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