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_bmap.h"
  33 #include "xfs_buf_item.h"
  34 #include "xfs_dir2_format.h"
  35 #include "xfs_dir2.h"
  36 #include "xfs_dir2_priv.h"
  37 #include "xfs_error.h"
  38 #include "xfs_trace.h"
  39 #include "xfs_cksum.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_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
  94                                           XFS_DIR3_DATA_CRC_OFF)) ||
  95             !xfs_dir3_block_verify(bp)) {
  96                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
  97                 xfs_buf_ioerror(bp, EFSCORRUPTED);
  98         }
  99 }
 100
 101 static void
 102 xfs_dir3_block_write_verify(
 103         struct xfs_buf  *bp)
 104 {
 105         struct xfs_mount        *mp = bp->b_target->bt_mount;
 106         struct xfs_buf_log_item *bip = bp->b_fspriv;
 107         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 108
 109         if (!xfs_dir3_block_verify(bp)) {
 110                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 111                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 112                 return;
 113         }
 114
 115         if (!xfs_sb_version_hascrc(&mp->m_sb))
 116                 return;
 117
 118         if (bip)
 119                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 120
 121         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_DATA_CRC_OFF);
 122 }
 123
 124 const struct xfs_buf_ops xfs_dir3_block_buf_ops = {
 125         .verify_read = xfs_dir3_block_read_verify,
 126         .verify_write = xfs_dir3_block_write_verify,
 127 };
 128
 129 int
 130 xfs_dir3_block_read(
 131         struct xfs_trans        *tp,
 132         struct xfs_inode        *dp,
 133         struct xfs_buf          **bpp)
 134 {
 135         struct xfs_mount        *mp = dp->i_mount;
 136         int                     err;
 137
 138         err = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, bpp,
 139                                 XFS_DATA_FORK, &xfs_dir3_block_buf_ops);
 140         if (!err && tp)
 141                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_BLOCK_BUF);
 142         return err;
 143 }
 144
 145 static void
 146 xfs_dir3_block_init(
 147         struct xfs_mount        *mp,
 148         struct xfs_trans        *tp,
 149         struct xfs_buf          *bp,
 150         struct xfs_inode        *dp)
 151 {
 152         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 153
 154         bp->b_ops = &xfs_dir3_block_buf_ops;
 155         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_BLOCK_BUF);
 156
 157         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 158                 memset(hdr3, 0, sizeof(*hdr3));
 159                 hdr3->magic = cpu_to_be32(XFS_DIR3_BLOCK_MAGIC);
 160                 hdr3->blkno = cpu_to_be64(bp->b_bn);
 161                 hdr3->owner = cpu_to_be64(dp->i_ino);
 162                 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
 163                 return;
 164
 165         }
 166         hdr3->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
 167 }
 168
 169 static void
 170 xfs_dir2_block_need_space(
 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 = xfs_dir3_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_buf                  *bp,
 284         struct xfs_dir2_data_hdr        *hdr,
 285         struct xfs_dir2_block_tail      *btp,
 286         struct xfs_dir2_leaf_entry      *blp,
 287         int                             *needlog,
 288         int                             *lfloghigh,
 289         int                             *lfloglow)
 290 {
 291         int                     fromidx;        /* source leaf index */
 292         int                     toidx;          /* target leaf index */
 293         int                     needscan = 0;
 294         int                     highstale;      /* high stale index */
 295
 296         fromidx = toidx = be32_to_cpu(btp->count) - 1;
 297         highstale = *lfloghigh = -1;
 298         for (; fromidx >= 0; fromidx--) {
 299                 if (blp[fromidx].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
 300                         if (highstale == -1)
 301                                 highstale = toidx;
 302                         else {
 303                                 if (*lfloghigh == -1)
 304                                         *lfloghigh = toidx;
 305                                 continue;
 306                         }
 307                 }
 308                 if (fromidx < toidx)
 309                         blp[toidx] = blp[fromidx];
 310                 toidx--;
 311         }
 312         *lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
 313         *lfloghigh -= be32_to_cpu(btp->stale) - 1;
 314         be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
 315         xfs_dir2_data_make_free(tp, bp,
 316                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 317                 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
 318                 needlog, &needscan);
 319         blp += be32_to_cpu(btp->stale) - 1;
 320         btp->stale = cpu_to_be32(1);
 321         /*
 322          * If we now need to rebuild the bestfree map, do so.
 323          * This needs to happen before the next call to use_free.
 324          */
 325         if (needscan)
 326                 xfs_dir2_data_freescan(tp->t_mountp, hdr, needlog);
 327 }
 328
 329 /*
 330  * Add an entry to a block directory.
 331  */
 332 int                                             /* error */
 333 xfs_dir2_block_addname(
 334         xfs_da_args_t           *args)          /* directory op arguments */
 335 {
 336         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 337         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 338         struct xfs_buf          *bp;            /* buffer for block */
 339         xfs_dir2_block_tail_t   *btp;           /* block tail */
 340         int                     compact;        /* need to compact leaf ents */
 341         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 342         xfs_inode_t             *dp;            /* directory inode */
 343         xfs_dir2_data_unused_t  *dup;           /* block unused entry */
 344         int                     error;          /* error return value */
 345         xfs_dir2_data_unused_t  *enddup=NULL;   /* unused at end of data */
 346         xfs_dahash_t            hash;           /* hash value of found entry */
 347         int                     high;           /* high index for binary srch */
 348         int                     highstale;      /* high stale index */
 349         int                     lfloghigh=0;    /* last final leaf to log */
 350         int                     lfloglow=0;     /* first final leaf to log */
 351         int                     len;            /* length of the new entry */
 352         int                     low;            /* low index for binary srch */
 353         int                     lowstale;       /* low stale index */
 354         int                     mid=0;          /* midpoint for binary srch */
 355         xfs_mount_t             *mp;            /* filesystem mount point */
 356         int                     needlog;        /* need to log header */
 357         int                     needscan;       /* need to rescan freespace */
 358         __be16                  *tagp;          /* pointer to tag value */
 359         xfs_trans_t             *tp;            /* transaction structure */
 360
 361         trace_xfs_dir2_block_addname(args);
 362
 363         dp = args->dp;
 364         tp = args->trans;
 365         mp = dp->i_mount;
 366
 367         /* Read the (one and only) directory block into bp. */
 368         error = xfs_dir3_block_read(tp, dp, &bp);
 369         if (error)
 370                 return error;
 371
 372         len = xfs_dir3_data_entsize(mp, args->namelen);
 373
 374         /*
 375          * Set up pointers to parts of the block.
 376          */
 377         hdr = bp->b_addr;
 378         btp = xfs_dir2_block_tail_p(mp, hdr);
 379         blp = xfs_dir2_block_leaf_p(btp);
 380
 381         /*
 382          * Find out if we can reuse stale entries or whether we need extra
 383          * space for entry and new leaf.
 384          */
 385         xfs_dir2_block_need_space(hdr, btp, blp, &tagp, &dup,
 386                                   &enddup, &compact, len);
 387
 388         /*
 389          * Done everything we need for a space check now.
 390          */
 391         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 392                 xfs_trans_brelse(tp, bp);
 393                 if (!dup)
 394                         return XFS_ERROR(ENOSPC);
 395                 return 0;
 396         }
 397
 398         /*
 399          * If we don't have space for the new entry & leaf ...
 400          */
 401         if (!dup) {
 402                 /* Don't have a space reservation: return no-space.  */
 403                 if (args->total == 0)
 404                         return XFS_ERROR(ENOSPC);
 405                 /*
 406                  * Convert to the next larger format.
 407                  * Then add the new entry in that format.
 408                  */
 409                 error = xfs_dir2_block_to_leaf(args, bp);
 410                 if (error)
 411                         return error;
 412                 return xfs_dir2_leaf_addname(args);
 413         }
 414
 415         needlog = needscan = 0;
 416
 417         /*
 418          * If need to compact the leaf entries, do it now.
 419          */
 420         if (compact) {
 421                 xfs_dir2_block_compact(tp, bp, hdr, btp, blp, &needlog,
 422                                       &lfloghigh, &lfloglow);
 423                 /* recalculate blp post-compaction */
 424                 blp = xfs_dir2_block_leaf_p(btp);
 425         } else if (btp->stale) {
 426                 /*
 427                  * Set leaf logging boundaries to impossible state.
 428                  * For the no-stale case they're set explicitly.
 429                  */
 430                 lfloglow = be32_to_cpu(btp->count);
 431                 lfloghigh = -1;
 432         }
 433
 434         /*
 435          * Find the slot that's first lower than our hash value, -1 if none.
 436          */
 437         for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
 438                 mid = (low + high) >> 1;
 439                 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
 440                         break;
 441                 if (hash < args->hashval)
 442                         low = mid + 1;
 443                 else
 444                         high = mid - 1;
 445         }
 446         while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
 447                 mid--;
 448         }
 449         /*
 450          * No stale entries, will use enddup space to hold new leaf.
 451          */
 452         if (!btp->stale) {
 453                 /*
 454                  * Mark the space needed for the new leaf entry, now in use.
 455                  */
 456                 xfs_dir2_data_use_free(tp, bp, enddup,
 457                         (xfs_dir2_data_aoff_t)
 458                         ((char *)enddup - (char *)hdr + be16_to_cpu(enddup->length) -
 459                          sizeof(*blp)),
 460                         (xfs_dir2_data_aoff_t)sizeof(*blp),
 461                         &needlog, &needscan);
 462                 /*
 463                  * Update the tail (entry count).
 464                  */
 465                 be32_add_cpu(&btp->count, 1);
 466                 /*
 467                  * If we now need to rebuild the bestfree map, do so.
 468                  * This needs to happen before the next call to use_free.
 469                  */
 470                 if (needscan) {
 471                         xfs_dir2_data_freescan(mp, hdr, &needlog);
 472                         needscan = 0;
 473                 }
 474                 /*
 475                  * Adjust pointer to the first leaf entry, we're about to move
 476                  * the table up one to open up space for the new leaf entry.
 477                  * Then adjust our index to match.
 478                  */
 479                 blp--;
 480                 mid++;
 481                 if (mid)
 482                         memmove(blp, &blp[1], mid * sizeof(*blp));
 483                 lfloglow = 0;
 484                 lfloghigh = mid;
 485         }
 486         /*
 487          * Use a stale leaf for our new entry.
 488          */
 489         else {
 490                 for (lowstale = mid;
 491                      lowstale >= 0 &&
 492                         blp[lowstale].address !=
 493                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 494                      lowstale--)
 495                         continue;
 496                 for (highstale = mid + 1;
 497                      highstale < be32_to_cpu(btp->count) &&
 498                         blp[highstale].address !=
 499                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
 500                         (lowstale < 0 || mid - lowstale > highstale - mid);
 501                      highstale++)
 502                         continue;
 503                 /*
 504                  * Move entries toward the low-numbered stale entry.
 505                  */
 506                 if (lowstale >= 0 &&
 507                     (highstale == be32_to_cpu(btp->count) ||
 508                      mid - lowstale <= highstale - mid)) {
 509                         if (mid - lowstale)
 510                                 memmove(&blp[lowstale], &blp[lowstale + 1],
 511                                         (mid - lowstale) * sizeof(*blp));
 512                         lfloglow = MIN(lowstale, lfloglow);
 513                         lfloghigh = MAX(mid, lfloghigh);
 514                 }
 515                 /*
 516                  * Move entries toward the high-numbered stale entry.
 517                  */
 518                 else {
 519                         ASSERT(highstale < be32_to_cpu(btp->count));
 520                         mid++;
 521                         if (highstale - mid)
 522                                 memmove(&blp[mid + 1], &blp[mid],
 523                                         (highstale - mid) * sizeof(*blp));
 524                         lfloglow = MIN(mid, lfloglow);
 525                         lfloghigh = MAX(highstale, lfloghigh);
 526                 }
 527                 be32_add_cpu(&btp->stale, -1);
 528         }
 529         /*
 530          * Point to the new data entry.
 531          */
 532         dep = (xfs_dir2_data_entry_t *)dup;
 533         /*
 534          * Fill in the leaf entry.
 535          */
 536         blp[mid].hashval = cpu_to_be32(args->hashval);
 537         blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
 538                                 (char *)dep - (char *)hdr));
 539         xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
 540         /*
 541          * Mark space for the data entry used.
 542          */
 543         xfs_dir2_data_use_free(tp, bp, dup,
 544                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
 545                 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
 546         /*
 547          * Create the new data entry.
 548          */
 549         dep->inumber = cpu_to_be64(args->inumber);
 550         dep->namelen = args->namelen;
 551         memcpy(dep->name, args->name, args->namelen);
 552         xfs_dir3_dirent_put_ftype(mp, dep, args->filetype);
 553         tagp = xfs_dir3_data_entry_tag_p(mp, dep);
 554         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 555         /*
 556          * Clean up the bestfree array and log the header, tail, and entry.
 557          */
 558         if (needscan)
 559                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 560         if (needlog)
 561                 xfs_dir2_data_log_header(tp, bp);
 562         xfs_dir2_block_log_tail(tp, bp);
 563         xfs_dir2_data_log_entry(tp, bp, dep);
 564         xfs_dir3_data_check(dp, bp);
 565         return 0;
 566 }
 567
 568 /*
 569  * Log leaf entries from the block.
 570  */
 571 static void
 572 xfs_dir2_block_log_leaf(
 573         xfs_trans_t             *tp,            /* transaction structure */
 574         struct xfs_buf          *bp,            /* block buffer */
 575         int                     first,          /* index of first logged leaf */
 576         int                     last)           /* index of last logged leaf */
 577 {
 578         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 579         xfs_dir2_leaf_entry_t   *blp;
 580         xfs_dir2_block_tail_t   *btp;
 581
 582         btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
 583         blp = xfs_dir2_block_leaf_p(btp);
 584         xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
 585                 (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
 586 }
 587
 588 /*
 589  * Log the block tail.
 590  */
 591 static void
 592 xfs_dir2_block_log_tail(
 593         xfs_trans_t             *tp,            /* transaction structure */
 594         struct xfs_buf          *bp)            /* block buffer */
 595 {
 596         xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 597         xfs_dir2_block_tail_t   *btp;
 598
 599         btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
 600         xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
 601                 (uint)((char *)(btp + 1) - (char *)hdr - 1));
 602 }
 603
 604 /*
 605  * Look up an entry in the block.  This is the external routine,
 606  * xfs_dir2_block_lookup_int does the real work.
 607  */
 608 int                                             /* error */
 609 xfs_dir2_block_lookup(
 610         xfs_da_args_t           *args)          /* dir lookup arguments */
 611 {
 612         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 613         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 614         struct xfs_buf          *bp;            /* block buffer */
 615         xfs_dir2_block_tail_t   *btp;           /* block tail */
 616         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 617         xfs_inode_t             *dp;            /* incore inode */
 618         int                     ent;            /* entry index */
 619         int                     error;          /* error return value */
 620         xfs_mount_t             *mp;            /* filesystem mount point */
 621
 622         trace_xfs_dir2_block_lookup(args);
 623
 624         /*
 625          * Get the buffer, look up the entry.
 626          * If not found (ENOENT) then return, have no buffer.
 627          */
 628         if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
 629                 return error;
 630         dp = args->dp;
 631         mp = dp->i_mount;
 632         hdr = bp->b_addr;
 633         xfs_dir3_data_check(dp, bp);
 634         btp = xfs_dir2_block_tail_p(mp, hdr);
 635         blp = xfs_dir2_block_leaf_p(btp);
 636         /*
 637          * Get the offset from the leaf entry, to point to the data.
 638          */
 639         dep = (xfs_dir2_data_entry_t *)((char *)hdr +
 640                 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 641         /*
 642          * Fill in inode number, CI name if appropriate, release the block.
 643          */
 644         args->inumber = be64_to_cpu(dep->inumber);
 645         args->filetype = xfs_dir3_dirent_get_ftype(mp, dep);
 646         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
 647         xfs_trans_brelse(args->trans, bp);
 648         return XFS_ERROR(error);
 649 }
 650
 651 /*
 652  * Internal block lookup routine.
 653  */
 654 static int                                      /* error */
 655 xfs_dir2_block_lookup_int(
 656         xfs_da_args_t           *args,          /* dir lookup arguments */
 657         struct xfs_buf          **bpp,          /* returned block buffer */
 658         int                     *entno)         /* returned entry number */
 659 {
 660         xfs_dir2_dataptr_t      addr;           /* data entry address */
 661         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 662         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 663         struct xfs_buf          *bp;            /* block buffer */
 664         xfs_dir2_block_tail_t   *btp;           /* block tail */
 665         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 666         xfs_inode_t             *dp;            /* incore inode */
 667         int                     error;          /* error return value */
 668         xfs_dahash_t            hash;           /* found hash value */
 669         int                     high;           /* binary search high index */
 670         int                     low;            /* binary search low index */
 671         int                     mid;            /* binary search current idx */
 672         xfs_mount_t             *mp;            /* filesystem mount point */
 673         xfs_trans_t             *tp;            /* transaction pointer */
 674         enum xfs_dacmp          cmp;            /* comparison result */
 675
 676         dp = args->dp;
 677         tp = args->trans;
 678         mp = dp->i_mount;
 679
 680         error = xfs_dir3_block_read(tp, dp, &bp);
 681         if (error)
 682                 return error;
 683
 684         hdr = bp->b_addr;
 685         xfs_dir3_data_check(dp, bp);
 686         btp = xfs_dir2_block_tail_p(mp, hdr);
 687         blp = xfs_dir2_block_leaf_p(btp);
 688         /*
 689          * Loop doing a binary search for our hash value.
 690          * Find our entry, ENOENT if it's not there.
 691          */
 692         for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
 693                 ASSERT(low <= high);
 694                 mid = (low + high) >> 1;
 695                 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
 696                         break;
 697                 if (hash < args->hashval)
 698                         low = mid + 1;
 699                 else
 700                         high = mid - 1;
 701                 if (low > high) {
 702                         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 703                         xfs_trans_brelse(tp, bp);
 704                         return XFS_ERROR(ENOENT);
 705                 }
 706         }
 707         /*
 708          * Back up to the first one with the right hash value.
 709          */
 710         while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
 711                 mid--;
 712         }
 713         /*
 714          * Now loop forward through all the entries with the
 715          * right hash value looking for our name.
 716          */
 717         do {
 718                 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
 719                         continue;
 720                 /*
 721                  * Get pointer to the entry from the leaf.
 722                  */
 723                 dep = (xfs_dir2_data_entry_t *)
 724                         ((char *)hdr + xfs_dir2_dataptr_to_off(mp, addr));
 725                 /*
 726                  * Compare name and if it's an exact match, return the index
 727                  * and buffer. If it's the first case-insensitive match, store
 728                  * the index and buffer and continue looking for an exact match.
 729                  */
 730                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 731                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 732                         args->cmpresult = cmp;
 733                         *bpp = bp;
 734                         *entno = mid;
 735                         if (cmp == XFS_CMP_EXACT)
 736                                 return 0;
 737                 }
 738         } while (++mid < be32_to_cpu(btp->count) &&
 739                         be32_to_cpu(blp[mid].hashval) == hash);
 740
 741         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 742         /*
 743          * Here, we can only be doing a lookup (not a rename or replace).
 744          * If a case-insensitive match was found earlier, return success.
 745          */
 746         if (args->cmpresult == XFS_CMP_CASE)
 747                 return 0;
 748         /*
 749          * No match, release the buffer and return ENOENT.
 750          */
 751         xfs_trans_brelse(tp, bp);
 752         return XFS_ERROR(ENOENT);
 753 }
 754
 755 /*
 756  * Remove an entry from a block format directory.
 757  * If that makes the block small enough to fit in shortform, transform it.
 758  */
 759 int                                             /* error */
 760 xfs_dir2_block_removename(
 761         xfs_da_args_t           *args)          /* directory operation args */
 762 {
 763         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 764         xfs_dir2_leaf_entry_t   *blp;           /* block leaf pointer */
 765         struct xfs_buf          *bp;            /* block buffer */
 766         xfs_dir2_block_tail_t   *btp;           /* block tail */
 767         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 768         xfs_inode_t             *dp;            /* incore inode */
 769         int                     ent;            /* block leaf entry index */
 770         int                     error;          /* error return value */
 771         xfs_mount_t             *mp;            /* filesystem mount point */
 772         int                     needlog;        /* need to log block header */
 773         int                     needscan;       /* need to fixup bestfree */
 774         xfs_dir2_sf_hdr_t       sfh;            /* shortform header */
 775         int                     size;           /* shortform size */
 776         xfs_trans_t             *tp;            /* transaction pointer */
 777
 778         trace_xfs_dir2_block_removename(args);
 779
 780         /*
 781          * Look up the entry in the block.  Gets the buffer and entry index.
 782          * It will always be there, the vnodeops level does a lookup first.
 783          */
 784         if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
 785                 return error;
 786         }
 787         dp = args->dp;
 788         tp = args->trans;
 789         mp = dp->i_mount;
 790         hdr = bp->b_addr;
 791         btp = xfs_dir2_block_tail_p(mp, hdr);
 792         blp = xfs_dir2_block_leaf_p(btp);
 793         /*
 794          * Point to the data entry using the leaf entry.
 795          */
 796         dep = (xfs_dir2_data_entry_t *)
 797               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 798         /*
 799          * Mark the data entry's space free.
 800          */
 801         needlog = needscan = 0;
 802         xfs_dir2_data_make_free(tp, bp,
 803                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
 804                 xfs_dir3_data_entsize(mp, dep->namelen), &needlog, &needscan);
 805         /*
 806          * Fix up the block tail.
 807          */
 808         be32_add_cpu(&btp->stale, 1);
 809         xfs_dir2_block_log_tail(tp, bp);
 810         /*
 811          * Remove the leaf entry by marking it stale.
 812          */
 813         blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
 814         xfs_dir2_block_log_leaf(tp, bp, ent, ent);
 815         /*
 816          * Fix up bestfree, log the header if necessary.
 817          */
 818         if (needscan)
 819                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 820         if (needlog)
 821                 xfs_dir2_data_log_header(tp, bp);
 822         xfs_dir3_data_check(dp, bp);
 823         /*
 824          * See if the size as a shortform is good enough.
 825          */
 826         size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
 827         if (size > XFS_IFORK_DSIZE(dp))
 828                 return 0;
 829
 830         /*
 831          * If it works, do the conversion.
 832          */
 833         return xfs_dir2_block_to_sf(args, bp, size, &sfh);
 834 }
 835
 836 /*
 837  * Replace an entry in a V2 block directory.
 838  * Change the inode number to the new value.
 839  */
 840 int                                             /* error */
 841 xfs_dir2_block_replace(
 842         xfs_da_args_t           *args)          /* directory operation args */
 843 {
 844         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 845         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
 846         struct xfs_buf          *bp;            /* block buffer */
 847         xfs_dir2_block_tail_t   *btp;           /* block tail */
 848         xfs_dir2_data_entry_t   *dep;           /* block data entry */
 849         xfs_inode_t             *dp;            /* incore inode */
 850         int                     ent;            /* leaf entry index */
 851         int                     error;          /* error return value */
 852         xfs_mount_t             *mp;            /* filesystem mount point */
 853
 854         trace_xfs_dir2_block_replace(args);
 855
 856         /*
 857          * Lookup the entry in the directory.  Get buffer and entry index.
 858          * This will always succeed since the caller has already done a lookup.
 859          */
 860         if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
 861                 return error;
 862         }
 863         dp = args->dp;
 864         mp = dp->i_mount;
 865         hdr = bp->b_addr;
 866         btp = xfs_dir2_block_tail_p(mp, hdr);
 867         blp = xfs_dir2_block_leaf_p(btp);
 868         /*
 869          * Point to the data entry we need to change.
 870          */
 871         dep = (xfs_dir2_data_entry_t *)
 872               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 873         ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
 874         /*
 875          * Change the inode number to the new value.
 876          */
 877         dep->inumber = cpu_to_be64(args->inumber);
 878         xfs_dir3_dirent_put_ftype(mp, dep, args->filetype);
 879         xfs_dir2_data_log_entry(args->trans, bp, dep);
 880         xfs_dir3_data_check(dp, bp);
 881         return 0;
 882 }
 883
 884 /*
 885  * Qsort comparison routine for the block leaf entries.
 886  */
 887 static int                                      /* sort order */
 888 xfs_dir2_block_sort(
 889         const void                      *a,     /* first leaf entry */
 890         const void                      *b)     /* second leaf entry */
 891 {
 892         const xfs_dir2_leaf_entry_t     *la;    /* first leaf entry */
 893         const xfs_dir2_leaf_entry_t     *lb;    /* second leaf entry */
 894
 895         la = a;
 896         lb = b;
 897         return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
 898                 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
 899 }
 900
 901 /*
 902  * Convert a V2 leaf directory to a V2 block directory if possible.
 903  */
 904 int                                             /* error */
 905 xfs_dir2_leaf_to_block(
 906         xfs_da_args_t           *args,          /* operation arguments */
 907         struct xfs_buf          *lbp,           /* leaf buffer */
 908         struct xfs_buf          *dbp)           /* data buffer */
 909 {
 910         __be16                  *bestsp;        /* leaf bests table */
 911         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 912         xfs_dir2_block_tail_t   *btp;           /* block tail */
 913         xfs_inode_t             *dp;            /* incore directory inode */
 914         xfs_dir2_data_unused_t  *dup;           /* unused data entry */
 915         int                     error;          /* error return value */
 916         int                     from;           /* leaf from index */
 917         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 918         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 919         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 920         xfs_mount_t             *mp;            /* file system mount point */
 921         int                     needlog;        /* need to log data header */
 922         int                     needscan;       /* need to scan for bestfree */
 923         xfs_dir2_sf_hdr_t       sfh;            /* shortform header */
 924         int                     size;           /* bytes used */
 925         __be16                  *tagp;          /* end of entry (tag) */
 926         int                     to;             /* block/leaf to index */
 927         xfs_trans_t             *tp;            /* transaction pointer */
 928         struct xfs_dir2_leaf_entry *ents;
 929         struct xfs_dir3_icleaf_hdr leafhdr;
 930
 931         trace_xfs_dir2_leaf_to_block(args);
 932
 933         dp = args->dp;
 934         tp = args->trans;
 935         mp = dp->i_mount;
 936         leaf = lbp->b_addr;
 937         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 938         ents = xfs_dir3_leaf_ents_p(leaf);
 939         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 940
 941         ASSERT(leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
 942                leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
 943         /*
 944          * If there are data blocks other than the first one, take this
 945          * opportunity to remove trailing empty data blocks that may have
 946          * been left behind during no-space-reservation operations.
 947          * These will show up in the leaf bests table.
 948          */
 949         while (dp->i_d.di_size > mp->m_dirblksize) {
 950                 int hdrsz;
 951
 952                 hdrsz = xfs_dir3_data_hdr_size(xfs_sb_version_hascrc(&mp->m_sb));
 953                 bestsp = xfs_dir2_leaf_bests_p(ltp);
 954                 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
 955                                             mp->m_dirblksize - hdrsz) {
 956                         if ((error =
 957                             xfs_dir2_leaf_trim_data(args, lbp,
 958                                     (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
 959                                 return error;
 960                 } else
 961                         return 0;
 962         }
 963         /*
 964          * Read the data block if we don't already have it, give up if it fails.
 965          */
 966         if (!dbp) {
 967                 error = xfs_dir3_data_read(tp, dp, mp->m_dirdatablk, -1, &dbp);
 968                 if (error)
 969                         return error;
 970         }
 971         hdr = dbp->b_addr;
 972         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 973                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
 974
 975         /*
 976          * Size of the "leaf" area in the block.
 977          */
 978         size = (uint)sizeof(xfs_dir2_block_tail_t) +
 979                (uint)sizeof(*lep) * (leafhdr.count - leafhdr.stale);
 980         /*
 981          * Look at the last data entry.
 982          */
 983         tagp = (__be16 *)((char *)hdr + mp->m_dirblksize) - 1;
 984         dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 985         /*
 986          * If it's not free or is too short we can't do it.
 987          */
 988         if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
 989             be16_to_cpu(dup->length) < size)
 990                 return 0;
 991
 992         /*
 993          * Start converting it to block form.
 994          */
 995         xfs_dir3_block_init(mp, tp, dbp, dp);
 996
 997         needlog = 1;
 998         needscan = 0;
 999         /*
1000          * Use up the space at the end of the block (blp/btp).
1001          */
1002         xfs_dir2_data_use_free(tp, dbp, dup, mp->m_dirblksize - size, size,
1003                 &needlog, &needscan);
1004         /*
1005          * Initialize the block tail.
1006          */
1007         btp = xfs_dir2_block_tail_p(mp, hdr);
1008         btp->count = cpu_to_be32(leafhdr.count - leafhdr.stale);
1009         btp->stale = 0;
1010         xfs_dir2_block_log_tail(tp, dbp);
1011         /*
1012          * Initialize the block leaf area.  We compact out stale entries.
1013          */
1014         lep = xfs_dir2_block_leaf_p(btp);
1015         for (from = to = 0; from < leafhdr.count; from++) {
1016                 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
1017                         continue;
1018                 lep[to++] = ents[from];
1019         }
1020         ASSERT(to == be32_to_cpu(btp->count));
1021         xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
1022         /*
1023          * Scan the bestfree if we need it and log the data block header.
1024          */
1025         if (needscan)
1026                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1027         if (needlog)
1028                 xfs_dir2_data_log_header(tp, dbp);
1029         /*
1030          * Pitch the old leaf block.
1031          */
1032         error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
1033         if (error)
1034                 return error;
1035
1036         /*
1037          * Now see if the resulting block can be shrunken to shortform.
1038          */
1039         size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1040         if (size > XFS_IFORK_DSIZE(dp))
1041                 return 0;
1042
1043         return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1044 }
1045
1046 /*
1047  * Convert the shortform directory to block form.
1048  */
1049 int                                             /* error */
1050 xfs_dir2_sf_to_block(
1051         xfs_da_args_t           *args)          /* operation arguments */
1052 {
1053         xfs_dir2_db_t           blkno;          /* dir-relative block # (0) */
1054         xfs_dir2_data_hdr_t     *hdr;           /* block header */
1055         xfs_dir2_leaf_entry_t   *blp;           /* block leaf entries */
1056         struct xfs_buf          *bp;            /* block buffer */
1057         xfs_dir2_block_tail_t   *btp;           /* block tail pointer */
1058         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1059         xfs_inode_t             *dp;            /* incore directory inode */
1060         int                     dummy;          /* trash */
1061         xfs_dir2_data_unused_t  *dup;           /* unused entry pointer */
1062         int                     endoffset;      /* end of data objects */
1063         int                     error;          /* error return value */
1064         int                     i;              /* index */
1065         xfs_mount_t             *mp;            /* filesystem mount point */
1066         int                     needlog;        /* need to log block header */
1067         int                     needscan;       /* need to scan block freespc */
1068         int                     newoffset;      /* offset from current entry */
1069         int                     offset;         /* target block offset */
1070         xfs_dir2_sf_entry_t     *sfep;          /* sf entry pointer */
1071         xfs_dir2_sf_hdr_t       *oldsfp;        /* old shortform header  */
1072         xfs_dir2_sf_hdr_t       *sfp;           /* shortform header  */
1073         __be16                  *tagp;          /* end of data entry */
1074         xfs_trans_t             *tp;            /* transaction pointer */
1075         struct xfs_name         name;
1076         struct xfs_ifork        *ifp;
1077
1078         trace_xfs_dir2_sf_to_block(args);
1079
1080         dp = args->dp;
1081         tp = args->trans;
1082         mp = dp->i_mount;
1083         ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
1084         ASSERT(ifp->if_flags & XFS_IFINLINE);
1085         /*
1086          * Bomb out if the shortform directory is way too short.
1087          */
1088         if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1089                 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1090                 return XFS_ERROR(EIO);
1091         }
1092
1093         oldsfp = (xfs_dir2_sf_hdr_t *)ifp->if_u1.if_data;
1094
1095         ASSERT(ifp->if_bytes == dp->i_d.di_size);
1096         ASSERT(ifp->if_u1.if_data != NULL);
1097         ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1098         ASSERT(dp->i_d.di_nextents == 0);
1099
1100         /*
1101          * Copy the directory into a temporary buffer.
1102          * Then pitch the incore inode data so we can make extents.
1103          */
1104         sfp = kmem_alloc(ifp->if_bytes, KM_SLEEP);
1105         memcpy(sfp, oldsfp, ifp->if_bytes);
1106
1107         xfs_idata_realloc(dp, -ifp->if_bytes, XFS_DATA_FORK);
1108         xfs_bmap_local_to_extents_empty(dp, XFS_DATA_FORK);
1109         dp->i_d.di_size = 0;
1110
1111         /*
1112          * Add block 0 to the inode.
1113          */
1114         error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1115         if (error) {
1116                 kmem_free(sfp);
1117                 return error;
1118         }
1119         /*
1120          * Initialize the data block, then convert it to block format.
1121          */
1122         error = xfs_dir3_data_init(args, blkno, &bp);
1123         if (error) {
1124                 kmem_free(sfp);
1125                 return error;
1126         }
1127         xfs_dir3_block_init(mp, tp, bp, dp);
1128         hdr = bp->b_addr;
1129
1130         /*
1131          * Compute size of block "tail" area.
1132          */
1133         i = (uint)sizeof(*btp) +
1134             (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1135         /*
1136          * The whole thing is initialized to free by the init routine.
1137          * Say we're using the leaf and tail area.
1138          */
1139         dup = xfs_dir3_data_unused_p(hdr);
1140         needlog = needscan = 0;
1141         xfs_dir2_data_use_free(tp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1142                 &needscan);
1143         ASSERT(needscan == 0);
1144         /*
1145          * Fill in the tail.
1146          */
1147         btp = xfs_dir2_block_tail_p(mp, hdr);
1148         btp->count = cpu_to_be32(sfp->count + 2);       /* ., .. */
1149         btp->stale = 0;
1150         blp = xfs_dir2_block_leaf_p(btp);
1151         endoffset = (uint)((char *)blp - (char *)hdr);
1152         /*
1153          * Remove the freespace, we'll manage it.
1154          */
1155         xfs_dir2_data_use_free(tp, bp, dup,
1156                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1157                 be16_to_cpu(dup->length), &needlog, &needscan);
1158         /*
1159          * Create entry for .
1160          */
1161         dep = xfs_dir3_data_dot_entry_p(hdr);
1162         dep->inumber = cpu_to_be64(dp->i_ino);
1163         dep->namelen = 1;
1164         dep->name[0] = '.';
1165         xfs_dir3_dirent_put_ftype(mp, dep, XFS_DIR3_FT_DIR);
1166         tagp = xfs_dir3_data_entry_tag_p(mp, dep);
1167         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1168         xfs_dir2_data_log_entry(tp, bp, dep);
1169         blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1170         blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1171                                 (char *)dep - (char *)hdr));
1172         /*
1173          * Create entry for ..
1174          */
1175         dep = xfs_dir3_data_dotdot_entry_p(hdr);
1176         dep->inumber = cpu_to_be64(xfs_dir2_sf_get_parent_ino(sfp));
1177         dep->namelen = 2;
1178         dep->name[0] = dep->name[1] = '.';
1179         xfs_dir3_dirent_put_ftype(mp, dep, XFS_DIR3_FT_DIR);
1180         tagp = xfs_dir3_data_entry_tag_p(mp, dep);
1181         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1182         xfs_dir2_data_log_entry(tp, bp, dep);
1183         blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1184         blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1185                                 (char *)dep - (char *)hdr));
1186         offset = xfs_dir3_data_first_offset(hdr);
1187         /*
1188          * Loop over existing entries, stuff them in.
1189          */
1190         i = 0;
1191         if (!sfp->count)
1192                 sfep = NULL;
1193         else
1194                 sfep = xfs_dir2_sf_firstentry(sfp);
1195         /*
1196          * Need to preserve the existing offset values in the sf directory.
1197          * Insert holes (unused entries) where necessary.
1198          */
1199         while (offset < endoffset) {
1200                 /*
1201                  * sfep is null when we reach the end of the list.
1202                  */
1203                 if (sfep == NULL)
1204                         newoffset = endoffset;
1205                 else
1206                         newoffset = xfs_dir2_sf_get_offset(sfep);
1207                 /*
1208                  * There should be a hole here, make one.
1209                  */
1210                 if (offset < newoffset) {
1211                         dup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1212                         dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1213                         dup->length = cpu_to_be16(newoffset - offset);
1214                         *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1215                                 ((char *)dup - (char *)hdr));
1216                         xfs_dir2_data_log_unused(tp, bp, dup);
1217                         xfs_dir2_data_freeinsert(hdr, dup, &dummy);
1218                         offset += be16_to_cpu(dup->length);
1219                         continue;
1220                 }
1221                 /*
1222                  * Copy a real entry.
1223                  */
1224                 dep = (xfs_dir2_data_entry_t *)((char *)hdr + newoffset);
1225                 dep->inumber = cpu_to_be64(xfs_dir3_sfe_get_ino(mp, sfp, sfep));
1226                 dep->namelen = sfep->namelen;
1227                 xfs_dir3_dirent_put_ftype(mp, dep,
1228                                         xfs_dir3_sfe_get_ftype(mp, sfp, sfep));
1229                 memcpy(dep->name, sfep->name, dep->namelen);
1230                 tagp = xfs_dir3_data_entry_tag_p(mp, dep);
1231                 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1232                 xfs_dir2_data_log_entry(tp, bp, dep);
1233                 name.name = sfep->name;
1234                 name.len = sfep->namelen;
1235                 blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1236                                                         hashname(&name));
1237                 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1238                                                  (char *)dep - (char *)hdr));
1239                 offset = (int)((char *)(tagp + 1) - (char *)hdr);
1240                 if (++i == sfp->count)
1241                         sfep = NULL;
1242                 else
1243                         sfep = xfs_dir3_sf_nextentry(mp, sfp, sfep);
1244         }
1245         /* Done with the temporary buffer */
1246         kmem_free(sfp);
1247         /*
1248          * Sort the leaf entries by hash value.
1249          */
1250         xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1251         /*
1252          * Log the leaf entry area and tail.
1253          * Already logged the header in data_init, ignore needlog.
1254          */
1255         ASSERT(needscan == 0);
1256         xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1257         xfs_dir2_block_log_tail(tp, bp);
1258         xfs_dir3_data_check(dp, bp);
1259         return 0;
1260 }