fs/xfs/xfs_dir2_data.c

   1 /*
   2  * Copyright (c) 2000-2002,2005 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_dir2_format.h"
  32 #include "xfs_dir2_priv.h"
  33 #include "xfs_error.h"
  34 #include "xfs_buf_item.h"
  35 #include "xfs_cksum.h"
  36
  37 STATIC xfs_dir2_data_free_t *
  38 xfs_dir2_data_freefind(xfs_dir2_data_hdr_t *hdr, xfs_dir2_data_unused_t *dup);
  39
  40 /*
  41  * Check the consistency of the data block.
  42  * The input can also be a block-format directory.
  43  * Return 0 is the buffer is good, otherwise an error.
  44  */
  45 int
  46 __xfs_dir3_data_check(
  47         struct xfs_inode        *dp,            /* incore inode pointer */
  48         struct xfs_buf          *bp)            /* data block's buffer */
  49 {
  50         xfs_dir2_dataptr_t      addr;           /* addr for leaf lookup */
  51         xfs_dir2_data_free_t    *bf;            /* bestfree table */
  52         xfs_dir2_block_tail_t   *btp=NULL;      /* block tail */
  53         int                     count;          /* count of entries found */
  54         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
  55         xfs_dir2_data_entry_t   *dep;           /* data entry */
  56         xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
  57         xfs_dir2_data_unused_t  *dup;           /* unused entry */
  58         char                    *endp;          /* end of useful data */
  59         int                     freeseen;       /* mask of bestfrees seen */
  60         xfs_dahash_t            hash;           /* hash of current name */
  61         int                     i;              /* leaf index */
  62         int                     lastfree;       /* last entry was unused */
  63         xfs_dir2_leaf_entry_t   *lep=NULL;      /* block leaf entries */
  64         xfs_mount_t             *mp;            /* filesystem mount point */
  65         char                    *p;             /* current data position */
  66         int                     stale;          /* count of stale leaves */
  67         struct xfs_name         name;
  68
  69         mp = bp->b_target->bt_mount;
  70         hdr = bp->b_addr;
  71         bf = xfs_dir3_data_bestfree_p(hdr);
  72         p = (char *)xfs_dir3_data_entry_p(hdr);
  73
  74         switch (hdr->magic) {
  75         case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
  76         case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
  77                 btp = xfs_dir2_block_tail_p(mp, hdr);
  78                 lep = xfs_dir2_block_leaf_p(btp);
  79                 endp = (char *)lep;
  80                 break;
  81         case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
  82         case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
  83                 endp = (char *)hdr + mp->m_dirblksize;
  84                 break;
  85         default:
  86                 XFS_ERROR_REPORT("Bad Magic", XFS_ERRLEVEL_LOW, mp);
  87                 return EFSCORRUPTED;
  88         }
  89
  90         count = lastfree = freeseen = 0;
  91         /*
  92          * Account for zero bestfree entries.
  93          */
  94         if (!bf[0].length) {
  95                 XFS_WANT_CORRUPTED_RETURN(!bf[0].offset);
  96                 freeseen |= 1 << 0;
  97         }
  98         if (!bf[1].length) {
  99                 XFS_WANT_CORRUPTED_RETURN(!bf[1].offset);
 100                 freeseen |= 1 << 1;
 101         }
 102         if (!bf[2].length) {
 103                 XFS_WANT_CORRUPTED_RETURN(!bf[2].offset);
 104                 freeseen |= 1 << 2;
 105         }
 106
 107         XFS_WANT_CORRUPTED_RETURN(be16_to_cpu(bf[0].length) >=
 108                                                 be16_to_cpu(bf[1].length));
 109         XFS_WANT_CORRUPTED_RETURN(be16_to_cpu(bf[1].length) >=
 110                                                 be16_to_cpu(bf[2].length));
 111         /*
 112          * Loop over the data/unused entries.
 113          */
 114         while (p < endp) {
 115                 dup = (xfs_dir2_data_unused_t *)p;
 116                 /*
 117                  * If it's unused, look for the space in the bestfree table.
 118                  * If we find it, account for that, else make sure it
 119                  * doesn't need to be there.
 120                  */
 121                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 122                         XFS_WANT_CORRUPTED_RETURN(lastfree == 0);
 123                         XFS_WANT_CORRUPTED_RETURN(
 124                                 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) ==
 125                                                (char *)dup - (char *)hdr);
 126                         dfp = xfs_dir2_data_freefind(hdr, dup);
 127                         if (dfp) {
 128                                 i = (int)(dfp - bf);
 129                                 XFS_WANT_CORRUPTED_RETURN(
 130                                         (freeseen & (1 << i)) == 0);
 131                                 freeseen |= 1 << i;
 132                         } else {
 133                                 XFS_WANT_CORRUPTED_RETURN(
 134                                         be16_to_cpu(dup->length) <=
 135                                                 be16_to_cpu(bf[2].length));
 136                         }
 137                         p += be16_to_cpu(dup->length);
 138                         lastfree = 1;
 139                         continue;
 140                 }
 141                 /*
 142                  * It's a real entry.  Validate the fields.
 143                  * If this is a block directory then make sure it's
 144                  * in the leaf section of the block.
 145                  * The linear search is crude but this is DEBUG code.
 146                  */
 147                 dep = (xfs_dir2_data_entry_t *)p;
 148                 XFS_WANT_CORRUPTED_RETURN(dep->namelen != 0);
 149                 XFS_WANT_CORRUPTED_RETURN(
 150                         !xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber)));
 151                 XFS_WANT_CORRUPTED_RETURN(
 152                         be16_to_cpu(*xfs_dir2_data_entry_tag_p(dep)) ==
 153                                                (char *)dep - (char *)hdr);
 154                 count++;
 155                 lastfree = 0;
 156                 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 157                     hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 158                         addr = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
 159                                 (xfs_dir2_data_aoff_t)
 160                                 ((char *)dep - (char *)hdr));
 161                         name.name = dep->name;
 162                         name.len = dep->namelen;
 163                         hash = mp->m_dirnameops->hashname(&name);
 164                         for (i = 0; i < be32_to_cpu(btp->count); i++) {
 165                                 if (be32_to_cpu(lep[i].address) == addr &&
 166                                     be32_to_cpu(lep[i].hashval) == hash)
 167                                         break;
 168                         }
 169                         XFS_WANT_CORRUPTED_RETURN(i < be32_to_cpu(btp->count));
 170                 }
 171                 p += xfs_dir2_data_entsize(dep->namelen);
 172         }
 173         /*
 174          * Need to have seen all the entries and all the bestfree slots.
 175          */
 176         XFS_WANT_CORRUPTED_RETURN(freeseen == 7);
 177         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 178             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 179                 for (i = stale = 0; i < be32_to_cpu(btp->count); i++) {
 180                         if (lep[i].address ==
 181                             cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 182                                 stale++;
 183                         if (i > 0)
 184                                 XFS_WANT_CORRUPTED_RETURN(
 185                                         be32_to_cpu(lep[i].hashval) >=
 186                                                 be32_to_cpu(lep[i - 1].hashval));
 187                 }
 188                 XFS_WANT_CORRUPTED_RETURN(count ==
 189                         be32_to_cpu(btp->count) - be32_to_cpu(btp->stale));
 190                 XFS_WANT_CORRUPTED_RETURN(stale == be32_to_cpu(btp->stale));
 191         }
 192         return 0;
 193 }
 194
 195 static bool
 196 xfs_dir3_data_verify(
 197         struct xfs_buf          *bp)
 198 {
 199         struct xfs_mount        *mp = bp->b_target->bt_mount;
 200         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 201
 202         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 203                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 204                         return false;
 205                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
 206                         return false;
 207                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 208                         return false;
 209         } else {
 210                 if (hdr3->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC))
 211                         return false;
 212         }
 213         if (__xfs_dir3_data_check(NULL, bp))
 214                 return false;
 215         return true;
 216 }
 217
 218 /*
 219  * Readahead of the first block of the directory when it is opened is completely
 220  * oblivious to the format of the directory. Hence we can either get a block
 221  * format buffer or a data format buffer on readahead.
 222  */
 223 static void
 224 xfs_dir3_data_reada_verify(
 225         struct xfs_buf          *bp)
 226 {
 227         struct xfs_mount        *mp = bp->b_target->bt_mount;
 228         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 229
 230         switch (hdr->magic) {
 231         case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
 232         case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
 233                 bp->b_ops = &xfs_dir3_block_buf_ops;
 234                 bp->b_ops->verify_read(bp);
 235                 return;
 236         case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
 237         case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
 238                 xfs_dir3_data_verify(bp);
 239                 return;
 240         default:
 241                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, hdr);
 242                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 243                 break;
 244         }
 245 }
 246
 247 static void
 248 xfs_dir3_data_read_verify(
 249         struct xfs_buf  *bp)
 250 {
 251         struct xfs_mount        *mp = bp->b_target->bt_mount;
 252
 253         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
 254              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
 255                                           XFS_DIR3_DATA_CRC_OFF)) ||
 256             !xfs_dir3_data_verify(bp)) {
 257                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 258                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 259         }
 260 }
 261
 262 static void
 263 xfs_dir3_data_write_verify(
 264         struct xfs_buf  *bp)
 265 {
 266         struct xfs_mount        *mp = bp->b_target->bt_mount;
 267         struct xfs_buf_log_item *bip = bp->b_fspriv;
 268         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 269
 270         if (!xfs_dir3_data_verify(bp)) {
 271                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 272                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 273                 return;
 274         }
 275
 276         if (!xfs_sb_version_hascrc(&mp->m_sb))
 277                 return;
 278
 279         if (bip)
 280                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 281
 282         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_DATA_CRC_OFF);
 283 }
 284
 285 const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
 286         .verify_read = xfs_dir3_data_read_verify,
 287         .verify_write = xfs_dir3_data_write_verify,
 288 };
 289
 290 static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
 291         .verify_read = xfs_dir3_data_reada_verify,
 292         .verify_write = xfs_dir3_data_write_verify,
 293 };
 294
 295
 296 int
 297 xfs_dir3_data_read(
 298         struct xfs_trans        *tp,
 299         struct xfs_inode        *dp,
 300         xfs_dablk_t             bno,
 301         xfs_daddr_t             mapped_bno,
 302         struct xfs_buf          **bpp)
 303 {
 304         int                     err;
 305
 306         err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
 307                                 XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
 308         if (!err && tp)
 309                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
 310         return err;
 311 }
 312
 313 int
 314 xfs_dir3_data_readahead(
 315         struct xfs_trans        *tp,
 316         struct xfs_inode        *dp,
 317         xfs_dablk_t             bno,
 318         xfs_daddr_t             mapped_bno)
 319 {
 320         return xfs_da_reada_buf(tp, dp, bno, mapped_bno,
 321                                 XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
 322 }
 323
 324 /*
 325  * Given a data block and an unused entry from that block,
 326  * return the bestfree entry if any that corresponds to it.
 327  */
 328 STATIC xfs_dir2_data_free_t *
 329 xfs_dir2_data_freefind(
 330         xfs_dir2_data_hdr_t     *hdr,           /* data block */
 331         xfs_dir2_data_unused_t  *dup)           /* data unused entry */
 332 {
 333         xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
 334         xfs_dir2_data_aoff_t    off;            /* offset value needed */
 335         struct xfs_dir2_data_free *bf;
 336 #if defined(DEBUG) && defined(__KERNEL__)
 337         int                     matched;        /* matched the value */
 338         int                     seenzero;       /* saw a 0 bestfree entry */
 339 #endif
 340
 341         off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
 342         bf = xfs_dir3_data_bestfree_p(hdr);
 343
 344 #if defined(DEBUG) && defined(__KERNEL__)
 345         /*
 346          * Validate some consistency in the bestfree table.
 347          * Check order, non-overlapping entries, and if we find the
 348          * one we're looking for it has to be exact.
 349          */
 350         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 351                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 352                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 353                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 354         for (dfp = &bf[0], seenzero = matched = 0;
 355              dfp < &bf[XFS_DIR2_DATA_FD_COUNT];
 356              dfp++) {
 357                 if (!dfp->offset) {
 358                         ASSERT(!dfp->length);
 359                         seenzero = 1;
 360                         continue;
 361                 }
 362                 ASSERT(seenzero == 0);
 363                 if (be16_to_cpu(dfp->offset) == off) {
 364                         matched = 1;
 365                         ASSERT(dfp->length == dup->length);
 366                 } else if (off < be16_to_cpu(dfp->offset))
 367                         ASSERT(off + be16_to_cpu(dup->length) <= be16_to_cpu(dfp->offset));
 368                 else
 369                         ASSERT(be16_to_cpu(dfp->offset) + be16_to_cpu(dfp->length) <= off);
 370                 ASSERT(matched || be16_to_cpu(dfp->length) >= be16_to_cpu(dup->length));
 371                 if (dfp > &bf[0])
 372                         ASSERT(be16_to_cpu(dfp[-1].length) >= be16_to_cpu(dfp[0].length));
 373         }
 374 #endif
 375         /*
 376          * If this is smaller than the smallest bestfree entry,
 377          * it can't be there since they're sorted.
 378          */
 379         if (be16_to_cpu(dup->length) <
 380             be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 381                 return NULL;
 382         /*
 383          * Look at the three bestfree entries for our guy.
 384          */
 385         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 386                 if (!dfp->offset)
 387                         return NULL;
 388                 if (be16_to_cpu(dfp->offset) == off)
 389                         return dfp;
 390         }
 391         /*
 392          * Didn't find it.  This only happens if there are duplicate lengths.
 393          */
 394         return NULL;
 395 }
 396
 397 /*
 398  * Insert an unused-space entry into the bestfree table.
 399  */
 400 xfs_dir2_data_free_t *                          /* entry inserted */
 401 xfs_dir2_data_freeinsert(
 402         xfs_dir2_data_hdr_t     *hdr,           /* data block pointer */
 403         xfs_dir2_data_unused_t  *dup,           /* unused space */
 404         int                     *loghead)       /* log the data header (out) */
 405 {
 406         xfs_dir2_data_free_t    *dfp;           /* bestfree table pointer */
 407         xfs_dir2_data_free_t    new;            /* new bestfree entry */
 408
 409         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 410                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 411                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 412                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 413
 414         dfp = xfs_dir3_data_bestfree_p(hdr);
 415         new.length = dup->length;
 416         new.offset = cpu_to_be16((char *)dup - (char *)hdr);
 417
 418         /*
 419          * Insert at position 0, 1, or 2; or not at all.
 420          */
 421         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
 422                 dfp[2] = dfp[1];
 423                 dfp[1] = dfp[0];
 424                 dfp[0] = new;
 425                 *loghead = 1;
 426                 return &dfp[0];
 427         }
 428         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
 429                 dfp[2] = dfp[1];
 430                 dfp[1] = new;
 431                 *loghead = 1;
 432                 return &dfp[1];
 433         }
 434         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
 435                 dfp[2] = new;
 436                 *loghead = 1;
 437                 return &dfp[2];
 438         }
 439         return NULL;
 440 }
 441
 442 /*
 443  * Remove a bestfree entry from the table.
 444  */
 445 STATIC void
 446 xfs_dir2_data_freeremove(
 447         xfs_dir2_data_hdr_t     *hdr,           /* data block header */
 448         xfs_dir2_data_free_t    *dfp,           /* bestfree entry pointer */
 449         int                     *loghead)       /* out: log data header */
 450 {
 451         struct xfs_dir2_data_free *bf;
 452
 453         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 454                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 455                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 456                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 457
 458         /*
 459          * It's the first entry, slide the next 2 up.
 460          */
 461         bf = xfs_dir3_data_bestfree_p(hdr);
 462         if (dfp == &bf[0]) {
 463                 bf[0] = bf[1];
 464                 bf[1] = bf[2];
 465         }
 466         /*
 467          * It's the second entry, slide the 3rd entry up.
 468          */
 469         else if (dfp == &bf[1])
 470                 bf[1] = bf[2];
 471         /*
 472          * Must be the last entry.
 473          */
 474         else
 475                 ASSERT(dfp == &bf[2]);
 476         /*
 477          * Clear the 3rd entry, must be zero now.
 478          */
 479         bf[2].length = 0;
 480         bf[2].offset = 0;
 481         *loghead = 1;
 482 }
 483
 484 /*
 485  * Given a data block, reconstruct its bestfree map.
 486  */
 487 void
 488 xfs_dir2_data_freescan(
 489         xfs_mount_t             *mp,            /* filesystem mount point */
 490         xfs_dir2_data_hdr_t     *hdr,           /* data block header */
 491         int                     *loghead)       /* out: log data header */
 492 {
 493         xfs_dir2_block_tail_t   *btp;           /* block tail */
 494         xfs_dir2_data_entry_t   *dep;           /* active data entry */
 495         xfs_dir2_data_unused_t  *dup;           /* unused data entry */
 496         struct xfs_dir2_data_free *bf;
 497         char                    *endp;          /* end of block's data */
 498         char                    *p;             /* current entry pointer */
 499
 500         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 501                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 502                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 503                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 504
 505         /*
 506          * Start by clearing the table.
 507          */
 508         bf = xfs_dir3_data_bestfree_p(hdr);
 509         memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
 510         *loghead = 1;
 511         /*
 512          * Set up pointers.
 513          */
 514         p = (char *)xfs_dir3_data_entry_p(hdr);
 515         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 516             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 517                 btp = xfs_dir2_block_tail_p(mp, hdr);
 518                 endp = (char *)xfs_dir2_block_leaf_p(btp);
 519         } else
 520                 endp = (char *)hdr + mp->m_dirblksize;
 521         /*
 522          * Loop over the block's entries.
 523          */
 524         while (p < endp) {
 525                 dup = (xfs_dir2_data_unused_t *)p;
 526                 /*
 527                  * If it's a free entry, insert it.
 528                  */
 529                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 530                         ASSERT((char *)dup - (char *)hdr ==
 531                                be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 532                         xfs_dir2_data_freeinsert(hdr, dup, loghead);
 533                         p += be16_to_cpu(dup->length);
 534                 }
 535                 /*
 536                  * For active entries, check their tags and skip them.
 537                  */
 538                 else {
 539                         dep = (xfs_dir2_data_entry_t *)p;
 540                         ASSERT((char *)dep - (char *)hdr ==
 541                                be16_to_cpu(*xfs_dir2_data_entry_tag_p(dep)));
 542                         p += xfs_dir2_data_entsize(dep->namelen);
 543                 }
 544         }
 545 }
 546
 547 /*
 548  * Initialize a data block at the given block number in the directory.
 549  * Give back the buffer for the created block.
 550  */
 551 int                                             /* error */
 552 xfs_dir3_data_init(
 553         xfs_da_args_t           *args,          /* directory operation args */
 554         xfs_dir2_db_t           blkno,          /* logical dir block number */
 555         struct xfs_buf          **bpp)          /* output block buffer */
 556 {
 557         struct xfs_buf          *bp;            /* block buffer */
 558         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 559         xfs_inode_t             *dp;            /* incore directory inode */
 560         xfs_dir2_data_unused_t  *dup;           /* unused entry pointer */
 561         struct xfs_dir2_data_free *bf;
 562         int                     error;          /* error return value */
 563         int                     i;              /* bestfree index */
 564         xfs_mount_t             *mp;            /* filesystem mount point */
 565         xfs_trans_t             *tp;            /* transaction pointer */
 566         int                     t;              /* temp */
 567
 568         dp = args->dp;
 569         mp = dp->i_mount;
 570         tp = args->trans;
 571         /*
 572          * Get the buffer set up for the block.
 573          */
 574         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, blkno), -1, &bp,
 575                 XFS_DATA_FORK);
 576         if (error)
 577                 return error;
 578         bp->b_ops = &xfs_dir3_data_buf_ops;
 579         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
 580
 581         /*
 582          * Initialize the header.
 583          */
 584         hdr = bp->b_addr;
 585         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 586                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 587
 588                 memset(hdr3, 0, sizeof(*hdr3));
 589                 hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 590                 hdr3->blkno = cpu_to_be64(bp->b_bn);
 591                 hdr3->owner = cpu_to_be64(dp->i_ino);
 592                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
 593
 594         } else
 595                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 596
 597         bf = xfs_dir3_data_bestfree_p(hdr);
 598         bf[0].offset = cpu_to_be16(xfs_dir3_data_entry_offset(hdr));
 599         for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
 600                 bf[i].length = 0;
 601                 bf[i].offset = 0;
 602         }
 603
 604         /*
 605          * Set up an unused entry for the block's body.
 606          */
 607         dup = xfs_dir3_data_unused_p(hdr);
 608         dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 609
 610         t = mp->m_dirblksize - (uint)xfs_dir3_data_entry_offset(hdr);
 611         bf[0].length = cpu_to_be16(t);
 612         dup->length = cpu_to_be16(t);
 613         *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
 614         /*
 615          * Log it and return it.
 616          */
 617         xfs_dir2_data_log_header(tp, bp);
 618         xfs_dir2_data_log_unused(tp, bp, dup);
 619         *bpp = bp;
 620         return 0;
 621 }
 622
 623 /*
 624  * Log an active data entry from the block.
 625  */
 626 void
 627 xfs_dir2_data_log_entry(
 628         struct xfs_trans        *tp,
 629         struct xfs_buf          *bp,
 630         xfs_dir2_data_entry_t   *dep)           /* data entry pointer */
 631 {
 632         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 633
 634         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 635                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 636                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 637                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 638
 639         xfs_trans_log_buf(tp, bp, (uint)((char *)dep - (char *)hdr),
 640                 (uint)((char *)(xfs_dir2_data_entry_tag_p(dep) + 1) -
 641                        (char *)hdr - 1));
 642 }
 643
 644 /*
 645  * Log a data block header.
 646  */
 647 void
 648 xfs_dir2_data_log_header(
 649         struct xfs_trans        *tp,
 650         struct xfs_buf          *bp)
 651 {
 652         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 653
 654         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 655                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 656                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 657                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 658
 659         xfs_trans_log_buf(tp, bp, 0, xfs_dir3_data_entry_offset(hdr) - 1);
 660 }
 661
 662 /*
 663  * Log a data unused entry.
 664  */
 665 void
 666 xfs_dir2_data_log_unused(
 667         struct xfs_trans        *tp,
 668         struct xfs_buf          *bp,
 669         xfs_dir2_data_unused_t  *dup)           /* data unused pointer */
 670 {
 671         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 672
 673         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 674                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 675                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 676                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 677
 678         /*
 679          * Log the first part of the unused entry.
 680          */
 681         xfs_trans_log_buf(tp, bp, (uint)((char *)dup - (char *)hdr),
 682                 (uint)((char *)&dup->length + sizeof(dup->length) -
 683                        1 - (char *)hdr));
 684         /*
 685          * Log the end (tag) of the unused entry.
 686          */
 687         xfs_trans_log_buf(tp, bp,
 688                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
 689                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
 690                        sizeof(xfs_dir2_data_off_t) - 1));
 691 }
 692
 693 /*
 694  * Make a byte range in the data block unused.
 695  * Its current contents are unimportant.
 696  */
 697 void
 698 xfs_dir2_data_make_free(
 699         struct xfs_trans        *tp,
 700         struct xfs_buf          *bp,
 701         xfs_dir2_data_aoff_t    offset,         /* starting byte offset */
 702         xfs_dir2_data_aoff_t    len,            /* length in bytes */
 703         int                     *needlogp,      /* out: log header */
 704         int                     *needscanp)     /* out: regen bestfree */
 705 {
 706         xfs_dir2_data_hdr_t     *hdr;           /* data block pointer */
 707         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 708         char                    *endptr;        /* end of data area */
 709         xfs_mount_t             *mp;            /* filesystem mount point */
 710         int                     needscan;       /* need to regen bestfree */
 711         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 712         xfs_dir2_data_unused_t  *postdup;       /* unused entry after us */
 713         xfs_dir2_data_unused_t  *prevdup;       /* unused entry before us */
 714         struct xfs_dir2_data_free *bf;
 715
 716         mp = tp->t_mountp;
 717         hdr = bp->b_addr;
 718
 719         /*
 720          * Figure out where the end of the data area is.
 721          */
 722         if (hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 723             hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 724                 endptr = (char *)hdr + mp->m_dirblksize;
 725         else {
 726                 xfs_dir2_block_tail_t   *btp;   /* block tail */
 727
 728                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 729                         hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 730                 btp = xfs_dir2_block_tail_p(mp, hdr);
 731                 endptr = (char *)xfs_dir2_block_leaf_p(btp);
 732         }
 733         /*
 734          * If this isn't the start of the block, then back up to
 735          * the previous entry and see if it's free.
 736          */
 737         if (offset > xfs_dir3_data_entry_offset(hdr)) {
 738                 __be16                  *tagp;  /* tag just before us */
 739
 740                 tagp = (__be16 *)((char *)hdr + offset) - 1;
 741                 prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 742                 if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 743                         prevdup = NULL;
 744         } else
 745                 prevdup = NULL;
 746         /*
 747          * If this isn't the end of the block, see if the entry after
 748          * us is free.
 749          */
 750         if ((char *)hdr + offset + len < endptr) {
 751                 postdup =
 752                         (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 753                 if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 754                         postdup = NULL;
 755         } else
 756                 postdup = NULL;
 757         ASSERT(*needscanp == 0);
 758         needscan = 0;
 759         /*
 760          * Previous and following entries are both free,
 761          * merge everything into a single free entry.
 762          */
 763         bf = xfs_dir3_data_bestfree_p(hdr);
 764         if (prevdup && postdup) {
 765                 xfs_dir2_data_free_t    *dfp2;  /* another bestfree pointer */
 766
 767                 /*
 768                  * See if prevdup and/or postdup are in bestfree table.
 769                  */
 770                 dfp = xfs_dir2_data_freefind(hdr, prevdup);
 771                 dfp2 = xfs_dir2_data_freefind(hdr, postdup);
 772                 /*
 773                  * We need a rescan unless there are exactly 2 free entries
 774                  * namely our two.  Then we know what's happening, otherwise
 775                  * since the third bestfree is there, there might be more
 776                  * entries.
 777                  */
 778                 needscan = (bf[2].length != 0);
 779                 /*
 780                  * Fix up the new big freespace.
 781                  */
 782                 be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
 783                 *xfs_dir2_data_unused_tag_p(prevdup) =
 784                         cpu_to_be16((char *)prevdup - (char *)hdr);
 785                 xfs_dir2_data_log_unused(tp, bp, prevdup);
 786                 if (!needscan) {
 787                         /*
 788                          * Has to be the case that entries 0 and 1 are
 789                          * dfp and dfp2 (don't know which is which), and
 790                          * entry 2 is empty.
 791                          * Remove entry 1 first then entry 0.
 792                          */
 793                         ASSERT(dfp && dfp2);
 794                         if (dfp == &bf[1]) {
 795                                 dfp = &bf[0];
 796                                 ASSERT(dfp2 == dfp);
 797                                 dfp2 = &bf[1];
 798                         }
 799                         xfs_dir2_data_freeremove(hdr, dfp2, needlogp);
 800                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 801                         /*
 802                          * Now insert the new entry.
 803                          */
 804                         dfp = xfs_dir2_data_freeinsert(hdr, prevdup, needlogp);
 805                         ASSERT(dfp == &bf[0]);
 806                         ASSERT(dfp->length == prevdup->length);
 807                         ASSERT(!dfp[1].length);
 808                         ASSERT(!dfp[2].length);
 809                 }
 810         }
 811         /*
 812          * The entry before us is free, merge with it.
 813          */
 814         else if (prevdup) {
 815                 dfp = xfs_dir2_data_freefind(hdr, prevdup);
 816                 be16_add_cpu(&prevdup->length, len);
 817                 *xfs_dir2_data_unused_tag_p(prevdup) =
 818                         cpu_to_be16((char *)prevdup - (char *)hdr);
 819                 xfs_dir2_data_log_unused(tp, bp, prevdup);
 820                 /*
 821                  * If the previous entry was in the table, the new entry
 822                  * is longer, so it will be in the table too.  Remove
 823                  * the old one and add the new one.
 824                  */
 825                 if (dfp) {
 826                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 827                         xfs_dir2_data_freeinsert(hdr, prevdup, needlogp);
 828                 }
 829                 /*
 830                  * Otherwise we need a scan if the new entry is big enough.
 831                  */
 832                 else {
 833                         needscan = be16_to_cpu(prevdup->length) >
 834                                    be16_to_cpu(bf[2].length);
 835                 }
 836         }
 837         /*
 838          * The following entry is free, merge with it.
 839          */
 840         else if (postdup) {
 841                 dfp = xfs_dir2_data_freefind(hdr, postdup);
 842                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 843                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 844                 newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
 845                 *xfs_dir2_data_unused_tag_p(newdup) =
 846                         cpu_to_be16((char *)newdup - (char *)hdr);
 847                 xfs_dir2_data_log_unused(tp, bp, newdup);
 848                 /*
 849                  * If the following entry was in the table, the new entry
 850                  * is longer, so it will be in the table too.  Remove
 851                  * the old one and add the new one.
 852                  */
 853                 if (dfp) {
 854                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 855                         xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 856                 }
 857                 /*
 858                  * Otherwise we need a scan if the new entry is big enough.
 859                  */
 860                 else {
 861                         needscan = be16_to_cpu(newdup->length) >
 862                                    be16_to_cpu(bf[2].length);
 863                 }
 864         }
 865         /*
 866          * Neither neighbor is free.  Make a new entry.
 867          */
 868         else {
 869                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 870                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 871                 newdup->length = cpu_to_be16(len);
 872                 *xfs_dir2_data_unused_tag_p(newdup) =
 873                         cpu_to_be16((char *)newdup - (char *)hdr);
 874                 xfs_dir2_data_log_unused(tp, bp, newdup);
 875                 xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 876         }
 877         *needscanp = needscan;
 878 }
 879
 880 /*
 881  * Take a byte range out of an existing unused space and make it un-free.
 882  */
 883 void
 884 xfs_dir2_data_use_free(
 885         struct xfs_trans        *tp,
 886         struct xfs_buf          *bp,
 887         xfs_dir2_data_unused_t  *dup,           /* unused entry */
 888         xfs_dir2_data_aoff_t    offset,         /* starting offset to use */
 889         xfs_dir2_data_aoff_t    len,            /* length to use */
 890         int                     *needlogp,      /* out: need to log header */
 891         int                     *needscanp)     /* out: need regen bestfree */
 892 {
 893         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 894         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 895         int                     matchback;      /* matches end of freespace */
 896         int                     matchfront;     /* matches start of freespace */
 897         int                     needscan;       /* need to regen bestfree */
 898         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 899         xfs_dir2_data_unused_t  *newdup2;       /* another new unused entry */
 900         int                     oldlen;         /* old unused entry's length */
 901         struct xfs_dir2_data_free *bf;
 902
 903         hdr = bp->b_addr;
 904         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 905                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 906                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 907                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 908         ASSERT(be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG);
 909         ASSERT(offset >= (char *)dup - (char *)hdr);
 910         ASSERT(offset + len <= (char *)dup + be16_to_cpu(dup->length) - (char *)hdr);
 911         ASSERT((char *)dup - (char *)hdr == be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 912         /*
 913          * Look up the entry in the bestfree table.
 914          */
 915         dfp = xfs_dir2_data_freefind(hdr, dup);
 916         oldlen = be16_to_cpu(dup->length);
 917         bf = xfs_dir3_data_bestfree_p(hdr);
 918         ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
 919         /*
 920          * Check for alignment with front and back of the entry.
 921          */
 922         matchfront = (char *)dup - (char *)hdr == offset;
 923         matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
 924         ASSERT(*needscanp == 0);
 925         needscan = 0;
 926         /*
 927          * If we matched it exactly we just need to get rid of it from
 928          * the bestfree table.
 929          */
 930         if (matchfront && matchback) {
 931                 if (dfp) {
 932                         needscan = (bf[2].offset != 0);
 933                         if (!needscan)
 934                                 xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 935                 }
 936         }
 937         /*
 938          * We match the first part of the entry.
 939          * Make a new entry with the remaining freespace.
 940          */
 941         else if (matchfront) {
 942                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 943                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 944                 newdup->length = cpu_to_be16(oldlen - len);
 945                 *xfs_dir2_data_unused_tag_p(newdup) =
 946                         cpu_to_be16((char *)newdup - (char *)hdr);
 947                 xfs_dir2_data_log_unused(tp, bp, newdup);
 948                 /*
 949                  * If it was in the table, remove it and add the new one.
 950                  */
 951                 if (dfp) {
 952                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 953                         dfp = xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 954                         ASSERT(dfp != NULL);
 955                         ASSERT(dfp->length == newdup->length);
 956                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 957                         /*
 958                          * If we got inserted at the last slot,
 959                          * that means we don't know if there was a better
 960                          * choice for the last slot, or not.  Rescan.
 961                          */
 962                         needscan = dfp == &bf[2];
 963                 }
 964         }
 965         /*
 966          * We match the last part of the entry.
 967          * Trim the allocated space off the tail of the entry.
 968          */
 969         else if (matchback) {
 970                 newdup = dup;
 971                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
 972                 *xfs_dir2_data_unused_tag_p(newdup) =
 973                         cpu_to_be16((char *)newdup - (char *)hdr);
 974                 xfs_dir2_data_log_unused(tp, bp, newdup);
 975                 /*
 976                  * If it was in the table, remove it and add the new one.
 977                  */
 978                 if (dfp) {
 979                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 980                         dfp = xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 981                         ASSERT(dfp != NULL);
 982                         ASSERT(dfp->length == newdup->length);
 983                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 984                         /*
 985                          * If we got inserted at the last slot,
 986                          * that means we don't know if there was a better
 987                          * choice for the last slot, or not.  Rescan.
 988                          */
 989                         needscan = dfp == &bf[2];
 990                 }
 991         }
 992         /*
 993          * Poking out the middle of an entry.
 994          * Make two new entries.
 995          */
 996         else {
 997                 newdup = dup;
 998                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
 999                 *xfs_dir2_data_unused_tag_p(newdup) =
1000                         cpu_to_be16((char *)newdup - (char *)hdr);
1001                 xfs_dir2_data_log_unused(tp, bp, newdup);
1002                 newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1003                 newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1004                 newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1005                 *xfs_dir2_data_unused_tag_p(newdup2) =
1006                         cpu_to_be16((char *)newdup2 - (char *)hdr);
1007                 xfs_dir2_data_log_unused(tp, bp, newdup2);
1008                 /*
1009                  * If the old entry was in the table, we need to scan
1010                  * if the 3rd entry was valid, since these entries
1011                  * are smaller than the old one.
1012                  * If we don't need to scan that means there were 1 or 2
1013                  * entries in the table, and removing the old and adding
1014                  * the 2 new will work.
1015                  */
1016                 if (dfp) {
1017                         needscan = (bf[2].length != 0);
1018                         if (!needscan) {
1019                                 xfs_dir2_data_freeremove(hdr, dfp, needlogp);
1020                                 xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
1021                                 xfs_dir2_data_freeinsert(hdr, newdup2,
1022                                                          needlogp);
1023                         }
1024                 }
1025         }
1026         *needscanp = needscan;
1027 }