Pileus Git - ~andy/linux/blob - 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         bp->b_iodone = NULL;
  67         xfs_buf_ioend(bp, 0);
  68 }
  69
  70 static void
  71 xfs_dir2_leaf1_verify(
  72         struct xfs_buf          *bp)
  73 {
  74         xfs_dir2_leaf_verify(bp, cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
  75 }
  76
  77 void
  78 xfs_dir2_leafn_verify(
  79         struct xfs_buf          *bp)
  80 {
  81         xfs_dir2_leaf_verify(bp, cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
  82 }
  83
  84 static int
  85 xfs_dir2_leaf_read(
  86         struct xfs_trans        *tp,
  87         struct xfs_inode        *dp,
  88         xfs_dablk_t             fbno,
  89         xfs_daddr_t             mappedbno,
  90         struct xfs_buf          **bpp)
  91 {
  92         return xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
  93                                         XFS_DATA_FORK, xfs_dir2_leaf1_verify);
  94 }
  95
  96 int
  97 xfs_dir2_leafn_read(
  98         struct xfs_trans        *tp,
  99         struct xfs_inode        *dp,
 100         xfs_dablk_t             fbno,
 101         xfs_daddr_t             mappedbno,
 102         struct xfs_buf          **bpp)
 103 {
 104         return xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 105                                         XFS_DATA_FORK, xfs_dir2_leafn_verify);
 106 }
 107
 108 /*
 109  * Convert a block form directory to a leaf form directory.
 110  */
 111 int                                             /* error */
 112 xfs_dir2_block_to_leaf(
 113         xfs_da_args_t           *args,          /* operation arguments */
 114         struct xfs_buf          *dbp)           /* input block's buffer */
 115 {
 116         __be16                  *bestsp;        /* leaf's bestsp entries */
 117         xfs_dablk_t             blkno;          /* leaf block's bno */
 118         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 119         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
 120         xfs_dir2_block_tail_t   *btp;           /* block's tail */
 121         xfs_inode_t             *dp;            /* incore directory inode */
 122         int                     error;          /* error return code */
 123         struct xfs_buf          *lbp;           /* leaf block's buffer */
 124         xfs_dir2_db_t           ldb;            /* leaf block's bno */
 125         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 126         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
 127         xfs_mount_t             *mp;            /* filesystem mount point */
 128         int                     needlog;        /* need to log block header */
 129         int                     needscan;       /* need to rescan bestfree */
 130         xfs_trans_t             *tp;            /* transaction pointer */
 131
 132         trace_xfs_dir2_block_to_leaf(args);
 133
 134         dp = args->dp;
 135         mp = dp->i_mount;
 136         tp = args->trans;
 137         /*
 138          * Add the leaf block to the inode.
 139          * This interface will only put blocks in the leaf/node range.
 140          * Since that's empty now, we'll get the root (block 0 in range).
 141          */
 142         if ((error = xfs_da_grow_inode(args, &blkno))) {
 143                 return error;
 144         }
 145         ldb = xfs_dir2_da_to_db(mp, blkno);
 146         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
 147         /*
 148          * Initialize the leaf block, get a buffer for it.
 149          */
 150         if ((error = xfs_dir2_leaf_init(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC))) {
 151                 return error;
 152         }
 153         ASSERT(lbp != NULL);
 154         leaf = lbp->b_addr;
 155         hdr = dbp->b_addr;
 156         xfs_dir2_data_check(dp, dbp);
 157         btp = xfs_dir2_block_tail_p(mp, hdr);
 158         blp = xfs_dir2_block_leaf_p(btp);
 159         /*
 160          * Set the counts in the leaf header.
 161          */
 162         leaf->hdr.count = cpu_to_be16(be32_to_cpu(btp->count));
 163         leaf->hdr.stale = cpu_to_be16(be32_to_cpu(btp->stale));
 164         /*
 165          * Could compact these but I think we always do the conversion
 166          * after squeezing out stale entries.
 167          */
 168         memcpy(leaf->ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
 169         xfs_dir2_leaf_log_ents(tp, lbp, 0, be16_to_cpu(leaf->hdr.count) - 1);
 170         needscan = 0;
 171         needlog = 1;
 172         /*
 173          * Make the space formerly occupied by the leaf entries and block
 174          * tail be free.
 175          */
 176         xfs_dir2_data_make_free(tp, dbp,
 177                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 178                 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
 179                                        (char *)blp),
 180                 &needlog, &needscan);
 181         /*
 182          * Fix up the block header, make it a data block.
 183          */
 184         hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 185         if (needscan)
 186                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 187         /*
 188          * Set up leaf tail and bests table.
 189          */
 190         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 191         ltp->bestcount = cpu_to_be32(1);
 192         bestsp = xfs_dir2_leaf_bests_p(ltp);
 193         bestsp[0] =  hdr->bestfree[0].length;
 194         /*
 195          * Log the data header and leaf bests table.
 196          */
 197         if (needlog)
 198                 xfs_dir2_data_log_header(tp, dbp);
 199         xfs_dir2_leaf_check(dp, lbp);
 200         xfs_dir2_data_check(dp, dbp);
 201         xfs_dir2_leaf_log_bests(tp, lbp, 0, 0);
 202         return 0;
 203 }
 204
 205 STATIC void
 206 xfs_dir2_leaf_find_stale(
 207         struct xfs_dir2_leaf    *leaf,
 208         int                     index,
 209         int                     *lowstale,
 210         int                     *highstale)
 211 {
 212         /*
 213          * Find the first stale entry before our index, if any.
 214          */
 215         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
 216                 if (leaf->ents[*lowstale].address ==
 217                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 218                         break;
 219         }
 220
 221         /*
 222          * Find the first stale entry at or after our index, if any.
 223          * Stop if the result would require moving more entries than using
 224          * lowstale.
 225          */
 226         for (*highstale = index;
 227              *highstale < be16_to_cpu(leaf->hdr.count);
 228              ++*highstale) {
 229                 if (leaf->ents[*highstale].address ==
 230                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 231                         break;
 232                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
 233                         break;
 234         }
 235 }
 236
 237 struct xfs_dir2_leaf_entry *
 238 xfs_dir2_leaf_find_entry(
 239         xfs_dir2_leaf_t         *leaf,          /* leaf structure */
 240         int                     index,          /* leaf table position */
 241         int                     compact,        /* need to compact leaves */
 242         int                     lowstale,       /* index of prev stale leaf */
 243         int                     highstale,      /* index of next stale leaf */
 244         int                     *lfloglow,      /* low leaf logging index */
 245         int                     *lfloghigh)     /* high leaf logging index */
 246 {
 247         if (!leaf->hdr.stale) {
 248                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
 249
 250                 /*
 251                  * Now we need to make room to insert the leaf entry.
 252                  *
 253                  * If there are no stale entries, just insert a hole at index.
 254                  */
 255                 lep = &leaf->ents[index];
 256                 if (index < be16_to_cpu(leaf->hdr.count))
 257                         memmove(lep + 1, lep,
 258                                 (be16_to_cpu(leaf->hdr.count) - index) *
 259                                  sizeof(*lep));
 260
 261                 /*
 262                  * Record low and high logging indices for the leaf.
 263                  */
 264                 *lfloglow = index;
 265                 *lfloghigh = be16_to_cpu(leaf->hdr.count);
 266                 be16_add_cpu(&leaf->hdr.count, 1);
 267                 return lep;
 268         }
 269
 270         /*
 271          * There are stale entries.
 272          *
 273          * We will use one of them for the new entry.  It's probably not at
 274          * the right location, so we'll have to shift some up or down first.
 275          *
 276          * If we didn't compact before, we need to find the nearest stale
 277          * entries before and after our insertion point.
 278          */
 279         if (compact == 0)
 280                 xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
 281
 282         /*
 283          * If the low one is better, use it.
 284          */
 285         if (lowstale >= 0 &&
 286             (highstale == be16_to_cpu(leaf->hdr.count) ||
 287              index - lowstale - 1 < highstale - index)) {
 288                 ASSERT(index - lowstale - 1 >= 0);
 289                 ASSERT(leaf->ents[lowstale].address ==
 290                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 291
 292                 /*
 293                  * Copy entries up to cover the stale entry and make room
 294                  * for the new entry.
 295                  */
 296                 if (index - lowstale - 1 > 0) {
 297                         memmove(&leaf->ents[lowstale],
 298                                 &leaf->ents[lowstale + 1],
 299                                 (index - lowstale - 1) *
 300                                 sizeof(xfs_dir2_leaf_entry_t));
 301                 }
 302                 *lfloglow = MIN(lowstale, *lfloglow);
 303                 *lfloghigh = MAX(index - 1, *lfloghigh);
 304                 be16_add_cpu(&leaf->hdr.stale, -1);
 305                 return &leaf->ents[index - 1];
 306         }
 307
 308         /*
 309          * The high one is better, so use that one.
 310          */
 311         ASSERT(highstale - index >= 0);
 312         ASSERT(leaf->ents[highstale].address ==
 313                cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 314
 315         /*
 316          * Copy entries down to cover the stale entry and make room for the
 317          * new entry.
 318          */
 319         if (highstale - index > 0) {
 320                 memmove(&leaf->ents[index + 1],
 321                         &leaf->ents[index],
 322                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
 323         }
 324         *lfloglow = MIN(index, *lfloglow);
 325         *lfloghigh = MAX(highstale, *lfloghigh);
 326         be16_add_cpu(&leaf->hdr.stale, -1);
 327         return &leaf->ents[index];
 328 }
 329
 330 /*
 331  * Add an entry to a leaf form directory.
 332  */
 333 int                                             /* error */
 334 xfs_dir2_leaf_addname(
 335         xfs_da_args_t           *args)          /* operation arguments */
 336 {
 337         __be16                  *bestsp;        /* freespace table in leaf */
 338         int                     compact;        /* need to compact leaves */
 339         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 340         struct xfs_buf          *dbp;           /* data block buffer */
 341         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 342         xfs_inode_t             *dp;            /* incore directory inode */
 343         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
 344         int                     error;          /* error return value */
 345         int                     grown;          /* allocated new data block */
 346         int                     highstale;      /* index of next stale leaf */
 347         int                     i;              /* temporary, index */
 348         int                     index;          /* leaf table position */
 349         struct xfs_buf          *lbp;           /* leaf's buffer */
 350         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 351         int                     length;         /* length of new entry */
 352         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
 353         int                     lfloglow;       /* low leaf logging index */
 354         int                     lfloghigh;      /* high leaf logging index */
 355         int                     lowstale;       /* index of prev stale leaf */
 356         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 357         xfs_mount_t             *mp;            /* filesystem mount point */
 358         int                     needbytes;      /* leaf block bytes needed */
 359         int                     needlog;        /* need to log data header */
 360         int                     needscan;       /* need to rescan data free */
 361         __be16                  *tagp;          /* end of data entry */
 362         xfs_trans_t             *tp;            /* transaction pointer */
 363         xfs_dir2_db_t           use_block;      /* data block number */
 364
 365         trace_xfs_dir2_leaf_addname(args);
 366
 367         dp = args->dp;
 368         tp = args->trans;
 369         mp = dp->i_mount;
 370
 371         error = xfs_dir2_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
 372         if (error)
 373                 return error;
 374
 375         /*
 376          * Look up the entry by hash value and name.
 377          * We know it's not there, our caller has already done a lookup.
 378          * So the index is of the entry to insert in front of.
 379          * But if there are dup hash values the index is of the first of those.
 380          */
 381         index = xfs_dir2_leaf_search_hash(args, lbp);
 382         leaf = lbp->b_addr;
 383         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 384         bestsp = xfs_dir2_leaf_bests_p(ltp);
 385         length = xfs_dir2_data_entsize(args->namelen);
 386         /*
 387          * See if there are any entries with the same hash value
 388          * and space in their block for the new entry.
 389          * This is good because it puts multiple same-hash value entries
 390          * in a data block, improving the lookup of those entries.
 391          */
 392         for (use_block = -1, lep = &leaf->ents[index];
 393              index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
 394              index++, lep++) {
 395                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 396                         continue;
 397                 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 398                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 399                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
 400                 if (be16_to_cpu(bestsp[i]) >= length) {
 401                         use_block = i;
 402                         break;
 403                 }
 404         }
 405         /*
 406          * Didn't find a block yet, linear search all the data blocks.
 407          */
 408         if (use_block == -1) {
 409                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 410                         /*
 411                          * Remember a block we see that's missing.
 412                          */
 413                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
 414                             use_block == -1)
 415                                 use_block = i;
 416                         else if (be16_to_cpu(bestsp[i]) >= length) {
 417                                 use_block = i;
 418                                 break;
 419                         }
 420                 }
 421         }
 422         /*
 423          * How many bytes do we need in the leaf block?
 424          */
 425         needbytes = 0;
 426         if (!leaf->hdr.stale)
 427                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
 428         if (use_block == -1)
 429                 needbytes += sizeof(xfs_dir2_data_off_t);
 430
 431         /*
 432          * Now kill use_block if it refers to a missing block, so we
 433          * can use it as an indication of allocation needed.
 434          */
 435         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
 436                 use_block = -1;
 437         /*
 438          * If we don't have enough free bytes but we can make enough
 439          * by compacting out stale entries, we'll do that.
 440          */
 441         if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <
 442                                 needbytes && be16_to_cpu(leaf->hdr.stale) > 1) {
 443                 compact = 1;
 444         }
 445         /*
 446          * Otherwise if we don't have enough free bytes we need to
 447          * convert to node form.
 448          */
 449         else if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(
 450                                                 leaf->hdr.count)] < needbytes) {
 451                 /*
 452                  * Just checking or no space reservation, give up.
 453                  */
 454                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 455                                                         args->total == 0) {
 456                         xfs_trans_brelse(tp, lbp);
 457                         return XFS_ERROR(ENOSPC);
 458                 }
 459                 /*
 460                  * Convert to node form.
 461                  */
 462                 error = xfs_dir2_leaf_to_node(args, lbp);
 463                 if (error)
 464                         return error;
 465                 /*
 466                  * Then add the new entry.
 467                  */
 468                 return xfs_dir2_node_addname(args);
 469         }
 470         /*
 471          * Otherwise it will fit without compaction.
 472          */
 473         else
 474                 compact = 0;
 475         /*
 476          * If just checking, then it will fit unless we needed to allocate
 477          * a new data block.
 478          */
 479         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 480                 xfs_trans_brelse(tp, lbp);
 481                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
 482         }
 483         /*
 484          * If no allocations are allowed, return now before we've
 485          * changed anything.
 486          */
 487         if (args->total == 0 && use_block == -1) {
 488                 xfs_trans_brelse(tp, lbp);
 489                 return XFS_ERROR(ENOSPC);
 490         }
 491         /*
 492          * Need to compact the leaf entries, removing stale ones.
 493          * Leave one stale entry behind - the one closest to our
 494          * insertion index - and we'll shift that one to our insertion
 495          * point later.
 496          */
 497         if (compact) {
 498                 xfs_dir2_leaf_compact_x1(lbp, &index, &lowstale, &highstale,
 499                         &lfloglow, &lfloghigh);
 500         }
 501         /*
 502          * There are stale entries, so we'll need log-low and log-high
 503          * impossibly bad values later.
 504          */
 505         else if (be16_to_cpu(leaf->hdr.stale)) {
 506                 lfloglow = be16_to_cpu(leaf->hdr.count);
 507                 lfloghigh = -1;
 508         }
 509         /*
 510          * If there was no data block space found, we need to allocate
 511          * a new one.
 512          */
 513         if (use_block == -1) {
 514                 /*
 515                  * Add the new data block.
 516                  */
 517                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 518                                 &use_block))) {
 519                         xfs_trans_brelse(tp, lbp);
 520                         return error;
 521                 }
 522                 /*
 523                  * Initialize the block.
 524                  */
 525                 if ((error = xfs_dir2_data_init(args, use_block, &dbp))) {
 526                         xfs_trans_brelse(tp, lbp);
 527                         return error;
 528                 }
 529                 /*
 530                  * If we're adding a new data block on the end we need to
 531                  * extend the bests table.  Copy it up one entry.
 532                  */
 533                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
 534                         bestsp--;
 535                         memmove(&bestsp[0], &bestsp[1],
 536                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 537                         be32_add_cpu(&ltp->bestcount, 1);
 538                         xfs_dir2_leaf_log_tail(tp, lbp);
 539                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
 540                 }
 541                 /*
 542                  * If we're filling in a previously empty block just log it.
 543                  */
 544                 else
 545                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 546                 hdr = dbp->b_addr;
 547                 bestsp[use_block] = hdr->bestfree[0].length;
 548                 grown = 1;
 549         } else {
 550                 /*
 551                  * Already had space in some data block.
 552                  * Just read that one in.
 553                  */
 554                 error = xfs_dir2_data_read(tp, dp,
 555                                            xfs_dir2_db_to_da(mp, use_block),
 556                                            -1, &dbp);
 557                 if (error) {
 558                         xfs_trans_brelse(tp, lbp);
 559                         return error;
 560                 }
 561                 hdr = dbp->b_addr;
 562                 grown = 0;
 563         }
 564         /*
 565          * Point to the biggest freespace in our data block.
 566          */
 567         dup = (xfs_dir2_data_unused_t *)
 568               ((char *)hdr + be16_to_cpu(hdr->bestfree[0].offset));
 569         ASSERT(be16_to_cpu(dup->length) >= length);
 570         needscan = needlog = 0;
 571         /*
 572          * Mark the initial part of our freespace in use for the new entry.
 573          */
 574         xfs_dir2_data_use_free(tp, dbp, dup,
 575                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
 576                 &needlog, &needscan);
 577         /*
 578          * Initialize our new entry (at last).
 579          */
 580         dep = (xfs_dir2_data_entry_t *)dup;
 581         dep->inumber = cpu_to_be64(args->inumber);
 582         dep->namelen = args->namelen;
 583         memcpy(dep->name, args->name, dep->namelen);
 584         tagp = xfs_dir2_data_entry_tag_p(dep);
 585         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 586         /*
 587          * Need to scan fix up the bestfree table.
 588          */
 589         if (needscan)
 590                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 591         /*
 592          * Need to log the data block's header.
 593          */
 594         if (needlog)
 595                 xfs_dir2_data_log_header(tp, dbp);
 596         xfs_dir2_data_log_entry(tp, dbp, dep);
 597         /*
 598          * If the bests table needs to be changed, do it.
 599          * Log the change unless we've already done that.
 600          */
 601         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(hdr->bestfree[0].length)) {
 602                 bestsp[use_block] = hdr->bestfree[0].length;
 603                 if (!grown)
 604                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 605         }
 606
 607         lep = xfs_dir2_leaf_find_entry(leaf, index, compact, lowstale,
 608                                        highstale, &lfloglow, &lfloghigh);
 609
 610         /*
 611          * Fill in the new leaf entry.
 612          */
 613         lep->hashval = cpu_to_be32(args->hashval);
 614         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
 615                                 be16_to_cpu(*tagp)));
 616         /*
 617          * Log the leaf fields and give up the buffers.
 618          */
 619         xfs_dir2_leaf_log_header(tp, lbp);
 620         xfs_dir2_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
 621         xfs_dir2_leaf_check(dp, lbp);
 622         xfs_dir2_data_check(dp, dbp);
 623         return 0;
 624 }
 625
 626 #ifdef DEBUG
 627 /*
 628  * Check the internal consistency of a leaf1 block.
 629  * Pop an assert if something is wrong.
 630  */
 631 STATIC void
 632 xfs_dir2_leaf_check(
 633         struct xfs_inode        *dp,            /* incore directory inode */
 634         struct xfs_buf          *bp)            /* leaf's buffer */
 635 {
 636         int                     i;              /* leaf index */
 637         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 638         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 639         xfs_mount_t             *mp;            /* filesystem mount point */
 640         int                     stale;          /* count of stale leaves */
 641
 642         leaf = bp->b_addr;
 643         mp = dp->i_mount;
 644         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
 645         /*
 646          * This value is not restrictive enough.
 647          * Should factor in the size of the bests table as well.
 648          * We can deduce a value for that from di_size.
 649          */
 650         ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
 651         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 652         /*
 653          * Leaves and bests don't overlap.
 654          */
 655         ASSERT((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <=
 656                (char *)xfs_dir2_leaf_bests_p(ltp));
 657         /*
 658          * Check hash value order, count stale entries.
 659          */
 660         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
 661                 if (i + 1 < be16_to_cpu(leaf->hdr.count))
 662                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
 663                                be32_to_cpu(leaf->ents[i + 1].hashval));
 664                 if (leaf->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 665                         stale++;
 666         }
 667         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
 668 }
 669 #endif  /* DEBUG */
 670
 671 /*
 672  * Compact out any stale entries in the leaf.
 673  * Log the header and changed leaf entries, if any.
 674  */
 675 void
 676 xfs_dir2_leaf_compact(
 677         xfs_da_args_t   *args,          /* operation arguments */
 678         struct xfs_buf  *bp)            /* leaf buffer */
 679 {
 680         int             from;           /* source leaf index */
 681         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 682         int             loglow;         /* first leaf entry to log */
 683         int             to;             /* target leaf index */
 684
 685         leaf = bp->b_addr;
 686         if (!leaf->hdr.stale) {
 687                 return;
 688         }
 689         /*
 690          * Compress out the stale entries in place.
 691          */
 692         for (from = to = 0, loglow = -1; from < be16_to_cpu(leaf->hdr.count); from++) {
 693                 if (leaf->ents[from].address ==
 694                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 695                         continue;
 696                 /*
 697                  * Only actually copy the entries that are different.
 698                  */
 699                 if (from > to) {
 700                         if (loglow == -1)
 701                                 loglow = to;
 702                         leaf->ents[to] = leaf->ents[from];
 703                 }
 704                 to++;
 705         }
 706         /*
 707          * Update and log the header, log the leaf entries.
 708          */
 709         ASSERT(be16_to_cpu(leaf->hdr.stale) == from - to);
 710         be16_add_cpu(&leaf->hdr.count, -(be16_to_cpu(leaf->hdr.stale)));
 711         leaf->hdr.stale = 0;
 712         xfs_dir2_leaf_log_header(args->trans, bp);
 713         if (loglow != -1)
 714                 xfs_dir2_leaf_log_ents(args->trans, bp, loglow, to - 1);
 715 }
 716
 717 /*
 718  * Compact the leaf entries, removing stale ones.
 719  * Leave one stale entry behind - the one closest to our
 720  * insertion index - and the caller will shift that one to our insertion
 721  * point later.
 722  * Return new insertion index, where the remaining stale entry is,
 723  * and leaf logging indices.
 724  */
 725 void
 726 xfs_dir2_leaf_compact_x1(
 727         struct xfs_buf  *bp,            /* leaf buffer */
 728         int             *indexp,        /* insertion index */
 729         int             *lowstalep,     /* out: stale entry before us */
 730         int             *highstalep,    /* out: stale entry after us */
 731         int             *lowlogp,       /* out: low log index */
 732         int             *highlogp)      /* out: high log index */
 733 {
 734         int             from;           /* source copy index */
 735         int             highstale;      /* stale entry at/after index */
 736         int             index;          /* insertion index */
 737         int             keepstale;      /* source index of kept stale */
 738         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 739         int             lowstale;       /* stale entry before index */
 740         int             newindex=0;     /* new insertion index */
 741         int             to;             /* destination copy index */
 742
 743         leaf = bp->b_addr;
 744         ASSERT(be16_to_cpu(leaf->hdr.stale) > 1);
 745         index = *indexp;
 746
 747         xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
 748
 749         /*
 750          * Pick the better of lowstale and highstale.
 751          */
 752         if (lowstale >= 0 &&
 753             (highstale == be16_to_cpu(leaf->hdr.count) ||
 754              index - lowstale <= highstale - index))
 755                 keepstale = lowstale;
 756         else
 757                 keepstale = highstale;
 758         /*
 759          * Copy the entries in place, removing all the stale entries
 760          * except keepstale.
 761          */
 762         for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
 763                 /*
 764                  * Notice the new value of index.
 765                  */
 766                 if (index == from)
 767                         newindex = to;
 768                 if (from != keepstale &&
 769                     leaf->ents[from].address ==
 770                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
 771                         if (from == to)
 772                                 *lowlogp = to;
 773                         continue;
 774                 }
 775                 /*
 776                  * Record the new keepstale value for the insertion.
 777                  */
 778                 if (from == keepstale)
 779                         lowstale = highstale = to;
 780                 /*
 781                  * Copy only the entries that have moved.
 782                  */
 783                 if (from > to)
 784                         leaf->ents[to] = leaf->ents[from];
 785                 to++;
 786         }
 787         ASSERT(from > to);
 788         /*
 789          * If the insertion point was past the last entry,
 790          * set the new insertion point accordingly.
 791          */
 792         if (index == from)
 793                 newindex = to;
 794         *indexp = newindex;
 795         /*
 796          * Adjust the leaf header values.
 797          */
 798         be16_add_cpu(&leaf->hdr.count, -(from - to));
 799         leaf->hdr.stale = cpu_to_be16(1);
 800         /*
 801          * Remember the low/high stale value only in the "right"
 802          * direction.
 803          */
 804         if (lowstale >= newindex)
 805                 lowstale = -1;
 806         else
 807                 highstale = be16_to_cpu(leaf->hdr.count);
 808         *highlogp = be16_to_cpu(leaf->hdr.count) - 1;
 809         *lowstalep = lowstale;
 810         *highstalep = highstale;
 811 }
 812
 813 struct xfs_dir2_leaf_map_info {
 814         xfs_extlen_t    map_blocks;     /* number of fsbs in map */
 815         xfs_dablk_t     map_off;        /* last mapped file offset */
 816         int             map_size;       /* total entries in *map */
 817         int             map_valid;      /* valid entries in *map */
 818         int             nmap;           /* mappings to ask xfs_bmapi */
 819         xfs_dir2_db_t   curdb;          /* db for current block */
 820         int             ra_current;     /* number of read-ahead blks */
 821         int             ra_index;       /* *map index for read-ahead */
 822         int             ra_offset;      /* map entry offset for ra */
 823         int             ra_want;        /* readahead count wanted */
 824         struct xfs_bmbt_irec map[];     /* map vector for blocks */
 825 };
 826
 827 STATIC int
 828 xfs_dir2_leaf_readbuf(
 829         struct xfs_inode        *dp,
 830         size_t                  bufsize,
 831         struct xfs_dir2_leaf_map_info *mip,
 832         xfs_dir2_off_t          *curoff,
 833         struct xfs_buf          **bpp)
 834 {
 835         struct xfs_mount        *mp = dp->i_mount;
 836         struct xfs_buf          *bp = *bpp;
 837         struct xfs_bmbt_irec    *map = mip->map;
 838         int                     error = 0;
 839         int                     length;
 840         int                     i;
 841         int                     j;
 842
 843         /*
 844          * If we have a buffer, we need to release it and
 845          * take it out of the mapping.
 846          */
 847
 848         if (bp) {
 849                 xfs_trans_brelse(NULL, bp);
 850                 bp = NULL;
 851                 mip->map_blocks -= mp->m_dirblkfsbs;
 852                 /*
 853                  * Loop to get rid of the extents for the
 854                  * directory block.
 855                  */
 856                 for (i = mp->m_dirblkfsbs; i > 0; ) {
 857                         j = min_t(int, map->br_blockcount, i);
 858                         map->br_blockcount -= j;
 859                         map->br_startblock += j;
 860                         map->br_startoff += j;
 861                         /*
 862                          * If mapping is done, pitch it from
 863                          * the table.
 864                          */
 865                         if (!map->br_blockcount && --mip->map_valid)
 866                                 memmove(&map[0], &map[1],
 867                                         sizeof(map[0]) * mip->map_valid);
 868                         i -= j;
 869                 }
 870         }
 871
 872         /*
 873          * Recalculate the readahead blocks wanted.
 874          */
 875         mip->ra_want = howmany(bufsize + mp->m_dirblksize,
 876                                mp->m_sb.sb_blocksize) - 1;
 877         ASSERT(mip->ra_want >= 0);
 878
 879         /*
 880          * If we don't have as many as we want, and we haven't
 881          * run out of data blocks, get some more mappings.
 882          */
 883         if (1 + mip->ra_want > mip->map_blocks &&
 884             mip->map_off < xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET)) {
 885                 /*
 886                  * Get more bmaps, fill in after the ones
 887                  * we already have in the table.
 888                  */
 889                 mip->nmap = mip->map_size - mip->map_valid;
 890                 error = xfs_bmapi_read(dp, mip->map_off,
 891                                 xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET) -
 892                                                                 mip->map_off,
 893                                 &map[mip->map_valid], &mip->nmap, 0);
 894
 895                 /*
 896                  * Don't know if we should ignore this or try to return an
 897                  * error.  The trouble with returning errors is that readdir
 898                  * will just stop without actually passing the error through.
 899                  */
 900                 if (error)
 901                         goto out;       /* XXX */
 902
 903                 /*
 904                  * If we got all the mappings we asked for, set the final map
 905                  * offset based on the last bmap value received.  Otherwise,
 906                  * we've reached the end.
 907                  */
 908                 if (mip->nmap == mip->map_size - mip->map_valid) {
 909                         i = mip->map_valid + mip->nmap - 1;
 910                         mip->map_off = map[i].br_startoff + map[i].br_blockcount;
 911                 } else
 912                         mip->map_off = xfs_dir2_byte_to_da(mp,
 913                                                         XFS_DIR2_LEAF_OFFSET);
 914
 915                 /*
 916                  * Look for holes in the mapping, and eliminate them.  Count up
 917                  * the valid blocks.
 918                  */
 919                 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
 920                         if (map[i].br_startblock == HOLESTARTBLOCK) {
 921                                 mip->nmap--;
 922                                 length = mip->map_valid + mip->nmap - i;
 923                                 if (length)
 924                                         memmove(&map[i], &map[i + 1],
 925                                                 sizeof(map[i]) * length);
 926                         } else {
 927                                 mip->map_blocks += map[i].br_blockcount;
 928                                 i++;
 929                         }
 930                 }
 931                 mip->map_valid += mip->nmap;
 932         }
 933
 934         /*
 935          * No valid mappings, so no more data blocks.
 936          */
 937         if (!mip->map_valid) {
 938                 *curoff = xfs_dir2_da_to_byte(mp, mip->map_off);
 939                 goto out;
 940         }
 941
 942         /*
 943          * Read the directory block starting at the first mapping.
 944          */
 945         mip->curdb = xfs_dir2_da_to_db(mp, map->br_startoff);
 946         error = xfs_dir2_data_read(NULL, dp, map->br_startoff,
 947                         map->br_blockcount >= mp->m_dirblkfsbs ?
 948                             XFS_FSB_TO_DADDR(mp, map->br_startblock) : -1, &bp);
 949
 950         /*
 951          * Should just skip over the data block instead of giving up.
 952          */
 953         if (error)
 954                 goto out;       /* XXX */
 955
 956         /*
 957          * Adjust the current amount of read-ahead: we just read a block that
 958          * was previously ra.
 959          */
 960         if (mip->ra_current)
 961                 mip->ra_current -= mp->m_dirblkfsbs;
 962
 963         /*
 964          * Do we need more readahead?
 965          */
 966         for (mip->ra_index = mip->ra_offset = i = 0;
 967              mip->ra_want > mip->ra_current && i < mip->map_blocks;
 968              i += mp->m_dirblkfsbs) {
 969                 ASSERT(mip->ra_index < mip->map_valid);
 970                 /*
 971                  * Read-ahead a contiguous directory block.
 972                  */
 973                 if (i > mip->ra_current &&
 974                     map[mip->ra_index].br_blockcount >= mp->m_dirblkfsbs) {
 975                         xfs_dir2_data_readahead(NULL, dp,
 976                                 map[mip->ra_index].br_startoff + mip->ra_offset,
 977                                 XFS_FSB_TO_DADDR(mp,
 978                                         map[mip->ra_index].br_startblock +
 979                                                         mip->ra_offset));
 980                         mip->ra_current = i;
 981                 }
 982
 983                 /*
 984                  * Read-ahead a non-contiguous directory block.  This doesn't
 985                  * use our mapping, but this is a very rare case.
 986                  */
 987                 else if (i > mip->ra_current) {
 988                         xfs_dir2_data_readahead(NULL, dp,
 989                                         map[mip->ra_index].br_startoff +
 990                                                         mip->ra_offset, -1);
 991                         mip->ra_current = i;
 992                 }
 993
 994                 /*
 995                  * Advance offset through the mapping table.
 996                  */
 997                 for (j = 0; j < mp->m_dirblkfsbs; j++) {
 998                         /*
 999                          * The rest of this extent but not more than a dir
1000                          * block.
1001                          */
1002                         length = min_t(int, mp->m_dirblkfsbs,
1003                                         map[mip->ra_index].br_blockcount -
1004                                                         mip->ra_offset);
1005                         j += length;
1006                         mip->ra_offset += length;
1007
1008                         /*
1009                          * Advance to the next mapping if this one is used up.
1010                          */
1011                         if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
1012                                 mip->ra_offset = 0;
1013                                 mip->ra_index++;
1014                         }
1015                 }
1016         }
1017
1018 out:
1019         *bpp = bp;
1020         return error;
1021 }
1022
1023 /*
1024  * Getdents (readdir) for leaf and node directories.
1025  * This reads the data blocks only, so is the same for both forms.
1026  */
1027 int                                             /* error */
1028 xfs_dir2_leaf_getdents(
1029         xfs_inode_t             *dp,            /* incore directory inode */
1030         void                    *dirent,
1031         size_t                  bufsize,
1032         xfs_off_t               *offset,
1033         filldir_t               filldir)
1034 {
1035         struct xfs_buf          *bp = NULL;     /* data block buffer */
1036         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1037         xfs_dir2_data_entry_t   *dep;           /* data entry */
1038         xfs_dir2_data_unused_t  *dup;           /* unused entry */
1039         int                     error = 0;      /* error return value */
1040         int                     length;         /* temporary length value */
1041         xfs_mount_t             *mp;            /* filesystem mount point */
1042         int                     byteoff;        /* offset in current block */
1043         xfs_dir2_off_t          curoff;         /* current overall offset */
1044         xfs_dir2_off_t          newoff;         /* new curoff after new blk */
1045         char                    *ptr = NULL;    /* pointer to current data */
1046         struct xfs_dir2_leaf_map_info *map_info;
1047
1048         /*
1049          * If the offset is at or past the largest allowed value,
1050          * give up right away.
1051          */
1052         if (*offset >= XFS_DIR2_MAX_DATAPTR)
1053                 return 0;
1054
1055         mp = dp->i_mount;
1056
1057         /*
1058          * Set up to bmap a number of blocks based on the caller's
1059          * buffer size, the directory block size, and the filesystem
1060          * block size.
1061          */
1062         length = howmany(bufsize + mp->m_dirblksize,
1063                                      mp->m_sb.sb_blocksize);
1064         map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
1065                                 (length * sizeof(struct xfs_bmbt_irec)),
1066                                KM_SLEEP);
1067         map_info->map_size = length;
1068
1069         /*
1070          * Inside the loop we keep the main offset value as a byte offset
1071          * in the directory file.
1072          */
1073         curoff = xfs_dir2_dataptr_to_byte(mp, *offset);
1074
1075         /*
1076          * Force this conversion through db so we truncate the offset
1077          * down to get the start of the data block.
1078          */
1079         map_info->map_off = xfs_dir2_db_to_da(mp,
1080                                               xfs_dir2_byte_to_db(mp, curoff));
1081
1082         /*
1083          * Loop over directory entries until we reach the end offset.
1084          * Get more blocks and readahead as necessary.
1085          */
1086         while (curoff < XFS_DIR2_LEAF_OFFSET) {
1087                 /*
1088                  * If we have no buffer, or we're off the end of the
1089                  * current buffer, need to get another one.
1090                  */
1091                 if (!bp || ptr >= (char *)bp->b_addr + mp->m_dirblksize) {
1092
1093                         error = xfs_dir2_leaf_readbuf(dp, bufsize, map_info,
1094                                                       &curoff, &bp);
1095                         if (error || !map_info->map_valid)
1096                                 break;
1097
1098                         /*
1099                          * Having done a read, we need to set a new offset.
1100                          */
1101                         newoff = xfs_dir2_db_off_to_byte(mp, map_info->curdb, 0);
1102                         /*
1103                          * Start of the current block.
1104                          */
1105                         if (curoff < newoff)
1106                                 curoff = newoff;
1107                         /*
1108                          * Make sure we're in the right block.
1109                          */
1110                         else if (curoff > newoff)
1111                                 ASSERT(xfs_dir2_byte_to_db(mp, curoff) ==
1112                                        map_info->curdb);
1113                         hdr = bp->b_addr;
1114                         xfs_dir2_data_check(dp, bp);
1115                         /*
1116                          * Find our position in the block.
1117                          */
1118                         ptr = (char *)(hdr + 1);
1119                         byteoff = xfs_dir2_byte_to_off(mp, curoff);
1120                         /*
1121                          * Skip past the header.
1122                          */
1123                         if (byteoff == 0)
1124                                 curoff += (uint)sizeof(*hdr);
1125                         /*
1126                          * Skip past entries until we reach our offset.
1127                          */
1128                         else {
1129                                 while ((char *)ptr - (char *)hdr < byteoff) {
1130                                         dup = (xfs_dir2_data_unused_t *)ptr;
1131
1132                                         if (be16_to_cpu(dup->freetag)
1133                                                   == XFS_DIR2_DATA_FREE_TAG) {
1134
1135                                                 length = be16_to_cpu(dup->length);
1136                                                 ptr += length;
1137                                                 continue;
1138                                         }
1139                                         dep = (xfs_dir2_data_entry_t *)ptr;
1140                                         length =
1141                                            xfs_dir2_data_entsize(dep->namelen);
1142                                         ptr += length;
1143                                 }
1144                                 /*
1145                                  * Now set our real offset.
1146                                  */
1147                                 curoff =
1148                                         xfs_dir2_db_off_to_byte(mp,
1149                                             xfs_dir2_byte_to_db(mp, curoff),
1150                                             (char *)ptr - (char *)hdr);
1151                                 if (ptr >= (char *)hdr + mp->m_dirblksize) {
1152                                         continue;
1153                                 }
1154                         }
1155                 }
1156                 /*
1157                  * We have a pointer to an entry.
1158                  * Is it a live one?
1159                  */
1160                 dup = (xfs_dir2_data_unused_t *)ptr;
1161                 /*
1162                  * No, it's unused, skip over it.
1163                  */
1164                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
1165                         length = be16_to_cpu(dup->length);
1166                         ptr += length;
1167                         curoff += length;
1168                         continue;
1169                 }
1170
1171                 dep = (xfs_dir2_data_entry_t *)ptr;
1172                 length = xfs_dir2_data_entsize(dep->namelen);
1173
1174                 if (filldir(dirent, (char *)dep->name, dep->namelen,
1175                             xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff,
1176                             be64_to_cpu(dep->inumber), DT_UNKNOWN))
1177                         break;
1178
1179                 /*
1180                  * Advance to next entry in the block.
1181                  */
1182                 ptr += length;
1183                 curoff += length;
1184                 /* bufsize may have just been a guess; don't go negative */
1185                 bufsize = bufsize > length ? bufsize - length : 0;
1186         }
1187
1188         /*
1189          * All done.  Set output offset value to current offset.
1190          */
1191         if (curoff > xfs_dir2_dataptr_to_byte(mp, XFS_DIR2_MAX_DATAPTR))
1192                 *offset = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
1193         else
1194                 *offset = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
1195         kmem_free(map_info);
1196         if (bp)
1197                 xfs_trans_brelse(NULL, bp);
1198         return error;
1199 }
1200
1201 /*
1202  * Initialize a new leaf block, leaf1 or leafn magic accepted.
1203  */
1204 int
1205 xfs_dir2_leaf_init(
1206         xfs_da_args_t           *args,          /* operation arguments */
1207         xfs_dir2_db_t           bno,            /* directory block number */
1208         struct xfs_buf          **bpp,          /* out: leaf buffer */
1209         int                     magic)          /* magic number for block */
1210 {
1211         struct xfs_buf          *bp;            /* leaf buffer */
1212         xfs_inode_t             *dp;            /* incore directory inode */
1213         int                     error;          /* error return code */
1214         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1215         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1216         xfs_mount_t             *mp;            /* filesystem mount point */
1217         xfs_trans_t             *tp;            /* transaction pointer */
1218
1219         dp = args->dp;
1220         ASSERT(dp != NULL);
1221         tp = args->trans;
1222         mp = dp->i_mount;
1223         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
1224                bno < XFS_DIR2_FREE_FIRSTDB(mp));
1225         /*
1226          * Get the buffer for the block.
1227          */
1228         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
1229                 XFS_DATA_FORK);
1230         if (error) {
1231                 return error;
1232         }
1233         ASSERT(bp != NULL);
1234         leaf = bp->b_addr;
1235         /*
1236          * Initialize the header.
1237          */
1238         leaf->hdr.info.magic = cpu_to_be16(magic);
1239         leaf->hdr.info.forw = 0;
1240         leaf->hdr.info.back = 0;
1241         leaf->hdr.count = 0;
1242         leaf->hdr.stale = 0;
1243         xfs_dir2_leaf_log_header(tp, bp);
1244         /*
1245          * If it's a leaf-format directory initialize the tail.
1246          * In this case our caller has the real bests table to copy into
1247          * the block.
1248          */
1249         if (magic == XFS_DIR2_LEAF1_MAGIC) {
1250                 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1251                 ltp->bestcount = 0;
1252                 xfs_dir2_leaf_log_tail(tp, bp);
1253         }
1254         *bpp = bp;
1255         return 0;
1256 }
1257
1258 /*
1259  * Log the bests entries indicated from a leaf1 block.
1260  */
1261 static void
1262 xfs_dir2_leaf_log_bests(
1263         xfs_trans_t             *tp,            /* transaction pointer */
1264         struct xfs_buf          *bp,            /* leaf buffer */
1265         int                     first,          /* first entry to log */
1266         int                     last)           /* last entry to log */
1267 {
1268         __be16                  *firstb;        /* pointer to first entry */
1269         __be16                  *lastb;         /* pointer to last entry */
1270         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1271         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1272
1273         leaf = bp->b_addr;
1274         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1275         ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1276         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1277         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1278         xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1279                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1280 }
1281
1282 /*
1283  * Log the leaf entries indicated from a leaf1 or leafn block.
1284  */
1285 void
1286 xfs_dir2_leaf_log_ents(
1287         xfs_trans_t             *tp,            /* transaction pointer */
1288         struct xfs_buf          *bp,            /* leaf buffer */
1289         int                     first,          /* first entry to log */
1290         int                     last)           /* last entry to log */
1291 {
1292         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1293         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1294         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1295
1296         leaf = bp->b_addr;
1297         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1298                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1299         firstlep = &leaf->ents[first];
1300         lastlep = &leaf->ents[last];
1301         xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1302                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1303 }
1304
1305 /*
1306  * Log the header of the leaf1 or leafn block.
1307  */
1308 void
1309 xfs_dir2_leaf_log_header(
1310         struct xfs_trans        *tp,
1311         struct xfs_buf          *bp)
1312 {
1313         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1314
1315         leaf = bp->b_addr;
1316         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1317                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1318         xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1319                 (uint)(sizeof(leaf->hdr) - 1));
1320 }
1321
1322 /*
1323  * Log the tail of the leaf1 block.
1324  */
1325 STATIC void
1326 xfs_dir2_leaf_log_tail(
1327         struct xfs_trans        *tp,
1328         struct xfs_buf          *bp)
1329 {
1330         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1331         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1332         xfs_mount_t             *mp;            /* filesystem mount point */
1333
1334         mp = tp->t_mountp;
1335         leaf = bp->b_addr;
1336         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1337         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1338         xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1339                 (uint)(mp->m_dirblksize - 1));
1340 }
1341
1342 /*
1343  * Look up the entry referred to by args in the leaf format directory.
1344  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1345  * is also used by the node-format code.
1346  */
1347 int
1348 xfs_dir2_leaf_lookup(
1349         xfs_da_args_t           *args)          /* operation arguments */
1350 {
1351         struct xfs_buf          *dbp;           /* data block buffer */
1352         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1353         xfs_inode_t             *dp;            /* incore directory inode */
1354         int                     error;          /* error return code */
1355         int                     index;          /* found entry index */
1356         struct xfs_buf          *lbp;           /* leaf buffer */
1357         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1358         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1359         xfs_trans_t             *tp;            /* transaction pointer */
1360
1361         trace_xfs_dir2_leaf_lookup(args);
1362
1363         /*
1364          * Look up name in the leaf block, returning both buffers and index.
1365          */
1366         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1367                 return error;
1368         }
1369         tp = args->trans;
1370         dp = args->dp;
1371         xfs_dir2_leaf_check(dp, lbp);
1372         leaf = lbp->b_addr;
1373         /*
1374          * Get to the leaf entry and contained data entry address.
1375          */
1376         lep = &leaf->ents[index];
1377         /*
1378          * Point to the data entry.
1379          */
1380         dep = (xfs_dir2_data_entry_t *)
1381               ((char *)dbp->b_addr +
1382                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1383         /*
1384          * Return the found inode number & CI name if appropriate
1385          */
1386         args->inumber = be64_to_cpu(dep->inumber);
1387         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1388         xfs_trans_brelse(tp, dbp);
1389         xfs_trans_brelse(tp, lbp);
1390         return XFS_ERROR(error);
1391 }
1392
1393 /*
1394  * Look up name/hash in the leaf block.
1395  * Fill in indexp with the found index, and dbpp with the data buffer.
1396  * If not found dbpp will be NULL, and ENOENT comes back.
1397  * lbpp will always be filled in with the leaf buffer unless there's an error.
1398  */
1399 static int                                      /* error */
1400 xfs_dir2_leaf_lookup_int(
1401         xfs_da_args_t           *args,          /* operation arguments */
1402         struct xfs_buf          **lbpp,         /* out: leaf buffer */
1403         int                     *indexp,        /* out: index in leaf block */
1404         struct xfs_buf          **dbpp)         /* out: data buffer */
1405 {
1406         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1407         struct xfs_buf          *dbp = NULL;    /* data buffer */
1408         xfs_dir2_data_entry_t   *dep;           /* data entry */
1409         xfs_inode_t             *dp;            /* incore directory inode */
1410         int                     error;          /* error return code */
1411         int                     index;          /* index in leaf block */
1412         struct xfs_buf          *lbp;           /* leaf buffer */
1413         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1414         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1415         xfs_mount_t             *mp;            /* filesystem mount point */
1416         xfs_dir2_db_t           newdb;          /* new data block number */
1417         xfs_trans_t             *tp;            /* transaction pointer */
1418         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1419         enum xfs_dacmp          cmp;            /* name compare result */
1420
1421         dp = args->dp;
1422         tp = args->trans;
1423         mp = dp->i_mount;
1424
1425         error = xfs_dir2_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
1426         if (error)
1427                 return error;
1428
1429         *lbpp = lbp;
1430         leaf = lbp->b_addr;
1431         xfs_dir2_leaf_check(dp, lbp);
1432         /*
1433          * Look for the first leaf entry with our hash value.
1434          */
1435         index = xfs_dir2_leaf_search_hash(args, lbp);
1436         /*
1437          * Loop over all the entries with the right hash value
1438          * looking to match the name.
1439          */
1440         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
1441                                 be32_to_cpu(lep->hashval) == args->hashval;
1442                                 lep++, index++) {
1443                 /*
1444                  * Skip over stale leaf entries.
1445                  */
1446                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1447                         continue;
1448                 /*
1449                  * Get the new data block number.
1450                  */
1451                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1452                 /*
1453                  * If it's not the same as the old data block number,
1454                  * need to pitch the old one and read the new one.
1455                  */
1456                 if (newdb != curdb) {
1457                         if (dbp)
1458                                 xfs_trans_brelse(tp, dbp);
1459                         error = xfs_dir2_data_read(tp, dp,
1460                                                    xfs_dir2_db_to_da(mp, newdb),
1461                                                    -1, &dbp);
1462                         if (error) {
1463                                 xfs_trans_brelse(tp, lbp);
1464                                 return error;
1465                         }
1466                         curdb = newdb;
1467                 }
1468                 /*
1469                  * Point to the data entry.
1470                  */
1471                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1472                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1473                 /*
1474                  * Compare name and if it's an exact match, return the index
1475                  * and buffer. If it's the first case-insensitive match, store
1476                  * the index and buffer and continue looking for an exact match.
1477                  */
1478                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1479                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1480                         args->cmpresult = cmp;
1481                         *indexp = index;
1482                         /* case exact match: return the current buffer. */
1483                         if (cmp == XFS_CMP_EXACT) {
1484                                 *dbpp = dbp;
1485                                 return 0;
1486                         }
1487                         cidb = curdb;
1488                 }
1489         }
1490         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1491         /*
1492          * Here, we can only be doing a lookup (not a rename or remove).
1493          * If a case-insensitive match was found earlier, re-read the
1494          * appropriate data block if required and return it.
1495          */
1496         if (args->cmpresult == XFS_CMP_CASE) {
1497                 ASSERT(cidb != -1);
1498                 if (cidb != curdb) {
1499                         xfs_trans_brelse(tp, dbp);
1500                         error = xfs_dir2_data_read(tp, dp,
1501                                                    xfs_dir2_db_to_da(mp, cidb),
1502                                                    -1, &dbp);
1503                         if (error) {
1504                                 xfs_trans_brelse(tp, lbp);
1505                                 return error;
1506                         }
1507                 }
1508                 *dbpp = dbp;
1509                 return 0;
1510         }
1511         /*
1512          * No match found, return ENOENT.
1513          */
1514         ASSERT(cidb == -1);
1515         if (dbp)
1516                 xfs_trans_brelse(tp, dbp);
1517         xfs_trans_brelse(tp, lbp);
1518         return XFS_ERROR(ENOENT);
1519 }
1520
1521 /*
1522  * Remove an entry from a leaf format directory.
1523  */
1524 int                                             /* error */
1525 xfs_dir2_leaf_removename(
1526         xfs_da_args_t           *args)          /* operation arguments */
1527 {
1528         __be16                  *bestsp;        /* leaf block best freespace */
1529         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1530         xfs_dir2_db_t           db;             /* data block number */
1531         struct xfs_buf          *dbp;           /* data block buffer */
1532         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1533         xfs_inode_t             *dp;            /* incore directory inode */
1534         int                     error;          /* error return code */
1535         xfs_dir2_db_t           i;              /* temporary data block # */
1536         int                     index;          /* index into leaf entries */
1537         struct xfs_buf          *lbp;           /* leaf buffer */
1538         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1539         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1540         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1541         xfs_mount_t             *mp;            /* filesystem mount point */
1542         int                     needlog;        /* need to log data header */
1543         int                     needscan;       /* need to rescan data frees */
1544         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1545         xfs_trans_t             *tp;            /* transaction pointer */
1546
1547         trace_xfs_dir2_leaf_removename(args);
1548
1549         /*
1550          * Lookup the leaf entry, get the leaf and data blocks read in.
1551          */
1552         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1553                 return error;
1554         }
1555         dp = args->dp;
1556         tp = args->trans;
1557         mp = dp->i_mount;
1558         leaf = lbp->b_addr;
1559         hdr = dbp->b_addr;
1560         xfs_dir2_data_check(dp, dbp);
1561         /*
1562          * Point to the leaf entry, use that to point to the data entry.
1563          */
1564         lep = &leaf->ents[index];
1565         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1566         dep = (xfs_dir2_data_entry_t *)
1567               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1568         needscan = needlog = 0;
1569         oldbest = be16_to_cpu(hdr->bestfree[0].length);
1570         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1571         bestsp = xfs_dir2_leaf_bests_p(ltp);
1572         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1573         /*
1574          * Mark the former data entry unused.
1575          */
1576         xfs_dir2_data_make_free(tp, dbp,
1577                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1578                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1579         /*
1580          * We just mark the leaf entry stale by putting a null in it.
1581          */
1582         be16_add_cpu(&leaf->hdr.stale, 1);
1583         xfs_dir2_leaf_log_header(tp, lbp);
1584         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1585         xfs_dir2_leaf_log_ents(tp, lbp, index, index);
1586         /*
1587          * Scan the freespace in the data block again if necessary,
1588          * log the data block header if necessary.
1589          */
1590         if (needscan)
1591                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1592         if (needlog)
1593                 xfs_dir2_data_log_header(tp, dbp);
1594         /*
1595          * If the longest freespace in the data block has changed,
1596          * put the new value in the bests table and log that.
1597          */
1598         if (be16_to_cpu(hdr->bestfree[0].length) != oldbest) {
1599                 bestsp[db] = hdr->bestfree[0].length;
1600                 xfs_dir2_leaf_log_bests(tp, lbp, db, db);
1601         }
1602         xfs_dir2_data_check(dp, dbp);
1603         /*
1604          * If the data block is now empty then get rid of the data block.
1605          */
1606         if (be16_to_cpu(hdr->bestfree[0].length) ==
1607             mp->m_dirblksize - (uint)sizeof(*hdr)) {
1608                 ASSERT(db != mp->m_dirdatablk);
1609                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1610                         /*
1611                          * Nope, can't get rid of it because it caused
1612                          * allocation of a bmap btree block to do so.
1613                          * Just go on, returning success, leaving the
1614                          * empty block in place.
1615                          */
1616                         if (error == ENOSPC && args->total == 0)
1617                                 error = 0;
1618                         xfs_dir2_leaf_check(dp, lbp);
1619                         return error;
1620                 }
1621                 dbp = NULL;
1622                 /*
1623                  * If this is the last data block then compact the
1624                  * bests table by getting rid of entries.
1625                  */
1626                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1627                         /*
1628                          * Look for the last active entry (i).
1629                          */
1630                         for (i = db - 1; i > 0; i--) {
1631                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1632                                         break;
1633                         }
1634                         /*
1635                          * Copy the table down so inactive entries at the
1636                          * end are removed.
1637                          */
1638                         memmove(&bestsp[db - i], bestsp,
1639                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1640                         be32_add_cpu(&ltp->bestcount, -(db - i));
1641                         xfs_dir2_leaf_log_tail(tp, lbp);
1642                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1643                 } else
1644                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1645         }
1646         /*
1647          * If the data block was not the first one, drop it.
1648          */
1649         else if (db != mp->m_dirdatablk)
1650                 dbp = NULL;
1651
1652         xfs_dir2_leaf_check(dp, lbp);
1653         /*
1654          * See if we can convert to block form.
1655          */
1656         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1657 }
1658
1659 /*
1660  * Replace the inode number in a leaf format directory entry.
1661  */
1662 int                                             /* error */
1663 xfs_dir2_leaf_replace(
1664         xfs_da_args_t           *args)          /* operation arguments */
1665 {
1666         struct xfs_buf          *dbp;           /* data block buffer */
1667         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1668         xfs_inode_t             *dp;            /* incore directory inode */
1669         int                     error;          /* error return code */
1670         int                     index;          /* index of leaf entry */
1671         struct xfs_buf          *lbp;           /* leaf buffer */
1672         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1673         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1674         xfs_trans_t             *tp;            /* transaction pointer */
1675
1676         trace_xfs_dir2_leaf_replace(args);
1677
1678         /*
1679          * Look up the entry.
1680          */
1681         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1682                 return error;
1683         }
1684         dp = args->dp;
1685         leaf = lbp->b_addr;
1686         /*
1687          * Point to the leaf entry, get data address from it.
1688          */
1689         lep = &leaf->ents[index];
1690         /*
1691          * Point to the data entry.
1692          */
1693         dep = (xfs_dir2_data_entry_t *)
1694               ((char *)dbp->b_addr +
1695                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1696         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1697         /*
1698          * Put the new inode number in, log it.
1699          */
1700         dep->inumber = cpu_to_be64(args->inumber);
1701         tp = args->trans;
1702         xfs_dir2_data_log_entry(tp, dbp, dep);
1703         xfs_dir2_leaf_check(dp, lbp);
1704         xfs_trans_brelse(tp, lbp);
1705         return 0;
1706 }
1707
1708 /*
1709  * Return index in the leaf block (lbp) which is either the first
1710  * one with this hash value, or if there are none, the insert point
1711  * for that hash value.
1712  */
1713 int                                             /* index value */
1714 xfs_dir2_leaf_search_hash(
1715         xfs_da_args_t           *args,          /* operation arguments */
1716         struct xfs_buf          *lbp)           /* leaf buffer */
1717 {
1718         xfs_dahash_t            hash=0;         /* hash from this entry */
1719         xfs_dahash_t            hashwant;       /* hash value looking for */
1720         int                     high;           /* high leaf index */
1721         int                     low;            /* low leaf index */
1722         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1723         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1724         int                     mid=0;          /* current leaf index */
1725
1726         leaf = lbp->b_addr;
1727 #ifndef __KERNEL__
1728         if (!leaf->hdr.count)
1729                 return 0;
1730 #endif
1731         /*
1732          * Note, the table cannot be empty, so we have to go through the loop.
1733          * Binary search the leaf entries looking for our hash value.
1734          */
1735         for (lep = leaf->ents, low = 0, high = be16_to_cpu(leaf->hdr.count) - 1,
1736                 hashwant = args->hashval;
1737              low <= high; ) {
1738                 mid = (low + high) >> 1;
1739                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1740                         break;
1741                 if (hash < hashwant)
1742                         low = mid + 1;
1743                 else
1744                         high = mid - 1;
1745         }
1746         /*
1747          * Found one, back up through all the equal hash values.
1748          */
1749         if (hash == hashwant) {
1750                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1751                         mid--;
1752                 }
1753         }
1754         /*
1755          * Need to point to an entry higher than ours.
1756          */
1757         else if (hash < hashwant)
1758                 mid++;
1759         return mid;
1760 }
1761
1762 /*
1763  * Trim off a trailing data block.  We know it's empty since the leaf
1764  * freespace table says so.
1765  */
1766 int                                             /* error */
1767 xfs_dir2_leaf_trim_data(
1768         xfs_da_args_t           *args,          /* operation arguments */
1769         struct xfs_buf          *lbp,           /* leaf buffer */
1770         xfs_dir2_db_t           db)             /* data block number */
1771 {
1772         __be16                  *bestsp;        /* leaf bests table */
1773         struct xfs_buf          *dbp;           /* data block buffer */
1774         xfs_inode_t             *dp;            /* incore directory inode */
1775         int                     error;          /* error return value */
1776         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1777         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1778         xfs_mount_t             *mp;            /* filesystem mount point */
1779         xfs_trans_t             *tp;            /* transaction pointer */
1780
1781         dp = args->dp;
1782         mp = dp->i_mount;
1783         tp = args->trans;
1784         /*
1785          * Read the offending data block.  We need its buffer.
1786          */
1787         error = xfs_dir2_data_read(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp);
1788         if (error)
1789                 return error;
1790
1791         leaf = lbp->b_addr;
1792         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1793
1794 #ifdef DEBUG
1795 {
1796         struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1797
1798         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
1799         ASSERT(be16_to_cpu(hdr->bestfree[0].length) ==
1800                mp->m_dirblksize - (uint)sizeof(*hdr));
1801         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1802 }
1803 #endif
1804
1805         /*
1806          * Get rid of the data block.
1807          */
1808         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1809                 ASSERT(error != ENOSPC);
1810                 xfs_trans_brelse(tp, dbp);
1811                 return error;
1812         }
1813         /*
1814          * Eliminate the last bests entry from the table.
1815          */
1816         bestsp = xfs_dir2_leaf_bests_p(ltp);
1817         be32_add_cpu(&ltp->bestcount, -1);
1818         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1819         xfs_dir2_leaf_log_tail(tp, lbp);
1820         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1821         return 0;
1822 }
1823
1824 static inline size_t
1825 xfs_dir2_leaf_size(
1826         struct xfs_dir2_leaf_hdr        *hdr,
1827         int                             counts)
1828 {
1829         int                     entries;
1830
1831         entries = be16_to_cpu(hdr->count) - be16_to_cpu(hdr->stale);
1832         return sizeof(xfs_dir2_leaf_hdr_t) +
1833             entries * sizeof(xfs_dir2_leaf_entry_t) +
1834             counts * sizeof(xfs_dir2_data_off_t) +
1835             sizeof(xfs_dir2_leaf_tail_t);
1836 }
1837
1838 /*
1839  * Convert node form directory to leaf form directory.
1840  * The root of the node form dir needs to already be a LEAFN block.
1841  * Just return if we can't do anything.
1842  */
1843 int                                             /* error */
1844 xfs_dir2_node_to_leaf(
1845         xfs_da_state_t          *state)         /* directory operation state */
1846 {
1847         xfs_da_args_t           *args;          /* operation arguments */
1848         xfs_inode_t             *dp;            /* incore directory inode */
1849         int                     error;          /* error return code */
1850         struct xfs_buf          *fbp;           /* buffer for freespace block */
1851         xfs_fileoff_t           fo;             /* freespace file offset */
1852         xfs_dir2_free_t         *free;          /* freespace structure */
1853         struct xfs_buf          *lbp;           /* buffer for leaf block */
1854         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1855         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1856         xfs_mount_t             *mp;            /* filesystem mount point */
1857         int                     rval;           /* successful free trim? */
1858         xfs_trans_t             *tp;            /* transaction pointer */
1859
1860         /*
1861          * There's more than a leaf level in the btree, so there must
1862          * be multiple leafn blocks.  Give up.
1863          */
1864         if (state->path.active > 1)
1865                 return 0;
1866         args = state->args;
1867
1868         trace_xfs_dir2_node_to_leaf(args);
1869
1870         mp = state->mp;
1871         dp = args->dp;
1872         tp = args->trans;
1873         /*
1874          * Get the last offset in the file.
1875          */
1876         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1877                 return error;
1878         }
1879         fo -= mp->m_dirblkfsbs;
1880         /*
1881          * If there are freespace blocks other than the first one,
1882          * take this opportunity to remove trailing empty freespace blocks
1883          * that may have been left behind during no-space-reservation
1884          * operations.
1885          */
1886         while (fo > mp->m_dirfreeblk) {
1887                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1888                         return error;
1889                 }
1890                 if (rval)
1891                         fo -= mp->m_dirblkfsbs;
1892                 else
1893                         return 0;
1894         }
1895         /*
1896          * Now find the block just before the freespace block.
1897          */
1898         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1899                 return error;
1900         }
1901         /*
1902          * If it's not the single leaf block, give up.
1903          */
1904         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1905                 return 0;
1906         lbp = state->path.blk[0].bp;
1907         leaf = lbp->b_addr;
1908         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1909         /*
1910          * Read the freespace block.
1911          */
1912         error = xfs_dir2_free_read(tp, dp,  mp->m_dirfreeblk, &fbp);
1913         if (error)
1914                 return error;
1915         free = fbp->b_addr;
1916         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1917         ASSERT(!free->hdr.firstdb);
1918
1919         /*
1920          * Now see if the leafn and free data will fit in a leaf1.
1921          * If not, release the buffer and give up.
1922          */
1923         if (xfs_dir2_leaf_size(&leaf->hdr, be32_to_cpu(free->hdr.nvalid)) >
1924                         mp->m_dirblksize) {
1925                 xfs_trans_brelse(tp, fbp);
1926                 return 0;
1927         }
1928
1929         /*
1930          * If the leaf has any stale entries in it, compress them out.
1931          * The compact routine will log the header.
1932          */
1933         if (be16_to_cpu(leaf->hdr.stale))
1934                 xfs_dir2_leaf_compact(args, lbp);
1935         else
1936                 xfs_dir2_leaf_log_header(tp, lbp);
1937         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAF1_MAGIC);
1938         /*
1939          * Set up the leaf tail from the freespace block.
1940          */
1941         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1942         ltp->bestcount = free->hdr.nvalid;
1943         /*
1944          * Set up the leaf bests table.
1945          */
1946         memcpy(xfs_dir2_leaf_bests_p(ltp), free->bests,
1947                 be32_to_cpu(ltp->bestcount) * sizeof(xfs_dir2_data_off_t));
1948         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1949         xfs_dir2_leaf_log_tail(tp, lbp);
1950         xfs_dir2_leaf_check(dp, lbp);
1951         /*
1952          * Get rid of the freespace block.
1953          */
1954         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1955         if (error) {
1956                 /*
1957                  * This can't fail here because it can only happen when
1958                  * punching out the middle of an extent, and this is an
1959                  * isolated block.
1960                  */
1961                 ASSERT(error != ENOSPC);
1962                 return error;
1963         }
1964         fbp = NULL;
1965         /*
1966          * Now see if we can convert the single-leaf directory
1967          * down to a block form directory.
1968          * This routine always kills the dabuf for the leaf, so
1969          * eliminate it from the path.
1970          */
1971         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1972         state->path.blk[0].bp = NULL;
1973         return error;
1974 }