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_mount.h"
  29 #include "xfs_da_btree.h"
  30 #include "xfs_bmap_btree.h"
  31 #include "xfs_dir2_sf.h"
  32 #include "xfs_dinode.h"
  33 #include "xfs_inode.h"
  34 #include "xfs_bmap.h"
  35 #include "xfs_dir2_data.h"
  36 #include "xfs_dir2_leaf.h"
  37 #include "xfs_dir2_block.h"
  38 #include "xfs_dir2_node.h"
  39 #include "xfs_error.h"
  40 #include "xfs_trace.h"
  41
  42 /*
  43  * Local function declarations.
  44  */
  45 #ifdef DEBUG
  46 static void xfs_dir2_leaf_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
  47 #else
  48 #define xfs_dir2_leaf_check(dp, bp)
  49 #endif
  50 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **lbpp,
  51                                     int *indexp, xfs_dabuf_t **dbpp);
  52 static void xfs_dir2_leaf_log_bests(struct xfs_trans *tp, struct xfs_dabuf *bp,
  53                                     int first, int last);
  54 static void xfs_dir2_leaf_log_tail(struct xfs_trans *tp, struct xfs_dabuf *bp);
  55
  56
  57 /*
  58  * Convert a block form directory to a leaf form directory.
  59  */
  60 int                                             /* error */
  61 xfs_dir2_block_to_leaf(
  62         xfs_da_args_t           *args,          /* operation arguments */
  63         xfs_dabuf_t             *dbp)           /* input block's buffer */
  64 {
  65         __be16                  *bestsp;        /* leaf's bestsp entries */
  66         xfs_dablk_t             blkno;          /* leaf block's bno */
  67         xfs_dir2_block_t        *block;         /* block structure */
  68         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
  69         xfs_dir2_block_tail_t   *btp;           /* block's tail */
  70         xfs_inode_t             *dp;            /* incore directory inode */
  71         int                     error;          /* error return code */
  72         xfs_dabuf_t             *lbp;           /* leaf block's buffer */
  73         xfs_dir2_db_t           ldb;            /* leaf block's bno */
  74         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
  75         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
  76         xfs_mount_t             *mp;            /* filesystem mount point */
  77         int                     needlog;        /* need to log block header */
  78         int                     needscan;       /* need to rescan bestfree */
  79         xfs_trans_t             *tp;            /* transaction pointer */
  80
  81         trace_xfs_dir2_block_to_leaf(args);
  82
  83         dp = args->dp;
  84         mp = dp->i_mount;
  85         tp = args->trans;
  86         /*
  87          * Add the leaf block to the inode.
  88          * This interface will only put blocks in the leaf/node range.
  89          * Since that's empty now, we'll get the root (block 0 in range).
  90          */
  91         if ((error = xfs_da_grow_inode(args, &blkno))) {
  92                 return error;
  93         }
  94         ldb = xfs_dir2_da_to_db(mp, blkno);
  95         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
  96         /*
  97          * Initialize the leaf block, get a buffer for it.
  98          */
  99         if ((error = xfs_dir2_leaf_init(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC))) {
 100                 return error;
 101         }
 102         ASSERT(lbp != NULL);
 103         leaf = lbp->data;
 104         block = dbp->data;
 105         xfs_dir2_data_check(dp, dbp);
 106         btp = xfs_dir2_block_tail_p(mp, block);
 107         blp = xfs_dir2_block_leaf_p(btp);
 108         /*
 109          * Set the counts in the leaf header.
 110          */
 111         leaf->hdr.count = cpu_to_be16(be32_to_cpu(btp->count));
 112         leaf->hdr.stale = cpu_to_be16(be32_to_cpu(btp->stale));
 113         /*
 114          * Could compact these but I think we always do the conversion
 115          * after squeezing out stale entries.
 116          */
 117         memcpy(leaf->ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
 118         xfs_dir2_leaf_log_ents(tp, lbp, 0, be16_to_cpu(leaf->hdr.count) - 1);
 119         needscan = 0;
 120         needlog = 1;
 121         /*
 122          * Make the space formerly occupied by the leaf entries and block
 123          * tail be free.
 124          */
 125         xfs_dir2_data_make_free(tp, dbp,
 126                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)block),
 127                 (xfs_dir2_data_aoff_t)((char *)block + mp->m_dirblksize -
 128                                        (char *)blp),
 129                 &needlog, &needscan);
 130         /*
 131          * Fix up the block header, make it a data block.
 132          */
 133         block->hdr.magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 134         if (needscan)
 135                 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
 136         /*
 137          * Set up leaf tail and bests table.
 138          */
 139         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 140         ltp->bestcount = cpu_to_be32(1);
 141         bestsp = xfs_dir2_leaf_bests_p(ltp);
 142         bestsp[0] =  block->hdr.bestfree[0].length;
 143         /*
 144          * Log the data header and leaf bests table.
 145          */
 146         if (needlog)
 147                 xfs_dir2_data_log_header(tp, dbp);
 148         xfs_dir2_leaf_check(dp, lbp);
 149         xfs_dir2_data_check(dp, dbp);
 150         xfs_dir2_leaf_log_bests(tp, lbp, 0, 0);
 151         xfs_da_buf_done(lbp);
 152         return 0;
 153 }
 154
 155 /*
 156  * Add an entry to a leaf form directory.
 157  */
 158 int                                             /* error */
 159 xfs_dir2_leaf_addname(
 160         xfs_da_args_t           *args)          /* operation arguments */
 161 {
 162         __be16                  *bestsp;        /* freespace table in leaf */
 163         int                     compact;        /* need to compact leaves */
 164         xfs_dir2_data_t         *data;          /* data block structure */
 165         xfs_dabuf_t             *dbp;           /* data block buffer */
 166         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 167         xfs_inode_t             *dp;            /* incore directory inode */
 168         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
 169         int                     error;          /* error return value */
 170         int                     grown;          /* allocated new data block */
 171         int                     highstale;      /* index of next stale leaf */
 172         int                     i;              /* temporary, index */
 173         int                     index;          /* leaf table position */
 174         xfs_dabuf_t             *lbp;           /* leaf's buffer */
 175         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 176         int                     length;         /* length of new entry */
 177         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
 178         int                     lfloglow;       /* low leaf logging index */
 179         int                     lfloghigh;      /* high leaf logging index */
 180         int                     lowstale;       /* index of prev stale leaf */
 181         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 182         xfs_mount_t             *mp;            /* filesystem mount point */
 183         int                     needbytes;      /* leaf block bytes needed */
 184         int                     needlog;        /* need to log data header */
 185         int                     needscan;       /* need to rescan data free */
 186         __be16                  *tagp;          /* end of data entry */
 187         xfs_trans_t             *tp;            /* transaction pointer */
 188         xfs_dir2_db_t           use_block;      /* data block number */
 189
 190         trace_xfs_dir2_leaf_addname(args);
 191
 192         dp = args->dp;
 193         tp = args->trans;
 194         mp = dp->i_mount;
 195         /*
 196          * Read the leaf block.
 197          */
 198         error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
 199                 XFS_DATA_FORK);
 200         if (error) {
 201                 return error;
 202         }
 203         ASSERT(lbp != NULL);
 204         /*
 205          * Look up the entry by hash value and name.
 206          * We know it's not there, our caller has already done a lookup.
 207          * So the index is of the entry to insert in front of.
 208          * But if there are dup hash values the index is of the first of those.
 209          */
 210         index = xfs_dir2_leaf_search_hash(args, lbp);
 211         leaf = lbp->data;
 212         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 213         bestsp = xfs_dir2_leaf_bests_p(ltp);
 214         length = xfs_dir2_data_entsize(args->namelen);
 215         /*
 216          * See if there are any entries with the same hash value
 217          * and space in their block for the new entry.
 218          * This is good because it puts multiple same-hash value entries
 219          * in a data block, improving the lookup of those entries.
 220          */
 221         for (use_block = -1, lep = &leaf->ents[index];
 222              index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
 223              index++, lep++) {
 224                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 225                         continue;
 226                 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 227                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 228                 ASSERT(be16_to_cpu(bestsp[i]) != NULLDATAOFF);
 229                 if (be16_to_cpu(bestsp[i]) >= length) {
 230                         use_block = i;
 231                         break;
 232                 }
 233         }
 234         /*
 235          * Didn't find a block yet, linear search all the data blocks.
 236          */
 237         if (use_block == -1) {
 238                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 239                         /*
 240                          * Remember a block we see that's missing.
 241                          */
 242                         if (be16_to_cpu(bestsp[i]) == NULLDATAOFF && use_block == -1)
 243                                 use_block = i;
 244                         else if (be16_to_cpu(bestsp[i]) >= length) {
 245                                 use_block = i;
 246                                 break;
 247                         }
 248                 }
 249         }
 250         /*
 251          * How many bytes do we need in the leaf block?
 252          */
 253         needbytes =
 254                 (leaf->hdr.stale ? 0 : (uint)sizeof(leaf->ents[0])) +
 255                 (use_block != -1 ? 0 : (uint)sizeof(leaf->bests[0]));
 256         /*
 257          * Now kill use_block if it refers to a missing block, so we
 258          * can use it as an indication of allocation needed.
 259          */
 260         if (use_block != -1 && be16_to_cpu(bestsp[use_block]) == NULLDATAOFF)
 261                 use_block = -1;
 262         /*
 263          * If we don't have enough free bytes but we can make enough
 264          * by compacting out stale entries, we'll do that.
 265          */
 266         if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <
 267                                 needbytes && be16_to_cpu(leaf->hdr.stale) > 1) {
 268                 compact = 1;
 269         }
 270         /*
 271          * Otherwise if we don't have enough free bytes we need to
 272          * convert to node form.
 273          */
 274         else if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(
 275                                                 leaf->hdr.count)] < needbytes) {
 276                 /*
 277                  * Just checking or no space reservation, give up.
 278                  */
 279                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 280                                                         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->op_flags & XFS_DA_OP_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_cpu(&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);
 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_cpu(&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_cpu(&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 STATIC 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_cpu(&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_cpu(&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_inode_t             *dp,            /* incore directory inode */
 754         void                    *dirent,
 755         size_t                  bufsize,
 756         xfs_off_t               *offset,
 757         filldir_t               filldir)
 758 {
 759         xfs_dabuf_t             *bp;            /* data block buffer */
 760         int                     byteoff;        /* offset in current block */
 761         xfs_dir2_db_t           curdb;          /* db for current block */
 762         xfs_dir2_off_t          curoff;         /* current overall offset */
 763         xfs_dir2_data_t         *data;          /* data block structure */
 764         xfs_dir2_data_entry_t   *dep;           /* data entry */
 765         xfs_dir2_data_unused_t  *dup;           /* unused entry */
 766         int                     error = 0;      /* error return value */
 767         int                     i;              /* temporary loop index */
 768         int                     j;              /* temporary loop index */
 769         int                     length;         /* temporary length value */
 770         xfs_bmbt_irec_t         *map;           /* map vector for blocks */
 771         xfs_extlen_t            map_blocks;     /* number of fsbs in map */
 772         xfs_dablk_t             map_off;        /* last mapped file offset */
 773         int                     map_size;       /* total entries in *map */
 774         int                     map_valid;      /* valid entries in *map */
 775         xfs_mount_t             *mp;            /* filesystem mount point */
 776         xfs_dir2_off_t          newoff;         /* new curoff after new blk */
 777         int                     nmap;           /* mappings to ask xfs_bmapi */
 778         char                    *ptr = NULL;    /* pointer to current data */
 779         int                     ra_current;     /* number of read-ahead blks */
 780         int                     ra_index;       /* *map index for read-ahead */
 781         int                     ra_offset;      /* map entry offset for ra */
 782         int                     ra_want;        /* readahead count wanted */
 783
 784         /*
 785          * If the offset is at or past the largest allowed value,
 786          * give up right away.
 787          */
 788         if (*offset >= XFS_DIR2_MAX_DATAPTR)
 789                 return 0;
 790
 791         mp = dp->i_mount;
 792
 793         /*
 794          * Set up to bmap a number of blocks based on the caller's
 795          * buffer size, the directory block size, and the filesystem
 796          * block size.
 797          */
 798         map_size = howmany(bufsize + mp->m_dirblksize, mp->m_sb.sb_blocksize);
 799         map = kmem_alloc(map_size * sizeof(*map), KM_SLEEP);
 800         map_valid = ra_index = ra_offset = ra_current = map_blocks = 0;
 801         bp = NULL;
 802
 803         /*
 804          * Inside the loop we keep the main offset value as a byte offset
 805          * in the directory file.
 806          */
 807         curoff = xfs_dir2_dataptr_to_byte(mp, *offset);
 808
 809         /*
 810          * Force this conversion through db so we truncate the offset
 811          * down to get the start of the data block.
 812          */
 813         map_off = xfs_dir2_db_to_da(mp, xfs_dir2_byte_to_db(mp, curoff));
 814         /*
 815          * Loop over directory entries until we reach the end offset.
 816          * Get more blocks and readahead as necessary.
 817          */
 818         while (curoff < XFS_DIR2_LEAF_OFFSET) {
 819                 /*
 820                  * If we have no buffer, or we're off the end of the
 821                  * current buffer, need to get another one.
 822                  */
 823                 if (!bp || ptr >= (char *)bp->data + mp->m_dirblksize) {
 824                         /*
 825                          * If we have a buffer, we need to release it and
 826                          * take it out of the mapping.
 827                          */
 828                         if (bp) {
 829                                 xfs_da_brelse(NULL, bp);
 830                                 bp = NULL;
 831                                 map_blocks -= mp->m_dirblkfsbs;
 832                                 /*
 833                                  * Loop to get rid of the extents for the
 834                                  * directory block.
 835                                  */
 836                                 for (i = mp->m_dirblkfsbs; i > 0; ) {
 837                                         j = MIN((int)map->br_blockcount, i);
 838                                         map->br_blockcount -= j;
 839                                         map->br_startblock += j;
 840                                         map->br_startoff += j;
 841                                         /*
 842                                          * If mapping is done, pitch it from
 843                                          * the table.
 844                                          */
 845                                         if (!map->br_blockcount && --map_valid)
 846                                                 memmove(&map[0], &map[1],
 847                                                         sizeof(map[0]) *
 848                                                         map_valid);
 849                                         i -= j;
 850                                 }
 851                         }
 852                         /*
 853                          * Recalculate the readahead blocks wanted.
 854                          */
 855                         ra_want = howmany(bufsize + mp->m_dirblksize,
 856                                           mp->m_sb.sb_blocksize) - 1;
 857                         ASSERT(ra_want >= 0);
 858
 859                         /*
 860                          * If we don't have as many as we want, and we haven't
 861                          * run out of data blocks, get some more mappings.
 862                          */
 863                         if (1 + ra_want > map_blocks &&
 864                             map_off <
 865                             xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET)) {
 866                                 /*
 867                                  * Get more bmaps, fill in after the ones
 868                                  * we already have in the table.
 869                                  */
 870                                 nmap = map_size - map_valid;
 871                                 error = xfs_bmapi(NULL, dp,
 872                                         map_off,
 873                                         xfs_dir2_byte_to_da(mp,
 874                                                 XFS_DIR2_LEAF_OFFSET) - map_off,
 875                                         XFS_BMAPI_METADATA, NULL, 0,
 876                                         &map[map_valid], &nmap, NULL);
 877                                 /*
 878                                  * Don't know if we should ignore this or
 879                                  * try to return an error.
 880                                  * The trouble with returning errors
 881                                  * is that readdir will just stop without
 882                                  * actually passing the error through.
 883                                  */
 884                                 if (error)
 885                                         break;  /* XXX */
 886                                 /*
 887                                  * If we got all the mappings we asked for,
 888                                  * set the final map offset based on the
 889                                  * last bmap value received.
 890                                  * Otherwise, we've reached the end.
 891                                  */
 892                                 if (nmap == map_size - map_valid)
 893                                         map_off =
 894                                         map[map_valid + nmap - 1].br_startoff +
 895                                         map[map_valid + nmap - 1].br_blockcount;
 896                                 else
 897                                         map_off =
 898                                                 xfs_dir2_byte_to_da(mp,
 899                                                         XFS_DIR2_LEAF_OFFSET);
 900                                 /*
 901                                  * Look for holes in the mapping, and
 902                                  * eliminate them.  Count up the valid blocks.
 903                                  */
 904                                 for (i = map_valid; i < map_valid + nmap; ) {
 905                                         if (map[i].br_startblock ==
 906                                             HOLESTARTBLOCK) {
 907                                                 nmap--;
 908                                                 length = map_valid + nmap - i;
 909                                                 if (length)
 910                                                         memmove(&map[i],
 911                                                                 &map[i + 1],
 912                                                                 sizeof(map[i]) *
 913                                                                 length);
 914                                         } else {
 915                                                 map_blocks +=
 916                                                         map[i].br_blockcount;
 917                                                 i++;
 918                                         }
 919                                 }
 920                                 map_valid += nmap;
 921                         }
 922                         /*
 923                          * No valid mappings, so no more data blocks.
 924                          */
 925                         if (!map_valid) {
 926                                 curoff = xfs_dir2_da_to_byte(mp, map_off);
 927                                 break;
 928                         }
 929                         /*
 930                          * Read the directory block starting at the first
 931                          * mapping.
 932                          */
 933                         curdb = xfs_dir2_da_to_db(mp, map->br_startoff);
 934                         error = xfs_da_read_buf(NULL, dp, map->br_startoff,
 935                                 map->br_blockcount >= mp->m_dirblkfsbs ?
 936                                     XFS_FSB_TO_DADDR(mp, map->br_startblock) :
 937                                     -1,
 938                                 &bp, XFS_DATA_FORK);
 939                         /*
 940                          * Should just skip over the data block instead
 941                          * of giving up.
 942                          */
 943                         if (error)
 944                                 break;  /* XXX */
 945                         /*
 946                          * Adjust the current amount of read-ahead: we just
 947                          * read a block that was previously ra.
 948                          */
 949                         if (ra_current)
 950                                 ra_current -= mp->m_dirblkfsbs;
 951                         /*
 952                          * Do we need more readahead?
 953                          */
 954                         for (ra_index = ra_offset = i = 0;
 955                              ra_want > ra_current && i < map_blocks;
 956                              i += mp->m_dirblkfsbs) {
 957                                 ASSERT(ra_index < map_valid);
 958                                 /*
 959                                  * Read-ahead a contiguous directory block.
 960                                  */
 961                                 if (i > ra_current &&
 962                                     map[ra_index].br_blockcount >=
 963                                     mp->m_dirblkfsbs) {
 964                                         xfs_buf_readahead(mp->m_ddev_targp,
 965                                                 XFS_FSB_TO_DADDR(mp,
 966                                                    map[ra_index].br_startblock +
 967                                                    ra_offset),
 968                                                 (int)BTOBB(mp->m_dirblksize));
 969                                         ra_current = i;
 970                                 }
 971                                 /*
 972                                  * Read-ahead a non-contiguous directory block.
 973                                  * This doesn't use our mapping, but this
 974                                  * is a very rare case.
 975                                  */
 976                                 else if (i > ra_current) {
 977                                         (void)xfs_da_reada_buf(NULL, dp,
 978                                                 map[ra_index].br_startoff +
 979                                                 ra_offset, XFS_DATA_FORK);
 980                                         ra_current = i;
 981                                 }
 982                                 /*
 983                                  * Advance offset through the mapping table.
 984                                  */
 985                                 for (j = 0; j < mp->m_dirblkfsbs; j++) {
 986                                         /*
 987                                          * The rest of this extent but not
 988                                          * more than a dir block.
 989                                          */
 990                                         length = MIN(mp->m_dirblkfsbs,
 991                                                 (int)(map[ra_index].br_blockcount -
 992                                                 ra_offset));
 993                                         j += length;
 994                                         ra_offset += length;
 995                                         /*
 996                                          * Advance to the next mapping if
 997                                          * this one is used up.
 998                                          */
 999                                         if (ra_offset ==
1000                                             map[ra_index].br_blockcount) {
1001                                                 ra_offset = 0;
1002                                                 ra_index++;
1003                                         }
1004                                 }
1005                         }
1006                         /*
1007                          * Having done a read, we need to set a new offset.
1008                          */
1009                         newoff = xfs_dir2_db_off_to_byte(mp, curdb, 0);
1010                         /*
1011                          * Start of the current block.
1012                          */
1013                         if (curoff < newoff)
1014                                 curoff = newoff;
1015                         /*
1016                          * Make sure we're in the right block.
1017                          */
1018                         else if (curoff > newoff)
1019                                 ASSERT(xfs_dir2_byte_to_db(mp, curoff) ==
1020                                        curdb);
1021                         data = bp->data;
1022                         xfs_dir2_data_check(dp, bp);
1023                         /*
1024                          * Find our position in the block.
1025                          */
1026                         ptr = (char *)&data->u;
1027                         byteoff = xfs_dir2_byte_to_off(mp, curoff);
1028                         /*
1029                          * Skip past the header.
1030                          */
1031                         if (byteoff == 0)
1032                                 curoff += (uint)sizeof(data->hdr);
1033                         /*
1034                          * Skip past entries until we reach our offset.
1035                          */
1036                         else {
1037                                 while ((char *)ptr - (char *)data < byteoff) {
1038                                         dup = (xfs_dir2_data_unused_t *)ptr;
1039
1040                                         if (be16_to_cpu(dup->freetag)
1041                                                   == XFS_DIR2_DATA_FREE_TAG) {
1042
1043                                                 length = be16_to_cpu(dup->length);
1044                                                 ptr += length;
1045                                                 continue;
1046                                         }
1047                                         dep = (xfs_dir2_data_entry_t *)ptr;
1048                                         length =
1049                                            xfs_dir2_data_entsize(dep->namelen);
1050                                         ptr += length;
1051                                 }
1052                                 /*
1053                                  * Now set our real offset.
1054                                  */
1055                                 curoff =
1056                                         xfs_dir2_db_off_to_byte(mp,
1057                                             xfs_dir2_byte_to_db(mp, curoff),
1058                                             (char *)ptr - (char *)data);
1059                                 if (ptr >= (char *)data + mp->m_dirblksize) {
1060                                         continue;
1061                                 }
1062                         }
1063                 }
1064                 /*
1065                  * We have a pointer to an entry.
1066                  * Is it a live one?
1067                  */
1068                 dup = (xfs_dir2_data_unused_t *)ptr;
1069                 /*
1070                  * No, it's unused, skip over it.
1071                  */
1072                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
1073                         length = be16_to_cpu(dup->length);
1074                         ptr += length;
1075                         curoff += length;
1076                         continue;
1077                 }
1078
1079                 dep = (xfs_dir2_data_entry_t *)ptr;
1080                 length = xfs_dir2_data_entsize(dep->namelen);
1081
1082                 if (filldir(dirent, (char *)dep->name, dep->namelen,
1083                             xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff,
1084                             be64_to_cpu(dep->inumber), DT_UNKNOWN))
1085                         break;
1086
1087                 /*
1088                  * Advance to next entry in the block.
1089                  */
1090                 ptr += length;
1091                 curoff += length;
1092                 /* bufsize may have just been a guess; don't go negative */
1093                 bufsize = bufsize > length ? bufsize - length : 0;
1094         }
1095
1096         /*
1097          * All done.  Set output offset value to current offset.
1098          */
1099         if (curoff > xfs_dir2_dataptr_to_byte(mp, XFS_DIR2_MAX_DATAPTR))
1100                 *offset = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
1101         else
1102                 *offset = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
1103         kmem_free(map);
1104         if (bp)
1105                 xfs_da_brelse(NULL, bp);
1106         return error;
1107 }
1108
1109 /*
1110  * Initialize a new leaf block, leaf1 or leafn magic accepted.
1111  */
1112 int
1113 xfs_dir2_leaf_init(
1114         xfs_da_args_t           *args,          /* operation arguments */
1115         xfs_dir2_db_t           bno,            /* directory block number */
1116         xfs_dabuf_t             **bpp,          /* out: leaf buffer */
1117         int                     magic)          /* magic number for block */
1118 {
1119         xfs_dabuf_t             *bp;            /* leaf buffer */
1120         xfs_inode_t             *dp;            /* incore directory inode */
1121         int                     error;          /* error return code */
1122         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1123         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1124         xfs_mount_t             *mp;            /* filesystem mount point */
1125         xfs_trans_t             *tp;            /* transaction pointer */
1126
1127         dp = args->dp;
1128         ASSERT(dp != NULL);
1129         tp = args->trans;
1130         mp = dp->i_mount;
1131         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
1132                bno < XFS_DIR2_FREE_FIRSTDB(mp));
1133         /*
1134          * Get the buffer for the block.
1135          */
1136         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
1137                 XFS_DATA_FORK);
1138         if (error) {
1139                 return error;
1140         }
1141         ASSERT(bp != NULL);
1142         leaf = bp->data;
1143         /*
1144          * Initialize the header.
1145          */
1146         leaf->hdr.info.magic = cpu_to_be16(magic);
1147         leaf->hdr.info.forw = 0;
1148         leaf->hdr.info.back = 0;
1149         leaf->hdr.count = 0;
1150         leaf->hdr.stale = 0;
1151         xfs_dir2_leaf_log_header(tp, bp);
1152         /*
1153          * If it's a leaf-format directory initialize the tail.
1154          * In this case our caller has the real bests table to copy into
1155          * the block.
1156          */
1157         if (magic == XFS_DIR2_LEAF1_MAGIC) {
1158                 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1159                 ltp->bestcount = 0;
1160                 xfs_dir2_leaf_log_tail(tp, bp);
1161         }
1162         *bpp = bp;
1163         return 0;
1164 }
1165
1166 /*
1167  * Log the bests entries indicated from a leaf1 block.
1168  */
1169 static void
1170 xfs_dir2_leaf_log_bests(
1171         xfs_trans_t             *tp,            /* transaction pointer */
1172         xfs_dabuf_t             *bp,            /* leaf buffer */
1173         int                     first,          /* first entry to log */
1174         int                     last)           /* last entry to log */
1175 {
1176         __be16                  *firstb;        /* pointer to first entry */
1177         __be16                  *lastb;         /* pointer to last entry */
1178         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1179         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1180
1181         leaf = bp->data;
1182         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
1183         ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1184         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1185         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1186         xfs_da_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1187                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1188 }
1189
1190 /*
1191  * Log the leaf entries indicated from a leaf1 or leafn block.
1192  */
1193 void
1194 xfs_dir2_leaf_log_ents(
1195         xfs_trans_t             *tp,            /* transaction pointer */
1196         xfs_dabuf_t             *bp,            /* leaf buffer */
1197         int                     first,          /* first entry to log */
1198         int                     last)           /* last entry to log */
1199 {
1200         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1201         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1202         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1203
1204         leaf = bp->data;
1205         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC ||
1206                be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1207         firstlep = &leaf->ents[first];
1208         lastlep = &leaf->ents[last];
1209         xfs_da_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1210                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1211 }
1212
1213 /*
1214  * Log the header of the leaf1 or leafn block.
1215  */
1216 void
1217 xfs_dir2_leaf_log_header(
1218         xfs_trans_t             *tp,            /* transaction pointer */
1219         xfs_dabuf_t             *bp)            /* leaf buffer */
1220 {
1221         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1222
1223         leaf = bp->data;
1224         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC ||
1225                be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1226         xfs_da_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1227                 (uint)(sizeof(leaf->hdr) - 1));
1228 }
1229
1230 /*
1231  * Log the tail of the leaf1 block.
1232  */
1233 STATIC void
1234 xfs_dir2_leaf_log_tail(
1235         xfs_trans_t             *tp,            /* transaction pointer */
1236         xfs_dabuf_t             *bp)            /* leaf buffer */
1237 {
1238         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1239         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1240         xfs_mount_t             *mp;            /* filesystem mount point */
1241
1242         mp = tp->t_mountp;
1243         leaf = bp->data;
1244         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
1245         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1246         xfs_da_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1247                 (uint)(mp->m_dirblksize - 1));
1248 }
1249
1250 /*
1251  * Look up the entry referred to by args in the leaf format directory.
1252  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1253  * is also used by the node-format code.
1254  */
1255 int
1256 xfs_dir2_leaf_lookup(
1257         xfs_da_args_t           *args)          /* operation arguments */
1258 {
1259         xfs_dabuf_t             *dbp;           /* data block buffer */
1260         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1261         xfs_inode_t             *dp;            /* incore directory inode */
1262         int                     error;          /* error return code */
1263         int                     index;          /* found entry index */
1264         xfs_dabuf_t             *lbp;           /* leaf buffer */
1265         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1266         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1267         xfs_trans_t             *tp;            /* transaction pointer */
1268
1269         trace_xfs_dir2_leaf_lookup(args);
1270
1271         /*
1272          * Look up name in the leaf block, returning both buffers and index.
1273          */
1274         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1275                 return error;
1276         }
1277         tp = args->trans;
1278         dp = args->dp;
1279         xfs_dir2_leaf_check(dp, lbp);
1280         leaf = lbp->data;
1281         /*
1282          * Get to the leaf entry and contained data entry address.
1283          */
1284         lep = &leaf->ents[index];
1285         /*
1286          * Point to the data entry.
1287          */
1288         dep = (xfs_dir2_data_entry_t *)
1289               ((char *)dbp->data +
1290                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1291         /*
1292          * Return the found inode number & CI name if appropriate
1293          */
1294         args->inumber = be64_to_cpu(dep->inumber);
1295         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1296         xfs_da_brelse(tp, dbp);
1297         xfs_da_brelse(tp, lbp);
1298         return XFS_ERROR(error);
1299 }
1300
1301 /*
1302  * Look up name/hash in the leaf block.
1303  * Fill in indexp with the found index, and dbpp with the data buffer.
1304  * If not found dbpp will be NULL, and ENOENT comes back.
1305  * lbpp will always be filled in with the leaf buffer unless there's an error.
1306  */
1307 static int                                      /* error */
1308 xfs_dir2_leaf_lookup_int(
1309         xfs_da_args_t           *args,          /* operation arguments */
1310         xfs_dabuf_t             **lbpp,         /* out: leaf buffer */
1311         int                     *indexp,        /* out: index in leaf block */
1312         xfs_dabuf_t             **dbpp)         /* out: data buffer */
1313 {
1314         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1315         xfs_dabuf_t             *dbp = NULL;    /* data buffer */
1316         xfs_dir2_data_entry_t   *dep;           /* data entry */
1317         xfs_inode_t             *dp;            /* incore directory inode */
1318         int                     error;          /* error return code */
1319         int                     index;          /* index in leaf block */
1320         xfs_dabuf_t             *lbp;           /* leaf buffer */
1321         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1322         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1323         xfs_mount_t             *mp;            /* filesystem mount point */
1324         xfs_dir2_db_t           newdb;          /* new data block number */
1325         xfs_trans_t             *tp;            /* transaction pointer */
1326         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1327         enum xfs_dacmp          cmp;            /* name compare result */
1328
1329         dp = args->dp;
1330         tp = args->trans;
1331         mp = dp->i_mount;
1332         /*
1333          * Read the leaf block into the buffer.
1334          */
1335         error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
1336                                                         XFS_DATA_FORK);
1337         if (error)
1338                 return error;
1339         *lbpp = lbp;
1340         leaf = lbp->data;
1341         xfs_dir2_leaf_check(dp, lbp);
1342         /*
1343          * Look for the first leaf entry with our hash value.
1344          */
1345         index = xfs_dir2_leaf_search_hash(args, lbp);
1346         /*
1347          * Loop over all the entries with the right hash value
1348          * looking to match the name.
1349          */
1350         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
1351                                 be32_to_cpu(lep->hashval) == args->hashval;
1352                                 lep++, index++) {
1353                 /*
1354                  * Skip over stale leaf entries.
1355                  */
1356                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1357                         continue;
1358                 /*
1359                  * Get the new data block number.
1360                  */
1361                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1362                 /*
1363                  * If it's not the same as the old data block number,
1364                  * need to pitch the old one and read the new one.
1365                  */
1366                 if (newdb != curdb) {
1367                         if (dbp)
1368                                 xfs_da_brelse(tp, dbp);
1369                         error = xfs_da_read_buf(tp, dp,
1370                                                 xfs_dir2_db_to_da(mp, newdb),
1371                                                 -1, &dbp, XFS_DATA_FORK);
1372                         if (error) {
1373                                 xfs_da_brelse(tp, lbp);
1374                                 return error;
1375                         }
1376                         xfs_dir2_data_check(dp, dbp);
1377                         curdb = newdb;
1378                 }
1379                 /*
1380                  * Point to the data entry.
1381                  */
1382                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->data +
1383                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1384                 /*
1385                  * Compare name and if it's an exact match, return the index
1386                  * and buffer. If it's the first case-insensitive match, store
1387                  * the index and buffer and continue looking for an exact match.
1388                  */
1389                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1390                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1391                         args->cmpresult = cmp;
1392                         *indexp = index;
1393                         /* case exact match: return the current buffer. */
1394                         if (cmp == XFS_CMP_EXACT) {
1395                                 *dbpp = dbp;
1396                                 return 0;
1397                         }
1398                         cidb = curdb;
1399                 }
1400         }
1401         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1402         /*
1403          * Here, we can only be doing a lookup (not a rename or remove).
1404          * If a case-insensitive match was found earlier, re-read the
1405          * appropriate data block if required and return it.
1406          */
1407         if (args->cmpresult == XFS_CMP_CASE) {
1408                 ASSERT(cidb != -1);
1409                 if (cidb != curdb) {
1410                         xfs_da_brelse(tp, dbp);
1411                         error = xfs_da_read_buf(tp, dp,
1412                                                 xfs_dir2_db_to_da(mp, cidb),
1413                                                 -1, &dbp, XFS_DATA_FORK);
1414                         if (error) {
1415                                 xfs_da_brelse(tp, lbp);
1416                                 return error;
1417                         }
1418                 }
1419                 *dbpp = dbp;
1420                 return 0;
1421         }
1422         /*
1423          * No match found, return ENOENT.
1424          */
1425         ASSERT(cidb == -1);
1426         if (dbp)
1427                 xfs_da_brelse(tp, dbp);
1428         xfs_da_brelse(tp, lbp);
1429         return XFS_ERROR(ENOENT);
1430 }
1431
1432 /*
1433  * Remove an entry from a leaf format directory.
1434  */
1435 int                                             /* error */
1436 xfs_dir2_leaf_removename(
1437         xfs_da_args_t           *args)          /* operation arguments */
1438 {
1439         __be16                  *bestsp;        /* leaf block best freespace */
1440         xfs_dir2_data_t         *data;          /* data block structure */
1441         xfs_dir2_db_t           db;             /* data block number */
1442         xfs_dabuf_t             *dbp;           /* data block buffer */
1443         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1444         xfs_inode_t             *dp;            /* incore directory inode */
1445         int                     error;          /* error return code */
1446         xfs_dir2_db_t           i;              /* temporary data block # */
1447         int                     index;          /* index into leaf entries */
1448         xfs_dabuf_t             *lbp;           /* leaf buffer */
1449         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1450         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1451         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1452         xfs_mount_t             *mp;            /* filesystem mount point */
1453         int                     needlog;        /* need to log data header */
1454         int                     needscan;       /* need to rescan data frees */
1455         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1456         xfs_trans_t             *tp;            /* transaction pointer */
1457
1458         trace_xfs_dir2_leaf_removename(args);
1459
1460         /*
1461          * Lookup the leaf entry, get the leaf and data blocks read in.
1462          */
1463         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1464                 return error;
1465         }
1466         dp = args->dp;
1467         tp = args->trans;
1468         mp = dp->i_mount;
1469         leaf = lbp->data;
1470         data = dbp->data;
1471         xfs_dir2_data_check(dp, dbp);
1472         /*
1473          * Point to the leaf entry, use that to point to the data entry.
1474          */
1475         lep = &leaf->ents[index];
1476         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1477         dep = (xfs_dir2_data_entry_t *)
1478               ((char *)data + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1479         needscan = needlog = 0;
1480         oldbest = be16_to_cpu(data->hdr.bestfree[0].length);
1481         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1482         bestsp = xfs_dir2_leaf_bests_p(ltp);
1483         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1484         /*
1485          * Mark the former data entry unused.
1486          */
1487         xfs_dir2_data_make_free(tp, dbp,
1488                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)data),
1489                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1490         /*
1491          * We just mark the leaf entry stale by putting a null in it.
1492          */
1493         be16_add_cpu(&leaf->hdr.stale, 1);
1494         xfs_dir2_leaf_log_header(tp, lbp);
1495         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1496         xfs_dir2_leaf_log_ents(tp, lbp, index, index);
1497         /*
1498          * Scan the freespace in the data block again if necessary,
1499          * log the data block header if necessary.
1500          */
1501         if (needscan)
1502                 xfs_dir2_data_freescan(mp, data, &needlog);
1503         if (needlog)
1504                 xfs_dir2_data_log_header(tp, dbp);
1505         /*
1506          * If the longest freespace in the data block has changed,
1507          * put the new value in the bests table and log that.
1508          */
1509         if (be16_to_cpu(data->hdr.bestfree[0].length) != oldbest) {
1510                 bestsp[db] = data->hdr.bestfree[0].length;
1511                 xfs_dir2_leaf_log_bests(tp, lbp, db, db);
1512         }
1513         xfs_dir2_data_check(dp, dbp);
1514         /*
1515          * If the data block is now empty then get rid of the data block.
1516          */
1517         if (be16_to_cpu(data->hdr.bestfree[0].length) ==
1518             mp->m_dirblksize - (uint)sizeof(data->hdr)) {
1519                 ASSERT(db != mp->m_dirdatablk);
1520                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1521                         /*
1522                          * Nope, can't get rid of it because it caused
1523                          * allocation of a bmap btree block to do so.
1524                          * Just go on, returning success, leaving the
1525                          * empty block in place.
1526                          */
1527                         if (error == ENOSPC && args->total == 0) {
1528                                 xfs_da_buf_done(dbp);
1529                                 error = 0;
1530                         }
1531                         xfs_dir2_leaf_check(dp, lbp);
1532                         xfs_da_buf_done(lbp);
1533                         return error;
1534                 }
1535                 dbp = NULL;
1536                 /*
1537                  * If this is the last data block then compact the
1538                  * bests table by getting rid of entries.
1539                  */
1540                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1541                         /*
1542                          * Look for the last active entry (i).
1543                          */
1544                         for (i = db - 1; i > 0; i--) {
1545                                 if (be16_to_cpu(bestsp[i]) != NULLDATAOFF)
1546                                         break;
1547                         }
1548                         /*
1549                          * Copy the table down so inactive entries at the
1550                          * end are removed.
1551                          */
1552                         memmove(&bestsp[db - i], bestsp,
1553                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1554                         be32_add_cpu(&ltp->bestcount, -(db - i));
1555                         xfs_dir2_leaf_log_tail(tp, lbp);
1556                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1557                 } else
1558                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1559         }
1560         /*
1561          * If the data block was not the first one, drop it.
1562          */
1563         else if (db != mp->m_dirdatablk && dbp != NULL) {
1564                 xfs_da_buf_done(dbp);
1565                 dbp = NULL;
1566         }
1567         xfs_dir2_leaf_check(dp, lbp);
1568         /*
1569          * See if we can convert to block form.
1570          */
1571         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1572 }
1573
1574 /*
1575  * Replace the inode number in a leaf format directory entry.
1576  */
1577 int                                             /* error */
1578 xfs_dir2_leaf_replace(
1579         xfs_da_args_t           *args)          /* operation arguments */
1580 {
1581         xfs_dabuf_t             *dbp;           /* data block buffer */
1582         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1583         xfs_inode_t             *dp;            /* incore directory inode */
1584         int                     error;          /* error return code */
1585         int                     index;          /* index of leaf entry */
1586         xfs_dabuf_t             *lbp;           /* leaf buffer */
1587         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1588         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1589         xfs_trans_t             *tp;            /* transaction pointer */
1590
1591         trace_xfs_dir2_leaf_replace(args);
1592
1593         /*
1594          * Look up the entry.
1595          */
1596         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1597                 return error;
1598         }
1599         dp = args->dp;
1600         leaf = lbp->data;
1601         /*
1602          * Point to the leaf entry, get data address from it.
1603          */
1604         lep = &leaf->ents[index];
1605         /*
1606          * Point to the data entry.
1607          */
1608         dep = (xfs_dir2_data_entry_t *)
1609               ((char *)dbp->data +
1610                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1611         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1612         /*
1613          * Put the new inode number in, log it.
1614          */
1615         dep->inumber = cpu_to_be64(args->inumber);
1616         tp = args->trans;
1617         xfs_dir2_data_log_entry(tp, dbp, dep);
1618         xfs_da_buf_done(dbp);
1619         xfs_dir2_leaf_check(dp, lbp);
1620         xfs_da_brelse(tp, lbp);
1621         return 0;
1622 }
1623
1624 /*
1625  * Return index in the leaf block (lbp) which is either the first
1626  * one with this hash value, or if there are none, the insert point
1627  * for that hash value.
1628  */
1629 int                                             /* index value */
1630 xfs_dir2_leaf_search_hash(
1631         xfs_da_args_t           *args,          /* operation arguments */
1632         xfs_dabuf_t             *lbp)           /* leaf buffer */
1633 {
1634         xfs_dahash_t            hash=0;         /* hash from this entry */
1635         xfs_dahash_t            hashwant;       /* hash value looking for */
1636         int                     high;           /* high leaf index */
1637         int                     low;            /* low leaf index */
1638         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1639         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1640         int                     mid=0;          /* current leaf index */
1641
1642         leaf = lbp->data;
1643 #ifndef __KERNEL__
1644         if (!leaf->hdr.count)
1645                 return 0;
1646 #endif
1647         /*
1648          * Note, the table cannot be empty, so we have to go through the loop.
1649          * Binary search the leaf entries looking for our hash value.
1650          */
1651         for (lep = leaf->ents, low = 0, high = be16_to_cpu(leaf->hdr.count) - 1,
1652                 hashwant = args->hashval;
1653              low <= high; ) {
1654                 mid = (low + high) >> 1;
1655                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1656                         break;
1657                 if (hash < hashwant)
1658                         low = mid + 1;
1659                 else
1660                         high = mid - 1;
1661         }
1662         /*
1663          * Found one, back up through all the equal hash values.
1664          */
1665         if (hash == hashwant) {
1666                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1667                         mid--;
1668                 }
1669         }
1670         /*
1671          * Need to point to an entry higher than ours.
1672          */
1673         else if (hash < hashwant)
1674                 mid++;
1675         return mid;
1676 }
1677
1678 /*
1679  * Trim off a trailing data block.  We know it's empty since the leaf
1680  * freespace table says so.
1681  */
1682 int                                             /* error */
1683 xfs_dir2_leaf_trim_data(
1684         xfs_da_args_t           *args,          /* operation arguments */
1685         xfs_dabuf_t             *lbp,           /* leaf buffer */
1686         xfs_dir2_db_t           db)             /* data block number */
1687 {
1688         __be16                  *bestsp;        /* leaf bests table */
1689 #ifdef DEBUG
1690         xfs_dir2_data_t         *data;          /* data block structure */
1691 #endif
1692         xfs_dabuf_t             *dbp;           /* data block buffer */
1693         xfs_inode_t             *dp;            /* incore directory inode */
1694         int                     error;          /* error return value */
1695         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1696         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1697         xfs_mount_t             *mp;            /* filesystem mount point */
1698         xfs_trans_t             *tp;            /* transaction pointer */
1699
1700         dp = args->dp;
1701         mp = dp->i_mount;
1702         tp = args->trans;
1703         /*
1704          * Read the offending data block.  We need its buffer.
1705          */
1706         if ((error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp,
1707                         XFS_DATA_FORK))) {
1708                 return error;
1709         }
1710 #ifdef DEBUG
1711         data = dbp->data;
1712         ASSERT(be32_to_cpu(data->hdr.magic) == XFS_DIR2_DATA_MAGIC);
1713 #endif
1714         /* this seems to be an error
1715          * data is only valid if DEBUG is defined?
1716          * RMC 09/08/1999
1717          */
1718
1719         leaf = lbp->data;
1720         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1721         ASSERT(be16_to_cpu(data->hdr.bestfree[0].length) ==
1722                mp->m_dirblksize - (uint)sizeof(data->hdr));
1723         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1724         /*
1725          * Get rid of the data block.
1726          */
1727         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1728                 ASSERT(error != ENOSPC);
1729                 xfs_da_brelse(tp, dbp);
1730                 return error;
1731         }
1732         /*
1733          * Eliminate the last bests entry from the table.
1734          */
1735         bestsp = xfs_dir2_leaf_bests_p(ltp);
1736         be32_add_cpu(&ltp->bestcount, -1);
1737         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1738         xfs_dir2_leaf_log_tail(tp, lbp);
1739         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1740         return 0;
1741 }
1742
1743 /*
1744  * Convert node form directory to leaf form directory.
1745  * The root of the node form dir needs to already be a LEAFN block.
1746  * Just return if we can't do anything.
1747  */
1748 int                                             /* error */
1749 xfs_dir2_node_to_leaf(
1750         xfs_da_state_t          *state)         /* directory operation state */
1751 {
1752         xfs_da_args_t           *args;          /* operation arguments */
1753         xfs_inode_t             *dp;            /* incore directory inode */
1754         int                     error;          /* error return code */
1755         xfs_dabuf_t             *fbp;           /* buffer for freespace block */
1756         xfs_fileoff_t           fo;             /* freespace file offset */
1757         xfs_dir2_free_t         *free;          /* freespace structure */
1758         xfs_dabuf_t             *lbp;           /* buffer for leaf block */
1759         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1760         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1761         xfs_mount_t             *mp;            /* filesystem mount point */
1762         int                     rval;           /* successful free trim? */
1763         xfs_trans_t             *tp;            /* transaction pointer */
1764
1765         /*
1766          * There's more than a leaf level in the btree, so there must
1767          * be multiple leafn blocks.  Give up.
1768          */
1769         if (state->path.active > 1)
1770                 return 0;
1771         args = state->args;
1772
1773         trace_xfs_dir2_node_to_leaf(args);
1774
1775         mp = state->mp;
1776         dp = args->dp;
1777         tp = args->trans;
1778         /*
1779          * Get the last offset in the file.
1780          */
1781         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1782                 return error;
1783         }
1784         fo -= mp->m_dirblkfsbs;
1785         /*
1786          * If there are freespace blocks other than the first one,
1787          * take this opportunity to remove trailing empty freespace blocks
1788          * that may have been left behind during no-space-reservation
1789          * operations.
1790          */
1791         while (fo > mp->m_dirfreeblk) {
1792                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1793                         return error;
1794                 }
1795                 if (rval)
1796                         fo -= mp->m_dirblkfsbs;
1797                 else
1798                         return 0;
1799         }
1800         /*
1801          * Now find the block just before the freespace block.
1802          */
1803         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1804                 return error;
1805         }
1806         /*
1807          * If it's not the single leaf block, give up.
1808          */
1809         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1810                 return 0;
1811         lbp = state->path.blk[0].bp;
1812         leaf = lbp->data;
1813         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1814         /*
1815          * Read the freespace block.
1816          */
1817         if ((error = xfs_da_read_buf(tp, dp, mp->m_dirfreeblk, -1, &fbp,
1818                         XFS_DATA_FORK))) {
1819                 return error;
1820         }
1821         free = fbp->data;
1822         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1823         ASSERT(!free->hdr.firstdb);
1824         /*
1825          * Now see if the leafn and free data will fit in a leaf1.
1826          * If not, release the buffer and give up.
1827          */
1828         if ((uint)sizeof(leaf->hdr) +
1829             (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale)) * (uint)sizeof(leaf->ents[0]) +
1830             be32_to_cpu(free->hdr.nvalid) * (uint)sizeof(leaf->bests[0]) +
1831             (uint)sizeof(leaf->tail) >
1832             mp->m_dirblksize) {
1833                 xfs_da_brelse(tp, fbp);
1834                 return 0;
1835         }
1836         /*
1837          * If the leaf has any stale entries in it, compress them out.
1838          * The compact routine will log the header.
1839          */
1840         if (be16_to_cpu(leaf->hdr.stale))
1841                 xfs_dir2_leaf_compact(args, lbp);
1842         else
1843                 xfs_dir2_leaf_log_header(tp, lbp);
1844         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAF1_MAGIC);
1845         /*
1846          * Set up the leaf tail from the freespace block.
1847          */
1848         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1849         ltp->bestcount = free->hdr.nvalid;
1850         /*
1851          * Set up the leaf bests table.
1852          */
1853         memcpy(xfs_dir2_leaf_bests_p(ltp), free->bests,
1854                 be32_to_cpu(ltp->bestcount) * sizeof(leaf->bests[0]));
1855         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1856         xfs_dir2_leaf_log_tail(tp, lbp);
1857         xfs_dir2_leaf_check(dp, lbp);
1858         /*
1859          * Get rid of the freespace block.
1860          */
1861         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1862         if (error) {
1863                 /*
1864                  * This can't fail here because it can only happen when
1865                  * punching out the middle of an extent, and this is an
1866                  * isolated block.
1867                  */
1868                 ASSERT(error != ENOSPC);
1869                 return error;
1870         }
1871         fbp = NULL;
1872         /*
1873          * Now see if we can convert the single-leaf directory
1874          * down to a block form directory.
1875          * This routine always kills the dabuf for the leaf, so
1876          * eliminate it from the path.
1877          */
1878         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1879         state->path.blk[0].bp = NULL;
1880         return error;
1881 }