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_dir.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_dir_sf.h"
  32 #include "xfs_dir2_sf.h"
  33 #include "xfs_attr_sf.h"
  34 #include "xfs_dinode.h"
  35 #include "xfs_inode.h"
  36 #include "xfs_bmap.h"
  37 #include "xfs_dir2_data.h"
  38 #include "xfs_dir2_leaf.h"
  39 #include "xfs_dir2_block.h"
  40 #include "xfs_dir2_node.h"
  41 #include "xfs_dir2_trace.h"
  42 #include "xfs_error.h"
  43
  44 /*
  45  * Function declarations.
  46  */
  47 static void xfs_dir2_free_log_header(xfs_trans_t *tp, xfs_dabuf_t *bp);
  48 static int xfs_dir2_leafn_add(xfs_dabuf_t *bp, xfs_da_args_t *args, int index);
  49 #ifdef DEBUG
  50 static void xfs_dir2_leafn_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
  51 #else
  52 #define xfs_dir2_leafn_check(dp, bp)
  53 #endif
  54 static void xfs_dir2_leafn_moveents(xfs_da_args_t *args, xfs_dabuf_t *bp_s,
  55                                     int start_s, xfs_dabuf_t *bp_d, int start_d,
  56                                     int count);
  57 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  58                                      xfs_da_state_blk_t *blk1,
  59                                      xfs_da_state_blk_t *blk2);
  60 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, xfs_dabuf_t *bp,
  61                                  int index, xfs_da_state_blk_t *dblk,
  62                                  int *rval);
  63 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  64                                      xfs_da_state_blk_t *fblk);
  65
  66 /*
  67  * Log entries from a freespace block.
  68  */
  69 void
  70 xfs_dir2_free_log_bests(
  71         xfs_trans_t             *tp,            /* transaction pointer */
  72         xfs_dabuf_t             *bp,            /* freespace buffer */
  73         int                     first,          /* first entry to log */
  74         int                     last)           /* last entry to log */
  75 {
  76         xfs_dir2_free_t         *free;          /* freespace structure */
  77
  78         free = bp->data;
  79         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
  80         xfs_da_log_buf(tp, bp,
  81                 (uint)((char *)&free->bests[first] - (char *)free),
  82                 (uint)((char *)&free->bests[last] - (char *)free +
  83                        sizeof(free->bests[0]) - 1));
  84 }
  85
  86 /*
  87  * Log header from a freespace block.
  88  */
  89 static void
  90 xfs_dir2_free_log_header(
  91         xfs_trans_t             *tp,            /* transaction pointer */
  92         xfs_dabuf_t             *bp)            /* freespace buffer */
  93 {
  94         xfs_dir2_free_t         *free;          /* freespace structure */
  95
  96         free = bp->data;
  97         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
  98         xfs_da_log_buf(tp, bp, (uint)((char *)&free->hdr - (char *)free),
  99                 (uint)(sizeof(xfs_dir2_free_hdr_t) - 1));
 100 }
 101
 102 /*
 103  * Convert a leaf-format directory to a node-format directory.
 104  * We need to change the magic number of the leaf block, and copy
 105  * the freespace table out of the leaf block into its own block.
 106  */
 107 int                                             /* error */
 108 xfs_dir2_leaf_to_node(
 109         xfs_da_args_t           *args,          /* operation arguments */
 110         xfs_dabuf_t             *lbp)           /* leaf buffer */
 111 {
 112         xfs_inode_t             *dp;            /* incore directory inode */
 113         int                     error;          /* error return value */
 114         xfs_dabuf_t             *fbp;           /* freespace buffer */
 115         xfs_dir2_db_t           fdb;            /* freespace block number */
 116         xfs_dir2_free_t         *free;          /* freespace structure */
 117         __be16                  *from;          /* pointer to freespace entry */
 118         int                     i;              /* leaf freespace index */
 119         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 120         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 121         xfs_mount_t             *mp;            /* filesystem mount point */
 122         int                     n;              /* count of live freespc ents */
 123         xfs_dir2_data_off_t     off;            /* freespace entry value */
 124         __be16                  *to;            /* pointer to freespace entry */
 125         xfs_trans_t             *tp;            /* transaction pointer */
 126
 127         xfs_dir2_trace_args_b("leaf_to_node", args, lbp);
 128         dp = args->dp;
 129         mp = dp->i_mount;
 130         tp = args->trans;
 131         /*
 132          * Add a freespace block to the directory.
 133          */
 134         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 135                 return error;
 136         }
 137         ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
 138         /*
 139          * Get the buffer for the new freespace block.
 140          */
 141         if ((error = xfs_da_get_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, fdb), -1, &fbp,
 142                         XFS_DATA_FORK))) {
 143                 return error;
 144         }
 145         ASSERT(fbp != NULL);
 146         free = fbp->data;
 147         leaf = lbp->data;
 148         ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
 149         /*
 150          * Initialize the freespace block header.
 151          */
 152         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
 153         free->hdr.firstdb = 0;
 154         ASSERT(be32_to_cpu(ltp->bestcount) <= (uint)dp->i_d.di_size / mp->m_dirblksize);
 155         free->hdr.nvalid = ltp->bestcount;
 156         /*
 157          * Copy freespace entries from the leaf block to the new block.
 158          * Count active entries.
 159          */
 160         for (i = n = 0, from = XFS_DIR2_LEAF_BESTS_P(ltp), to = free->bests;
 161              i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 162                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 163                         n++;
 164                 *to = cpu_to_be16(off);
 165         }
 166         free->hdr.nused = cpu_to_be32(n);
 167         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 168         /*
 169          * Log everything.
 170          */
 171         xfs_dir2_leaf_log_header(tp, lbp);
 172         xfs_dir2_free_log_header(tp, fbp);
 173         xfs_dir2_free_log_bests(tp, fbp, 0, be32_to_cpu(free->hdr.nvalid) - 1);
 174         xfs_da_buf_done(fbp);
 175         xfs_dir2_leafn_check(dp, lbp);
 176         return 0;
 177 }
 178
 179 /*
 180  * Add a leaf entry to a leaf block in a node-form directory.
 181  * The other work necessary is done from the caller.
 182  */
 183 static int                                      /* error */
 184 xfs_dir2_leafn_add(
 185         xfs_dabuf_t             *bp,            /* leaf buffer */
 186         xfs_da_args_t           *args,          /* operation arguments */
 187         int                     index)          /* insertion pt for new entry */
 188 {
 189         int                     compact;        /* compacting stale leaves */
 190         xfs_inode_t             *dp;            /* incore directory inode */
 191         int                     highstale;      /* next stale entry */
 192         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 193         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 194         int                     lfloghigh;      /* high leaf entry logging */
 195         int                     lfloglow;       /* low leaf entry logging */
 196         int                     lowstale;       /* previous stale entry */
 197         xfs_mount_t             *mp;            /* filesystem mount point */
 198         xfs_trans_t             *tp;            /* transaction pointer */
 199
 200         xfs_dir2_trace_args_sb("leafn_add", args, index, bp);
 201         dp = args->dp;
 202         mp = dp->i_mount;
 203         tp = args->trans;
 204         leaf = bp->data;
 205
 206         /*
 207          * Quick check just to make sure we are not going to index
 208          * into other peoples memory
 209          */
 210         if (index < 0)
 211                 return XFS_ERROR(EFSCORRUPTED);
 212
 213         /*
 214          * If there are already the maximum number of leaf entries in
 215          * the block, if there are no stale entries it won't fit.
 216          * Caller will do a split.  If there are stale entries we'll do
 217          * a compact.
 218          */
 219
 220         if (be16_to_cpu(leaf->hdr.count) == XFS_DIR2_MAX_LEAF_ENTS(mp)) {
 221                 if (!leaf->hdr.stale)
 222                         return XFS_ERROR(ENOSPC);
 223                 compact = be16_to_cpu(leaf->hdr.stale) > 1;
 224         } else
 225                 compact = 0;
 226         ASSERT(index == 0 || be32_to_cpu(leaf->ents[index - 1].hashval) <= args->hashval);
 227         ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
 228                be32_to_cpu(leaf->ents[index].hashval) >= args->hashval);
 229
 230         if (args->justcheck)
 231                 return 0;
 232
 233         /*
 234          * Compact out all but one stale leaf entry.  Leaves behind
 235          * the entry closest to index.
 236          */
 237         if (compact) {
 238                 xfs_dir2_leaf_compact_x1(bp, &index, &lowstale, &highstale,
 239                         &lfloglow, &lfloghigh);
 240         }
 241         /*
 242          * Set impossible logging indices for this case.
 243          */
 244         else if (leaf->hdr.stale) {
 245                 lfloglow = be16_to_cpu(leaf->hdr.count);
 246                 lfloghigh = -1;
 247         }
 248         /*
 249          * No stale entries, just insert a space for the new entry.
 250          */
 251         if (!leaf->hdr.stale) {
 252                 lep = &leaf->ents[index];
 253                 if (index < be16_to_cpu(leaf->hdr.count))
 254                         memmove(lep + 1, lep,
 255                                 (be16_to_cpu(leaf->hdr.count) - index) * sizeof(*lep));
 256                 lfloglow = index;
 257                 lfloghigh = be16_to_cpu(leaf->hdr.count);
 258                 be16_add(&leaf->hdr.count, 1);
 259         }
 260         /*
 261          * There are stale entries.  We'll use one for the new entry.
 262          */
 263         else {
 264                 /*
 265                  * If we didn't do a compact then we need to figure out
 266                  * which stale entry will be used.
 267                  */
 268                 if (compact == 0) {
 269                         /*
 270                          * Find first stale entry before our insertion point.
 271                          */
 272                         for (lowstale = index - 1;
 273                              lowstale >= 0 &&
 274                                 be32_to_cpu(leaf->ents[lowstale].address) !=
 275                                 XFS_DIR2_NULL_DATAPTR;
 276                              lowstale--)
 277                                 continue;
 278                         /*
 279                          * Find next stale entry after insertion point.
 280                          * Stop looking if the answer would be worse than
 281                          * lowstale already found.
 282                          */
 283                         for (highstale = index;
 284                              highstale < be16_to_cpu(leaf->hdr.count) &&
 285                                 be32_to_cpu(leaf->ents[highstale].address) !=
 286                                 XFS_DIR2_NULL_DATAPTR &&
 287                                 (lowstale < 0 ||
 288                                  index - lowstale - 1 >= highstale - index);
 289                              highstale++)
 290                                 continue;
 291                 }
 292                 /*
 293                  * Using the low stale entry.
 294                  * Shift entries up toward the stale slot.
 295                  */
 296                 if (lowstale >= 0 &&
 297                     (highstale == be16_to_cpu(leaf->hdr.count) ||
 298                      index - lowstale - 1 < highstale - index)) {
 299                         ASSERT(be32_to_cpu(leaf->ents[lowstale].address) ==
 300                                XFS_DIR2_NULL_DATAPTR);
 301                         ASSERT(index - lowstale - 1 >= 0);
 302                         if (index - lowstale - 1 > 0)
 303                                 memmove(&leaf->ents[lowstale],
 304                                         &leaf->ents[lowstale + 1],
 305                                         (index - lowstale - 1) * sizeof(*lep));
 306                         lep = &leaf->ents[index - 1];
 307                         lfloglow = MIN(lowstale, lfloglow);
 308                         lfloghigh = MAX(index - 1, lfloghigh);
 309                 }
 310                 /*
 311                  * Using the high stale entry.
 312                  * Shift entries down toward the stale slot.
 313                  */
 314                 else {
 315                         ASSERT(be32_to_cpu(leaf->ents[highstale].address) ==
 316                                XFS_DIR2_NULL_DATAPTR);
 317                         ASSERT(highstale - index >= 0);
 318                         if (highstale - index > 0)
 319                                 memmove(&leaf->ents[index + 1],
 320                                         &leaf->ents[index],
 321                                         (highstale - index) * sizeof(*lep));
 322                         lep = &leaf->ents[index];
 323                         lfloglow = MIN(index, lfloglow);
 324                         lfloghigh = MAX(highstale, lfloghigh);
 325                 }
 326                 be16_add(&leaf->hdr.stale, -1);
 327         }
 328         /*
 329          * Insert the new entry, log everything.
 330          */
 331         lep->hashval = cpu_to_be32(args->hashval);
 332         lep->address = cpu_to_be32(XFS_DIR2_DB_OFF_TO_DATAPTR(mp,
 333                                 args->blkno, args->index));
 334         xfs_dir2_leaf_log_header(tp, bp);
 335         xfs_dir2_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
 336         xfs_dir2_leafn_check(dp, bp);
 337         return 0;
 338 }
 339
 340 #ifdef DEBUG
 341 /*
 342  * Check internal consistency of a leafn block.
 343  */
 344 void
 345 xfs_dir2_leafn_check(
 346         xfs_inode_t     *dp,                    /* incore directory inode */
 347         xfs_dabuf_t     *bp)                    /* leaf buffer */
 348 {
 349         int             i;                      /* leaf index */
 350         xfs_dir2_leaf_t *leaf;                  /* leaf structure */
 351         xfs_mount_t     *mp;                    /* filesystem mount point */
 352         int             stale;                  /* count of stale leaves */
 353
 354         leaf = bp->data;
 355         mp = dp->i_mount;
 356         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 357         ASSERT(be16_to_cpu(leaf->hdr.count) <= XFS_DIR2_MAX_LEAF_ENTS(mp));
 358         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
 359                 if (i + 1 < be16_to_cpu(leaf->hdr.count)) {
 360                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
 361                                be32_to_cpu(leaf->ents[i + 1].hashval));
 362                 }
 363                 if (be32_to_cpu(leaf->ents[i].address) == XFS_DIR2_NULL_DATAPTR)
 364                         stale++;
 365         }
 366         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
 367 }
 368 #endif  /* DEBUG */
 369
 370 /*
 371  * Return the last hash value in the leaf.
 372  * Stale entries are ok.
 373  */
 374 xfs_dahash_t                                    /* hash value */
 375 xfs_dir2_leafn_lasthash(
 376         xfs_dabuf_t     *bp,                    /* leaf buffer */
 377         int             *count)                 /* count of entries in leaf */
 378 {
 379         xfs_dir2_leaf_t *leaf;                  /* leaf structure */
 380
 381         leaf = bp->data;
 382         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 383         if (count)
 384                 *count = be16_to_cpu(leaf->hdr.count);
 385         if (!leaf->hdr.count)
 386                 return 0;
 387         return be32_to_cpu(leaf->ents[be16_to_cpu(leaf->hdr.count) - 1].hashval);
 388 }
 389
 390 /*
 391  * Look up a leaf entry in a node-format leaf block.
 392  * If this is an addname then the extrablk in state is a freespace block,
 393  * otherwise it's a data block.
 394  */
 395 int
 396 xfs_dir2_leafn_lookup_int(
 397         xfs_dabuf_t             *bp,            /* leaf buffer */
 398         xfs_da_args_t           *args,          /* operation arguments */
 399         int                     *indexp,        /* out: leaf entry index */
 400         xfs_da_state_t          *state)         /* state to fill in */
 401 {
 402         xfs_dabuf_t             *curbp;         /* current data/free buffer */
 403         xfs_dir2_db_t           curdb;          /* current data block number */
 404         xfs_dir2_db_t           curfdb;         /* current free block number */
 405         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 406         xfs_inode_t             *dp;            /* incore directory inode */
 407         int                     error;          /* error return value */
 408         int                     fi;             /* free entry index */
 409         xfs_dir2_free_t         *free=NULL;     /* free block structure */
 410         int                     index;          /* leaf entry index */
 411         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 412         int                     length=0;       /* length of new data entry */
 413         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 414         xfs_mount_t             *mp;            /* filesystem mount point */
 415         xfs_dir2_db_t           newdb;          /* new data block number */
 416         xfs_dir2_db_t           newfdb;         /* new free block number */
 417         xfs_trans_t             *tp;            /* transaction pointer */
 418
 419         dp = args->dp;
 420         tp = args->trans;
 421         mp = dp->i_mount;
 422         leaf = bp->data;
 423         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 424 #ifdef __KERNEL__
 425         ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
 426 #endif
 427         xfs_dir2_leafn_check(dp, bp);
 428         /*
 429          * Look up the hash value in the leaf entries.
 430          */
 431         index = xfs_dir2_leaf_search_hash(args, bp);
 432         /*
 433          * Do we have a buffer coming in?
 434          */
 435         if (state->extravalid)
 436                 curbp = state->extrablk.bp;
 437         else
 438                 curbp = NULL;
 439         /*
 440          * For addname, it's a free block buffer, get the block number.
 441          */
 442         if (args->addname) {
 443                 curfdb = curbp ? state->extrablk.blkno : -1;
 444                 curdb = -1;
 445                 length = XFS_DIR2_DATA_ENTSIZE(args->namelen);
 446                 if ((free = (curbp ? curbp->data : NULL)))
 447                         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
 448         }
 449         /*
 450          * For others, it's a data block buffer, get the block number.
 451          */
 452         else {
 453                 curfdb = -1;
 454                 curdb = curbp ? state->extrablk.blkno : -1;
 455         }
 456         /*
 457          * Loop over leaf entries with the right hash value.
 458          */
 459         for (lep = &leaf->ents[index];
 460              index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
 461              lep++, index++) {
 462                 /*
 463                  * Skip stale leaf entries.
 464                  */
 465                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 466                         continue;
 467                 /*
 468                  * Pull the data block number from the entry.
 469                  */
 470                 newdb = XFS_DIR2_DATAPTR_TO_DB(mp, be32_to_cpu(lep->address));
 471                 /*
 472                  * For addname, we're looking for a place to put the new entry.
 473                  * We want to use a data block with an entry of equal
 474                  * hash value to ours if there is one with room.
 475                  */
 476                 if (args->addname) {
 477                         /*
 478                          * If this block isn't the data block we already have
 479                          * in hand, take a look at it.
 480                          */
 481                         if (newdb != curdb) {
 482                                 curdb = newdb;
 483                                 /*
 484                                  * Convert the data block to the free block
 485                                  * holding its freespace information.
 486                                  */
 487                                 newfdb = XFS_DIR2_DB_TO_FDB(mp, newdb);
 488                                 /*
 489                                  * If it's not the one we have in hand,
 490                                  * read it in.
 491                                  */
 492                                 if (newfdb != curfdb) {
 493                                         /*
 494                                          * If we had one before, drop it.
 495                                          */
 496                                         if (curbp)
 497                                                 xfs_da_brelse(tp, curbp);
 498                                         /*
 499                                          * Read the free block.
 500                                          */
 501                                         if ((error = xfs_da_read_buf(tp, dp,
 502                                                         XFS_DIR2_DB_TO_DA(mp,
 503                                                                 newfdb),
 504                                                         -1, &curbp,
 505                                                         XFS_DATA_FORK))) {
 506                                                 return error;
 507                                         }
 508                                         free = curbp->data;
 509                                         ASSERT(be32_to_cpu(free->hdr.magic) ==
 510                                                XFS_DIR2_FREE_MAGIC);
 511                                         ASSERT((be32_to_cpu(free->hdr.firstdb) %
 512                                                 XFS_DIR2_MAX_FREE_BESTS(mp)) ==
 513                                                0);
 514                                         ASSERT(be32_to_cpu(free->hdr.firstdb) <= curdb);
 515                                         ASSERT(curdb <
 516                                                be32_to_cpu(free->hdr.firstdb) +
 517                                                be32_to_cpu(free->hdr.nvalid));
 518                                 }
 519                                 /*
 520                                  * Get the index for our entry.
 521                                  */
 522                                 fi = XFS_DIR2_DB_TO_FDINDEX(mp, curdb);
 523                                 /*
 524                                  * If it has room, return it.
 525                                  */
 526                                 if (unlikely(be16_to_cpu(free->bests[fi]) == NULLDATAOFF)) {
 527                                         XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 528                                                          XFS_ERRLEVEL_LOW, mp);
 529                                         if (curfdb != newfdb)
 530                                                 xfs_da_brelse(tp, curbp);
 531                                         return XFS_ERROR(EFSCORRUPTED);
 532                                 }
 533                                 curfdb = newfdb;
 534                                 if (be16_to_cpu(free->bests[fi]) >= length) {
 535                                         *indexp = index;
 536                                         state->extravalid = 1;
 537                                         state->extrablk.bp = curbp;
 538                                         state->extrablk.blkno = curfdb;
 539                                         state->extrablk.index = fi;
 540                                         state->extrablk.magic =
 541                                                 XFS_DIR2_FREE_MAGIC;
 542                                         ASSERT(args->oknoent);
 543                                         return XFS_ERROR(ENOENT);
 544                                 }
 545                         }
 546                 }
 547                 /*
 548                  * Not adding a new entry, so we really want to find
 549                  * the name given to us.
 550                  */
 551                 else {
 552                         /*
 553                          * If it's a different data block, go get it.
 554                          */
 555                         if (newdb != curdb) {
 556                                 /*
 557                                  * If we had a block before, drop it.
 558                                  */
 559                                 if (curbp)
 560                                         xfs_da_brelse(tp, curbp);
 561                                 /*
 562                                  * Read the data block.
 563                                  */
 564                                 if ((error =
 565                                     xfs_da_read_buf(tp, dp,
 566                                             XFS_DIR2_DB_TO_DA(mp, newdb), -1,
 567                                             &curbp, XFS_DATA_FORK))) {
 568                                         return error;
 569                                 }
 570                                 xfs_dir2_data_check(dp, curbp);
 571                                 curdb = newdb;
 572                         }
 573                         /*
 574                          * Point to the data entry.
 575                          */
 576                         dep = (xfs_dir2_data_entry_t *)
 577                               ((char *)curbp->data +
 578                                XFS_DIR2_DATAPTR_TO_OFF(mp, be32_to_cpu(lep->address)));
 579                         /*
 580                          * Compare the entry, return it if it matches.
 581                          */
 582                         if (dep->namelen == args->namelen &&
 583                             dep->name[0] == args->name[0] &&
 584                             memcmp(dep->name, args->name, args->namelen) == 0) {
 585                                 args->inumber = be64_to_cpu(dep->inumber);
 586                                 *indexp = index;
 587                                 state->extravalid = 1;
 588                                 state->extrablk.bp = curbp;
 589                                 state->extrablk.blkno = curdb;
 590                                 state->extrablk.index =
 591                                         (int)((char *)dep -
 592                                               (char *)curbp->data);
 593                                 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 594                                 return XFS_ERROR(EEXIST);
 595                         }
 596                 }
 597         }
 598         /*
 599          * Didn't find a match.
 600          * If we are holding a buffer, give it back in case our caller
 601          * finds it useful.
 602          */
 603         if ((state->extravalid = (curbp != NULL))) {
 604                 state->extrablk.bp = curbp;
 605                 state->extrablk.index = -1;
 606                 /*
 607                  * For addname, giving back a free block.
 608                  */
 609                 if (args->addname) {
 610                         state->extrablk.blkno = curfdb;
 611                         state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 612                 }
 613                 /*
 614                  * For other callers, giving back a data block.
 615                  */
 616                 else {
 617                         state->extrablk.blkno = curdb;
 618                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 619                 }
 620         }
 621         /*
 622          * Return the final index, that will be the insertion point.
 623          */
 624         *indexp = index;
 625         ASSERT(index == be16_to_cpu(leaf->hdr.count) || args->oknoent);
 626         return XFS_ERROR(ENOENT);
 627 }
 628
 629 /*
 630  * Move count leaf entries from source to destination leaf.
 631  * Log entries and headers.  Stale entries are preserved.
 632  */
 633 static void
 634 xfs_dir2_leafn_moveents(
 635         xfs_da_args_t   *args,                  /* operation arguments */
 636         xfs_dabuf_t     *bp_s,                  /* source leaf buffer */
 637         int             start_s,                /* source leaf index */
 638         xfs_dabuf_t     *bp_d,                  /* destination leaf buffer */
 639         int             start_d,                /* destination leaf index */
 640         int             count)                  /* count of leaves to copy */
 641 {
 642         xfs_dir2_leaf_t *leaf_d;                /* destination leaf structure */
 643         xfs_dir2_leaf_t *leaf_s;                /* source leaf structure */
 644         int             stale;                  /* count stale leaves copied */
 645         xfs_trans_t     *tp;                    /* transaction pointer */
 646
 647         xfs_dir2_trace_args_bibii("leafn_moveents", args, bp_s, start_s, bp_d,
 648                 start_d, count);
 649         /*
 650          * Silently return if nothing to do.
 651          */
 652         if (count == 0) {
 653                 return;
 654         }
 655         tp = args->trans;
 656         leaf_s = bp_s->data;
 657         leaf_d = bp_d->data;
 658         /*
 659          * If the destination index is not the end of the current
 660          * destination leaf entries, open up a hole in the destination
 661          * to hold the new entries.
 662          */
 663         if (start_d < be16_to_cpu(leaf_d->hdr.count)) {
 664                 memmove(&leaf_d->ents[start_d + count], &leaf_d->ents[start_d],
 665                         (be16_to_cpu(leaf_d->hdr.count) - start_d) *
 666                         sizeof(xfs_dir2_leaf_entry_t));
 667                 xfs_dir2_leaf_log_ents(tp, bp_d, start_d + count,
 668                         count + be16_to_cpu(leaf_d->hdr.count) - 1);
 669         }
 670         /*
 671          * If the source has stale leaves, count the ones in the copy range
 672          * so we can update the header correctly.
 673          */
 674         if (leaf_s->hdr.stale) {
 675                 int     i;                      /* temp leaf index */
 676
 677                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 678                         if (be32_to_cpu(leaf_s->ents[i].address) == XFS_DIR2_NULL_DATAPTR)
 679                                 stale++;
 680                 }
 681         } else
 682                 stale = 0;
 683         /*
 684          * Copy the leaf entries from source to destination.
 685          */
 686         memcpy(&leaf_d->ents[start_d], &leaf_s->ents[start_s],
 687                 count * sizeof(xfs_dir2_leaf_entry_t));
 688         xfs_dir2_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
 689         /*
 690          * If there are source entries after the ones we copied,
 691          * delete the ones we copied by sliding the next ones down.
 692          */
 693         if (start_s + count < be16_to_cpu(leaf_s->hdr.count)) {
 694                 memmove(&leaf_s->ents[start_s], &leaf_s->ents[start_s + count],
 695                         count * sizeof(xfs_dir2_leaf_entry_t));
 696                 xfs_dir2_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
 697         }
 698         /*
 699          * Update the headers and log them.
 700          */
 701         be16_add(&leaf_s->hdr.count, -(count));
 702         be16_add(&leaf_s->hdr.stale, -(stale));
 703         be16_add(&leaf_d->hdr.count, count);
 704         be16_add(&leaf_d->hdr.stale, stale);
 705         xfs_dir2_leaf_log_header(tp, bp_s);
 706         xfs_dir2_leaf_log_header(tp, bp_d);
 707         xfs_dir2_leafn_check(args->dp, bp_s);
 708         xfs_dir2_leafn_check(args->dp, bp_d);
 709 }
 710
 711 /*
 712  * Determine the sort order of two leaf blocks.
 713  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 714  */
 715 int                                             /* sort order */
 716 xfs_dir2_leafn_order(
 717         xfs_dabuf_t     *leaf1_bp,              /* leaf1 buffer */
 718         xfs_dabuf_t     *leaf2_bp)              /* leaf2 buffer */
 719 {
 720         xfs_dir2_leaf_t *leaf1;                 /* leaf1 structure */
 721         xfs_dir2_leaf_t *leaf2;                 /* leaf2 structure */
 722
 723         leaf1 = leaf1_bp->data;
 724         leaf2 = leaf2_bp->data;
 725         ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 726         ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 727         if (be16_to_cpu(leaf1->hdr.count) > 0 &&
 728             be16_to_cpu(leaf2->hdr.count) > 0 &&
 729             (be32_to_cpu(leaf2->ents[0].hashval) < be32_to_cpu(leaf1->ents[0].hashval) ||
 730              be32_to_cpu(leaf2->ents[be16_to_cpu(leaf2->hdr.count) - 1].hashval) <
 731              be32_to_cpu(leaf1->ents[be16_to_cpu(leaf1->hdr.count) - 1].hashval)))
 732                 return 1;
 733         return 0;
 734 }
 735
 736 /*
 737  * Rebalance leaf entries between two leaf blocks.
 738  * This is actually only called when the second block is new,
 739  * though the code deals with the general case.
 740  * A new entry will be inserted in one of the blocks, and that
 741  * entry is taken into account when balancing.
 742  */
 743 static void
 744 xfs_dir2_leafn_rebalance(
 745         xfs_da_state_t          *state,         /* btree cursor */
 746         xfs_da_state_blk_t      *blk1,          /* first btree block */
 747         xfs_da_state_blk_t      *blk2)          /* second btree block */
 748 {
 749         xfs_da_args_t           *args;          /* operation arguments */
 750         int                     count;          /* count (& direction) leaves */
 751         int                     isleft;         /* new goes in left leaf */
 752         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 753         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 754         int                     mid;            /* midpoint leaf index */
 755 #ifdef DEBUG
 756         int                     oldstale;       /* old count of stale leaves */
 757 #endif
 758         int                     oldsum;         /* old total leaf count */
 759         int                     swap;           /* swapped leaf blocks */
 760
 761         args = state->args;
 762         /*
 763          * If the block order is wrong, swap the arguments.
 764          */
 765         if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
 766                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
 767
 768                 tmp = blk1;
 769                 blk1 = blk2;
 770                 blk2 = tmp;
 771         }
 772         leaf1 = blk1->bp->data;
 773         leaf2 = blk2->bp->data;
 774         oldsum = be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count);
 775 #ifdef DEBUG
 776         oldstale = be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale);
 777 #endif
 778         mid = oldsum >> 1;
 779         /*
 780          * If the old leaf count was odd then the new one will be even,
 781          * so we need to divide the new count evenly.
 782          */
 783         if (oldsum & 1) {
 784                 xfs_dahash_t    midhash;        /* middle entry hash value */
 785
 786                 if (mid >= be16_to_cpu(leaf1->hdr.count))
 787                         midhash = be32_to_cpu(leaf2->ents[mid - be16_to_cpu(leaf1->hdr.count)].hashval);
 788                 else
 789                         midhash = be32_to_cpu(leaf1->ents[mid].hashval);
 790                 isleft = args->hashval <= midhash;
 791         }
 792         /*
 793          * If the old count is even then the new count is odd, so there's
 794          * no preferred side for the new entry.
 795          * Pick the left one.
 796          */
 797         else
 798                 isleft = 1;
 799         /*
 800          * Calculate moved entry count.  Positive means left-to-right,
 801          * negative means right-to-left.  Then move the entries.
 802          */
 803         count = be16_to_cpu(leaf1->hdr.count) - mid + (isleft == 0);
 804         if (count > 0)
 805                 xfs_dir2_leafn_moveents(args, blk1->bp,
 806                         be16_to_cpu(leaf1->hdr.count) - count, blk2->bp, 0, count);
 807         else if (count < 0)
 808                 xfs_dir2_leafn_moveents(args, blk2->bp, 0, blk1->bp,
 809                         be16_to_cpu(leaf1->hdr.count), count);
 810         ASSERT(be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count) == oldsum);
 811         ASSERT(be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale) == oldstale);
 812         /*
 813          * Mark whether we're inserting into the old or new leaf.
 814          */
 815         if (be16_to_cpu(leaf1->hdr.count) < be16_to_cpu(leaf2->hdr.count))
 816                 state->inleaf = swap;
 817         else if (be16_to_cpu(leaf1->hdr.count) > be16_to_cpu(leaf2->hdr.count))
 818                 state->inleaf = !swap;
 819         else
 820                 state->inleaf =
 821                         swap ^ (blk1->index <= be16_to_cpu(leaf1->hdr.count));
 822         /*
 823          * Adjust the expected index for insertion.
 824          */
 825         if (!state->inleaf)
 826                 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
 827
 828         /*
 829          * Finally sanity check just to make sure we are not returning a negative index
 830          */
 831         if(blk2->index < 0) {
 832                 state->inleaf = 1;
 833                 blk2->index = 0;
 834                 cmn_err(CE_ALERT,
 835                         "xfs_dir2_leafn_rebalance: picked the wrong leaf? reverting original leaf: "
 836                         "blk1->index %d\n",
 837                         blk1->index);
 838         }
 839 }
 840
 841 /*
 842  * Remove an entry from a node directory.
 843  * This removes the leaf entry and the data entry,
 844  * and updates the free block if necessary.
 845  */
 846 static int                                      /* error */
 847 xfs_dir2_leafn_remove(
 848         xfs_da_args_t           *args,          /* operation arguments */
 849         xfs_dabuf_t             *bp,            /* leaf buffer */
 850         int                     index,          /* leaf entry index */
 851         xfs_da_state_blk_t      *dblk,          /* data block */
 852         int                     *rval)          /* resulting block needs join */
 853 {
 854         xfs_dir2_data_t         *data;          /* data block structure */
 855         xfs_dir2_db_t           db;             /* data block number */
 856         xfs_dabuf_t             *dbp;           /* data block buffer */
 857         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 858         xfs_inode_t             *dp;            /* incore directory inode */
 859         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 860         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 861         int                     longest;        /* longest data free entry */
 862         int                     off;            /* data block entry offset */
 863         xfs_mount_t             *mp;            /* filesystem mount point */
 864         int                     needlog;        /* need to log data header */
 865         int                     needscan;       /* need to rescan data frees */
 866         xfs_trans_t             *tp;            /* transaction pointer */
 867
 868         xfs_dir2_trace_args_sb("leafn_remove", args, index, bp);
 869         dp = args->dp;
 870         tp = args->trans;
 871         mp = dp->i_mount;
 872         leaf = bp->data;
 873         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 874         /*
 875          * Point to the entry we're removing.
 876          */
 877         lep = &leaf->ents[index];
 878         /*
 879          * Extract the data block and offset from the entry.
 880          */
 881         db = XFS_DIR2_DATAPTR_TO_DB(mp, be32_to_cpu(lep->address));
 882         ASSERT(dblk->blkno == db);
 883         off = XFS_DIR2_DATAPTR_TO_OFF(mp, be32_to_cpu(lep->address));
 884         ASSERT(dblk->index == off);
 885         /*
 886          * Kill the leaf entry by marking it stale.
 887          * Log the leaf block changes.
 888          */
 889         be16_add(&leaf->hdr.stale, 1);
 890         xfs_dir2_leaf_log_header(tp, bp);
 891         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 892         xfs_dir2_leaf_log_ents(tp, bp, index, index);
 893         /*
 894          * Make the data entry free.  Keep track of the longest freespace
 895          * in the data block in case it changes.
 896          */
 897         dbp = dblk->bp;
 898         data = dbp->data;
 899         dep = (xfs_dir2_data_entry_t *)((char *)data + off);
 900         longest = be16_to_cpu(data->hdr.bestfree[0].length);
 901         needlog = needscan = 0;
 902         xfs_dir2_data_make_free(tp, dbp, off,
 903                 XFS_DIR2_DATA_ENTSIZE(dep->namelen), &needlog, &needscan);
 904         /*
 905          * Rescan the data block freespaces for bestfree.
 906          * Log the data block header if needed.
 907          */
 908         if (needscan)
 909                 xfs_dir2_data_freescan(mp, data, &needlog, NULL);
 910         if (needlog)
 911                 xfs_dir2_data_log_header(tp, dbp);
 912         xfs_dir2_data_check(dp, dbp);
 913         /*
 914          * If the longest data block freespace changes, need to update
 915          * the corresponding freeblock entry.
 916          */
 917         if (longest < be16_to_cpu(data->hdr.bestfree[0].length)) {
 918                 int             error;          /* error return value */
 919                 xfs_dabuf_t     *fbp;           /* freeblock buffer */
 920                 xfs_dir2_db_t   fdb;            /* freeblock block number */
 921                 int             findex;         /* index in freeblock entries */
 922                 xfs_dir2_free_t *free;          /* freeblock structure */
 923                 int             logfree;        /* need to log free entry */
 924
 925                 /*
 926                  * Convert the data block number to a free block,
 927                  * read in the free block.
 928                  */
 929                 fdb = XFS_DIR2_DB_TO_FDB(mp, db);
 930                 if ((error = xfs_da_read_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, fdb),
 931                                 -1, &fbp, XFS_DATA_FORK))) {
 932                         return error;
 933                 }
 934                 free = fbp->data;
 935                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
 936                 ASSERT(be32_to_cpu(free->hdr.firstdb) ==
 937                        XFS_DIR2_MAX_FREE_BESTS(mp) *
 938                        (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
 939                 /*
 940                  * Calculate which entry we need to fix.
 941                  */
 942                 findex = XFS_DIR2_DB_TO_FDINDEX(mp, db);
 943                 longest = be16_to_cpu(data->hdr.bestfree[0].length);
 944                 /*
 945                  * If the data block is now empty we can get rid of it
 946                  * (usually).
 947                  */
 948                 if (longest == mp->m_dirblksize - (uint)sizeof(data->hdr)) {
 949                         /*
 950                          * Try to punch out the data block.
 951                          */
 952                         error = xfs_dir2_shrink_inode(args, db, dbp);
 953                         if (error == 0) {
 954                                 dblk->bp = NULL;
 955                                 data = NULL;
 956                         }
 957                         /*
 958                          * We can get ENOSPC if there's no space reservation.
 959                          * In this case just drop the buffer and some one else
 960                          * will eventually get rid of the empty block.
 961                          */
 962                         else if (error == ENOSPC && args->total == 0)
 963                                 xfs_da_buf_done(dbp);
 964                         else
 965                                 return error;
 966                 }
 967                 /*
 968                  * If we got rid of the data block, we can eliminate that entry
 969                  * in the free block.
 970                  */
 971                 if (data == NULL) {
 972                         /*
 973                          * One less used entry in the free table.
 974                          */
 975                         be32_add(&free->hdr.nused, -1);
 976                         xfs_dir2_free_log_header(tp, fbp);
 977                         /*
 978                          * If this was the last entry in the table, we can
 979                          * trim the table size back.  There might be other
 980                          * entries at the end referring to non-existent
 981                          * data blocks, get those too.
 982                          */
 983                         if (findex == be32_to_cpu(free->hdr.nvalid) - 1) {
 984                                 int     i;              /* free entry index */
 985
 986                                 for (i = findex - 1;
 987                                      i >= 0 && be16_to_cpu(free->bests[i]) == NULLDATAOFF;
 988                                      i--)
 989                                         continue;
 990                                 free->hdr.nvalid = cpu_to_be32(i + 1);
 991                                 logfree = 0;
 992                         }
 993                         /*
 994                          * Not the last entry, just punch it out.
 995                          */
 996                         else {
 997                                 free->bests[findex] = cpu_to_be16(NULLDATAOFF);
 998                                 logfree = 1;
 999                         }
1000                         /*
1001                          * If there are no useful entries left in the block,
1002                          * get rid of the block if we can.
1003                          */
1004                         if (!free->hdr.nused) {
1005                                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1006                                 if (error == 0) {
1007                                         fbp = NULL;
1008                                         logfree = 0;
1009                                 } else if (error != ENOSPC || args->total != 0)
1010                                         return error;
1011                                 /*
1012                                  * It's possible to get ENOSPC if there is no
1013                                  * space reservation.  In this case some one
1014                                  * else will eventually get rid of this block.
1015                                  */
1016                         }
1017                 }
1018                 /*
1019                  * Data block is not empty, just set the free entry to
1020                  * the new value.
1021                  */
1022                 else {
1023                         free->bests[findex] = cpu_to_be16(longest);
1024                         logfree = 1;
1025                 }
1026                 /*
1027                  * Log the free entry that changed, unless we got rid of it.
1028                  */
1029                 if (logfree)
1030                         xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1031                 /*
1032                  * Drop the buffer if we still have it.
1033                  */
1034                 if (fbp)
1035                         xfs_da_buf_done(fbp);
1036         }
1037         xfs_dir2_leafn_check(dp, bp);
1038         /*
1039          * Return indication of whether this leaf block is emtpy enough
1040          * to justify trying to join it with a neighbor.
1041          */
1042         *rval =
1043                 ((uint)sizeof(leaf->hdr) +
1044                  (uint)sizeof(leaf->ents[0]) *
1045                  (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale))) <
1046                 mp->m_dir_magicpct;
1047         return 0;
1048 }
1049
1050 /*
1051  * Split the leaf entries in the old block into old and new blocks.
1052  */
1053 int                                             /* error */
1054 xfs_dir2_leafn_split(
1055         xfs_da_state_t          *state,         /* btree cursor */
1056         xfs_da_state_blk_t      *oldblk,        /* original block */
1057         xfs_da_state_blk_t      *newblk)        /* newly created block */
1058 {
1059         xfs_da_args_t           *args;          /* operation arguments */
1060         xfs_dablk_t             blkno;          /* new leaf block number */
1061         int                     error;          /* error return value */
1062         xfs_mount_t             *mp;            /* filesystem mount point */
1063
1064         /*
1065          * Allocate space for a new leaf node.
1066          */
1067         args = state->args;
1068         mp = args->dp->i_mount;
1069         ASSERT(args != NULL);
1070         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1071         error = xfs_da_grow_inode(args, &blkno);
1072         if (error) {
1073                 return error;
1074         }
1075         /*
1076          * Initialize the new leaf block.
1077          */
1078         error = xfs_dir2_leaf_init(args, XFS_DIR2_DA_TO_DB(mp, blkno),
1079                 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1080         if (error) {
1081                 return error;
1082         }
1083         newblk->blkno = blkno;
1084         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1085         /*
1086          * Rebalance the entries across the two leaves, link the new
1087          * block into the leaves.
1088          */
1089         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1090         error = xfs_da_blk_link(state, oldblk, newblk);
1091         if (error) {
1092                 return error;
1093         }
1094         /*
1095          * Insert the new entry in the correct block.
1096          */
1097         if (state->inleaf)
1098                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1099         else
1100                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1101         /*
1102          * Update last hashval in each block since we added the name.
1103          */
1104         oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1105         newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1106         xfs_dir2_leafn_check(args->dp, oldblk->bp);
1107         xfs_dir2_leafn_check(args->dp, newblk->bp);
1108         return error;
1109 }
1110
1111 /*
1112  * Check a leaf block and its neighbors to see if the block should be
1113  * collapsed into one or the other neighbor.  Always keep the block
1114  * with the smaller block number.
1115  * If the current block is over 50% full, don't try to join it, return 0.
1116  * If the block is empty, fill in the state structure and return 2.
1117  * If it can be collapsed, fill in the state structure and return 1.
1118  * If nothing can be done, return 0.
1119  */
1120 int                                             /* error */
1121 xfs_dir2_leafn_toosmall(
1122         xfs_da_state_t          *state,         /* btree cursor */
1123         int                     *action)        /* resulting action to take */
1124 {
1125         xfs_da_state_blk_t      *blk;           /* leaf block */
1126         xfs_dablk_t             blkno;          /* leaf block number */
1127         xfs_dabuf_t             *bp;            /* leaf buffer */
1128         int                     bytes;          /* bytes in use */
1129         int                     count;          /* leaf live entry count */
1130         int                     error;          /* error return value */
1131         int                     forward;        /* sibling block direction */
1132         int                     i;              /* sibling counter */
1133         xfs_da_blkinfo_t        *info;          /* leaf block header */
1134         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1135         int                     rval;           /* result from path_shift */
1136
1137         /*
1138          * Check for the degenerate case of the block being over 50% full.
1139          * If so, it's not worth even looking to see if we might be able
1140          * to coalesce with a sibling.
1141          */
1142         blk = &state->path.blk[state->path.active - 1];
1143         info = blk->bp->data;
1144         ASSERT(be16_to_cpu(info->magic) == XFS_DIR2_LEAFN_MAGIC);
1145         leaf = (xfs_dir2_leaf_t *)info;
1146         count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1147         bytes = (uint)sizeof(leaf->hdr) + count * (uint)sizeof(leaf->ents[0]);
1148         if (bytes > (state->blocksize >> 1)) {
1149                 /*
1150                  * Blk over 50%, don't try to join.
1151                  */
1152                 *action = 0;
1153                 return 0;
1154         }
1155         /*
1156          * Check for the degenerate case of the block being empty.
1157          * If the block is empty, we'll simply delete it, no need to
1158          * coalesce it with a sibling block.  We choose (arbitrarily)
1159          * to merge with the forward block unless it is NULL.
1160          */
1161         if (count == 0) {
1162                 /*
1163                  * Make altpath point to the block we want to keep and
1164                  * path point to the block we want to drop (this one).
1165                  */
1166                 forward = (info->forw != 0);
1167                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1168                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1169                         &rval);
1170                 if (error)
1171                         return error;
1172                 *action = rval ? 2 : 0;
1173                 return 0;
1174         }
1175         /*
1176          * Examine each sibling block to see if we can coalesce with
1177          * at least 25% free space to spare.  We need to figure out
1178          * whether to merge with the forward or the backward block.
1179          * We prefer coalescing with the lower numbered sibling so as
1180          * to shrink a directory over time.
1181          */
1182         forward = be32_to_cpu(info->forw) < be32_to_cpu(info->back);
1183         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1184                 blkno = forward ? be32_to_cpu(info->forw) : be32_to_cpu(info->back);
1185                 if (blkno == 0)
1186                         continue;
1187                 /*
1188                  * Read the sibling leaf block.
1189                  */
1190                 if ((error =
1191                     xfs_da_read_buf(state->args->trans, state->args->dp, blkno,
1192                             -1, &bp, XFS_DATA_FORK))) {
1193                         return error;
1194                 }
1195                 ASSERT(bp != NULL);
1196                 /*
1197                  * Count bytes in the two blocks combined.
1198                  */
1199                 leaf = (xfs_dir2_leaf_t *)info;
1200                 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1201                 bytes = state->blocksize - (state->blocksize >> 2);
1202                 leaf = bp->data;
1203                 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1204                 count += be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1205                 bytes -= count * (uint)sizeof(leaf->ents[0]);
1206                 /*
1207                  * Fits with at least 25% to spare.
1208                  */
1209                 if (bytes >= 0)
1210                         break;
1211                 xfs_da_brelse(state->args->trans, bp);
1212         }
1213         /*
1214          * Didn't like either block, give up.
1215          */
1216         if (i >= 2) {
1217                 *action = 0;
1218                 return 0;
1219         }
1220         /*
1221          * Done with the sibling leaf block here, drop the dabuf
1222          * so path_shift can get it.
1223          */
1224         xfs_da_buf_done(bp);
1225         /*
1226          * Make altpath point to the block we want to keep (the lower
1227          * numbered block) and path point to the block we want to drop.
1228          */
1229         memcpy(&state->altpath, &state->path, sizeof(state->path));
1230         if (blkno < blk->blkno)
1231                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1232                         &rval);
1233         else
1234                 error = xfs_da_path_shift(state, &state->path, forward, 0,
1235                         &rval);
1236         if (error) {
1237                 return error;
1238         }
1239         *action = rval ? 0 : 1;
1240         return 0;
1241 }
1242
1243 /*
1244  * Move all the leaf entries from drop_blk to save_blk.
1245  * This is done as part of a join operation.
1246  */
1247 void
1248 xfs_dir2_leafn_unbalance(
1249         xfs_da_state_t          *state,         /* cursor */
1250         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1251         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1252 {
1253         xfs_da_args_t           *args;          /* operation arguments */
1254         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1255         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1256
1257         args = state->args;
1258         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1259         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1260         drop_leaf = drop_blk->bp->data;
1261         save_leaf = save_blk->bp->data;
1262         ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1263         ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1264         /*
1265          * If there are any stale leaf entries, take this opportunity
1266          * to purge them.
1267          */
1268         if (drop_leaf->hdr.stale)
1269                 xfs_dir2_leaf_compact(args, drop_blk->bp);
1270         if (save_leaf->hdr.stale)
1271                 xfs_dir2_leaf_compact(args, save_blk->bp);
1272         /*
1273          * Move the entries from drop to the appropriate end of save.
1274          */
1275         drop_blk->hashval = be32_to_cpu(drop_leaf->ents[be16_to_cpu(drop_leaf->hdr.count) - 1].hashval);
1276         if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1277                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp, 0,
1278                         be16_to_cpu(drop_leaf->hdr.count));
1279         else
1280                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp,
1281                         be16_to_cpu(save_leaf->hdr.count), be16_to_cpu(drop_leaf->hdr.count));
1282         save_blk->hashval = be32_to_cpu(save_leaf->ents[be16_to_cpu(save_leaf->hdr.count) - 1].hashval);
1283         xfs_dir2_leafn_check(args->dp, save_blk->bp);
1284 }
1285
1286 /*
1287  * Top-level node form directory addname routine.
1288  */
1289 int                                             /* error */
1290 xfs_dir2_node_addname(
1291         xfs_da_args_t           *args)          /* operation arguments */
1292 {
1293         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1294         int                     error;          /* error return value */
1295         int                     rval;           /* sub-return value */
1296         xfs_da_state_t          *state;         /* btree cursor */
1297
1298         xfs_dir2_trace_args("node_addname", args);
1299         /*
1300          * Allocate and initialize the state (btree cursor).
1301          */
1302         state = xfs_da_state_alloc();
1303         state->args = args;
1304         state->mp = args->dp->i_mount;
1305         state->blocksize = state->mp->m_dirblksize;
1306         state->node_ents = state->mp->m_dir_node_ents;
1307         /*
1308          * Look up the name.  We're not supposed to find it, but
1309          * this gives us the insertion point.
1310          */
1311         error = xfs_da_node_lookup_int(state, &rval);
1312         if (error)
1313                 rval = error;
1314         if (rval != ENOENT) {
1315                 goto done;
1316         }
1317         /*
1318          * Add the data entry to a data block.
1319          * Extravalid is set to a freeblock found by lookup.
1320          */
1321         rval = xfs_dir2_node_addname_int(args,
1322                 state->extravalid ? &state->extrablk : NULL);
1323         if (rval) {
1324                 goto done;
1325         }
1326         blk = &state->path.blk[state->path.active - 1];
1327         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1328         /*
1329          * Add the new leaf entry.
1330          */
1331         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1332         if (rval == 0) {
1333                 /*
1334                  * It worked, fix the hash values up the btree.
1335                  */
1336                 if (!args->justcheck)
1337                         xfs_da_fixhashpath(state, &state->path);
1338         } else {
1339                 /*
1340                  * It didn't work, we need to split the leaf block.
1341                  */
1342                 if (args->total == 0) {
1343                         ASSERT(rval == ENOSPC);
1344                         goto done;
1345                 }
1346                 /*
1347                  * Split the leaf block and insert the new entry.
1348                  */
1349                 rval = xfs_da_split(state);
1350         }
1351 done:
1352         xfs_da_state_free(state);
1353         return rval;
1354 }
1355
1356 /*
1357  * Add the data entry for a node-format directory name addition.
1358  * The leaf entry is added in xfs_dir2_leafn_add.
1359  * We may enter with a freespace block that the lookup found.
1360  */
1361 static int                                      /* error */
1362 xfs_dir2_node_addname_int(
1363         xfs_da_args_t           *args,          /* operation arguments */
1364         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1365 {
1366         xfs_dir2_data_t         *data;          /* data block structure */
1367         xfs_dir2_db_t           dbno;           /* data block number */
1368         xfs_dabuf_t             *dbp;           /* data block buffer */
1369         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1370         xfs_inode_t             *dp;            /* incore directory inode */
1371         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1372         int                     error;          /* error return value */
1373         xfs_dir2_db_t           fbno;           /* freespace block number */
1374         xfs_dabuf_t             *fbp;           /* freespace buffer */
1375         int                     findex;         /* freespace entry index */
1376         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1377         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1378         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1379         int                     length;         /* length of the new entry */
1380         int                     logfree;        /* need to log free entry */
1381         xfs_mount_t             *mp;            /* filesystem mount point */
1382         int                     needlog;        /* need to log data header */
1383         int                     needscan;       /* need to rescan data frees */
1384         __be16                  *tagp;          /* data entry tag pointer */
1385         xfs_trans_t             *tp;            /* transaction pointer */
1386
1387         dp = args->dp;
1388         mp = dp->i_mount;
1389         tp = args->trans;
1390         length = XFS_DIR2_DATA_ENTSIZE(args->namelen);
1391         /*
1392          * If we came in with a freespace block that means that lookup
1393          * found an entry with our hash value.  This is the freespace
1394          * block for that data entry.
1395          */
1396         if (fblk) {
1397                 fbp = fblk->bp;
1398                 /*
1399                  * Remember initial freespace block number.
1400                  */
1401                 ifbno = fblk->blkno;
1402                 free = fbp->data;
1403                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1404                 findex = fblk->index;
1405                 /*
1406                  * This means the free entry showed that the data block had
1407                  * space for our entry, so we remembered it.
1408                  * Use that data block.
1409                  */
1410                 if (findex >= 0) {
1411                         ASSERT(findex < be32_to_cpu(free->hdr.nvalid));
1412                         ASSERT(be16_to_cpu(free->bests[findex]) != NULLDATAOFF);
1413                         ASSERT(be16_to_cpu(free->bests[findex]) >= length);
1414                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1415                 }
1416                 /*
1417                  * The data block looked at didn't have enough room.
1418                  * We'll start at the beginning of the freespace entries.
1419                  */
1420                 else {
1421                         dbno = -1;
1422                         findex = 0;
1423                 }
1424         }
1425         /*
1426          * Didn't come in with a freespace block, so don't have a data block.
1427          */
1428         else {
1429                 ifbno = dbno = -1;
1430                 fbp = NULL;
1431                 findex = 0;
1432         }
1433         /*
1434          * If we don't have a data block yet, we're going to scan the
1435          * freespace blocks looking for one.  Figure out what the
1436          * highest freespace block number is.
1437          */
1438         if (dbno == -1) {
1439                 xfs_fileoff_t   fo;             /* freespace block number */
1440
1441                 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1442                         return error;
1443                 lastfbno = XFS_DIR2_DA_TO_DB(mp, (xfs_dablk_t)fo);
1444                 fbno = ifbno;
1445         }
1446         /*
1447          * While we haven't identified a data block, search the freeblock
1448          * data for a good data block.  If we find a null freeblock entry,
1449          * indicating a hole in the data blocks, remember that.
1450          */
1451         while (dbno == -1) {
1452                 /*
1453                  * If we don't have a freeblock in hand, get the next one.
1454                  */
1455                 if (fbp == NULL) {
1456                         /*
1457                          * Happens the first time through unless lookup gave
1458                          * us a freespace block to start with.
1459                          */
1460                         if (++fbno == 0)
1461                                 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1462                         /*
1463                          * If it's ifbno we already looked at it.
1464                          */
1465                         if (fbno == ifbno)
1466                                 fbno++;
1467                         /*
1468                          * If it's off the end we're done.
1469                          */
1470                         if (fbno >= lastfbno)
1471                                 break;
1472                         /*
1473                          * Read the block.  There can be holes in the
1474                          * freespace blocks, so this might not succeed.
1475                          * This should be really rare, so there's no reason
1476                          * to avoid it.
1477                          */
1478                         if ((error = xfs_da_read_buf(tp, dp,
1479                                         XFS_DIR2_DB_TO_DA(mp, fbno), -2, &fbp,
1480                                         XFS_DATA_FORK))) {
1481                                 return error;
1482                         }
1483                         if (unlikely(fbp == NULL)) {
1484                                 continue;
1485                         }
1486                         free = fbp->data;
1487                         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1488                         findex = 0;
1489                 }
1490                 /*
1491                  * Look at the current free entry.  Is it good enough?
1492                  */
1493                 if (be16_to_cpu(free->bests[findex]) != NULLDATAOFF &&
1494                     be16_to_cpu(free->bests[findex]) >= length)
1495                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1496                 else {
1497                         /*
1498                          * Are we done with the freeblock?
1499                          */
1500                         if (++findex == be32_to_cpu(free->hdr.nvalid)) {
1501                                 /*
1502                                  * Drop the block.
1503                                  */
1504                                 xfs_da_brelse(tp, fbp);
1505                                 fbp = NULL;
1506                                 if (fblk && fblk->bp)
1507                                         fblk->bp = NULL;
1508                         }
1509                 }
1510         }
1511         /*
1512          * If we don't have a data block, we need to allocate one and make
1513          * the freespace entries refer to it.
1514          */
1515         if (unlikely(dbno == -1)) {
1516                 /*
1517                  * Not allowed to allocate, return failure.
1518                  */
1519                 if (args->justcheck || args->total == 0) {
1520                         /*
1521                          * Drop the freespace buffer unless it came from our
1522                          * caller.
1523                          */
1524                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1525                                 xfs_da_buf_done(fbp);
1526                         return XFS_ERROR(ENOSPC);
1527                 }
1528                 /*
1529                  * Allocate and initialize the new data block.
1530                  */
1531                 if (unlikely((error = xfs_dir2_grow_inode(args,
1532                                                          XFS_DIR2_DATA_SPACE,
1533                                                          &dbno)) ||
1534                     (error = xfs_dir2_data_init(args, dbno, &dbp)))) {
1535                         /*
1536                          * Drop the freespace buffer unless it came from our
1537                          * caller.
1538                          */
1539                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1540                                 xfs_da_buf_done(fbp);
1541                         return error;
1542                 }
1543                 /*
1544                  * If (somehow) we have a freespace block, get rid of it.
1545                  */
1546                 if (fbp)
1547                         xfs_da_brelse(tp, fbp);
1548                 if (fblk && fblk->bp)
1549                         fblk->bp = NULL;
1550
1551                 /*
1552                  * Get the freespace block corresponding to the data block
1553                  * that was just allocated.
1554                  */
1555                 fbno = XFS_DIR2_DB_TO_FDB(mp, dbno);
1556                 if (unlikely(error = xfs_da_read_buf(tp, dp,
1557                                 XFS_DIR2_DB_TO_DA(mp, fbno), -2, &fbp,
1558                                 XFS_DATA_FORK))) {
1559                         xfs_da_buf_done(dbp);
1560                         return error;
1561                 }
1562                 /*
1563                  * If there wasn't a freespace block, the read will
1564                  * return a NULL fbp.  Allocate and initialize a new one.
1565                  */
1566                 if( fbp == NULL ) {
1567                         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1568                                                         &fbno))) {
1569                                 return error;
1570                         }
1571
1572                         if (unlikely(XFS_DIR2_DB_TO_FDB(mp, dbno) != fbno)) {
1573                                 cmn_err(CE_ALERT,
1574                                         "xfs_dir2_node_addname_int: dir ino "
1575                                         "%llu needed freesp block %lld for\n"
1576                                         "  data block %lld, got %lld\n"
1577                                         "  ifbno %llu lastfbno %d\n",
1578                                         (unsigned long long)dp->i_ino,
1579                                         (long long)XFS_DIR2_DB_TO_FDB(mp, dbno),
1580                                         (long long)dbno, (long long)fbno,
1581                                         (unsigned long long)ifbno, lastfbno);
1582                                 if (fblk) {
1583                                         cmn_err(CE_ALERT,
1584                                                 " fblk 0x%p blkno %llu "
1585                                                 "index %d magic 0x%x\n",
1586                                                 fblk,
1587                                                 (unsigned long long)fblk->blkno,
1588                                                 fblk->index,
1589                                                 fblk->magic);
1590                                 } else {
1591                                         cmn_err(CE_ALERT,
1592                                                 " ... fblk is NULL\n");
1593                                 }
1594                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1595                                                  XFS_ERRLEVEL_LOW, mp);
1596                                 return XFS_ERROR(EFSCORRUPTED);
1597                         }
1598
1599                         /*
1600                          * Get a buffer for the new block.
1601                          */
1602                         if ((error = xfs_da_get_buf(tp, dp,
1603                                                    XFS_DIR2_DB_TO_DA(mp, fbno),
1604                                                    -1, &fbp, XFS_DATA_FORK))) {
1605                                 return error;
1606                         }
1607                         ASSERT(fbp != NULL);
1608
1609                         /*
1610                          * Initialize the new block to be empty, and remember
1611                          * its first slot as our empty slot.
1612                          */
1613                         free = fbp->data;
1614                         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
1615                         free->hdr.firstdb = cpu_to_be32(
1616                                 (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1617                                 XFS_DIR2_MAX_FREE_BESTS(mp));
1618                         free->hdr.nvalid = 0;
1619                         free->hdr.nused = 0;
1620                 } else {
1621                         free = fbp->data;
1622                         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1623                 }
1624
1625                 /*
1626                  * Set the freespace block index from the data block number.
1627                  */
1628                 findex = XFS_DIR2_DB_TO_FDINDEX(mp, dbno);
1629                 /*
1630                  * If it's after the end of the current entries in the
1631                  * freespace block, extend that table.
1632                  */
1633                 if (findex >= be32_to_cpu(free->hdr.nvalid)) {
1634                         ASSERT(findex < XFS_DIR2_MAX_FREE_BESTS(mp));
1635                         free->hdr.nvalid = cpu_to_be32(findex + 1);
1636                         /*
1637                          * Tag new entry so nused will go up.
1638                          */
1639                         free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1640                 }
1641                 /*
1642                  * If this entry was for an empty data block
1643                  * (this should always be true) then update the header.
1644                  */
1645                 if (be16_to_cpu(free->bests[findex]) == NULLDATAOFF) {
1646                         be32_add(&free->hdr.nused, 1);
1647                         xfs_dir2_free_log_header(tp, fbp);
1648                 }
1649                 /*
1650                  * Update the real value in the table.
1651                  * We haven't allocated the data entry yet so this will
1652                  * change again.
1653                  */
1654                 data = dbp->data;
1655                 free->bests[findex] = data->hdr.bestfree[0].length;
1656                 logfree = 1;
1657         }
1658         /*
1659          * We had a data block so we don't have to make a new one.
1660          */
1661         else {
1662                 /*
1663                  * If just checking, we succeeded.
1664                  */
1665                 if (args->justcheck) {
1666                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1667                                 xfs_da_buf_done(fbp);
1668                         return 0;
1669                 }
1670                 /*
1671                  * Read the data block in.
1672                  */
1673                 if (unlikely(
1674                     error = xfs_da_read_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, dbno),
1675                                 -1, &dbp, XFS_DATA_FORK))) {
1676                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1677                                 xfs_da_buf_done(fbp);
1678                         return error;
1679                 }
1680                 data = dbp->data;
1681                 logfree = 0;
1682         }
1683         ASSERT(be16_to_cpu(data->hdr.bestfree[0].length) >= length);
1684         /*
1685          * Point to the existing unused space.
1686          */
1687         dup = (xfs_dir2_data_unused_t *)
1688               ((char *)data + be16_to_cpu(data->hdr.bestfree[0].offset));
1689         needscan = needlog = 0;
1690         /*
1691          * Mark the first part of the unused space, inuse for us.
1692          */
1693         xfs_dir2_data_use_free(tp, dbp, dup,
1694                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)data), length,
1695                 &needlog, &needscan);
1696         /*
1697          * Fill in the new entry and log it.
1698          */
1699         dep = (xfs_dir2_data_entry_t *)dup;
1700         dep->inumber = cpu_to_be64(args->inumber);
1701         dep->namelen = args->namelen;
1702         memcpy(dep->name, args->name, dep->namelen);
1703         tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
1704         *tagp = cpu_to_be16((char *)dep - (char *)data);
1705         xfs_dir2_data_log_entry(tp, dbp, dep);
1706         /*
1707          * Rescan the block for bestfree if needed.
1708          */
1709         if (needscan)
1710                 xfs_dir2_data_freescan(mp, data, &needlog, NULL);
1711         /*
1712          * Log the data block header if needed.
1713          */
1714         if (needlog)
1715                 xfs_dir2_data_log_header(tp, dbp);
1716         /*
1717          * If the freespace entry is now wrong, update it.
1718          */
1719         if (be16_to_cpu(free->bests[findex]) != be16_to_cpu(data->hdr.bestfree[0].length)) {
1720                 free->bests[findex] = data->hdr.bestfree[0].length;
1721                 logfree = 1;
1722         }
1723         /*
1724          * Log the freespace entry if needed.
1725          */
1726         if (logfree)
1727                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1728         /*
1729          * If the caller didn't hand us the freespace block, drop it.
1730          */
1731         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1732                 xfs_da_buf_done(fbp);
1733         /*
1734          * Return the data block and offset in args, then drop the data block.
1735          */
1736         args->blkno = (xfs_dablk_t)dbno;
1737         args->index = be16_to_cpu(*tagp);
1738         xfs_da_buf_done(dbp);
1739         return 0;
1740 }
1741
1742 /*
1743  * Lookup an entry in a node-format directory.
1744  * All the real work happens in xfs_da_node_lookup_int.
1745  * The only real output is the inode number of the entry.
1746  */
1747 int                                             /* error */
1748 xfs_dir2_node_lookup(
1749         xfs_da_args_t   *args)                  /* operation arguments */
1750 {
1751         int             error;                  /* error return value */
1752         int             i;                      /* btree level */
1753         int             rval;                   /* operation return value */
1754         xfs_da_state_t  *state;                 /* btree cursor */
1755
1756         xfs_dir2_trace_args("node_lookup", args);
1757         /*
1758          * Allocate and initialize the btree cursor.
1759          */
1760         state = xfs_da_state_alloc();
1761         state->args = args;
1762         state->mp = args->dp->i_mount;
1763         state->blocksize = state->mp->m_dirblksize;
1764         state->node_ents = state->mp->m_dir_node_ents;
1765         /*
1766          * Fill in the path to the entry in the cursor.
1767          */
1768         error = xfs_da_node_lookup_int(state, &rval);
1769         if (error)
1770                 rval = error;
1771         /*
1772          * Release the btree blocks and leaf block.
1773          */
1774         for (i = 0; i < state->path.active; i++) {
1775                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1776                 state->path.blk[i].bp = NULL;
1777         }
1778         /*
1779          * Release the data block if we have it.
1780          */
1781         if (state->extravalid && state->extrablk.bp) {
1782                 xfs_da_brelse(args->trans, state->extrablk.bp);
1783                 state->extrablk.bp = NULL;
1784         }
1785         xfs_da_state_free(state);
1786         return rval;
1787 }
1788
1789 /*
1790  * Remove an entry from a node-format directory.
1791  */
1792 int                                             /* error */
1793 xfs_dir2_node_removename(
1794         xfs_da_args_t           *args)          /* operation arguments */
1795 {
1796         xfs_da_state_blk_t      *blk;           /* leaf block */
1797         int                     error;          /* error return value */
1798         int                     rval;           /* operation return value */
1799         xfs_da_state_t          *state;         /* btree cursor */
1800
1801         xfs_dir2_trace_args("node_removename", 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          * Look up the entry we're deleting, set up the cursor.
1812          */
1813         error = xfs_da_node_lookup_int(state, &rval);
1814         if (error) {
1815                 rval = error;
1816         }
1817         /*
1818          * Didn't find it, upper layer screwed up.
1819          */
1820         if (rval != EEXIST) {
1821                 xfs_da_state_free(state);
1822                 return rval;
1823         }
1824         blk = &state->path.blk[state->path.active - 1];
1825         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1826         ASSERT(state->extravalid);
1827         /*
1828          * Remove the leaf and data entries.
1829          * Extrablk refers to the data block.
1830          */
1831         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
1832                 &state->extrablk, &rval);
1833         if (error) {
1834                 return error;
1835         }
1836         /*
1837          * Fix the hash values up the btree.
1838          */
1839         xfs_da_fixhashpath(state, &state->path);
1840         /*
1841          * If we need to join leaf blocks, do it.
1842          */
1843         if (rval && state->path.active > 1)
1844                 error = xfs_da_join(state);
1845         /*
1846          * If no errors so far, try conversion to leaf format.
1847          */
1848         if (!error)
1849                 error = xfs_dir2_node_to_leaf(state);
1850         xfs_da_state_free(state);
1851         return error;
1852 }
1853
1854 /*
1855  * Replace an entry's inode number in a node-format directory.
1856  */
1857 int                                             /* error */
1858 xfs_dir2_node_replace(
1859         xfs_da_args_t           *args)          /* operation arguments */
1860 {
1861         xfs_da_state_blk_t      *blk;           /* leaf block */
1862         xfs_dir2_data_t         *data;          /* data block structure */
1863         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
1864         int                     error;          /* error return value */
1865         int                     i;              /* btree level */
1866         xfs_ino_t               inum;           /* new inode number */
1867         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1868         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
1869         int                     rval;           /* internal return value */
1870         xfs_da_state_t          *state;         /* btree cursor */
1871
1872         xfs_dir2_trace_args("node_replace", args);
1873         /*
1874          * Allocate and initialize the btree cursor.
1875          */
1876         state = xfs_da_state_alloc();
1877         state->args = args;
1878         state->mp = args->dp->i_mount;
1879         state->blocksize = state->mp->m_dirblksize;
1880         state->node_ents = state->mp->m_dir_node_ents;
1881         inum = args->inumber;
1882         /*
1883          * Lookup the entry to change in the btree.
1884          */
1885         error = xfs_da_node_lookup_int(state, &rval);
1886         if (error) {
1887                 rval = error;
1888         }
1889         /*
1890          * It should be found, since the vnodeops layer has looked it up
1891          * and locked it.  But paranoia is good.
1892          */
1893         if (rval == EEXIST) {
1894                 /*
1895                  * Find the leaf entry.
1896                  */
1897                 blk = &state->path.blk[state->path.active - 1];
1898                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1899                 leaf = blk->bp->data;
1900                 lep = &leaf->ents[blk->index];
1901                 ASSERT(state->extravalid);
1902                 /*
1903                  * Point to the data entry.
1904                  */
1905                 data = state->extrablk.bp->data;
1906                 ASSERT(be32_to_cpu(data->hdr.magic) == XFS_DIR2_DATA_MAGIC);
1907                 dep = (xfs_dir2_data_entry_t *)
1908                       ((char *)data +
1909                        XFS_DIR2_DATAPTR_TO_OFF(state->mp, be32_to_cpu(lep->address)));
1910                 ASSERT(inum != be64_to_cpu(dep->inumber));
1911                 /*
1912                  * Fill in the new inode number and log the entry.
1913                  */
1914                 dep->inumber = cpu_to_be64(inum);
1915                 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
1916                 rval = 0;
1917         }
1918         /*
1919          * Didn't find it, and we're holding a data block.  Drop it.
1920          */
1921         else if (state->extravalid) {
1922                 xfs_da_brelse(args->trans, state->extrablk.bp);
1923                 state->extrablk.bp = NULL;
1924         }
1925         /*
1926          * Release all the buffers in the cursor.
1927          */
1928         for (i = 0; i < state->path.active; i++) {
1929                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1930                 state->path.blk[i].bp = NULL;
1931         }
1932         xfs_da_state_free(state);
1933         return rval;
1934 }
1935
1936 /*
1937  * Trim off a trailing empty freespace block.
1938  * Return (in rvalp) 1 if we did it, 0 if not.
1939  */
1940 int                                             /* error */
1941 xfs_dir2_node_trim_free(
1942         xfs_da_args_t           *args,          /* operation arguments */
1943         xfs_fileoff_t           fo,             /* free block number */
1944         int                     *rvalp)         /* out: did something */
1945 {
1946         xfs_dabuf_t             *bp;            /* freespace buffer */
1947         xfs_inode_t             *dp;            /* incore directory inode */
1948         int                     error;          /* error return code */
1949         xfs_dir2_free_t         *free;          /* freespace structure */
1950         xfs_mount_t             *mp;            /* filesystem mount point */
1951         xfs_trans_t             *tp;            /* transaction pointer */
1952
1953         dp = args->dp;
1954         mp = dp->i_mount;
1955         tp = args->trans;
1956         /*
1957          * Read the freespace block.
1958          */
1959         if (unlikely(error = xfs_da_read_buf(tp, dp, (xfs_dablk_t)fo, -2, &bp,
1960                         XFS_DATA_FORK))) {
1961                 return error;
1962         }
1963
1964         /*
1965          * There can be holes in freespace.  If fo is a hole, there's
1966          * nothing to do.
1967          */
1968         if (bp == NULL) {
1969                 return 0;
1970         }
1971         free = bp->data;
1972         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1973         /*
1974          * If there are used entries, there's nothing to do.
1975          */
1976         if (be32_to_cpu(free->hdr.nused) > 0) {
1977                 xfs_da_brelse(tp, bp);
1978                 *rvalp = 0;
1979                 return 0;
1980         }
1981         /*
1982          * Blow the block away.
1983          */
1984         if ((error =
1985             xfs_dir2_shrink_inode(args, XFS_DIR2_DA_TO_DB(mp, (xfs_dablk_t)fo),
1986                     bp))) {
1987                 /*
1988                  * Can't fail with ENOSPC since that only happens with no
1989                  * space reservation, when breaking up an extent into two
1990                  * pieces.  This is the last block of an extent.
1991                  */
1992                 ASSERT(error != ENOSPC);
1993                 xfs_da_brelse(tp, bp);
1994                 return error;
1995         }
1996         /*
1997          * Return that we succeeded.
1998          */
1999         *rvalp = 1;
2000         return 0;
2001 }