 398
 399         /*
 400          * Look up the entry by hash value and name.
 401          * We know it's not there, our caller has already done a lookup.
 402          * So the index is of the entry to insert in front of.
 403          * But if there are dup hash values the index is of the first of those.
 404          */
 405         index = xfs_dir2_leaf_search_hash(args, lbp);
 406         leaf = lbp->b_addr;
 407         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 408         bestsp = xfs_dir2_leaf_bests_p(ltp);
 409         length = xfs_dir2_data_entsize(args->namelen);
 410         /*
 411          * See if there are any entries with the same hash value
 412          * and space in their block for the new entry.
 413          * This is good because it puts multiple same-hash value entries
 414          * in a data block, improving the lookup of those entries.
 415          */
 416         for (use_block = -1, lep = &leaf->ents[index];
 417              index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
 418              index++, lep++) {
 419                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 420                         continue;
 421                 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 422                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 423                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
 424                 if (be16_to_cpu(bestsp[i]) >= length) {
 425                         use_block = i;
 426                         break;
 427                 }
 428         }
 429         /*
 430          * Didn't find a block yet, linear search all the data blocks.
 431          */
 432         if (use_block == -1) {
 433                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 434                         /*
 435                          * Remember a block we see that's missing.
 436                          */
 437                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
 438                             use_block == -1)
 439                                 use_block = i;
 440                         else if (be16_to_cpu(bestsp[i]) >= length) {
 441                                 use_block = i;
 442                                 break;
 443                         }
 444                 }
 445         }
 446         /*
 447          * How many bytes do we need in the leaf block?
 448          */
 449         needbytes = 0;
 450         if (!leaf->hdr.stale)
 451                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
 452         if (use_block == -1)
 453                 needbytes += sizeof(xfs_dir2_data_off_t);
 454
 455         /*
 456          * Now kill use_block if it refers to a missing block, so we
 457          * can use it as an indication of allocation needed.
 458          */
 459         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
 460                 use_block = -1;
 461         /*
 462          * If we don't have enough free bytes but we can make enough
 463          * by compacting out stale entries, we'll do that.
 464          */
 465         if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <
 466                                 needbytes && be16_to_cpu(leaf->hdr.stale) > 1) {
 467                 compact = 1;
 468         }
 469         /*
 470          * Otherwise if we don't have enough free bytes we need to
 471          * convert to node form.
 472          */
 473         else if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(
 474                                                 leaf->hdr.count)] < needbytes) {
 475                 /*
 476                  * Just checking or no space reservation, give up.
 477                  */
 478                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 479                                                         args->total == 0) {
 480                         xfs_trans_brelse(tp, lbp);
 481                         return XFS_ERROR(ENOSPC);
 482                 }
 483                 /*
 484                  * Convert to node form.
 485                  */
 486                 error = xfs_dir2_leaf_to_node(args, lbp);
 487                 if (error)
 488                         return error;
 489                 /*
 490                  * Then add the new entry.
 491                  */
 492                 return xfs_dir2_node_addname(args);
 493         }
 494         /*
 495          * Otherwise it will fit without compaction.
 496          */
 497         else
 498                 compact = 0;
 499         /*
 500          * If just checking, then it will fit unless we needed to allocate
 501          * a new data block.
 502          */
 503         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 504                 xfs_trans_brelse(tp, lbp);
 505                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
 506         }
 507         /*
 508          * If no allocations are allowed, return now before we've
 509          * changed anything.
 510          */
 511         if (args->total == 0 && use_block == -1) {
 512                 xfs_trans_brelse(tp, lbp);
 513                 return XFS_ERROR(ENOSPC);
 514         }
 515         /*
 516          * Need to compact the leaf entries, removing stale ones.
 517          * Leave one stale entry behind - the one closest to our
 518          * insertion index - and we'll shift that one to our insertion
 519          * point later.
 520          */
 521         if (compact) {
 522                 xfs_dir2_leaf_compact_x1(lbp, &index, &lowstale, &highstale,
 523                         &lfloglow, &lfloghigh);
 524         }
 525         /*
 526          * There are stale entries, so we'll need log-low and log-high
 527          * impossibly bad values later.
 528          */
 529         else if (be16_to_cpu(leaf->hdr.stale)) {
 530                 lfloglow = be16_to_cpu(leaf->hdr.count);
 531                 lfloghigh = -1;
 532         }
 533         /*
 534          * If there was no data block space found, we need to allocate
 535          * a new one.
 536          */
 537         if (use_block == -1) {
 538                 /*
 539                  * Add the new data block.
 540                  */
 541                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 542                                 &use_block))) {
 543                         xfs_trans_brelse(tp, lbp);
 544                         return error;
 545                 }
 546                 /*
 547                  * Initialize the block.
 548                  */
 549                 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
 550                         xfs_trans_brelse(tp, lbp);
 551                         return error;
 552                 }
 553                 /*
 554                  * If we're adding a new data block on the end we need to
 555                  * extend the bests table.  Copy it up one entry.
 556                  */
 557                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
 558                         bestsp--;
 559                         memmove(&bestsp[0], &bestsp[1],
 560                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 561                         be32_add_cpu(&ltp->bestcount, 1);
 562                         xfs_dir2_leaf_log_tail(tp, lbp);
 563                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
 564                 }
 565                 /*
 566                  * If we're filling in a previously empty block just log it.
 567                  */
 568                 else
 569                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 570                 hdr = dbp->b_addr;
 571                 bestsp[use_block] = hdr->bestfree[0].length;
 572                 grown = 1;
 573         } else {
 574                 /*
 575                  * Already had space in some data block.
 576                  * Just read that one in.
 577                  */
 578                 error = xfs_dir2_data_read(tp, dp,
 579                                            xfs_dir2_db_to_da(mp, use_block),
 580                                            -1, &dbp);
 581                 if (error) {
 582                         xfs_trans_brelse(tp, lbp);
 583                         return error;
 584                 }
 585                 hdr = dbp->b_addr;
 586                 grown = 0;
 587         }
 588         /*
 589          * Point to the biggest freespace in our data block.
 590          */
 591         dup = (xfs_dir2_data_unused_t *)
 592               ((char *)hdr + be16_to_cpu(hdr->bestfree[0].offset));
 593         ASSERT(be16_to_cpu(dup->length) >= length);
 594         needscan = needlog = 0;
 595         /*
 596          * Mark the initial part of our freespace in use for the new entry.
 597          */
 598         xfs_dir2_data_use_free(tp, dbp, dup,
 599                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
 600                 &needlog, &needscan);
 601         /*
 602          * Initialize our new entry (at last).
 603          */
 604         dep = (xfs_dir2_data_entry_t *)dup;
 605         dep->inumber = cpu_to_be64(args->inumber);
 606         dep->namelen = args->namelen;
 607         memcpy(dep->name, args->name, dep->namelen);
 608         tagp = xfs_dir2_data_entry_tag_p(dep);
 609         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 610         /*
 611          * Need to scan fix up the bestfree table.
 612          */
 613         if (needscan)
 614                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 615         /*
 616          * Need to log the data block's header.
 617          */
 618         if (needlog)
 619                 xfs_dir2_data_log_header(tp, dbp);
 620         xfs_dir2_data_log_entry(tp, dbp, dep);
 621         /*
 622          * If the bests table needs to be changed, do it.
 623          * Log the change unless we've already done that.
 624          */
 625         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(hdr->bestfree[0].length)) {
 626                 bestsp[use_block] = hdr->bestfree[0].length;
 627                 if (!grown)
 628                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 629         }
 630
 631         lep = xfs_dir2_leaf_find_entry(leaf, index, compact, lowstale,
 632                                        highstale, &lfloglow, &lfloghigh);
 633
 634         /*
 635          * Fill in the new leaf entry.
 636          */
 637         lep->hashval = cpu_to_be32(args->hashval);
 638         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
 639                                 be16_to_cpu(*tagp)));
 640         /*
 641          * Log the leaf fields and give up the buffers.
 642          */
 643         xfs_dir2_leaf_log_header(tp, lbp);
 644         xfs_dir2_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
 645         xfs_dir2_leaf_check(dp, lbp);
 646         xfs_dir2_data_check(dp, dbp);
 647         return 0;
 648 }
 649
 650 #ifdef DEBUG
 651 /*
 652  * Check the internal consistency of a leaf1 block.
 653  * Pop an assert if something is wrong.
 654  */
 655 STATIC void
 656 xfs_dir2_leaf_check(
 657         struct xfs_inode        *dp,            /* incore directory inode */
 658         struct xfs_buf          *bp)            /* leaf's buffer */
 659 {
 660         int                     i;              /* leaf index */
 661         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 662         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 663         xfs_mount_t             *mp;            /* filesystem mount point */
 664         int                     stale;          /* count of stale leaves */
 665
 666         leaf = bp->b_addr;
 667         mp = dp->i_mount;
 668         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
 669         /*
 670          * This value is not restrictive enough.
 671          * Should factor in the size of the bests table as well.
 672          * We can deduce a value for that from di_size.
 673          */
 674         ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
 675         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 676         /*
 677          * Leaves and bests don't overlap.
 678          */
 679         ASSERT((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <=
 680                (char *)xfs_dir2_leaf_bests_p(ltp));
 681         /*
 682          * Check hash value order, count stale entries.
 683          */
 684         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
 685                 if (i + 1 < be16_to_cpu(leaf->hdr.count))
 686                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
 687                                be32_to_cpu(leaf->ents[i + 1].hashval));
 688                 if (leaf->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 689                         stale++;
 690         }
 691         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
 692 }
 693 #endif  /* DEBUG */
 694
 695 /*
 696  * Compact out any stale entries in the leaf.
 697  * Log the header and changed leaf entries, if any.
 698  */
 699 void
 700 xfs_dir2_leaf_compact(
 701         xfs_da_args_t   *args,          /* operation arguments */
 702         struct xfs_buf  *bp)            /* leaf buffer */
 703 {
 704         int             from;           /* source leaf index */
 705         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 706         int             loglow;         /* first leaf entry to log */
 707         int             to;             /* target leaf index */
 708
 709         leaf = bp->b_addr;
 710         if (!leaf->hdr.stale) {
 711                 return;
 712         }
 713         /*
 714          * Compress out the stale entries in place.
 715          */
 716         for (from = to = 0, loglow = -1; from < be16_to_cpu(leaf->hdr.count); from++) {
 717                 if (leaf->ents[from].address ==
 718                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 719                         continue;
 720                 /*
 721                  * Only actually copy the entries that are different.
 722                  */
 723                 if (from > to) {
 724                         if (loglow == -1)
 725                                 loglow = to;
 726                         leaf->ents[to] = leaf->ents[from];
 727                 }
 728                 to++;
 729         }
 730         /*
 731          * Update and log the header, log the leaf entries.
 732          */
 733         ASSERT(be16_to_cpu(leaf->hdr.stale) == from - to);
 734         be16_add_cpu(&leaf->hdr.count, -(be16_to_cpu(leaf->hdr.stale)));
 735         leaf->hdr.stale = 0;
 736         xfs_dir2_leaf_log_header(args->trans, bp);
 737         if (loglow != -1)
 738                 xfs_dir2_leaf_log_ents(args->trans, bp, loglow, to - 1);
 739 }
 740
 741 /*
 742  * Compact the leaf entries, removing stale ones.
 743  * Leave one stale entry behind - the one closest to our
 744  * insertion index - and the caller will shift that one to our insertion
 745  * point later.
 746  * Return new insertion index, where the remaining stale entry is,
 747  * and leaf logging indices.
 748  */
 749 void
 750 xfs_dir2_leaf_compact_x1(
 751         struct xfs_buf  *bp,            /* leaf buffer */
 752         int             *indexp,        /* insertion index */
 753         int             *lowstalep,     /* out: stale entry before us */
 754         int             *highstalep,    /* out: stale entry after us */
 755         int             *lowlogp,       /* out: low log index */
 756         int             *highlogp)      /* out: high log index */
 757 {
 758         int             from;           /* source copy index */
 759         int             highstale;      /* stale entry at/after index */
 760         int             index;          /* insertion index */
 761         int             keepstale;      /* source index of kept stale */
 762         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 763         int             lowstale;       /* stale entry before index */
 764         int             newindex=0;     /* new insertion index */
 765         int             to;             /* destination copy index */
 766
 767         leaf = bp->b_addr;
 768         ASSERT(be16_to_cpu(leaf->hdr.stale) > 1);
 769         index = *indexp;
 770
 771         xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
 772
 773         /*
 774          * Pick the better of lowstale and highstale.
 775          */
 776         if (lowstale >= 0 &&
 777             (highstale == be16_to_cpu(leaf->hdr.count) ||
 778              index - lowstale <= highstale - index))
 779                 keepstale = lowstale;
 780         else
 781                 keepstale = highstale;
 782         /*
 783          * Copy the entries in place, removing all the stale entries
 784          * except keepstale.
 785          */
 786         for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
 787                 /*
 788                  * Notice the new value of index.
 789                  */
 790                 if (index == from)
 791                         newindex = to;
 792                 if (from != keepstale &&
 793                     leaf->ents[from].address ==
 794                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
 795                         if (from == to)
 796                                 *lowlogp = to;
 797                         continue;
 798                 }
 799                 /*
 800                  * Record the new keepstale value for the insertion.
 801                  */
 802                 if (from == keepstale)
 803                         lowstale = highstale = to;
 804                 /*
 805                  * Copy only the entries that have moved.
 806                  */
 807                 if (from > to)
 808                         leaf->ents[to] = leaf->ents[from];
 809                 to++;
 810         }
 811         ASSERT(from > to);
 812         /*
 813          * If the insertion point was past the last entry,
 814          * set the new insertion point accordingly.
 815          */
 816         if (index == from)
 817                 newindex = to;
 818         *indexp = newindex;
 819         /*
 820          * Adjust the leaf header values.
 821          */
 822         be16_add_cpu(&leaf->hdr.count, -(from - to));
 823         leaf->hdr.stale = cpu_to_be16(1);
 824         /*
 825          * Remember the low/high stale value only in the "right"
 826          * direction.
 827          */
 828         if (lowstale >= newindex)
 829                 lowstale = -1;
 830         else
 831                 highstale = be16_to_cpu(leaf->hdr.count);
 832         *highlogp = be16_to_cpu(leaf->hdr.count) - 1;
 833         *lowstalep = lowstale;
 834         *highstalep = highstale;
 835 }
 836
 837 struct xfs_dir2_leaf_map_info {
 838         xfs_extlen_t    map_blocks;     /* number of fsbs in map */
 839         xfs_dablk_t     map_off;        /* last mapped file offset */
 840         int             map_size;       /* total entries in *map */
 841         int             map_valid;      /* valid entries in *map */
 842         int             nmap;           /* mappings to ask xfs_bmapi */
 843         xfs_dir2_db_t   curdb;          /* db for current block */
 844         int             ra_current;     /* number of read-ahead blks */
 845         int             ra_index;       /* *map index for read-ahead */
 846         int             ra_offset;      /* map entry offset for ra */
 847         int             ra_want;        /* readahead count wanted */
 848         struct xfs_bmbt_irec map[];     /* map vector for blocks */
 849 };
 850
 851 STATIC int
 852 xfs_dir2_leaf_readbuf(
 853         struct xfs_inode        *dp,
 854         size_t                  bufsize,
 855         struct xfs_dir2_leaf_map_info *mip,
 856         xfs_dir2_off_t          *curoff,
 857         struct xfs_buf          **bpp)
 858 {
 859         struct xfs_mount        *mp = dp->i_mount;
 860         struct xfs_buf          *bp = *bpp;
 861         struct xfs_bmbt_irec    *map = mip->map;
 862         int                     error = 0;
 863         int                     length;
 864         int                     i;
 865         int                     j;
 866
 867         /*
 868          * If we have a buffer, we need to release it and
 869          * take it out of the mapping.
 870          */
 871
 872         if (bp) {
 873                 xfs_trans_brelse(NULL, bp);
 874                 bp = NULL;
 875                 mip->map_blocks -= mp->m_dirblkfsbs;
 876                 /*
 877                  * Loop to get rid of the extents for the
 878                  * directory block.
 879                  */
 880                 for (i = mp->m_dirblkfsbs; i > 0; ) {
 881                         j = min_t(int, map->br_blockcount, i);
 882                         map->br_blockcount -= j;
 883                         map->br_startblock += j;
 884                         map->br_startoff += j;
 885                         /*
 886                          * If mapping is done, pitch it from
 887                          * the table.
 888                          */
 889                         if (!map->br_blockcount && --mip->map_valid)
 890                                 memmove(&map[0], &map[1],
 891                                         sizeof(map[0]) * mip->map_valid);
 892                         i -= j;
 893                 }
 894         }
 895
 896         /*
 897          * Recalculate the readahead blocks wanted.
 898          */
 899         mip->ra_want = howmany(bufsize + mp->m_dirblksize,
 900                                mp->m_sb.sb_blocksize) - 1;
 901         ASSERT(mip->ra_want >= 0);
 902
 903         /*
 904          * If we don't have as many as we want, and we haven't
 905          * run out of data blocks, get some more mappings.
 906          */
 907         if (1 + mip->ra_want > mip->map_blocks &&
 908             mip->map_off < xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET)) {
 909                 /*
 910                  * Get more bmaps, fill in after the ones
 911                  * we already have in the table.
 912                  */
 913                 mip->nmap = mip->map_size - mip->map_valid;
 914                 error = xfs_bmapi_read(dp, mip->map_off,
 915                                 xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET) -
 916                                                                 mip->map_off,
 917                                 &map[mip->map_valid], &mip->nmap, 0);
 918
 919                 /*
 920                  * Don't know if we should ignore this or try to return an
 921                  * error.  The trouble with returning errors is that readdir
 922                  * will just stop without actually passing the error through.
 923                  */
 924                 if (error)
 925                         goto out;       /* XXX */
 926
 927                 /*
 928                  * If we got all the mappings we asked for, set the final map
 929                  * offset based on the last bmap value received.  Otherwise,
 930                  * we've reached the end.
 931                  */
 932                 if (mip->nmap == mip->map_size - mip->map_valid) {
 933                         i = mip->map_valid + mip->nmap - 1;
 934                         mip->map_off = map[i].br_startoff + map[i].br_blockcount;
 935                 } else
 936                         mip->map_off = xfs_dir2_byte_to_da(mp,
 937                                                         XFS_DIR2_LEAF_OFFSET);
 938
 939                 /*
 940                  * Look for holes in the mapping, and eliminate them.  Count up
 941                  * the valid blocks.
 942                  */
 943                 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
 944                         if (map[i].br_startblock == HOLESTARTBLOCK) {
 945                                 mip->nmap--;
 946                                 length = mip->map_valid + mip->nmap - i;
 947                                 if (length)
 948                                         memmove(&map[i], &map[i + 1],
 949                                                 sizeof(map[i]) * length);
 950                         } else {
 951                                 mip->map_blocks += map[i].br_blockcount;
 952                                 i++;
 953                         }
 954                 }
 955                 mip->map_valid += mip->nmap;
 956         }
 957
 958         /*
 959          * No valid mappings, so no more data blocks.
 960          */
 961         if (!mip->map_valid) {
 962                 *curoff = xfs_dir2_da_to_byte(mp, mip->map_off);
 963                 goto out;
 964         }
 965
 966         /*
 967          * Read the directory block starting at the first mapping.
 968          */
 969         mip->curdb = xfs_dir2_da_to_db(mp, map->br_startoff);
 970         error = xfs_dir2_data_read(NULL, dp, map->br_startoff,
 971                         map->br_blockcount >= mp->m_dirblkfsbs ?
 972                             XFS_FSB_TO_DADDR(mp, map->br_startblock) : -1, &bp);
 973
 974         /*
 975          * Should just skip over the data block instead of giving up.
 976          */
 977         if (error)
 978                 goto out;       /* XXX */
 979
 980         /*
 981          * Adjust the current amount of read-ahead: we just read a block that
 982          * was previously ra.
 983          */
 984         if (mip->ra_current)
 985                 mip->ra_current -= mp->m_dirblkfsbs;
 986
 987         /*
 988          * Do we need more readahead?
 989          */
 990         for (mip->ra_index = mip->ra_offset = i = 0;
 991              mip->ra_want > mip->ra_current && i < mip->map_blocks;
 992              i += mp->m_dirblkfsbs) {
 993                 ASSERT(mip->ra_index < mip->map_valid);
 994                 /*
 995                  * Read-ahead a contiguous directory block.
 996                  */
 997                 if (i > mip->ra_current &&
 998                     map[mip->ra_index].br_blockcount >= mp->m_dirblkfsbs) {
 999                         xfs_dir2_data_readahead(NULL, dp,
1000                                 map[mip->ra_index].br_startoff + mip->ra_offset,
1001                                 XFS_FSB_TO_DADDR(mp,
1002                                         map[mip->ra_index].br_startblock +
1003                                                         mip->ra_offset));
1004                         mip->ra_current = i;
1005                 }
1006
1007                 /*
1008                  * Read-ahead a non-contiguous directory block.  This doesn't
1009                  * use our mapping, but this is a very rare case.
1010                  */
1011                 else if (i > mip->ra_current) {
1012                         xfs_dir2_data_readahead(NULL, dp,
1013                                         map[mip->ra_index].br_startoff +
1014                                                         mip->ra_offset, -1);
1015                         mip->ra_current = i;
1016                 }
1017
1018                 /*
1019                  * Advance offset through the mapping table.
1020                  */
1021                 for (j = 0; j < mp->m_dirblkfsbs; j++) {
1022                         /*
1023                          * The rest of this extent but not more than a dir
1024                          * block.
1025                          */
1026                         length = min_t(int, mp->m_dirblkfsbs,
1027                                         map[mip->ra_index].br_blockcount -
1028                                                         mip->ra_offset);
1029                         j += length;
1030                         mip->ra_offset += length;
1031
1032                         /*
1033                          * Advance to the next mapping if this one is used up.
1034                          */
1035                         if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
1036                                 mip->ra_offset = 0;
1037                                 mip->ra_index++;
1038                         }
1039                 }
1040         }
1041
1042 out:
1043         *bpp = bp;
1044         return error;
1045 }
1046
1047 /*
1048  * Getdents (readdir) for leaf and node directories.
1049  * This reads the data blocks only, so is the same for both forms.
1050  */
1051 int                                             /* error */
1052 xfs_dir2_leaf_getdents(
1053         xfs_inode_t             *dp,            /* incore directory inode */
1054         void                    *dirent,
1055         size_t                  bufsize,
1056         xfs_off_t               *offset,
1057         filldir_t               filldir)
1058 {
1059         struct xfs_buf          *bp = NULL;     /* data block buffer */
1060         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1061         xfs_dir2_data_entry_t   *dep;           /* data entry */
1062         xfs_dir2_data_unused_t  *dup;           /* unused entry */
1063         int                     error = 0;      /* error return value */
1064         int                     length;         /* temporary length value */
1065         xfs_mount_t             *mp;            /* filesystem mount point */
1066         int                     byteoff;        /* offset in current block */
1067         xfs_dir2_off_t          curoff;         /* current overall offset */
1068         xfs_dir2_off_t          newoff;         /* new curoff after new blk */
1069         char                    *ptr = NULL;    /* pointer to current data */
1070         struct xfs_dir2_leaf_map_info *map_info;
1071
1072         /*
1073          * If the offset is at or past the largest allowed value,
1074          * give up right away.
1075          */
1076         if (*offset >= XFS_DIR2_MAX_DATAPTR)
1077                 return 0;
1078
1079         mp = dp->i_mount;
1080
1081         /*
1082          * Set up to bmap a number of blocks based on the caller's
1083          * buffer size, the directory block size, and the filesystem
1084          * block size.
1085          */
1086         length = howmany(bufsize + mp->m_dirblksize,
1087                                      mp->m_sb.sb_blocksize);
1088         map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
1089                                 (length * sizeof(struct xfs_bmbt_irec)),
1090                                KM_SLEEP);
1091         map_info->map_size = length;
1092
1093         /*
1094          * Inside the loop we keep the main offset value as a byte offset
1095          * in the directory file.
1096          */
1097         curoff = xfs_dir2_dataptr_to_byte(mp, *offset);
1098
1099         /*
1100          * Force this conversion through db so we truncate the offset
1101          * down to get the start of the data block.
1102          */
1103         map_info->map_off = xfs_dir2_db_to_da(mp,
1104                                               xfs_dir2_byte_to_db(mp, curoff));
1105
1106         /*
1107          * Loop over directory entries until we reach the end offset.
1108          * Get more blocks and readahead as necessary.
1109          */
1110         while (curoff < XFS_DIR2_LEAF_OFFSET) {
1111                 /*
1112                  * If we have no buffer, or we're off the end of the
1113                  * current buffer, need to get another one.
1114                  */
1115                 if (!bp || ptr >= (char *)bp->b_addr + mp->m_dirblksize) {
1116
1117                         error = xfs_dir2_leaf_readbuf(dp, bufsize, map_info,
1118                                                       &curoff, &bp);
1119                         if (error || !map_info->map_valid)
1120                                 break;
1121
1122                         /*
1123                          * Having done a read, we need to set a new offset.
1124                          */
1125                         newoff = xfs_dir2_db_off_to_byte(mp, map_info->curdb, 0);
1126                         /*
1127                          * Start of the current block.
1128                          */
1129                         if (curoff < newoff)
1130                                 curoff = newoff;
1131                         /*
1132                          * Make sure we're in the right block.
1133                          */
1134                         else if (curoff > newoff)
1135                                 ASSERT(xfs_dir2_byte_to_db(mp, curoff) ==
1136                                        map_info->curdb);
1137                         hdr = bp->b_addr;
1138                         xfs_dir2_data_check(dp, bp);
1139                         /*
1140                          * Find our position in the block.
1141                          */
1142                         ptr = (char *)(hdr + 1);
1143                         byteoff = xfs_dir2_byte_to_off(mp, curoff);
1144                         /*
1145                          * Skip past the header.
1146                          */
1147                         if (byteoff == 0)
1148                                 curoff += (uint)sizeof(*hdr);
1149                         /*
1150                          * Skip past entries until we reach our offset.
1151                          */
1152                         else {
1153                                 while ((char *)ptr - (char *)hdr < byteoff) {
1154                                         dup = (xfs_dir2_data_unused_t *)ptr;
1155
1156                                         if (be16_to_cpu(dup->freetag)
1157                                                   == XFS_DIR2_DATA_FREE_TAG) {
1158
1159                                                 length = be16_to_cpu(dup->length);
1160                                                 ptr += length;
1161                                                 continue;
1162                                         }
1163                                         dep = (xfs_dir2_data_entry_t *)ptr;
1164                                         length =
1165                                            xfs_dir2_data_entsize(dep->namelen);
1166                                         ptr += length;
1167                                 }
1168                                 /*
1169                                  * Now set our real offset.
1170                                  */
1171                                 curoff =
1172                                         xfs_dir2_db_off_to_byte(mp,
1173                                             xfs_dir2_byte_to_db(mp, curoff),
1174                                             (char *)ptr - (char *)hdr);
1175                                 if (ptr >= (char *)hdr + mp->m_dirblksize) {
1176                                         continue;
1177                                 }
1178                         }
1179                 }
1180                 /*
1181                  * We have a pointer to an entry.
1182                  * Is it a live one?
1183                  */
1184                 dup = (xfs_dir2_data_unused_t *)ptr;
1185                 /*
1186                  * No, it's unused, skip over it.
1187                  */
1188                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
1189                         length = be16_to_cpu(dup->length);
1190                         ptr += length;
1191                         curoff += length;
1192                         continue;
1193                 }
1194
1195                 dep = (xfs_dir2_data_entry_t *)ptr;
1196                 length = xfs_dir2_data_entsize(dep->namelen);
1197
1198                 if (filldir(dirent, (char *)dep->name, dep->namelen,
1199                             xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff,
1200                             be64_to_cpu(dep->inumber), DT_UNKNOWN))
1201                         break;
1202
1203                 /*
1204                  * Advance to next entry in the block.
1205                  */
1206                 ptr += length;
1207                 curoff += length;
1208                 /* bufsize may have just been a guess; don't go negative */
1209                 bufsize = bufsize > length ? bufsize - length : 0;
1210         }
1211
1212         /*
1213          * All done.  Set output offset value to current offset.
1214          */
1215         if (curoff > xfs_dir2_dataptr_to_byte(mp, XFS_DIR2_MAX_DATAPTR))
1216                 *offset = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
1217         else
1218                 *offset = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
1219         kmem_free(map_info);
1220         if (bp)
1221                 xfs_trans_brelse(NULL, bp);
1222         return error;
1223 }
1224
1225 /*
1226  * Initialize a new leaf block, leaf1 or leafn magic accepted.
1227  */
1228 int
1229 xfs_dir2_leaf_init(
1230         xfs_da_args_t           *args,          /* operation arguments */
1231         xfs_dir2_db_t           bno,            /* directory block number */
1232         struct xfs_buf          **bpp,          /* out: leaf buffer */
1233         int                     magic)          /* magic number for block */
1234 {
1235         struct xfs_buf          *bp;            /* leaf buffer */
1236         xfs_inode_t             *dp;            /* incore directory inode */
1237         int                     error;          /* error return code */
1238         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1239         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1240         xfs_mount_t             *mp;            /* filesystem mount point */
1241         xfs_trans_t             *tp;            /* transaction pointer */
1242
1243         dp = args->dp;
1244         ASSERT(dp != NULL);
1245         tp = args->trans;
1246         mp = dp->i_mount;
1247         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
1248                bno < XFS_DIR2_FREE_FIRSTDB(mp));
1249         /*
1250          * Get the buffer for the block.
1251          */
1252         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
1253                                XFS_DATA_FORK);
1254         if (error)
1255                 return error;
1256
1257         /*
1258          * Initialize the header.
1259          */
1260         leaf = bp->b_addr;
1261         leaf->hdr.info.magic = cpu_to_be16(magic);
1262         leaf->hdr.info.forw = 0;
1263         leaf->hdr.info.back = 0;
1264         leaf->hdr.count = 0;
1265         leaf->hdr.stale = 0;
1266         xfs_dir2_leaf_log_header(tp, bp);
1267         /*
1268          * If it's a leaf-format directory initialize the tail.
1269          * In this case our caller has the real bests table to copy into
1270          * the block.
1271          */
1272         if (magic == XFS_DIR2_LEAF1_MAGIC) {
1273                 bp->b_ops = &xfs_dir2_leaf1_buf_ops;
1274                 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1275                 ltp->bestcount = 0;
1276                 xfs_dir2_leaf_log_tail(tp, bp);
1277         } else
1278                 bp->b_ops = &xfs_dir2_leafn_buf_ops;
1279         *bpp = bp;
1280         return 0;
1281 }
1282
1283 /*
1284  * Log the bests entries indicated from a leaf1 block.
1285  */
1286 static void
1287 xfs_dir2_leaf_log_bests(
1288         xfs_trans_t             *tp,            /* transaction pointer */
1289         struct xfs_buf          *bp,            /* leaf buffer */
1290         int                     first,          /* first entry to log */
1291         int                     last)           /* last entry to log */
1292 {
1293         __be16                  *firstb;        /* pointer to first entry */
1294         __be16                  *lastb;         /* pointer to last entry */
1295         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1296         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1297
1298         leaf = bp->b_addr;
1299         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1300         ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1301         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1302         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1303         xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1304                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1305 }
1306
1307 /*
1308  * Log the leaf entries indicated from a leaf1 or leafn block.
1309  */
1310 void
1311 xfs_dir2_leaf_log_ents(
1312         xfs_trans_t             *tp,            /* transaction pointer */
1313         struct xfs_buf          *bp,            /* leaf buffer */
1314         int                     first,          /* first entry to log */
1315         int                     last)           /* last entry to log */
1316 {
1317         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1318         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1319         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1320
1321         leaf = bp->b_addr;
1322         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1323                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1324         firstlep = &leaf->ents[first];
1325         lastlep = &leaf->ents[last];
1326         xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1327                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1328 }
1329
1330 /*
1331  * Log the header of the leaf1 or leafn block.
1332  */
1333 void
1334 xfs_dir2_leaf_log_header(
1335         struct xfs_trans        *tp,
1336         struct xfs_buf          *bp)
1337 {
1338         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1339
1340         leaf = bp->b_addr;
1341         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1342                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1343         xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1344                 (uint)(sizeof(leaf->hdr) - 1));
1345 }
1346
1347 /*
1348  * Log the tail of the leaf1 block.
1349  */
1350 STATIC void
1351 xfs_dir2_leaf_log_tail(
1352         struct xfs_trans        *tp,
1353         struct xfs_buf          *bp)
1354 {
1355         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1356         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1357         xfs_mount_t             *mp;            /* filesystem mount point */
1358
1359         mp = tp->t_mountp;
1360         leaf = bp->b_addr;
1361         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1362         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1363         xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1364                 (uint)(mp->m_dirblksize - 1));
1365 }
1366
1367 /*
1368  * Look up the entry referred to by args in the leaf format directory.
1369  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1370  * is also used by the node-format code.
1371  */
1372 int
1373 xfs_dir2_leaf_lookup(
1374         xfs_da_args_t           *args)          /* operation arguments */
1375 {
1376         struct xfs_buf          *dbp;           /* data block buffer */
1377         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1378         xfs_inode_t             *dp;            /* incore directory inode */
1379         int                     error;          /* error return code */
1380         int                     index;          /* found entry index */
1381         struct xfs_buf          *lbp;           /* leaf buffer */
1382         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1383         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1384         xfs_trans_t             *tp;            /* transaction pointer */
1385
1386         trace_xfs_dir2_leaf_lookup(args);
1387
1388         /*
1389          * Look up name in the leaf block, returning both buffers and index.
1390          */
1391         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1392                 return error;
1393         }
1394         tp = args->trans;
1395         dp = args->dp;
1396         xfs_dir2_leaf_check(dp, lbp);
1397         leaf = lbp->b_addr;
1398         /*
1399          * Get to the leaf entry and contained data entry address.
1400          */
1401         lep = &leaf->ents[index];
1402         /*
1403          * Point to the data entry.
1404          */
1405         dep = (xfs_dir2_data_entry_t *)
1406               ((char *)dbp->b_addr +
1407                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1408         /*
1409          * Return the found inode number & CI name if appropriate
1410          */
1411         args->inumber = be64_to_cpu(dep->inumber);
1412         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1413         xfs_trans_brelse(tp, dbp);
1414         xfs_trans_brelse(tp, lbp);
1415         return XFS_ERROR(error);
1416 }
1417
1418 /*
1419  * Look up name/hash in the leaf block.
1420  * Fill in indexp with the found index, and dbpp with the data buffer.
1421  * If not found dbpp will be NULL, and ENOENT comes back.
1422  * lbpp will always be filled in with the leaf buffer unless there's an error.
1423  */
1424 static int                                      /* error */
1425 xfs_dir2_leaf_lookup_int(
1426         xfs_da_args_t           *args,          /* operation arguments */
1427         struct xfs_buf          **lbpp,         /* out: leaf buffer */
1428         int                     *indexp,        /* out: index in leaf block */
1429         struct xfs_buf          **dbpp)         /* out: data buffer */
1430 {
1431         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1432         struct xfs_buf          *dbp = NULL;    /* data buffer */
1433         xfs_dir2_data_entry_t   *dep;           /* data entry */
1434         xfs_inode_t             *dp;            /* incore directory inode */
1435         int                     error;          /* error return code */
1436         int                     index;          /* index in leaf block */
1437         struct xfs_buf          *lbp;           /* leaf buffer */
1438         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1439         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1440         xfs_mount_t             *mp;            /* filesystem mount point */
1441         xfs_dir2_db_t           newdb;          /* new data block number */
1442         xfs_trans_t             *tp;            /* transaction pointer */
1443         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1444         enum xfs_dacmp          cmp;            /* name compare result */
1445
1446         dp = args->dp;
1447         tp = args->trans;
1448         mp = dp->i_mount;
1449
1450         error = xfs_dir2_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
1451         if (error)
1452                 return error;
1453
1454         *lbpp = lbp;
1455         leaf = lbp->b_addr;
1456         xfs_dir2_leaf_check(dp, lbp);
1457         /*
1458          * Look for the first leaf entry with our hash value.
1459          */
1460         index = xfs_dir2_leaf_search_hash(args, lbp);
1461         /*
1462          * Loop over all the entries with the right hash value
1463          * looking to match the name.
1464          */
1465         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
1466                                 be32_to_cpu(lep->hashval) == args->hashval;
1467                                 lep++, index++) {
1468                 /*
1469                  * Skip over stale leaf entries.
1470                  */
1471                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1472                         continue;
1473                 /*
1474                  * Get the new data block number.
1475                  */
1476                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1477                 /*
1478                  * If it's not the same as the old data block number,
1479                  * need to pitch the old one and read the new one.
1480                  */
1481                 if (newdb != curdb) {
1482                         if (dbp)
1483                                 xfs_trans_brelse(tp, dbp);
1484                         error = xfs_dir2_data_read(tp, dp,
1485                                                    xfs_dir2_db_to_da(mp, newdb),
1486                                                    -1, &dbp);
1487                         if (error) {
1488                                 xfs_trans_brelse(tp, lbp);
1489                                 return error;
1490                         }
1491                         curdb = newdb;
1492                 }
1493                 /*
1494                  * Point to the data entry.
1495                  */
1496                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1497                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1498                 /*
1499                  * Compare name and if it's an exact match, return the index
1500                  * and buffer. If it's the first case-insensitive match, store
1501                  * the index and buffer and continue looking for an exact match.
1502                  */
1503                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1504                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1505                         args->cmpresult = cmp;
1506                         *indexp = index;
1507                         /* case exact match: return the current buffer. */
1508                         if (cmp == XFS_CMP_EXACT) {
1509                                 *dbpp = dbp;
1510                                 return 0;
1511                         }
1512                         cidb = curdb;
1513                 }
1514         }
1515         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1516         /*
1517          * Here, we can only be doing a lookup (not a rename or remove).
1518          * If a case-insensitive match was found earlier, re-read the
1519          * appropriate data block if required and return it.
1520          */
1521         if (args->cmpresult == XFS_CMP_CASE) {
1522                 ASSERT(cidb != -1);
1523                 if (cidb != curdb) {
1524                         xfs_trans_brelse(tp, dbp);
1525                         error = xfs_dir2_data_read(tp, dp,
1526                                                    xfs_dir2_db_to_da(mp, cidb),
1527                                                    -1, &dbp);
1528                         if (error) {
1529                                 xfs_trans_brelse(tp, lbp);
1530                                 return error;
1531                         }
1532                 }
1533                 *dbpp = dbp;
1534                 return 0;
1535         }
1536         /*
1537          * No match found, return ENOENT.
1538          */
1539         ASSERT(cidb == -1);
1540         if (dbp)
1541                 xfs_trans_brelse(tp, dbp);
1542         xfs_trans_brelse(tp, lbp);
1543         return XFS_ERROR(ENOENT);
1544 }
1545
1546 /*
1547  * Remove an entry from a leaf format directory.
1548  */
1549 int                                             /* error */
1550 xfs_dir2_leaf_removename(
1551         xfs_da_args_t           *args)          /* operation arguments */
1552 {
1553         __be16                  *bestsp;        /* leaf block best freespace */
1554         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1555         xfs_dir2_db_t           db;             /* data block number */
1556         struct xfs_buf          *dbp;           /* data block buffer */
1557         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1558         xfs_inode_t             *dp;            /* incore directory inode */
1559         int                     error;          /* error return code */
1560         xfs_dir2_db_t           i;              /* temporary data block # */
1561         int                     index;          /* index into leaf entries */
1562         struct xfs_buf          *lbp;           /* leaf buffer */
1563         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1564         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1565         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1566         xfs_mount_t             *mp;            /* filesystem mount point */
1567         int                     needlog;        /* need to log data header */
1568         int                     needscan;       /* need to rescan data frees */
1569         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1570         xfs_trans_t             *tp;            /* transaction pointer */
1571
1572         trace_xfs_dir2_leaf_removename(args);
1573
1574         /*
1575          * Lookup the leaf entry, get the leaf and data blocks read in.
1576          */
1577         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1578                 return error;
1579         }
1580         dp = args->dp;
1581         tp = args->trans;
1582         mp = dp->i_mount;
1583         leaf = lbp->b_addr;
1584         hdr = dbp->b_addr;
1585         xfs_dir2_data_check(dp, dbp);
1586         /*
1587          * Point to the leaf entry, use that to point to the data entry.
1588          */
1589         lep = &leaf->ents[index];
1590         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1591         dep = (xfs_dir2_data_entry_t *)
1592               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1593         needscan = needlog = 0;
1594         oldbest = be16_to_cpu(hdr->bestfree[0].length);
1595         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1596         bestsp = xfs_dir2_leaf_bests_p(ltp);
1597         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1598         /*
1599          * Mark the former data entry unused.
1600          */
1601         xfs_dir2_data_make_free(tp, dbp,
1602                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1603                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1604         /*
1605          * We just mark the leaf entry stale by putting a null in it.
1606          */
1607         be16_add_cpu(&leaf->hdr.stale, 1);
1608         xfs_dir2_leaf_log_header(tp, lbp);
1609         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1610         xfs_dir2_leaf_log_ents(tp, lbp, index, index);
1611         /*
1612          * Scan the freespace in the data block again if necessary,
1613          * log the data block header if necessary.
1614          */
1615         if (needscan)
1616                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1617         if (needlog)
1618                 xfs_dir2_data_log_header(tp, dbp);
1619         /*
1620          * If the longest freespace in the data block has changed,
1621          * put the new value in the bests table and log that.
1622          */
1623         if (be16_to_cpu(hdr->bestfree[0].length) != oldbest) {
1624                 bestsp[db] = hdr->bestfree[0].length;
1625                 xfs_dir2_leaf_log_bests(tp, lbp, db, db);
1626         }
1627         xfs_dir2_data_check(dp, dbp);
1628         /*
1629          * If the data block is now empty then get rid of the data block.
1630          */
1631         if (be16_to_cpu(hdr->bestfree[0].length) ==
1632             mp->m_dirblksize - (uint)sizeof(*hdr)) {
1633                 ASSERT(db != mp->m_dirdatablk);
1634                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1635                         /*
1636                          * Nope, can't get rid of it because it caused
1637                          * allocation of a bmap btree block to do so.
1638                          * Just go on, returning success, leaving the
1639                          * empty block in place.
1640                          */
1641                         if (error == ENOSPC && args->total == 0)
1642                                 error = 0;
1643                         xfs_dir2_leaf_check(dp, lbp);
1644                         return error;
1645                 }
1646                 dbp = NULL;
1647                 /*
1648                  * If this is the last data block then compact the
1649                  * bests table by getting rid of entries.
1650                  */
1651                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1652                         /*
1653                          * Look for the last active entry (i).
1654                          */
1655                         for (i = db - 1; i > 0; i--) {
1656                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1657                                         break;
1658                         }
1659                         /*
1660                          * Copy the table down so inactive entries at the
1661                          * end are removed.
1662                          */
1663                         memmove(&bestsp[db - i], bestsp,
1664                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1665                         be32_add_cpu(&ltp->bestcount, -(db - i));
1666                         xfs_dir2_leaf_log_tail(tp, lbp);
1667                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1668                 } else
1669                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1670         }
1671         /*
1672          * If the data block was not the first one, drop it.
1673          */
1674         else if (db != mp->m_dirdatablk)
1675                 dbp = NULL;
1676
1677         xfs_dir2_leaf_check(dp, lbp);
1678         /*
1679          * See if we can convert to block form.
1680          */
1681         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1682 }
1683
1684 /*
1685  * Replace the inode number in a leaf format directory entry.
1686  */
1687 int                                             /* error */
1688 xfs_dir2_leaf_replace(
1689         xfs_da_args_t           *args)          /* operation arguments */
1690 {
1691         struct xfs_buf          *dbp;           /* data block buffer */
1692         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1693         xfs_inode_t             *dp;            /* incore directory inode */
1694         int                     error;          /* error return code */
1695         int                     index;          /* index of leaf entry */
1696         struct xfs_buf          *lbp;           /* leaf buffer */
1697         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1698         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1699         xfs_trans_t             *tp;            /* transaction pointer */
1700
1701         trace_xfs_dir2_leaf_replace(args);
1702
1703         /*
1704          * Look up the entry.
1705          */
1706         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1707                 return error;
1708         }
1709         dp = args->dp;
1710         leaf = lbp->b_addr;
1711         /*
1712          * Point to the leaf entry, get data address from it.
1713          */
1714         lep = &leaf->ents[index];
1715         /*
1716          * Point to the data entry.
1717          */
1718         dep = (xfs_dir2_data_entry_t *)
1719               ((char *)dbp->b_addr +
1720                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1721         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1722         /*
1723          * Put the new inode number in, log it.
1724          */
1725         dep->inumber = cpu_to_be64(args->inumber);
1726         tp = args->trans;
1727         xfs_dir2_data_log_entry(tp, dbp, dep);
1728         xfs_dir2_leaf_check(dp, lbp);
1729         xfs_trans_brelse(tp, lbp);
1730         return 0;
1731 }
1732
1733 /*
1734  * Return index in the leaf block (lbp) which is either the first
1735  * one with this hash value, or if there are none, the insert point
1736  * for that hash value.
1737  */
1738 int                                             /* index value */
1739 xfs_dir2_leaf_search_hash(
1740         xfs_da_args_t           *args,          /* operation arguments */
1741         struct xfs_buf          *lbp)           /* leaf buffer */
1742 {
1743         xfs_dahash_t            hash=0;         /* hash from this entry */
1744         xfs_dahash_t            hashwant;       /* hash value looking for */
1745         int                     high;           /* high leaf index */
1746         int                     low;            /* low leaf index */
1747         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1748         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1749         int                     mid=0;          /* current leaf index */
1750
1751         leaf = lbp->b_addr;
1752 #ifndef __KERNEL__
1753         if (!leaf->hdr.count)
1754                 return 0;
1755 #endif
1756         /*
1757          * Note, the table cannot be empty, so we have to go through the loop.
1758          * Binary search the leaf entries looking for our hash value.
1759          */
1760         for (lep = leaf->ents, low = 0, high = be16_to_cpu(leaf->hdr.count) - 1,
1761                 hashwant = args->hashval;
1762              low <= high; ) {
1763                 mid = (low + high) >> 1;
1764                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1765                         break;
1766                 if (hash < hashwant)
1767                         low = mid + 1;
1768                 else
1769                         high = mid - 1;
1770         }
1771         /*
1772          * Found one, back up through all the equal hash values.
1773          */
1774         if (hash == hashwant) {
1775                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1776                         mid--;
1777                 }
1778         }
1779         /*
1780          * Need to point to an entry higher than ours.
1781          */
1782         else if (hash < hashwant)
1783                 mid++;
1784         return mid;
1785 }
1786
1787 /*
1788  * Trim off a trailing data block.  We know it's empty since the leaf
1789  * freespace table says so.
1790  */
1791 int                                             /* error */
1792 xfs_dir2_leaf_trim_data(
1793         xfs_da_args_t           *args,          /* operation arguments */
1794         struct xfs_buf          *lbp,           /* leaf buffer */
1795         xfs_dir2_db_t           db)             /* data block number */
1796 {
1797         __be16                  *bestsp;        /* leaf bests table */
1798         struct xfs_buf          *dbp;           /* data block buffer */
1799         xfs_inode_t             *dp;            /* incore directory inode */
1800         int                     error;          /* error return value */
1801         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1802         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1803         xfs_mount_t             *mp;            /* filesystem mount point */
1804         xfs_trans_t             *tp;            /* transaction pointer */
1805
1806         dp = args->dp;
1807         mp = dp->i_mount;
1808         tp = args->trans;
1809         /*
1810          * Read the offending data block.  We need its buffer.
1811          */
1812         error = xfs_dir2_data_read(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp);
1813         if (error)
1814                 return error;
1815
1816         leaf = lbp->b_addr;
1817         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1818
1819 #ifdef DEBUG
1820 {
1821         struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1822
1823         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
1824         ASSERT(be16_to_cpu(hdr->bestfree[0].length) ==
1825                mp->m_dirblksize - (uint)sizeof(*hdr));
1826         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1827 }
1828 #endif
1829
1830         /*
1831          * Get rid of the data block.
1832          */
1833         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1834                 ASSERT(error != ENOSPC);
1835                 xfs_trans_brelse(tp, dbp);
1836                 return error;
1837         }
1838         /*
1839          * Eliminate the last bests entry from the table.
1840          */
1841         bestsp = xfs_dir2_leaf_bests_p(ltp);
1842         be32_add_cpu(&ltp->bestcount, -1);
1843         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1844         xfs_dir2_leaf_log_tail(tp, lbp);
1845         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1846         return 0;
1847 }
1848
1849 static inline size_t
1850 xfs_dir2_leaf_size(
1851         struct xfs_dir2_leaf_hdr        *hdr,
1852         int                             counts)
1853 {
1854         int                     entries;
1855
1856         entries = be16_to_cpu(hdr->count) - be16_to_cpu(hdr->stale);
1857         return sizeof(xfs_dir2_leaf_hdr_t) +
1858             entries * sizeof(xfs_dir2_leaf_entry_t) +
1859             counts * sizeof(xfs_dir2_data_off_t) +
1860             sizeof(xfs_dir2_leaf_tail_t);
1861 }
1862
1863 /*
1864  * Convert node form directory to leaf form directory.
1865  * The root of the node form dir needs to already be a LEAFN block.
1866  * Just return if we can't do anything.
1867  */
1868 int                                             /* error */
1869 xfs_dir2_node_to_leaf(
1870         xfs_da_state_t          *state)         /* directory operation state */
1871 {
1872         xfs_da_args_t           *args;          /* operation arguments */
1873         xfs_inode_t             *dp;            /* incore directory inode */
1874         int                     error;          /* error return code */
1875         struct xfs_buf          *fbp;           /* buffer for freespace block */
1876         xfs_fileoff_t           fo;             /* freespace file offset */
1877         xfs_dir2_free_t         *free;          /* freespace structure */
1878         struct xfs_buf          *lbp;           /* buffer for leaf block */
1879         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1880         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1881         xfs_mount_t             *mp;            /* filesystem mount point */
1882         int                     rval;           /* successful free trim? */
1883         xfs_trans_t             *tp;            /* transaction pointer */
1884
1885         /*
1886          * There's more than a leaf level in the btree, so there must
1887          * be multiple leafn blocks.  Give up.
1888          */
1889         if (state->path.active > 1)
1890                 return 0;
1891         args = state->args;
1892
1893         trace_xfs_dir2_node_to_leaf(args);
1894
1895         mp = state->mp;
1896         dp = args->dp;
1897         tp = args->trans;
1898         /*
1899          * Get the last offset in the file.
1900          */
1901         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1902                 return error;
1903         }
1904         fo -= mp->m_dirblkfsbs;
1905         /*
1906          * If there are freespace blocks other than the first one,
1907          * take this opportunity to remove trailing empty freespace blocks
1908          * that may have been left behind during no-space-reservation
1909          * operations.
1910          */
1911         while (fo > mp->m_dirfreeblk) {
1912                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1913                         return error;
1914                 }
1915                 if (rval)
1916                         fo -= mp->m_dirblkfsbs;
1917                 else
1918                         return 0;
1919         }
1920         /*
1921          * Now find the block just before the freespace block.
1922          */
1923         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1924                 return error;
1925         }
1926         /*
1927          * If it's not the single leaf block, give up.
1928          */
1929         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1930                 return 0;
1931         lbp = state->path.blk[0].bp;
1932         leaf = lbp->b_addr;
1933         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1934         /*
1935          * Read the freespace block.
1936          */
1937         error = xfs_dir2_free_read(tp, dp,  mp->m_dirfreeblk, &fbp);
1938         if (error)
1939                 return error;
1940         free = fbp->b_addr;
1941         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1942         ASSERT(!free->hdr.firstdb);
1943
1944         /*
1945          * Now see if the leafn and free data will fit in a leaf1.
1946          * If not, release the buffer and give up.
1947          */
1948         if (xfs_dir2_leaf_size(&leaf->hdr, be32_to_cpu(free->hdr.nvalid)) >
1949                         mp->m_dirblksize) {
1950                 xfs_trans_brelse(tp, fbp);
1951                 return 0;
1952         }
1953
1954         /*
1955          * If the leaf has any stale entries in it, compress them out.
1956          * The compact routine will log the header.
1957          */
1958         if (be16_to_cpu(leaf->hdr.stale))
1959                 xfs_dir2_leaf_compact(args, lbp);
1960         else
1961                 xfs_dir2_leaf_log_header(tp, lbp);
1962
1963         lbp->b_ops = &xfs_dir2_leaf1_buf_ops;
1964         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAF1_MAGIC);
1965
1966         /*
1967          * Set up the leaf tail from the freespace block.
1968          */
1969         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1970         ltp->bestcount = free->hdr.nvalid;
1971         /*
1972          * Set up the leaf bests table.
1973          */
1974         memcpy(xfs_dir2_leaf_bests_p(ltp), free->bests,
1975                 be32_to_cpu(ltp->bestcount) * sizeof(xfs_dir2_data_off_t));
1976         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1977         xfs_dir2_leaf_log_tail(tp, lbp);
1978         xfs_dir2_leaf_check(dp, lbp);
1979         /*
1980          * Get rid of the freespace block.
1981          */
1982         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1983         if (error) {
1984                 /*
1985                  * This can't fail here because it can only happen when
1986                  * punching out the middle of an extent, and this is an
1987                  * isolated block.
1988                  */
1989                 ASSERT(error != ENOSPC);
1990                 return error;
1991         }
1992         fbp = NULL;
1993         /*
1994          * Now see if we can convert the single-leaf directory
1995          * down to a block form directory.
1996          * This routine always kills the dabuf for the leaf, so
1997          * eliminate it from the path.
1998          */
1999         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
2000         state->path.blk[0].bp = NULL;
2001         return error;
2002 }