fs/xfs/xfs_dir2_data.c

   1 /*
   2  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_inode.h"
  30 #include "xfs_dir2.h"
  31 #include "xfs_dir2_priv.h"
  32 #include "xfs_error.h"
  33 #include "xfs_trans.h"
  34 #include "xfs_buf_item.h"
  35 #include "xfs_cksum.h"
  36
  37 /*
  38  * Check the consistency of the data block.
  39  * The input can also be a block-format directory.
  40  * Return 0 is the buffer is good, otherwise an error.
  41  */
  42 int
  43 __xfs_dir3_data_check(
  44         struct xfs_inode        *dp,            /* incore inode pointer */
  45         struct xfs_buf          *bp)            /* data block's buffer */
  46 {
  47         xfs_dir2_dataptr_t      addr;           /* addr for leaf lookup */
  48         xfs_dir2_data_free_t    *bf;            /* bestfree table */
  49         xfs_dir2_block_tail_t   *btp=NULL;      /* block tail */
  50         int                     count;          /* count of entries found */
  51         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
  52         xfs_dir2_data_entry_t   *dep;           /* data entry */
  53         xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
  54         xfs_dir2_data_unused_t  *dup;           /* unused entry */
  55         char                    *endp;          /* end of useful data */
  56         int                     freeseen;       /* mask of bestfrees seen */
  57         xfs_dahash_t            hash;           /* hash of current name */
  58         int                     i;              /* leaf index */
  59         int                     lastfree;       /* last entry was unused */
  60         xfs_dir2_leaf_entry_t   *lep=NULL;      /* block leaf entries */
  61         xfs_mount_t             *mp;            /* filesystem mount point */
  62         char                    *p;             /* current data position */
  63         int                     stale;          /* count of stale leaves */
  64         struct xfs_name         name;
  65         const struct xfs_dir_ops *ops;
  66
  67         mp = bp->b_target->bt_mount;
  68         hdr = bp->b_addr;
  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         switch (hdr->magic) {
  77         case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
  78         case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
  79                 btp = xfs_dir2_block_tail_p(mp, hdr);
  80                 lep = xfs_dir2_block_leaf_p(btp);
  81                 endp = (char *)lep;
  82                 break;
  83         case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
  84         case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
  85                 endp = (char *)hdr + mp->m_dirblksize;
  86                 break;
  87         default:
  88                 XFS_ERROR_REPORT("Bad Magic", XFS_ERRLEVEL_LOW, mp);
  89                 return EFSCORRUPTED;
  90         }
  91         bf = ops->data_bestfree_p(hdr);
  92         p = (char *)ops->data_entry_p(hdr);
  93
  94         count = lastfree = freeseen = 0;
  95         /*
  96          * Account for zero bestfree entries.
  97          */
  98         if (!bf[0].length) {
  99                 XFS_WANT_CORRUPTED_RETURN(!bf[0].offset);
 100                 freeseen |= 1 << 0;
 101         }
 102         if (!bf[1].length) {
 103                 XFS_WANT_CORRUPTED_RETURN(!bf[1].offset);
 104                 freeseen |= 1 << 1;
 105         }
 106         if (!bf[2].length) {
 107                 XFS_WANT_CORRUPTED_RETURN(!bf[2].offset);
 108                 freeseen |= 1 << 2;
 109         }
 110
 111         XFS_WANT_CORRUPTED_RETURN(be16_to_cpu(bf[0].length) >=
 112                                                 be16_to_cpu(bf[1].length));
 113         XFS_WANT_CORRUPTED_RETURN(be16_to_cpu(bf[1].length) >=
 114                                                 be16_to_cpu(bf[2].length));
 115         /*
 116          * Loop over the data/unused entries.
 117          */
 118         while (p < endp) {
 119                 dup = (xfs_dir2_data_unused_t *)p;
 120                 /*
 121                  * If it's unused, look for the space in the bestfree table.
 122                  * If we find it, account for that, else make sure it
 123                  * doesn't need to be there.
 124                  */
 125                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 126                         XFS_WANT_CORRUPTED_RETURN(lastfree == 0);
 127                         XFS_WANT_CORRUPTED_RETURN(
 128                                 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) ==
 129                                                (char *)dup - (char *)hdr);
 130                         dfp = xfs_dir2_data_freefind(hdr, bf, dup);
 131                         if (dfp) {
 132                                 i = (int)(dfp - bf);
 133                                 XFS_WANT_CORRUPTED_RETURN(
 134                                         (freeseen & (1 << i)) == 0);
 135                                 freeseen |= 1 << i;
 136                         } else {
 137                                 XFS_WANT_CORRUPTED_RETURN(
 138                                         be16_to_cpu(dup->length) <=
 139                                                 be16_to_cpu(bf[2].length));
 140                         }
 141                         p += be16_to_cpu(dup->length);
 142                         lastfree = 1;
 143                         continue;
 144                 }
 145                 /*
 146                  * It's a real entry.  Validate the fields.
 147                  * If this is a block directory then make sure it's
 148                  * in the leaf section of the block.
 149                  * The linear search is crude but this is DEBUG code.
 150                  */
 151                 dep = (xfs_dir2_data_entry_t *)p;
 152                 XFS_WANT_CORRUPTED_RETURN(dep->namelen != 0);
 153                 XFS_WANT_CORRUPTED_RETURN(
 154                         !xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber)));
 155                 XFS_WANT_CORRUPTED_RETURN(
 156                         be16_to_cpu(*ops->data_entry_tag_p(dep)) ==
 157                                                (char *)dep - (char *)hdr);
 158                 XFS_WANT_CORRUPTED_RETURN(
 159                                 ops->data_get_ftype(dep) < XFS_DIR3_FT_MAX);
 160                 count++;
 161                 lastfree = 0;
 162                 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 163                     hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 164                         addr = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
 165                                 (xfs_dir2_data_aoff_t)
 166                                 ((char *)dep - (char *)hdr));
 167                         name.name = dep->name;
 168                         name.len = dep->namelen;
 169                         hash = mp->m_dirnameops->hashname(&name);
 170                         for (i = 0; i < be32_to_cpu(btp->count); i++) {
 171                                 if (be32_to_cpu(lep[i].address) == addr &&
 172                                     be32_to_cpu(lep[i].hashval) == hash)
 173                                         break;
 174                         }
 175                         XFS_WANT_CORRUPTED_RETURN(i < be32_to_cpu(btp->count));
 176                 }
 177                 p += ops->data_entsize(dep->namelen);
 178         }
 179         /*
 180          * Need to have seen all the entries and all the bestfree slots.
 181          */
 182         XFS_WANT_CORRUPTED_RETURN(freeseen == 7);
 183         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 184             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 185                 for (i = stale = 0; i < be32_to_cpu(btp->count); i++) {
 186                         if (lep[i].address ==
 187                             cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 188                                 stale++;
 189                         if (i > 0)
 190                                 XFS_WANT_CORRUPTED_RETURN(
 191                                         be32_to_cpu(lep[i].hashval) >=
 192                                                 be32_to_cpu(lep[i - 1].hashval));
 193                 }
 194                 XFS_WANT_CORRUPTED_RETURN(count ==
 195                         be32_to_cpu(btp->count) - be32_to_cpu(btp->stale));
 196                 XFS_WANT_CORRUPTED_RETURN(stale == be32_to_cpu(btp->stale));
 197         }
 198         return 0;
 199 }
 200
 201 static bool
 202 xfs_dir3_data_verify(
 203         struct xfs_buf          *bp)
 204 {
 205         struct xfs_mount        *mp = bp->b_target->bt_mount;
 206         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 207
 208         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 209                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 210                         return false;
 211                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
 212                         return false;
 213                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 214                         return false;
 215         } else {
 216                 if (hdr3->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC))
 217                         return false;
 218         }
 219         if (__xfs_dir3_data_check(NULL, bp))
 220                 return false;
 221         return true;
 222 }
 223
 224 /*
 225  * Readahead of the first block of the directory when it is opened is completely
 226  * oblivious to the format of the directory. Hence we can either get a block
 227  * format buffer or a data format buffer on readahead.
 228  */
 229 static void
 230 xfs_dir3_data_reada_verify(
 231         struct xfs_buf          *bp)
 232 {
 233         struct xfs_mount        *mp = bp->b_target->bt_mount;
 234         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 235
 236         switch (hdr->magic) {
 237         case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
 238         case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
 239                 bp->b_ops = &xfs_dir3_block_buf_ops;
 240                 bp->b_ops->verify_read(bp);
 241                 return;
 242         case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
 243         case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
 244                 xfs_dir3_data_verify(bp);
 245                 return;
 246         default:
 247                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, hdr);
 248                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 249                 break;
 250         }
 251 }
 252
 253 static void
 254 xfs_dir3_data_read_verify(
 255         struct xfs_buf  *bp)
 256 {
 257         struct xfs_mount        *mp = bp->b_target->bt_mount;
 258
 259         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
 260              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
 261                                           XFS_DIR3_DATA_CRC_OFF)) ||
 262             !xfs_dir3_data_verify(bp)) {
 263                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 264                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 265         }
 266 }
 267
 268 static void
 269 xfs_dir3_data_write_verify(
 270         struct xfs_buf  *bp)
 271 {
 272         struct xfs_mount        *mp = bp->b_target->bt_mount;
 273         struct xfs_buf_log_item *bip = bp->b_fspriv;
 274         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 275
 276         if (!xfs_dir3_data_verify(bp)) {
 277                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 278                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 279                 return;
 280         }
 281
 282         if (!xfs_sb_version_hascrc(&mp->m_sb))
 283                 return;
 284
 285         if (bip)
 286                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 287
 288         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_DATA_CRC_OFF);
 289 }
 290
 291 const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
 292         .verify_read = xfs_dir3_data_read_verify,
 293         .verify_write = xfs_dir3_data_write_verify,
 294 };
 295
 296 static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
 297         .verify_read = xfs_dir3_data_reada_verify,
 298         .verify_write = xfs_dir3_data_write_verify,
 299 };
 300
 301
 302 int
 303 xfs_dir3_data_read(
 304         struct xfs_trans        *tp,
 305         struct xfs_inode        *dp,
 306         xfs_dablk_t             bno,
 307         xfs_daddr_t             mapped_bno,
 308         struct xfs_buf          **bpp)
 309 {
 310         int                     err;
 311
 312         err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
 313                                 XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
 314         if (!err && tp)
 315                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
 316         return err;
 317 }
 318
 319 int
 320 xfs_dir3_data_readahead(
 321         struct xfs_trans        *tp,
 322         struct xfs_inode        *dp,
 323         xfs_dablk_t             bno,
 324         xfs_daddr_t             mapped_bno)
 325 {
 326         return xfs_da_reada_buf(tp, dp, bno, mapped_bno,
 327                                 XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
 328 }
 329
 330 /*
 331  * Given a data block and an unused entry from that block,
 332  * return the bestfree entry if any that corresponds to it.
 333  */
 334 xfs_dir2_data_free_t *
 335 xfs_dir2_data_freefind(
 336         struct xfs_dir2_data_hdr *hdr,          /* data block header */
 337         struct xfs_dir2_data_free *bf,          /* bestfree table pointer */
 338         struct xfs_dir2_data_unused *dup)       /* unused space */
 339 {
 340         xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
 341         xfs_dir2_data_aoff_t    off;            /* offset value needed */
 342 #ifdef DEBUG
 343         int                     matched;        /* matched the value */
 344         int                     seenzero;       /* saw a 0 bestfree entry */
 345 #endif
 346
 347         off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
 348
 349 #ifdef DEBUG
 350         /*
 351          * Validate some consistency in the bestfree table.
 352          * Check order, non-overlapping entries, and if we find the
 353          * one we're looking for it has to be exact.
 354          */
 355         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 356                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 357                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 358                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 359         for (dfp = &bf[0], seenzero = matched = 0;
 360              dfp < &bf[XFS_DIR2_DATA_FD_COUNT];
 361              dfp++) {
 362                 if (!dfp->offset) {
 363                         ASSERT(!dfp->length);
 364                         seenzero = 1;
 365                         continue;
 366                 }
 367                 ASSERT(seenzero == 0);
 368                 if (be16_to_cpu(dfp->offset) == off) {
 369                         matched = 1;
 370                         ASSERT(dfp->length == dup->length);
 371                 } else if (off < be16_to_cpu(dfp->offset))
 372                         ASSERT(off + be16_to_cpu(dup->length) <= be16_to_cpu(dfp->offset));
 373                 else
 374                         ASSERT(be16_to_cpu(dfp->offset) + be16_to_cpu(dfp->length) <= off);
 375                 ASSERT(matched || be16_to_cpu(dfp->length) >= be16_to_cpu(dup->length));
 376                 if (dfp > &bf[0])
 377                         ASSERT(be16_to_cpu(dfp[-1].length) >= be16_to_cpu(dfp[0].length));
 378         }
 379 #endif
 380         /*
 381          * If this is smaller than the smallest bestfree entry,
 382          * it can't be there since they're sorted.
 383          */
 384         if (be16_to_cpu(dup->length) <
 385             be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 386                 return NULL;
 387         /*
 388          * Look at the three bestfree entries for our guy.
 389          */
 390         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 391                 if (!dfp->offset)
 392                         return NULL;
 393                 if (be16_to_cpu(dfp->offset) == off)
 394                         return dfp;
 395         }
 396         /*
 397          * Didn't find it.  This only happens if there are duplicate lengths.
 398          */
 399         return NULL;
 400 }
 401
 402 /*
 403  * Insert an unused-space entry into the bestfree table.
 404  */
 405 xfs_dir2_data_free_t *                          /* entry inserted */
 406 xfs_dir2_data_freeinsert(
 407         struct xfs_dir2_data_hdr *hdr,          /* data block pointer */
 408         struct xfs_dir2_data_free *dfp,         /* bestfree table pointer */
 409         struct xfs_dir2_data_unused *dup,       /* unused space */
 410         int                     *loghead)       /* log the data header (out) */
 411 {
 412         xfs_dir2_data_free_t    new;            /* new bestfree entry */
 413
 414         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 415                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 416                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 417                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 418
 419         new.length = dup->length;
 420         new.offset = cpu_to_be16((char *)dup - (char *)hdr);
 421
 422         /*
 423          * Insert at position 0, 1, or 2; or not at all.
 424          */
 425         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
 426                 dfp[2] = dfp[1];
 427                 dfp[1] = dfp[0];
 428                 dfp[0] = new;
 429                 *loghead = 1;
 430                 return &dfp[0];
 431         }
 432         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
 433                 dfp[2] = dfp[1];
 434                 dfp[1] = new;
 435                 *loghead = 1;
 436                 return &dfp[1];
 437         }
 438         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
 439                 dfp[2] = new;
 440                 *loghead = 1;
 441                 return &dfp[2];
 442         }
 443         return NULL;
 444 }
 445
 446 /*
 447  * Remove a bestfree entry from the table.
 448  */
 449 STATIC void
 450 xfs_dir2_data_freeremove(
 451         struct xfs_dir2_data_hdr *hdr,          /* data block header */
 452         struct xfs_dir2_data_free *bf,          /* bestfree table pointer */
 453         struct xfs_dir2_data_free *dfp,         /* bestfree entry pointer */
 454         int                     *loghead)       /* out: log data header */
 455 {
 456
 457         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 458                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 459                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 460                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 461
 462         /*
 463          * It's the first entry, slide the next 2 up.
 464          */
 465         if (dfp == &bf[0]) {
 466                 bf[0] = bf[1];
 467                 bf[1] = bf[2];
 468         }
 469         /*
 470          * It's the second entry, slide the 3rd entry up.
 471          */
 472         else if (dfp == &bf[1])
 473                 bf[1] = bf[2];
 474         /*
 475          * Must be the last entry.
 476          */
 477         else
 478                 ASSERT(dfp == &bf[2]);
 479         /*
 480          * Clear the 3rd entry, must be zero now.
 481          */
 482         bf[2].length = 0;
 483         bf[2].offset = 0;
 484         *loghead = 1;
 485 }
 486
 487 /*
 488  * Given a data block, reconstruct its bestfree map.
 489  */
 490 void
 491 xfs_dir2_data_freescan(
 492         struct xfs_inode        *dp,
 493         struct xfs_dir2_data_hdr *hdr,
 494         int                     *loghead)
 495 {
 496         xfs_dir2_block_tail_t   *btp;           /* block tail */
 497         xfs_dir2_data_entry_t   *dep;           /* active data entry */
 498         xfs_dir2_data_unused_t  *dup;           /* unused data entry */
 499         struct xfs_dir2_data_free *bf;
 500         char                    *endp;          /* end of block's data */
 501         char                    *p;             /* current entry pointer */
 502
 503         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 504                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 505                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 506                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 507
 508         /*
 509          * Start by clearing the table.
 510          */
 511         bf = dp->d_ops->data_bestfree_p(hdr);
 512         memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
 513         *loghead = 1;
 514         /*
 515          * Set up pointers.
 516          */
 517         p = (char *)dp->d_ops->data_entry_p(hdr);
 518         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 519             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 520                 btp = xfs_dir2_block_tail_p(dp->i_mount, hdr);
 521                 endp = (char *)xfs_dir2_block_leaf_p(btp);
 522         } else
 523                 endp = (char *)hdr + dp->i_mount->m_dirblksize;
 524         /*
 525          * Loop over the block's entries.
 526          */
 527         while (p < endp) {
 528                 dup = (xfs_dir2_data_unused_t *)p;
 529                 /*
 530                  * If it's a free entry, insert it.
 531                  */
 532                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 533                         ASSERT((char *)dup - (char *)hdr ==
 534                                be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 535                         xfs_dir2_data_freeinsert(hdr, bf, dup, loghead);
 536                         p += be16_to_cpu(dup->length);
 537                 }
 538                 /*
 539                  * For active entries, check their tags and skip them.
 540                  */
 541                 else {
 542                         dep = (xfs_dir2_data_entry_t *)p;
 543                         ASSERT((char *)dep - (char *)hdr ==
 544                                be16_to_cpu(*dp->d_ops->data_entry_tag_p(dep)));
 545                         p += dp->d_ops->data_entsize(dep->namelen);
 546                 }
 547         }
 548 }
 549
 550 /*
 551  * Initialize a data block at the given block number in the directory.
 552  * Give back the buffer for the created block.
 553  */
 554 int                                             /* error */
 555 xfs_dir3_data_init(
 556         xfs_da_args_t           *args,          /* directory operation args */
 557         xfs_dir2_db_t           blkno,          /* logical dir block number */
 558         struct xfs_buf          **bpp)          /* output block buffer */
 559 {
 560         struct xfs_buf          *bp;            /* block buffer */
 561         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 562         xfs_inode_t             *dp;            /* incore directory inode */
 563         xfs_dir2_data_unused_t  *dup;           /* unused entry pointer */
 564         struct xfs_dir2_data_free *bf;
 565         int                     error;          /* error return value */
 566         int                     i;              /* bestfree index */
 567         xfs_mount_t             *mp;            /* filesystem mount point */
 568         xfs_trans_t             *tp;            /* transaction pointer */
 569         int                     t;              /* temp */
 570
 571         dp = args->dp;
 572         mp = dp->i_mount;
 573         tp = args->trans;
 574         /*
 575          * Get the buffer set up for the block.
 576          */
 577         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, blkno), -1, &bp,
 578                 XFS_DATA_FORK);
 579         if (error)
 580                 return error;
 581         bp->b_ops = &xfs_dir3_data_buf_ops;
 582         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
 583
 584         /*
 585          * Initialize the header.
 586          */
 587         hdr = bp->b_addr;
 588         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 589                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 590
 591                 memset(hdr3, 0, sizeof(*hdr3));
 592                 hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 593                 hdr3->blkno = cpu_to_be64(bp->b_bn);
 594                 hdr3->owner = cpu_to_be64(dp->i_ino);
 595                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
 596
 597         } else
 598                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 599
 600         bf = dp->d_ops->data_bestfree_p(hdr);
 601         bf[0].offset = cpu_to_be16(dp->d_ops->data_entry_offset);
 602         for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
 603                 bf[i].length = 0;
 604                 bf[i].offset = 0;
 605         }
 606
 607         /*
 608          * Set up an unused entry for the block's body.
 609          */
 610         dup = dp->d_ops->data_unused_p(hdr);
 611         dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 612
 613         t = mp->m_dirblksize - (uint)dp->d_ops->data_entry_offset;
 614         bf[0].length = cpu_to_be16(t);
 615         dup->length = cpu_to_be16(t);
 616         *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
 617         /*
 618          * Log it and return it.
 619          */
 620         xfs_dir2_data_log_header(tp, dp, bp);
 621         xfs_dir2_data_log_unused(tp, bp, dup);
 622         *bpp = bp;
 623         return 0;
 624 }
 625
 626 /*
 627  * Log an active data entry from the block.
 628  */
 629 void
 630 xfs_dir2_data_log_entry(
 631         struct xfs_trans        *tp,
 632         struct xfs_inode        *dp,
 633         struct xfs_buf          *bp,
 634         xfs_dir2_data_entry_t   *dep)           /* data entry pointer */
 635 {
 636         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 637
 638         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 639                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 640                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 641                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 642
 643         xfs_trans_log_buf(tp, bp, (uint)((char *)dep - (char *)hdr),
 644                 (uint)((char *)(dp->d_ops->data_entry_tag_p(dep) + 1) -
 645                        (char *)hdr - 1));
 646 }
 647
 648 /*
 649  * Log a data block header.
 650  */
 651 void
 652 xfs_dir2_data_log_header(
 653         struct xfs_trans        *tp,
 654         struct xfs_inode        *dp,
 655         struct xfs_buf          *bp)
 656 {
 657 #ifdef DEBUG
 658         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 659
 660         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 661                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 662                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 663                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 664 #endif
 665
 666         xfs_trans_log_buf(tp, bp, 0, dp->d_ops->data_entry_offset - 1);
 667 }
 668
 669 /*
 670  * Log a data unused entry.
 671  */
 672 void
 673 xfs_dir2_data_log_unused(
 674         struct xfs_trans        *tp,
 675         struct xfs_buf          *bp,
 676         xfs_dir2_data_unused_t  *dup)           /* data unused pointer */
 677 {
 678         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 679
 680         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 681                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 682                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 683                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 684
 685         /*
 686          * Log the first part of the unused entry.
 687          */
 688         xfs_trans_log_buf(tp, bp, (uint)((char *)dup - (char *)hdr),
 689                 (uint)((char *)&dup->length + sizeof(dup->length) -
 690                        1 - (char *)hdr));
 691         /*
 692          * Log the end (tag) of the unused entry.
 693          */
 694         xfs_trans_log_buf(tp, bp,
 695                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
 696                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
 697                        sizeof(xfs_dir2_data_off_t) - 1));
 698 }
 699
 700 /*
 701  * Make a byte range in the data block unused.
 702  * Its current contents are unimportant.
 703  */
 704 void
 705 xfs_dir2_data_make_free(
 706         struct xfs_trans        *tp,
 707         struct xfs_inode        *dp,
 708         struct xfs_buf          *bp,
 709         xfs_dir2_data_aoff_t    offset,         /* starting byte offset */
 710         xfs_dir2_data_aoff_t    len,            /* length in bytes */
 711         int                     *needlogp,      /* out: log header */
 712         int                     *needscanp)     /* out: regen bestfree */
 713 {
 714         xfs_dir2_data_hdr_t     *hdr;           /* data block pointer */
 715         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 716         char                    *endptr;        /* end of data area */
 717         xfs_mount_t             *mp;            /* filesystem mount point */
 718         int                     needscan;       /* need to regen bestfree */
 719         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 720         xfs_dir2_data_unused_t  *postdup;       /* unused entry after us */
 721         xfs_dir2_data_unused_t  *prevdup;       /* unused entry before us */
 722         struct xfs_dir2_data_free *bf;
 723
 724         mp = tp->t_mountp;
 725         hdr = bp->b_addr;
 726
 727         /*
 728          * Figure out where the end of the data area is.
 729          */
 730         if (hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 731             hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 732                 endptr = (char *)hdr + mp->m_dirblksize;
 733         else {
 734                 xfs_dir2_block_tail_t   *btp;   /* block tail */
 735
 736                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 737                         hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 738                 btp = xfs_dir2_block_tail_p(mp, hdr);
 739                 endptr = (char *)xfs_dir2_block_leaf_p(btp);
 740         }
 741         /*
 742          * If this isn't the start of the block, then back up to
 743          * the previous entry and see if it's free.
 744          */
 745         if (offset > dp->d_ops->data_entry_offset) {
 746                 __be16                  *tagp;  /* tag just before us */
 747
 748                 tagp = (__be16 *)((char *)hdr + offset) - 1;
 749                 prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 750                 if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 751                         prevdup = NULL;
 752         } else
 753                 prevdup = NULL;
 754         /*
 755          * If this isn't the end of the block, see if the entry after
 756          * us is free.
 757          */
 758         if ((char *)hdr + offset + len < endptr) {
 759                 postdup =
 760                         (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 761                 if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 762                         postdup = NULL;
 763         } else
 764                 postdup = NULL;
 765         ASSERT(*needscanp == 0);
 766         needscan = 0;
 767         /*
 768          * Previous and following entries are both free,
 769          * merge everything into a single free entry.
 770          */
 771         bf = dp->d_ops->data_bestfree_p(hdr);
 772         if (prevdup && postdup) {
 773                 xfs_dir2_data_free_t    *dfp2;  /* another bestfree pointer */
 774
 775                 /*
 776                  * See if prevdup and/or postdup are in bestfree table.
 777                  */
 778                 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
 779                 dfp2 = xfs_dir2_data_freefind(hdr, bf, postdup);
 780                 /*
 781                  * We need a rescan unless there are exactly 2 free entries
 782                  * namely our two.  Then we know what's happening, otherwise
 783                  * since the third bestfree is there, there might be more
 784                  * entries.
 785                  */
 786                 needscan = (bf[2].length != 0);
 787                 /*
 788                  * Fix up the new big freespace.
 789                  */
 790                 be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
 791                 *xfs_dir2_data_unused_tag_p(prevdup) =
 792                         cpu_to_be16((char *)prevdup - (char *)hdr);
 793                 xfs_dir2_data_log_unused(tp, bp, prevdup);
 794                 if (!needscan) {
 795                         /*
 796                          * Has to be the case that entries 0 and 1 are
 797                          * dfp and dfp2 (don't know which is which), and
 798                          * entry 2 is empty.
 799                          * Remove entry 1 first then entry 0.
 800                          */
 801                         ASSERT(dfp && dfp2);
 802                         if (dfp == &bf[1]) {
 803                                 dfp = &bf[0];
 804                                 ASSERT(dfp2 == dfp);
 805                                 dfp2 = &bf[1];
 806                         }
 807                         xfs_dir2_data_freeremove(hdr, bf, dfp2, needlogp);
 808                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 809                         /*
 810                          * Now insert the new entry.
 811                          */
 812                         dfp = xfs_dir2_data_freeinsert(hdr, bf, prevdup,
 813                                                        needlogp);
 814                         ASSERT(dfp == &bf[0]);
 815                         ASSERT(dfp->length == prevdup->length);
 816                         ASSERT(!dfp[1].length);
 817                         ASSERT(!dfp[2].length);
 818                 }
 819         }
 820         /*
 821          * The entry before us is free, merge with it.
 822          */
 823         else if (prevdup) {
 824                 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
 825                 be16_add_cpu(&prevdup->length, len);
 826                 *xfs_dir2_data_unused_tag_p(prevdup) =
 827                         cpu_to_be16((char *)prevdup - (char *)hdr);
 828                 xfs_dir2_data_log_unused(tp, bp, prevdup);
 829                 /*
 830                  * If the previous entry was in the table, the new entry
 831                  * is longer, so it will be in the table too.  Remove
 832                  * the old one and add the new one.
 833                  */
 834                 if (dfp) {
 835                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 836                         xfs_dir2_data_freeinsert(hdr, bf, prevdup, needlogp);
 837                 }
 838                 /*
 839                  * Otherwise we need a scan if the new entry is big enough.
 840                  */
 841                 else {
 842                         needscan = be16_to_cpu(prevdup->length) >
 843                                    be16_to_cpu(bf[2].length);
 844                 }
 845         }
 846         /*
 847          * The following entry is free, merge with it.
 848          */
 849         else if (postdup) {
 850                 dfp = xfs_dir2_data_freefind(hdr, bf, postdup);
 851                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 852                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 853                 newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
 854                 *xfs_dir2_data_unused_tag_p(newdup) =
 855                         cpu_to_be16((char *)newdup - (char *)hdr);
 856                 xfs_dir2_data_log_unused(tp, bp, newdup);
 857                 /*
 858                  * If the following entry was in the table, the new entry
 859                  * is longer, so it will be in the table too.  Remove
 860                  * the old one and add the new one.
 861                  */
 862                 if (dfp) {
 863                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 864                         xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
 865                 }
 866                 /*
 867                  * Otherwise we need a scan if the new entry is big enough.
 868                  */
 869                 else {
 870                         needscan = be16_to_cpu(newdup->length) >
 871                                    be16_to_cpu(bf[2].length);
 872                 }
 873         }
 874         /*
 875          * Neither neighbor is free.  Make a new entry.
 876          */
 877         else {
 878                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 879                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 880                 newdup->length = cpu_to_be16(len);
 881                 *xfs_dir2_data_unused_tag_p(newdup) =
 882                         cpu_to_be16((char *)newdup - (char *)hdr);
 883                 xfs_dir2_data_log_unused(tp, bp, newdup);
 884                 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
 885         }
 886         *needscanp = needscan;
 887 }
 888
 889 /*
 890  * Take a byte range out of an existing unused space and make it un-free.
 891  */
 892 void
 893 xfs_dir2_data_use_free(
 894         struct xfs_trans        *tp,
 895         struct xfs_inode        *dp,
 896         struct xfs_buf          *bp,
 897         xfs_dir2_data_unused_t  *dup,           /* unused entry */
 898         xfs_dir2_data_aoff_t    offset,         /* starting offset to use */
 899         xfs_dir2_data_aoff_t    len,            /* length to use */
 900         int                     *needlogp,      /* out: need to log header */
 901         int                     *needscanp)     /* out: need regen bestfree */
 902 {
 903         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 904         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 905         int                     matchback;      /* matches end of freespace */
 906         int                     matchfront;     /* matches start of freespace */
 907         int                     needscan;       /* need to regen bestfree */
 908         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 909         xfs_dir2_data_unused_t  *newdup2;       /* another new unused entry */
 910         int                     oldlen;         /* old unused entry's length */
 911         struct xfs_dir2_data_free *bf;
 912
 913         hdr = bp->b_addr;
 914         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 915                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 916                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 917                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 918         ASSERT(be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG);
 919         ASSERT(offset >= (char *)dup - (char *)hdr);
 920         ASSERT(offset + len <= (char *)dup + be16_to_cpu(dup->length) - (char *)hdr);
 921         ASSERT((char *)dup - (char *)hdr == be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 922         /*
 923          * Look up the entry in the bestfree table.
 924          */
 925         oldlen = be16_to_cpu(dup->length);
 926         bf = dp->d_ops->data_bestfree_p(hdr);
 927         dfp = xfs_dir2_data_freefind(hdr, bf, dup);
 928         ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
 929         /*
 930          * Check for alignment with front and back of the entry.
 931          */
 932         matchfront = (char *)dup - (char *)hdr == offset;
 933         matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
 934         ASSERT(*needscanp == 0);
 935         needscan = 0;
 936         /*
 937          * If we matched it exactly we just need to get rid of it from
 938          * the bestfree table.
 939          */
 940         if (matchfront && matchback) {
 941                 if (dfp) {
 942                         needscan = (bf[2].offset != 0);
 943                         if (!needscan)
 944                                 xfs_dir2_data_freeremove(hdr, bf, dfp,
 945                                                          needlogp);
 946                 }
 947         }
 948         /*
 949          * We match the first part of the entry.
 950          * Make a new entry with the remaining freespace.
 951          */
 952         else if (matchfront) {
 953                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 954                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 955                 newdup->length = cpu_to_be16(oldlen - len);
 956                 *xfs_dir2_data_unused_tag_p(newdup) =
 957                         cpu_to_be16((char *)newdup - (char *)hdr);
 958                 xfs_dir2_data_log_unused(tp, bp, newdup);
 959                 /*
 960                  * If it was in the table, remove it and add the new one.
 961                  */
 962                 if (dfp) {
 963                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 964                         dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
 965                                                        needlogp);
 966                         ASSERT(dfp != NULL);
 967                         ASSERT(dfp->length == newdup->length);
 968                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 969                         /*
 970                          * If we got inserted at the last slot,
 971                          * that means we don't know if there was a better
 972                          * choice for the last slot, or not.  Rescan.
 973                          */
 974                         needscan = dfp == &bf[2];
 975                 }
 976         }
 977         /*
 978          * We match the last part of the entry.
 979          * Trim the allocated space off the tail of the entry.
 980          */
 981         else if (matchback) {
 982                 newdup = dup;
 983                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
 984                 *xfs_dir2_data_unused_tag_p(newdup) =
 985                         cpu_to_be16((char *)newdup - (char *)hdr);
 986                 xfs_dir2_data_log_unused(tp, bp, newdup);
 987                 /*
 988                  * If it was in the table, remove it and add the new one.
 989                  */
 990                 if (dfp) {
 991                         xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 992                         dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
 993                                                        needlogp);
 994                         ASSERT(dfp != NULL);
 995                         ASSERT(dfp->length == newdup->length);
 996                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 997                         /*
 998                          * If we got inserted at the last slot,
 999                          * that means we don't know if there was a better
1000                          * choice for the last slot, or not.  Rescan.
1001                          */
1002                         needscan = dfp == &bf[2];
1003                 }
1004         }
1005         /*
1006          * Poking out the middle of an entry.
1007          * Make two new entries.
1008          */
1009         else {
1010                 newdup = dup;
1011                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1012                 *xfs_dir2_data_unused_tag_p(newdup) =
1013                         cpu_to_be16((char *)newdup - (char *)hdr);
1014                 xfs_dir2_data_log_unused(tp, bp, newdup);
1015                 newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1016                 newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1017                 newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1018                 *xfs_dir2_data_unused_tag_p(newdup2) =
1019                         cpu_to_be16((char *)newdup2 - (char *)hdr);
1020                 xfs_dir2_data_log_unused(tp, bp, newdup2);
1021                 /*
1022                  * If the old entry was in the table, we need to scan
1023                  * if the 3rd entry was valid, since these entries
1024                  * are smaller than the old one.
1025                  * If we don't need to scan that means there were 1 or 2
1026                  * entries in the table, and removing the old and adding
1027                  * the 2 new will work.
1028                  */
1029                 if (dfp) {
1030                         needscan = (bf[2].length != 0);
1031                         if (!needscan) {
1032                                 xfs_dir2_data_freeremove(hdr, bf, dfp,
1033                                                          needlogp);
1034                                 xfs_dir2_data_freeinsert(hdr, bf, newdup,
1035                                                          needlogp);
1036                                 xfs_dir2_data_freeinsert(hdr, bf, newdup2,
1037                                                          needlogp);
1038                         }
1039                 }
1040         }
1041         *needscanp = needscan;
1042 }