fs/xfs/xfs_dir2_data.c

   1 /*
   2  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_types.h"
  22 #include "xfs_log.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_btree.h"
  28 #include "xfs_bmap_btree.h"
  29 #include "xfs_dinode.h"
  30 #include "xfs_inode.h"
  31 #include "xfs_dir2_format.h"
  32 #include "xfs_dir2.h"
  33 #include "xfs_dir2_priv.h"
  34 #include "xfs_error.h"
  35 #include "xfs_buf_item.h"
  36 #include "xfs_cksum.h"
  37
  38 /*
  39  * Check the consistency of the data block.
  40  * The input can also be a block-format directory.
  41  * Return 0 is the buffer is good, otherwise an error.
  42  */
  43 int
  44 __xfs_dir3_data_check(
  45         struct xfs_inode        *dp,            /* incore inode pointer */
  46         struct xfs_buf          *bp)            /* data block's buffer */
  47 {
  48         xfs_dir2_dataptr_t      addr;           /* addr for leaf lookup */
  49         xfs_dir2_data_free_t    *bf;            /* bestfree table */
  50         xfs_dir2_block_tail_t   *btp=NULL;      /* block tail */
  51         int                     count;          /* count of entries found */
  52         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
  53         xfs_dir2_data_entry_t   *dep;           /* data entry */
  54         xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
  55         xfs_dir2_data_unused_t  *dup;           /* unused entry */
  56         char                    *endp;          /* end of useful data */
  57         int                     freeseen;       /* mask of bestfrees seen */
  58         xfs_dahash_t            hash;           /* hash of current name */
  59         int                     i;              /* leaf index */
  60         int                     lastfree;       /* last entry was unused */
  61         xfs_dir2_leaf_entry_t   *lep=NULL;      /* block leaf entries */
  62         xfs_mount_t             *mp;            /* filesystem mount point */
  63         char                    *p;             /* current data position */
  64         int                     stale;          /* count of stale leaves */
  65         struct xfs_name         name;
  66
  67         mp = bp->b_target->bt_mount;
  68         hdr = bp->b_addr;
  69         bf = xfs_dir3_data_bestfree_p(hdr);
  70         p = (char *)xfs_dir3_data_entry_p(hdr);
  71
  72         switch (hdr->magic) {
  73         case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
  74         case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
  75                 btp = xfs_dir2_block_tail_p(mp, hdr);
  76                 lep = xfs_dir2_block_leaf_p(btp);
  77                 endp = (char *)lep;
  78                 break;
  79         case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
  80         case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
  81                 endp = (char *)hdr + mp->m_dirblksize;
  82                 break;
  83         default:
  84                 XFS_ERROR_REPORT("Bad Magic", XFS_ERRLEVEL_LOW, mp);
  85                 return EFSCORRUPTED;
  86         }
  87
  88         count = lastfree = freeseen = 0;
  89         /*
  90          * Account for zero bestfree entries.
  91          */
  92         if (!bf[0].length) {
  93                 XFS_WANT_CORRUPTED_RETURN(!bf[0].offset);
  94                 freeseen |= 1 << 0;
  95         }
  96         if (!bf[1].length) {
  97                 XFS_WANT_CORRUPTED_RETURN(!bf[1].offset);
  98                 freeseen |= 1 << 1;
  99         }
 100         if (!bf[2].length) {
 101                 XFS_WANT_CORRUPTED_RETURN(!bf[2].offset);
 102                 freeseen |= 1 << 2;
 103         }
 104
 105         XFS_WANT_CORRUPTED_RETURN(be16_to_cpu(bf[0].length) >=
 106                                                 be16_to_cpu(bf[1].length));
 107         XFS_WANT_CORRUPTED_RETURN(be16_to_cpu(bf[1].length) >=
 108                                                 be16_to_cpu(bf[2].length));
 109         /*
 110          * Loop over the data/unused entries.
 111          */
 112         while (p < endp) {
 113                 dup = (xfs_dir2_data_unused_t *)p;
 114                 /*
 115                  * If it's unused, look for the space in the bestfree table.
 116                  * If we find it, account for that, else make sure it
 117                  * doesn't need to be there.
 118                  */
 119                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 120                         XFS_WANT_CORRUPTED_RETURN(lastfree == 0);
 121                         XFS_WANT_CORRUPTED_RETURN(
 122                                 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) ==
 123                                                (char *)dup - (char *)hdr);
 124                         dfp = xfs_dir2_data_freefind(hdr, dup);
 125                         if (dfp) {
 126                                 i = (int)(dfp - bf);
 127                                 XFS_WANT_CORRUPTED_RETURN(
 128                                         (freeseen & (1 << i)) == 0);
 129                                 freeseen |= 1 << i;
 130                         } else {
 131                                 XFS_WANT_CORRUPTED_RETURN(
 132                                         be16_to_cpu(dup->length) <=
 133                                                 be16_to_cpu(bf[2].length));
 134                         }
 135                         p += be16_to_cpu(dup->length);
 136                         lastfree = 1;
 137                         continue;
 138                 }
 139                 /*
 140                  * It's a real entry.  Validate the fields.
 141                  * If this is a block directory then make sure it's
 142                  * in the leaf section of the block.
 143                  * The linear search is crude but this is DEBUG code.
 144                  */
 145                 dep = (xfs_dir2_data_entry_t *)p;
 146                 XFS_WANT_CORRUPTED_RETURN(dep->namelen != 0);
 147                 XFS_WANT_CORRUPTED_RETURN(
 148                         !xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber)));
 149                 XFS_WANT_CORRUPTED_RETURN(
 150                         be16_to_cpu(*xfs_dir2_data_entry_tag_p(dep)) ==
 151                                                (char *)dep - (char *)hdr);
 152                 count++;
 153                 lastfree = 0;
 154                 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 155                     hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 156                         addr = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
 157                                 (xfs_dir2_data_aoff_t)
 158                                 ((char *)dep - (char *)hdr));
 159                         name.name = dep->name;
 160                         name.len = dep->namelen;
 161                         hash = mp->m_dirnameops->hashname(&name);
 162                         for (i = 0; i < be32_to_cpu(btp->count); i++) {
 163                                 if (be32_to_cpu(lep[i].address) == addr &&
 164                                     be32_to_cpu(lep[i].hashval) == hash)
 165                                         break;
 166                         }
 167                         XFS_WANT_CORRUPTED_RETURN(i < be32_to_cpu(btp->count));
 168                 }
 169                 p += xfs_dir2_data_entsize(dep->namelen);
 170         }
 171         /*
 172          * Need to have seen all the entries and all the bestfree slots.
 173          */
 174         XFS_WANT_CORRUPTED_RETURN(freeseen == 7);
 175         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 176             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 177                 for (i = stale = 0; i < be32_to_cpu(btp->count); i++) {
 178                         if (lep[i].address ==
 179                             cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 180                                 stale++;
 181                         if (i > 0)
 182                                 XFS_WANT_CORRUPTED_RETURN(
 183                                         be32_to_cpu(lep[i].hashval) >=
 184                                                 be32_to_cpu(lep[i - 1].hashval));
 185                 }
 186                 XFS_WANT_CORRUPTED_RETURN(count ==
 187                         be32_to_cpu(btp->count) - be32_to_cpu(btp->stale));
 188                 XFS_WANT_CORRUPTED_RETURN(stale == be32_to_cpu(btp->stale));
 189         }
 190         return 0;
 191 }
 192
 193 static bool
 194 xfs_dir3_data_verify(
 195         struct xfs_buf          *bp)
 196 {
 197         struct xfs_mount        *mp = bp->b_target->bt_mount;
 198         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 199
 200         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 201                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 202                         return false;
 203                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
 204                         return false;
 205                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 206                         return false;
 207         } else {
 208                 if (hdr3->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC))
 209                         return false;
 210         }
 211         if (__xfs_dir3_data_check(NULL, bp))
 212                 return false;
 213         return true;
 214 }
 215
 216 /*
 217  * Readahead of the first block of the directory when it is opened is completely
 218  * oblivious to the format of the directory. Hence we can either get a block
 219  * format buffer or a data format buffer on readahead.
 220  */
 221 static void
 222 xfs_dir3_data_reada_verify(
 223         struct xfs_buf          *bp)
 224 {
 225         struct xfs_mount        *mp = bp->b_target->bt_mount;
 226         struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 227
 228         switch (hdr->magic) {
 229         case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
 230         case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
 231                 bp->b_ops = &xfs_dir3_block_buf_ops;
 232                 bp->b_ops->verify_read(bp);
 233                 return;
 234         case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
 235         case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
 236                 xfs_dir3_data_verify(bp);
 237                 return;
 238         default:
 239                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, hdr);
 240                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 241                 break;
 242         }
 243 }
 244
 245 static void
 246 xfs_dir3_data_read_verify(
 247         struct xfs_buf  *bp)
 248 {
 249         struct xfs_mount        *mp = bp->b_target->bt_mount;
 250
 251         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
 252              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
 253                                           XFS_DIR3_DATA_CRC_OFF)) ||
 254             !xfs_dir3_data_verify(bp)) {
 255                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 256                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 257         }
 258 }
 259
 260 static void
 261 xfs_dir3_data_write_verify(
 262         struct xfs_buf  *bp)
 263 {
 264         struct xfs_mount        *mp = bp->b_target->bt_mount;
 265         struct xfs_buf_log_item *bip = bp->b_fspriv;
 266         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 267
 268         if (!xfs_dir3_data_verify(bp)) {
 269                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 270                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 271                 return;
 272         }
 273
 274         if (!xfs_sb_version_hascrc(&mp->m_sb))
 275                 return;
 276
 277         if (bip)
 278                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 279
 280         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_DATA_CRC_OFF);
 281 }
 282
 283 const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
 284         .verify_read = xfs_dir3_data_read_verify,
 285         .verify_write = xfs_dir3_data_write_verify,
 286 };
 287
 288 static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
 289         .verify_read = xfs_dir3_data_reada_verify,
 290         .verify_write = xfs_dir3_data_write_verify,
 291 };
 292
 293
 294 int
 295 xfs_dir3_data_read(
 296         struct xfs_trans        *tp,
 297         struct xfs_inode        *dp,
 298         xfs_dablk_t             bno,
 299         xfs_daddr_t             mapped_bno,
 300         struct xfs_buf          **bpp)
 301 {
 302         int                     err;
 303
 304         err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
 305                                 XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
 306         if (!err && tp)
 307                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
 308         return err;
 309 }
 310
 311 int
 312 xfs_dir3_data_readahead(
 313         struct xfs_trans        *tp,
 314         struct xfs_inode        *dp,
 315         xfs_dablk_t             bno,
 316         xfs_daddr_t             mapped_bno)
 317 {
 318         return xfs_da_reada_buf(tp, dp, bno, mapped_bno,
 319                                 XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
 320 }
 321
 322 /*
 323  * Given a data block and an unused entry from that block,
 324  * return the bestfree entry if any that corresponds to it.
 325  */
 326 xfs_dir2_data_free_t *
 327 xfs_dir2_data_freefind(
 328         xfs_dir2_data_hdr_t     *hdr,           /* data block */
 329         xfs_dir2_data_unused_t  *dup)           /* data unused entry */
 330 {
 331         xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
 332         xfs_dir2_data_aoff_t    off;            /* offset value needed */
 333         struct xfs_dir2_data_free *bf;
 334 #if defined(DEBUG) && defined(__KERNEL__)
 335         int                     matched;        /* matched the value */
 336         int                     seenzero;       /* saw a 0 bestfree entry */
 337 #endif
 338
 339         off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
 340         bf = xfs_dir3_data_bestfree_p(hdr);
 341
 342 #if defined(DEBUG) && defined(__KERNEL__)
 343         /*
 344          * Validate some consistency in the bestfree table.
 345          * Check order, non-overlapping entries, and if we find the
 346          * one we're looking for it has to be exact.
 347          */
 348         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 349                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 350                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 351                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 352         for (dfp = &bf[0], seenzero = matched = 0;
 353              dfp < &bf[XFS_DIR2_DATA_FD_COUNT];
 354              dfp++) {
 355                 if (!dfp->offset) {
 356                         ASSERT(!dfp->length);
 357                         seenzero = 1;
 358                         continue;
 359                 }
 360                 ASSERT(seenzero == 0);
 361                 if (be16_to_cpu(dfp->offset) == off) {
 362                         matched = 1;
 363                         ASSERT(dfp->length == dup->length);
 364                 } else if (off < be16_to_cpu(dfp->offset))
 365                         ASSERT(off + be16_to_cpu(dup->length) <= be16_to_cpu(dfp->offset));
 366                 else
 367                         ASSERT(be16_to_cpu(dfp->offset) + be16_to_cpu(dfp->length) <= off);
 368                 ASSERT(matched || be16_to_cpu(dfp->length) >= be16_to_cpu(dup->length));
 369                 if (dfp > &bf[0])
 370                         ASSERT(be16_to_cpu(dfp[-1].length) >= be16_to_cpu(dfp[0].length));
 371         }
 372 #endif
 373         /*
 374          * If this is smaller than the smallest bestfree entry,
 375          * it can't be there since they're sorted.
 376          */
 377         if (be16_to_cpu(dup->length) <
 378             be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 379                 return NULL;
 380         /*
 381          * Look at the three bestfree entries for our guy.
 382          */
 383         for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 384                 if (!dfp->offset)
 385                         return NULL;
 386                 if (be16_to_cpu(dfp->offset) == off)
 387                         return dfp;
 388         }
 389         /*
 390          * Didn't find it.  This only happens if there are duplicate lengths.
 391          */
 392         return NULL;
 393 }
 394
 395 /*
 396  * Insert an unused-space entry into the bestfree table.
 397  */
 398 xfs_dir2_data_free_t *                          /* entry inserted */
 399 xfs_dir2_data_freeinsert(
 400         xfs_dir2_data_hdr_t     *hdr,           /* data block pointer */
 401         xfs_dir2_data_unused_t  *dup,           /* unused space */
 402         int                     *loghead)       /* log the data header (out) */
 403 {
 404         xfs_dir2_data_free_t    *dfp;           /* bestfree table pointer */
 405         xfs_dir2_data_free_t    new;            /* new bestfree entry */
 406
 407         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 408                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 409                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 410                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 411
 412         dfp = xfs_dir3_data_bestfree_p(hdr);
 413         new.length = dup->length;
 414         new.offset = cpu_to_be16((char *)dup - (char *)hdr);
 415
 416         /*
 417          * Insert at position 0, 1, or 2; or not at all.
 418          */
 419         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
 420                 dfp[2] = dfp[1];
 421                 dfp[1] = dfp[0];
 422                 dfp[0] = new;
 423                 *loghead = 1;
 424                 return &dfp[0];
 425         }
 426         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
 427                 dfp[2] = dfp[1];
 428                 dfp[1] = new;
 429                 *loghead = 1;
 430                 return &dfp[1];
 431         }
 432         if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
 433                 dfp[2] = new;
 434                 *loghead = 1;
 435                 return &dfp[2];
 436         }
 437         return NULL;
 438 }
 439
 440 /*
 441  * Remove a bestfree entry from the table.
 442  */
 443 STATIC void
 444 xfs_dir2_data_freeremove(
 445         xfs_dir2_data_hdr_t     *hdr,           /* data block header */
 446         xfs_dir2_data_free_t    *dfp,           /* bestfree entry pointer */
 447         int                     *loghead)       /* out: log data header */
 448 {
 449         struct xfs_dir2_data_free *bf;
 450
 451         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 452                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 453                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 454                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 455
 456         /*
 457          * It's the first entry, slide the next 2 up.
 458          */
 459         bf = xfs_dir3_data_bestfree_p(hdr);
 460         if (dfp == &bf[0]) {
 461                 bf[0] = bf[1];
 462                 bf[1] = bf[2];
 463         }
 464         /*
 465          * It's the second entry, slide the 3rd entry up.
 466          */
 467         else if (dfp == &bf[1])
 468                 bf[1] = bf[2];
 469         /*
 470          * Must be the last entry.
 471          */
 472         else
 473                 ASSERT(dfp == &bf[2]);
 474         /*
 475          * Clear the 3rd entry, must be zero now.
 476          */
 477         bf[2].length = 0;
 478         bf[2].offset = 0;
 479         *loghead = 1;
 480 }
 481
 482 /*
 483  * Given a data block, reconstruct its bestfree map.
 484  */
 485 void
 486 xfs_dir2_data_freescan(
 487         xfs_mount_t             *mp,            /* filesystem mount point */
 488         xfs_dir2_data_hdr_t     *hdr,           /* data block header */
 489         int                     *loghead)       /* out: log data header */
 490 {
 491         xfs_dir2_block_tail_t   *btp;           /* block tail */
 492         xfs_dir2_data_entry_t   *dep;           /* active data entry */
 493         xfs_dir2_data_unused_t  *dup;           /* unused data entry */
 494         struct xfs_dir2_data_free *bf;
 495         char                    *endp;          /* end of block's data */
 496         char                    *p;             /* current entry pointer */
 497
 498         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 499                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 500                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 501                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 502
 503         /*
 504          * Start by clearing the table.
 505          */
 506         bf = xfs_dir3_data_bestfree_p(hdr);
 507         memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
 508         *loghead = 1;
 509         /*
 510          * Set up pointers.
 511          */
 512         p = (char *)xfs_dir3_data_entry_p(hdr);
 513         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 514             hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 515                 btp = xfs_dir2_block_tail_p(mp, hdr);
 516                 endp = (char *)xfs_dir2_block_leaf_p(btp);
 517         } else
 518                 endp = (char *)hdr + mp->m_dirblksize;
 519         /*
 520          * Loop over the block's entries.
 521          */
 522         while (p < endp) {
 523                 dup = (xfs_dir2_data_unused_t *)p;
 524                 /*
 525                  * If it's a free entry, insert it.
 526                  */
 527                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 528                         ASSERT((char *)dup - (char *)hdr ==
 529                                be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 530                         xfs_dir2_data_freeinsert(hdr, dup, loghead);
 531                         p += be16_to_cpu(dup->length);
 532                 }
 533                 /*
 534                  * For active entries, check their tags and skip them.
 535                  */
 536                 else {
 537                         dep = (xfs_dir2_data_entry_t *)p;
 538                         ASSERT((char *)dep - (char *)hdr ==
 539                                be16_to_cpu(*xfs_dir2_data_entry_tag_p(dep)));
 540                         p += xfs_dir2_data_entsize(dep->namelen);
 541                 }
 542         }
 543 }
 544
 545 /*
 546  * Initialize a data block at the given block number in the directory.
 547  * Give back the buffer for the created block.
 548  */
 549 int                                             /* error */
 550 xfs_dir3_data_init(
 551         xfs_da_args_t           *args,          /* directory operation args */
 552         xfs_dir2_db_t           blkno,          /* logical dir block number */
 553         struct xfs_buf          **bpp)          /* output block buffer */
 554 {
 555         struct xfs_buf          *bp;            /* block buffer */
 556         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 557         xfs_inode_t             *dp;            /* incore directory inode */
 558         xfs_dir2_data_unused_t  *dup;           /* unused entry pointer */
 559         struct xfs_dir2_data_free *bf;
 560         int                     error;          /* error return value */
 561         int                     i;              /* bestfree index */
 562         xfs_mount_t             *mp;            /* filesystem mount point */
 563         xfs_trans_t             *tp;            /* transaction pointer */
 564         int                     t;              /* temp */
 565
 566         dp = args->dp;
 567         mp = dp->i_mount;
 568         tp = args->trans;
 569         /*
 570          * Get the buffer set up for the block.
 571          */
 572         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, blkno), -1, &bp,
 573                 XFS_DATA_FORK);
 574         if (error)
 575                 return error;
 576         bp->b_ops = &xfs_dir3_data_buf_ops;
 577         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
 578
 579         /*
 580          * Initialize the header.
 581          */
 582         hdr = bp->b_addr;
 583         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 584                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 585
 586                 memset(hdr3, 0, sizeof(*hdr3));
 587                 hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 588                 hdr3->blkno = cpu_to_be64(bp->b_bn);
 589                 hdr3->owner = cpu_to_be64(dp->i_ino);
 590                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
 591
 592         } else
 593                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 594
 595         bf = xfs_dir3_data_bestfree_p(hdr);
 596         bf[0].offset = cpu_to_be16(xfs_dir3_data_entry_offset(hdr));
 597         for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
 598                 bf[i].length = 0;
 599                 bf[i].offset = 0;
 600         }
 601
 602         /*
 603          * Set up an unused entry for the block's body.
 604          */
 605         dup = xfs_dir3_data_unused_p(hdr);
 606         dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 607
 608         t = mp->m_dirblksize - (uint)xfs_dir3_data_entry_offset(hdr);
 609         bf[0].length = cpu_to_be16(t);
 610         dup->length = cpu_to_be16(t);
 611         *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
 612         /*
 613          * Log it and return it.
 614          */
 615         xfs_dir2_data_log_header(tp, bp);
 616         xfs_dir2_data_log_unused(tp, bp, dup);
 617         *bpp = bp;
 618         return 0;
 619 }
 620
 621 /*
 622  * Log an active data entry from the block.
 623  */
 624 void
 625 xfs_dir2_data_log_entry(
 626         struct xfs_trans        *tp,
 627         struct xfs_buf          *bp,
 628         xfs_dir2_data_entry_t   *dep)           /* data entry pointer */
 629 {
 630         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 631
 632         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 633                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 634                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 635                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 636
 637         xfs_trans_log_buf(tp, bp, (uint)((char *)dep - (char *)hdr),
 638                 (uint)((char *)(xfs_dir2_data_entry_tag_p(dep) + 1) -
 639                        (char *)hdr - 1));
 640 }
 641
 642 /*
 643  * Log a data block header.
 644  */
 645 void
 646 xfs_dir2_data_log_header(
 647         struct xfs_trans        *tp,
 648         struct xfs_buf          *bp)
 649 {
 650         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 651
 652         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 653                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 654                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 655                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 656
 657         xfs_trans_log_buf(tp, bp, 0, xfs_dir3_data_entry_offset(hdr) - 1);
 658 }
 659
 660 /*
 661  * Log a data unused entry.
 662  */
 663 void
 664 xfs_dir2_data_log_unused(
 665         struct xfs_trans        *tp,
 666         struct xfs_buf          *bp,
 667         xfs_dir2_data_unused_t  *dup)           /* data unused pointer */
 668 {
 669         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 670
 671         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 672                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 673                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 674                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 675
 676         /*
 677          * Log the first part of the unused entry.
 678          */
 679         xfs_trans_log_buf(tp, bp, (uint)((char *)dup - (char *)hdr),
 680                 (uint)((char *)&dup->length + sizeof(dup->length) -
 681                        1 - (char *)hdr));
 682         /*
 683          * Log the end (tag) of the unused entry.
 684          */
 685         xfs_trans_log_buf(tp, bp,
 686                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
 687                 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
 688                        sizeof(xfs_dir2_data_off_t) - 1));
 689 }
 690
 691 /*
 692  * Make a byte range in the data block unused.
 693  * Its current contents are unimportant.
 694  */
 695 void
 696 xfs_dir2_data_make_free(
 697         struct xfs_trans        *tp,
 698         struct xfs_buf          *bp,
 699         xfs_dir2_data_aoff_t    offset,         /* starting byte offset */
 700         xfs_dir2_data_aoff_t    len,            /* length in bytes */
 701         int                     *needlogp,      /* out: log header */
 702         int                     *needscanp)     /* out: regen bestfree */
 703 {
 704         xfs_dir2_data_hdr_t     *hdr;           /* data block pointer */
 705         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 706         char                    *endptr;        /* end of data area */
 707         xfs_mount_t             *mp;            /* filesystem mount point */
 708         int                     needscan;       /* need to regen bestfree */
 709         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 710         xfs_dir2_data_unused_t  *postdup;       /* unused entry after us */
 711         xfs_dir2_data_unused_t  *prevdup;       /* unused entry before us */
 712         struct xfs_dir2_data_free *bf;
 713
 714         mp = tp->t_mountp;
 715         hdr = bp->b_addr;
 716
 717         /*
 718          * Figure out where the end of the data area is.
 719          */
 720         if (hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 721             hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 722                 endptr = (char *)hdr + mp->m_dirblksize;
 723         else {
 724                 xfs_dir2_block_tail_t   *btp;   /* block tail */
 725
 726                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 727                         hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 728                 btp = xfs_dir2_block_tail_p(mp, hdr);
 729                 endptr = (char *)xfs_dir2_block_leaf_p(btp);
 730         }
 731         /*
 732          * If this isn't the start of the block, then back up to
 733          * the previous entry and see if it's free.
 734          */
 735         if (offset > xfs_dir3_data_entry_offset(hdr)) {
 736                 __be16                  *tagp;  /* tag just before us */
 737
 738                 tagp = (__be16 *)((char *)hdr + offset) - 1;
 739                 prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 740                 if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 741                         prevdup = NULL;
 742         } else
 743                 prevdup = NULL;
 744         /*
 745          * If this isn't the end of the block, see if the entry after
 746          * us is free.
 747          */
 748         if ((char *)hdr + offset + len < endptr) {
 749                 postdup =
 750                         (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 751                 if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 752                         postdup = NULL;
 753         } else
 754                 postdup = NULL;
 755         ASSERT(*needscanp == 0);
 756         needscan = 0;
 757         /*
 758          * Previous and following entries are both free,
 759          * merge everything into a single free entry.
 760          */
 761         bf = xfs_dir3_data_bestfree_p(hdr);
 762         if (prevdup && postdup) {
 763                 xfs_dir2_data_free_t    *dfp2;  /* another bestfree pointer */
 764
 765                 /*
 766                  * See if prevdup and/or postdup are in bestfree table.
 767                  */
 768                 dfp = xfs_dir2_data_freefind(hdr, prevdup);
 769                 dfp2 = xfs_dir2_data_freefind(hdr, postdup);
 770                 /*
 771                  * We need a rescan unless there are exactly 2 free entries
 772                  * namely our two.  Then we know what's happening, otherwise
 773                  * since the third bestfree is there, there might be more
 774                  * entries.
 775                  */
 776                 needscan = (bf[2].length != 0);
 777                 /*
 778                  * Fix up the new big freespace.
 779                  */
 780                 be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
 781                 *xfs_dir2_data_unused_tag_p(prevdup) =
 782                         cpu_to_be16((char *)prevdup - (char *)hdr);
 783                 xfs_dir2_data_log_unused(tp, bp, prevdup);
 784                 if (!needscan) {
 785                         /*
 786                          * Has to be the case that entries 0 and 1 are
 787                          * dfp and dfp2 (don't know which is which), and
 788                          * entry 2 is empty.
 789                          * Remove entry 1 first then entry 0.
 790                          */
 791                         ASSERT(dfp && dfp2);
 792                         if (dfp == &bf[1]) {
 793                                 dfp = &bf[0];
 794                                 ASSERT(dfp2 == dfp);
 795                                 dfp2 = &bf[1];
 796                         }
 797                         xfs_dir2_data_freeremove(hdr, dfp2, needlogp);
 798                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 799                         /*
 800                          * Now insert the new entry.
 801                          */
 802                         dfp = xfs_dir2_data_freeinsert(hdr, prevdup, needlogp);
 803                         ASSERT(dfp == &bf[0]);
 804                         ASSERT(dfp->length == prevdup->length);
 805                         ASSERT(!dfp[1].length);
 806                         ASSERT(!dfp[2].length);
 807                 }
 808         }
 809         /*
 810          * The entry before us is free, merge with it.
 811          */
 812         else if (prevdup) {
 813                 dfp = xfs_dir2_data_freefind(hdr, prevdup);
 814                 be16_add_cpu(&prevdup->length, len);
 815                 *xfs_dir2_data_unused_tag_p(prevdup) =
 816                         cpu_to_be16((char *)prevdup - (char *)hdr);
 817                 xfs_dir2_data_log_unused(tp, bp, prevdup);
 818                 /*
 819                  * If the previous entry was in the table, the new entry
 820                  * is longer, so it will be in the table too.  Remove
 821                  * the old one and add the new one.
 822                  */
 823                 if (dfp) {
 824                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 825                         xfs_dir2_data_freeinsert(hdr, prevdup, needlogp);
 826                 }
 827                 /*
 828                  * Otherwise we need a scan if the new entry is big enough.
 829                  */
 830                 else {
 831                         needscan = be16_to_cpu(prevdup->length) >
 832                                    be16_to_cpu(bf[2].length);
 833                 }
 834         }
 835         /*
 836          * The following entry is free, merge with it.
 837          */
 838         else if (postdup) {
 839                 dfp = xfs_dir2_data_freefind(hdr, postdup);
 840                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 841                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 842                 newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
 843                 *xfs_dir2_data_unused_tag_p(newdup) =
 844                         cpu_to_be16((char *)newdup - (char *)hdr);
 845                 xfs_dir2_data_log_unused(tp, bp, newdup);
 846                 /*
 847                  * If the following entry was in the table, the new entry
 848                  * is longer, so it will be in the table too.  Remove
 849                  * the old one and add the new one.
 850                  */
 851                 if (dfp) {
 852                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 853                         xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 854                 }
 855                 /*
 856                  * Otherwise we need a scan if the new entry is big enough.
 857                  */
 858                 else {
 859                         needscan = be16_to_cpu(newdup->length) >
 860                                    be16_to_cpu(bf[2].length);
 861                 }
 862         }
 863         /*
 864          * Neither neighbor is free.  Make a new entry.
 865          */
 866         else {
 867                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 868                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 869                 newdup->length = cpu_to_be16(len);
 870                 *xfs_dir2_data_unused_tag_p(newdup) =
 871                         cpu_to_be16((char *)newdup - (char *)hdr);
 872                 xfs_dir2_data_log_unused(tp, bp, newdup);
 873                 xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 874         }
 875         *needscanp = needscan;
 876 }
 877
 878 /*
 879  * Take a byte range out of an existing unused space and make it un-free.
 880  */
 881 void
 882 xfs_dir2_data_use_free(
 883         struct xfs_trans        *tp,
 884         struct xfs_buf          *bp,
 885         xfs_dir2_data_unused_t  *dup,           /* unused entry */
 886         xfs_dir2_data_aoff_t    offset,         /* starting offset to use */
 887         xfs_dir2_data_aoff_t    len,            /* length to use */
 888         int                     *needlogp,      /* out: need to log header */
 889         int                     *needscanp)     /* out: need regen bestfree */
 890 {
 891         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 892         xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 893         int                     matchback;      /* matches end of freespace */
 894         int                     matchfront;     /* matches start of freespace */
 895         int                     needscan;       /* need to regen bestfree */
 896         xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 897         xfs_dir2_data_unused_t  *newdup2;       /* another new unused entry */
 898         int                     oldlen;         /* old unused entry's length */
 899         struct xfs_dir2_data_free *bf;
 900
 901         hdr = bp->b_addr;
 902         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 903                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 904                hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 905                hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 906         ASSERT(be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG);
 907         ASSERT(offset >= (char *)dup - (char *)hdr);
 908         ASSERT(offset + len <= (char *)dup + be16_to_cpu(dup->length) - (char *)hdr);
 909         ASSERT((char *)dup - (char *)hdr == be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 910         /*
 911          * Look up the entry in the bestfree table.
 912          */
 913         dfp = xfs_dir2_data_freefind(hdr, dup);
 914         oldlen = be16_to_cpu(dup->length);
 915         bf = xfs_dir3_data_bestfree_p(hdr);
 916         ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
 917         /*
 918          * Check for alignment with front and back of the entry.
 919          */
 920         matchfront = (char *)dup - (char *)hdr == offset;
 921         matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
 922         ASSERT(*needscanp == 0);
 923         needscan = 0;
 924         /*
 925          * If we matched it exactly we just need to get rid of it from
 926          * the bestfree table.
 927          */
 928         if (matchfront && matchback) {
 929                 if (dfp) {
 930                         needscan = (bf[2].offset != 0);
 931                         if (!needscan)
 932                                 xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 933                 }
 934         }
 935         /*
 936          * We match the first part of the entry.
 937          * Make a new entry with the remaining freespace.
 938          */
 939         else if (matchfront) {
 940                 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 941                 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 942                 newdup->length = cpu_to_be16(oldlen - len);
 943                 *xfs_dir2_data_unused_tag_p(newdup) =
 944                         cpu_to_be16((char *)newdup - (char *)hdr);
 945                 xfs_dir2_data_log_unused(tp, bp, newdup);
 946                 /*
 947                  * If it was in the table, remove it and add the new one.
 948                  */
 949                 if (dfp) {
 950                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 951                         dfp = xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 952                         ASSERT(dfp != NULL);
 953                         ASSERT(dfp->length == newdup->length);
 954                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 955                         /*
 956                          * If we got inserted at the last slot,
 957                          * that means we don't know if there was a better
 958                          * choice for the last slot, or not.  Rescan.
 959                          */
 960                         needscan = dfp == &bf[2];
 961                 }
 962         }
 963         /*
 964          * We match the last part of the entry.
 965          * Trim the allocated space off the tail of the entry.
 966          */
 967         else if (matchback) {
 968                 newdup = dup;
 969                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
 970                 *xfs_dir2_data_unused_tag_p(newdup) =
 971                         cpu_to_be16((char *)newdup - (char *)hdr);
 972                 xfs_dir2_data_log_unused(tp, bp, newdup);
 973                 /*
 974                  * If it was in the table, remove it and add the new one.
 975                  */
 976                 if (dfp) {
 977                         xfs_dir2_data_freeremove(hdr, dfp, needlogp);
 978                         dfp = xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
 979                         ASSERT(dfp != NULL);
 980                         ASSERT(dfp->length == newdup->length);
 981                         ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 982                         /*
 983                          * If we got inserted at the last slot,
 984                          * that means we don't know if there was a better
 985                          * choice for the last slot, or not.  Rescan.
 986                          */
 987                         needscan = dfp == &bf[2];
 988                 }
 989         }
 990         /*
 991          * Poking out the middle of an entry.
 992          * Make two new entries.
 993          */
 994         else {
 995                 newdup = dup;
 996                 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
 997                 *xfs_dir2_data_unused_tag_p(newdup) =
 998                         cpu_to_be16((char *)newdup - (char *)hdr);
 999                 xfs_dir2_data_log_unused(tp, bp, newdup);
1000                 newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1001                 newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1002                 newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1003                 *xfs_dir2_data_unused_tag_p(newdup2) =
1004                         cpu_to_be16((char *)newdup2 - (char *)hdr);
1005                 xfs_dir2_data_log_unused(tp, bp, newdup2);
1006                 /*
1007                  * If the old entry was in the table, we need to scan
1008                  * if the 3rd entry was valid, since these entries
1009                  * are smaller than the old one.
1010                  * If we don't need to scan that means there were 1 or 2
1011                  * entries in the table, and removing the old and adding
1012                  * the 2 new will work.
1013                  */
1014                 if (dfp) {
1015                         needscan = (bf[2].length != 0);
1016                         if (!needscan) {
1017                                 xfs_dir2_data_freeremove(hdr, dfp, needlogp);
1018                                 xfs_dir2_data_freeinsert(hdr, newdup, needlogp);
1019                                 xfs_dir2_data_freeinsert(hdr, newdup2,
1020                                                          needlogp);
1021                         }
1022                 }
1023         }
1024         *needscanp = needscan;
1025 }