fs/xfs/xfs_dir2_leaf.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_bmap.h"
  31 #include "xfs_dir2.h"
  32 #include "xfs_dir2_priv.h"
  33 #include "xfs_error.h"
  34 #include "xfs_trace.h"
  35 #include "xfs_trans.h"
  36 #include "xfs_buf_item.h"
  37 #include "xfs_cksum.h"
  38
  39 /*
  40  * Local function declarations.
  41  */
  42 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
  43                                     int *indexp, struct xfs_buf **dbpp);
  44 static void xfs_dir3_leaf_log_bests(struct xfs_trans *tp, struct xfs_buf *bp,
  45                                     int first, int last);
  46 static void xfs_dir3_leaf_log_tail(struct xfs_trans *tp, struct xfs_buf *bp);
  47
  48 /*
  49  * Check the internal consistency of a leaf1 block.
  50  * Pop an assert if something is wrong.
  51  */
  52 #ifdef DEBUG
  53 #define xfs_dir3_leaf_check(dp, bp) \
  54 do { \
  55         if (!xfs_dir3_leaf1_check((dp), (bp))) \
  56                 ASSERT(0); \
  57 } while (0);
  58
  59 STATIC bool
  60 xfs_dir3_leaf1_check(
  61         struct xfs_inode        *dp,
  62         struct xfs_buf          *bp)
  63 {
  64         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  65         struct xfs_dir3_icleaf_hdr leafhdr;
  66
  67         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
  68
  69         if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
  70                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  71                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  72                         return false;
  73         } else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
  74                 return false;
  75
  76         return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
  77 }
  78 #else
  79 #define xfs_dir3_leaf_check(dp, bp)
  80 #endif
  81
  82 bool
  83 xfs_dir3_leaf_check_int(
  84         struct xfs_mount        *mp,
  85         struct xfs_inode        *dp,
  86         struct xfs_dir3_icleaf_hdr *hdr,
  87         struct xfs_dir2_leaf    *leaf)
  88 {
  89         struct xfs_dir2_leaf_entry *ents;
  90         xfs_dir2_leaf_tail_t    *ltp;
  91         int                     stale;
  92         int                     i;
  93         const struct xfs_dir_ops *ops;
  94         struct xfs_dir3_icleaf_hdr leafhdr;
  95
  96         /*
  97          * we can be passed a null dp here from a verifier, so we need to go the
  98          * hard way to get them.
  99          */
 100         ops = xfs_dir_get_ops(mp, dp);
 101
 102         if (!hdr) {
 103                 ops->leaf_hdr_from_disk(&leafhdr, leaf);
 104                 hdr = &leafhdr;
 105         }
 106
 107         ents = ops->leaf_ents_p(leaf);
 108         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 109
 110         /*
 111          * XXX (dgc): This value is not restrictive enough.
 112          * Should factor in the size of the bests table as well.
 113          * We can deduce a value for that from di_size.
 114          */
 115         if (hdr->count > ops->leaf_max_ents(mp))
 116                 return false;
 117
 118         /* Leaves and bests don't overlap in leaf format. */
 119         if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
 120              hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
 121             (char *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
 122                 return false;
 123
 124         /* Check hash value order, count stale entries.  */
 125         for (i = stale = 0; i < hdr->count; i++) {
 126                 if (i + 1 < hdr->count) {
 127                         if (be32_to_cpu(ents[i].hashval) >
 128                                         be32_to_cpu(ents[i + 1].hashval))
 129                                 return false;
 130                 }
 131                 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 132                         stale++;
 133         }
 134         if (hdr->stale != stale)
 135                 return false;
 136         return true;
 137 }
 138
 139 /*
 140  * We verify the magic numbers before decoding the leaf header so that on debug
 141  * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
 142  * to incorrect magic numbers.
 143  */
 144 static bool
 145 xfs_dir3_leaf_verify(
 146         struct xfs_buf          *bp,
 147         __uint16_t              magic)
 148 {
 149         struct xfs_mount        *mp = bp->b_target->bt_mount;
 150         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 151
 152         ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
 153
 154         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 155                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 156                 __uint16_t              magic3;
 157
 158                 magic3 = (magic == XFS_DIR2_LEAF1_MAGIC) ? XFS_DIR3_LEAF1_MAGIC
 159                                                          : XFS_DIR3_LEAFN_MAGIC;
 160
 161                 if (leaf3->info.hdr.magic != cpu_to_be16(magic3))
 162                         return false;
 163                 if (!uuid_equal(&leaf3->info.uuid, &mp->m_sb.sb_uuid))
 164                         return false;
 165                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
 166                         return false;
 167         } else {
 168                 if (leaf->hdr.info.magic != cpu_to_be16(magic))
 169                         return false;
 170         }
 171
 172         return xfs_dir3_leaf_check_int(mp, NULL, NULL, leaf);
 173 }
 174
 175 static void
 176 __read_verify(
 177         struct xfs_buf  *bp,
 178         __uint16_t      magic)
 179 {
 180         struct xfs_mount        *mp = bp->b_target->bt_mount;
 181
 182         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
 183              !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF)) ||
 184             !xfs_dir3_leaf_verify(bp, magic)) {
 185                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 186                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 187         }
 188 }
 189
 190 static void
 191 __write_verify(
 192         struct xfs_buf  *bp,
 193         __uint16_t      magic)
 194 {
 195         struct xfs_mount        *mp = bp->b_target->bt_mount;
 196         struct xfs_buf_log_item *bip = bp->b_fspriv;
 197         struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
 198
 199         if (!xfs_dir3_leaf_verify(bp, magic)) {
 200                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 201                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 202                 return;
 203         }
 204
 205         if (!xfs_sb_version_hascrc(&mp->m_sb))
 206                 return;
 207
 208         if (bip)
 209                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 210
 211         xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
 212 }
 213
 214 static void
 215 xfs_dir3_leaf1_read_verify(
 216         struct xfs_buf  *bp)
 217 {
 218         __read_verify(bp, XFS_DIR2_LEAF1_MAGIC);
 219 }
 220
 221 static void
 222 xfs_dir3_leaf1_write_verify(
 223         struct xfs_buf  *bp)
 224 {
 225         __write_verify(bp, XFS_DIR2_LEAF1_MAGIC);
 226 }
 227
 228 static void
 229 xfs_dir3_leafn_read_verify(
 230         struct xfs_buf  *bp)
 231 {
 232         __read_verify(bp, XFS_DIR2_LEAFN_MAGIC);
 233 }
 234
 235 static void
 236 xfs_dir3_leafn_write_verify(
 237         struct xfs_buf  *bp)
 238 {
 239         __write_verify(bp, XFS_DIR2_LEAFN_MAGIC);
 240 }
 241
 242 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
 243         .verify_read = xfs_dir3_leaf1_read_verify,
 244         .verify_write = xfs_dir3_leaf1_write_verify,
 245 };
 246
 247 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
 248         .verify_read = xfs_dir3_leafn_read_verify,
 249         .verify_write = xfs_dir3_leafn_write_verify,
 250 };
 251
 252 static int
 253 xfs_dir3_leaf_read(
 254         struct xfs_trans        *tp,
 255         struct xfs_inode        *dp,
 256         xfs_dablk_t             fbno,
 257         xfs_daddr_t             mappedbno,
 258         struct xfs_buf          **bpp)
 259 {
 260         int                     err;
 261
 262         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 263                                 XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
 264         if (!err && tp)
 265                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
 266         return err;
 267 }
 268
 269 int
 270 xfs_dir3_leafn_read(
 271         struct xfs_trans        *tp,
 272         struct xfs_inode        *dp,
 273         xfs_dablk_t             fbno,
 274         xfs_daddr_t             mappedbno,
 275         struct xfs_buf          **bpp)
 276 {
 277         int                     err;
 278
 279         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 280                                 XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
 281         if (!err && tp)
 282                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
 283         return err;
 284 }
 285
 286 /*
 287  * Initialize a new leaf block, leaf1 or leafn magic accepted.
 288  */
 289 static void
 290 xfs_dir3_leaf_init(
 291         struct xfs_mount        *mp,
 292         struct xfs_trans        *tp,
 293         struct xfs_buf          *bp,
 294         xfs_ino_t               owner,
 295         __uint16_t              type)
 296 {
 297         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 298
 299         ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
 300
 301         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 302                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 303
 304                 memset(leaf3, 0, sizeof(*leaf3));
 305
 306                 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
 307                                          ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
 308                                          : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 309                 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
 310                 leaf3->info.owner = cpu_to_be64(owner);
 311                 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_uuid);
 312         } else {
 313                 memset(leaf, 0, sizeof(*leaf));
 314                 leaf->hdr.info.magic = cpu_to_be16(type);
 315         }
 316
 317         /*
 318          * If it's a leaf-format directory initialize the tail.
 319          * Caller is responsible for initialising the bests table.
 320          */
 321         if (type == XFS_DIR2_LEAF1_MAGIC) {
 322                 struct xfs_dir2_leaf_tail *ltp;
 323
 324                 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 325                 ltp->bestcount = 0;
 326                 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
 327                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
 328         } else {
 329                 bp->b_ops = &xfs_dir3_leafn_buf_ops;
 330                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
 331         }
 332 }
 333
 334 int
 335 xfs_dir3_leaf_get_buf(
 336         xfs_da_args_t           *args,
 337         xfs_dir2_db_t           bno,
 338         struct xfs_buf          **bpp,
 339         __uint16_t              magic)
 340 {
 341         struct xfs_inode        *dp = args->dp;
 342         struct xfs_trans        *tp = args->trans;
 343         struct xfs_mount        *mp = dp->i_mount;
 344         struct xfs_buf          *bp;
 345         int                     error;
 346
 347         ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
 348         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
 349                bno < XFS_DIR2_FREE_FIRSTDB(mp));
 350
 351         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
 352                                XFS_DATA_FORK);
 353         if (error)
 354                 return error;
 355
 356         xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
 357         xfs_dir3_leaf_log_header(tp, dp, bp);
 358         if (magic == XFS_DIR2_LEAF1_MAGIC)
 359                 xfs_dir3_leaf_log_tail(tp, bp);
 360         *bpp = bp;
 361         return 0;
 362 }
 363
 364 /*
 365  * Convert a block form directory to a leaf form directory.
 366  */
 367 int                                             /* error */
 368 xfs_dir2_block_to_leaf(
 369         xfs_da_args_t           *args,          /* operation arguments */
 370         struct xfs_buf          *dbp)           /* input block's buffer */
 371 {
 372         __be16                  *bestsp;        /* leaf's bestsp entries */
 373         xfs_dablk_t             blkno;          /* leaf block's bno */
 374         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 375         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
 376         xfs_dir2_block_tail_t   *btp;           /* block's tail */
 377         xfs_inode_t             *dp;            /* incore directory inode */
 378         int                     error;          /* error return code */
 379         struct xfs_buf          *lbp;           /* leaf block's buffer */
 380         xfs_dir2_db_t           ldb;            /* leaf block's bno */
 381         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 382         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
 383         xfs_mount_t             *mp;            /* filesystem mount point */
 384         int                     needlog;        /* need to log block header */
 385         int                     needscan;       /* need to rescan bestfree */
 386         xfs_trans_t             *tp;            /* transaction pointer */
 387         struct xfs_dir2_data_free *bf;
 388         struct xfs_dir2_leaf_entry *ents;
 389         struct xfs_dir3_icleaf_hdr leafhdr;
 390
 391         trace_xfs_dir2_block_to_leaf(args);
 392
 393         dp = args->dp;
 394         mp = dp->i_mount;
 395         tp = args->trans;
 396         /*
 397          * Add the leaf block to the inode.
 398          * This interface will only put blocks in the leaf/node range.
 399          * Since that's empty now, we'll get the root (block 0 in range).
 400          */
 401         if ((error = xfs_da_grow_inode(args, &blkno))) {
 402                 return error;
 403         }
 404         ldb = xfs_dir2_da_to_db(mp, blkno);
 405         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
 406         /*
 407          * Initialize the leaf block, get a buffer for it.
 408          */
 409         error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
 410         if (error)
 411                 return error;
 412
 413         leaf = lbp->b_addr;
 414         hdr = dbp->b_addr;
 415         xfs_dir3_data_check(dp, dbp);
 416         btp = xfs_dir2_block_tail_p(mp, hdr);
 417         blp = xfs_dir2_block_leaf_p(btp);
 418         bf = dp->d_ops->data_bestfree_p(hdr);
 419         ents = dp->d_ops->leaf_ents_p(leaf);
 420
 421         /*
 422          * Set the counts in the leaf header.
 423          */
 424         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 425         leafhdr.count = be32_to_cpu(btp->count);
 426         leafhdr.stale = be32_to_cpu(btp->stale);
 427         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
 428         xfs_dir3_leaf_log_header(tp, dp, lbp);
 429
 430         /*
 431          * Could compact these but I think we always do the conversion
 432          * after squeezing out stale entries.
 433          */
 434         memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
 435         xfs_dir3_leaf_log_ents(tp, dp, lbp, 0, leafhdr.count - 1);
 436         needscan = 0;
 437         needlog = 1;
 438         /*
 439          * Make the space formerly occupied by the leaf entries and block
 440          * tail be free.
 441          */
 442         xfs_dir2_data_make_free(tp, dp, dbp,
 443                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 444                 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
 445                                        (char *)blp),
 446                 &needlog, &needscan);
 447         /*
 448          * Fix up the block header, make it a data block.
 449          */
 450         dbp->b_ops = &xfs_dir3_data_buf_ops;
 451         xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
 452         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
 453                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 454         else
 455                 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 456
 457         if (needscan)
 458                 xfs_dir2_data_freescan(dp, hdr, &needlog);
 459         /*
 460          * Set up leaf tail and bests table.
 461          */
 462         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 463         ltp->bestcount = cpu_to_be32(1);
 464         bestsp = xfs_dir2_leaf_bests_p(ltp);
 465         bestsp[0] =  bf[0].length;
 466         /*
 467          * Log the data header and leaf bests table.
 468          */
 469         if (needlog)
 470                 xfs_dir2_data_log_header(tp, dp, dbp);
 471         xfs_dir3_leaf_check(dp, lbp);
 472         xfs_dir3_data_check(dp, dbp);
 473         xfs_dir3_leaf_log_bests(tp, lbp, 0, 0);
 474         return 0;
 475 }
 476
 477 STATIC void
 478 xfs_dir3_leaf_find_stale(
 479         struct xfs_dir3_icleaf_hdr *leafhdr,
 480         struct xfs_dir2_leaf_entry *ents,
 481         int                     index,
 482         int                     *lowstale,
 483         int                     *highstale)
 484 {
 485         /*
 486          * Find the first stale entry before our index, if any.
 487          */
 488         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
 489                 if (ents[*lowstale].address ==
 490                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 491                         break;
 492         }
 493
 494         /*
 495          * Find the first stale entry at or after our index, if any.
 496          * Stop if the result would require moving more entries than using
 497          * lowstale.
 498          */
 499         for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
 500                 if (ents[*highstale].address ==
 501                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 502                         break;
 503                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
 504                         break;
 505         }
 506 }
 507
 508 struct xfs_dir2_leaf_entry *
 509 xfs_dir3_leaf_find_entry(
 510         struct xfs_dir3_icleaf_hdr *leafhdr,
 511         struct xfs_dir2_leaf_entry *ents,
 512         int                     index,          /* leaf table position */
 513         int                     compact,        /* need to compact leaves */
 514         int                     lowstale,       /* index of prev stale leaf */
 515         int                     highstale,      /* index of next stale leaf */
 516         int                     *lfloglow,      /* low leaf logging index */
 517         int                     *lfloghigh)     /* high leaf logging index */
 518 {
 519         if (!leafhdr->stale) {
 520                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
 521
 522                 /*
 523                  * Now we need to make room to insert the leaf entry.
 524                  *
 525                  * If there are no stale entries, just insert a hole at index.
 526                  */
 527                 lep = &ents[index];
 528                 if (index < leafhdr->count)
 529                         memmove(lep + 1, lep,
 530                                 (leafhdr->count - index) * sizeof(*lep));
 531
 532                 /*
 533                  * Record low and high logging indices for the leaf.
 534                  */
 535                 *lfloglow = index;
 536                 *lfloghigh = leafhdr->count++;
 537                 return lep;
 538         }
 539
 540         /*
 541          * There are stale entries.
 542          *
 543          * We will use one of them for the new entry.  It's probably not at
 544          * the right location, so we'll have to shift some up or down first.
 545          *
 546          * If we didn't compact before, we need to find the nearest stale
 547          * entries before and after our insertion point.
 548          */
 549         if (compact == 0)
 550                 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
 551                                          &lowstale, &highstale);
 552
 553         /*
 554          * If the low one is better, use it.
 555          */
 556         if (lowstale >= 0 &&
 557             (highstale == leafhdr->count ||
 558              index - lowstale - 1 < highstale - index)) {
 559                 ASSERT(index - lowstale - 1 >= 0);
 560                 ASSERT(ents[lowstale].address ==
 561                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 562
 563                 /*
 564                  * Copy entries up to cover the stale entry and make room
 565                  * for the new entry.
 566                  */
 567                 if (index - lowstale - 1 > 0) {
 568                         memmove(&ents[lowstale], &ents[lowstale + 1],
 569                                 (index - lowstale - 1) *
 570                                         sizeof(xfs_dir2_leaf_entry_t));
 571                 }
 572                 *lfloglow = MIN(lowstale, *lfloglow);
 573                 *lfloghigh = MAX(index - 1, *lfloghigh);
 574                 leafhdr->stale--;
 575                 return &ents[index - 1];
 576         }
 577
 578         /*
 579          * The high one is better, so use that one.
 580          */
 581         ASSERT(highstale - index >= 0);
 582         ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 583
 584         /*
 585          * Copy entries down to cover the stale entry and make room for the
 586          * new entry.
 587          */
 588         if (highstale - index > 0) {
 589                 memmove(&ents[index + 1], &ents[index],
 590                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
 591         }
 592         *lfloglow = MIN(index, *lfloglow);
 593         *lfloghigh = MAX(highstale, *lfloghigh);
 594         leafhdr->stale--;
 595         return &ents[index];
 596 }
 597
 598 /*
 599  * Add an entry to a leaf form directory.
 600  */
 601 int                                             /* error */
 602 xfs_dir2_leaf_addname(
 603         xfs_da_args_t           *args)          /* operation arguments */
 604 {
 605         __be16                  *bestsp;        /* freespace table in leaf */
 606         int                     compact;        /* need to compact leaves */
 607         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 608         struct xfs_buf          *dbp;           /* data block buffer */
 609         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 610         xfs_inode_t             *dp;            /* incore directory inode */
 611         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
 612         int                     error;          /* error return value */
 613         int                     grown;          /* allocated new data block */
 614         int                     highstale;      /* index of next stale leaf */
 615         int                     i;              /* temporary, index */
 616         int                     index;          /* leaf table position */
 617         struct xfs_buf          *lbp;           /* leaf's buffer */
 618         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 619         int                     length;         /* length of new entry */
 620         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
 621         int                     lfloglow;       /* low leaf logging index */
 622         int                     lfloghigh;      /* high leaf logging index */
 623         int                     lowstale;       /* index of prev stale leaf */
 624         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 625         xfs_mount_t             *mp;            /* filesystem mount point */
 626         int                     needbytes;      /* leaf block bytes needed */
 627         int                     needlog;        /* need to log data header */
 628         int                     needscan;       /* need to rescan data free */
 629         __be16                  *tagp;          /* end of data entry */
 630         xfs_trans_t             *tp;            /* transaction pointer */
 631         xfs_dir2_db_t           use_block;      /* data block number */
 632         struct xfs_dir2_data_free *bf;          /* bestfree table */
 633         struct xfs_dir2_leaf_entry *ents;
 634         struct xfs_dir3_icleaf_hdr leafhdr;
 635
 636         trace_xfs_dir2_leaf_addname(args);
 637
 638         dp = args->dp;
 639         tp = args->trans;
 640         mp = dp->i_mount;
 641
 642         error = xfs_dir3_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
 643         if (error)
 644                 return error;
 645
 646         /*
 647          * Look up the entry by hash value and name.
 648          * We know it's not there, our caller has already done a lookup.
 649          * So the index is of the entry to insert in front of.
 650          * But if there are dup hash values the index is of the first of those.
 651          */
 652         index = xfs_dir2_leaf_search_hash(args, lbp);
 653         leaf = lbp->b_addr;
 654         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 655         ents = dp->d_ops->leaf_ents_p(leaf);
 656         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 657         bestsp = xfs_dir2_leaf_bests_p(ltp);
 658         length = dp->d_ops->data_entsize(args->namelen);
 659
 660         /*
 661          * See if there are any entries with the same hash value
 662          * and space in their block for the new entry.
 663          * This is good because it puts multiple same-hash value entries
 664          * in a data block, improving the lookup of those entries.
 665          */
 666         for (use_block = -1, lep = &ents[index];
 667              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 668              index++, lep++) {
 669                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 670                         continue;
 671                 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 672                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 673                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
 674                 if (be16_to_cpu(bestsp[i]) >= length) {
 675                         use_block = i;
 676                         break;
 677                 }
 678         }
 679         /*
 680          * Didn't find a block yet, linear search all the data blocks.
 681          */
 682         if (use_block == -1) {
 683                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 684                         /*
 685                          * Remember a block we see that's missing.
 686                          */
 687                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
 688                             use_block == -1)
 689                                 use_block = i;
 690                         else if (be16_to_cpu(bestsp[i]) >= length) {
 691                                 use_block = i;
 692                                 break;
 693                         }
 694                 }
 695         }
 696         /*
 697          * How many bytes do we need in the leaf block?
 698          */
 699         needbytes = 0;
 700         if (!leafhdr.stale)
 701                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
 702         if (use_block == -1)
 703                 needbytes += sizeof(xfs_dir2_data_off_t);
 704
 705         /*
 706          * Now kill use_block if it refers to a missing block, so we
 707          * can use it as an indication of allocation needed.
 708          */
 709         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
 710                 use_block = -1;
 711         /*
 712          * If we don't have enough free bytes but we can make enough
 713          * by compacting out stale entries, we'll do that.
 714          */
 715         if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
 716             leafhdr.stale > 1)
 717                 compact = 1;
 718
 719         /*
 720          * Otherwise if we don't have enough free bytes we need to
 721          * convert to node form.
 722          */
 723         else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
 724                 /*
 725                  * Just checking or no space reservation, give up.
 726                  */
 727                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 728                                                         args->total == 0) {
 729                         xfs_trans_brelse(tp, lbp);
 730                         return XFS_ERROR(ENOSPC);
 731                 }
 732                 /*
 733                  * Convert to node form.
 734                  */
 735                 error = xfs_dir2_leaf_to_node(args, lbp);
 736                 if (error)
 737                         return error;
 738                 /*
 739                  * Then add the new entry.
 740                  */
 741                 return xfs_dir2_node_addname(args);
 742         }
 743         /*
 744          * Otherwise it will fit without compaction.
 745          */
 746         else
 747                 compact = 0;
 748         /*
 749          * If just checking, then it will fit unless we needed to allocate
 750          * a new data block.
 751          */
 752         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 753                 xfs_trans_brelse(tp, lbp);
 754                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
 755         }
 756         /*
 757          * If no allocations are allowed, return now before we've
 758          * changed anything.
 759          */
 760         if (args->total == 0 && use_block == -1) {
 761                 xfs_trans_brelse(tp, lbp);
 762                 return XFS_ERROR(ENOSPC);
 763         }
 764         /*
 765          * Need to compact the leaf entries, removing stale ones.
 766          * Leave one stale entry behind - the one closest to our
 767          * insertion index - and we'll shift that one to our insertion
 768          * point later.
 769          */
 770         if (compact) {
 771                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 772                         &highstale, &lfloglow, &lfloghigh);
 773         }
 774         /*
 775          * There are stale entries, so we'll need log-low and log-high
 776          * impossibly bad values later.
 777          */
 778         else if (leafhdr.stale) {
 779                 lfloglow = leafhdr.count;
 780                 lfloghigh = -1;
 781         }
 782         /*
 783          * If there was no data block space found, we need to allocate
 784          * a new one.
 785          */
 786         if (use_block == -1) {
 787                 /*
 788                  * Add the new data block.
 789                  */
 790                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 791                                 &use_block))) {
 792                         xfs_trans_brelse(tp, lbp);
 793                         return error;
 794                 }
 795                 /*
 796                  * Initialize the block.
 797                  */
 798                 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
 799                         xfs_trans_brelse(tp, lbp);
 800                         return error;
 801                 }
 802                 /*
 803                  * If we're adding a new data block on the end we need to
 804                  * extend the bests table.  Copy it up one entry.
 805                  */
 806                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
 807                         bestsp--;
 808                         memmove(&bestsp[0], &bestsp[1],
 809                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 810                         be32_add_cpu(&ltp->bestcount, 1);
 811                         xfs_dir3_leaf_log_tail(tp, lbp);
 812                         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
 813                 }
 814                 /*
 815                  * If we're filling in a previously empty block just log it.
 816                  */
 817                 else
 818                         xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
 819                 hdr = dbp->b_addr;
 820                 bf = dp->d_ops->data_bestfree_p(hdr);
 821                 bestsp[use_block] = bf[0].length;
 822                 grown = 1;
 823         } else {
 824                 /*
 825                  * Already had space in some data block.
 826                  * Just read that one in.
 827                  */
 828                 error = xfs_dir3_data_read(tp, dp,
 829                                            xfs_dir2_db_to_da(mp, use_block),
 830                                            -1, &dbp);
 831                 if (error) {
 832                         xfs_trans_brelse(tp, lbp);
 833                         return error;
 834                 }
 835                 hdr = dbp->b_addr;
 836                 bf = dp->d_ops->data_bestfree_p(hdr);
 837                 grown = 0;
 838         }
 839         /*
 840          * Point to the biggest freespace in our data block.
 841          */
 842         dup = (xfs_dir2_data_unused_t *)
 843               ((char *)hdr + be16_to_cpu(bf[0].offset));
 844         ASSERT(be16_to_cpu(dup->length) >= length);
 845         needscan = needlog = 0;
 846         /*
 847          * Mark the initial part of our freespace in use for the new entry.
 848          */
 849         xfs_dir2_data_use_free(tp, dp, dbp, dup,
 850                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
 851                 &needlog, &needscan);
 852         /*
 853          * Initialize our new entry (at last).
 854          */
 855         dep = (xfs_dir2_data_entry_t *)dup;
 856         dep->inumber = cpu_to_be64(args->inumber);
 857         dep->namelen = args->namelen;
 858         memcpy(dep->name, args->name, dep->namelen);
 859         dp->d_ops->data_put_ftype(dep, args->filetype);
 860         tagp = dp->d_ops->data_entry_tag_p(dep);
 861         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 862         /*
 863          * Need to scan fix up the bestfree table.
 864          */
 865         if (needscan)
 866                 xfs_dir2_data_freescan(dp, hdr, &needlog);
 867         /*
 868          * Need to log the data block's header.
 869          */
 870         if (needlog)
 871                 xfs_dir2_data_log_header(tp, dp, dbp);
 872         xfs_dir2_data_log_entry(tp, dp, dbp, dep);
 873         /*
 874          * If the bests table needs to be changed, do it.
 875          * Log the change unless we've already done that.
 876          */
 877         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
 878                 bestsp[use_block] = bf[0].length;
 879                 if (!grown)
 880                         xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
 881         }
 882
 883         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 884                                        highstale, &lfloglow, &lfloghigh);
 885
 886         /*
 887          * Fill in the new leaf entry.
 888          */
 889         lep->hashval = cpu_to_be32(args->hashval);
 890         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
 891                                 be16_to_cpu(*tagp)));
 892         /*
 893          * Log the leaf fields and give up the buffers.
 894          */
 895         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
 896         xfs_dir3_leaf_log_header(tp, dp, lbp);
 897         xfs_dir3_leaf_log_ents(tp, dp, lbp, lfloglow, lfloghigh);
 898         xfs_dir3_leaf_check(dp, lbp);
 899         xfs_dir3_data_check(dp, dbp);
 900         return 0;
 901 }
 902
 903 /*
 904  * Compact out any stale entries in the leaf.
 905  * Log the header and changed leaf entries, if any.
 906  */
 907 void
 908 xfs_dir3_leaf_compact(
 909         xfs_da_args_t   *args,          /* operation arguments */
 910         struct xfs_dir3_icleaf_hdr *leafhdr,
 911         struct xfs_buf  *bp)            /* leaf buffer */
 912 {
 913         int             from;           /* source leaf index */
 914         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 915         int             loglow;         /* first leaf entry to log */
 916         int             to;             /* target leaf index */
 917         struct xfs_dir2_leaf_entry *ents;
 918         struct xfs_inode *dp = args->dp;
 919
 920         leaf = bp->b_addr;
 921         if (!leafhdr->stale)
 922                 return;
 923
 924         /*
 925          * Compress out the stale entries in place.
 926          */
 927         ents = dp->d_ops->leaf_ents_p(leaf);
 928         for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
 929                 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 930                         continue;
 931                 /*
 932                  * Only actually copy the entries that are different.
 933                  */
 934                 if (from > to) {
 935                         if (loglow == -1)
 936                                 loglow = to;
 937                         ents[to] = ents[from];
 938                 }
 939                 to++;
 940         }
 941         /*
 942          * Update and log the header, log the leaf entries.
 943          */
 944         ASSERT(leafhdr->stale == from - to);
 945         leafhdr->count -= leafhdr->stale;
 946         leafhdr->stale = 0;
 947
 948         dp->d_ops->leaf_hdr_to_disk(leaf, leafhdr);
 949         xfs_dir3_leaf_log_header(args->trans, dp, bp);
 950         if (loglow != -1)
 951                 xfs_dir3_leaf_log_ents(args->trans, dp, bp, loglow, to - 1);
 952 }
 953
 954 /*
 955  * Compact the leaf entries, removing stale ones.
 956  * Leave one stale entry behind - the one closest to our
 957  * insertion index - and the caller will shift that one to our insertion
 958  * point later.
 959  * Return new insertion index, where the remaining stale entry is,
 960  * and leaf logging indices.
 961  */
 962 void
 963 xfs_dir3_leaf_compact_x1(
 964         struct xfs_dir3_icleaf_hdr *leafhdr,
 965         struct xfs_dir2_leaf_entry *ents,
 966         int             *indexp,        /* insertion index */
 967         int             *lowstalep,     /* out: stale entry before us */
 968         int             *highstalep,    /* out: stale entry after us */
 969         int             *lowlogp,       /* out: low log index */
 970         int             *highlogp)      /* out: high log index */
 971 {
 972         int             from;           /* source copy index */
 973         int             highstale;      /* stale entry at/after index */
 974         int             index;          /* insertion index */
 975         int             keepstale;      /* source index of kept stale */
 976         int             lowstale;       /* stale entry before index */
 977         int             newindex=0;     /* new insertion index */
 978         int             to;             /* destination copy index */
 979
 980         ASSERT(leafhdr->stale > 1);
 981         index = *indexp;
 982
 983         xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
 984
 985         /*
 986          * Pick the better of lowstale and highstale.
 987          */
 988         if (lowstale >= 0 &&
 989             (highstale == leafhdr->count ||
 990              index - lowstale <= highstale - index))
 991                 keepstale = lowstale;
 992         else
 993                 keepstale = highstale;
 994         /*
 995          * Copy the entries in place, removing all the stale entries
 996          * except keepstale.
 997          */
 998         for (from = to = 0; from < leafhdr->count; from++) {
 999                 /*
1000                  * Notice the new value of index.
1001                  */
1002                 if (index == from)
1003                         newindex = to;
1004                 if (from != keepstale &&
1005                     ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1006                         if (from == to)
1007                                 *lowlogp = to;
1008                         continue;
1009                 }
1010                 /*
1011                  * Record the new keepstale value for the insertion.
1012                  */
1013                 if (from == keepstale)
1014                         lowstale = highstale = to;
1015                 /*
1016                  * Copy only the entries that have moved.
1017                  */
1018                 if (from > to)
1019                         ents[to] = ents[from];
1020                 to++;
1021         }
1022         ASSERT(from > to);
1023         /*
1024          * If the insertion point was past the last entry,
1025          * set the new insertion point accordingly.
1026          */
1027         if (index == from)
1028                 newindex = to;
1029         *indexp = newindex;
1030         /*
1031          * Adjust the leaf header values.
1032          */
1033         leafhdr->count -= from - to;
1034         leafhdr->stale = 1;
1035         /*
1036          * Remember the low/high stale value only in the "right"
1037          * direction.
1038          */
1039         if (lowstale >= newindex)
1040                 lowstale = -1;
1041         else
1042                 highstale = leafhdr->count;
1043         *highlogp = leafhdr->count - 1;
1044         *lowstalep = lowstale;
1045         *highstalep = highstale;
1046 }
1047
1048 /*
1049  * Log the bests entries indicated from a leaf1 block.
1050  */
1051 static void
1052 xfs_dir3_leaf_log_bests(
1053         xfs_trans_t             *tp,            /* transaction pointer */
1054         struct xfs_buf          *bp,            /* leaf buffer */
1055         int                     first,          /* first entry to log */
1056         int                     last)           /* last entry to log */
1057 {
1058         __be16                  *firstb;        /* pointer to first entry */
1059         __be16                  *lastb;         /* pointer to last entry */
1060         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1061         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1062
1063         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1064                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1065
1066         ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1067         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1068         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1069         xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1070                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1071 }
1072
1073 /*
1074  * Log the leaf entries indicated from a leaf1 or leafn block.
1075  */
1076 void
1077 xfs_dir3_leaf_log_ents(
1078         struct xfs_trans        *tp,
1079         struct xfs_inode        *dp,
1080         struct xfs_buf          *bp,
1081         int                     first,
1082         int                     last)
1083 {
1084         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1085         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1086         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1087         struct xfs_dir2_leaf_entry *ents;
1088
1089         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1090                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1091                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1092                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1093
1094         ents = dp->d_ops->leaf_ents_p(leaf);
1095         firstlep = &ents[first];
1096         lastlep = &ents[last];
1097         xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1098                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1099 }
1100
1101 /*
1102  * Log the header of the leaf1 or leafn block.
1103  */
1104 void
1105 xfs_dir3_leaf_log_header(
1106         struct xfs_trans        *tp,
1107         struct xfs_inode        *dp,
1108         struct xfs_buf          *bp)
1109 {
1110         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1111
1112         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1113                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1114                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1115                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1116
1117         xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1118                           dp->d_ops->leaf_hdr_size - 1);
1119 }
1120
1121 /*
1122  * Log the tail of the leaf1 block.
1123  */
1124 STATIC void
1125 xfs_dir3_leaf_log_tail(
1126         struct xfs_trans        *tp,
1127         struct xfs_buf          *bp)
1128 {
1129         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1130         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1131         struct xfs_mount        *mp = tp->t_mountp;
1132
1133         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1134                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1135                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1136                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1137
1138         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1139         xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1140                 (uint)(mp->m_dirblksize - 1));
1141 }
1142
1143 /*
1144  * Look up the entry referred to by args in the leaf format directory.
1145  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1146  * is also used by the node-format code.
1147  */
1148 int
1149 xfs_dir2_leaf_lookup(
1150         xfs_da_args_t           *args)          /* operation arguments */
1151 {
1152         struct xfs_buf          *dbp;           /* data block buffer */
1153         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1154         xfs_inode_t             *dp;            /* incore directory inode */
1155         int                     error;          /* error return code */
1156         int                     index;          /* found entry index */
1157         struct xfs_buf          *lbp;           /* leaf buffer */
1158         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1159         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1160         xfs_trans_t             *tp;            /* transaction pointer */
1161         struct xfs_dir2_leaf_entry *ents;
1162
1163         trace_xfs_dir2_leaf_lookup(args);
1164
1165         /*
1166          * Look up name in the leaf block, returning both buffers and index.
1167          */
1168         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1169                 return error;
1170         }
1171         tp = args->trans;
1172         dp = args->dp;
1173         xfs_dir3_leaf_check(dp, lbp);
1174         leaf = lbp->b_addr;
1175         ents = dp->d_ops->leaf_ents_p(leaf);
1176         /*
1177          * Get to the leaf entry and contained data entry address.
1178          */
1179         lep = &ents[index];
1180
1181         /*
1182          * Point to the data entry.
1183          */
1184         dep = (xfs_dir2_data_entry_t *)
1185               ((char *)dbp->b_addr +
1186                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1187         /*
1188          * Return the found inode number & CI name if appropriate
1189          */
1190         args->inumber = be64_to_cpu(dep->inumber);
1191         args->filetype = dp->d_ops->data_get_ftype(dep);
1192         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1193         xfs_trans_brelse(tp, dbp);
1194         xfs_trans_brelse(tp, lbp);
1195         return XFS_ERROR(error);
1196 }
1197
1198 /*
1199  * Look up name/hash in the leaf block.
1200  * Fill in indexp with the found index, and dbpp with the data buffer.
1201  * If not found dbpp will be NULL, and ENOENT comes back.
1202  * lbpp will always be filled in with the leaf buffer unless there's an error.
1203  */
1204 static int                                      /* error */
1205 xfs_dir2_leaf_lookup_int(
1206         xfs_da_args_t           *args,          /* operation arguments */
1207         struct xfs_buf          **lbpp,         /* out: leaf buffer */
1208         int                     *indexp,        /* out: index in leaf block */
1209         struct xfs_buf          **dbpp)         /* out: data buffer */
1210 {
1211         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1212         struct xfs_buf          *dbp = NULL;    /* data buffer */
1213         xfs_dir2_data_entry_t   *dep;           /* data entry */
1214         xfs_inode_t             *dp;            /* incore directory inode */
1215         int                     error;          /* error return code */
1216         int                     index;          /* index in leaf block */
1217         struct xfs_buf          *lbp;           /* leaf buffer */
1218         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1219         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1220         xfs_mount_t             *mp;            /* filesystem mount point */
1221         xfs_dir2_db_t           newdb;          /* new data block number */
1222         xfs_trans_t             *tp;            /* transaction pointer */
1223         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1224         enum xfs_dacmp          cmp;            /* name compare result */
1225         struct xfs_dir2_leaf_entry *ents;
1226         struct xfs_dir3_icleaf_hdr leafhdr;
1227
1228         dp = args->dp;
1229         tp = args->trans;
1230         mp = dp->i_mount;
1231
1232         error = xfs_dir3_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
1233         if (error)
1234                 return error;
1235
1236         *lbpp = lbp;
1237         leaf = lbp->b_addr;
1238         xfs_dir3_leaf_check(dp, lbp);
1239         ents = dp->d_ops->leaf_ents_p(leaf);
1240         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1241
1242         /*
1243          * Look for the first leaf entry with our hash value.
1244          */
1245         index = xfs_dir2_leaf_search_hash(args, lbp);
1246         /*
1247          * Loop over all the entries with the right hash value
1248          * looking to match the name.
1249          */
1250         for (lep = &ents[index];
1251              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1252              lep++, index++) {
1253                 /*
1254                  * Skip over stale leaf entries.
1255                  */
1256                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1257                         continue;
1258                 /*
1259                  * Get the new data block number.
1260                  */
1261                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1262                 /*
1263                  * If it's not the same as the old data block number,
1264                  * need to pitch the old one and read the new one.
1265                  */
1266                 if (newdb != curdb) {
1267                         if (dbp)
1268                                 xfs_trans_brelse(tp, dbp);
1269                         error = xfs_dir3_data_read(tp, dp,
1270                                                    xfs_dir2_db_to_da(mp, newdb),
1271                                                    -1, &dbp);
1272                         if (error) {
1273                                 xfs_trans_brelse(tp, lbp);
1274                                 return error;
1275                         }
1276                         curdb = newdb;
1277                 }
1278                 /*
1279                  * Point to the data entry.
1280                  */
1281                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1282                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1283                 /*
1284                  * Compare name and if it's an exact match, return the index
1285                  * and buffer. If it's the first case-insensitive match, store
1286                  * the index and buffer and continue looking for an exact match.
1287                  */
1288                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1289                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1290                         args->cmpresult = cmp;
1291                         *indexp = index;
1292                         /* case exact match: return the current buffer. */
1293                         if (cmp == XFS_CMP_EXACT) {
1294                                 *dbpp = dbp;
1295                                 return 0;
1296                         }
1297                         cidb = curdb;
1298                 }
1299         }
1300         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1301         /*
1302          * Here, we can only be doing a lookup (not a rename or remove).
1303          * If a case-insensitive match was found earlier, re-read the
1304          * appropriate data block if required and return it.
1305          */
1306         if (args->cmpresult == XFS_CMP_CASE) {
1307                 ASSERT(cidb != -1);
1308                 if (cidb != curdb) {
1309                         xfs_trans_brelse(tp, dbp);
1310                         error = xfs_dir3_data_read(tp, dp,
1311                                                    xfs_dir2_db_to_da(mp, cidb),
1312                                                    -1, &dbp);
1313                         if (error) {
1314                                 xfs_trans_brelse(tp, lbp);
1315                                 return error;
1316                         }
1317                 }
1318                 *dbpp = dbp;
1319                 return 0;
1320         }
1321         /*
1322          * No match found, return ENOENT.
1323          */
1324         ASSERT(cidb == -1);
1325         if (dbp)
1326                 xfs_trans_brelse(tp, dbp);
1327         xfs_trans_brelse(tp, lbp);
1328         return XFS_ERROR(ENOENT);
1329 }
1330
1331 /*
1332  * Remove an entry from a leaf format directory.
1333  */
1334 int                                             /* error */
1335 xfs_dir2_leaf_removename(
1336         xfs_da_args_t           *args)          /* operation arguments */
1337 {
1338         __be16                  *bestsp;        /* leaf block best freespace */
1339         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1340         xfs_dir2_db_t           db;             /* data block number */
1341         struct xfs_buf          *dbp;           /* data block buffer */
1342         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1343         xfs_inode_t             *dp;            /* incore directory inode */
1344         int                     error;          /* error return code */
1345         xfs_dir2_db_t           i;              /* temporary data block # */
1346         int                     index;          /* index into leaf entries */
1347         struct xfs_buf          *lbp;           /* leaf buffer */
1348         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1349         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1350         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1351         xfs_mount_t             *mp;            /* filesystem mount point */
1352         int                     needlog;        /* need to log data header */
1353         int                     needscan;       /* need to rescan data frees */
1354         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1355         xfs_trans_t             *tp;            /* transaction pointer */
1356         struct xfs_dir2_data_free *bf;          /* bestfree table */
1357         struct xfs_dir2_leaf_entry *ents;
1358         struct xfs_dir3_icleaf_hdr leafhdr;
1359
1360         trace_xfs_dir2_leaf_removename(args);
1361
1362         /*
1363          * Lookup the leaf entry, get the leaf and data blocks read in.
1364          */
1365         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1366                 return error;
1367         }
1368         dp = args->dp;
1369         tp = args->trans;
1370         mp = dp->i_mount;
1371         leaf = lbp->b_addr;
1372         hdr = dbp->b_addr;
1373         xfs_dir3_data_check(dp, dbp);
1374         bf = dp->d_ops->data_bestfree_p(hdr);
1375         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1376         ents = dp->d_ops->leaf_ents_p(leaf);
1377         /*
1378          * Point to the leaf entry, use that to point to the data entry.
1379          */
1380         lep = &ents[index];
1381         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1382         dep = (xfs_dir2_data_entry_t *)
1383               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1384         needscan = needlog = 0;
1385         oldbest = be16_to_cpu(bf[0].length);
1386         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1387         bestsp = xfs_dir2_leaf_bests_p(ltp);
1388         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1389         /*
1390          * Mark the former data entry unused.
1391          */
1392         xfs_dir2_data_make_free(tp, dp, dbp,
1393                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1394                 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1395         /*
1396          * We just mark the leaf entry stale by putting a null in it.
1397          */
1398         leafhdr.stale++;
1399         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1400         xfs_dir3_leaf_log_header(tp, dp, lbp);
1401
1402         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1403         xfs_dir3_leaf_log_ents(tp, dp, lbp, index, index);
1404
1405         /*
1406          * Scan the freespace in the data block again if necessary,
1407          * log the data block header if necessary.
1408          */
1409         if (needscan)
1410                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1411         if (needlog)
1412                 xfs_dir2_data_log_header(tp, dp, dbp);
1413         /*
1414          * If the longest freespace in the data block has changed,
1415          * put the new value in the bests table and log that.
1416          */
1417         if (be16_to_cpu(bf[0].length) != oldbest) {
1418                 bestsp[db] = bf[0].length;
1419                 xfs_dir3_leaf_log_bests(tp, lbp, db, db);
1420         }
1421         xfs_dir3_data_check(dp, dbp);
1422         /*
1423          * If the data block is now empty then get rid of the data block.
1424          */
1425         if (be16_to_cpu(bf[0].length) ==
1426                         mp->m_dirblksize - dp->d_ops->data_entry_offset) {
1427                 ASSERT(db != mp->m_dirdatablk);
1428                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1429                         /*
1430                          * Nope, can't get rid of it because it caused
1431                          * allocation of a bmap btree block to do so.
1432                          * Just go on, returning success, leaving the
1433                          * empty block in place.
1434                          */
1435                         if (error == ENOSPC && args->total == 0)
1436                                 error = 0;
1437                         xfs_dir3_leaf_check(dp, lbp);
1438                         return error;
1439                 }
1440                 dbp = NULL;
1441                 /*
1442                  * If this is the last data block then compact the
1443                  * bests table by getting rid of entries.
1444                  */
1445                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1446                         /*
1447                          * Look for the last active entry (i).
1448                          */
1449                         for (i = db - 1; i > 0; i--) {
1450                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1451                                         break;
1452                         }
1453                         /*
1454                          * Copy the table down so inactive entries at the
1455                          * end are removed.
1456                          */
1457                         memmove(&bestsp[db - i], bestsp,
1458                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1459                         be32_add_cpu(&ltp->bestcount, -(db - i));
1460                         xfs_dir3_leaf_log_tail(tp, lbp);
1461                         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1462                 } else
1463                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1464         }
1465         /*
1466          * If the data block was not the first one, drop it.
1467          */
1468         else if (db != mp->m_dirdatablk)
1469                 dbp = NULL;
1470
1471         xfs_dir3_leaf_check(dp, lbp);
1472         /*
1473          * See if we can convert to block form.
1474          */
1475         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1476 }
1477
1478 /*
1479  * Replace the inode number in a leaf format directory entry.
1480  */
1481 int                                             /* error */
1482 xfs_dir2_leaf_replace(
1483         xfs_da_args_t           *args)          /* operation arguments */
1484 {
1485         struct xfs_buf          *dbp;           /* data block buffer */
1486         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1487         xfs_inode_t             *dp;            /* incore directory inode */
1488         int                     error;          /* error return code */
1489         int                     index;          /* index of leaf entry */
1490         struct xfs_buf          *lbp;           /* leaf buffer */
1491         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1492         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1493         xfs_trans_t             *tp;            /* transaction pointer */
1494         struct xfs_dir2_leaf_entry *ents;
1495
1496         trace_xfs_dir2_leaf_replace(args);
1497
1498         /*
1499          * Look up the entry.
1500          */
1501         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1502                 return error;
1503         }
1504         dp = args->dp;
1505         leaf = lbp->b_addr;
1506         ents = dp->d_ops->leaf_ents_p(leaf);
1507         /*
1508          * Point to the leaf entry, get data address from it.
1509          */
1510         lep = &ents[index];
1511         /*
1512          * Point to the data entry.
1513          */
1514         dep = (xfs_dir2_data_entry_t *)
1515               ((char *)dbp->b_addr +
1516                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1517         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1518         /*
1519          * Put the new inode number in, log it.
1520          */
1521         dep->inumber = cpu_to_be64(args->inumber);
1522         dp->d_ops->data_put_ftype(dep, args->filetype);
1523         tp = args->trans;
1524         xfs_dir2_data_log_entry(tp, dp, dbp, dep);
1525         xfs_dir3_leaf_check(dp, lbp);
1526         xfs_trans_brelse(tp, lbp);
1527         return 0;
1528 }
1529
1530 /*
1531  * Return index in the leaf block (lbp) which is either the first
1532  * one with this hash value, or if there are none, the insert point
1533  * for that hash value.
1534  */
1535 int                                             /* index value */
1536 xfs_dir2_leaf_search_hash(
1537         xfs_da_args_t           *args,          /* operation arguments */
1538         struct xfs_buf          *lbp)           /* leaf buffer */
1539 {
1540         xfs_dahash_t            hash=0;         /* hash from this entry */
1541         xfs_dahash_t            hashwant;       /* hash value looking for */
1542         int                     high;           /* high leaf index */
1543         int                     low;            /* low leaf index */
1544         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1545         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1546         int                     mid=0;          /* current leaf index */
1547         struct xfs_dir2_leaf_entry *ents;
1548         struct xfs_dir3_icleaf_hdr leafhdr;
1549
1550         leaf = lbp->b_addr;
1551         ents = args->dp->d_ops->leaf_ents_p(leaf);
1552         args->dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1553
1554         /*
1555          * Note, the table cannot be empty, so we have to go through the loop.
1556          * Binary search the leaf entries looking for our hash value.
1557          */
1558         for (lep = ents, low = 0, high = leafhdr.count - 1,
1559                 hashwant = args->hashval;
1560              low <= high; ) {
1561                 mid = (low + high) >> 1;
1562                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1563                         break;
1564                 if (hash < hashwant)
1565                         low = mid + 1;
1566                 else
1567                         high = mid - 1;
1568         }
1569         /*
1570          * Found one, back up through all the equal hash values.
1571          */
1572         if (hash == hashwant) {
1573                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1574                         mid--;
1575                 }
1576         }
1577         /*
1578          * Need to point to an entry higher than ours.
1579          */
1580         else if (hash < hashwant)
1581                 mid++;
1582         return mid;
1583 }
1584
1585 /*
1586  * Trim off a trailing data block.  We know it's empty since the leaf
1587  * freespace table says so.
1588  */
1589 int                                             /* error */
1590 xfs_dir2_leaf_trim_data(
1591         xfs_da_args_t           *args,          /* operation arguments */
1592         struct xfs_buf          *lbp,           /* leaf buffer */
1593         xfs_dir2_db_t           db)             /* data block number */
1594 {
1595         __be16                  *bestsp;        /* leaf bests table */
1596         struct xfs_buf          *dbp;           /* data block buffer */
1597         xfs_inode_t             *dp;            /* incore directory inode */
1598         int                     error;          /* error return value */
1599         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1600         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1601         xfs_mount_t             *mp;            /* filesystem mount point */
1602         xfs_trans_t             *tp;            /* transaction pointer */
1603
1604         dp = args->dp;
1605         mp = dp->i_mount;
1606         tp = args->trans;
1607         /*
1608          * Read the offending data block.  We need its buffer.
1609          */
1610         error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp);
1611         if (error)
1612                 return error;
1613
1614         leaf = lbp->b_addr;
1615         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1616
1617 #ifdef DEBUG
1618 {
1619         struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1620         struct xfs_dir2_data_free *bf = dp->d_ops->data_bestfree_p(hdr);
1621
1622         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1623                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1624         ASSERT(be16_to_cpu(bf[0].length) ==
1625                mp->m_dirblksize - dp->d_ops->data_entry_offset);
1626         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1627 }
1628 #endif
1629
1630         /*
1631          * Get rid of the data block.
1632          */
1633         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1634                 ASSERT(error != ENOSPC);
1635                 xfs_trans_brelse(tp, dbp);
1636                 return error;
1637         }
1638         /*
1639          * Eliminate the last bests entry from the table.
1640          */
1641         bestsp = xfs_dir2_leaf_bests_p(ltp);
1642         be32_add_cpu(&ltp->bestcount, -1);
1643         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1644         xfs_dir3_leaf_log_tail(tp, lbp);
1645         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1646         return 0;
1647 }
1648
1649 static inline size_t
1650 xfs_dir3_leaf_size(
1651         struct xfs_dir3_icleaf_hdr      *hdr,
1652         int                             counts)
1653 {
1654         int     entries;
1655         int     hdrsize;
1656
1657         entries = hdr->count - hdr->stale;
1658         if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1659             hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1660                 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1661         else
1662                 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1663
1664         return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1665                        + counts * sizeof(xfs_dir2_data_off_t)
1666                        + sizeof(xfs_dir2_leaf_tail_t);
1667 }
1668
1669 /*
1670  * Convert node form directory to leaf form directory.
1671  * The root of the node form dir needs to already be a LEAFN block.
1672  * Just return if we can't do anything.
1673  */
1674 int                                             /* error */
1675 xfs_dir2_node_to_leaf(
1676         xfs_da_state_t          *state)         /* directory operation state */
1677 {
1678         xfs_da_args_t           *args;          /* operation arguments */
1679         xfs_inode_t             *dp;            /* incore directory inode */
1680         int                     error;          /* error return code */
1681         struct xfs_buf          *fbp;           /* buffer for freespace block */
1682         xfs_fileoff_t           fo;             /* freespace file offset */
1683         xfs_dir2_free_t         *free;          /* freespace structure */
1684         struct xfs_buf          *lbp;           /* buffer for leaf block */
1685         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1686         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1687         xfs_mount_t             *mp;            /* filesystem mount point */
1688         int                     rval;           /* successful free trim? */
1689         xfs_trans_t             *tp;            /* transaction pointer */
1690         struct xfs_dir3_icleaf_hdr leafhdr;
1691         struct xfs_dir3_icfree_hdr freehdr;
1692
1693         /*
1694          * There's more than a leaf level in the btree, so there must
1695          * be multiple leafn blocks.  Give up.
1696          */
1697         if (state->path.active > 1)
1698                 return 0;
1699         args = state->args;
1700
1701         trace_xfs_dir2_node_to_leaf(args);
1702
1703         mp = state->mp;
1704         dp = args->dp;
1705         tp = args->trans;
1706         /*
1707          * Get the last offset in the file.
1708          */
1709         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1710                 return error;
1711         }
1712         fo -= mp->m_dirblkfsbs;
1713         /*
1714          * If there are freespace blocks other than the first one,
1715          * take this opportunity to remove trailing empty freespace blocks
1716          * that may have been left behind during no-space-reservation
1717          * operations.
1718          */
1719         while (fo > mp->m_dirfreeblk) {
1720                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1721                         return error;
1722                 }
1723                 if (rval)
1724                         fo -= mp->m_dirblkfsbs;
1725                 else
1726                         return 0;
1727         }
1728         /*
1729          * Now find the block just before the freespace block.
1730          */
1731         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1732                 return error;
1733         }
1734         /*
1735          * If it's not the single leaf block, give up.
1736          */
1737         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1738                 return 0;
1739         lbp = state->path.blk[0].bp;
1740         leaf = lbp->b_addr;
1741         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1742
1743         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1744                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1745
1746         /*
1747          * Read the freespace block.
1748          */
1749         error = xfs_dir2_free_read(tp, dp,  mp->m_dirfreeblk, &fbp);
1750         if (error)
1751                 return error;
1752         free = fbp->b_addr;
1753         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1754
1755         ASSERT(!freehdr.firstdb);
1756
1757         /*
1758          * Now see if the leafn and free data will fit in a leaf1.
1759          * If not, release the buffer and give up.
1760          */
1761         if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > mp->m_dirblksize) {
1762                 xfs_trans_brelse(tp, fbp);
1763                 return 0;
1764         }
1765
1766         /*
1767          * If the leaf has any stale entries in it, compress them out.
1768          */
1769         if (leafhdr.stale)
1770                 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1771
1772         lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1773         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1774         leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1775                                         ? XFS_DIR2_LEAF1_MAGIC
1776                                         : XFS_DIR3_LEAF1_MAGIC;
1777
1778         /*
1779          * Set up the leaf tail from the freespace block.
1780          */
1781         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1782         ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1783
1784         /*
1785          * Set up the leaf bests table.
1786          */
1787         memcpy(xfs_dir2_leaf_bests_p(ltp), dp->d_ops->free_bests_p(free),
1788                 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1789
1790         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1791         xfs_dir3_leaf_log_header(tp, dp, lbp);
1792         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1793         xfs_dir3_leaf_log_tail(tp, lbp);
1794         xfs_dir3_leaf_check(dp, lbp);
1795
1796         /*
1797          * Get rid of the freespace block.
1798          */
1799         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1800         if (error) {
1801                 /*
1802                  * This can't fail here because it can only happen when
1803                  * punching out the middle of an extent, and this is an
1804                  * isolated block.
1805                  */
1806                 ASSERT(error != ENOSPC);
1807                 return error;
1808         }
1809         fbp = NULL;
1810         /*
1811          * Now see if we can convert the single-leaf directory
1812          * down to a block form directory.
1813          * This routine always kills the dabuf for the leaf, so
1814          * eliminate it from the path.
1815          */
1816         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1817         state->path.blk[0].bp = NULL;
1818         return error;
1819 }