fs/xfs/xfs_dir2_leaf.c

   1 /*
   2  * Copyright (c) 2000-2003 Silicon Graphics, Inc.  All Rights Reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it would be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  11  *
  12  * Further, this software is distributed without any warranty that it is
  13  * free of the rightful claim of any third person regarding infringement
  14  * or the like.  Any license provided herein, whether implied or
  15  * otherwise, applies only to this software file.  Patent licenses, if
  16  * any, provided herein do not apply to combinations of this program with
  17  * other software, or any other product whatsoever.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write the Free Software Foundation, Inc., 59
  21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  22  *
  23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  24  * Mountain View, CA  94043, or:
  25  *
  26  * http://www.sgi.com
  27  *
  28  * For further information regarding this notice, see:
  29  *
  30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  31  */
  32 #include "xfs.h"
  33 #include "xfs_fs.h"
  34 #include "xfs_types.h"
  35 #include "xfs_bit.h"
  36 #include "xfs_log.h"
  37 #include "xfs_inum.h"
  38 #include "xfs_trans.h"
  39 #include "xfs_sb.h"
  40 #include "xfs_ag.h"
  41 #include "xfs_dir.h"
  42 #include "xfs_dir2.h"
  43 #include "xfs_dmapi.h"
  44 #include "xfs_mount.h"
  45 #include "xfs_da_btree.h"
  46 #include "xfs_bmap_btree.h"
  47 #include "xfs_attr_sf.h"
  48 #include "xfs_dir_sf.h"
  49 #include "xfs_dir2_sf.h"
  50 #include "xfs_dinode.h"
  51 #include "xfs_inode.h"
  52 #include "xfs_bmap.h"
  53 #include "xfs_dir2_data.h"
  54 #include "xfs_dir2_leaf.h"
  55 #include "xfs_dir2_block.h"
  56 #include "xfs_dir2_node.h"
  57 #include "xfs_dir2_trace.h"
  58 #include "xfs_error.h"
  59
  60 /*
  61  * Local function declarations.
  62  */
  63 #ifdef DEBUG
  64 static void xfs_dir2_leaf_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
  65 #else
  66 #define xfs_dir2_leaf_check(dp, bp)
  67 #endif
  68 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **lbpp,
  69                                     int *indexp, xfs_dabuf_t **dbpp);
  70 static void xfs_dir2_leaf_log_bests(struct xfs_trans *tp, struct xfs_dabuf *bp,
  71                                     int first, int last);
  72 static void xfs_dir2_leaf_log_tail(struct xfs_trans *tp, struct xfs_dabuf *bp);
  73
  74
  75 /*
  76  * Convert a block form directory to a leaf form directory.
  77  */
  78 int                                             /* error */
  79 xfs_dir2_block_to_leaf(
  80         xfs_da_args_t           *args,          /* operation arguments */
  81         xfs_dabuf_t             *dbp)           /* input block's buffer */
  82 {
  83         xfs_dir2_data_off_t     *bestsp;        /* leaf's bestsp entries */
  84         xfs_dablk_t             blkno;          /* leaf block's bno */
  85         xfs_dir2_block_t        *block;         /* block structure */
  86         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
  87         xfs_dir2_block_tail_t   *btp;           /* block's tail */
  88         xfs_inode_t             *dp;            /* incore directory inode */
  89         int                     error;          /* error return code */
  90         xfs_dabuf_t             *lbp;           /* leaf block's buffer */
  91         xfs_dir2_db_t           ldb;            /* leaf block's bno */
  92         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
  93         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
  94         xfs_mount_t             *mp;            /* filesystem mount point */
  95         int                     needlog;        /* need to log block header */
  96         int                     needscan;       /* need to rescan bestfree */
  97         xfs_trans_t             *tp;            /* transaction pointer */
  98
  99         xfs_dir2_trace_args_b("block_to_leaf", args, dbp);
 100         dp = args->dp;
 101         mp = dp->i_mount;
 102         tp = args->trans;
 103         /*
 104          * Add the leaf block to the inode.
 105          * This interface will only put blocks in the leaf/node range.
 106          * Since that's empty now, we'll get the root (block 0 in range).
 107          */
 108         if ((error = xfs_da_grow_inode(args, &blkno))) {
 109                 return error;
 110         }
 111         ldb = XFS_DIR2_DA_TO_DB(mp, blkno);
 112         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
 113         /*
 114          * Initialize the leaf block, get a buffer for it.
 115          */
 116         if ((error = xfs_dir2_leaf_init(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC))) {
 117                 return error;
 118         }
 119         ASSERT(lbp != NULL);
 120         leaf = lbp->data;
 121         block = dbp->data;
 122         xfs_dir2_data_check(dp, dbp);
 123         btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
 124         blp = XFS_DIR2_BLOCK_LEAF_P(btp);
 125         /*
 126          * Set the counts in the leaf header.
 127          */
 128         INT_COPY(leaf->hdr.count, btp->count, ARCH_CONVERT); /* INT_: type change */
 129         INT_COPY(leaf->hdr.stale, btp->stale, ARCH_CONVERT); /* INT_: type change */
 130         /*
 131          * Could compact these but I think we always do the conversion
 132          * after squeezing out stale entries.
 133          */
 134         memcpy(leaf->ents, blp, INT_GET(btp->count, ARCH_CONVERT) * sizeof(xfs_dir2_leaf_entry_t));
 135         xfs_dir2_leaf_log_ents(tp, lbp, 0, INT_GET(leaf->hdr.count, ARCH_CONVERT) - 1);
 136         needscan = 0;
 137         needlog = 1;
 138         /*
 139          * Make the space formerly occupied by the leaf entries and block
 140          * tail be free.
 141          */
 142         xfs_dir2_data_make_free(tp, dbp,
 143                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)block),
 144                 (xfs_dir2_data_aoff_t)((char *)block + mp->m_dirblksize -
 145                                        (char *)blp),
 146                 &needlog, &needscan);
 147         /*
 148          * Fix up the block header, make it a data block.
 149          */
 150         INT_SET(block->hdr.magic, ARCH_CONVERT, XFS_DIR2_DATA_MAGIC);
 151         if (needscan)
 152                 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
 153                         NULL);
 154         /*
 155          * Set up leaf tail and bests table.
 156          */
 157         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
 158         INT_SET(ltp->bestcount, ARCH_CONVERT, 1);
 159         bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
 160         INT_COPY(bestsp[0], block->hdr.bestfree[0].length, ARCH_CONVERT);
 161         /*
 162          * Log the data header and leaf bests table.
 163          */
 164         if (needlog)
 165                 xfs_dir2_data_log_header(tp, dbp);
 166         xfs_dir2_leaf_check(dp, lbp);
 167         xfs_dir2_data_check(dp, dbp);
 168         xfs_dir2_leaf_log_bests(tp, lbp, 0, 0);
 169         xfs_da_buf_done(lbp);
 170         return 0;
 171 }
 172
 173 /*
 174  * Add an entry to a leaf form directory.
 175  */
 176 int                                             /* error */
 177 xfs_dir2_leaf_addname(
 178         xfs_da_args_t           *args)          /* operation arguments */
 179 {
 180         xfs_dir2_data_off_t     *bestsp;        /* freespace table in leaf */
 181         int                     compact;        /* need to compact leaves */
 182         xfs_dir2_data_t         *data;          /* data block structure */
 183         xfs_dabuf_t             *dbp;           /* data block buffer */
 184         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 185         xfs_inode_t             *dp;            /* incore directory inode */
 186         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
 187         int                     error;          /* error return value */
 188         int                     grown;          /* allocated new data block */
 189         int                     highstale;      /* index of next stale leaf */
 190         int                     i;              /* temporary, index */
 191         int                     index;          /* leaf table position */
 192         xfs_dabuf_t             *lbp;           /* leaf's buffer */
 193         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 194         int                     length;         /* length of new entry */
 195         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
 196         int                     lfloglow;       /* low leaf logging index */
 197         int                     lfloghigh;      /* high leaf logging index */
 198         int                     lowstale;       /* index of prev stale leaf */
 199         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 200         xfs_mount_t             *mp;            /* filesystem mount point */
 201         int                     needbytes;      /* leaf block bytes needed */
 202         int                     needlog;        /* need to log data header */
 203         int                     needscan;       /* need to rescan data free */
 204         xfs_dir2_data_off_t     *tagp;          /* end of data entry */
 205         xfs_trans_t             *tp;            /* transaction pointer */
 206         xfs_dir2_db_t           use_block;      /* data block number */
 207
 208         xfs_dir2_trace_args("leaf_addname", args);
 209         dp = args->dp;
 210         tp = args->trans;
 211         mp = dp->i_mount;
 212         /*
 213          * Read the leaf block.
 214          */
 215         error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
 216                 XFS_DATA_FORK);
 217         if (error) {
 218                 return error;
 219         }
 220         ASSERT(lbp != NULL);
 221         /*
 222          * Look up the entry by hash value and name.
 223          * We know it's not there, our caller has already done a lookup.
 224          * So the index is of the entry to insert in front of.
 225          * But if there are dup hash values the index is of the first of those.
 226          */
 227         index = xfs_dir2_leaf_search_hash(args, lbp);
 228         leaf = lbp->data;
 229         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
 230         bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
 231         length = XFS_DIR2_DATA_ENTSIZE(args->namelen);
 232         /*
 233          * See if there are any entries with the same hash value
 234          * and space in their block for the new entry.
 235          * This is good because it puts multiple same-hash value entries
 236          * in a data block, improving the lookup of those entries.
 237          */
 238         for (use_block = -1, lep = &leaf->ents[index];
 239              index < INT_GET(leaf->hdr.count, ARCH_CONVERT) && INT_GET(lep->hashval, ARCH_CONVERT) == args->hashval;
 240              index++, lep++) {
 241                 if (INT_GET(lep->address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR)
 242                         continue;
 243                 i = XFS_DIR2_DATAPTR_TO_DB(mp, INT_GET(lep->address, ARCH_CONVERT));
 244                 ASSERT(i < INT_GET(ltp->bestcount, ARCH_CONVERT));
 245                 ASSERT(INT_GET(bestsp[i], ARCH_CONVERT) != NULLDATAOFF);
 246                 if (INT_GET(bestsp[i], ARCH_CONVERT) >= length) {
 247                         use_block = i;
 248                         break;
 249                 }
 250         }
 251         /*
 252          * Didn't find a block yet, linear search all the data blocks.
 253          */
 254         if (use_block == -1) {
 255                 for (i = 0; i < INT_GET(ltp->bestcount, ARCH_CONVERT); i++) {
 256                         /*
 257                          * Remember a block we see that's missing.
 258                          */
 259                         if (INT_GET(bestsp[i], ARCH_CONVERT) == NULLDATAOFF && use_block == -1)
 260                                 use_block = i;
 261                         else if (INT_GET(bestsp[i], ARCH_CONVERT) >= length) {
 262                                 use_block = i;
 263                                 break;
 264                         }
 265                 }
 266         }
 267         /*
 268          * How many bytes do we need in the leaf block?
 269          */
 270         needbytes =
 271                 (leaf->hdr.stale ? 0 : (uint)sizeof(leaf->ents[0])) +
 272                 (use_block != -1 ? 0 : (uint)sizeof(leaf->bests[0]));
 273         /*
 274          * Now kill use_block if it refers to a missing block, so we
 275          * can use it as an indication of allocation needed.
 276          */
 277         if (use_block != -1 && INT_GET(bestsp[use_block], ARCH_CONVERT) == NULLDATAOFF)
 278                 use_block = -1;
 279         /*
 280          * If we don't have enough free bytes but we can make enough
 281          * by compacting out stale entries, we'll do that.
 282          */
 283         if ((char *)bestsp - (char *)&leaf->ents[INT_GET(leaf->hdr.count, ARCH_CONVERT)] < needbytes &&
 284             INT_GET(leaf->hdr.stale, ARCH_CONVERT) > 1) {
 285                 compact = 1;
 286         }
 287         /*
 288          * Otherwise if we don't have enough free bytes we need to
 289          * convert to node form.
 290          */
 291         else if ((char *)bestsp - (char *)&leaf->ents[INT_GET(leaf->hdr.count, ARCH_CONVERT)] <
 292                  needbytes) {
 293                 /*
 294                  * Just checking or no space reservation, give up.
 295                  */
 296                 if (args->justcheck || args->total == 0) {
 297                         xfs_da_brelse(tp, lbp);
 298                         return XFS_ERROR(ENOSPC);
 299                 }
 300                 /*
 301                  * Convert to node form.
 302                  */
 303                 error = xfs_dir2_leaf_to_node(args, lbp);
 304                 xfs_da_buf_done(lbp);
 305                 if (error)
 306                         return error;
 307                 /*
 308                  * Then add the new entry.
 309                  */
 310                 return xfs_dir2_node_addname(args);
 311         }
 312         /*
 313          * Otherwise it will fit without compaction.
 314          */
 315         else
 316                 compact = 0;
 317         /*
 318          * If just checking, then it will fit unless we needed to allocate
 319          * a new data block.
 320          */
 321         if (args->justcheck) {
 322                 xfs_da_brelse(tp, lbp);
 323                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
 324         }
 325         /*
 326          * If no allocations are allowed, return now before we've
 327          * changed anything.
 328          */
 329         if (args->total == 0 && use_block == -1) {
 330                 xfs_da_brelse(tp, lbp);
 331                 return XFS_ERROR(ENOSPC);
 332         }
 333         /*
 334          * Need to compact the leaf entries, removing stale ones.
 335          * Leave one stale entry behind - the one closest to our
 336          * insertion index - and we'll shift that one to our insertion
 337          * point later.
 338          */
 339         if (compact) {
 340                 xfs_dir2_leaf_compact_x1(lbp, &index, &lowstale, &highstale,
 341                         &lfloglow, &lfloghigh);
 342         }
 343         /*
 344          * There are stale entries, so we'll need log-low and log-high
 345          * impossibly bad values later.
 346          */
 347         else if (INT_GET(leaf->hdr.stale, ARCH_CONVERT)) {
 348                 lfloglow = INT_GET(leaf->hdr.count, ARCH_CONVERT);
 349                 lfloghigh = -1;
 350         }
 351         /*
 352          * If there was no data block space found, we need to allocate
 353          * a new one.
 354          */
 355         if (use_block == -1) {
 356                 /*
 357                  * Add the new data block.
 358                  */
 359                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 360                                 &use_block))) {
 361                         xfs_da_brelse(tp, lbp);
 362                         return error;
 363                 }
 364                 /*
 365                  * Initialize the block.
 366                  */
 367                 if ((error = xfs_dir2_data_init(args, use_block, &dbp))) {
 368                         xfs_da_brelse(tp, lbp);
 369                         return error;
 370                 }
 371                 /*
 372                  * If we're adding a new data block on the end we need to
 373                  * extend the bests table.  Copy it up one entry.
 374                  */
 375                 if (use_block >= INT_GET(ltp->bestcount, ARCH_CONVERT)) {
 376                         bestsp--;
 377                         memmove(&bestsp[0], &bestsp[1],
 378                                 INT_GET(ltp->bestcount, ARCH_CONVERT) * sizeof(bestsp[0]));
 379                         INT_MOD(ltp->bestcount, ARCH_CONVERT, +1);
 380                         xfs_dir2_leaf_log_tail(tp, lbp);
 381                         xfs_dir2_leaf_log_bests(tp, lbp, 0, INT_GET(ltp->bestcount, ARCH_CONVERT) - 1);
 382                 }
 383                 /*
 384                  * If we're filling in a previously empty block just log it.
 385                  */
 386                 else
 387                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 388                 data = dbp->data;
 389                 INT_COPY(bestsp[use_block], data->hdr.bestfree[0].length, ARCH_CONVERT);
 390                 grown = 1;
 391         }
 392         /*
 393          * Already had space in some data block.
 394          * Just read that one in.
 395          */
 396         else {
 397                 if ((error =
 398                     xfs_da_read_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, use_block),
 399                             -1, &dbp, XFS_DATA_FORK))) {
 400                         xfs_da_brelse(tp, lbp);
 401                         return error;
 402                 }
 403                 data = dbp->data;
 404                 grown = 0;
 405         }
 406         xfs_dir2_data_check(dp, dbp);
 407         /*
 408          * Point to the biggest freespace in our data block.
 409          */
 410         dup = (xfs_dir2_data_unused_t *)
 411               ((char *)data + INT_GET(data->hdr.bestfree[0].offset, ARCH_CONVERT));
 412         ASSERT(INT_GET(dup->length, ARCH_CONVERT) >= length);
 413         needscan = needlog = 0;
 414         /*
 415          * Mark the initial part of our freespace in use for the new entry.
 416          */
 417         xfs_dir2_data_use_free(tp, dbp, dup,
 418                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)data), length,
 419                 &needlog, &needscan);
 420         /*
 421          * Initialize our new entry (at last).
 422          */
 423         dep = (xfs_dir2_data_entry_t *)dup;
 424         INT_SET(dep->inumber, ARCH_CONVERT, args->inumber);
 425         dep->namelen = args->namelen;
 426         memcpy(dep->name, args->name, dep->namelen);
 427         tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
 428         INT_SET(*tagp, ARCH_CONVERT, (xfs_dir2_data_off_t)((char *)dep - (char *)data));
 429         /*
 430          * Need to scan fix up the bestfree table.
 431          */
 432         if (needscan)
 433                 xfs_dir2_data_freescan(mp, data, &needlog, NULL);
 434         /*
 435          * Need to log the data block's header.
 436          */
 437         if (needlog)
 438                 xfs_dir2_data_log_header(tp, dbp);
 439         xfs_dir2_data_log_entry(tp, dbp, dep);
 440         /*
 441          * If the bests table needs to be changed, do it.
 442          * Log the change unless we've already done that.
 443          */
 444         if (INT_GET(bestsp[use_block], ARCH_CONVERT) != INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT)) {
 445                 INT_COPY(bestsp[use_block], data->hdr.bestfree[0].length, ARCH_CONVERT);
 446                 if (!grown)
 447                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 448         }
 449         /*
 450          * Now we need to make room to insert the leaf entry.
 451          * If there are no stale entries, we just insert a hole at index.
 452          */
 453         if (!leaf->hdr.stale) {
 454                 /*
 455                  * lep is still good as the index leaf entry.
 456                  */
 457                 if (index < INT_GET(leaf->hdr.count, ARCH_CONVERT))
 458                         memmove(lep + 1, lep,
 459                                 (INT_GET(leaf->hdr.count, ARCH_CONVERT) - index) * sizeof(*lep));
 460                 /*
 461                  * Record low and high logging indices for the leaf.
 462                  */
 463                 lfloglow = index;
 464                 lfloghigh = INT_GET(leaf->hdr.count, ARCH_CONVERT);
 465                 INT_MOD(leaf->hdr.count, ARCH_CONVERT, +1);
 466         }
 467         /*
 468          * There are stale entries.
 469          * We will use one of them for the new entry.
 470          * It's probably not at the right location, so we'll have to
 471          * shift some up or down first.
 472          */
 473         else {
 474                 /*
 475                  * If we didn't compact before, we need to find the nearest
 476                  * stale entries before and after our insertion point.
 477                  */
 478                 if (compact == 0) {
 479                         /*
 480                          * Find the first stale entry before the insertion
 481                          * point, if any.
 482                          */
 483                         for (lowstale = index - 1;
 484                              lowstale >= 0 &&
 485                                 INT_GET(leaf->ents[lowstale].address, ARCH_CONVERT) !=
 486                                 XFS_DIR2_NULL_DATAPTR;
 487                              lowstale--)
 488                                 continue;
 489                         /*
 490                          * Find the next stale entry at or after the insertion
 491                          * point, if any.   Stop if we go so far that the
 492                          * lowstale entry would be better.
 493                          */
 494                         for (highstale = index;
 495                              highstale < INT_GET(leaf->hdr.count, ARCH_CONVERT) &&
 496                                 INT_GET(leaf->ents[highstale].address, ARCH_CONVERT) !=
 497                                 XFS_DIR2_NULL_DATAPTR &&
 498                                 (lowstale < 0 ||
 499                                  index - lowstale - 1 >= highstale - index);
 500                              highstale++)
 501                                 continue;
 502                 }
 503                 /*
 504                  * If the low one is better, use it.
 505                  */
 506                 if (lowstale >= 0 &&
 507                     (highstale == INT_GET(leaf->hdr.count, ARCH_CONVERT) ||
 508                      index - lowstale - 1 < highstale - index)) {
 509                         ASSERT(index - lowstale - 1 >= 0);
 510                         ASSERT(INT_GET(leaf->ents[lowstale].address, ARCH_CONVERT) ==
 511                                XFS_DIR2_NULL_DATAPTR);
 512                         /*
 513                          * Copy entries up to cover the stale entry
 514                          * and make room for the new entry.
 515                          */
 516                         if (index - lowstale - 1 > 0)
 517                                 memmove(&leaf->ents[lowstale],
 518                                         &leaf->ents[lowstale + 1],
 519                                         (index - lowstale - 1) * sizeof(*lep));
 520                         lep = &leaf->ents[index - 1];
 521                         lfloglow = MIN(lowstale, lfloglow);
 522                         lfloghigh = MAX(index - 1, lfloghigh);
 523                 }
 524                 /*
 525                  * The high one is better, so use that one.
 526                  */
 527                 else {
 528                         ASSERT(highstale - index >= 0);
 529                         ASSERT(INT_GET(leaf->ents[highstale].address, ARCH_CONVERT) ==
 530                                XFS_DIR2_NULL_DATAPTR);
 531                         /*
 532                          * Copy entries down to copver the stale entry
 533                          * and make room for the new entry.
 534                          */
 535                         if (highstale - index > 0)
 536                                 memmove(&leaf->ents[index + 1],
 537                                         &leaf->ents[index],
 538                                         (highstale - index) * sizeof(*lep));
 539                         lep = &leaf->ents[index];
 540                         lfloglow = MIN(index, lfloglow);
 541                         lfloghigh = MAX(highstale, lfloghigh);
 542                 }
 543                 INT_MOD(leaf->hdr.stale, ARCH_CONVERT, -1);
 544         }
 545         /*
 546          * Fill in the new leaf entry.
 547          */
 548         INT_SET(lep->hashval, ARCH_CONVERT, args->hashval);
 549         INT_SET(lep->address, ARCH_CONVERT, XFS_DIR2_DB_OFF_TO_DATAPTR(mp, use_block, INT_GET(*tagp, ARCH_CONVERT)));
 550         /*
 551          * Log the leaf fields and give up the buffers.
 552          */
 553         xfs_dir2_leaf_log_header(tp, lbp);
 554         xfs_dir2_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
 555         xfs_dir2_leaf_check(dp, lbp);
 556         xfs_da_buf_done(lbp);
 557         xfs_dir2_data_check(dp, dbp);
 558         xfs_da_buf_done(dbp);
 559         return 0;
 560 }
 561
 562 #ifdef DEBUG
 563 /*
 564  * Check the internal consistency of a leaf1 block.
 565  * Pop an assert if something is wrong.
 566  */
 567 void
 568 xfs_dir2_leaf_check(
 569         xfs_inode_t             *dp,            /* incore directory inode */
 570         xfs_dabuf_t             *bp)            /* leaf's buffer */
 571 {
 572         int                     i;              /* leaf index */
 573         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 574         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 575         xfs_mount_t             *mp;            /* filesystem mount point */
 576         int                     stale;          /* count of stale leaves */
 577
 578         leaf = bp->data;
 579         mp = dp->i_mount;
 580         ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAF1_MAGIC);
 581         /*
 582          * This value is not restrictive enough.
 583          * Should factor in the size of the bests table as well.
 584          * We can deduce a value for that from di_size.
 585          */
 586         ASSERT(INT_GET(leaf->hdr.count, ARCH_CONVERT) <= XFS_DIR2_MAX_LEAF_ENTS(mp));
 587         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
 588         /*
 589          * Leaves and bests don't overlap.
 590          */
 591         ASSERT((char *)&leaf->ents[INT_GET(leaf->hdr.count, ARCH_CONVERT)] <=
 592                (char *)XFS_DIR2_LEAF_BESTS_P(ltp));
 593         /*
 594          * Check hash value order, count stale entries.
 595          */
 596         for (i = stale = 0; i < INT_GET(leaf->hdr.count, ARCH_CONVERT); i++) {
 597                 if (i + 1 < INT_GET(leaf->hdr.count, ARCH_CONVERT))
 598                         ASSERT(INT_GET(leaf->ents[i].hashval, ARCH_CONVERT) <=
 599                                INT_GET(leaf->ents[i + 1].hashval, ARCH_CONVERT));
 600                 if (INT_GET(leaf->ents[i].address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR)
 601                         stale++;
 602         }
 603         ASSERT(INT_GET(leaf->hdr.stale, ARCH_CONVERT) == stale);
 604 }
 605 #endif  /* DEBUG */
 606
 607 /*
 608  * Compact out any stale entries in the leaf.
 609  * Log the header and changed leaf entries, if any.
 610  */
 611 void
 612 xfs_dir2_leaf_compact(
 613         xfs_da_args_t   *args,          /* operation arguments */
 614         xfs_dabuf_t     *bp)            /* leaf buffer */
 615 {
 616         int             from;           /* source leaf index */
 617         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 618         int             loglow;         /* first leaf entry to log */
 619         int             to;             /* target leaf index */
 620
 621         leaf = bp->data;
 622         if (!leaf->hdr.stale) {
 623                 return;
 624         }
 625         /*
 626          * Compress out the stale entries in place.
 627          */
 628         for (from = to = 0, loglow = -1; from < INT_GET(leaf->hdr.count, ARCH_CONVERT); from++) {
 629                 if (INT_GET(leaf->ents[from].address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR)
 630                         continue;
 631                 /*
 632                  * Only actually copy the entries that are different.
 633                  */
 634                 if (from > to) {
 635                         if (loglow == -1)
 636                                 loglow = to;
 637                         leaf->ents[to] = leaf->ents[from];
 638                 }
 639                 to++;
 640         }
 641         /*
 642          * Update and log the header, log the leaf entries.
 643          */
 644         ASSERT(INT_GET(leaf->hdr.stale, ARCH_CONVERT) == from - to);
 645         INT_MOD(leaf->hdr.count, ARCH_CONVERT, -(INT_GET(leaf->hdr.stale, ARCH_CONVERT)));
 646         leaf->hdr.stale = 0;
 647         xfs_dir2_leaf_log_header(args->trans, bp);
 648         if (loglow != -1)
 649                 xfs_dir2_leaf_log_ents(args->trans, bp, loglow, to - 1);
 650 }
 651
 652 /*
 653  * Compact the leaf entries, removing stale ones.
 654  * Leave one stale entry behind - the one closest to our
 655  * insertion index - and the caller will shift that one to our insertion
 656  * point later.
 657  * Return new insertion index, where the remaining stale entry is,
 658  * and leaf logging indices.
 659  */
 660 void
 661 xfs_dir2_leaf_compact_x1(
 662         xfs_dabuf_t     *bp,            /* leaf buffer */
 663         int             *indexp,        /* insertion index */
 664         int             *lowstalep,     /* out: stale entry before us */
 665         int             *highstalep,    /* out: stale entry after us */
 666         int             *lowlogp,       /* out: low log index */
 667         int             *highlogp)      /* out: high log index */
 668 {
 669         int             from;           /* source copy index */
 670         int             highstale;      /* stale entry at/after index */
 671         int             index;          /* insertion index */
 672         int             keepstale;      /* source index of kept stale */
 673         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 674         int             lowstale;       /* stale entry before index */
 675         int             newindex=0;     /* new insertion index */
 676         int             to;             /* destination copy index */
 677
 678         leaf = bp->data;
 679         ASSERT(INT_GET(leaf->hdr.stale, ARCH_CONVERT) > 1);
 680         index = *indexp;
 681         /*
 682          * Find the first stale entry before our index, if any.
 683          */
 684         for (lowstale = index - 1;
 685              lowstale >= 0 &&
 686                 INT_GET(leaf->ents[lowstale].address, ARCH_CONVERT) != XFS_DIR2_NULL_DATAPTR;
 687              lowstale--)
 688                 continue;
 689         /*
 690          * Find the first stale entry at or after our index, if any.
 691          * Stop if the answer would be worse than lowstale.
 692          */
 693         for (highstale = index;
 694              highstale < INT_GET(leaf->hdr.count, ARCH_CONVERT) &&
 695                 INT_GET(leaf->ents[highstale].address, ARCH_CONVERT) != XFS_DIR2_NULL_DATAPTR &&
 696                 (lowstale < 0 || index - lowstale > highstale - index);
 697              highstale++)
 698                 continue;
 699         /*
 700          * Pick the better of lowstale and highstale.
 701          */
 702         if (lowstale >= 0 &&
 703             (highstale == INT_GET(leaf->hdr.count, ARCH_CONVERT) ||
 704              index - lowstale <= highstale - index))
 705                 keepstale = lowstale;
 706         else
 707                 keepstale = highstale;
 708         /*
 709          * Copy the entries in place, removing all the stale entries
 710          * except keepstale.
 711          */
 712         for (from = to = 0; from < INT_GET(leaf->hdr.count, ARCH_CONVERT); from++) {
 713                 /*
 714                  * Notice the new value of index.
 715                  */
 716                 if (index == from)
 717                         newindex = to;
 718                 if (from != keepstale &&
 719                     INT_GET(leaf->ents[from].address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR) {
 720                         if (from == to)
 721                                 *lowlogp = to;
 722                         continue;
 723                 }
 724                 /*
 725                  * Record the new keepstale value for the insertion.
 726                  */
 727                 if (from == keepstale)
 728                         lowstale = highstale = to;
 729                 /*
 730                  * Copy only the entries that have moved.
 731                  */
 732                 if (from > to)
 733                         leaf->ents[to] = leaf->ents[from];
 734                 to++;
 735         }
 736         ASSERT(from > to);
 737         /*
 738          * If the insertion point was past the last entry,
 739          * set the new insertion point accordingly.
 740          */
 741         if (index == from)
 742                 newindex = to;
 743         *indexp = newindex;
 744         /*
 745          * Adjust the leaf header values.
 746          */
 747         INT_MOD(leaf->hdr.count, ARCH_CONVERT, -(from - to));
 748         INT_SET(leaf->hdr.stale, ARCH_CONVERT, 1);
 749         /*
 750          * Remember the low/high stale value only in the "right"
 751          * direction.
 752          */
 753         if (lowstale >= newindex)
 754                 lowstale = -1;
 755         else
 756                 highstale = INT_GET(leaf->hdr.count, ARCH_CONVERT);
 757         *highlogp = INT_GET(leaf->hdr.count, ARCH_CONVERT) - 1;
 758         *lowstalep = lowstale;
 759         *highstalep = highstale;
 760 }
 761
 762 /*
 763  * Getdents (readdir) for leaf and node directories.
 764  * This reads the data blocks only, so is the same for both forms.
 765  */
 766 int                                             /* error */
 767 xfs_dir2_leaf_getdents(
 768         xfs_trans_t             *tp,            /* transaction pointer */
 769         xfs_inode_t             *dp,            /* incore directory inode */
 770         uio_t                   *uio,           /* I/O control & vectors */
 771         int                     *eofp,          /* out: reached end of dir */
 772         xfs_dirent_t            *dbp,           /* caller's buffer */
 773         xfs_dir2_put_t          put)            /* ABI formatting routine */
 774 {
 775         xfs_dabuf_t             *bp;            /* data block buffer */
 776         int                     byteoff;        /* offset in current block */
 777         xfs_dir2_db_t           curdb;          /* db for current block */
 778         xfs_dir2_off_t          curoff;         /* current overall offset */
 779         xfs_dir2_data_t         *data;          /* data block structure */
 780         xfs_dir2_data_entry_t   *dep;           /* data entry */
 781         xfs_dir2_data_unused_t  *dup;           /* unused entry */
 782         int                     eof;            /* reached end of directory */
 783         int                     error=0;                /* error return value */
 784         int                     i;              /* temporary loop index */
 785         int                     j;              /* temporary loop index */
 786         int                     length;         /* temporary length value */
 787         xfs_bmbt_irec_t         *map;           /* map vector for blocks */
 788         xfs_extlen_t            map_blocks;     /* number of fsbs in map */
 789         xfs_dablk_t             map_off;        /* last mapped file offset */
 790         int                     map_size;       /* total entries in *map */
 791         int                     map_valid;      /* valid entries in *map */
 792         xfs_mount_t             *mp;            /* filesystem mount point */
 793         xfs_dir2_off_t          newoff;         /* new curoff after new blk */
 794         int                     nmap;           /* mappings to ask xfs_bmapi */
 795         xfs_dir2_put_args_t     p;              /* formatting arg bundle */
 796         char                    *ptr=NULL;              /* pointer to current data */
 797         int                     ra_current;     /* number of read-ahead blks */
 798         int                     ra_index;       /* *map index for read-ahead */
 799         int                     ra_offset;      /* map entry offset for ra */
 800         int                     ra_want;        /* readahead count wanted */
 801
 802         /*
 803          * If the offset is at or past the largest allowed value,
 804          * give up right away, return eof.
 805          */
 806         if (uio->uio_offset >= XFS_DIR2_MAX_DATAPTR) {
 807                 *eofp = 1;
 808                 return 0;
 809         }
 810         mp = dp->i_mount;
 811         /*
 812          * Setup formatting arguments.
 813          */
 814         p.dbp = dbp;
 815         p.put = put;
 816         p.uio = uio;
 817         /*
 818          * Set up to bmap a number of blocks based on the caller's
 819          * buffer size, the directory block size, and the filesystem
 820          * block size.
 821          */
 822         map_size =
 823                 howmany(uio->uio_resid + mp->m_dirblksize,
 824                         mp->m_sb.sb_blocksize);
 825         map = kmem_alloc(map_size * sizeof(*map), KM_SLEEP);
 826         map_valid = ra_index = ra_offset = ra_current = map_blocks = 0;
 827         bp = NULL;
 828         eof = 1;
 829         /*
 830          * Inside the loop we keep the main offset value as a byte offset
 831          * in the directory file.
 832          */
 833         curoff = XFS_DIR2_DATAPTR_TO_BYTE(mp, uio->uio_offset);
 834         /*
 835          * Force this conversion through db so we truncate the offset
 836          * down to get the start of the data block.
 837          */
 838         map_off = XFS_DIR2_DB_TO_DA(mp, XFS_DIR2_BYTE_TO_DB(mp, curoff));
 839         /*
 840          * Loop over directory entries until we reach the end offset.
 841          * Get more blocks and readahead as necessary.
 842          */
 843         while (curoff < XFS_DIR2_LEAF_OFFSET) {
 844                 /*
 845                  * If we have no buffer, or we're off the end of the
 846                  * current buffer, need to get another one.
 847                  */
 848                 if (!bp || ptr >= (char *)bp->data + mp->m_dirblksize) {
 849                         /*
 850                          * If we have a buffer, we need to release it and
 851                          * take it out of the mapping.
 852                          */
 853                         if (bp) {
 854                                 xfs_da_brelse(tp, bp);
 855                                 bp = NULL;
 856                                 map_blocks -= mp->m_dirblkfsbs;
 857                                 /*
 858                                  * Loop to get rid of the extents for the
 859                                  * directory block.
 860                                  */
 861                                 for (i = mp->m_dirblkfsbs; i > 0; ) {
 862                                         j = MIN((int)map->br_blockcount, i);
 863                                         map->br_blockcount -= j;
 864                                         map->br_startblock += j;
 865                                         map->br_startoff += j;
 866                                         /*
 867                                          * If mapping is done, pitch it from
 868                                          * the table.
 869                                          */
 870                                         if (!map->br_blockcount && --map_valid)
 871                                                 memmove(&map[0], &map[1],
 872                                                         sizeof(map[0]) *
 873                                                         map_valid);
 874                                         i -= j;
 875                                 }
 876                         }
 877                         /*
 878                          * Recalculate the readahead blocks wanted.
 879                          */
 880                         ra_want = howmany(uio->uio_resid + mp->m_dirblksize,
 881                                           mp->m_sb.sb_blocksize) - 1;
 882                         /*
 883                          * If we don't have as many as we want, and we haven't
 884                          * run out of data blocks, get some more mappings.
 885                          */
 886                         if (1 + ra_want > map_blocks &&
 887                             map_off <
 888                             XFS_DIR2_BYTE_TO_DA(mp, XFS_DIR2_LEAF_OFFSET)) {
 889                                 /*
 890                                  * Get more bmaps, fill in after the ones
 891                                  * we already have in the table.
 892                                  */
 893                                 nmap = map_size - map_valid;
 894                                 error = xfs_bmapi(tp, dp,
 895                                         map_off,
 896                                         XFS_DIR2_BYTE_TO_DA(mp,
 897                                                 XFS_DIR2_LEAF_OFFSET) - map_off,
 898                                         XFS_BMAPI_METADATA, NULL, 0,
 899                                         &map[map_valid], &nmap, NULL);
 900                                 /*
 901                                  * Don't know if we should ignore this or
 902                                  * try to return an error.
 903                                  * The trouble with returning errors
 904                                  * is that readdir will just stop without
 905                                  * actually passing the error through.
 906                                  */
 907                                 if (error)
 908                                         break;  /* XXX */
 909                                 /*
 910                                  * If we got all the mappings we asked for,
 911                                  * set the final map offset based on the
 912                                  * last bmap value received.
 913                                  * Otherwise, we've reached the end.
 914                                  */
 915                                 if (nmap == map_size - map_valid)
 916                                         map_off =
 917                                         map[map_valid + nmap - 1].br_startoff +
 918                                         map[map_valid + nmap - 1].br_blockcount;
 919                                 else
 920                                         map_off =
 921                                                 XFS_DIR2_BYTE_TO_DA(mp,
 922                                                         XFS_DIR2_LEAF_OFFSET);
 923                                 /*
 924                                  * Look for holes in the mapping, and
 925                                  * eliminate them.  Count up the valid blocks.
 926                                  */
 927                                 for (i = map_valid; i < map_valid + nmap; ) {
 928                                         if (map[i].br_startblock ==
 929                                             HOLESTARTBLOCK) {
 930                                                 nmap--;
 931                                                 length = map_valid + nmap - i;
 932                                                 if (length)
 933                                                         memmove(&map[i],
 934                                                                 &map[i + 1],
 935                                                                 sizeof(map[i]) *
 936                                                                 length);
 937                                         } else {
 938                                                 map_blocks +=
 939                                                         map[i].br_blockcount;
 940                                                 i++;
 941                                         }
 942                                 }
 943                                 map_valid += nmap;
 944                         }
 945                         /*
 946                          * No valid mappings, so no more data blocks.
 947                          */
 948                         if (!map_valid) {
 949                                 curoff = XFS_DIR2_DA_TO_BYTE(mp, map_off);
 950                                 break;
 951                         }
 952                         /*
 953                          * Read the directory block starting at the first
 954                          * mapping.
 955                          */
 956                         curdb = XFS_DIR2_DA_TO_DB(mp, map->br_startoff);
 957                         error = xfs_da_read_buf(tp, dp, map->br_startoff,
 958                                 map->br_blockcount >= mp->m_dirblkfsbs ?
 959                                     XFS_FSB_TO_DADDR(mp, map->br_startblock) :
 960                                     -1,
 961                                 &bp, XFS_DATA_FORK);
 962                         /*
 963                          * Should just skip over the data block instead
 964                          * of giving up.
 965                          */
 966                         if (error)
 967                                 break;  /* XXX */
 968                         /*
 969                          * Adjust the current amount of read-ahead: we just
 970                          * read a block that was previously ra.
 971                          */
 972                         if (ra_current)
 973                                 ra_current -= mp->m_dirblkfsbs;
 974                         /*
 975                          * Do we need more readahead?
 976                          */
 977                         for (ra_index = ra_offset = i = 0;
 978                              ra_want > ra_current && i < map_blocks;
 979                              i += mp->m_dirblkfsbs) {
 980                                 ASSERT(ra_index < map_valid);
 981                                 /*
 982                                  * Read-ahead a contiguous directory block.
 983                                  */
 984                                 if (i > ra_current &&
 985                                     map[ra_index].br_blockcount >=
 986                                     mp->m_dirblkfsbs) {
 987                                         xfs_baread(mp->m_ddev_targp,
 988                                                 XFS_FSB_TO_DADDR(mp,
 989                                                    map[ra_index].br_startblock +
 990                                                    ra_offset),
 991                                                 (int)BTOBB(mp->m_dirblksize));
 992                                         ra_current = i;
 993                                 }
 994                                 /*
 995                                  * Read-ahead a non-contiguous directory block.
 996                                  * This doesn't use our mapping, but this
 997                                  * is a very rare case.
 998                                  */
 999                                 else if (i > ra_current) {
1000                                         (void)xfs_da_reada_buf(tp, dp,
1001                                                 map[ra_index].br_startoff +
1002                                                 ra_offset, XFS_DATA_FORK);
1003                                         ra_current = i;
1004                                 }
1005                                 /*
1006                                  * Advance offset through the mapping table.
1007                                  */
1008                                 for (j = 0; j < mp->m_dirblkfsbs; j++) {
1009                                         /*
1010                                          * The rest of this extent but not
1011                                          * more than a dir block.
1012                                          */
1013                                         length = MIN(mp->m_dirblkfsbs,
1014                                                 (int)(map[ra_index].br_blockcount -
1015                                                 ra_offset));
1016                                         j += length;
1017                                         ra_offset += length;
1018                                         /*
1019                                          * Advance to the next mapping if
1020                                          * this one is used up.
1021                                          */
1022                                         if (ra_offset ==
1023                                             map[ra_index].br_blockcount) {
1024                                                 ra_offset = 0;
1025                                                 ra_index++;
1026                                         }
1027                                 }
1028                         }
1029                         /*
1030                          * Having done a read, we need to set a new offset.
1031                          */
1032                         newoff = XFS_DIR2_DB_OFF_TO_BYTE(mp, curdb, 0);
1033                         /*
1034                          * Start of the current block.
1035                          */
1036                         if (curoff < newoff)
1037                                 curoff = newoff;
1038                         /*
1039                          * Make sure we're in the right block.
1040                          */
1041                         else if (curoff > newoff)
1042                                 ASSERT(XFS_DIR2_BYTE_TO_DB(mp, curoff) ==
1043                                        curdb);
1044                         data = bp->data;
1045                         xfs_dir2_data_check(dp, bp);
1046                         /*
1047                          * Find our position in the block.
1048                          */
1049                         ptr = (char *)&data->u;
1050                         byteoff = XFS_DIR2_BYTE_TO_OFF(mp, curoff);
1051                         /*
1052                          * Skip past the header.
1053                          */
1054                         if (byteoff == 0)
1055                                 curoff += (uint)sizeof(data->hdr);
1056                         /*
1057                          * Skip past entries until we reach our offset.
1058                          */
1059                         else {
1060                                 while ((char *)ptr - (char *)data < byteoff) {
1061                                         dup = (xfs_dir2_data_unused_t *)ptr;
1062
1063                                         if (INT_GET(dup->freetag, ARCH_CONVERT)
1064                                                   == XFS_DIR2_DATA_FREE_TAG) {
1065
1066                                                 length = INT_GET(dup->length,
1067                                                                  ARCH_CONVERT);
1068                                                 ptr += length;
1069                                                 continue;
1070                                         }
1071                                         dep = (xfs_dir2_data_entry_t *)ptr;
1072                                         length =
1073                                            XFS_DIR2_DATA_ENTSIZE(dep->namelen);
1074                                         ptr += length;
1075                                 }
1076                                 /*
1077                                  * Now set our real offset.
1078                                  */
1079                                 curoff =
1080                                         XFS_DIR2_DB_OFF_TO_BYTE(mp,
1081                                             XFS_DIR2_BYTE_TO_DB(mp, curoff),
1082                                             (char *)ptr - (char *)data);
1083                                 if (ptr >= (char *)data + mp->m_dirblksize) {
1084                                         continue;
1085                                 }
1086                         }
1087                 }
1088                 /*
1089                  * We have a pointer to an entry.
1090                  * Is it a live one?
1091                  */
1092                 dup = (xfs_dir2_data_unused_t *)ptr;
1093                 /*
1094                  * No, it's unused, skip over it.
1095                  */
1096                 if (INT_GET(dup->freetag, ARCH_CONVERT)
1097                                                 == XFS_DIR2_DATA_FREE_TAG) {
1098                         length = INT_GET(dup->length, ARCH_CONVERT);
1099                         ptr += length;
1100                         curoff += length;
1101                         continue;
1102                 }
1103
1104                 /*
1105                  * Copy the entry into the putargs, and try formatting it.
1106                  */
1107                 dep = (xfs_dir2_data_entry_t *)ptr;
1108
1109                 p.namelen = dep->namelen;
1110
1111                 length = XFS_DIR2_DATA_ENTSIZE(p.namelen);
1112
1113                 p.cook = XFS_DIR2_BYTE_TO_DATAPTR(mp, curoff + length);
1114
1115                 p.ino = INT_GET(dep->inumber, ARCH_CONVERT);
1116 #if XFS_BIG_INUMS
1117                 p.ino += mp->m_inoadd;
1118 #endif
1119                 p.name = (char *)dep->name;
1120
1121                 error = p.put(&p);
1122
1123                 /*
1124                  * Won't fit.  Return to caller.
1125                  */
1126                 if (!p.done) {
1127                         eof = 0;
1128                         break;
1129                 }
1130                 /*
1131                  * Advance to next entry in the block.
1132                  */
1133                 ptr += length;
1134                 curoff += length;
1135         }
1136
1137         /*
1138          * All done.  Set output offset value to current offset.
1139          */
1140         *eofp = eof;
1141         if (curoff > XFS_DIR2_DATAPTR_TO_BYTE(mp, XFS_DIR2_MAX_DATAPTR))
1142                 uio->uio_offset = XFS_DIR2_MAX_DATAPTR;
1143         else
1144                 uio->uio_offset = XFS_DIR2_BYTE_TO_DATAPTR(mp, curoff);
1145         kmem_free(map, map_size * sizeof(*map));
1146         if (bp)
1147                 xfs_da_brelse(tp, bp);
1148         return error;
1149 }
1150
1151 /*
1152  * Initialize a new leaf block, leaf1 or leafn magic accepted.
1153  */
1154 int
1155 xfs_dir2_leaf_init(
1156         xfs_da_args_t           *args,          /* operation arguments */
1157         xfs_dir2_db_t           bno,            /* directory block number */
1158         xfs_dabuf_t             **bpp,          /* out: leaf buffer */
1159         int                     magic)          /* magic number for block */
1160 {
1161         xfs_dabuf_t             *bp;            /* leaf buffer */
1162         xfs_inode_t             *dp;            /* incore directory inode */
1163         int                     error;          /* error return code */
1164         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1165         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1166         xfs_mount_t             *mp;            /* filesystem mount point */
1167         xfs_trans_t             *tp;            /* transaction pointer */
1168
1169         dp = args->dp;
1170         ASSERT(dp != NULL);
1171         tp = args->trans;
1172         mp = dp->i_mount;
1173         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
1174                bno < XFS_DIR2_FREE_FIRSTDB(mp));
1175         /*
1176          * Get the buffer for the block.
1177          */
1178         error = xfs_da_get_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, bno), -1, &bp,
1179                 XFS_DATA_FORK);
1180         if (error) {
1181                 return error;
1182         }
1183         ASSERT(bp != NULL);
1184         leaf = bp->data;
1185         /*
1186          * Initialize the header.
1187          */
1188         INT_SET(leaf->hdr.info.magic, ARCH_CONVERT, magic);
1189         leaf->hdr.info.forw = 0;
1190         leaf->hdr.info.back = 0;
1191         leaf->hdr.count = 0;
1192         leaf->hdr.stale = 0;
1193         xfs_dir2_leaf_log_header(tp, bp);
1194         /*
1195          * If it's a leaf-format directory initialize the tail.
1196          * In this case our caller has the real bests table to copy into
1197          * the block.
1198          */
1199         if (magic == XFS_DIR2_LEAF1_MAGIC) {
1200                 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1201                 ltp->bestcount = 0;
1202                 xfs_dir2_leaf_log_tail(tp, bp);
1203         }
1204         *bpp = bp;
1205         return 0;
1206 }
1207
1208 /*
1209  * Log the bests entries indicated from a leaf1 block.
1210  */
1211 static void
1212 xfs_dir2_leaf_log_bests(
1213         xfs_trans_t             *tp,            /* transaction pointer */
1214         xfs_dabuf_t             *bp,            /* leaf buffer */
1215         int                     first,          /* first entry to log */
1216         int                     last)           /* last entry to log */
1217 {
1218         xfs_dir2_data_off_t     *firstb;        /* pointer to first entry */
1219         xfs_dir2_data_off_t     *lastb;         /* pointer to last entry */
1220         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1221         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1222
1223         leaf = bp->data;
1224         ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAF1_MAGIC);
1225         ltp = XFS_DIR2_LEAF_TAIL_P(tp->t_mountp, leaf);
1226         firstb = XFS_DIR2_LEAF_BESTS_P(ltp) + first;
1227         lastb = XFS_DIR2_LEAF_BESTS_P(ltp) + last;
1228         xfs_da_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1229                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1230 }
1231
1232 /*
1233  * Log the leaf entries indicated from a leaf1 or leafn block.
1234  */
1235 void
1236 xfs_dir2_leaf_log_ents(
1237         xfs_trans_t             *tp,            /* transaction pointer */
1238         xfs_dabuf_t             *bp,            /* leaf buffer */
1239         int                     first,          /* first entry to log */
1240         int                     last)           /* last entry to log */
1241 {
1242         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1243         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1244         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1245
1246         leaf = bp->data;
1247         ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAF1_MAGIC ||
1248                INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
1249         firstlep = &leaf->ents[first];
1250         lastlep = &leaf->ents[last];
1251         xfs_da_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1252                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1253 }
1254
1255 /*
1256  * Log the header of the leaf1 or leafn block.
1257  */
1258 void
1259 xfs_dir2_leaf_log_header(
1260         xfs_trans_t             *tp,            /* transaction pointer */
1261         xfs_dabuf_t             *bp)            /* leaf buffer */
1262 {
1263         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1264
1265         leaf = bp->data;
1266         ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAF1_MAGIC ||
1267                INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
1268         xfs_da_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1269                 (uint)(sizeof(leaf->hdr) - 1));
1270 }
1271
1272 /*
1273  * Log the tail of the leaf1 block.
1274  */
1275 STATIC void
1276 xfs_dir2_leaf_log_tail(
1277         xfs_trans_t             *tp,            /* transaction pointer */
1278         xfs_dabuf_t             *bp)            /* leaf buffer */
1279 {
1280         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1281         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1282         xfs_mount_t             *mp;            /* filesystem mount point */
1283
1284         mp = tp->t_mountp;
1285         leaf = bp->data;
1286         ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAF1_MAGIC);
1287         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1288         xfs_da_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1289                 (uint)(mp->m_dirblksize - 1));
1290 }
1291
1292 /*
1293  * Look up the entry referred to by args in the leaf format directory.
1294  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1295  * is also used by the node-format code.
1296  */
1297 int
1298 xfs_dir2_leaf_lookup(
1299         xfs_da_args_t           *args)          /* operation arguments */
1300 {
1301         xfs_dabuf_t             *dbp;           /* data block buffer */
1302         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1303         xfs_inode_t             *dp;            /* incore directory inode */
1304         int                     error;          /* error return code */
1305         int                     index;          /* found entry index */
1306         xfs_dabuf_t             *lbp;           /* leaf buffer */
1307         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1308         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1309         xfs_trans_t             *tp;            /* transaction pointer */
1310
1311         xfs_dir2_trace_args("leaf_lookup", args);
1312         /*
1313          * Look up name in the leaf block, returning both buffers and index.
1314          */
1315         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1316                 return error;
1317         }
1318         tp = args->trans;
1319         dp = args->dp;
1320         xfs_dir2_leaf_check(dp, lbp);
1321         leaf = lbp->data;
1322         /*
1323          * Get to the leaf entry and contained data entry address.
1324          */
1325         lep = &leaf->ents[index];
1326         /*
1327          * Point to the data entry.
1328          */
1329         dep = (xfs_dir2_data_entry_t *)
1330               ((char *)dbp->data +
1331                XFS_DIR2_DATAPTR_TO_OFF(dp->i_mount, INT_GET(lep->address, ARCH_CONVERT)));
1332         /*
1333          * Return the found inode number.
1334          */
1335         args->inumber = INT_GET(dep->inumber, ARCH_CONVERT);
1336         xfs_da_brelse(tp, dbp);
1337         xfs_da_brelse(tp, lbp);
1338         return XFS_ERROR(EEXIST);
1339 }
1340
1341 /*
1342  * Look up name/hash in the leaf block.
1343  * Fill in indexp with the found index, and dbpp with the data buffer.
1344  * If not found dbpp will be NULL, and ENOENT comes back.
1345  * lbpp will always be filled in with the leaf buffer unless there's an error.
1346  */
1347 static int                                      /* error */
1348 xfs_dir2_leaf_lookup_int(
1349         xfs_da_args_t           *args,          /* operation arguments */
1350         xfs_dabuf_t             **lbpp,         /* out: leaf buffer */
1351         int                     *indexp,        /* out: index in leaf block */
1352         xfs_dabuf_t             **dbpp)         /* out: data buffer */
1353 {
1354         xfs_dir2_db_t           curdb;          /* current data block number */
1355         xfs_dabuf_t             *dbp;           /* data buffer */
1356         xfs_dir2_data_entry_t   *dep;           /* data entry */
1357         xfs_inode_t             *dp;            /* incore directory inode */
1358         int                     error;          /* error return code */
1359         int                     index;          /* index in leaf block */
1360         xfs_dabuf_t             *lbp;           /* leaf buffer */
1361         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1362         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1363         xfs_mount_t             *mp;            /* filesystem mount point */
1364         xfs_dir2_db_t           newdb;          /* new data block number */
1365         xfs_trans_t             *tp;            /* transaction pointer */
1366
1367         dp = args->dp;
1368         tp = args->trans;
1369         mp = dp->i_mount;
1370         /*
1371          * Read the leaf block into the buffer.
1372          */
1373         if ((error =
1374             xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
1375                     XFS_DATA_FORK))) {
1376                 return error;
1377         }
1378         *lbpp = lbp;
1379         leaf = lbp->data;
1380         xfs_dir2_leaf_check(dp, lbp);
1381         /*
1382          * Look for the first leaf entry with our hash value.
1383          */
1384         index = xfs_dir2_leaf_search_hash(args, lbp);
1385         /*
1386          * Loop over all the entries with the right hash value
1387          * looking to match the name.
1388          */
1389         for (lep = &leaf->ents[index], dbp = NULL, curdb = -1;
1390              index < INT_GET(leaf->hdr.count, ARCH_CONVERT) && INT_GET(lep->hashval, ARCH_CONVERT) == args->hashval;
1391              lep++, index++) {
1392                 /*
1393                  * Skip over stale leaf entries.
1394                  */
1395                 if (INT_GET(lep->address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR)
1396                         continue;
1397                 /*
1398                  * Get the new data block number.
1399                  */
1400                 newdb = XFS_DIR2_DATAPTR_TO_DB(mp, INT_GET(lep->address, ARCH_CONVERT));
1401                 /*
1402                  * If it's not the same as the old data block number,
1403                  * need to pitch the old one and read the new one.
1404                  */
1405                 if (newdb != curdb) {
1406                         if (dbp)
1407                                 xfs_da_brelse(tp, dbp);
1408                         if ((error =
1409                             xfs_da_read_buf(tp, dp,
1410                                     XFS_DIR2_DB_TO_DA(mp, newdb), -1, &dbp,
1411                                     XFS_DATA_FORK))) {
1412                                 xfs_da_brelse(tp, lbp);
1413                                 return error;
1414                         }
1415                         xfs_dir2_data_check(dp, dbp);
1416                         curdb = newdb;
1417                 }
1418                 /*
1419                  * Point to the data entry.
1420                  */
1421                 dep = (xfs_dir2_data_entry_t *)
1422                       ((char *)dbp->data +
1423                        XFS_DIR2_DATAPTR_TO_OFF(mp, INT_GET(lep->address, ARCH_CONVERT)));
1424                 /*
1425                  * If it matches then return it.
1426                  */
1427                 if (dep->namelen == args->namelen &&
1428                     dep->name[0] == args->name[0] &&
1429                     memcmp(dep->name, args->name, args->namelen) == 0) {
1430                         *dbpp = dbp;
1431                         *indexp = index;
1432                         return 0;
1433                 }
1434         }
1435         /*
1436          * No match found, return ENOENT.
1437          */
1438         ASSERT(args->oknoent);
1439         if (dbp)
1440                 xfs_da_brelse(tp, dbp);
1441         xfs_da_brelse(tp, lbp);
1442         return XFS_ERROR(ENOENT);
1443 }
1444
1445 /*
1446  * Remove an entry from a leaf format directory.
1447  */
1448 int                                             /* error */
1449 xfs_dir2_leaf_removename(
1450         xfs_da_args_t           *args)          /* operation arguments */
1451 {
1452         xfs_dir2_data_off_t     *bestsp;        /* leaf block best freespace */
1453         xfs_dir2_data_t         *data;          /* data block structure */
1454         xfs_dir2_db_t           db;             /* data block number */
1455         xfs_dabuf_t             *dbp;           /* data block buffer */
1456         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1457         xfs_inode_t             *dp;            /* incore directory inode */
1458         int                     error;          /* error return code */
1459         xfs_dir2_db_t           i;              /* temporary data block # */
1460         int                     index;          /* index into leaf entries */
1461         xfs_dabuf_t             *lbp;           /* leaf buffer */
1462         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1463         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1464         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1465         xfs_mount_t             *mp;            /* filesystem mount point */
1466         int                     needlog;        /* need to log data header */
1467         int                     needscan;       /* need to rescan data frees */
1468         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1469         xfs_trans_t             *tp;            /* transaction pointer */
1470
1471         xfs_dir2_trace_args("leaf_removename", args);
1472         /*
1473          * Lookup the leaf entry, get the leaf and data blocks read in.
1474          */
1475         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1476                 return error;
1477         }
1478         dp = args->dp;
1479         tp = args->trans;
1480         mp = dp->i_mount;
1481         leaf = lbp->data;
1482         data = dbp->data;
1483         xfs_dir2_data_check(dp, dbp);
1484         /*
1485          * Point to the leaf entry, use that to point to the data entry.
1486          */
1487         lep = &leaf->ents[index];
1488         db = XFS_DIR2_DATAPTR_TO_DB(mp, INT_GET(lep->address, ARCH_CONVERT));
1489         dep = (xfs_dir2_data_entry_t *)
1490               ((char *)data + XFS_DIR2_DATAPTR_TO_OFF(mp, INT_GET(lep->address, ARCH_CONVERT)));
1491         needscan = needlog = 0;
1492         oldbest = INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT);
1493         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1494         bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
1495         ASSERT(INT_GET(bestsp[db], ARCH_CONVERT) == oldbest);
1496         /*
1497          * Mark the former data entry unused.
1498          */
1499         xfs_dir2_data_make_free(tp, dbp,
1500                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)data),
1501                 XFS_DIR2_DATA_ENTSIZE(dep->namelen), &needlog, &needscan);
1502         /*
1503          * We just mark the leaf entry stale by putting a null in it.
1504          */
1505         INT_MOD(leaf->hdr.stale, ARCH_CONVERT, +1);
1506         xfs_dir2_leaf_log_header(tp, lbp);
1507         INT_SET(lep->address, ARCH_CONVERT, XFS_DIR2_NULL_DATAPTR);
1508         xfs_dir2_leaf_log_ents(tp, lbp, index, index);
1509         /*
1510          * Scan the freespace in the data block again if necessary,
1511          * log the data block header if necessary.
1512          */
1513         if (needscan)
1514                 xfs_dir2_data_freescan(mp, data, &needlog, NULL);
1515         if (needlog)
1516                 xfs_dir2_data_log_header(tp, dbp);
1517         /*
1518          * If the longest freespace in the data block has changed,
1519          * put the new value in the bests table and log that.
1520          */
1521         if (INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT) != oldbest) {
1522                 INT_COPY(bestsp[db], data->hdr.bestfree[0].length, ARCH_CONVERT);
1523                 xfs_dir2_leaf_log_bests(tp, lbp, db, db);
1524         }
1525         xfs_dir2_data_check(dp, dbp);
1526         /*
1527          * If the data block is now empty then get rid of the data block.
1528          */
1529         if (INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT) ==
1530             mp->m_dirblksize - (uint)sizeof(data->hdr)) {
1531                 ASSERT(db != mp->m_dirdatablk);
1532                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1533                         /*
1534                          * Nope, can't get rid of it because it caused
1535                          * allocation of a bmap btree block to do so.
1536                          * Just go on, returning success, leaving the
1537                          * empty block in place.
1538                          */
1539                         if (error == ENOSPC && args->total == 0) {
1540                                 xfs_da_buf_done(dbp);
1541                                 error = 0;
1542                         }
1543                         xfs_dir2_leaf_check(dp, lbp);
1544                         xfs_da_buf_done(lbp);
1545                         return error;
1546                 }
1547                 dbp = NULL;
1548                 /*
1549                  * If this is the last data block then compact the
1550                  * bests table by getting rid of entries.
1551                  */
1552                 if (db == INT_GET(ltp->bestcount, ARCH_CONVERT) - 1) {
1553                         /*
1554                          * Look for the last active entry (i).
1555                          */
1556                         for (i = db - 1; i > 0; i--) {
1557                                 if (INT_GET(bestsp[i], ARCH_CONVERT) != NULLDATAOFF)
1558                                         break;
1559                         }
1560                         /*
1561                          * Copy the table down so inactive entries at the
1562                          * end are removed.
1563                          */
1564                         memmove(&bestsp[db - i], bestsp,
1565                                 (INT_GET(ltp->bestcount, ARCH_CONVERT) - (db - i)) * sizeof(*bestsp));
1566                         INT_MOD(ltp->bestcount, ARCH_CONVERT, -(db - i));
1567                         xfs_dir2_leaf_log_tail(tp, lbp);
1568                         xfs_dir2_leaf_log_bests(tp, lbp, 0, INT_GET(ltp->bestcount, ARCH_CONVERT) - 1);
1569                 } else
1570                         INT_SET(bestsp[db], ARCH_CONVERT, NULLDATAOFF);
1571         }
1572         /*
1573          * If the data block was not the first one, drop it.
1574          */
1575         else if (db != mp->m_dirdatablk && dbp != NULL) {
1576                 xfs_da_buf_done(dbp);
1577                 dbp = NULL;
1578         }
1579         xfs_dir2_leaf_check(dp, lbp);
1580         /*
1581          * See if we can convert to block form.
1582          */
1583         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1584 }
1585
1586 /*
1587  * Replace the inode number in a leaf format directory entry.
1588  */
1589 int                                             /* error */
1590 xfs_dir2_leaf_replace(
1591         xfs_da_args_t           *args)          /* operation arguments */
1592 {
1593         xfs_dabuf_t             *dbp;           /* data block buffer */
1594         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1595         xfs_inode_t             *dp;            /* incore directory inode */
1596         int                     error;          /* error return code */
1597         int                     index;          /* index of leaf entry */
1598         xfs_dabuf_t             *lbp;           /* leaf buffer */
1599         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1600         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1601         xfs_trans_t             *tp;            /* transaction pointer */
1602
1603         xfs_dir2_trace_args("leaf_replace", args);
1604         /*
1605          * Look up the entry.
1606          */
1607         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1608                 return error;
1609         }
1610         dp = args->dp;
1611         leaf = lbp->data;
1612         /*
1613          * Point to the leaf entry, get data address from it.
1614          */
1615         lep = &leaf->ents[index];
1616         /*
1617          * Point to the data entry.
1618          */
1619         dep = (xfs_dir2_data_entry_t *)
1620               ((char *)dbp->data +
1621                XFS_DIR2_DATAPTR_TO_OFF(dp->i_mount, INT_GET(lep->address, ARCH_CONVERT)));
1622         ASSERT(args->inumber != INT_GET(dep->inumber, ARCH_CONVERT));
1623         /*
1624          * Put the new inode number in, log it.
1625          */
1626         INT_SET(dep->inumber, ARCH_CONVERT, args->inumber);
1627         tp = args->trans;
1628         xfs_dir2_data_log_entry(tp, dbp, dep);
1629         xfs_da_buf_done(dbp);
1630         xfs_dir2_leaf_check(dp, lbp);
1631         xfs_da_brelse(tp, lbp);
1632         return 0;
1633 }
1634
1635 /*
1636  * Return index in the leaf block (lbp) which is either the first
1637  * one with this hash value, or if there are none, the insert point
1638  * for that hash value.
1639  */
1640 int                                             /* index value */
1641 xfs_dir2_leaf_search_hash(
1642         xfs_da_args_t           *args,          /* operation arguments */
1643         xfs_dabuf_t             *lbp)           /* leaf buffer */
1644 {
1645         xfs_dahash_t            hash=0;         /* hash from this entry */
1646         xfs_dahash_t            hashwant;       /* hash value looking for */
1647         int                     high;           /* high leaf index */
1648         int                     low;            /* low leaf index */
1649         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1650         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1651         int                     mid=0;          /* current leaf index */
1652
1653         leaf = lbp->data;
1654 #ifndef __KERNEL__
1655         if (!leaf->hdr.count)
1656                 return 0;
1657 #endif
1658         /*
1659          * Note, the table cannot be empty, so we have to go through the loop.
1660          * Binary search the leaf entries looking for our hash value.
1661          */
1662         for (lep = leaf->ents, low = 0, high = INT_GET(leaf->hdr.count, ARCH_CONVERT) - 1,
1663                 hashwant = args->hashval;
1664              low <= high; ) {
1665                 mid = (low + high) >> 1;
1666                 if ((hash = INT_GET(lep[mid].hashval, ARCH_CONVERT)) == hashwant)
1667                         break;
1668                 if (hash < hashwant)
1669                         low = mid + 1;
1670                 else
1671                         high = mid - 1;
1672         }
1673         /*
1674          * Found one, back up through all the equal hash values.
1675          */
1676         if (hash == hashwant) {
1677                 while (mid > 0 && INT_GET(lep[mid - 1].hashval, ARCH_CONVERT) == hashwant) {
1678                         mid--;
1679                 }
1680         }
1681         /*
1682          * Need to point to an entry higher than ours.
1683          */
1684         else if (hash < hashwant)
1685                 mid++;
1686         return mid;
1687 }
1688
1689 /*
1690  * Trim off a trailing data block.  We know it's empty since the leaf
1691  * freespace table says so.
1692  */
1693 int                                             /* error */
1694 xfs_dir2_leaf_trim_data(
1695         xfs_da_args_t           *args,          /* operation arguments */
1696         xfs_dabuf_t             *lbp,           /* leaf buffer */
1697         xfs_dir2_db_t           db)             /* data block number */
1698 {
1699         xfs_dir2_data_off_t     *bestsp;        /* leaf bests table */
1700 #ifdef DEBUG
1701         xfs_dir2_data_t         *data;          /* data block structure */
1702 #endif
1703         xfs_dabuf_t             *dbp;           /* data block buffer */
1704         xfs_inode_t             *dp;            /* incore directory inode */
1705         int                     error;          /* error return value */
1706         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1707         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1708         xfs_mount_t             *mp;            /* filesystem mount point */
1709         xfs_trans_t             *tp;            /* transaction pointer */
1710
1711         dp = args->dp;
1712         mp = dp->i_mount;
1713         tp = args->trans;
1714         /*
1715          * Read the offending data block.  We need its buffer.
1716          */
1717         if ((error = xfs_da_read_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, db), -1, &dbp,
1718                         XFS_DATA_FORK))) {
1719                 return error;
1720         }
1721 #ifdef DEBUG
1722         data = dbp->data;
1723         ASSERT(INT_GET(data->hdr.magic, ARCH_CONVERT) == XFS_DIR2_DATA_MAGIC);
1724 #endif
1725         /* this seems to be an error
1726          * data is only valid if DEBUG is defined?
1727          * RMC 09/08/1999
1728          */
1729
1730         leaf = lbp->data;
1731         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1732         ASSERT(INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT) ==
1733                mp->m_dirblksize - (uint)sizeof(data->hdr));
1734         ASSERT(db == INT_GET(ltp->bestcount, ARCH_CONVERT) - 1);
1735         /*
1736          * Get rid of the data block.
1737          */
1738         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1739                 ASSERT(error != ENOSPC);
1740                 xfs_da_brelse(tp, dbp);
1741                 return error;
1742         }
1743         /*
1744          * Eliminate the last bests entry from the table.
1745          */
1746         bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
1747         INT_MOD(ltp->bestcount, ARCH_CONVERT, -1);
1748         memmove(&bestsp[1], &bestsp[0], INT_GET(ltp->bestcount, ARCH_CONVERT) * sizeof(*bestsp));
1749         xfs_dir2_leaf_log_tail(tp, lbp);
1750         xfs_dir2_leaf_log_bests(tp, lbp, 0, INT_GET(ltp->bestcount, ARCH_CONVERT) - 1);
1751         return 0;
1752 }
1753
1754 /*
1755  * Convert node form directory to leaf form directory.
1756  * The root of the node form dir needs to already be a LEAFN block.
1757  * Just return if we can't do anything.
1758  */
1759 int                                             /* error */
1760 xfs_dir2_node_to_leaf(
1761         xfs_da_state_t          *state)         /* directory operation state */
1762 {
1763         xfs_da_args_t           *args;          /* operation arguments */
1764         xfs_inode_t             *dp;            /* incore directory inode */
1765         int                     error;          /* error return code */
1766         xfs_dabuf_t             *fbp;           /* buffer for freespace block */
1767         xfs_fileoff_t           fo;             /* freespace file offset */
1768         xfs_dir2_free_t         *free;          /* freespace structure */
1769         xfs_dabuf_t             *lbp;           /* buffer for leaf block */
1770         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1771         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1772         xfs_mount_t             *mp;            /* filesystem mount point */
1773         int                     rval;           /* successful free trim? */
1774         xfs_trans_t             *tp;            /* transaction pointer */
1775
1776         /*
1777          * There's more than a leaf level in the btree, so there must
1778          * be multiple leafn blocks.  Give up.
1779          */
1780         if (state->path.active > 1)
1781                 return 0;
1782         args = state->args;
1783         xfs_dir2_trace_args("node_to_leaf", args);
1784         mp = state->mp;
1785         dp = args->dp;
1786         tp = args->trans;
1787         /*
1788          * Get the last offset in the file.
1789          */
1790         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1791                 return error;
1792         }
1793         fo -= mp->m_dirblkfsbs;
1794         /*
1795          * If there are freespace blocks other than the first one,
1796          * take this opportunity to remove trailing empty freespace blocks
1797          * that may have been left behind during no-space-reservation
1798          * operations.
1799          */
1800         while (fo > mp->m_dirfreeblk) {
1801                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1802                         return error;
1803                 }
1804                 if (rval)
1805                         fo -= mp->m_dirblkfsbs;
1806                 else
1807                         return 0;
1808         }
1809         /*
1810          * Now find the block just before the freespace block.
1811          */
1812         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1813                 return error;
1814         }
1815         /*
1816          * If it's not the single leaf block, give up.
1817          */
1818         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1819                 return 0;
1820         lbp = state->path.blk[0].bp;
1821         leaf = lbp->data;
1822         ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
1823         /*
1824          * Read the freespace block.
1825          */
1826         if ((error = xfs_da_read_buf(tp, dp, mp->m_dirfreeblk, -1, &fbp,
1827                         XFS_DATA_FORK))) {
1828                 return error;
1829         }
1830         free = fbp->data;
1831         ASSERT(INT_GET(free->hdr.magic, ARCH_CONVERT) == XFS_DIR2_FREE_MAGIC);
1832         ASSERT(!free->hdr.firstdb);
1833         /*
1834          * Now see if the leafn and free data will fit in a leaf1.
1835          * If not, release the buffer and give up.
1836          */
1837         if ((uint)sizeof(leaf->hdr) +
1838             (INT_GET(leaf->hdr.count, ARCH_CONVERT) - INT_GET(leaf->hdr.stale, ARCH_CONVERT)) * (uint)sizeof(leaf->ents[0]) +
1839             INT_GET(free->hdr.nvalid, ARCH_CONVERT) * (uint)sizeof(leaf->bests[0]) +
1840             (uint)sizeof(leaf->tail) >
1841             mp->m_dirblksize) {
1842                 xfs_da_brelse(tp, fbp);
1843                 return 0;
1844         }
1845         /*
1846          * If the leaf has any stale entries in it, compress them out.
1847          * The compact routine will log the header.
1848          */
1849         if (INT_GET(leaf->hdr.stale, ARCH_CONVERT))
1850                 xfs_dir2_leaf_compact(args, lbp);
1851         else
1852                 xfs_dir2_leaf_log_header(tp, lbp);
1853         INT_SET(leaf->hdr.info.magic, ARCH_CONVERT, XFS_DIR2_LEAF1_MAGIC);
1854         /*
1855          * Set up the leaf tail from the freespace block.
1856          */
1857         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1858         INT_COPY(ltp->bestcount, free->hdr.nvalid, ARCH_CONVERT);
1859         /*
1860          * Set up the leaf bests table.
1861          */
1862         memcpy(XFS_DIR2_LEAF_BESTS_P(ltp), free->bests,
1863                 INT_GET(ltp->bestcount, ARCH_CONVERT) * sizeof(leaf->bests[0]));
1864         xfs_dir2_leaf_log_bests(tp, lbp, 0, INT_GET(ltp->bestcount, ARCH_CONVERT) - 1);
1865         xfs_dir2_leaf_log_tail(tp, lbp);
1866         xfs_dir2_leaf_check(dp, lbp);
1867         /*
1868          * Get rid of the freespace block.
1869          */
1870         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1871         if (error) {
1872                 /*
1873                  * This can't fail here because it can only happen when
1874                  * punching out the middle of an extent, and this is an
1875                  * isolated block.
1876                  */
1877                 ASSERT(error != ENOSPC);
1878                 return error;
1879         }
1880         fbp = NULL;
1881         /*
1882          * Now see if we can convert the single-leaf directory
1883          * down to a block form directory.
1884          * This routine always kills the dabuf for the leaf, so
1885          * eliminate it from the path.
1886          */
1887         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1888         state->path.blk[0].bp = NULL;
1889         return error;
1890 }