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_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_inode.h"
  30 #include "xfs_trans.h"
  31 #include "xfs_inode_item.h"
  32 #include "xfs_bmap.h"
  33 #include "xfs_buf_item.h"
  34 #include "xfs_dir2.h"
  35 #include "xfs_dir2_priv.h"
  36 #include "xfs_error.h"
  37 #include "xfs_trace.h"
  38 #include "xfs_cksum.h"
  39 #include "xfs_dinode.h"
  40
  41 /*
  42  * Local function prototypes.
  43  */
  44 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, struct xfs_buf *bp,
  45                                     int first, int last);
  46 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, struct xfs_buf *bp);
  47 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, struct xfs_buf **bpp,
  48                                      int *entno);
  49 static int xfs_dir2_block_sort(const void *a, const void *b);
  50
  51 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
  52
  53 /*
  54  * One-time startup routine called from xfs_init().
  55  */
  56 void
  57 xfs_dir_startup(void)
  58 {
  59         xfs_dir_hash_dot = xfs_da_hashname((unsigned char *)".", 1);
  60         xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
  61 }
  62
  63 static bool
  64 xfs_dir3_block_verify(
  65         struct xfs_buf          *bp)
  66 {
  67         struct xfs_mount        *mp = bp->b_target->bt_mount;
  68         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
  69
  70         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  71                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_BLOCK_MAGIC))
  72                         return false;
  73                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
  74                         return false;
  75                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
  76                         return false;
  77         } else {
  78                 if (hdr3->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
  79                         return false;
  80         }
  81         if (__xfs_dir3_data_check(NULL, bp))
  82                 return false;
  83         return true;
  84 }
  85
  86 static void
  87 xfs_dir3_block_read_verify(
  88         struct xfs_buf  *bp)
  89 {
  90         struct xfs_mount        *mp = bp->b_target->bt_mount;
  91
  92         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
  93              !xfs_buf_verify_cksum(bp, 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_buf_update_cksum(bp, 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 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_inode                *dp,
 171         struct xfs_dir2_data_hdr        *hdr,
 172         struct xfs_dir2_block_tail      *btp,
 173         struct xfs_dir2_leaf_entry      *blp,
 174         __be16                          **tagpp,
 175         struct xfs_dir2_data_unused     **dupp,
 176         struct xfs_dir2_data_unused     **enddupp,
 177         int                             *compact,
 178         int                             len)
 179 {
 180         struct xfs_dir2_data_free       *bf;
 181         __be16                          *tagp = NULL;
 182         struct xfs_dir2_data_unused     *dup = NULL;
 183         struct xfs_dir2_data_unused     *enddup = NULL;
 184
 185         *compact = 0;
 186         bf = dp->d_ops->data_bestfree_p(hdr);
 187
 188         /*
 189          * If there are stale entries we'll use one for the leaf.
 190          */
 191         if (btp->stale) {
 192                 if (be16_to_cpu(bf[0].length) >= len) {
 193                         /*
 194                          * The biggest entry enough to avoid compaction.
 195                          */
 196                         dup = (xfs_dir2_data_unused_t *)
 197                               ((char *)hdr + be16_to_cpu(bf[0].offset));
 198                         goto out;
 199                 }
 200
 201                 /*
 202                  * Will need to compact to make this work.
 203                  * Tag just before the first leaf entry.
 204                  */
 205                 *compact = 1;
 206                 tagp = (__be16 *)blp - 1;
 207
 208                 /* Data object just before the first leaf entry.  */
 209                 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 210
 211                 /*
 212                  * If it's not free then the data will go where the
 213                  * leaf data starts now, if it works at all.
 214                  */
 215                 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 216                         if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
 217                             (uint)sizeof(*blp) < len)
 218                                 dup = NULL;
 219                 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
 220                         dup = NULL;
 221                 else
 222                         dup = (xfs_dir2_data_unused_t *)blp;
 223                 goto out;
 224         }
 225
 226         /*
 227          * no stale entries, so just use free space.
 228          * Tag just before the first leaf entry.
 229          */
 230         tagp = (__be16 *)blp - 1;
 231
 232         /* Data object just before the first leaf entry.  */
 233         enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 234
 235         /*
 236          * If it's not free then can't do this add without cleaning up:
 237          * the space before the first leaf entry needs to be free so it
 238          * can be expanded to hold the pointer to the new entry.
 239          */
 240         if (be16_to_cpu(enddup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 241                 /*
 242                  * Check out the biggest freespace and see if it's the same one.
 243                  */
 244                 dup = (xfs_dir2_data_unused_t *)
 245                       ((char *)hdr + be16_to_cpu(bf[0].offset));
 246                 if (dup != enddup) {
 247                         /*
 248                          * Not the same free entry, just check its length.
 249                          */
 250                         if (be16_to_cpu(dup->length) < len)
 251                                 dup = NULL;
 252                         goto out;
 253                 }
 254
 255                 /*
 256                  * It is the biggest freespace, can it hold the leaf too?
 257                  */
 258                 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
 259                         /*
 260                          * Yes, use the second-largest entry instead if it works.
 261                          */
 262                         if (be16_to_cpu(bf[1].length) >= len)
 263                                 dup = (xfs_dir2_data_unused_t *)
 264                                       ((char *)hdr + be16_to_cpu(bf[1].offset));
 265                         else
 266                                 dup = NULL;
 267                 }
 268         }
 269 out:
 270         *tagpp = tagp;
 271         *dupp = dup;
 272         *enddupp = enddup;
 273 }
 274
 275 /*
 276  * compact the leaf entries.
 277  * Leave the highest-numbered stale entry stale.
 278  * XXX should be the one closest to mid but mid is not yet computed.
 279  */
 280 static void
 281 xfs_dir2_block_compact(
 282         struct xfs_trans                *tp,
 283         struct xfs_inode                *dp,
 284         struct xfs_buf                  *bp,
 285         struct xfs_dir2_data_hdr        *hdr,
 286         struct xfs_dir2_block_tail      *btp,
 287         struct xfs_dir2_leaf_entry      *blp,
 288         int                             *needlog,
 289         int                             *lfloghigh,
 290         int                             *lfloglow)
 291 {
 292         int                     fromidx;        /* source leaf index */
 293         int                     toidx;          /* target leaf index */
 294         int                     needscan = 0;
 295         int                     highstale;      /* high stale index */
 296
 297         fromidx = toidx = be32_to_cpu(btp->count) - 1;
 298         highstale = *lfloghigh = -1;
 299         for (; fromidx >= 0; fromidx--) {
 300                 if (blp[fromidx].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
 301                         if (highstale == -1)
 302                                 highstale = toidx;
 303                         else {
 304                                 if (*lfloghigh == -1)
 305                                         *lfloghigh = toidx;
 306                                 continue;
 307                         }
 308                 }
 309                 if (fromidx < toidx)
 310                         blp[toidx] = blp[fromidx];
 311                 toidx--;
 312         }
 313         *lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
 314         *lfloghigh -= be32_to_cpu(btp->stale) - 1;
 315         be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
 316         xfs_dir2_data_make_free(tp, dp, bp,
 317                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 318                 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
 319                 needlog, &needscan);
 320         blp += be32_to_cpu(btp->stale) - 1;
 321         btp->stale = cpu_to_be32(1);
 322         /*
 323          * If we now need to rebuild the bestfree map, do so.
 324          * This needs to happen before the next call to use_free.
 325          */
 326         if (needscan)
 327                 xfs_dir2_data_freescan(dp, hdr, needlog);
 328 }
 329
 330 /*
 331  * Add an entry to a block directory.
 332  */
 333 int                                             /* error */
 334 xfs_dir2_block_addname(
 335         xfs_da_args_t           *args)          /* directory op arguments */
 336 {
 337         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 338         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 339         struct xfs_buf          *bp;            /* buffer for block */
 340         xfs_dir2_block_tail_t   *btp;           /* block tail */
 341         int                     compact;        /* need to compact leaf ents */
 342         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 343         xfs_inode_t             *dp;            /* directory inode */
 344         xfs_dir2_data_unused_t  *dup;           /* block unused entry */
 345         int                     error;          /* error return value */
 346         xfs_dir2_data_unused_t  *enddup=NULL;   /* unused at end of data */
 347         xfs_dahash_t            hash;           /* hash value of found entry */
 348         int                     high;           /* high index for binary srch */
 349         int                     highstale;      /* high stale index */
 350         int                     lfloghigh=0;    /* last final leaf to log */
 351         int                     lfloglow=0;     /* first final leaf to log */
 352         int                     len;            /* length of the new entry */
 353         int                     low;            /* low index for binary srch */
 354         int                     lowstale;       /* low stale index */
 355         int                     mid=0;          /* midpoint for binary srch */
 356         xfs_mount_t             *mp;            /* filesystem mount point */
 357         int                     needlog;        /* need to log header */
 358         int                     needscan;       /* need to rescan freespace */
 359         __be16                  *tagp;          /* pointer to tag value */
 360         xfs_trans_t             *tp;            /* transaction structure */
 361
 362         trace_xfs_dir2_block_addname(args);
 363
 364         dp = args->dp;
 365         tp = args->trans;
 366         mp = dp->i_mount;
 367
 368         /* Read the (one and only) directory block into bp. */
 369         error = xfs_dir3_block_read(tp, dp, &bp);
 370         if (error)
 371                 return error;
 372
 373         len = dp->d_ops->data_entsize(args->namelen);
 374
 375         /*
 376          * Set up pointers to parts of the block.
 377          */
 378         hdr = bp->b_addr;
 379         btp = xfs_dir2_block_tail_p(mp, hdr);
 380         blp = xfs_dir2_block_leaf_p(btp);
 381
 382         /*
 383          * Find out if we can reuse stale entries or whether we need extra
 384          * space for entry and new leaf.
 385          */
 386         xfs_dir2_block_need_space(dp, hdr, btp, blp, &tagp, &dup,
 387                                   &enddup, &compact, len);
 388
 389         /*
 390          * Done everything we need for a space check now.
 391          */
 392         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 393                 xfs_trans_brelse(tp, bp);
 394                 if (!dup)
 395                         return XFS_ERROR(ENOSPC);
 396                 return 0;
 397         }
 398
 399         /*
 400          * If we don't have space for the new entry & leaf ...
 401          */
 402         if (!dup) {
 403                 /* Don't have a space reservation: return no-space.  */
 404                 if (args->total == 0)
 405                         return XFS_ERROR(ENOSPC);
 406                 /*
 407                  * Convert to the next larger format.
 408                  * Then add the new entry in that format.
 409                  */
 410                 error = xfs_dir2_block_to_leaf(args, bp);
 411                 if (error)
 412                         return error;
 413                 return xfs_dir2_leaf_addname(args);
 414         }
 415
 416         needlog = needscan = 0;
 417
 418         /*
 419          * If need to compact the leaf entries, do it now.
 420          */
 421         if (compact) {
 422                 xfs_dir2_block_compact(tp, dp, bp, hdr, btp, blp, &needlog,
 423                                       &lfloghigh, &lfloglow);
 424                 /* recalculate blp post-compaction */
 425                 blp = xfs_dir2_block_leaf_p(btp);
 426         } else if (btp->stale) {
 427                 /*
 428                  * Set leaf logging boundaries to impossible state.
 429                  * For the no-stale case they're set explicitly.
 430                  */
 431                 lfloglow = be32_to_cpu(btp->count);
 432                 lfloghigh = -1;
 433         }
 434
 435         /*
 436          * Find the slot that's first lower than our hash value, -1 if none.
 437          */
 438         for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
 439                 mid = (low + high) >> 1;
 440                 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
 441                         break;
 442                 if (hash < args->hashval)
 443                         low = mid + 1;
 444                 else
 445                         high = mid - 1;
 446         }
 447         while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
 448                 mid--;
 449         }
 450         /*
 451          * No stale entries, will use enddup space to hold new leaf.
 452          */
 453         if (!btp->stale) {
 454                 /*
 455                  * Mark the space needed for the new leaf entry, now in use.
 456                  */
 457                 xfs_dir2_data_use_free(tp, dp, bp, enddup,
 458                         (xfs_dir2_data_aoff_t)
 459                         ((char *)enddup - (char *)hdr + be16_to_cpu(enddup->length) -
 460                          sizeof(*blp)),
 461                         (xfs_dir2_data_aoff_t)sizeof(*blp),
 462                         &needlog, &needscan);
 463                 /*
 464                  * Update the tail (entry count).
 465                  */
 466                 be32_add_cpu(&btp->count, 1);
 467                 /*
 468                  * If we now need to rebuild the bestfree map, do so.
 469                  * This needs to happen before the next call to use_free.
 470                  */
 471                 if (needscan) {
 472                         xfs_dir2_data_freescan(dp, hdr, &needlog);
 473                         needscan = 0;
 474                 }
 475                 /*
 476                  * Adjust pointer to the first leaf entry, we're about to move
 477                  * the table up one to open up space for the new leaf entry.
 478                  * Then adjust our index to match.
 479                  */
 480                 blp--;
 481                 mid++;
 482                 if (mid)
 483                         memmove(blp, &blp[1], mid * sizeof(*blp));
 484                 lfloglow = 0;
 485                 lfloghigh = mid;
 486         }
 487         /*
 488          * Use a stale leaf for our new entry.
 489          */
 490         else {
 491                 for (lowstale = mid;
 492                      lowstale >= 0 &&
 493                         blp[lowstale].address !=
 494                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 495                      lowstale--)
 496                         continue;
 497                 for (highstale = mid + 1;
 498                      highstale < be32_to_cpu(btp->count) &&
 499                         blp[highstale].address !=
 500                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
 501                         (lowstale < 0 || mid - lowstale > highstale - mid);
 502                      highstale++)
 503                         continue;
 504                 /*
 505                  * Move entries toward the low-numbered stale entry.
 506                  */
 507                 if (lowstale >= 0 &&
 508                     (highstale == be32_to_cpu(btp->count) ||
 509                      mid - lowstale <= highstale - mid)) {
 510                         if (mid - lowstale)
 511                                 memmove(&blp[lowstale], &blp[lowstale + 1],
 512                                         (mid - lowstale) * sizeof(*blp));
 513                         lfloglow = MIN(lowstale, lfloglow);
 514                         lfloghigh = MAX(mid, lfloghigh);
 515                 }
 516                 /*
 517                  * Move entries toward the high-numbered stale entry.
 518                  */
 519                 else {
 520                         ASSERT(highstale < be32_to_cpu(btp->count));
 521                         mid++;
 522                         if (highstale - mid)
 523                                 memmove(&blp[mid + 1], &blp[mid],
 524                                         (highstale - mid) * sizeof(*blp));
 525                         lfloglow = MIN(mid, lfloglow);
 526                         lfloghigh = MAX(highstale, lfloghigh);
 527                 }
 528                 be32_add_cpu(&btp->stale, -1);
 529         }
 530         /*
 531          * Point to the new data entry.
 532          */
 533         dep = (xfs_dir2_data_entry_t *)dup;
 534         /*
 535          * Fill in the leaf entry.
 536          */
 537         blp[mid].hashval = cpu_to_be32(args->hashval);
 538         blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
 539                                 (char *)dep - (char *)hdr));
 540         xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
 541         /*
 542          * Mark space for the data entry used.
 543          */
 544         xfs_dir2_data_use_free(tp, dp, bp, dup,
 545                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
 546                 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
 547         /*
 548          * Create the new data entry.
 549          */
 550         dep->inumber = cpu_to_be64(args->inumber);
 551         dep->namelen = args->namelen;
 552         memcpy(dep->name, args->name, args->namelen);
 553         dp->d_ops->data_put_ftype(dep, args->filetype);
 554         tagp = dp->d_ops->data_entry_tag_p(dep);
 555         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 556         /*
 557          * Clean up the bestfree array and log the header, tail, and entry.
 558          */
 559         if (needscan)
 560                 xfs_dir2_data_freescan(dp, hdr, &needlog);
 561         if (needlog)
 562                 xfs_dir2_data_log_header(tp, dp, bp);
 563         xfs_dir2_block_log_tail(tp, bp);
 564         xfs_dir2_data_log_entry(tp, dp, bp, dep);
 565         xfs_dir3_data_check(dp, bp);
 566         return 0;
 567 }
 568
 569 /*
 570  * Log leaf entries from the block.
 571  */
 572 static void
 573 xfs_dir2_block_log_leaf(
 574         xfs_trans_t             *tp,            /* transaction structure */
 575         struct xfs_buf          *bp,            /* block buffer */
 576         int                     first,          /* index of first logged leaf */
 577         int                     last)           /* index of last logged leaf */
 578 {
 579         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 580         xfs_dir2_leaf_entry_t   *blp;
 581         xfs_dir2_block_tail_t   *btp;
 582
 583         btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
 584         blp = xfs_dir2_block_leaf_p(btp);
 585         xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
 586                 (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
 587 }
 588
 589 /*
 590  * Log the block tail.
 591  */
 592 static void
 593 xfs_dir2_block_log_tail(
 594         xfs_trans_t             *tp,            /* transaction structure */
 595         struct xfs_buf          *bp)            /* block buffer */
 596 {
 597         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 598         xfs_dir2_block_tail_t   *btp;
 599
 600         btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
 601         xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
 602                 (uint)((char *)(btp + 1) - (char *)hdr - 1));
 603 }
 604
 605 /*
 606  * Look up an entry in the block.  This is the external routine,
 607  * xfs_dir2_block_lookup_int does the real work.
 608  */
 609 int                                             /* error */
 610 xfs_dir2_block_lookup(
 611         xfs_da_args_t           *args)          /* dir lookup arguments */
 612 {
 613         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 614         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 615         struct xfs_buf          *bp;            /* block buffer */
 616         xfs_dir2_block_tail_t   *btp;           /* block tail */
 617         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 618         xfs_inode_t             *dp;            /* incore inode */
 619         int                     ent;            /* entry index */
 620         int                     error;          /* error return value */
 621         xfs_mount_t             *mp;            /* filesystem mount point */
 622
 623         trace_xfs_dir2_block_lookup(args);
 624
 625         /*
 626          * Get the buffer, look up the entry.
 627          * If not found (ENOENT) then return, have no buffer.
 628          */
 629         if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
 630                 return error;
 631         dp = args->dp;
 632         mp = dp->i_mount;
 633         hdr = bp->b_addr;
 634         xfs_dir3_data_check(dp, bp);
 635         btp = xfs_dir2_block_tail_p(mp, hdr);
 636         blp = xfs_dir2_block_leaf_p(btp);
 637         /*
 638          * Get the offset from the leaf entry, to point to the data.
 639          */
 640         dep = (xfs_dir2_data_entry_t *)((char *)hdr +
 641                 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 642         /*
 643          * Fill in inode number, CI name if appropriate, release the block.
 644          */
 645         args->inumber = be64_to_cpu(dep->inumber);
 646         args->filetype = dp->d_ops->data_get_ftype(dep);
 647         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
 648         xfs_trans_brelse(args->trans, bp);
 649         return XFS_ERROR(error);
 650 }
 651
 652 /*
 653  * Internal block lookup routine.
 654  */
 655 static int                                      /* error */
 656 xfs_dir2_block_lookup_int(
 657         xfs_da_args_t           *args,          /* dir lookup arguments */
 658         struct xfs_buf          **bpp,          /* returned block buffer */
 659         int                     *entno)         /* returned entry number */
 660 {
 661         xfs_dir2_dataptr_t      addr;           /* data entry address */
 662         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 663         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 664         struct xfs_buf          *bp;            /* block buffer */
 665         xfs_dir2_block_tail_t   *btp;           /* block tail */
 666         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 667         xfs_inode_t             *dp;            /* incore inode */
 668         int                     error;          /* error return value */
 669         xfs_dahash_t            hash;           /* found hash value */
 670         int                     high;           /* binary search high index */
 671         int                     low;            /* binary search low index */
 672         int                     mid;            /* binary search current idx */
 673         xfs_mount_t             *mp;            /* filesystem mount point */
 674         xfs_trans_t             *tp;            /* transaction pointer */
 675         enum xfs_dacmp          cmp;            /* comparison result */
 676
 677         dp = args->dp;
 678         tp = args->trans;
 679         mp = dp->i_mount;
 680
 681         error = xfs_dir3_block_read(tp, dp, &bp);
 682         if (error)
 683                 return error;
 684
 685         hdr = bp->b_addr;
 686         xfs_dir3_data_check(dp, bp);
 687         btp = xfs_dir2_block_tail_p(mp, hdr);
 688         blp = xfs_dir2_block_leaf_p(btp);
 689         /*
 690          * Loop doing a binary search for our hash value.
 691          * Find our entry, ENOENT if it's not there.
 692          */
 693         for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
 694                 ASSERT(low <= high);
 695                 mid = (low + high) >> 1;
 696                 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
 697                         break;
 698                 if (hash < args->hashval)
 699                         low = mid + 1;
 700                 else
 701                         high = mid - 1;
 702                 if (low > high) {
 703                         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 704                         xfs_trans_brelse(tp, bp);
 705                         return XFS_ERROR(ENOENT);
 706                 }
 707         }
 708         /*
 709          * Back up to the first one with the right hash value.
 710          */
 711         while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
 712                 mid--;
 713         }
 714         /*
 715          * Now loop forward through all the entries with the
 716          * right hash value looking for our name.
 717          */
 718         do {
 719                 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
 720                         continue;
 721                 /*
 722                  * Get pointer to the entry from the leaf.
 723                  */
 724                 dep = (xfs_dir2_data_entry_t *)
 725                         ((char *)hdr + xfs_dir2_dataptr_to_off(mp, addr));
 726                 /*
 727                  * Compare name and if it's an exact match, return the index
 728                  * and buffer. If it's the first case-insensitive match, store
 729                  * the index and buffer and continue looking for an exact match.
 730                  */
 731                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 732                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 733                         args->cmpresult = cmp;
 734                         *bpp = bp;
 735                         *entno = mid;
 736                         if (cmp == XFS_CMP_EXACT)
 737                                 return 0;
 738                 }
 739         } while (++mid < be32_to_cpu(btp->count) &&
 740                         be32_to_cpu(blp[mid].hashval) == hash);
 741
 742         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 743         /*
 744          * Here, we can only be doing a lookup (not a rename or replace).
 745          * If a case-insensitive match was found earlier, return success.
 746          */
 747         if (args->cmpresult == XFS_CMP_CASE)
 748                 return 0;
 749         /*
 750          * No match, release the buffer and return ENOENT.
 751          */
 752         xfs_trans_brelse(tp, bp);
 753         return XFS_ERROR(ENOENT);
 754 }
 755
 756 /*
 757  * Remove an entry from a block format directory.
 758  * If that makes the block small enough to fit in shortform, transform it.
 759  */
 760 int                                             /* error */
 761 xfs_dir2_block_removename(
 762         xfs_da_args_t           *args)          /* directory operation args */
 763 {
 764         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 765         xfs_dir2_leaf_entry_t   *blp;           /* block leaf pointer */
 766         struct xfs_buf          *bp;            /* block buffer */
 767         xfs_dir2_block_tail_t   *btp;           /* block tail */
 768         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 769         xfs_inode_t             *dp;            /* incore inode */
 770         int                     ent;            /* block leaf entry index */
 771         int                     error;          /* error return value */
 772         xfs_mount_t             *mp;            /* filesystem mount point */
 773         int                     needlog;        /* need to log block header */
 774         int                     needscan;       /* need to fixup bestfree */
 775         xfs_dir2_sf_hdr_t       sfh;            /* shortform header */
 776         int                     size;           /* shortform size */
 777         xfs_trans_t             *tp;            /* transaction pointer */
 778
 779         trace_xfs_dir2_block_removename(args);
 780
 781         /*
 782          * Look up the entry in the block.  Gets the buffer and entry index.
 783          * It will always be there, the vnodeops level does a lookup first.
 784          */
 785         if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
 786                 return error;
 787         }
 788         dp = args->dp;
 789         tp = args->trans;
 790         mp = dp->i_mount;
 791         hdr = bp->b_addr;
 792         btp = xfs_dir2_block_tail_p(mp, hdr);
 793         blp = xfs_dir2_block_leaf_p(btp);
 794         /*
 795          * Point to the data entry using the leaf entry.
 796          */
 797         dep = (xfs_dir2_data_entry_t *)
 798               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 799         /*
 800          * Mark the data entry's space free.
 801          */
 802         needlog = needscan = 0;
 803         xfs_dir2_data_make_free(tp, dp, bp,
 804                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
 805                 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
 806         /*
 807          * Fix up the block tail.
 808          */
 809         be32_add_cpu(&btp->stale, 1);
 810         xfs_dir2_block_log_tail(tp, bp);
 811         /*
 812          * Remove the leaf entry by marking it stale.
 813          */
 814         blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 815         xfs_dir2_block_log_leaf(tp, bp, ent, ent);
 816         /*
 817          * Fix up bestfree, log the header if necessary.
 818          */
 819         if (needscan)
 820                 xfs_dir2_data_freescan(dp, hdr, &needlog);
 821         if (needlog)
 822                 xfs_dir2_data_log_header(tp, dp, bp);
 823         xfs_dir3_data_check(dp, bp);
 824         /*
 825          * See if the size as a shortform is good enough.
 826          */
 827         size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
 828         if (size > XFS_IFORK_DSIZE(dp))
 829                 return 0;
 830
 831         /*
 832          * If it works, do the conversion.
 833          */
 834         return xfs_dir2_block_to_sf(args, bp, size, &sfh);
 835 }
 836
 837 /*
 838  * Replace an entry in a V2 block directory.
 839  * Change the inode number to the new value.
 840  */
 841 int                                             /* error */
 842 xfs_dir2_block_replace(
 843         xfs_da_args_t           *args)          /* directory operation args */
 844 {
 845         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 846         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 847         struct xfs_buf          *bp;            /* block buffer */
 848         xfs_dir2_block_tail_t   *btp;           /* block tail */
 849         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 850         xfs_inode_t             *dp;            /* incore inode */
 851         int                     ent;            /* leaf entry index */
 852         int                     error;          /* error return value */
 853         xfs_mount_t             *mp;            /* filesystem mount point */
 854
 855         trace_xfs_dir2_block_replace(args);
 856
 857         /*
 858          * Lookup the entry in the directory.  Get buffer and entry index.
 859          * This will always succeed since the caller has already done a lookup.
 860          */
 861         if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
 862                 return error;
 863         }
 864         dp = args->dp;
 865         mp = dp->i_mount;
 866         hdr = bp->b_addr;
 867         btp = xfs_dir2_block_tail_p(mp, hdr);
 868         blp = xfs_dir2_block_leaf_p(btp);
 869         /*
 870          * Point to the data entry we need to change.
 871          */
 872         dep = (xfs_dir2_data_entry_t *)
 873               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 874         ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
 875         /*
 876          * Change the inode number to the new value.
 877          */
 878         dep->inumber = cpu_to_be64(args->inumber);
 879         dp->d_ops->data_put_ftype(dep, args->filetype);
 880         xfs_dir2_data_log_entry(args->trans, dp, bp, dep);
 881         xfs_dir3_data_check(dp, bp);
 882         return 0;
 883 }
 884
 885 /*
 886  * Qsort comparison routine for the block leaf entries.
 887  */
 888 static int                                      /* sort order */
 889 xfs_dir2_block_sort(
 890         const void                      *a,     /* first leaf entry */
 891         const void                      *b)     /* second leaf entry */
 892 {
 893         const xfs_dir2_leaf_entry_t     *la;    /* first leaf entry */
 894         const xfs_dir2_leaf_entry_t     *lb;    /* second leaf entry */
 895
 896         la = a;
 897         lb = b;
 898         return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
 899                 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
 900 }
 901
 902 /*
 903  * Convert a V2 leaf directory to a V2 block directory if possible.
 904  */
 905 int                                             /* error */
 906 xfs_dir2_leaf_to_block(
 907         xfs_da_args_t           *args,          /* operation arguments */
 908         struct xfs_buf          *lbp,           /* leaf buffer */
 909         struct xfs_buf          *dbp)           /* data buffer */
 910 {
 911         __be16                  *bestsp;        /* leaf bests table */
 912         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 913         xfs_dir2_block_tail_t   *btp;           /* block tail */
 914         xfs_inode_t             *dp;            /* incore directory inode */
 915         xfs_dir2_data_unused_t  *dup;           /* unused data entry */
 916         int                     error;          /* error return value */
 917         int                     from;           /* leaf from index */
 918         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 919         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 920         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 921         xfs_mount_t             *mp;            /* file system mount point */
 922         int                     needlog;        /* need to log data header */
 923         int                     needscan;       /* need to scan for bestfree */
 924         xfs_dir2_sf_hdr_t       sfh;            /* shortform header */
 925         int                     size;           /* bytes used */
 926         __be16                  *tagp;          /* end of entry (tag) */
 927         int                     to;             /* block/leaf to index */
 928         xfs_trans_t             *tp;            /* transaction pointer */
 929         struct xfs_dir2_leaf_entry *ents;
 930         struct xfs_dir3_icleaf_hdr leafhdr;
 931
 932         trace_xfs_dir2_leaf_to_block(args);
 933
 934         dp = args->dp;
 935         tp = args->trans;
 936         mp = dp->i_mount;
 937         leaf = lbp->b_addr;
 938         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 939         ents = dp->d_ops->leaf_ents_p(leaf);
 940         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 941
 942         ASSERT(leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
 943                leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
 944         /*
 945          * If there are data blocks other than the first one, take this
 946          * opportunity to remove trailing empty data blocks that may have
 947          * been left behind during no-space-reservation operations.
 948          * These will show up in the leaf bests table.
 949          */
 950         while (dp->i_d.di_size > mp->m_dirblksize) {
 951                 int hdrsz;
 952
 953                 hdrsz = dp->d_ops->data_entry_offset;
 954                 bestsp = xfs_dir2_leaf_bests_p(ltp);
 955                 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
 956                                             mp->m_dirblksize - hdrsz) {
 957                         if ((error =
 958                             xfs_dir2_leaf_trim_data(args, lbp,
 959                                     (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
 960                                 return error;
 961                 } else
 962                         return 0;
 963         }
 964         /*
 965          * Read the data block if we don't already have it, give up if it fails.
 966          */
 967         if (!dbp) {
 968                 error = xfs_dir3_data_read(tp, dp, mp->m_dirdatablk, -1, &dbp);
 969                 if (error)
 970                         return error;
 971         }
 972         hdr = dbp->b_addr;
 973         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 974                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
 975
 976         /*
 977          * Size of the "leaf" area in the block.
 978          */
 979         size = (uint)sizeof(xfs_dir2_block_tail_t) +
 980                (uint)sizeof(*lep) * (leafhdr.count - leafhdr.stale);
 981         /*
 982          * Look at the last data entry.
 983          */
 984         tagp = (__be16 *)((char *)hdr + mp->m_dirblksize) - 1;
 985         dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 986         /*
 987          * If it's not free or is too short we can't do it.
 988          */
 989         if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
 990             be16_to_cpu(dup->length) < size)
 991                 return 0;
 992
 993         /*
 994          * Start converting it to block form.
 995          */
 996         xfs_dir3_block_init(mp, tp, dbp, dp);
 997
 998         needlog = 1;
 999         needscan = 0;
1000         /*
1001          * Use up the space at the end of the block (blp/btp).
1002          */
1003         xfs_dir2_data_use_free(tp, dp, dbp, dup, mp->m_dirblksize - size, size,
1004                 &needlog, &needscan);
1005         /*
1006          * Initialize the block tail.
1007          */
1008         btp = xfs_dir2_block_tail_p(mp, hdr);
1009         btp->count = cpu_to_be32(leafhdr.count - leafhdr.stale);
1010         btp->stale = 0;
1011         xfs_dir2_block_log_tail(tp, dbp);
1012         /*
1013          * Initialize the block leaf area.  We compact out stale entries.
1014          */
1015         lep = xfs_dir2_block_leaf_p(btp);
1016         for (from = to = 0; from < leafhdr.count; from++) {
1017                 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
1018                         continue;
1019                 lep[to++] = ents[from];
1020         }
1021         ASSERT(to == be32_to_cpu(btp->count));
1022         xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
1023         /*
1024          * Scan the bestfree if we need it and log the data block header.
1025          */
1026         if (needscan)
1027                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1028         if (needlog)
1029                 xfs_dir2_data_log_header(tp, dp, dbp);
1030         /*
1031          * Pitch the old leaf block.
1032          */
1033         error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
1034         if (error)
1035                 return error;
1036
1037         /*
1038          * Now see if the resulting block can be shrunken to shortform.
1039          */
1040         size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1041         if (size > XFS_IFORK_DSIZE(dp))
1042                 return 0;
1043
1044         return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1045 }
1046
1047 /*
1048  * Convert the shortform directory to block form.
1049  */
1050 int                                             /* error */
1051 xfs_dir2_sf_to_block(
1052         xfs_da_args_t           *args)          /* operation arguments */
1053 {
1054         xfs_dir2_db_t           blkno;          /* dir-relative block # (0) */
1055         xfs_dir2_data_hdr_t     *hdr;           /* block header */
1056         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
1057         struct xfs_buf          *bp;            /* block buffer */
1058         xfs_dir2_block_tail_t   *btp;           /* block tail pointer */
1059         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1060         xfs_inode_t             *dp;            /* incore directory inode */
1061         int                     dummy;          /* trash */
1062         xfs_dir2_data_unused_t  *dup;           /* unused entry pointer */
1063         int                     endoffset;      /* end of data objects */
1064         int                     error;          /* error return value */
1065         int                     i;              /* index */
1066         xfs_mount_t             *mp;            /* filesystem mount point */
1067         int                     needlog;        /* need to log block header */
1068         int                     needscan;       /* need to scan block freespc */
1069         int                     newoffset;      /* offset from current entry */
1070         int                     offset;         /* target block offset */
1071         xfs_dir2_sf_entry_t     *sfep;          /* sf entry pointer */
1072         xfs_dir2_sf_hdr_t       *oldsfp;        /* old shortform header  */
1073         xfs_dir2_sf_hdr_t       *sfp;           /* shortform header  */
1074         __be16                  *tagp;          /* end of data entry */
1075         xfs_trans_t             *tp;            /* transaction pointer */
1076         struct xfs_name         name;
1077         struct xfs_ifork        *ifp;
1078
1079         trace_xfs_dir2_sf_to_block(args);
1080
1081         dp = args->dp;
1082         tp = args->trans;
1083         mp = dp->i_mount;
1084         ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
1085         ASSERT(ifp->if_flags & XFS_IFINLINE);
1086         /*
1087          * Bomb out if the shortform directory is way too short.
1088          */
1089         if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1090                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1091                 return XFS_ERROR(EIO);
1092         }
1093
1094         oldsfp = (xfs_dir2_sf_hdr_t *)ifp->if_u1.if_data;
1095
1096         ASSERT(ifp->if_bytes == dp->i_d.di_size);
1097         ASSERT(ifp->if_u1.if_data != NULL);
1098         ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1099         ASSERT(dp->i_d.di_nextents == 0);
1100
1101         /*
1102          * Copy the directory into a temporary buffer.
1103          * Then pitch the incore inode data so we can make extents.
1104          */
1105         sfp = kmem_alloc(ifp->if_bytes, KM_SLEEP);
1106         memcpy(sfp, oldsfp, ifp->if_bytes);
1107
1108         xfs_idata_realloc(dp, -ifp->if_bytes, XFS_DATA_FORK);
1109         xfs_bmap_local_to_extents_empty(dp, XFS_DATA_FORK);
1110         dp->i_d.di_size = 0;
1111
1112         /*
1113          * Add block 0 to the inode.
1114          */
1115         error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1116         if (error) {
1117                 kmem_free(sfp);
1118                 return error;
1119         }
1120         /*
1121          * Initialize the data block, then convert it to block format.
1122          */
1123         error = xfs_dir3_data_init(args, blkno, &bp);
1124         if (error) {
1125                 kmem_free(sfp);
1126                 return error;
1127         }
1128         xfs_dir3_block_init(mp, tp, bp, dp);
1129         hdr = bp->b_addr;
1130
1131         /*
1132          * Compute size of block "tail" area.
1133          */
1134         i = (uint)sizeof(*btp) +
1135             (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1136         /*
1137          * The whole thing is initialized to free by the init routine.
1138          * Say we're using the leaf and tail area.
1139          */
1140         dup = dp->d_ops->data_unused_p(hdr);
1141         needlog = needscan = 0;
1142         xfs_dir2_data_use_free(tp, dp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1143                 &needscan);
1144         ASSERT(needscan == 0);
1145         /*
1146          * Fill in the tail.
1147          */
1148         btp = xfs_dir2_block_tail_p(mp, hdr);
1149         btp->count = cpu_to_be32(sfp->count + 2);       /* ., .. */
1150         btp->stale = 0;
1151         blp = xfs_dir2_block_leaf_p(btp);
1152         endoffset = (uint)((char *)blp - (char *)hdr);
1153         /*
1154          * Remove the freespace, we'll manage it.
1155          */
1156         xfs_dir2_data_use_free(tp, dp, bp, dup,
1157                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1158                 be16_to_cpu(dup->length), &needlog, &needscan);
1159         /*
1160          * Create entry for .
1161          */
1162         dep = dp->d_ops->data_dot_entry_p(hdr);
1163         dep->inumber = cpu_to_be64(dp->i_ino);
1164         dep->namelen = 1;
1165         dep->name[0] = '.';
1166         dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1167         tagp = dp->d_ops->data_entry_tag_p(dep);
1168         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1169         xfs_dir2_data_log_entry(tp, dp, bp, dep);
1170         blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1171         blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1172                                 (char *)dep - (char *)hdr));
1173         /*
1174          * Create entry for ..
1175          */
1176         dep = dp->d_ops->data_dotdot_entry_p(hdr);
1177         dep->inumber = cpu_to_be64(dp->d_ops->sf_get_parent_ino(sfp));
1178         dep->namelen = 2;
1179         dep->name[0] = dep->name[1] = '.';
1180         dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1181         tagp = dp->d_ops->data_entry_tag_p(dep);
1182         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1183         xfs_dir2_data_log_entry(tp, dp, bp, dep);
1184         blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1185         blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1186                                 (char *)dep - (char *)hdr));
1187         offset = dp->d_ops->data_first_offset;
1188         /*
1189          * Loop over existing entries, stuff them in.
1190          */
1191         i = 0;
1192         if (!sfp->count)
1193                 sfep = NULL;
1194         else
1195                 sfep = xfs_dir2_sf_firstentry(sfp);
1196         /*
1197          * Need to preserve the existing offset values in the sf directory.
1198          * Insert holes (unused entries) where necessary.
1199          */
1200         while (offset < endoffset) {
1201                 /*
1202                  * sfep is null when we reach the end of the list.
1203                  */
1204                 if (sfep == NULL)
1205                         newoffset = endoffset;
1206                 else
1207                         newoffset = xfs_dir2_sf_get_offset(sfep);
1208                 /*
1209                  * There should be a hole here, make one.
1210                  */
1211                 if (offset < newoffset) {
1212                         dup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1213                         dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1214                         dup->length = cpu_to_be16(newoffset - offset);
1215                         *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1216                                 ((char *)dup - (char *)hdr));
1217                         xfs_dir2_data_log_unused(tp, bp, dup);
1218                         xfs_dir2_data_freeinsert(hdr,
1219                                                  dp->d_ops->data_bestfree_p(hdr),
1220                                                  dup, &dummy);
1221                         offset += be16_to_cpu(dup->length);
1222                         continue;
1223                 }
1224                 /*
1225                  * Copy a real entry.
1226                  */
1227                 dep = (xfs_dir2_data_entry_t *)((char *)hdr + newoffset);
1228                 dep->inumber = cpu_to_be64(dp->d_ops->sf_get_ino(sfp, sfep));
1229                 dep->namelen = sfep->namelen;
1230                 dp->d_ops->data_put_ftype(dep, dp->d_ops->sf_get_ftype(sfep));
1231                 memcpy(dep->name, sfep->name, dep->namelen);
1232                 tagp = dp->d_ops->data_entry_tag_p(dep);
1233                 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1234                 xfs_dir2_data_log_entry(tp, dp, bp, dep);
1235                 name.name = sfep->name;
1236                 name.len = sfep->namelen;
1237                 blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1238                                                         hashname(&name));
1239                 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1240                                                  (char *)dep - (char *)hdr));
1241                 offset = (int)((char *)(tagp + 1) - (char *)hdr);
1242                 if (++i == sfp->count)
1243                         sfep = NULL;
1244                 else
1245                         sfep = dp->d_ops->sf_nextentry(sfp, sfep);
1246         }
1247         /* Done with the temporary buffer */
1248         kmem_free(sfp);
1249         /*
1250          * Sort the leaf entries by hash value.
1251          */
1252         xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1253         /*
1254          * Log the leaf entry area and tail.
1255          * Already logged the header in data_init, ignore needlog.
1256          */
1257         ASSERT(needscan == 0);
1258         xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1259         xfs_dir2_block_log_tail(tp, bp);
1260         xfs_dir3_data_check(dp, bp);
1261         return 0;
1262 }