kernel/fs/xfs/libxfs/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_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_mount.h"
  25 #include "xfs_da_format.h"
  26 #include "xfs_da_btree.h"
  27 #include "xfs_inode.h"
  28 #include "xfs_dir2.h"
  29 #include "xfs_dir2_priv.h"
  30 #include "xfs_error.h"
  31 #include "xfs_trans.h"
  32 #include "xfs_buf_item.h"
  33 #include "xfs_cksum.h"
  34
  35 /*
  36  * Check the consistency of the data block.
  37  * The input can also be a block-format directory.
  38  * Return 0 is the buffer is good, otherwise an error.
  39  */
  40 int
  41 __xfs_dir3_data_check(
  42         struct xfs_inode        *dp,            /* incore inode pointer */
  43         struct xfs_buf          *bp)            /* data block's buffer */
  44 {
  45         xfs_dir2_dataptr_t      addr;           /* addr for leaf lookup */
  46         xfs_dir2_data_free_t    *bf;            /* bestfree table */
  47         xfs_dir2_block_tail_t   *btp=NULL;      /* block tail */
  48         int                     count;          /* count of entries found */
  49         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
  50         xfs_dir2_data_entry_t   *dep;           /* data entry */
  51         xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
  52         xfs_dir2_data_unused_t  *dup;           /* unused entry */
  53         char                    *endp;          /* end of useful data */
  54         int                     freeseen;       /* mask of bestfrees seen */
  55         xfs_dahash_t            hash;           /* hash of current name */
  56         int                     i;              /* leaf index */
  57         int                     lastfree;       /* last entry was unused */
  58         xfs_dir2_leaf_entry_t   *lep=NULL;      /* block leaf entries */
  59         xfs_mount_t             *mp;            /* filesystem mount point */
  60         char                    *p;             /* current data position */
  61         int                     stale;          /* count of stale leaves */
  62         struct xfs_name         name;
  63         const struct xfs_dir_ops *ops;
  64         struct xfs_da_geometry  *geo;
  65
  66         mp = bp->b_target->bt_mount;
  67         geo = mp->m_dir_geo;
  68
  69         /*
  70          * We can be passed a null dp here from a verifier, so we need to go the
  71          * hard way to get them.
  72          */
  73         ops = xfs_dir_get_ops(mp, dp);
  74
  75         hdr = bp->b_addr;
  76         p = (char *)ops->data_entry_p(hdr);
  77
  78         switch (hdr->magic) {
  79         case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
  80         case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
  81                 btp = xfs_dir2_block_tail_p(geo, hdr);
  82                 lep = xfs_dir2_block_leaf_p(btp);
  83                 endp = (char *)lep;
  84
  85                 /*
  86                  * The number of leaf entries is limited by the size of the
  87                  * block and the amount of space used by the data entries.
  88                  * We don't know how much space is used by the data entries yet,
  89                  * so just ensure that the count falls somewhere inside the
  90                  * block right now.
  91                  */
  92                 XFS_WANT_CORRUPTED_RETURN(mp, be32_to_cpu(btp->count) <
  93                         ((char *)btp - p) / sizeof(struct xfs_dir2_leaf_entry));
  94                 break;
  95         case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
  96         case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
  97                 endp = (char *)hdr + geo->blksize;
  98                 break;
  99         default:
 100                 XFS_ERROR_REPORT("Bad Magic", XFS_ERRLEVEL_LOW, mp);
 101                 return -EFSCORRUPTED;
 102         }
 103
 104         /*
 105          * Account for zero bestfree entries.
 106          */
 107         bf = ops->data_bestfree_p(hdr);
 108         count = lastfree = freeseen = 0;
 109         if (!bf[0].length) {
 110                 XFS_WANT_CORRUPTED_RETURN(mp, !bf[0].offset);
 111                 freeseen |= 1 << 0;
 112         }
 113         if (!bf[1].length) {
 114                 XFS_WANT_CORRUPTED_RETURN(mp, !bf[1].offset);
 115                 freeseen |= 1 << 1;
 116         }
 117         if (!bf[2].length) {
 118                 XFS_WANT_CORRUPTED_RETURN(mp, !bf[2].offset);
 119                 freeseen |= 1 << 2;
 120         }
 121
 122         XFS_WANT_CORRUPTED_RETURN(mp, be16_to_cpu(bf[0].length) >=
 123                                                 be16_to_cpu(bf[1].length));
 124         XFS_WANT_CORRUPTED_RETURN(mp, be16_to_cpu(bf[1].length) >=
 125                                                 be16_to_cpu(bf[2].length));
 126         /*
 127          * Loop over the data/unused entries.
 128          */
 129         while (p < endp) {
 130                 dup = (xfs_dir2_data_unused_t *)p;
 131                 /*
 132                  * If it's unused, look for the space in the bestfree table.
 133                  * If we find it, account for that, else make sure it
 134                  * doesn't need to be there.
 135                  */
 136                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 137                         XFS_WANT_CORRUPTED_RETURN(mp, lastfree == 0);
 138                         XFS_WANT_CORRUPTED_RETURN(mp,
 139                                 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) ==
 140                                                (char *)dup - (char *)hdr);
 141                         dfp = xfs_dir2_data_freefind(hdr, bf, dup);
 142                         if (dfp) {
 143                                 i = (int)(dfp - bf);
 144                                 XFS_WANT_CORRUPTED_RETURN(mp,
 145                                         (freeseen & (1 << i)) == 0);
 146                                 freeseen |= 1 << i;
 147                         } else {
 148                                 XFS_WANT_CORRUPTED_RETURN(mp,
 149                                         be16_to_cpu(dup->length) <=
 150                                                 be16_to_cpu(bf[2].length));
 151                         }
 152                         p += be16_to_cpu(dup->length);
 153                         lastfree = 1;
 154                         continue;
 155                 }
 156                 /*
 157                  * It's a real entry.  Validate the fields.
 158                  * If this is a block directory then make sure it's
 159                  * in the leaf section of the block.
 160                  * The linear search is crude but this is DEBUG code.
 161                  */
 162                 dep = (xfs_dir2_data_entry_t *)p;
 163                 XFS_WANT_CORRUPTED_RETURN(mp, dep->namelen != 0);
 164                 XFS_WANT_CORRUPTED_RETURN(mp,
 165                         !xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber)));
 166                 XFS_WANT_CORRUPTED_RETURN(mp,
 167                         be16_to_cpu(*ops->data_entry_tag_p(dep)) ==
 168                                                (char *)dep - (char *)hdr);
 169                 XFS_WANT_CORRUPTED_RETURN(mp,
 170                                 ops->data_get_ftype(dep) < XFS_DIR3_FT_MAX);
 171                 count++;
 172                 lastfree = 0;
 173                 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 174                     hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 175                         addr = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
 176                                                 (xfs_dir2_data_aoff_t)
 177                                                 ((char *)dep - (char *)hdr));
 178                         name.name = dep->name;
 179                         name.len = dep->namelen;
 180                         hash = mp->m_dirnameops->hashname(&name);
 181                         for (i = 0; i < be32_to_cpu(btp->count); i++) {
 182                                 if (be32_to_cpu(lep[i].address) == addr &&
 183                                     be32_to_cpu(lep[i].hashval) == hash)
 184                                         break;
 185                         }
 186                         XFS_WANT_CORRUPTED_RETURN(mp,
 187                                                   i < be32_to_cpu(btp->count));
 188                 }
 189                 p += ops->data_entsize(dep->namelen);
 190         }
 191         /*
 192          * Need to have seen all the entries and all the bestfree slots.
 193          */
 194         XFS_WANT_CORRUPTED_RETURN(mp, freeseen == 7);
 195         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 196             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 197                 for (i = stale = 0; i < be32_to_cpu(btp->count); i++) {
 198                         if (lep[i].address ==
 199                             cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 200                                 stale++;
 201                         if (i > 0)
 202                                 XFS_WANT_CORRUPTED_RETURN(mp,
 203                                         be32_to_cpu(lep[i].hashval) >=
 204                                                 be32_to_cpu(lep[i - 1].hashval));
 205                 }
 206                 XFS_WANT_CORRUPTED_RETURN(mp, count ==
 207                         be32_to_cpu(btp->count) - be32_to_cpu(btp->stale));
 208                 XFS_WANT_CORRUPTED_RETURN(mp, stale == be32_to_cpu(btp->stale));
 209         }
 210         return 0;
 211 }
 212
 213 static bool
 214 xfs_dir3_data_verify(
 215         struct xfs_buf          *bp)
 216 {
 217         struct xfs_mount        *mp = bp->b_target->bt_mount;
 218         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 219
 220         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 221                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 222                         return false;
 223                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
 224                         return false;
 225                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 226                         return false;
 227         } else {
 228                 if (hdr3->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC))
 229                         return false;
 230         }
 231         if (__xfs_dir3_data_check(NULL, bp))
 232                 return false;
 233         return true;
 234 }
 235
 236 /*
 237  * Readahead of the first block of the directory when it is opened is completely
 238  * oblivious to the format of the directory. Hence we can either get a block
 239  * format buffer or a data format buffer on readahead.
 240  */
 241 static void
 242 xfs_dir3_data_reada_verify(
 243         struct xfs_buf          *bp)
 244 {
 245         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 246
 247         switch (hdr->magic) {
 248         case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
 249         case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
 250                 bp->b_ops = &xfs_dir3_block_buf_ops;
 251                 bp->b_ops->verify_read(bp);
 252                 return;
 253         case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
 254         case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
 255                 xfs_dir3_data_verify(bp);
 256                 return;
 257         default:
 258                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 259                 xfs_verifier_error(bp);
 260                 break;
 261         }
 262 }
 263
 264 static void
 265 xfs_dir3_data_read_verify(
 266         struct xfs_buf  *bp)
 267 {
 268         struct xfs_mount        *mp = bp->b_target->bt_mount;
 269
 270         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 271              !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
 272                  xfs_buf_ioerror(bp, -EFSBADCRC);
 273         else if (!xfs_dir3_data_verify(bp))
 274                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 275
 276         if (bp->b_error)
 277                 xfs_verifier_error(bp);
 278 }
 279
 280 static void
 281 xfs_dir3_data_write_verify(
 282         struct xfs_buf  *bp)
 283 {
 284         struct xfs_mount        *mp = bp->b_target->bt_mount;
 285         struct xfs_buf_log_item *bip = bp->b_fspriv;
 286         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 287
 288         if (!xfs_dir3_data_verify(bp)) {
 289                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 290                 xfs_verifier_error(bp);
 291                 return;
 292         }
 293
 294         if (!xfs_sb_version_hascrc(&mp->m_sb))
 295                 return;
 296
 297         if (bip)
 298                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 299
 300         xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
 301 }
 302
 303 const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
 304         .verify_read = xfs_dir3_data_read_verify,
 305         .verify_write = xfs_dir3_data_write_verify,
 306 };
 307
 308 static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
 309         .verify_read = xfs_dir3_data_reada_verify,
 310         .verify_write = xfs_dir3_data_write_verify,
 311 };
 312
 313
 314 int
 315 xfs_dir3_data_read(
 316         struct xfs_trans        *tp,
 317         struct xfs_inode        *dp,
 318         xfs_dablk_t             bno,
 319         xfs_daddr_t             mapped_bno,
 320         struct xfs_buf          **bpp)
 321 {
 322         int                     err;
 323
 324         err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
 325                                 XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
 326         if (!err && tp)
 327                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
 328         return err;
 329 }
 330
 331 int
 332 xfs_dir3_data_readahead(
 333         struct xfs_inode        *dp,
 334         xfs_dablk_t             bno,
 335         xfs_daddr_t             mapped_bno)
 336 {
 337         return xfs_da_reada_buf(dp, bno, mapped_bno,
 338                                 XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
 339 }
 340
 341 /*
 342  * Given a data block and an unused entry from that block,
 343  * return the bestfree entry if any that corresponds to it.
 344  */
 345 xfs_dir2_data_free_t *
 346 xfs_dir2_data_freefind(
 347         struct xfs_dir2_data_hdr *hdr,          /* data block header */
 348         struct xfs_dir2_data_free *bf,          /* bestfree table pointer */
 349         struct xfs_dir2_data_unused *dup)       /* unused space */
 350 {
 351         xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
 352         xfs_dir2_data_aoff_t    off;            /* offset value needed */
 353 #ifdef DEBUG
 354         int                     matched;        /* matched the value */
 355         int                     seenzero;       /* saw a 0 bestfree entry */
 356 #endif
 357
 358         off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
 359
 360 #ifdef DEBUG
 361         /*
 362          * Validate some consistency in the bestfree table.
 363          * Check order, non-overlapping entries, and if we find the
 364          * one we're looking for it has to be exact.
 365          */
 366         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 367                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 368                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 369                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 370         for (dfp = &bf[0], seenzero = matched = 0;
 371              dfp < &bf[XFS_DIR2_DATA_FD_COUNT];
 372              dfp++) {
 373                 if (!dfp->offset) {
 374                         ASSERT(!dfp->length);
 375                         seenzero = 1;
 376                         continue;
 377                 }
 378                 ASSERT(seenzero == 0);
 379                 if (be16_to_cpu(dfp->offset) == off) {
 380                         matched = 1;
 381                         ASSERT(dfp->length == dup->length);
 382                 } else if (off < be16_to_cpu(dfp->offset))
 383                         ASSERT(off + be16_to_cpu(dup->length) <= be16_to_cpu(dfp->offset));
 384                 else
 385                         ASSERT(be16_to_cpu(dfp->offset) + be16_to_cpu(dfp->length) <= off);
 386                 ASSERT(matched || be16_to_cpu(dfp->length) >= be16_to_cpu(dup->length));
 387                 if (dfp > &bf[0])
 388                         ASSERT(be16_to_cpu(dfp[-1].length) >= be16_to_cpu(dfp[0].length));
 389         }
 390 #endif
 391         /*
 392          * If this is smaller than the smallest bestfree entry,
 393          * it can't be there since they're sorted.
 394          */
 395         if (be16_to_cpu(dup->length) <
 396             be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 397                 return NULL;
 398         /*
 399          * Look at the three bestfree entries for our guy.
 400          */
 401         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 402                 if (!dfp->offset)
 403                         return NULL;
 404                 if (be16_to_cpu(dfp->offset) == off)
 405                         return dfp;
 406         }
 407         /*
 408          * Didn't find it.  This only happens if there are duplicate lengths.
 409          */
 410         return NULL;
 411 }
 412
 413 /*
 414  * Insert an unused-space entry into the bestfree table.
 415  */
 416 xfs_dir2_data_free_t *                          /* entry inserted */
 417 xfs_dir2_data_freeinsert(
 418         struct xfs_dir2_data_hdr *hdr,          /* data block pointer */
 419         struct xfs_dir2_data_free *dfp,         /* bestfree table pointer */
 420         struct xfs_dir2_data_unused *dup,       /* unused space */
 421         int                     *loghead)       /* log the data header (out) */
 422 {
 423         xfs_dir2_data_free_t    new;            /* new bestfree entry */
 424
 425         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 426                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 427                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 428                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 429
 430         new.length = dup->length;
 431         new.offset = cpu_to_be16((char *)dup - (char *)hdr);
 432
 433         /*
 434          * Insert at position 0, 1, or 2; or not at all.
 435          */
 436         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
 437                 dfp[2] = dfp[1];
 438                 dfp[1] = dfp[0];
 439                 dfp[0] = new;
 440                 *loghead = 1;
 441                 return &dfp[0];
 442         }
 443         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
 444                 dfp[2] = dfp[1];
 445                 dfp[1] = new;
 446                 *loghead = 1;
 447                 return &dfp[1];
 448         }
 449         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
 450                 dfp[2] = new;
 451                 *loghead = 1;
 452                 return &dfp[2];
 453         }
 454         return NULL;
 455 }
 456
 457 /*
 458  * Remove a bestfree entry from the table.
 459  */
 460 STATIC void
 461 xfs_dir2_data_freeremove(
 462         struct xfs_dir2_data_hdr *hdr,          /* data block header */
 463         struct xfs_dir2_data_free *bf,          /* bestfree table pointer */
 464         struct xfs_dir2_data_free *dfp,         /* bestfree entry pointer */
 465         int                     *loghead)       /* out: log data header */
 466 {
 467
 468         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 469                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 470                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 471                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 472
 473         /*
 474          * It's the first entry, slide the next 2 up.
 475          */
 476         if (dfp == &bf[0]) {
 477                 bf[0] = bf[1];
 478                 bf[1] = bf[2];
 479         }
 480         /*
 481          * It's the second entry, slide the 3rd entry up.
 482          */
 483         else if (dfp == &bf[1])
 484                 bf[1] = bf[2];
 485         /*
 486          * Must be the last entry.
 487          */
 488         else
 489                 ASSERT(dfp == &bf[2]);
 490         /*
 491          * Clear the 3rd entry, must be zero now.
 492          */
 493         bf[2].length = 0;
 494         bf[2].offset = 0;
 495         *loghead = 1;
 496 }
 497
 498 /*
 499  * Given a data block, reconstruct its bestfree map.
 500  */
 501 void
 502 xfs_dir2_data_freescan(
 503         struct xfs_inode        *dp,
 504         struct xfs_dir2_data_hdr *hdr,
 505         int                     *loghead)
 506 {
 507         xfs_dir2_block_tail_t   *btp;           /* block tail */
 508         xfs_dir2_data_entry_t   *dep;           /* active data entry */
 509         xfs_dir2_data_unused_t  *dup;           /* unused data entry */
 510         struct xfs_dir2_data_free *bf;
 511         char                    *endp;          /* end of block's data */
 512         char                    *p;             /* current entry pointer */
 513         struct xfs_da_geometry  *geo = dp->i_mount->m_dir_geo;
 514
 515         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 516                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 517                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 518                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 519
 520         /*
 521          * Start by clearing the table.
 522          */
 523         bf = dp->d_ops->data_bestfree_p(hdr);
 524         memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
 525         *loghead = 1;
 526         /*
 527          * Set up pointers.
 528          */
 529         p = (char *)dp->d_ops->data_entry_p(hdr);
 530         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 531             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 532                 btp = xfs_dir2_block_tail_p(geo, hdr);
 533                 endp = (char *)xfs_dir2_block_leaf_p(btp);
 534         } else
 535                 endp = (char *)hdr + geo->blksize;
 536         /*
 537          * Loop over the block's entries.
 538          */
 539         while (p < endp) {
 540                 dup = (xfs_dir2_data_unused_t *)p;
 541                 /*
 542                  * If it's a free entry, insert it.
 543                  */
 544                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 545                         ASSERT((char *)dup - (char *)hdr ==
 546                                be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 547                         xfs_dir2_data_freeinsert(hdr, bf, dup, loghead);
 548                         p += be16_to_cpu(dup->length);
 549                 }
 550                 /*
 551                  * For active entries, check their tags and skip them.
 552                  */
 553                 else {
 554                         dep = (xfs_dir2_data_entry_t *)p;
 555                         ASSERT((char *)dep - (char *)hdr ==
 556                                be16_to_cpu(*dp->d_ops->data_entry_tag_p(dep)));
 557                         p += dp->d_ops->data_entsize(dep->namelen);
 558                 }
 559         }
 560 }
 561
 562 /*
 563  * Initialize a data block at the given block number in the directory.
 564  * Give back the buffer for the created block.
 565  */
 566 int                                             /* error */
 567 xfs_dir3_data_init(
 568         xfs_da_args_t           *args,          /* directory operation args */
 569         xfs_dir2_db_t           blkno,          /* logical dir block number */
 570         struct xfs_buf          **bpp)          /* output block buffer */
 571 {
 572         struct xfs_buf          *bp;            /* block buffer */
 573         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 574         xfs_inode_t             *dp;            /* incore directory inode */
 575         xfs_dir2_data_unused_t  *dup;           /* unused entry pointer */
 576         struct xfs_dir2_data_free *bf;
 577         int                     error;          /* error return value */
 578         int                     i;              /* bestfree index */
 579         xfs_mount_t             *mp;            /* filesystem mount point */
 580         xfs_trans_t             *tp;            /* transaction pointer */
 581         int                     t;              /* temp */
 582
 583         dp = args->dp;
 584         mp = dp->i_mount;
 585         tp = args->trans;
 586         /*
 587          * Get the buffer set up for the block.
 588          */
 589         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, blkno),
 590                                -1, &bp, XFS_DATA_FORK);
 591         if (error)
 592                 return error;
 593         bp->b_ops = &xfs_dir3_data_buf_ops;
 594         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
 595
 596         /*
 597          * Initialize the header.
 598          */
 599         hdr = bp->b_addr;
 600         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 601                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 602
 603                 memset(hdr3, 0, sizeof(*hdr3));
 604                 hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 605                 hdr3->blkno = cpu_to_be64(bp->b_bn);
 606                 hdr3->owner = cpu_to_be64(dp->i_ino);
 607                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
 608
 609         } else
 610                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 611
 612         bf = dp->d_ops->data_bestfree_p(hdr);
 613         bf[0].offset = cpu_to_be16(dp->d_ops->data_entry_offset);
 614         for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
 615                 bf[i].length = 0;
 616                 bf[i].offset = 0;
 617         }
 618
 619         /*
 620          * Set up an unused entry for the block's body.
 621          */
 622         dup = dp->d_ops->data_unused_p(hdr);
 623         dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 624
 625         t = args->geo->blksize - (uint)dp->d_ops->data_entry_offset;
 626         bf[0].length = cpu_to_be16(t);
 627         dup->length = cpu_to_be16(t);
 628         *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
 629         /*
 630          * Log it and return it.
 631          */
 632         xfs_dir2_data_log_header(args, bp);
 633         xfs_dir2_data_log_unused(args, bp, dup);
 634         *bpp = bp;
 635         return 0;
 636 }
 637
 638 /*
 639  * Log an active data entry from the block.
 640  */
 641 void
 642 xfs_dir2_data_log_entry(
 643         struct xfs_da_args      *args,
 644         struct xfs_buf          *bp,
 645         xfs_dir2_data_entry_t   *dep)           /* data entry pointer */
 646 {
 647         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 648
 649         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 650                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 651                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 652                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 653
 654         xfs_trans_log_buf(args->trans, bp, (uint)((char *)dep - (char *)hdr),
 655                 (uint)((char *)(args->dp->d_ops->data_entry_tag_p(dep) + 1) -
 656                        (char *)hdr - 1));
 657 }
 658
 659 /*
 660  * Log a data block header.
 661  */
 662 void
 663 xfs_dir2_data_log_header(
 664         struct xfs_da_args      *args,
 665         struct xfs_buf          *bp)
 666 {
 667 #ifdef DEBUG
 668         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 669
 670         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 671                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 672                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 673                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 674 #endif
 675
 676         xfs_trans_log_buf(args->trans, bp, 0,
 677                           args->dp->d_ops->data_entry_offset - 1);
 678 }
 679
 680 /*
 681  * Log a data unused entry.
 682  */
 683 void
 684 xfs_dir2_data_log_unused(
 685         struct xfs_da_args      *args,
 686         struct xfs_buf          *bp,
 687         xfs_dir2_data_unused_t  *dup)           /* data unused pointer */
 688 {
 689         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 690
 691         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 692                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 693                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 694                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 695
 696         /*
 697          * Log the first part of the unused entry.
 698          */
 699         xfs_trans_log_buf(args->trans, bp, (uint)((char *)dup - (char *)hdr),
 700                 (uint)((char *)&dup->length + sizeof(dup->length) -
 701                        1 - (char *)hdr));
 702         /*
 703          * Log the end (tag) of the unused entry.
 704          */
 705         xfs_trans_log_buf(args->trans, bp,
 706                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
 707                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
 708                        sizeof(xfs_dir2_data_off_t) - 1));
 709 }
 710
 711 /*
 712  * Make a byte range in the data block unused.
 713  * Its current contents are unimportant.
 714  */
 715 void
 716 xfs_dir2_data_make_free(
 717         struct xfs_da_args      *args,
 718         struct xfs_buf          *bp,
 719         xfs_dir2_data_aoff_t    offset,         /* starting byte offset */
 720         xfs_dir2_data_aoff_t    len,            /* length in bytes */
 721         int                     *needlogp,      /* out: log header */
 722         int                     *needscanp)     /* out: regen bestfree */
 723 {
 724         xfs_dir2_data_hdr_t     *hdr;           /* data block pointer */
 725         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 726         char                    *endptr;        /* end of data area */
 727         int                     needscan;       /* need to regen bestfree */
 728         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 729         xfs_dir2_data_unused_t  *postdup;       /* unused entry after us */
 730         xfs_dir2_data_unused_t  *prevdup;       /* unused entry before us */
 731         struct xfs_dir2_data_free *bf;
 732
 733         hdr = bp->b_addr;
 734
 735         /*
 736          * Figure out where the end of the data area is.
 737          */
 738         if (hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 739             hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 740                 endptr = (char *)hdr + args->geo->blksize;
 741         else {
 742                 xfs_dir2_block_tail_t   *btp;   /* block tail */
 743
 744                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 745                         hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 746                 btp = xfs_dir2_block_tail_p(args->geo, hdr);
 747                 endptr = (char *)xfs_dir2_block_leaf_p(btp);
 748         }
 749         /*
 750          * If this isn't the start of the block, then back up to
 751          * the previous entry and see if it's free.
 752          */
 753         if (offset > args->dp->d_ops->data_entry_offset) {
 754                 __be16                  *tagp;  /* tag just before us */
 755
 756                 tagp = (__be16 *)((char *)hdr + offset) - 1;
 757                 prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 758                 if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 759                         prevdup = NULL;
 760         } else
 761                 prevdup = NULL;
 762         /*
 763          * If this isn't the end of the block, see if the entry after
 764          * us is free.
 765          */
 766         if ((char *)hdr + offset + len < endptr) {
 767                 postdup =
 768                         (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 769                 if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 770                         postdup = NULL;
 771         } else
 772                 postdup = NULL;
 773         ASSERT(*needscanp == 0);
 774         needscan = 0;
 775         /*
 776          * Previous and following entries are both free,
 777          * merge everything into a single free entry.
 778          */
 779         bf = args->dp->d_ops->data_bestfree_p(hdr);
 780         if (prevdup && postdup) {
 781                 xfs_dir2_data_free_t    *dfp2;  /* another bestfree pointer */
 782
 783                 /*
 784                  * See if prevdup and/or postdup are in bestfree table.
 785                  */
 786                 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
 787                 dfp2 = xfs_dir2_data_freefind(hdr, bf, postdup);
 788                 /*
 789                  * We need a rescan unless there are exactly 2 free entries
 790                  * namely our two.  Then we know what's happening, otherwise
 791                  * since the third bestfree is there, there might be more
 792                  * entries.
 793                  */
 794                 needscan = (bf[2].length != 0);
 795                 /*
 796                  * Fix up the new big freespace.
 797                  */
 798                 be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
 799                 *xfs_dir2_data_unused_tag_p(prevdup) =
 800                         cpu_to_be16((char *)prevdup - (char *)hdr);
 801                 xfs_dir2_data_log_unused(args, bp, prevdup);
 802                 if (!needscan) {
 803                         /*
 804                          * Has to be the case that entries 0 and 1 are
 805                          * dfp and dfp2 (don't know which is which), and
 806                          * entry 2 is empty.
 807                          * Remove entry 1 first then entry 0.
 808                          */
 809                         ASSERT(dfp && dfp2);
 810                         if (dfp == &bf[1]) {
 811                                 dfp = &bf[0];
 812                                 ASSERT(dfp2 == dfp);
 813                                 dfp2 = &bf[1];
 814                         }
 815                         xfs_dir2_data_freeremove(hdr, bf, dfp2, needlogp);
 816                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 817                         /*
 818                          * Now insert the new entry.
 819                          */
 820                         dfp = xfs_dir2_data_freeinsert(hdr, bf, prevdup,
 821                                                        needlogp);
 822                         ASSERT(dfp == &bf[0]);
 823                         ASSERT(dfp->length == prevdup->length);
 824                         ASSERT(!dfp[1].length);
 825                         ASSERT(!dfp[2].length);
 826                 }
 827         }
 828         /*
 829          * The entry before us is free, merge with it.
 830          */
 831         else if (prevdup) {
 832                 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
 833                 be16_add_cpu(&prevdup->length, len);
 834                 *xfs_dir2_data_unused_tag_p(prevdup) =
 835                         cpu_to_be16((char *)prevdup - (char *)hdr);
 836                 xfs_dir2_data_log_unused(args, bp, prevdup);
 837                 /*
 838                  * If the previous entry was in the table, the new entry
 839                  * is longer, so it will be in the table too.  Remove
 840                  * the old one and add the new one.
 841                  */
 842                 if (dfp) {
 843                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 844                         xfs_dir2_data_freeinsert(hdr, bf, prevdup, needlogp);
 845                 }
 846                 /*
 847                  * Otherwise we need a scan if the new entry is big enough.
 848                  */
 849                 else {
 850                         needscan = be16_to_cpu(prevdup->length) >
 851                                    be16_to_cpu(bf[2].length);
 852                 }
 853         }
 854         /*
 855          * The following entry is free, merge with it.
 856          */
 857         else if (postdup) {
 858                 dfp = xfs_dir2_data_freefind(hdr, bf, postdup);
 859                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 860                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 861                 newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
 862                 *xfs_dir2_data_unused_tag_p(newdup) =
 863                         cpu_to_be16((char *)newdup - (char *)hdr);
 864                 xfs_dir2_data_log_unused(args, bp, newdup);
 865                 /*
 866                  * If the following entry was in the table, the new entry
 867                  * is longer, so it will be in the table too.  Remove
 868                  * the old one and add the new one.
 869                  */
 870                 if (dfp) {
 871                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 872                         xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
 873                 }
 874                 /*
 875                  * Otherwise we need a scan if the new entry is big enough.
 876                  */
 877                 else {
 878                         needscan = be16_to_cpu(newdup->length) >
 879                                    be16_to_cpu(bf[2].length);
 880                 }
 881         }
 882         /*
 883          * Neither neighbor is free.  Make a new entry.
 884          */
 885         else {
 886                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 887                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 888                 newdup->length = cpu_to_be16(len);
 889                 *xfs_dir2_data_unused_tag_p(newdup) =
 890                         cpu_to_be16((char *)newdup - (char *)hdr);
 891                 xfs_dir2_data_log_unused(args, bp, newdup);
 892                 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
 893         }
 894         *needscanp = needscan;
 895 }
 896
 897 /*
 898  * Take a byte range out of an existing unused space and make it un-free.
 899  */
 900 void
 901 xfs_dir2_data_use_free(
 902         struct xfs_da_args      *args,
 903         struct xfs_buf          *bp,
 904         xfs_dir2_data_unused_t  *dup,           /* unused entry */
 905         xfs_dir2_data_aoff_t    offset,         /* starting offset to use */
 906         xfs_dir2_data_aoff_t    len,            /* length to use */
 907         int                     *needlogp,      /* out: need to log header */
 908         int                     *needscanp)     /* out: need regen bestfree */
 909 {
 910         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 911         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 912         int                     matchback;      /* matches end of freespace */
 913         int                     matchfront;     /* matches start of freespace */
 914         int                     needscan;       /* need to regen bestfree */
 915         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 916         xfs_dir2_data_unused_t  *newdup2;       /* another new unused entry */
 917         int                     oldlen;         /* old unused entry's length */
 918         struct xfs_dir2_data_free *bf;
 919
 920         hdr = bp->b_addr;
 921         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 922                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 923                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 924                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 925         ASSERT(be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG);
 926         ASSERT(offset >= (char *)dup - (char *)hdr);
 927         ASSERT(offset + len <= (char *)dup + be16_to_cpu(dup->length) - (char *)hdr);
 928         ASSERT((char *)dup - (char *)hdr == be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 929         /*
 930          * Look up the entry in the bestfree table.
 931          */
 932         oldlen = be16_to_cpu(dup->length);
 933         bf = args->dp->d_ops->data_bestfree_p(hdr);
 934         dfp = xfs_dir2_data_freefind(hdr, bf, dup);
 935         ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
 936         /*
 937          * Check for alignment with front and back of the entry.
 938          */
 939         matchfront = (char *)dup - (char *)hdr == offset;
 940         matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
 941         ASSERT(*needscanp == 0);
 942         needscan = 0;
 943         /*
 944          * If we matched it exactly we just need to get rid of it from
 945          * the bestfree table.
 946          */
 947         if (matchfront && matchback) {
 948                 if (dfp) {
 949                         needscan = (bf[2].offset != 0);
 950                         if (!needscan)
 951                                 xfs_dir2_data_freeremove(hdr, bf, dfp,
 952                                                          needlogp);
 953                 }
 954         }
 955         /*
 956          * We match the first part of the entry.
 957          * Make a new entry with the remaining freespace.
 958          */
 959         else if (matchfront) {
 960                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 961                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 962                 newdup->length = cpu_to_be16(oldlen - len);
 963                 *xfs_dir2_data_unused_tag_p(newdup) =
 964                         cpu_to_be16((char *)newdup - (char *)hdr);
 965                 xfs_dir2_data_log_unused(args, bp, newdup);
 966                 /*
 967                  * If it was in the table, remove it and add the new one.
 968                  */
 969                 if (dfp) {
 970                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 971                         dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
 972                                                        needlogp);
 973                         ASSERT(dfp != NULL);
 974                         ASSERT(dfp->length == newdup->length);
 975                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 976                         /*
 977                          * If we got inserted at the last slot,
 978                          * that means we don't know if there was a better
 979                          * choice for the last slot, or not.  Rescan.
 980                          */
 981                         needscan = dfp == &bf[2];
 982                 }
 983         }
 984         /*
 985          * We match the last part of the entry.
 986          * Trim the allocated space off the tail of the entry.
 987          */
 988         else if (matchback) {
 989                 newdup = dup;
 990                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
 991                 *xfs_dir2_data_unused_tag_p(newdup) =
 992                         cpu_to_be16((char *)newdup - (char *)hdr);
 993                 xfs_dir2_data_log_unused(args, bp, newdup);
 994                 /*
 995                  * If it was in the table, remove it and add the new one.
 996                  */
 997                 if (dfp) {
 998                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 999                         dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
1000                                                        needlogp);
1001                         ASSERT(dfp != NULL);
1002                         ASSERT(dfp->length == newdup->length);
1003                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
1004                         /*
1005                          * If we got inserted at the last slot,
1006                          * that means we don't know if there was a better
1007                          * choice for the last slot, or not.  Rescan.
1008                          */
1009                         needscan = dfp == &bf[2];
1010                 }
1011         }
1012         /*
1013          * Poking out the middle of an entry.
1014          * Make two new entries.
1015          */
1016         else {
1017                 newdup = dup;
1018                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1019                 *xfs_dir2_data_unused_tag_p(newdup) =
1020                         cpu_to_be16((char *)newdup - (char *)hdr);
1021                 xfs_dir2_data_log_unused(args, bp, newdup);
1022                 newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1023                 newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1024                 newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1025                 *xfs_dir2_data_unused_tag_p(newdup2) =
1026                         cpu_to_be16((char *)newdup2 - (char *)hdr);
1027                 xfs_dir2_data_log_unused(args, bp, newdup2);
1028                 /*
1029                  * If the old entry was in the table, we need to scan
1030                  * if the 3rd entry was valid, since these entries
1031                  * are smaller than the old one.
1032                  * If we don't need to scan that means there were 1 or 2
1033                  * entries in the table, and removing the old and adding
1034                  * the 2 new will work.
1035                  */
1036                 if (dfp) {
1037                         needscan = (bf[2].length != 0);
1038                         if (!needscan) {
1039                                 xfs_dir2_data_freeremove(hdr, bf, dfp,
1040                                                          needlogp);
1041                                 xfs_dir2_data_freeinsert(hdr, bf, newdup,
1042                                                          needlogp);
1043                                 xfs_dir2_data_freeinsert(hdr, bf, newdup2,
1044                                                          needlogp);
1045                         }
1046                 }
1047         }
1048         *needscanp = needscan;
1049 }