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_trans.h"
  23 #include "xfs_sb.h"
  24 #include "xfs_ag.h"
  25 #include "xfs_mount.h"
  26 #include "xfs_da_btree.h"
  27 #include "xfs_bmap_btree.h"
  28 #include "xfs_dinode.h"
  29 #include "xfs_inode.h"
  30 #include "xfs_bmap.h"
  31 #include "xfs_dir2_format.h"
  32 #include "xfs_dir2_priv.h"
  33 #include "xfs_error.h"
  34 #include "xfs_trace.h"
  35
  36 /*
  37  * Function declarations.
  38  */
  39 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
  40                               int index);
  41 #ifdef DEBUG
  42 static void xfs_dir2_leafn_check(struct xfs_inode *dp, struct xfs_buf *bp);
  43 #else
  44 #define xfs_dir2_leafn_check(dp, bp)
  45 #endif
  46 static void xfs_dir2_leafn_moveents(xfs_da_args_t *args, struct xfs_buf *bp_s,
  47                                     int start_s, struct xfs_buf *bp_d,
  48                                     int start_d, int count);
  49 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  50                                      xfs_da_state_blk_t *blk1,
  51                                      xfs_da_state_blk_t *blk2);
  52 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
  53                                  int index, xfs_da_state_blk_t *dblk,
  54                                  int *rval);
  55 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  56                                      xfs_da_state_blk_t *fblk);
  57
  58 static void
  59 xfs_dir2_free_verify(
  60         struct xfs_buf          *bp)
  61 {
  62         struct xfs_mount        *mp = bp->b_target->bt_mount;
  63         struct xfs_dir2_free_hdr *hdr = bp->b_addr;
  64         int                     block_ok = 0;
  65
  66         block_ok = hdr->magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC);
  67         if (!block_ok) {
  68                 XFS_CORRUPTION_ERROR("xfs_dir2_free_verify magic",
  69                                      XFS_ERRLEVEL_LOW, mp, hdr);
  70                 xfs_buf_ioerror(bp, EFSCORRUPTED);
  71         }
  72
  73         bp->b_iodone = NULL;
  74         xfs_buf_ioend(bp, 0);
  75 }
  76
  77 static int
  78 __xfs_dir2_free_read(
  79         struct xfs_trans        *tp,
  80         struct xfs_inode        *dp,
  81         xfs_dablk_t             fbno,
  82         xfs_daddr_t             mappedbno,
  83         struct xfs_buf          **bpp)
  84 {
  85         return xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
  86                                         XFS_DATA_FORK, xfs_dir2_free_verify);
  87 }
  88
  89 int
  90 xfs_dir2_free_read(
  91         struct xfs_trans        *tp,
  92         struct xfs_inode        *dp,
  93         xfs_dablk_t             fbno,
  94         struct xfs_buf          **bpp)
  95 {
  96         return __xfs_dir2_free_read(tp, dp, fbno, -1, bpp);
  97 }
  98
  99 static int
 100 xfs_dir2_free_try_read(
 101         struct xfs_trans        *tp,
 102         struct xfs_inode        *dp,
 103         xfs_dablk_t             fbno,
 104         struct xfs_buf          **bpp)
 105 {
 106         return __xfs_dir2_free_read(tp, dp, fbno, -2, bpp);
 107 }
 108
 109 /*
 110  * Log entries from a freespace block.
 111  */
 112 STATIC void
 113 xfs_dir2_free_log_bests(
 114         struct xfs_trans        *tp,
 115         struct xfs_buf          *bp,
 116         int                     first,          /* first entry to log */
 117         int                     last)           /* last entry to log */
 118 {
 119         xfs_dir2_free_t         *free;          /* freespace structure */
 120
 121         free = bp->b_addr;
 122         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
 123         xfs_trans_log_buf(tp, bp,
 124                 (uint)((char *)&free->bests[first] - (char *)free),
 125                 (uint)((char *)&free->bests[last] - (char *)free +
 126                        sizeof(free->bests[0]) - 1));
 127 }
 128
 129 /*
 130  * Log header from a freespace block.
 131  */
 132 static void
 133 xfs_dir2_free_log_header(
 134         struct xfs_trans        *tp,
 135         struct xfs_buf          *bp)
 136 {
 137         xfs_dir2_free_t         *free;          /* freespace structure */
 138
 139         free = bp->b_addr;
 140         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
 141         xfs_trans_log_buf(tp, bp, (uint)((char *)&free->hdr - (char *)free),
 142                 (uint)(sizeof(xfs_dir2_free_hdr_t) - 1));
 143 }
 144
 145 /*
 146  * Convert a leaf-format directory to a node-format directory.
 147  * We need to change the magic number of the leaf block, and copy
 148  * the freespace table out of the leaf block into its own block.
 149  */
 150 int                                             /* error */
 151 xfs_dir2_leaf_to_node(
 152         xfs_da_args_t           *args,          /* operation arguments */
 153         struct xfs_buf          *lbp)           /* leaf buffer */
 154 {
 155         xfs_inode_t             *dp;            /* incore directory inode */
 156         int                     error;          /* error return value */
 157         struct xfs_buf          *fbp;           /* freespace buffer */
 158         xfs_dir2_db_t           fdb;            /* freespace block number */
 159         xfs_dir2_free_t         *free;          /* freespace structure */
 160         __be16                  *from;          /* pointer to freespace entry */
 161         int                     i;              /* leaf freespace index */
 162         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 163         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 164         xfs_mount_t             *mp;            /* filesystem mount point */
 165         int                     n;              /* count of live freespc ents */
 166         xfs_dir2_data_off_t     off;            /* freespace entry value */
 167         __be16                  *to;            /* pointer to freespace entry */
 168         xfs_trans_t             *tp;            /* transaction pointer */
 169
 170         trace_xfs_dir2_leaf_to_node(args);
 171
 172         dp = args->dp;
 173         mp = dp->i_mount;
 174         tp = args->trans;
 175         /*
 176          * Add a freespace block to the directory.
 177          */
 178         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 179                 return error;
 180         }
 181         ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
 182         /*
 183          * Get the buffer for the new freespace block.
 184          */
 185         if ((error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb), -1, &fbp,
 186                         XFS_DATA_FORK))) {
 187                 return error;
 188         }
 189         ASSERT(fbp != NULL);
 190         free = fbp->b_addr;
 191         leaf = lbp->b_addr;
 192         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 193         /*
 194          * Initialize the freespace block header.
 195          */
 196         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
 197         free->hdr.firstdb = 0;
 198         ASSERT(be32_to_cpu(ltp->bestcount) <= (uint)dp->i_d.di_size / mp->m_dirblksize);
 199         free->hdr.nvalid = ltp->bestcount;
 200         /*
 201          * Copy freespace entries from the leaf block to the new block.
 202          * Count active entries.
 203          */
 204         for (i = n = 0, from = xfs_dir2_leaf_bests_p(ltp), to = free->bests;
 205              i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 206                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 207                         n++;
 208                 *to = cpu_to_be16(off);
 209         }
 210         free->hdr.nused = cpu_to_be32(n);
 211         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 212         /*
 213          * Log everything.
 214          */
 215         xfs_dir2_leaf_log_header(tp, lbp);
 216         xfs_dir2_free_log_header(tp, fbp);
 217         xfs_dir2_free_log_bests(tp, fbp, 0, be32_to_cpu(free->hdr.nvalid) - 1);
 218         xfs_dir2_leafn_check(dp, lbp);
 219         return 0;
 220 }
 221
 222 /*
 223  * Add a leaf entry to a leaf block in a node-form directory.
 224  * The other work necessary is done from the caller.
 225  */
 226 static int                                      /* error */
 227 xfs_dir2_leafn_add(
 228         struct xfs_buf          *bp,            /* leaf buffer */
 229         xfs_da_args_t           *args,          /* operation arguments */
 230         int                     index)          /* insertion pt for new entry */
 231 {
 232         int                     compact;        /* compacting stale leaves */
 233         xfs_inode_t             *dp;            /* incore directory inode */
 234         int                     highstale;      /* next stale entry */
 235         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 236         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 237         int                     lfloghigh;      /* high leaf entry logging */
 238         int                     lfloglow;       /* low leaf entry logging */
 239         int                     lowstale;       /* previous stale entry */
 240         xfs_mount_t             *mp;            /* filesystem mount point */
 241         xfs_trans_t             *tp;            /* transaction pointer */
 242
 243         trace_xfs_dir2_leafn_add(args, index);
 244
 245         dp = args->dp;
 246         mp = dp->i_mount;
 247         tp = args->trans;
 248         leaf = bp->b_addr;
 249
 250         /*
 251          * Quick check just to make sure we are not going to index
 252          * into other peoples memory
 253          */
 254         if (index < 0)
 255                 return XFS_ERROR(EFSCORRUPTED);
 256
 257         /*
 258          * If there are already the maximum number of leaf entries in
 259          * the block, if there are no stale entries it won't fit.
 260          * Caller will do a split.  If there are stale entries we'll do
 261          * a compact.
 262          */
 263
 264         if (be16_to_cpu(leaf->hdr.count) == xfs_dir2_max_leaf_ents(mp)) {
 265                 if (!leaf->hdr.stale)
 266                         return XFS_ERROR(ENOSPC);
 267                 compact = be16_to_cpu(leaf->hdr.stale) > 1;
 268         } else
 269                 compact = 0;
 270         ASSERT(index == 0 || be32_to_cpu(leaf->ents[index - 1].hashval) <= args->hashval);
 271         ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
 272                be32_to_cpu(leaf->ents[index].hashval) >= args->hashval);
 273
 274         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 275                 return 0;
 276
 277         /*
 278          * Compact out all but one stale leaf entry.  Leaves behind
 279          * the entry closest to index.
 280          */
 281         if (compact) {
 282                 xfs_dir2_leaf_compact_x1(bp, &index, &lowstale, &highstale,
 283                         &lfloglow, &lfloghigh);
 284         }
 285         /*
 286          * Set impossible logging indices for this case.
 287          */
 288         else if (leaf->hdr.stale) {
 289                 lfloglow = be16_to_cpu(leaf->hdr.count);
 290                 lfloghigh = -1;
 291         }
 292
 293         /*
 294          * Insert the new entry, log everything.
 295          */
 296         lep = xfs_dir2_leaf_find_entry(leaf, index, compact, lowstale,
 297                                        highstale, &lfloglow, &lfloghigh);
 298
 299         lep->hashval = cpu_to_be32(args->hashval);
 300         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
 301                                 args->blkno, args->index));
 302         xfs_dir2_leaf_log_header(tp, bp);
 303         xfs_dir2_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
 304         xfs_dir2_leafn_check(dp, bp);
 305         return 0;
 306 }
 307
 308 #ifdef DEBUG
 309 /*
 310  * Check internal consistency of a leafn block.
 311  */
 312 void
 313 xfs_dir2_leafn_check(
 314         struct xfs_inode *dp,
 315         struct xfs_buf  *bp)
 316 {
 317         int             i;                      /* leaf index */
 318         xfs_dir2_leaf_t *leaf;                  /* leaf structure */
 319         xfs_mount_t     *mp;                    /* filesystem mount point */
 320         int             stale;                  /* count of stale leaves */
 321
 322         leaf = bp->b_addr;
 323         mp = dp->i_mount;
 324         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 325         ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
 326         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
 327                 if (i + 1 < be16_to_cpu(leaf->hdr.count)) {
 328                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
 329                                be32_to_cpu(leaf->ents[i + 1].hashval));
 330                 }
 331                 if (leaf->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 332                         stale++;
 333         }
 334         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
 335 }
 336 #endif  /* DEBUG */
 337
 338 /*
 339  * Return the last hash value in the leaf.
 340  * Stale entries are ok.
 341  */
 342 xfs_dahash_t                                    /* hash value */
 343 xfs_dir2_leafn_lasthash(
 344         struct xfs_buf  *bp,                    /* leaf buffer */
 345         int             *count)                 /* count of entries in leaf */
 346 {
 347         xfs_dir2_leaf_t *leaf;                  /* leaf structure */
 348
 349         leaf = bp->b_addr;
 350         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 351         if (count)
 352                 *count = be16_to_cpu(leaf->hdr.count);
 353         if (!leaf->hdr.count)
 354                 return 0;
 355         return be32_to_cpu(leaf->ents[be16_to_cpu(leaf->hdr.count) - 1].hashval);
 356 }
 357
 358 /*
 359  * Look up a leaf entry for space to add a name in a node-format leaf block.
 360  * The extrablk in state is a freespace block.
 361  */
 362 STATIC int
 363 xfs_dir2_leafn_lookup_for_addname(
 364         struct xfs_buf          *bp,            /* leaf buffer */
 365         xfs_da_args_t           *args,          /* operation arguments */
 366         int                     *indexp,        /* out: leaf entry index */
 367         xfs_da_state_t          *state)         /* state to fill in */
 368 {
 369         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 370         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 371         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 372         xfs_inode_t             *dp;            /* incore directory inode */
 373         int                     error;          /* error return value */
 374         int                     fi;             /* free entry index */
 375         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 376         int                     index;          /* leaf entry index */
 377         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 378         int                     length;         /* length of new data entry */
 379         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 380         xfs_mount_t             *mp;            /* filesystem mount point */
 381         xfs_dir2_db_t           newdb;          /* new data block number */
 382         xfs_dir2_db_t           newfdb;         /* new free block number */
 383         xfs_trans_t             *tp;            /* transaction pointer */
 384
 385         dp = args->dp;
 386         tp = args->trans;
 387         mp = dp->i_mount;
 388         leaf = bp->b_addr;
 389         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 390 #ifdef __KERNEL__
 391         ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
 392 #endif
 393         xfs_dir2_leafn_check(dp, bp);
 394         /*
 395          * Look up the hash value in the leaf entries.
 396          */
 397         index = xfs_dir2_leaf_search_hash(args, bp);
 398         /*
 399          * Do we have a buffer coming in?
 400          */
 401         if (state->extravalid) {
 402                 /* If so, it's a free block buffer, get the block number. */
 403                 curbp = state->extrablk.bp;
 404                 curfdb = state->extrablk.blkno;
 405                 free = curbp->b_addr;
 406                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
 407         }
 408         length = xfs_dir2_data_entsize(args->namelen);
 409         /*
 410          * Loop over leaf entries with the right hash value.
 411          */
 412         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
 413                                 be32_to_cpu(lep->hashval) == args->hashval;
 414                                 lep++, index++) {
 415                 /*
 416                  * Skip stale leaf entries.
 417                  */
 418                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 419                         continue;
 420                 /*
 421                  * Pull the data block number from the entry.
 422                  */
 423                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 424                 /*
 425                  * For addname, we're looking for a place to put the new entry.
 426                  * We want to use a data block with an entry of equal
 427                  * hash value to ours if there is one with room.
 428                  *
 429                  * If this block isn't the data block we already have
 430                  * in hand, take a look at it.
 431                  */
 432                 if (newdb != curdb) {
 433                         curdb = newdb;
 434                         /*
 435                          * Convert the data block to the free block
 436                          * holding its freespace information.
 437                          */
 438                         newfdb = xfs_dir2_db_to_fdb(mp, newdb);
 439                         /*
 440                          * If it's not the one we have in hand, read it in.
 441                          */
 442                         if (newfdb != curfdb) {
 443                                 /*
 444                                  * If we had one before, drop it.
 445                                  */
 446                                 if (curbp)
 447                                         xfs_trans_brelse(tp, curbp);
 448
 449                                 error = xfs_dir2_free_read(tp, dp,
 450                                                 xfs_dir2_db_to_da(mp, newfdb),
 451                                                 &curbp);
 452                                 if (error)
 453                                         return error;
 454                                 free = curbp->b_addr;
 455                                 ASSERT(be32_to_cpu(free->hdr.magic) ==
 456                                         XFS_DIR2_FREE_MAGIC);
 457                                 ASSERT((be32_to_cpu(free->hdr.firstdb) %
 458                                         xfs_dir2_free_max_bests(mp)) == 0);
 459                                 ASSERT(be32_to_cpu(free->hdr.firstdb) <= curdb);
 460                                 ASSERT(curdb < be32_to_cpu(free->hdr.firstdb) +
 461                                         be32_to_cpu(free->hdr.nvalid));
 462                         }
 463                         /*
 464                          * Get the index for our entry.
 465                          */
 466                         fi = xfs_dir2_db_to_fdindex(mp, curdb);
 467                         /*
 468                          * If it has room, return it.
 469                          */
 470                         if (unlikely(free->bests[fi] ==
 471                             cpu_to_be16(NULLDATAOFF))) {
 472                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 473                                                         XFS_ERRLEVEL_LOW, mp);
 474                                 if (curfdb != newfdb)
 475                                         xfs_trans_brelse(tp, curbp);
 476                                 return XFS_ERROR(EFSCORRUPTED);
 477                         }
 478                         curfdb = newfdb;
 479                         if (be16_to_cpu(free->bests[fi]) >= length)
 480                                 goto out;
 481                 }
 482         }
 483         /* Didn't find any space */
 484         fi = -1;
 485 out:
 486         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 487         if (curbp) {
 488                 /* Giving back a free block. */
 489                 state->extravalid = 1;
 490                 state->extrablk.bp = curbp;
 491                 state->extrablk.index = fi;
 492                 state->extrablk.blkno = curfdb;
 493                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 494         } else {
 495                 state->extravalid = 0;
 496         }
 497         /*
 498          * Return the index, that will be the insertion point.
 499          */
 500         *indexp = index;
 501         return XFS_ERROR(ENOENT);
 502 }
 503
 504 /*
 505  * Look up a leaf entry in a node-format leaf block.
 506  * The extrablk in state a data block.
 507  */
 508 STATIC int
 509 xfs_dir2_leafn_lookup_for_entry(
 510         struct xfs_buf          *bp,            /* leaf buffer */
 511         xfs_da_args_t           *args,          /* operation arguments */
 512         int                     *indexp,        /* out: leaf entry index */
 513         xfs_da_state_t          *state)         /* state to fill in */
 514 {
 515         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 516         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 517         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 518         xfs_inode_t             *dp;            /* incore directory inode */
 519         int                     error;          /* error return value */
 520         int                     index;          /* leaf entry index */
 521         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 522         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 523         xfs_mount_t             *mp;            /* filesystem mount point */
 524         xfs_dir2_db_t           newdb;          /* new data block number */
 525         xfs_trans_t             *tp;            /* transaction pointer */
 526         enum xfs_dacmp          cmp;            /* comparison result */
 527
 528         dp = args->dp;
 529         tp = args->trans;
 530         mp = dp->i_mount;
 531         leaf = bp->b_addr;
 532         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 533 #ifdef __KERNEL__
 534         ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
 535 #endif
 536         xfs_dir2_leafn_check(dp, bp);
 537         /*
 538          * Look up the hash value in the leaf entries.
 539          */
 540         index = xfs_dir2_leaf_search_hash(args, bp);
 541         /*
 542          * Do we have a buffer coming in?
 543          */
 544         if (state->extravalid) {
 545                 curbp = state->extrablk.bp;
 546                 curdb = state->extrablk.blkno;
 547         }
 548         /*
 549          * Loop over leaf entries with the right hash value.
 550          */
 551         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
 552                                 be32_to_cpu(lep->hashval) == args->hashval;
 553                                 lep++, index++) {
 554                 /*
 555                  * Skip stale leaf entries.
 556                  */
 557                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 558                         continue;
 559                 /*
 560                  * Pull the data block number from the entry.
 561                  */
 562                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 563                 /*
 564                  * Not adding a new entry, so we really want to find
 565                  * the name given to us.
 566                  *
 567                  * If it's a different data block, go get it.
 568                  */
 569                 if (newdb != curdb) {
 570                         /*
 571                          * If we had a block before that we aren't saving
 572                          * for a CI name, drop it
 573                          */
 574                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 575                                                 curdb != state->extrablk.blkno))
 576                                 xfs_trans_brelse(tp, curbp);
 577                         /*
 578                          * If needing the block that is saved with a CI match,
 579                          * use it otherwise read in the new data block.
 580                          */
 581                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 582                                         newdb == state->extrablk.blkno) {
 583                                 ASSERT(state->extravalid);
 584                                 curbp = state->extrablk.bp;
 585                         } else {
 586                                 error = xfs_da_read_buf(tp, dp,
 587                                                 xfs_dir2_db_to_da(mp, newdb),
 588                                                 -1, &curbp, XFS_DATA_FORK, NULL);
 589                                 if (error)
 590                                         return error;
 591                         }
 592                         xfs_dir2_data_check(dp, curbp);
 593                         curdb = newdb;
 594                 }
 595                 /*
 596                  * Point to the data entry.
 597                  */
 598                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 599                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
 600                 /*
 601                  * Compare the entry and if it's an exact match, return
 602                  * EEXIST immediately. If it's the first case-insensitive
 603                  * match, store the block & inode number and continue looking.
 604                  */
 605                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 606                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 607                         /* If there is a CI match block, drop it */
 608                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 609                                                 curdb != state->extrablk.blkno)
 610                                 xfs_trans_brelse(tp, state->extrablk.bp);
 611                         args->cmpresult = cmp;
 612                         args->inumber = be64_to_cpu(dep->inumber);
 613                         *indexp = index;
 614                         state->extravalid = 1;
 615                         state->extrablk.bp = curbp;
 616                         state->extrablk.blkno = curdb;
 617                         state->extrablk.index = (int)((char *)dep -
 618                                                         (char *)curbp->b_addr);
 619                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 620                         if (cmp == XFS_CMP_EXACT)
 621                                 return XFS_ERROR(EEXIST);
 622                 }
 623         }
 624         ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
 625                                         (args->op_flags & XFS_DA_OP_OKNOENT));
 626         if (curbp) {
 627                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 628                         /* Giving back last used data block. */
 629                         state->extravalid = 1;
 630                         state->extrablk.bp = curbp;
 631                         state->extrablk.index = -1;
 632                         state->extrablk.blkno = curdb;
 633                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 634                 } else {
 635                         /* If the curbp is not the CI match block, drop it */
 636                         if (state->extrablk.bp != curbp)
 637                                 xfs_trans_brelse(tp, curbp);
 638                 }
 639         } else {
 640                 state->extravalid = 0;
 641         }
 642         *indexp = index;
 643         return XFS_ERROR(ENOENT);
 644 }
 645
 646 /*
 647  * Look up a leaf entry in a node-format leaf block.
 648  * If this is an addname then the extrablk in state is a freespace block,
 649  * otherwise it's a data block.
 650  */
 651 int
 652 xfs_dir2_leafn_lookup_int(
 653         struct xfs_buf          *bp,            /* leaf buffer */
 654         xfs_da_args_t           *args,          /* operation arguments */
 655         int                     *indexp,        /* out: leaf entry index */
 656         xfs_da_state_t          *state)         /* state to fill in */
 657 {
 658         if (args->op_flags & XFS_DA_OP_ADDNAME)
 659                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 660                                                         state);
 661         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 662 }
 663
 664 /*
 665  * Move count leaf entries from source to destination leaf.
 666  * Log entries and headers.  Stale entries are preserved.
 667  */
 668 static void
 669 xfs_dir2_leafn_moveents(
 670         xfs_da_args_t   *args,                  /* operation arguments */
 671         struct xfs_buf  *bp_s,                  /* source leaf buffer */
 672         int             start_s,                /* source leaf index */
 673         struct xfs_buf  *bp_d,                  /* destination leaf buffer */
 674         int             start_d,                /* destination leaf index */
 675         int             count)                  /* count of leaves to copy */
 676 {
 677         xfs_dir2_leaf_t *leaf_d;                /* destination leaf structure */
 678         xfs_dir2_leaf_t *leaf_s;                /* source leaf structure */
 679         int             stale;                  /* count stale leaves copied */
 680         xfs_trans_t     *tp;                    /* transaction pointer */
 681
 682         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 683
 684         /*
 685          * Silently return if nothing to do.
 686          */
 687         if (count == 0) {
 688                 return;
 689         }
 690         tp = args->trans;
 691         leaf_s = bp_s->b_addr;
 692         leaf_d = bp_d->b_addr;
 693         /*
 694          * If the destination index is not the end of the current
 695          * destination leaf entries, open up a hole in the destination
 696          * to hold the new entries.
 697          */
 698         if (start_d < be16_to_cpu(leaf_d->hdr.count)) {
 699                 memmove(&leaf_d->ents[start_d + count], &leaf_d->ents[start_d],
 700                         (be16_to_cpu(leaf_d->hdr.count) - start_d) *
 701                         sizeof(xfs_dir2_leaf_entry_t));
 702                 xfs_dir2_leaf_log_ents(tp, bp_d, start_d + count,
 703                         count + be16_to_cpu(leaf_d->hdr.count) - 1);
 704         }
 705         /*
 706          * If the source has stale leaves, count the ones in the copy range
 707          * so we can update the header correctly.
 708          */
 709         if (leaf_s->hdr.stale) {
 710                 int     i;                      /* temp leaf index */
 711
 712                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 713                         if (leaf_s->ents[i].address ==
 714                             cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 715                                 stale++;
 716                 }
 717         } else
 718                 stale = 0;
 719         /*
 720          * Copy the leaf entries from source to destination.
 721          */
 722         memcpy(&leaf_d->ents[start_d], &leaf_s->ents[start_s],
 723                 count * sizeof(xfs_dir2_leaf_entry_t));
 724         xfs_dir2_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
 725         /*
 726          * If there are source entries after the ones we copied,
 727          * delete the ones we copied by sliding the next ones down.
 728          */
 729         if (start_s + count < be16_to_cpu(leaf_s->hdr.count)) {
 730                 memmove(&leaf_s->ents[start_s], &leaf_s->ents[start_s + count],
 731                         count * sizeof(xfs_dir2_leaf_entry_t));
 732                 xfs_dir2_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
 733         }
 734         /*
 735          * Update the headers and log them.
 736          */
 737         be16_add_cpu(&leaf_s->hdr.count, -(count));
 738         be16_add_cpu(&leaf_s->hdr.stale, -(stale));
 739         be16_add_cpu(&leaf_d->hdr.count, count);
 740         be16_add_cpu(&leaf_d->hdr.stale, stale);
 741         xfs_dir2_leaf_log_header(tp, bp_s);
 742         xfs_dir2_leaf_log_header(tp, bp_d);
 743         xfs_dir2_leafn_check(args->dp, bp_s);
 744         xfs_dir2_leafn_check(args->dp, bp_d);
 745 }
 746
 747 /*
 748  * Determine the sort order of two leaf blocks.
 749  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 750  */
 751 int                                             /* sort order */
 752 xfs_dir2_leafn_order(
 753         struct xfs_buf  *leaf1_bp,              /* leaf1 buffer */
 754         struct xfs_buf  *leaf2_bp)              /* leaf2 buffer */
 755 {
 756         xfs_dir2_leaf_t *leaf1;                 /* leaf1 structure */
 757         xfs_dir2_leaf_t *leaf2;                 /* leaf2 structure */
 758
 759         leaf1 = leaf1_bp->b_addr;
 760         leaf2 = leaf2_bp->b_addr;
 761         ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 762         ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 763         if (be16_to_cpu(leaf1->hdr.count) > 0 &&
 764             be16_to_cpu(leaf2->hdr.count) > 0 &&
 765             (be32_to_cpu(leaf2->ents[0].hashval) < be32_to_cpu(leaf1->ents[0].hashval) ||
 766              be32_to_cpu(leaf2->ents[be16_to_cpu(leaf2->hdr.count) - 1].hashval) <
 767              be32_to_cpu(leaf1->ents[be16_to_cpu(leaf1->hdr.count) - 1].hashval)))
 768                 return 1;
 769         return 0;
 770 }
 771
 772 /*
 773  * Rebalance leaf entries between two leaf blocks.
 774  * This is actually only called when the second block is new,
 775  * though the code deals with the general case.
 776  * A new entry will be inserted in one of the blocks, and that
 777  * entry is taken into account when balancing.
 778  */
 779 static void
 780 xfs_dir2_leafn_rebalance(
 781         xfs_da_state_t          *state,         /* btree cursor */
 782         xfs_da_state_blk_t      *blk1,          /* first btree block */
 783         xfs_da_state_blk_t      *blk2)          /* second btree block */
 784 {
 785         xfs_da_args_t           *args;          /* operation arguments */
 786         int                     count;          /* count (& direction) leaves */
 787         int                     isleft;         /* new goes in left leaf */
 788         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 789         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 790         int                     mid;            /* midpoint leaf index */
 791 #ifdef DEBUG
 792         int                     oldstale;       /* old count of stale leaves */
 793 #endif
 794         int                     oldsum;         /* old total leaf count */
 795         int                     swap;           /* swapped leaf blocks */
 796
 797         args = state->args;
 798         /*
 799          * If the block order is wrong, swap the arguments.
 800          */
 801         if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
 802                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
 803
 804                 tmp = blk1;
 805                 blk1 = blk2;
 806                 blk2 = tmp;
 807         }
 808         leaf1 = blk1->bp->b_addr;
 809         leaf2 = blk2->bp->b_addr;
 810         oldsum = be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count);
 811 #ifdef DEBUG
 812         oldstale = be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale);
 813 #endif
 814         mid = oldsum >> 1;
 815         /*
 816          * If the old leaf count was odd then the new one will be even,
 817          * so we need to divide the new count evenly.
 818          */
 819         if (oldsum & 1) {
 820                 xfs_dahash_t    midhash;        /* middle entry hash value */
 821
 822                 if (mid >= be16_to_cpu(leaf1->hdr.count))
 823                         midhash = be32_to_cpu(leaf2->ents[mid - be16_to_cpu(leaf1->hdr.count)].hashval);
 824                 else
 825                         midhash = be32_to_cpu(leaf1->ents[mid].hashval);
 826                 isleft = args->hashval <= midhash;
 827         }
 828         /*
 829          * If the old count is even then the new count is odd, so there's
 830          * no preferred side for the new entry.
 831          * Pick the left one.
 832          */
 833         else
 834                 isleft = 1;
 835         /*
 836          * Calculate moved entry count.  Positive means left-to-right,
 837          * negative means right-to-left.  Then move the entries.
 838          */
 839         count = be16_to_cpu(leaf1->hdr.count) - mid + (isleft == 0);
 840         if (count > 0)
 841                 xfs_dir2_leafn_moveents(args, blk1->bp,
 842                         be16_to_cpu(leaf1->hdr.count) - count, blk2->bp, 0, count);
 843         else if (count < 0)
 844                 xfs_dir2_leafn_moveents(args, blk2->bp, 0, blk1->bp,
 845                         be16_to_cpu(leaf1->hdr.count), count);
 846         ASSERT(be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count) == oldsum);
 847         ASSERT(be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale) == oldstale);
 848         /*
 849          * Mark whether we're inserting into the old or new leaf.
 850          */
 851         if (be16_to_cpu(leaf1->hdr.count) < be16_to_cpu(leaf2->hdr.count))
 852                 state->inleaf = swap;
 853         else if (be16_to_cpu(leaf1->hdr.count) > be16_to_cpu(leaf2->hdr.count))
 854                 state->inleaf = !swap;
 855         else
 856                 state->inleaf =
 857                         swap ^ (blk1->index <= be16_to_cpu(leaf1->hdr.count));
 858         /*
 859          * Adjust the expected index for insertion.
 860          */
 861         if (!state->inleaf)
 862                 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
 863
 864         /*
 865          * Finally sanity check just to make sure we are not returning a
 866          * negative index
 867          */
 868         if(blk2->index < 0) {
 869                 state->inleaf = 1;
 870                 blk2->index = 0;
 871                 xfs_alert(args->dp->i_mount,
 872         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d\n",
 873                         __func__, blk1->index);
 874         }
 875 }
 876
 877 static int
 878 xfs_dir2_data_block_free(
 879         xfs_da_args_t           *args,
 880         struct xfs_dir2_data_hdr *hdr,
 881         struct xfs_dir2_free    *free,
 882         xfs_dir2_db_t           fdb,
 883         int                     findex,
 884         struct xfs_buf          *fbp,
 885         int                     longest)
 886 {
 887         struct xfs_trans        *tp = args->trans;
 888         int                     logfree = 0;
 889
 890         if (!hdr) {
 891                 /* One less used entry in the free table.  */
 892                 be32_add_cpu(&free->hdr.nused, -1);
 893                 xfs_dir2_free_log_header(tp, fbp);
 894
 895                 /*
 896                  * If this was the last entry in the table, we can trim the
 897                  * table size back.  There might be other entries at the end
 898                  * referring to non-existent data blocks, get those too.
 899                  */
 900                 if (findex == be32_to_cpu(free->hdr.nvalid) - 1) {
 901                         int     i;              /* free entry index */
 902
 903                         for (i = findex - 1; i >= 0; i--) {
 904                                 if (free->bests[i] != cpu_to_be16(NULLDATAOFF))
 905                                         break;
 906                         }
 907                         free->hdr.nvalid = cpu_to_be32(i + 1);
 908                         logfree = 0;
 909                 } else {
 910                         /* Not the last entry, just punch it out.  */
 911                         free->bests[findex] = cpu_to_be16(NULLDATAOFF);
 912                         logfree = 1;
 913                 }
 914                 /*
 915                  * If there are no useful entries left in the block,
 916                  * get rid of the block if we can.
 917                  */
 918                 if (!free->hdr.nused) {
 919                         int error;
 920
 921                         error = xfs_dir2_shrink_inode(args, fdb, fbp);
 922                         if (error == 0) {
 923                                 fbp = NULL;
 924                                 logfree = 0;
 925                         } else if (error != ENOSPC || args->total != 0)
 926                                 return error;
 927                         /*
 928                          * It's possible to get ENOSPC if there is no
 929                          * space reservation.  In this case some one
 930                          * else will eventually get rid of this block.
 931                          */
 932                 }
 933         } else {
 934                 /*
 935                  * Data block is not empty, just set the free entry to the new
 936                  * value.
 937                  */
 938                 free->bests[findex] = cpu_to_be16(longest);
 939                 logfree = 1;
 940         }
 941
 942         /* Log the free entry that changed, unless we got rid of it.  */
 943         if (logfree)
 944                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
 945         return 0;
 946 }
 947
 948 /*
 949  * Remove an entry from a node directory.
 950  * This removes the leaf entry and the data entry,
 951  * and updates the free block if necessary.
 952  */
 953 static int                                      /* error */
 954 xfs_dir2_leafn_remove(
 955         xfs_da_args_t           *args,          /* operation arguments */
 956         struct xfs_buf          *bp,            /* leaf buffer */
 957         int                     index,          /* leaf entry index */
 958         xfs_da_state_blk_t      *dblk,          /* data block */
 959         int                     *rval)          /* resulting block needs join */
 960 {
 961         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 962         xfs_dir2_db_t           db;             /* data block number */
 963         struct xfs_buf          *dbp;           /* data block buffer */
 964         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 965         xfs_inode_t             *dp;            /* incore directory inode */
 966         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 967         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 968         int                     longest;        /* longest data free entry */
 969         int                     off;            /* data block entry offset */
 970         xfs_mount_t             *mp;            /* filesystem mount point */
 971         int                     needlog;        /* need to log data header */
 972         int                     needscan;       /* need to rescan data frees */
 973         xfs_trans_t             *tp;            /* transaction pointer */
 974
 975         trace_xfs_dir2_leafn_remove(args, index);
 976
 977         dp = args->dp;
 978         tp = args->trans;
 979         mp = dp->i_mount;
 980         leaf = bp->b_addr;
 981         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 982         /*
 983          * Point to the entry we're removing.
 984          */
 985         lep = &leaf->ents[index];
 986         /*
 987          * Extract the data block and offset from the entry.
 988          */
 989         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 990         ASSERT(dblk->blkno == db);
 991         off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
 992         ASSERT(dblk->index == off);
 993         /*
 994          * Kill the leaf entry by marking it stale.
 995          * Log the leaf block changes.
 996          */
 997         be16_add_cpu(&leaf->hdr.stale, 1);
 998         xfs_dir2_leaf_log_header(tp, bp);
 999         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1000         xfs_dir2_leaf_log_ents(tp, bp, index, index);
1001         /*
1002          * Make the data entry free.  Keep track of the longest freespace
1003          * in the data block in case it changes.
1004          */
1005         dbp = dblk->bp;
1006         hdr = dbp->b_addr;
1007         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1008         longest = be16_to_cpu(hdr->bestfree[0].length);
1009         needlog = needscan = 0;
1010         xfs_dir2_data_make_free(tp, dbp, off,
1011                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1012         /*
1013          * Rescan the data block freespaces for bestfree.
1014          * Log the data block header if needed.
1015          */
1016         if (needscan)
1017                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1018         if (needlog)
1019                 xfs_dir2_data_log_header(tp, dbp);
1020         xfs_dir2_data_check(dp, dbp);
1021         /*
1022          * If the longest data block freespace changes, need to update
1023          * the corresponding freeblock entry.
1024          */
1025         if (longest < be16_to_cpu(hdr->bestfree[0].length)) {
1026                 int             error;          /* error return value */
1027                 struct xfs_buf  *fbp;           /* freeblock buffer */
1028                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1029                 int             findex;         /* index in freeblock entries */
1030                 xfs_dir2_free_t *free;          /* freeblock structure */
1031
1032                 /*
1033                  * Convert the data block number to a free block,
1034                  * read in the free block.
1035                  */
1036                 fdb = xfs_dir2_db_to_fdb(mp, db);
1037                 error = xfs_dir2_free_read(tp, dp, xfs_dir2_db_to_da(mp, fdb),
1038                                            &fbp);
1039                 if (error)
1040                         return error;
1041                 free = fbp->b_addr;
1042                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1043                 ASSERT(be32_to_cpu(free->hdr.firstdb) ==
1044                        xfs_dir2_free_max_bests(mp) *
1045                        (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
1046                 /*
1047                  * Calculate which entry we need to fix.
1048                  */
1049                 findex = xfs_dir2_db_to_fdindex(mp, db);
1050                 longest = be16_to_cpu(hdr->bestfree[0].length);
1051                 /*
1052                  * If the data block is now empty we can get rid of it
1053                  * (usually).
1054                  */
1055                 if (longest == mp->m_dirblksize - (uint)sizeof(*hdr)) {
1056                         /*
1057                          * Try to punch out the data block.
1058                          */
1059                         error = xfs_dir2_shrink_inode(args, db, dbp);
1060                         if (error == 0) {
1061                                 dblk->bp = NULL;
1062                                 hdr = NULL;
1063                         }
1064                         /*
1065                          * We can get ENOSPC if there's no space reservation.
1066                          * In this case just drop the buffer and some one else
1067                          * will eventually get rid of the empty block.
1068                          */
1069                         else if (!(error == ENOSPC && args->total == 0))
1070                                 return error;
1071                 }
1072                 /*
1073                  * If we got rid of the data block, we can eliminate that entry
1074                  * in the free block.
1075                  */
1076                 error = xfs_dir2_data_block_free(args, hdr, free,
1077                                                  fdb, findex, fbp, longest);
1078                 if (error)
1079                         return error;
1080         }
1081
1082         xfs_dir2_leafn_check(dp, bp);
1083         /*
1084          * Return indication of whether this leaf block is empty enough
1085          * to justify trying to join it with a neighbor.
1086          */
1087         *rval =
1088                 ((uint)sizeof(leaf->hdr) +
1089                  (uint)sizeof(leaf->ents[0]) *
1090                  (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale))) <
1091                 mp->m_dir_magicpct;
1092         return 0;
1093 }
1094
1095 /*
1096  * Split the leaf entries in the old block into old and new blocks.
1097  */
1098 int                                             /* error */
1099 xfs_dir2_leafn_split(
1100         xfs_da_state_t          *state,         /* btree cursor */
1101         xfs_da_state_blk_t      *oldblk,        /* original block */
1102         xfs_da_state_blk_t      *newblk)        /* newly created block */
1103 {
1104         xfs_da_args_t           *args;          /* operation arguments */
1105         xfs_dablk_t             blkno;          /* new leaf block number */
1106         int                     error;          /* error return value */
1107         xfs_mount_t             *mp;            /* filesystem mount point */
1108
1109         /*
1110          * Allocate space for a new leaf node.
1111          */
1112         args = state->args;
1113         mp = args->dp->i_mount;
1114         ASSERT(args != NULL);
1115         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1116         error = xfs_da_grow_inode(args, &blkno);
1117         if (error) {
1118                 return error;
1119         }
1120         /*
1121          * Initialize the new leaf block.
1122          */
1123         error = xfs_dir2_leaf_init(args, xfs_dir2_da_to_db(mp, blkno),
1124                 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1125         if (error) {
1126                 return error;
1127         }
1128         newblk->blkno = blkno;
1129         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1130         /*
1131          * Rebalance the entries across the two leaves, link the new
1132          * block into the leaves.
1133          */
1134         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1135         error = xfs_da_blk_link(state, oldblk, newblk);
1136         if (error) {
1137                 return error;
1138         }
1139         /*
1140          * Insert the new entry in the correct block.
1141          */
1142         if (state->inleaf)
1143                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1144         else
1145                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1146         /*
1147          * Update last hashval in each block since we added the name.
1148          */
1149         oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1150         newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1151         xfs_dir2_leafn_check(args->dp, oldblk->bp);
1152         xfs_dir2_leafn_check(args->dp, newblk->bp);
1153         return error;
1154 }
1155
1156 /*
1157  * Check a leaf block and its neighbors to see if the block should be
1158  * collapsed into one or the other neighbor.  Always keep the block
1159  * with the smaller block number.
1160  * If the current block is over 50% full, don't try to join it, return 0.
1161  * If the block is empty, fill in the state structure and return 2.
1162  * If it can be collapsed, fill in the state structure and return 1.
1163  * If nothing can be done, return 0.
1164  */
1165 int                                             /* error */
1166 xfs_dir2_leafn_toosmall(
1167         xfs_da_state_t          *state,         /* btree cursor */
1168         int                     *action)        /* resulting action to take */
1169 {
1170         xfs_da_state_blk_t      *blk;           /* leaf block */
1171         xfs_dablk_t             blkno;          /* leaf block number */
1172         struct xfs_buf          *bp;            /* leaf buffer */
1173         int                     bytes;          /* bytes in use */
1174         int                     count;          /* leaf live entry count */
1175         int                     error;          /* error return value */
1176         int                     forward;        /* sibling block direction */
1177         int                     i;              /* sibling counter */
1178         xfs_da_blkinfo_t        *info;          /* leaf block header */
1179         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1180         int                     rval;           /* result from path_shift */
1181
1182         /*
1183          * Check for the degenerate case of the block being over 50% full.
1184          * If so, it's not worth even looking to see if we might be able
1185          * to coalesce with a sibling.
1186          */
1187         blk = &state->path.blk[state->path.active - 1];
1188         info = blk->bp->b_addr;
1189         ASSERT(info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1190         leaf = (xfs_dir2_leaf_t *)info;
1191         count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1192         bytes = (uint)sizeof(leaf->hdr) + count * (uint)sizeof(leaf->ents[0]);
1193         if (bytes > (state->blocksize >> 1)) {
1194                 /*
1195                  * Blk over 50%, don't try to join.
1196                  */
1197                 *action = 0;
1198                 return 0;
1199         }
1200         /*
1201          * Check for the degenerate case of the block being empty.
1202          * If the block is empty, we'll simply delete it, no need to
1203          * coalesce it with a sibling block.  We choose (arbitrarily)
1204          * to merge with the forward block unless it is NULL.
1205          */
1206         if (count == 0) {
1207                 /*
1208                  * Make altpath point to the block we want to keep and
1209                  * path point to the block we want to drop (this one).
1210                  */
1211                 forward = (info->forw != 0);
1212                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1213                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1214                         &rval);
1215                 if (error)
1216                         return error;
1217                 *action = rval ? 2 : 0;
1218                 return 0;
1219         }
1220         /*
1221          * Examine each sibling block to see if we can coalesce with
1222          * at least 25% free space to spare.  We need to figure out
1223          * whether to merge with the forward or the backward block.
1224          * We prefer coalescing with the lower numbered sibling so as
1225          * to shrink a directory over time.
1226          */
1227         forward = be32_to_cpu(info->forw) < be32_to_cpu(info->back);
1228         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1229                 blkno = forward ? be32_to_cpu(info->forw) : be32_to_cpu(info->back);
1230                 if (blkno == 0)
1231                         continue;
1232                 /*
1233                  * Read the sibling leaf block.
1234                  */
1235                 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1236                                         blkno, -1, &bp, XFS_DATA_FORK, NULL);
1237                 if (error)
1238                         return error;
1239                 ASSERT(bp != NULL);
1240                 /*
1241                  * Count bytes in the two blocks combined.
1242                  */
1243                 leaf = (xfs_dir2_leaf_t *)info;
1244                 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1245                 bytes = state->blocksize - (state->blocksize >> 2);
1246                 leaf = bp->b_addr;
1247                 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1248                 count += be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1249                 bytes -= count * (uint)sizeof(leaf->ents[0]);
1250                 /*
1251                  * Fits with at least 25% to spare.
1252                  */
1253                 if (bytes >= 0)
1254                         break;
1255                 xfs_trans_brelse(state->args->trans, bp);
1256         }
1257         /*
1258          * Didn't like either block, give up.
1259          */
1260         if (i >= 2) {
1261                 *action = 0;
1262                 return 0;
1263         }
1264
1265         /*
1266          * Make altpath point to the block we want to keep (the lower
1267          * numbered block) and path point to the block we want to drop.
1268          */
1269         memcpy(&state->altpath, &state->path, sizeof(state->path));
1270         if (blkno < blk->blkno)
1271                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1272                         &rval);
1273         else
1274                 error = xfs_da_path_shift(state, &state->path, forward, 0,
1275                         &rval);
1276         if (error) {
1277                 return error;
1278         }
1279         *action = rval ? 0 : 1;
1280         return 0;
1281 }
1282
1283 /*
1284  * Move all the leaf entries from drop_blk to save_blk.
1285  * This is done as part of a join operation.
1286  */
1287 void
1288 xfs_dir2_leafn_unbalance(
1289         xfs_da_state_t          *state,         /* cursor */
1290         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1291         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1292 {
1293         xfs_da_args_t           *args;          /* operation arguments */
1294         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1295         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1296
1297         args = state->args;
1298         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1299         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1300         drop_leaf = drop_blk->bp->b_addr;
1301         save_leaf = save_blk->bp->b_addr;
1302         ASSERT(drop_leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1303         ASSERT(save_leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1304         /*
1305          * If there are any stale leaf entries, take this opportunity
1306          * to purge them.
1307          */
1308         if (drop_leaf->hdr.stale)
1309                 xfs_dir2_leaf_compact(args, drop_blk->bp);
1310         if (save_leaf->hdr.stale)
1311                 xfs_dir2_leaf_compact(args, save_blk->bp);
1312         /*
1313          * Move the entries from drop to the appropriate end of save.
1314          */
1315         drop_blk->hashval = be32_to_cpu(drop_leaf->ents[be16_to_cpu(drop_leaf->hdr.count) - 1].hashval);
1316         if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1317                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp, 0,
1318                         be16_to_cpu(drop_leaf->hdr.count));
1319         else
1320                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp,
1321                         be16_to_cpu(save_leaf->hdr.count), be16_to_cpu(drop_leaf->hdr.count));
1322         save_blk->hashval = be32_to_cpu(save_leaf->ents[be16_to_cpu(save_leaf->hdr.count) - 1].hashval);
1323         xfs_dir2_leafn_check(args->dp, save_blk->bp);
1324 }
1325
1326 /*
1327  * Top-level node form directory addname routine.
1328  */
1329 int                                             /* error */
1330 xfs_dir2_node_addname(
1331         xfs_da_args_t           *args)          /* operation arguments */
1332 {
1333         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1334         int                     error;          /* error return value */
1335         int                     rval;           /* sub-return value */
1336         xfs_da_state_t          *state;         /* btree cursor */
1337
1338         trace_xfs_dir2_node_addname(args);
1339
1340         /*
1341          * Allocate and initialize the state (btree cursor).
1342          */
1343         state = xfs_da_state_alloc();
1344         state->args = args;
1345         state->mp = args->dp->i_mount;
1346         state->blocksize = state->mp->m_dirblksize;
1347         state->node_ents = state->mp->m_dir_node_ents;
1348         /*
1349          * Look up the name.  We're not supposed to find it, but
1350          * this gives us the insertion point.
1351          */
1352         error = xfs_da_node_lookup_int(state, &rval);
1353         if (error)
1354                 rval = error;
1355         if (rval != ENOENT) {
1356                 goto done;
1357         }
1358         /*
1359          * Add the data entry to a data block.
1360          * Extravalid is set to a freeblock found by lookup.
1361          */
1362         rval = xfs_dir2_node_addname_int(args,
1363                 state->extravalid ? &state->extrablk : NULL);
1364         if (rval) {
1365                 goto done;
1366         }
1367         blk = &state->path.blk[state->path.active - 1];
1368         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1369         /*
1370          * Add the new leaf entry.
1371          */
1372         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1373         if (rval == 0) {
1374                 /*
1375                  * It worked, fix the hash values up the btree.
1376                  */
1377                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1378                         xfs_da_fixhashpath(state, &state->path);
1379         } else {
1380                 /*
1381                  * It didn't work, we need to split the leaf block.
1382                  */
1383                 if (args->total == 0) {
1384                         ASSERT(rval == ENOSPC);
1385                         goto done;
1386                 }
1387                 /*
1388                  * Split the leaf block and insert the new entry.
1389                  */
1390                 rval = xfs_da_split(state);
1391         }
1392 done:
1393         xfs_da_state_free(state);
1394         return rval;
1395 }
1396
1397 /*
1398  * Add the data entry for a node-format directory name addition.
1399  * The leaf entry is added in xfs_dir2_leafn_add.
1400  * We may enter with a freespace block that the lookup found.
1401  */
1402 static int                                      /* error */
1403 xfs_dir2_node_addname_int(
1404         xfs_da_args_t           *args,          /* operation arguments */
1405         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1406 {
1407         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1408         xfs_dir2_db_t           dbno;           /* data block number */
1409         struct xfs_buf          *dbp;           /* data block buffer */
1410         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1411         xfs_inode_t             *dp;            /* incore directory inode */
1412         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1413         int                     error;          /* error return value */
1414         xfs_dir2_db_t           fbno;           /* freespace block number */
1415         struct xfs_buf          *fbp;           /* freespace buffer */
1416         int                     findex;         /* freespace entry index */
1417         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1418         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1419         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1420         int                     length;         /* length of the new entry */
1421         int                     logfree;        /* need to log free entry */
1422         xfs_mount_t             *mp;            /* filesystem mount point */
1423         int                     needlog;        /* need to log data header */
1424         int                     needscan;       /* need to rescan data frees */
1425         __be16                  *tagp;          /* data entry tag pointer */
1426         xfs_trans_t             *tp;            /* transaction pointer */
1427
1428         dp = args->dp;
1429         mp = dp->i_mount;
1430         tp = args->trans;
1431         length = xfs_dir2_data_entsize(args->namelen);
1432         /*
1433          * If we came in with a freespace block that means that lookup
1434          * found an entry with our hash value.  This is the freespace
1435          * block for that data entry.
1436          */
1437         if (fblk) {
1438                 fbp = fblk->bp;
1439                 /*
1440                  * Remember initial freespace block number.
1441                  */
1442                 ifbno = fblk->blkno;
1443                 free = fbp->b_addr;
1444                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1445                 findex = fblk->index;
1446                 /*
1447                  * This means the free entry showed that the data block had
1448                  * space for our entry, so we remembered it.
1449                  * Use that data block.
1450                  */
1451                 if (findex >= 0) {
1452                         ASSERT(findex < be32_to_cpu(free->hdr.nvalid));
1453                         ASSERT(be16_to_cpu(free->bests[findex]) != NULLDATAOFF);
1454                         ASSERT(be16_to_cpu(free->bests[findex]) >= length);
1455                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1456                 }
1457                 /*
1458                  * The data block looked at didn't have enough room.
1459                  * We'll start at the beginning of the freespace entries.
1460                  */
1461                 else {
1462                         dbno = -1;
1463                         findex = 0;
1464                 }
1465         }
1466         /*
1467          * Didn't come in with a freespace block, so don't have a data block.
1468          */
1469         else {
1470                 ifbno = dbno = -1;
1471                 fbp = NULL;
1472                 findex = 0;
1473         }
1474         /*
1475          * If we don't have a data block yet, we're going to scan the
1476          * freespace blocks looking for one.  Figure out what the
1477          * highest freespace block number is.
1478          */
1479         if (dbno == -1) {
1480                 xfs_fileoff_t   fo;             /* freespace block number */
1481
1482                 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1483                         return error;
1484                 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1485                 fbno = ifbno;
1486         }
1487         /*
1488          * While we haven't identified a data block, search the freeblock
1489          * data for a good data block.  If we find a null freeblock entry,
1490          * indicating a hole in the data blocks, remember that.
1491          */
1492         while (dbno == -1) {
1493                 /*
1494                  * If we don't have a freeblock in hand, get the next one.
1495                  */
1496                 if (fbp == NULL) {
1497                         /*
1498                          * Happens the first time through unless lookup gave
1499                          * us a freespace block to start with.
1500                          */
1501                         if (++fbno == 0)
1502                                 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1503                         /*
1504                          * If it's ifbno we already looked at it.
1505                          */
1506                         if (fbno == ifbno)
1507                                 fbno++;
1508                         /*
1509                          * If it's off the end we're done.
1510                          */
1511                         if (fbno >= lastfbno)
1512                                 break;
1513                         /*
1514                          * Read the block.  There can be holes in the
1515                          * freespace blocks, so this might not succeed.
1516                          * This should be really rare, so there's no reason
1517                          * to avoid it.
1518                          */
1519                         error = xfs_dir2_free_try_read(tp, dp,
1520                                                 xfs_dir2_db_to_da(mp, fbno),
1521                                                 &fbp);
1522                         if (error)
1523                                 return error;
1524                         if (!fbp)
1525                                 continue;
1526                         free = fbp->b_addr;
1527                         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1528                         findex = 0;
1529                 }
1530                 /*
1531                  * Look at the current free entry.  Is it good enough?
1532                  */
1533                 if (be16_to_cpu(free->bests[findex]) != NULLDATAOFF &&
1534                     be16_to_cpu(free->bests[findex]) >= length)
1535                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1536                 else {
1537                         /*
1538                          * Are we done with the freeblock?
1539                          */
1540                         if (++findex == be32_to_cpu(free->hdr.nvalid)) {
1541                                 /*
1542                                  * Drop the block.
1543                                  */
1544                                 xfs_trans_brelse(tp, fbp);
1545                                 fbp = NULL;
1546                                 if (fblk && fblk->bp)
1547                                         fblk->bp = NULL;
1548                         }
1549                 }
1550         }
1551         /*
1552          * If we don't have a data block, we need to allocate one and make
1553          * the freespace entries refer to it.
1554          */
1555         if (unlikely(dbno == -1)) {
1556                 /*
1557                  * Not allowed to allocate, return failure.
1558                  */
1559                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1560                         return XFS_ERROR(ENOSPC);
1561
1562                 /*
1563                  * Allocate and initialize the new data block.
1564                  */
1565                 if (unlikely((error = xfs_dir2_grow_inode(args,
1566                                                          XFS_DIR2_DATA_SPACE,
1567                                                          &dbno)) ||
1568                     (error = xfs_dir2_data_init(args, dbno, &dbp))))
1569                         return error;
1570
1571                 /*
1572                  * If (somehow) we have a freespace block, get rid of it.
1573                  */
1574                 if (fbp)
1575                         xfs_trans_brelse(tp, fbp);
1576                 if (fblk && fblk->bp)
1577                         fblk->bp = NULL;
1578
1579                 /*
1580                  * Get the freespace block corresponding to the data block
1581                  * that was just allocated.
1582                  */
1583                 fbno = xfs_dir2_db_to_fdb(mp, dbno);
1584                 error = xfs_dir2_free_try_read(tp, dp,
1585                                                xfs_dir2_db_to_da(mp, fbno),
1586                                                &fbp);
1587                 if (error)
1588                         return error;
1589
1590                 /*
1591                  * If there wasn't a freespace block, the read will
1592                  * return a NULL fbp.  Allocate and initialize a new one.
1593                  */
1594                 if( fbp == NULL ) {
1595                         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1596                                                         &fbno))) {
1597                                 return error;
1598                         }
1599
1600                         if (unlikely(xfs_dir2_db_to_fdb(mp, dbno) != fbno)) {
1601                                 xfs_alert(mp,
1602                         "%s: dir ino %llu needed freesp block %lld for\n"
1603                         "  data block %lld, got %lld ifbno %llu lastfbno %d",
1604                                         __func__, (unsigned long long)dp->i_ino,
1605                                         (long long)xfs_dir2_db_to_fdb(mp, dbno),
1606                                         (long long)dbno, (long long)fbno,
1607                                         (unsigned long long)ifbno, lastfbno);
1608                                 if (fblk) {
1609                                         xfs_alert(mp,
1610                                 " fblk 0x%p blkno %llu index %d magic 0x%x",
1611                                                 fblk,
1612                                                 (unsigned long long)fblk->blkno,
1613                                                 fblk->index,
1614                                                 fblk->magic);
1615                                 } else {
1616                                         xfs_alert(mp, " ... fblk is NULL");
1617                                 }
1618                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1619                                                  XFS_ERRLEVEL_LOW, mp);
1620                                 return XFS_ERROR(EFSCORRUPTED);
1621                         }
1622
1623                         /*
1624                          * Get a buffer for the new block.
1625                          */
1626                         if ((error = xfs_da_get_buf(tp, dp,
1627                                                    xfs_dir2_db_to_da(mp, fbno),
1628                                                    -1, &fbp, XFS_DATA_FORK))) {
1629                                 return error;
1630                         }
1631                         ASSERT(fbp != NULL);
1632
1633                         /*
1634                          * Initialize the new block to be empty, and remember
1635                          * its first slot as our empty slot.
1636                          */
1637                         free = fbp->b_addr;
1638                         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
1639                         free->hdr.firstdb = cpu_to_be32(
1640                                 (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1641                                 xfs_dir2_free_max_bests(mp));
1642                         free->hdr.nvalid = 0;
1643                         free->hdr.nused = 0;
1644                 } else {
1645                         free = fbp->b_addr;
1646                         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1647                 }
1648
1649                 /*
1650                  * Set the freespace block index from the data block number.
1651                  */
1652                 findex = xfs_dir2_db_to_fdindex(mp, dbno);
1653                 /*
1654                  * If it's after the end of the current entries in the
1655                  * freespace block, extend that table.
1656                  */
1657                 if (findex >= be32_to_cpu(free->hdr.nvalid)) {
1658                         ASSERT(findex < xfs_dir2_free_max_bests(mp));
1659                         free->hdr.nvalid = cpu_to_be32(findex + 1);
1660                         /*
1661                          * Tag new entry so nused will go up.
1662                          */
1663                         free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1664                 }
1665                 /*
1666                  * If this entry was for an empty data block
1667                  * (this should always be true) then update the header.
1668                  */
1669                 if (free->bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1670                         be32_add_cpu(&free->hdr.nused, 1);
1671                         xfs_dir2_free_log_header(tp, fbp);
1672                 }
1673                 /*
1674                  * Update the real value in the table.
1675                  * We haven't allocated the data entry yet so this will
1676                  * change again.
1677                  */
1678                 hdr = dbp->b_addr;
1679                 free->bests[findex] = hdr->bestfree[0].length;
1680                 logfree = 1;
1681         }
1682         /*
1683          * We had a data block so we don't have to make a new one.
1684          */
1685         else {
1686                 /*
1687                  * If just checking, we succeeded.
1688                  */
1689                 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1690                         return 0;
1691
1692                 /*
1693                  * Read the data block in.
1694                  */
1695                 error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1696                                         -1, &dbp, XFS_DATA_FORK, NULL);
1697                 if (error)
1698                         return error;
1699                 hdr = dbp->b_addr;
1700                 logfree = 0;
1701         }
1702         ASSERT(be16_to_cpu(hdr->bestfree[0].length) >= length);
1703         /*
1704          * Point to the existing unused space.
1705          */
1706         dup = (xfs_dir2_data_unused_t *)
1707               ((char *)hdr + be16_to_cpu(hdr->bestfree[0].offset));
1708         needscan = needlog = 0;
1709         /*
1710          * Mark the first part of the unused space, inuse for us.
1711          */
1712         xfs_dir2_data_use_free(tp, dbp, dup,
1713                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
1714                 &needlog, &needscan);
1715         /*
1716          * Fill in the new entry and log it.
1717          */
1718         dep = (xfs_dir2_data_entry_t *)dup;
1719         dep->inumber = cpu_to_be64(args->inumber);
1720         dep->namelen = args->namelen;
1721         memcpy(dep->name, args->name, dep->namelen);
1722         tagp = xfs_dir2_data_entry_tag_p(dep);
1723         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1724         xfs_dir2_data_log_entry(tp, dbp, dep);
1725         /*
1726          * Rescan the block for bestfree if needed.
1727          */
1728         if (needscan)
1729                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1730         /*
1731          * Log the data block header if needed.
1732          */
1733         if (needlog)
1734                 xfs_dir2_data_log_header(tp, dbp);
1735         /*
1736          * If the freespace entry is now wrong, update it.
1737          */
1738         if (be16_to_cpu(free->bests[findex]) != be16_to_cpu(hdr->bestfree[0].length)) {
1739                 free->bests[findex] = hdr->bestfree[0].length;
1740                 logfree = 1;
1741         }
1742         /*
1743          * Log the freespace entry if needed.
1744          */
1745         if (logfree)
1746                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1747         /*
1748          * Return the data block and offset in args, then drop the data block.
1749          */
1750         args->blkno = (xfs_dablk_t)dbno;
1751         args->index = be16_to_cpu(*tagp);
1752         return 0;
1753 }
1754
1755 /*
1756  * Lookup an entry in a node-format directory.
1757  * All the real work happens in xfs_da_node_lookup_int.
1758  * The only real output is the inode number of the entry.
1759  */
1760 int                                             /* error */
1761 xfs_dir2_node_lookup(
1762         xfs_da_args_t   *args)                  /* operation arguments */
1763 {
1764         int             error;                  /* error return value */
1765         int             i;                      /* btree level */
1766         int             rval;                   /* operation return value */
1767         xfs_da_state_t  *state;                 /* btree cursor */
1768
1769         trace_xfs_dir2_node_lookup(args);
1770
1771         /*
1772          * Allocate and initialize the btree cursor.
1773          */
1774         state = xfs_da_state_alloc();
1775         state->args = args;
1776         state->mp = args->dp->i_mount;
1777         state->blocksize = state->mp->m_dirblksize;
1778         state->node_ents = state->mp->m_dir_node_ents;
1779         /*
1780          * Fill in the path to the entry in the cursor.
1781          */
1782         error = xfs_da_node_lookup_int(state, &rval);
1783         if (error)
1784                 rval = error;
1785         else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
1786                 /* If a CI match, dup the actual name and return EEXIST */
1787                 xfs_dir2_data_entry_t   *dep;
1788
1789                 dep = (xfs_dir2_data_entry_t *)
1790                         ((char *)state->extrablk.bp->b_addr +
1791                                                  state->extrablk.index);
1792                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1793         }
1794         /*
1795          * Release the btree blocks and leaf block.
1796          */
1797         for (i = 0; i < state->path.active; i++) {
1798                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
1799                 state->path.blk[i].bp = NULL;
1800         }
1801         /*
1802          * Release the data block if we have it.
1803          */
1804         if (state->extravalid && state->extrablk.bp) {
1805                 xfs_trans_brelse(args->trans, state->extrablk.bp);
1806                 state->extrablk.bp = NULL;
1807         }
1808         xfs_da_state_free(state);
1809         return rval;
1810 }
1811
1812 /*
1813  * Remove an entry from a node-format directory.
1814  */
1815 int                                             /* error */
1816 xfs_dir2_node_removename(
1817         xfs_da_args_t           *args)          /* operation arguments */
1818 {
1819         xfs_da_state_blk_t      *blk;           /* leaf block */
1820         int                     error;          /* error return value */
1821         int                     rval;           /* operation return value */
1822         xfs_da_state_t          *state;         /* btree cursor */
1823
1824         trace_xfs_dir2_node_removename(args);
1825
1826         /*
1827          * Allocate and initialize the btree cursor.
1828          */
1829         state = xfs_da_state_alloc();
1830         state->args = args;
1831         state->mp = args->dp->i_mount;
1832         state->blocksize = state->mp->m_dirblksize;
1833         state->node_ents = state->mp->m_dir_node_ents;
1834         /*
1835          * Look up the entry we're deleting, set up the cursor.
1836          */
1837         error = xfs_da_node_lookup_int(state, &rval);
1838         if (error)
1839                 rval = error;
1840         /*
1841          * Didn't find it, upper layer screwed up.
1842          */
1843         if (rval != EEXIST) {
1844                 xfs_da_state_free(state);
1845                 return rval;
1846         }
1847         blk = &state->path.blk[state->path.active - 1];
1848         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1849         ASSERT(state->extravalid);
1850         /*
1851          * Remove the leaf and data entries.
1852          * Extrablk refers to the data block.
1853          */
1854         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
1855                 &state->extrablk, &rval);
1856         if (error)
1857                 return error;
1858         /*
1859          * Fix the hash values up the btree.
1860          */
1861         xfs_da_fixhashpath(state, &state->path);
1862         /*
1863          * If we need to join leaf blocks, do it.
1864          */
1865         if (rval && state->path.active > 1)
1866                 error = xfs_da_join(state);
1867         /*
1868          * If no errors so far, try conversion to leaf format.
1869          */
1870         if (!error)
1871                 error = xfs_dir2_node_to_leaf(state);
1872         xfs_da_state_free(state);
1873         return error;
1874 }
1875
1876 /*
1877  * Replace an entry's inode number in a node-format directory.
1878  */
1879 int                                             /* error */
1880 xfs_dir2_node_replace(
1881         xfs_da_args_t           *args)          /* operation arguments */
1882 {
1883         xfs_da_state_blk_t      *blk;           /* leaf block */
1884         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1885         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
1886         int                     error;          /* error return value */
1887         int                     i;              /* btree level */
1888         xfs_ino_t               inum;           /* new inode number */
1889         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1890         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
1891         int                     rval;           /* internal return value */
1892         xfs_da_state_t          *state;         /* btree cursor */
1893
1894         trace_xfs_dir2_node_replace(args);
1895
1896         /*
1897          * Allocate and initialize the btree cursor.
1898          */
1899         state = xfs_da_state_alloc();
1900         state->args = args;
1901         state->mp = args->dp->i_mount;
1902         state->blocksize = state->mp->m_dirblksize;
1903         state->node_ents = state->mp->m_dir_node_ents;
1904         inum = args->inumber;
1905         /*
1906          * Lookup the entry to change in the btree.
1907          */
1908         error = xfs_da_node_lookup_int(state, &rval);
1909         if (error) {
1910                 rval = error;
1911         }
1912         /*
1913          * It should be found, since the vnodeops layer has looked it up
1914          * and locked it.  But paranoia is good.
1915          */
1916         if (rval == EEXIST) {
1917                 /*
1918                  * Find the leaf entry.
1919                  */
1920                 blk = &state->path.blk[state->path.active - 1];
1921                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1922                 leaf = blk->bp->b_addr;
1923                 lep = &leaf->ents[blk->index];
1924                 ASSERT(state->extravalid);
1925                 /*
1926                  * Point to the data entry.
1927                  */
1928                 hdr = state->extrablk.bp->b_addr;
1929                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
1930                 dep = (xfs_dir2_data_entry_t *)
1931                       ((char *)hdr +
1932                        xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
1933                 ASSERT(inum != be64_to_cpu(dep->inumber));
1934                 /*
1935                  * Fill in the new inode number and log the entry.
1936                  */
1937                 dep->inumber = cpu_to_be64(inum);
1938                 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
1939                 rval = 0;
1940         }
1941         /*
1942          * Didn't find it, and we're holding a data block.  Drop it.
1943          */
1944         else if (state->extravalid) {
1945                 xfs_trans_brelse(args->trans, state->extrablk.bp);
1946                 state->extrablk.bp = NULL;
1947         }
1948         /*
1949          * Release all the buffers in the cursor.
1950          */
1951         for (i = 0; i < state->path.active; i++) {
1952                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
1953                 state->path.blk[i].bp = NULL;
1954         }
1955         xfs_da_state_free(state);
1956         return rval;
1957 }
1958
1959 /*
1960  * Trim off a trailing empty freespace block.
1961  * Return (in rvalp) 1 if we did it, 0 if not.
1962  */
1963 int                                             /* error */
1964 xfs_dir2_node_trim_free(
1965         xfs_da_args_t           *args,          /* operation arguments */
1966         xfs_fileoff_t           fo,             /* free block number */
1967         int                     *rvalp)         /* out: did something */
1968 {
1969         struct xfs_buf          *bp;            /* freespace buffer */
1970         xfs_inode_t             *dp;            /* incore directory inode */
1971         int                     error;          /* error return code */
1972         xfs_dir2_free_t         *free;          /* freespace structure */
1973         xfs_mount_t             *mp;            /* filesystem mount point */
1974         xfs_trans_t             *tp;            /* transaction pointer */
1975
1976         dp = args->dp;
1977         mp = dp->i_mount;
1978         tp = args->trans;
1979         /*
1980          * Read the freespace block.
1981          */
1982         error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
1983         if (error)
1984                 return error;
1985         /*
1986          * There can be holes in freespace.  If fo is a hole, there's
1987          * nothing to do.
1988          */
1989         if (!bp)
1990                 return 0;
1991         free = bp->b_addr;
1992         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1993         /*
1994          * If there are used entries, there's nothing to do.
1995          */
1996         if (be32_to_cpu(free->hdr.nused) > 0) {
1997                 xfs_trans_brelse(tp, bp);
1998                 *rvalp = 0;
1999                 return 0;
2000         }
2001         /*
2002          * Blow the block away.
2003          */
2004         if ((error =
2005             xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
2006                     bp))) {
2007                 /*
2008                  * Can't fail with ENOSPC since that only happens with no
2009                  * space reservation, when breaking up an extent into two
2010                  * pieces.  This is the last block of an extent.
2011                  */
2012                 ASSERT(error != ENOSPC);
2013                 xfs_trans_brelse(tp, bp);
2014                 return error;
2015         }
2016         /*
2017          * Return that we succeeded.
2018          */
2019         *rvalp = 1;
2020         return 0;
2021 }