fs/xfs/xfs_dir2_node.c

   1 /*
   2  * Copyright (c) 2000-2005 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_log.h"
  22 #include "xfs_inum.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_dir2.h"
  27 #include "xfs_dmapi.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_attr_sf.h"
  33 #include "xfs_dinode.h"
  34 #include "xfs_inode.h"
  35 #include "xfs_bmap.h"
  36 #include "xfs_dir2_data.h"
  37 #include "xfs_dir2_leaf.h"
  38 #include "xfs_dir2_block.h"
  39 #include "xfs_dir2_node.h"
  40 #include "xfs_dir2_trace.h"
  41 #include "xfs_error.h"
  42
  43 /*
  44  * Function declarations.
  45  */
  46 static void xfs_dir2_free_log_header(xfs_trans_t *tp, xfs_dabuf_t *bp);
  47 static int xfs_dir2_leafn_add(xfs_dabuf_t *bp, xfs_da_args_t *args, int index);
  48 #ifdef DEBUG
  49 static void xfs_dir2_leafn_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
  50 #else
  51 #define xfs_dir2_leafn_check(dp, bp)
  52 #endif
  53 static void xfs_dir2_leafn_moveents(xfs_da_args_t *args, xfs_dabuf_t *bp_s,
  54                                     int start_s, xfs_dabuf_t *bp_d, int start_d,
  55                                     int count);
  56 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  57                                      xfs_da_state_blk_t *blk1,
  58                                      xfs_da_state_blk_t *blk2);
  59 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, xfs_dabuf_t *bp,
  60                                  int index, xfs_da_state_blk_t *dblk,
  61                                  int *rval);
  62 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  63                                      xfs_da_state_blk_t *fblk);
  64
  65 /*
  66  * Log entries from a freespace block.
  67  */
  68 void
  69 xfs_dir2_free_log_bests(
  70         xfs_trans_t             *tp,            /* transaction pointer */
  71         xfs_dabuf_t             *bp,            /* freespace buffer */
  72         int                     first,          /* first entry to log */
  73         int                     last)           /* last entry to log */
  74 {
  75         xfs_dir2_free_t         *free;          /* freespace structure */
  76
  77         free = bp->data;
  78         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
  79         xfs_da_log_buf(tp, bp,
  80                 (uint)((char *)&free->bests[first] - (char *)free),
  81                 (uint)((char *)&free->bests[last] - (char *)free +
  82                        sizeof(free->bests[0]) - 1));
  83 }
  84
  85 /*
  86  * Log header from a freespace block.
  87  */
  88 static void
  89 xfs_dir2_free_log_header(
  90         xfs_trans_t             *tp,            /* transaction pointer */
  91         xfs_dabuf_t             *bp)            /* freespace buffer */
  92 {
  93         xfs_dir2_free_t         *free;          /* freespace structure */
  94
  95         free = bp->data;
  96         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
  97         xfs_da_log_buf(tp, bp, (uint)((char *)&free->hdr - (char *)free),
  98                 (uint)(sizeof(xfs_dir2_free_hdr_t) - 1));
  99 }
 100
 101 /*
 102  * Convert a leaf-format directory to a node-format directory.
 103  * We need to change the magic number of the leaf block, and copy
 104  * the freespace table out of the leaf block into its own block.
 105  */
 106 int                                             /* error */
 107 xfs_dir2_leaf_to_node(
 108         xfs_da_args_t           *args,          /* operation arguments */
 109         xfs_dabuf_t             *lbp)           /* leaf buffer */
 110 {
 111         xfs_inode_t             *dp;            /* incore directory inode */
 112         int                     error;          /* error return value */
 113         xfs_dabuf_t             *fbp;           /* freespace buffer */
 114         xfs_dir2_db_t           fdb;            /* freespace block number */
 115         xfs_dir2_free_t         *free;          /* freespace structure */
 116         __be16                  *from;          /* pointer to freespace entry */
 117         int                     i;              /* leaf freespace index */
 118         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 119         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 120         xfs_mount_t             *mp;            /* filesystem mount point */
 121         int                     n;              /* count of live freespc ents */
 122         xfs_dir2_data_off_t     off;            /* freespace entry value */
 123         __be16                  *to;            /* pointer to freespace entry */
 124         xfs_trans_t             *tp;            /* transaction pointer */
 125
 126         xfs_dir2_trace_args_b("leaf_to_node", args, lbp);
 127         dp = args->dp;
 128         mp = dp->i_mount;
 129         tp = args->trans;
 130         /*
 131          * Add a freespace block to the directory.
 132          */
 133         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 134                 return error;
 135         }
 136         ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
 137         /*
 138          * Get the buffer for the new freespace block.
 139          */
 140         if ((error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb), -1, &fbp,
 141                         XFS_DATA_FORK))) {
 142                 return error;
 143         }
 144         ASSERT(fbp != NULL);
 145         free = fbp->data;
 146         leaf = lbp->data;
 147         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 148         /*
 149          * Initialize the freespace block header.
 150          */
 151         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
 152         free->hdr.firstdb = 0;
 153         ASSERT(be32_to_cpu(ltp->bestcount) <= (uint)dp->i_d.di_size / mp->m_dirblksize);
 154         free->hdr.nvalid = ltp->bestcount;
 155         /*
 156          * Copy freespace entries from the leaf block to the new block.
 157          * Count active entries.
 158          */
 159         for (i = n = 0, from = xfs_dir2_leaf_bests_p(ltp), to = free->bests;
 160              i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 161                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 162                         n++;
 163                 *to = cpu_to_be16(off);
 164         }
 165         free->hdr.nused = cpu_to_be32(n);
 166         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 167         /*
 168          * Log everything.
 169          */
 170         xfs_dir2_leaf_log_header(tp, lbp);
 171         xfs_dir2_free_log_header(tp, fbp);
 172         xfs_dir2_free_log_bests(tp, fbp, 0, be32_to_cpu(free->hdr.nvalid) - 1);
 173         xfs_da_buf_done(fbp);
 174         xfs_dir2_leafn_check(dp, lbp);
 175         return 0;
 176 }
 177
 178 /*
 179  * Add a leaf entry to a leaf block in a node-form directory.
 180  * The other work necessary is done from the caller.
 181  */
 182 static int                                      /* error */
 183 xfs_dir2_leafn_add(
 184         xfs_dabuf_t             *bp,            /* leaf buffer */
 185         xfs_da_args_t           *args,          /* operation arguments */
 186         int                     index)          /* insertion pt for new entry */
 187 {
 188         int                     compact;        /* compacting stale leaves */
 189         xfs_inode_t             *dp;            /* incore directory inode */
 190         int                     highstale;      /* next stale entry */
 191         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 192         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 193         int                     lfloghigh;      /* high leaf entry logging */
 194         int                     lfloglow;       /* low leaf entry logging */
 195         int                     lowstale;       /* previous stale entry */
 196         xfs_mount_t             *mp;            /* filesystem mount point */
 197         xfs_trans_t             *tp;            /* transaction pointer */
 198
 199         xfs_dir2_trace_args_sb("leafn_add", args, index, bp);
 200         dp = args->dp;
 201         mp = dp->i_mount;
 202         tp = args->trans;
 203         leaf = bp->data;
 204
 205         /*
 206          * Quick check just to make sure we are not going to index
 207          * into other peoples memory
 208          */
 209         if (index < 0)
 210                 return XFS_ERROR(EFSCORRUPTED);
 211
 212         /*
 213          * If there are already the maximum number of leaf entries in
 214          * the block, if there are no stale entries it won't fit.
 215          * Caller will do a split.  If there are stale entries we'll do
 216          * a compact.
 217          */
 218
 219         if (be16_to_cpu(leaf->hdr.count) == xfs_dir2_max_leaf_ents(mp)) {
 220                 if (!leaf->hdr.stale)
 221                         return XFS_ERROR(ENOSPC);
 222                 compact = be16_to_cpu(leaf->hdr.stale) > 1;
 223         } else
 224                 compact = 0;
 225         ASSERT(index == 0 || be32_to_cpu(leaf->ents[index - 1].hashval) <= args->hashval);
 226         ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
 227                be32_to_cpu(leaf->ents[index].hashval) >= args->hashval);
 228
 229         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 230                 return 0;
 231
 232         /*
 233          * Compact out all but one stale leaf entry.  Leaves behind
 234          * the entry closest to index.
 235          */
 236         if (compact) {
 237                 xfs_dir2_leaf_compact_x1(bp, &index, &lowstale, &highstale,
 238                         &lfloglow, &lfloghigh);
 239         }
 240         /*
 241          * Set impossible logging indices for this case.
 242          */
 243         else if (leaf->hdr.stale) {
 244                 lfloglow = be16_to_cpu(leaf->hdr.count);
 245                 lfloghigh = -1;
 246         }
 247         /*
 248          * No stale entries, just insert a space for the new entry.
 249          */
 250         if (!leaf->hdr.stale) {
 251                 lep = &leaf->ents[index];
 252                 if (index < be16_to_cpu(leaf->hdr.count))
 253                         memmove(lep + 1, lep,
 254                                 (be16_to_cpu(leaf->hdr.count) - index) * sizeof(*lep));
 255                 lfloglow = index;
 256                 lfloghigh = be16_to_cpu(leaf->hdr.count);
 257                 be16_add_cpu(&leaf->hdr.count, 1);
 258         }
 259         /*
 260          * There are stale entries.  We'll use one for the new entry.
 261          */
 262         else {
 263                 /*
 264                  * If we didn't do a compact then we need to figure out
 265                  * which stale entry will be used.
 266                  */
 267                 if (compact == 0) {
 268                         /*
 269                          * Find first stale entry before our insertion point.
 270                          */
 271                         for (lowstale = index - 1;
 272                              lowstale >= 0 &&
 273                                 be32_to_cpu(leaf->ents[lowstale].address) !=
 274                                 XFS_DIR2_NULL_DATAPTR;
 275                              lowstale--)
 276                                 continue;
 277                         /*
 278                          * Find next stale entry after insertion point.
 279                          * Stop looking if the answer would be worse than
 280                          * lowstale already found.
 281                          */
 282                         for (highstale = index;
 283                              highstale < be16_to_cpu(leaf->hdr.count) &&
 284                                 be32_to_cpu(leaf->ents[highstale].address) !=
 285                                 XFS_DIR2_NULL_DATAPTR &&
 286                                 (lowstale < 0 ||
 287                                  index - lowstale - 1 >= highstale - index);
 288                              highstale++)
 289                                 continue;
 290                 }
 291                 /*
 292                  * Using the low stale entry.
 293                  * Shift entries up toward the stale slot.
 294                  */
 295                 if (lowstale >= 0 &&
 296                     (highstale == be16_to_cpu(leaf->hdr.count) ||
 297                      index - lowstale - 1 < highstale - index)) {
 298                         ASSERT(be32_to_cpu(leaf->ents[lowstale].address) ==
 299                                XFS_DIR2_NULL_DATAPTR);
 300                         ASSERT(index - lowstale - 1 >= 0);
 301                         if (index - lowstale - 1 > 0)
 302                                 memmove(&leaf->ents[lowstale],
 303                                         &leaf->ents[lowstale + 1],
 304                                         (index - lowstale - 1) * sizeof(*lep));
 305                         lep = &leaf->ents[index - 1];
 306                         lfloglow = MIN(lowstale, lfloglow);
 307                         lfloghigh = MAX(index - 1, lfloghigh);
 308                 }
 309                 /*
 310                  * Using the high stale entry.
 311                  * Shift entries down toward the stale slot.
 312                  */
 313                 else {
 314                         ASSERT(be32_to_cpu(leaf->ents[highstale].address) ==
 315                                XFS_DIR2_NULL_DATAPTR);
 316                         ASSERT(highstale - index >= 0);
 317                         if (highstale - index > 0)
 318                                 memmove(&leaf->ents[index + 1],
 319                                         &leaf->ents[index],
 320                                         (highstale - index) * sizeof(*lep));
 321                         lep = &leaf->ents[index];
 322                         lfloglow = MIN(index, lfloglow);
 323                         lfloghigh = MAX(highstale, lfloghigh);
 324                 }
 325                 be16_add_cpu(&leaf->hdr.stale, -1);
 326         }
 327         /*
 328          * Insert the new entry, log everything.
 329          */
 330         lep->hashval = cpu_to_be32(args->hashval);
 331         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
 332                                 args->blkno, args->index));
 333         xfs_dir2_leaf_log_header(tp, bp);
 334         xfs_dir2_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
 335         xfs_dir2_leafn_check(dp, bp);
 336         return 0;
 337 }
 338
 339 #ifdef DEBUG
 340 /*
 341  * Check internal consistency of a leafn block.
 342  */
 343 void
 344 xfs_dir2_leafn_check(
 345         xfs_inode_t     *dp,                    /* incore directory inode */
 346         xfs_dabuf_t     *bp)                    /* leaf buffer */
 347 {
 348         int             i;                      /* leaf index */
 349         xfs_dir2_leaf_t *leaf;                  /* leaf structure */
 350         xfs_mount_t     *mp;                    /* filesystem mount point */
 351         int             stale;                  /* count of stale leaves */
 352
 353         leaf = bp->data;
 354         mp = dp->i_mount;
 355         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 356         ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
 357         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
 358                 if (i + 1 < be16_to_cpu(leaf->hdr.count)) {
 359                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
 360                                be32_to_cpu(leaf->ents[i + 1].hashval));
 361                 }
 362                 if (be32_to_cpu(leaf->ents[i].address) == XFS_DIR2_NULL_DATAPTR)
 363                         stale++;
 364         }
 365         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
 366 }
 367 #endif  /* DEBUG */
 368
 369 /*
 370  * Return the last hash value in the leaf.
 371  * Stale entries are ok.
 372  */
 373 xfs_dahash_t                                    /* hash value */
 374 xfs_dir2_leafn_lasthash(
 375         xfs_dabuf_t     *bp,                    /* leaf buffer */
 376         int             *count)                 /* count of entries in leaf */
 377 {
 378         xfs_dir2_leaf_t *leaf;                  /* leaf structure */
 379
 380         leaf = bp->data;
 381         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 382         if (count)
 383                 *count = be16_to_cpu(leaf->hdr.count);
 384         if (!leaf->hdr.count)
 385                 return 0;
 386         return be32_to_cpu(leaf->ents[be16_to_cpu(leaf->hdr.count) - 1].hashval);
 387 }
 388
 389 /*
 390  * Look up a leaf entry for space to add a name in a node-format leaf block.
 391  * The extrablk in state is a freespace block.
 392  */
 393 STATIC int
 394 xfs_dir2_leafn_lookup_for_addname(
 395         xfs_dabuf_t             *bp,            /* leaf buffer */
 396         xfs_da_args_t           *args,          /* operation arguments */
 397         int                     *indexp,        /* out: leaf entry index */
 398         xfs_da_state_t          *state)         /* state to fill in */
 399 {
 400         xfs_dabuf_t             *curbp = NULL;  /* current data/free buffer */
 401         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 402         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 403         xfs_inode_t             *dp;            /* incore directory inode */
 404         int                     error;          /* error return value */
 405         int                     fi;             /* free entry index */
 406         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 407         int                     index;          /* leaf entry index */
 408         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 409         int                     length;         /* length of new data entry */
 410         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 411         xfs_mount_t             *mp;            /* filesystem mount point */
 412         xfs_dir2_db_t           newdb;          /* new data block number */
 413         xfs_dir2_db_t           newfdb;         /* new free block number */
 414         xfs_trans_t             *tp;            /* transaction pointer */
 415
 416         dp = args->dp;
 417         tp = args->trans;
 418         mp = dp->i_mount;
 419         leaf = bp->data;
 420         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 421 #ifdef __KERNEL__
 422         ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
 423 #endif
 424         xfs_dir2_leafn_check(dp, bp);
 425         /*
 426          * Look up the hash value in the leaf entries.
 427          */
 428         index = xfs_dir2_leaf_search_hash(args, bp);
 429         /*
 430          * Do we have a buffer coming in?
 431          */
 432         if (state->extravalid) {
 433                 /* If so, it's a free block buffer, get the block number. */
 434                 curbp = state->extrablk.bp;
 435                 curfdb = state->extrablk.blkno;
 436                 free = curbp->data;
 437                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
 438         }
 439         length = xfs_dir2_data_entsize(args->namelen);
 440         /*
 441          * Loop over leaf entries with the right hash value.
 442          */
 443         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
 444                                 be32_to_cpu(lep->hashval) == args->hashval;
 445                                 lep++, index++) {
 446                 /*
 447                  * Skip stale leaf entries.
 448                  */
 449                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 450                         continue;
 451                 /*
 452                  * Pull the data block number from the entry.
 453                  */
 454                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 455                 /*
 456                  * For addname, we're looking for a place to put the new entry.
 457                  * We want to use a data block with an entry of equal
 458                  * hash value to ours if there is one with room.
 459                  *
 460                  * If this block isn't the data block we already have
 461                  * in hand, take a look at it.
 462                  */
 463                 if (newdb != curdb) {
 464                         curdb = newdb;
 465                         /*
 466                          * Convert the data block to the free block
 467                          * holding its freespace information.
 468                          */
 469                         newfdb = xfs_dir2_db_to_fdb(mp, newdb);
 470                         /*
 471                          * If it's not the one we have in hand, read it in.
 472                          */
 473                         if (newfdb != curfdb) {
 474                                 /*
 475                                  * If we had one before, drop it.
 476                                  */
 477                                 if (curbp)
 478                                         xfs_da_brelse(tp, curbp);
 479                                 /*
 480                                  * Read the free block.
 481                                  */
 482                                 error = xfs_da_read_buf(tp, dp,
 483                                                 xfs_dir2_db_to_da(mp, newfdb),
 484                                                 -1, &curbp, XFS_DATA_FORK);
 485                                 if (error)
 486                                         return error;
 487                                 free = curbp->data;
 488                                 ASSERT(be32_to_cpu(free->hdr.magic) ==
 489                                         XFS_DIR2_FREE_MAGIC);
 490                                 ASSERT((be32_to_cpu(free->hdr.firstdb) %
 491                                         XFS_DIR2_MAX_FREE_BESTS(mp)) == 0);
 492                                 ASSERT(be32_to_cpu(free->hdr.firstdb) <= curdb);
 493                                 ASSERT(curdb < be32_to_cpu(free->hdr.firstdb) +
 494                                         be32_to_cpu(free->hdr.nvalid));
 495                         }
 496                         /*
 497                          * Get the index for our entry.
 498                          */
 499                         fi = xfs_dir2_db_to_fdindex(mp, curdb);
 500                         /*
 501                          * If it has room, return it.
 502                          */
 503                         if (unlikely(be16_to_cpu(free->bests[fi]) == NULLDATAOFF)) {
 504                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 505                                                         XFS_ERRLEVEL_LOW, mp);
 506                                 if (curfdb != newfdb)
 507                                         xfs_da_brelse(tp, curbp);
 508                                 return XFS_ERROR(EFSCORRUPTED);
 509                         }
 510                         curfdb = newfdb;
 511                         if (be16_to_cpu(free->bests[fi]) >= length)
 512                                 goto out;
 513                 }
 514         }
 515         /* Didn't find any space */
 516         fi = -1;
 517 out:
 518         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 519         if (curbp) {
 520                 /* Giving back a free block. */
 521                 state->extravalid = 1;
 522                 state->extrablk.bp = curbp;
 523                 state->extrablk.index = fi;
 524                 state->extrablk.blkno = curfdb;
 525                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 526         } else {
 527                 state->extravalid = 0;
 528         }
 529         /*
 530          * Return the index, that will be the insertion point.
 531          */
 532         *indexp = index;
 533         return XFS_ERROR(ENOENT);
 534 }
 535
 536 /*
 537  * Look up a leaf entry in a node-format leaf block.
 538  * The extrablk in state a data block.
 539  */
 540 STATIC int
 541 xfs_dir2_leafn_lookup_for_entry(
 542         xfs_dabuf_t             *bp,            /* leaf buffer */
 543         xfs_da_args_t           *args,          /* operation arguments */
 544         int                     *indexp,        /* out: leaf entry index */
 545         xfs_da_state_t          *state)         /* state to fill in */
 546 {
 547         xfs_dabuf_t             *curbp = NULL;  /* current data/free buffer */
 548         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 549         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 550         xfs_inode_t             *dp;            /* incore directory inode */
 551         int                     error;          /* error return value */
 552         int                     di;             /* data entry index */
 553         int                     index;          /* leaf entry index */
 554         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 555         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 556         xfs_mount_t             *mp;            /* filesystem mount point */
 557         xfs_dir2_db_t           newdb;          /* new data block number */
 558         xfs_trans_t             *tp;            /* transaction pointer */
 559         enum xfs_dacmp          cmp;            /* comparison result */
 560
 561         dp = args->dp;
 562         tp = args->trans;
 563         mp = dp->i_mount;
 564         leaf = bp->data;
 565         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 566 #ifdef __KERNEL__
 567         ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
 568 #endif
 569         xfs_dir2_leafn_check(dp, bp);
 570         /*
 571          * Look up the hash value in the leaf entries.
 572          */
 573         index = xfs_dir2_leaf_search_hash(args, bp);
 574         /*
 575          * Do we have a buffer coming in?
 576          */
 577         if (state->extravalid) {
 578                 curbp = state->extrablk.bp;
 579                 curdb = state->extrablk.blkno;
 580         }
 581         /*
 582          * Loop over leaf entries with the right hash value.
 583          */
 584         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
 585                                 be32_to_cpu(lep->hashval) == args->hashval;
 586                                 lep++, index++) {
 587                 /*
 588                  * Skip stale leaf entries.
 589                  */
 590                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 591                         continue;
 592                 /*
 593                  * Pull the data block number from the entry.
 594                  */
 595                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 596                 /*
 597                  * Not adding a new entry, so we really want to find
 598                  * the name given to us.
 599                  *
 600                  * If it's a different data block, go get it.
 601                  */
 602                 if (newdb != curdb) {
 603                         /*
 604                          * If we had a block before, drop it.
 605                          */
 606                         if (curbp)
 607                                 xfs_da_brelse(tp, curbp);
 608                         /*
 609                          * Read the data block.
 610                          */
 611                         error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp,
 612                                         newdb), -1, &curbp, XFS_DATA_FORK);
 613                         if (error)
 614                                 return error;
 615                         xfs_dir2_data_check(dp, curbp);
 616                         curdb = newdb;
 617                 }
 618                 /*
 619                  * Point to the data entry.
 620                  */
 621                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->data +
 622                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
 623                 /*
 624                  * Compare the entry and if it's an exact match, return
 625                  * EEXIST immediately. If it's the first case-insensitive
 626                  * match, store the inode number and continue looking.
 627                  */
 628                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 629                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 630                         args->cmpresult = cmp;
 631                         args->inumber = be64_to_cpu(dep->inumber);
 632                         di = (int)((char *)dep - (char *)curbp->data);
 633                         error = EEXIST;
 634                         if (cmp == XFS_CMP_EXACT)
 635                                 goto out;
 636                 }
 637         }
 638         /* Didn't find an exact match. */
 639         error = ENOENT;
 640         di = -1;
 641         ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
 642                                         (args->op_flags & XFS_DA_OP_OKNOENT));
 643 out:
 644         if (curbp) {
 645                 /* Giving back a data block. */
 646                 state->extravalid = 1;
 647                 state->extrablk.bp = curbp;
 648                 state->extrablk.index = di;
 649                 state->extrablk.blkno = curdb;
 650                 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 651         } else {
 652                 state->extravalid = 0;
 653         }
 654         /*
 655          * Return the index, that will be the insertion point.
 656          */
 657         *indexp = index;
 658         return XFS_ERROR(error);
 659 }
 660
 661 /*
 662  * Look up a leaf entry in a node-format leaf block.
 663  * If this is an addname then the extrablk in state is a freespace block,
 664  * otherwise it's a data block.
 665  */
 666 int
 667 xfs_dir2_leafn_lookup_int(
 668         xfs_dabuf_t             *bp,            /* leaf buffer */
 669         xfs_da_args_t           *args,          /* operation arguments */
 670         int                     *indexp,        /* out: leaf entry index */
 671         xfs_da_state_t          *state)         /* state to fill in */
 672 {
 673         if (args->op_flags & XFS_DA_OP_ADDNAME)
 674                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 675                                                         state);
 676         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 677 }
 678
 679 /*
 680  * Move count leaf entries from source to destination leaf.
 681  * Log entries and headers.  Stale entries are preserved.
 682  */
 683 static void
 684 xfs_dir2_leafn_moveents(
 685         xfs_da_args_t   *args,                  /* operation arguments */
 686         xfs_dabuf_t     *bp_s,                  /* source leaf buffer */
 687         int             start_s,                /* source leaf index */
 688         xfs_dabuf_t     *bp_d,                  /* destination leaf buffer */
 689         int             start_d,                /* destination leaf index */
 690         int             count)                  /* count of leaves to copy */
 691 {
 692         xfs_dir2_leaf_t *leaf_d;                /* destination leaf structure */
 693         xfs_dir2_leaf_t *leaf_s;                /* source leaf structure */
 694         int             stale;                  /* count stale leaves copied */
 695         xfs_trans_t     *tp;                    /* transaction pointer */
 696
 697         xfs_dir2_trace_args_bibii("leafn_moveents", args, bp_s, start_s, bp_d,
 698                 start_d, count);
 699         /*
 700          * Silently return if nothing to do.
 701          */
 702         if (count == 0) {
 703                 return;
 704         }
 705         tp = args->trans;
 706         leaf_s = bp_s->data;
 707         leaf_d = bp_d->data;
 708         /*
 709          * If the destination index is not the end of the current
 710          * destination leaf entries, open up a hole in the destination
 711          * to hold the new entries.
 712          */
 713         if (start_d < be16_to_cpu(leaf_d->hdr.count)) {
 714                 memmove(&leaf_d->ents[start_d + count], &leaf_d->ents[start_d],
 715                         (be16_to_cpu(leaf_d->hdr.count) - start_d) *
 716                         sizeof(xfs_dir2_leaf_entry_t));
 717                 xfs_dir2_leaf_log_ents(tp, bp_d, start_d + count,
 718                         count + be16_to_cpu(leaf_d->hdr.count) - 1);
 719         }
 720         /*
 721          * If the source has stale leaves, count the ones in the copy range
 722          * so we can update the header correctly.
 723          */
 724         if (leaf_s->hdr.stale) {
 725                 int     i;                      /* temp leaf index */
 726
 727                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 728                         if (be32_to_cpu(leaf_s->ents[i].address) == XFS_DIR2_NULL_DATAPTR)
 729                                 stale++;
 730                 }
 731         } else
 732                 stale = 0;
 733         /*
 734          * Copy the leaf entries from source to destination.
 735          */
 736         memcpy(&leaf_d->ents[start_d], &leaf_s->ents[start_s],
 737                 count * sizeof(xfs_dir2_leaf_entry_t));
 738         xfs_dir2_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
 739         /*
 740          * If there are source entries after the ones we copied,
 741          * delete the ones we copied by sliding the next ones down.
 742          */
 743         if (start_s + count < be16_to_cpu(leaf_s->hdr.count)) {
 744                 memmove(&leaf_s->ents[start_s], &leaf_s->ents[start_s + count],
 745                         count * sizeof(xfs_dir2_leaf_entry_t));
 746                 xfs_dir2_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
 747         }
 748         /*
 749          * Update the headers and log them.
 750          */
 751         be16_add_cpu(&leaf_s->hdr.count, -(count));
 752         be16_add_cpu(&leaf_s->hdr.stale, -(stale));
 753         be16_add_cpu(&leaf_d->hdr.count, count);
 754         be16_add_cpu(&leaf_d->hdr.stale, stale);
 755         xfs_dir2_leaf_log_header(tp, bp_s);
 756         xfs_dir2_leaf_log_header(tp, bp_d);
 757         xfs_dir2_leafn_check(args->dp, bp_s);
 758         xfs_dir2_leafn_check(args->dp, bp_d);
 759 }
 760
 761 /*
 762  * Determine the sort order of two leaf blocks.
 763  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 764  */
 765 int                                             /* sort order */
 766 xfs_dir2_leafn_order(
 767         xfs_dabuf_t     *leaf1_bp,              /* leaf1 buffer */
 768         xfs_dabuf_t     *leaf2_bp)              /* leaf2 buffer */
 769 {
 770         xfs_dir2_leaf_t *leaf1;                 /* leaf1 structure */
 771         xfs_dir2_leaf_t *leaf2;                 /* leaf2 structure */
 772
 773         leaf1 = leaf1_bp->data;
 774         leaf2 = leaf2_bp->data;
 775         ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 776         ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 777         if (be16_to_cpu(leaf1->hdr.count) > 0 &&
 778             be16_to_cpu(leaf2->hdr.count) > 0 &&
 779             (be32_to_cpu(leaf2->ents[0].hashval) < be32_to_cpu(leaf1->ents[0].hashval) ||
 780              be32_to_cpu(leaf2->ents[be16_to_cpu(leaf2->hdr.count) - 1].hashval) <
 781              be32_to_cpu(leaf1->ents[be16_to_cpu(leaf1->hdr.count) - 1].hashval)))
 782                 return 1;
 783         return 0;
 784 }
 785
 786 /*
 787  * Rebalance leaf entries between two leaf blocks.
 788  * This is actually only called when the second block is new,
 789  * though the code deals with the general case.
 790  * A new entry will be inserted in one of the blocks, and that
 791  * entry is taken into account when balancing.
 792  */
 793 static void
 794 xfs_dir2_leafn_rebalance(
 795         xfs_da_state_t          *state,         /* btree cursor */
 796         xfs_da_state_blk_t      *blk1,          /* first btree block */
 797         xfs_da_state_blk_t      *blk2)          /* second btree block */
 798 {
 799         xfs_da_args_t           *args;          /* operation arguments */
 800         int                     count;          /* count (& direction) leaves */
 801         int                     isleft;         /* new goes in left leaf */
 802         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 803         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 804         int                     mid;            /* midpoint leaf index */
 805 #ifdef DEBUG
 806         int                     oldstale;       /* old count of stale leaves */
 807 #endif
 808         int                     oldsum;         /* old total leaf count */
 809         int                     swap;           /* swapped leaf blocks */
 810
 811         args = state->args;
 812         /*
 813          * If the block order is wrong, swap the arguments.
 814          */
 815         if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
 816                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
 817
 818                 tmp = blk1;
 819                 blk1 = blk2;
 820                 blk2 = tmp;
 821         }
 822         leaf1 = blk1->bp->data;
 823         leaf2 = blk2->bp->data;
 824         oldsum = be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count);
 825 #ifdef DEBUG
 826         oldstale = be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale);
 827 #endif
 828         mid = oldsum >> 1;
 829         /*
 830          * If the old leaf count was odd then the new one will be even,
 831          * so we need to divide the new count evenly.
 832          */
 833         if (oldsum & 1) {
 834                 xfs_dahash_t    midhash;        /* middle entry hash value */
 835
 836                 if (mid >= be16_to_cpu(leaf1->hdr.count))
 837                         midhash = be32_to_cpu(leaf2->ents[mid - be16_to_cpu(leaf1->hdr.count)].hashval);
 838                 else
 839                         midhash = be32_to_cpu(leaf1->ents[mid].hashval);
 840                 isleft = args->hashval <= midhash;
 841         }
 842         /*
 843          * If the old count is even then the new count is odd, so there's
 844          * no preferred side for the new entry.
 845          * Pick the left one.
 846          */
 847         else
 848                 isleft = 1;
 849         /*
 850          * Calculate moved entry count.  Positive means left-to-right,
 851          * negative means right-to-left.  Then move the entries.
 852          */
 853         count = be16_to_cpu(leaf1->hdr.count) - mid + (isleft == 0);
 854         if (count > 0)
 855                 xfs_dir2_leafn_moveents(args, blk1->bp,
 856                         be16_to_cpu(leaf1->hdr.count) - count, blk2->bp, 0, count);
 857         else if (count < 0)
 858                 xfs_dir2_leafn_moveents(args, blk2->bp, 0, blk1->bp,
 859                         be16_to_cpu(leaf1->hdr.count), count);
 860         ASSERT(be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count) == oldsum);
 861         ASSERT(be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale) == oldstale);
 862         /*
 863          * Mark whether we're inserting into the old or new leaf.
 864          */
 865         if (be16_to_cpu(leaf1->hdr.count) < be16_to_cpu(leaf2->hdr.count))
 866                 state->inleaf = swap;
 867         else if (be16_to_cpu(leaf1->hdr.count) > be16_to_cpu(leaf2->hdr.count))
 868                 state->inleaf = !swap;
 869         else
 870                 state->inleaf =
 871                         swap ^ (blk1->index <= be16_to_cpu(leaf1->hdr.count));
 872         /*
 873          * Adjust the expected index for insertion.
 874          */
 875         if (!state->inleaf)
 876                 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
 877
 878         /*
 879          * Finally sanity check just to make sure we are not returning a
 880          * negative index
 881          */
 882         if(blk2->index < 0) {
 883                 state->inleaf = 1;
 884                 blk2->index = 0;
 885                 cmn_err(CE_ALERT,
 886                         "xfs_dir2_leafn_rebalance: picked the wrong leaf? reverting original leaf: "
 887                         "blk1->index %d\n",
 888                         blk1->index);
 889         }
 890 }
 891
 892 /*
 893  * Remove an entry from a node directory.
 894  * This removes the leaf entry and the data entry,
 895  * and updates the free block if necessary.
 896  */
 897 static int                                      /* error */
 898 xfs_dir2_leafn_remove(
 899         xfs_da_args_t           *args,          /* operation arguments */
 900         xfs_dabuf_t             *bp,            /* leaf buffer */
 901         int                     index,          /* leaf entry index */
 902         xfs_da_state_blk_t      *dblk,          /* data block */
 903         int                     *rval)          /* resulting block needs join */
 904 {
 905         xfs_dir2_data_t         *data;          /* data block structure */
 906         xfs_dir2_db_t           db;             /* data block number */
 907         xfs_dabuf_t             *dbp;           /* data block buffer */
 908         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 909         xfs_inode_t             *dp;            /* incore directory inode */
 910         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 911         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 912         int                     longest;        /* longest data free entry */
 913         int                     off;            /* data block entry offset */
 914         xfs_mount_t             *mp;            /* filesystem mount point */
 915         int                     needlog;        /* need to log data header */
 916         int                     needscan;       /* need to rescan data frees */
 917         xfs_trans_t             *tp;            /* transaction pointer */
 918
 919         xfs_dir2_trace_args_sb("leafn_remove", args, index, bp);
 920         dp = args->dp;
 921         tp = args->trans;
 922         mp = dp->i_mount;
 923         leaf = bp->data;
 924         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 925         /*
 926          * Point to the entry we're removing.
 927          */
 928         lep = &leaf->ents[index];
 929         /*
 930          * Extract the data block and offset from the entry.
 931          */
 932         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 933         ASSERT(dblk->blkno == db);
 934         off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
 935         ASSERT(dblk->index == off);
 936         /*
 937          * Kill the leaf entry by marking it stale.
 938          * Log the leaf block changes.
 939          */
 940         be16_add_cpu(&leaf->hdr.stale, 1);
 941         xfs_dir2_leaf_log_header(tp, bp);
 942         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 943         xfs_dir2_leaf_log_ents(tp, bp, index, index);
 944         /*
 945          * Make the data entry free.  Keep track of the longest freespace
 946          * in the data block in case it changes.
 947          */
 948         dbp = dblk->bp;
 949         data = dbp->data;
 950         dep = (xfs_dir2_data_entry_t *)((char *)data + off);
 951         longest = be16_to_cpu(data->hdr.bestfree[0].length);
 952         needlog = needscan = 0;
 953         xfs_dir2_data_make_free(tp, dbp, off,
 954                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
 955         /*
 956          * Rescan the data block freespaces for bestfree.
 957          * Log the data block header if needed.
 958          */
 959         if (needscan)
 960                 xfs_dir2_data_freescan(mp, data, &needlog);
 961         if (needlog)
 962                 xfs_dir2_data_log_header(tp, dbp);
 963         xfs_dir2_data_check(dp, dbp);
 964         /*
 965          * If the longest data block freespace changes, need to update
 966          * the corresponding freeblock entry.
 967          */
 968         if (longest < be16_to_cpu(data->hdr.bestfree[0].length)) {
 969                 int             error;          /* error return value */
 970                 xfs_dabuf_t     *fbp;           /* freeblock buffer */
 971                 xfs_dir2_db_t   fdb;            /* freeblock block number */
 972                 int             findex;         /* index in freeblock entries */
 973                 xfs_dir2_free_t *free;          /* freeblock structure */
 974                 int             logfree;        /* need to log free entry */
 975
 976                 /*
 977                  * Convert the data block number to a free block,
 978                  * read in the free block.
 979                  */
 980                 fdb = xfs_dir2_db_to_fdb(mp, db);
 981                 if ((error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb),
 982                                 -1, &fbp, XFS_DATA_FORK))) {
 983                         return error;
 984                 }
 985                 free = fbp->data;
 986                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
 987                 ASSERT(be32_to_cpu(free->hdr.firstdb) ==
 988                        XFS_DIR2_MAX_FREE_BESTS(mp) *
 989                        (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
 990                 /*
 991                  * Calculate which entry we need to fix.
 992                  */
 993                 findex = xfs_dir2_db_to_fdindex(mp, db);
 994                 longest = be16_to_cpu(data->hdr.bestfree[0].length);
 995                 /*
 996                  * If the data block is now empty we can get rid of it
 997                  * (usually).
 998                  */
 999                 if (longest == mp->m_dirblksize - (uint)sizeof(data->hdr)) {
1000                         /*
1001                          * Try to punch out the data block.
1002                          */
1003                         error = xfs_dir2_shrink_inode(args, db, dbp);
1004                         if (error == 0) {
1005                                 dblk->bp = NULL;
1006                                 data = NULL;
1007                         }
1008                         /*
1009                          * We can get ENOSPC if there's no space reservation.
1010                          * In this case just drop the buffer and some one else
1011                          * will eventually get rid of the empty block.
1012                          */
1013                         else if (error == ENOSPC && args->total == 0)
1014                                 xfs_da_buf_done(dbp);
1015                         else
1016                                 return error;
1017                 }
1018                 /*
1019                  * If we got rid of the data block, we can eliminate that entry
1020                  * in the free block.
1021                  */
1022                 if (data == NULL) {
1023                         /*
1024                          * One less used entry in the free table.
1025                          */
1026                         be32_add_cpu(&free->hdr.nused, -1);
1027                         xfs_dir2_free_log_header(tp, fbp);
1028                         /*
1029                          * If this was the last entry in the table, we can
1030                          * trim the table size back.  There might be other
1031                          * entries at the end referring to non-existent
1032                          * data blocks, get those too.
1033                          */
1034                         if (findex == be32_to_cpu(free->hdr.nvalid) - 1) {
1035                                 int     i;              /* free entry index */
1036
1037                                 for (i = findex - 1;
1038                                      i >= 0 && be16_to_cpu(free->bests[i]) == NULLDATAOFF;
1039                                      i--)
1040                                         continue;
1041                                 free->hdr.nvalid = cpu_to_be32(i + 1);
1042                                 logfree = 0;
1043                         }
1044                         /*
1045                          * Not the last entry, just punch it out.
1046                          */
1047                         else {
1048                                 free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1049                                 logfree = 1;
1050                         }
1051                         /*
1052                          * If there are no useful entries left in the block,
1053                          * get rid of the block if we can.
1054                          */
1055                         if (!free->hdr.nused) {
1056                                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1057                                 if (error == 0) {
1058                                         fbp = NULL;
1059                                         logfree = 0;
1060                                 } else if (error != ENOSPC || args->total != 0)
1061                                         return error;
1062                                 /*
1063                                  * It's possible to get ENOSPC if there is no
1064                                  * space reservation.  In this case some one
1065                                  * else will eventually get rid of this block.
1066                                  */
1067                         }
1068                 }
1069                 /*
1070                  * Data block is not empty, just set the free entry to
1071                  * the new value.
1072                  */
1073                 else {
1074                         free->bests[findex] = cpu_to_be16(longest);
1075                         logfree = 1;
1076                 }
1077                 /*
1078                  * Log the free entry that changed, unless we got rid of it.
1079                  */
1080                 if (logfree)
1081                         xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1082                 /*
1083                  * Drop the buffer if we still have it.
1084                  */
1085                 if (fbp)
1086                         xfs_da_buf_done(fbp);
1087         }
1088         xfs_dir2_leafn_check(dp, bp);
1089         /*
1090          * Return indication of whether this leaf block is emtpy enough
1091          * to justify trying to join it with a neighbor.
1092          */
1093         *rval =
1094                 ((uint)sizeof(leaf->hdr) +
1095                  (uint)sizeof(leaf->ents[0]) *
1096                  (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale))) <
1097                 mp->m_dir_magicpct;
1098         return 0;
1099 }
1100
1101 /*
1102  * Split the leaf entries in the old block into old and new blocks.
1103  */
1104 int                                             /* error */
1105 xfs_dir2_leafn_split(
1106         xfs_da_state_t          *state,         /* btree cursor */
1107         xfs_da_state_blk_t      *oldblk,        /* original block */
1108         xfs_da_state_blk_t      *newblk)        /* newly created block */
1109 {
1110         xfs_da_args_t           *args;          /* operation arguments */
1111         xfs_dablk_t             blkno;          /* new leaf block number */
1112         int                     error;          /* error return value */
1113         xfs_mount_t             *mp;            /* filesystem mount point */
1114
1115         /*
1116          * Allocate space for a new leaf node.
1117          */
1118         args = state->args;
1119         mp = args->dp->i_mount;
1120         ASSERT(args != NULL);
1121         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1122         error = xfs_da_grow_inode(args, &blkno);
1123         if (error) {
1124                 return error;
1125         }
1126         /*
1127          * Initialize the new leaf block.
1128          */
1129         error = xfs_dir2_leaf_init(args, xfs_dir2_da_to_db(mp, blkno),
1130                 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1131         if (error) {
1132                 return error;
1133         }
1134         newblk->blkno = blkno;
1135         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1136         /*
1137          * Rebalance the entries across the two leaves, link the new
1138          * block into the leaves.
1139          */
1140         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1141         error = xfs_da_blk_link(state, oldblk, newblk);
1142         if (error) {
1143                 return error;
1144         }
1145         /*
1146          * Insert the new entry in the correct block.
1147          */
1148         if (state->inleaf)
1149                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1150         else
1151                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1152         /*
1153          * Update last hashval in each block since we added the name.
1154          */
1155         oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1156         newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1157         xfs_dir2_leafn_check(args->dp, oldblk->bp);
1158         xfs_dir2_leafn_check(args->dp, newblk->bp);
1159         return error;
1160 }
1161
1162 /*
1163  * Check a leaf block and its neighbors to see if the block should be
1164  * collapsed into one or the other neighbor.  Always keep the block
1165  * with the smaller block number.
1166  * If the current block is over 50% full, don't try to join it, return 0.
1167  * If the block is empty, fill in the state structure and return 2.
1168  * If it can be collapsed, fill in the state structure and return 1.
1169  * If nothing can be done, return 0.
1170  */
1171 int                                             /* error */
1172 xfs_dir2_leafn_toosmall(
1173         xfs_da_state_t          *state,         /* btree cursor */
1174         int                     *action)        /* resulting action to take */
1175 {
1176         xfs_da_state_blk_t      *blk;           /* leaf block */
1177         xfs_dablk_t             blkno;          /* leaf block number */
1178         xfs_dabuf_t             *bp;            /* leaf buffer */
1179         int                     bytes;          /* bytes in use */
1180         int                     count;          /* leaf live entry count */
1181         int                     error;          /* error return value */
1182         int                     forward;        /* sibling block direction */
1183         int                     i;              /* sibling counter */
1184         xfs_da_blkinfo_t        *info;          /* leaf block header */
1185         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1186         int                     rval;           /* result from path_shift */
1187
1188         /*
1189          * Check for the degenerate case of the block being over 50% full.
1190          * If so, it's not worth even looking to see if we might be able
1191          * to coalesce with a sibling.
1192          */
1193         blk = &state->path.blk[state->path.active - 1];
1194         info = blk->bp->data;
1195         ASSERT(be16_to_cpu(info->magic) == XFS_DIR2_LEAFN_MAGIC);
1196         leaf = (xfs_dir2_leaf_t *)info;
1197         count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1198         bytes = (uint)sizeof(leaf->hdr) + count * (uint)sizeof(leaf->ents[0]);
1199         if (bytes > (state->blocksize >> 1)) {
1200                 /*
1201                  * Blk over 50%, don't try to join.
1202                  */
1203                 *action = 0;
1204                 return 0;
1205         }
1206         /*
1207          * Check for the degenerate case of the block being empty.
1208          * If the block is empty, we'll simply delete it, no need to
1209          * coalesce it with a sibling block.  We choose (arbitrarily)
1210          * to merge with the forward block unless it is NULL.
1211          */
1212         if (count == 0) {
1213                 /*
1214                  * Make altpath point to the block we want to keep and
1215                  * path point to the block we want to drop (this one).
1216                  */
1217                 forward = (info->forw != 0);
1218                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1219                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1220                         &rval);
1221                 if (error)
1222                         return error;
1223                 *action = rval ? 2 : 0;
1224                 return 0;
1225         }
1226         /*
1227          * Examine each sibling block to see if we can coalesce with
1228          * at least 25% free space to spare.  We need to figure out
1229          * whether to merge with the forward or the backward block.
1230          * We prefer coalescing with the lower numbered sibling so as
1231          * to shrink a directory over time.
1232          */
1233         forward = be32_to_cpu(info->forw) < be32_to_cpu(info->back);
1234         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1235                 blkno = forward ? be32_to_cpu(info->forw) : be32_to_cpu(info->back);
1236                 if (blkno == 0)
1237                         continue;
1238                 /*
1239                  * Read the sibling leaf block.
1240                  */
1241                 if ((error =
1242                     xfs_da_read_buf(state->args->trans, state->args->dp, blkno,
1243                             -1, &bp, XFS_DATA_FORK))) {
1244                         return error;
1245                 }
1246                 ASSERT(bp != NULL);
1247                 /*
1248                  * Count bytes in the two blocks combined.
1249                  */
1250                 leaf = (xfs_dir2_leaf_t *)info;
1251                 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1252                 bytes = state->blocksize - (state->blocksize >> 2);
1253                 leaf = bp->data;
1254                 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1255                 count += be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1256                 bytes -= count * (uint)sizeof(leaf->ents[0]);
1257                 /*
1258                  * Fits with at least 25% to spare.
1259                  */
1260                 if (bytes >= 0)
1261                         break;
1262                 xfs_da_brelse(state->args->trans, bp);
1263         }
1264         /*
1265          * Didn't like either block, give up.
1266          */
1267         if (i >= 2) {
1268                 *action = 0;
1269                 return 0;
1270         }
1271         /*
1272          * Done with the sibling leaf block here, drop the dabuf
1273          * so path_shift can get it.
1274          */
1275         xfs_da_buf_done(bp);
1276         /*
1277          * Make altpath point to the block we want to keep (the lower
1278          * numbered block) and path point to the block we want to drop.
1279          */
1280         memcpy(&state->altpath, &state->path, sizeof(state->path));
1281         if (blkno < blk->blkno)
1282                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1283                         &rval);
1284         else
1285                 error = xfs_da_path_shift(state, &state->path, forward, 0,
1286                         &rval);
1287         if (error) {
1288                 return error;
1289         }
1290         *action = rval ? 0 : 1;
1291         return 0;
1292 }
1293
1294 /*
1295  * Move all the leaf entries from drop_blk to save_blk.
1296  * This is done as part of a join operation.
1297  */
1298 void
1299 xfs_dir2_leafn_unbalance(
1300         xfs_da_state_t          *state,         /* cursor */
1301         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1302         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1303 {
1304         xfs_da_args_t           *args;          /* operation arguments */
1305         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1306         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1307
1308         args = state->args;
1309         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1310         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1311         drop_leaf = drop_blk->bp->data;
1312         save_leaf = save_blk->bp->data;
1313         ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1314         ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1315         /*
1316          * If there are any stale leaf entries, take this opportunity
1317          * to purge them.
1318          */
1319         if (drop_leaf->hdr.stale)
1320                 xfs_dir2_leaf_compact(args, drop_blk->bp);
1321         if (save_leaf->hdr.stale)
1322                 xfs_dir2_leaf_compact(args, save_blk->bp);
1323         /*
1324          * Move the entries from drop to the appropriate end of save.
1325          */
1326         drop_blk->hashval = be32_to_cpu(drop_leaf->ents[be16_to_cpu(drop_leaf->hdr.count) - 1].hashval);
1327         if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1328                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp, 0,
1329                         be16_to_cpu(drop_leaf->hdr.count));
1330         else
1331                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp,
1332                         be16_to_cpu(save_leaf->hdr.count), be16_to_cpu(drop_leaf->hdr.count));
1333         save_blk->hashval = be32_to_cpu(save_leaf->ents[be16_to_cpu(save_leaf->hdr.count) - 1].hashval);
1334         xfs_dir2_leafn_check(args->dp, save_blk->bp);
1335 }
1336
1337 /*
1338  * Top-level node form directory addname routine.
1339  */
1340 int                                             /* error */
1341 xfs_dir2_node_addname(
1342         xfs_da_args_t           *args)          /* operation arguments */
1343 {
1344         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1345         int                     error;          /* error return value */
1346         int                     rval;           /* sub-return value */
1347         xfs_da_state_t          *state;         /* btree cursor */
1348
1349         xfs_dir2_trace_args("node_addname", args);
1350         /*
1351          * Allocate and initialize the state (btree cursor).
1352          */
1353         state = xfs_da_state_alloc();
1354         state->args = args;
1355         state->mp = args->dp->i_mount;
1356         state->blocksize = state->mp->m_dirblksize;
1357         state->node_ents = state->mp->m_dir_node_ents;
1358         /*
1359          * Look up the name.  We're not supposed to find it, but
1360          * this gives us the insertion point.
1361          */
1362         error = xfs_da_node_lookup_int(state, &rval);
1363         if (error)
1364                 rval = error;
1365         if (rval != ENOENT) {
1366                 goto done;
1367         }
1368         /*
1369          * Add the data entry to a data block.
1370          * Extravalid is set to a freeblock found by lookup.
1371          */
1372         rval = xfs_dir2_node_addname_int(args,
1373                 state->extravalid ? &state->extrablk : NULL);
1374         if (rval) {
1375                 goto done;
1376         }
1377         blk = &state->path.blk[state->path.active - 1];
1378         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1379         /*
1380          * Add the new leaf entry.
1381          */
1382         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1383         if (rval == 0) {
1384                 /*
1385                  * It worked, fix the hash values up the btree.
1386                  */
1387                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1388                         xfs_da_fixhashpath(state, &state->path);
1389         } else {
1390                 /*
1391                  * It didn't work, we need to split the leaf block.
1392                  */
1393                 if (args->total == 0) {
1394                         ASSERT(rval == ENOSPC);
1395                         goto done;
1396                 }
1397                 /*
1398                  * Split the leaf block and insert the new entry.
1399                  */
1400                 rval = xfs_da_split(state);
1401         }
1402 done:
1403         xfs_da_state_free(state);
1404         return rval;
1405 }
1406
1407 /*
1408  * Add the data entry for a node-format directory name addition.
1409  * The leaf entry is added in xfs_dir2_leafn_add.
1410  * We may enter with a freespace block that the lookup found.
1411  */
1412 static int                                      /* error */
1413 xfs_dir2_node_addname_int(
1414         xfs_da_args_t           *args,          /* operation arguments */
1415         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1416 {
1417         xfs_dir2_data_t         *data;          /* data block structure */
1418         xfs_dir2_db_t           dbno;           /* data block number */
1419         xfs_dabuf_t             *dbp;           /* data block buffer */
1420         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1421         xfs_inode_t             *dp;            /* incore directory inode */
1422         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1423         int                     error;          /* error return value */
1424         xfs_dir2_db_t           fbno;           /* freespace block number */
1425         xfs_dabuf_t             *fbp;           /* freespace buffer */
1426         int                     findex;         /* freespace entry index */
1427         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1428         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1429         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1430         int                     length;         /* length of the new entry */
1431         int                     logfree;        /* need to log free entry */
1432         xfs_mount_t             *mp;            /* filesystem mount point */
1433         int                     needlog;        /* need to log data header */
1434         int                     needscan;       /* need to rescan data frees */
1435         __be16                  *tagp;          /* data entry tag pointer */
1436         xfs_trans_t             *tp;            /* transaction pointer */
1437
1438         dp = args->dp;
1439         mp = dp->i_mount;
1440         tp = args->trans;
1441         length = xfs_dir2_data_entsize(args->namelen);
1442         /*
1443          * If we came in with a freespace block that means that lookup
1444          * found an entry with our hash value.  This is the freespace
1445          * block for that data entry.
1446          */
1447         if (fblk) {
1448                 fbp = fblk->bp;
1449                 /*
1450                  * Remember initial freespace block number.
1451                  */
1452                 ifbno = fblk->blkno;
1453                 free = fbp->data;
1454                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1455                 findex = fblk->index;
1456                 /*
1457                  * This means the free entry showed that the data block had
1458                  * space for our entry, so we remembered it.
1459                  * Use that data block.
1460                  */
1461                 if (findex >= 0) {
1462                         ASSERT(findex < be32_to_cpu(free->hdr.nvalid));
1463                         ASSERT(be16_to_cpu(free->bests[findex]) != NULLDATAOFF);
1464                         ASSERT(be16_to_cpu(free->bests[findex]) >= length);
1465                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1466                 }
1467                 /*
1468                  * The data block looked at didn't have enough room.
1469                  * We'll start at the beginning of the freespace entries.
1470                  */
1471                 else {
1472                         dbno = -1;
1473                         findex = 0;
1474                 }
1475         }
1476         /*
1477          * Didn't come in with a freespace block, so don't have a data block.
1478          */
1479         else {
1480                 ifbno = dbno = -1;
1481                 fbp = NULL;
1482                 findex = 0;
1483         }
1484         /*
1485          * If we don't have a data block yet, we're going to scan the
1486          * freespace blocks looking for one.  Figure out what the
1487          * highest freespace block number is.
1488          */
1489         if (dbno == -1) {
1490                 xfs_fileoff_t   fo;             /* freespace block number */
1491
1492                 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1493                         return error;
1494                 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1495                 fbno = ifbno;
1496         }
1497         /*
1498          * While we haven't identified a data block, search the freeblock
1499          * data for a good data block.  If we find a null freeblock entry,
1500          * indicating a hole in the data blocks, remember that.
1501          */
1502         while (dbno == -1) {
1503                 /*
1504                  * If we don't have a freeblock in hand, get the next one.
1505                  */
1506                 if (fbp == NULL) {
1507                         /*
1508                          * Happens the first time through unless lookup gave
1509                          * us a freespace block to start with.
1510                          */
1511                         if (++fbno == 0)
1512                                 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1513                         /*
1514                          * If it's ifbno we already looked at it.
1515                          */
1516                         if (fbno == ifbno)
1517                                 fbno++;
1518                         /*
1519                          * If it's off the end we're done.
1520                          */
1521                         if (fbno >= lastfbno)
1522                                 break;
1523                         /*
1524                          * Read the block.  There can be holes in the
1525                          * freespace blocks, so this might not succeed.
1526                          * This should be really rare, so there's no reason
1527                          * to avoid it.
1528                          */
1529                         if ((error = xfs_da_read_buf(tp, dp,
1530                                         xfs_dir2_db_to_da(mp, fbno), -2, &fbp,
1531                                         XFS_DATA_FORK))) {
1532                                 return error;
1533                         }
1534                         if (unlikely(fbp == NULL)) {
1535                                 continue;
1536                         }
1537                         free = fbp->data;
1538                         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1539                         findex = 0;
1540                 }
1541                 /*
1542                  * Look at the current free entry.  Is it good enough?
1543                  */
1544                 if (be16_to_cpu(free->bests[findex]) != NULLDATAOFF &&
1545                     be16_to_cpu(free->bests[findex]) >= length)
1546                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1547                 else {
1548                         /*
1549                          * Are we done with the freeblock?
1550                          */
1551                         if (++findex == be32_to_cpu(free->hdr.nvalid)) {
1552                                 /*
1553                                  * Drop the block.
1554                                  */
1555                                 xfs_da_brelse(tp, fbp);
1556                                 fbp = NULL;
1557                                 if (fblk && fblk->bp)
1558                                         fblk->bp = NULL;
1559                         }
1560                 }
1561         }
1562         /*
1563          * If we don't have a data block, we need to allocate one and make
1564          * the freespace entries refer to it.
1565          */
1566         if (unlikely(dbno == -1)) {
1567                 /*
1568                  * Not allowed to allocate, return failure.
1569                  */
1570                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
1571                                                         args->total == 0) {
1572                         /*
1573                          * Drop the freespace buffer unless it came from our
1574                          * caller.
1575                          */
1576                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1577                                 xfs_da_buf_done(fbp);
1578                         return XFS_ERROR(ENOSPC);
1579                 }
1580                 /*
1581                  * Allocate and initialize the new data block.
1582                  */
1583                 if (unlikely((error = xfs_dir2_grow_inode(args,
1584                                                          XFS_DIR2_DATA_SPACE,
1585                                                          &dbno)) ||
1586                     (error = xfs_dir2_data_init(args, dbno, &dbp)))) {
1587                         /*
1588                          * Drop the freespace buffer unless it came from our
1589                          * caller.
1590                          */
1591                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1592                                 xfs_da_buf_done(fbp);
1593                         return error;
1594                 }
1595                 /*
1596                  * If (somehow) we have a freespace block, get rid of it.
1597                  */
1598                 if (fbp)
1599                         xfs_da_brelse(tp, fbp);
1600                 if (fblk && fblk->bp)
1601                         fblk->bp = NULL;
1602
1603                 /*
1604                  * Get the freespace block corresponding to the data block
1605                  * that was just allocated.
1606                  */
1607                 fbno = xfs_dir2_db_to_fdb(mp, dbno);
1608                 if (unlikely(error = xfs_da_read_buf(tp, dp,
1609                                 xfs_dir2_db_to_da(mp, fbno), -2, &fbp,
1610                                 XFS_DATA_FORK))) {
1611                         xfs_da_buf_done(dbp);
1612                         return error;
1613                 }
1614                 /*
1615                  * If there wasn't a freespace block, the read will
1616                  * return a NULL fbp.  Allocate and initialize a new one.
1617                  */
1618                 if( fbp == NULL ) {
1619                         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1620                                                         &fbno))) {
1621                                 return error;
1622                         }
1623
1624                         if (unlikely(xfs_dir2_db_to_fdb(mp, dbno) != fbno)) {
1625                                 cmn_err(CE_ALERT,
1626                                         "xfs_dir2_node_addname_int: dir ino "
1627                                         "%llu needed freesp block %lld for\n"
1628                                         "  data block %lld, got %lld\n"
1629                                         "  ifbno %llu lastfbno %d\n",
1630                                         (unsigned long long)dp->i_ino,
1631                                         (long long)xfs_dir2_db_to_fdb(mp, dbno),
1632                                         (long long)dbno, (long long)fbno,
1633                                         (unsigned long long)ifbno, lastfbno);
1634                                 if (fblk) {
1635                                         cmn_err(CE_ALERT,
1636                                                 " fblk 0x%p blkno %llu "
1637                                                 "index %d magic 0x%x\n",
1638                                                 fblk,
1639                                                 (unsigned long long)fblk->blkno,
1640                                                 fblk->index,
1641                                                 fblk->magic);
1642                                 } else {
1643                                         cmn_err(CE_ALERT,
1644                                                 " ... fblk is NULL\n");
1645                                 }
1646                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1647                                                  XFS_ERRLEVEL_LOW, mp);
1648                                 return XFS_ERROR(EFSCORRUPTED);
1649                         }
1650
1651                         /*
1652                          * Get a buffer for the new block.
1653                          */
1654                         if ((error = xfs_da_get_buf(tp, dp,
1655                                                    xfs_dir2_db_to_da(mp, fbno),
1656                                                    -1, &fbp, XFS_DATA_FORK))) {
1657                                 return error;
1658                         }
1659                         ASSERT(fbp != NULL);
1660
1661                         /*
1662                          * Initialize the new block to be empty, and remember
1663                          * its first slot as our empty slot.
1664                          */
1665                         free = fbp->data;
1666                         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
1667                         free->hdr.firstdb = cpu_to_be32(
1668                                 (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1669                                 XFS_DIR2_MAX_FREE_BESTS(mp));
1670                         free->hdr.nvalid = 0;
1671                         free->hdr.nused = 0;
1672                 } else {
1673                         free = fbp->data;
1674                         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1675                 }
1676
1677                 /*
1678                  * Set the freespace block index from the data block number.
1679                  */
1680                 findex = xfs_dir2_db_to_fdindex(mp, dbno);
1681                 /*
1682                  * If it's after the end of the current entries in the
1683                  * freespace block, extend that table.
1684                  */
1685                 if (findex >= be32_to_cpu(free->hdr.nvalid)) {
1686                         ASSERT(findex < XFS_DIR2_MAX_FREE_BESTS(mp));
1687                         free->hdr.nvalid = cpu_to_be32(findex + 1);
1688                         /*
1689                          * Tag new entry so nused will go up.
1690                          */
1691                         free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1692                 }
1693                 /*
1694                  * If this entry was for an empty data block
1695                  * (this should always be true) then update the header.
1696                  */
1697                 if (be16_to_cpu(free->bests[findex]) == NULLDATAOFF) {
1698                         be32_add_cpu(&free->hdr.nused, 1);
1699                         xfs_dir2_free_log_header(tp, fbp);
1700                 }
1701                 /*
1702                  * Update the real value in the table.
1703                  * We haven't allocated the data entry yet so this will
1704                  * change again.
1705                  */
1706                 data = dbp->data;
1707                 free->bests[findex] = data->hdr.bestfree[0].length;
1708                 logfree = 1;
1709         }
1710         /*
1711          * We had a data block so we don't have to make a new one.
1712          */
1713         else {
1714                 /*
1715                  * If just checking, we succeeded.
1716                  */
1717                 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
1718                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1719                                 xfs_da_buf_done(fbp);
1720                         return 0;
1721                 }
1722                 /*
1723                  * Read the data block in.
1724                  */
1725                 if (unlikely(
1726                     error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1727                                 -1, &dbp, XFS_DATA_FORK))) {
1728                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1729                                 xfs_da_buf_done(fbp);
1730                         return error;
1731                 }
1732                 data = dbp->data;
1733                 logfree = 0;
1734         }
1735         ASSERT(be16_to_cpu(data->hdr.bestfree[0].length) >= length);
1736         /*
1737          * Point to the existing unused space.
1738          */
1739         dup = (xfs_dir2_data_unused_t *)
1740               ((char *)data + be16_to_cpu(data->hdr.bestfree[0].offset));
1741         needscan = needlog = 0;
1742         /*
1743          * Mark the first part of the unused space, inuse for us.
1744          */
1745         xfs_dir2_data_use_free(tp, dbp, dup,
1746                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)data), length,
1747                 &needlog, &needscan);
1748         /*
1749          * Fill in the new entry and log it.
1750          */
1751         dep = (xfs_dir2_data_entry_t *)dup;
1752         dep->inumber = cpu_to_be64(args->inumber);
1753         dep->namelen = args->namelen;
1754         memcpy(dep->name, args->name, dep->namelen);
1755         tagp = xfs_dir2_data_entry_tag_p(dep);
1756         *tagp = cpu_to_be16((char *)dep - (char *)data);
1757         xfs_dir2_data_log_entry(tp, dbp, dep);
1758         /*
1759          * Rescan the block for bestfree if needed.
1760          */
1761         if (needscan)
1762                 xfs_dir2_data_freescan(mp, data, &needlog);
1763         /*
1764          * Log the data block header if needed.
1765          */
1766         if (needlog)
1767                 xfs_dir2_data_log_header(tp, dbp);
1768         /*
1769          * If the freespace entry is now wrong, update it.
1770          */
1771         if (be16_to_cpu(free->bests[findex]) != be16_to_cpu(data->hdr.bestfree[0].length)) {
1772                 free->bests[findex] = data->hdr.bestfree[0].length;
1773                 logfree = 1;
1774         }
1775         /*
1776          * Log the freespace entry if needed.
1777          */
1778         if (logfree)
1779                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1780         /*
1781          * If the caller didn't hand us the freespace block, drop it.
1782          */
1783         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1784                 xfs_da_buf_done(fbp);
1785         /*
1786          * Return the data block and offset in args, then drop the data block.
1787          */
1788         args->blkno = (xfs_dablk_t)dbno;
1789         args->index = be16_to_cpu(*tagp);
1790         xfs_da_buf_done(dbp);
1791         return 0;
1792 }
1793
1794 /*
1795  * Lookup an entry in a node-format directory.
1796  * All the real work happens in xfs_da_node_lookup_int.
1797  * The only real output is the inode number of the entry.
1798  */
1799 int                                             /* error */
1800 xfs_dir2_node_lookup(
1801         xfs_da_args_t   *args)                  /* operation arguments */
1802 {
1803         int             error;                  /* error return value */
1804         int             i;                      /* btree level */
1805         int             rval;                   /* operation return value */
1806         xfs_da_state_t  *state;                 /* btree cursor */
1807
1808         xfs_dir2_trace_args("node_lookup", args);
1809         /*
1810          * Allocate and initialize the btree cursor.
1811          */
1812         state = xfs_da_state_alloc();
1813         state->args = args;
1814         state->mp = args->dp->i_mount;
1815         state->blocksize = state->mp->m_dirblksize;
1816         state->node_ents = state->mp->m_dir_node_ents;
1817         /*
1818          * Fill in the path to the entry in the cursor.
1819          */
1820         error = xfs_da_node_lookup_int(state, &rval);
1821         if (error)
1822                 rval = error;
1823         else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE)
1824                 rval = EEXIST;  /* a case-insensitive match was found */
1825         /*
1826          * Release the btree blocks and leaf block.
1827          */
1828         for (i = 0; i < state->path.active; i++) {
1829                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1830                 state->path.blk[i].bp = NULL;
1831         }
1832         /*
1833          * Release the data block if we have it.
1834          */
1835         if (state->extravalid && state->extrablk.bp) {
1836                 xfs_da_brelse(args->trans, state->extrablk.bp);
1837                 state->extrablk.bp = NULL;
1838         }
1839         xfs_da_state_free(state);
1840         return rval;
1841 }
1842
1843 /*
1844  * Remove an entry from a node-format directory.
1845  */
1846 int                                             /* error */
1847 xfs_dir2_node_removename(
1848         xfs_da_args_t           *args)          /* operation arguments */
1849 {
1850         xfs_da_state_blk_t      *blk;           /* leaf block */
1851         int                     error;          /* error return value */
1852         int                     rval;           /* operation return value */
1853         xfs_da_state_t          *state;         /* btree cursor */
1854
1855         xfs_dir2_trace_args("node_removename", args);
1856         /*
1857          * Allocate and initialize the btree cursor.
1858          */
1859         state = xfs_da_state_alloc();
1860         state->args = args;
1861         state->mp = args->dp->i_mount;
1862         state->blocksize = state->mp->m_dirblksize;
1863         state->node_ents = state->mp->m_dir_node_ents;
1864         /*
1865          * Look up the entry we're deleting, set up the cursor.
1866          */
1867         error = xfs_da_node_lookup_int(state, &rval);
1868         if (error)
1869                 rval = error;
1870         /*
1871          * Didn't find it, upper layer screwed up.
1872          */
1873         if (rval != EEXIST) {
1874                 xfs_da_state_free(state);
1875                 return rval;
1876         }
1877         blk = &state->path.blk[state->path.active - 1];
1878         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1879         ASSERT(state->extravalid);
1880         /*
1881          * Remove the leaf and data entries.
1882          * Extrablk refers to the data block.
1883          */
1884         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
1885                 &state->extrablk, &rval);
1886         if (error)
1887                 return error;
1888         /*
1889          * Fix the hash values up the btree.
1890          */
1891         xfs_da_fixhashpath(state, &state->path);
1892         /*
1893          * If we need to join leaf blocks, do it.
1894          */
1895         if (rval && state->path.active > 1)
1896                 error = xfs_da_join(state);
1897         /*
1898          * If no errors so far, try conversion to leaf format.
1899          */
1900         if (!error)
1901                 error = xfs_dir2_node_to_leaf(state);
1902         xfs_da_state_free(state);
1903         return error;
1904 }
1905
1906 /*
1907  * Replace an entry's inode number in a node-format directory.
1908  */
1909 int                                             /* error */
1910 xfs_dir2_node_replace(
1911         xfs_da_args_t           *args)          /* operation arguments */
1912 {
1913         xfs_da_state_blk_t      *blk;           /* leaf block */
1914         xfs_dir2_data_t         *data;          /* data block structure */
1915         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
1916         int                     error;          /* error return value */
1917         int                     i;              /* btree level */
1918         xfs_ino_t               inum;           /* new inode number */
1919         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1920         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
1921         int                     rval;           /* internal return value */
1922         xfs_da_state_t          *state;         /* btree cursor */
1923
1924         xfs_dir2_trace_args("node_replace", args);
1925         /*
1926          * Allocate and initialize the btree cursor.
1927          */
1928         state = xfs_da_state_alloc();
1929         state->args = args;
1930         state->mp = args->dp->i_mount;
1931         state->blocksize = state->mp->m_dirblksize;
1932         state->node_ents = state->mp->m_dir_node_ents;
1933         inum = args->inumber;
1934         /*
1935          * Lookup the entry to change in the btree.
1936          */
1937         error = xfs_da_node_lookup_int(state, &rval);
1938         if (error) {
1939                 rval = error;
1940         }
1941         /*
1942          * It should be found, since the vnodeops layer has looked it up
1943          * and locked it.  But paranoia is good.
1944          */
1945         if (rval == EEXIST) {
1946                 /*
1947                  * Find the leaf entry.
1948                  */
1949                 blk = &state->path.blk[state->path.active - 1];
1950                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1951                 leaf = blk->bp->data;
1952                 lep = &leaf->ents[blk->index];
1953                 ASSERT(state->extravalid);
1954                 /*
1955                  * Point to the data entry.
1956                  */
1957                 data = state->extrablk.bp->data;
1958                 ASSERT(be32_to_cpu(data->hdr.magic) == XFS_DIR2_DATA_MAGIC);
1959                 dep = (xfs_dir2_data_entry_t *)
1960                       ((char *)data +
1961                        xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
1962                 ASSERT(inum != be64_to_cpu(dep->inumber));
1963                 /*
1964                  * Fill in the new inode number and log the entry.
1965                  */
1966                 dep->inumber = cpu_to_be64(inum);
1967                 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
1968                 rval = 0;
1969         }
1970         /*
1971          * Didn't find it, and we're holding a data block.  Drop it.
1972          */
1973         else if (state->extravalid) {
1974                 xfs_da_brelse(args->trans, state->extrablk.bp);
1975                 state->extrablk.bp = NULL;
1976         }
1977         /*
1978          * Release all the buffers in the cursor.
1979          */
1980         for (i = 0; i < state->path.active; i++) {
1981                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1982                 state->path.blk[i].bp = NULL;
1983         }
1984         xfs_da_state_free(state);
1985         return rval;
1986 }
1987
1988 /*
1989  * Trim off a trailing empty freespace block.
1990  * Return (in rvalp) 1 if we did it, 0 if not.
1991  */
1992 int                                             /* error */
1993 xfs_dir2_node_trim_free(
1994         xfs_da_args_t           *args,          /* operation arguments */
1995         xfs_fileoff_t           fo,             /* free block number */
1996         int                     *rvalp)         /* out: did something */
1997 {
1998         xfs_dabuf_t             *bp;            /* freespace buffer */
1999         xfs_inode_t             *dp;            /* incore directory inode */
2000         int                     error;          /* error return code */
2001         xfs_dir2_free_t         *free;          /* freespace structure */
2002         xfs_mount_t             *mp;            /* filesystem mount point */
2003         xfs_trans_t             *tp;            /* transaction pointer */
2004
2005         dp = args->dp;
2006         mp = dp->i_mount;
2007         tp = args->trans;
2008         /*
2009          * Read the freespace block.
2010          */
2011         if (unlikely(error = xfs_da_read_buf(tp, dp, (xfs_dablk_t)fo, -2, &bp,
2012                         XFS_DATA_FORK))) {
2013                 return error;
2014         }
2015
2016         /*
2017          * There can be holes in freespace.  If fo is a hole, there's
2018          * nothing to do.
2019          */
2020         if (bp == NULL) {
2021                 return 0;
2022         }
2023         free = bp->data;
2024         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
2025         /*
2026          * If there are used entries, there's nothing to do.
2027          */
2028         if (be32_to_cpu(free->hdr.nused) > 0) {
2029                 xfs_da_brelse(tp, bp);
2030                 *rvalp = 0;
2031                 return 0;
2032         }
2033         /*
2034          * Blow the block away.
2035          */
2036         if ((error =
2037             xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
2038                     bp))) {
2039                 /*
2040                  * Can't fail with ENOSPC since that only happens with no
2041                  * space reservation, when breaking up an extent into two
2042                  * pieces.  This is the last block of an extent.
2043                  */
2044                 ASSERT(error != ENOSPC);
2045                 xfs_da_brelse(tp, bp);
2046                 return error;
2047         }
2048         /*
2049          * Return that we succeeded.
2050          */
2051         *rvalp = 1;
2052         return 0;
2053 }