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 #include "xfs_log.h"
  35
  36 /*
  37  * Check the consistency of the data block.
  38  * The input can also be a block-format directory.
  39  * Return 0 is the buffer is good, otherwise an error.
  40  */
  41 int
  42 __xfs_dir3_data_check(
  43         struct xfs_inode        *dp,            /* incore inode pointer */
  44         struct xfs_buf          *bp)            /* data block's buffer */
  45 {
  46         xfs_dir2_dataptr_t      addr;           /* addr for leaf lookup */
  47         xfs_dir2_data_free_t    *bf;            /* bestfree table */
  48         xfs_dir2_block_tail_t   *btp=NULL;      /* block tail */
  49         int                     count;          /* count of entries found */
  50         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
  51         xfs_dir2_data_entry_t   *dep;           /* data entry */
  52         xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
  53         xfs_dir2_data_unused_t  *dup;           /* unused entry */
  54         char                    *endp;          /* end of useful data */
  55         int                     freeseen;       /* mask of bestfrees seen */
  56         xfs_dahash_t            hash;           /* hash of current name */
  57         int                     i;              /* leaf index */
  58         int                     lastfree;       /* last entry was unused */
  59         xfs_dir2_leaf_entry_t   *lep=NULL;      /* block leaf entries */
  60         xfs_mount_t             *mp;            /* filesystem mount point */
  61         char                    *p;             /* current data position */
  62         int                     stale;          /* count of stale leaves */
  63         struct xfs_name         name;
  64         const struct xfs_dir_ops *ops;
  65         struct xfs_da_geometry  *geo;
  66
  67         mp = bp->b_target->bt_mount;
  68         geo = mp->m_dir_geo;
  69
  70         /*
  71          * We can be passed a null dp here from a verifier, so we need to go the
  72          * hard way to get them.
  73          */
  74         ops = xfs_dir_get_ops(mp, dp);
  75
  76         hdr = bp->b_addr;
  77         p = (char *)ops->data_entry_p(hdr);
  78
  79         switch (hdr->magic) {
  80         case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
  81         case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
  82                 btp = xfs_dir2_block_tail_p(geo, hdr);
  83                 lep = xfs_dir2_block_leaf_p(btp);
  84                 endp = (char *)lep;
  85
  86                 /*
  87                  * The number of leaf entries is limited by the size of the
  88                  * block and the amount of space used by the data entries.
  89                  * We don't know how much space is used by the data entries yet,
  90                  * so just ensure that the count falls somewhere inside the
  91                  * block right now.
  92                  */
  93                 XFS_WANT_CORRUPTED_RETURN(mp, be32_to_cpu(btp->count) <
  94                         ((char *)btp - p) / sizeof(struct xfs_dir2_leaf_entry));
  95                 break;
  96         case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
  97         case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
  98                 endp = (char *)hdr + geo->blksize;
  99                 break;
 100         default:
 101                 XFS_ERROR_REPORT("Bad Magic", XFS_ERRLEVEL_LOW, mp);
 102                 return -EFSCORRUPTED;
 103         }
 104
 105         /*
 106          * Account for zero bestfree entries.
 107          */
 108         bf = ops->data_bestfree_p(hdr);
 109         count = lastfree = freeseen = 0;
 110         if (!bf[0].length) {
 111                 XFS_WANT_CORRUPTED_RETURN(mp, !bf[0].offset);
 112                 freeseen |= 1 << 0;
 113         }
 114         if (!bf[1].length) {
 115                 XFS_WANT_CORRUPTED_RETURN(mp, !bf[1].offset);
 116                 freeseen |= 1 << 1;
 117         }
 118         if (!bf[2].length) {
 119                 XFS_WANT_CORRUPTED_RETURN(mp, !bf[2].offset);
 120                 freeseen |= 1 << 2;
 121         }
 122
 123         XFS_WANT_CORRUPTED_RETURN(mp, be16_to_cpu(bf[0].length) >=
 124                                                 be16_to_cpu(bf[1].length));
 125         XFS_WANT_CORRUPTED_RETURN(mp, be16_to_cpu(bf[1].length) >=
 126                                                 be16_to_cpu(bf[2].length));
 127         /*
 128          * Loop over the data/unused entries.
 129          */
 130         while (p < endp) {
 131                 dup = (xfs_dir2_data_unused_t *)p;
 132                 /*
 133                  * If it's unused, look for the space in the bestfree table.
 134                  * If we find it, account for that, else make sure it
 135                  * doesn't need to be there.
 136                  */
 137                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 138                         XFS_WANT_CORRUPTED_RETURN(mp, lastfree == 0);
 139                         XFS_WANT_CORRUPTED_RETURN(mp,
 140                                 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) ==
 141                                                (char *)dup - (char *)hdr);
 142                         dfp = xfs_dir2_data_freefind(hdr, bf, dup);
 143                         if (dfp) {
 144                                 i = (int)(dfp - bf);
 145                                 XFS_WANT_CORRUPTED_RETURN(mp,
 146                                         (freeseen & (1 << i)) == 0);
 147                                 freeseen |= 1 << i;
 148                         } else {
 149                                 XFS_WANT_CORRUPTED_RETURN(mp,
 150                                         be16_to_cpu(dup->length) <=
 151                                                 be16_to_cpu(bf[2].length));
 152                         }
 153                         p += be16_to_cpu(dup->length);
 154                         lastfree = 1;
 155                         continue;
 156                 }
 157                 /*
 158                  * It's a real entry.  Validate the fields.
 159                  * If this is a block directory then make sure it's
 160                  * in the leaf section of the block.
 161                  * The linear search is crude but this is DEBUG code.
 162                  */
 163                 dep = (xfs_dir2_data_entry_t *)p;
 164                 XFS_WANT_CORRUPTED_RETURN(mp, dep->namelen != 0);
 165                 XFS_WANT_CORRUPTED_RETURN(mp,
 166                         !xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber)));
 167                 XFS_WANT_CORRUPTED_RETURN(mp,
 168                         be16_to_cpu(*ops->data_entry_tag_p(dep)) ==
 169                                                (char *)dep - (char *)hdr);
 170                 XFS_WANT_CORRUPTED_RETURN(mp,
 171                                 ops->data_get_ftype(dep) < XFS_DIR3_FT_MAX);
 172                 count++;
 173                 lastfree = 0;
 174                 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 175                     hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 176                         addr = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
 177                                                 (xfs_dir2_data_aoff_t)
 178                                                 ((char *)dep - (char *)hdr));
 179                         name.name = dep->name;
 180                         name.len = dep->namelen;
 181                         hash = mp->m_dirnameops->hashname(&name);
 182                         for (i = 0; i < be32_to_cpu(btp->count); i++) {
 183                                 if (be32_to_cpu(lep[i].address) == addr &&
 184                                     be32_to_cpu(lep[i].hashval) == hash)
 185                                         break;
 186                         }
 187                         XFS_WANT_CORRUPTED_RETURN(mp,
 188                                                   i < be32_to_cpu(btp->count));
 189                 }
 190                 p += ops->data_entsize(dep->namelen);
 191         }
 192         /*
 193          * Need to have seen all the entries and all the bestfree slots.
 194          */
 195         XFS_WANT_CORRUPTED_RETURN(mp, freeseen == 7);
 196         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 197             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 198                 for (i = stale = 0; i < be32_to_cpu(btp->count); i++) {
 199                         if (lep[i].address ==
 200                             cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 201                                 stale++;
 202                         if (i > 0)
 203                                 XFS_WANT_CORRUPTED_RETURN(mp,
 204                                         be32_to_cpu(lep[i].hashval) >=
 205                                                 be32_to_cpu(lep[i - 1].hashval));
 206                 }
 207                 XFS_WANT_CORRUPTED_RETURN(mp, count ==
 208                         be32_to_cpu(btp->count) - be32_to_cpu(btp->stale));
 209                 XFS_WANT_CORRUPTED_RETURN(mp, stale == be32_to_cpu(btp->stale));
 210         }
 211         return 0;
 212 }
 213
 214 static bool
 215 xfs_dir3_data_verify(
 216         struct xfs_buf          *bp)
 217 {
 218         struct xfs_mount        *mp = bp->b_target->bt_mount;
 219         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 220
 221         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 222                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 223                         return false;
 224                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
 225                         return false;
 226                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 227                         return false;
 228                 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
 229                         return false;
 230         } else {
 231                 if (hdr3->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC))
 232                         return false;
 233         }
 234         if (__xfs_dir3_data_check(NULL, bp))
 235                 return false;
 236         return true;
 237 }
 238
 239 /*
 240  * Readahead of the first block of the directory when it is opened is completely
 241  * oblivious to the format of the directory. Hence we can either get a block
 242  * format buffer or a data format buffer on readahead.
 243  */
 244 static void
 245 xfs_dir3_data_reada_verify(
 246         struct xfs_buf          *bp)
 247 {
 248         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 249
 250         switch (hdr->magic) {
 251         case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
 252         case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
 253                 bp->b_ops = &xfs_dir3_block_buf_ops;
 254                 bp->b_ops->verify_read(bp);
 255                 return;
 256         case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
 257         case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
 258                 bp->b_ops = &xfs_dir3_data_buf_ops;
 259                 bp->b_ops->verify_read(bp);
 260                 return;
 261         default:
 262                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 263                 xfs_verifier_error(bp);
 264                 break;
 265         }
 266 }
 267
 268 static void
 269 xfs_dir3_data_read_verify(
 270         struct xfs_buf  *bp)
 271 {
 272         struct xfs_mount        *mp = bp->b_target->bt_mount;
 273
 274         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 275              !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
 276                  xfs_buf_ioerror(bp, -EFSBADCRC);
 277         else if (!xfs_dir3_data_verify(bp))
 278                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 279
 280         if (bp->b_error)
 281                 xfs_verifier_error(bp);
 282 }
 283
 284 static void
 285 xfs_dir3_data_write_verify(
 286         struct xfs_buf  *bp)
 287 {
 288         struct xfs_mount        *mp = bp->b_target->bt_mount;
 289         struct xfs_buf_log_item *bip = bp->b_fspriv;
 290         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 291
 292         if (!xfs_dir3_data_verify(bp)) {
 293                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 294                 xfs_verifier_error(bp);
 295                 return;
 296         }
 297
 298         if (!xfs_sb_version_hascrc(&mp->m_sb))
 299                 return;
 300
 301         if (bip)
 302                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 303
 304         xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
 305 }
 306
 307 const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
 308         .verify_read = xfs_dir3_data_read_verify,
 309         .verify_write = xfs_dir3_data_write_verify,
 310 };
 311
 312 static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
 313         .verify_read = xfs_dir3_data_reada_verify,
 314         .verify_write = xfs_dir3_data_write_verify,
 315 };
 316
 317
 318 int
 319 xfs_dir3_data_read(
 320         struct xfs_trans        *tp,
 321         struct xfs_inode        *dp,
 322         xfs_dablk_t             bno,
 323         xfs_daddr_t             mapped_bno,
 324         struct xfs_buf          **bpp)
 325 {
 326         int                     err;
 327
 328         err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
 329                                 XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
 330         if (!err && tp)
 331                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
 332         return err;
 333 }
 334
 335 int
 336 xfs_dir3_data_readahead(
 337         struct xfs_inode        *dp,
 338         xfs_dablk_t             bno,
 339         xfs_daddr_t             mapped_bno)
 340 {
 341         return xfs_da_reada_buf(dp, bno, mapped_bno,
 342                                 XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
 343 }
 344
 345 /*
 346  * Given a data block and an unused entry from that block,
 347  * return the bestfree entry if any that corresponds to it.
 348  */
 349 xfs_dir2_data_free_t *
 350 xfs_dir2_data_freefind(
 351         struct xfs_dir2_data_hdr *hdr,          /* data block header */
 352         struct xfs_dir2_data_free *bf,          /* bestfree table pointer */
 353         struct xfs_dir2_data_unused *dup)       /* unused space */
 354 {
 355         xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
 356         xfs_dir2_data_aoff_t    off;            /* offset value needed */
 357 #ifdef DEBUG
 358         int                     matched;        /* matched the value */
 359         int                     seenzero;       /* saw a 0 bestfree entry */
 360 #endif
 361
 362         off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
 363
 364 #ifdef DEBUG
 365         /*
 366          * Validate some consistency in the bestfree table.
 367          * Check order, non-overlapping entries, and if we find the
 368          * one we're looking for it has to be exact.
 369          */
 370         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 371                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 372                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 373                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 374         for (dfp = &bf[0], seenzero = matched = 0;
 375              dfp < &bf[XFS_DIR2_DATA_FD_COUNT];
 376              dfp++) {
 377                 if (!dfp->offset) {
 378                         ASSERT(!dfp->length);
 379                         seenzero = 1;
 380                         continue;
 381                 }
 382                 ASSERT(seenzero == 0);
 383                 if (be16_to_cpu(dfp->offset) == off) {
 384                         matched = 1;
 385                         ASSERT(dfp->length == dup->length);
 386                 } else if (off < be16_to_cpu(dfp->offset))
 387                         ASSERT(off + be16_to_cpu(dup->length) <= be16_to_cpu(dfp->offset));
 388                 else
 389                         ASSERT(be16_to_cpu(dfp->offset) + be16_to_cpu(dfp->length) <= off);
 390                 ASSERT(matched || be16_to_cpu(dfp->length) >= be16_to_cpu(dup->length));
 391                 if (dfp > &bf[0])
 392                         ASSERT(be16_to_cpu(dfp[-1].length) >= be16_to_cpu(dfp[0].length));
 393         }
 394 #endif
 395         /*
 396          * If this is smaller than the smallest bestfree entry,
 397          * it can't be there since they're sorted.
 398          */
 399         if (be16_to_cpu(dup->length) <
 400             be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 401                 return NULL;
 402         /*
 403          * Look at the three bestfree entries for our guy.
 404          */
 405         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 406                 if (!dfp->offset)
 407                         return NULL;
 408                 if (be16_to_cpu(dfp->offset) == off)
 409                         return dfp;
 410         }
 411         /*
 412          * Didn't find it.  This only happens if there are duplicate lengths.
 413          */
 414         return NULL;
 415 }
 416
 417 /*
 418  * Insert an unused-space entry into the bestfree table.
 419  */
 420 xfs_dir2_data_free_t *                          /* entry inserted */
 421 xfs_dir2_data_freeinsert(
 422         struct xfs_dir2_data_hdr *hdr,          /* data block pointer */
 423         struct xfs_dir2_data_free *dfp,         /* bestfree table pointer */
 424         struct xfs_dir2_data_unused *dup,       /* unused space */
 425         int                     *loghead)       /* log the data header (out) */
 426 {
 427         xfs_dir2_data_free_t    new;            /* new bestfree entry */
 428
 429         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 430                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 431                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 432                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 433
 434         new.length = dup->length;
 435         new.offset = cpu_to_be16((char *)dup - (char *)hdr);
 436
 437         /*
 438          * Insert at position 0, 1, or 2; or not at all.
 439          */
 440         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
 441                 dfp[2] = dfp[1];
 442                 dfp[1] = dfp[0];
 443                 dfp[0] = new;
 444                 *loghead = 1;
 445                 return &dfp[0];
 446         }
 447         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
 448                 dfp[2] = dfp[1];
 449                 dfp[1] = new;
 450                 *loghead = 1;
 451                 return &dfp[1];
 452         }
 453         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
 454                 dfp[2] = new;
 455                 *loghead = 1;
 456                 return &dfp[2];
 457         }
 458         return NULL;
 459 }
 460
 461 /*
 462  * Remove a bestfree entry from the table.
 463  */
 464 STATIC void
 465 xfs_dir2_data_freeremove(
 466         struct xfs_dir2_data_hdr *hdr,          /* data block header */
 467         struct xfs_dir2_data_free *bf,          /* bestfree table pointer */
 468         struct xfs_dir2_data_free *dfp,         /* bestfree entry pointer */
 469         int                     *loghead)       /* out: log data header */
 470 {
 471
 472         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 473                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 474                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 475                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 476
 477         /*
 478          * It's the first entry, slide the next 2 up.
 479          */
 480         if (dfp == &bf[0]) {
 481                 bf[0] = bf[1];
 482                 bf[1] = bf[2];
 483         }
 484         /*
 485          * It's the second entry, slide the 3rd entry up.
 486          */
 487         else if (dfp == &bf[1])
 488                 bf[1] = bf[2];
 489         /*
 490          * Must be the last entry.
 491          */
 492         else
 493                 ASSERT(dfp == &bf[2]);
 494         /*
 495          * Clear the 3rd entry, must be zero now.
 496          */
 497         bf[2].length = 0;
 498         bf[2].offset = 0;
 499         *loghead = 1;
 500 }
 501
 502 /*
 503  * Given a data block, reconstruct its bestfree map.
 504  */
 505 void
 506 xfs_dir2_data_freescan(
 507         struct xfs_inode        *dp,
 508         struct xfs_dir2_data_hdr *hdr,
 509         int                     *loghead)
 510 {
 511         xfs_dir2_block_tail_t   *btp;           /* block tail */
 512         xfs_dir2_data_entry_t   *dep;           /* active data entry */
 513         xfs_dir2_data_unused_t  *dup;           /* unused data entry */
 514         struct xfs_dir2_data_free *bf;
 515         char                    *endp;          /* end of block's data */
 516         char                    *p;             /* current entry pointer */
 517         struct xfs_da_geometry  *geo = dp->i_mount->m_dir_geo;
 518
 519         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 520                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 521                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 522                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 523
 524         /*
 525          * Start by clearing the table.
 526          */
 527         bf = dp->d_ops->data_bestfree_p(hdr);
 528         memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
 529         *loghead = 1;
 530         /*
 531          * Set up pointers.
 532          */
 533         p = (char *)dp->d_ops->data_entry_p(hdr);
 534         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 535             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 536                 btp = xfs_dir2_block_tail_p(geo, hdr);
 537                 endp = (char *)xfs_dir2_block_leaf_p(btp);
 538         } else
 539                 endp = (char *)hdr + geo->blksize;
 540         /*
 541          * Loop over the block's entries.
 542          */
 543         while (p < endp) {
 544                 dup = (xfs_dir2_data_unused_t *)p;
 545                 /*
 546                  * If it's a free entry, insert it.
 547                  */
 548                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 549                         ASSERT((char *)dup - (char *)hdr ==
 550                                be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 551                         xfs_dir2_data_freeinsert(hdr, bf, dup, loghead);
 552                         p += be16_to_cpu(dup->length);
 553                 }
 554                 /*
 555                  * For active entries, check their tags and skip them.
 556                  */
 557                 else {
 558                         dep = (xfs_dir2_data_entry_t *)p;
 559                         ASSERT((char *)dep - (char *)hdr ==
 560                                be16_to_cpu(*dp->d_ops->data_entry_tag_p(dep)));
 561                         p += dp->d_ops->data_entsize(dep->namelen);
 562                 }
 563         }
 564 }
 565
 566 /*
 567  * Initialize a data block at the given block number in the directory.
 568  * Give back the buffer for the created block.
 569  */
 570 int                                             /* error */
 571 xfs_dir3_data_init(
 572         xfs_da_args_t           *args,          /* directory operation args */
 573         xfs_dir2_db_t           blkno,          /* logical dir block number */
 574         struct xfs_buf          **bpp)          /* output block buffer */
 575 {
 576         struct xfs_buf          *bp;            /* block buffer */
 577         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 578         xfs_inode_t             *dp;            /* incore directory inode */
 579         xfs_dir2_data_unused_t  *dup;           /* unused entry pointer */
 580         struct xfs_dir2_data_free *bf;
 581         int                     error;          /* error return value */
 582         int                     i;              /* bestfree index */
 583         xfs_mount_t             *mp;            /* filesystem mount point */
 584         xfs_trans_t             *tp;            /* transaction pointer */
 585         int                     t;              /* temp */
 586
 587         dp = args->dp;
 588         mp = dp->i_mount;
 589         tp = args->trans;
 590         /*
 591          * Get the buffer set up for the block.
 592          */
 593         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, blkno),
 594                                -1, &bp, XFS_DATA_FORK);
 595         if (error)
 596                 return error;
 597         bp->b_ops = &xfs_dir3_data_buf_ops;
 598         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
 599
 600         /*
 601          * Initialize the header.
 602          */
 603         hdr = bp->b_addr;
 604         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 605                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 606
 607                 memset(hdr3, 0, sizeof(*hdr3));
 608                 hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 609                 hdr3->blkno = cpu_to_be64(bp->b_bn);
 610                 hdr3->owner = cpu_to_be64(dp->i_ino);
 611                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
 612
 613         } else
 614                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 615
 616         bf = dp->d_ops->data_bestfree_p(hdr);
 617         bf[0].offset = cpu_to_be16(dp->d_ops->data_entry_offset);
 618         for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
 619                 bf[i].length = 0;
 620                 bf[i].offset = 0;
 621         }
 622
 623         /*
 624          * Set up an unused entry for the block's body.
 625          */
 626         dup = dp->d_ops->data_unused_p(hdr);
 627         dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 628
 629         t = args->geo->blksize - (uint)dp->d_ops->data_entry_offset;
 630         bf[0].length = cpu_to_be16(t);
 631         dup->length = cpu_to_be16(t);
 632         *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
 633         /*
 634          * Log it and return it.
 635          */
 636         xfs_dir2_data_log_header(args, bp);
 637         xfs_dir2_data_log_unused(args, bp, dup);
 638         *bpp = bp;
 639         return 0;
 640 }
 641
 642 /*
 643  * Log an active data entry from the block.
 644  */
 645 void
 646 xfs_dir2_data_log_entry(
 647         struct xfs_da_args      *args,
 648         struct xfs_buf          *bp,
 649         xfs_dir2_data_entry_t   *dep)           /* data entry pointer */
 650 {
 651         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 652
 653         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 654                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 655                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 656                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 657
 658         xfs_trans_log_buf(args->trans, bp, (uint)((char *)dep - (char *)hdr),
 659                 (uint)((char *)(args->dp->d_ops->data_entry_tag_p(dep) + 1) -
 660                        (char *)hdr - 1));
 661 }
 662
 663 /*
 664  * Log a data block header.
 665  */
 666 void
 667 xfs_dir2_data_log_header(
 668         struct xfs_da_args      *args,
 669         struct xfs_buf          *bp)
 670 {
 671 #ifdef DEBUG
 672         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 673
 674         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 675                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 676                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 677                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 678 #endif
 679
 680         xfs_trans_log_buf(args->trans, bp, 0,
 681                           args->dp->d_ops->data_entry_offset - 1);
 682 }
 683
 684 /*
 685  * Log a data unused entry.
 686  */
 687 void
 688 xfs_dir2_data_log_unused(
 689         struct xfs_da_args      *args,
 690         struct xfs_buf          *bp,
 691         xfs_dir2_data_unused_t  *dup)           /* data unused pointer */
 692 {
 693         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 694
 695         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 696                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 697                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 698                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 699
 700         /*
 701          * Log the first part of the unused entry.
 702          */
 703         xfs_trans_log_buf(args->trans, bp, (uint)((char *)dup - (char *)hdr),
 704                 (uint)((char *)&dup->length + sizeof(dup->length) -
 705                        1 - (char *)hdr));
 706         /*
 707          * Log the end (tag) of the unused entry.
 708          */
 709         xfs_trans_log_buf(args->trans, bp,
 710                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
 711                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
 712                        sizeof(xfs_dir2_data_off_t) - 1));
 713 }
 714
 715 /*
 716  * Make a byte range in the data block unused.
 717  * Its current contents are unimportant.
 718  */
 719 void
 720 xfs_dir2_data_make_free(
 721         struct xfs_da_args      *args,
 722         struct xfs_buf          *bp,
 723         xfs_dir2_data_aoff_t    offset,         /* starting byte offset */
 724         xfs_dir2_data_aoff_t    len,            /* length in bytes */
 725         int                     *needlogp,      /* out: log header */
 726         int                     *needscanp)     /* out: regen bestfree */
 727 {
 728         xfs_dir2_data_hdr_t     *hdr;           /* data block pointer */
 729         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 730         char                    *endptr;        /* end of data area */
 731         int                     needscan;       /* need to regen bestfree */
 732         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 733         xfs_dir2_data_unused_t  *postdup;       /* unused entry after us */
 734         xfs_dir2_data_unused_t  *prevdup;       /* unused entry before us */
 735         struct xfs_dir2_data_free *bf;
 736
 737         hdr = bp->b_addr;
 738
 739         /*
 740          * Figure out where the end of the data area is.
 741          */
 742         if (hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 743             hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 744                 endptr = (char *)hdr + args->geo->blksize;
 745         else {
 746                 xfs_dir2_block_tail_t   *btp;   /* block tail */
 747
 748                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 749                         hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 750                 btp = xfs_dir2_block_tail_p(args->geo, hdr);
 751                 endptr = (char *)xfs_dir2_block_leaf_p(btp);
 752         }
 753         /*
 754          * If this isn't the start of the block, then back up to
 755          * the previous entry and see if it's free.
 756          */
 757         if (offset > args->dp->d_ops->data_entry_offset) {
 758                 __be16                  *tagp;  /* tag just before us */
 759
 760                 tagp = (__be16 *)((char *)hdr + offset) - 1;
 761                 prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 762                 if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 763                         prevdup = NULL;
 764         } else
 765                 prevdup = NULL;
 766         /*
 767          * If this isn't the end of the block, see if the entry after
 768          * us is free.
 769          */
 770         if ((char *)hdr + offset + len < endptr) {
 771                 postdup =
 772                         (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 773                 if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 774                         postdup = NULL;
 775         } else
 776                 postdup = NULL;
 777         ASSERT(*needscanp == 0);
 778         needscan = 0;
 779         /*
 780          * Previous and following entries are both free,
 781          * merge everything into a single free entry.
 782          */
 783         bf = args->dp->d_ops->data_bestfree_p(hdr);
 784         if (prevdup && postdup) {
 785                 xfs_dir2_data_free_t    *dfp2;  /* another bestfree pointer */
 786
 787                 /*
 788                  * See if prevdup and/or postdup are in bestfree table.
 789                  */
 790                 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
 791                 dfp2 = xfs_dir2_data_freefind(hdr, bf, postdup);
 792                 /*
 793                  * We need a rescan unless there are exactly 2 free entries
 794                  * namely our two.  Then we know what's happening, otherwise
 795                  * since the third bestfree is there, there might be more
 796                  * entries.
 797                  */
 798                 needscan = (bf[2].length != 0);
 799                 /*
 800                  * Fix up the new big freespace.
 801                  */
 802                 be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
 803                 *xfs_dir2_data_unused_tag_p(prevdup) =
 804                         cpu_to_be16((char *)prevdup - (char *)hdr);
 805                 xfs_dir2_data_log_unused(args, bp, prevdup);
 806                 if (!needscan) {
 807                         /*
 808                          * Has to be the case that entries 0 and 1 are
 809                          * dfp and dfp2 (don't know which is which), and
 810                          * entry 2 is empty.
 811                          * Remove entry 1 first then entry 0.
 812                          */
 813                         ASSERT(dfp && dfp2);
 814                         if (dfp == &bf[1]) {
 815                                 dfp = &bf[0];
 816                                 ASSERT(dfp2 == dfp);
 817                                 dfp2 = &bf[1];
 818                         }
 819                         xfs_dir2_data_freeremove(hdr, bf, dfp2, needlogp);
 820                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 821                         /*
 822                          * Now insert the new entry.
 823                          */
 824                         dfp = xfs_dir2_data_freeinsert(hdr, bf, prevdup,
 825                                                        needlogp);
 826                         ASSERT(dfp == &bf[0]);
 827                         ASSERT(dfp->length == prevdup->length);
 828                         ASSERT(!dfp[1].length);
 829                         ASSERT(!dfp[2].length);
 830                 }
 831         }
 832         /*
 833          * The entry before us is free, merge with it.
 834          */
 835         else if (prevdup) {
 836                 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
 837                 be16_add_cpu(&prevdup->length, len);
 838                 *xfs_dir2_data_unused_tag_p(prevdup) =
 839                         cpu_to_be16((char *)prevdup - (char *)hdr);
 840                 xfs_dir2_data_log_unused(args, bp, prevdup);
 841                 /*
 842                  * If the previous entry was in the table, the new entry
 843                  * is longer, so it will be in the table too.  Remove
 844                  * the old one and add the new one.
 845                  */
 846                 if (dfp) {
 847                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 848                         xfs_dir2_data_freeinsert(hdr, bf, prevdup, needlogp);
 849                 }
 850                 /*
 851                  * Otherwise we need a scan if the new entry is big enough.
 852                  */
 853                 else {
 854                         needscan = be16_to_cpu(prevdup->length) >
 855                                    be16_to_cpu(bf[2].length);
 856                 }
 857         }
 858         /*
 859          * The following entry is free, merge with it.
 860          */
 861         else if (postdup) {
 862                 dfp = xfs_dir2_data_freefind(hdr, bf, postdup);
 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 + be16_to_cpu(postdup->length));
 866                 *xfs_dir2_data_unused_tag_p(newdup) =
 867                         cpu_to_be16((char *)newdup - (char *)hdr);
 868                 xfs_dir2_data_log_unused(args, bp, newdup);
 869                 /*
 870                  * If the following entry was in the table, the new entry
 871                  * is longer, so it will be in the table too.  Remove
 872                  * the old one and add the new one.
 873                  */
 874                 if (dfp) {
 875                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 876                         xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
 877                 }
 878                 /*
 879                  * Otherwise we need a scan if the new entry is big enough.
 880                  */
 881                 else {
 882                         needscan = be16_to_cpu(newdup->length) >
 883                                    be16_to_cpu(bf[2].length);
 884                 }
 885         }
 886         /*
 887          * Neither neighbor is free.  Make a new entry.
 888          */
 889         else {
 890                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 891                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 892                 newdup->length = cpu_to_be16(len);
 893                 *xfs_dir2_data_unused_tag_p(newdup) =
 894                         cpu_to_be16((char *)newdup - (char *)hdr);
 895                 xfs_dir2_data_log_unused(args, bp, newdup);
 896                 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
 897         }
 898         *needscanp = needscan;
 899 }
 900
 901 /*
 902  * Take a byte range out of an existing unused space and make it un-free.
 903  */
 904 void
 905 xfs_dir2_data_use_free(
 906         struct xfs_da_args      *args,
 907         struct xfs_buf          *bp,
 908         xfs_dir2_data_unused_t  *dup,           /* unused entry */
 909         xfs_dir2_data_aoff_t    offset,         /* starting offset to use */
 910         xfs_dir2_data_aoff_t    len,            /* length to use */
 911         int                     *needlogp,      /* out: need to log header */
 912         int                     *needscanp)     /* out: need regen bestfree */
 913 {
 914         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 915         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 916         int                     matchback;      /* matches end of freespace */
 917         int                     matchfront;     /* matches start of freespace */
 918         int                     needscan;       /* need to regen bestfree */
 919         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 920         xfs_dir2_data_unused_t  *newdup2;       /* another new unused entry */
 921         int                     oldlen;         /* old unused entry's length */
 922         struct xfs_dir2_data_free *bf;
 923
 924         hdr = bp->b_addr;
 925         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 926                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 927                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 928                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 929         ASSERT(be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG);
 930         ASSERT(offset >= (char *)dup - (char *)hdr);
 931         ASSERT(offset + len <= (char *)dup + be16_to_cpu(dup->length) - (char *)hdr);
 932         ASSERT((char *)dup - (char *)hdr == be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 933         /*
 934          * Look up the entry in the bestfree table.
 935          */
 936         oldlen = be16_to_cpu(dup->length);
 937         bf = args->dp->d_ops->data_bestfree_p(hdr);
 938         dfp = xfs_dir2_data_freefind(hdr, bf, dup);
 939         ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
 940         /*
 941          * Check for alignment with front and back of the entry.
 942          */
 943         matchfront = (char *)dup - (char *)hdr == offset;
 944         matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
 945         ASSERT(*needscanp == 0);
 946         needscan = 0;
 947         /*
 948          * If we matched it exactly we just need to get rid of it from
 949          * the bestfree table.
 950          */
 951         if (matchfront && matchback) {
 952                 if (dfp) {
 953                         needscan = (bf[2].offset != 0);
 954                         if (!needscan)
 955                                 xfs_dir2_data_freeremove(hdr, bf, dfp,
 956                                                          needlogp);
 957                 }
 958         }
 959         /*
 960          * We match the first part of the entry.
 961          * Make a new entry with the remaining freespace.
 962          */
 963         else if (matchfront) {
 964                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 965                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 966                 newdup->length = cpu_to_be16(oldlen - len);
 967                 *xfs_dir2_data_unused_tag_p(newdup) =
 968                         cpu_to_be16((char *)newdup - (char *)hdr);
 969                 xfs_dir2_data_log_unused(args, bp, newdup);
 970                 /*
 971                  * If it was in the table, remove it and add the new one.
 972                  */
 973                 if (dfp) {
 974                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 975                         dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
 976                                                        needlogp);
 977                         ASSERT(dfp != NULL);
 978                         ASSERT(dfp->length == newdup->length);
 979                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 980                         /*
 981                          * If we got inserted at the last slot,
 982                          * that means we don't know if there was a better
 983                          * choice for the last slot, or not.  Rescan.
 984                          */
 985                         needscan = dfp == &bf[2];
 986                 }
 987         }
 988         /*
 989          * We match the last part of the entry.
 990          * Trim the allocated space off the tail of the entry.
 991          */
 992         else if (matchback) {
 993                 newdup = dup;
 994                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
 995                 *xfs_dir2_data_unused_tag_p(newdup) =
 996                         cpu_to_be16((char *)newdup - (char *)hdr);
 997                 xfs_dir2_data_log_unused(args, bp, newdup);
 998                 /*
 999                  * If it was in the table, remove it and add the new one.
1000                  */
1001                 if (dfp) {
1002                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
1003                         dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
1004                                                        needlogp);
1005                         ASSERT(dfp != NULL);
1006                         ASSERT(dfp->length == newdup->length);
1007                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
1008                         /*
1009                          * If we got inserted at the last slot,
1010                          * that means we don't know if there was a better
1011                          * choice for the last slot, or not.  Rescan.
1012                          */
1013                         needscan = dfp == &bf[2];
1014                 }
1015         }
1016         /*
1017          * Poking out the middle of an entry.
1018          * Make two new entries.
1019          */
1020         else {
1021                 newdup = dup;
1022                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1023                 *xfs_dir2_data_unused_tag_p(newdup) =
1024                         cpu_to_be16((char *)newdup - (char *)hdr);
1025                 xfs_dir2_data_log_unused(args, bp, newdup);
1026                 newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1027                 newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1028                 newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1029                 *xfs_dir2_data_unused_tag_p(newdup2) =
1030                         cpu_to_be16((char *)newdup2 - (char *)hdr);
1031                 xfs_dir2_data_log_unused(args, bp, newdup2);
1032                 /*
1033                  * If the old entry was in the table, we need to scan
1034                  * if the 3rd entry was valid, since these entries
1035                  * are smaller than the old one.
1036                  * If we don't need to scan that means there were 1 or 2
1037                  * entries in the table, and removing the old and adding
1038                  * the 2 new will work.
1039                  */
1040                 if (dfp) {
1041                         needscan = (bf[2].length != 0);
1042                         if (!needscan) {
1043                                 xfs_dir2_data_freeremove(hdr, bf, dfp,
1044                                                          needlogp);
1045                                 xfs_dir2_data_freeinsert(hdr, bf, newdup,
1046                                                          needlogp);
1047                                 xfs_dir2_data_freeinsert(hdr, bf, newdup2,
1048                                                          needlogp);
1049                         }
1050                 }
1051         }
1052         *needscanp = needscan;
1053 }