fs/xfs/xfs_dir2_node.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_log.h"
  22 #include "xfs_inum.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_dir2.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_bmap_btree.h"
  30 #include "xfs_dir2_sf.h"
  31 #include "xfs_dinode.h"
  32 #include "xfs_inode.h"
  33 #include "xfs_bmap.h"
  34 #include "xfs_dir2_data.h"
  35 #include "xfs_dir2_leaf.h"
  36 #include "xfs_dir2_block.h"
  37 #include "xfs_dir2_node.h"
  38 #include "xfs_error.h"
  39 #include "xfs_trace.h"
  40
  41 /*
  42  * Function declarations.
  43  */
  44 static void xfs_dir2_free_log_header(xfs_trans_t *tp, xfs_dabuf_t *bp);
  45 static int xfs_dir2_leafn_add(xfs_dabuf_t *bp, xfs_da_args_t *args, int index);
  46 #ifdef DEBUG
  47 static void xfs_dir2_leafn_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
  48 #else
  49 #define xfs_dir2_leafn_check(dp, bp)
  50 #endif
  51 static void xfs_dir2_leafn_moveents(xfs_da_args_t *args, xfs_dabuf_t *bp_s,
  52                                     int start_s, xfs_dabuf_t *bp_d, int start_d,
  53                                     int count);
  54 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  55                                      xfs_da_state_blk_t *blk1,
  56                                      xfs_da_state_blk_t *blk2);
  57 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, xfs_dabuf_t *bp,
  58                                  int index, xfs_da_state_blk_t *dblk,
  59                                  int *rval);
  60 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  61                                      xfs_da_state_blk_t *fblk);
  62
  63 /*
  64  * Log entries from a freespace block.
  65  */
  66 STATIC void
  67 xfs_dir2_free_log_bests(
  68         xfs_trans_t             *tp,            /* transaction pointer */
  69         xfs_dabuf_t             *bp,            /* freespace buffer */
  70         int                     first,          /* first entry to log */
  71         int                     last)           /* last entry to log */
  72 {
  73         xfs_dir2_free_t         *free;          /* freespace structure */
  74
  75         free = bp->data;
  76         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
  77         xfs_da_log_buf(tp, bp,
  78                 (uint)((char *)&free->bests[first] - (char *)free),
  79                 (uint)((char *)&free->bests[last] - (char *)free +
  80                        sizeof(free->bests[0]) - 1));
  81 }
  82
  83 /*
  84  * Log header from a freespace block.
  85  */
  86 static void
  87 xfs_dir2_free_log_header(
  88         xfs_trans_t             *tp,            /* transaction pointer */
  89         xfs_dabuf_t             *bp)            /* freespace buffer */
  90 {
  91         xfs_dir2_free_t         *free;          /* freespace structure */
  92
  93         free = bp->data;
  94         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
  95         xfs_da_log_buf(tp, bp, (uint)((char *)&free->hdr - (char *)free),
  96                 (uint)(sizeof(xfs_dir2_free_hdr_t) - 1));
  97 }
  98
  99 /*
 100  * Convert a leaf-format directory to a node-format directory.
 101  * We need to change the magic number of the leaf block, and copy
 102  * the freespace table out of the leaf block into its own block.
 103  */
 104 int                                             /* error */
 105 xfs_dir2_leaf_to_node(
 106         xfs_da_args_t           *args,          /* operation arguments */
 107         xfs_dabuf_t             *lbp)           /* leaf buffer */
 108 {
 109         xfs_inode_t             *dp;            /* incore directory inode */
 110         int                     error;          /* error return value */
 111         xfs_dabuf_t             *fbp;           /* freespace buffer */
 112         xfs_dir2_db_t           fdb;            /* freespace block number */
 113         xfs_dir2_free_t         *free;          /* freespace structure */
 114         __be16                  *from;          /* pointer to freespace entry */
 115         int                     i;              /* leaf freespace index */
 116         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 117         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 118         xfs_mount_t             *mp;            /* filesystem mount point */
 119         int                     n;              /* count of live freespc ents */
 120         xfs_dir2_data_off_t     off;            /* freespace entry value */
 121         __be16                  *to;            /* pointer to freespace entry */
 122         xfs_trans_t             *tp;            /* transaction pointer */
 123
 124         trace_xfs_dir2_leaf_to_node(args);
 125
 126         dp = args->dp;
 127         mp = dp->i_mount;
 128         tp = args->trans;
 129         /*
 130          * Add a freespace block to the directory.
 131          */
 132         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 133                 return error;
 134         }
 135         ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
 136         /*
 137          * Get the buffer for the new freespace block.
 138          */
 139         if ((error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb), -1, &fbp,
 140                         XFS_DATA_FORK))) {
 141                 return error;
 142         }
 143         ASSERT(fbp != NULL);
 144         free = fbp->data;
 145         leaf = lbp->data;
 146         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 147         /*
 148          * Initialize the freespace block header.
 149          */
 150         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
 151         free->hdr.firstdb = 0;
 152         ASSERT(be32_to_cpu(ltp->bestcount) <= (uint)dp->i_d.di_size / mp->m_dirblksize);
 153         free->hdr.nvalid = ltp->bestcount;
 154         /*
 155          * Copy freespace entries from the leaf block to the new block.
 156          * Count active entries.
 157          */
 158         for (i = n = 0, from = xfs_dir2_leaf_bests_p(ltp), to = free->bests;
 159              i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 160                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 161                         n++;
 162                 *to = cpu_to_be16(off);
 163         }
 164         free->hdr.nused = cpu_to_be32(n);
 165         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 166         /*
 167          * Log everything.
 168          */
 169         xfs_dir2_leaf_log_header(tp, lbp);
 170         xfs_dir2_free_log_header(tp, fbp);
 171         xfs_dir2_free_log_bests(tp, fbp, 0, be32_to_cpu(free->hdr.nvalid) - 1);
 172         xfs_da_buf_done(fbp);
 173         xfs_dir2_leafn_check(dp, lbp);
 174         return 0;
 175 }
 176
 177 /*
 178  * Add a leaf entry to a leaf block in a node-form directory.
 179  * The other work necessary is done from the caller.
 180  */
 181 static int                                      /* error */
 182 xfs_dir2_leafn_add(
 183         xfs_dabuf_t             *bp,            /* leaf buffer */
 184         xfs_da_args_t           *args,          /* operation arguments */
 185         int                     index)          /* insertion pt for new entry */
 186 {
 187         int                     compact;        /* compacting stale leaves */
 188         xfs_inode_t             *dp;            /* incore directory inode */
 189         int                     highstale;      /* next stale entry */
 190         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 191         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 192         int                     lfloghigh;      /* high leaf entry logging */
 193         int                     lfloglow;       /* low leaf entry logging */
 194         int                     lowstale;       /* previous stale entry */
 195         xfs_mount_t             *mp;            /* filesystem mount point */
 196         xfs_trans_t             *tp;            /* transaction pointer */
 197
 198         trace_xfs_dir2_leafn_add(args, index);
 199
 200         dp = args->dp;
 201         mp = dp->i_mount;
 202         tp = args->trans;
 203         leaf = bp->data;
 204
 205         /*
 206          * Quick check just to make sure we are not going to index
 207          * into other peoples memory
 208          */
 209         if (index < 0)
 210                 return XFS_ERROR(EFSCORRUPTED);
 211
 212         /*
 213          * If there are already the maximum number of leaf entries in
 214          * the block, if there are no stale entries it won't fit.
 215          * Caller will do a split.  If there are stale entries we'll do
 216          * a compact.
 217          */
 218
 219         if (be16_to_cpu(leaf->hdr.count) == xfs_dir2_max_leaf_ents(mp)) {
 220                 if (!leaf->hdr.stale)
 221                         return XFS_ERROR(ENOSPC);
 222                 compact = be16_to_cpu(leaf->hdr.stale) > 1;
 223         } else
 224                 compact = 0;
 225         ASSERT(index == 0 || be32_to_cpu(leaf->ents[index - 1].hashval) <= args->hashval);
 226         ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
 227                be32_to_cpu(leaf->ents[index].hashval) >= args->hashval);
 228
 229         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 230                 return 0;
 231
 232         /*
 233          * Compact out all but one stale leaf entry.  Leaves behind
 234          * the entry closest to index.
 235          */
 236         if (compact) {
 237                 xfs_dir2_leaf_compact_x1(bp, &index, &lowstale, &highstale,
 238                         &lfloglow, &lfloghigh);
 239         }
 240         /*
 241          * Set impossible logging indices for this case.
 242          */
 243         else if (leaf->hdr.stale) {
 244                 lfloglow = be16_to_cpu(leaf->hdr.count);
 245                 lfloghigh = -1;
 246         }
 247         /*
 248          * No stale entries, just insert a space for the new entry.
 249          */
 250         if (!leaf->hdr.stale) {
 251                 lep = &leaf->ents[index];
 252                 if (index < be16_to_cpu(leaf->hdr.count))
 253                         memmove(lep + 1, lep,
 254                                 (be16_to_cpu(leaf->hdr.count) - index) * sizeof(*lep));
 255                 lfloglow = index;
 256                 lfloghigh = be16_to_cpu(leaf->hdr.count);
 257                 be16_add_cpu(&leaf->hdr.count, 1);
 258         }
 259         /*
 260          * There are stale entries.  We'll use one for the new entry.
 261          */
 262         else {
 263                 /*
 264                  * If we didn't do a compact then we need to figure out
 265                  * which stale entry will be used.
 266                  */
 267                 if (compact == 0) {
 268                         /*
 269                          * Find first stale entry before our insertion point.
 270                          */
 271                         for (lowstale = index - 1;
 272                              lowstale >= 0 &&
 273                                 be32_to_cpu(leaf->ents[lowstale].address) !=
 274                                 XFS_DIR2_NULL_DATAPTR;
 275                              lowstale--)
 276                                 continue;
 277                         /*
 278                          * Find next stale entry after insertion point.
 279                          * Stop looking if the answer would be worse than
 280                          * lowstale already found.
 281                          */
 282                         for (highstale = index;
 283                              highstale < be16_to_cpu(leaf->hdr.count) &&
 284                                 be32_to_cpu(leaf->ents[highstale].address) !=
 285                                 XFS_DIR2_NULL_DATAPTR &&
 286                                 (lowstale < 0 ||
 287                                  index - lowstale - 1 >= highstale - index);
 288                              highstale++)
 289                                 continue;
 290                 }
 291                 /*
 292                  * Using the low stale entry.
 293                  * Shift entries up toward the stale slot.
 294                  */
 295                 if (lowstale >= 0 &&
 296                     (highstale == be16_to_cpu(leaf->hdr.count) ||
 297                      index - lowstale - 1 < highstale - index)) {
 298                         ASSERT(be32_to_cpu(leaf->ents[lowstale].address) ==
 299                                XFS_DIR2_NULL_DATAPTR);
 300                         ASSERT(index - lowstale - 1 >= 0);
 301                         if (index - lowstale - 1 > 0)
 302                                 memmove(&leaf->ents[lowstale],
 303                                         &leaf->ents[lowstale + 1],
 304                                         (index - lowstale - 1) * sizeof(*lep));
 305                         lep = &leaf->ents[index - 1];
 306                         lfloglow = MIN(lowstale, lfloglow);
 307                         lfloghigh = MAX(index - 1, lfloghigh);
 308                 }
 309                 /*
 310                  * Using the high stale entry.
 311                  * Shift entries down toward the stale slot.
 312                  */
 313                 else {
 314                         ASSERT(be32_to_cpu(leaf->ents[highstale].address) ==
 315                                XFS_DIR2_NULL_DATAPTR);
 316                         ASSERT(highstale - index >= 0);
 317                         if (highstale - index > 0)
 318                                 memmove(&leaf->ents[index + 1],
 319                                         &leaf->ents[index],
 320                                         (highstale - index) * sizeof(*lep));
 321                         lep = &leaf->ents[index];
 322                         lfloglow = MIN(index, lfloglow);
 323                         lfloghigh = MAX(highstale, lfloghigh);
 324                 }
 325                 be16_add_cpu(&leaf->hdr.stale, -1);
 326         }
 327         /*
 328          * Insert the new entry, log everything.
 329          */
 330         lep->hashval = cpu_to_be32(args->hashval);
 331         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
 332                                 args->blkno, args->index));
 333         xfs_dir2_leaf_log_header(tp, bp);
 334         xfs_dir2_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
 335         xfs_dir2_leafn_check(dp, bp);
 336         return 0;
 337 }
 338
 339 #ifdef DEBUG
 340 /*
 341  * Check internal consistency of a leafn block.
 342  */
 343 void
 344 xfs_dir2_leafn_check(
 345         xfs_inode_t     *dp,                    /* incore directory inode */
 346         xfs_dabuf_t     *bp)                    /* leaf buffer */
 347 {
 348         int             i;                      /* leaf index */
 349         xfs_dir2_leaf_t *leaf;                  /* leaf structure */
 350         xfs_mount_t     *mp;                    /* filesystem mount point */
 351         int             stale;                  /* count of stale leaves */
 352
 353         leaf = bp->data;
 354         mp = dp->i_mount;
 355         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 356         ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
 357         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
 358                 if (i + 1 < be16_to_cpu(leaf->hdr.count)) {
 359                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
 360                                be32_to_cpu(leaf->ents[i + 1].hashval));
 361                 }
 362                 if (be32_to_cpu(leaf->ents[i].address) == XFS_DIR2_NULL_DATAPTR)
 363                         stale++;
 364         }
 365         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
 366 }
 367 #endif  /* DEBUG */
 368
 369 /*
 370  * Return the last hash value in the leaf.
 371  * Stale entries are ok.
 372  */
 373 xfs_dahash_t                                    /* hash value */
 374 xfs_dir2_leafn_lasthash(
 375         xfs_dabuf_t     *bp,                    /* leaf buffer */
 376         int             *count)                 /* count of entries in leaf */
 377 {
 378         xfs_dir2_leaf_t *leaf;                  /* leaf structure */
 379
 380         leaf = bp->data;
 381         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 382         if (count)
 383                 *count = be16_to_cpu(leaf->hdr.count);
 384         if (!leaf->hdr.count)
 385                 return 0;
 386         return be32_to_cpu(leaf->ents[be16_to_cpu(leaf->hdr.count) - 1].hashval);
 387 }
 388
 389 /*
 390  * Look up a leaf entry for space to add a name in a node-format leaf block.
 391  * The extrablk in state is a freespace block.
 392  */
 393 STATIC int
 394 xfs_dir2_leafn_lookup_for_addname(
 395         xfs_dabuf_t             *bp,            /* leaf buffer */
 396         xfs_da_args_t           *args,          /* operation arguments */
 397         int                     *indexp,        /* out: leaf entry index */
 398         xfs_da_state_t          *state)         /* state to fill in */
 399 {
 400         xfs_dabuf_t             *curbp = NULL;  /* current data/free buffer */
 401         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 402         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 403         xfs_inode_t             *dp;            /* incore directory inode */
 404         int                     error;          /* error return value */
 405         int                     fi;             /* free entry index */
 406         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 407         int                     index;          /* leaf entry index */
 408         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 409         int                     length;         /* length of new data entry */
 410         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 411         xfs_mount_t             *mp;            /* filesystem mount point */
 412         xfs_dir2_db_t           newdb;          /* new data block number */
 413         xfs_dir2_db_t           newfdb;         /* new free block number */
 414         xfs_trans_t             *tp;            /* transaction pointer */
 415
 416         dp = args->dp;
 417         tp = args->trans;
 418         mp = dp->i_mount;
 419         leaf = bp->data;
 420         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 421 #ifdef __KERNEL__
 422         ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
 423 #endif
 424         xfs_dir2_leafn_check(dp, bp);
 425         /*
 426          * Look up the hash value in the leaf entries.
 427          */
 428         index = xfs_dir2_leaf_search_hash(args, bp);
 429         /*
 430          * Do we have a buffer coming in?
 431          */
 432         if (state->extravalid) {
 433                 /* If so, it's a free block buffer, get the block number. */
 434                 curbp = state->extrablk.bp;
 435                 curfdb = state->extrablk.blkno;
 436                 free = curbp->data;
 437                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
 438         }
 439         length = xfs_dir2_data_entsize(args->namelen);
 440         /*
 441          * Loop over leaf entries with the right hash value.
 442          */
 443         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
 444                                 be32_to_cpu(lep->hashval) == args->hashval;
 445                                 lep++, index++) {
 446                 /*
 447                  * Skip stale leaf entries.
 448                  */
 449                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 450                         continue;
 451                 /*
 452                  * Pull the data block number from the entry.
 453                  */
 454                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 455                 /*
 456                  * For addname, we're looking for a place to put the new entry.
 457                  * We want to use a data block with an entry of equal
 458                  * hash value to ours if there is one with room.
 459                  *
 460                  * If this block isn't the data block we already have
 461                  * in hand, take a look at it.
 462                  */
 463                 if (newdb != curdb) {
 464                         curdb = newdb;
 465                         /*
 466                          * Convert the data block to the free block
 467                          * holding its freespace information.
 468                          */
 469                         newfdb = xfs_dir2_db_to_fdb(mp, newdb);
 470                         /*
 471                          * If it's not the one we have in hand, read it in.
 472                          */
 473                         if (newfdb != curfdb) {
 474                                 /*
 475                                  * If we had one before, drop it.
 476                                  */
 477                                 if (curbp)
 478                                         xfs_da_brelse(tp, curbp);
 479                                 /*
 480                                  * Read the free block.
 481                                  */
 482                                 error = xfs_da_read_buf(tp, dp,
 483                                                 xfs_dir2_db_to_da(mp, newfdb),
 484                                                 -1, &curbp, XFS_DATA_FORK);
 485                                 if (error)
 486                                         return error;
 487                                 free = curbp->data;
 488                                 ASSERT(be32_to_cpu(free->hdr.magic) ==
 489                                         XFS_DIR2_FREE_MAGIC);
 490                                 ASSERT((be32_to_cpu(free->hdr.firstdb) %
 491                                         XFS_DIR2_MAX_FREE_BESTS(mp)) == 0);
 492                                 ASSERT(be32_to_cpu(free->hdr.firstdb) <= curdb);
 493                                 ASSERT(curdb < be32_to_cpu(free->hdr.firstdb) +
 494                                         be32_to_cpu(free->hdr.nvalid));
 495                         }
 496                         /*
 497                          * Get the index for our entry.
 498                          */
 499                         fi = xfs_dir2_db_to_fdindex(mp, curdb);
 500                         /*
 501                          * If it has room, return it.
 502                          */
 503                         if (unlikely(be16_to_cpu(free->bests[fi]) == NULLDATAOFF)) {
 504                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 505                                                         XFS_ERRLEVEL_LOW, mp);
 506                                 if (curfdb != newfdb)
 507                                         xfs_da_brelse(tp, curbp);
 508                                 return XFS_ERROR(EFSCORRUPTED);
 509                         }
 510                         curfdb = newfdb;
 511                         if (be16_to_cpu(free->bests[fi]) >= length)
 512                                 goto out;
 513                 }
 514         }
 515         /* Didn't find any space */
 516         fi = -1;
 517 out:
 518         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 519         if (curbp) {
 520                 /* Giving back a free block. */
 521                 state->extravalid = 1;
 522                 state->extrablk.bp = curbp;
 523                 state->extrablk.index = fi;
 524                 state->extrablk.blkno = curfdb;
 525                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 526         } else {
 527                 state->extravalid = 0;
 528         }
 529         /*
 530          * Return the index, that will be the insertion point.
 531          */
 532         *indexp = index;
 533         return XFS_ERROR(ENOENT);
 534 }
 535
 536 /*
 537  * Look up a leaf entry in a node-format leaf block.
 538  * The extrablk in state a data block.
 539  */
 540 STATIC int
 541 xfs_dir2_leafn_lookup_for_entry(
 542         xfs_dabuf_t             *bp,            /* leaf buffer */
 543         xfs_da_args_t           *args,          /* operation arguments */
 544         int                     *indexp,        /* out: leaf entry index */
 545         xfs_da_state_t          *state)         /* state to fill in */
 546 {
 547         xfs_dabuf_t             *curbp = NULL;  /* current data/free buffer */
 548         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 549         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 550         xfs_inode_t             *dp;            /* incore directory inode */
 551         int                     error;          /* error return value */
 552         int                     index;          /* leaf entry index */
 553         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 554         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 555         xfs_mount_t             *mp;            /* filesystem mount point */
 556         xfs_dir2_db_t           newdb;          /* new data block number */
 557         xfs_trans_t             *tp;            /* transaction pointer */
 558         enum xfs_dacmp          cmp;            /* comparison result */
 559
 560         dp = args->dp;
 561         tp = args->trans;
 562         mp = dp->i_mount;
 563         leaf = bp->data;
 564         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 565 #ifdef __KERNEL__
 566         ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
 567 #endif
 568         xfs_dir2_leafn_check(dp, bp);
 569         /*
 570          * Look up the hash value in the leaf entries.
 571          */
 572         index = xfs_dir2_leaf_search_hash(args, bp);
 573         /*
 574          * Do we have a buffer coming in?
 575          */
 576         if (state->extravalid) {
 577                 curbp = state->extrablk.bp;
 578                 curdb = state->extrablk.blkno;
 579         }
 580         /*
 581          * Loop over leaf entries with the right hash value.
 582          */
 583         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
 584                                 be32_to_cpu(lep->hashval) == args->hashval;
 585                                 lep++, index++) {
 586                 /*
 587                  * Skip stale leaf entries.
 588                  */
 589                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 590                         continue;
 591                 /*
 592                  * Pull the data block number from the entry.
 593                  */
 594                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 595                 /*
 596                  * Not adding a new entry, so we really want to find
 597                  * the name given to us.
 598                  *
 599                  * If it's a different data block, go get it.
 600                  */
 601                 if (newdb != curdb) {
 602                         /*
 603                          * If we had a block before that we aren't saving
 604                          * for a CI name, drop it
 605                          */
 606                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 607                                                 curdb != state->extrablk.blkno))
 608                                 xfs_da_brelse(tp, curbp);
 609                         /*
 610                          * If needing the block that is saved with a CI match,
 611                          * use it otherwise read in the new data block.
 612                          */
 613                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 614                                         newdb == state->extrablk.blkno) {
 615                                 ASSERT(state->extravalid);
 616                                 curbp = state->extrablk.bp;
 617                         } else {
 618                                 error = xfs_da_read_buf(tp, dp,
 619                                                 xfs_dir2_db_to_da(mp, newdb),
 620                                                 -1, &curbp, XFS_DATA_FORK);
 621                                 if (error)
 622                                         return error;
 623                         }
 624                         xfs_dir2_data_check(dp, curbp);
 625                         curdb = newdb;
 626                 }
 627                 /*
 628                  * Point to the data entry.
 629                  */
 630                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->data +
 631                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
 632                 /*
 633                  * Compare the entry and if it's an exact match, return
 634                  * EEXIST immediately. If it's the first case-insensitive
 635                  * match, store the block & inode number and continue looking.
 636                  */
 637                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 638                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 639                         /* If there is a CI match block, drop it */
 640                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 641                                                 curdb != state->extrablk.blkno)
 642                                 xfs_da_brelse(tp, state->extrablk.bp);
 643                         args->cmpresult = cmp;
 644                         args->inumber = be64_to_cpu(dep->inumber);
 645                         *indexp = index;
 646                         state->extravalid = 1;
 647                         state->extrablk.bp = curbp;
 648                         state->extrablk.blkno = curdb;
 649                         state->extrablk.index = (int)((char *)dep -
 650                                                         (char *)curbp->data);
 651                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 652                         if (cmp == XFS_CMP_EXACT)
 653                                 return XFS_ERROR(EEXIST);
 654                 }
 655         }
 656         ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
 657                                         (args->op_flags & XFS_DA_OP_OKNOENT));
 658         if (curbp) {
 659                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 660                         /* Giving back last used data block. */
 661                         state->extravalid = 1;
 662                         state->extrablk.bp = curbp;
 663                         state->extrablk.index = -1;
 664                         state->extrablk.blkno = curdb;
 665                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 666                 } else {
 667                         /* If the curbp is not the CI match block, drop it */
 668                         if (state->extrablk.bp != curbp)
 669                                 xfs_da_brelse(tp, curbp);
 670                 }
 671         } else {
 672                 state->extravalid = 0;
 673         }
 674         *indexp = index;
 675         return XFS_ERROR(ENOENT);
 676 }
 677
 678 /*
 679  * Look up a leaf entry in a node-format leaf block.
 680  * If this is an addname then the extrablk in state is a freespace block,
 681  * otherwise it's a data block.
 682  */
 683 int
 684 xfs_dir2_leafn_lookup_int(
 685         xfs_dabuf_t             *bp,            /* leaf buffer */
 686         xfs_da_args_t           *args,          /* operation arguments */
 687         int                     *indexp,        /* out: leaf entry index */
 688         xfs_da_state_t          *state)         /* state to fill in */
 689 {
 690         if (args->op_flags & XFS_DA_OP_ADDNAME)
 691                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 692                                                         state);
 693         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 694 }
 695
 696 /*
 697  * Move count leaf entries from source to destination leaf.
 698  * Log entries and headers.  Stale entries are preserved.
 699  */
 700 static void
 701 xfs_dir2_leafn_moveents(
 702         xfs_da_args_t   *args,                  /* operation arguments */
 703         xfs_dabuf_t     *bp_s,                  /* source leaf buffer */
 704         int             start_s,                /* source leaf index */
 705         xfs_dabuf_t     *bp_d,                  /* destination leaf buffer */
 706         int             start_d,                /* destination leaf index */
 707         int             count)                  /* count of leaves to copy */
 708 {
 709         xfs_dir2_leaf_t *leaf_d;                /* destination leaf structure */
 710         xfs_dir2_leaf_t *leaf_s;                /* source leaf structure */
 711         int             stale;                  /* count stale leaves copied */
 712         xfs_trans_t     *tp;                    /* transaction pointer */
 713
 714         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 715
 716         /*
 717          * Silently return if nothing to do.
 718          */
 719         if (count == 0) {
 720                 return;
 721         }
 722         tp = args->trans;
 723         leaf_s = bp_s->data;
 724         leaf_d = bp_d->data;
 725         /*
 726          * If the destination index is not the end of the current
 727          * destination leaf entries, open up a hole in the destination
 728          * to hold the new entries.
 729          */
 730         if (start_d < be16_to_cpu(leaf_d->hdr.count)) {
 731                 memmove(&leaf_d->ents[start_d + count], &leaf_d->ents[start_d],
 732                         (be16_to_cpu(leaf_d->hdr.count) - start_d) *
 733                         sizeof(xfs_dir2_leaf_entry_t));
 734                 xfs_dir2_leaf_log_ents(tp, bp_d, start_d + count,
 735                         count + be16_to_cpu(leaf_d->hdr.count) - 1);
 736         }
 737         /*
 738          * If the source has stale leaves, count the ones in the copy range
 739          * so we can update the header correctly.
 740          */
 741         if (leaf_s->hdr.stale) {
 742                 int     i;                      /* temp leaf index */
 743
 744                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 745                         if (be32_to_cpu(leaf_s->ents[i].address) == XFS_DIR2_NULL_DATAPTR)
 746                                 stale++;
 747                 }
 748         } else
 749                 stale = 0;
 750         /*
 751          * Copy the leaf entries from source to destination.
 752          */
 753         memcpy(&leaf_d->ents[start_d], &leaf_s->ents[start_s],
 754                 count * sizeof(xfs_dir2_leaf_entry_t));
 755         xfs_dir2_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
 756         /*
 757          * If there are source entries after the ones we copied,
 758          * delete the ones we copied by sliding the next ones down.
 759          */
 760         if (start_s + count < be16_to_cpu(leaf_s->hdr.count)) {
 761                 memmove(&leaf_s->ents[start_s], &leaf_s->ents[start_s + count],
 762                         count * sizeof(xfs_dir2_leaf_entry_t));
 763                 xfs_dir2_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
 764         }
 765         /*
 766          * Update the headers and log them.
 767          */
 768         be16_add_cpu(&leaf_s->hdr.count, -(count));
 769         be16_add_cpu(&leaf_s->hdr.stale, -(stale));
 770         be16_add_cpu(&leaf_d->hdr.count, count);
 771         be16_add_cpu(&leaf_d->hdr.stale, stale);
 772         xfs_dir2_leaf_log_header(tp, bp_s);
 773         xfs_dir2_leaf_log_header(tp, bp_d);
 774         xfs_dir2_leafn_check(args->dp, bp_s);
 775         xfs_dir2_leafn_check(args->dp, bp_d);
 776 }
 777
 778 /*
 779  * Determine the sort order of two leaf blocks.
 780  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 781  */
 782 int                                             /* sort order */
 783 xfs_dir2_leafn_order(
 784         xfs_dabuf_t     *leaf1_bp,              /* leaf1 buffer */
 785         xfs_dabuf_t     *leaf2_bp)              /* leaf2 buffer */
 786 {
 787         xfs_dir2_leaf_t *leaf1;                 /* leaf1 structure */
 788         xfs_dir2_leaf_t *leaf2;                 /* leaf2 structure */
 789
 790         leaf1 = leaf1_bp->data;
 791         leaf2 = leaf2_bp->data;
 792         ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 793         ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 794         if (be16_to_cpu(leaf1->hdr.count) > 0 &&
 795             be16_to_cpu(leaf2->hdr.count) > 0 &&
 796             (be32_to_cpu(leaf2->ents[0].hashval) < be32_to_cpu(leaf1->ents[0].hashval) ||
 797              be32_to_cpu(leaf2->ents[be16_to_cpu(leaf2->hdr.count) - 1].hashval) <
 798              be32_to_cpu(leaf1->ents[be16_to_cpu(leaf1->hdr.count) - 1].hashval)))
 799                 return 1;
 800         return 0;
 801 }
 802
 803 /*
 804  * Rebalance leaf entries between two leaf blocks.
 805  * This is actually only called when the second block is new,
 806  * though the code deals with the general case.
 807  * A new entry will be inserted in one of the blocks, and that
 808  * entry is taken into account when balancing.
 809  */
 810 static void
 811 xfs_dir2_leafn_rebalance(
 812         xfs_da_state_t          *state,         /* btree cursor */
 813         xfs_da_state_blk_t      *blk1,          /* first btree block */
 814         xfs_da_state_blk_t      *blk2)          /* second btree block */
 815 {
 816         xfs_da_args_t           *args;          /* operation arguments */
 817         int                     count;          /* count (& direction) leaves */
 818         int                     isleft;         /* new goes in left leaf */
 819         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 820         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 821         int                     mid;            /* midpoint leaf index */
 822 #ifdef DEBUG
 823         int                     oldstale;       /* old count of stale leaves */
 824 #endif
 825         int                     oldsum;         /* old total leaf count */
 826         int                     swap;           /* swapped leaf blocks */
 827
 828         args = state->args;
 829         /*
 830          * If the block order is wrong, swap the arguments.
 831          */
 832         if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
 833                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
 834
 835                 tmp = blk1;
 836                 blk1 = blk2;
 837                 blk2 = tmp;
 838         }
 839         leaf1 = blk1->bp->data;
 840         leaf2 = blk2->bp->data;
 841         oldsum = be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count);
 842 #ifdef DEBUG
 843         oldstale = be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale);
 844 #endif
 845         mid = oldsum >> 1;
 846         /*
 847          * If the old leaf count was odd then the new one will be even,
 848          * so we need to divide the new count evenly.
 849          */
 850         if (oldsum & 1) {
 851                 xfs_dahash_t    midhash;        /* middle entry hash value */
 852
 853                 if (mid >= be16_to_cpu(leaf1->hdr.count))
 854                         midhash = be32_to_cpu(leaf2->ents[mid - be16_to_cpu(leaf1->hdr.count)].hashval);
 855                 else
 856                         midhash = be32_to_cpu(leaf1->ents[mid].hashval);
 857                 isleft = args->hashval <= midhash;
 858         }
 859         /*
 860          * If the old count is even then the new count is odd, so there's
 861          * no preferred side for the new entry.
 862          * Pick the left one.
 863          */
 864         else
 865                 isleft = 1;
 866         /*
 867          * Calculate moved entry count.  Positive means left-to-right,
 868          * negative means right-to-left.  Then move the entries.
 869          */
 870         count = be16_to_cpu(leaf1->hdr.count) - mid + (isleft == 0);
 871         if (count > 0)
 872                 xfs_dir2_leafn_moveents(args, blk1->bp,
 873                         be16_to_cpu(leaf1->hdr.count) - count, blk2->bp, 0, count);
 874         else if (count < 0)
 875                 xfs_dir2_leafn_moveents(args, blk2->bp, 0, blk1->bp,
 876                         be16_to_cpu(leaf1->hdr.count), count);
 877         ASSERT(be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count) == oldsum);
 878         ASSERT(be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale) == oldstale);
 879         /*
 880          * Mark whether we're inserting into the old or new leaf.
 881          */
 882         if (be16_to_cpu(leaf1->hdr.count) < be16_to_cpu(leaf2->hdr.count))
 883                 state->inleaf = swap;
 884         else if (be16_to_cpu(leaf1->hdr.count) > be16_to_cpu(leaf2->hdr.count))
 885                 state->inleaf = !swap;
 886         else
 887                 state->inleaf =
 888                         swap ^ (blk1->index <= be16_to_cpu(leaf1->hdr.count));
 889         /*
 890          * Adjust the expected index for insertion.
 891          */
 892         if (!state->inleaf)
 893                 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
 894
 895         /*
 896          * Finally sanity check just to make sure we are not returning a
 897          * negative index
 898          */
 899         if(blk2->index < 0) {
 900                 state->inleaf = 1;
 901                 blk2->index = 0;
 902                 xfs_alert(args->dp->i_mount,
 903         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d\n",
 904                         __func__, blk1->index);
 905         }
 906 }
 907
 908 /*
 909  * Remove an entry from a node directory.
 910  * This removes the leaf entry and the data entry,
 911  * and updates the free block if necessary.
 912  */
 913 static int                                      /* error */
 914 xfs_dir2_leafn_remove(
 915         xfs_da_args_t           *args,          /* operation arguments */
 916         xfs_dabuf_t             *bp,            /* leaf buffer */
 917         int                     index,          /* leaf entry index */
 918         xfs_da_state_blk_t      *dblk,          /* data block */
 919         int                     *rval)          /* resulting block needs join */
 920 {
 921         xfs_dir2_data_t         *data;          /* data block structure */
 922         xfs_dir2_db_t           db;             /* data block number */
 923         xfs_dabuf_t             *dbp;           /* data block buffer */
 924         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 925         xfs_inode_t             *dp;            /* incore directory inode */
 926         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 927         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 928         int                     longest;        /* longest data free entry */
 929         int                     off;            /* data block entry offset */
 930         xfs_mount_t             *mp;            /* filesystem mount point */
 931         int                     needlog;        /* need to log data header */
 932         int                     needscan;       /* need to rescan data frees */
 933         xfs_trans_t             *tp;            /* transaction pointer */
 934
 935         trace_xfs_dir2_leafn_remove(args, index);
 936
 937         dp = args->dp;
 938         tp = args->trans;
 939         mp = dp->i_mount;
 940         leaf = bp->data;
 941         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 942         /*
 943          * Point to the entry we're removing.
 944          */
 945         lep = &leaf->ents[index];
 946         /*
 947          * Extract the data block and offset from the entry.
 948          */
 949         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 950         ASSERT(dblk->blkno == db);
 951         off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
 952         ASSERT(dblk->index == off);
 953         /*
 954          * Kill the leaf entry by marking it stale.
 955          * Log the leaf block changes.
 956          */
 957         be16_add_cpu(&leaf->hdr.stale, 1);
 958         xfs_dir2_leaf_log_header(tp, bp);
 959         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 960         xfs_dir2_leaf_log_ents(tp, bp, index, index);
 961         /*
 962          * Make the data entry free.  Keep track of the longest freespace
 963          * in the data block in case it changes.
 964          */
 965         dbp = dblk->bp;
 966         data = dbp->data;
 967         dep = (xfs_dir2_data_entry_t *)((char *)data + off);
 968         longest = be16_to_cpu(data->hdr.bestfree[0].length);
 969         needlog = needscan = 0;
 970         xfs_dir2_data_make_free(tp, dbp, off,
 971                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
 972         /*
 973          * Rescan the data block freespaces for bestfree.
 974          * Log the data block header if needed.
 975          */
 976         if (needscan)
 977                 xfs_dir2_data_freescan(mp, data, &needlog);
 978         if (needlog)
 979                 xfs_dir2_data_log_header(tp, dbp);
 980         xfs_dir2_data_check(dp, dbp);
 981         /*
 982          * If the longest data block freespace changes, need to update
 983          * the corresponding freeblock entry.
 984          */
 985         if (longest < be16_to_cpu(data->hdr.bestfree[0].length)) {
 986                 int             error;          /* error return value */
 987                 xfs_dabuf_t     *fbp;           /* freeblock buffer */
 988                 xfs_dir2_db_t   fdb;            /* freeblock block number */
 989                 int             findex;         /* index in freeblock entries */
 990                 xfs_dir2_free_t *free;          /* freeblock structure */
 991                 int             logfree;        /* need to log free entry */
 992
 993                 /*
 994                  * Convert the data block number to a free block,
 995                  * read in the free block.
 996                  */
 997                 fdb = xfs_dir2_db_to_fdb(mp, db);
 998                 if ((error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb),
 999                                 -1, &fbp, XFS_DATA_FORK))) {
1000                         return error;
1001                 }
1002                 free = fbp->data;
1003                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1004                 ASSERT(be32_to_cpu(free->hdr.firstdb) ==
1005                        XFS_DIR2_MAX_FREE_BESTS(mp) *
1006                        (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
1007                 /*
1008                  * Calculate which entry we need to fix.
1009                  */
1010                 findex = xfs_dir2_db_to_fdindex(mp, db);
1011                 longest = be16_to_cpu(data->hdr.bestfree[0].length);
1012                 /*
1013                  * If the data block is now empty we can get rid of it
1014                  * (usually).
1015                  */
1016                 if (longest == mp->m_dirblksize - (uint)sizeof(data->hdr)) {
1017                         /*
1018                          * Try to punch out the data block.
1019                          */
1020                         error = xfs_dir2_shrink_inode(args, db, dbp);
1021                         if (error == 0) {
1022                                 dblk->bp = NULL;
1023                                 data = NULL;
1024                         }
1025                         /*
1026                          * We can get ENOSPC if there's no space reservation.
1027                          * In this case just drop the buffer and some one else
1028                          * will eventually get rid of the empty block.
1029                          */
1030                         else if (error == ENOSPC && args->total == 0)
1031                                 xfs_da_buf_done(dbp);
1032                         else
1033                                 return error;
1034                 }
1035                 /*
1036                  * If we got rid of the data block, we can eliminate that entry
1037                  * in the free block.
1038                  */
1039                 if (data == NULL) {
1040                         /*
1041                          * One less used entry in the free table.
1042                          */
1043                         be32_add_cpu(&free->hdr.nused, -1);
1044                         xfs_dir2_free_log_header(tp, fbp);
1045                         /*
1046                          * If this was the last entry in the table, we can
1047                          * trim the table size back.  There might be other
1048                          * entries at the end referring to non-existent
1049                          * data blocks, get those too.
1050                          */
1051                         if (findex == be32_to_cpu(free->hdr.nvalid) - 1) {
1052                                 int     i;              /* free entry index */
1053
1054                                 for (i = findex - 1;
1055                                      i >= 0 && be16_to_cpu(free->bests[i]) == NULLDATAOFF;
1056                                      i--)
1057                                         continue;
1058                                 free->hdr.nvalid = cpu_to_be32(i + 1);
1059                                 logfree = 0;
1060                         }
1061                         /*
1062                          * Not the last entry, just punch it out.
1063                          */
1064                         else {
1065                                 free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1066                                 logfree = 1;
1067                         }
1068                         /*
1069                          * If there are no useful entries left in the block,
1070                          * get rid of the block if we can.
1071                          */
1072                         if (!free->hdr.nused) {
1073                                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1074                                 if (error == 0) {
1075                                         fbp = NULL;
1076                                         logfree = 0;
1077                                 } else if (error != ENOSPC || args->total != 0)
1078                                         return error;
1079                                 /*
1080                                  * It's possible to get ENOSPC if there is no
1081                                  * space reservation.  In this case some one
1082                                  * else will eventually get rid of this block.
1083                                  */
1084                         }
1085                 }
1086                 /*
1087                  * Data block is not empty, just set the free entry to
1088                  * the new value.
1089                  */
1090                 else {
1091                         free->bests[findex] = cpu_to_be16(longest);
1092                         logfree = 1;
1093                 }
1094                 /*
1095                  * Log the free entry that changed, unless we got rid of it.
1096                  */
1097                 if (logfree)
1098                         xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1099                 /*
1100                  * Drop the buffer if we still have it.
1101                  */
1102                 if (fbp)
1103                         xfs_da_buf_done(fbp);
1104         }
1105         xfs_dir2_leafn_check(dp, bp);
1106         /*
1107          * Return indication of whether this leaf block is empty enough
1108          * to justify trying to join it with a neighbor.
1109          */
1110         *rval =
1111                 ((uint)sizeof(leaf->hdr) +
1112                  (uint)sizeof(leaf->ents[0]) *
1113                  (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale))) <
1114                 mp->m_dir_magicpct;
1115         return 0;
1116 }
1117
1118 /*
1119  * Split the leaf entries in the old block into old and new blocks.
1120  */
1121 int                                             /* error */
1122 xfs_dir2_leafn_split(
1123         xfs_da_state_t          *state,         /* btree cursor */
1124         xfs_da_state_blk_t      *oldblk,        /* original block */
1125         xfs_da_state_blk_t      *newblk)        /* newly created block */
1126 {
1127         xfs_da_args_t           *args;          /* operation arguments */
1128         xfs_dablk_t             blkno;          /* new leaf block number */
1129         int                     error;          /* error return value */
1130         xfs_mount_t             *mp;            /* filesystem mount point */
1131
1132         /*
1133          * Allocate space for a new leaf node.
1134          */
1135         args = state->args;
1136         mp = args->dp->i_mount;
1137         ASSERT(args != NULL);
1138         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1139         error = xfs_da_grow_inode(args, &blkno);
1140         if (error) {
1141                 return error;
1142         }
1143         /*
1144          * Initialize the new leaf block.
1145          */
1146         error = xfs_dir2_leaf_init(args, xfs_dir2_da_to_db(mp, blkno),
1147                 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1148         if (error) {
1149                 return error;
1150         }
1151         newblk->blkno = blkno;
1152         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1153         /*
1154          * Rebalance the entries across the two leaves, link the new
1155          * block into the leaves.
1156          */
1157         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1158         error = xfs_da_blk_link(state, oldblk, newblk);
1159         if (error) {
1160                 return error;
1161         }
1162         /*
1163          * Insert the new entry in the correct block.
1164          */
1165         if (state->inleaf)
1166                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1167         else
1168                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1169         /*
1170          * Update last hashval in each block since we added the name.
1171          */
1172         oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1173         newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1174         xfs_dir2_leafn_check(args->dp, oldblk->bp);
1175         xfs_dir2_leafn_check(args->dp, newblk->bp);
1176         return error;
1177 }
1178
1179 /*
1180  * Check a leaf block and its neighbors to see if the block should be
1181  * collapsed into one or the other neighbor.  Always keep the block
1182  * with the smaller block number.
1183  * If the current block is over 50% full, don't try to join it, return 0.
1184  * If the block is empty, fill in the state structure and return 2.
1185  * If it can be collapsed, fill in the state structure and return 1.
1186  * If nothing can be done, return 0.
1187  */
1188 int                                             /* error */
1189 xfs_dir2_leafn_toosmall(
1190         xfs_da_state_t          *state,         /* btree cursor */
1191         int                     *action)        /* resulting action to take */
1192 {
1193         xfs_da_state_blk_t      *blk;           /* leaf block */
1194         xfs_dablk_t             blkno;          /* leaf block number */
1195         xfs_dabuf_t             *bp;            /* leaf buffer */
1196         int                     bytes;          /* bytes in use */
1197         int                     count;          /* leaf live entry count */
1198         int                     error;          /* error return value */
1199         int                     forward;        /* sibling block direction */
1200         int                     i;              /* sibling counter */
1201         xfs_da_blkinfo_t        *info;          /* leaf block header */
1202         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1203         int                     rval;           /* result from path_shift */
1204
1205         /*
1206          * Check for the degenerate case of the block being over 50% full.
1207          * If so, it's not worth even looking to see if we might be able
1208          * to coalesce with a sibling.
1209          */
1210         blk = &state->path.blk[state->path.active - 1];
1211         info = blk->bp->data;
1212         ASSERT(be16_to_cpu(info->magic) == XFS_DIR2_LEAFN_MAGIC);
1213         leaf = (xfs_dir2_leaf_t *)info;
1214         count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1215         bytes = (uint)sizeof(leaf->hdr) + count * (uint)sizeof(leaf->ents[0]);
1216         if (bytes > (state->blocksize >> 1)) {
1217                 /*
1218                  * Blk over 50%, don't try to join.
1219                  */
1220                 *action = 0;
1221                 return 0;
1222         }
1223         /*
1224          * Check for the degenerate case of the block being empty.
1225          * If the block is empty, we'll simply delete it, no need to
1226          * coalesce it with a sibling block.  We choose (arbitrarily)
1227          * to merge with the forward block unless it is NULL.
1228          */
1229         if (count == 0) {
1230                 /*
1231                  * Make altpath point to the block we want to keep and
1232                  * path point to the block we want to drop (this one).
1233                  */
1234                 forward = (info->forw != 0);
1235                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1236                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1237                         &rval);
1238                 if (error)
1239                         return error;
1240                 *action = rval ? 2 : 0;
1241                 return 0;
1242         }
1243         /*
1244          * Examine each sibling block to see if we can coalesce with
1245          * at least 25% free space to spare.  We need to figure out
1246          * whether to merge with the forward or the backward block.
1247          * We prefer coalescing with the lower numbered sibling so as
1248          * to shrink a directory over time.
1249          */
1250         forward = be32_to_cpu(info->forw) < be32_to_cpu(info->back);
1251         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1252                 blkno = forward ? be32_to_cpu(info->forw) : be32_to_cpu(info->back);
1253                 if (blkno == 0)
1254                         continue;
1255                 /*
1256                  * Read the sibling leaf block.
1257                  */
1258                 if ((error =
1259                     xfs_da_read_buf(state->args->trans, state->args->dp, blkno,
1260                             -1, &bp, XFS_DATA_FORK))) {
1261                         return error;
1262                 }
1263                 ASSERT(bp != NULL);
1264                 /*
1265                  * Count bytes in the two blocks combined.
1266                  */
1267                 leaf = (xfs_dir2_leaf_t *)info;
1268                 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1269                 bytes = state->blocksize - (state->blocksize >> 2);
1270                 leaf = bp->data;
1271                 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1272                 count += be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1273                 bytes -= count * (uint)sizeof(leaf->ents[0]);
1274                 /*
1275                  * Fits with at least 25% to spare.
1276                  */
1277                 if (bytes >= 0)
1278                         break;
1279                 xfs_da_brelse(state->args->trans, bp);
1280         }
1281         /*
1282          * Didn't like either block, give up.
1283          */
1284         if (i >= 2) {
1285                 *action = 0;
1286                 return 0;
1287         }
1288         /*
1289          * Done with the sibling leaf block here, drop the dabuf
1290          * so path_shift can get it.
1291          */
1292         xfs_da_buf_done(bp);
1293         /*
1294          * Make altpath point to the block we want to keep (the lower
1295          * numbered block) and path point to the block we want to drop.
1296          */
1297         memcpy(&state->altpath, &state->path, sizeof(state->path));
1298         if (blkno < blk->blkno)
1299                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1300                         &rval);
1301         else
1302                 error = xfs_da_path_shift(state, &state->path, forward, 0,
1303                         &rval);
1304         if (error) {
1305                 return error;
1306         }
1307         *action = rval ? 0 : 1;
1308         return 0;
1309 }
1310
1311 /*
1312  * Move all the leaf entries from drop_blk to save_blk.
1313  * This is done as part of a join operation.
1314  */
1315 void
1316 xfs_dir2_leafn_unbalance(
1317         xfs_da_state_t          *state,         /* cursor */
1318         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1319         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1320 {
1321         xfs_da_args_t           *args;          /* operation arguments */
1322         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1323         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1324
1325         args = state->args;
1326         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1327         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1328         drop_leaf = drop_blk->bp->data;
1329         save_leaf = save_blk->bp->data;
1330         ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1331         ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1332         /*
1333          * If there are any stale leaf entries, take this opportunity
1334          * to purge them.
1335          */
1336         if (drop_leaf->hdr.stale)
1337                 xfs_dir2_leaf_compact(args, drop_blk->bp);
1338         if (save_leaf->hdr.stale)
1339                 xfs_dir2_leaf_compact(args, save_blk->bp);
1340         /*
1341          * Move the entries from drop to the appropriate end of save.
1342          */
1343         drop_blk->hashval = be32_to_cpu(drop_leaf->ents[be16_to_cpu(drop_leaf->hdr.count) - 1].hashval);
1344         if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1345                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp, 0,
1346                         be16_to_cpu(drop_leaf->hdr.count));
1347         else
1348                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp,
1349                         be16_to_cpu(save_leaf->hdr.count), be16_to_cpu(drop_leaf->hdr.count));
1350         save_blk->hashval = be32_to_cpu(save_leaf->ents[be16_to_cpu(save_leaf->hdr.count) - 1].hashval);
1351         xfs_dir2_leafn_check(args->dp, save_blk->bp);
1352 }
1353
1354 /*
1355  * Top-level node form directory addname routine.
1356  */
1357 int                                             /* error */
1358 xfs_dir2_node_addname(
1359         xfs_da_args_t           *args)          /* operation arguments */
1360 {
1361         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1362         int                     error;          /* error return value */
1363         int                     rval;           /* sub-return value */
1364         xfs_da_state_t          *state;         /* btree cursor */
1365
1366         trace_xfs_dir2_node_addname(args);
1367
1368         /*
1369          * Allocate and initialize the state (btree cursor).
1370          */
1371         state = xfs_da_state_alloc();
1372         state->args = args;
1373         state->mp = args->dp->i_mount;
1374         state->blocksize = state->mp->m_dirblksize;
1375         state->node_ents = state->mp->m_dir_node_ents;
1376         /*
1377          * Look up the name.  We're not supposed to find it, but
1378          * this gives us the insertion point.
1379          */
1380         error = xfs_da_node_lookup_int(state, &rval);
1381         if (error)
1382                 rval = error;
1383         if (rval != ENOENT) {
1384                 goto done;
1385         }
1386         /*
1387          * Add the data entry to a data block.
1388          * Extravalid is set to a freeblock found by lookup.
1389          */
1390         rval = xfs_dir2_node_addname_int(args,
1391                 state->extravalid ? &state->extrablk : NULL);
1392         if (rval) {
1393                 goto done;
1394         }
1395         blk = &state->path.blk[state->path.active - 1];
1396         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1397         /*
1398          * Add the new leaf entry.
1399          */
1400         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1401         if (rval == 0) {
1402                 /*
1403                  * It worked, fix the hash values up the btree.
1404                  */
1405                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1406                         xfs_da_fixhashpath(state, &state->path);
1407         } else {
1408                 /*
1409                  * It didn't work, we need to split the leaf block.
1410                  */
1411                 if (args->total == 0) {
1412                         ASSERT(rval == ENOSPC);
1413                         goto done;
1414                 }
1415                 /*
1416                  * Split the leaf block and insert the new entry.
1417                  */
1418                 rval = xfs_da_split(state);
1419         }
1420 done:
1421         xfs_da_state_free(state);
1422         return rval;
1423 }
1424
1425 /*
1426  * Add the data entry for a node-format directory name addition.
1427  * The leaf entry is added in xfs_dir2_leafn_add.
1428  * We may enter with a freespace block that the lookup found.
1429  */
1430 static int                                      /* error */
1431 xfs_dir2_node_addname_int(
1432         xfs_da_args_t           *args,          /* operation arguments */
1433         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1434 {
1435         xfs_dir2_data_t         *data;          /* data block structure */
1436         xfs_dir2_db_t           dbno;           /* data block number */
1437         xfs_dabuf_t             *dbp;           /* data block buffer */
1438         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1439         xfs_inode_t             *dp;            /* incore directory inode */
1440         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1441         int                     error;          /* error return value */
1442         xfs_dir2_db_t           fbno;           /* freespace block number */
1443         xfs_dabuf_t             *fbp;           /* freespace buffer */
1444         int                     findex;         /* freespace entry index */
1445         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1446         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1447         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1448         int                     length;         /* length of the new entry */
1449         int                     logfree;        /* need to log free entry */
1450         xfs_mount_t             *mp;            /* filesystem mount point */
1451         int                     needlog;        /* need to log data header */
1452         int                     needscan;       /* need to rescan data frees */
1453         __be16                  *tagp;          /* data entry tag pointer */
1454         xfs_trans_t             *tp;            /* transaction pointer */
1455
1456         dp = args->dp;
1457         mp = dp->i_mount;
1458         tp = args->trans;
1459         length = xfs_dir2_data_entsize(args->namelen);
1460         /*
1461          * If we came in with a freespace block that means that lookup
1462          * found an entry with our hash value.  This is the freespace
1463          * block for that data entry.
1464          */
1465         if (fblk) {
1466                 fbp = fblk->bp;
1467                 /*
1468                  * Remember initial freespace block number.
1469                  */
1470                 ifbno = fblk->blkno;
1471                 free = fbp->data;
1472                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1473                 findex = fblk->index;
1474                 /*
1475                  * This means the free entry showed that the data block had
1476                  * space for our entry, so we remembered it.
1477                  * Use that data block.
1478                  */
1479                 if (findex >= 0) {
1480                         ASSERT(findex < be32_to_cpu(free->hdr.nvalid));
1481                         ASSERT(be16_to_cpu(free->bests[findex]) != NULLDATAOFF);
1482                         ASSERT(be16_to_cpu(free->bests[findex]) >= length);
1483                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1484                 }
1485                 /*
1486                  * The data block looked at didn't have enough room.
1487                  * We'll start at the beginning of the freespace entries.
1488                  */
1489                 else {
1490                         dbno = -1;
1491                         findex = 0;
1492                 }
1493         }
1494         /*
1495          * Didn't come in with a freespace block, so don't have a data block.
1496          */
1497         else {
1498                 ifbno = dbno = -1;
1499                 fbp = NULL;
1500                 findex = 0;
1501         }
1502         /*
1503          * If we don't have a data block yet, we're going to scan the
1504          * freespace blocks looking for one.  Figure out what the
1505          * highest freespace block number is.
1506          */
1507         if (dbno == -1) {
1508                 xfs_fileoff_t   fo;             /* freespace block number */
1509
1510                 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1511                         return error;
1512                 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1513                 fbno = ifbno;
1514         }
1515         /*
1516          * While we haven't identified a data block, search the freeblock
1517          * data for a good data block.  If we find a null freeblock entry,
1518          * indicating a hole in the data blocks, remember that.
1519          */
1520         while (dbno == -1) {
1521                 /*
1522                  * If we don't have a freeblock in hand, get the next one.
1523                  */
1524                 if (fbp == NULL) {
1525                         /*
1526                          * Happens the first time through unless lookup gave
1527                          * us a freespace block to start with.
1528                          */
1529                         if (++fbno == 0)
1530                                 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1531                         /*
1532                          * If it's ifbno we already looked at it.
1533                          */
1534                         if (fbno == ifbno)
1535                                 fbno++;
1536                         /*
1537                          * If it's off the end we're done.
1538                          */
1539                         if (fbno >= lastfbno)
1540                                 break;
1541                         /*
1542                          * Read the block.  There can be holes in the
1543                          * freespace blocks, so this might not succeed.
1544                          * This should be really rare, so there's no reason
1545                          * to avoid it.
1546                          */
1547                         if ((error = xfs_da_read_buf(tp, dp,
1548                                         xfs_dir2_db_to_da(mp, fbno), -2, &fbp,
1549                                         XFS_DATA_FORK))) {
1550                                 return error;
1551                         }
1552                         if (unlikely(fbp == NULL)) {
1553                                 continue;
1554                         }
1555                         free = fbp->data;
1556                         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1557                         findex = 0;
1558                 }
1559                 /*
1560                  * Look at the current free entry.  Is it good enough?
1561                  */
1562                 if (be16_to_cpu(free->bests[findex]) != NULLDATAOFF &&
1563                     be16_to_cpu(free->bests[findex]) >= length)
1564                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1565                 else {
1566                         /*
1567                          * Are we done with the freeblock?
1568                          */
1569                         if (++findex == be32_to_cpu(free->hdr.nvalid)) {
1570                                 /*
1571                                  * Drop the block.
1572                                  */
1573                                 xfs_da_brelse(tp, fbp);
1574                                 fbp = NULL;
1575                                 if (fblk && fblk->bp)
1576                                         fblk->bp = NULL;
1577                         }
1578                 }
1579         }
1580         /*
1581          * If we don't have a data block, we need to allocate one and make
1582          * the freespace entries refer to it.
1583          */
1584         if (unlikely(dbno == -1)) {
1585                 /*
1586                  * Not allowed to allocate, return failure.
1587                  */
1588                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
1589                                                         args->total == 0) {
1590                         /*
1591                          * Drop the freespace buffer unless it came from our
1592                          * caller.
1593                          */
1594                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1595                                 xfs_da_buf_done(fbp);
1596                         return XFS_ERROR(ENOSPC);
1597                 }
1598                 /*
1599                  * Allocate and initialize the new data block.
1600                  */
1601                 if (unlikely((error = xfs_dir2_grow_inode(args,
1602                                                          XFS_DIR2_DATA_SPACE,
1603                                                          &dbno)) ||
1604                     (error = xfs_dir2_data_init(args, dbno, &dbp)))) {
1605                         /*
1606                          * Drop the freespace buffer unless it came from our
1607                          * caller.
1608                          */
1609                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1610                                 xfs_da_buf_done(fbp);
1611                         return error;
1612                 }
1613                 /*
1614                  * If (somehow) we have a freespace block, get rid of it.
1615                  */
1616                 if (fbp)
1617                         xfs_da_brelse(tp, fbp);
1618                 if (fblk && fblk->bp)
1619                         fblk->bp = NULL;
1620
1621                 /*
1622                  * Get the freespace block corresponding to the data block
1623                  * that was just allocated.
1624                  */
1625                 fbno = xfs_dir2_db_to_fdb(mp, dbno);
1626                 if (unlikely(error = xfs_da_read_buf(tp, dp,
1627                                 xfs_dir2_db_to_da(mp, fbno), -2, &fbp,
1628                                 XFS_DATA_FORK))) {
1629                         xfs_da_buf_done(dbp);
1630                         return error;
1631                 }
1632                 /*
1633                  * If there wasn't a freespace block, the read will
1634                  * return a NULL fbp.  Allocate and initialize a new one.
1635                  */
1636                 if( fbp == NULL ) {
1637                         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1638                                                         &fbno))) {
1639                                 return error;
1640                         }
1641
1642                         if (unlikely(xfs_dir2_db_to_fdb(mp, dbno) != fbno)) {
1643                                 xfs_alert(mp,
1644                         "%s: dir ino " "%llu needed freesp block %lld for\n"
1645                         "  data block %lld, got %lld ifbno %llu lastfbno %d",
1646                                         __func__, (unsigned long long)dp->i_ino,
1647                                         (long long)xfs_dir2_db_to_fdb(mp, dbno),
1648                                         (long long)dbno, (long long)fbno,
1649                                         (unsigned long long)ifbno, lastfbno);
1650                                 if (fblk) {
1651                                         xfs_alert(mp,
1652                                 " fblk 0x%p blkno %llu index %d magic 0x%x",
1653                                                 fblk,
1654                                                 (unsigned long long)fblk->blkno,
1655                                                 fblk->index,
1656                                                 fblk->magic);
1657                                 } else {
1658                                         xfs_alert(mp, " ... fblk is NULL");
1659                                 }
1660                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1661                                                  XFS_ERRLEVEL_LOW, mp);
1662                                 return XFS_ERROR(EFSCORRUPTED);
1663                         }
1664
1665                         /*
1666                          * Get a buffer for the new block.
1667                          */
1668                         if ((error = xfs_da_get_buf(tp, dp,
1669                                                    xfs_dir2_db_to_da(mp, fbno),
1670                                                    -1, &fbp, XFS_DATA_FORK))) {
1671                                 return error;
1672                         }
1673                         ASSERT(fbp != NULL);
1674
1675                         /*
1676                          * Initialize the new block to be empty, and remember
1677                          * its first slot as our empty slot.
1678                          */
1679                         free = fbp->data;
1680                         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
1681                         free->hdr.firstdb = cpu_to_be32(
1682                                 (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1683                                 XFS_DIR2_MAX_FREE_BESTS(mp));
1684                         free->hdr.nvalid = 0;
1685                         free->hdr.nused = 0;
1686                 } else {
1687                         free = fbp->data;
1688                         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1689                 }
1690
1691                 /*
1692                  * Set the freespace block index from the data block number.
1693                  */
1694                 findex = xfs_dir2_db_to_fdindex(mp, dbno);
1695                 /*
1696                  * If it's after the end of the current entries in the
1697                  * freespace block, extend that table.
1698                  */
1699                 if (findex >= be32_to_cpu(free->hdr.nvalid)) {
1700                         ASSERT(findex < XFS_DIR2_MAX_FREE_BESTS(mp));
1701                         free->hdr.nvalid = cpu_to_be32(findex + 1);
1702                         /*
1703                          * Tag new entry so nused will go up.
1704                          */
1705                         free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1706                 }
1707                 /*
1708                  * If this entry was for an empty data block
1709                  * (this should always be true) then update the header.
1710                  */
1711                 if (be16_to_cpu(free->bests[findex]) == NULLDATAOFF) {
1712                         be32_add_cpu(&free->hdr.nused, 1);
1713                         xfs_dir2_free_log_header(tp, fbp);
1714                 }
1715                 /*
1716                  * Update the real value in the table.
1717                  * We haven't allocated the data entry yet so this will
1718                  * change again.
1719                  */
1720                 data = dbp->data;
1721                 free->bests[findex] = data->hdr.bestfree[0].length;
1722                 logfree = 1;
1723         }
1724         /*
1725          * We had a data block so we don't have to make a new one.
1726          */
1727         else {
1728                 /*
1729                  * If just checking, we succeeded.
1730                  */
1731                 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
1732                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1733                                 xfs_da_buf_done(fbp);
1734                         return 0;
1735                 }
1736                 /*
1737                  * Read the data block in.
1738                  */
1739                 if (unlikely(
1740                     error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1741                                 -1, &dbp, XFS_DATA_FORK))) {
1742                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1743                                 xfs_da_buf_done(fbp);
1744                         return error;
1745                 }
1746                 data = dbp->data;
1747                 logfree = 0;
1748         }
1749         ASSERT(be16_to_cpu(data->hdr.bestfree[0].length) >= length);
1750         /*
1751          * Point to the existing unused space.
1752          */
1753         dup = (xfs_dir2_data_unused_t *)
1754               ((char *)data + be16_to_cpu(data->hdr.bestfree[0].offset));
1755         needscan = needlog = 0;
1756         /*
1757          * Mark the first part of the unused space, inuse for us.
1758          */
1759         xfs_dir2_data_use_free(tp, dbp, dup,
1760                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)data), length,
1761                 &needlog, &needscan);
1762         /*
1763          * Fill in the new entry and log it.
1764          */
1765         dep = (xfs_dir2_data_entry_t *)dup;
1766         dep->inumber = cpu_to_be64(args->inumber);
1767         dep->namelen = args->namelen;
1768         memcpy(dep->name, args->name, dep->namelen);
1769         tagp = xfs_dir2_data_entry_tag_p(dep);
1770         *tagp = cpu_to_be16((char *)dep - (char *)data);
1771         xfs_dir2_data_log_entry(tp, dbp, dep);
1772         /*
1773          * Rescan the block for bestfree if needed.
1774          */
1775         if (needscan)
1776                 xfs_dir2_data_freescan(mp, data, &needlog);
1777         /*
1778          * Log the data block header if needed.
1779          */
1780         if (needlog)
1781                 xfs_dir2_data_log_header(tp, dbp);
1782         /*
1783          * If the freespace entry is now wrong, update it.
1784          */
1785         if (be16_to_cpu(free->bests[findex]) != be16_to_cpu(data->hdr.bestfree[0].length)) {
1786                 free->bests[findex] = data->hdr.bestfree[0].length;
1787                 logfree = 1;
1788         }
1789         /*
1790          * Log the freespace entry if needed.
1791          */
1792         if (logfree)
1793                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1794         /*
1795          * If the caller didn't hand us the freespace block, drop it.
1796          */
1797         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1798                 xfs_da_buf_done(fbp);
1799         /*
1800          * Return the data block and offset in args, then drop the data block.
1801          */
1802         args->blkno = (xfs_dablk_t)dbno;
1803         args->index = be16_to_cpu(*tagp);
1804         xfs_da_buf_done(dbp);
1805         return 0;
1806 }
1807
1808 /*
1809  * Lookup an entry in a node-format directory.
1810  * All the real work happens in xfs_da_node_lookup_int.
1811  * The only real output is the inode number of the entry.
1812  */
1813 int                                             /* error */
1814 xfs_dir2_node_lookup(
1815         xfs_da_args_t   *args)                  /* operation arguments */
1816 {
1817         int             error;                  /* error return value */
1818         int             i;                      /* btree level */
1819         int             rval;                   /* operation return value */
1820         xfs_da_state_t  *state;                 /* btree cursor */
1821
1822         trace_xfs_dir2_node_lookup(args);
1823
1824         /*
1825          * Allocate and initialize the btree cursor.
1826          */
1827         state = xfs_da_state_alloc();
1828         state->args = args;
1829         state->mp = args->dp->i_mount;
1830         state->blocksize = state->mp->m_dirblksize;
1831         state->node_ents = state->mp->m_dir_node_ents;
1832         /*
1833          * Fill in the path to the entry in the cursor.
1834          */
1835         error = xfs_da_node_lookup_int(state, &rval);
1836         if (error)
1837                 rval = error;
1838         else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
1839                 /* If a CI match, dup the actual name and return EEXIST */
1840                 xfs_dir2_data_entry_t   *dep;
1841
1842                 dep = (xfs_dir2_data_entry_t *)((char *)state->extrablk.bp->
1843                                                 data + state->extrablk.index);
1844                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1845         }
1846         /*
1847          * Release the btree blocks and leaf block.
1848          */
1849         for (i = 0; i < state->path.active; i++) {
1850                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1851                 state->path.blk[i].bp = NULL;
1852         }
1853         /*
1854          * Release the data block if we have it.
1855          */
1856         if (state->extravalid && state->extrablk.bp) {
1857                 xfs_da_brelse(args->trans, state->extrablk.bp);
1858                 state->extrablk.bp = NULL;
1859         }
1860         xfs_da_state_free(state);
1861         return rval;
1862 }
1863
1864 /*
1865  * Remove an entry from a node-format directory.
1866  */
1867 int                                             /* error */
1868 xfs_dir2_node_removename(
1869         xfs_da_args_t           *args)          /* operation arguments */
1870 {
1871         xfs_da_state_blk_t      *blk;           /* leaf block */
1872         int                     error;          /* error return value */
1873         int                     rval;           /* operation return value */
1874         xfs_da_state_t          *state;         /* btree cursor */
1875
1876         trace_xfs_dir2_node_removename(args);
1877
1878         /*
1879          * Allocate and initialize the btree cursor.
1880          */
1881         state = xfs_da_state_alloc();
1882         state->args = args;
1883         state->mp = args->dp->i_mount;
1884         state->blocksize = state->mp->m_dirblksize;
1885         state->node_ents = state->mp->m_dir_node_ents;
1886         /*
1887          * Look up the entry we're deleting, set up the cursor.
1888          */
1889         error = xfs_da_node_lookup_int(state, &rval);
1890         if (error)
1891                 rval = error;
1892         /*
1893          * Didn't find it, upper layer screwed up.
1894          */
1895         if (rval != EEXIST) {
1896                 xfs_da_state_free(state);
1897                 return rval;
1898         }
1899         blk = &state->path.blk[state->path.active - 1];
1900         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1901         ASSERT(state->extravalid);
1902         /*
1903          * Remove the leaf and data entries.
1904          * Extrablk refers to the data block.
1905          */
1906         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
1907                 &state->extrablk, &rval);
1908         if (error)
1909                 return error;
1910         /*
1911          * Fix the hash values up the btree.
1912          */
1913         xfs_da_fixhashpath(state, &state->path);
1914         /*
1915          * If we need to join leaf blocks, do it.
1916          */
1917         if (rval && state->path.active > 1)
1918                 error = xfs_da_join(state);
1919         /*
1920          * If no errors so far, try conversion to leaf format.
1921          */
1922         if (!error)
1923                 error = xfs_dir2_node_to_leaf(state);
1924         xfs_da_state_free(state);
1925         return error;
1926 }
1927
1928 /*
1929  * Replace an entry's inode number in a node-format directory.
1930  */
1931 int                                             /* error */
1932 xfs_dir2_node_replace(
1933         xfs_da_args_t           *args)          /* operation arguments */
1934 {
1935         xfs_da_state_blk_t      *blk;           /* leaf block */
1936         xfs_dir2_data_t         *data;          /* data block structure */
1937         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
1938         int                     error;          /* error return value */
1939         int                     i;              /* btree level */
1940         xfs_ino_t               inum;           /* new inode number */
1941         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1942         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
1943         int                     rval;           /* internal return value */
1944         xfs_da_state_t          *state;         /* btree cursor */
1945
1946         trace_xfs_dir2_node_replace(args);
1947
1948         /*
1949          * Allocate and initialize the btree cursor.
1950          */
1951         state = xfs_da_state_alloc();
1952         state->args = args;
1953         state->mp = args->dp->i_mount;
1954         state->blocksize = state->mp->m_dirblksize;
1955         state->node_ents = state->mp->m_dir_node_ents;
1956         inum = args->inumber;
1957         /*
1958          * Lookup the entry to change in the btree.
1959          */
1960         error = xfs_da_node_lookup_int(state, &rval);
1961         if (error) {
1962                 rval = error;
1963         }
1964         /*
1965          * It should be found, since the vnodeops layer has looked it up
1966          * and locked it.  But paranoia is good.
1967          */
1968         if (rval == EEXIST) {
1969                 /*
1970                  * Find the leaf entry.
1971                  */
1972                 blk = &state->path.blk[state->path.active - 1];
1973                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1974                 leaf = blk->bp->data;
1975                 lep = &leaf->ents[blk->index];
1976                 ASSERT(state->extravalid);
1977                 /*
1978                  * Point to the data entry.
1979                  */
1980                 data = state->extrablk.bp->data;
1981                 ASSERT(be32_to_cpu(data->hdr.magic) == XFS_DIR2_DATA_MAGIC);
1982                 dep = (xfs_dir2_data_entry_t *)
1983                       ((char *)data +
1984                        xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
1985                 ASSERT(inum != be64_to_cpu(dep->inumber));
1986                 /*
1987                  * Fill in the new inode number and log the entry.
1988                  */
1989                 dep->inumber = cpu_to_be64(inum);
1990                 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
1991                 rval = 0;
1992         }
1993         /*
1994          * Didn't find it, and we're holding a data block.  Drop it.
1995          */
1996         else if (state->extravalid) {
1997                 xfs_da_brelse(args->trans, state->extrablk.bp);
1998                 state->extrablk.bp = NULL;
1999         }
2000         /*
2001          * Release all the buffers in the cursor.
2002          */
2003         for (i = 0; i < state->path.active; i++) {
2004                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
2005                 state->path.blk[i].bp = NULL;
2006         }
2007         xfs_da_state_free(state);
2008         return rval;
2009 }
2010
2011 /*
2012  * Trim off a trailing empty freespace block.
2013  * Return (in rvalp) 1 if we did it, 0 if not.
2014  */
2015 int                                             /* error */
2016 xfs_dir2_node_trim_free(
2017         xfs_da_args_t           *args,          /* operation arguments */
2018         xfs_fileoff_t           fo,             /* free block number */
2019         int                     *rvalp)         /* out: did something */
2020 {
2021         xfs_dabuf_t             *bp;            /* freespace buffer */
2022         xfs_inode_t             *dp;            /* incore directory inode */
2023         int                     error;          /* error return code */
2024         xfs_dir2_free_t         *free;          /* freespace structure */
2025         xfs_mount_t             *mp;            /* filesystem mount point */
2026         xfs_trans_t             *tp;            /* transaction pointer */
2027
2028         dp = args->dp;
2029         mp = dp->i_mount;
2030         tp = args->trans;
2031         /*
2032          * Read the freespace block.
2033          */
2034         if (unlikely(error = xfs_da_read_buf(tp, dp, (xfs_dablk_t)fo, -2, &bp,
2035                         XFS_DATA_FORK))) {
2036                 return error;
2037         }
2038
2039         /*
2040          * There can be holes in freespace.  If fo is a hole, there's
2041          * nothing to do.
2042          */
2043         if (bp == NULL) {
2044                 return 0;
2045         }
2046         free = bp->data;
2047         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
2048         /*
2049          * If there are used entries, there's nothing to do.
2050          */
2051         if (be32_to_cpu(free->hdr.nused) > 0) {
2052                 xfs_da_brelse(tp, bp);
2053                 *rvalp = 0;
2054                 return 0;
2055         }
2056         /*
2057          * Blow the block away.
2058          */
2059         if ((error =
2060             xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
2061                     bp))) {
2062                 /*
2063                  * Can't fail with ENOSPC since that only happens with no
2064                  * space reservation, when breaking up an extent into two
2065                  * pieces.  This is the last block of an extent.
2066                  */
2067                 ASSERT(error != ENOSPC);
2068                 xfs_da_brelse(tp, bp);
2069                 return error;
2070         }
2071         /*
2072          * Return that we succeeded.
2073          */
2074         *rvalp = 1;
2075         return 0;
2076 }