fs/xfs/xfs_dir2_node.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_log.h"
  22 #include "xfs_inum.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_dir2.h"
  27 #include "xfs_dmapi.h"
  28 #include "xfs_mount.h"
  29 #include "xfs_da_btree.h"
  30 #include "xfs_bmap_btree.h"
  31 #include "xfs_dir2_sf.h"
  32 #include "xfs_attr_sf.h"
  33 #include "xfs_dinode.h"
  34 #include "xfs_inode.h"
  35 #include "xfs_bmap.h"
  36 #include "xfs_dir2_data.h"
  37 #include "xfs_dir2_leaf.h"
  38 #include "xfs_dir2_block.h"
  39 #include "xfs_dir2_node.h"
  40 #include "xfs_error.h"
  41 #include "xfs_trace.h"
  42
  43 /*
  44  * Function declarations.
  45  */
  46 static void xfs_dir2_free_log_header(xfs_trans_t *tp, xfs_dabuf_t *bp);
  47 static int xfs_dir2_leafn_add(xfs_dabuf_t *bp, xfs_da_args_t *args, int index);
  48 #ifdef DEBUG
  49 static void xfs_dir2_leafn_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
  50 #else
  51 #define xfs_dir2_leafn_check(dp, bp)
  52 #endif
  53 static void xfs_dir2_leafn_moveents(xfs_da_args_t *args, xfs_dabuf_t *bp_s,
  54                                     int start_s, xfs_dabuf_t *bp_d, int start_d,
  55                                     int count);
  56 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  57                                      xfs_da_state_blk_t *blk1,
  58                                      xfs_da_state_blk_t *blk2);
  59 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, xfs_dabuf_t *bp,
  60                                  int index, xfs_da_state_blk_t *dblk,
  61                                  int *rval);
  62 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  63                                      xfs_da_state_blk_t *fblk);
  64
  65 /*
  66  * Log entries from a freespace block.
  67  */
  68 STATIC void
  69 xfs_dir2_free_log_bests(
  70         xfs_trans_t             *tp,            /* transaction pointer */
  71         xfs_dabuf_t             *bp,            /* freespace buffer */
  72         int                     first,          /* first entry to log */
  73         int                     last)           /* last entry to log */
  74 {
  75         xfs_dir2_free_t         *free;          /* freespace structure */
  76
  77         free = bp->data;
  78         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
  79         xfs_da_log_buf(tp, bp,
  80                 (uint)((char *)&free->bests[first] - (char *)free),
  81                 (uint)((char *)&free->bests[last] - (char *)free +
  82                        sizeof(free->bests[0]) - 1));
  83 }
  84
  85 /*
  86  * Log header from a freespace block.
  87  */
  88 static void
  89 xfs_dir2_free_log_header(
  90         xfs_trans_t             *tp,            /* transaction pointer */
  91         xfs_dabuf_t             *bp)            /* freespace buffer */
  92 {
  93         xfs_dir2_free_t         *free;          /* freespace structure */
  94
  95         free = bp->data;
  96         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
  97         xfs_da_log_buf(tp, bp, (uint)((char *)&free->hdr - (char *)free),
  98                 (uint)(sizeof(xfs_dir2_free_hdr_t) - 1));
  99 }
 100
 101 /*
 102  * Convert a leaf-format directory to a node-format directory.
 103  * We need to change the magic number of the leaf block, and copy
 104  * the freespace table out of the leaf block into its own block.
 105  */
 106 int                                             /* error */
 107 xfs_dir2_leaf_to_node(
 108         xfs_da_args_t           *args,          /* operation arguments */
 109         xfs_dabuf_t             *lbp)           /* leaf buffer */
 110 {
 111         xfs_inode_t             *dp;            /* incore directory inode */
 112         int                     error;          /* error return value */
 113         xfs_dabuf_t             *fbp;           /* freespace buffer */
 114         xfs_dir2_db_t           fdb;            /* freespace block number */
 115         xfs_dir2_free_t         *free;          /* freespace structure */
 116         __be16                  *from;          /* pointer to freespace entry */
 117         int                     i;              /* leaf freespace index */
 118         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 119         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 120         xfs_mount_t             *mp;            /* filesystem mount point */
 121         int                     n;              /* count of live freespc ents */
 122         xfs_dir2_data_off_t     off;            /* freespace entry value */
 123         __be16                  *to;            /* pointer to freespace entry */
 124         xfs_trans_t             *tp;            /* transaction pointer */
 125
 126         trace_xfs_dir2_leaf_to_node(args);
 127
 128         dp = args->dp;
 129         mp = dp->i_mount;
 130         tp = args->trans;
 131         /*
 132          * Add a freespace block to the directory.
 133          */
 134         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 135                 return error;
 136         }
 137         ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
 138         /*
 139          * Get the buffer for the new freespace block.
 140          */
 141         if ((error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb), -1, &fbp,
 142                         XFS_DATA_FORK))) {
 143                 return error;
 144         }
 145         ASSERT(fbp != NULL);
 146         free = fbp->data;
 147         leaf = lbp->data;
 148         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 149         /*
 150          * Initialize the freespace block header.
 151          */
 152         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
 153         free->hdr.firstdb = 0;
 154         ASSERT(be32_to_cpu(ltp->bestcount) <= (uint)dp->i_d.di_size / mp->m_dirblksize);
 155         free->hdr.nvalid = ltp->bestcount;
 156         /*
 157          * Copy freespace entries from the leaf block to the new block.
 158          * Count active entries.
 159          */
 160         for (i = n = 0, from = xfs_dir2_leaf_bests_p(ltp), to = free->bests;
 161              i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 162                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 163                         n++;
 164                 *to = cpu_to_be16(off);
 165         }
 166         free->hdr.nused = cpu_to_be32(n);
 167         leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 168         /*
 169          * Log everything.
 170          */
 171         xfs_dir2_leaf_log_header(tp, lbp);
 172         xfs_dir2_free_log_header(tp, fbp);
 173         xfs_dir2_free_log_bests(tp, fbp, 0, be32_to_cpu(free->hdr.nvalid) - 1);
 174         xfs_da_buf_done(fbp);
 175         xfs_dir2_leafn_check(dp, lbp);
 176         return 0;
 177 }
 178
 179 /*
 180  * Add a leaf entry to a leaf block in a node-form directory.
 181  * The other work necessary is done from the caller.
 182  */
 183 static int                                      /* error */
 184 xfs_dir2_leafn_add(
 185         xfs_dabuf_t             *bp,            /* leaf buffer */
 186         xfs_da_args_t           *args,          /* operation arguments */
 187         int                     index)          /* insertion pt for new entry */
 188 {
 189         int                     compact;        /* compacting stale leaves */
 190         xfs_inode_t             *dp;            /* incore directory inode */
 191         int                     highstale;      /* next stale entry */
 192         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 193         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 194         int                     lfloghigh;      /* high leaf entry logging */
 195         int                     lfloglow;       /* low leaf entry logging */
 196         int                     lowstale;       /* previous stale entry */
 197         xfs_mount_t             *mp;            /* filesystem mount point */
 198         xfs_trans_t             *tp;            /* transaction pointer */
 199
 200         trace_xfs_dir2_leafn_add(args, index);
 201
 202         dp = args->dp;
 203         mp = dp->i_mount;
 204         tp = args->trans;
 205         leaf = bp->data;
 206
 207         /*
 208          * Quick check just to make sure we are not going to index
 209          * into other peoples memory
 210          */
 211         if (index < 0)
 212                 return XFS_ERROR(EFSCORRUPTED);
 213
 214         /*
 215          * If there are already the maximum number of leaf entries in
 216          * the block, if there are no stale entries it won't fit.
 217          * Caller will do a split.  If there are stale entries we'll do
 218          * a compact.
 219          */
 220
 221         if (be16_to_cpu(leaf->hdr.count) == xfs_dir2_max_leaf_ents(mp)) {
 222                 if (!leaf->hdr.stale)
 223                         return XFS_ERROR(ENOSPC);
 224                 compact = be16_to_cpu(leaf->hdr.stale) > 1;
 225         } else
 226                 compact = 0;
 227         ASSERT(index == 0 || be32_to_cpu(leaf->ents[index - 1].hashval) <= args->hashval);
 228         ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
 229                be32_to_cpu(leaf->ents[index].hashval) >= args->hashval);
 230
 231         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 232                 return 0;
 233
 234         /*
 235          * Compact out all but one stale leaf entry.  Leaves behind
 236          * the entry closest to index.
 237          */
 238         if (compact) {
 239                 xfs_dir2_leaf_compact_x1(bp, &index, &lowstale, &highstale,
 240                         &lfloglow, &lfloghigh);
 241         }
 242         /*
 243          * Set impossible logging indices for this case.
 244          */
 245         else if (leaf->hdr.stale) {
 246                 lfloglow = be16_to_cpu(leaf->hdr.count);
 247                 lfloghigh = -1;
 248         }
 249         /*
 250          * No stale entries, just insert a space for the new entry.
 251          */
 252         if (!leaf->hdr.stale) {
 253                 lep = &leaf->ents[index];
 254                 if (index < be16_to_cpu(leaf->hdr.count))
 255                         memmove(lep + 1, lep,
 256                                 (be16_to_cpu(leaf->hdr.count) - index) * sizeof(*lep));
 257                 lfloglow = index;
 258                 lfloghigh = be16_to_cpu(leaf->hdr.count);
 259                 be16_add_cpu(&leaf->hdr.count, 1);
 260         }
 261         /*
 262          * There are stale entries.  We'll use one for the new entry.
 263          */
 264         else {
 265                 /*
 266                  * If we didn't do a compact then we need to figure out
 267                  * which stale entry will be used.
 268                  */
 269                 if (compact == 0) {
 270                         /*
 271                          * Find first stale entry before our insertion point.
 272                          */
 273                         for (lowstale = index - 1;
 274                              lowstale >= 0 &&
 275                                 be32_to_cpu(leaf->ents[lowstale].address) !=
 276                                 XFS_DIR2_NULL_DATAPTR;
 277                              lowstale--)
 278                                 continue;
 279                         /*
 280                          * Find next stale entry after insertion point.
 281                          * Stop looking if the answer would be worse than
 282                          * lowstale already found.
 283                          */
 284                         for (highstale = index;
 285                              highstale < be16_to_cpu(leaf->hdr.count) &&
 286                                 be32_to_cpu(leaf->ents[highstale].address) !=
 287                                 XFS_DIR2_NULL_DATAPTR &&
 288                                 (lowstale < 0 ||
 289                                  index - lowstale - 1 >= highstale - index);
 290                              highstale++)
 291                                 continue;
 292                 }
 293                 /*
 294                  * Using the low stale entry.
 295                  * Shift entries up toward the stale slot.
 296                  */
 297                 if (lowstale >= 0 &&
 298                     (highstale == be16_to_cpu(leaf->hdr.count) ||
 299                      index - lowstale - 1 < highstale - index)) {
 300                         ASSERT(be32_to_cpu(leaf->ents[lowstale].address) ==
 301                                XFS_DIR2_NULL_DATAPTR);
 302                         ASSERT(index - lowstale - 1 >= 0);
 303                         if (index - lowstale - 1 > 0)
 304                                 memmove(&leaf->ents[lowstale],
 305                                         &leaf->ents[lowstale + 1],
 306                                         (index - lowstale - 1) * sizeof(*lep));
 307                         lep = &leaf->ents[index - 1];
 308                         lfloglow = MIN(lowstale, lfloglow);
 309                         lfloghigh = MAX(index - 1, lfloghigh);
 310                 }
 311                 /*
 312                  * Using the high stale entry.
 313                  * Shift entries down toward the stale slot.
 314                  */
 315                 else {
 316                         ASSERT(be32_to_cpu(leaf->ents[highstale].address) ==
 317                                XFS_DIR2_NULL_DATAPTR);
 318                         ASSERT(highstale - index >= 0);
 319                         if (highstale - index > 0)
 320                                 memmove(&leaf->ents[index + 1],
 321                                         &leaf->ents[index],
 322                                         (highstale - index) * sizeof(*lep));
 323                         lep = &leaf->ents[index];
 324                         lfloglow = MIN(index, lfloglow);
 325                         lfloghigh = MAX(highstale, lfloghigh);
 326                 }
 327                 be16_add_cpu(&leaf->hdr.stale, -1);
 328         }
 329         /*
 330          * Insert the new entry, log everything.
 331          */
 332         lep->hashval = cpu_to_be32(args->hashval);
 333         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
 334                                 args->blkno, args->index));
 335         xfs_dir2_leaf_log_header(tp, bp);
 336         xfs_dir2_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
 337         xfs_dir2_leafn_check(dp, bp);
 338         return 0;
 339 }
 340
 341 #ifdef DEBUG
 342 /*
 343  * Check internal consistency of a leafn block.
 344  */
 345 void
 346 xfs_dir2_leafn_check(
 347         xfs_inode_t     *dp,                    /* incore directory inode */
 348         xfs_dabuf_t     *bp)                    /* leaf buffer */
 349 {
 350         int             i;                      /* leaf index */
 351         xfs_dir2_leaf_t *leaf;                  /* leaf structure */
 352         xfs_mount_t     *mp;                    /* filesystem mount point */
 353         int             stale;                  /* count of stale leaves */
 354
 355         leaf = bp->data;
 356         mp = dp->i_mount;
 357         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 358         ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
 359         for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
 360                 if (i + 1 < be16_to_cpu(leaf->hdr.count)) {
 361                         ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
 362                                be32_to_cpu(leaf->ents[i + 1].hashval));
 363                 }
 364                 if (be32_to_cpu(leaf->ents[i].address) == XFS_DIR2_NULL_DATAPTR)
 365                         stale++;
 366         }
 367         ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
 368 }
 369 #endif  /* DEBUG */
 370
 371 /*
 372  * Return the last hash value in the leaf.
 373  * Stale entries are ok.
 374  */
 375 xfs_dahash_t                                    /* hash value */
 376 xfs_dir2_leafn_lasthash(
 377         xfs_dabuf_t     *bp,                    /* leaf buffer */
 378         int             *count)                 /* count of entries in leaf */
 379 {
 380         xfs_dir2_leaf_t *leaf;                  /* leaf structure */
 381
 382         leaf = bp->data;
 383         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 384         if (count)
 385                 *count = be16_to_cpu(leaf->hdr.count);
 386         if (!leaf->hdr.count)
 387                 return 0;
 388         return be32_to_cpu(leaf->ents[be16_to_cpu(leaf->hdr.count) - 1].hashval);
 389 }
 390
 391 /*
 392  * Look up a leaf entry for space to add a name in a node-format leaf block.
 393  * The extrablk in state is a freespace block.
 394  */
 395 STATIC int
 396 xfs_dir2_leafn_lookup_for_addname(
 397         xfs_dabuf_t             *bp,            /* leaf buffer */
 398         xfs_da_args_t           *args,          /* operation arguments */
 399         int                     *indexp,        /* out: leaf entry index */
 400         xfs_da_state_t          *state)         /* state to fill in */
 401 {
 402         xfs_dabuf_t             *curbp = NULL;  /* current data/free buffer */
 403         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 404         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 405         xfs_inode_t             *dp;            /* incore directory inode */
 406         int                     error;          /* error return value */
 407         int                     fi;             /* free entry index */
 408         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 409         int                     index;          /* leaf entry index */
 410         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 411         int                     length;         /* length of new data entry */
 412         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 413         xfs_mount_t             *mp;            /* filesystem mount point */
 414         xfs_dir2_db_t           newdb;          /* new data block number */
 415         xfs_dir2_db_t           newfdb;         /* new free block number */
 416         xfs_trans_t             *tp;            /* transaction pointer */
 417
 418         dp = args->dp;
 419         tp = args->trans;
 420         mp = dp->i_mount;
 421         leaf = bp->data;
 422         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 423 #ifdef __KERNEL__
 424         ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
 425 #endif
 426         xfs_dir2_leafn_check(dp, bp);
 427         /*
 428          * Look up the hash value in the leaf entries.
 429          */
 430         index = xfs_dir2_leaf_search_hash(args, bp);
 431         /*
 432          * Do we have a buffer coming in?
 433          */
 434         if (state->extravalid) {
 435                 /* If so, it's a free block buffer, get the block number. */
 436                 curbp = state->extrablk.bp;
 437                 curfdb = state->extrablk.blkno;
 438                 free = curbp->data;
 439                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
 440         }
 441         length = xfs_dir2_data_entsize(args->namelen);
 442         /*
 443          * Loop over leaf entries with the right hash value.
 444          */
 445         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
 446                                 be32_to_cpu(lep->hashval) == args->hashval;
 447                                 lep++, index++) {
 448                 /*
 449                  * Skip stale leaf entries.
 450                  */
 451                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 452                         continue;
 453                 /*
 454                  * Pull the data block number from the entry.
 455                  */
 456                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 457                 /*
 458                  * For addname, we're looking for a place to put the new entry.
 459                  * We want to use a data block with an entry of equal
 460                  * hash value to ours if there is one with room.
 461                  *
 462                  * If this block isn't the data block we already have
 463                  * in hand, take a look at it.
 464                  */
 465                 if (newdb != curdb) {
 466                         curdb = newdb;
 467                         /*
 468                          * Convert the data block to the free block
 469                          * holding its freespace information.
 470                          */
 471                         newfdb = xfs_dir2_db_to_fdb(mp, newdb);
 472                         /*
 473                          * If it's not the one we have in hand, read it in.
 474                          */
 475                         if (newfdb != curfdb) {
 476                                 /*
 477                                  * If we had one before, drop it.
 478                                  */
 479                                 if (curbp)
 480                                         xfs_da_brelse(tp, curbp);
 481                                 /*
 482                                  * Read the free block.
 483                                  */
 484                                 error = xfs_da_read_buf(tp, dp,
 485                                                 xfs_dir2_db_to_da(mp, newfdb),
 486                                                 -1, &curbp, XFS_DATA_FORK);
 487                                 if (error)
 488                                         return error;
 489                                 free = curbp->data;
 490                                 ASSERT(be32_to_cpu(free->hdr.magic) ==
 491                                         XFS_DIR2_FREE_MAGIC);
 492                                 ASSERT((be32_to_cpu(free->hdr.firstdb) %
 493                                         XFS_DIR2_MAX_FREE_BESTS(mp)) == 0);
 494                                 ASSERT(be32_to_cpu(free->hdr.firstdb) <= curdb);
 495                                 ASSERT(curdb < be32_to_cpu(free->hdr.firstdb) +
 496                                         be32_to_cpu(free->hdr.nvalid));
 497                         }
 498                         /*
 499                          * Get the index for our entry.
 500                          */
 501                         fi = xfs_dir2_db_to_fdindex(mp, curdb);
 502                         /*
 503                          * If it has room, return it.
 504                          */
 505                         if (unlikely(be16_to_cpu(free->bests[fi]) == NULLDATAOFF)) {
 506                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 507                                                         XFS_ERRLEVEL_LOW, mp);
 508                                 if (curfdb != newfdb)
 509                                         xfs_da_brelse(tp, curbp);
 510                                 return XFS_ERROR(EFSCORRUPTED);
 511                         }
 512                         curfdb = newfdb;
 513                         if (be16_to_cpu(free->bests[fi]) >= length)
 514                                 goto out;
 515                 }
 516         }
 517         /* Didn't find any space */
 518         fi = -1;
 519 out:
 520         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 521         if (curbp) {
 522                 /* Giving back a free block. */
 523                 state->extravalid = 1;
 524                 state->extrablk.bp = curbp;
 525                 state->extrablk.index = fi;
 526                 state->extrablk.blkno = curfdb;
 527                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 528         } else {
 529                 state->extravalid = 0;
 530         }
 531         /*
 532          * Return the index, that will be the insertion point.
 533          */
 534         *indexp = index;
 535         return XFS_ERROR(ENOENT);
 536 }
 537
 538 /*
 539  * Look up a leaf entry in a node-format leaf block.
 540  * The extrablk in state a data block.
 541  */
 542 STATIC int
 543 xfs_dir2_leafn_lookup_for_entry(
 544         xfs_dabuf_t             *bp,            /* leaf buffer */
 545         xfs_da_args_t           *args,          /* operation arguments */
 546         int                     *indexp,        /* out: leaf entry index */
 547         xfs_da_state_t          *state)         /* state to fill in */
 548 {
 549         xfs_dabuf_t             *curbp = NULL;  /* current data/free buffer */
 550         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 551         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 552         xfs_inode_t             *dp;            /* incore directory inode */
 553         int                     error;          /* error return value */
 554         int                     index;          /* leaf entry index */
 555         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 556         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 557         xfs_mount_t             *mp;            /* filesystem mount point */
 558         xfs_dir2_db_t           newdb;          /* new data block number */
 559         xfs_trans_t             *tp;            /* transaction pointer */
 560         enum xfs_dacmp          cmp;            /* comparison result */
 561
 562         dp = args->dp;
 563         tp = args->trans;
 564         mp = dp->i_mount;
 565         leaf = bp->data;
 566         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 567 #ifdef __KERNEL__
 568         ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
 569 #endif
 570         xfs_dir2_leafn_check(dp, bp);
 571         /*
 572          * Look up the hash value in the leaf entries.
 573          */
 574         index = xfs_dir2_leaf_search_hash(args, bp);
 575         /*
 576          * Do we have a buffer coming in?
 577          */
 578         if (state->extravalid) {
 579                 curbp = state->extrablk.bp;
 580                 curdb = state->extrablk.blkno;
 581         }
 582         /*
 583          * Loop over leaf entries with the right hash value.
 584          */
 585         for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
 586                                 be32_to_cpu(lep->hashval) == args->hashval;
 587                                 lep++, index++) {
 588                 /*
 589                  * Skip stale leaf entries.
 590                  */
 591                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 592                         continue;
 593                 /*
 594                  * Pull the data block number from the entry.
 595                  */
 596                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 597                 /*
 598                  * Not adding a new entry, so we really want to find
 599                  * the name given to us.
 600                  *
 601                  * If it's a different data block, go get it.
 602                  */
 603                 if (newdb != curdb) {
 604                         /*
 605                          * If we had a block before that we aren't saving
 606                          * for a CI name, drop it
 607                          */
 608                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 609                                                 curdb != state->extrablk.blkno))
 610                                 xfs_da_brelse(tp, curbp);
 611                         /*
 612                          * If needing the block that is saved with a CI match,
 613                          * use it otherwise read in the new data block.
 614                          */
 615                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 616                                         newdb == state->extrablk.blkno) {
 617                                 ASSERT(state->extravalid);
 618                                 curbp = state->extrablk.bp;
 619                         } else {
 620                                 error = xfs_da_read_buf(tp, dp,
 621                                                 xfs_dir2_db_to_da(mp, newdb),
 622                                                 -1, &curbp, XFS_DATA_FORK);
 623                                 if (error)
 624                                         return error;
 625                         }
 626                         xfs_dir2_data_check(dp, curbp);
 627                         curdb = newdb;
 628                 }
 629                 /*
 630                  * Point to the data entry.
 631                  */
 632                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->data +
 633                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
 634                 /*
 635                  * Compare the entry and if it's an exact match, return
 636                  * EEXIST immediately. If it's the first case-insensitive
 637                  * match, store the block & inode number and continue looking.
 638                  */
 639                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 640                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 641                         /* If there is a CI match block, drop it */
 642                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 643                                                 curdb != state->extrablk.blkno)
 644                                 xfs_da_brelse(tp, state->extrablk.bp);
 645                         args->cmpresult = cmp;
 646                         args->inumber = be64_to_cpu(dep->inumber);
 647                         *indexp = index;
 648                         state->extravalid = 1;
 649                         state->extrablk.bp = curbp;
 650                         state->extrablk.blkno = curdb;
 651                         state->extrablk.index = (int)((char *)dep -
 652                                                         (char *)curbp->data);
 653                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 654                         if (cmp == XFS_CMP_EXACT)
 655                                 return XFS_ERROR(EEXIST);
 656                 }
 657         }
 658         ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
 659                                         (args->op_flags & XFS_DA_OP_OKNOENT));
 660         if (curbp) {
 661                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 662                         /* Giving back last used data block. */
 663                         state->extravalid = 1;
 664                         state->extrablk.bp = curbp;
 665                         state->extrablk.index = -1;
 666                         state->extrablk.blkno = curdb;
 667                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 668                 } else {
 669                         /* If the curbp is not the CI match block, drop it */
 670                         if (state->extrablk.bp != curbp)
 671                                 xfs_da_brelse(tp, curbp);
 672                 }
 673         } else {
 674                 state->extravalid = 0;
 675         }
 676         *indexp = index;
 677         return XFS_ERROR(ENOENT);
 678 }
 679
 680 /*
 681  * Look up a leaf entry in a node-format leaf block.
 682  * If this is an addname then the extrablk in state is a freespace block,
 683  * otherwise it's a data block.
 684  */
 685 int
 686 xfs_dir2_leafn_lookup_int(
 687         xfs_dabuf_t             *bp,            /* leaf buffer */
 688         xfs_da_args_t           *args,          /* operation arguments */
 689         int                     *indexp,        /* out: leaf entry index */
 690         xfs_da_state_t          *state)         /* state to fill in */
 691 {
 692         if (args->op_flags & XFS_DA_OP_ADDNAME)
 693                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 694                                                         state);
 695         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 696 }
 697
 698 /*
 699  * Move count leaf entries from source to destination leaf.
 700  * Log entries and headers.  Stale entries are preserved.
 701  */
 702 static void
 703 xfs_dir2_leafn_moveents(
 704         xfs_da_args_t   *args,                  /* operation arguments */
 705         xfs_dabuf_t     *bp_s,                  /* source leaf buffer */
 706         int             start_s,                /* source leaf index */
 707         xfs_dabuf_t     *bp_d,                  /* destination leaf buffer */
 708         int             start_d,                /* destination leaf index */
 709         int             count)                  /* count of leaves to copy */
 710 {
 711         xfs_dir2_leaf_t *leaf_d;                /* destination leaf structure */
 712         xfs_dir2_leaf_t *leaf_s;                /* source leaf structure */
 713         int             stale;                  /* count stale leaves copied */
 714         xfs_trans_t     *tp;                    /* transaction pointer */
 715
 716         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 717
 718         /*
 719          * Silently return if nothing to do.
 720          */
 721         if (count == 0) {
 722                 return;
 723         }
 724         tp = args->trans;
 725         leaf_s = bp_s->data;
 726         leaf_d = bp_d->data;
 727         /*
 728          * If the destination index is not the end of the current
 729          * destination leaf entries, open up a hole in the destination
 730          * to hold the new entries.
 731          */
 732         if (start_d < be16_to_cpu(leaf_d->hdr.count)) {
 733                 memmove(&leaf_d->ents[start_d + count], &leaf_d->ents[start_d],
 734                         (be16_to_cpu(leaf_d->hdr.count) - start_d) *
 735                         sizeof(xfs_dir2_leaf_entry_t));
 736                 xfs_dir2_leaf_log_ents(tp, bp_d, start_d + count,
 737                         count + be16_to_cpu(leaf_d->hdr.count) - 1);
 738         }
 739         /*
 740          * If the source has stale leaves, count the ones in the copy range
 741          * so we can update the header correctly.
 742          */
 743         if (leaf_s->hdr.stale) {
 744                 int     i;                      /* temp leaf index */
 745
 746                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 747                         if (be32_to_cpu(leaf_s->ents[i].address) == XFS_DIR2_NULL_DATAPTR)
 748                                 stale++;
 749                 }
 750         } else
 751                 stale = 0;
 752         /*
 753          * Copy the leaf entries from source to destination.
 754          */
 755         memcpy(&leaf_d->ents[start_d], &leaf_s->ents[start_s],
 756                 count * sizeof(xfs_dir2_leaf_entry_t));
 757         xfs_dir2_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
 758         /*
 759          * If there are source entries after the ones we copied,
 760          * delete the ones we copied by sliding the next ones down.
 761          */
 762         if (start_s + count < be16_to_cpu(leaf_s->hdr.count)) {
 763                 memmove(&leaf_s->ents[start_s], &leaf_s->ents[start_s + count],
 764                         count * sizeof(xfs_dir2_leaf_entry_t));
 765                 xfs_dir2_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
 766         }
 767         /*
 768          * Update the headers and log them.
 769          */
 770         be16_add_cpu(&leaf_s->hdr.count, -(count));
 771         be16_add_cpu(&leaf_s->hdr.stale, -(stale));
 772         be16_add_cpu(&leaf_d->hdr.count, count);
 773         be16_add_cpu(&leaf_d->hdr.stale, stale);
 774         xfs_dir2_leaf_log_header(tp, bp_s);
 775         xfs_dir2_leaf_log_header(tp, bp_d);
 776         xfs_dir2_leafn_check(args->dp, bp_s);
 777         xfs_dir2_leafn_check(args->dp, bp_d);
 778 }
 779
 780 /*
 781  * Determine the sort order of two leaf blocks.
 782  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 783  */
 784 int                                             /* sort order */
 785 xfs_dir2_leafn_order(
 786         xfs_dabuf_t     *leaf1_bp,              /* leaf1 buffer */
 787         xfs_dabuf_t     *leaf2_bp)              /* leaf2 buffer */
 788 {
 789         xfs_dir2_leaf_t *leaf1;                 /* leaf1 structure */
 790         xfs_dir2_leaf_t *leaf2;                 /* leaf2 structure */
 791
 792         leaf1 = leaf1_bp->data;
 793         leaf2 = leaf2_bp->data;
 794         ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 795         ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 796         if (be16_to_cpu(leaf1->hdr.count) > 0 &&
 797             be16_to_cpu(leaf2->hdr.count) > 0 &&
 798             (be32_to_cpu(leaf2->ents[0].hashval) < be32_to_cpu(leaf1->ents[0].hashval) ||
 799              be32_to_cpu(leaf2->ents[be16_to_cpu(leaf2->hdr.count) - 1].hashval) <
 800              be32_to_cpu(leaf1->ents[be16_to_cpu(leaf1->hdr.count) - 1].hashval)))
 801                 return 1;
 802         return 0;
 803 }
 804
 805 /*
 806  * Rebalance leaf entries between two leaf blocks.
 807  * This is actually only called when the second block is new,
 808  * though the code deals with the general case.
 809  * A new entry will be inserted in one of the blocks, and that
 810  * entry is taken into account when balancing.
 811  */
 812 static void
 813 xfs_dir2_leafn_rebalance(
 814         xfs_da_state_t          *state,         /* btree cursor */
 815         xfs_da_state_blk_t      *blk1,          /* first btree block */
 816         xfs_da_state_blk_t      *blk2)          /* second btree block */
 817 {
 818         xfs_da_args_t           *args;          /* operation arguments */
 819         int                     count;          /* count (& direction) leaves */
 820         int                     isleft;         /* new goes in left leaf */
 821         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 822         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 823         int                     mid;            /* midpoint leaf index */
 824 #ifdef DEBUG
 825         int                     oldstale;       /* old count of stale leaves */
 826 #endif
 827         int                     oldsum;         /* old total leaf count */
 828         int                     swap;           /* swapped leaf blocks */
 829
 830         args = state->args;
 831         /*
 832          * If the block order is wrong, swap the arguments.
 833          */
 834         if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
 835                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
 836
 837                 tmp = blk1;
 838                 blk1 = blk2;
 839                 blk2 = tmp;
 840         }
 841         leaf1 = blk1->bp->data;
 842         leaf2 = blk2->bp->data;
 843         oldsum = be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count);
 844 #ifdef DEBUG
 845         oldstale = be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale);
 846 #endif
 847         mid = oldsum >> 1;
 848         /*
 849          * If the old leaf count was odd then the new one will be even,
 850          * so we need to divide the new count evenly.
 851          */
 852         if (oldsum & 1) {
 853                 xfs_dahash_t    midhash;        /* middle entry hash value */
 854
 855                 if (mid >= be16_to_cpu(leaf1->hdr.count))
 856                         midhash = be32_to_cpu(leaf2->ents[mid - be16_to_cpu(leaf1->hdr.count)].hashval);
 857                 else
 858                         midhash = be32_to_cpu(leaf1->ents[mid].hashval);
 859                 isleft = args->hashval <= midhash;
 860         }
 861         /*
 862          * If the old count is even then the new count is odd, so there's
 863          * no preferred side for the new entry.
 864          * Pick the left one.
 865          */
 866         else
 867                 isleft = 1;
 868         /*
 869          * Calculate moved entry count.  Positive means left-to-right,
 870          * negative means right-to-left.  Then move the entries.
 871          */
 872         count = be16_to_cpu(leaf1->hdr.count) - mid + (isleft == 0);
 873         if (count > 0)
 874                 xfs_dir2_leafn_moveents(args, blk1->bp,
 875                         be16_to_cpu(leaf1->hdr.count) - count, blk2->bp, 0, count);
 876         else if (count < 0)
 877                 xfs_dir2_leafn_moveents(args, blk2->bp, 0, blk1->bp,
 878                         be16_to_cpu(leaf1->hdr.count), count);
 879         ASSERT(be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count) == oldsum);
 880         ASSERT(be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale) == oldstale);
 881         /*
 882          * Mark whether we're inserting into the old or new leaf.
 883          */
 884         if (be16_to_cpu(leaf1->hdr.count) < be16_to_cpu(leaf2->hdr.count))
 885                 state->inleaf = swap;
 886         else if (be16_to_cpu(leaf1->hdr.count) > be16_to_cpu(leaf2->hdr.count))
 887                 state->inleaf = !swap;
 888         else
 889                 state->inleaf =
 890                         swap ^ (blk1->index <= be16_to_cpu(leaf1->hdr.count));
 891         /*
 892          * Adjust the expected index for insertion.
 893          */
 894         if (!state->inleaf)
 895                 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
 896
 897         /*
 898          * Finally sanity check just to make sure we are not returning a
 899          * negative index
 900          */
 901         if(blk2->index < 0) {
 902                 state->inleaf = 1;
 903                 blk2->index = 0;
 904                 cmn_err(CE_ALERT,
 905                         "xfs_dir2_leafn_rebalance: picked the wrong leaf? reverting original leaf: "
 906                         "blk1->index %d\n",
 907                         blk1->index);
 908         }
 909 }
 910
 911 /*
 912  * Remove an entry from a node directory.
 913  * This removes the leaf entry and the data entry,
 914  * and updates the free block if necessary.
 915  */
 916 static int                                      /* error */
 917 xfs_dir2_leafn_remove(
 918         xfs_da_args_t           *args,          /* operation arguments */
 919         xfs_dabuf_t             *bp,            /* leaf buffer */
 920         int                     index,          /* leaf entry index */
 921         xfs_da_state_blk_t      *dblk,          /* data block */
 922         int                     *rval)          /* resulting block needs join */
 923 {
 924         xfs_dir2_data_t         *data;          /* data block structure */
 925         xfs_dir2_db_t           db;             /* data block number */
 926         xfs_dabuf_t             *dbp;           /* data block buffer */
 927         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 928         xfs_inode_t             *dp;            /* incore directory inode */
 929         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 930         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 931         int                     longest;        /* longest data free entry */
 932         int                     off;            /* data block entry offset */
 933         xfs_mount_t             *mp;            /* filesystem mount point */
 934         int                     needlog;        /* need to log data header */
 935         int                     needscan;       /* need to rescan data frees */
 936         xfs_trans_t             *tp;            /* transaction pointer */
 937
 938         trace_xfs_dir2_leafn_remove(args, index);
 939
 940         dp = args->dp;
 941         tp = args->trans;
 942         mp = dp->i_mount;
 943         leaf = bp->data;
 944         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
 945         /*
 946          * Point to the entry we're removing.
 947          */
 948         lep = &leaf->ents[index];
 949         /*
 950          * Extract the data block and offset from the entry.
 951          */
 952         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 953         ASSERT(dblk->blkno == db);
 954         off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
 955         ASSERT(dblk->index == off);
 956         /*
 957          * Kill the leaf entry by marking it stale.
 958          * Log the leaf block changes.
 959          */
 960         be16_add_cpu(&leaf->hdr.stale, 1);
 961         xfs_dir2_leaf_log_header(tp, bp);
 962         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 963         xfs_dir2_leaf_log_ents(tp, bp, index, index);
 964         /*
 965          * Make the data entry free.  Keep track of the longest freespace
 966          * in the data block in case it changes.
 967          */
 968         dbp = dblk->bp;
 969         data = dbp->data;
 970         dep = (xfs_dir2_data_entry_t *)((char *)data + off);
 971         longest = be16_to_cpu(data->hdr.bestfree[0].length);
 972         needlog = needscan = 0;
 973         xfs_dir2_data_make_free(tp, dbp, off,
 974                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
 975         /*
 976          * Rescan the data block freespaces for bestfree.
 977          * Log the data block header if needed.
 978          */
 979         if (needscan)
 980                 xfs_dir2_data_freescan(mp, data, &needlog);
 981         if (needlog)
 982                 xfs_dir2_data_log_header(tp, dbp);
 983         xfs_dir2_data_check(dp, dbp);
 984         /*
 985          * If the longest data block freespace changes, need to update
 986          * the corresponding freeblock entry.
 987          */
 988         if (longest < be16_to_cpu(data->hdr.bestfree[0].length)) {
 989                 int             error;          /* error return value */
 990                 xfs_dabuf_t     *fbp;           /* freeblock buffer */
 991                 xfs_dir2_db_t   fdb;            /* freeblock block number */
 992                 int             findex;         /* index in freeblock entries */
 993                 xfs_dir2_free_t *free;          /* freeblock structure */
 994                 int             logfree;        /* need to log free entry */
 995
 996                 /*
 997                  * Convert the data block number to a free block,
 998                  * read in the free block.
 999                  */
1000                 fdb = xfs_dir2_db_to_fdb(mp, db);
1001                 if ((error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb),
1002                                 -1, &fbp, XFS_DATA_FORK))) {
1003                         return error;
1004                 }
1005                 free = fbp->data;
1006                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1007                 ASSERT(be32_to_cpu(free->hdr.firstdb) ==
1008                        XFS_DIR2_MAX_FREE_BESTS(mp) *
1009                        (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
1010                 /*
1011                  * Calculate which entry we need to fix.
1012                  */
1013                 findex = xfs_dir2_db_to_fdindex(mp, db);
1014                 longest = be16_to_cpu(data->hdr.bestfree[0].length);
1015                 /*
1016                  * If the data block is now empty we can get rid of it
1017                  * (usually).
1018                  */
1019                 if (longest == mp->m_dirblksize - (uint)sizeof(data->hdr)) {
1020                         /*
1021                          * Try to punch out the data block.
1022                          */
1023                         error = xfs_dir2_shrink_inode(args, db, dbp);
1024                         if (error == 0) {
1025                                 dblk->bp = NULL;
1026                                 data = NULL;
1027                         }
1028                         /*
1029                          * We can get ENOSPC if there's no space reservation.
1030                          * In this case just drop the buffer and some one else
1031                          * will eventually get rid of the empty block.
1032                          */
1033                         else if (error == ENOSPC && args->total == 0)
1034                                 xfs_da_buf_done(dbp);
1035                         else
1036                                 return error;
1037                 }
1038                 /*
1039                  * If we got rid of the data block, we can eliminate that entry
1040                  * in the free block.
1041                  */
1042                 if (data == NULL) {
1043                         /*
1044                          * One less used entry in the free table.
1045                          */
1046                         be32_add_cpu(&free->hdr.nused, -1);
1047                         xfs_dir2_free_log_header(tp, fbp);
1048                         /*
1049                          * If this was the last entry in the table, we can
1050                          * trim the table size back.  There might be other
1051                          * entries at the end referring to non-existent
1052                          * data blocks, get those too.
1053                          */
1054                         if (findex == be32_to_cpu(free->hdr.nvalid) - 1) {
1055                                 int     i;              /* free entry index */
1056
1057                                 for (i = findex - 1;
1058                                      i >= 0 && be16_to_cpu(free->bests[i]) == NULLDATAOFF;
1059                                      i--)
1060                                         continue;
1061                                 free->hdr.nvalid = cpu_to_be32(i + 1);
1062                                 logfree = 0;
1063                         }
1064                         /*
1065                          * Not the last entry, just punch it out.
1066                          */
1067                         else {
1068                                 free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1069                                 logfree = 1;
1070                         }
1071                         /*
1072                          * If there are no useful entries left in the block,
1073                          * get rid of the block if we can.
1074                          */
1075                         if (!free->hdr.nused) {
1076                                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1077                                 if (error == 0) {
1078                                         fbp = NULL;
1079                                         logfree = 0;
1080                                 } else if (error != ENOSPC || args->total != 0)
1081                                         return error;
1082                                 /*
1083                                  * It's possible to get ENOSPC if there is no
1084                                  * space reservation.  In this case some one
1085                                  * else will eventually get rid of this block.
1086                                  */
1087                         }
1088                 }
1089                 /*
1090                  * Data block is not empty, just set the free entry to
1091                  * the new value.
1092                  */
1093                 else {
1094                         free->bests[findex] = cpu_to_be16(longest);
1095                         logfree = 1;
1096                 }
1097                 /*
1098                  * Log the free entry that changed, unless we got rid of it.
1099                  */
1100                 if (logfree)
1101                         xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1102                 /*
1103                  * Drop the buffer if we still have it.
1104                  */
1105                 if (fbp)
1106                         xfs_da_buf_done(fbp);
1107         }
1108         xfs_dir2_leafn_check(dp, bp);
1109         /*
1110          * Return indication of whether this leaf block is empty enough
1111          * to justify trying to join it with a neighbor.
1112          */
1113         *rval =
1114                 ((uint)sizeof(leaf->hdr) +
1115                  (uint)sizeof(leaf->ents[0]) *
1116                  (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale))) <
1117                 mp->m_dir_magicpct;
1118         return 0;
1119 }
1120
1121 /*
1122  * Split the leaf entries in the old block into old and new blocks.
1123  */
1124 int                                             /* error */
1125 xfs_dir2_leafn_split(
1126         xfs_da_state_t          *state,         /* btree cursor */
1127         xfs_da_state_blk_t      *oldblk,        /* original block */
1128         xfs_da_state_blk_t      *newblk)        /* newly created block */
1129 {
1130         xfs_da_args_t           *args;          /* operation arguments */
1131         xfs_dablk_t             blkno;          /* new leaf block number */
1132         int                     error;          /* error return value */
1133         xfs_mount_t             *mp;            /* filesystem mount point */
1134
1135         /*
1136          * Allocate space for a new leaf node.
1137          */
1138         args = state->args;
1139         mp = args->dp->i_mount;
1140         ASSERT(args != NULL);
1141         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1142         error = xfs_da_grow_inode(args, &blkno);
1143         if (error) {
1144                 return error;
1145         }
1146         /*
1147          * Initialize the new leaf block.
1148          */
1149         error = xfs_dir2_leaf_init(args, xfs_dir2_da_to_db(mp, blkno),
1150                 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1151         if (error) {
1152                 return error;
1153         }
1154         newblk->blkno = blkno;
1155         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1156         /*
1157          * Rebalance the entries across the two leaves, link the new
1158          * block into the leaves.
1159          */
1160         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1161         error = xfs_da_blk_link(state, oldblk, newblk);
1162         if (error) {
1163                 return error;
1164         }
1165         /*
1166          * Insert the new entry in the correct block.
1167          */
1168         if (state->inleaf)
1169                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1170         else
1171                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1172         /*
1173          * Update last hashval in each block since we added the name.
1174          */
1175         oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1176         newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1177         xfs_dir2_leafn_check(args->dp, oldblk->bp);
1178         xfs_dir2_leafn_check(args->dp, newblk->bp);
1179         return error;
1180 }
1181
1182 /*
1183  * Check a leaf block and its neighbors to see if the block should be
1184  * collapsed into one or the other neighbor.  Always keep the block
1185  * with the smaller block number.
1186  * If the current block is over 50% full, don't try to join it, return 0.
1187  * If the block is empty, fill in the state structure and return 2.
1188  * If it can be collapsed, fill in the state structure and return 1.
1189  * If nothing can be done, return 0.
1190  */
1191 int                                             /* error */
1192 xfs_dir2_leafn_toosmall(
1193         xfs_da_state_t          *state,         /* btree cursor */
1194         int                     *action)        /* resulting action to take */
1195 {
1196         xfs_da_state_blk_t      *blk;           /* leaf block */
1197         xfs_dablk_t             blkno;          /* leaf block number */
1198         xfs_dabuf_t             *bp;            /* leaf buffer */
1199         int                     bytes;          /* bytes in use */
1200         int                     count;          /* leaf live entry count */
1201         int                     error;          /* error return value */
1202         int                     forward;        /* sibling block direction */
1203         int                     i;              /* sibling counter */
1204         xfs_da_blkinfo_t        *info;          /* leaf block header */
1205         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1206         int                     rval;           /* result from path_shift */
1207
1208         /*
1209          * Check for the degenerate case of the block being over 50% full.
1210          * If so, it's not worth even looking to see if we might be able
1211          * to coalesce with a sibling.
1212          */
1213         blk = &state->path.blk[state->path.active - 1];
1214         info = blk->bp->data;
1215         ASSERT(be16_to_cpu(info->magic) == XFS_DIR2_LEAFN_MAGIC);
1216         leaf = (xfs_dir2_leaf_t *)info;
1217         count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1218         bytes = (uint)sizeof(leaf->hdr) + count * (uint)sizeof(leaf->ents[0]);
1219         if (bytes > (state->blocksize >> 1)) {
1220                 /*
1221                  * Blk over 50%, don't try to join.
1222                  */
1223                 *action = 0;
1224                 return 0;
1225         }
1226         /*
1227          * Check for the degenerate case of the block being empty.
1228          * If the block is empty, we'll simply delete it, no need to
1229          * coalesce it with a sibling block.  We choose (arbitrarily)
1230          * to merge with the forward block unless it is NULL.
1231          */
1232         if (count == 0) {
1233                 /*
1234                  * Make altpath point to the block we want to keep and
1235                  * path point to the block we want to drop (this one).
1236                  */
1237                 forward = (info->forw != 0);
1238                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1239                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1240                         &rval);
1241                 if (error)
1242                         return error;
1243                 *action = rval ? 2 : 0;
1244                 return 0;
1245         }
1246         /*
1247          * Examine each sibling block to see if we can coalesce with
1248          * at least 25% free space to spare.  We need to figure out
1249          * whether to merge with the forward or the backward block.
1250          * We prefer coalescing with the lower numbered sibling so as
1251          * to shrink a directory over time.
1252          */
1253         forward = be32_to_cpu(info->forw) < be32_to_cpu(info->back);
1254         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1255                 blkno = forward ? be32_to_cpu(info->forw) : be32_to_cpu(info->back);
1256                 if (blkno == 0)
1257                         continue;
1258                 /*
1259                  * Read the sibling leaf block.
1260                  */
1261                 if ((error =
1262                     xfs_da_read_buf(state->args->trans, state->args->dp, blkno,
1263                             -1, &bp, XFS_DATA_FORK))) {
1264                         return error;
1265                 }
1266                 ASSERT(bp != NULL);
1267                 /*
1268                  * Count bytes in the two blocks combined.
1269                  */
1270                 leaf = (xfs_dir2_leaf_t *)info;
1271                 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1272                 bytes = state->blocksize - (state->blocksize >> 2);
1273                 leaf = bp->data;
1274                 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1275                 count += be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1276                 bytes -= count * (uint)sizeof(leaf->ents[0]);
1277                 /*
1278                  * Fits with at least 25% to spare.
1279                  */
1280                 if (bytes >= 0)
1281                         break;
1282                 xfs_da_brelse(state->args->trans, bp);
1283         }
1284         /*
1285          * Didn't like either block, give up.
1286          */
1287         if (i >= 2) {
1288                 *action = 0;
1289                 return 0;
1290         }
1291         /*
1292          * Done with the sibling leaf block here, drop the dabuf
1293          * so path_shift can get it.
1294          */
1295         xfs_da_buf_done(bp);
1296         /*
1297          * Make altpath point to the block we want to keep (the lower
1298          * numbered block) and path point to the block we want to drop.
1299          */
1300         memcpy(&state->altpath, &state->path, sizeof(state->path));
1301         if (blkno < blk->blkno)
1302                 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1303                         &rval);
1304         else
1305                 error = xfs_da_path_shift(state, &state->path, forward, 0,
1306                         &rval);
1307         if (error) {
1308                 return error;
1309         }
1310         *action = rval ? 0 : 1;
1311         return 0;
1312 }
1313
1314 /*
1315  * Move all the leaf entries from drop_blk to save_blk.
1316  * This is done as part of a join operation.
1317  */
1318 void
1319 xfs_dir2_leafn_unbalance(
1320         xfs_da_state_t          *state,         /* cursor */
1321         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1322         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1323 {
1324         xfs_da_args_t           *args;          /* operation arguments */
1325         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1326         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1327
1328         args = state->args;
1329         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1330         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1331         drop_leaf = drop_blk->bp->data;
1332         save_leaf = save_blk->bp->data;
1333         ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1334         ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_DIR2_LEAFN_MAGIC);
1335         /*
1336          * If there are any stale leaf entries, take this opportunity
1337          * to purge them.
1338          */
1339         if (drop_leaf->hdr.stale)
1340                 xfs_dir2_leaf_compact(args, drop_blk->bp);
1341         if (save_leaf->hdr.stale)
1342                 xfs_dir2_leaf_compact(args, save_blk->bp);
1343         /*
1344          * Move the entries from drop to the appropriate end of save.
1345          */
1346         drop_blk->hashval = be32_to_cpu(drop_leaf->ents[be16_to_cpu(drop_leaf->hdr.count) - 1].hashval);
1347         if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1348                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp, 0,
1349                         be16_to_cpu(drop_leaf->hdr.count));
1350         else
1351                 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp,
1352                         be16_to_cpu(save_leaf->hdr.count), be16_to_cpu(drop_leaf->hdr.count));
1353         save_blk->hashval = be32_to_cpu(save_leaf->ents[be16_to_cpu(save_leaf->hdr.count) - 1].hashval);
1354         xfs_dir2_leafn_check(args->dp, save_blk->bp);
1355 }
1356
1357 /*
1358  * Top-level node form directory addname routine.
1359  */
1360 int                                             /* error */
1361 xfs_dir2_node_addname(
1362         xfs_da_args_t           *args)          /* operation arguments */
1363 {
1364         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1365         int                     error;          /* error return value */
1366         int                     rval;           /* sub-return value */
1367         xfs_da_state_t          *state;         /* btree cursor */
1368
1369         trace_xfs_dir2_node_addname(args);
1370
1371         /*
1372          * Allocate and initialize the state (btree cursor).
1373          */
1374         state = xfs_da_state_alloc();
1375         state->args = args;
1376         state->mp = args->dp->i_mount;
1377         state->blocksize = state->mp->m_dirblksize;
1378         state->node_ents = state->mp->m_dir_node_ents;
1379         /*
1380          * Look up the name.  We're not supposed to find it, but
1381          * this gives us the insertion point.
1382          */
1383         error = xfs_da_node_lookup_int(state, &rval);
1384         if (error)
1385                 rval = error;
1386         if (rval != ENOENT) {
1387                 goto done;
1388         }
1389         /*
1390          * Add the data entry to a data block.
1391          * Extravalid is set to a freeblock found by lookup.
1392          */
1393         rval = xfs_dir2_node_addname_int(args,
1394                 state->extravalid ? &state->extrablk : NULL);
1395         if (rval) {
1396                 goto done;
1397         }
1398         blk = &state->path.blk[state->path.active - 1];
1399         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1400         /*
1401          * Add the new leaf entry.
1402          */
1403         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1404         if (rval == 0) {
1405                 /*
1406                  * It worked, fix the hash values up the btree.
1407                  */
1408                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1409                         xfs_da_fixhashpath(state, &state->path);
1410         } else {
1411                 /*
1412                  * It didn't work, we need to split the leaf block.
1413                  */
1414                 if (args->total == 0) {
1415                         ASSERT(rval == ENOSPC);
1416                         goto done;
1417                 }
1418                 /*
1419                  * Split the leaf block and insert the new entry.
1420                  */
1421                 rval = xfs_da_split(state);
1422         }
1423 done:
1424         xfs_da_state_free(state);
1425         return rval;
1426 }
1427
1428 /*
1429  * Add the data entry for a node-format directory name addition.
1430  * The leaf entry is added in xfs_dir2_leafn_add.
1431  * We may enter with a freespace block that the lookup found.
1432  */
1433 static int                                      /* error */
1434 xfs_dir2_node_addname_int(
1435         xfs_da_args_t           *args,          /* operation arguments */
1436         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1437 {
1438         xfs_dir2_data_t         *data;          /* data block structure */
1439         xfs_dir2_db_t           dbno;           /* data block number */
1440         xfs_dabuf_t             *dbp;           /* data block buffer */
1441         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1442         xfs_inode_t             *dp;            /* incore directory inode */
1443         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1444         int                     error;          /* error return value */
1445         xfs_dir2_db_t           fbno;           /* freespace block number */
1446         xfs_dabuf_t             *fbp;           /* freespace buffer */
1447         int                     findex;         /* freespace entry index */
1448         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1449         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1450         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1451         int                     length;         /* length of the new entry */
1452         int                     logfree;        /* need to log free entry */
1453         xfs_mount_t             *mp;            /* filesystem mount point */
1454         int                     needlog;        /* need to log data header */
1455         int                     needscan;       /* need to rescan data frees */
1456         __be16                  *tagp;          /* data entry tag pointer */
1457         xfs_trans_t             *tp;            /* transaction pointer */
1458
1459         dp = args->dp;
1460         mp = dp->i_mount;
1461         tp = args->trans;
1462         length = xfs_dir2_data_entsize(args->namelen);
1463         /*
1464          * If we came in with a freespace block that means that lookup
1465          * found an entry with our hash value.  This is the freespace
1466          * block for that data entry.
1467          */
1468         if (fblk) {
1469                 fbp = fblk->bp;
1470                 /*
1471                  * Remember initial freespace block number.
1472                  */
1473                 ifbno = fblk->blkno;
1474                 free = fbp->data;
1475                 ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1476                 findex = fblk->index;
1477                 /*
1478                  * This means the free entry showed that the data block had
1479                  * space for our entry, so we remembered it.
1480                  * Use that data block.
1481                  */
1482                 if (findex >= 0) {
1483                         ASSERT(findex < be32_to_cpu(free->hdr.nvalid));
1484                         ASSERT(be16_to_cpu(free->bests[findex]) != NULLDATAOFF);
1485                         ASSERT(be16_to_cpu(free->bests[findex]) >= length);
1486                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1487                 }
1488                 /*
1489                  * The data block looked at didn't have enough room.
1490                  * We'll start at the beginning of the freespace entries.
1491                  */
1492                 else {
1493                         dbno = -1;
1494                         findex = 0;
1495                 }
1496         }
1497         /*
1498          * Didn't come in with a freespace block, so don't have a data block.
1499          */
1500         else {
1501                 ifbno = dbno = -1;
1502                 fbp = NULL;
1503                 findex = 0;
1504         }
1505         /*
1506          * If we don't have a data block yet, we're going to scan the
1507          * freespace blocks looking for one.  Figure out what the
1508          * highest freespace block number is.
1509          */
1510         if (dbno == -1) {
1511                 xfs_fileoff_t   fo;             /* freespace block number */
1512
1513                 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1514                         return error;
1515                 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1516                 fbno = ifbno;
1517         }
1518         /*
1519          * While we haven't identified a data block, search the freeblock
1520          * data for a good data block.  If we find a null freeblock entry,
1521          * indicating a hole in the data blocks, remember that.
1522          */
1523         while (dbno == -1) {
1524                 /*
1525                  * If we don't have a freeblock in hand, get the next one.
1526                  */
1527                 if (fbp == NULL) {
1528                         /*
1529                          * Happens the first time through unless lookup gave
1530                          * us a freespace block to start with.
1531                          */
1532                         if (++fbno == 0)
1533                                 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1534                         /*
1535                          * If it's ifbno we already looked at it.
1536                          */
1537                         if (fbno == ifbno)
1538                                 fbno++;
1539                         /*
1540                          * If it's off the end we're done.
1541                          */
1542                         if (fbno >= lastfbno)
1543                                 break;
1544                         /*
1545                          * Read the block.  There can be holes in the
1546                          * freespace blocks, so this might not succeed.
1547                          * This should be really rare, so there's no reason
1548                          * to avoid it.
1549                          */
1550                         if ((error = xfs_da_read_buf(tp, dp,
1551                                         xfs_dir2_db_to_da(mp, fbno), -2, &fbp,
1552                                         XFS_DATA_FORK))) {
1553                                 return error;
1554                         }
1555                         if (unlikely(fbp == NULL)) {
1556                                 continue;
1557                         }
1558                         free = fbp->data;
1559                         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1560                         findex = 0;
1561                 }
1562                 /*
1563                  * Look at the current free entry.  Is it good enough?
1564                  */
1565                 if (be16_to_cpu(free->bests[findex]) != NULLDATAOFF &&
1566                     be16_to_cpu(free->bests[findex]) >= length)
1567                         dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1568                 else {
1569                         /*
1570                          * Are we done with the freeblock?
1571                          */
1572                         if (++findex == be32_to_cpu(free->hdr.nvalid)) {
1573                                 /*
1574                                  * Drop the block.
1575                                  */
1576                                 xfs_da_brelse(tp, fbp);
1577                                 fbp = NULL;
1578                                 if (fblk && fblk->bp)
1579                                         fblk->bp = NULL;
1580                         }
1581                 }
1582         }
1583         /*
1584          * If we don't have a data block, we need to allocate one and make
1585          * the freespace entries refer to it.
1586          */
1587         if (unlikely(dbno == -1)) {
1588                 /*
1589                  * Not allowed to allocate, return failure.
1590                  */
1591                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
1592                                                         args->total == 0) {
1593                         /*
1594                          * Drop the freespace buffer unless it came from our
1595                          * caller.
1596                          */
1597                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1598                                 xfs_da_buf_done(fbp);
1599                         return XFS_ERROR(ENOSPC);
1600                 }
1601                 /*
1602                  * Allocate and initialize the new data block.
1603                  */
1604                 if (unlikely((error = xfs_dir2_grow_inode(args,
1605                                                          XFS_DIR2_DATA_SPACE,
1606                                                          &dbno)) ||
1607                     (error = xfs_dir2_data_init(args, dbno, &dbp)))) {
1608                         /*
1609                          * Drop the freespace buffer unless it came from our
1610                          * caller.
1611                          */
1612                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1613                                 xfs_da_buf_done(fbp);
1614                         return error;
1615                 }
1616                 /*
1617                  * If (somehow) we have a freespace block, get rid of it.
1618                  */
1619                 if (fbp)
1620                         xfs_da_brelse(tp, fbp);
1621                 if (fblk && fblk->bp)
1622                         fblk->bp = NULL;
1623
1624                 /*
1625                  * Get the freespace block corresponding to the data block
1626                  * that was just allocated.
1627                  */
1628                 fbno = xfs_dir2_db_to_fdb(mp, dbno);
1629                 if (unlikely(error = xfs_da_read_buf(tp, dp,
1630                                 xfs_dir2_db_to_da(mp, fbno), -2, &fbp,
1631                                 XFS_DATA_FORK))) {
1632                         xfs_da_buf_done(dbp);
1633                         return error;
1634                 }
1635                 /*
1636                  * If there wasn't a freespace block, the read will
1637                  * return a NULL fbp.  Allocate and initialize a new one.
1638                  */
1639                 if( fbp == NULL ) {
1640                         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1641                                                         &fbno))) {
1642                                 return error;
1643                         }
1644
1645                         if (unlikely(xfs_dir2_db_to_fdb(mp, dbno) != fbno)) {
1646                                 cmn_err(CE_ALERT,
1647                                         "xfs_dir2_node_addname_int: dir ino "
1648                                         "%llu needed freesp block %lld for\n"
1649                                         "  data block %lld, got %lld\n"
1650                                         "  ifbno %llu lastfbno %d\n",
1651                                         (unsigned long long)dp->i_ino,
1652                                         (long long)xfs_dir2_db_to_fdb(mp, dbno),
1653                                         (long long)dbno, (long long)fbno,
1654                                         (unsigned long long)ifbno, lastfbno);
1655                                 if (fblk) {
1656                                         cmn_err(CE_ALERT,
1657                                                 " fblk 0x%p blkno %llu "
1658                                                 "index %d magic 0x%x\n",
1659                                                 fblk,
1660                                                 (unsigned long long)fblk->blkno,
1661                                                 fblk->index,
1662                                                 fblk->magic);
1663                                 } else {
1664                                         cmn_err(CE_ALERT,
1665                                                 " ... fblk is NULL\n");
1666                                 }
1667                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1668                                                  XFS_ERRLEVEL_LOW, mp);
1669                                 return XFS_ERROR(EFSCORRUPTED);
1670                         }
1671
1672                         /*
1673                          * Get a buffer for the new block.
1674                          */
1675                         if ((error = xfs_da_get_buf(tp, dp,
1676                                                    xfs_dir2_db_to_da(mp, fbno),
1677                                                    -1, &fbp, XFS_DATA_FORK))) {
1678                                 return error;
1679                         }
1680                         ASSERT(fbp != NULL);
1681
1682                         /*
1683                          * Initialize the new block to be empty, and remember
1684                          * its first slot as our empty slot.
1685                          */
1686                         free = fbp->data;
1687                         free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
1688                         free->hdr.firstdb = cpu_to_be32(
1689                                 (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1690                                 XFS_DIR2_MAX_FREE_BESTS(mp));
1691                         free->hdr.nvalid = 0;
1692                         free->hdr.nused = 0;
1693                 } else {
1694                         free = fbp->data;
1695                         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
1696                 }
1697
1698                 /*
1699                  * Set the freespace block index from the data block number.
1700                  */
1701                 findex = xfs_dir2_db_to_fdindex(mp, dbno);
1702                 /*
1703                  * If it's after the end of the current entries in the
1704                  * freespace block, extend that table.
1705                  */
1706                 if (findex >= be32_to_cpu(free->hdr.nvalid)) {
1707                         ASSERT(findex < XFS_DIR2_MAX_FREE_BESTS(mp));
1708                         free->hdr.nvalid = cpu_to_be32(findex + 1);
1709                         /*
1710                          * Tag new entry so nused will go up.
1711                          */
1712                         free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1713                 }
1714                 /*
1715                  * If this entry was for an empty data block
1716                  * (this should always be true) then update the header.
1717                  */
1718                 if (be16_to_cpu(free->bests[findex]) == NULLDATAOFF) {
1719                         be32_add_cpu(&free->hdr.nused, 1);
1720                         xfs_dir2_free_log_header(tp, fbp);
1721                 }
1722                 /*
1723                  * Update the real value in the table.
1724                  * We haven't allocated the data entry yet so this will
1725                  * change again.
1726                  */
1727                 data = dbp->data;
1728                 free->bests[findex] = data->hdr.bestfree[0].length;
1729                 logfree = 1;
1730         }
1731         /*
1732          * We had a data block so we don't have to make a new one.
1733          */
1734         else {
1735                 /*
1736                  * If just checking, we succeeded.
1737                  */
1738                 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
1739                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1740                                 xfs_da_buf_done(fbp);
1741                         return 0;
1742                 }
1743                 /*
1744                  * Read the data block in.
1745                  */
1746                 if (unlikely(
1747                     error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1748                                 -1, &dbp, XFS_DATA_FORK))) {
1749                         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1750                                 xfs_da_buf_done(fbp);
1751                         return error;
1752                 }
1753                 data = dbp->data;
1754                 logfree = 0;
1755         }
1756         ASSERT(be16_to_cpu(data->hdr.bestfree[0].length) >= length);
1757         /*
1758          * Point to the existing unused space.
1759          */
1760         dup = (xfs_dir2_data_unused_t *)
1761               ((char *)data + be16_to_cpu(data->hdr.bestfree[0].offset));
1762         needscan = needlog = 0;
1763         /*
1764          * Mark the first part of the unused space, inuse for us.
1765          */
1766         xfs_dir2_data_use_free(tp, dbp, dup,
1767                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)data), length,
1768                 &needlog, &needscan);
1769         /*
1770          * Fill in the new entry and log it.
1771          */
1772         dep = (xfs_dir2_data_entry_t *)dup;
1773         dep->inumber = cpu_to_be64(args->inumber);
1774         dep->namelen = args->namelen;
1775         memcpy(dep->name, args->name, dep->namelen);
1776         tagp = xfs_dir2_data_entry_tag_p(dep);
1777         *tagp = cpu_to_be16((char *)dep - (char *)data);
1778         xfs_dir2_data_log_entry(tp, dbp, dep);
1779         /*
1780          * Rescan the block for bestfree if needed.
1781          */
1782         if (needscan)
1783                 xfs_dir2_data_freescan(mp, data, &needlog);
1784         /*
1785          * Log the data block header if needed.
1786          */
1787         if (needlog)
1788                 xfs_dir2_data_log_header(tp, dbp);
1789         /*
1790          * If the freespace entry is now wrong, update it.
1791          */
1792         if (be16_to_cpu(free->bests[findex]) != be16_to_cpu(data->hdr.bestfree[0].length)) {
1793                 free->bests[findex] = data->hdr.bestfree[0].length;
1794                 logfree = 1;
1795         }
1796         /*
1797          * Log the freespace entry if needed.
1798          */
1799         if (logfree)
1800                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1801         /*
1802          * If the caller didn't hand us the freespace block, drop it.
1803          */
1804         if ((fblk == NULL || fblk->bp == NULL) && fbp != NULL)
1805                 xfs_da_buf_done(fbp);
1806         /*
1807          * Return the data block and offset in args, then drop the data block.
1808          */
1809         args->blkno = (xfs_dablk_t)dbno;
1810         args->index = be16_to_cpu(*tagp);
1811         xfs_da_buf_done(dbp);
1812         return 0;
1813 }
1814
1815 /*
1816  * Lookup an entry in a node-format directory.
1817  * All the real work happens in xfs_da_node_lookup_int.
1818  * The only real output is the inode number of the entry.
1819  */
1820 int                                             /* error */
1821 xfs_dir2_node_lookup(
1822         xfs_da_args_t   *args)                  /* operation arguments */
1823 {
1824         int             error;                  /* error return value */
1825         int             i;                      /* btree level */
1826         int             rval;                   /* operation return value */
1827         xfs_da_state_t  *state;                 /* btree cursor */
1828
1829         trace_xfs_dir2_node_lookup(args);
1830
1831         /*
1832          * Allocate and initialize the btree cursor.
1833          */
1834         state = xfs_da_state_alloc();
1835         state->args = args;
1836         state->mp = args->dp->i_mount;
1837         state->blocksize = state->mp->m_dirblksize;
1838         state->node_ents = state->mp->m_dir_node_ents;
1839         /*
1840          * Fill in the path to the entry in the cursor.
1841          */
1842         error = xfs_da_node_lookup_int(state, &rval);
1843         if (error)
1844                 rval = error;
1845         else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
1846                 /* If a CI match, dup the actual name and return EEXIST */
1847                 xfs_dir2_data_entry_t   *dep;
1848
1849                 dep = (xfs_dir2_data_entry_t *)((char *)state->extrablk.bp->
1850                                                 data + state->extrablk.index);
1851                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1852         }
1853         /*
1854          * Release the btree blocks and leaf block.
1855          */
1856         for (i = 0; i < state->path.active; i++) {
1857                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1858                 state->path.blk[i].bp = NULL;
1859         }
1860         /*
1861          * Release the data block if we have it.
1862          */
1863         if (state->extravalid && state->extrablk.bp) {
1864                 xfs_da_brelse(args->trans, state->extrablk.bp);
1865                 state->extrablk.bp = NULL;
1866         }
1867         xfs_da_state_free(state);
1868         return rval;
1869 }
1870
1871 /*
1872  * Remove an entry from a node-format directory.
1873  */
1874 int                                             /* error */
1875 xfs_dir2_node_removename(
1876         xfs_da_args_t           *args)          /* operation arguments */
1877 {
1878         xfs_da_state_blk_t      *blk;           /* leaf block */
1879         int                     error;          /* error return value */
1880         int                     rval;           /* operation return value */
1881         xfs_da_state_t          *state;         /* btree cursor */
1882
1883         trace_xfs_dir2_node_removename(args);
1884
1885         /*
1886          * Allocate and initialize the btree cursor.
1887          */
1888         state = xfs_da_state_alloc();
1889         state->args = args;
1890         state->mp = args->dp->i_mount;
1891         state->blocksize = state->mp->m_dirblksize;
1892         state->node_ents = state->mp->m_dir_node_ents;
1893         /*
1894          * Look up the entry we're deleting, set up the cursor.
1895          */
1896         error = xfs_da_node_lookup_int(state, &rval);
1897         if (error)
1898                 rval = error;
1899         /*
1900          * Didn't find it, upper layer screwed up.
1901          */
1902         if (rval != EEXIST) {
1903                 xfs_da_state_free(state);
1904                 return rval;
1905         }
1906         blk = &state->path.blk[state->path.active - 1];
1907         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1908         ASSERT(state->extravalid);
1909         /*
1910          * Remove the leaf and data entries.
1911          * Extrablk refers to the data block.
1912          */
1913         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
1914                 &state->extrablk, &rval);
1915         if (error)
1916                 return error;
1917         /*
1918          * Fix the hash values up the btree.
1919          */
1920         xfs_da_fixhashpath(state, &state->path);
1921         /*
1922          * If we need to join leaf blocks, do it.
1923          */
1924         if (rval && state->path.active > 1)
1925                 error = xfs_da_join(state);
1926         /*
1927          * If no errors so far, try conversion to leaf format.
1928          */
1929         if (!error)
1930                 error = xfs_dir2_node_to_leaf(state);
1931         xfs_da_state_free(state);
1932         return error;
1933 }
1934
1935 /*
1936  * Replace an entry's inode number in a node-format directory.
1937  */
1938 int                                             /* error */
1939 xfs_dir2_node_replace(
1940         xfs_da_args_t           *args)          /* operation arguments */
1941 {
1942         xfs_da_state_blk_t      *blk;           /* leaf block */
1943         xfs_dir2_data_t         *data;          /* data block structure */
1944         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
1945         int                     error;          /* error return value */
1946         int                     i;              /* btree level */
1947         xfs_ino_t               inum;           /* new inode number */
1948         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1949         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
1950         int                     rval;           /* internal return value */
1951         xfs_da_state_t          *state;         /* btree cursor */
1952
1953         trace_xfs_dir2_node_replace(args);
1954
1955         /*
1956          * Allocate and initialize the btree cursor.
1957          */
1958         state = xfs_da_state_alloc();
1959         state->args = args;
1960         state->mp = args->dp->i_mount;
1961         state->blocksize = state->mp->m_dirblksize;
1962         state->node_ents = state->mp->m_dir_node_ents;
1963         inum = args->inumber;
1964         /*
1965          * Lookup the entry to change in the btree.
1966          */
1967         error = xfs_da_node_lookup_int(state, &rval);
1968         if (error) {
1969                 rval = error;
1970         }
1971         /*
1972          * It should be found, since the vnodeops layer has looked it up
1973          * and locked it.  But paranoia is good.
1974          */
1975         if (rval == EEXIST) {
1976                 /*
1977                  * Find the leaf entry.
1978                  */
1979                 blk = &state->path.blk[state->path.active - 1];
1980                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1981                 leaf = blk->bp->data;
1982                 lep = &leaf->ents[blk->index];
1983                 ASSERT(state->extravalid);
1984                 /*
1985                  * Point to the data entry.
1986                  */
1987                 data = state->extrablk.bp->data;
1988                 ASSERT(be32_to_cpu(data->hdr.magic) == XFS_DIR2_DATA_MAGIC);
1989                 dep = (xfs_dir2_data_entry_t *)
1990                       ((char *)data +
1991                        xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
1992                 ASSERT(inum != be64_to_cpu(dep->inumber));
1993                 /*
1994                  * Fill in the new inode number and log the entry.
1995                  */
1996                 dep->inumber = cpu_to_be64(inum);
1997                 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
1998                 rval = 0;
1999         }
2000         /*
2001          * Didn't find it, and we're holding a data block.  Drop it.
2002          */
2003         else if (state->extravalid) {
2004                 xfs_da_brelse(args->trans, state->extrablk.bp);
2005                 state->extrablk.bp = NULL;
2006         }
2007         /*
2008          * Release all the buffers in the cursor.
2009          */
2010         for (i = 0; i < state->path.active; i++) {
2011                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
2012                 state->path.blk[i].bp = NULL;
2013         }
2014         xfs_da_state_free(state);
2015         return rval;
2016 }
2017
2018 /*
2019  * Trim off a trailing empty freespace block.
2020  * Return (in rvalp) 1 if we did it, 0 if not.
2021  */
2022 int                                             /* error */
2023 xfs_dir2_node_trim_free(
2024         xfs_da_args_t           *args,          /* operation arguments */
2025         xfs_fileoff_t           fo,             /* free block number */
2026         int                     *rvalp)         /* out: did something */
2027 {
2028         xfs_dabuf_t             *bp;            /* freespace buffer */
2029         xfs_inode_t             *dp;            /* incore directory inode */
2030         int                     error;          /* error return code */
2031         xfs_dir2_free_t         *free;          /* freespace structure */
2032         xfs_mount_t             *mp;            /* filesystem mount point */
2033         xfs_trans_t             *tp;            /* transaction pointer */
2034
2035         dp = args->dp;
2036         mp = dp->i_mount;
2037         tp = args->trans;
2038         /*
2039          * Read the freespace block.
2040          */
2041         if (unlikely(error = xfs_da_read_buf(tp, dp, (xfs_dablk_t)fo, -2, &bp,
2042                         XFS_DATA_FORK))) {
2043                 return error;
2044         }
2045
2046         /*
2047          * There can be holes in freespace.  If fo is a hole, there's
2048          * nothing to do.
2049          */
2050         if (bp == NULL) {
2051                 return 0;
2052         }
2053         free = bp->data;
2054         ASSERT(be32_to_cpu(free->hdr.magic) == XFS_DIR2_FREE_MAGIC);
2055         /*
2056          * If there are used entries, there's nothing to do.
2057          */
2058         if (be32_to_cpu(free->hdr.nused) > 0) {
2059                 xfs_da_brelse(tp, bp);
2060                 *rvalp = 0;
2061                 return 0;
2062         }
2063         /*
2064          * Blow the block away.
2065          */
2066         if ((error =
2067             xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
2068                     bp))) {
2069                 /*
2070                  * Can't fail with ENOSPC since that only happens with no
2071                  * space reservation, when breaking up an extent into two
2072                  * pieces.  This is the last block of an extent.
2073                  */
2074                 ASSERT(error != ENOSPC);
2075                 xfs_da_brelse(tp, bp);
2076                 return error;
2077         }
2078         /*
2079          * Return that we succeeded.
2080          */
2081         *rvalp = 1;
2082         return 0;
2083 }