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
  66         mp = bp->b_target->bt_mount;
  67         hdr = bp->b_addr;
  68         bf = xfs_dir3_data_bestfree_p(hdr);
  69         p = (char *)xfs_dir3_data_entry_p(hdr);
  70
  71         switch (hdr->magic) {
  72         case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
  73         case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
  74                 btp = xfs_dir2_block_tail_p(mp, hdr);
  75                 lep = xfs_dir2_block_leaf_p(btp);
  76                 endp = (char *)lep;
  77                 break;
  78         case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
  79         case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
  80                 endp = (char *)hdr + mp->m_dirblksize;
  81                 break;
  82         default:
  83                 XFS_ERROR_REPORT("Bad Magic", XFS_ERRLEVEL_LOW, mp);
  84                 return EFSCORRUPTED;
  85         }
  86
  87         count = lastfree = freeseen = 0;
  88         /*
  89          * Account for zero bestfree entries.
  90          */
  91         if (!bf[0].length) {
  92                 XFS_WANT_CORRUPTED_RETURN(!bf[0].offset);
  93                 freeseen |= 1 << 0;
  94         }
  95         if (!bf[1].length) {
  96                 XFS_WANT_CORRUPTED_RETURN(!bf[1].offset);
  97                 freeseen |= 1 << 1;
  98         }
  99         if (!bf[2].length) {
 100                 XFS_WANT_CORRUPTED_RETURN(!bf[2].offset);
 101                 freeseen |= 1 << 2;
 102         }
 103
 104         XFS_WANT_CORRUPTED_RETURN(be16_to_cpu(bf[0].length) >=
 105                                                 be16_to_cpu(bf[1].length));
 106         XFS_WANT_CORRUPTED_RETURN(be16_to_cpu(bf[1].length) >=
 107                                                 be16_to_cpu(bf[2].length));
 108         /*
 109          * Loop over the data/unused entries.
 110          */
 111         while (p < endp) {
 112                 dup = (xfs_dir2_data_unused_t *)p;
 113                 /*
 114                  * If it's unused, look for the space in the bestfree table.
 115                  * If we find it, account for that, else make sure it
 116                  * doesn't need to be there.
 117                  */
 118                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 119                         XFS_WANT_CORRUPTED_RETURN(lastfree == 0);
 120                         XFS_WANT_CORRUPTED_RETURN(
 121                                 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) ==
 122                                                (char *)dup - (char *)hdr);
 123                         dfp = xfs_dir2_data_freefind(hdr, dup);
 124                         if (dfp) {
 125                                 i = (int)(dfp - bf);
 126                                 XFS_WANT_CORRUPTED_RETURN(
 127                                         (freeseen & (1 << i)) == 0);
 128                                 freeseen |= 1 << i;
 129                         } else {
 130                                 XFS_WANT_CORRUPTED_RETURN(
 131                                         be16_to_cpu(dup->length) <=
 132                                                 be16_to_cpu(bf[2].length));
 133                         }
 134                         p += be16_to_cpu(dup->length);
 135                         lastfree = 1;
 136                         continue;
 137                 }
 138                 /*
 139                  * It's a real entry.  Validate the fields.
 140                  * If this is a block directory then make sure it's
 141                  * in the leaf section of the block.
 142                  * The linear search is crude but this is DEBUG code.
 143                  */
 144                 dep = (xfs_dir2_data_entry_t *)p;
 145                 XFS_WANT_CORRUPTED_RETURN(dep->namelen != 0);
 146                 XFS_WANT_CORRUPTED_RETURN(
 147                         !xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber)));
 148                 XFS_WANT_CORRUPTED_RETURN(
 149                         be16_to_cpu(*xfs_dir3_data_entry_tag_p(mp, dep)) ==
 150                                                (char *)dep - (char *)hdr);
 151                 XFS_WANT_CORRUPTED_RETURN(
 152                         xfs_dir3_dirent_get_ftype(mp, dep) < XFS_DIR3_FT_MAX);
 153                 count++;
 154                 lastfree = 0;
 155                 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 156                     hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 157                         addr = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
 158                                 (xfs_dir2_data_aoff_t)
 159                                 ((char *)dep - (char *)hdr));
 160                         name.name = dep->name;
 161                         name.len = dep->namelen;
 162                         hash = mp->m_dirnameops->hashname(&name);
 163                         for (i = 0; i < be32_to_cpu(btp->count); i++) {
 164                                 if (be32_to_cpu(lep[i].address) == addr &&
 165                                     be32_to_cpu(lep[i].hashval) == hash)
 166                                         break;
 167                         }
 168                         XFS_WANT_CORRUPTED_RETURN(i < be32_to_cpu(btp->count));
 169                 }
 170                 p += xfs_dir3_data_entsize(mp, dep->namelen);
 171         }
 172         /*
 173          * Need to have seen all the entries and all the bestfree slots.
 174          */
 175         XFS_WANT_CORRUPTED_RETURN(freeseen == 7);
 176         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 177             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 178                 for (i = stale = 0; i < be32_to_cpu(btp->count); i++) {
 179                         if (lep[i].address ==
 180                             cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 181                                 stale++;
 182                         if (i > 0)
 183                                 XFS_WANT_CORRUPTED_RETURN(
 184                                         be32_to_cpu(lep[i].hashval) >=
 185                                                 be32_to_cpu(lep[i - 1].hashval));
 186                 }
 187                 XFS_WANT_CORRUPTED_RETURN(count ==
 188                         be32_to_cpu(btp->count) - be32_to_cpu(btp->stale));
 189                 XFS_WANT_CORRUPTED_RETURN(stale == be32_to_cpu(btp->stale));
 190         }
 191         return 0;
 192 }
 193
 194 static bool
 195 xfs_dir3_data_verify(
 196         struct xfs_buf          *bp)
 197 {
 198         struct xfs_mount        *mp = bp->b_target->bt_mount;
 199         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 200
 201         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 202                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 203                         return false;
 204                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
 205                         return false;
 206                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 207                         return false;
 208         } else {
 209                 if (hdr3->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC))
 210                         return false;
 211         }
 212         if (__xfs_dir3_data_check(NULL, bp))
 213                 return false;
 214         return true;
 215 }
 216
 217 /*
 218  * Readahead of the first block of the directory when it is opened is completely
 219  * oblivious to the format of the directory. Hence we can either get a block
 220  * format buffer or a data format buffer on readahead.
 221  */
 222 static void
 223 xfs_dir3_data_reada_verify(
 224         struct xfs_buf          *bp)
 225 {
 226         struct xfs_mount        *mp = bp->b_target->bt_mount;
 227         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 228
 229         switch (hdr->magic) {
 230         case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
 231         case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
 232                 bp->b_ops = &xfs_dir3_block_buf_ops;
 233                 bp->b_ops->verify_read(bp);
 234                 return;
 235         case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
 236         case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
 237                 xfs_dir3_data_verify(bp);
 238                 return;
 239         default:
 240                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, hdr);
 241                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 242                 break;
 243         }
 244 }
 245
 246 static void
 247 xfs_dir3_data_read_verify(
 248         struct xfs_buf  *bp)
 249 {
 250         struct xfs_mount        *mp = bp->b_target->bt_mount;
 251
 252         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
 253              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
 254                                           XFS_DIR3_DATA_CRC_OFF)) ||
 255             !xfs_dir3_data_verify(bp)) {
 256                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 257                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 258         }
 259 }
 260
 261 static void
 262 xfs_dir3_data_write_verify(
 263         struct xfs_buf  *bp)
 264 {
 265         struct xfs_mount        *mp = bp->b_target->bt_mount;
 266         struct xfs_buf_log_item *bip = bp->b_fspriv;
 267         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 268
 269         if (!xfs_dir3_data_verify(bp)) {
 270                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 271                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 272                 return;
 273         }
 274
 275         if (!xfs_sb_version_hascrc(&mp->m_sb))
 276                 return;
 277
 278         if (bip)
 279                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 280
 281         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_DATA_CRC_OFF);
 282 }
 283
 284 const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
 285         .verify_read = xfs_dir3_data_read_verify,
 286         .verify_write = xfs_dir3_data_write_verify,
 287 };
 288
 289 static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
 290         .verify_read = xfs_dir3_data_reada_verify,
 291         .verify_write = xfs_dir3_data_write_verify,
 292 };
 293
 294
 295 int
 296 xfs_dir3_data_read(
 297         struct xfs_trans        *tp,
 298         struct xfs_inode        *dp,
 299         xfs_dablk_t             bno,
 300         xfs_daddr_t             mapped_bno,
 301         struct xfs_buf          **bpp)
 302 {
 303         int                     err;
 304
 305         err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
 306                                 XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
 307         if (!err && tp)
 308                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
 309         return err;
 310 }
 311
 312 int
 313 xfs_dir3_data_readahead(
 314         struct xfs_trans        *tp,
 315         struct xfs_inode        *dp,
 316         xfs_dablk_t             bno,
 317         xfs_daddr_t             mapped_bno)
 318 {
 319         return xfs_da_reada_buf(tp, dp, bno, mapped_bno,
 320                                 XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
 321 }
 322
 323 /*
 324  * Given a data block and an unused entry from that block,
 325  * return the bestfree entry if any that corresponds to it.
 326  */
 327 xfs_dir2_data_free_t *
 328 xfs_dir2_data_freefind(
 329         xfs_dir2_data_hdr_t     *hdr,           /* data block */
 330         xfs_dir2_data_unused_t  *dup)           /* data unused entry */
 331 {
 332         xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
 333         xfs_dir2_data_aoff_t    off;            /* offset value needed */
 334         struct xfs_dir2_data_free *bf;
 335 #ifdef DEBUG
 336         int                     matched;        /* matched the value */
 337         int                     seenzero;       /* saw a 0 bestfree entry */
 338 #endif
 339
 340         off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
 341         bf = xfs_dir3_data_bestfree_p(hdr);
 342
 343 #ifdef DEBUG
 344         /*
 345          * Validate some consistency in the bestfree table.
 346          * Check order, non-overlapping entries, and if we find the
 347          * one we're looking for it has to be exact.
 348          */
 349         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 350                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 351                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 352                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 353         for (dfp = &bf[0], seenzero = matched = 0;
 354              dfp < &bf[XFS_DIR2_DATA_FD_COUNT];
 355              dfp++) {
 356                 if (!dfp->offset) {
 357                         ASSERT(!dfp->length);
 358                         seenzero = 1;
 359                         continue;
 360                 }
 361                 ASSERT(seenzero == 0);
 362                 if (be16_to_cpu(dfp->offset) == off) {
 363                         matched = 1;
 364                         ASSERT(dfp->length == dup->length);
 365                 } else if (off < be16_to_cpu(dfp->offset))
 366                         ASSERT(off + be16_to_cpu(dup->length) <= be16_to_cpu(dfp->offset));
 367                 else
 368                         ASSERT(be16_to_cpu(dfp->offset) + be16_to_cpu(dfp->length) <= off);
 369                 ASSERT(matched || be16_to_cpu(dfp->length) >= be16_to_cpu(dup->length));
 370                 if (dfp > &bf[0])
 371                         ASSERT(be16_to_cpu(dfp[-1].length) >= be16_to_cpu(dfp[0].length));
 372         }
 373 #endif
 374         /*
 375          * If this is smaller than the smallest bestfree entry,
 376          * it can't be there since they're sorted.
 377          */
 378         if (be16_to_cpu(dup->length) <
 379             be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 380                 return NULL;
 381         /*
 382          * Look at the three bestfree entries for our guy.
 383          */
 384         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 385                 if (!dfp->offset)
 386                         return NULL;
 387                 if (be16_to_cpu(dfp->offset) == off)
 388                         return dfp;
 389         }
 390         /*
 391          * Didn't find it.  This only happens if there are duplicate lengths.
 392          */
 393         return NULL;
 394 }
 395
 396 /*
 397  * Insert an unused-space entry into the bestfree table.
 398  */
 399 xfs_dir2_data_free_t *                          /* entry inserted */
 400 xfs_dir2_data_freeinsert(
 401         xfs_dir2_data_hdr_t     *hdr,           /* data block pointer */
 402         xfs_dir2_data_unused_t  *dup,           /* unused space */
 403         int                     *loghead)       /* log the data header (out) */
 404 {
 405         xfs_dir2_data_free_t    *dfp;           /* bestfree table pointer */
 406         xfs_dir2_data_free_t    new;            /* new bestfree entry */
 407
 408         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 409                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 410                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 411                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 412
 413         dfp = xfs_dir3_data_bestfree_p(hdr);
 414         new.length = dup->length;
 415         new.offset = cpu_to_be16((char *)dup - (char *)hdr);
 416
 417         /*
 418          * Insert at position 0, 1, or 2; or not at all.
 419          */
 420         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
 421                 dfp[2] = dfp[1];
 422                 dfp[1] = dfp[0];
 423                 dfp[0] = new;
 424                 *loghead = 1;
 425                 return &dfp[0];
 426         }
 427         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
 428                 dfp[2] = dfp[1];
 429                 dfp[1] = new;
 430                 *loghead = 1;
 431                 return &dfp[1];
 432         }
 433         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
 434                 dfp[2] = new;
 435                 *loghead = 1;
 436                 return &dfp[2];
 437         }
 438         return NULL;
 439 }
 440
 441 /*
 442  * Remove a bestfree entry from the table.
 443  */
 444 STATIC void
 445 xfs_dir2_data_freeremove(
 446         xfs_dir2_data_hdr_t     *hdr,           /* data block header */
 447         xfs_dir2_data_free_t    *dfp,           /* bestfree entry pointer */
 448         int                     *loghead)       /* out: log data header */
 449 {
 450         struct xfs_dir2_data_free *bf;
 451
 452         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 453                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 454                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 455                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 456
 457         /*
 458          * It's the first entry, slide the next 2 up.
 459          */
 460         bf = xfs_dir3_data_bestfree_p(hdr);
 461         if (dfp == &bf[0]) {
 462                 bf[0] = bf[1];
 463                 bf[1] = bf[2];
 464         }
 465         /*
 466          * It's the second entry, slide the 3rd entry up.
 467          */
 468         else if (dfp == &bf[1])
 469                 bf[1] = bf[2];
 470         /*
 471          * Must be the last entry.
 472          */
 473         else
 474                 ASSERT(dfp == &bf[2]);
 475         /*
 476          * Clear the 3rd entry, must be zero now.
 477          */
 478         bf[2].length = 0;
 479         bf[2].offset = 0;
 480         *loghead = 1;
 481 }
 482
 483 /*
 484  * Given a data block, reconstruct its bestfree map.
 485  */
 486 void
 487 xfs_dir2_data_freescan(
 488         xfs_mount_t             *mp,            /* filesystem mount point */
 489         xfs_dir2_data_hdr_t     *hdr,           /* data block header */
 490         int                     *loghead)       /* out: log data header */
 491 {
 492         xfs_dir2_block_tail_t   *btp;           /* block tail */
 493         xfs_dir2_data_entry_t   *dep;           /* active data entry */
 494         xfs_dir2_data_unused_t  *dup;           /* unused data entry */
 495         struct xfs_dir2_data_free *bf;
 496         char                    *endp;          /* end of block's data */
 497         char                    *p;             /* current entry pointer */
 498
 499         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 500                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 501                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 502                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 503
 504         /*
 505          * Start by clearing the table.
 506          */
 507         bf = xfs_dir3_data_bestfree_p(hdr);
 508         memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
 509         *loghead = 1;
 510         /*
 511          * Set up pointers.
 512          */
 513         p = (char *)xfs_dir3_data_entry_p(hdr);
 514         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 515             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 516                 btp = xfs_dir2_block_tail_p(mp, hdr);
 517                 endp = (char *)xfs_dir2_block_leaf_p(btp);
 518         } else
 519                 endp = (char *)hdr + mp->m_dirblksize;
 520         /*
 521          * Loop over the block's entries.
 522          */
 523         while (p < endp) {
 524                 dup = (xfs_dir2_data_unused_t *)p;
 525                 /*
 526                  * If it's a free entry, insert it.
 527                  */
 528                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 529                         ASSERT((char *)dup - (char *)hdr ==
 530                                be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 531                         xfs_dir2_data_freeinsert(hdr, dup, loghead);
 532                         p += be16_to_cpu(dup->length);
 533                 }
 534                 /*
 535                  * For active entries, check their tags and skip them.
 536                  */
 537                 else {
 538                         dep = (xfs_dir2_data_entry_t *)p;
 539                         ASSERT((char *)dep - (char *)hdr ==
 540                                be16_to_cpu(*xfs_dir3_data_entry_tag_p(mp, dep)));
 541                         p += xfs_dir3_data_entsize(mp, dep->namelen);
 542                 }
 543         }
 544 }
 545
 546 /*
 547  * Initialize a data block at the given block number in the directory.
 548  * Give back the buffer for the created block.
 549  */
 550 int                                             /* error */
 551 xfs_dir3_data_init(
 552         xfs_da_args_t           *args,          /* directory operation args */
 553         xfs_dir2_db_t           blkno,          /* logical dir block number */
 554         struct xfs_buf          **bpp)          /* output block buffer */
 555 {
 556         struct xfs_buf          *bp;            /* block buffer */
 557         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 558         xfs_inode_t             *dp;            /* incore directory inode */
 559         xfs_dir2_data_unused_t  *dup;           /* unused entry pointer */
 560         struct xfs_dir2_data_free *bf;
 561         int                     error;          /* error return value */
 562         int                     i;              /* bestfree index */
 563         xfs_mount_t             *mp;            /* filesystem mount point */
 564         xfs_trans_t             *tp;            /* transaction pointer */
 565         int                     t;              /* temp */
 566
 567         dp = args->dp;
 568         mp = dp->i_mount;
 569         tp = args->trans;
 570         /*
 571          * Get the buffer set up for the block.
 572          */
 573         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, blkno), -1, &bp,
 574                 XFS_DATA_FORK);
 575         if (error)
 576                 return error;
 577         bp->b_ops = &xfs_dir3_data_buf_ops;
 578         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
 579
 580         /*
 581          * Initialize the header.
 582          */
 583         hdr = bp->b_addr;
 584         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 585                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 586
 587                 memset(hdr3, 0, sizeof(*hdr3));
 588                 hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 589                 hdr3->blkno = cpu_to_be64(bp->b_bn);
 590                 hdr3->owner = cpu_to_be64(dp->i_ino);
 591                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
 592
 593         } else
 594                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 595
 596         bf = xfs_dir3_data_bestfree_p(hdr);
 597         bf[0].offset = cpu_to_be16(xfs_dir3_data_entry_offset(hdr));
 598         for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
 599                 bf[i].length = 0;
 600                 bf[i].offset = 0;
 601         }
 602
 603         /*
 604          * Set up an unused entry for the block's body.
 605          */
 606         dup = xfs_dir3_data_unused_p(hdr);
 607         dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 608
 609         t = mp->m_dirblksize - (uint)xfs_dir3_data_entry_offset(hdr);
 610         bf[0].length = cpu_to_be16(t);
 611         dup->length = cpu_to_be16(t);
 612         *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
 613         /*
 614          * Log it and return it.
 615          */
 616         xfs_dir2_data_log_header(tp, bp);
 617         xfs_dir2_data_log_unused(tp, bp, dup);
 618         *bpp = bp;
 619         return 0;
 620 }
 621
 622 /*
 623  * Log an active data entry from the block.
 624  */
 625 void
 626 xfs_dir2_data_log_entry(
 627         struct xfs_trans        *tp,
 628         struct xfs_buf          *bp,
 629         xfs_dir2_data_entry_t   *dep)           /* data entry pointer */
 630 {
 631         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 632         struct xfs_mount        *mp = tp->t_mountp;
 633
 634         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 635                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 636                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 637                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 638
 639         xfs_trans_log_buf(tp, bp, (uint)((char *)dep - (char *)hdr),
 640                 (uint)((char *)(xfs_dir3_data_entry_tag_p(mp, dep) + 1) -
 641                        (char *)hdr - 1));
 642 }
 643
 644 /*
 645  * Log a data block header.
 646  */
 647 void
 648 xfs_dir2_data_log_header(
 649         struct xfs_trans        *tp,
 650         struct xfs_buf          *bp)
 651 {
 652         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 653
 654         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 655                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 656                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 657                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 658
 659         xfs_trans_log_buf(tp, bp, 0, xfs_dir3_data_entry_offset(hdr) - 1);
 660 }
 661
 662 /*
 663  * Log a data unused entry.
 664  */
 665 void
 666 xfs_dir2_data_log_unused(
 667         struct xfs_trans        *tp,
 668         struct xfs_buf          *bp,
 669         xfs_dir2_data_unused_t  *dup)           /* data unused pointer */
 670 {
 671         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 672
 673         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 674                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 675                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 676                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 677
 678         /*
 679          * Log the first part of the unused entry.
 680          */
 681         xfs_trans_log_buf(tp, bp, (uint)((char *)dup - (char *)hdr),
 682                 (uint)((char *)&dup->length + sizeof(dup->length) -
 683                        1 - (char *)hdr));
 684         /*
 685          * Log the end (tag) of the unused entry.
 686          */
 687         xfs_trans_log_buf(tp, bp,
 688                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
 689                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
 690                        sizeof(xfs_dir2_data_off_t) - 1));
 691 }
 692
 693 /*
 694  * Make a byte range in the data block unused.
 695  * Its current contents are unimportant.
 696  */
 697 void
 698 xfs_dir2_data_make_free(
 699         struct xfs_trans        *tp,
 700         struct xfs_buf          *bp,
 701         xfs_dir2_data_aoff_t    offset,         /* starting byte offset */
 702         xfs_dir2_data_aoff_t    len,            /* length in bytes */
 703         int                     *needlogp,      /* out: log header */
 704         int                     *needscanp)     /* out: regen bestfree */
 705 {
 706         xfs_dir2_data_hdr_t     *hdr;           /* data block pointer */
 707         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 708         char                    *endptr;        /* end of data area */
 709         xfs_mount_t             *mp;            /* filesystem mount point */
 710         int                     needscan;       /* need to regen bestfree */
 711         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 712         xfs_dir2_data_unused_t  *postdup;       /* unused entry after us */
 713         xfs_dir2_data_unused_t  *prevdup;       /* unused entry before us */
 714         struct xfs_dir2_data_free *bf;
 715
 716         mp = tp->t_mountp;
 717         hdr = bp->b_addr;
 718
 719         /*
 720          * Figure out where the end of the data area is.
 721          */
 722         if (hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 723             hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 724                 endptr = (char *)hdr + mp->m_dirblksize;
 725         else {
 726                 xfs_dir2_block_tail_t   *btp;   /* block tail */
 727
 728                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 729                         hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 730                 btp = xfs_dir2_block_tail_p(mp, hdr);
 731                 endptr = (char *)xfs_dir2_block_leaf_p(btp);
 732         }
 733         /*
 734          * If this isn't the start of the block, then back up to
 735          * the previous entry and see if it's free.
 736          */
 737         if (offset > xfs_dir3_data_entry_offset(hdr)) {
 738                 __be16                  *tagp;  /* tag just before us */
 739
 740                 tagp = (__be16 *)((char *)hdr + offset) - 1;
 741                 prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 742                 if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 743                         prevdup = NULL;
 744         } else
 745                 prevdup = NULL;
 746         /*
 747          * If this isn't the end of the block, see if the entry after
 748          * us is free.
 749          */
 750         if ((char *)hdr + offset + len < endptr) {
 751                 postdup =
 752                         (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 753                 if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 754                         postdup = NULL;
 755         } else
 756                 postdup = NULL;
 757         ASSERT(*needscanp == 0);
 758         needscan = 0;
 759         /*
 760          * Previous and following entries are both free,
 761          * merge everything into a single free entry.
 762          */
 763         bf = xfs_dir3_data_bestfree_p(hdr);
 764         if (prevdup && postdup) {
 765                 xfs_dir2_data_free_t    *dfp2;  /* another bestfree pointer */
 766
 767                 /*
 768                  * See if prevdup and/or postdup are in bestfree table.
 769                  */
 770                 dfp = xfs_dir2_data_freefind(hdr, prevdup);
 771                 dfp2 = xfs_dir2_data_freefind(hdr, postdup);
 772                 /*
 773                  * We need a rescan unless there are exactly 2 free entries
 774                  * namely our two.  Then we know what's happening, otherwise
 775                  * since the third bestfree is there, there might be more
 776                  * entries.
 777                  */
 778                 needscan = (bf[2].length != 0);
 779                 /*
 780                  * Fix up the new big freespace.
 781                  */
 782                 be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
 783                 *xfs_dir2_data_unused_tag_p(prevdup) =
 784                         cpu_to_be16((char *)prevdup - (char *)hdr);
 785                 xfs_dir2_data_log_unused(tp, bp, prevdup);
 786                 if (!needscan) {
 787                         /*
 788                          * Has to be the case that entries 0 and 1 are
 789                          * dfp and dfp2 (don't know which is which), and
 790                          * entry 2 is empty.
 791                          * Remove entry 1 first then entry 0.
 792                          */
 793                         ASSERT(dfp && dfp2);
 794                         if (dfp == &bf[1]) {
 795                                 dfp = &bf[0];
 796                                 ASSERT(dfp2 == dfp);
 797                                 dfp2 = &bf[1];
 798                         }
 799                         xfs_dir2_data_freeremove(hdr, dfp2, needlogp);
 800                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 801                         /*
 802                          * Now insert the new entry.
 803                          */
 804                         dfp = xfs_dir2_data_freeinsert(hdr, prevdup, needlogp);
 805                         ASSERT(dfp == &bf[0]);
 806                         ASSERT(dfp->length == prevdup->length);
 807                         ASSERT(!dfp[1].length);
 808                         ASSERT(!dfp[2].length);
 809                 }
 810         }
 811         /*
 812          * The entry before us is free, merge with it.
 813          */
 814         else if (prevdup) {
 815                 dfp = xfs_dir2_data_freefind(hdr, prevdup);
 816                 be16_add_cpu(&prevdup->length, len);
 817                 *xfs_dir2_data_unused_tag_p(prevdup) =
 818                         cpu_to_be16((char *)prevdup - (char *)hdr);
 819                 xfs_dir2_data_log_unused(tp, bp, prevdup);
 820                 /*
 821                  * If the previous entry was in the table, the new entry
 822                  * is longer, so it will be in the table too.  Remove
 823                  * the old one and add the new one.
 824                  */
 825                 if (dfp) {
 826                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 827                         xfs_dir2_data_freeinsert(hdr, prevdup, needlogp);
 828                 }
 829                 /*
 830                  * Otherwise we need a scan if the new entry is big enough.
 831                  */
 832                 else {
 833                         needscan = be16_to_cpu(prevdup->length) >
 834                                    be16_to_cpu(bf[2].length);
 835                 }
 836         }
 837         /*
 838          * The following entry is free, merge with it.
 839          */
 840         else if (postdup) {
 841                 dfp = xfs_dir2_data_freefind(hdr, postdup);
 842                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 843                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 844                 newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
 845                 *xfs_dir2_data_unused_tag_p(newdup) =
 846                         cpu_to_be16((char *)newdup - (char *)hdr);
 847                 xfs_dir2_data_log_unused(tp, bp, newdup);
 848                 /*
 849                  * If the following entry was in the table, the new entry
 850                  * is longer, so it will be in the table too.  Remove
 851                  * the old one and add the new one.
 852                  */
 853                 if (dfp) {
 854                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 855                         xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 856                 }
 857                 /*
 858                  * Otherwise we need a scan if the new entry is big enough.
 859                  */
 860                 else {
 861                         needscan = be16_to_cpu(newdup->length) >
 862                                    be16_to_cpu(bf[2].length);
 863                 }
 864         }
 865         /*
 866          * Neither neighbor is free.  Make a new entry.
 867          */
 868         else {
 869                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 870                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 871                 newdup->length = cpu_to_be16(len);
 872                 *xfs_dir2_data_unused_tag_p(newdup) =
 873                         cpu_to_be16((char *)newdup - (char *)hdr);
 874                 xfs_dir2_data_log_unused(tp, bp, newdup);
 875                 xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 876         }
 877         *needscanp = needscan;
 878 }
 879
 880 /*
 881  * Take a byte range out of an existing unused space and make it un-free.
 882  */
 883 void
 884 xfs_dir2_data_use_free(
 885         struct xfs_trans        *tp,
 886         struct xfs_buf          *bp,
 887         xfs_dir2_data_unused_t  *dup,           /* unused entry */
 888         xfs_dir2_data_aoff_t    offset,         /* starting offset to use */
 889         xfs_dir2_data_aoff_t    len,            /* length to use */
 890         int                     *needlogp,      /* out: need to log header */
 891         int                     *needscanp)     /* out: need regen bestfree */
 892 {
 893         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 894         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 895         int                     matchback;      /* matches end of freespace */
 896         int                     matchfront;     /* matches start of freespace */
 897         int                     needscan;       /* need to regen bestfree */
 898         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 899         xfs_dir2_data_unused_t  *newdup2;       /* another new unused entry */
 900         int                     oldlen;         /* old unused entry's length */
 901         struct xfs_dir2_data_free *bf;
 902
 903         hdr = bp->b_addr;
 904         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 905                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 906                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 907                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 908         ASSERT(be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG);
 909         ASSERT(offset >= (char *)dup - (char *)hdr);
 910         ASSERT(offset + len <= (char *)dup + be16_to_cpu(dup->length) - (char *)hdr);
 911         ASSERT((char *)dup - (char *)hdr == be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 912         /*
 913          * Look up the entry in the bestfree table.
 914          */
 915         dfp = xfs_dir2_data_freefind(hdr, dup);
 916         oldlen = be16_to_cpu(dup->length);
 917         bf = xfs_dir3_data_bestfree_p(hdr);
 918         ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
 919         /*
 920          * Check for alignment with front and back of the entry.
 921          */
 922         matchfront = (char *)dup - (char *)hdr == offset;
 923         matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
 924         ASSERT(*needscanp == 0);
 925         needscan = 0;
 926         /*
 927          * If we matched it exactly we just need to get rid of it from
 928          * the bestfree table.
 929          */
 930         if (matchfront && matchback) {
 931                 if (dfp) {
 932                         needscan = (bf[2].offset != 0);
 933                         if (!needscan)
 934                                 xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 935                 }
 936         }
 937         /*
 938          * We match the first part of the entry.
 939          * Make a new entry with the remaining freespace.
 940          */
 941         else if (matchfront) {
 942                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 943                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 944                 newdup->length = cpu_to_be16(oldlen - len);
 945                 *xfs_dir2_data_unused_tag_p(newdup) =
 946                         cpu_to_be16((char *)newdup - (char *)hdr);
 947                 xfs_dir2_data_log_unused(tp, bp, newdup);
 948                 /*
 949                  * If it was in the table, remove it and add the new one.
 950                  */
 951                 if (dfp) {
 952                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 953                         dfp = xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 954                         ASSERT(dfp != NULL);
 955                         ASSERT(dfp->length == newdup->length);
 956                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 957                         /*
 958                          * If we got inserted at the last slot,
 959                          * that means we don't know if there was a better
 960                          * choice for the last slot, or not.  Rescan.
 961                          */
 962                         needscan = dfp == &bf[2];
 963                 }
 964         }
 965         /*
 966          * We match the last part of the entry.
 967          * Trim the allocated space off the tail of the entry.
 968          */
 969         else if (matchback) {
 970                 newdup = dup;
 971                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
 972                 *xfs_dir2_data_unused_tag_p(newdup) =
 973                         cpu_to_be16((char *)newdup - (char *)hdr);
 974                 xfs_dir2_data_log_unused(tp, bp, newdup);
 975                 /*
 976                  * If it was in the table, remove it and add the new one.
 977                  */
 978                 if (dfp) {
 979                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 980                         dfp = xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 981                         ASSERT(dfp != NULL);
 982                         ASSERT(dfp->length == newdup->length);
 983                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 984                         /*
 985                          * If we got inserted at the last slot,
 986                          * that means we don't know if there was a better
 987                          * choice for the last slot, or not.  Rescan.
 988                          */
 989                         needscan = dfp == &bf[2];
 990                 }
 991         }
 992         /*
 993          * Poking out the middle of an entry.
 994          * Make two new entries.
 995          */
 996         else {
 997                 newdup = dup;
 998                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
 999                 *xfs_dir2_data_unused_tag_p(newdup) =
1000                         cpu_to_be16((char *)newdup - (char *)hdr);
1001                 xfs_dir2_data_log_unused(tp, bp, newdup);
1002                 newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1003                 newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1004                 newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1005                 *xfs_dir2_data_unused_tag_p(newdup2) =
1006                         cpu_to_be16((char *)newdup2 - (char *)hdr);
1007                 xfs_dir2_data_log_unused(tp, bp, newdup2);
1008                 /*
1009                  * If the old entry was in the table, we need to scan
1010                  * if the 3rd entry was valid, since these entries
1011                  * are smaller than the old one.
1012                  * If we don't need to scan that means there were 1 or 2
1013                  * entries in the table, and removing the old and adding
1014                  * the 2 new will work.
1015                  */
1016                 if (dfp) {
1017                         needscan = (bf[2].length != 0);
1018                         if (!needscan) {
1019                                 xfs_dir2_data_freeremove(hdr, dfp, needlogp);
1020                                 xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
1021                                 xfs_dir2_data_freeinsert(hdr, newdup2,
1022                                                          needlogp);
1023                         }
1024                 }
1025         }
1026         *needscanp = needscan;
1027 }