git.oblomov.eu Git - linux-2.6/blob - 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->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->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
 560         dp = args->dp;
 561         tp = args->trans;
 562         mp = dp->i_mount;
 563         leaf = bp->data;
 564         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 565 #ifdef __KERNEL__
 566         ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
 567 #endif
 568         xfs_dir2_leafn_check(dp, bp);
 569         /*
 570          * Look up the hash value in the leaf entries.
 571          */
 572         index = xfs_dir2_leaf_search_hash(args, bp);
 573         /*
 574          * Do we have a buffer coming in?
 575          */
 576         if (state->extravalid) {
 577                 curbp = state->extrablk.bp;
 578                 curdb = state->extrablk.blkno;
 579         }
 580         /*
 581          * Loop over leaf entries with the right hash value.
 582          */
 583         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
 584                                 be32_to_cpu(lep->hashval) == args->hashval;
 585                                 lep++, index++) {
 586                 /*
 587                  * Skip stale leaf entries.
 588                  */
 589                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 590                         continue;
 591                 /*
 592                  * Pull the data block number from the entry.
 593                  */
 594                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 595                 /*
 596                  * Not adding a new entry, so we really want to find
 597                  * the name given to us.
 598                  *
 599                  * If it's a different data block, go get it.
 600                  */
 601                 if (newdb != curdb) {
 602                         /*
 603                          * If we had a block before, drop it.
 604                          */
 605                         if (curbp)
 606                                 xfs_da_brelse(tp, curbp);
 607                         /*
 608                          * Read the data block.
 609                          */
 610                         error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp,
 611                                         newdb), -1, &curbp, XFS_DATA_FORK);
 612                         if (error)
 613                                 return error;
 614                         xfs_dir2_data_check(dp, curbp);
 615                         curdb = newdb;
 616                 }
 617                 /*
 618                  * Point to the data entry.
 619                  */
 620                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->data +
 621                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
 622                 /*
 623                  * Compare the entry, return it if it matches.
 624                  */
 625                 if (dep->namelen == args->namelen && memcmp(dep->name,
 626                                         args->name, args->namelen) == 0) {
 627                         args->inumber = be64_to_cpu(dep->inumber);
 628                         di = (int)((char *)dep - (char *)curbp->data);
 629                         error = EEXIST;
 630                         goto out;
 631                 }
 632         }
 633         /* Didn't find a match. */
 634         error = ENOENT;
 635         di = -1;
 636         ASSERT(index == be16_to_cpu(leaf->hdr.count) || args->oknoent);
 637 out:
 638         if (curbp) {
 639                 /* Giving back a data block. */
 640                 state->extravalid = 1;
 641                 state->extrablk.bp = curbp;
 642                 state->extrablk.index = di;
 643                 state->extrablk.blkno = curdb;
 644                 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 645         } else {
 646                 state->extravalid = 0;
 647         }
 648         /*
 649          * Return the index, that will be the insertion point.
 650          */
 651         *indexp = index;
 652         return XFS_ERROR(error);
 653 }
 654
 655 /*
 656  * Look up a leaf entry in a node-format leaf block.
 657  * If this is an addname then the extrablk in state is a freespace block,
 658  * otherwise it's a data block.
 659  */
 660 int
 661 xfs_dir2_leafn_lookup_int(
 662         xfs_dabuf_t             *bp,            /* leaf buffer */
 663         xfs_da_args_t           *args,          /* operation arguments */
 664         int                     *indexp,        /* out: leaf entry index */
 665         xfs_da_state_t          *state)         /* state to fill in */
 666 {
 667         if (args->addname)
 668                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 669                                                         state);
 670         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 671 }
 672
 673 /*
 674  * Move count leaf entries from source to destination leaf.
 675  * Log entries and headers.  Stale entries are preserved.
 676  */
 677 static void
 678 xfs_dir2_leafn_moveents(
 679         xfs_da_args_t   *args,                  /* operation arguments */
 680         xfs_dabuf_t     *bp_s,                  /* source leaf buffer */
 681         int             start_s,                /* source leaf index */
 682         xfs_dabuf_t     *bp_d,                  /* destination leaf buffer */
 683         int             start_d,                /* destination leaf index */
 684         int             count)                  /* count of leaves to copy */
 685 {
 686         xfs_dir2_leaf_t *leaf_d;                /* destination leaf structure */
 687         xfs_dir2_leaf_t *leaf_s;                /* source leaf structure */
 688         int             stale;                  /* count stale leaves copied */
 689         xfs_trans_t     *tp;                    /* transaction pointer */
 690
 691         xfs_dir2_trace_args_bibii("leafn_moveents", args, bp_s, start_s, bp_d,
 692                 start_d, count);
 693         /*
 694          * Silently return if nothing to do.
 695          */
 696         if (count == 0) {
 697                 return;
 698         }
 699         tp = args->trans;
 700         leaf_s = bp_s->data;
 701         leaf_d = bp_d->data;
 702         /*
 703          * If the destination index is not the end of the current
 704          * destination leaf entries, open up a hole in the destination
 705          * to hold the new entries.
 706          */
 707         if (start_d < be16_to_cpu(leaf_d->hdr.count)) {
 708                 memmove(&leaf_d->ents[start_d + count], &leaf_d->ents[start_d],
 709                         (be16_to_cpu(leaf_d->hdr.count) - start_d) *
 710                         sizeof(xfs_dir2_leaf_entry_t));
 711                 xfs_dir2_leaf_log_ents(tp, bp_d, start_d + count,
 712                         count + be16_to_cpu(leaf_d->hdr.count) - 1);
 713         }
 714         /*
 715          * If the source has stale leaves, count the ones in the copy range
 716          * so we can update the header correctly.
 717          */
 718         if (leaf_s->hdr.stale) {
 719                 int     i;                      /* temp leaf index */
 720
 721                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 722                         if (be32_to_cpu(leaf_s->ents[i].address) == XFS_DIR2_NULL_DATAPTR)
 723                                 stale++;
 724                 }
 725         } else
 726                 stale = 0;
 727         /*
 728          * Copy the leaf entries from source to destination.
 729          */
 730         memcpy(&leaf_d->ents[start_d], &leaf_s->ents[start_s],
 731                 count * sizeof(xfs_dir2_leaf_entry_t));
 732         xfs_dir2_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
 733         /*
 734          * If there are source entries after the ones we copied,
 735          * delete the ones we copied by sliding the next ones down.
 736          */
 737         if (start_s + count < be16_to_cpu(leaf_s->hdr.count)) {
 738                 memmove(&leaf_s->ents[start_s], &leaf_s->ents[start_s + count],
 739                         count * sizeof(xfs_dir2_leaf_entry_t));
 740                 xfs_dir2_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
 741         }
 742         /*
 743          * Update the headers and log them.
 744          */
 745         be16_add_cpu(&leaf_s->hdr.count, -(count));
 746         be16_add_cpu(&leaf_s->hdr.stale, -(stale));
 747         be16_add_cpu(&leaf_d->hdr.count, count);
 748         be16_add_cpu(&leaf_d->hdr.stale, stale);
 749         xfs_dir2_leaf_log_header(tp, bp_s);
 750         xfs_dir2_leaf_log_header(tp, bp_d);
 751         xfs_dir2_leafn_check(args->dp, bp_s);
 752         xfs_dir2_leafn_check(args->dp, bp_d);
 753 }
 754
 755 /*
 756  * Determine the sort order of two leaf blocks.
 757  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 758  */
 759 int                                             /* sort order */
 760 xfs_dir2_leafn_order(
 761         xfs_dabuf_t     *leaf1_bp,              /* leaf1 buffer */
 762         xfs_dabuf_t     *leaf2_bp)              /* leaf2 buffer */
 763 {
 764         xfs_dir2_leaf_t *leaf1;                 /* leaf1 structure */
 765         xfs_dir2_leaf_t *leaf2;                 /* leaf2 structure */
 766
 767         leaf1 = leaf1_bp->data;
 768         leaf2 = leaf2_bp->data;
 769         ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 770         ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 771         if (be16_to_cpu(leaf1->hdr.count) > 0 &&
 772             be16_to_cpu(leaf2->hdr.count) > 0 &&
 773             (be32_to_cpu(leaf2->ents[0].hashval) < be32_to_cpu(leaf1->ents[0].hashval) ||
 774              be32_to_cpu(leaf2->ents[be16_to_cpu(leaf2->hdr.count) - 1].hashval) <
 775              be32_to_cpu(leaf1->ents[be16_to_cpu(leaf1->hdr.count) - 1].hashval)))
 776                 return 1;
 777         return 0;
 778 }
 779
 780 /*
 781  * Rebalance leaf entries between two leaf blocks.
 782  * This is actually only called when the second block is new,
 783  * though the code deals with the general case.
 784  * A new entry will be inserted in one of the blocks, and that
 785  * entry is taken into account when balancing.
 786  */
 787 static void
 788 xfs_dir2_leafn_rebalance(
 789         xfs_da_state_t          *state,         /* btree cursor */
 790         xfs_da_state_blk_t      *blk1,          /* first btree block */
 791         xfs_da_state_blk_t      *blk2)          /* second btree block */
 792 {
 793         xfs_da_args_t           *args;          /* operation arguments */
 794         int                     count;          /* count (& direction) leaves */
 795         int                     isleft;         /* new goes in left leaf */
 796         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 797         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 798         int                     mid;            /* midpoint leaf index */
 799 #ifdef DEBUG
 800         int                     oldstale;       /* old count of stale leaves */
 801 #endif
 802         int                     oldsum;         /* old total leaf count */
 803         int                     swap;           /* swapped leaf blocks */
 804
 805         args = state->args;
 806         /*
 807          * If the block order is wrong, swap the arguments.
 808          */
 809         if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
 810                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
 811
 812                 tmp = blk1;
 813                 blk1 = blk2;
 814                 blk2 = tmp;
 815         }
 816         leaf1 = blk1->bp->data;
 817         leaf2 = blk2->bp->data;
 818         oldsum = be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count);
 819 #ifdef DEBUG
 820         oldstale = be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale);
 821 #endif
 822         mid = oldsum >> 1;
 823         /*
 824          * If the old leaf count was odd then the new one will be even,
 825          * so we need to divide the new count evenly.
 826          */
 827         if (oldsum & 1) {
 828                 xfs_dahash_t    midhash;        /* middle entry hash value */
 829
 830                 if (mid >= be16_to_cpu(leaf1->hdr.count))
 831                         midhash = be32_to_cpu(leaf2->ents[mid - be16_to_cpu(leaf1->hdr.count)].hashval);
 832                 else
 833                         midhash = be32_to_cpu(leaf1->ents[mid].hashval);
 834                 isleft = args->hashval <= midhash;
 835         }
 836         /*
 837          * If the old count is even then the new count is odd, so there's
 838          * no preferred side for the new entry.
 839          * Pick the left one.
 840          */
 841         else
 842                 isleft = 1;
 843         /*
 844          * Calculate moved entry count.  Positive means left-to-right,
 845          * negative means right-to-left.  Then move the entries.
 846          */
 847         count = be16_to_cpu(leaf1->hdr.count) - mid + (isleft == 0);
 848         if (count > 0)
 849                 xfs_dir2_leafn_moveents(args, blk1->bp,
 850                         be16_to_cpu(leaf1->hdr.count) - count, blk2->bp, 0, count);
 851         else if (count < 0)
 852                 xfs_dir2_leafn_moveents(args, blk2->bp, 0, blk1->bp,
 853                         be16_to_cpu(leaf1->hdr.count), count);
 854         ASSERT(be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count) == oldsum);
 855         ASSERT(be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale) == oldstale);
 856         /*
 857          * Mark whether we're inserting into the old or new leaf.
 858          */
 859         if (be16_to_cpu(leaf1->hdr.count) < be16_to_cpu(leaf2->hdr.count))
 860                 state->inleaf = swap;
 861         else if (be16_to_cpu(leaf1->hdr.count) > be16_to_cpu(leaf2->hdr.count))
 862                 state->inleaf = !swap;
 863         else
 864                 state->inleaf =
 865                         swap ^ (blk1->index <= be16_to_cpu(leaf1->hdr.count));
 866         /*
 867          * Adjust the expected index for insertion.
 868          */
 869         if (!state->inleaf)
 870                 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
 871
 872         /*
 873          * Finally sanity check just to make sure we are not returning a
 874          * negative index
 875          */
 876         if(blk2->index < 0) {
 877                 state->inleaf = 1;
 878                 blk2->index = 0;
 879                 cmn_err(CE_ALERT,
 880                         "xfs_dir2_leafn_rebalance: picked the wrong leaf? reverting original leaf: "
 881                         "blk1->index %d\n",
 882                         blk1->index);
 883         }
 884 }
 885
 886 /*
 887  * Remove an entry from a node directory.
 888  * This removes the leaf entry and the data entry,
 889  * and updates the free block if necessary.
 890  */
 891 static int                                      /* error */
 892 xfs_dir2_leafn_remove(
 893         xfs_da_args_t           *args,          /* operation arguments */
 894         xfs_dabuf_t             *bp,            /* leaf buffer */
 895         int                     index,          /* leaf entry index */
 896         xfs_da_state_blk_t      *dblk,          /* data block */
 897         int                     *rval)          /* resulting block needs join */
 898 {
 899         xfs_dir2_data_t         *data;          /* data block structure */
 900         xfs_dir2_db_t           db;             /* data block number */
 901         xfs_dabuf_t             *dbp;           /* data block buffer */
 902         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 903         xfs_inode_t             *dp;            /* incore directory inode */
 904         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 905         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 906         int                     longest;        /* longest data free entry */
 907         int                     off;            /* data block entry offset */
 908         xfs_mount_t             *mp;            /* filesystem mount point */
 909         int                     needlog;        /* need to log data header */
 910         int                     needscan;       /* need to rescan data frees */
 911         xfs_trans_t             *tp;            /* transaction pointer */
 912
 913         xfs_dir2_trace_args_sb("leafn_remove", args, index, bp);
 914         dp = args->dp;
 915         tp = args->trans;
 916         mp = dp->i_mount;
 917         leaf = bp->data;
 918         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 919         /*
 920          * Point to the entry we're removing.
 921          */
 922         lep = &leaf->ents[index];
 923         /*
 924          * Extract the data block and offset from the entry.
 925          */
 926         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 927         ASSERT(dblk->blkno == db);
 928         off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
 929         ASSERT(dblk->index == off);
 930         /*
 931          * Kill the leaf entry by marking it stale.
 932          * Log the leaf block changes.
 933          */
 934         be16_add_cpu(&leaf->hdr.stale, 1);
 935         xfs_dir2_leaf_log_header(tp, bp);
 936         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 937         xfs_dir2_leaf_log_ents(tp, bp, index, index);
 938         /*
 939          * Make the data entry free.  Keep track of the longest freespace
 940          * in the data block in case it changes.
 941          */
 942         dbp = dblk->bp;
 943         data = dbp->data;
 944         dep = (xfs_dir2_data_entry_t *)((char *)data + off);
 945         longest = be16_to_cpu(data->hdr.bestfree[0].length);
 946         needlog = needscan = 0;
 947         xfs_dir2_data_make_free(tp, dbp, off,
 948                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
 949         /*
 950          * Rescan the data block freespaces for bestfree.
 951          * Log the data block header if needed.
 952          */
 953         if (needscan)
 954                 xfs_dir2_data_freescan(mp, data, &needlog);
 955         if (needlog)
 956                 xfs_dir2_data_log_header(tp, dbp);
 957         xfs_dir2_data_check(dp, dbp);
 958         /*
 959          * If the longest data block freespace changes, need to update
 960          * the corresponding freeblock entry.
 961          */
 962         if (longest < be16_to_cpu(data->hdr.bestfree[0].length)) {
 963                 int             error;          /* error return value */
 964                 xfs_dabuf_t     *fbp;           /* freeblock buffer */
 965                 xfs_dir2_db_t   fdb;            /* freeblock block number */
 966                 int             findex;         /* index in freeblock entries */
 967                 xfs_dir2_free_t *free;          /* freeblock structure */
 968                 int             logfree;        /* need to log free entry */
 969
 970                 /*
 971                  * Convert the data block number to a free block,
 972                  * read in the free block.
 973                  */
 974                 fdb = xfs_dir2_db_to_fdb(mp, db);
 975                 if ((error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb),
 976                                 -1, &fbp, XFS_DATA_FORK))) {
 977                         return error;
 978                 }
 979                 free = fbp->data;
 980                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
 981                 ASSERT(be32_to_cpu(free->hdr.firstdb) ==
 982                        XFS_DIR2_MAX_FREE_BESTS(mp) *
 983                        (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
 984                 /*
 985                  * Calculate which entry we need to fix.
 986                  */
 987                 findex = xfs_dir2_db_to_fdindex(mp, db);
 988                 longest = be16_to_cpu(data->hdr.bestfree[0].length);
 989                 /*
 990                  * If the data block is now empty we can get rid of it
 991                  * (usually).
 992                  */
 993                 if (longest == mp->m_dirblksize - (uint)sizeof(data->hdr)) {
 994                         /*
 995                          * Try to punch out the data block.
 996                          */
 997                         error = xfs_dir2_shrink_inode(args, db, dbp);
 998                         if (error == 0) {
 999                                 dblk->bp = NULL;
1000                                 data = NULL;
1001                         }
1002                         /*
1003                          * We can get ENOSPC if there's no space reservation.
1004                          * In this case just drop the buffer and some one else
1005                          * will eventually get rid of the empty block.
1006                          */
1007                         else if (error == ENOSPC && args->total == 0)
1008                                 xfs_da_buf_done(dbp);
1009                         else
1010                                 return error;
1011                 }
1012                 /*
1013                  * If we got rid of the data block, we can eliminate that entry
1014                  * in the free block.
1015                  */
1016                 if (data == NULL) {
1017                         /*
1018                          * One less used entry in the free table.
1019                          */
1020                         be32_add_cpu(&free->hdr.nused, -1);
1021                         xfs_dir2_free_log_header(tp, fbp);
1022                         /*
1023                          * If this was the last entry in the table, we can
1024                          * trim the table size back.  There might be other
1025                          * entries at the end referring to non-existent
1026                          * data blocks, get those too.
1027                          */
1028                         if (findex == be32_to_cpu(free->hdr.nvalid) - 1) {
1029                                 int     i;              /* free entry index */
1030
1031                                 for (i = findex - 1;
1032                                      i >= 0 && be16_to_cpu(free->bests[i]) == NULLDATAOFF;
1033                                      i--)
1034                                         continue;
1035                                 free->hdr.nvalid = cpu_to_be32(i + 1);
1036                                 logfree = 0;
1037                         }
1038                         /*
1039                          * Not the last entry, just punch it out.
1040                          */
1041                         else {
1042                                 free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1043                                 logfree = 1;
1044                         }
1045                         /*
1046                          * If there are no useful entries left in the block,
1047                          * get rid of the block if we can.
1048                          */
1049                         if (!free->hdr.nused) {
1050                                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1051                                 if (error == 0) {
1052                                         fbp = NULL;
1053                                         logfree = 0;
1054                                 } else if (error != ENOSPC || args->total != 0)
1055                                         return error;
1056                                 /*
1057                                  * It's possible to get ENOSPC if there is no
1058                                  * space reservation.  In this case some one
1059                                  * else will eventually get rid of this block.
1060                                  */
1061                         }
1062                 }
1063                 /*
1064                  * Data block is not empty, just set the free entry to
1065                  * the new value.
1066                  */
1067                 else {
1068                         free->bests[findex] = cpu_to_be16(longest);
1069                         logfree = 1;
1070                 }
1071                 /*
1072                  * Log the free entry that changed, unless we got rid of it.
1073                  */
1074                 if (logfree)
1075                         xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1076                 /*
1077                  * Drop the buffer if we still have it.
1078                  */
1079                 if (fbp)
1080                         xfs_da_buf_done(fbp);
1081         }
1082         xfs_dir2_leafn_check(dp, bp);
1083         /*
1084          * Return indication of whether this leaf block is emtpy enough
1085          * to justify trying to join it with a neighbor.
1086          */
1087         *rval =
1088                 ((uint)sizeof(leaf->hdr) +
1089                  (uint)sizeof(leaf->ents[0]) *
1090                  (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale))) <
1091                 mp->m_dir_magicpct;
1092         return 0;
1093 }
1094
1095 /*
1096  * Split the leaf entries in the old block into old and new blocks.
1097  */
1098 int                                             /* error */
1099 xfs_dir2_leafn_split(
1100         xfs_da_state_t          *state,         /* btree cursor */
1101         xfs_da_state_blk_t      *oldblk,        /* original block */
1102         xfs_da_state_blk_t      *newblk)        /* newly created block */
1103 {
1104         xfs_da_args_t           *args;          /* operation arguments */
1105         xfs_dablk_t             blkno;          /* new leaf block number */
1106         int                     error;          /* error return value */
1107         xfs_mount_t             *mp;            /* filesystem mount point */
1108
1109         /*
1110          * Allocate space for a new leaf node.
1111          */
1112         args = state->args;
1113         mp = args->dp->i_mount;
1114         ASSERT(args != NULL);
1115         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1116         error = xfs_da_grow_inode(args, &blkno);
1117         if (error) {
1118                 return error;
1119         }
1120         /*
1121          * Initialize the new leaf block.
1122          */
1123         error = xfs_dir2_leaf_init(args, xfs_dir2_da_to_db(mp, blkno),
1124                 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1125         if (error) {
1126                 return error;
1127         }
1128         newblk->blkno = blkno;
1129         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1130         /*
1131          * Rebalance the entries across the two leaves, link the new
1132          * block into the leaves.
1133          */
1134         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1135         error = xfs_da_blk_link(state, oldblk, newblk);
1136         if (error) {
1137                 return error;
1138         }
1139         /*
1140          * Insert the new entry in the correct block.
1141          */
1142         if (state->inleaf)
1143                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1144         else
1145                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1146         /*
1147          * Update last hashval in each block since we added the name.
1148          */
1149         oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1150         newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1151         xfs_dir2_leafn_check(args->dp, oldblk->bp);
1152         xfs_dir2_leafn_check(args->dp, newblk->bp);
1153         return error;
1154 }
1155
1156 /*
1157  * Check a leaf block and its neighbors to see if the block should be
1158  * collapsed into one or the other neighbor.  Always keep the block
1159  * with the smaller block number.
1160  * If the current block is over 50% full, don't try to join it, return 0.
1161  * If the block is empty, fill in the state structure and return 2.
1162  * If it can be collapsed, fill in the state structure and return 1.
1163  * If nothing can be done, return 0.
1164  */
1165 int                                             /* error */
1166 xfs_dir2_leafn_toosmall(
1167         xfs_da_state_t          *state,         /* btree cursor */
1168         int                     *action)        /* resulting action to take */
1169 {
1170         xfs_da_state_blk_t      *blk;           /* leaf block */
1171         xfs_dablk_t             blkno;          /* leaf block number */
1172         xfs_dabuf_t             *bp;            /* leaf buffer */
1173         int                     bytes;          /* bytes in use */
1174         int                     count;          /* leaf live entry count */
1175         int                     error;          /* error return value */
1176         int                     forward;        /* sibling block direction */
1177         int                     i;              /* sibling counter */
1178         xfs_da_blkinfo_t        *info;          /* leaf block header */
1179         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1180         int                     rval;           /* result from path_shift */
1181
1182         /*
1183          * Check for the degenerate case of the block being over 50% full.
1184          * If so, it's not worth even looking to see if we might be able
1185          * to coalesce with a sibling.
1186          */
1187         blk = &state->path.blk[state->path.active - 1];
1188         info = blk->bp->data;
1189         ASSERT(be16_to_cpu(info->magic) == XFS_DIR2_LEAFN_MAGIC);
1190         leaf = (xfs_dir2_leaf_t *)info;
1191         count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1192         bytes = (uint)sizeof(leaf->hdr) + count * (uint)sizeof(leaf->ents[0]);
1193         if (bytes > (state->blocksize >> 1)) {
1194                 /*
1195                  * Blk over 50%, don't try to join.
1196                  */
1197                 *action = 0;
1198                 return 0;
1199         }
1200         /*
1201          * Check for the degenerate case of the block being empty.
1202          * If the block is empty, we'll simply delete it, no need to
1203          * coalesce it with a sibling block.  We choose (arbitrarily)
1204          * to merge with the forward block unless it is NULL.
1205          */
1206         if (count == 0) {
1207                 /*
1208                  * Make altpath point to the block we want to keep and
1209                  * path point to the block we want to drop (this one).
1210                  */
1211                 forward = (info->forw != 0);
1212                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1213                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1214                         &rval);
1215                 if (error)
1216                         return error;
1217                 *action = rval ? 2 : 0;
1218                 return 0;
1219         }
1220         /*
1221          * Examine each sibling block to see if we can coalesce with
1222          * at least 25% free space to spare.  We need to figure out
1223          * whether to merge with the forward or the backward block.
1224          * We prefer coalescing with the lower numbered sibling so as
1225          * to shrink a directory over time.
1226          */
1227         forward = be32_to_cpu(info->forw) < be32_to_cpu(info->back);
1228         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1229                 blkno = forward ? be32_to_cpu(info->forw) : be32_to_cpu(info->back);
1230                 if (blkno == 0)
1231                         continue;
1232                 /*
1233                  * Read the sibling leaf block.
1234                  */
1235                 if ((error =
1236                     xfs_da_read_buf(state->args->trans, state->args->dp, blkno,
1237                             -1, &bp, XFS_DATA_FORK))) {
1238                         return error;
1239                 }
1240                 ASSERT(bp != NULL);
1241                 /*
1242                  * Count bytes in the two blocks combined.
1243                  */
1244                 leaf = (xfs_dir2_leaf_t *)info;
1245                 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1246                 bytes = state->blocksize - (state->blocksize >> 2);
1247                 leaf = bp->data;
1248                 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1249                 count += be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1250                 bytes -= count * (uint)sizeof(leaf->ents[0]);
1251                 /*
1252                  * Fits with at least 25% to spare.
1253                  */
1254                 if (bytes >= 0)
1255                         break;
1256                 xfs_da_brelse(state->args->trans, bp);
1257         }
1258         /*
1259          * Didn't like either block, give up.
1260          */
1261         if (i >= 2) {
1262                 *action = 0;
1263                 return 0;
1264         }
1265         /*
1266          * Done with the sibling leaf block here, drop the dabuf
1267          * so path_shift can get it.
1268          */
1269         xfs_da_buf_done(bp);
1270         /*
1271          * Make altpath point to the block we want to keep (the lower
1272          * numbered block) and path point to the block we want to drop.
1273          */
1274         memcpy(&state->altpath, &state->path, sizeof(state->path));
1275         if (blkno < blk->blkno)
1276                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1277                         &rval);
1278         else
1279                 error = xfs_da_path_shift(state, &state->path, forward, 0,
1280                         &rval);
1281         if (error) {
1282                 return error;
1283         }
1284         *action = rval ? 0 : 1;
1285         return 0;
1286 }
1287
1288 /*
1289  * Move all the leaf entries from drop_blk to save_blk.
1290  * This is done as part of a join operation.
1291  */
1292 void
1293 xfs_dir2_leafn_unbalance(
1294         xfs_da_state_t          *state,         /* cursor */
1295         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1296         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1297 {
1298         xfs_da_args_t           *args;          /* operation arguments */
1299         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1300         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1301
1302         args = state->args;
1303         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1304         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1305         drop_leaf = drop_blk->bp->data;
1306         save_leaf = save_blk->bp->data;
1307         ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1308         ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1309         /*
1310          * If there are any stale leaf entries, take this opportunity
1311          * to purge them.
1312          */
1313         if (drop_leaf->hdr.stale)
1314                 xfs_dir2_leaf_compact(args, drop_blk->bp);
1315         if (save_leaf->hdr.stale)
1316                 xfs_dir2_leaf_compact(args, save_blk->bp);
1317         /*
1318          * Move the entries from drop to the appropriate end of save.
1319          */
1320         drop_blk->hashval = be32_to_cpu(drop_leaf->ents[be16_to_cpu(drop_leaf->hdr.count) - 1].hashval);
1321         if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1322                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp, 0,
1323                         be16_to_cpu(drop_leaf->hdr.count));
1324         else
1325                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp,
1326                         be16_to_cpu(save_leaf->hdr.count), be16_to_cpu(drop_leaf->hdr.count));
1327         save_blk->hashval = be32_to_cpu(save_leaf->ents[be16_to_cpu(save_leaf->hdr.count) - 1].hashval);
1328         xfs_dir2_leafn_check(args->dp, save_blk->bp);
1329 }
1330
1331 /*
1332  * Top-level node form directory addname routine.
1333  */
1334 int                                             /* error */
1335 xfs_dir2_node_addname(
1336         xfs_da_args_t           *args)          /* operation arguments */
1337 {
1338         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1339         int                     error;          /* error return value */
1340         int                     rval;           /* sub-return value */
1341         xfs_da_state_t          *state;         /* btree cursor */
1342
1343         xfs_dir2_trace_args("node_addname", args);
1344         /*
1345          * Allocate and initialize the state (btree cursor).
1346          */
1347         state = xfs_da_state_alloc();
1348         state->args = args;
1349         state->mp = args->dp->i_mount;
1350         state->blocksize = state->mp->m_dirblksize;
1351         state->node_ents = state->mp->m_dir_node_ents;
1352         /*
1353          * Look up the name.  We're not supposed to find it, but
1354          * this gives us the insertion point.
1355          */
1356         error = xfs_da_node_lookup_int(state, &rval);
1357         if (error)
1358                 rval = error;
1359         if (rval != ENOENT) {
1360                 goto done;
1361         }
1362         /*
1363          * Add the data entry to a data block.
1364          * Extravalid is set to a freeblock found by lookup.
1365          */
1366         rval = xfs_dir2_node_addname_int(args,
1367                 state->extravalid ? &state->extrablk : NULL);
1368         if (rval) {
1369                 goto done;
1370         }
1371         blk = &state->path.blk[state->path.active - 1];
1372         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1373         /*
1374          * Add the new leaf entry.
1375          */
1376         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1377         if (rval == 0) {
1378                 /*
1379                  * It worked, fix the hash values up the btree.
1380                  */
1381                 if (!args->justcheck)
1382                         xfs_da_fixhashpath(state, &state->path);
1383         } else {
1384                 /*
1385                  * It didn't work, we need to split the leaf block.
1386                  */
1387                 if (args->total == 0) {
1388                         ASSERT(rval == ENOSPC);
1389                         goto done;
1390                 }
1391                 /*
1392                  * Split the leaf block and insert the new entry.
1393                  */
1394                 rval = xfs_da_split(state);
1395         }
1396 done:
1397         xfs_da_state_free(state);
1398         return rval;
1399 }
1400
1401 /*
1402  * Add the data entry for a node-format directory name addition.
1403  * The leaf entry is added in xfs_dir2_leafn_add.
1404  * We may enter with a freespace block that the lookup found.
1405  */
1406 static int                                      /* error */
1407 xfs_dir2_node_addname_int(
1408         xfs_da_args_t           *args,          /* operation arguments */
1409         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1410 {
1411         xfs_dir2_data_t         *data;          /* data block structure */
1412         xfs_dir2_db_t           dbno;           /* data block number */
1413         xfs_dabuf_t             *dbp;           /* data block buffer */
1414         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1415         xfs_inode_t             *dp;            /* incore directory inode */
1416         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1417         int                     error;          /* error return value */
1418         xfs_dir2_db_t           fbno;           /* freespace block number */
1419         xfs_dabuf_t             *fbp;           /* freespace buffer */
1420         int                     findex;         /* freespace entry index */
1421         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1422         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1423         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1424         int                     length;         /* length of the new entry */
1425         int                     logfree;        /* need to log free entry */
1426         xfs_mount_t             *mp;            /* filesystem mount point */
1427         int                     needlog;        /* need to log data header */
1428         int                     needscan;       /* need to rescan data frees */
1429         __be16                  *tagp;          /* data entry tag pointer */
1430         xfs_trans_t             *tp;            /* transaction pointer */
1431
1432         dp = args->dp;
1433         mp = dp->i_mount;
1434         tp = args->trans;
1435         length = xfs_dir2_data_entsize(args->namelen);
1436         /*
1437          * If we came in with a freespace block that means that lookup
1438          * found an entry with our hash value.  This is the freespace
1439          * block for that data entry.
1440          */
1441         if (fblk) {
1442                 fbp = fblk->bp;
1443                 /*
1444                  * Remember initial freespace block number.
1445                  */
1446                 ifbno = fblk->blkno;
1447                 free = fbp->data;
1448                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1449                 findex = fblk->index;
1450                 /*
1451                  * This means the free entry showed that the data block had
1452                  * space for our entry, so we remembered it.
1453                  * Use that data block.
1454                  */
1455                 if (findex >= 0) {
1456                         ASSERT(findex < be32_to_cpu(free->hdr.nvalid));
1457                         ASSERT(be16_to_cpu(free->bests[findex]) != NULLDATAOFF);
1458                         ASSERT(be16_to_cpu(free->bests[findex]) >= length);
1459                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1460                 }
1461                 /*
1462                  * The data block looked at didn't have enough room.
1463                  * We'll start at the beginning of the freespace entries.
1464                  */
1465                 else {
1466                         dbno = -1;
1467                         findex = 0;
1468                 }
1469         }
1470         /*
1471          * Didn't come in with a freespace block, so don't have a data block.
1472          */
1473         else {
1474                 ifbno = dbno = -1;
1475                 fbp = NULL;
1476                 findex = 0;
1477         }
1478         /*
1479          * If we don't have a data block yet, we're going to scan the
1480          * freespace blocks looking for one.  Figure out what the
1481          * highest freespace block number is.
1482          */
1483         if (dbno == -1) {
1484                 xfs_fileoff_t   fo;             /* freespace block number */
1485
1486                 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1487                         return error;
1488                 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1489                 fbno = ifbno;
1490         }
1491         /*
1492          * While we haven't identified a data block, search the freeblock
1493          * data for a good data block.  If we find a null freeblock entry,
1494          * indicating a hole in the data blocks, remember that.
1495          */
1496         while (dbno == -1) {
1497                 /*
1498                  * If we don't have a freeblock in hand, get the next one.
1499                  */
1500                 if (fbp == NULL) {
1501                         /*
1502                          * Happens the first time through unless lookup gave
1503                          * us a freespace block to start with.
1504                          */
1505                         if (++fbno == 0)
1506                                 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1507                         /*
1508                          * If it's ifbno we already looked at it.
1509                          */
1510                         if (fbno == ifbno)
1511                                 fbno++;
1512                         /*
1513                          * If it's off the end we're done.
1514                          */
1515                         if (fbno >= lastfbno)
1516                                 break;
1517                         /*
1518                          * Read the block.  There can be holes in the
1519                          * freespace blocks, so this might not succeed.
1520                          * This should be really rare, so there's no reason
1521                          * to avoid it.
1522                          */
1523                         if ((error = xfs_da_read_buf(tp, dp,
1524                                         xfs_dir2_db_to_da(mp, fbno), -2, &fbp,
1525                                         XFS_DATA_FORK))) {
1526                                 return error;
1527                         }
1528                         if (unlikely(fbp == NULL)) {
1529                                 continue;
1530                         }
1531                         free = fbp->data;
1532                         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1533                         findex = 0;
1534                 }
1535                 /*
1536                  * Look at the current free entry.  Is it good enough?
1537                  */
1538                 if (be16_to_cpu(free->bests[findex]) != NULLDATAOFF &&
1539                     be16_to_cpu(free->bests[findex]) >= length)
1540                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1541                 else {
1542                         /*
1543                          * Are we done with the freeblock?
1544                          */
1545                         if (++findex == be32_to_cpu(free->hdr.nvalid)) {
1546                                 /*
1547                                  * Drop the block.
1548                                  */
1549                                 xfs_da_brelse(tp, fbp);
1550                                 fbp = NULL;
1551                                 if (fblk && fblk->bp)
1552                                         fblk->bp = NULL;
1553                         }
1554                 }
1555         }
1556         /*
1557          * If we don't have a data block, we need to allocate one and make
1558          * the freespace entries refer to it.
1559          */
1560         if (unlikely(dbno == -1)) {
1561                 /*
1562                  * Not allowed to allocate, return failure.
1563                  */
1564                 if (args->justcheck || args->total == 0) {
1565                         /*
1566                          * Drop the freespace buffer unless it came from our
1567                          * caller.
1568                          */
1569                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1570                                 xfs_da_buf_done(fbp);
1571                         return XFS_ERROR(ENOSPC);
1572                 }
1573                 /*
1574                  * Allocate and initialize the new data block.
1575                  */
1576                 if (unlikely((error = xfs_dir2_grow_inode(args,
1577                                                          XFS_DIR2_DATA_SPACE,
1578                                                          &dbno)) ||
1579                     (error = xfs_dir2_data_init(args, dbno, &dbp)))) {
1580                         /*
1581                          * Drop the freespace buffer unless it came from our
1582                          * caller.
1583                          */
1584                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1585                                 xfs_da_buf_done(fbp);
1586                         return error;
1587                 }
1588                 /*
1589                  * If (somehow) we have a freespace block, get rid of it.
1590                  */
1591                 if (fbp)
1592                         xfs_da_brelse(tp, fbp);
1593                 if (fblk && fblk->bp)
1594                         fblk->bp = NULL;
1595
1596                 /*
1597                  * Get the freespace block corresponding to the data block
1598                  * that was just allocated.
1599                  */
1600                 fbno = xfs_dir2_db_to_fdb(mp, dbno);
1601                 if (unlikely(error = xfs_da_read_buf(tp, dp,
1602                                 xfs_dir2_db_to_da(mp, fbno), -2, &fbp,
1603                                 XFS_DATA_FORK))) {
1604                         xfs_da_buf_done(dbp);
1605                         return error;
1606                 }
1607                 /*
1608                  * If there wasn't a freespace block, the read will
1609                  * return a NULL fbp.  Allocate and initialize a new one.
1610                  */
1611                 if( fbp == NULL ) {
1612                         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1613                                                         &fbno))) {
1614                                 return error;
1615                         }
1616
1617                         if (unlikely(xfs_dir2_db_to_fdb(mp, dbno) != fbno)) {
1618                                 cmn_err(CE_ALERT,
1619                                         "xfs_dir2_node_addname_int: dir ino "
1620                                         "%llu needed freesp block %lld for\n"
1621                                         "  data block %lld, got %lld\n"
1622                                         "  ifbno %llu lastfbno %d\n",
1623                                         (unsigned long long)dp->i_ino,
1624                                         (long long)xfs_dir2_db_to_fdb(mp, dbno),
1625                                         (long long)dbno, (long long)fbno,
1626                                         (unsigned long long)ifbno, lastfbno);
1627                                 if (fblk) {
1628                                         cmn_err(CE_ALERT,
1629                                                 " fblk 0x%p blkno %llu "
1630                                                 "index %d magic 0x%x\n",
1631                                                 fblk,
1632                                                 (unsigned long long)fblk->blkno,
1633                                                 fblk->index,
1634                                                 fblk->magic);
1635                                 } else {
1636                                         cmn_err(CE_ALERT,
1637                                                 " ... fblk is NULL\n");
1638                                 }
1639                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1640                                                  XFS_ERRLEVEL_LOW, mp);
1641                                 return XFS_ERROR(EFSCORRUPTED);
1642                         }
1643
1644                         /*
1645                          * Get a buffer for the new block.
1646                          */
1647                         if ((error = xfs_da_get_buf(tp, dp,
1648                                                    xfs_dir2_db_to_da(mp, fbno),
1649                                                    -1, &fbp, XFS_DATA_FORK))) {
1650                                 return error;
1651                         }
1652                         ASSERT(fbp != NULL);
1653
1654                         /*
1655                          * Initialize the new block to be empty, and remember
1656                          * its first slot as our empty slot.
1657                          */
1658                         free = fbp->data;
1659                         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
1660                         free->hdr.firstdb = cpu_to_be32(
1661                                 (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1662                                 XFS_DIR2_MAX_FREE_BESTS(mp));
1663                         free->hdr.nvalid = 0;
1664                         free->hdr.nused = 0;
1665                 } else {
1666                         free = fbp->data;
1667                         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1668                 }
1669
1670                 /*
1671                  * Set the freespace block index from the data block number.
1672                  */
1673                 findex = xfs_dir2_db_to_fdindex(mp, dbno);
1674                 /*
1675                  * If it's after the end of the current entries in the
1676                  * freespace block, extend that table.
1677                  */
1678                 if (findex >= be32_to_cpu(free->hdr.nvalid)) {
1679                         ASSERT(findex < XFS_DIR2_MAX_FREE_BESTS(mp));
1680                         free->hdr.nvalid = cpu_to_be32(findex + 1);
1681                         /*
1682                          * Tag new entry so nused will go up.
1683                          */
1684                         free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1685                 }
1686                 /*
1687                  * If this entry was for an empty data block
1688                  * (this should always be true) then update the header.
1689                  */
1690                 if (be16_to_cpu(free->bests[findex]) == NULLDATAOFF) {
1691                         be32_add_cpu(&free->hdr.nused, 1);
1692                         xfs_dir2_free_log_header(tp, fbp);
1693                 }
1694                 /*
1695                  * Update the real value in the table.
1696                  * We haven't allocated the data entry yet so this will
1697                  * change again.
1698                  */
1699                 data = dbp->data;
1700                 free->bests[findex] = data->hdr.bestfree[0].length;
1701                 logfree = 1;
1702         }
1703         /*
1704          * We had a data block so we don't have to make a new one.
1705          */
1706         else {
1707                 /*
1708                  * If just checking, we succeeded.
1709                  */
1710                 if (args->justcheck) {
1711                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1712                                 xfs_da_buf_done(fbp);
1713                         return 0;
1714                 }
1715                 /*
1716                  * Read the data block in.
1717                  */
1718                 if (unlikely(
1719                     error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1720                                 -1, &dbp, XFS_DATA_FORK))) {
1721                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1722                                 xfs_da_buf_done(fbp);
1723                         return error;
1724                 }
1725                 data = dbp->data;
1726                 logfree = 0;
1727         }
1728         ASSERT(be16_to_cpu(data->hdr.bestfree[0].length) >= length);
1729         /*
1730          * Point to the existing unused space.
1731          */
1732         dup = (xfs_dir2_data_unused_t *)
1733               ((char *)data + be16_to_cpu(data->hdr.bestfree[0].offset));
1734         needscan = needlog = 0;
1735         /*
1736          * Mark the first part of the unused space, inuse for us.
1737          */
1738         xfs_dir2_data_use_free(tp, dbp, dup,
1739                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)data), length,
1740                 &needlog, &needscan);
1741         /*
1742          * Fill in the new entry and log it.
1743          */
1744         dep = (xfs_dir2_data_entry_t *)dup;
1745         dep->inumber = cpu_to_be64(args->inumber);
1746         dep->namelen = args->namelen;
1747         memcpy(dep->name, args->name, dep->namelen);
1748         tagp = xfs_dir2_data_entry_tag_p(dep);
1749         *tagp = cpu_to_be16((char *)dep - (char *)data);
1750         xfs_dir2_data_log_entry(tp, dbp, dep);
1751         /*
1752          * Rescan the block for bestfree if needed.
1753          */
1754         if (needscan)
1755                 xfs_dir2_data_freescan(mp, data, &needlog);
1756         /*
1757          * Log the data block header if needed.
1758          */
1759         if (needlog)
1760                 xfs_dir2_data_log_header(tp, dbp);
1761         /*
1762          * If the freespace entry is now wrong, update it.
1763          */
1764         if (be16_to_cpu(free->bests[findex]) != be16_to_cpu(data->hdr.bestfree[0].length)) {
1765                 free->bests[findex] = data->hdr.bestfree[0].length;
1766                 logfree = 1;
1767         }
1768         /*
1769          * Log the freespace entry if needed.
1770          */
1771         if (logfree)
1772                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1773         /*
1774          * If the caller didn't hand us the freespace block, drop it.
1775          */
1776         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1777                 xfs_da_buf_done(fbp);
1778         /*
1779          * Return the data block and offset in args, then drop the data block.
1780          */
1781         args->blkno = (xfs_dablk_t)dbno;
1782         args->index = be16_to_cpu(*tagp);
1783         xfs_da_buf_done(dbp);
1784         return 0;
1785 }
1786
1787 /*
1788  * Lookup an entry in a node-format directory.
1789  * All the real work happens in xfs_da_node_lookup_int.
1790  * The only real output is the inode number of the entry.
1791  */
1792 int                                             /* error */
1793 xfs_dir2_node_lookup(
1794         xfs_da_args_t   *args)                  /* operation arguments */
1795 {
1796         int             error;                  /* error return value */
1797         int             i;                      /* btree level */
1798         int             rval;                   /* operation return value */
1799         xfs_da_state_t  *state;                 /* btree cursor */
1800
1801         xfs_dir2_trace_args("node_lookup", args);
1802         /*
1803          * Allocate and initialize the btree cursor.
1804          */
1805         state = xfs_da_state_alloc();
1806         state->args = args;
1807         state->mp = args->dp->i_mount;
1808         state->blocksize = state->mp->m_dirblksize;
1809         state->node_ents = state->mp->m_dir_node_ents;
1810         /*
1811          * Fill in the path to the entry in the cursor.
1812          */
1813         error = xfs_da_node_lookup_int(state, &rval);
1814         if (error)
1815                 rval = error;
1816         /*
1817          * Release the btree blocks and leaf block.
1818          */
1819         for (i = 0; i < state->path.active; i++) {
1820                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1821                 state->path.blk[i].bp = NULL;
1822         }
1823         /*
1824          * Release the data block if we have it.
1825          */
1826         if (state->extravalid && state->extrablk.bp) {
1827                 xfs_da_brelse(args->trans, state->extrablk.bp);
1828                 state->extrablk.bp = NULL;
1829         }
1830         xfs_da_state_free(state);
1831         return rval;
1832 }
1833
1834 /*
1835  * Remove an entry from a node-format directory.
1836  */
1837 int                                             /* error */
1838 xfs_dir2_node_removename(
1839         xfs_da_args_t           *args)          /* operation arguments */
1840 {
1841         xfs_da_state_blk_t      *blk;           /* leaf block */
1842         int                     error;          /* error return value */
1843         int                     rval;           /* operation return value */
1844         xfs_da_state_t          *state;         /* btree cursor */
1845
1846         xfs_dir2_trace_args("node_removename", args);
1847         /*
1848          * Allocate and initialize the btree cursor.
1849          */
1850         state = xfs_da_state_alloc();
1851         state->args = args;
1852         state->mp = args->dp->i_mount;
1853         state->blocksize = state->mp->m_dirblksize;
1854         state->node_ents = state->mp->m_dir_node_ents;
1855         /*
1856          * Look up the entry we're deleting, set up the cursor.
1857          */
1858         error = xfs_da_node_lookup_int(state, &rval);
1859         if (error) {
1860                 rval = error;
1861         }
1862         /*
1863          * Didn't find it, upper layer screwed up.
1864          */
1865         if (rval != EEXIST) {
1866                 xfs_da_state_free(state);
1867                 return rval;
1868         }
1869         blk = &state->path.blk[state->path.active - 1];
1870         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1871         ASSERT(state->extravalid);
1872         /*
1873          * Remove the leaf and data entries.
1874          * Extrablk refers to the data block.
1875          */
1876         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
1877                 &state->extrablk, &rval);
1878         if (error) {
1879                 return error;
1880         }
1881         /*
1882          * Fix the hash values up the btree.
1883          */
1884         xfs_da_fixhashpath(state, &state->path);
1885         /*
1886          * If we need to join leaf blocks, do it.
1887          */
1888         if (rval && state->path.active > 1)
1889                 error = xfs_da_join(state);
1890         /*
1891          * If no errors so far, try conversion to leaf format.
1892          */
1893         if (!error)
1894                 error = xfs_dir2_node_to_leaf(state);
1895         xfs_da_state_free(state);
1896         return error;
1897 }
1898
1899 /*
1900  * Replace an entry's inode number in a node-format directory.
1901  */
1902 int                                             /* error */
1903 xfs_dir2_node_replace(
1904         xfs_da_args_t           *args)          /* operation arguments */
1905 {
1906         xfs_da_state_blk_t      *blk;           /* leaf block */
1907         xfs_dir2_data_t         *data;          /* data block structure */
1908         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
1909         int                     error;          /* error return value */
1910         int                     i;              /* btree level */
1911         xfs_ino_t               inum;           /* new inode number */
1912         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1913         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
1914         int                     rval;           /* internal return value */
1915         xfs_da_state_t          *state;         /* btree cursor */
1916
1917         xfs_dir2_trace_args("node_replace", args);
1918         /*
1919          * Allocate and initialize the btree cursor.
1920          */
1921         state = xfs_da_state_alloc();
1922         state->args = args;
1923         state->mp = args->dp->i_mount;
1924         state->blocksize = state->mp->m_dirblksize;
1925         state->node_ents = state->mp->m_dir_node_ents;
1926         inum = args->inumber;
1927         /*
1928          * Lookup the entry to change in the btree.
1929          */
1930         error = xfs_da_node_lookup_int(state, &rval);
1931         if (error) {
1932                 rval = error;
1933         }
1934         /*
1935          * It should be found, since the vnodeops layer has looked it up
1936          * and locked it.  But paranoia is good.
1937          */
1938         if (rval == EEXIST) {
1939                 /*
1940                  * Find the leaf entry.
1941                  */
1942                 blk = &state->path.blk[state->path.active - 1];
1943                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1944                 leaf = blk->bp->data;
1945                 lep = &leaf->ents[blk->index];
1946                 ASSERT(state->extravalid);
1947                 /*
1948                  * Point to the data entry.
1949                  */
1950                 data = state->extrablk.bp->data;
1951                 ASSERT(be32_to_cpu(data->hdr.magic) == XFS_DIR2_DATA_MAGIC);
1952                 dep = (xfs_dir2_data_entry_t *)
1953                       ((char *)data +
1954                        xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
1955                 ASSERT(inum != be64_to_cpu(dep->inumber));
1956                 /*
1957                  * Fill in the new inode number and log the entry.
1958                  */
1959                 dep->inumber = cpu_to_be64(inum);
1960                 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
1961                 rval = 0;
1962         }
1963         /*
1964          * Didn't find it, and we're holding a data block.  Drop it.
1965          */
1966         else if (state->extravalid) {
1967                 xfs_da_brelse(args->trans, state->extrablk.bp);
1968                 state->extrablk.bp = NULL;
1969         }
1970         /*
1971          * Release all the buffers in the cursor.
1972          */
1973         for (i = 0; i < state->path.active; i++) {
1974                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1975                 state->path.blk[i].bp = NULL;
1976         }
1977         xfs_da_state_free(state);
1978         return rval;
1979 }
1980
1981 /*
1982  * Trim off a trailing empty freespace block.
1983  * Return (in rvalp) 1 if we did it, 0 if not.
1984  */
1985 int                                             /* error */
1986 xfs_dir2_node_trim_free(
1987         xfs_da_args_t           *args,          /* operation arguments */
1988         xfs_fileoff_t           fo,             /* free block number */
1989         int                     *rvalp)         /* out: did something */
1990 {
1991         xfs_dabuf_t             *bp;            /* freespace buffer */
1992         xfs_inode_t             *dp;            /* incore directory inode */
1993         int                     error;          /* error return code */
1994         xfs_dir2_free_t         *free;          /* freespace structure */
1995         xfs_mount_t             *mp;            /* filesystem mount point */
1996         xfs_trans_t             *tp;            /* transaction pointer */
1997
1998         dp = args->dp;
1999         mp = dp->i_mount;
2000         tp = args->trans;
2001         /*
2002          * Read the freespace block.
2003          */
2004         if (unlikely(error = xfs_da_read_buf(tp, dp, (xfs_dablk_t)fo, -2, &bp,
2005                         XFS_DATA_FORK))) {
2006                 return error;
2007         }
2008
2009         /*
2010          * There can be holes in freespace.  If fo is a hole, there's
2011          * nothing to do.
2012          */
2013         if (bp == NULL) {
2014                 return 0;
2015         }
2016         free = bp->data;
2017         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
2018         /*
2019          * If there are used entries, there's nothing to do.
2020          */
2021         if (be32_to_cpu(free->hdr.nused) > 0) {
2022                 xfs_da_brelse(tp, bp);
2023                 *rvalp = 0;
2024                 return 0;
2025         }
2026         /*
2027          * Blow the block away.
2028          */
2029         if ((error =
2030             xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
2031                     bp))) {
2032                 /*
2033                  * Can't fail with ENOSPC since that only happens with no
2034                  * space reservation, when breaking up an extent into two
2035                  * pieces.  This is the last block of an extent.
2036                  */
2037                 ASSERT(error != ENOSPC);
2038                 xfs_da_brelse(tp, bp);
2039                 return error;
2040         }
2041         /*
2042          * Return that we succeeded.
2043          */
2044         *rvalp = 1;
2045         return 0;
2046 }