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         return xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
 305                                 XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
 306 }
 307
 308 int
 309 xfs_dir3_data_readahead(
 310         struct xfs_trans        *tp,
 311         struct xfs_inode        *dp,
 312         xfs_dablk_t             bno,
 313         xfs_daddr_t             mapped_bno)
 314 {
 315         return xfs_da_reada_buf(tp, dp, bno, mapped_bno,
 316                                 XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
 317 }
 318
 319 /*
 320  * Given a data block and an unused entry from that block,
 321  * return the bestfree entry if any that corresponds to it.
 322  */
 323 STATIC xfs_dir2_data_free_t *
 324 xfs_dir2_data_freefind(
 325         xfs_dir2_data_hdr_t     *hdr,           /* data block */
 326         xfs_dir2_data_unused_t  *dup)           /* data unused entry */
 327 {
 328         xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
 329         xfs_dir2_data_aoff_t    off;            /* offset value needed */
 330         struct xfs_dir2_data_free *bf;
 331 #if defined(DEBUG) && defined(__KERNEL__)
 332         int                     matched;        /* matched the value */
 333         int                     seenzero;       /* saw a 0 bestfree entry */
 334 #endif
 335
 336         off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
 337         bf = xfs_dir3_data_bestfree_p(hdr);
 338
 339 #if defined(DEBUG) && defined(__KERNEL__)
 340         /*
 341          * Validate some consistency in the bestfree table.
 342          * Check order, non-overlapping entries, and if we find the
 343          * one we're looking for it has to be exact.
 344          */
 345         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 346                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 347                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 348                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 349         for (dfp = &bf[0], seenzero = matched = 0;
 350              dfp < &bf[XFS_DIR2_DATA_FD_COUNT];
 351              dfp++) {
 352                 if (!dfp->offset) {
 353                         ASSERT(!dfp->length);
 354                         seenzero = 1;
 355                         continue;
 356                 }
 357                 ASSERT(seenzero == 0);
 358                 if (be16_to_cpu(dfp->offset) == off) {
 359                         matched = 1;
 360                         ASSERT(dfp->length == dup->length);
 361                 } else if (off < be16_to_cpu(dfp->offset))
 362                         ASSERT(off + be16_to_cpu(dup->length) <= be16_to_cpu(dfp->offset));
 363                 else
 364                         ASSERT(be16_to_cpu(dfp->offset) + be16_to_cpu(dfp->length) <= off);
 365                 ASSERT(matched || be16_to_cpu(dfp->length) >= be16_to_cpu(dup->length));
 366                 if (dfp > &bf[0])
 367                         ASSERT(be16_to_cpu(dfp[-1].length) >= be16_to_cpu(dfp[0].length));
 368         }
 369 #endif
 370         /*
 371          * If this is smaller than the smallest bestfree entry,
 372          * it can't be there since they're sorted.
 373          */
 374         if (be16_to_cpu(dup->length) <
 375             be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 376                 return NULL;
 377         /*
 378          * Look at the three bestfree entries for our guy.
 379          */
 380         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 381                 if (!dfp->offset)
 382                         return NULL;
 383                 if (be16_to_cpu(dfp->offset) == off)
 384                         return dfp;
 385         }
 386         /*
 387          * Didn't find it.  This only happens if there are duplicate lengths.
 388          */
 389         return NULL;
 390 }
 391
 392 /*
 393  * Insert an unused-space entry into the bestfree table.
 394  */
 395 xfs_dir2_data_free_t *                          /* entry inserted */
 396 xfs_dir2_data_freeinsert(
 397         xfs_dir2_data_hdr_t     *hdr,           /* data block pointer */
 398         xfs_dir2_data_unused_t  *dup,           /* unused space */
 399         int                     *loghead)       /* log the data header (out) */
 400 {
 401         xfs_dir2_data_free_t    *dfp;           /* bestfree table pointer */
 402         xfs_dir2_data_free_t    new;            /* new bestfree entry */
 403
 404         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 405                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 406                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 407                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 408
 409         dfp = xfs_dir3_data_bestfree_p(hdr);
 410         new.length = dup->length;
 411         new.offset = cpu_to_be16((char *)dup - (char *)hdr);
 412
 413         /*
 414          * Insert at position 0, 1, or 2; or not at all.
 415          */
 416         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
 417                 dfp[2] = dfp[1];
 418                 dfp[1] = dfp[0];
 419                 dfp[0] = new;
 420                 *loghead = 1;
 421                 return &dfp[0];
 422         }
 423         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
 424                 dfp[2] = dfp[1];
 425                 dfp[1] = new;
 426                 *loghead = 1;
 427                 return &dfp[1];
 428         }
 429         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
 430                 dfp[2] = new;
 431                 *loghead = 1;
 432                 return &dfp[2];
 433         }
 434         return NULL;
 435 }
 436
 437 /*
 438  * Remove a bestfree entry from the table.
 439  */
 440 STATIC void
 441 xfs_dir2_data_freeremove(
 442         xfs_dir2_data_hdr_t     *hdr,           /* data block header */
 443         xfs_dir2_data_free_t    *dfp,           /* bestfree entry pointer */
 444         int                     *loghead)       /* out: log data header */
 445 {
 446         struct xfs_dir2_data_free *bf;
 447
 448         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 449                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 450                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 451                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 452
 453         /*
 454          * It's the first entry, slide the next 2 up.
 455          */
 456         bf = xfs_dir3_data_bestfree_p(hdr);
 457         if (dfp == &bf[0]) {
 458                 bf[0] = bf[1];
 459                 bf[1] = bf[2];
 460         }
 461         /*
 462          * It's the second entry, slide the 3rd entry up.
 463          */
 464         else if (dfp == &bf[1])
 465                 bf[1] = bf[2];
 466         /*
 467          * Must be the last entry.
 468          */
 469         else
 470                 ASSERT(dfp == &bf[2]);
 471         /*
 472          * Clear the 3rd entry, must be zero now.
 473          */
 474         bf[2].length = 0;
 475         bf[2].offset = 0;
 476         *loghead = 1;
 477 }
 478
 479 /*
 480  * Given a data block, reconstruct its bestfree map.
 481  */
 482 void
 483 xfs_dir2_data_freescan(
 484         xfs_mount_t             *mp,            /* filesystem mount point */
 485         xfs_dir2_data_hdr_t     *hdr,           /* data block header */
 486         int                     *loghead)       /* out: log data header */
 487 {
 488         xfs_dir2_block_tail_t   *btp;           /* block tail */
 489         xfs_dir2_data_entry_t   *dep;           /* active data entry */
 490         xfs_dir2_data_unused_t  *dup;           /* unused data entry */
 491         struct xfs_dir2_data_free *bf;
 492         char                    *endp;          /* end of block's data */
 493         char                    *p;             /* current entry pointer */
 494
 495         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 496                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 497                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 498                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 499
 500         /*
 501          * Start by clearing the table.
 502          */
 503         bf = xfs_dir3_data_bestfree_p(hdr);
 504         memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
 505         *loghead = 1;
 506         /*
 507          * Set up pointers.
 508          */
 509         p = (char *)xfs_dir3_data_entry_p(hdr);
 510         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 511             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 512                 btp = xfs_dir2_block_tail_p(mp, hdr);
 513                 endp = (char *)xfs_dir2_block_leaf_p(btp);
 514         } else
 515                 endp = (char *)hdr + mp->m_dirblksize;
 516         /*
 517          * Loop over the block's entries.
 518          */
 519         while (p < endp) {
 520                 dup = (xfs_dir2_data_unused_t *)p;
 521                 /*
 522                  * If it's a free entry, insert it.
 523                  */
 524                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 525                         ASSERT((char *)dup - (char *)hdr ==
 526                                be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 527                         xfs_dir2_data_freeinsert(hdr, dup, loghead);
 528                         p += be16_to_cpu(dup->length);
 529                 }
 530                 /*
 531                  * For active entries, check their tags and skip them.
 532                  */
 533                 else {
 534                         dep = (xfs_dir2_data_entry_t *)p;
 535                         ASSERT((char *)dep - (char *)hdr ==
 536                                be16_to_cpu(*xfs_dir2_data_entry_tag_p(dep)));
 537                         p += xfs_dir2_data_entsize(dep->namelen);
 538                 }
 539         }
 540 }
 541
 542 /*
 543  * Initialize a data block at the given block number in the directory.
 544  * Give back the buffer for the created block.
 545  */
 546 int                                             /* error */
 547 xfs_dir3_data_init(
 548         xfs_da_args_t           *args,          /* directory operation args */
 549         xfs_dir2_db_t           blkno,          /* logical dir block number */
 550         struct xfs_buf          **bpp)          /* output block buffer */
 551 {
 552         struct xfs_buf          *bp;            /* block buffer */
 553         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 554         xfs_inode_t             *dp;            /* incore directory inode */
 555         xfs_dir2_data_unused_t  *dup;           /* unused entry pointer */
 556         struct xfs_dir2_data_free *bf;
 557         int                     error;          /* error return value */
 558         int                     i;              /* bestfree index */
 559         xfs_mount_t             *mp;            /* filesystem mount point */
 560         xfs_trans_t             *tp;            /* transaction pointer */
 561         int                     t;              /* temp */
 562
 563         dp = args->dp;
 564         mp = dp->i_mount;
 565         tp = args->trans;
 566         /*
 567          * Get the buffer set up for the block.
 568          */
 569         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, blkno), -1, &bp,
 570                 XFS_DATA_FORK);
 571         if (error)
 572                 return error;
 573         bp->b_ops = &xfs_dir3_data_buf_ops;
 574
 575         /*
 576          * Initialize the header.
 577          */
 578         hdr = bp->b_addr;
 579         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 580                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 581
 582                 memset(hdr3, 0, sizeof(*hdr3));
 583                 hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 584                 hdr3->blkno = cpu_to_be64(bp->b_bn);
 585                 hdr3->owner = cpu_to_be64(dp->i_ino);
 586                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
 587
 588         } else
 589                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 590
 591         bf = xfs_dir3_data_bestfree_p(hdr);
 592         bf[0].offset = cpu_to_be16(xfs_dir3_data_entry_offset(hdr));
 593         for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
 594                 bf[i].length = 0;
 595                 bf[i].offset = 0;
 596         }
 597
 598         /*
 599          * Set up an unused entry for the block's body.
 600          */
 601         dup = xfs_dir3_data_unused_p(hdr);
 602         dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 603
 604         t = mp->m_dirblksize - (uint)xfs_dir3_data_entry_offset(hdr);
 605         bf[0].length = cpu_to_be16(t);
 606         dup->length = cpu_to_be16(t);
 607         *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
 608         /*
 609          * Log it and return it.
 610          */
 611         xfs_dir2_data_log_header(tp, bp);
 612         xfs_dir2_data_log_unused(tp, bp, dup);
 613         *bpp = bp;
 614         return 0;
 615 }
 616
 617 /*
 618  * Log an active data entry from the block.
 619  */
 620 void
 621 xfs_dir2_data_log_entry(
 622         struct xfs_trans        *tp,
 623         struct xfs_buf          *bp,
 624         xfs_dir2_data_entry_t   *dep)           /* data entry pointer */
 625 {
 626         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 627
 628         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 629                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 630                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 631                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 632
 633         xfs_trans_log_buf(tp, bp, (uint)((char *)dep - (char *)hdr),
 634                 (uint)((char *)(xfs_dir2_data_entry_tag_p(dep) + 1) -
 635                        (char *)hdr - 1));
 636 }
 637
 638 /*
 639  * Log a data block header.
 640  */
 641 void
 642 xfs_dir2_data_log_header(
 643         struct xfs_trans        *tp,
 644         struct xfs_buf          *bp)
 645 {
 646         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 647
 648         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 649                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 650                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 651                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 652
 653         xfs_trans_log_buf(tp, bp, 0, xfs_dir3_data_entry_offset(hdr) - 1);
 654 }
 655
 656 /*
 657  * Log a data unused entry.
 658  */
 659 void
 660 xfs_dir2_data_log_unused(
 661         struct xfs_trans        *tp,
 662         struct xfs_buf          *bp,
 663         xfs_dir2_data_unused_t  *dup)           /* data unused pointer */
 664 {
 665         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 666
 667         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 668                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 669                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 670                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 671
 672         /*
 673          * Log the first part of the unused entry.
 674          */
 675         xfs_trans_log_buf(tp, bp, (uint)((char *)dup - (char *)hdr),
 676                 (uint)((char *)&dup->length + sizeof(dup->length) -
 677                        1 - (char *)hdr));
 678         /*
 679          * Log the end (tag) of the unused entry.
 680          */
 681         xfs_trans_log_buf(tp, bp,
 682                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
 683                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
 684                        sizeof(xfs_dir2_data_off_t) - 1));
 685 }
 686
 687 /*
 688  * Make a byte range in the data block unused.
 689  * Its current contents are unimportant.
 690  */
 691 void
 692 xfs_dir2_data_make_free(
 693         struct xfs_trans        *tp,
 694         struct xfs_buf          *bp,
 695         xfs_dir2_data_aoff_t    offset,         /* starting byte offset */
 696         xfs_dir2_data_aoff_t    len,            /* length in bytes */
 697         int                     *needlogp,      /* out: log header */
 698         int                     *needscanp)     /* out: regen bestfree */
 699 {
 700         xfs_dir2_data_hdr_t     *hdr;           /* data block pointer */
 701         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 702         char                    *endptr;        /* end of data area */
 703         xfs_mount_t             *mp;            /* filesystem mount point */
 704         int                     needscan;       /* need to regen bestfree */
 705         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 706         xfs_dir2_data_unused_t  *postdup;       /* unused entry after us */
 707         xfs_dir2_data_unused_t  *prevdup;       /* unused entry before us */
 708         struct xfs_dir2_data_free *bf;
 709
 710         mp = tp->t_mountp;
 711         hdr = bp->b_addr;
 712
 713         /*
 714          * Figure out where the end of the data area is.
 715          */
 716         if (hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 717             hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 718                 endptr = (char *)hdr + mp->m_dirblksize;
 719         else {
 720                 xfs_dir2_block_tail_t   *btp;   /* block tail */
 721
 722                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 723                         hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 724                 btp = xfs_dir2_block_tail_p(mp, hdr);
 725                 endptr = (char *)xfs_dir2_block_leaf_p(btp);
 726         }
 727         /*
 728          * If this isn't the start of the block, then back up to
 729          * the previous entry and see if it's free.
 730          */
 731         if (offset > xfs_dir3_data_entry_offset(hdr)) {
 732                 __be16                  *tagp;  /* tag just before us */
 733
 734                 tagp = (__be16 *)((char *)hdr + offset) - 1;
 735                 prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 736                 if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 737                         prevdup = NULL;
 738         } else
 739                 prevdup = NULL;
 740         /*
 741          * If this isn't the end of the block, see if the entry after
 742          * us is free.
 743          */
 744         if ((char *)hdr + offset + len < endptr) {
 745                 postdup =
 746                         (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 747                 if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 748                         postdup = NULL;
 749         } else
 750                 postdup = NULL;
 751         ASSERT(*needscanp == 0);
 752         needscan = 0;
 753         /*
 754          * Previous and following entries are both free,
 755          * merge everything into a single free entry.
 756          */
 757         bf = xfs_dir3_data_bestfree_p(hdr);
 758         if (prevdup && postdup) {
 759                 xfs_dir2_data_free_t    *dfp2;  /* another bestfree pointer */
 760
 761                 /*
 762                  * See if prevdup and/or postdup are in bestfree table.
 763                  */
 764                 dfp = xfs_dir2_data_freefind(hdr, prevdup);
 765                 dfp2 = xfs_dir2_data_freefind(hdr, postdup);
 766                 /*
 767                  * We need a rescan unless there are exactly 2 free entries
 768                  * namely our two.  Then we know what's happening, otherwise
 769                  * since the third bestfree is there, there might be more
 770                  * entries.
 771                  */
 772                 needscan = (bf[2].length != 0);
 773                 /*
 774                  * Fix up the new big freespace.
 775                  */
 776                 be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
 777                 *xfs_dir2_data_unused_tag_p(prevdup) =
 778                         cpu_to_be16((char *)prevdup - (char *)hdr);
 779                 xfs_dir2_data_log_unused(tp, bp, prevdup);
 780                 if (!needscan) {
 781                         /*
 782                          * Has to be the case that entries 0 and 1 are
 783                          * dfp and dfp2 (don't know which is which), and
 784                          * entry 2 is empty.
 785                          * Remove entry 1 first then entry 0.
 786                          */
 787                         ASSERT(dfp && dfp2);
 788                         if (dfp == &bf[1]) {
 789                                 dfp = &bf[0];
 790                                 ASSERT(dfp2 == dfp);
 791                                 dfp2 = &bf[1];
 792                         }
 793                         xfs_dir2_data_freeremove(hdr, dfp2, needlogp);
 794                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 795                         /*
 796                          * Now insert the new entry.
 797                          */
 798                         dfp = xfs_dir2_data_freeinsert(hdr, prevdup, needlogp);
 799                         ASSERT(dfp == &bf[0]);
 800                         ASSERT(dfp->length == prevdup->length);
 801                         ASSERT(!dfp[1].length);
 802                         ASSERT(!dfp[2].length);
 803                 }
 804         }
 805         /*
 806          * The entry before us is free, merge with it.
 807          */
 808         else if (prevdup) {
 809                 dfp = xfs_dir2_data_freefind(hdr, prevdup);
 810                 be16_add_cpu(&prevdup->length, len);
 811                 *xfs_dir2_data_unused_tag_p(prevdup) =
 812                         cpu_to_be16((char *)prevdup - (char *)hdr);
 813                 xfs_dir2_data_log_unused(tp, bp, prevdup);
 814                 /*
 815                  * If the previous entry was in the table, the new entry
 816                  * is longer, so it will be in the table too.  Remove
 817                  * the old one and add the new one.
 818                  */
 819                 if (dfp) {
 820                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 821                         xfs_dir2_data_freeinsert(hdr, prevdup, needlogp);
 822                 }
 823                 /*
 824                  * Otherwise we need a scan if the new entry is big enough.
 825                  */
 826                 else {
 827                         needscan = be16_to_cpu(prevdup->length) >
 828                                    be16_to_cpu(bf[2].length);
 829                 }
 830         }
 831         /*
 832          * The following entry is free, merge with it.
 833          */
 834         else if (postdup) {
 835                 dfp = xfs_dir2_data_freefind(hdr, postdup);
 836                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 837                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 838                 newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
 839                 *xfs_dir2_data_unused_tag_p(newdup) =
 840                         cpu_to_be16((char *)newdup - (char *)hdr);
 841                 xfs_dir2_data_log_unused(tp, bp, newdup);
 842                 /*
 843                  * If the following entry was in the table, the new entry
 844                  * is longer, so it will be in the table too.  Remove
 845                  * the old one and add the new one.
 846                  */
 847                 if (dfp) {
 848                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 849                         xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 850                 }
 851                 /*
 852                  * Otherwise we need a scan if the new entry is big enough.
 853                  */
 854                 else {
 855                         needscan = be16_to_cpu(newdup->length) >
 856                                    be16_to_cpu(bf[2].length);
 857                 }
 858         }
 859         /*
 860          * Neither neighbor is free.  Make a new entry.
 861          */
 862         else {
 863                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 864                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 865                 newdup->length = cpu_to_be16(len);
 866                 *xfs_dir2_data_unused_tag_p(newdup) =
 867                         cpu_to_be16((char *)newdup - (char *)hdr);
 868                 xfs_dir2_data_log_unused(tp, bp, newdup);
 869                 xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 870         }
 871         *needscanp = needscan;
 872 }
 873
 874 /*
 875  * Take a byte range out of an existing unused space and make it un-free.
 876  */
 877 void
 878 xfs_dir2_data_use_free(
 879         struct xfs_trans        *tp,
 880         struct xfs_buf          *bp,
 881         xfs_dir2_data_unused_t  *dup,           /* unused entry */
 882         xfs_dir2_data_aoff_t    offset,         /* starting offset to use */
 883         xfs_dir2_data_aoff_t    len,            /* length to use */
 884         int                     *needlogp,      /* out: need to log header */
 885         int                     *needscanp)     /* out: need regen bestfree */
 886 {
 887         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 888         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 889         int                     matchback;      /* matches end of freespace */
 890         int                     matchfront;     /* matches start of freespace */
 891         int                     needscan;       /* need to regen bestfree */
 892         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 893         xfs_dir2_data_unused_t  *newdup2;       /* another new unused entry */
 894         int                     oldlen;         /* old unused entry's length */
 895         struct xfs_dir2_data_free *bf;
 896
 897         hdr = bp->b_addr;
 898         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 899                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 900                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 901                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 902         ASSERT(be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG);
 903         ASSERT(offset >= (char *)dup - (char *)hdr);
 904         ASSERT(offset + len <= (char *)dup + be16_to_cpu(dup->length) - (char *)hdr);
 905         ASSERT((char *)dup - (char *)hdr == be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 906         /*
 907          * Look up the entry in the bestfree table.
 908          */
 909         dfp = xfs_dir2_data_freefind(hdr, dup);
 910         oldlen = be16_to_cpu(dup->length);
 911         bf = xfs_dir3_data_bestfree_p(hdr);
 912         ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
 913         /*
 914          * Check for alignment with front and back of the entry.
 915          */
 916         matchfront = (char *)dup - (char *)hdr == offset;
 917         matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
 918         ASSERT(*needscanp == 0);
 919         needscan = 0;
 920         /*
 921          * If we matched it exactly we just need to get rid of it from
 922          * the bestfree table.
 923          */
 924         if (matchfront && matchback) {
 925                 if (dfp) {
 926                         needscan = (bf[2].offset != 0);
 927                         if (!needscan)
 928                                 xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 929                 }
 930         }
 931         /*
 932          * We match the first part of the entry.
 933          * Make a new entry with the remaining freespace.
 934          */
 935         else if (matchfront) {
 936                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 937                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 938                 newdup->length = cpu_to_be16(oldlen - len);
 939                 *xfs_dir2_data_unused_tag_p(newdup) =
 940                         cpu_to_be16((char *)newdup - (char *)hdr);
 941                 xfs_dir2_data_log_unused(tp, bp, newdup);
 942                 /*
 943                  * If it was in the table, remove it and add the new one.
 944                  */
 945                 if (dfp) {
 946                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 947                         dfp = xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 948                         ASSERT(dfp != NULL);
 949                         ASSERT(dfp->length == newdup->length);
 950                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 951                         /*
 952                          * If we got inserted at the last slot,
 953                          * that means we don't know if there was a better
 954                          * choice for the last slot, or not.  Rescan.
 955                          */
 956                         needscan = dfp == &bf[2];
 957                 }
 958         }
 959         /*
 960          * We match the last part of the entry.
 961          * Trim the allocated space off the tail of the entry.
 962          */
 963         else if (matchback) {
 964                 newdup = dup;
 965                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
 966                 *xfs_dir2_data_unused_tag_p(newdup) =
 967                         cpu_to_be16((char *)newdup - (char *)hdr);
 968                 xfs_dir2_data_log_unused(tp, bp, newdup);
 969                 /*
 970                  * If it was in the table, remove it and add the new one.
 971                  */
 972                 if (dfp) {
 973                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 974                         dfp = xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 975                         ASSERT(dfp != NULL);
 976                         ASSERT(dfp->length == newdup->length);
 977                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 978                         /*
 979                          * If we got inserted at the last slot,
 980                          * that means we don't know if there was a better
 981                          * choice for the last slot, or not.  Rescan.
 982                          */
 983                         needscan = dfp == &bf[2];
 984                 }
 985         }
 986         /*
 987          * Poking out the middle of an entry.
 988          * Make two new entries.
 989          */
 990         else {
 991                 newdup = dup;
 992                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
 993                 *xfs_dir2_data_unused_tag_p(newdup) =
 994                         cpu_to_be16((char *)newdup - (char *)hdr);
 995                 xfs_dir2_data_log_unused(tp, bp, newdup);
 996                 newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 997                 newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 998                 newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
 999                 *xfs_dir2_data_unused_tag_p(newdup2) =
1000                         cpu_to_be16((char *)newdup2 - (char *)hdr);
1001                 xfs_dir2_data_log_unused(tp, bp, newdup2);
1002                 /*
1003                  * If the old entry was in the table, we need to scan
1004                  * if the 3rd entry was valid, since these entries
1005                  * are smaller than the old one.
1006                  * If we don't need to scan that means there were 1 or 2
1007                  * entries in the table, and removing the old and adding
1008                  * the 2 new will work.
1009                  */
1010                 if (dfp) {
1011                         needscan = (bf[2].length != 0);
1012                         if (!needscan) {
1013                                 xfs_dir2_data_freeremove(hdr, dfp, needlogp);
1014                                 xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
1015                                 xfs_dir2_data_freeinsert(hdr, newdup2,
1016                                                          needlogp);
1017                         }
1018                 }
1019         }
1020         *needscanp = needscan;
1021 }