fs/xfs/xfs_dir2_block.c

   1 /*
   2  * Copyright (c) 2000-2003,2005 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_inode_item.h"
  32 #include "xfs_buf_item.h"
  33 #include "xfs_dir2.h"
  34 #include "xfs_dir2_format.h"
  35 #include "xfs_dir2_priv.h"
  36 #include "xfs_error.h"
  37 #include "xfs_trace.h"
  38 #include "xfs_cksum.h"
  39
  40 /*
  41  * Local function prototypes.
  42  */
  43 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, struct xfs_buf *bp,
  44                                     int first, int last);
  45 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, struct xfs_buf *bp);
  46 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, struct xfs_buf **bpp,
  47                                      int *entno);
  48 static int xfs_dir2_block_sort(const void *a, const void *b);
  49
  50 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
  51
  52 /*
  53  * One-time startup routine called from xfs_init().
  54  */
  55 void
  56 xfs_dir_startup(void)
  57 {
  58         xfs_dir_hash_dot = xfs_da_hashname((unsigned char *)".", 1);
  59         xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
  60 }
  61
  62 static bool
  63 xfs_dir3_block_verify(
  64         struct xfs_buf          *bp)
  65 {
  66         struct xfs_mount        *mp = bp->b_target->bt_mount;
  67         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
  68
  69         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  70                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_BLOCK_MAGIC))
  71                         return false;
  72                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
  73                         return false;
  74                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
  75                         return false;
  76         } else {
  77                 if (hdr3->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
  78                         return false;
  79         }
  80         if (__xfs_dir3_data_check(NULL, bp))
  81                 return false;
  82         return true;
  83 }
  84
  85 static void
  86 xfs_dir3_block_read_verify(
  87         struct xfs_buf  *bp)
  88 {
  89         struct xfs_mount        *mp = bp->b_target->bt_mount;
  90
  91         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
  92              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
  93                                           XFS_DIR3_DATA_CRC_OFF)) ||
  94             !xfs_dir3_block_verify(bp)) {
  95                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
  96                 xfs_buf_ioerror(bp, EFSCORRUPTED);
  97         }
  98 }
  99
 100 static void
 101 xfs_dir3_block_write_verify(
 102         struct xfs_buf  *bp)
 103 {
 104         struct xfs_mount        *mp = bp->b_target->bt_mount;
 105         struct xfs_buf_log_item *bip = bp->b_fspriv;
 106         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 107
 108         if (!xfs_dir3_block_verify(bp)) {
 109                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 110                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 111                 return;
 112         }
 113
 114         if (!xfs_sb_version_hascrc(&mp->m_sb))
 115                 return;
 116
 117         if (bip)
 118                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 119
 120         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_DATA_CRC_OFF);
 121 }
 122
 123 const struct xfs_buf_ops xfs_dir3_block_buf_ops = {
 124         .verify_read = xfs_dir3_block_read_verify,
 125         .verify_write = xfs_dir3_block_write_verify,
 126 };
 127
 128 static int
 129 xfs_dir3_block_read(
 130         struct xfs_trans        *tp,
 131         struct xfs_inode        *dp,
 132         struct xfs_buf          **bpp)
 133 {
 134         struct xfs_mount        *mp = dp->i_mount;
 135         int                     err;
 136
 137         err = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, bpp,
 138                                 XFS_DATA_FORK, &xfs_dir3_block_buf_ops);
 139         if (!err && tp)
 140                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_BLOCK_BUF);
 141         return err;
 142 }
 143
 144 static void
 145 xfs_dir3_block_init(
 146         struct xfs_mount        *mp,
 147         struct xfs_trans        *tp,
 148         struct xfs_buf          *bp,
 149         struct xfs_inode        *dp)
 150 {
 151         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 152
 153         bp->b_ops = &xfs_dir3_block_buf_ops;
 154         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_BLOCK_BUF);
 155
 156         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 157                 memset(hdr3, 0, sizeof(*hdr3));
 158                 hdr3->magic = cpu_to_be32(XFS_DIR3_BLOCK_MAGIC);
 159                 hdr3->blkno = cpu_to_be64(bp->b_bn);
 160                 hdr3->owner = cpu_to_be64(dp->i_ino);
 161                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
 162                 return;
 163
 164         }
 165         hdr3->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
 166 }
 167
 168 static void
 169 xfs_dir2_block_need_space(
 170         struct xfs_dir2_data_hdr        *hdr,
 171         struct xfs_dir2_block_tail      *btp,
 172         struct xfs_dir2_leaf_entry      *blp,
 173         __be16                          **tagpp,
 174         struct xfs_dir2_data_unused     **dupp,
 175         struct xfs_dir2_data_unused     **enddupp,
 176         int                             *compact,
 177         int                             len)
 178 {
 179         struct xfs_dir2_data_free       *bf;
 180         __be16                          *tagp = NULL;
 181         struct xfs_dir2_data_unused     *dup = NULL;
 182         struct xfs_dir2_data_unused     *enddup = NULL;
 183
 184         *compact = 0;
 185         bf = xfs_dir3_data_bestfree_p(hdr);
 186
 187         /*
 188          * If there are stale entries we'll use one for the leaf.
 189          */
 190         if (btp->stale) {
 191                 if (be16_to_cpu(bf[0].length) >= len) {
 192                         /*
 193                          * The biggest entry enough to avoid compaction.
 194                          */
 195                         dup = (xfs_dir2_data_unused_t *)
 196                               ((char *)hdr + be16_to_cpu(bf[0].offset));
 197                         goto out;
 198                 }
 199
 200                 /*
 201                  * Will need to compact to make this work.
 202                  * Tag just before the first leaf entry.
 203                  */
 204                 *compact = 1;
 205                 tagp = (__be16 *)blp - 1;
 206
 207                 /* Data object just before the first leaf entry.  */
 208                 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 209
 210                 /*
 211                  * If it's not free then the data will go where the
 212                  * leaf data starts now, if it works at all.
 213                  */
 214                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 215                         if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
 216                             (uint)sizeof(*blp) < len)
 217                                 dup = NULL;
 218                 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
 219                         dup = NULL;
 220                 else
 221                         dup = (xfs_dir2_data_unused_t *)blp;
 222                 goto out;
 223         }
 224
 225         /*
 226          * no stale entries, so just use free space.
 227          * Tag just before the first leaf entry.
 228          */
 229         tagp = (__be16 *)blp - 1;
 230
 231         /* Data object just before the first leaf entry.  */
 232         enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 233
 234         /*
 235          * If it's not free then can't do this add without cleaning up:
 236          * the space before the first leaf entry needs to be free so it
 237          * can be expanded to hold the pointer to the new entry.
 238          */
 239         if (be16_to_cpu(enddup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 240                 /*
 241                  * Check out the biggest freespace and see if it's the same one.
 242                  */
 243                 dup = (xfs_dir2_data_unused_t *)
 244                       ((char *)hdr + be16_to_cpu(bf[0].offset));
 245                 if (dup != enddup) {
 246                         /*
 247                          * Not the same free entry, just check its length.
 248                          */
 249                         if (be16_to_cpu(dup->length) < len)
 250                                 dup = NULL;
 251                         goto out;
 252                 }
 253
 254                 /*
 255                  * It is the biggest freespace, can it hold the leaf too?
 256                  */
 257                 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
 258                         /*
 259                          * Yes, use the second-largest entry instead if it works.
 260                          */
 261                         if (be16_to_cpu(bf[1].length) >= len)
 262                                 dup = (xfs_dir2_data_unused_t *)
 263                                       ((char *)hdr + be16_to_cpu(bf[1].offset));
 264                         else
 265                                 dup = NULL;
 266                 }
 267         }
 268 out:
 269         *tagpp = tagp;
 270         *dupp = dup;
 271         *enddupp = enddup;
 272 }
 273
 274 /*
 275  * compact the leaf entries.
 276  * Leave the highest-numbered stale entry stale.
 277  * XXX should be the one closest to mid but mid is not yet computed.
 278  */
 279 static void
 280 xfs_dir2_block_compact(
 281         struct xfs_trans                *tp,
 282         struct xfs_buf                  *bp,
 283         struct xfs_dir2_data_hdr        *hdr,
 284         struct xfs_dir2_block_tail      *btp,
 285         struct xfs_dir2_leaf_entry      *blp,
 286         int                             *needlog,
 287         int                             *lfloghigh,
 288         int                             *lfloglow)
 289 {
 290         int                     fromidx;        /* source leaf index */
 291         int                     toidx;          /* target leaf index */
 292         int                     needscan = 0;
 293         int                     highstale;      /* high stale index */
 294
 295         fromidx = toidx = be32_to_cpu(btp->count) - 1;
 296         highstale = *lfloghigh = -1;
 297         for (; fromidx >= 0; fromidx--) {
 298                 if (blp[fromidx].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
 299                         if (highstale == -1)
 300                                 highstale = toidx;
 301                         else {
 302                                 if (*lfloghigh == -1)
 303                                         *lfloghigh = toidx;
 304                                 continue;
 305                         }
 306                 }
 307                 if (fromidx < toidx)
 308                         blp[toidx] = blp[fromidx];
 309                 toidx--;
 310         }
 311         *lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
 312         *lfloghigh -= be32_to_cpu(btp->stale) - 1;
 313         be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
 314         xfs_dir2_data_make_free(tp, bp,
 315                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 316                 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
 317                 needlog, &needscan);
 318         blp += be32_to_cpu(btp->stale) - 1;
 319         btp->stale = cpu_to_be32(1);
 320         /*
 321          * If we now need to rebuild the bestfree map, do so.
 322          * This needs to happen before the next call to use_free.
 323          */
 324         if (needscan)
 325                 xfs_dir2_data_freescan(tp->t_mountp, hdr, needlog);
 326 }
 327
 328 /*
 329  * Add an entry to a block directory.
 330  */
 331 int                                             /* error */
 332 xfs_dir2_block_addname(
 333         xfs_da_args_t           *args)          /* directory op arguments */
 334 {
 335         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 336         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 337         struct xfs_buf          *bp;            /* buffer for block */
 338         xfs_dir2_block_tail_t   *btp;           /* block tail */
 339         int                     compact;        /* need to compact leaf ents */
 340         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 341         xfs_inode_t             *dp;            /* directory inode */
 342         xfs_dir2_data_unused_t  *dup;           /* block unused entry */
 343         int                     error;          /* error return value */
 344         xfs_dir2_data_unused_t  *enddup=NULL;   /* unused at end of data */
 345         xfs_dahash_t            hash;           /* hash value of found entry */
 346         int                     high;           /* high index for binary srch */
 347         int                     highstale;      /* high stale index */
 348         int                     lfloghigh=0;    /* last final leaf to log */
 349         int                     lfloglow=0;     /* first final leaf to log */
 350         int                     len;            /* length of the new entry */
 351         int                     low;            /* low index for binary srch */
 352         int                     lowstale;       /* low stale index */
 353         int                     mid=0;          /* midpoint for binary srch */
 354         xfs_mount_t             *mp;            /* filesystem mount point */
 355         int                     needlog;        /* need to log header */
 356         int                     needscan;       /* need to rescan freespace */
 357         __be16                  *tagp;          /* pointer to tag value */
 358         xfs_trans_t             *tp;            /* transaction structure */
 359
 360         trace_xfs_dir2_block_addname(args);
 361
 362         dp = args->dp;
 363         tp = args->trans;
 364         mp = dp->i_mount;
 365
 366         /* Read the (one and only) directory block into bp. */
 367         error = xfs_dir3_block_read(tp, dp, &bp);
 368         if (error)
 369                 return error;
 370
 371         len = xfs_dir2_data_entsize(args->namelen);
 372
 373         /*
 374          * Set up pointers to parts of the block.
 375          */
 376         hdr = bp->b_addr;
 377         btp = xfs_dir2_block_tail_p(mp, hdr);
 378         blp = xfs_dir2_block_leaf_p(btp);
 379
 380         /*
 381          * Find out if we can reuse stale entries or whether we need extra
 382          * space for entry and new leaf.
 383          */
 384         xfs_dir2_block_need_space(hdr, btp, blp, &tagp, &dup,
 385                                   &enddup, &compact, len);
 386
 387         /*
 388          * Done everything we need for a space check now.
 389          */
 390         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 391                 xfs_trans_brelse(tp, bp);
 392                 if (!dup)
 393                         return XFS_ERROR(ENOSPC);
 394                 return 0;
 395         }
 396
 397         /*
 398          * If we don't have space for the new entry & leaf ...
 399          */
 400         if (!dup) {
 401                 /* Don't have a space reservation: return no-space.  */
 402                 if (args->total == 0)
 403                         return XFS_ERROR(ENOSPC);
 404                 /*
 405                  * Convert to the next larger format.
 406                  * Then add the new entry in that format.
 407                  */
 408                 error = xfs_dir2_block_to_leaf(args, bp);
 409                 if (error)
 410                         return error;
 411                 return xfs_dir2_leaf_addname(args);
 412         }
 413
 414         needlog = needscan = 0;
 415
 416         /*
 417          * If need to compact the leaf entries, do it now.
 418          */
 419         if (compact) {
 420                 xfs_dir2_block_compact(tp, bp, hdr, btp, blp, &needlog,
 421                                       &lfloghigh, &lfloglow);
 422                 /* recalculate blp post-compaction */
 423                 blp = xfs_dir2_block_leaf_p(btp);
 424         } else if (btp->stale) {
 425                 /*
 426                  * Set leaf logging boundaries to impossible state.
 427                  * For the no-stale case they're set explicitly.
 428                  */
 429                 lfloglow = be32_to_cpu(btp->count);
 430                 lfloghigh = -1;
 431         }
 432
 433         /*
 434          * Find the slot that's first lower than our hash value, -1 if none.
 435          */
 436         for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
 437                 mid = (low + high) >> 1;
 438                 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
 439                         break;
 440                 if (hash < args->hashval)
 441                         low = mid + 1;
 442                 else
 443                         high = mid - 1;
 444         }
 445         while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
 446                 mid--;
 447         }
 448         /*
 449          * No stale entries, will use enddup space to hold new leaf.
 450          */
 451         if (!btp->stale) {
 452                 /*
 453                  * Mark the space needed for the new leaf entry, now in use.
 454                  */
 455                 xfs_dir2_data_use_free(tp, bp, enddup,
 456                         (xfs_dir2_data_aoff_t)
 457                         ((char *)enddup - (char *)hdr + be16_to_cpu(enddup->length) -
 458                          sizeof(*blp)),
 459                         (xfs_dir2_data_aoff_t)sizeof(*blp),
 460                         &needlog, &needscan);
 461                 /*
 462                  * Update the tail (entry count).
 463                  */
 464                 be32_add_cpu(&btp->count, 1);
 465                 /*
 466                  * If we now need to rebuild the bestfree map, do so.
 467                  * This needs to happen before the next call to use_free.
 468                  */
 469                 if (needscan) {
 470                         xfs_dir2_data_freescan(mp, hdr, &needlog);
 471                         needscan = 0;
 472                 }
 473                 /*
 474                  * Adjust pointer to the first leaf entry, we're about to move
 475                  * the table up one to open up space for the new leaf entry.
 476                  * Then adjust our index to match.
 477                  */
 478                 blp--;
 479                 mid++;
 480                 if (mid)
 481                         memmove(blp, &blp[1], mid * sizeof(*blp));
 482                 lfloglow = 0;
 483                 lfloghigh = mid;
 484         }
 485         /*
 486          * Use a stale leaf for our new entry.
 487          */
 488         else {
 489                 for (lowstale = mid;
 490                      lowstale >= 0 &&
 491                         blp[lowstale].address !=
 492                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 493                      lowstale--)
 494                         continue;
 495                 for (highstale = mid + 1;
 496                      highstale < be32_to_cpu(btp->count) &&
 497                         blp[highstale].address !=
 498                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
 499                         (lowstale < 0 || mid - lowstale > highstale - mid);
 500                      highstale++)
 501                         continue;
 502                 /*
 503                  * Move entries toward the low-numbered stale entry.
 504                  */
 505                 if (lowstale >= 0 &&
 506                     (highstale == be32_to_cpu(btp->count) ||
 507                      mid - lowstale <= highstale - mid)) {
 508                         if (mid - lowstale)
 509                                 memmove(&blp[lowstale], &blp[lowstale + 1],
 510                                         (mid - lowstale) * sizeof(*blp));
 511                         lfloglow = MIN(lowstale, lfloglow);
 512                         lfloghigh = MAX(mid, lfloghigh);
 513                 }
 514                 /*
 515                  * Move entries toward the high-numbered stale entry.
 516                  */
 517                 else {
 518                         ASSERT(highstale < be32_to_cpu(btp->count));
 519                         mid++;
 520                         if (highstale - mid)
 521                                 memmove(&blp[mid + 1], &blp[mid],
 522                                         (highstale - mid) * sizeof(*blp));
 523                         lfloglow = MIN(mid, lfloglow);
 524                         lfloghigh = MAX(highstale, lfloghigh);
 525                 }
 526                 be32_add_cpu(&btp->stale, -1);
 527         }
 528         /*
 529          * Point to the new data entry.
 530          */
 531         dep = (xfs_dir2_data_entry_t *)dup;
 532         /*
 533          * Fill in the leaf entry.
 534          */
 535         blp[mid].hashval = cpu_to_be32(args->hashval);
 536         blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
 537                                 (char *)dep - (char *)hdr));
 538         xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
 539         /*
 540          * Mark space for the data entry used.
 541          */
 542         xfs_dir2_data_use_free(tp, bp, dup,
 543                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
 544                 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
 545         /*
 546          * Create the new data entry.
 547          */
 548         dep->inumber = cpu_to_be64(args->inumber);
 549         dep->namelen = args->namelen;
 550         memcpy(dep->name, args->name, args->namelen);
 551         tagp = xfs_dir2_data_entry_tag_p(dep);
 552         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 553         /*
 554          * Clean up the bestfree array and log the header, tail, and entry.
 555          */
 556         if (needscan)
 557                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 558         if (needlog)
 559                 xfs_dir2_data_log_header(tp, bp);
 560         xfs_dir2_block_log_tail(tp, bp);
 561         xfs_dir2_data_log_entry(tp, bp, dep);
 562         xfs_dir3_data_check(dp, bp);
 563         return 0;
 564 }
 565
 566 /*
 567  * Readdir for block directories.
 568  */
 569 int                                             /* error */
 570 xfs_dir2_block_getdents(
 571         xfs_inode_t             *dp,            /* incore inode */
 572         void                    *dirent,
 573         xfs_off_t               *offset,
 574         filldir_t               filldir)
 575 {
 576         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 577         struct xfs_buf          *bp;            /* buffer for block */
 578         xfs_dir2_block_tail_t   *btp;           /* block tail */
 579         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 580         xfs_dir2_data_unused_t  *dup;           /* block unused entry */
 581         char                    *endptr;        /* end of the data entries */
 582         int                     error;          /* error return value */
 583         xfs_mount_t             *mp;            /* filesystem mount point */
 584         char                    *ptr;           /* current data entry */
 585         int                     wantoff;        /* starting block offset */
 586         xfs_off_t               cook;
 587
 588         mp = dp->i_mount;
 589         /*
 590          * If the block number in the offset is out of range, we're done.
 591          */
 592         if (xfs_dir2_dataptr_to_db(mp, *offset) > mp->m_dirdatablk)
 593                 return 0;
 594
 595         error = xfs_dir3_block_read(NULL, dp, &bp);
 596         if (error)
 597                 return error;
 598
 599         /*
 600          * Extract the byte offset we start at from the seek pointer.
 601          * We'll skip entries before this.
 602          */
 603         wantoff = xfs_dir2_dataptr_to_off(mp, *offset);
 604         hdr = bp->b_addr;
 605         xfs_dir3_data_check(dp, bp);
 606         /*
 607          * Set up values for the loop.
 608          */
 609         btp = xfs_dir2_block_tail_p(mp, hdr);
 610         ptr = (char *)xfs_dir3_data_entry_p(hdr);
 611         endptr = (char *)xfs_dir2_block_leaf_p(btp);
 612
 613         /*
 614          * Loop over the data portion of the block.
 615          * Each object is a real entry (dep) or an unused one (dup).
 616          */
 617         while (ptr < endptr) {
 618                 dup = (xfs_dir2_data_unused_t *)ptr;
 619                 /*
 620                  * Unused, skip it.
 621                  */
 622                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 623                         ptr += be16_to_cpu(dup->length);
 624                         continue;
 625                 }
 626
 627                 dep = (xfs_dir2_data_entry_t *)ptr;
 628
 629                 /*
 630                  * Bump pointer for the next iteration.
 631                  */
 632                 ptr += xfs_dir2_data_entsize(dep->namelen);
 633                 /*
 634                  * The entry is before the desired starting point, skip it.
 635                  */
 636                 if ((char *)dep - (char *)hdr < wantoff)
 637                         continue;
 638
 639                 cook = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
 640                                             (char *)dep - (char *)hdr);
 641
 642                 /*
 643                  * If it didn't fit, set the final offset to here & return.
 644                  */
 645                 if (filldir(dirent, (char *)dep->name, dep->namelen,
 646                             cook & 0x7fffffff, be64_to_cpu(dep->inumber),
 647                             DT_UNKNOWN)) {
 648                         *offset = cook & 0x7fffffff;
 649                         xfs_trans_brelse(NULL, bp);
 650                         return 0;
 651                 }
 652         }
 653
 654         /*
 655          * Reached the end of the block.
 656          * Set the offset to a non-existent block 1 and return.
 657          */
 658         *offset = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk + 1, 0) &
 659                         0x7fffffff;
 660         xfs_trans_brelse(NULL, bp);
 661         return 0;
 662 }
 663
 664 /*
 665  * Log leaf entries from the block.
 666  */
 667 static void
 668 xfs_dir2_block_log_leaf(
 669         xfs_trans_t             *tp,            /* transaction structure */
 670         struct xfs_buf          *bp,            /* block buffer */
 671         int                     first,          /* index of first logged leaf */
 672         int                     last)           /* index of last logged leaf */
 673 {
 674         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 675         xfs_dir2_leaf_entry_t   *blp;
 676         xfs_dir2_block_tail_t   *btp;
 677
 678         btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
 679         blp = xfs_dir2_block_leaf_p(btp);
 680         xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
 681                 (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
 682 }
 683
 684 /*
 685  * Log the block tail.
 686  */
 687 static void
 688 xfs_dir2_block_log_tail(
 689         xfs_trans_t             *tp,            /* transaction structure */
 690         struct xfs_buf          *bp)            /* block buffer */
 691 {
 692         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 693         xfs_dir2_block_tail_t   *btp;
 694
 695         btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
 696         xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
 697                 (uint)((char *)(btp + 1) - (char *)hdr - 1));
 698 }
 699
 700 /*
 701  * Look up an entry in the block.  This is the external routine,
 702  * xfs_dir2_block_lookup_int does the real work.
 703  */
 704 int                                             /* error */
 705 xfs_dir2_block_lookup(
 706         xfs_da_args_t           *args)          /* dir lookup arguments */
 707 {
 708         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 709         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 710         struct xfs_buf          *bp;            /* block buffer */
 711         xfs_dir2_block_tail_t   *btp;           /* block tail */
 712         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 713         xfs_inode_t             *dp;            /* incore inode */
 714         int                     ent;            /* entry index */
 715         int                     error;          /* error return value */
 716         xfs_mount_t             *mp;            /* filesystem mount point */
 717
 718         trace_xfs_dir2_block_lookup(args);
 719
 720         /*
 721          * Get the buffer, look up the entry.
 722          * If not found (ENOENT) then return, have no buffer.
 723          */
 724         if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
 725                 return error;
 726         dp = args->dp;
 727         mp = dp->i_mount;
 728         hdr = bp->b_addr;
 729         xfs_dir3_data_check(dp, bp);
 730         btp = xfs_dir2_block_tail_p(mp, hdr);
 731         blp = xfs_dir2_block_leaf_p(btp);
 732         /*
 733          * Get the offset from the leaf entry, to point to the data.
 734          */
 735         dep = (xfs_dir2_data_entry_t *)((char *)hdr +
 736                 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 737         /*
 738          * Fill in inode number, CI name if appropriate, release the block.
 739          */
 740         args->inumber = be64_to_cpu(dep->inumber);
 741         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
 742         xfs_trans_brelse(args->trans, bp);
 743         return XFS_ERROR(error);
 744 }
 745
 746 /*
 747  * Internal block lookup routine.
 748  */
 749 static int                                      /* error */
 750 xfs_dir2_block_lookup_int(
 751         xfs_da_args_t           *args,          /* dir lookup arguments */
 752         struct xfs_buf          **bpp,          /* returned block buffer */
 753         int                     *entno)         /* returned entry number */
 754 {
 755         xfs_dir2_dataptr_t      addr;           /* data entry address */
 756         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 757         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 758         struct xfs_buf          *bp;            /* block buffer */
 759         xfs_dir2_block_tail_t   *btp;           /* block tail */
 760         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 761         xfs_inode_t             *dp;            /* incore inode */
 762         int                     error;          /* error return value */
 763         xfs_dahash_t            hash;           /* found hash value */
 764         int                     high;           /* binary search high index */
 765         int                     low;            /* binary search low index */
 766         int                     mid;            /* binary search current idx */
 767         xfs_mount_t             *mp;            /* filesystem mount point */
 768         xfs_trans_t             *tp;            /* transaction pointer */
 769         enum xfs_dacmp          cmp;            /* comparison result */
 770
 771         dp = args->dp;
 772         tp = args->trans;
 773         mp = dp->i_mount;
 774
 775         error = xfs_dir3_block_read(tp, dp, &bp);
 776         if (error)
 777                 return error;
 778
 779         hdr = bp->b_addr;
 780         xfs_dir3_data_check(dp, bp);
 781         btp = xfs_dir2_block_tail_p(mp, hdr);
 782         blp = xfs_dir2_block_leaf_p(btp);
 783         /*
 784          * Loop doing a binary search for our hash value.
 785          * Find our entry, ENOENT if it's not there.
 786          */
 787         for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
 788                 ASSERT(low <= high);
 789                 mid = (low + high) >> 1;
 790                 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
 791                         break;
 792                 if (hash < args->hashval)
 793                         low = mid + 1;
 794                 else
 795                         high = mid - 1;
 796                 if (low > high) {
 797                         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 798                         xfs_trans_brelse(tp, bp);
 799                         return XFS_ERROR(ENOENT);
 800                 }
 801         }
 802         /*
 803          * Back up to the first one with the right hash value.
 804          */
 805         while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
 806                 mid--;
 807         }
 808         /*
 809          * Now loop forward through all the entries with the
 810          * right hash value looking for our name.
 811          */
 812         do {
 813                 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
 814                         continue;
 815                 /*
 816                  * Get pointer to the entry from the leaf.
 817                  */
 818                 dep = (xfs_dir2_data_entry_t *)
 819                         ((char *)hdr + xfs_dir2_dataptr_to_off(mp, addr));
 820                 /*
 821                  * Compare name and if it's an exact match, return the index
 822                  * and buffer. If it's the first case-insensitive match, store
 823                  * the index and buffer and continue looking for an exact match.
 824                  */
 825                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 826                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 827                         args->cmpresult = cmp;
 828                         *bpp = bp;
 829                         *entno = mid;
 830                         if (cmp == XFS_CMP_EXACT)
 831                                 return 0;
 832                 }
 833         } while (++mid < be32_to_cpu(btp->count) &&
 834                         be32_to_cpu(blp[mid].hashval) == hash);
 835
 836         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 837         /*
 838          * Here, we can only be doing a lookup (not a rename or replace).
 839          * If a case-insensitive match was found earlier, return success.
 840          */
 841         if (args->cmpresult == XFS_CMP_CASE)
 842                 return 0;
 843         /*
 844          * No match, release the buffer and return ENOENT.
 845          */
 846         xfs_trans_brelse(tp, bp);
 847         return XFS_ERROR(ENOENT);
 848 }
 849
 850 /*
 851  * Remove an entry from a block format directory.
 852  * If that makes the block small enough to fit in shortform, transform it.
 853  */
 854 int                                             /* error */
 855 xfs_dir2_block_removename(
 856         xfs_da_args_t           *args)          /* directory operation args */
 857 {
 858         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 859         xfs_dir2_leaf_entry_t   *blp;           /* block leaf pointer */
 860         struct xfs_buf          *bp;            /* block buffer */
 861         xfs_dir2_block_tail_t   *btp;           /* block tail */
 862         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 863         xfs_inode_t             *dp;            /* incore inode */
 864         int                     ent;            /* block leaf entry index */
 865         int                     error;          /* error return value */
 866         xfs_mount_t             *mp;            /* filesystem mount point */
 867         int                     needlog;        /* need to log block header */
 868         int                     needscan;       /* need to fixup bestfree */
 869         xfs_dir2_sf_hdr_t       sfh;            /* shortform header */
 870         int                     size;           /* shortform size */
 871         xfs_trans_t             *tp;            /* transaction pointer */
 872
 873         trace_xfs_dir2_block_removename(args);
 874
 875         /*
 876          * Look up the entry in the block.  Gets the buffer and entry index.
 877          * It will always be there, the vnodeops level does a lookup first.
 878          */
 879         if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
 880                 return error;
 881         }
 882         dp = args->dp;
 883         tp = args->trans;
 884         mp = dp->i_mount;
 885         hdr = bp->b_addr;
 886         btp = xfs_dir2_block_tail_p(mp, hdr);
 887         blp = xfs_dir2_block_leaf_p(btp);
 888         /*
 889          * Point to the data entry using the leaf entry.
 890          */
 891         dep = (xfs_dir2_data_entry_t *)
 892               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 893         /*
 894          * Mark the data entry's space free.
 895          */
 896         needlog = needscan = 0;
 897         xfs_dir2_data_make_free(tp, bp,
 898                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
 899                 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
 900         /*
 901          * Fix up the block tail.
 902          */
 903         be32_add_cpu(&btp->stale, 1);
 904         xfs_dir2_block_log_tail(tp, bp);
 905         /*
 906          * Remove the leaf entry by marking it stale.
 907          */
 908         blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 909         xfs_dir2_block_log_leaf(tp, bp, ent, ent);
 910         /*
 911          * Fix up bestfree, log the header if necessary.
 912          */
 913         if (needscan)
 914                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 915         if (needlog)
 916                 xfs_dir2_data_log_header(tp, bp);
 917         xfs_dir3_data_check(dp, bp);
 918         /*
 919          * See if the size as a shortform is good enough.
 920          */
 921         size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
 922         if (size > XFS_IFORK_DSIZE(dp))
 923                 return 0;
 924
 925         /*
 926          * If it works, do the conversion.
 927          */
 928         return xfs_dir2_block_to_sf(args, bp, size, &sfh);
 929 }
 930
 931 /*
 932  * Replace an entry in a V2 block directory.
 933  * Change the inode number to the new value.
 934  */
 935 int                                             /* error */
 936 xfs_dir2_block_replace(
 937         xfs_da_args_t           *args)          /* directory operation args */
 938 {
 939         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 940         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 941         struct xfs_buf          *bp;            /* block buffer */
 942         xfs_dir2_block_tail_t   *btp;           /* block tail */
 943         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 944         xfs_inode_t             *dp;            /* incore inode */
 945         int                     ent;            /* leaf entry index */
 946         int                     error;          /* error return value */
 947         xfs_mount_t             *mp;            /* filesystem mount point */
 948
 949         trace_xfs_dir2_block_replace(args);
 950
 951         /*
 952          * Lookup the entry in the directory.  Get buffer and entry index.
 953          * This will always succeed since the caller has already done a lookup.
 954          */
 955         if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
 956                 return error;
 957         }
 958         dp = args->dp;
 959         mp = dp->i_mount;
 960         hdr = bp->b_addr;
 961         btp = xfs_dir2_block_tail_p(mp, hdr);
 962         blp = xfs_dir2_block_leaf_p(btp);
 963         /*
 964          * Point to the data entry we need to change.
 965          */
 966         dep = (xfs_dir2_data_entry_t *)
 967               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 968         ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
 969         /*
 970          * Change the inode number to the new value.
 971          */
 972         dep->inumber = cpu_to_be64(args->inumber);
 973         xfs_dir2_data_log_entry(args->trans, bp, dep);
 974         xfs_dir3_data_check(dp, bp);
 975         return 0;
 976 }
 977
 978 /*
 979  * Qsort comparison routine for the block leaf entries.
 980  */
 981 static int                                      /* sort order */
 982 xfs_dir2_block_sort(
 983         const void                      *a,     /* first leaf entry */
 984         const void                      *b)     /* second leaf entry */
 985 {
 986         const xfs_dir2_leaf_entry_t     *la;    /* first leaf entry */
 987         const xfs_dir2_leaf_entry_t     *lb;    /* second leaf entry */
 988
 989         la = a;
 990         lb = b;
 991         return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
 992                 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
 993 }
 994
 995 /*
 996  * Convert a V2 leaf directory to a V2 block directory if possible.
 997  */
 998 int                                             /* error */
 999 xfs_dir2_leaf_to_block(
1000         xfs_da_args_t           *args,          /* operation arguments */
1001         struct xfs_buf          *lbp,           /* leaf buffer */
1002         struct xfs_buf          *dbp)           /* data buffer */
1003 {
1004         __be16                  *bestsp;        /* leaf bests table */
1005         xfs_dir2_data_hdr_t     *hdr;           /* block header */
1006         xfs_dir2_block_tail_t   *btp;           /* block tail */
1007         xfs_inode_t             *dp;            /* incore directory inode */
1008         xfs_dir2_data_unused_t  *dup;           /* unused data entry */
1009         int                     error;          /* error return value */
1010         int                     from;           /* leaf from index */
1011         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1012         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1013         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1014         xfs_mount_t             *mp;            /* file system mount point */
1015         int                     needlog;        /* need to log data header */
1016         int                     needscan;       /* need to scan for bestfree */
1017         xfs_dir2_sf_hdr_t       sfh;            /* shortform header */
1018         int                     size;           /* bytes used */
1019         __be16                  *tagp;          /* end of entry (tag) */
1020         int                     to;             /* block/leaf to index */
1021         xfs_trans_t             *tp;            /* transaction pointer */
1022         struct xfs_dir2_leaf_entry *ents;
1023         struct xfs_dir3_icleaf_hdr leafhdr;
1024
1025         trace_xfs_dir2_leaf_to_block(args);
1026
1027         dp = args->dp;
1028         tp = args->trans;
1029         mp = dp->i_mount;
1030         leaf = lbp->b_addr;
1031         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1032         ents = xfs_dir3_leaf_ents_p(leaf);
1033         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1034
1035         ASSERT(leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
1036                leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
1037         /*
1038          * If there are data blocks other than the first one, take this
1039          * opportunity to remove trailing empty data blocks that may have
1040          * been left behind during no-space-reservation operations.
1041          * These will show up in the leaf bests table.
1042          */
1043         while (dp->i_d.di_size > mp->m_dirblksize) {
1044                 int hdrsz;
1045
1046                 hdrsz = xfs_dir3_data_hdr_size(xfs_sb_version_hascrc(&mp->m_sb));
1047                 bestsp = xfs_dir2_leaf_bests_p(ltp);
1048                 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
1049                                             mp->m_dirblksize - hdrsz) {
1050                         if ((error =
1051                             xfs_dir2_leaf_trim_data(args, lbp,
1052                                     (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
1053                                 return error;
1054                 } else
1055                         return 0;
1056         }
1057         /*
1058          * Read the data block if we don't already have it, give up if it fails.
1059          */
1060         if (!dbp) {
1061                 error = xfs_dir3_data_read(tp, dp, mp->m_dirdatablk, -1, &dbp);
1062                 if (error)
1063                         return error;
1064         }
1065         hdr = dbp->b_addr;
1066         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1067                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1068
1069         /*
1070          * Size of the "leaf" area in the block.
1071          */
1072         size = (uint)sizeof(xfs_dir2_block_tail_t) +
1073                (uint)sizeof(*lep) * (leafhdr.count - leafhdr.stale);
1074         /*
1075          * Look at the last data entry.
1076          */
1077         tagp = (__be16 *)((char *)hdr + mp->m_dirblksize) - 1;
1078         dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
1079         /*
1080          * If it's not free or is too short we can't do it.
1081          */
1082         if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
1083             be16_to_cpu(dup->length) < size)
1084                 return 0;
1085
1086         /*
1087          * Start converting it to block form.
1088          */
1089         xfs_dir3_block_init(mp, tp, dbp, dp);
1090
1091         needlog = 1;
1092         needscan = 0;
1093         /*
1094          * Use up the space at the end of the block (blp/btp).
1095          */
1096         xfs_dir2_data_use_free(tp, dbp, dup, mp->m_dirblksize - size, size,
1097                 &needlog, &needscan);
1098         /*
1099          * Initialize the block tail.
1100          */
1101         btp = xfs_dir2_block_tail_p(mp, hdr);
1102         btp->count = cpu_to_be32(leafhdr.count - leafhdr.stale);
1103         btp->stale = 0;
1104         xfs_dir2_block_log_tail(tp, dbp);
1105         /*
1106          * Initialize the block leaf area.  We compact out stale entries.
1107          */
1108         lep = xfs_dir2_block_leaf_p(btp);
1109         for (from = to = 0; from < leafhdr.count; from++) {
1110                 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
1111                         continue;
1112                 lep[to++] = ents[from];
1113         }
1114         ASSERT(to == be32_to_cpu(btp->count));
1115         xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
1116         /*
1117          * Scan the bestfree if we need it and log the data block header.
1118          */
1119         if (needscan)
1120                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1121         if (needlog)
1122                 xfs_dir2_data_log_header(tp, dbp);
1123         /*
1124          * Pitch the old leaf block.
1125          */
1126         error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
1127         if (error)
1128                 return error;
1129
1130         /*
1131          * Now see if the resulting block can be shrunken to shortform.
1132          */
1133         size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1134         if (size > XFS_IFORK_DSIZE(dp))
1135                 return 0;
1136
1137         return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1138 }
1139
1140 /*
1141  * Convert the shortform directory to block form.
1142  */
1143 int                                             /* error */
1144 xfs_dir2_sf_to_block(
1145         xfs_da_args_t           *args)          /* operation arguments */
1146 {
1147         xfs_dir2_db_t           blkno;          /* dir-relative block # (0) */
1148         xfs_dir2_data_hdr_t     *hdr;           /* block header */
1149         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
1150         struct xfs_buf          *bp;            /* block buffer */
1151         xfs_dir2_block_tail_t   *btp;           /* block tail pointer */
1152         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1153         xfs_inode_t             *dp;            /* incore directory inode */
1154         int                     dummy;          /* trash */
1155         xfs_dir2_data_unused_t  *dup;           /* unused entry pointer */
1156         int                     endoffset;      /* end of data objects */
1157         int                     error;          /* error return value */
1158         int                     i;              /* index */
1159         xfs_mount_t             *mp;            /* filesystem mount point */
1160         int                     needlog;        /* need to log block header */
1161         int                     needscan;       /* need to scan block freespc */
1162         int                     newoffset;      /* offset from current entry */
1163         int                     offset;         /* target block offset */
1164         xfs_dir2_sf_entry_t     *sfep;          /* sf entry pointer */
1165         xfs_dir2_sf_hdr_t       *oldsfp;        /* old shortform header  */
1166         xfs_dir2_sf_hdr_t       *sfp;           /* shortform header  */
1167         __be16                  *tagp;          /* end of data entry */
1168         xfs_trans_t             *tp;            /* transaction pointer */
1169         struct xfs_name         name;
1170
1171         trace_xfs_dir2_sf_to_block(args);
1172
1173         dp = args->dp;
1174         tp = args->trans;
1175         mp = dp->i_mount;
1176         ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
1177         /*
1178          * Bomb out if the shortform directory is way too short.
1179          */
1180         if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1181                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1182                 return XFS_ERROR(EIO);
1183         }
1184
1185         oldsfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
1186
1187         ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
1188         ASSERT(dp->i_df.if_u1.if_data != NULL);
1189         ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1190
1191         /*
1192          * Copy the directory into a temporary buffer.
1193          * Then pitch the incore inode data so we can make extents.
1194          */
1195         sfp = kmem_alloc(dp->i_df.if_bytes, KM_SLEEP);
1196         memcpy(sfp, oldsfp, dp->i_df.if_bytes);
1197
1198         xfs_idata_realloc(dp, -dp->i_df.if_bytes, XFS_DATA_FORK);
1199         dp->i_d.di_size = 0;
1200         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1201
1202         /*
1203          * Add block 0 to the inode.
1204          */
1205         error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1206         if (error) {
1207                 kmem_free(sfp);
1208                 return error;
1209         }
1210         /*
1211          * Initialize the data block, then convert it to block format.
1212          */
1213         error = xfs_dir3_data_init(args, blkno, &bp);
1214         if (error) {
1215                 kmem_free(sfp);
1216                 return error;
1217         }
1218         xfs_dir3_block_init(mp, tp, bp, dp);
1219         hdr = bp->b_addr;
1220
1221         /*
1222          * Compute size of block "tail" area.
1223          */
1224         i = (uint)sizeof(*btp) +
1225             (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1226         /*
1227          * The whole thing is initialized to free by the init routine.
1228          * Say we're using the leaf and tail area.
1229          */
1230         dup = xfs_dir3_data_unused_p(hdr);
1231         needlog = needscan = 0;
1232         xfs_dir2_data_use_free(tp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1233                 &needscan);
1234         ASSERT(needscan == 0);
1235         /*
1236          * Fill in the tail.
1237          */
1238         btp = xfs_dir2_block_tail_p(mp, hdr);
1239         btp->count = cpu_to_be32(sfp->count + 2);       /* ., .. */
1240         btp->stale = 0;
1241         blp = xfs_dir2_block_leaf_p(btp);
1242         endoffset = (uint)((char *)blp - (char *)hdr);
1243         /*
1244          * Remove the freespace, we'll manage it.
1245          */
1246         xfs_dir2_data_use_free(tp, bp, dup,
1247                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1248                 be16_to_cpu(dup->length), &needlog, &needscan);
1249         /*
1250          * Create entry for .
1251          */
1252         dep = xfs_dir3_data_dot_entry_p(hdr);
1253         dep->inumber = cpu_to_be64(dp->i_ino);
1254         dep->namelen = 1;
1255         dep->name[0] = '.';
1256         tagp = xfs_dir2_data_entry_tag_p(dep);
1257         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1258         xfs_dir2_data_log_entry(tp, bp, dep);
1259         blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1260         blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1261                                 (char *)dep - (char *)hdr));
1262         /*
1263          * Create entry for ..
1264          */
1265         dep = xfs_dir3_data_dotdot_entry_p(hdr);
1266         dep->inumber = cpu_to_be64(xfs_dir2_sf_get_parent_ino(sfp));
1267         dep->namelen = 2;
1268         dep->name[0] = dep->name[1] = '.';
1269         tagp = xfs_dir2_data_entry_tag_p(dep);
1270         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1271         xfs_dir2_data_log_entry(tp, bp, dep);
1272         blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1273         blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1274                                 (char *)dep - (char *)hdr));
1275         offset = xfs_dir3_data_first_offset(hdr);
1276         /*
1277          * Loop over existing entries, stuff them in.
1278          */
1279         i = 0;
1280         if (!sfp->count)
1281                 sfep = NULL;
1282         else
1283                 sfep = xfs_dir2_sf_firstentry(sfp);
1284         /*
1285          * Need to preserve the existing offset values in the sf directory.
1286          * Insert holes (unused entries) where necessary.
1287          */
1288         while (offset < endoffset) {
1289                 /*
1290                  * sfep is null when we reach the end of the list.
1291                  */
1292                 if (sfep == NULL)
1293                         newoffset = endoffset;
1294                 else
1295                         newoffset = xfs_dir2_sf_get_offset(sfep);
1296                 /*
1297                  * There should be a hole here, make one.
1298                  */
1299                 if (offset < newoffset) {
1300                         dup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1301                         dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1302                         dup->length = cpu_to_be16(newoffset - offset);
1303                         *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1304                                 ((char *)dup - (char *)hdr));
1305                         xfs_dir2_data_log_unused(tp, bp, dup);
1306                         xfs_dir2_data_freeinsert(hdr, dup, &dummy);
1307                         offset += be16_to_cpu(dup->length);
1308                         continue;
1309                 }
1310                 /*
1311                  * Copy a real entry.
1312                  */
1313                 dep = (xfs_dir2_data_entry_t *)((char *)hdr + newoffset);
1314                 dep->inumber = cpu_to_be64(xfs_dir2_sfe_get_ino(sfp, sfep));
1315                 dep->namelen = sfep->namelen;
1316                 memcpy(dep->name, sfep->name, dep->namelen);
1317                 tagp = xfs_dir2_data_entry_tag_p(dep);
1318                 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1319                 xfs_dir2_data_log_entry(tp, bp, dep);
1320                 name.name = sfep->name;
1321                 name.len = sfep->namelen;
1322                 blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1323                                                         hashname(&name));
1324                 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1325                                                  (char *)dep - (char *)hdr));
1326                 offset = (int)((char *)(tagp + 1) - (char *)hdr);
1327                 if (++i == sfp->count)
1328                         sfep = NULL;
1329                 else
1330                         sfep = xfs_dir2_sf_nextentry(sfp, sfep);
1331         }
1332         /* Done with the temporary buffer */
1333         kmem_free(sfp);
1334         /*
1335          * Sort the leaf entries by hash value.
1336          */
1337         xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1338         /*
1339          * Log the leaf entry area and tail.
1340          * Already logged the header in data_init, ignore needlog.
1341          */
1342         ASSERT(needscan == 0);
1343         xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1344         xfs_dir2_block_log_tail(tp, bp);
1345         xfs_dir3_data_check(dp, bp);
1346         return 0;
1347 }