fs/xfs/xfs_dir2_node.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_types.h"
  22 #include "xfs_log.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_btree.h"
  28 #include "xfs_bmap_btree.h"
  29 #include "xfs_dinode.h"
  30 #include "xfs_inode.h"
  31 #include "xfs_bmap.h"
  32 #include "xfs_dir2_format.h"
  33 #include "xfs_dir2_priv.h"
  34 #include "xfs_error.h"
  35 #include "xfs_trace.h"
  36 #include "xfs_buf_item.h"
  37 #include "xfs_cksum.h"
  38
  39 /*
  40  * Function declarations.
  41  */
  42 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
  43                               int index);
  44 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  45                                      xfs_da_state_blk_t *blk1,
  46                                      xfs_da_state_blk_t *blk2);
  47 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
  48                                  int index, xfs_da_state_blk_t *dblk,
  49                                  int *rval);
  50 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  51                                      xfs_da_state_blk_t *fblk);
  52
  53 /*
  54  * Check internal consistency of a leafn block.
  55  */
  56 #ifdef DEBUG
  57 #define xfs_dir3_leaf_check(mp, bp) \
  58 do { \
  59         if (!xfs_dir3_leafn_check((mp), (bp))) \
  60                 ASSERT(0); \
  61 } while (0);
  62
  63 static bool
  64 xfs_dir3_leafn_check(
  65         struct xfs_mount        *mp,
  66         struct xfs_buf          *bp)
  67 {
  68         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  69         struct xfs_dir3_icleaf_hdr leafhdr;
  70
  71         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
  72
  73         if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
  74                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  75                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  76                         return false;
  77         } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
  78                 return false;
  79
  80         return xfs_dir3_leaf_check_int(mp, &leafhdr, leaf);
  81 }
  82 #else
  83 #define xfs_dir3_leaf_check(mp, bp)
  84 #endif
  85
  86 static bool
  87 xfs_dir3_free_verify(
  88         struct xfs_buf          *bp)
  89 {
  90         struct xfs_mount        *mp = bp->b_target->bt_mount;
  91         struct xfs_dir2_free_hdr *hdr = bp->b_addr;
  92
  93         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  94                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
  95
  96                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_FREE_MAGIC))
  97                         return false;
  98                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
  99                         return false;
 100                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 101                         return false;
 102         } else {
 103                 if (hdr->magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC))
 104                         return false;
 105         }
 106
 107         /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
 108
 109         return true;
 110 }
 111
 112 static void
 113 xfs_dir3_free_read_verify(
 114         struct xfs_buf  *bp)
 115 {
 116         struct xfs_mount        *mp = bp->b_target->bt_mount;
 117
 118         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
 119              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
 120                                           XFS_DIR3_FREE_CRC_OFF)) ||
 121             !xfs_dir3_free_verify(bp)) {
 122                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 123                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 124         }
 125 }
 126
 127 static void
 128 xfs_dir3_free_write_verify(
 129         struct xfs_buf  *bp)
 130 {
 131         struct xfs_mount        *mp = bp->b_target->bt_mount;
 132         struct xfs_buf_log_item *bip = bp->b_fspriv;
 133         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 134
 135         if (!xfs_dir3_free_verify(bp)) {
 136                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 137                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 138                 return;
 139         }
 140
 141         if (!xfs_sb_version_hascrc(&mp->m_sb))
 142                 return;
 143
 144         if (bip)
 145                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 146
 147         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_FREE_CRC_OFF);
 148 }
 149
 150 const struct xfs_buf_ops xfs_dir3_free_buf_ops = {
 151         .verify_read = xfs_dir3_free_read_verify,
 152         .verify_write = xfs_dir3_free_write_verify,
 153 };
 154
 155
 156 static int
 157 __xfs_dir3_free_read(
 158         struct xfs_trans        *tp,
 159         struct xfs_inode        *dp,
 160         xfs_dablk_t             fbno,
 161         xfs_daddr_t             mappedbno,
 162         struct xfs_buf          **bpp)
 163 {
 164         int                     err;
 165
 166         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 167                                 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
 168
 169         /* try read returns without an error or *bpp if it lands in a hole */
 170         if (!err && tp && *bpp)
 171                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
 172         return err;
 173 }
 174
 175 int
 176 xfs_dir2_free_read(
 177         struct xfs_trans        *tp,
 178         struct xfs_inode        *dp,
 179         xfs_dablk_t             fbno,
 180         struct xfs_buf          **bpp)
 181 {
 182         return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
 183 }
 184
 185 static int
 186 xfs_dir2_free_try_read(
 187         struct xfs_trans        *tp,
 188         struct xfs_inode        *dp,
 189         xfs_dablk_t             fbno,
 190         struct xfs_buf          **bpp)
 191 {
 192         return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
 193 }
 194
 195
 196 void
 197 xfs_dir3_free_hdr_from_disk(
 198         struct xfs_dir3_icfree_hdr      *to,
 199         struct xfs_dir2_free            *from)
 200 {
 201         if (from->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC)) {
 202                 to->magic = be32_to_cpu(from->hdr.magic);
 203                 to->firstdb = be32_to_cpu(from->hdr.firstdb);
 204                 to->nvalid = be32_to_cpu(from->hdr.nvalid);
 205                 to->nused = be32_to_cpu(from->hdr.nused);
 206         } else {
 207                 struct xfs_dir3_free_hdr *hdr3 = (struct xfs_dir3_free_hdr *)from;
 208
 209                 to->magic = be32_to_cpu(hdr3->hdr.magic);
 210                 to->firstdb = be32_to_cpu(hdr3->firstdb);
 211                 to->nvalid = be32_to_cpu(hdr3->nvalid);
 212                 to->nused = be32_to_cpu(hdr3->nused);
 213         }
 214
 215         ASSERT(to->magic == XFS_DIR2_FREE_MAGIC ||
 216                to->magic == XFS_DIR3_FREE_MAGIC);
 217 }
 218
 219 static void
 220 xfs_dir3_free_hdr_to_disk(
 221         struct xfs_dir2_free            *to,
 222         struct xfs_dir3_icfree_hdr      *from)
 223 {
 224         ASSERT(from->magic == XFS_DIR2_FREE_MAGIC ||
 225                from->magic == XFS_DIR3_FREE_MAGIC);
 226
 227         if (from->magic == XFS_DIR2_FREE_MAGIC) {
 228                 to->hdr.magic = cpu_to_be32(from->magic);
 229                 to->hdr.firstdb = cpu_to_be32(from->firstdb);
 230                 to->hdr.nvalid = cpu_to_be32(from->nvalid);
 231                 to->hdr.nused = cpu_to_be32(from->nused);
 232         } else {
 233                 struct xfs_dir3_free_hdr *hdr3 = (struct xfs_dir3_free_hdr *)to;
 234
 235                 hdr3->hdr.magic = cpu_to_be32(from->magic);
 236                 hdr3->firstdb = cpu_to_be32(from->firstdb);
 237                 hdr3->nvalid = cpu_to_be32(from->nvalid);
 238                 hdr3->nused = cpu_to_be32(from->nused);
 239         }
 240 }
 241
 242 static int
 243 xfs_dir3_free_get_buf(
 244         struct xfs_trans        *tp,
 245         struct xfs_inode        *dp,
 246         xfs_dir2_db_t           fbno,
 247         struct xfs_buf          **bpp)
 248 {
 249         struct xfs_mount        *mp = dp->i_mount;
 250         struct xfs_buf          *bp;
 251         int                     error;
 252         struct xfs_dir3_icfree_hdr hdr;
 253
 254         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, fbno),
 255                                    -1, &bp, XFS_DATA_FORK);
 256         if (error)
 257                 return error;
 258
 259         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
 260         bp->b_ops = &xfs_dir3_free_buf_ops;
 261
 262         /*
 263          * Initialize the new block to be empty, and remember
 264          * its first slot as our empty slot.
 265          */
 266         memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
 267         memset(&hdr, 0, sizeof(hdr));
 268
 269         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 270                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 271
 272                 hdr.magic = XFS_DIR3_FREE_MAGIC;
 273
 274                 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
 275                 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
 276                 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_uuid);
 277         } else
 278                 hdr.magic = XFS_DIR2_FREE_MAGIC;
 279         xfs_dir3_free_hdr_to_disk(bp->b_addr, &hdr);
 280         *bpp = bp;
 281         return 0;
 282 }
 283
 284 /*
 285  * Log entries from a freespace block.
 286  */
 287 STATIC void
 288 xfs_dir2_free_log_bests(
 289         struct xfs_trans        *tp,
 290         struct xfs_buf          *bp,
 291         int                     first,          /* first entry to log */
 292         int                     last)           /* last entry to log */
 293 {
 294         xfs_dir2_free_t         *free;          /* freespace structure */
 295         __be16                  *bests;
 296
 297         free = bp->b_addr;
 298         bests = xfs_dir3_free_bests_p(tp->t_mountp, free);
 299         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 300                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 301         xfs_trans_log_buf(tp, bp,
 302                 (uint)((char *)&bests[first] - (char *)free),
 303                 (uint)((char *)&bests[last] - (char *)free +
 304                        sizeof(bests[0]) - 1));
 305 }
 306
 307 /*
 308  * Log header from a freespace block.
 309  */
 310 static void
 311 xfs_dir2_free_log_header(
 312         struct xfs_trans        *tp,
 313         struct xfs_buf          *bp)
 314 {
 315         xfs_dir2_free_t         *free;          /* freespace structure */
 316
 317         free = bp->b_addr;
 318         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 319                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 320         xfs_trans_log_buf(tp, bp, 0, xfs_dir3_free_hdr_size(tp->t_mountp) - 1);
 321 }
 322
 323 /*
 324  * Convert a leaf-format directory to a node-format directory.
 325  * We need to change the magic number of the leaf block, and copy
 326  * the freespace table out of the leaf block into its own block.
 327  */
 328 int                                             /* error */
 329 xfs_dir2_leaf_to_node(
 330         xfs_da_args_t           *args,          /* operation arguments */
 331         struct xfs_buf          *lbp)           /* leaf buffer */
 332 {
 333         xfs_inode_t             *dp;            /* incore directory inode */
 334         int                     error;          /* error return value */
 335         struct xfs_buf          *fbp;           /* freespace buffer */
 336         xfs_dir2_db_t           fdb;            /* freespace block number */
 337         xfs_dir2_free_t         *free;          /* freespace structure */
 338         __be16                  *from;          /* pointer to freespace entry */
 339         int                     i;              /* leaf freespace index */
 340         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 341         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 342         xfs_mount_t             *mp;            /* filesystem mount point */
 343         int                     n;              /* count of live freespc ents */
 344         xfs_dir2_data_off_t     off;            /* freespace entry value */
 345         __be16                  *to;            /* pointer to freespace entry */
 346         xfs_trans_t             *tp;            /* transaction pointer */
 347         struct xfs_dir3_icfree_hdr freehdr;
 348
 349         trace_xfs_dir2_leaf_to_node(args);
 350
 351         dp = args->dp;
 352         mp = dp->i_mount;
 353         tp = args->trans;
 354         /*
 355          * Add a freespace block to the directory.
 356          */
 357         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 358                 return error;
 359         }
 360         ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
 361         /*
 362          * Get the buffer for the new freespace block.
 363          */
 364         error = xfs_dir3_free_get_buf(tp, dp, fdb, &fbp);
 365         if (error)
 366                 return error;
 367
 368         free = fbp->b_addr;
 369         xfs_dir3_free_hdr_from_disk(&freehdr, free);
 370         leaf = lbp->b_addr;
 371         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 372         ASSERT(be32_to_cpu(ltp->bestcount) <=
 373                                 (uint)dp->i_d.di_size / mp->m_dirblksize);
 374
 375         /*
 376          * Copy freespace entries from the leaf block to the new block.
 377          * Count active entries.
 378          */
 379         from = xfs_dir2_leaf_bests_p(ltp);
 380         to = xfs_dir3_free_bests_p(mp, free);
 381         for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 382                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 383                         n++;
 384                 *to = cpu_to_be16(off);
 385         }
 386
 387         /*
 388          * Now initialize the freespace block header.
 389          */
 390         freehdr.nused = n;
 391         freehdr.nvalid = be32_to_cpu(ltp->bestcount);
 392
 393         xfs_dir3_free_hdr_to_disk(fbp->b_addr, &freehdr);
 394         xfs_dir2_free_log_bests(tp, fbp, 0, freehdr.nvalid - 1);
 395         xfs_dir2_free_log_header(tp, fbp);
 396
 397         /*
 398          * Converting the leaf to a leafnode is just a matter of changing the
 399          * magic number and the ops. Do the change directly to the buffer as
 400          * it's less work (and less code) than decoding the header to host
 401          * format and back again.
 402          */
 403         if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
 404                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 405         else
 406                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 407         lbp->b_ops = &xfs_dir3_leafn_buf_ops;
 408         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
 409         xfs_dir3_leaf_log_header(tp, lbp);
 410         xfs_dir3_leaf_check(mp, lbp);
 411         return 0;
 412 }
 413
 414 /*
 415  * Add a leaf entry to a leaf block in a node-form directory.
 416  * The other work necessary is done from the caller.
 417  */
 418 static int                                      /* error */
 419 xfs_dir2_leafn_add(
 420         struct xfs_buf          *bp,            /* leaf buffer */
 421         xfs_da_args_t           *args,          /* operation arguments */
 422         int                     index)          /* insertion pt for new entry */
 423 {
 424         int                     compact;        /* compacting stale leaves */
 425         xfs_inode_t             *dp;            /* incore directory inode */
 426         int                     highstale;      /* next stale entry */
 427         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 428         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 429         int                     lfloghigh;      /* high leaf entry logging */
 430         int                     lfloglow;       /* low leaf entry logging */
 431         int                     lowstale;       /* previous stale entry */
 432         xfs_mount_t             *mp;            /* filesystem mount point */
 433         xfs_trans_t             *tp;            /* transaction pointer */
 434         struct xfs_dir3_icleaf_hdr leafhdr;
 435         struct xfs_dir2_leaf_entry *ents;
 436
 437         trace_xfs_dir2_leafn_add(args, index);
 438
 439         dp = args->dp;
 440         mp = dp->i_mount;
 441         tp = args->trans;
 442         leaf = bp->b_addr;
 443         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 444         ents = xfs_dir3_leaf_ents_p(leaf);
 445
 446         /*
 447          * Quick check just to make sure we are not going to index
 448          * into other peoples memory
 449          */
 450         if (index < 0)
 451                 return XFS_ERROR(EFSCORRUPTED);
 452
 453         /*
 454          * If there are already the maximum number of leaf entries in
 455          * the block, if there are no stale entries it won't fit.
 456          * Caller will do a split.  If there are stale entries we'll do
 457          * a compact.
 458          */
 459
 460         if (leafhdr.count == xfs_dir3_max_leaf_ents(mp, leaf)) {
 461                 if (!leafhdr.stale)
 462                         return XFS_ERROR(ENOSPC);
 463                 compact = leafhdr.stale > 1;
 464         } else
 465                 compact = 0;
 466         ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
 467         ASSERT(index == leafhdr.count ||
 468                be32_to_cpu(ents[index].hashval) >= args->hashval);
 469
 470         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 471                 return 0;
 472
 473         /*
 474          * Compact out all but one stale leaf entry.  Leaves behind
 475          * the entry closest to index.
 476          */
 477         if (compact)
 478                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 479                                          &highstale, &lfloglow, &lfloghigh);
 480         else if (leafhdr.stale) {
 481                 /*
 482                  * Set impossible logging indices for this case.
 483                  */
 484                 lfloglow = leafhdr.count;
 485                 lfloghigh = -1;
 486         }
 487
 488         /*
 489          * Insert the new entry, log everything.
 490          */
 491         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 492                                        highstale, &lfloglow, &lfloghigh);
 493
 494         lep->hashval = cpu_to_be32(args->hashval);
 495         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
 496                                 args->blkno, args->index));
 497
 498         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
 499         xfs_dir3_leaf_log_header(tp, bp);
 500         xfs_dir3_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
 501         xfs_dir3_leaf_check(mp, bp);
 502         return 0;
 503 }
 504
 505 #ifdef DEBUG
 506 static void
 507 xfs_dir2_free_hdr_check(
 508         struct xfs_mount *mp,
 509         struct xfs_buf  *bp,
 510         xfs_dir2_db_t   db)
 511 {
 512         struct xfs_dir3_icfree_hdr hdr;
 513
 514         xfs_dir3_free_hdr_from_disk(&hdr, bp->b_addr);
 515
 516         ASSERT((hdr.firstdb % xfs_dir3_free_max_bests(mp)) == 0);
 517         ASSERT(hdr.firstdb <= db);
 518         ASSERT(db < hdr.firstdb + hdr.nvalid);
 519 }
 520 #else
 521 #define xfs_dir2_free_hdr_check(mp, dp, db)
 522 #endif  /* DEBUG */
 523
 524 /*
 525  * Return the last hash value in the leaf.
 526  * Stale entries are ok.
 527  */
 528 xfs_dahash_t                                    /* hash value */
 529 xfs_dir2_leafn_lasthash(
 530         struct xfs_buf  *bp,                    /* leaf buffer */
 531         int             *count)                 /* count of entries in leaf */
 532 {
 533         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 534         struct xfs_dir2_leaf_entry *ents;
 535         struct xfs_dir3_icleaf_hdr leafhdr;
 536
 537         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 538
 539         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 540                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
 541
 542         if (count)
 543                 *count = leafhdr.count;
 544         if (!leafhdr.count)
 545                 return 0;
 546
 547         ents = xfs_dir3_leaf_ents_p(leaf);
 548         return be32_to_cpu(ents[leafhdr.count - 1].hashval);
 549 }
 550
 551 /*
 552  * Look up a leaf entry for space to add a name in a node-format leaf block.
 553  * The extrablk in state is a freespace block.
 554  */
 555 STATIC int
 556 xfs_dir2_leafn_lookup_for_addname(
 557         struct xfs_buf          *bp,            /* leaf buffer */
 558         xfs_da_args_t           *args,          /* operation arguments */
 559         int                     *indexp,        /* out: leaf entry index */
 560         xfs_da_state_t          *state)         /* state to fill in */
 561 {
 562         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 563         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 564         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 565         xfs_inode_t             *dp;            /* incore directory inode */
 566         int                     error;          /* error return value */
 567         int                     fi;             /* free entry index */
 568         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 569         int                     index;          /* leaf entry index */
 570         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 571         int                     length;         /* length of new data entry */
 572         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 573         xfs_mount_t             *mp;            /* filesystem mount point */
 574         xfs_dir2_db_t           newdb;          /* new data block number */
 575         xfs_dir2_db_t           newfdb;         /* new free block number */
 576         xfs_trans_t             *tp;            /* transaction pointer */
 577         struct xfs_dir2_leaf_entry *ents;
 578         struct xfs_dir3_icleaf_hdr leafhdr;
 579
 580         dp = args->dp;
 581         tp = args->trans;
 582         mp = dp->i_mount;
 583         leaf = bp->b_addr;
 584         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 585         ents = xfs_dir3_leaf_ents_p(leaf);
 586
 587         xfs_dir3_leaf_check(mp, bp);
 588         ASSERT(leafhdr.count > 0);
 589
 590         /*
 591          * Look up the hash value in the leaf entries.
 592          */
 593         index = xfs_dir2_leaf_search_hash(args, bp);
 594         /*
 595          * Do we have a buffer coming in?
 596          */
 597         if (state->extravalid) {
 598                 /* If so, it's a free block buffer, get the block number. */
 599                 curbp = state->extrablk.bp;
 600                 curfdb = state->extrablk.blkno;
 601                 free = curbp->b_addr;
 602                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 603                        free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 604         }
 605         length = xfs_dir2_data_entsize(args->namelen);
 606         /*
 607          * Loop over leaf entries with the right hash value.
 608          */
 609         for (lep = &ents[index];
 610              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 611              lep++, index++) {
 612                 /*
 613                  * Skip stale leaf entries.
 614                  */
 615                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 616                         continue;
 617                 /*
 618                  * Pull the data block number from the entry.
 619                  */
 620                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 621                 /*
 622                  * For addname, we're looking for a place to put the new entry.
 623                  * We want to use a data block with an entry of equal
 624                  * hash value to ours if there is one with room.
 625                  *
 626                  * If this block isn't the data block we already have
 627                  * in hand, take a look at it.
 628                  */
 629                 if (newdb != curdb) {
 630                         __be16 *bests;
 631
 632                         curdb = newdb;
 633                         /*
 634                          * Convert the data block to the free block
 635                          * holding its freespace information.
 636                          */
 637                         newfdb = xfs_dir2_db_to_fdb(mp, newdb);
 638                         /*
 639                          * If it's not the one we have in hand, read it in.
 640                          */
 641                         if (newfdb != curfdb) {
 642                                 /*
 643                                  * If we had one before, drop it.
 644                                  */
 645                                 if (curbp)
 646                                         xfs_trans_brelse(tp, curbp);
 647
 648                                 error = xfs_dir2_free_read(tp, dp,
 649                                                 xfs_dir2_db_to_da(mp, newfdb),
 650                                                 &curbp);
 651                                 if (error)
 652                                         return error;
 653                                 free = curbp->b_addr;
 654
 655                                 xfs_dir2_free_hdr_check(mp, curbp, curdb);
 656                         }
 657                         /*
 658                          * Get the index for our entry.
 659                          */
 660                         fi = xfs_dir2_db_to_fdindex(mp, curdb);
 661                         /*
 662                          * If it has room, return it.
 663                          */
 664                         bests = xfs_dir3_free_bests_p(mp, free);
 665                         if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
 666                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 667                                                         XFS_ERRLEVEL_LOW, mp);
 668                                 if (curfdb != newfdb)
 669                                         xfs_trans_brelse(tp, curbp);
 670                                 return XFS_ERROR(EFSCORRUPTED);
 671                         }
 672                         curfdb = newfdb;
 673                         if (be16_to_cpu(bests[fi]) >= length)
 674                                 goto out;
 675                 }
 676         }
 677         /* Didn't find any space */
 678         fi = -1;
 679 out:
 680         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 681         if (curbp) {
 682                 /* Giving back a free block. */
 683                 state->extravalid = 1;
 684                 state->extrablk.bp = curbp;
 685                 state->extrablk.index = fi;
 686                 state->extrablk.blkno = curfdb;
 687
 688                 /*
 689                  * Important: this magic number is not in the buffer - it's for
 690                  * buffer type information and therefore only the free/data type
 691                  * matters here, not whether CRCs are enabled or not.
 692                  */
 693                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 694         } else {
 695                 state->extravalid = 0;
 696         }
 697         /*
 698          * Return the index, that will be the insertion point.
 699          */
 700         *indexp = index;
 701         return XFS_ERROR(ENOENT);
 702 }
 703
 704 /*
 705  * Look up a leaf entry in a node-format leaf block.
 706  * The extrablk in state a data block.
 707  */
 708 STATIC int
 709 xfs_dir2_leafn_lookup_for_entry(
 710         struct xfs_buf          *bp,            /* leaf buffer */
 711         xfs_da_args_t           *args,          /* operation arguments */
 712         int                     *indexp,        /* out: leaf entry index */
 713         xfs_da_state_t          *state)         /* state to fill in */
 714 {
 715         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 716         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 717         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 718         xfs_inode_t             *dp;            /* incore directory inode */
 719         int                     error;          /* error return value */
 720         int                     index;          /* leaf entry index */
 721         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 722         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 723         xfs_mount_t             *mp;            /* filesystem mount point */
 724         xfs_dir2_db_t           newdb;          /* new data block number */
 725         xfs_trans_t             *tp;            /* transaction pointer */
 726         enum xfs_dacmp          cmp;            /* comparison result */
 727         struct xfs_dir2_leaf_entry *ents;
 728         struct xfs_dir3_icleaf_hdr leafhdr;
 729
 730         dp = args->dp;
 731         tp = args->trans;
 732         mp = dp->i_mount;
 733         leaf = bp->b_addr;
 734         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 735         ents = xfs_dir3_leaf_ents_p(leaf);
 736
 737         xfs_dir3_leaf_check(mp, bp);
 738         ASSERT(leafhdr.count > 0);
 739
 740         /*
 741          * Look up the hash value in the leaf entries.
 742          */
 743         index = xfs_dir2_leaf_search_hash(args, bp);
 744         /*
 745          * Do we have a buffer coming in?
 746          */
 747         if (state->extravalid) {
 748                 curbp = state->extrablk.bp;
 749                 curdb = state->extrablk.blkno;
 750         }
 751         /*
 752          * Loop over leaf entries with the right hash value.
 753          */
 754         for (lep = &ents[index];
 755              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 756              lep++, index++) {
 757                 /*
 758                  * Skip stale leaf entries.
 759                  */
 760                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 761                         continue;
 762                 /*
 763                  * Pull the data block number from the entry.
 764                  */
 765                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 766                 /*
 767                  * Not adding a new entry, so we really want to find
 768                  * the name given to us.
 769                  *
 770                  * If it's a different data block, go get it.
 771                  */
 772                 if (newdb != curdb) {
 773                         /*
 774                          * If we had a block before that we aren't saving
 775                          * for a CI name, drop it
 776                          */
 777                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 778                                                 curdb != state->extrablk.blkno))
 779                                 xfs_trans_brelse(tp, curbp);
 780                         /*
 781                          * If needing the block that is saved with a CI match,
 782                          * use it otherwise read in the new data block.
 783                          */
 784                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 785                                         newdb == state->extrablk.blkno) {
 786                                 ASSERT(state->extravalid);
 787                                 curbp = state->extrablk.bp;
 788                         } else {
 789                                 error = xfs_dir3_data_read(tp, dp,
 790                                                 xfs_dir2_db_to_da(mp, newdb),
 791                                                 -1, &curbp);
 792                                 if (error)
 793                                         return error;
 794                         }
 795                         xfs_dir3_data_check(dp, curbp);
 796                         curdb = newdb;
 797                 }
 798                 /*
 799                  * Point to the data entry.
 800                  */
 801                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 802                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
 803                 /*
 804                  * Compare the entry and if it's an exact match, return
 805                  * EEXIST immediately. If it's the first case-insensitive
 806                  * match, store the block & inode number and continue looking.
 807                  */
 808                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 809                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 810                         /* If there is a CI match block, drop it */
 811                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 812                                                 curdb != state->extrablk.blkno)
 813                                 xfs_trans_brelse(tp, state->extrablk.bp);
 814                         args->cmpresult = cmp;
 815                         args->inumber = be64_to_cpu(dep->inumber);
 816                         *indexp = index;
 817                         state->extravalid = 1;
 818                         state->extrablk.bp = curbp;
 819                         state->extrablk.blkno = curdb;
 820                         state->extrablk.index = (int)((char *)dep -
 821                                                         (char *)curbp->b_addr);
 822                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 823                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 824                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 825                         if (cmp == XFS_CMP_EXACT)
 826                                 return XFS_ERROR(EEXIST);
 827                 }
 828         }
 829         ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
 830         if (curbp) {
 831                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 832                         /* Giving back last used data block. */
 833                         state->extravalid = 1;
 834                         state->extrablk.bp = curbp;
 835                         state->extrablk.index = -1;
 836                         state->extrablk.blkno = curdb;
 837                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 838                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 839                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 840                 } else {
 841                         /* If the curbp is not the CI match block, drop it */
 842                         if (state->extrablk.bp != curbp)
 843                                 xfs_trans_brelse(tp, curbp);
 844                 }
 845         } else {
 846                 state->extravalid = 0;
 847         }
 848         *indexp = index;
 849         return XFS_ERROR(ENOENT);
 850 }
 851
 852 /*
 853  * Look up a leaf entry in a node-format leaf block.
 854  * If this is an addname then the extrablk in state is a freespace block,
 855  * otherwise it's a data block.
 856  */
 857 int
 858 xfs_dir2_leafn_lookup_int(
 859         struct xfs_buf          *bp,            /* leaf buffer */
 860         xfs_da_args_t           *args,          /* operation arguments */
 861         int                     *indexp,        /* out: leaf entry index */
 862         xfs_da_state_t          *state)         /* state to fill in */
 863 {
 864         if (args->op_flags & XFS_DA_OP_ADDNAME)
 865                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 866                                                         state);
 867         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 868 }
 869
 870 /*
 871  * Move count leaf entries from source to destination leaf.
 872  * Log entries and headers.  Stale entries are preserved.
 873  */
 874 static void
 875 xfs_dir3_leafn_moveents(
 876         xfs_da_args_t                   *args,  /* operation arguments */
 877         struct xfs_buf                  *bp_s,  /* source */
 878         struct xfs_dir3_icleaf_hdr      *shdr,
 879         struct xfs_dir2_leaf_entry      *sents,
 880         int                             start_s,/* source leaf index */
 881         struct xfs_buf                  *bp_d,  /* destination */
 882         struct xfs_dir3_icleaf_hdr      *dhdr,
 883         struct xfs_dir2_leaf_entry      *dents,
 884         int                             start_d,/* destination leaf index */
 885         int                             count)  /* count of leaves to copy */
 886 {
 887         struct xfs_trans                *tp = args->trans;
 888         int                             stale;  /* count stale leaves copied */
 889
 890         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 891
 892         /*
 893          * Silently return if nothing to do.
 894          */
 895         if (count == 0)
 896                 return;
 897
 898         /*
 899          * If the destination index is not the end of the current
 900          * destination leaf entries, open up a hole in the destination
 901          * to hold the new entries.
 902          */
 903         if (start_d < dhdr->count) {
 904                 memmove(&dents[start_d + count], &dents[start_d],
 905                         (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
 906                 xfs_dir3_leaf_log_ents(tp, bp_d, start_d + count,
 907                                        count + dhdr->count - 1);
 908         }
 909         /*
 910          * If the source has stale leaves, count the ones in the copy range
 911          * so we can update the header correctly.
 912          */
 913         if (shdr->stale) {
 914                 int     i;                      /* temp leaf index */
 915
 916                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 917                         if (sents[i].address ==
 918                                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 919                                 stale++;
 920                 }
 921         } else
 922                 stale = 0;
 923         /*
 924          * Copy the leaf entries from source to destination.
 925          */
 926         memcpy(&dents[start_d], &sents[start_s],
 927                 count * sizeof(xfs_dir2_leaf_entry_t));
 928         xfs_dir3_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
 929
 930         /*
 931          * If there are source entries after the ones we copied,
 932          * delete the ones we copied by sliding the next ones down.
 933          */
 934         if (start_s + count < shdr->count) {
 935                 memmove(&sents[start_s], &sents[start_s + count],
 936                         count * sizeof(xfs_dir2_leaf_entry_t));
 937                 xfs_dir3_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
 938         }
 939
 940         /*
 941          * Update the headers and log them.
 942          */
 943         shdr->count -= count;
 944         shdr->stale -= stale;
 945         dhdr->count += count;
 946         dhdr->stale += stale;
 947 }
 948
 949 /*
 950  * Determine the sort order of two leaf blocks.
 951  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 952  */
 953 int                                             /* sort order */
 954 xfs_dir2_leafn_order(
 955         struct xfs_buf          *leaf1_bp,              /* leaf1 buffer */
 956         struct xfs_buf          *leaf2_bp)              /* leaf2 buffer */
 957 {
 958         struct xfs_dir2_leaf    *leaf1 = leaf1_bp->b_addr;
 959         struct xfs_dir2_leaf    *leaf2 = leaf2_bp->b_addr;
 960         struct xfs_dir2_leaf_entry *ents1;
 961         struct xfs_dir2_leaf_entry *ents2;
 962         struct xfs_dir3_icleaf_hdr hdr1;
 963         struct xfs_dir3_icleaf_hdr hdr2;
 964
 965         xfs_dir3_leaf_hdr_from_disk(&hdr1, leaf1);
 966         xfs_dir3_leaf_hdr_from_disk(&hdr2, leaf2);
 967         ents1 = xfs_dir3_leaf_ents_p(leaf1);
 968         ents2 = xfs_dir3_leaf_ents_p(leaf2);
 969
 970         if (hdr1.count > 0 && hdr2.count > 0 &&
 971             (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
 972              be32_to_cpu(ents2[hdr2.count - 1].hashval) <
 973                                 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
 974                 return 1;
 975         return 0;
 976 }
 977
 978 /*
 979  * Rebalance leaf entries between two leaf blocks.
 980  * This is actually only called when the second block is new,
 981  * though the code deals with the general case.
 982  * A new entry will be inserted in one of the blocks, and that
 983  * entry is taken into account when balancing.
 984  */
 985 static void
 986 xfs_dir2_leafn_rebalance(
 987         xfs_da_state_t          *state,         /* btree cursor */
 988         xfs_da_state_blk_t      *blk1,          /* first btree block */
 989         xfs_da_state_blk_t      *blk2)          /* second btree block */
 990 {
 991         xfs_da_args_t           *args;          /* operation arguments */
 992         int                     count;          /* count (& direction) leaves */
 993         int                     isleft;         /* new goes in left leaf */
 994         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 995         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 996         int                     mid;            /* midpoint leaf index */
 997 #if defined(DEBUG) || defined(XFS_WARN)
 998         int                     oldstale;       /* old count of stale leaves */
 999 #endif
1000         int                     oldsum;         /* old total leaf count */
1001         int                     swap;           /* swapped leaf blocks */
1002         struct xfs_dir2_leaf_entry *ents1;
1003         struct xfs_dir2_leaf_entry *ents2;
1004         struct xfs_dir3_icleaf_hdr hdr1;
1005         struct xfs_dir3_icleaf_hdr hdr2;
1006
1007         args = state->args;
1008         /*
1009          * If the block order is wrong, swap the arguments.
1010          */
1011         if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
1012                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
1013
1014                 tmp = blk1;
1015                 blk1 = blk2;
1016                 blk2 = tmp;
1017         }
1018         leaf1 = blk1->bp->b_addr;
1019         leaf2 = blk2->bp->b_addr;
1020         xfs_dir3_leaf_hdr_from_disk(&hdr1, leaf1);
1021         xfs_dir3_leaf_hdr_from_disk(&hdr2, leaf2);
1022         ents1 = xfs_dir3_leaf_ents_p(leaf1);
1023         ents2 = xfs_dir3_leaf_ents_p(leaf2);
1024
1025         oldsum = hdr1.count + hdr2.count;
1026 #if defined(DEBUG) || defined(XFS_WARN)
1027         oldstale = hdr1.stale + hdr2.stale;
1028 #endif
1029         mid = oldsum >> 1;
1030
1031         /*
1032          * If the old leaf count was odd then the new one will be even,
1033          * so we need to divide the new count evenly.
1034          */
1035         if (oldsum & 1) {
1036                 xfs_dahash_t    midhash;        /* middle entry hash value */
1037
1038                 if (mid >= hdr1.count)
1039                         midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1040                 else
1041                         midhash = be32_to_cpu(ents1[mid].hashval);
1042                 isleft = args->hashval <= midhash;
1043         }
1044         /*
1045          * If the old count is even then the new count is odd, so there's
1046          * no preferred side for the new entry.
1047          * Pick the left one.
1048          */
1049         else
1050                 isleft = 1;
1051         /*
1052          * Calculate moved entry count.  Positive means left-to-right,
1053          * negative means right-to-left.  Then move the entries.
1054          */
1055         count = hdr1.count - mid + (isleft == 0);
1056         if (count > 0)
1057                 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1058                                         hdr1.count - count, blk2->bp,
1059                                         &hdr2, ents2, 0, count);
1060         else if (count < 0)
1061                 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1062                                         blk1->bp, &hdr1, ents1,
1063                                         hdr1.count, count);
1064
1065         ASSERT(hdr1.count + hdr2.count == oldsum);
1066         ASSERT(hdr1.stale + hdr2.stale == oldstale);
1067
1068         /* log the changes made when moving the entries */
1069         xfs_dir3_leaf_hdr_to_disk(leaf1, &hdr1);
1070         xfs_dir3_leaf_hdr_to_disk(leaf2, &hdr2);
1071         xfs_dir3_leaf_log_header(args->trans, blk1->bp);
1072         xfs_dir3_leaf_log_header(args->trans, blk2->bp);
1073
1074         xfs_dir3_leaf_check(args->dp->i_mount, blk1->bp);
1075         xfs_dir3_leaf_check(args->dp->i_mount, blk2->bp);
1076
1077         /*
1078          * Mark whether we're inserting into the old or new leaf.
1079          */
1080         if (hdr1.count < hdr2.count)
1081                 state->inleaf = swap;
1082         else if (hdr1.count > hdr2.count)
1083                 state->inleaf = !swap;
1084         else
1085                 state->inleaf = swap ^ (blk1->index <= hdr1.count);
1086         /*
1087          * Adjust the expected index for insertion.
1088          */
1089         if (!state->inleaf)
1090                 blk2->index = blk1->index - hdr1.count;
1091
1092         /*
1093          * Finally sanity check just to make sure we are not returning a
1094          * negative index
1095          */
1096         if(blk2->index < 0) {
1097                 state->inleaf = 1;
1098                 blk2->index = 0;
1099                 xfs_alert(args->dp->i_mount,
1100         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d\n",
1101                         __func__, blk1->index);
1102         }
1103 }
1104
1105 static int
1106 xfs_dir3_data_block_free(
1107         xfs_da_args_t           *args,
1108         struct xfs_dir2_data_hdr *hdr,
1109         struct xfs_dir2_free    *free,
1110         xfs_dir2_db_t           fdb,
1111         int                     findex,
1112         struct xfs_buf          *fbp,
1113         int                     longest)
1114 {
1115         struct xfs_trans        *tp = args->trans;
1116         int                     logfree = 0;
1117         __be16                  *bests;
1118         struct xfs_dir3_icfree_hdr freehdr;
1119
1120         xfs_dir3_free_hdr_from_disk(&freehdr, free);
1121
1122         bests = xfs_dir3_free_bests_p(tp->t_mountp, free);
1123         if (hdr) {
1124                 /*
1125                  * Data block is not empty, just set the free entry to the new
1126                  * value.
1127                  */
1128                 bests[findex] = cpu_to_be16(longest);
1129                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1130                 return 0;
1131         }
1132
1133         /* One less used entry in the free table. */
1134         freehdr.nused--;
1135
1136         /*
1137          * If this was the last entry in the table, we can trim the table size
1138          * back.  There might be other entries at the end referring to
1139          * non-existent data blocks, get those too.
1140          */
1141         if (findex == freehdr.nvalid - 1) {
1142                 int     i;              /* free entry index */
1143
1144                 for (i = findex - 1; i >= 0; i--) {
1145                         if (bests[i] != cpu_to_be16(NULLDATAOFF))
1146                                 break;
1147                 }
1148                 freehdr.nvalid = i + 1;
1149                 logfree = 0;
1150         } else {
1151                 /* Not the last entry, just punch it out.  */
1152                 bests[findex] = cpu_to_be16(NULLDATAOFF);
1153                 logfree = 1;
1154         }
1155
1156         xfs_dir3_free_hdr_to_disk(free, &freehdr);
1157         xfs_dir2_free_log_header(tp, fbp);
1158
1159         /*
1160          * If there are no useful entries left in the block, get rid of the
1161          * block if we can.
1162          */
1163         if (!freehdr.nused) {
1164                 int error;
1165
1166                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1167                 if (error == 0) {
1168                         fbp = NULL;
1169                         logfree = 0;
1170                 } else if (error != ENOSPC || args->total != 0)
1171                         return error;
1172                 /*
1173                  * It's possible to get ENOSPC if there is no
1174                  * space reservation.  In this case some one
1175                  * else will eventually get rid of this block.
1176                  */
1177         }
1178
1179         /* Log the free entry that changed, unless we got rid of it.  */
1180         if (logfree)
1181                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1182         return 0;
1183 }
1184
1185 /*
1186  * Remove an entry from a node directory.
1187  * This removes the leaf entry and the data entry,
1188  * and updates the free block if necessary.
1189  */
1190 static int                                      /* error */
1191 xfs_dir2_leafn_remove(
1192         xfs_da_args_t           *args,          /* operation arguments */
1193         struct xfs_buf          *bp,            /* leaf buffer */
1194         int                     index,          /* leaf entry index */
1195         xfs_da_state_blk_t      *dblk,          /* data block */
1196         int                     *rval)          /* resulting block needs join */
1197 {
1198         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1199         xfs_dir2_db_t           db;             /* data block number */
1200         struct xfs_buf          *dbp;           /* data block buffer */
1201         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1202         xfs_inode_t             *dp;            /* incore directory inode */
1203         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1204         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1205         int                     longest;        /* longest data free entry */
1206         int                     off;            /* data block entry offset */
1207         xfs_mount_t             *mp;            /* filesystem mount point */
1208         int                     needlog;        /* need to log data header */
1209         int                     needscan;       /* need to rescan data frees */
1210         xfs_trans_t             *tp;            /* transaction pointer */
1211         struct xfs_dir2_data_free *bf;          /* bestfree table */
1212         struct xfs_dir3_icleaf_hdr leafhdr;
1213         struct xfs_dir2_leaf_entry *ents;
1214
1215         trace_xfs_dir2_leafn_remove(args, index);
1216
1217         dp = args->dp;
1218         tp = args->trans;
1219         mp = dp->i_mount;
1220         leaf = bp->b_addr;
1221         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1222         ents = xfs_dir3_leaf_ents_p(leaf);
1223
1224         /*
1225          * Point to the entry we're removing.
1226          */
1227         lep = &ents[index];
1228
1229         /*
1230          * Extract the data block and offset from the entry.
1231          */
1232         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1233         ASSERT(dblk->blkno == db);
1234         off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
1235         ASSERT(dblk->index == off);
1236
1237         /*
1238          * Kill the leaf entry by marking it stale.
1239          * Log the leaf block changes.
1240          */
1241         leafhdr.stale++;
1242         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
1243         xfs_dir3_leaf_log_header(tp, bp);
1244
1245         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1246         xfs_dir3_leaf_log_ents(tp, bp, index, index);
1247
1248         /*
1249          * Make the data entry free.  Keep track of the longest freespace
1250          * in the data block in case it changes.
1251          */
1252         dbp = dblk->bp;
1253         hdr = dbp->b_addr;
1254         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1255         bf = xfs_dir3_data_bestfree_p(hdr);
1256         longest = be16_to_cpu(bf[0].length);
1257         needlog = needscan = 0;
1258         xfs_dir2_data_make_free(tp, dbp, off,
1259                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1260         /*
1261          * Rescan the data block freespaces for bestfree.
1262          * Log the data block header if needed.
1263          */
1264         if (needscan)
1265                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1266         if (needlog)
1267                 xfs_dir2_data_log_header(tp, dbp);
1268         xfs_dir3_data_check(dp, dbp);
1269         /*
1270          * If the longest data block freespace changes, need to update
1271          * the corresponding freeblock entry.
1272          */
1273         if (longest < be16_to_cpu(bf[0].length)) {
1274                 int             error;          /* error return value */
1275                 struct xfs_buf  *fbp;           /* freeblock buffer */
1276                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1277                 int             findex;         /* index in freeblock entries */
1278                 xfs_dir2_free_t *free;          /* freeblock structure */
1279
1280                 /*
1281                  * Convert the data block number to a free block,
1282                  * read in the free block.
1283                  */
1284                 fdb = xfs_dir2_db_to_fdb(mp, db);
1285                 error = xfs_dir2_free_read(tp, dp, xfs_dir2_db_to_da(mp, fdb),
1286                                            &fbp);
1287                 if (error)
1288                         return error;
1289                 free = fbp->b_addr;
1290 #ifdef DEBUG
1291         {
1292                 struct xfs_dir3_icfree_hdr freehdr;
1293                 xfs_dir3_free_hdr_from_disk(&freehdr, free);
1294                 ASSERT(freehdr.firstdb == xfs_dir3_free_max_bests(mp) *
1295                                           (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
1296         }
1297 #endif
1298                 /*
1299                  * Calculate which entry we need to fix.
1300                  */
1301                 findex = xfs_dir2_db_to_fdindex(mp, db);
1302                 longest = be16_to_cpu(bf[0].length);
1303                 /*
1304                  * If the data block is now empty we can get rid of it
1305                  * (usually).
1306                  */
1307                 if (longest == mp->m_dirblksize -
1308                                xfs_dir3_data_entry_offset(hdr)) {
1309                         /*
1310                          * Try to punch out the data block.
1311                          */
1312                         error = xfs_dir2_shrink_inode(args, db, dbp);
1313                         if (error == 0) {
1314                                 dblk->bp = NULL;
1315                                 hdr = NULL;
1316                         }
1317                         /*
1318                          * We can get ENOSPC if there's no space reservation.
1319                          * In this case just drop the buffer and some one else
1320                          * will eventually get rid of the empty block.
1321                          */
1322                         else if (!(error == ENOSPC && args->total == 0))
1323                                 return error;
1324                 }
1325                 /*
1326                  * If we got rid of the data block, we can eliminate that entry
1327                  * in the free block.
1328                  */
1329                 error = xfs_dir3_data_block_free(args, hdr, free,
1330                                                  fdb, findex, fbp, longest);
1331                 if (error)
1332                         return error;
1333         }
1334
1335         xfs_dir3_leaf_check(mp, bp);
1336         /*
1337          * Return indication of whether this leaf block is empty enough
1338          * to justify trying to join it with a neighbor.
1339          */
1340         *rval = (xfs_dir3_leaf_hdr_size(leaf) +
1341                  (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1342                 mp->m_dir_magicpct;
1343         return 0;
1344 }
1345
1346 /*
1347  * Split the leaf entries in the old block into old and new blocks.
1348  */
1349 int                                             /* error */
1350 xfs_dir2_leafn_split(
1351         xfs_da_state_t          *state,         /* btree cursor */
1352         xfs_da_state_blk_t      *oldblk,        /* original block */
1353         xfs_da_state_blk_t      *newblk)        /* newly created block */
1354 {
1355         xfs_da_args_t           *args;          /* operation arguments */
1356         xfs_dablk_t             blkno;          /* new leaf block number */
1357         int                     error;          /* error return value */
1358         xfs_mount_t             *mp;            /* filesystem mount point */
1359
1360         /*
1361          * Allocate space for a new leaf node.
1362          */
1363         args = state->args;
1364         mp = args->dp->i_mount;
1365         ASSERT(args != NULL);
1366         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1367         error = xfs_da_grow_inode(args, &blkno);
1368         if (error) {
1369                 return error;
1370         }
1371         /*
1372          * Initialize the new leaf block.
1373          */
1374         error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(mp, blkno),
1375                                       &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1376         if (error)
1377                 return error;
1378
1379         newblk->blkno = blkno;
1380         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1381         /*
1382          * Rebalance the entries across the two leaves, link the new
1383          * block into the leaves.
1384          */
1385         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1386         error = xfs_da3_blk_link(state, oldblk, newblk);
1387         if (error) {
1388                 return error;
1389         }
1390         /*
1391          * Insert the new entry in the correct block.
1392          */
1393         if (state->inleaf)
1394                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1395         else
1396                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1397         /*
1398          * Update last hashval in each block since we added the name.
1399          */
1400         oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1401         newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1402         xfs_dir3_leaf_check(mp, oldblk->bp);
1403         xfs_dir3_leaf_check(mp, newblk->bp);
1404         return error;
1405 }
1406
1407 /*
1408  * Check a leaf block and its neighbors to see if the block should be
1409  * collapsed into one or the other neighbor.  Always keep the block
1410  * with the smaller block number.
1411  * If the current block is over 50% full, don't try to join it, return 0.
1412  * If the block is empty, fill in the state structure and return 2.
1413  * If it can be collapsed, fill in the state structure and return 1.
1414  * If nothing can be done, return 0.
1415  */
1416 int                                             /* error */
1417 xfs_dir2_leafn_toosmall(
1418         xfs_da_state_t          *state,         /* btree cursor */
1419         int                     *action)        /* resulting action to take */
1420 {
1421         xfs_da_state_blk_t      *blk;           /* leaf block */
1422         xfs_dablk_t             blkno;          /* leaf block number */
1423         struct xfs_buf          *bp;            /* leaf buffer */
1424         int                     bytes;          /* bytes in use */
1425         int                     count;          /* leaf live entry count */
1426         int                     error;          /* error return value */
1427         int                     forward;        /* sibling block direction */
1428         int                     i;              /* sibling counter */
1429         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1430         int                     rval;           /* result from path_shift */
1431         struct xfs_dir3_icleaf_hdr leafhdr;
1432         struct xfs_dir2_leaf_entry *ents;
1433
1434         /*
1435          * Check for the degenerate case of the block being over 50% full.
1436          * If so, it's not worth even looking to see if we might be able
1437          * to coalesce with a sibling.
1438          */
1439         blk = &state->path.blk[state->path.active - 1];
1440         leaf = blk->bp->b_addr;
1441         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1442         ents = xfs_dir3_leaf_ents_p(leaf);
1443         xfs_dir3_leaf_check(state->args->dp->i_mount, blk->bp);
1444
1445         count = leafhdr.count - leafhdr.stale;
1446         bytes = xfs_dir3_leaf_hdr_size(leaf) + count * sizeof(ents[0]);
1447         if (bytes > (state->blocksize >> 1)) {
1448                 /*
1449                  * Blk over 50%, don't try to join.
1450                  */
1451                 *action = 0;
1452                 return 0;
1453         }
1454         /*
1455          * Check for the degenerate case of the block being empty.
1456          * If the block is empty, we'll simply delete it, no need to
1457          * coalesce it with a sibling block.  We choose (arbitrarily)
1458          * to merge with the forward block unless it is NULL.
1459          */
1460         if (count == 0) {
1461                 /*
1462                  * Make altpath point to the block we want to keep and
1463                  * path point to the block we want to drop (this one).
1464                  */
1465                 forward = (leafhdr.forw != 0);
1466                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1467                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1468                         &rval);
1469                 if (error)
1470                         return error;
1471                 *action = rval ? 2 : 0;
1472                 return 0;
1473         }
1474         /*
1475          * Examine each sibling block to see if we can coalesce with
1476          * at least 25% free space to spare.  We need to figure out
1477          * whether to merge with the forward or the backward block.
1478          * We prefer coalescing with the lower numbered sibling so as
1479          * to shrink a directory over time.
1480          */
1481         forward = leafhdr.forw < leafhdr.back;
1482         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1483                 struct xfs_dir3_icleaf_hdr hdr2;
1484
1485                 blkno = forward ? leafhdr.forw : leafhdr.back;
1486                 if (blkno == 0)
1487                         continue;
1488                 /*
1489                  * Read the sibling leaf block.
1490                  */
1491                 error = xfs_dir3_leafn_read(state->args->trans, state->args->dp,
1492                                             blkno, -1, &bp);
1493                 if (error)
1494                         return error;
1495
1496                 /*
1497                  * Count bytes in the two blocks combined.
1498                  */
1499                 count = leafhdr.count - leafhdr.stale;
1500                 bytes = state->blocksize - (state->blocksize >> 2);
1501
1502                 leaf = bp->b_addr;
1503                 xfs_dir3_leaf_hdr_from_disk(&hdr2, leaf);
1504                 ents = xfs_dir3_leaf_ents_p(leaf);
1505                 count += hdr2.count - hdr2.stale;
1506                 bytes -= count * sizeof(ents[0]);
1507
1508                 /*
1509                  * Fits with at least 25% to spare.
1510                  */
1511                 if (bytes >= 0)
1512                         break;
1513                 xfs_trans_brelse(state->args->trans, bp);
1514         }
1515         /*
1516          * Didn't like either block, give up.
1517          */
1518         if (i >= 2) {
1519                 *action = 0;
1520                 return 0;
1521         }
1522
1523         /*
1524          * Make altpath point to the block we want to keep (the lower
1525          * numbered block) and path point to the block we want to drop.
1526          */
1527         memcpy(&state->altpath, &state->path, sizeof(state->path));
1528         if (blkno < blk->blkno)
1529                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1530                         &rval);
1531         else
1532                 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1533                         &rval);
1534         if (error) {
1535                 return error;
1536         }
1537         *action = rval ? 0 : 1;
1538         return 0;
1539 }
1540
1541 /*
1542  * Move all the leaf entries from drop_blk to save_blk.
1543  * This is done as part of a join operation.
1544  */
1545 void
1546 xfs_dir2_leafn_unbalance(
1547         xfs_da_state_t          *state,         /* cursor */
1548         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1549         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1550 {
1551         xfs_da_args_t           *args;          /* operation arguments */
1552         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1553         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1554         struct xfs_dir3_icleaf_hdr savehdr;
1555         struct xfs_dir3_icleaf_hdr drophdr;
1556         struct xfs_dir2_leaf_entry *sents;
1557         struct xfs_dir2_leaf_entry *dents;
1558
1559         args = state->args;
1560         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1561         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1562         drop_leaf = drop_blk->bp->b_addr;
1563         save_leaf = save_blk->bp->b_addr;
1564
1565         xfs_dir3_leaf_hdr_from_disk(&savehdr, save_leaf);
1566         xfs_dir3_leaf_hdr_from_disk(&drophdr, drop_leaf);
1567         sents = xfs_dir3_leaf_ents_p(save_leaf);
1568         dents = xfs_dir3_leaf_ents_p(drop_leaf);
1569
1570         /*
1571          * If there are any stale leaf entries, take this opportunity
1572          * to purge them.
1573          */
1574         if (drophdr.stale)
1575                 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1576         if (savehdr.stale)
1577                 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1578
1579         /*
1580          * Move the entries from drop to the appropriate end of save.
1581          */
1582         drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1583         if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1584                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1585                                         save_blk->bp, &savehdr, sents, 0,
1586                                         drophdr.count);
1587         else
1588                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1589                                         save_blk->bp, &savehdr, sents,
1590                                         savehdr.count, drophdr.count);
1591         save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1592
1593         /* log the changes made when moving the entries */
1594         xfs_dir3_leaf_hdr_to_disk(save_leaf, &savehdr);
1595         xfs_dir3_leaf_hdr_to_disk(drop_leaf, &drophdr);
1596         xfs_dir3_leaf_log_header(args->trans, save_blk->bp);
1597         xfs_dir3_leaf_log_header(args->trans, drop_blk->bp);
1598
1599         xfs_dir3_leaf_check(args->dp->i_mount, save_blk->bp);
1600         xfs_dir3_leaf_check(args->dp->i_mount, drop_blk->bp);
1601 }
1602
1603 /*
1604  * Top-level node form directory addname routine.
1605  */
1606 int                                             /* error */
1607 xfs_dir2_node_addname(
1608         xfs_da_args_t           *args)          /* operation arguments */
1609 {
1610         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1611         int                     error;          /* error return value */
1612         int                     rval;           /* sub-return value */
1613         xfs_da_state_t          *state;         /* btree cursor */
1614
1615         trace_xfs_dir2_node_addname(args);
1616
1617         /*
1618          * Allocate and initialize the state (btree cursor).
1619          */
1620         state = xfs_da_state_alloc();
1621         state->args = args;
1622         state->mp = args->dp->i_mount;
1623         state->blocksize = state->mp->m_dirblksize;
1624         state->node_ents = state->mp->m_dir_node_ents;
1625         /*
1626          * Look up the name.  We're not supposed to find it, but
1627          * this gives us the insertion point.
1628          */
1629         error = xfs_da3_node_lookup_int(state, &rval);
1630         if (error)
1631                 rval = error;
1632         if (rval != ENOENT) {
1633                 goto done;
1634         }
1635         /*
1636          * Add the data entry to a data block.
1637          * Extravalid is set to a freeblock found by lookup.
1638          */
1639         rval = xfs_dir2_node_addname_int(args,
1640                 state->extravalid ? &state->extrablk : NULL);
1641         if (rval) {
1642                 goto done;
1643         }
1644         blk = &state->path.blk[state->path.active - 1];
1645         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1646         /*
1647          * Add the new leaf entry.
1648          */
1649         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1650         if (rval == 0) {
1651                 /*
1652                  * It worked, fix the hash values up the btree.
1653                  */
1654                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1655                         xfs_da3_fixhashpath(state, &state->path);
1656         } else {
1657                 /*
1658                  * It didn't work, we need to split the leaf block.
1659                  */
1660                 if (args->total == 0) {
1661                         ASSERT(rval == ENOSPC);
1662                         goto done;
1663                 }
1664                 /*
1665                  * Split the leaf block and insert the new entry.
1666                  */
1667                 rval = xfs_da3_split(state);
1668         }
1669 done:
1670         xfs_da_state_free(state);
1671         return rval;
1672 }
1673
1674 /*
1675  * Add the data entry for a node-format directory name addition.
1676  * The leaf entry is added in xfs_dir2_leafn_add.
1677  * We may enter with a freespace block that the lookup found.
1678  */
1679 static int                                      /* error */
1680 xfs_dir2_node_addname_int(
1681         xfs_da_args_t           *args,          /* operation arguments */
1682         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1683 {
1684         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1685         xfs_dir2_db_t           dbno;           /* data block number */
1686         struct xfs_buf          *dbp;           /* data block buffer */
1687         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1688         xfs_inode_t             *dp;            /* incore directory inode */
1689         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1690         int                     error;          /* error return value */
1691         xfs_dir2_db_t           fbno;           /* freespace block number */
1692         struct xfs_buf          *fbp;           /* freespace buffer */
1693         int                     findex;         /* freespace entry index */
1694         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1695         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1696         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1697         int                     length;         /* length of the new entry */
1698         int                     logfree;        /* need to log free entry */
1699         xfs_mount_t             *mp;            /* filesystem mount point */
1700         int                     needlog;        /* need to log data header */
1701         int                     needscan;       /* need to rescan data frees */
1702         __be16                  *tagp;          /* data entry tag pointer */
1703         xfs_trans_t             *tp;            /* transaction pointer */
1704         __be16                  *bests;
1705         struct xfs_dir3_icfree_hdr freehdr;
1706         struct xfs_dir2_data_free *bf;
1707
1708         dp = args->dp;
1709         mp = dp->i_mount;
1710         tp = args->trans;
1711         length = xfs_dir2_data_entsize(args->namelen);
1712         /*
1713          * If we came in with a freespace block that means that lookup
1714          * found an entry with our hash value.  This is the freespace
1715          * block for that data entry.
1716          */
1717         if (fblk) {
1718                 fbp = fblk->bp;
1719                 /*
1720                  * Remember initial freespace block number.
1721                  */
1722                 ifbno = fblk->blkno;
1723                 free = fbp->b_addr;
1724                 findex = fblk->index;
1725                 bests = xfs_dir3_free_bests_p(mp, free);
1726                 xfs_dir3_free_hdr_from_disk(&freehdr, free);
1727
1728                 /*
1729                  * This means the free entry showed that the data block had
1730                  * space for our entry, so we remembered it.
1731                  * Use that data block.
1732                  */
1733                 if (findex >= 0) {
1734                         ASSERT(findex < freehdr.nvalid);
1735                         ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1736                         ASSERT(be16_to_cpu(bests[findex]) >= length);
1737                         dbno = freehdr.firstdb + findex;
1738                 } else {
1739                         /*
1740                          * The data block looked at didn't have enough room.
1741                          * We'll start at the beginning of the freespace entries.
1742                          */
1743                         dbno = -1;
1744                         findex = 0;
1745                 }
1746         } else {
1747                 /*
1748                  * Didn't come in with a freespace block, so no data block.
1749                  */
1750                 ifbno = dbno = -1;
1751                 fbp = NULL;
1752                 findex = 0;
1753         }
1754
1755         /*
1756          * If we don't have a data block yet, we're going to scan the
1757          * freespace blocks looking for one.  Figure out what the
1758          * highest freespace block number is.
1759          */
1760         if (dbno == -1) {
1761                 xfs_fileoff_t   fo;             /* freespace block number */
1762
1763                 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1764                         return error;
1765                 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1766                 fbno = ifbno;
1767         }
1768         /*
1769          * While we haven't identified a data block, search the freeblock
1770          * data for a good data block.  If we find a null freeblock entry,
1771          * indicating a hole in the data blocks, remember that.
1772          */
1773         while (dbno == -1) {
1774                 /*
1775                  * If we don't have a freeblock in hand, get the next one.
1776                  */
1777                 if (fbp == NULL) {
1778                         /*
1779                          * Happens the first time through unless lookup gave
1780                          * us a freespace block to start with.
1781                          */
1782                         if (++fbno == 0)
1783                                 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1784                         /*
1785                          * If it's ifbno we already looked at it.
1786                          */
1787                         if (fbno == ifbno)
1788                                 fbno++;
1789                         /*
1790                          * If it's off the end we're done.
1791                          */
1792                         if (fbno >= lastfbno)
1793                                 break;
1794                         /*
1795                          * Read the block.  There can be holes in the
1796                          * freespace blocks, so this might not succeed.
1797                          * This should be really rare, so there's no reason
1798                          * to avoid it.
1799                          */
1800                         error = xfs_dir2_free_try_read(tp, dp,
1801                                                 xfs_dir2_db_to_da(mp, fbno),
1802                                                 &fbp);
1803                         if (error)
1804                                 return error;
1805                         if (!fbp)
1806                                 continue;
1807                         free = fbp->b_addr;
1808                         findex = 0;
1809                 }
1810                 /*
1811                  * Look at the current free entry.  Is it good enough?
1812                  *
1813                  * The bests initialisation should be where the bufer is read in
1814                  * the above branch. But gcc is too stupid to realise that bests
1815                  * and the freehdr are actually initialised if they are placed
1816                  * there, so we have to do it here to avoid warnings. Blech.
1817                  */
1818                 bests = xfs_dir3_free_bests_p(mp, free);
1819                 xfs_dir3_free_hdr_from_disk(&freehdr, free);
1820                 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1821                     be16_to_cpu(bests[findex]) >= length)
1822                         dbno = freehdr.firstdb + findex;
1823                 else {
1824                         /*
1825                          * Are we done with the freeblock?
1826                          */
1827                         if (++findex == freehdr.nvalid) {
1828                                 /*
1829                                  * Drop the block.
1830                                  */
1831                                 xfs_trans_brelse(tp, fbp);
1832                                 fbp = NULL;
1833                                 if (fblk && fblk->bp)
1834                                         fblk->bp = NULL;
1835                         }
1836                 }
1837         }
1838         /*
1839          * If we don't have a data block, we need to allocate one and make
1840          * the freespace entries refer to it.
1841          */
1842         if (unlikely(dbno == -1)) {
1843                 /*
1844                  * Not allowed to allocate, return failure.
1845                  */
1846                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1847                         return XFS_ERROR(ENOSPC);
1848
1849                 /*
1850                  * Allocate and initialize the new data block.
1851                  */
1852                 if (unlikely((error = xfs_dir2_grow_inode(args,
1853                                                          XFS_DIR2_DATA_SPACE,
1854                                                          &dbno)) ||
1855                     (error = xfs_dir3_data_init(args, dbno, &dbp))))
1856                         return error;
1857
1858                 /*
1859                  * If (somehow) we have a freespace block, get rid of it.
1860                  */
1861                 if (fbp)
1862                         xfs_trans_brelse(tp, fbp);
1863                 if (fblk && fblk->bp)
1864                         fblk->bp = NULL;
1865
1866                 /*
1867                  * Get the freespace block corresponding to the data block
1868                  * that was just allocated.
1869                  */
1870                 fbno = xfs_dir2_db_to_fdb(mp, dbno);
1871                 error = xfs_dir2_free_try_read(tp, dp,
1872                                                xfs_dir2_db_to_da(mp, fbno),
1873                                                &fbp);
1874                 if (error)
1875                         return error;
1876
1877                 /*
1878                  * If there wasn't a freespace block, the read will
1879                  * return a NULL fbp.  Allocate and initialize a new one.
1880                  */
1881                 if (!fbp) {
1882                         error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1883                                                     &fbno);
1884                         if (error)
1885                                 return error;
1886
1887                         if (unlikely(xfs_dir2_db_to_fdb(mp, dbno) != fbno)) {
1888                                 xfs_alert(mp,
1889                         "%s: dir ino %llu needed freesp block %lld for\n"
1890                         "  data block %lld, got %lld ifbno %llu lastfbno %d",
1891                                         __func__, (unsigned long long)dp->i_ino,
1892                                         (long long)xfs_dir2_db_to_fdb(mp, dbno),
1893                                         (long long)dbno, (long long)fbno,
1894                                         (unsigned long long)ifbno, lastfbno);
1895                                 if (fblk) {
1896                                         xfs_alert(mp,
1897                                 " fblk 0x%p blkno %llu index %d magic 0x%x",
1898                                                 fblk,
1899                                                 (unsigned long long)fblk->blkno,
1900                                                 fblk->index,
1901                                                 fblk->magic);
1902                                 } else {
1903                                         xfs_alert(mp, " ... fblk is NULL");
1904                                 }
1905                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1906                                                  XFS_ERRLEVEL_LOW, mp);
1907                                 return XFS_ERROR(EFSCORRUPTED);
1908                         }
1909
1910                         /*
1911                          * Get a buffer for the new block.
1912                          */
1913                         error = xfs_dir3_free_get_buf(tp, dp, fbno, &fbp);
1914                         if (error)
1915                                 return error;
1916                         free = fbp->b_addr;
1917                         bests = xfs_dir3_free_bests_p(mp, free);
1918                         xfs_dir3_free_hdr_from_disk(&freehdr, free);
1919
1920                         /*
1921                          * Remember the first slot as our empty slot.
1922                          */
1923                         freehdr.firstdb = (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1924                                         xfs_dir3_free_max_bests(mp);
1925                 } else {
1926                         free = fbp->b_addr;
1927                         bests = xfs_dir3_free_bests_p(mp, free);
1928                         xfs_dir3_free_hdr_from_disk(&freehdr, free);
1929                 }
1930
1931                 /*
1932                  * Set the freespace block index from the data block number.
1933                  */
1934                 findex = xfs_dir2_db_to_fdindex(mp, dbno);
1935                 /*
1936                  * If it's after the end of the current entries in the
1937                  * freespace block, extend that table.
1938                  */
1939                 if (findex >= freehdr.nvalid) {
1940                         ASSERT(findex < xfs_dir3_free_max_bests(mp));
1941                         freehdr.nvalid = findex + 1;
1942                         /*
1943                          * Tag new entry so nused will go up.
1944                          */
1945                         bests[findex] = cpu_to_be16(NULLDATAOFF);
1946                 }
1947                 /*
1948                  * If this entry was for an empty data block
1949                  * (this should always be true) then update the header.
1950                  */
1951                 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1952                         freehdr.nused++;
1953                         xfs_dir3_free_hdr_to_disk(fbp->b_addr, &freehdr);
1954                         xfs_dir2_free_log_header(tp, fbp);
1955                 }
1956                 /*
1957                  * Update the real value in the table.
1958                  * We haven't allocated the data entry yet so this will
1959                  * change again.
1960                  */
1961                 hdr = dbp->b_addr;
1962                 bf = xfs_dir3_data_bestfree_p(hdr);
1963                 bests[findex] = bf[0].length;
1964                 logfree = 1;
1965         }
1966         /*
1967          * We had a data block so we don't have to make a new one.
1968          */
1969         else {
1970                 /*
1971                  * If just checking, we succeeded.
1972                  */
1973                 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1974                         return 0;
1975
1976                 /*
1977                  * Read the data block in.
1978                  */
1979                 error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1980                                            -1, &dbp);
1981                 if (error)
1982                         return error;
1983                 hdr = dbp->b_addr;
1984                 bf = xfs_dir3_data_bestfree_p(hdr);
1985                 logfree = 0;
1986         }
1987         ASSERT(be16_to_cpu(bf[0].length) >= length);
1988         /*
1989          * Point to the existing unused space.
1990          */
1991         dup = (xfs_dir2_data_unused_t *)
1992               ((char *)hdr + be16_to_cpu(bf[0].offset));
1993         needscan = needlog = 0;
1994         /*
1995          * Mark the first part of the unused space, inuse for us.
1996          */
1997         xfs_dir2_data_use_free(tp, dbp, dup,
1998                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
1999                 &needlog, &needscan);
2000         /*
2001          * Fill in the new entry and log it.
2002          */
2003         dep = (xfs_dir2_data_entry_t *)dup;
2004         dep->inumber = cpu_to_be64(args->inumber);
2005         dep->namelen = args->namelen;
2006         memcpy(dep->name, args->name, dep->namelen);
2007         tagp = xfs_dir2_data_entry_tag_p(dep);
2008         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
2009         xfs_dir2_data_log_entry(tp, dbp, dep);
2010         /*
2011          * Rescan the block for bestfree if needed.
2012          */
2013         if (needscan)
2014                 xfs_dir2_data_freescan(mp, hdr, &needlog);
2015         /*
2016          * Log the data block header if needed.
2017          */
2018         if (needlog)
2019                 xfs_dir2_data_log_header(tp, dbp);
2020         /*
2021          * If the freespace entry is now wrong, update it.
2022          */
2023         bests = xfs_dir3_free_bests_p(mp, free); /* gcc is so stupid */
2024         if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
2025                 bests[findex] = bf[0].length;
2026                 logfree = 1;
2027         }
2028         /*
2029          * Log the freespace entry if needed.
2030          */
2031         if (logfree)
2032                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
2033         /*
2034          * Return the data block and offset in args, then drop the data block.
2035          */
2036         args->blkno = (xfs_dablk_t)dbno;
2037         args->index = be16_to_cpu(*tagp);
2038         return 0;
2039 }
2040
2041 /*
2042  * Lookup an entry in a node-format directory.
2043  * All the real work happens in xfs_da3_node_lookup_int.
2044  * The only real output is the inode number of the entry.
2045  */
2046 int                                             /* error */
2047 xfs_dir2_node_lookup(
2048         xfs_da_args_t   *args)                  /* operation arguments */
2049 {
2050         int             error;                  /* error return value */
2051         int             i;                      /* btree level */
2052         int             rval;                   /* operation return value */
2053         xfs_da_state_t  *state;                 /* btree cursor */
2054
2055         trace_xfs_dir2_node_lookup(args);
2056
2057         /*
2058          * Allocate and initialize the btree cursor.
2059          */
2060         state = xfs_da_state_alloc();
2061         state->args = args;
2062         state->mp = args->dp->i_mount;
2063         state->blocksize = state->mp->m_dirblksize;
2064         state->node_ents = state->mp->m_dir_node_ents;
2065         /*
2066          * Fill in the path to the entry in the cursor.
2067          */
2068         error = xfs_da3_node_lookup_int(state, &rval);
2069         if (error)
2070                 rval = error;
2071         else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
2072                 /* If a CI match, dup the actual name and return EEXIST */
2073                 xfs_dir2_data_entry_t   *dep;
2074
2075                 dep = (xfs_dir2_data_entry_t *)
2076                         ((char *)state->extrablk.bp->b_addr +
2077                                                  state->extrablk.index);
2078                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2079         }
2080         /*
2081          * Release the btree blocks and leaf block.
2082          */
2083         for (i = 0; i < state->path.active; i++) {
2084                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2085                 state->path.blk[i].bp = NULL;
2086         }
2087         /*
2088          * Release the data block if we have it.
2089          */
2090         if (state->extravalid && state->extrablk.bp) {
2091                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2092                 state->extrablk.bp = NULL;
2093         }
2094         xfs_da_state_free(state);
2095         return rval;
2096 }
2097
2098 /*
2099  * Remove an entry from a node-format directory.
2100  */
2101 int                                             /* error */
2102 xfs_dir2_node_removename(
2103         xfs_da_args_t           *args)          /* operation arguments */
2104 {
2105         xfs_da_state_blk_t      *blk;           /* leaf block */
2106         int                     error;          /* error return value */
2107         int                     rval;           /* operation return value */
2108         xfs_da_state_t          *state;         /* btree cursor */
2109
2110         trace_xfs_dir2_node_removename(args);
2111
2112         /*
2113          * Allocate and initialize the btree cursor.
2114          */
2115         state = xfs_da_state_alloc();
2116         state->args = args;
2117         state->mp = args->dp->i_mount;
2118         state->blocksize = state->mp->m_dirblksize;
2119         state->node_ents = state->mp->m_dir_node_ents;
2120         /*
2121          * Look up the entry we're deleting, set up the cursor.
2122          */
2123         error = xfs_da3_node_lookup_int(state, &rval);
2124         if (error)
2125                 rval = error;
2126         /*
2127          * Didn't find it, upper layer screwed up.
2128          */
2129         if (rval != EEXIST) {
2130                 xfs_da_state_free(state);
2131                 return rval;
2132         }
2133         blk = &state->path.blk[state->path.active - 1];
2134         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2135         ASSERT(state->extravalid);
2136         /*
2137          * Remove the leaf and data entries.
2138          * Extrablk refers to the data block.
2139          */
2140         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2141                 &state->extrablk, &rval);
2142         if (error)
2143                 return error;
2144         /*
2145          * Fix the hash values up the btree.
2146          */
2147         xfs_da3_fixhashpath(state, &state->path);
2148         /*
2149          * If we need to join leaf blocks, do it.
2150          */
2151         if (rval && state->path.active > 1)
2152                 error = xfs_da3_join(state);
2153         /*
2154          * If no errors so far, try conversion to leaf format.
2155          */
2156         if (!error)
2157                 error = xfs_dir2_node_to_leaf(state);
2158         xfs_da_state_free(state);
2159         return error;
2160 }
2161
2162 /*
2163  * Replace an entry's inode number in a node-format directory.
2164  */
2165 int                                             /* error */
2166 xfs_dir2_node_replace(
2167         xfs_da_args_t           *args)          /* operation arguments */
2168 {
2169         xfs_da_state_blk_t      *blk;           /* leaf block */
2170         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
2171         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
2172         int                     error;          /* error return value */
2173         int                     i;              /* btree level */
2174         xfs_ino_t               inum;           /* new inode number */
2175         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
2176         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
2177         int                     rval;           /* internal return value */
2178         xfs_da_state_t          *state;         /* btree cursor */
2179
2180         trace_xfs_dir2_node_replace(args);
2181
2182         /*
2183          * Allocate and initialize the btree cursor.
2184          */
2185         state = xfs_da_state_alloc();
2186         state->args = args;
2187         state->mp = args->dp->i_mount;
2188         state->blocksize = state->mp->m_dirblksize;
2189         state->node_ents = state->mp->m_dir_node_ents;
2190         inum = args->inumber;
2191         /*
2192          * Lookup the entry to change in the btree.
2193          */
2194         error = xfs_da3_node_lookup_int(state, &rval);
2195         if (error) {
2196                 rval = error;
2197         }
2198         /*
2199          * It should be found, since the vnodeops layer has looked it up
2200          * and locked it.  But paranoia is good.
2201          */
2202         if (rval == EEXIST) {
2203                 struct xfs_dir2_leaf_entry *ents;
2204                 /*
2205                  * Find the leaf entry.
2206                  */
2207                 blk = &state->path.blk[state->path.active - 1];
2208                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2209                 leaf = blk->bp->b_addr;
2210                 ents = xfs_dir3_leaf_ents_p(leaf);
2211                 lep = &ents[blk->index];
2212                 ASSERT(state->extravalid);
2213                 /*
2214                  * Point to the data entry.
2215                  */
2216                 hdr = state->extrablk.bp->b_addr;
2217                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2218                        hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2219                 dep = (xfs_dir2_data_entry_t *)
2220                       ((char *)hdr +
2221                        xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
2222                 ASSERT(inum != be64_to_cpu(dep->inumber));
2223                 /*
2224                  * Fill in the new inode number and log the entry.
2225                  */
2226                 dep->inumber = cpu_to_be64(inum);
2227                 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
2228                 rval = 0;
2229         }
2230         /*
2231          * Didn't find it, and we're holding a data block.  Drop it.
2232          */
2233         else if (state->extravalid) {
2234                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2235                 state->extrablk.bp = NULL;
2236         }
2237         /*
2238          * Release all the buffers in the cursor.
2239          */
2240         for (i = 0; i < state->path.active; i++) {
2241                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2242                 state->path.blk[i].bp = NULL;
2243         }
2244         xfs_da_state_free(state);
2245         return rval;
2246 }
2247
2248 /*
2249  * Trim off a trailing empty freespace block.
2250  * Return (in rvalp) 1 if we did it, 0 if not.
2251  */
2252 int                                             /* error */
2253 xfs_dir2_node_trim_free(
2254         xfs_da_args_t           *args,          /* operation arguments */
2255         xfs_fileoff_t           fo,             /* free block number */
2256         int                     *rvalp)         /* out: did something */
2257 {
2258         struct xfs_buf          *bp;            /* freespace buffer */
2259         xfs_inode_t             *dp;            /* incore directory inode */
2260         int                     error;          /* error return code */
2261         xfs_dir2_free_t         *free;          /* freespace structure */
2262         xfs_mount_t             *mp;            /* filesystem mount point */
2263         xfs_trans_t             *tp;            /* transaction pointer */
2264         struct xfs_dir3_icfree_hdr freehdr;
2265
2266         dp = args->dp;
2267         mp = dp->i_mount;
2268         tp = args->trans;
2269         /*
2270          * Read the freespace block.
2271          */
2272         error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2273         if (error)
2274                 return error;
2275         /*
2276          * There can be holes in freespace.  If fo is a hole, there's
2277          * nothing to do.
2278          */
2279         if (!bp)
2280                 return 0;
2281         free = bp->b_addr;
2282         xfs_dir3_free_hdr_from_disk(&freehdr, free);
2283
2284         /*
2285          * If there are used entries, there's nothing to do.
2286          */
2287         if (freehdr.nused > 0) {
2288                 xfs_trans_brelse(tp, bp);
2289                 *rvalp = 0;
2290                 return 0;
2291         }
2292         /*
2293          * Blow the block away.
2294          */
2295         if ((error =
2296             xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
2297                     bp))) {
2298                 /*
2299                  * Can't fail with ENOSPC since that only happens with no
2300                  * space reservation, when breaking up an extent into two
2301                  * pieces.  This is the last block of an extent.
2302                  */
2303                 ASSERT(error != ENOSPC);
2304                 xfs_trans_brelse(tp, bp);
2305                 return error;
2306         }
2307         /*
2308          * Return that we succeeded.
2309          */
2310         *rvalp = 1;
2311         return 0;
2312 }