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_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_da_btree.h"
  31 #include "xfs_bmap_btree.h"
  32 #include "xfs_attr_sf.h"
  33 #include "xfs_dir2_sf.h"
  34 #include "xfs_dinode.h"
  35 #include "xfs_inode.h"
  36 #include "xfs_bmap.h"
  37 #include "xfs_dir2_data.h"
  38 #include "xfs_dir2_leaf.h"
  39 #include "xfs_dir2_block.h"
  40 #include "xfs_dir2_node.h"
  41 #include "xfs_dir2_trace.h"
  42 #include "xfs_error.h"
  43
  44 /*
  45  * Local function declarations.
  46  */
  47 #ifdef DEBUG
  48 static void xfs_dir2_leaf_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
  49 #else
  50 #define xfs_dir2_leaf_check(dp, bp)
  51 #endif
  52 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **lbpp,
  53                                     int *indexp, xfs_dabuf_t **dbpp);
  54 static void xfs_dir2_leaf_log_bests(struct xfs_trans *tp, struct xfs_dabuf *bp,
  55                                     int first, int last);
  56 static void xfs_dir2_leaf_log_tail(struct xfs_trans *tp, struct xfs_dabuf *bp);
  57
  58
  59 /*
  60  * Convert a block form directory to a leaf form directory.
  61  */
  62 int                                             /* error */
  63 xfs_dir2_block_to_leaf(
  64         xfs_da_args_t           *args,          /* operation arguments */
  65         xfs_dabuf_t             *dbp)           /* input block's buffer */
  66 {
  67         __be16                  *bestsp;        /* leaf's bestsp entries */
  68         xfs_dablk_t             blkno;          /* leaf block's bno */
  69         xfs_dir2_block_t        *block;         /* block structure */
  70         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
  71         xfs_dir2_block_tail_t   *btp;           /* block's tail */
  72         xfs_inode_t             *dp;            /* incore directory inode */
  73         int                     error;          /* error return code */
  74         xfs_dabuf_t             *lbp;           /* leaf block's buffer */
  75         xfs_dir2_db_t           ldb;            /* leaf block's bno */
  76         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
  77         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
  78         xfs_mount_t             *mp;            /* filesystem mount point */
  79         int                     needlog;        /* need to log block header */
  80         int                     needscan;       /* need to rescan bestfree */
  81         xfs_trans_t             *tp;            /* transaction pointer */
  82
  83         xfs_dir2_trace_args_b("block_to_leaf", args, dbp);
  84         dp = args->dp;
  85         mp = dp->i_mount;
  86         tp = args->trans;
  87         /*
  88          * Add the leaf block to the inode.
  89          * This interface will only put blocks in the leaf/node range.
  90          * Since that's empty now, we'll get the root (block 0 in range).
  91          */
  92         if ((error = xfs_da_grow_inode(args, &blkno))) {
  93                 return error;
  94         }
  95         ldb = XFS_DIR2_DA_TO_DB(mp, blkno);
  96         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
  97         /*
  98          * Initialize the leaf block, get a buffer for it.
  99          */
 100         if ((error = xfs_dir2_leaf_init(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC))) {
 101                 return error;
 102         }
 103         ASSERT(lbp != NULL);
 104         leaf = lbp->data;
 105         block = dbp->data;
 106         xfs_dir2_data_check(dp, dbp);
 107         btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
 108         blp = XFS_DIR2_BLOCK_LEAF_P(btp);
 109         /*
 110          * Set the counts in the leaf header.
 111          */
 112         leaf->hdr.count = cpu_to_be16(be32_to_cpu(btp->count));
 113         leaf->hdr.stale = cpu_to_be16(be32_to_cpu(btp->stale));
 114         /*
 115          * Could compact these but I think we always do the conversion
 116          * after squeezing out stale entries.
 117          */
 118         memcpy(leaf->ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
 119         xfs_dir2_leaf_log_ents(tp, lbp, 0, be16_to_cpu(leaf->hdr.count) - 1);
 120         needscan = 0;
 121         needlog = 1;
 122         /*
 123          * Make the space formerly occupied by the leaf entries and block
 124          * tail be free.
 125          */
 126         xfs_dir2_data_make_free(tp, dbp,
 127                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)block),
 128                 (xfs_dir2_data_aoff_t)((char *)block + mp->m_dirblksize -
 129                                        (char *)blp),
 130                 &needlog, &needscan);
 131         /*
 132          * Fix up the block header, make it a data block.
 133          */
 134         block->hdr.magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 135         if (needscan)
 136                 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
 137                         NULL);
 138         /*
 139          * Set up leaf tail and bests table.
 140          */
 141         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
 142         ltp->bestcount = cpu_to_be32(1);
 143         bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
 144         bestsp[0] =  block->hdr.bestfree[0].length;
 145         /*
 146          * Log the data header and leaf bests table.
 147          */
 148         if (needlog)
 149                 xfs_dir2_data_log_header(tp, dbp);
 150         xfs_dir2_leaf_check(dp, lbp);
 151         xfs_dir2_data_check(dp, dbp);
 152         xfs_dir2_leaf_log_bests(tp, lbp, 0, 0);
 153         xfs_da_buf_done(lbp);
 154         return 0;
 155 }
 156
 157 /*
 158  * Add an entry to a leaf form directory.
 159  */
 160 int                                             /* error */
 161 xfs_dir2_leaf_addname(
 162         xfs_da_args_t           *args)          /* operation arguments */
 163 {
 164         __be16                  *bestsp;        /* freespace table in leaf */
 165         int                     compact;        /* need to compact leaves */
 166         xfs_dir2_data_t         *data;          /* data block structure */
 167         xfs_dabuf_t             *dbp;           /* data block buffer */
 168         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 169         xfs_inode_t             *dp;            /* incore directory inode */
 170         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
 171         int                     error;          /* error return value */
 172         int                     grown;          /* allocated new data block */
 173         int                     highstale;      /* index of next stale leaf */
 174         int                     i;              /* temporary, index */
 175         int                     index;          /* leaf table position */
 176         xfs_dabuf_t             *lbp;           /* leaf's buffer */
 177         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 178         int                     length;         /* length of new entry */
 179         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
 180         int                     lfloglow;       /* low leaf logging index */
 181         int                     lfloghigh;      /* high leaf logging index */
 182         int                     lowstale;       /* index of prev stale leaf */
 183         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 184         xfs_mount_t             *mp;            /* filesystem mount point */
 185         int                     needbytes;      /* leaf block bytes needed */
 186         int                     needlog;        /* need to log data header */
 187         int                     needscan;       /* need to rescan data free */
 188         __be16                  *tagp;          /* end of data entry */
 189         xfs_trans_t             *tp;            /* transaction pointer */
 190         xfs_dir2_db_t           use_block;      /* data block number */
 191
 192         xfs_dir2_trace_args("leaf_addname", args);
 193         dp = args->dp;
 194         tp = args->trans;
 195         mp = dp->i_mount;
 196         /*
 197          * Read the leaf block.
 198          */
 199         error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
 200                 XFS_DATA_FORK);
 201         if (error) {
 202                 return error;
 203         }
 204         ASSERT(lbp != NULL);
 205         /*
 206          * Look up the entry by hash value and name.
 207          * We know it's not there, our caller has already done a lookup.
 208          * So the index is of the entry to insert in front of.
 209          * But if there are dup hash values the index is of the first of those.
 210          */
 211         index = xfs_dir2_leaf_search_hash(args, lbp);
 212         leaf = lbp->data;
 213         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
 214         bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
 215         length = XFS_DIR2_DATA_ENTSIZE(args->namelen);
 216         /*
 217          * See if there are any entries with the same hash value
 218          * and space in their block for the new entry.
 219          * This is good because it puts multiple same-hash value entries
 220          * in a data block, improving the lookup of those entries.
 221          */
 222         for (use_block = -1, lep = &leaf->ents[index];
 223              index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
 224              index++, lep++) {
 225                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 226                         continue;
 227                 i = XFS_DIR2_DATAPTR_TO_DB(mp, be32_to_cpu(lep->address));
 228                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 229                 ASSERT(be16_to_cpu(bestsp[i]) != NULLDATAOFF);
 230                 if (be16_to_cpu(bestsp[i]) >= length) {
 231                         use_block = i;
 232                         break;
 233                 }
 234         }
 235         /*
 236          * Didn't find a block yet, linear search all the data blocks.
 237          */
 238         if (use_block == -1) {
 239                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 240                         /*
 241                          * Remember a block we see that's missing.
 242                          */
 243                         if (be16_to_cpu(bestsp[i]) == NULLDATAOFF && use_block == -1)
 244                                 use_block = i;
 245                         else if (be16_to_cpu(bestsp[i]) >= length) {
 246                                 use_block = i;
 247                                 break;
 248                         }
 249                 }
 250         }
 251         /*
 252          * How many bytes do we need in the leaf block?
 253          */
 254         needbytes =
 255                 (leaf->hdr.stale ? 0 : (uint)sizeof(leaf->ents[0])) +
 256                 (use_block != -1 ? 0 : (uint)sizeof(leaf->bests[0]));
 257         /*
 258          * Now kill use_block if it refers to a missing block, so we
 259          * can use it as an indication of allocation needed.
 260          */
 261         if (use_block != -1 && be16_to_cpu(bestsp[use_block]) == NULLDATAOFF)
 262                 use_block = -1;
 263         /*
 264          * If we don't have enough free bytes but we can make enough
 265          * by compacting out stale entries, we'll do that.
 266          */
 267         if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] < needbytes &&
 268             be16_to_cpu(leaf->hdr.stale) > 1) {
 269                 compact = 1;
 270         }
 271         /*
 272          * Otherwise if we don't have enough free bytes we need to
 273          * convert to node form.
 274          */
 275         else if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <
 276                  needbytes) {
 277                 /*
 278                  * Just checking or no space reservation, give up.
 279                  */
 280                 if (args->justcheck || args->total == 0) {
 281                         xfs_da_brelse(tp, lbp);
 282                         return XFS_ERROR(ENOSPC);
 283                 }
 284                 /*
 285                  * Convert to node form.
 286                  */
 287                 error = xfs_dir2_leaf_to_node(args, lbp);
 288                 xfs_da_buf_done(lbp);
 289                 if (error)
 290                         return error;
 291                 /*
 292                  * Then add the new entry.
 293                  */
 294                 return xfs_dir2_node_addname(args);
 295         }
 296         /*
 297          * Otherwise it will fit without compaction.
 298          */
 299         else
 300                 compact = 0;
 301         /*
 302          * If just checking, then it will fit unless we needed to allocate
 303          * a new data block.
 304          */
 305         if (args->justcheck) {
 306                 xfs_da_brelse(tp, lbp);
 307                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
 308         }
 309         /*
 310          * If no allocations are allowed, return now before we've
 311          * changed anything.
 312          */
 313         if (args->total == 0 && use_block == -1) {
 314                 xfs_da_brelse(tp, lbp);
 315                 return XFS_ERROR(ENOSPC);
 316         }
 317         /*
 318          * Need to compact the leaf entries, removing stale ones.
 319          * Leave one stale entry behind - the one closest to our
 320          * insertion index - and we'll shift that one to our insertion
 321          * point later.
 322          */
 323         if (compact) {
 324                 xfs_dir2_leaf_compact_x1(lbp, &index, &lowstale, &highstale,
 325                         &lfloglow, &lfloghigh);
 326         }
 327         /*
 328          * There are stale entries, so we'll need log-low and log-high
 329          * impossibly bad values later.
 330          */
 331         else if (be16_to_cpu(leaf->hdr.stale)) {
 332                 lfloglow = be16_to_cpu(leaf->hdr.count);
 333                 lfloghigh = -1;
 334         }
 335         /*
 336          * If there was no data block space found, we need to allocate
 337          * a new one.
 338          */
 339         if (use_block == -1) {
 340                 /*
 341                  * Add the new data block.
 342                  */
 343                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 344                                 &use_block))) {
 345                         xfs_da_brelse(tp, lbp);
 346                         return error;
 347                 }
 348                 /*
 349                  * Initialize the block.
 350                  */
 351                 if ((error = xfs_dir2_data_init(args, use_block, &dbp))) {
 352                         xfs_da_brelse(tp, lbp);
 353                         return error;
 354                 }
 355                 /*
 356                  * If we're adding a new data block on the end we need to
 357                  * extend the bests table.  Copy it up one entry.
 358                  */
 359                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
 360                         bestsp--;
 361                         memmove(&bestsp[0], &bestsp[1],
 362                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 363                         be32_add(&ltp->bestcount, 1);
 364                         xfs_dir2_leaf_log_tail(tp, lbp);
 365                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
 366                 }
 367                 /*
 368                  * If we're filling in a previously empty block just log it.
 369                  */
 370                 else
 371                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 372                 data = dbp->data;
 373                 bestsp[use_block] = data->hdr.bestfree[0].length;
 374                 grown = 1;
 375         }
 376         /*
 377          * Already had space in some data block.
 378          * Just read that one in.
 379          */
 380         else {
 381                 if ((error =
 382                     xfs_da_read_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, use_block),
 383                             -1, &dbp, XFS_DATA_FORK))) {
 384                         xfs_da_brelse(tp, lbp);
 385                         return error;
 386                 }
 387                 data = dbp->data;
 388                 grown = 0;
 389         }
 390         xfs_dir2_data_check(dp, dbp);
 391         /*
 392          * Point to the biggest freespace in our data block.
 393          */
 394         dup = (xfs_dir2_data_unused_t *)
 395               ((char *)data + be16_to_cpu(data->hdr.bestfree[0].offset));
 396         ASSERT(be16_to_cpu(dup->length) >= length);
 397         needscan = needlog = 0;
 398         /*
 399          * Mark the initial part of our freespace in use for the new entry.
 400          */
 401         xfs_dir2_data_use_free(tp, dbp, dup,
 402                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)data), length,
 403                 &needlog, &needscan);
 404         /*
 405          * Initialize our new entry (at last).
 406          */
 407         dep = (xfs_dir2_data_entry_t *)dup;
 408         dep->inumber = cpu_to_be64(args->inumber);
 409         dep->namelen = args->namelen;
 410         memcpy(dep->name, args->name, dep->namelen);
 411         tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
 412         *tagp = cpu_to_be16((char *)dep - (char *)data);
 413         /*
 414          * Need to scan fix up the bestfree table.
 415          */
 416         if (needscan)
 417                 xfs_dir2_data_freescan(mp, data, &needlog, NULL);
 418         /*
 419          * Need to log the data block's header.
 420          */
 421         if (needlog)
 422                 xfs_dir2_data_log_header(tp, dbp);
 423         xfs_dir2_data_log_entry(tp, dbp, dep);
 424         /*
 425          * If the bests table needs to be changed, do it.
 426          * Log the change unless we've already done that.
 427          */
 428         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(data->hdr.bestfree[0].length)) {
 429                 bestsp[use_block] = data->hdr.bestfree[0].length;
 430                 if (!grown)
 431                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 432         }
 433         /*
 434          * Now we need to make room to insert the leaf entry.
 435          * If there are no stale entries, we just insert a hole at index.
 436          */
 437         if (!leaf->hdr.stale) {
 438                 /*
 439                  * lep is still good as the index leaf entry.
 440                  */
 441                 if (index < be16_to_cpu(leaf->hdr.count))
 442                         memmove(lep + 1, lep,
 443                                 (be16_to_cpu(leaf->hdr.count) - index) * sizeof(*lep));
 444                 /*
 445                  * Record low and high logging indices for the leaf.
 446                  */
 447                 lfloglow = index;
 448                 lfloghigh = be16_to_cpu(leaf->hdr.count);
 449                 be16_add(&leaf->hdr.count, 1);
 450         }
 451         /*
 452          * There are stale entries.
 453          * We will use one of them for the new entry.
 454          * It's probably not at the right location, so we'll have to
 455          * shift some up or down first.
 456          */
 457         else {
 458                 /*
 459                  * If we didn't compact before, we need to find the nearest
 460                  * stale entries before and after our insertion point.
 461                  */
 462                 if (compact == 0) {
 463                         /*
 464                          * Find the first stale entry before the insertion
 465                          * point, if any.
 466                          */
 467                         for (lowstale = index - 1;
 468                              lowstale >= 0 &&
 469                                 be32_to_cpu(leaf->ents[lowstale].address) !=
 470                                 XFS_DIR2_NULL_DATAPTR;
 471                              lowstale--)
 472                                 continue;
 473                         /*
 474                          * Find the next stale entry at or after the insertion
 475                          * point, if any.   Stop if we go so far that the
 476                          * lowstale entry would be better.
 477                          */
 478                         for (highstale = index;
 479                              highstale < be16_to_cpu(leaf->hdr.count) &&
 480                                 be32_to_cpu(leaf->ents[highstale].address) !=
 481                                 XFS_DIR2_NULL_DATAPTR &&
 482                                 (lowstale < 0 ||
 483                                  index - lowstale - 1 >= highstale - index);
 484                              highstale++)
 485                                 continue;
 486                 }
 487                 /*
 488                  * If the low one is better, use it.
 489                  */
 490                 if (lowstale >= 0 &&
 491                     (highstale == be16_to_cpu(leaf->hdr.count) ||
 492                      index - lowstale - 1 < highstale - index)) {
 493                         ASSERT(index - lowstale - 1 >= 0);
 494                         ASSERT(be32_to_cpu(leaf->ents[lowstale].address) ==
 495                                XFS_DIR2_NULL_DATAPTR);
 496                         /*
 497                          * Copy entries up to cover the stale entry
 498                          * and make room for the new entry.
 499                          */
 500                         if (index - lowstale - 1 > 0)
 501                                 memmove(&leaf->ents[lowstale],
 502                                         &leaf->ents[lowstale + 1],
 503                                         (index - lowstale - 1) * sizeof(*lep));
 504                         lep = &leaf->ents[index - 1];
 505                         lfloglow = MIN(lowstale, lfloglow);
 506                         lfloghigh = MAX(index - 1, lfloghigh);
 507                 }
 508                 /*
 509                  * The high one is better, so use that one.
 510                  */
 511                 else {
 512                         ASSERT(highstale - index >= 0);
 513                         ASSERT(be32_to_cpu(leaf->ents[highstale].address) ==
 514                                XFS_DIR2_NULL_DATAPTR);
 515                         /*
 516                          * Copy entries down to cover the stale entry
 517                          * and make room for the new entry.
 518                          */
 519                         if (highstale - index > 0)
 520                                 memmove(&leaf->ents[index + 1],
 521                                         &leaf->ents[index],
 522                                         (highstale - index) * sizeof(*lep));
 523                         lep = &leaf->ents[index];
 524                         lfloglow = MIN(index, lfloglow);
 525                         lfloghigh = MAX(highstale, lfloghigh);
 526                 }
 527                 be16_add(&leaf->hdr.stale, -1);
 528         }
 529         /*
 530          * Fill in the new leaf entry.
 531          */
 532         lep->hashval = cpu_to_be32(args->hashval);
 533         lep->address = cpu_to_be32(XFS_DIR2_DB_OFF_TO_DATAPTR(mp, use_block,
 534                                 be16_to_cpu(*tagp)));
 535         /*
 536          * Log the leaf fields and give up the buffers.
 537          */
 538         xfs_dir2_leaf_log_header(tp, lbp);
 539         xfs_dir2_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
 540         xfs_dir2_leaf_check(dp, lbp);
 541         xfs_da_buf_done(lbp);
 542         xfs_dir2_data_check(dp, dbp);
 543         xfs_da_buf_done(dbp);
 544         return 0;
 545 }
 546
 547 #ifdef DEBUG
 548 /*
 549  * Check the internal consistency of a leaf1 block.
 550  * Pop an assert if something is wrong.
 551  */
 552 void
 553 xfs_dir2_leaf_check(
 554         xfs_inode_t             *dp,            /* incore directory inode */
 555         xfs_dabuf_t             *bp)            /* leaf's buffer */
 556 {
 557         int                     i;              /* leaf index */
 558         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 559         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 560         xfs_mount_t             *mp;            /* filesystem mount point */
 561         int                     stale;          /* count of stale leaves */
 562
 563         leaf = bp->data;
 564         mp = dp->i_mount;
 565         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
 566         /*
 567          * This value is not restrictive enough.
 568          * Should factor in the size of the bests table as well.
 569          * We can deduce a value for that from di_size.
 570          */
 571         ASSERT(be16_to_cpu(leaf->hdr.count) <= XFS_DIR2_MAX_LEAF_ENTS(mp));
 572         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
 573         /*
 574          * Leaves and bests don't overlap.
 575          */
 576         ASSERT((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <=
 577                (char *)XFS_DIR2_LEAF_BESTS_P(ltp));
 578         /*
 579          * Check hash value order, count stale entries.
 580          */
 581         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
 582                 if (i + 1 < be16_to_cpu(leaf->hdr.count))
 583                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
 584                                be32_to_cpu(leaf->ents[i + 1].hashval));
 585                 if (be32_to_cpu(leaf->ents[i].address) == XFS_DIR2_NULL_DATAPTR)
 586                         stale++;
 587         }
 588         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
 589 }
 590 #endif  /* DEBUG */
 591
 592 /*
 593  * Compact out any stale entries in the leaf.
 594  * Log the header and changed leaf entries, if any.
 595  */
 596 void
 597 xfs_dir2_leaf_compact(
 598         xfs_da_args_t   *args,          /* operation arguments */
 599         xfs_dabuf_t     *bp)            /* leaf buffer */
 600 {
 601         int             from;           /* source leaf index */
 602         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 603         int             loglow;         /* first leaf entry to log */
 604         int             to;             /* target leaf index */
 605
 606         leaf = bp->data;
 607         if (!leaf->hdr.stale) {
 608                 return;
 609         }
 610         /*
 611          * Compress out the stale entries in place.
 612          */
 613         for (from = to = 0, loglow = -1; from < be16_to_cpu(leaf->hdr.count); from++) {
 614                 if (be32_to_cpu(leaf->ents[from].address) == XFS_DIR2_NULL_DATAPTR)
 615                         continue;
 616                 /*
 617                  * Only actually copy the entries that are different.
 618                  */
 619                 if (from > to) {
 620                         if (loglow == -1)
 621                                 loglow = to;
 622                         leaf->ents[to] = leaf->ents[from];
 623                 }
 624                 to++;
 625         }
 626         /*
 627          * Update and log the header, log the leaf entries.
 628          */
 629         ASSERT(be16_to_cpu(leaf->hdr.stale) == from - to);
 630         be16_add(&leaf->hdr.count, -(be16_to_cpu(leaf->hdr.stale)));
 631         leaf->hdr.stale = 0;
 632         xfs_dir2_leaf_log_header(args->trans, bp);
 633         if (loglow != -1)
 634                 xfs_dir2_leaf_log_ents(args->trans, bp, loglow, to - 1);
 635 }
 636
 637 /*
 638  * Compact the leaf entries, removing stale ones.
 639  * Leave one stale entry behind - the one closest to our
 640  * insertion index - and the caller will shift that one to our insertion
 641  * point later.
 642  * Return new insertion index, where the remaining stale entry is,
 643  * and leaf logging indices.
 644  */
 645 void
 646 xfs_dir2_leaf_compact_x1(
 647         xfs_dabuf_t     *bp,            /* leaf buffer */
 648         int             *indexp,        /* insertion index */
 649         int             *lowstalep,     /* out: stale entry before us */
 650         int             *highstalep,    /* out: stale entry after us */
 651         int             *lowlogp,       /* out: low log index */
 652         int             *highlogp)      /* out: high log index */
 653 {
 654         int             from;           /* source copy index */
 655         int             highstale;      /* stale entry at/after index */
 656         int             index;          /* insertion index */
 657         int             keepstale;      /* source index of kept stale */
 658         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 659         int             lowstale;       /* stale entry before index */
 660         int             newindex=0;     /* new insertion index */
 661         int             to;             /* destination copy index */
 662
 663         leaf = bp->data;
 664         ASSERT(be16_to_cpu(leaf->hdr.stale) > 1);
 665         index = *indexp;
 666         /*
 667          * Find the first stale entry before our index, if any.
 668          */
 669         for (lowstale = index - 1;
 670              lowstale >= 0 &&
 671                 be32_to_cpu(leaf->ents[lowstale].address) != XFS_DIR2_NULL_DATAPTR;
 672              lowstale--)
 673                 continue;
 674         /*
 675          * Find the first stale entry at or after our index, if any.
 676          * Stop if the answer would be worse than lowstale.
 677          */
 678         for (highstale = index;
 679              highstale < be16_to_cpu(leaf->hdr.count) &&
 680                 be32_to_cpu(leaf->ents[highstale].address) != XFS_DIR2_NULL_DATAPTR &&
 681                 (lowstale < 0 || index - lowstale > highstale - index);
 682              highstale++)
 683                 continue;
 684         /*
 685          * Pick the better of lowstale and highstale.
 686          */
 687         if (lowstale >= 0 &&
 688             (highstale == be16_to_cpu(leaf->hdr.count) ||
 689              index - lowstale <= highstale - index))
 690                 keepstale = lowstale;
 691         else
 692                 keepstale = highstale;
 693         /*
 694          * Copy the entries in place, removing all the stale entries
 695          * except keepstale.
 696          */
 697         for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
 698                 /*
 699                  * Notice the new value of index.
 700                  */
 701                 if (index == from)
 702                         newindex = to;
 703                 if (from != keepstale &&
 704                     be32_to_cpu(leaf->ents[from].address) == XFS_DIR2_NULL_DATAPTR) {
 705                         if (from == to)
 706                                 *lowlogp = to;
 707                         continue;
 708                 }
 709                 /*
 710                  * Record the new keepstale value for the insertion.
 711                  */
 712                 if (from == keepstale)
 713                         lowstale = highstale = to;
 714                 /*
 715                  * Copy only the entries that have moved.
 716                  */
 717                 if (from > to)
 718                         leaf->ents[to] = leaf->ents[from];
 719                 to++;
 720         }
 721         ASSERT(from > to);
 722         /*
 723          * If the insertion point was past the last entry,
 724          * set the new insertion point accordingly.
 725          */
 726         if (index == from)
 727                 newindex = to;
 728         *indexp = newindex;
 729         /*
 730          * Adjust the leaf header values.
 731          */
 732         be16_add(&leaf->hdr.count, -(from - to));
 733         leaf->hdr.stale = cpu_to_be16(1);
 734         /*
 735          * Remember the low/high stale value only in the "right"
 736          * direction.
 737          */
 738         if (lowstale >= newindex)
 739                 lowstale = -1;
 740         else
 741                 highstale = be16_to_cpu(leaf->hdr.count);
 742         *highlogp = be16_to_cpu(leaf->hdr.count) - 1;
 743         *lowstalep = lowstale;
 744         *highstalep = highstale;
 745 }
 746
 747 /*
 748  * Getdents (readdir) for leaf and node directories.
 749  * This reads the data blocks only, so is the same for both forms.
 750  */
 751 int                                             /* error */
 752 xfs_dir2_leaf_getdents(
 753         xfs_trans_t             *tp,            /* transaction pointer */
 754         xfs_inode_t             *dp,            /* incore directory inode */
 755         uio_t                   *uio,           /* I/O control & vectors */
 756         int                     *eofp,          /* out: reached end of dir */
 757         xfs_dirent_t            *dbp,           /* caller's buffer */
 758         xfs_dir2_put_t          put)            /* ABI formatting routine */
 759 {
 760         xfs_dabuf_t             *bp;            /* data block buffer */
 761         int                     byteoff;        /* offset in current block */
 762         xfs_dir2_db_t           curdb;          /* db for current block */
 763         xfs_dir2_off_t          curoff;         /* current overall offset */
 764         xfs_dir2_data_t         *data;          /* data block structure */
 765         xfs_dir2_data_entry_t   *dep;           /* data entry */
 766         xfs_dir2_data_unused_t  *dup;           /* unused entry */
 767         int                     eof;            /* reached end of directory */
 768         int                     error = 0;      /* error return value */
 769         int                     i;              /* temporary loop index */
 770         int                     j;              /* temporary loop index */
 771         int                     length;         /* temporary length value */
 772         xfs_bmbt_irec_t         *map;           /* map vector for blocks */
 773         xfs_extlen_t            map_blocks;     /* number of fsbs in map */
 774         xfs_dablk_t             map_off;        /* last mapped file offset */
 775         int                     map_size;       /* total entries in *map */
 776         int                     map_valid;      /* valid entries in *map */
 777         xfs_mount_t             *mp;            /* filesystem mount point */
 778         xfs_dir2_off_t          newoff;         /* new curoff after new blk */
 779         int                     nmap;           /* mappings to ask xfs_bmapi */
 780         xfs_dir2_put_args_t     *p;             /* formatting arg bundle */
 781         char                    *ptr = NULL;    /* pointer to current data */
 782         int                     ra_current;     /* number of read-ahead blks */
 783         int                     ra_index;       /* *map index for read-ahead */
 784         int                     ra_offset;      /* map entry offset for ra */
 785         int                     ra_want;        /* readahead count wanted */
 786
 787         /*
 788          * If the offset is at or past the largest allowed value,
 789          * give up right away, return eof.
 790          */
 791         if (uio->uio_offset >= XFS_DIR2_MAX_DATAPTR) {
 792                 *eofp = 1;
 793                 return 0;
 794         }
 795         mp = dp->i_mount;
 796         /*
 797          * Setup formatting arguments.
 798          */
 799         p = kmem_alloc(sizeof(*p), KM_SLEEP);
 800         p->dbp = dbp;
 801         p->put = put;
 802         p->uio = uio;
 803         /*
 804          * Set up to bmap a number of blocks based on the caller's
 805          * buffer size, the directory block size, and the filesystem
 806          * block size.
 807          */
 808         map_size =
 809                 howmany(uio->uio_resid + mp->m_dirblksize,
 810                         mp->m_sb.sb_blocksize);
 811         map = kmem_alloc(map_size * sizeof(*map), KM_SLEEP);
 812         map_valid = ra_index = ra_offset = ra_current = map_blocks = 0;
 813         bp = NULL;
 814         eof = 1;
 815         /*
 816          * Inside the loop we keep the main offset value as a byte offset
 817          * in the directory file.
 818          */
 819         curoff = XFS_DIR2_DATAPTR_TO_BYTE(mp, uio->uio_offset);
 820         /*
 821          * Force this conversion through db so we truncate the offset
 822          * down to get the start of the data block.
 823          */
 824         map_off = XFS_DIR2_DB_TO_DA(mp, XFS_DIR2_BYTE_TO_DB(mp, curoff));
 825         /*
 826          * Loop over directory entries until we reach the end offset.
 827          * Get more blocks and readahead as necessary.
 828          */
 829         while (curoff < XFS_DIR2_LEAF_OFFSET) {
 830                 /*
 831                  * If we have no buffer, or we're off the end of the
 832                  * current buffer, need to get another one.
 833                  */
 834                 if (!bp || ptr >= (char *)bp->data + mp->m_dirblksize) {
 835                         /*
 836                          * If we have a buffer, we need to release it and
 837                          * take it out of the mapping.
 838                          */
 839                         if (bp) {
 840                                 xfs_da_brelse(tp, bp);
 841                                 bp = NULL;
 842                                 map_blocks -= mp->m_dirblkfsbs;
 843                                 /*
 844                                  * Loop to get rid of the extents for the
 845                                  * directory block.
 846                                  */
 847                                 for (i = mp->m_dirblkfsbs; i > 0; ) {
 848                                         j = MIN((int)map->br_blockcount, i);
 849                                         map->br_blockcount -= j;
 850                                         map->br_startblock += j;
 851                                         map->br_startoff += j;
 852                                         /*
 853                                          * If mapping is done, pitch it from
 854                                          * the table.
 855                                          */
 856                                         if (!map->br_blockcount && --map_valid)
 857                                                 memmove(&map[0], &map[1],
 858                                                         sizeof(map[0]) *
 859                                                         map_valid);
 860                                         i -= j;
 861                                 }
 862                         }
 863                         /*
 864                          * Recalculate the readahead blocks wanted.
 865                          */
 866                         ra_want = howmany(uio->uio_resid + mp->m_dirblksize,
 867                                           mp->m_sb.sb_blocksize) - 1;
 868                         /*
 869                          * If we don't have as many as we want, and we haven't
 870                          * run out of data blocks, get some more mappings.
 871                          */
 872                         if (1 + ra_want > map_blocks &&
 873                             map_off <
 874                             XFS_DIR2_BYTE_TO_DA(mp, XFS_DIR2_LEAF_OFFSET)) {
 875                                 /*
 876                                  * Get more bmaps, fill in after the ones
 877                                  * we already have in the table.
 878                                  */
 879                                 nmap = map_size - map_valid;
 880                                 error = xfs_bmapi(tp, dp,
 881                                         map_off,
 882                                         XFS_DIR2_BYTE_TO_DA(mp,
 883                                                 XFS_DIR2_LEAF_OFFSET) - map_off,
 884                                         XFS_BMAPI_METADATA, NULL, 0,
 885                                         &map[map_valid], &nmap, NULL, NULL);
 886                                 /*
 887                                  * Don't know if we should ignore this or
 888                                  * try to return an error.
 889                                  * The trouble with returning errors
 890                                  * is that readdir will just stop without
 891                                  * actually passing the error through.
 892                                  */
 893                                 if (error)
 894                                         break;  /* XXX */
 895                                 /*
 896                                  * If we got all the mappings we asked for,
 897                                  * set the final map offset based on the
 898                                  * last bmap value received.
 899                                  * Otherwise, we've reached the end.
 900                                  */
 901                                 if (nmap == map_size - map_valid)
 902                                         map_off =
 903                                         map[map_valid + nmap - 1].br_startoff +
 904                                         map[map_valid + nmap - 1].br_blockcount;
 905                                 else
 906                                         map_off =
 907                                                 XFS_DIR2_BYTE_TO_DA(mp,
 908                                                         XFS_DIR2_LEAF_OFFSET);
 909                                 /*
 910                                  * Look for holes in the mapping, and
 911                                  * eliminate them.  Count up the valid blocks.
 912                                  */
 913                                 for (i = map_valid; i < map_valid + nmap; ) {
 914                                         if (map[i].br_startblock ==
 915                                             HOLESTARTBLOCK) {
 916                                                 nmap--;
 917                                                 length = map_valid + nmap - i;
 918                                                 if (length)
 919                                                         memmove(&map[i],
 920                                                                 &map[i + 1],
 921                                                                 sizeof(map[i]) *
 922                                                                 length);
 923                                         } else {
 924                                                 map_blocks +=
 925                                                         map[i].br_blockcount;
 926                                                 i++;
 927                                         }
 928                                 }
 929                                 map_valid += nmap;
 930                         }
 931                         /*
 932                          * No valid mappings, so no more data blocks.
 933                          */
 934                         if (!map_valid) {
 935                                 curoff = XFS_DIR2_DA_TO_BYTE(mp, map_off);
 936                                 break;
 937                         }
 938                         /*
 939                          * Read the directory block starting at the first
 940                          * mapping.
 941                          */
 942                         curdb = XFS_DIR2_DA_TO_DB(mp, map->br_startoff);
 943                         error = xfs_da_read_buf(tp, dp, map->br_startoff,
 944                                 map->br_blockcount >= mp->m_dirblkfsbs ?
 945                                     XFS_FSB_TO_DADDR(mp, map->br_startblock) :
 946                                     -1,
 947                                 &bp, XFS_DATA_FORK);
 948                         /*
 949                          * Should just skip over the data block instead
 950                          * of giving up.
 951                          */
 952                         if (error)
 953                                 break;  /* XXX */
 954                         /*
 955                          * Adjust the current amount of read-ahead: we just
 956                          * read a block that was previously ra.
 957                          */
 958                         if (ra_current)
 959                                 ra_current -= mp->m_dirblkfsbs;
 960                         /*
 961                          * Do we need more readahead?
 962                          */
 963                         for (ra_index = ra_offset = i = 0;
 964                              ra_want > ra_current && i < map_blocks;
 965                              i += mp->m_dirblkfsbs) {
 966                                 ASSERT(ra_index < map_valid);
 967                                 /*
 968                                  * Read-ahead a contiguous directory block.
 969                                  */
 970                                 if (i > ra_current &&
 971                                     map[ra_index].br_blockcount >=
 972                                     mp->m_dirblkfsbs) {
 973                                         xfs_baread(mp->m_ddev_targp,
 974                                                 XFS_FSB_TO_DADDR(mp,
 975                                                    map[ra_index].br_startblock +
 976                                                    ra_offset),
 977                                                 (int)BTOBB(mp->m_dirblksize));
 978                                         ra_current = i;
 979                                 }
 980                                 /*
 981                                  * Read-ahead a non-contiguous directory block.
 982                                  * This doesn't use our mapping, but this
 983                                  * is a very rare case.
 984                                  */
 985                                 else if (i > ra_current) {
 986                                         (void)xfs_da_reada_buf(tp, dp,
 987                                                 map[ra_index].br_startoff +
 988                                                 ra_offset, XFS_DATA_FORK);
 989                                         ra_current = i;
 990                                 }
 991                                 /*
 992                                  * Advance offset through the mapping table.
 993                                  */
 994                                 for (j = 0; j < mp->m_dirblkfsbs; j++) {
 995                                         /*
 996                                          * The rest of this extent but not
 997                                          * more than a dir block.
 998                                          */
 999                                         length = MIN(mp->m_dirblkfsbs,
1000                                                 (int)(map[ra_index].br_blockcount -
1001                                                 ra_offset));
1002                                         j += length;
1003                                         ra_offset += length;
1004                                         /*
1005                                          * Advance to the next mapping if
1006                                          * this one is used up.
1007                                          */
1008                                         if (ra_offset ==
1009                                             map[ra_index].br_blockcount) {
1010                                                 ra_offset = 0;
1011                                                 ra_index++;
1012                                         }
1013                                 }
1014                         }
1015                         /*
1016                          * Having done a read, we need to set a new offset.
1017                          */
1018                         newoff = XFS_DIR2_DB_OFF_TO_BYTE(mp, curdb, 0);
1019                         /*
1020                          * Start of the current block.
1021                          */
1022                         if (curoff < newoff)
1023                                 curoff = newoff;
1024                         /*
1025                          * Make sure we're in the right block.
1026                          */
1027                         else if (curoff > newoff)
1028                                 ASSERT(XFS_DIR2_BYTE_TO_DB(mp, curoff) ==
1029                                        curdb);
1030                         data = bp->data;
1031                         xfs_dir2_data_check(dp, bp);
1032                         /*
1033                          * Find our position in the block.
1034                          */
1035                         ptr = (char *)&data->u;
1036                         byteoff = XFS_DIR2_BYTE_TO_OFF(mp, curoff);
1037                         /*
1038                          * Skip past the header.
1039                          */
1040                         if (byteoff == 0)
1041                                 curoff += (uint)sizeof(data->hdr);
1042                         /*
1043                          * Skip past entries until we reach our offset.
1044                          */
1045                         else {
1046                                 while ((char *)ptr - (char *)data < byteoff) {
1047                                         dup = (xfs_dir2_data_unused_t *)ptr;
1048
1049                                         if (be16_to_cpu(dup->freetag)
1050                                                   == XFS_DIR2_DATA_FREE_TAG) {
1051
1052                                                 length = be16_to_cpu(dup->length);
1053                                                 ptr += length;
1054                                                 continue;
1055                                         }
1056                                         dep = (xfs_dir2_data_entry_t *)ptr;
1057                                         length =
1058                                            XFS_DIR2_DATA_ENTSIZE(dep->namelen);
1059                                         ptr += length;
1060                                 }
1061                                 /*
1062                                  * Now set our real offset.
1063                                  */
1064                                 curoff =
1065                                         XFS_DIR2_DB_OFF_TO_BYTE(mp,
1066                                             XFS_DIR2_BYTE_TO_DB(mp, curoff),
1067                                             (char *)ptr - (char *)data);
1068                                 if (ptr >= (char *)data + mp->m_dirblksize) {
1069                                         continue;
1070                                 }
1071                         }
1072                 }
1073                 /*
1074                  * We have a pointer to an entry.
1075                  * Is it a live one?
1076                  */
1077                 dup = (xfs_dir2_data_unused_t *)ptr;
1078                 /*
1079                  * No, it's unused, skip over it.
1080                  */
1081                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
1082                         length = be16_to_cpu(dup->length);
1083                         ptr += length;
1084                         curoff += length;
1085                         continue;
1086                 }
1087
1088                 /*
1089                  * Copy the entry into the putargs, and try formatting it.
1090                  */
1091                 dep = (xfs_dir2_data_entry_t *)ptr;
1092
1093                 p->namelen = dep->namelen;
1094
1095                 length = XFS_DIR2_DATA_ENTSIZE(p->namelen);
1096
1097                 p->cook = XFS_DIR2_BYTE_TO_DATAPTR(mp, curoff + length);
1098
1099                 p->ino = be64_to_cpu(dep->inumber);
1100 #if XFS_BIG_INUMS
1101                 p->ino += mp->m_inoadd;
1102 #endif
1103                 p->name = (char *)dep->name;
1104
1105                 error = p->put(p);
1106
1107                 /*
1108                  * Won't fit.  Return to caller.
1109                  */
1110                 if (!p->done) {
1111                         eof = 0;
1112                         break;
1113                 }
1114                 /*
1115                  * Advance to next entry in the block.
1116                  */
1117                 ptr += length;
1118                 curoff += length;
1119         }
1120
1121         /*
1122          * All done.  Set output offset value to current offset.
1123          */
1124         *eofp = eof;
1125         if (curoff > XFS_DIR2_DATAPTR_TO_BYTE(mp, XFS_DIR2_MAX_DATAPTR))
1126                 uio->uio_offset = XFS_DIR2_MAX_DATAPTR;
1127         else
1128                 uio->uio_offset = XFS_DIR2_BYTE_TO_DATAPTR(mp, curoff);
1129         kmem_free(map, map_size * sizeof(*map));
1130         kmem_free(p, sizeof(*p));
1131         if (bp)
1132                 xfs_da_brelse(tp, bp);
1133         return error;
1134 }
1135
1136 /*
1137  * Initialize a new leaf block, leaf1 or leafn magic accepted.
1138  */
1139 int
1140 xfs_dir2_leaf_init(
1141         xfs_da_args_t           *args,          /* operation arguments */
1142         xfs_dir2_db_t           bno,            /* directory block number */
1143         xfs_dabuf_t             **bpp,          /* out: leaf buffer */
1144         int                     magic)          /* magic number for block */
1145 {
1146         xfs_dabuf_t             *bp;            /* leaf buffer */
1147         xfs_inode_t             *dp;            /* incore directory inode */
1148         int                     error;          /* error return code */
1149         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1150         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1151         xfs_mount_t             *mp;            /* filesystem mount point */
1152         xfs_trans_t             *tp;            /* transaction pointer */
1153
1154         dp = args->dp;
1155         ASSERT(dp != NULL);
1156         tp = args->trans;
1157         mp = dp->i_mount;
1158         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
1159                bno < XFS_DIR2_FREE_FIRSTDB(mp));
1160         /*
1161          * Get the buffer for the block.
1162          */
1163         error = xfs_da_get_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, bno), -1, &bp,
1164                 XFS_DATA_FORK);
1165         if (error) {
1166                 return error;
1167         }
1168         ASSERT(bp != NULL);
1169         leaf = bp->data;
1170         /*
1171          * Initialize the header.
1172          */
1173         leaf->hdr.info.magic = cpu_to_be16(magic);
1174         leaf->hdr.info.forw = 0;
1175         leaf->hdr.info.back = 0;
1176         leaf->hdr.count = 0;
1177         leaf->hdr.stale = 0;
1178         xfs_dir2_leaf_log_header(tp, bp);
1179         /*
1180          * If it's a leaf-format directory initialize the tail.
1181          * In this case our caller has the real bests table to copy into
1182          * the block.
1183          */
1184         if (magic == XFS_DIR2_LEAF1_MAGIC) {
1185                 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1186                 ltp->bestcount = 0;
1187                 xfs_dir2_leaf_log_tail(tp, bp);
1188         }
1189         *bpp = bp;
1190         return 0;
1191 }
1192
1193 /*
1194  * Log the bests entries indicated from a leaf1 block.
1195  */
1196 static void
1197 xfs_dir2_leaf_log_bests(
1198         xfs_trans_t             *tp,            /* transaction pointer */
1199         xfs_dabuf_t             *bp,            /* leaf buffer */
1200         int                     first,          /* first entry to log */
1201         int                     last)           /* last entry to log */
1202 {
1203         __be16                  *firstb;        /* pointer to first entry */
1204         __be16                  *lastb;         /* pointer to last entry */
1205         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1206         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1207
1208         leaf = bp->data;
1209         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
1210         ltp = XFS_DIR2_LEAF_TAIL_P(tp->t_mountp, leaf);
1211         firstb = XFS_DIR2_LEAF_BESTS_P(ltp) + first;
1212         lastb = XFS_DIR2_LEAF_BESTS_P(ltp) + last;
1213         xfs_da_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1214                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1215 }
1216
1217 /*
1218  * Log the leaf entries indicated from a leaf1 or leafn block.
1219  */
1220 void
1221 xfs_dir2_leaf_log_ents(
1222         xfs_trans_t             *tp,            /* transaction pointer */
1223         xfs_dabuf_t             *bp,            /* leaf buffer */
1224         int                     first,          /* first entry to log */
1225         int                     last)           /* last entry to log */
1226 {
1227         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1228         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1229         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1230
1231         leaf = bp->data;
1232         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC ||
1233                be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1234         firstlep = &leaf->ents[first];
1235         lastlep = &leaf->ents[last];
1236         xfs_da_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1237                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1238 }
1239
1240 /*
1241  * Log the header of the leaf1 or leafn block.
1242  */
1243 void
1244 xfs_dir2_leaf_log_header(
1245         xfs_trans_t             *tp,            /* transaction pointer */
1246         xfs_dabuf_t             *bp)            /* leaf buffer */
1247 {
1248         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1249
1250         leaf = bp->data;
1251         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC ||
1252                be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1253         xfs_da_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1254                 (uint)(sizeof(leaf->hdr) - 1));
1255 }
1256
1257 /*
1258  * Log the tail of the leaf1 block.
1259  */
1260 STATIC void
1261 xfs_dir2_leaf_log_tail(
1262         xfs_trans_t             *tp,            /* transaction pointer */
1263         xfs_dabuf_t             *bp)            /* leaf buffer */
1264 {
1265         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1266         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1267         xfs_mount_t             *mp;            /* filesystem mount point */
1268
1269         mp = tp->t_mountp;
1270         leaf = bp->data;
1271         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
1272         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1273         xfs_da_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1274                 (uint)(mp->m_dirblksize - 1));
1275 }
1276
1277 /*
1278  * Look up the entry referred to by args in the leaf format directory.
1279  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1280  * is also used by the node-format code.
1281  */
1282 int
1283 xfs_dir2_leaf_lookup(
1284         xfs_da_args_t           *args)          /* operation arguments */
1285 {
1286         xfs_dabuf_t             *dbp;           /* data block buffer */
1287         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1288         xfs_inode_t             *dp;            /* incore directory inode */
1289         int                     error;          /* error return code */
1290         int                     index;          /* found entry index */
1291         xfs_dabuf_t             *lbp;           /* leaf buffer */
1292         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1293         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1294         xfs_trans_t             *tp;            /* transaction pointer */
1295
1296         xfs_dir2_trace_args("leaf_lookup", args);
1297         /*
1298          * Look up name in the leaf block, returning both buffers and index.
1299          */
1300         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1301                 return error;
1302         }
1303         tp = args->trans;
1304         dp = args->dp;
1305         xfs_dir2_leaf_check(dp, lbp);
1306         leaf = lbp->data;
1307         /*
1308          * Get to the leaf entry and contained data entry address.
1309          */
1310         lep = &leaf->ents[index];
1311         /*
1312          * Point to the data entry.
1313          */
1314         dep = (xfs_dir2_data_entry_t *)
1315               ((char *)dbp->data +
1316                XFS_DIR2_DATAPTR_TO_OFF(dp->i_mount, be32_to_cpu(lep->address)));
1317         /*
1318          * Return the found inode number.
1319          */
1320         args->inumber = be64_to_cpu(dep->inumber);
1321         xfs_da_brelse(tp, dbp);
1322         xfs_da_brelse(tp, lbp);
1323         return XFS_ERROR(EEXIST);
1324 }
1325
1326 /*
1327  * Look up name/hash in the leaf block.
1328  * Fill in indexp with the found index, and dbpp with the data buffer.
1329  * If not found dbpp will be NULL, and ENOENT comes back.
1330  * lbpp will always be filled in with the leaf buffer unless there's an error.
1331  */
1332 static int                                      /* error */
1333 xfs_dir2_leaf_lookup_int(
1334         xfs_da_args_t           *args,          /* operation arguments */
1335         xfs_dabuf_t             **lbpp,         /* out: leaf buffer */
1336         int                     *indexp,        /* out: index in leaf block */
1337         xfs_dabuf_t             **dbpp)         /* out: data buffer */
1338 {
1339         xfs_dir2_db_t           curdb;          /* current data block number */
1340         xfs_dabuf_t             *dbp;           /* data buffer */
1341         xfs_dir2_data_entry_t   *dep;           /* data entry */
1342         xfs_inode_t             *dp;            /* incore directory inode */
1343         int                     error;          /* error return code */
1344         int                     index;          /* index in leaf block */
1345         xfs_dabuf_t             *lbp;           /* leaf buffer */
1346         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1347         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1348         xfs_mount_t             *mp;            /* filesystem mount point */
1349         xfs_dir2_db_t           newdb;          /* new data block number */
1350         xfs_trans_t             *tp;            /* transaction pointer */
1351
1352         dp = args->dp;
1353         tp = args->trans;
1354         mp = dp->i_mount;
1355         /*
1356          * Read the leaf block into the buffer.
1357          */
1358         if ((error =
1359             xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
1360                     XFS_DATA_FORK))) {
1361                 return error;
1362         }
1363         *lbpp = lbp;
1364         leaf = lbp->data;
1365         xfs_dir2_leaf_check(dp, lbp);
1366         /*
1367          * Look for the first leaf entry with our hash value.
1368          */
1369         index = xfs_dir2_leaf_search_hash(args, lbp);
1370         /*
1371          * Loop over all the entries with the right hash value
1372          * looking to match the name.
1373          */
1374         for (lep = &leaf->ents[index], dbp = NULL, curdb = -1;
1375              index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
1376              lep++, index++) {
1377                 /*
1378                  * Skip over stale leaf entries.
1379                  */
1380                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1381                         continue;
1382                 /*
1383                  * Get the new data block number.
1384                  */
1385                 newdb = XFS_DIR2_DATAPTR_TO_DB(mp, be32_to_cpu(lep->address));
1386                 /*
1387                  * If it's not the same as the old data block number,
1388                  * need to pitch the old one and read the new one.
1389                  */
1390                 if (newdb != curdb) {
1391                         if (dbp)
1392                                 xfs_da_brelse(tp, dbp);
1393                         if ((error =
1394                             xfs_da_read_buf(tp, dp,
1395                                     XFS_DIR2_DB_TO_DA(mp, newdb), -1, &dbp,
1396                                     XFS_DATA_FORK))) {
1397                                 xfs_da_brelse(tp, lbp);
1398                                 return error;
1399                         }
1400                         xfs_dir2_data_check(dp, dbp);
1401                         curdb = newdb;
1402                 }
1403                 /*
1404                  * Point to the data entry.
1405                  */
1406                 dep = (xfs_dir2_data_entry_t *)
1407                       ((char *)dbp->data +
1408                        XFS_DIR2_DATAPTR_TO_OFF(mp, be32_to_cpu(lep->address)));
1409                 /*
1410                  * If it matches then return it.
1411                  */
1412                 if (dep->namelen == args->namelen &&
1413                     dep->name[0] == args->name[0] &&
1414                     memcmp(dep->name, args->name, args->namelen) == 0) {
1415                         *dbpp = dbp;
1416                         *indexp = index;
1417                         return 0;
1418                 }
1419         }
1420         /*
1421          * No match found, return ENOENT.
1422          */
1423         ASSERT(args->oknoent);
1424         if (dbp)
1425                 xfs_da_brelse(tp, dbp);
1426         xfs_da_brelse(tp, lbp);
1427         return XFS_ERROR(ENOENT);
1428 }
1429
1430 /*
1431  * Remove an entry from a leaf format directory.
1432  */
1433 int                                             /* error */
1434 xfs_dir2_leaf_removename(
1435         xfs_da_args_t           *args)          /* operation arguments */
1436 {
1437         __be16                  *bestsp;        /* leaf block best freespace */
1438         xfs_dir2_data_t         *data;          /* data block structure */
1439         xfs_dir2_db_t           db;             /* data block number */
1440         xfs_dabuf_t             *dbp;           /* data block buffer */
1441         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1442         xfs_inode_t             *dp;            /* incore directory inode */
1443         int                     error;          /* error return code */
1444         xfs_dir2_db_t           i;              /* temporary data block # */
1445         int                     index;          /* index into leaf entries */
1446         xfs_dabuf_t             *lbp;           /* leaf buffer */
1447         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1448         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1449         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1450         xfs_mount_t             *mp;            /* filesystem mount point */
1451         int                     needlog;        /* need to log data header */
1452         int                     needscan;       /* need to rescan data frees */
1453         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1454         xfs_trans_t             *tp;            /* transaction pointer */
1455
1456         xfs_dir2_trace_args("leaf_removename", args);
1457         /*
1458          * Lookup the leaf entry, get the leaf and data blocks read in.
1459          */
1460         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1461                 return error;
1462         }
1463         dp = args->dp;
1464         tp = args->trans;
1465         mp = dp->i_mount;
1466         leaf = lbp->data;
1467         data = dbp->data;
1468         xfs_dir2_data_check(dp, dbp);
1469         /*
1470          * Point to the leaf entry, use that to point to the data entry.
1471          */
1472         lep = &leaf->ents[index];
1473         db = XFS_DIR2_DATAPTR_TO_DB(mp, be32_to_cpu(lep->address));
1474         dep = (xfs_dir2_data_entry_t *)
1475               ((char *)data + XFS_DIR2_DATAPTR_TO_OFF(mp, be32_to_cpu(lep->address)));
1476         needscan = needlog = 0;
1477         oldbest = be16_to_cpu(data->hdr.bestfree[0].length);
1478         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1479         bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
1480         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1481         /*
1482          * Mark the former data entry unused.
1483          */
1484         xfs_dir2_data_make_free(tp, dbp,
1485                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)data),
1486                 XFS_DIR2_DATA_ENTSIZE(dep->namelen), &needlog, &needscan);
1487         /*
1488          * We just mark the leaf entry stale by putting a null in it.
1489          */
1490         be16_add(&leaf->hdr.stale, 1);
1491         xfs_dir2_leaf_log_header(tp, lbp);
1492         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1493         xfs_dir2_leaf_log_ents(tp, lbp, index, index);
1494         /*
1495          * Scan the freespace in the data block again if necessary,
1496          * log the data block header if necessary.
1497          */
1498         if (needscan)
1499                 xfs_dir2_data_freescan(mp, data, &needlog, NULL);
1500         if (needlog)
1501                 xfs_dir2_data_log_header(tp, dbp);
1502         /*
1503          * If the longest freespace in the data block has changed,
1504          * put the new value in the bests table and log that.
1505          */
1506         if (be16_to_cpu(data->hdr.bestfree[0].length) != oldbest) {
1507                 bestsp[db] = data->hdr.bestfree[0].length;
1508                 xfs_dir2_leaf_log_bests(tp, lbp, db, db);
1509         }
1510         xfs_dir2_data_check(dp, dbp);
1511         /*
1512          * If the data block is now empty then get rid of the data block.
1513          */
1514         if (be16_to_cpu(data->hdr.bestfree[0].length) ==
1515             mp->m_dirblksize - (uint)sizeof(data->hdr)) {
1516                 ASSERT(db != mp->m_dirdatablk);
1517                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1518                         /*
1519                          * Nope, can't get rid of it because it caused
1520                          * allocation of a bmap btree block to do so.
1521                          * Just go on, returning success, leaving the
1522                          * empty block in place.
1523                          */
1524                         if (error == ENOSPC && args->total == 0) {
1525                                 xfs_da_buf_done(dbp);
1526                                 error = 0;
1527                         }
1528                         xfs_dir2_leaf_check(dp, lbp);
1529                         xfs_da_buf_done(lbp);
1530                         return error;
1531                 }
1532                 dbp = NULL;
1533                 /*
1534                  * If this is the last data block then compact the
1535                  * bests table by getting rid of entries.
1536                  */
1537                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1538                         /*
1539                          * Look for the last active entry (i).
1540                          */
1541                         for (i = db - 1; i > 0; i--) {
1542                                 if (be16_to_cpu(bestsp[i]) != NULLDATAOFF)
1543                                         break;
1544                         }
1545                         /*
1546                          * Copy the table down so inactive entries at the
1547                          * end are removed.
1548                          */
1549                         memmove(&bestsp[db - i], bestsp,
1550                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1551                         be32_add(&ltp->bestcount, -(db - i));
1552                         xfs_dir2_leaf_log_tail(tp, lbp);
1553                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1554                 } else
1555                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1556         }
1557         /*
1558          * If the data block was not the first one, drop it.
1559          */
1560         else if (db != mp->m_dirdatablk && dbp != NULL) {
1561                 xfs_da_buf_done(dbp);
1562                 dbp = NULL;
1563         }
1564         xfs_dir2_leaf_check(dp, lbp);
1565         /*
1566          * See if we can convert to block form.
1567          */
1568         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1569 }
1570
1571 /*
1572  * Replace the inode number in a leaf format directory entry.
1573  */
1574 int                                             /* error */
1575 xfs_dir2_leaf_replace(
1576         xfs_da_args_t           *args)          /* operation arguments */
1577 {
1578         xfs_dabuf_t             *dbp;           /* data block buffer */
1579         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1580         xfs_inode_t             *dp;            /* incore directory inode */
1581         int                     error;          /* error return code */
1582         int                     index;          /* index of leaf entry */
1583         xfs_dabuf_t             *lbp;           /* leaf buffer */
1584         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1585         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1586         xfs_trans_t             *tp;            /* transaction pointer */
1587
1588         xfs_dir2_trace_args("leaf_replace", args);
1589         /*
1590          * Look up the entry.
1591          */
1592         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1593                 return error;
1594         }
1595         dp = args->dp;
1596         leaf = lbp->data;
1597         /*
1598          * Point to the leaf entry, get data address from it.
1599          */
1600         lep = &leaf->ents[index];
1601         /*
1602          * Point to the data entry.
1603          */
1604         dep = (xfs_dir2_data_entry_t *)
1605               ((char *)dbp->data +
1606                XFS_DIR2_DATAPTR_TO_OFF(dp->i_mount, be32_to_cpu(lep->address)));
1607         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1608         /*
1609          * Put the new inode number in, log it.
1610          */
1611         dep->inumber = cpu_to_be64(args->inumber);
1612         tp = args->trans;
1613         xfs_dir2_data_log_entry(tp, dbp, dep);
1614         xfs_da_buf_done(dbp);
1615         xfs_dir2_leaf_check(dp, lbp);
1616         xfs_da_brelse(tp, lbp);
1617         return 0;
1618 }
1619
1620 /*
1621  * Return index in the leaf block (lbp) which is either the first
1622  * one with this hash value, or if there are none, the insert point
1623  * for that hash value.
1624  */
1625 int                                             /* index value */
1626 xfs_dir2_leaf_search_hash(
1627         xfs_da_args_t           *args,          /* operation arguments */
1628         xfs_dabuf_t             *lbp)           /* leaf buffer */
1629 {
1630         xfs_dahash_t            hash=0;         /* hash from this entry */
1631         xfs_dahash_t            hashwant;       /* hash value looking for */
1632         int                     high;           /* high leaf index */
1633         int                     low;            /* low leaf index */
1634         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1635         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1636         int                     mid=0;          /* current leaf index */
1637
1638         leaf = lbp->data;
1639 #ifndef __KERNEL__
1640         if (!leaf->hdr.count)
1641                 return 0;
1642 #endif
1643         /*
1644          * Note, the table cannot be empty, so we have to go through the loop.
1645          * Binary search the leaf entries looking for our hash value.
1646          */
1647         for (lep = leaf->ents, low = 0, high = be16_to_cpu(leaf->hdr.count) - 1,
1648                 hashwant = args->hashval;
1649              low <= high; ) {
1650                 mid = (low + high) >> 1;
1651                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1652                         break;
1653                 if (hash < hashwant)
1654                         low = mid + 1;
1655                 else
1656                         high = mid - 1;
1657         }
1658         /*
1659          * Found one, back up through all the equal hash values.
1660          */
1661         if (hash == hashwant) {
1662                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1663                         mid--;
1664                 }
1665         }
1666         /*
1667          * Need to point to an entry higher than ours.
1668          */
1669         else if (hash < hashwant)
1670                 mid++;
1671         return mid;
1672 }
1673
1674 /*
1675  * Trim off a trailing data block.  We know it's empty since the leaf
1676  * freespace table says so.
1677  */
1678 int                                             /* error */
1679 xfs_dir2_leaf_trim_data(
1680         xfs_da_args_t           *args,          /* operation arguments */
1681         xfs_dabuf_t             *lbp,           /* leaf buffer */
1682         xfs_dir2_db_t           db)             /* data block number */
1683 {
1684         __be16                  *bestsp;        /* leaf bests table */
1685 #ifdef DEBUG
1686         xfs_dir2_data_t         *data;          /* data block structure */
1687 #endif
1688         xfs_dabuf_t             *dbp;           /* data block buffer */
1689         xfs_inode_t             *dp;            /* incore directory inode */
1690         int                     error;          /* error return value */
1691         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1692         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1693         xfs_mount_t             *mp;            /* filesystem mount point */
1694         xfs_trans_t             *tp;            /* transaction pointer */
1695
1696         dp = args->dp;
1697         mp = dp->i_mount;
1698         tp = args->trans;
1699         /*
1700          * Read the offending data block.  We need its buffer.
1701          */
1702         if ((error = xfs_da_read_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, db), -1, &dbp,
1703                         XFS_DATA_FORK))) {
1704                 return error;
1705         }
1706 #ifdef DEBUG
1707         data = dbp->data;
1708         ASSERT(be32_to_cpu(data->hdr.magic) == XFS_DIR2_DATA_MAGIC);
1709 #endif
1710         /* this seems to be an error
1711          * data is only valid if DEBUG is defined?
1712          * RMC 09/08/1999
1713          */
1714
1715         leaf = lbp->data;
1716         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1717         ASSERT(be16_to_cpu(data->hdr.bestfree[0].length) ==
1718                mp->m_dirblksize - (uint)sizeof(data->hdr));
1719         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1720         /*
1721          * Get rid of the data block.
1722          */
1723         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1724                 ASSERT(error != ENOSPC);
1725                 xfs_da_brelse(tp, dbp);
1726                 return error;
1727         }
1728         /*
1729          * Eliminate the last bests entry from the table.
1730          */
1731         bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
1732         be32_add(&ltp->bestcount, -1);
1733         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1734         xfs_dir2_leaf_log_tail(tp, lbp);
1735         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1736         return 0;
1737 }
1738
1739 /*
1740  * Convert node form directory to leaf form directory.
1741  * The root of the node form dir needs to already be a LEAFN block.
1742  * Just return if we can't do anything.
1743  */
1744 int                                             /* error */
1745 xfs_dir2_node_to_leaf(
1746         xfs_da_state_t          *state)         /* directory operation state */
1747 {
1748         xfs_da_args_t           *args;          /* operation arguments */
1749         xfs_inode_t             *dp;            /* incore directory inode */
1750         int                     error;          /* error return code */
1751         xfs_dabuf_t             *fbp;           /* buffer for freespace block */
1752         xfs_fileoff_t           fo;             /* freespace file offset */
1753         xfs_dir2_free_t         *free;          /* freespace structure */
1754         xfs_dabuf_t             *lbp;           /* buffer for leaf block */
1755         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1756         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1757         xfs_mount_t             *mp;            /* filesystem mount point */
1758         int                     rval;           /* successful free trim? */
1759         xfs_trans_t             *tp;            /* transaction pointer */
1760
1761         /*
1762          * There's more than a leaf level in the btree, so there must
1763          * be multiple leafn blocks.  Give up.
1764          */
1765         if (state->path.active > 1)
1766                 return 0;
1767         args = state->args;
1768         xfs_dir2_trace_args("node_to_leaf", args);
1769         mp = state->mp;
1770         dp = args->dp;
1771         tp = args->trans;
1772         /*
1773          * Get the last offset in the file.
1774          */
1775         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1776                 return error;
1777         }
1778         fo -= mp->m_dirblkfsbs;
1779         /*
1780          * If there are freespace blocks other than the first one,
1781          * take this opportunity to remove trailing empty freespace blocks
1782          * that may have been left behind during no-space-reservation
1783          * operations.
1784          */
1785         while (fo > mp->m_dirfreeblk) {
1786                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1787                         return error;
1788                 }
1789                 if (rval)
1790                         fo -= mp->m_dirblkfsbs;
1791                 else
1792                         return 0;
1793         }
1794         /*
1795          * Now find the block just before the freespace block.
1796          */
1797         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1798                 return error;
1799         }
1800         /*
1801          * If it's not the single leaf block, give up.
1802          */
1803         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1804                 return 0;
1805         lbp = state->path.blk[0].bp;
1806         leaf = lbp->data;
1807         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1808         /*
1809          * Read the freespace block.
1810          */
1811         if ((error = xfs_da_read_buf(tp, dp, mp->m_dirfreeblk, -1, &fbp,
1812                         XFS_DATA_FORK))) {
1813                 return error;
1814         }
1815         free = fbp->data;
1816         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1817         ASSERT(!free->hdr.firstdb);
1818         /*
1819          * Now see if the leafn and free data will fit in a leaf1.
1820          * If not, release the buffer and give up.
1821          */
1822         if ((uint)sizeof(leaf->hdr) +
1823             (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale)) * (uint)sizeof(leaf->ents[0]) +
1824             be32_to_cpu(free->hdr.nvalid) * (uint)sizeof(leaf->bests[0]) +
1825             (uint)sizeof(leaf->tail) >
1826             mp->m_dirblksize) {
1827                 xfs_da_brelse(tp, fbp);
1828                 return 0;
1829         }
1830         /*
1831          * If the leaf has any stale entries in it, compress them out.
1832          * The compact routine will log the header.
1833          */
1834         if (be16_to_cpu(leaf->hdr.stale))
1835                 xfs_dir2_leaf_compact(args, lbp);
1836         else
1837                 xfs_dir2_leaf_log_header(tp, lbp);
1838         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAF1_MAGIC);
1839         /*
1840          * Set up the leaf tail from the freespace block.
1841          */
1842         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1843         ltp->bestcount = free->hdr.nvalid;
1844         /*
1845          * Set up the leaf bests table.
1846          */
1847         memcpy(XFS_DIR2_LEAF_BESTS_P(ltp), free->bests,
1848                 be32_to_cpu(ltp->bestcount) * sizeof(leaf->bests[0]));
1849         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1850         xfs_dir2_leaf_log_tail(tp, lbp);
1851         xfs_dir2_leaf_check(dp, lbp);
1852         /*
1853          * Get rid of the freespace block.
1854          */
1855         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1856         if (error) {
1857                 /*
1858                  * This can't fail here because it can only happen when
1859                  * punching out the middle of an extent, and this is an
1860                  * isolated block.
1861                  */
1862                 ASSERT(error != ENOSPC);
1863                 return error;
1864         }
1865         fbp = NULL;
1866         /*
1867          * Now see if we can convert the single-leaf directory
1868          * down to a block form directory.
1869          * This routine always kills the dabuf for the leaf, so
1870          * eliminate it from the path.
1871          */
1872         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1873         state->path.blk[0].bp = NULL;
1874         return error;
1875 }