Pileus Git - ~andy/linux/blob - fs/xfs/xfs_dir2_node.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_log.h"
  22 #include "xfs_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
  74 static void
  75 xfs_dir2_free_read_verify(
  76         struct xfs_buf  *bp)
  77 {
  78         xfs_dir2_free_verify(bp);
  79 }
  80
  81 static void
  82 xfs_dir2_free_write_verify(
  83         struct xfs_buf  *bp)
  84 {
  85         xfs_dir2_free_verify(bp);
  86 }
  87
  88 static const struct xfs_buf_ops xfs_dir2_free_buf_ops = {
  89         .verify_read = xfs_dir2_free_read_verify,
  90         .verify_write = xfs_dir2_free_write_verify,
  91 };
  92
  93
  94 static int
  95 __xfs_dir2_free_read(
  96         struct xfs_trans        *tp,
  97         struct xfs_inode        *dp,
  98         xfs_dablk_t             fbno,
  99         xfs_daddr_t             mappedbno,
 100         struct xfs_buf          **bpp)
 101 {
 102         return xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 103                                 XFS_DATA_FORK, &xfs_dir2_free_buf_ops);
 104 }
 105
 106 int
 107 xfs_dir2_free_read(
 108         struct xfs_trans        *tp,
 109         struct xfs_inode        *dp,
 110         xfs_dablk_t             fbno,
 111         struct xfs_buf          **bpp)
 112 {
 113         return __xfs_dir2_free_read(tp, dp, fbno, -1, bpp);
 114 }
 115
 116 static int
 117 xfs_dir2_free_try_read(
 118         struct xfs_trans        *tp,
 119         struct xfs_inode        *dp,
 120         xfs_dablk_t             fbno,
 121         struct xfs_buf          **bpp)
 122 {
 123         return __xfs_dir2_free_read(tp, dp, fbno, -2, bpp);
 124 }
 125
 126 /*
 127  * Log entries from a freespace block.
 128  */
 129 STATIC void
 130 xfs_dir2_free_log_bests(
 131         struct xfs_trans        *tp,
 132         struct xfs_buf          *bp,
 133         int                     first,          /* first entry to log */
 134         int                     last)           /* last entry to log */
 135 {
 136         xfs_dir2_free_t         *free;          /* freespace structure */
 137
 138         free = bp->b_addr;
 139         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
 140         xfs_trans_log_buf(tp, bp,
 141                 (uint)((char *)&free->bests[first] - (char *)free),
 142                 (uint)((char *)&free->bests[last] - (char *)free +
 143                        sizeof(free->bests[0]) - 1));
 144 }
 145
 146 /*
 147  * Log header from a freespace block.
 148  */
 149 static void
 150 xfs_dir2_free_log_header(
 151         struct xfs_trans        *tp,
 152         struct xfs_buf          *bp)
 153 {
 154         xfs_dir2_free_t         *free;          /* freespace structure */
 155
 156         free = bp->b_addr;
 157         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
 158         xfs_trans_log_buf(tp, bp, (uint)((char *)&free->hdr - (char *)free),
 159                 (uint)(sizeof(xfs_dir2_free_hdr_t) - 1));
 160 }
 161
 162 /*
 163  * Convert a leaf-format directory to a node-format directory.
 164  * We need to change the magic number of the leaf block, and copy
 165  * the freespace table out of the leaf block into its own block.
 166  */
 167 int                                             /* error */
 168 xfs_dir2_leaf_to_node(
 169         xfs_da_args_t           *args,          /* operation arguments */
 170         struct xfs_buf          *lbp)           /* leaf buffer */
 171 {
 172         xfs_inode_t             *dp;            /* incore directory inode */
 173         int                     error;          /* error return value */
 174         struct xfs_buf          *fbp;           /* freespace buffer */
 175         xfs_dir2_db_t           fdb;            /* freespace block number */
 176         xfs_dir2_free_t         *free;          /* freespace structure */
 177         __be16                  *from;          /* pointer to freespace entry */
 178         int                     i;              /* leaf freespace index */
 179         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 180         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 181         xfs_mount_t             *mp;            /* filesystem mount point */
 182         int                     n;              /* count of live freespc ents */
 183         xfs_dir2_data_off_t     off;            /* freespace entry value */
 184         __be16                  *to;            /* pointer to freespace entry */
 185         xfs_trans_t             *tp;            /* transaction pointer */
 186
 187         trace_xfs_dir2_leaf_to_node(args);
 188
 189         dp = args->dp;
 190         mp = dp->i_mount;
 191         tp = args->trans;
 192         /*
 193          * Add a freespace block to the directory.
 194          */
 195         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 196                 return error;
 197         }
 198         ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
 199         /*
 200          * Get the buffer for the new freespace block.
 201          */
 202         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb), -1, &fbp,
 203                                 XFS_DATA_FORK);
 204         if (error)
 205                 return error;
 206         fbp->b_ops = &xfs_dir2_free_buf_ops;
 207
 208         free = fbp->b_addr;
 209         leaf = lbp->b_addr;
 210         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 211         /*
 212          * Initialize the freespace block header.
 213          */
 214         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
 215         free->hdr.firstdb = 0;
 216         ASSERT(be32_to_cpu(ltp->bestcount) <= (uint)dp->i_d.di_size / mp->m_dirblksize);
 217         free->hdr.nvalid = ltp->bestcount;
 218         /*
 219          * Copy freespace entries from the leaf block to the new block.
 220          * Count active entries.
 221          */
 222         for (i = n = 0, from = xfs_dir2_leaf_bests_p(ltp), to = free->bests;
 223              i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 224                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 225                         n++;
 226                 *to = cpu_to_be16(off);
 227         }
 228         free->hdr.nused = cpu_to_be32(n);
 229
 230         lbp->b_ops = &xfs_dir2_leafn_buf_ops;
 231         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 232
 233         /*
 234          * Log everything.
 235          */
 236         xfs_dir2_leaf_log_header(tp, lbp);
 237         xfs_dir2_free_log_header(tp, fbp);
 238         xfs_dir2_free_log_bests(tp, fbp, 0, be32_to_cpu(free->hdr.nvalid) - 1);
 239         xfs_dir2_leafn_check(dp, lbp);
 240         return 0;
 241 }
 242
 243 /*
 244  * Add a leaf entry to a leaf block in a node-form directory.
 245  * The other work necessary is done from the caller.
 246  */
 247 static int                                      /* error */
 248 xfs_dir2_leafn_add(
 249         struct xfs_buf          *bp,            /* leaf buffer */
 250         xfs_da_args_t           *args,          /* operation arguments */
 251         int                     index)          /* insertion pt for new entry */
 252 {
 253         int                     compact;        /* compacting stale leaves */
 254         xfs_inode_t             *dp;            /* incore directory inode */
 255         int                     highstale;      /* next stale entry */
 256         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 257         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 258         int                     lfloghigh;      /* high leaf entry logging */
 259         int                     lfloglow;       /* low leaf entry logging */
 260         int                     lowstale;       /* previous stale entry */
 261         xfs_mount_t             *mp;            /* filesystem mount point */
 262         xfs_trans_t             *tp;            /* transaction pointer */
 263
 264         trace_xfs_dir2_leafn_add(args, index);
 265
 266         dp = args->dp;
 267         mp = dp->i_mount;
 268         tp = args->trans;
 269         leaf = bp->b_addr;
 270
 271         /*
 272          * Quick check just to make sure we are not going to index
 273          * into other peoples memory
 274          */
 275         if (index < 0)
 276                 return XFS_ERROR(EFSCORRUPTED);
 277
 278         /*
 279          * If there are already the maximum number of leaf entries in
 280          * the block, if there are no stale entries it won't fit.
 281          * Caller will do a split.  If there are stale entries we'll do
 282          * a compact.
 283          */
 284
 285         if (be16_to_cpu(leaf->hdr.count) == xfs_dir2_max_leaf_ents(mp)) {
 286                 if (!leaf->hdr.stale)
 287                         return XFS_ERROR(ENOSPC);
 288                 compact = be16_to_cpu(leaf->hdr.stale) > 1;
 289         } else
 290                 compact = 0;
 291         ASSERT(index == 0 || be32_to_cpu(leaf->ents[index - 1].hashval) <= args->hashval);
 292         ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
 293                be32_to_cpu(leaf->ents[index].hashval) >= args->hashval);
 294
 295         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 296                 return 0;
 297
 298         /*
 299          * Compact out all but one stale leaf entry.  Leaves behind
 300          * the entry closest to index.
 301          */
 302         if (compact) {
 303                 xfs_dir2_leaf_compact_x1(bp, &index, &lowstale, &highstale,
 304                         &lfloglow, &lfloghigh);
 305         }
 306         /*
 307          * Set impossible logging indices for this case.
 308          */
 309         else if (leaf->hdr.stale) {
 310                 lfloglow = be16_to_cpu(leaf->hdr.count);
 311                 lfloghigh = -1;
 312         }
 313
 314         /*
 315          * Insert the new entry, log everything.
 316          */
 317         lep = xfs_dir2_leaf_find_entry(leaf, index, compact, lowstale,
 318                                        highstale, &lfloglow, &lfloghigh);
 319
 320         lep->hashval = cpu_to_be32(args->hashval);
 321         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
 322                                 args->blkno, args->index));
 323         xfs_dir2_leaf_log_header(tp, bp);
 324         xfs_dir2_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
 325         xfs_dir2_leafn_check(dp, bp);
 326         return 0;
 327 }
 328
 329 #ifdef DEBUG
 330 /*
 331  * Check internal consistency of a leafn block.
 332  */
 333 void
 334 xfs_dir2_leafn_check(
 335         struct xfs_inode *dp,
 336         struct xfs_buf  *bp)
 337 {
 338         int             i;                      /* leaf index */
 339         xfs_dir2_leaf_t *leaf;                  /* leaf structure */
 340         xfs_mount_t     *mp;                    /* filesystem mount point */
 341         int             stale;                  /* count of stale leaves */
 342
 343         leaf = bp->b_addr;
 344         mp = dp->i_mount;
 345         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 346         ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
 347         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
 348                 if (i + 1 < be16_to_cpu(leaf->hdr.count)) {
 349                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
 350                                be32_to_cpu(leaf->ents[i + 1].hashval));
 351                 }
 352                 if (leaf->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 353                         stale++;
 354         }
 355         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
 356 }
 357 #endif  /* DEBUG */
 358
 359 /*
 360  * Return the last hash value in the leaf.
 361  * Stale entries are ok.
 362  */
 363 xfs_dahash_t                                    /* hash value */
 364 xfs_dir2_leafn_lasthash(
 365         struct xfs_buf  *bp,                    /* leaf buffer */
 366         int             *count)                 /* count of entries in leaf */
 367 {
 368         xfs_dir2_leaf_t *leaf;                  /* leaf structure */
 369
 370         leaf = bp->b_addr;
 371         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 372         if (count)
 373                 *count = be16_to_cpu(leaf->hdr.count);
 374         if (!leaf->hdr.count)
 375                 return 0;
 376         return be32_to_cpu(leaf->ents[be16_to_cpu(leaf->hdr.count) - 1].hashval);
 377 }
 378
 379 /*
 380  * Look up a leaf entry for space to add a name in a node-format leaf block.
 381  * The extrablk in state is a freespace block.
 382  */
 383 STATIC int
 384 xfs_dir2_leafn_lookup_for_addname(
 385         struct xfs_buf          *bp,            /* leaf buffer */
 386         xfs_da_args_t           *args,          /* operation arguments */
 387         int                     *indexp,        /* out: leaf entry index */
 388         xfs_da_state_t          *state)         /* state to fill in */
 389 {
 390         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 391         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 392         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 393         xfs_inode_t             *dp;            /* incore directory inode */
 394         int                     error;          /* error return value */
 395         int                     fi;             /* free entry index */
 396         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 397         int                     index;          /* leaf entry index */
 398         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 399         int                     length;         /* length of new data entry */
 400         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 401         xfs_mount_t             *mp;            /* filesystem mount point */
 402         xfs_dir2_db_t           newdb;          /* new data block number */
 403         xfs_dir2_db_t           newfdb;         /* new free block number */
 404         xfs_trans_t             *tp;            /* transaction pointer */
 405
 406         dp = args->dp;
 407         tp = args->trans;
 408         mp = dp->i_mount;
 409         leaf = bp->b_addr;
 410         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 411 #ifdef __KERNEL__
 412         ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
 413 #endif
 414         xfs_dir2_leafn_check(dp, bp);
 415         /*
 416          * Look up the hash value in the leaf entries.
 417          */
 418         index = xfs_dir2_leaf_search_hash(args, bp);
 419         /*
 420          * Do we have a buffer coming in?
 421          */
 422         if (state->extravalid) {
 423                 /* If so, it's a free block buffer, get the block number. */
 424                 curbp = state->extrablk.bp;
 425                 curfdb = state->extrablk.blkno;
 426                 free = curbp->b_addr;
 427                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
 428         }
 429         length = xfs_dir2_data_entsize(args->namelen);
 430         /*
 431          * Loop over leaf entries with the right hash value.
 432          */
 433         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
 434                                 be32_to_cpu(lep->hashval) == args->hashval;
 435                                 lep++, index++) {
 436                 /*
 437                  * Skip stale leaf entries.
 438                  */
 439                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 440                         continue;
 441                 /*
 442                  * Pull the data block number from the entry.
 443                  */
 444                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 445                 /*
 446                  * For addname, we're looking for a place to put the new entry.
 447                  * We want to use a data block with an entry of equal
 448                  * hash value to ours if there is one with room.
 449                  *
 450                  * If this block isn't the data block we already have
 451                  * in hand, take a look at it.
 452                  */
 453                 if (newdb != curdb) {
 454                         curdb = newdb;
 455                         /*
 456                          * Convert the data block to the free block
 457                          * holding its freespace information.
 458                          */
 459                         newfdb = xfs_dir2_db_to_fdb(mp, newdb);
 460                         /*
 461                          * If it's not the one we have in hand, read it in.
 462                          */
 463                         if (newfdb != curfdb) {
 464                                 /*
 465                                  * If we had one before, drop it.
 466                                  */
 467                                 if (curbp)
 468                                         xfs_trans_brelse(tp, curbp);
 469
 470                                 error = xfs_dir2_free_read(tp, dp,
 471                                                 xfs_dir2_db_to_da(mp, newfdb),
 472                                                 &curbp);
 473                                 if (error)
 474                                         return error;
 475                                 free = curbp->b_addr;
 476                                 ASSERT(be32_to_cpu(free->hdr.magic) ==
 477                                         XFS_DIR2_FREE_MAGIC);
 478                                 ASSERT((be32_to_cpu(free->hdr.firstdb) %
 479                                         xfs_dir2_free_max_bests(mp)) == 0);
 480                                 ASSERT(be32_to_cpu(free->hdr.firstdb) <= curdb);
 481                                 ASSERT(curdb < be32_to_cpu(free->hdr.firstdb) +
 482                                         be32_to_cpu(free->hdr.nvalid));
 483                         }
 484                         /*
 485                          * Get the index for our entry.
 486                          */
 487                         fi = xfs_dir2_db_to_fdindex(mp, curdb);
 488                         /*
 489                          * If it has room, return it.
 490                          */
 491                         if (unlikely(free->bests[fi] ==
 492                             cpu_to_be16(NULLDATAOFF))) {
 493                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 494                                                         XFS_ERRLEVEL_LOW, mp);
 495                                 if (curfdb != newfdb)
 496                                         xfs_trans_brelse(tp, curbp);
 497                                 return XFS_ERROR(EFSCORRUPTED);
 498                         }
 499                         curfdb = newfdb;
 500                         if (be16_to_cpu(free->bests[fi]) >= length)
 501                                 goto out;
 502                 }
 503         }
 504         /* Didn't find any space */
 505         fi = -1;
 506 out:
 507         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 508         if (curbp) {
 509                 /* Giving back a free block. */
 510                 state->extravalid = 1;
 511                 state->extrablk.bp = curbp;
 512                 state->extrablk.index = fi;
 513                 state->extrablk.blkno = curfdb;
 514                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 515         } else {
 516                 state->extravalid = 0;
 517         }
 518         /*
 519          * Return the index, that will be the insertion point.
 520          */
 521         *indexp = index;
 522         return XFS_ERROR(ENOENT);
 523 }
 524
 525 /*
 526  * Look up a leaf entry in a node-format leaf block.
 527  * The extrablk in state a data block.
 528  */
 529 STATIC int
 530 xfs_dir2_leafn_lookup_for_entry(
 531         struct xfs_buf          *bp,            /* leaf buffer */
 532         xfs_da_args_t           *args,          /* operation arguments */
 533         int                     *indexp,        /* out: leaf entry index */
 534         xfs_da_state_t          *state)         /* state to fill in */
 535 {
 536         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 537         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 538         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 539         xfs_inode_t             *dp;            /* incore directory inode */
 540         int                     error;          /* error return value */
 541         int                     index;          /* leaf entry index */
 542         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 543         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 544         xfs_mount_t             *mp;            /* filesystem mount point */
 545         xfs_dir2_db_t           newdb;          /* new data block number */
 546         xfs_trans_t             *tp;            /* transaction pointer */
 547         enum xfs_dacmp          cmp;            /* comparison result */
 548
 549         dp = args->dp;
 550         tp = args->trans;
 551         mp = dp->i_mount;
 552         leaf = bp->b_addr;
 553         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 554 #ifdef __KERNEL__
 555         ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
 556 #endif
 557         xfs_dir2_leafn_check(dp, bp);
 558         /*
 559          * Look up the hash value in the leaf entries.
 560          */
 561         index = xfs_dir2_leaf_search_hash(args, bp);
 562         /*
 563          * Do we have a buffer coming in?
 564          */
 565         if (state->extravalid) {
 566                 curbp = state->extrablk.bp;
 567                 curdb = state->extrablk.blkno;
 568         }
 569         /*
 570          * Loop over leaf entries with the right hash value.
 571          */
 572         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
 573                                 be32_to_cpu(lep->hashval) == args->hashval;
 574                                 lep++, index++) {
 575                 /*
 576                  * Skip stale leaf entries.
 577                  */
 578                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 579                         continue;
 580                 /*
 581                  * Pull the data block number from the entry.
 582                  */
 583                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 584                 /*
 585                  * Not adding a new entry, so we really want to find
 586                  * the name given to us.
 587                  *
 588                  * If it's a different data block, go get it.
 589                  */
 590                 if (newdb != curdb) {
 591                         /*
 592                          * If we had a block before that we aren't saving
 593                          * for a CI name, drop it
 594                          */
 595                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 596                                                 curdb != state->extrablk.blkno))
 597                                 xfs_trans_brelse(tp, curbp);
 598                         /*
 599                          * If needing the block that is saved with a CI match,
 600                          * use it otherwise read in the new data block.
 601                          */
 602                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 603                                         newdb == state->extrablk.blkno) {
 604                                 ASSERT(state->extravalid);
 605                                 curbp = state->extrablk.bp;
 606                         } else {
 607                                 error = xfs_dir2_data_read(tp, dp,
 608                                                 xfs_dir2_db_to_da(mp, newdb),
 609                                                 -1, &curbp);
 610                                 if (error)
 611                                         return error;
 612                         }
 613                         xfs_dir2_data_check(dp, curbp);
 614                         curdb = newdb;
 615                 }
 616                 /*
 617                  * Point to the data entry.
 618                  */
 619                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 620                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
 621                 /*
 622                  * Compare the entry and if it's an exact match, return
 623                  * EEXIST immediately. If it's the first case-insensitive
 624                  * match, store the block & inode number and continue looking.
 625                  */
 626                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 627                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 628                         /* If there is a CI match block, drop it */
 629                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 630                                                 curdb != state->extrablk.blkno)
 631                                 xfs_trans_brelse(tp, state->extrablk.bp);
 632                         args->cmpresult = cmp;
 633                         args->inumber = be64_to_cpu(dep->inumber);
 634                         *indexp = index;
 635                         state->extravalid = 1;
 636                         state->extrablk.bp = curbp;
 637                         state->extrablk.blkno = curdb;
 638                         state->extrablk.index = (int)((char *)dep -
 639                                                         (char *)curbp->b_addr);
 640                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 641                         curbp->b_ops = &xfs_dir2_data_buf_ops;
 642                         if (cmp == XFS_CMP_EXACT)
 643                                 return XFS_ERROR(EEXIST);
 644                 }
 645         }
 646         ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
 647                                         (args->op_flags & XFS_DA_OP_OKNOENT));
 648         if (curbp) {
 649                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 650                         /* Giving back last used data block. */
 651                         state->extravalid = 1;
 652                         state->extrablk.bp = curbp;
 653                         state->extrablk.index = -1;
 654                         state->extrablk.blkno = curdb;
 655                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 656                         curbp->b_ops = &xfs_dir2_data_buf_ops;
 657                 } else {
 658                         /* If the curbp is not the CI match block, drop it */
 659                         if (state->extrablk.bp != curbp)
 660                                 xfs_trans_brelse(tp, curbp);
 661                 }
 662         } else {
 663                 state->extravalid = 0;
 664         }
 665         *indexp = index;
 666         return XFS_ERROR(ENOENT);
 667 }
 668
 669 /*
 670  * Look up a leaf entry in a node-format leaf block.
 671  * If this is an addname then the extrablk in state is a freespace block,
 672  * otherwise it's a data block.
 673  */
 674 int
 675 xfs_dir2_leafn_lookup_int(
 676         struct xfs_buf          *bp,            /* leaf buffer */
 677         xfs_da_args_t           *args,          /* operation arguments */
 678         int                     *indexp,        /* out: leaf entry index */
 679         xfs_da_state_t          *state)         /* state to fill in */
 680 {
 681         if (args->op_flags & XFS_DA_OP_ADDNAME)
 682                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 683                                                         state);
 684         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 685 }
 686
 687 /*
 688  * Move count leaf entries from source to destination leaf.
 689  * Log entries and headers.  Stale entries are preserved.
 690  */
 691 static void
 692 xfs_dir2_leafn_moveents(
 693         xfs_da_args_t   *args,                  /* operation arguments */
 694         struct xfs_buf  *bp_s,                  /* source leaf buffer */
 695         int             start_s,                /* source leaf index */
 696         struct xfs_buf  *bp_d,                  /* destination leaf buffer */
 697         int             start_d,                /* destination leaf index */
 698         int             count)                  /* count of leaves to copy */
 699 {
 700         xfs_dir2_leaf_t *leaf_d;                /* destination leaf structure */
 701         xfs_dir2_leaf_t *leaf_s;                /* source leaf structure */
 702         int             stale;                  /* count stale leaves copied */
 703         xfs_trans_t     *tp;                    /* transaction pointer */
 704
 705         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 706
 707         /*
 708          * Silently return if nothing to do.
 709          */
 710         if (count == 0) {
 711                 return;
 712         }
 713         tp = args->trans;
 714         leaf_s = bp_s->b_addr;
 715         leaf_d = bp_d->b_addr;
 716         /*
 717          * If the destination index is not the end of the current
 718          * destination leaf entries, open up a hole in the destination
 719          * to hold the new entries.
 720          */
 721         if (start_d < be16_to_cpu(leaf_d->hdr.count)) {
 722                 memmove(&leaf_d->ents[start_d + count], &leaf_d->ents[start_d],
 723                         (be16_to_cpu(leaf_d->hdr.count) - start_d) *
 724                         sizeof(xfs_dir2_leaf_entry_t));
 725                 xfs_dir2_leaf_log_ents(tp, bp_d, start_d + count,
 726                         count + be16_to_cpu(leaf_d->hdr.count) - 1);
 727         }
 728         /*
 729          * If the source has stale leaves, count the ones in the copy range
 730          * so we can update the header correctly.
 731          */
 732         if (leaf_s->hdr.stale) {
 733                 int     i;                      /* temp leaf index */
 734
 735                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 736                         if (leaf_s->ents[i].address ==
 737                             cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 738                                 stale++;
 739                 }
 740         } else
 741                 stale = 0;
 742         /*
 743          * Copy the leaf entries from source to destination.
 744          */
 745         memcpy(&leaf_d->ents[start_d], &leaf_s->ents[start_s],
 746                 count * sizeof(xfs_dir2_leaf_entry_t));
 747         xfs_dir2_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
 748         /*
 749          * If there are source entries after the ones we copied,
 750          * delete the ones we copied by sliding the next ones down.
 751          */
 752         if (start_s + count < be16_to_cpu(leaf_s->hdr.count)) {
 753                 memmove(&leaf_s->ents[start_s], &leaf_s->ents[start_s + count],
 754                         count * sizeof(xfs_dir2_leaf_entry_t));
 755                 xfs_dir2_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
 756         }
 757         /*
 758          * Update the headers and log them.
 759          */
 760         be16_add_cpu(&leaf_s->hdr.count, -(count));
 761         be16_add_cpu(&leaf_s->hdr.stale, -(stale));
 762         be16_add_cpu(&leaf_d->hdr.count, count);
 763         be16_add_cpu(&leaf_d->hdr.stale, stale);
 764         xfs_dir2_leaf_log_header(tp, bp_s);
 765         xfs_dir2_leaf_log_header(tp, bp_d);
 766         xfs_dir2_leafn_check(args->dp, bp_s);
 767         xfs_dir2_leafn_check(args->dp, bp_d);
 768 }
 769
 770 /*
 771  * Determine the sort order of two leaf blocks.
 772  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 773  */
 774 int                                             /* sort order */
 775 xfs_dir2_leafn_order(
 776         struct xfs_buf  *leaf1_bp,              /* leaf1 buffer */
 777         struct xfs_buf  *leaf2_bp)              /* leaf2 buffer */
 778 {
 779         xfs_dir2_leaf_t *leaf1;                 /* leaf1 structure */
 780         xfs_dir2_leaf_t *leaf2;                 /* leaf2 structure */
 781
 782         leaf1 = leaf1_bp->b_addr;
 783         leaf2 = leaf2_bp->b_addr;
 784         ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 785         ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
 786         if (be16_to_cpu(leaf1->hdr.count) > 0 &&
 787             be16_to_cpu(leaf2->hdr.count) > 0 &&
 788             (be32_to_cpu(leaf2->ents[0].hashval) < be32_to_cpu(leaf1->ents[0].hashval) ||
 789              be32_to_cpu(leaf2->ents[be16_to_cpu(leaf2->hdr.count) - 1].hashval) <
 790              be32_to_cpu(leaf1->ents[be16_to_cpu(leaf1->hdr.count) - 1].hashval)))
 791                 return 1;
 792         return 0;
 793 }
 794
 795 /*
 796  * Rebalance leaf entries between two leaf blocks.
 797  * This is actually only called when the second block is new,
 798  * though the code deals with the general case.
 799  * A new entry will be inserted in one of the blocks, and that
 800  * entry is taken into account when balancing.
 801  */
 802 static void
 803 xfs_dir2_leafn_rebalance(
 804         xfs_da_state_t          *state,         /* btree cursor */
 805         xfs_da_state_blk_t      *blk1,          /* first btree block */
 806         xfs_da_state_blk_t      *blk2)          /* second btree block */
 807 {
 808         xfs_da_args_t           *args;          /* operation arguments */
 809         int                     count;          /* count (& direction) leaves */
 810         int                     isleft;         /* new goes in left leaf */
 811         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 812         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 813         int                     mid;            /* midpoint leaf index */
 814 #ifdef DEBUG
 815         int                     oldstale;       /* old count of stale leaves */
 816 #endif
 817         int                     oldsum;         /* old total leaf count */
 818         int                     swap;           /* swapped leaf blocks */
 819
 820         args = state->args;
 821         /*
 822          * If the block order is wrong, swap the arguments.
 823          */
 824         if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
 825                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
 826
 827                 tmp = blk1;
 828                 blk1 = blk2;
 829                 blk2 = tmp;
 830         }
 831         leaf1 = blk1->bp->b_addr;
 832         leaf2 = blk2->bp->b_addr;
 833         oldsum = be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count);
 834 #ifdef DEBUG
 835         oldstale = be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale);
 836 #endif
 837         mid = oldsum >> 1;
 838         /*
 839          * If the old leaf count was odd then the new one will be even,
 840          * so we need to divide the new count evenly.
 841          */
 842         if (oldsum & 1) {
 843                 xfs_dahash_t    midhash;        /* middle entry hash value */
 844
 845                 if (mid >= be16_to_cpu(leaf1->hdr.count))
 846                         midhash = be32_to_cpu(leaf2->ents[mid - be16_to_cpu(leaf1->hdr.count)].hashval);
 847                 else
 848                         midhash = be32_to_cpu(leaf1->ents[mid].hashval);
 849                 isleft = args->hashval <= midhash;
 850         }
 851         /*
 852          * If the old count is even then the new count is odd, so there's
 853          * no preferred side for the new entry.
 854          * Pick the left one.
 855          */
 856         else
 857                 isleft = 1;
 858         /*
 859          * Calculate moved entry count.  Positive means left-to-right,
 860          * negative means right-to-left.  Then move the entries.
 861          */
 862         count = be16_to_cpu(leaf1->hdr.count) - mid + (isleft == 0);
 863         if (count > 0)
 864                 xfs_dir2_leafn_moveents(args, blk1->bp,
 865                         be16_to_cpu(leaf1->hdr.count) - count, blk2->bp, 0, count);
 866         else if (count < 0)
 867                 xfs_dir2_leafn_moveents(args, blk2->bp, 0, blk1->bp,
 868                         be16_to_cpu(leaf1->hdr.count), count);
 869         ASSERT(be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count) == oldsum);
 870         ASSERT(be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale) == oldstale);
 871         /*
 872          * Mark whether we're inserting into the old or new leaf.
 873          */
 874         if (be16_to_cpu(leaf1->hdr.count) < be16_to_cpu(leaf2->hdr.count))
 875                 state->inleaf = swap;
 876         else if (be16_to_cpu(leaf1->hdr.count) > be16_to_cpu(leaf2->hdr.count))
 877                 state->inleaf = !swap;
 878         else
 879                 state->inleaf =
 880                         swap ^ (blk1->index <= be16_to_cpu(leaf1->hdr.count));
 881         /*
 882          * Adjust the expected index for insertion.
 883          */
 884         if (!state->inleaf)
 885                 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
 886
 887         /*
 888          * Finally sanity check just to make sure we are not returning a
 889          * negative index
 890          */
 891         if(blk2->index < 0) {
 892                 state->inleaf = 1;
 893                 blk2->index = 0;
 894                 xfs_alert(args->dp->i_mount,
 895         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d\n",
 896                         __func__, blk1->index);
 897         }
 898 }
 899
 900 static int
 901 xfs_dir2_data_block_free(
 902         xfs_da_args_t           *args,
 903         struct xfs_dir2_data_hdr *hdr,
 904         struct xfs_dir2_free    *free,
 905         xfs_dir2_db_t           fdb,
 906         int                     findex,
 907         struct xfs_buf          *fbp,
 908         int                     longest)
 909 {
 910         struct xfs_trans        *tp = args->trans;
 911         int                     logfree = 0;
 912
 913         if (!hdr) {
 914                 /* One less used entry in the free table.  */
 915                 be32_add_cpu(&free->hdr.nused, -1);
 916                 xfs_dir2_free_log_header(tp, fbp);
 917
 918                 /*
 919                  * If this was the last entry in the table, we can trim the
 920                  * table size back.  There might be other entries at the end
 921                  * referring to non-existent data blocks, get those too.
 922                  */
 923                 if (findex == be32_to_cpu(free->hdr.nvalid) - 1) {
 924                         int     i;              /* free entry index */
 925
 926                         for (i = findex - 1; i >= 0; i--) {
 927                                 if (free->bests[i] != cpu_to_be16(NULLDATAOFF))
 928                                         break;
 929                         }
 930                         free->hdr.nvalid = cpu_to_be32(i + 1);
 931                         logfree = 0;
 932                 } else {
 933                         /* Not the last entry, just punch it out.  */
 934                         free->bests[findex] = cpu_to_be16(NULLDATAOFF);
 935                         logfree = 1;
 936                 }
 937                 /*
 938                  * If there are no useful entries left in the block,
 939                  * get rid of the block if we can.
 940                  */
 941                 if (!free->hdr.nused) {
 942                         int error;
 943
 944                         error = xfs_dir2_shrink_inode(args, fdb, fbp);
 945                         if (error == 0) {
 946                                 fbp = NULL;
 947                                 logfree = 0;
 948                         } else if (error != ENOSPC || args->total != 0)
 949                                 return error;
 950                         /*
 951                          * It's possible to get ENOSPC if there is no
 952                          * space reservation.  In this case some one
 953                          * else will eventually get rid of this block.
 954                          */
 955                 }
 956         } else {
 957                 /*
 958                  * Data block is not empty, just set the free entry to the new
 959                  * value.
 960                  */
 961                 free->bests[findex] = cpu_to_be16(longest);
 962                 logfree = 1;
 963         }
 964
 965         /* Log the free entry that changed, unless we got rid of it.  */
 966         if (logfree)
 967                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
 968         return 0;
 969 }
 970
 971 /*
 972  * Remove an entry from a node directory.
 973  * This removes the leaf entry and the data entry,
 974  * and updates the free block if necessary.
 975  */
 976 static int                                      /* error */
 977 xfs_dir2_leafn_remove(
 978         xfs_da_args_t           *args,          /* operation arguments */
 979         struct xfs_buf          *bp,            /* leaf buffer */
 980         int                     index,          /* leaf entry index */
 981         xfs_da_state_blk_t      *dblk,          /* data block */
 982         int                     *rval)          /* resulting block needs join */
 983 {
 984         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 985         xfs_dir2_db_t           db;             /* data block number */
 986         struct xfs_buf          *dbp;           /* data block buffer */
 987         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 988         xfs_inode_t             *dp;            /* incore directory inode */
 989         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 990         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 991         int                     longest;        /* longest data free entry */
 992         int                     off;            /* data block entry offset */
 993         xfs_mount_t             *mp;            /* filesystem mount point */
 994         int                     needlog;        /* need to log data header */
 995         int                     needscan;       /* need to rescan data frees */
 996         xfs_trans_t             *tp;            /* transaction pointer */
 997
 998         trace_xfs_dir2_leafn_remove(args, index);
 999
1000         dp = args->dp;
1001         tp = args->trans;
1002         mp = dp->i_mount;
1003         leaf = bp->b_addr;
1004         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1005         /*
1006          * Point to the entry we're removing.
1007          */
1008         lep = &leaf->ents[index];
1009         /*
1010          * Extract the data block and offset from the entry.
1011          */
1012         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1013         ASSERT(dblk->blkno == db);
1014         off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
1015         ASSERT(dblk->index == off);
1016         /*
1017          * Kill the leaf entry by marking it stale.
1018          * Log the leaf block changes.
1019          */
1020         be16_add_cpu(&leaf->hdr.stale, 1);
1021         xfs_dir2_leaf_log_header(tp, bp);
1022         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1023         xfs_dir2_leaf_log_ents(tp, bp, index, index);
1024         /*
1025          * Make the data entry free.  Keep track of the longest freespace
1026          * in the data block in case it changes.
1027          */
1028         dbp = dblk->bp;
1029         hdr = dbp->b_addr;
1030         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1031         longest = be16_to_cpu(hdr->bestfree[0].length);
1032         needlog = needscan = 0;
1033         xfs_dir2_data_make_free(tp, dbp, off,
1034                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1035         /*
1036          * Rescan the data block freespaces for bestfree.
1037          * Log the data block header if needed.
1038          */
1039         if (needscan)
1040                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1041         if (needlog)
1042                 xfs_dir2_data_log_header(tp, dbp);
1043         xfs_dir2_data_check(dp, dbp);
1044         /*
1045          * If the longest data block freespace changes, need to update
1046          * the corresponding freeblock entry.
1047          */
1048         if (longest < be16_to_cpu(hdr->bestfree[0].length)) {
1049                 int             error;          /* error return value */
1050                 struct xfs_buf  *fbp;           /* freeblock buffer */
1051                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1052                 int             findex;         /* index in freeblock entries */
1053                 xfs_dir2_free_t *free;          /* freeblock structure */
1054
1055                 /*
1056                  * Convert the data block number to a free block,
1057                  * read in the free block.
1058                  */
1059                 fdb = xfs_dir2_db_to_fdb(mp, db);
1060                 error = xfs_dir2_free_read(tp, dp, xfs_dir2_db_to_da(mp, fdb),
1061                                            &fbp);
1062                 if (error)
1063                         return error;
1064                 free = fbp->b_addr;
1065                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1066                 ASSERT(be32_to_cpu(free->hdr.firstdb) ==
1067                        xfs_dir2_free_max_bests(mp) *
1068                        (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
1069                 /*
1070                  * Calculate which entry we need to fix.
1071                  */
1072                 findex = xfs_dir2_db_to_fdindex(mp, db);
1073                 longest = be16_to_cpu(hdr->bestfree[0].length);
1074                 /*
1075                  * If the data block is now empty we can get rid of it
1076                  * (usually).
1077                  */
1078                 if (longest == mp->m_dirblksize - (uint)sizeof(*hdr)) {
1079                         /*
1080                          * Try to punch out the data block.
1081                          */
1082                         error = xfs_dir2_shrink_inode(args, db, dbp);
1083                         if (error == 0) {
1084                                 dblk->bp = NULL;
1085                                 hdr = NULL;
1086                         }
1087                         /*
1088                          * We can get ENOSPC if there's no space reservation.
1089                          * In this case just drop the buffer and some one else
1090                          * will eventually get rid of the empty block.
1091                          */
1092                         else if (!(error == ENOSPC && args->total == 0))
1093                                 return error;
1094                 }
1095                 /*
1096                  * If we got rid of the data block, we can eliminate that entry
1097                  * in the free block.
1098                  */
1099                 error = xfs_dir2_data_block_free(args, hdr, free,
1100                                                  fdb, findex, fbp, longest);
1101                 if (error)
1102                         return error;
1103         }
1104
1105         xfs_dir2_leafn_check(dp, bp);
1106         /*
1107          * Return indication of whether this leaf block is empty enough
1108          * to justify trying to join it with a neighbor.
1109          */
1110         *rval =
1111                 ((uint)sizeof(leaf->hdr) +
1112                  (uint)sizeof(leaf->ents[0]) *
1113                  (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale))) <
1114                 mp->m_dir_magicpct;
1115         return 0;
1116 }
1117
1118 /*
1119  * Split the leaf entries in the old block into old and new blocks.
1120  */
1121 int                                             /* error */
1122 xfs_dir2_leafn_split(
1123         xfs_da_state_t          *state,         /* btree cursor */
1124         xfs_da_state_blk_t      *oldblk,        /* original block */
1125         xfs_da_state_blk_t      *newblk)        /* newly created block */
1126 {
1127         xfs_da_args_t           *args;          /* operation arguments */
1128         xfs_dablk_t             blkno;          /* new leaf block number */
1129         int                     error;          /* error return value */
1130         xfs_mount_t             *mp;            /* filesystem mount point */
1131
1132         /*
1133          * Allocate space for a new leaf node.
1134          */
1135         args = state->args;
1136         mp = args->dp->i_mount;
1137         ASSERT(args != NULL);
1138         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1139         error = xfs_da_grow_inode(args, &blkno);
1140         if (error) {
1141                 return error;
1142         }
1143         /*
1144          * Initialize the new leaf block.
1145          */
1146         error = xfs_dir2_leaf_init(args, xfs_dir2_da_to_db(mp, blkno),
1147                 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1148         if (error) {
1149                 return error;
1150         }
1151         newblk->blkno = blkno;
1152         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1153         /*
1154          * Rebalance the entries across the two leaves, link the new
1155          * block into the leaves.
1156          */
1157         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1158         error = xfs_da_blk_link(state, oldblk, newblk);
1159         if (error) {
1160                 return error;
1161         }
1162         /*
1163          * Insert the new entry in the correct block.
1164          */
1165         if (state->inleaf)
1166                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1167         else
1168                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1169         /*
1170          * Update last hashval in each block since we added the name.
1171          */
1172         oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1173         newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1174         xfs_dir2_leafn_check(args->dp, oldblk->bp);
1175         xfs_dir2_leafn_check(args->dp, newblk->bp);
1176         return error;
1177 }
1178
1179 /*
1180  * Check a leaf block and its neighbors to see if the block should be
1181  * collapsed into one or the other neighbor.  Always keep the block
1182  * with the smaller block number.
1183  * If the current block is over 50% full, don't try to join it, return 0.
1184  * If the block is empty, fill in the state structure and return 2.
1185  * If it can be collapsed, fill in the state structure and return 1.
1186  * If nothing can be done, return 0.
1187  */
1188 int                                             /* error */
1189 xfs_dir2_leafn_toosmall(
1190         xfs_da_state_t          *state,         /* btree cursor */
1191         int                     *action)        /* resulting action to take */
1192 {
1193         xfs_da_state_blk_t      *blk;           /* leaf block */
1194         xfs_dablk_t             blkno;          /* leaf block number */
1195         struct xfs_buf          *bp;            /* leaf buffer */
1196         int                     bytes;          /* bytes in use */
1197         int                     count;          /* leaf live entry count */
1198         int                     error;          /* error return value */
1199         int                     forward;        /* sibling block direction */
1200         int                     i;              /* sibling counter */
1201         xfs_da_blkinfo_t        *info;          /* leaf block header */
1202         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1203         int                     rval;           /* result from path_shift */
1204
1205         /*
1206          * Check for the degenerate case of the block being over 50% full.
1207          * If so, it's not worth even looking to see if we might be able
1208          * to coalesce with a sibling.
1209          */
1210         blk = &state->path.blk[state->path.active - 1];
1211         info = blk->bp->b_addr;
1212         ASSERT(info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1213         leaf = (xfs_dir2_leaf_t *)info;
1214         count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1215         bytes = (uint)sizeof(leaf->hdr) + count * (uint)sizeof(leaf->ents[0]);
1216         if (bytes > (state->blocksize >> 1)) {
1217                 /*
1218                  * Blk over 50%, don't try to join.
1219                  */
1220                 *action = 0;
1221                 return 0;
1222         }
1223         /*
1224          * Check for the degenerate case of the block being empty.
1225          * If the block is empty, we'll simply delete it, no need to
1226          * coalesce it with a sibling block.  We choose (arbitrarily)
1227          * to merge with the forward block unless it is NULL.
1228          */
1229         if (count == 0) {
1230                 /*
1231                  * Make altpath point to the block we want to keep and
1232                  * path point to the block we want to drop (this one).
1233                  */
1234                 forward = (info->forw != 0);
1235                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1236                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1237                         &rval);
1238                 if (error)
1239                         return error;
1240                 *action = rval ? 2 : 0;
1241                 return 0;
1242         }
1243         /*
1244          * Examine each sibling block to see if we can coalesce with
1245          * at least 25% free space to spare.  We need to figure out
1246          * whether to merge with the forward or the backward block.
1247          * We prefer coalescing with the lower numbered sibling so as
1248          * to shrink a directory over time.
1249          */
1250         forward = be32_to_cpu(info->forw) < be32_to_cpu(info->back);
1251         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1252                 blkno = forward ? be32_to_cpu(info->forw) : be32_to_cpu(info->back);
1253                 if (blkno == 0)
1254                         continue;
1255                 /*
1256                  * Read the sibling leaf block.
1257                  */
1258                 error = xfs_dir2_leafn_read(state->args->trans, state->args->dp,
1259                                             blkno, -1, &bp);
1260                 if (error)
1261                         return error;
1262
1263                 /*
1264                  * Count bytes in the two blocks combined.
1265                  */
1266                 leaf = (xfs_dir2_leaf_t *)info;
1267                 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1268                 bytes = state->blocksize - (state->blocksize >> 2);
1269                 leaf = bp->b_addr;
1270                 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1271                 count += be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1272                 bytes -= count * (uint)sizeof(leaf->ents[0]);
1273                 /*
1274                  * Fits with at least 25% to spare.
1275                  */
1276                 if (bytes >= 0)
1277                         break;
1278                 xfs_trans_brelse(state->args->trans, bp);
1279         }
1280         /*
1281          * Didn't like either block, give up.
1282          */
1283         if (i >= 2) {
1284                 *action = 0;
1285                 return 0;
1286         }
1287
1288         /*
1289          * Make altpath point to the block we want to keep (the lower
1290          * numbered block) and path point to the block we want to drop.
1291          */
1292         memcpy(&state->altpath, &state->path, sizeof(state->path));
1293         if (blkno < blk->blkno)
1294                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1295                         &rval);
1296         else
1297                 error = xfs_da_path_shift(state, &state->path, forward, 0,
1298                         &rval);
1299         if (error) {
1300                 return error;
1301         }
1302         *action = rval ? 0 : 1;
1303         return 0;
1304 }
1305
1306 /*
1307  * Move all the leaf entries from drop_blk to save_blk.
1308  * This is done as part of a join operation.
1309  */
1310 void
1311 xfs_dir2_leafn_unbalance(
1312         xfs_da_state_t          *state,         /* cursor */
1313         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1314         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1315 {
1316         xfs_da_args_t           *args;          /* operation arguments */
1317         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1318         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1319
1320         args = state->args;
1321         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1322         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1323         drop_leaf = drop_blk->bp->b_addr;
1324         save_leaf = save_blk->bp->b_addr;
1325         ASSERT(drop_leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1326         ASSERT(save_leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1327         /*
1328          * If there are any stale leaf entries, take this opportunity
1329          * to purge them.
1330          */
1331         if (drop_leaf->hdr.stale)
1332                 xfs_dir2_leaf_compact(args, drop_blk->bp);
1333         if (save_leaf->hdr.stale)
1334                 xfs_dir2_leaf_compact(args, save_blk->bp);
1335         /*
1336          * Move the entries from drop to the appropriate end of save.
1337          */
1338         drop_blk->hashval = be32_to_cpu(drop_leaf->ents[be16_to_cpu(drop_leaf->hdr.count) - 1].hashval);
1339         if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1340                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp, 0,
1341                         be16_to_cpu(drop_leaf->hdr.count));
1342         else
1343                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp,
1344                         be16_to_cpu(save_leaf->hdr.count), be16_to_cpu(drop_leaf->hdr.count));
1345         save_blk->hashval = be32_to_cpu(save_leaf->ents[be16_to_cpu(save_leaf->hdr.count) - 1].hashval);
1346         xfs_dir2_leafn_check(args->dp, save_blk->bp);
1347 }
1348
1349 /*
1350  * Top-level node form directory addname routine.
1351  */
1352 int                                             /* error */
1353 xfs_dir2_node_addname(
1354         xfs_da_args_t           *args)          /* operation arguments */
1355 {
1356         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1357         int                     error;          /* error return value */
1358         int                     rval;           /* sub-return value */
1359         xfs_da_state_t          *state;         /* btree cursor */
1360
1361         trace_xfs_dir2_node_addname(args);
1362
1363         /*
1364          * Allocate and initialize the state (btree cursor).
1365          */
1366         state = xfs_da_state_alloc();
1367         state->args = args;
1368         state->mp = args->dp->i_mount;
1369         state->blocksize = state->mp->m_dirblksize;
1370         state->node_ents = state->mp->m_dir_node_ents;
1371         /*
1372          * Look up the name.  We're not supposed to find it, but
1373          * this gives us the insertion point.
1374          */
1375         error = xfs_da_node_lookup_int(state, &rval);
1376         if (error)
1377                 rval = error;
1378         if (rval != ENOENT) {
1379                 goto done;
1380         }
1381         /*
1382          * Add the data entry to a data block.
1383          * Extravalid is set to a freeblock found by lookup.
1384          */
1385         rval = xfs_dir2_node_addname_int(args,
1386                 state->extravalid ? &state->extrablk : NULL);
1387         if (rval) {
1388                 goto done;
1389         }
1390         blk = &state->path.blk[state->path.active - 1];
1391         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1392         /*
1393          * Add the new leaf entry.
1394          */
1395         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1396         if (rval == 0) {
1397                 /*
1398                  * It worked, fix the hash values up the btree.
1399                  */
1400                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1401                         xfs_da_fixhashpath(state, &state->path);
1402         } else {
1403                 /*
1404                  * It didn't work, we need to split the leaf block.
1405                  */
1406                 if (args->total == 0) {
1407                         ASSERT(rval == ENOSPC);
1408                         goto done;
1409                 }
1410                 /*
1411                  * Split the leaf block and insert the new entry.
1412                  */
1413                 rval = xfs_da_split(state);
1414         }
1415 done:
1416         xfs_da_state_free(state);
1417         return rval;
1418 }
1419
1420 /*
1421  * Add the data entry for a node-format directory name addition.
1422  * The leaf entry is added in xfs_dir2_leafn_add.
1423  * We may enter with a freespace block that the lookup found.
1424  */
1425 static int                                      /* error */
1426 xfs_dir2_node_addname_int(
1427         xfs_da_args_t           *args,          /* operation arguments */
1428         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1429 {
1430         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1431         xfs_dir2_db_t           dbno;           /* data block number */
1432         struct xfs_buf          *dbp;           /* data block buffer */
1433         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1434         xfs_inode_t             *dp;            /* incore directory inode */
1435         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1436         int                     error;          /* error return value */
1437         xfs_dir2_db_t           fbno;           /* freespace block number */
1438         struct xfs_buf          *fbp;           /* freespace buffer */
1439         int                     findex;         /* freespace entry index */
1440         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1441         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1442         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1443         int                     length;         /* length of the new entry */
1444         int                     logfree;        /* need to log free entry */
1445         xfs_mount_t             *mp;            /* filesystem mount point */
1446         int                     needlog;        /* need to log data header */
1447         int                     needscan;       /* need to rescan data frees */
1448         __be16                  *tagp;          /* data entry tag pointer */
1449         xfs_trans_t             *tp;            /* transaction pointer */
1450
1451         dp = args->dp;
1452         mp = dp->i_mount;
1453         tp = args->trans;
1454         length = xfs_dir2_data_entsize(args->namelen);
1455         /*
1456          * If we came in with a freespace block that means that lookup
1457          * found an entry with our hash value.  This is the freespace
1458          * block for that data entry.
1459          */
1460         if (fblk) {
1461                 fbp = fblk->bp;
1462                 /*
1463                  * Remember initial freespace block number.
1464                  */
1465                 ifbno = fblk->blkno;
1466                 free = fbp->b_addr;
1467                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1468                 findex = fblk->index;
1469                 /*
1470                  * This means the free entry showed that the data block had
1471                  * space for our entry, so we remembered it.
1472                  * Use that data block.
1473                  */
1474                 if (findex >= 0) {
1475                         ASSERT(findex < be32_to_cpu(free->hdr.nvalid));
1476                         ASSERT(be16_to_cpu(free->bests[findex]) != NULLDATAOFF);
1477                         ASSERT(be16_to_cpu(free->bests[findex]) >= length);
1478                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1479                 }
1480                 /*
1481                  * The data block looked at didn't have enough room.
1482                  * We'll start at the beginning of the freespace entries.
1483                  */
1484                 else {
1485                         dbno = -1;
1486                         findex = 0;
1487                 }
1488         }
1489         /*
1490          * Didn't come in with a freespace block, so don't have a data block.
1491          */
1492         else {
1493                 ifbno = dbno = -1;
1494                 fbp = NULL;
1495                 findex = 0;
1496         }
1497         /*
1498          * If we don't have a data block yet, we're going to scan the
1499          * freespace blocks looking for one.  Figure out what the
1500          * highest freespace block number is.
1501          */
1502         if (dbno == -1) {
1503                 xfs_fileoff_t   fo;             /* freespace block number */
1504
1505                 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1506                         return error;
1507                 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1508                 fbno = ifbno;
1509         }
1510         /*
1511          * While we haven't identified a data block, search the freeblock
1512          * data for a good data block.  If we find a null freeblock entry,
1513          * indicating a hole in the data blocks, remember that.
1514          */
1515         while (dbno == -1) {
1516                 /*
1517                  * If we don't have a freeblock in hand, get the next one.
1518                  */
1519                 if (fbp == NULL) {
1520                         /*
1521                          * Happens the first time through unless lookup gave
1522                          * us a freespace block to start with.
1523                          */
1524                         if (++fbno == 0)
1525                                 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1526                         /*
1527                          * If it's ifbno we already looked at it.
1528                          */
1529                         if (fbno == ifbno)
1530                                 fbno++;
1531                         /*
1532                          * If it's off the end we're done.
1533                          */
1534                         if (fbno >= lastfbno)
1535                                 break;
1536                         /*
1537                          * Read the block.  There can be holes in the
1538                          * freespace blocks, so this might not succeed.
1539                          * This should be really rare, so there's no reason
1540                          * to avoid it.
1541                          */
1542                         error = xfs_dir2_free_try_read(tp, dp,
1543                                                 xfs_dir2_db_to_da(mp, fbno),
1544                                                 &fbp);
1545                         if (error)
1546                                 return error;
1547                         if (!fbp)
1548                                 continue;
1549                         free = fbp->b_addr;
1550                         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1551                         findex = 0;
1552                 }
1553                 /*
1554                  * Look at the current free entry.  Is it good enough?
1555                  */
1556                 if (be16_to_cpu(free->bests[findex]) != NULLDATAOFF &&
1557                     be16_to_cpu(free->bests[findex]) >= length)
1558                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1559                 else {
1560                         /*
1561                          * Are we done with the freeblock?
1562                          */
1563                         if (++findex == be32_to_cpu(free->hdr.nvalid)) {
1564                                 /*
1565                                  * Drop the block.
1566                                  */
1567                                 xfs_trans_brelse(tp, fbp);
1568                                 fbp = NULL;
1569                                 if (fblk && fblk->bp)
1570                                         fblk->bp = NULL;
1571                         }
1572                 }
1573         }
1574         /*
1575          * If we don't have a data block, we need to allocate one and make
1576          * the freespace entries refer to it.
1577          */
1578         if (unlikely(dbno == -1)) {
1579                 /*
1580                  * Not allowed to allocate, return failure.
1581                  */
1582                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1583                         return XFS_ERROR(ENOSPC);
1584
1585                 /*
1586                  * Allocate and initialize the new data block.
1587                  */
1588                 if (unlikely((error = xfs_dir2_grow_inode(args,
1589                                                          XFS_DIR2_DATA_SPACE,
1590                                                          &dbno)) ||
1591                     (error = xfs_dir2_data_init(args, dbno, &dbp))))
1592                         return error;
1593
1594                 /*
1595                  * If (somehow) we have a freespace block, get rid of it.
1596                  */
1597                 if (fbp)
1598                         xfs_trans_brelse(tp, fbp);
1599                 if (fblk && fblk->bp)
1600                         fblk->bp = NULL;
1601
1602                 /*
1603                  * Get the freespace block corresponding to the data block
1604                  * that was just allocated.
1605                  */
1606                 fbno = xfs_dir2_db_to_fdb(mp, dbno);
1607                 error = xfs_dir2_free_try_read(tp, dp,
1608                                                xfs_dir2_db_to_da(mp, fbno),
1609                                                &fbp);
1610                 if (error)
1611                         return error;
1612
1613                 /*
1614                  * If there wasn't a freespace block, the read will
1615                  * return a NULL fbp.  Allocate and initialize a new one.
1616                  */
1617                 if( fbp == NULL ) {
1618                         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1619                                                         &fbno))) {
1620                                 return error;
1621                         }
1622
1623                         if (unlikely(xfs_dir2_db_to_fdb(mp, dbno) != fbno)) {
1624                                 xfs_alert(mp,
1625                         "%s: dir ino %llu needed freesp block %lld for\n"
1626                         "  data block %lld, got %lld ifbno %llu lastfbno %d",
1627                                         __func__, (unsigned long long)dp->i_ino,
1628                                         (long long)xfs_dir2_db_to_fdb(mp, dbno),
1629                                         (long long)dbno, (long long)fbno,
1630                                         (unsigned long long)ifbno, lastfbno);
1631                                 if (fblk) {
1632                                         xfs_alert(mp,
1633                                 " fblk 0x%p blkno %llu index %d magic 0x%x",
1634                                                 fblk,
1635                                                 (unsigned long long)fblk->blkno,
1636                                                 fblk->index,
1637                                                 fblk->magic);
1638                                 } else {
1639                                         xfs_alert(mp, " ... fblk is NULL");
1640                                 }
1641                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1642                                                  XFS_ERRLEVEL_LOW, mp);
1643                                 return XFS_ERROR(EFSCORRUPTED);
1644                         }
1645
1646                         /*
1647                          * Get a buffer for the new block.
1648                          */
1649                         error = xfs_da_get_buf(tp, dp,
1650                                                xfs_dir2_db_to_da(mp, fbno),
1651                                                -1, &fbp, XFS_DATA_FORK);
1652                         if (error)
1653                                 return error;
1654                         fbp->b_ops = &xfs_dir2_free_buf_ops;
1655
1656                         /*
1657                          * Initialize the new block to be empty, and remember
1658                          * its first slot as our empty slot.
1659                          */
1660                         free = fbp->b_addr;
1661                         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
1662                         free->hdr.firstdb = cpu_to_be32(
1663                                 (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1664                                 xfs_dir2_free_max_bests(mp));
1665                         free->hdr.nvalid = 0;
1666                         free->hdr.nused = 0;
1667                 } else {
1668                         free = fbp->b_addr;
1669                         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1670                 }
1671
1672                 /*
1673                  * Set the freespace block index from the data block number.
1674                  */
1675                 findex = xfs_dir2_db_to_fdindex(mp, dbno);
1676                 /*
1677                  * If it's after the end of the current entries in the
1678                  * freespace block, extend that table.
1679                  */
1680                 if (findex >= be32_to_cpu(free->hdr.nvalid)) {
1681                         ASSERT(findex < xfs_dir2_free_max_bests(mp));
1682                         free->hdr.nvalid = cpu_to_be32(findex + 1);
1683                         /*
1684                          * Tag new entry so nused will go up.
1685                          */
1686                         free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1687                 }
1688                 /*
1689                  * If this entry was for an empty data block
1690                  * (this should always be true) then update the header.
1691                  */
1692                 if (free->bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1693                         be32_add_cpu(&free->hdr.nused, 1);
1694                         xfs_dir2_free_log_header(tp, fbp);
1695                 }
1696                 /*
1697                  * Update the real value in the table.
1698                  * We haven't allocated the data entry yet so this will
1699                  * change again.
1700                  */
1701                 hdr = dbp->b_addr;
1702                 free->bests[findex] = hdr->bestfree[0].length;
1703                 logfree = 1;
1704         }
1705         /*
1706          * We had a data block so we don't have to make a new one.
1707          */
1708         else {
1709                 /*
1710                  * If just checking, we succeeded.
1711                  */
1712                 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1713                         return 0;
1714
1715                 /*
1716                  * Read the data block in.
1717                  */
1718                 error = xfs_dir2_data_read(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1719                                            -1, &dbp);
1720                 if (error)
1721                         return error;
1722                 hdr = dbp->b_addr;
1723                 logfree = 0;
1724         }
1725         ASSERT(be16_to_cpu(hdr->bestfree[0].length) >= length);
1726         /*
1727          * Point to the existing unused space.
1728          */
1729         dup = (xfs_dir2_data_unused_t *)
1730               ((char *)hdr + be16_to_cpu(hdr->bestfree[0].offset));
1731         needscan = needlog = 0;
1732         /*
1733          * Mark the first part of the unused space, inuse for us.
1734          */
1735         xfs_dir2_data_use_free(tp, dbp, dup,
1736                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
1737                 &needlog, &needscan);
1738         /*
1739          * Fill in the new entry and log it.
1740          */
1741         dep = (xfs_dir2_data_entry_t *)dup;
1742         dep->inumber = cpu_to_be64(args->inumber);
1743         dep->namelen = args->namelen;
1744         memcpy(dep->name, args->name, dep->namelen);
1745         tagp = xfs_dir2_data_entry_tag_p(dep);
1746         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1747         xfs_dir2_data_log_entry(tp, dbp, dep);
1748         /*
1749          * Rescan the block for bestfree if needed.
1750          */
1751         if (needscan)
1752                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1753         /*
1754          * Log the data block header if needed.
1755          */
1756         if (needlog)
1757                 xfs_dir2_data_log_header(tp, dbp);
1758         /*
1759          * If the freespace entry is now wrong, update it.
1760          */
1761         if (be16_to_cpu(free->bests[findex]) != be16_to_cpu(hdr->bestfree[0].length)) {
1762                 free->bests[findex] = hdr->bestfree[0].length;
1763                 logfree = 1;
1764         }
1765         /*
1766          * Log the freespace entry if needed.
1767          */
1768         if (logfree)
1769                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1770         /*
1771          * Return the data block and offset in args, then drop the data block.
1772          */
1773         args->blkno = (xfs_dablk_t)dbno;
1774         args->index = be16_to_cpu(*tagp);
1775         return 0;
1776 }
1777
1778 /*
1779  * Lookup an entry in a node-format directory.
1780  * All the real work happens in xfs_da_node_lookup_int.
1781  * The only real output is the inode number of the entry.
1782  */
1783 int                                             /* error */
1784 xfs_dir2_node_lookup(
1785         xfs_da_args_t   *args)                  /* operation arguments */
1786 {
1787         int             error;                  /* error return value */
1788         int             i;                      /* btree level */
1789         int             rval;                   /* operation return value */
1790         xfs_da_state_t  *state;                 /* btree cursor */
1791
1792         trace_xfs_dir2_node_lookup(args);
1793
1794         /*
1795          * Allocate and initialize the btree cursor.
1796          */
1797         state = xfs_da_state_alloc();
1798         state->args = args;
1799         state->mp = args->dp->i_mount;
1800         state->blocksize = state->mp->m_dirblksize;
1801         state->node_ents = state->mp->m_dir_node_ents;
1802         /*
1803          * Fill in the path to the entry in the cursor.
1804          */
1805         error = xfs_da_node_lookup_int(state, &rval);
1806         if (error)
1807                 rval = error;
1808         else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
1809                 /* If a CI match, dup the actual name and return EEXIST */
1810                 xfs_dir2_data_entry_t   *dep;
1811
1812                 dep = (xfs_dir2_data_entry_t *)
1813                         ((char *)state->extrablk.bp->b_addr +
1814                                                  state->extrablk.index);
1815                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1816         }
1817         /*
1818          * Release the btree blocks and leaf block.
1819          */
1820         for (i = 0; i < state->path.active; i++) {
1821                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
1822                 state->path.blk[i].bp = NULL;
1823         }
1824         /*
1825          * Release the data block if we have it.
1826          */
1827         if (state->extravalid && state->extrablk.bp) {
1828                 xfs_trans_brelse(args->trans, state->extrablk.bp);
1829                 state->extrablk.bp = NULL;
1830         }
1831         xfs_da_state_free(state);
1832         return rval;
1833 }
1834
1835 /*
1836  * Remove an entry from a node-format directory.
1837  */
1838 int                                             /* error */
1839 xfs_dir2_node_removename(
1840         xfs_da_args_t           *args)          /* operation arguments */
1841 {
1842         xfs_da_state_blk_t      *blk;           /* leaf block */
1843         int                     error;          /* error return value */
1844         int                     rval;           /* operation return value */
1845         xfs_da_state_t          *state;         /* btree cursor */
1846
1847         trace_xfs_dir2_node_removename(args);
1848
1849         /*
1850          * Allocate and initialize the btree cursor.
1851          */
1852         state = xfs_da_state_alloc();
1853         state->args = args;
1854         state->mp = args->dp->i_mount;
1855         state->blocksize = state->mp->m_dirblksize;
1856         state->node_ents = state->mp->m_dir_node_ents;
1857         /*
1858          * Look up the entry we're deleting, set up the cursor.
1859          */
1860         error = xfs_da_node_lookup_int(state, &rval);
1861         if (error)
1862                 rval = error;
1863         /*
1864          * Didn't find it, upper layer screwed up.
1865          */
1866         if (rval != EEXIST) {
1867                 xfs_da_state_free(state);
1868                 return rval;
1869         }
1870         blk = &state->path.blk[state->path.active - 1];
1871         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1872         ASSERT(state->extravalid);
1873         /*
1874          * Remove the leaf and data entries.
1875          * Extrablk refers to the data block.
1876          */
1877         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
1878                 &state->extrablk, &rval);
1879         if (error)
1880                 return error;
1881         /*
1882          * Fix the hash values up the btree.
1883          */
1884         xfs_da_fixhashpath(state, &state->path);
1885         /*
1886          * If we need to join leaf blocks, do it.
1887          */
1888         if (rval && state->path.active > 1)
1889                 error = xfs_da_join(state);
1890         /*
1891          * If no errors so far, try conversion to leaf format.
1892          */
1893         if (!error)
1894                 error = xfs_dir2_node_to_leaf(state);
1895         xfs_da_state_free(state);
1896         return error;
1897 }
1898
1899 /*
1900  * Replace an entry's inode number in a node-format directory.
1901  */
1902 int                                             /* error */
1903 xfs_dir2_node_replace(
1904         xfs_da_args_t           *args)          /* operation arguments */
1905 {
1906         xfs_da_state_blk_t      *blk;           /* leaf block */
1907         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1908         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
1909         int                     error;          /* error return value */
1910         int                     i;              /* btree level */
1911         xfs_ino_t               inum;           /* new inode number */
1912         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1913         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
1914         int                     rval;           /* internal return value */
1915         xfs_da_state_t          *state;         /* btree cursor */
1916
1917         trace_xfs_dir2_node_replace(args);
1918
1919         /*
1920          * Allocate and initialize the btree cursor.
1921          */
1922         state = xfs_da_state_alloc();
1923         state->args = args;
1924         state->mp = args->dp->i_mount;
1925         state->blocksize = state->mp->m_dirblksize;
1926         state->node_ents = state->mp->m_dir_node_ents;
1927         inum = args->inumber;
1928         /*
1929          * Lookup the entry to change in the btree.
1930          */
1931         error = xfs_da_node_lookup_int(state, &rval);
1932         if (error) {
1933                 rval = error;
1934         }
1935         /*
1936          * It should be found, since the vnodeops layer has looked it up
1937          * and locked it.  But paranoia is good.
1938          */
1939         if (rval == EEXIST) {
1940                 /*
1941                  * Find the leaf entry.
1942                  */
1943                 blk = &state->path.blk[state->path.active - 1];
1944                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1945                 leaf = blk->bp->b_addr;
1946                 lep = &leaf->ents[blk->index];
1947                 ASSERT(state->extravalid);
1948                 /*
1949                  * Point to the data entry.
1950                  */
1951                 hdr = state->extrablk.bp->b_addr;
1952                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
1953                 dep = (xfs_dir2_data_entry_t *)
1954                       ((char *)hdr +
1955                        xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
1956                 ASSERT(inum != be64_to_cpu(dep->inumber));
1957                 /*
1958                  * Fill in the new inode number and log the entry.
1959                  */
1960                 dep->inumber = cpu_to_be64(inum);
1961                 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
1962                 rval = 0;
1963         }
1964         /*
1965          * Didn't find it, and we're holding a data block.  Drop it.
1966          */
1967         else if (state->extravalid) {
1968                 xfs_trans_brelse(args->trans, state->extrablk.bp);
1969                 state->extrablk.bp = NULL;
1970         }
1971         /*
1972          * Release all the buffers in the cursor.
1973          */
1974         for (i = 0; i < state->path.active; i++) {
1975                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
1976                 state->path.blk[i].bp = NULL;
1977         }
1978         xfs_da_state_free(state);
1979         return rval;
1980 }
1981
1982 /*
1983  * Trim off a trailing empty freespace block.
1984  * Return (in rvalp) 1 if we did it, 0 if not.
1985  */
1986 int                                             /* error */
1987 xfs_dir2_node_trim_free(
1988         xfs_da_args_t           *args,          /* operation arguments */
1989         xfs_fileoff_t           fo,             /* free block number */
1990         int                     *rvalp)         /* out: did something */
1991 {
1992         struct xfs_buf          *bp;            /* freespace buffer */
1993         xfs_inode_t             *dp;            /* incore directory inode */
1994         int                     error;          /* error return code */
1995         xfs_dir2_free_t         *free;          /* freespace structure */
1996         xfs_mount_t             *mp;            /* filesystem mount point */
1997         xfs_trans_t             *tp;            /* transaction pointer */
1998
1999         dp = args->dp;
2000         mp = dp->i_mount;
2001         tp = args->trans;
2002         /*
2003          * Read the freespace block.
2004          */
2005         error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2006         if (error)
2007                 return error;
2008         /*
2009          * There can be holes in freespace.  If fo is a hole, there's
2010          * nothing to do.
2011          */
2012         if (!bp)
2013                 return 0;
2014         free = bp->b_addr;
2015         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
2016         /*
2017          * If there are used entries, there's nothing to do.
2018          */
2019         if (be32_to_cpu(free->hdr.nused) > 0) {
2020                 xfs_trans_brelse(tp, bp);
2021                 *rvalp = 0;
2022                 return 0;
2023         }
2024         /*
2025          * Blow the block away.
2026          */
2027         if ((error =
2028             xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
2029                     bp))) {
2030                 /*
2031                  * Can't fail with ENOSPC since that only happens with no
2032                  * space reservation, when breaking up an extent into two
2033                  * pieces.  This is the last block of an extent.
2034                  */
2035                 ASSERT(error != ENOSPC);
2036                 xfs_trans_brelse(tp, bp);
2037                 return error;
2038         }
2039         /*
2040          * Return that we succeeded.
2041          */
2042         *rvalp = 1;
2043         return 0;
2044 }