fs/xfs/xfs_dir2_leaf.c

   1 /*
   2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_btree.h"
  28 #include "xfs_bmap_btree.h"
  29 #include "xfs_dinode.h"
  30 #include "xfs_inode.h"
  31 #include "xfs_bmap.h"
  32 #include "xfs_dir2_format.h"
  33 #include "xfs_dir2_priv.h"
  34 #include "xfs_error.h"
  35 #include "xfs_trace.h"
  36
  37 /*
  38  * Local function declarations.
  39  */
  40 #ifdef DEBUG
  41 static void xfs_dir2_leaf_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
  42 #else
  43 #define xfs_dir2_leaf_check(dp, bp)
  44 #endif
  45 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **lbpp,
  46                                     int *indexp, xfs_dabuf_t **dbpp);
  47 static void xfs_dir2_leaf_log_bests(struct xfs_trans *tp, struct xfs_dabuf *bp,
  48                                     int first, int last);
  49 static void xfs_dir2_leaf_log_tail(struct xfs_trans *tp, struct xfs_dabuf *bp);
  50
  51
  52 /*
  53  * Convert a block form directory to a leaf form directory.
  54  */
  55 int                                             /* error */
  56 xfs_dir2_block_to_leaf(
  57         xfs_da_args_t           *args,          /* operation arguments */
  58         xfs_dabuf_t             *dbp)           /* input block's buffer */
  59 {
  60         __be16                  *bestsp;        /* leaf's bestsp entries */
  61         xfs_dablk_t             blkno;          /* leaf block's bno */
  62         xfs_dir2_data_hdr_t     *hdr;           /* block header */
  63         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
  64         xfs_dir2_block_tail_t   *btp;           /* block's tail */
  65         xfs_inode_t             *dp;            /* incore directory inode */
  66         int                     error;          /* error return code */
  67         xfs_dabuf_t             *lbp;           /* leaf block's buffer */
  68         xfs_dir2_db_t           ldb;            /* leaf block's bno */
  69         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
  70         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
  71         xfs_mount_t             *mp;            /* filesystem mount point */
  72         int                     needlog;        /* need to log block header */
  73         int                     needscan;       /* need to rescan bestfree */
  74         xfs_trans_t             *tp;            /* transaction pointer */
  75
  76         trace_xfs_dir2_block_to_leaf(args);
  77
  78         dp = args->dp;
  79         mp = dp->i_mount;
  80         tp = args->trans;
  81         /*
  82          * Add the leaf block to the inode.
  83          * This interface will only put blocks in the leaf/node range.
  84          * Since that's empty now, we'll get the root (block 0 in range).
  85          */
  86         if ((error = xfs_da_grow_inode(args, &blkno))) {
  87                 return error;
  88         }
  89         ldb = xfs_dir2_da_to_db(mp, blkno);
  90         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
  91         /*
  92          * Initialize the leaf block, get a buffer for it.
  93          */
  94         if ((error = xfs_dir2_leaf_init(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC))) {
  95                 return error;
  96         }
  97         ASSERT(lbp != NULL);
  98         leaf = lbp->data;
  99         hdr = dbp->data;
 100         xfs_dir2_data_check(dp, dbp);
 101         btp = xfs_dir2_block_tail_p(mp, hdr);
 102         blp = xfs_dir2_block_leaf_p(btp);
 103         /*
 104          * Set the counts in the leaf header.
 105          */
 106         leaf->hdr.count = cpu_to_be16(be32_to_cpu(btp->count));
 107         leaf->hdr.stale = cpu_to_be16(be32_to_cpu(btp->stale));
 108         /*
 109          * Could compact these but I think we always do the conversion
 110          * after squeezing out stale entries.
 111          */
 112         memcpy(leaf->ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
 113         xfs_dir2_leaf_log_ents(tp, lbp, 0, be16_to_cpu(leaf->hdr.count) - 1);
 114         needscan = 0;
 115         needlog = 1;
 116         /*
 117          * Make the space formerly occupied by the leaf entries and block
 118          * tail be free.
 119          */
 120         xfs_dir2_data_make_free(tp, dbp,
 121                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 122                 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
 123                                        (char *)blp),
 124                 &needlog, &needscan);
 125         /*
 126          * Fix up the block header, make it a data block.
 127          */
 128         hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 129         if (needscan)
 130                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 131         /*
 132          * Set up leaf tail and bests table.
 133          */
 134         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 135         ltp->bestcount = cpu_to_be32(1);
 136         bestsp = xfs_dir2_leaf_bests_p(ltp);
 137         bestsp[0] =  hdr->bestfree[0].length;
 138         /*
 139          * Log the data header and leaf bests table.
 140          */
 141         if (needlog)
 142                 xfs_dir2_data_log_header(tp, dbp);
 143         xfs_dir2_leaf_check(dp, lbp);
 144         xfs_dir2_data_check(dp, dbp);
 145         xfs_dir2_leaf_log_bests(tp, lbp, 0, 0);
 146         xfs_da_buf_done(lbp);
 147         return 0;
 148 }
 149
 150 STATIC void
 151 xfs_dir2_leaf_find_stale(
 152         struct xfs_dir2_leaf    *leaf,
 153         int                     index,
 154         int                     *lowstale,
 155         int                     *highstale)
 156 {
 157         /*
 158          * Find the first stale entry before our index, if any.
 159          */
 160         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
 161                 if (leaf->ents[*lowstale].address ==
 162                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 163                         break;
 164         }
 165
 166         /*
 167          * Find the first stale entry at or after our index, if any.
 168          * Stop if the result would require moving more entries than using
 169          * lowstale.
 170          */
 171         for (*highstale = index;
 172              *highstale < be16_to_cpu(leaf->hdr.count);
 173              ++*highstale) {
 174                 if (leaf->ents[*highstale].address ==
 175                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 176                         break;
 177                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
 178                         break;
 179         }
 180 }
 181
 182 struct xfs_dir2_leaf_entry *
 183 xfs_dir2_leaf_find_entry(
 184         xfs_dir2_leaf_t         *leaf,          /* leaf structure */
 185         int                     index,          /* leaf table position */
 186         int                     compact,        /* need to compact leaves */
 187         int                     lowstale,       /* index of prev stale leaf */
 188         int                     highstale,      /* index of next stale leaf */
 189         int                     *lfloglow,      /* low leaf logging index */
 190         int                     *lfloghigh)     /* high leaf logging index */
 191 {
 192         if (!leaf->hdr.stale) {
 193                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
 194
 195                 /*
 196                  * Now we need to make room to insert the leaf entry.
 197                  *
 198                  * If there are no stale entries, just insert a hole at index.
 199                  */
 200                 lep = &leaf->ents[index];
 201                 if (index < be16_to_cpu(leaf->hdr.count))
 202                         memmove(lep + 1, lep,
 203                                 (be16_to_cpu(leaf->hdr.count) - index) *
 204                                  sizeof(*lep));
 205
 206                 /*
 207                  * Record low and high logging indices for the leaf.
 208                  */
 209                 *lfloglow = index;
 210                 *lfloghigh = be16_to_cpu(leaf->hdr.count);
 211                 be16_add_cpu(&leaf->hdr.count, 1);
 212                 return lep;
 213         }
 214
 215         /*
 216          * There are stale entries.
 217          *
 218          * We will use one of them for the new entry.  It's probably not at
 219          * the right location, so we'll have to shift some up or down first.
 220          *
 221          * If we didn't compact before, we need to find the nearest stale
 222          * entries before and after our insertion point.
 223          */
 224         if (compact == 0)
 225                 xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
 226
 227         /*
 228          * If the low one is better, use it.
 229          */
 230         if (lowstale >= 0 &&
 231             (highstale == be16_to_cpu(leaf->hdr.count) ||
 232              index - lowstale - 1 < highstale - index)) {
 233                 ASSERT(index - lowstale - 1 >= 0);
 234                 ASSERT(leaf->ents[lowstale].address ==
 235                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 236
 237                 /*
 238                  * Copy entries up to cover the stale entry and make room
 239                  * for the new entry.
 240                  */
 241                 if (index - lowstale - 1 > 0) {
 242                         memmove(&leaf->ents[lowstale],
 243                                 &leaf->ents[lowstale + 1],
 244                                 (index - lowstale - 1) *
 245                                 sizeof(xfs_dir2_leaf_entry_t));
 246                 }
 247                 *lfloglow = MIN(lowstale, *lfloglow);
 248                 *lfloghigh = MAX(index - 1, *lfloghigh);
 249                 be16_add_cpu(&leaf->hdr.stale, -1);
 250                 return &leaf->ents[index - 1];
 251         }
 252
 253         /*
 254          * The high one is better, so use that one.
 255          */
 256         ASSERT(highstale - index >= 0);
 257         ASSERT(leaf->ents[highstale].address ==
 258                cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 259
 260         /*
 261          * Copy entries down to cover the stale entry and make room for the
 262          * new entry.
 263          */
 264         if (highstale - index > 0) {
 265                 memmove(&leaf->ents[index + 1],
 266                         &leaf->ents[index],
 267                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
 268         }
 269         *lfloglow = MIN(index, *lfloglow);
 270         *lfloghigh = MAX(highstale, *lfloghigh);
 271         be16_add_cpu(&leaf->hdr.stale, -1);
 272         return &leaf->ents[index];
 273 }
 274
 275 /*
 276  * Add an entry to a leaf form directory.
 277  */
 278 int                                             /* error */
 279 xfs_dir2_leaf_addname(
 280         xfs_da_args_t           *args)          /* operation arguments */
 281 {
 282         __be16                  *bestsp;        /* freespace table in leaf */
 283         int                     compact;        /* need to compact leaves */
 284         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 285         xfs_dabuf_t             *dbp;           /* data block buffer */
 286         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 287         xfs_inode_t             *dp;            /* incore directory inode */
 288         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
 289         int                     error;          /* error return value */
 290         int                     grown;          /* allocated new data block */
 291         int                     highstale;      /* index of next stale leaf */
 292         int                     i;              /* temporary, index */
 293         int                     index;          /* leaf table position */
 294         xfs_dabuf_t             *lbp;           /* leaf's buffer */
 295         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 296         int                     length;         /* length of new entry */
 297         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
 298         int                     lfloglow;       /* low leaf logging index */
 299         int                     lfloghigh;      /* high leaf logging index */
 300         int                     lowstale;       /* index of prev stale leaf */
 301         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 302         xfs_mount_t             *mp;            /* filesystem mount point */
 303         int                     needbytes;      /* leaf block bytes needed */
 304         int                     needlog;        /* need to log data header */
 305         int                     needscan;       /* need to rescan data free */
 306         __be16                  *tagp;          /* end of data entry */
 307         xfs_trans_t             *tp;            /* transaction pointer */
 308         xfs_dir2_db_t           use_block;      /* data block number */
 309
 310         trace_xfs_dir2_leaf_addname(args);
 311
 312         dp = args->dp;
 313         tp = args->trans;
 314         mp = dp->i_mount;
 315         /*
 316          * Read the leaf block.
 317          */
 318         error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
 319                 XFS_DATA_FORK);
 320         if (error) {
 321                 return error;
 322         }
 323         ASSERT(lbp != NULL);
 324         /*
 325          * Look up the entry by hash value and name.
 326          * We know it's not there, our caller has already done a lookup.
 327          * So the index is of the entry to insert in front of.
 328          * But if there are dup hash values the index is of the first of those.
 329          */
 330         index = xfs_dir2_leaf_search_hash(args, lbp);
 331         leaf = lbp->data;
 332         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 333         bestsp = xfs_dir2_leaf_bests_p(ltp);
 334         length = xfs_dir2_data_entsize(args->namelen);
 335         /*
 336          * See if there are any entries with the same hash value
 337          * and space in their block for the new entry.
 338          * This is good because it puts multiple same-hash value entries
 339          * in a data block, improving the lookup of those entries.
 340          */
 341         for (use_block = -1, lep = &leaf->ents[index];
 342              index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
 343              index++, lep++) {
 344                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 345                         continue;
 346                 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 347                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 348                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
 349                 if (be16_to_cpu(bestsp[i]) >= length) {
 350                         use_block = i;
 351                         break;
 352                 }
 353         }
 354         /*
 355          * Didn't find a block yet, linear search all the data blocks.
 356          */
 357         if (use_block == -1) {
 358                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 359                         /*
 360                          * Remember a block we see that's missing.
 361                          */
 362                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
 363                             use_block == -1)
 364                                 use_block = i;
 365                         else if (be16_to_cpu(bestsp[i]) >= length) {
 366                                 use_block = i;
 367                                 break;
 368                         }
 369                 }
 370         }
 371         /*
 372          * How many bytes do we need in the leaf block?
 373          */
 374         needbytes = 0;
 375         if (!leaf->hdr.stale)
 376                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
 377         if (use_block == -1)
 378                 needbytes += sizeof(xfs_dir2_data_off_t);
 379
 380         /*
 381          * Now kill use_block if it refers to a missing block, so we
 382          * can use it as an indication of allocation needed.
 383          */
 384         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
 385                 use_block = -1;
 386         /*
 387          * If we don't have enough free bytes but we can make enough
 388          * by compacting out stale entries, we'll do that.
 389          */
 390         if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <
 391                                 needbytes && be16_to_cpu(leaf->hdr.stale) > 1) {
 392                 compact = 1;
 393         }
 394         /*
 395          * Otherwise if we don't have enough free bytes we need to
 396          * convert to node form.
 397          */
 398         else if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(
 399                                                 leaf->hdr.count)] < needbytes) {
 400                 /*
 401                  * Just checking or no space reservation, give up.
 402                  */
 403                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 404                                                         args->total == 0) {
 405                         xfs_da_brelse(tp, lbp);
 406                         return XFS_ERROR(ENOSPC);
 407                 }
 408                 /*
 409                  * Convert to node form.
 410                  */
 411                 error = xfs_dir2_leaf_to_node(args, lbp);
 412                 xfs_da_buf_done(lbp);
 413                 if (error)
 414                         return error;
 415                 /*
 416                  * Then add the new entry.
 417                  */
 418                 return xfs_dir2_node_addname(args);
 419         }
 420         /*
 421          * Otherwise it will fit without compaction.
 422          */
 423         else
 424                 compact = 0;
 425         /*
 426          * If just checking, then it will fit unless we needed to allocate
 427          * a new data block.
 428          */
 429         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 430                 xfs_da_brelse(tp, lbp);
 431                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
 432         }
 433         /*
 434          * If no allocations are allowed, return now before we've
 435          * changed anything.
 436          */
 437         if (args->total == 0 && use_block == -1) {
 438                 xfs_da_brelse(tp, lbp);
 439                 return XFS_ERROR(ENOSPC);
 440         }
 441         /*
 442          * Need to compact the leaf entries, removing stale ones.
 443          * Leave one stale entry behind - the one closest to our
 444          * insertion index - and we'll shift that one to our insertion
 445          * point later.
 446          */
 447         if (compact) {
 448                 xfs_dir2_leaf_compact_x1(lbp, &index, &lowstale, &highstale,
 449                         &lfloglow, &lfloghigh);
 450         }
 451         /*
 452          * There are stale entries, so we'll need log-low and log-high
 453          * impossibly bad values later.
 454          */
 455         else if (be16_to_cpu(leaf->hdr.stale)) {
 456                 lfloglow = be16_to_cpu(leaf->hdr.count);
 457                 lfloghigh = -1;
 458         }
 459         /*
 460          * If there was no data block space found, we need to allocate
 461          * a new one.
 462          */
 463         if (use_block == -1) {
 464                 /*
 465                  * Add the new data block.
 466                  */
 467                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 468                                 &use_block))) {
 469                         xfs_da_brelse(tp, lbp);
 470                         return error;
 471                 }
 472                 /*
 473                  * Initialize the block.
 474                  */
 475                 if ((error = xfs_dir2_data_init(args, use_block, &dbp))) {
 476                         xfs_da_brelse(tp, lbp);
 477                         return error;
 478                 }
 479                 /*
 480                  * If we're adding a new data block on the end we need to
 481                  * extend the bests table.  Copy it up one entry.
 482                  */
 483                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
 484                         bestsp--;
 485                         memmove(&bestsp[0], &bestsp[1],
 486                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 487                         be32_add_cpu(&ltp->bestcount, 1);
 488                         xfs_dir2_leaf_log_tail(tp, lbp);
 489                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
 490                 }
 491                 /*
 492                  * If we're filling in a previously empty block just log it.
 493                  */
 494                 else
 495                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 496                 hdr = dbp->data;
 497                 bestsp[use_block] = hdr->bestfree[0].length;
 498                 grown = 1;
 499         }
 500         /*
 501          * Already had space in some data block.
 502          * Just read that one in.
 503          */
 504         else {
 505                 if ((error =
 506                     xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, use_block),
 507                             -1, &dbp, XFS_DATA_FORK))) {
 508                         xfs_da_brelse(tp, lbp);
 509                         return error;
 510                 }
 511                 hdr = dbp->data;
 512                 grown = 0;
 513         }
 514         xfs_dir2_data_check(dp, dbp);
 515         /*
 516          * Point to the biggest freespace in our data block.
 517          */
 518         dup = (xfs_dir2_data_unused_t *)
 519               ((char *)hdr + be16_to_cpu(hdr->bestfree[0].offset));
 520         ASSERT(be16_to_cpu(dup->length) >= length);
 521         needscan = needlog = 0;
 522         /*
 523          * Mark the initial part of our freespace in use for the new entry.
 524          */
 525         xfs_dir2_data_use_free(tp, dbp, dup,
 526                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
 527                 &needlog, &needscan);
 528         /*
 529          * Initialize our new entry (at last).
 530          */
 531         dep = (xfs_dir2_data_entry_t *)dup;
 532         dep->inumber = cpu_to_be64(args->inumber);
 533         dep->namelen = args->namelen;
 534         memcpy(dep->name, args->name, dep->namelen);
 535         tagp = xfs_dir2_data_entry_tag_p(dep);
 536         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 537         /*
 538          * Need to scan fix up the bestfree table.
 539          */
 540         if (needscan)
 541                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 542         /*
 543          * Need to log the data block's header.
 544          */
 545         if (needlog)
 546                 xfs_dir2_data_log_header(tp, dbp);
 547         xfs_dir2_data_log_entry(tp, dbp, dep);
 548         /*
 549          * If the bests table needs to be changed, do it.
 550          * Log the change unless we've already done that.
 551          */
 552         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(hdr->bestfree[0].length)) {
 553                 bestsp[use_block] = hdr->bestfree[0].length;
 554                 if (!grown)
 555                         xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
 556         }
 557
 558         lep = xfs_dir2_leaf_find_entry(leaf, index, compact, lowstale,
 559                                        highstale, &lfloglow, &lfloghigh);
 560
 561         /*
 562          * Fill in the new leaf entry.
 563          */
 564         lep->hashval = cpu_to_be32(args->hashval);
 565         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
 566                                 be16_to_cpu(*tagp)));
 567         /*
 568          * Log the leaf fields and give up the buffers.
 569          */
 570         xfs_dir2_leaf_log_header(tp, lbp);
 571         xfs_dir2_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
 572         xfs_dir2_leaf_check(dp, lbp);
 573         xfs_da_buf_done(lbp);
 574         xfs_dir2_data_check(dp, dbp);
 575         xfs_da_buf_done(dbp);
 576         return 0;
 577 }
 578
 579 #ifdef DEBUG
 580 /*
 581  * Check the internal consistency of a leaf1 block.
 582  * Pop an assert if something is wrong.
 583  */
 584 STATIC void
 585 xfs_dir2_leaf_check(
 586         xfs_inode_t             *dp,            /* incore directory inode */
 587         xfs_dabuf_t             *bp)            /* leaf's buffer */
 588 {
 589         int                     i;              /* leaf index */
 590         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 591         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 592         xfs_mount_t             *mp;            /* filesystem mount point */
 593         int                     stale;          /* count of stale leaves */
 594
 595         leaf = bp->data;
 596         mp = dp->i_mount;
 597         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
 598         /*
 599          * This value is not restrictive enough.
 600          * Should factor in the size of the bests table as well.
 601          * We can deduce a value for that from di_size.
 602          */
 603         ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
 604         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 605         /*
 606          * Leaves and bests don't overlap.
 607          */
 608         ASSERT((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <=
 609                (char *)xfs_dir2_leaf_bests_p(ltp));
 610         /*
 611          * Check hash value order, count stale entries.
 612          */
 613         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
 614                 if (i + 1 < be16_to_cpu(leaf->hdr.count))
 615                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
 616                                be32_to_cpu(leaf->ents[i + 1].hashval));
 617                 if (leaf->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 618                         stale++;
 619         }
 620         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
 621 }
 622 #endif  /* DEBUG */
 623
 624 /*
 625  * Compact out any stale entries in the leaf.
 626  * Log the header and changed leaf entries, if any.
 627  */
 628 void
 629 xfs_dir2_leaf_compact(
 630         xfs_da_args_t   *args,          /* operation arguments */
 631         xfs_dabuf_t     *bp)            /* leaf buffer */
 632 {
 633         int             from;           /* source leaf index */
 634         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 635         int             loglow;         /* first leaf entry to log */
 636         int             to;             /* target leaf index */
 637
 638         leaf = bp->data;
 639         if (!leaf->hdr.stale) {
 640                 return;
 641         }
 642         /*
 643          * Compress out the stale entries in place.
 644          */
 645         for (from = to = 0, loglow = -1; from < be16_to_cpu(leaf->hdr.count); from++) {
 646                 if (leaf->ents[from].address ==
 647                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 648                         continue;
 649                 /*
 650                  * Only actually copy the entries that are different.
 651                  */
 652                 if (from > to) {
 653                         if (loglow == -1)
 654                                 loglow = to;
 655                         leaf->ents[to] = leaf->ents[from];
 656                 }
 657                 to++;
 658         }
 659         /*
 660          * Update and log the header, log the leaf entries.
 661          */
 662         ASSERT(be16_to_cpu(leaf->hdr.stale) == from - to);
 663         be16_add_cpu(&leaf->hdr.count, -(be16_to_cpu(leaf->hdr.stale)));
 664         leaf->hdr.stale = 0;
 665         xfs_dir2_leaf_log_header(args->trans, bp);
 666         if (loglow != -1)
 667                 xfs_dir2_leaf_log_ents(args->trans, bp, loglow, to - 1);
 668 }
 669
 670 /*
 671  * Compact the leaf entries, removing stale ones.
 672  * Leave one stale entry behind - the one closest to our
 673  * insertion index - and the caller will shift that one to our insertion
 674  * point later.
 675  * Return new insertion index, where the remaining stale entry is,
 676  * and leaf logging indices.
 677  */
 678 void
 679 xfs_dir2_leaf_compact_x1(
 680         xfs_dabuf_t     *bp,            /* leaf buffer */
 681         int             *indexp,        /* insertion index */
 682         int             *lowstalep,     /* out: stale entry before us */
 683         int             *highstalep,    /* out: stale entry after us */
 684         int             *lowlogp,       /* out: low log index */
 685         int             *highlogp)      /* out: high log index */
 686 {
 687         int             from;           /* source copy index */
 688         int             highstale;      /* stale entry at/after index */
 689         int             index;          /* insertion index */
 690         int             keepstale;      /* source index of kept stale */
 691         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 692         int             lowstale;       /* stale entry before index */
 693         int             newindex=0;     /* new insertion index */
 694         int             to;             /* destination copy index */
 695
 696         leaf = bp->data;
 697         ASSERT(be16_to_cpu(leaf->hdr.stale) > 1);
 698         index = *indexp;
 699
 700         xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
 701
 702         /*
 703          * Pick the better of lowstale and highstale.
 704          */
 705         if (lowstale >= 0 &&
 706             (highstale == be16_to_cpu(leaf->hdr.count) ||
 707              index - lowstale <= highstale - index))
 708                 keepstale = lowstale;
 709         else
 710                 keepstale = highstale;
 711         /*
 712          * Copy the entries in place, removing all the stale entries
 713          * except keepstale.
 714          */
 715         for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
 716                 /*
 717                  * Notice the new value of index.
 718                  */
 719                 if (index == from)
 720                         newindex = to;
 721                 if (from != keepstale &&
 722                     leaf->ents[from].address ==
 723                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
 724                         if (from == to)
 725                                 *lowlogp = to;
 726                         continue;
 727                 }
 728                 /*
 729                  * Record the new keepstale value for the insertion.
 730                  */
 731                 if (from == keepstale)
 732                         lowstale = highstale = to;
 733                 /*
 734                  * Copy only the entries that have moved.
 735                  */
 736                 if (from > to)
 737                         leaf->ents[to] = leaf->ents[from];
 738                 to++;
 739         }
 740         ASSERT(from > to);
 741         /*
 742          * If the insertion point was past the last entry,
 743          * set the new insertion point accordingly.
 744          */
 745         if (index == from)
 746                 newindex = to;
 747         *indexp = newindex;
 748         /*
 749          * Adjust the leaf header values.
 750          */
 751         be16_add_cpu(&leaf->hdr.count, -(from - to));
 752         leaf->hdr.stale = cpu_to_be16(1);
 753         /*
 754          * Remember the low/high stale value only in the "right"
 755          * direction.
 756          */
 757         if (lowstale >= newindex)
 758                 lowstale = -1;
 759         else
 760                 highstale = be16_to_cpu(leaf->hdr.count);
 761         *highlogp = be16_to_cpu(leaf->hdr.count) - 1;
 762         *lowstalep = lowstale;
 763         *highstalep = highstale;
 764 }
 765
 766 /*
 767  * Getdents (readdir) for leaf and node directories.
 768  * This reads the data blocks only, so is the same for both forms.
 769  */
 770 int                                             /* error */
 771 xfs_dir2_leaf_getdents(
 772         xfs_inode_t             *dp,            /* incore directory inode */
 773         void                    *dirent,
 774         size_t                  bufsize,
 775         xfs_off_t               *offset,
 776         filldir_t               filldir)
 777 {
 778         xfs_dabuf_t             *bp;            /* data block buffer */
 779         int                     byteoff;        /* offset in current block */
 780         xfs_dir2_db_t           curdb;          /* db for current block */
 781         xfs_dir2_off_t          curoff;         /* current overall offset */
 782         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 783         xfs_dir2_data_entry_t   *dep;           /* data entry */
 784         xfs_dir2_data_unused_t  *dup;           /* unused entry */
 785         int                     error = 0;      /* error return value */
 786         int                     i;              /* temporary loop index */
 787         int                     j;              /* temporary loop index */
 788         int                     length;         /* temporary length value */
 789         xfs_bmbt_irec_t         *map;           /* map vector for blocks */
 790         xfs_extlen_t            map_blocks;     /* number of fsbs in map */
 791         xfs_dablk_t             map_off;        /* last mapped file offset */
 792         int                     map_size;       /* total entries in *map */
 793         int                     map_valid;      /* valid entries in *map */
 794         xfs_mount_t             *mp;            /* filesystem mount point */
 795         xfs_dir2_off_t          newoff;         /* new curoff after new blk */
 796         int                     nmap;           /* mappings to ask xfs_bmapi */
 797         char                    *ptr = NULL;    /* pointer to current data */
 798         int                     ra_current;     /* number of read-ahead blks */
 799         int                     ra_index;       /* *map index for read-ahead */
 800         int                     ra_offset;      /* map entry offset for ra */
 801         int                     ra_want;        /* readahead count wanted */
 802
 803         /*
 804          * If the offset is at or past the largest allowed value,
 805          * give up right away.
 806          */
 807         if (*offset >= XFS_DIR2_MAX_DATAPTR)
 808                 return 0;
 809
 810         mp = dp->i_mount;
 811
 812         /*
 813          * Set up to bmap a number of blocks based on the caller's
 814          * buffer size, the directory block size, and the filesystem
 815          * block size.
 816          */
 817         map_size = howmany(bufsize + mp->m_dirblksize, mp->m_sb.sb_blocksize);
 818         map = kmem_alloc(map_size * sizeof(*map), KM_SLEEP);
 819         map_valid = ra_index = ra_offset = ra_current = map_blocks = 0;
 820         bp = NULL;
 821
 822         /*
 823          * Inside the loop we keep the main offset value as a byte offset
 824          * in the directory file.
 825          */
 826         curoff = xfs_dir2_dataptr_to_byte(mp, *offset);
 827
 828         /*
 829          * Force this conversion through db so we truncate the offset
 830          * down to get the start of the data block.
 831          */
 832         map_off = xfs_dir2_db_to_da(mp, xfs_dir2_byte_to_db(mp, curoff));
 833         /*
 834          * Loop over directory entries until we reach the end offset.
 835          * Get more blocks and readahead as necessary.
 836          */
 837         while (curoff < XFS_DIR2_LEAF_OFFSET) {
 838                 /*
 839                  * If we have no buffer, or we're off the end of the
 840                  * current buffer, need to get another one.
 841                  */
 842                 if (!bp || ptr >= (char *)bp->data + mp->m_dirblksize) {
 843                         /*
 844                          * If we have a buffer, we need to release it and
 845                          * take it out of the mapping.
 846                          */
 847                         if (bp) {
 848                                 xfs_da_brelse(NULL, bp);
 849                                 bp = NULL;
 850                                 map_blocks -= mp->m_dirblkfsbs;
 851                                 /*
 852                                  * Loop to get rid of the extents for the
 853                                  * directory block.
 854                                  */
 855                                 for (i = mp->m_dirblkfsbs; i > 0; ) {
 856                                         j = MIN((int)map->br_blockcount, i);
 857                                         map->br_blockcount -= j;
 858                                         map->br_startblock += j;
 859                                         map->br_startoff += j;
 860                                         /*
 861                                          * If mapping is done, pitch it from
 862                                          * the table.
 863                                          */
 864                                         if (!map->br_blockcount && --map_valid)
 865                                                 memmove(&map[0], &map[1],
 866                                                         sizeof(map[0]) *
 867                                                         map_valid);
 868                                         i -= j;
 869                                 }
 870                         }
 871                         /*
 872                          * Recalculate the readahead blocks wanted.
 873                          */
 874                         ra_want = howmany(bufsize + mp->m_dirblksize,
 875                                           mp->m_sb.sb_blocksize) - 1;
 876                         ASSERT(ra_want >= 0);
 877
 878                         /*
 879                          * If we don't have as many as we want, and we haven't
 880                          * run out of data blocks, get some more mappings.
 881                          */
 882                         if (1 + ra_want > map_blocks &&
 883                             map_off <
 884                             xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET)) {
 885                                 /*
 886                                  * Get more bmaps, fill in after the ones
 887                                  * we already have in the table.
 888                                  */
 889                                 nmap = map_size - map_valid;
 890                                 error = xfs_bmapi_read(dp, map_off,
 891                                         xfs_dir2_byte_to_da(mp,
 892                                                 XFS_DIR2_LEAF_OFFSET) - map_off,
 893                                         &map[map_valid], &nmap, 0);
 894                                 /*
 895                                  * Don't know if we should ignore this or
 896                                  * try to return an error.
 897                                  * The trouble with returning errors
 898                                  * is that readdir will just stop without
 899                                  * actually passing the error through.
 900                                  */
 901                                 if (error)
 902                                         break;  /* XXX */
 903                                 /*
 904                                  * If we got all the mappings we asked for,
 905                                  * set the final map offset based on the
 906                                  * last bmap value received.
 907                                  * Otherwise, we've reached the end.
 908                                  */
 909                                 if (nmap == map_size - map_valid)
 910                                         map_off =
 911                                         map[map_valid + nmap - 1].br_startoff +
 912                                         map[map_valid + nmap - 1].br_blockcount;
 913                                 else
 914                                         map_off =
 915                                                 xfs_dir2_byte_to_da(mp,
 916                                                         XFS_DIR2_LEAF_OFFSET);
 917                                 /*
 918                                  * Look for holes in the mapping, and
 919                                  * eliminate them.  Count up the valid blocks.
 920                                  */
 921                                 for (i = map_valid; i < map_valid + nmap; ) {
 922                                         if (map[i].br_startblock ==
 923                                             HOLESTARTBLOCK) {
 924                                                 nmap--;
 925                                                 length = map_valid + nmap - i;
 926                                                 if (length)
 927                                                         memmove(&map[i],
 928                                                                 &map[i + 1],
 929                                                                 sizeof(map[i]) *
 930                                                                 length);
 931                                         } else {
 932                                                 map_blocks +=
 933                                                         map[i].br_blockcount;
 934                                                 i++;
 935                                         }
 936                                 }
 937                                 map_valid += nmap;
 938                         }
 939                         /*
 940                          * No valid mappings, so no more data blocks.
 941                          */
 942                         if (!map_valid) {
 943                                 curoff = xfs_dir2_da_to_byte(mp, map_off);
 944                                 break;
 945                         }
 946                         /*
 947                          * Read the directory block starting at the first
 948                          * mapping.
 949                          */
 950                         curdb = xfs_dir2_da_to_db(mp, map->br_startoff);
 951                         error = xfs_da_read_buf(NULL, dp, map->br_startoff,
 952                                 map->br_blockcount >= mp->m_dirblkfsbs ?
 953                                     XFS_FSB_TO_DADDR(mp, map->br_startblock) :
 954                                     -1,
 955                                 &bp, XFS_DATA_FORK);
 956                         /*
 957                          * Should just skip over the data block instead
 958                          * of giving up.
 959                          */
 960                         if (error)
 961                                 break;  /* XXX */
 962                         /*
 963                          * Adjust the current amount of read-ahead: we just
 964                          * read a block that was previously ra.
 965                          */
 966                         if (ra_current)
 967                                 ra_current -= mp->m_dirblkfsbs;
 968                         /*
 969                          * Do we need more readahead?
 970                          */
 971                         for (ra_index = ra_offset = i = 0;
 972                              ra_want > ra_current && i < map_blocks;
 973                              i += mp->m_dirblkfsbs) {
 974                                 ASSERT(ra_index < map_valid);
 975                                 /*
 976                                  * Read-ahead a contiguous directory block.
 977                                  */
 978                                 if (i > ra_current &&
 979                                     map[ra_index].br_blockcount >=
 980                                     mp->m_dirblkfsbs) {
 981                                         xfs_buf_readahead(mp->m_ddev_targp,
 982                                                 XFS_FSB_TO_DADDR(mp,
 983                                                    map[ra_index].br_startblock +
 984                                                    ra_offset),
 985                                                 (int)BTOBB(mp->m_dirblksize));
 986                                         ra_current = i;
 987                                 }
 988                                 /*
 989                                  * Read-ahead a non-contiguous directory block.
 990                                  * This doesn't use our mapping, but this
 991                                  * is a very rare case.
 992                                  */
 993                                 else if (i > ra_current) {
 994                                         (void)xfs_da_reada_buf(NULL, dp,
 995                                                 map[ra_index].br_startoff +
 996                                                 ra_offset, XFS_DATA_FORK);
 997                                         ra_current = i;
 998                                 }
 999                                 /*
1000                                  * Advance offset through the mapping table.
1001                                  */
1002                                 for (j = 0; j < mp->m_dirblkfsbs; j++) {
1003                                         /*
1004                                          * The rest of this extent but not
1005                                          * more than a dir block.
1006                                          */
1007                                         length = MIN(mp->m_dirblkfsbs,
1008                                                 (int)(map[ra_index].br_blockcount -
1009                                                 ra_offset));
1010                                         j += length;
1011                                         ra_offset += length;
1012                                         /*
1013                                          * Advance to the next mapping if
1014                                          * this one is used up.
1015                                          */
1016                                         if (ra_offset ==
1017                                             map[ra_index].br_blockcount) {
1018                                                 ra_offset = 0;
1019                                                 ra_index++;
1020                                         }
1021                                 }
1022                         }
1023                         /*
1024                          * Having done a read, we need to set a new offset.
1025                          */
1026                         newoff = xfs_dir2_db_off_to_byte(mp, curdb, 0);
1027                         /*
1028                          * Start of the current block.
1029                          */
1030                         if (curoff < newoff)
1031                                 curoff = newoff;
1032                         /*
1033                          * Make sure we're in the right block.
1034                          */
1035                         else if (curoff > newoff)
1036                                 ASSERT(xfs_dir2_byte_to_db(mp, curoff) ==
1037                                        curdb);
1038                         hdr = bp->data;
1039                         xfs_dir2_data_check(dp, bp);
1040                         /*
1041                          * Find our position in the block.
1042                          */
1043                         ptr = (char *)(hdr + 1);
1044                         byteoff = xfs_dir2_byte_to_off(mp, curoff);
1045                         /*
1046                          * Skip past the header.
1047                          */
1048                         if (byteoff == 0)
1049                                 curoff += (uint)sizeof(*hdr);
1050                         /*
1051                          * Skip past entries until we reach our offset.
1052                          */
1053                         else {
1054                                 while ((char *)ptr - (char *)hdr < byteoff) {
1055                                         dup = (xfs_dir2_data_unused_t *)ptr;
1056
1057                                         if (be16_to_cpu(dup->freetag)
1058                                                   == XFS_DIR2_DATA_FREE_TAG) {
1059
1060                                                 length = be16_to_cpu(dup->length);
1061                                                 ptr += length;
1062                                                 continue;
1063                                         }
1064                                         dep = (xfs_dir2_data_entry_t *)ptr;
1065                                         length =
1066                                            xfs_dir2_data_entsize(dep->namelen);
1067                                         ptr += length;
1068                                 }
1069                                 /*
1070                                  * Now set our real offset.
1071                                  */
1072                                 curoff =
1073                                         xfs_dir2_db_off_to_byte(mp,
1074                                             xfs_dir2_byte_to_db(mp, curoff),
1075                                             (char *)ptr - (char *)hdr);
1076                                 if (ptr >= (char *)hdr + mp->m_dirblksize) {
1077                                         continue;
1078                                 }
1079                         }
1080                 }
1081                 /*
1082                  * We have a pointer to an entry.
1083                  * Is it a live one?
1084                  */
1085                 dup = (xfs_dir2_data_unused_t *)ptr;
1086                 /*
1087                  * No, it's unused, skip over it.
1088                  */
1089                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
1090                         length = be16_to_cpu(dup->length);
1091                         ptr += length;
1092                         curoff += length;
1093                         continue;
1094                 }
1095
1096                 dep = (xfs_dir2_data_entry_t *)ptr;
1097                 length = xfs_dir2_data_entsize(dep->namelen);
1098
1099                 if (filldir(dirent, (char *)dep->name, dep->namelen,
1100                             xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff,
1101                             be64_to_cpu(dep->inumber), DT_UNKNOWN))
1102                         break;
1103
1104                 /*
1105                  * Advance to next entry in the block.
1106                  */
1107                 ptr += length;
1108                 curoff += length;
1109                 /* bufsize may have just been a guess; don't go negative */
1110                 bufsize = bufsize > length ? bufsize - length : 0;
1111         }
1112
1113         /*
1114          * All done.  Set output offset value to current offset.
1115          */
1116         if (curoff > xfs_dir2_dataptr_to_byte(mp, XFS_DIR2_MAX_DATAPTR))
1117                 *offset = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
1118         else
1119                 *offset = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
1120         kmem_free(map);
1121         if (bp)
1122                 xfs_da_brelse(NULL, bp);
1123         return error;
1124 }
1125
1126 /*
1127  * Initialize a new leaf block, leaf1 or leafn magic accepted.
1128  */
1129 int
1130 xfs_dir2_leaf_init(
1131         xfs_da_args_t           *args,          /* operation arguments */
1132         xfs_dir2_db_t           bno,            /* directory block number */
1133         xfs_dabuf_t             **bpp,          /* out: leaf buffer */
1134         int                     magic)          /* magic number for block */
1135 {
1136         xfs_dabuf_t             *bp;            /* leaf buffer */
1137         xfs_inode_t             *dp;            /* incore directory inode */
1138         int                     error;          /* error return code */
1139         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1140         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1141         xfs_mount_t             *mp;            /* filesystem mount point */
1142         xfs_trans_t             *tp;            /* transaction pointer */
1143
1144         dp = args->dp;
1145         ASSERT(dp != NULL);
1146         tp = args->trans;
1147         mp = dp->i_mount;
1148         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
1149                bno < XFS_DIR2_FREE_FIRSTDB(mp));
1150         /*
1151          * Get the buffer for the block.
1152          */
1153         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
1154                 XFS_DATA_FORK);
1155         if (error) {
1156                 return error;
1157         }
1158         ASSERT(bp != NULL);
1159         leaf = bp->data;
1160         /*
1161          * Initialize the header.
1162          */
1163         leaf->hdr.info.magic = cpu_to_be16(magic);
1164         leaf->hdr.info.forw = 0;
1165         leaf->hdr.info.back = 0;
1166         leaf->hdr.count = 0;
1167         leaf->hdr.stale = 0;
1168         xfs_dir2_leaf_log_header(tp, bp);
1169         /*
1170          * If it's a leaf-format directory initialize the tail.
1171          * In this case our caller has the real bests table to copy into
1172          * the block.
1173          */
1174         if (magic == XFS_DIR2_LEAF1_MAGIC) {
1175                 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1176                 ltp->bestcount = 0;
1177                 xfs_dir2_leaf_log_tail(tp, bp);
1178         }
1179         *bpp = bp;
1180         return 0;
1181 }
1182
1183 /*
1184  * Log the bests entries indicated from a leaf1 block.
1185  */
1186 static void
1187 xfs_dir2_leaf_log_bests(
1188         xfs_trans_t             *tp,            /* transaction pointer */
1189         xfs_dabuf_t             *bp,            /* leaf buffer */
1190         int                     first,          /* first entry to log */
1191         int                     last)           /* last entry to log */
1192 {
1193         __be16                  *firstb;        /* pointer to first entry */
1194         __be16                  *lastb;         /* pointer to last entry */
1195         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1196         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1197
1198         leaf = bp->data;
1199         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1200         ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1201         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1202         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1203         xfs_da_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1204                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1205 }
1206
1207 /*
1208  * Log the leaf entries indicated from a leaf1 or leafn block.
1209  */
1210 void
1211 xfs_dir2_leaf_log_ents(
1212         xfs_trans_t             *tp,            /* transaction pointer */
1213         xfs_dabuf_t             *bp,            /* leaf buffer */
1214         int                     first,          /* first entry to log */
1215         int                     last)           /* last entry to log */
1216 {
1217         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1218         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1219         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1220
1221         leaf = bp->data;
1222         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1223                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1224         firstlep = &leaf->ents[first];
1225         lastlep = &leaf->ents[last];
1226         xfs_da_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1227                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1228 }
1229
1230 /*
1231  * Log the header of the leaf1 or leafn block.
1232  */
1233 void
1234 xfs_dir2_leaf_log_header(
1235         xfs_trans_t             *tp,            /* transaction pointer */
1236         xfs_dabuf_t             *bp)            /* leaf buffer */
1237 {
1238         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1239
1240         leaf = bp->data;
1241         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1242                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1243         xfs_da_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1244                 (uint)(sizeof(leaf->hdr) - 1));
1245 }
1246
1247 /*
1248  * Log the tail of the leaf1 block.
1249  */
1250 STATIC void
1251 xfs_dir2_leaf_log_tail(
1252         xfs_trans_t             *tp,            /* transaction pointer */
1253         xfs_dabuf_t             *bp)            /* leaf buffer */
1254 {
1255         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1256         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1257         xfs_mount_t             *mp;            /* filesystem mount point */
1258
1259         mp = tp->t_mountp;
1260         leaf = bp->data;
1261         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1262         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1263         xfs_da_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1264                 (uint)(mp->m_dirblksize - 1));
1265 }
1266
1267 /*
1268  * Look up the entry referred to by args in the leaf format directory.
1269  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1270  * is also used by the node-format code.
1271  */
1272 int
1273 xfs_dir2_leaf_lookup(
1274         xfs_da_args_t           *args)          /* operation arguments */
1275 {
1276         xfs_dabuf_t             *dbp;           /* data block buffer */
1277         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1278         xfs_inode_t             *dp;            /* incore directory inode */
1279         int                     error;          /* error return code */
1280         int                     index;          /* found entry index */
1281         xfs_dabuf_t             *lbp;           /* leaf buffer */
1282         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1283         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1284         xfs_trans_t             *tp;            /* transaction pointer */
1285
1286         trace_xfs_dir2_leaf_lookup(args);
1287
1288         /*
1289          * Look up name in the leaf block, returning both buffers and index.
1290          */
1291         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1292                 return error;
1293         }
1294         tp = args->trans;
1295         dp = args->dp;
1296         xfs_dir2_leaf_check(dp, lbp);
1297         leaf = lbp->data;
1298         /*
1299          * Get to the leaf entry and contained data entry address.
1300          */
1301         lep = &leaf->ents[index];
1302         /*
1303          * Point to the data entry.
1304          */
1305         dep = (xfs_dir2_data_entry_t *)
1306               ((char *)dbp->data +
1307                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1308         /*
1309          * Return the found inode number & CI name if appropriate
1310          */
1311         args->inumber = be64_to_cpu(dep->inumber);
1312         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1313         xfs_da_brelse(tp, dbp);
1314         xfs_da_brelse(tp, lbp);
1315         return XFS_ERROR(error);
1316 }
1317
1318 /*
1319  * Look up name/hash in the leaf block.
1320  * Fill in indexp with the found index, and dbpp with the data buffer.
1321  * If not found dbpp will be NULL, and ENOENT comes back.
1322  * lbpp will always be filled in with the leaf buffer unless there's an error.
1323  */
1324 static int                                      /* error */
1325 xfs_dir2_leaf_lookup_int(
1326         xfs_da_args_t           *args,          /* operation arguments */
1327         xfs_dabuf_t             **lbpp,         /* out: leaf buffer */
1328         int                     *indexp,        /* out: index in leaf block */
1329         xfs_dabuf_t             **dbpp)         /* out: data buffer */
1330 {
1331         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1332         xfs_dabuf_t             *dbp = NULL;    /* data buffer */
1333         xfs_dir2_data_entry_t   *dep;           /* data entry */
1334         xfs_inode_t             *dp;            /* incore directory inode */
1335         int                     error;          /* error return code */
1336         int                     index;          /* index in leaf block */
1337         xfs_dabuf_t             *lbp;           /* leaf buffer */
1338         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1339         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1340         xfs_mount_t             *mp;            /* filesystem mount point */
1341         xfs_dir2_db_t           newdb;          /* new data block number */
1342         xfs_trans_t             *tp;            /* transaction pointer */
1343         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1344         enum xfs_dacmp          cmp;            /* name compare result */
1345
1346         dp = args->dp;
1347         tp = args->trans;
1348         mp = dp->i_mount;
1349         /*
1350          * Read the leaf block into the buffer.
1351          */
1352         error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
1353                                                         XFS_DATA_FORK);
1354         if (error)
1355                 return error;
1356         *lbpp = lbp;
1357         leaf = lbp->data;
1358         xfs_dir2_leaf_check(dp, lbp);
1359         /*
1360          * Look for the first leaf entry with our hash value.
1361          */
1362         index = xfs_dir2_leaf_search_hash(args, lbp);
1363         /*
1364          * Loop over all the entries with the right hash value
1365          * looking to match the name.
1366          */
1367         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
1368                                 be32_to_cpu(lep->hashval) == args->hashval;
1369                                 lep++, index++) {
1370                 /*
1371                  * Skip over stale leaf entries.
1372                  */
1373                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1374                         continue;
1375                 /*
1376                  * Get the new data block number.
1377                  */
1378                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1379                 /*
1380                  * If it's not the same as the old data block number,
1381                  * need to pitch the old one and read the new one.
1382                  */
1383                 if (newdb != curdb) {
1384                         if (dbp)
1385                                 xfs_da_brelse(tp, dbp);
1386                         error = xfs_da_read_buf(tp, dp,
1387                                                 xfs_dir2_db_to_da(mp, newdb),
1388                                                 -1, &dbp, XFS_DATA_FORK);
1389                         if (error) {
1390                                 xfs_da_brelse(tp, lbp);
1391                                 return error;
1392                         }
1393                         xfs_dir2_data_check(dp, dbp);
1394                         curdb = newdb;
1395                 }
1396                 /*
1397                  * Point to the data entry.
1398                  */
1399                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->data +
1400                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1401                 /*
1402                  * Compare name and if it's an exact match, return the index
1403                  * and buffer. If it's the first case-insensitive match, store
1404                  * the index and buffer and continue looking for an exact match.
1405                  */
1406                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1407                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1408                         args->cmpresult = cmp;
1409                         *indexp = index;
1410                         /* case exact match: return the current buffer. */
1411                         if (cmp == XFS_CMP_EXACT) {
1412                                 *dbpp = dbp;
1413                                 return 0;
1414                         }
1415                         cidb = curdb;
1416                 }
1417         }
1418         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1419         /*
1420          * Here, we can only be doing a lookup (not a rename or remove).
1421          * If a case-insensitive match was found earlier, re-read the
1422          * appropriate data block if required and return it.
1423          */
1424         if (args->cmpresult == XFS_CMP_CASE) {
1425                 ASSERT(cidb != -1);
1426                 if (cidb != curdb) {
1427                         xfs_da_brelse(tp, dbp);
1428                         error = xfs_da_read_buf(tp, dp,
1429                                                 xfs_dir2_db_to_da(mp, cidb),
1430                                                 -1, &dbp, XFS_DATA_FORK);
1431                         if (error) {
1432                                 xfs_da_brelse(tp, lbp);
1433                                 return error;
1434                         }
1435                 }
1436                 *dbpp = dbp;
1437                 return 0;
1438         }
1439         /*
1440          * No match found, return ENOENT.
1441          */
1442         ASSERT(cidb == -1);
1443         if (dbp)
1444                 xfs_da_brelse(tp, dbp);
1445         xfs_da_brelse(tp, lbp);
1446         return XFS_ERROR(ENOENT);
1447 }
1448
1449 /*
1450  * Remove an entry from a leaf format directory.
1451  */
1452 int                                             /* error */
1453 xfs_dir2_leaf_removename(
1454         xfs_da_args_t           *args)          /* operation arguments */
1455 {
1456         __be16                  *bestsp;        /* leaf block best freespace */
1457         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1458         xfs_dir2_db_t           db;             /* data block number */
1459         xfs_dabuf_t             *dbp;           /* data block buffer */
1460         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1461         xfs_inode_t             *dp;            /* incore directory inode */
1462         int                     error;          /* error return code */
1463         xfs_dir2_db_t           i;              /* temporary data block # */
1464         int                     index;          /* index into leaf entries */
1465         xfs_dabuf_t             *lbp;           /* leaf buffer */
1466         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1467         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1468         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1469         xfs_mount_t             *mp;            /* filesystem mount point */
1470         int                     needlog;        /* need to log data header */
1471         int                     needscan;       /* need to rescan data frees */
1472         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1473         xfs_trans_t             *tp;            /* transaction pointer */
1474
1475         trace_xfs_dir2_leaf_removename(args);
1476
1477         /*
1478          * Lookup the leaf entry, get the leaf and data blocks read in.
1479          */
1480         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1481                 return error;
1482         }
1483         dp = args->dp;
1484         tp = args->trans;
1485         mp = dp->i_mount;
1486         leaf = lbp->data;
1487         hdr = dbp->data;
1488         xfs_dir2_data_check(dp, dbp);
1489         /*
1490          * Point to the leaf entry, use that to point to the data entry.
1491          */
1492         lep = &leaf->ents[index];
1493         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1494         dep = (xfs_dir2_data_entry_t *)
1495               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1496         needscan = needlog = 0;
1497         oldbest = be16_to_cpu(hdr->bestfree[0].length);
1498         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1499         bestsp = xfs_dir2_leaf_bests_p(ltp);
1500         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1501         /*
1502          * Mark the former data entry unused.
1503          */
1504         xfs_dir2_data_make_free(tp, dbp,
1505                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1506                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1507         /*
1508          * We just mark the leaf entry stale by putting a null in it.
1509          */
1510         be16_add_cpu(&leaf->hdr.stale, 1);
1511         xfs_dir2_leaf_log_header(tp, lbp);
1512         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1513         xfs_dir2_leaf_log_ents(tp, lbp, index, index);
1514         /*
1515          * Scan the freespace in the data block again if necessary,
1516          * log the data block header if necessary.
1517          */
1518         if (needscan)
1519                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1520         if (needlog)
1521                 xfs_dir2_data_log_header(tp, dbp);
1522         /*
1523          * If the longest freespace in the data block has changed,
1524          * put the new value in the bests table and log that.
1525          */
1526         if (be16_to_cpu(hdr->bestfree[0].length) != oldbest) {
1527                 bestsp[db] = hdr->bestfree[0].length;
1528                 xfs_dir2_leaf_log_bests(tp, lbp, db, db);
1529         }
1530         xfs_dir2_data_check(dp, dbp);
1531         /*
1532          * If the data block is now empty then get rid of the data block.
1533          */
1534         if (be16_to_cpu(hdr->bestfree[0].length) ==
1535             mp->m_dirblksize - (uint)sizeof(*hdr)) {
1536                 ASSERT(db != mp->m_dirdatablk);
1537                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1538                         /*
1539                          * Nope, can't get rid of it because it caused
1540                          * allocation of a bmap btree block to do so.
1541                          * Just go on, returning success, leaving the
1542                          * empty block in place.
1543                          */
1544                         if (error == ENOSPC && args->total == 0) {
1545                                 xfs_da_buf_done(dbp);
1546                                 error = 0;
1547                         }
1548                         xfs_dir2_leaf_check(dp, lbp);
1549                         xfs_da_buf_done(lbp);
1550                         return error;
1551                 }
1552                 dbp = NULL;
1553                 /*
1554                  * If this is the last data block then compact the
1555                  * bests table by getting rid of entries.
1556                  */
1557                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1558                         /*
1559                          * Look for the last active entry (i).
1560                          */
1561                         for (i = db - 1; i > 0; i--) {
1562                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1563                                         break;
1564                         }
1565                         /*
1566                          * Copy the table down so inactive entries at the
1567                          * end are removed.
1568                          */
1569                         memmove(&bestsp[db - i], bestsp,
1570                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1571                         be32_add_cpu(&ltp->bestcount, -(db - i));
1572                         xfs_dir2_leaf_log_tail(tp, lbp);
1573                         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1574                 } else
1575                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1576         }
1577         /*
1578          * If the data block was not the first one, drop it.
1579          */
1580         else if (db != mp->m_dirdatablk && dbp != NULL) {
1581                 xfs_da_buf_done(dbp);
1582                 dbp = NULL;
1583         }
1584         xfs_dir2_leaf_check(dp, lbp);
1585         /*
1586          * See if we can convert to block form.
1587          */
1588         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1589 }
1590
1591 /*
1592  * Replace the inode number in a leaf format directory entry.
1593  */
1594 int                                             /* error */
1595 xfs_dir2_leaf_replace(
1596         xfs_da_args_t           *args)          /* operation arguments */
1597 {
1598         xfs_dabuf_t             *dbp;           /* data block buffer */
1599         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1600         xfs_inode_t             *dp;            /* incore directory inode */
1601         int                     error;          /* error return code */
1602         int                     index;          /* index of leaf entry */
1603         xfs_dabuf_t             *lbp;           /* leaf buffer */
1604         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1605         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1606         xfs_trans_t             *tp;            /* transaction pointer */
1607
1608         trace_xfs_dir2_leaf_replace(args);
1609
1610         /*
1611          * Look up the entry.
1612          */
1613         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1614                 return error;
1615         }
1616         dp = args->dp;
1617         leaf = lbp->data;
1618         /*
1619          * Point to the leaf entry, get data address from it.
1620          */
1621         lep = &leaf->ents[index];
1622         /*
1623          * Point to the data entry.
1624          */
1625         dep = (xfs_dir2_data_entry_t *)
1626               ((char *)dbp->data +
1627                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1628         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1629         /*
1630          * Put the new inode number in, log it.
1631          */
1632         dep->inumber = cpu_to_be64(args->inumber);
1633         tp = args->trans;
1634         xfs_dir2_data_log_entry(tp, dbp, dep);
1635         xfs_da_buf_done(dbp);
1636         xfs_dir2_leaf_check(dp, lbp);
1637         xfs_da_brelse(tp, lbp);
1638         return 0;
1639 }
1640
1641 /*
1642  * Return index in the leaf block (lbp) which is either the first
1643  * one with this hash value, or if there are none, the insert point
1644  * for that hash value.
1645  */
1646 int                                             /* index value */
1647 xfs_dir2_leaf_search_hash(
1648         xfs_da_args_t           *args,          /* operation arguments */
1649         xfs_dabuf_t             *lbp)           /* leaf buffer */
1650 {
1651         xfs_dahash_t            hash=0;         /* hash from this entry */
1652         xfs_dahash_t            hashwant;       /* hash value looking for */
1653         int                     high;           /* high leaf index */
1654         int                     low;            /* low leaf index */
1655         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1656         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1657         int                     mid=0;          /* current leaf index */
1658
1659         leaf = lbp->data;
1660 #ifndef __KERNEL__
1661         if (!leaf->hdr.count)
1662                 return 0;
1663 #endif
1664         /*
1665          * Note, the table cannot be empty, so we have to go through the loop.
1666          * Binary search the leaf entries looking for our hash value.
1667          */
1668         for (lep = leaf->ents, low = 0, high = be16_to_cpu(leaf->hdr.count) - 1,
1669                 hashwant = args->hashval;
1670              low <= high; ) {
1671                 mid = (low + high) >> 1;
1672                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1673                         break;
1674                 if (hash < hashwant)
1675                         low = mid + 1;
1676                 else
1677                         high = mid - 1;
1678         }
1679         /*
1680          * Found one, back up through all the equal hash values.
1681          */
1682         if (hash == hashwant) {
1683                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1684                         mid--;
1685                 }
1686         }
1687         /*
1688          * Need to point to an entry higher than ours.
1689          */
1690         else if (hash < hashwant)
1691                 mid++;
1692         return mid;
1693 }
1694
1695 /*
1696  * Trim off a trailing data block.  We know it's empty since the leaf
1697  * freespace table says so.
1698  */
1699 int                                             /* error */
1700 xfs_dir2_leaf_trim_data(
1701         xfs_da_args_t           *args,          /* operation arguments */
1702         xfs_dabuf_t             *lbp,           /* leaf buffer */
1703         xfs_dir2_db_t           db)             /* data block number */
1704 {
1705         __be16                  *bestsp;        /* leaf bests table */
1706         xfs_dabuf_t             *dbp;           /* data block buffer */
1707         xfs_inode_t             *dp;            /* incore directory inode */
1708         int                     error;          /* error return value */
1709         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1710         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1711         xfs_mount_t             *mp;            /* filesystem mount point */
1712         xfs_trans_t             *tp;            /* transaction pointer */
1713
1714         dp = args->dp;
1715         mp = dp->i_mount;
1716         tp = args->trans;
1717         /*
1718          * Read the offending data block.  We need its buffer.
1719          */
1720         if ((error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp,
1721                         XFS_DATA_FORK))) {
1722                 return error;
1723         }
1724
1725         leaf = lbp->data;
1726         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1727
1728 #ifdef DEBUG
1729 {
1730         struct xfs_dir2_data_hdr *hdr = dbp->data;
1731
1732         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
1733         ASSERT(be16_to_cpu(hdr->bestfree[0].length) ==
1734                mp->m_dirblksize - (uint)sizeof(*hdr));
1735         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1736 }
1737 #endif
1738
1739         /*
1740          * Get rid of the data block.
1741          */
1742         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1743                 ASSERT(error != ENOSPC);
1744                 xfs_da_brelse(tp, dbp);
1745                 return error;
1746         }
1747         /*
1748          * Eliminate the last bests entry from the table.
1749          */
1750         bestsp = xfs_dir2_leaf_bests_p(ltp);
1751         be32_add_cpu(&ltp->bestcount, -1);
1752         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1753         xfs_dir2_leaf_log_tail(tp, lbp);
1754         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1755         return 0;
1756 }
1757
1758 static inline size_t
1759 xfs_dir2_leaf_size(
1760         struct xfs_dir2_leaf_hdr        *hdr,
1761         int                             counts)
1762 {
1763         int                     entries;
1764
1765         entries = be16_to_cpu(hdr->count) - be16_to_cpu(hdr->stale);
1766         return sizeof(xfs_dir2_leaf_hdr_t) +
1767             entries * sizeof(xfs_dir2_leaf_entry_t) +
1768             counts * sizeof(xfs_dir2_data_off_t) +
1769             sizeof(xfs_dir2_leaf_tail_t);
1770 }
1771
1772 /*
1773  * Convert node form directory to leaf form directory.
1774  * The root of the node form dir needs to already be a LEAFN block.
1775  * Just return if we can't do anything.
1776  */
1777 int                                             /* error */
1778 xfs_dir2_node_to_leaf(
1779         xfs_da_state_t          *state)         /* directory operation state */
1780 {
1781         xfs_da_args_t           *args;          /* operation arguments */
1782         xfs_inode_t             *dp;            /* incore directory inode */
1783         int                     error;          /* error return code */
1784         xfs_dabuf_t             *fbp;           /* buffer for freespace block */
1785         xfs_fileoff_t           fo;             /* freespace file offset */
1786         xfs_dir2_free_t         *free;          /* freespace structure */
1787         xfs_dabuf_t             *lbp;           /* buffer for leaf block */
1788         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1789         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1790         xfs_mount_t             *mp;            /* filesystem mount point */
1791         int                     rval;           /* successful free trim? */
1792         xfs_trans_t             *tp;            /* transaction pointer */
1793
1794         /*
1795          * There's more than a leaf level in the btree, so there must
1796          * be multiple leafn blocks.  Give up.
1797          */
1798         if (state->path.active > 1)
1799                 return 0;
1800         args = state->args;
1801
1802         trace_xfs_dir2_node_to_leaf(args);
1803
1804         mp = state->mp;
1805         dp = args->dp;
1806         tp = args->trans;
1807         /*
1808          * Get the last offset in the file.
1809          */
1810         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1811                 return error;
1812         }
1813         fo -= mp->m_dirblkfsbs;
1814         /*
1815          * If there are freespace blocks other than the first one,
1816          * take this opportunity to remove trailing empty freespace blocks
1817          * that may have been left behind during no-space-reservation
1818          * operations.
1819          */
1820         while (fo > mp->m_dirfreeblk) {
1821                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1822                         return error;
1823                 }
1824                 if (rval)
1825                         fo -= mp->m_dirblkfsbs;
1826                 else
1827                         return 0;
1828         }
1829         /*
1830          * Now find the block just before the freespace block.
1831          */
1832         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1833                 return error;
1834         }
1835         /*
1836          * If it's not the single leaf block, give up.
1837          */
1838         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1839                 return 0;
1840         lbp = state->path.blk[0].bp;
1841         leaf = lbp->data;
1842         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1843         /*
1844          * Read the freespace block.
1845          */
1846         if ((error = xfs_da_read_buf(tp, dp, mp->m_dirfreeblk, -1, &fbp,
1847                         XFS_DATA_FORK))) {
1848                 return error;
1849         }
1850         free = fbp->data;
1851         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1852         ASSERT(!free->hdr.firstdb);
1853
1854         /*
1855          * Now see if the leafn and free data will fit in a leaf1.
1856          * If not, release the buffer and give up.
1857          */
1858         if (xfs_dir2_leaf_size(&leaf->hdr, be32_to_cpu(free->hdr.nvalid)) >
1859                         mp->m_dirblksize) {
1860                 xfs_da_brelse(tp, fbp);
1861                 return 0;
1862         }
1863
1864         /*
1865          * If the leaf has any stale entries in it, compress them out.
1866          * The compact routine will log the header.
1867          */
1868         if (be16_to_cpu(leaf->hdr.stale))
1869                 xfs_dir2_leaf_compact(args, lbp);
1870         else
1871                 xfs_dir2_leaf_log_header(tp, lbp);
1872         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAF1_MAGIC);
1873         /*
1874          * Set up the leaf tail from the freespace block.
1875          */
1876         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1877         ltp->bestcount = free->hdr.nvalid;
1878         /*
1879          * Set up the leaf bests table.
1880          */
1881         memcpy(xfs_dir2_leaf_bests_p(ltp), free->bests,
1882                 be32_to_cpu(ltp->bestcount) * sizeof(xfs_dir2_data_off_t));
1883         xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1884         xfs_dir2_leaf_log_tail(tp, lbp);
1885         xfs_dir2_leaf_check(dp, lbp);
1886         /*
1887          * Get rid of the freespace block.
1888          */
1889         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1890         if (error) {
1891                 /*
1892                  * This can't fail here because it can only happen when
1893                  * punching out the middle of an extent, and this is an
1894                  * isolated block.
1895                  */
1896                 ASSERT(error != ENOSPC);
1897                 return error;
1898         }
1899         fbp = NULL;
1900         /*
1901          * Now see if we can convert the single-leaf directory
1902          * down to a block form directory.
1903          * This routine always kills the dabuf for the leaf, so
1904          * eliminate it from the path.
1905          */
1906         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1907         state->path.blk[0].bp = NULL;
1908         return error;
1909 }