fs/xfs/libxfs/xfs_dir2_leaf.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   4  * Copyright (c) 2013 Red Hat, Inc.
   5  * All Rights Reserved.
   6  */
   7 #include "xfs.h"
   8 #include "xfs_fs.h"
   9 #include "xfs_shared.h"
  10 #include "xfs_format.h"
  11 #include "xfs_log_format.h"
  12 #include "xfs_trans_resv.h"
  13 #include "xfs_mount.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_bmap.h"
  16 #include "xfs_dir2.h"
  17 #include "xfs_dir2_priv.h"
  18 #include "xfs_error.h"
  19 #include "xfs_trace.h"
  20 #include "xfs_trans.h"
  21 #include "xfs_buf_item.h"
  22
  23 /*
  24  * Local function declarations.
  25  */
  26 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
  27                                     int *indexp, struct xfs_buf **dbpp,
  28                                     struct xfs_dir3_icleaf_hdr *leafhdr);
  29 static void xfs_dir3_leaf_log_bests(struct xfs_da_args *args,
  30                                     struct xfs_buf *bp, int first, int last);
  31 static void xfs_dir3_leaf_log_tail(struct xfs_da_args *args,
  32                                    struct xfs_buf *bp);
  33
  34 void
  35 xfs_dir2_leaf_hdr_from_disk(
  36         struct xfs_mount                *mp,
  37         struct xfs_dir3_icleaf_hdr      *to,
  38         struct xfs_dir2_leaf            *from)
  39 {
  40         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  41                 struct xfs_dir3_leaf *from3 = (struct xfs_dir3_leaf *)from;
  42
  43                 to->forw = be32_to_cpu(from3->hdr.info.hdr.forw);
  44                 to->back = be32_to_cpu(from3->hdr.info.hdr.back);
  45                 to->magic = be16_to_cpu(from3->hdr.info.hdr.magic);
  46                 to->count = be16_to_cpu(from3->hdr.count);
  47                 to->stale = be16_to_cpu(from3->hdr.stale);
  48                 to->ents = from3->__ents;
  49
  50                 ASSERT(to->magic == XFS_DIR3_LEAF1_MAGIC ||
  51                        to->magic == XFS_DIR3_LEAFN_MAGIC);
  52         } else {
  53                 to->forw = be32_to_cpu(from->hdr.info.forw);
  54                 to->back = be32_to_cpu(from->hdr.info.back);
  55                 to->magic = be16_to_cpu(from->hdr.info.magic);
  56                 to->count = be16_to_cpu(from->hdr.count);
  57                 to->stale = be16_to_cpu(from->hdr.stale);
  58                 to->ents = from->__ents;
  59
  60                 ASSERT(to->magic == XFS_DIR2_LEAF1_MAGIC ||
  61                        to->magic == XFS_DIR2_LEAFN_MAGIC);
  62         }
  63 }
  64
  65 void
  66 xfs_dir2_leaf_hdr_to_disk(
  67         struct xfs_mount                *mp,
  68         struct xfs_dir2_leaf            *to,
  69         struct xfs_dir3_icleaf_hdr      *from)
  70 {
  71         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  72                 struct xfs_dir3_leaf *to3 = (struct xfs_dir3_leaf *)to;
  73
  74                 ASSERT(from->magic == XFS_DIR3_LEAF1_MAGIC ||
  75                        from->magic == XFS_DIR3_LEAFN_MAGIC);
  76
  77                 to3->hdr.info.hdr.forw = cpu_to_be32(from->forw);
  78                 to3->hdr.info.hdr.back = cpu_to_be32(from->back);
  79                 to3->hdr.info.hdr.magic = cpu_to_be16(from->magic);
  80                 to3->hdr.count = cpu_to_be16(from->count);
  81                 to3->hdr.stale = cpu_to_be16(from->stale);
  82         } else {
  83                 ASSERT(from->magic == XFS_DIR2_LEAF1_MAGIC ||
  84                        from->magic == XFS_DIR2_LEAFN_MAGIC);
  85
  86                 to->hdr.info.forw = cpu_to_be32(from->forw);
  87                 to->hdr.info.back = cpu_to_be32(from->back);
  88                 to->hdr.info.magic = cpu_to_be16(from->magic);
  89                 to->hdr.count = cpu_to_be16(from->count);
  90                 to->hdr.stale = cpu_to_be16(from->stale);
  91         }
  92 }
  93
  94 /*
  95  * Check the internal consistency of a leaf1 block.
  96  * Pop an assert if something is wrong.
  97  */
  98 #ifdef DEBUG
  99 static xfs_failaddr_t
 100 xfs_dir3_leaf1_check(
 101         struct xfs_inode        *dp,
 102         struct xfs_buf          *bp)
 103 {
 104         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 105         struct xfs_dir3_icleaf_hdr leafhdr;
 106
 107         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
 108
 109         if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
 110                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 111                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
 112                         return __this_address;
 113         } else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
 114                 return __this_address;
 115
 116         return xfs_dir3_leaf_check_int(dp->i_mount, &leafhdr, leaf);
 117 }
 118
 119 static inline void
 120 xfs_dir3_leaf_check(
 121         struct xfs_inode        *dp,
 122         struct xfs_buf          *bp)
 123 {
 124         xfs_failaddr_t          fa;
 125
 126         fa = xfs_dir3_leaf1_check(dp, bp);
 127         if (!fa)
 128                 return;
 129         xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
 130                         bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
 131                         fa);
 132         ASSERT(0);
 133 }
 134 #else
 135 #define xfs_dir3_leaf_check(dp, bp)
 136 #endif
 137
 138 xfs_failaddr_t
 139 xfs_dir3_leaf_check_int(
 140         struct xfs_mount                *mp,
 141         struct xfs_dir3_icleaf_hdr      *hdr,
 142         struct xfs_dir2_leaf            *leaf)
 143 {
 144         struct xfs_da_geometry          *geo = mp->m_dir_geo;
 145         xfs_dir2_leaf_tail_t            *ltp;
 146         int                             stale;
 147         int                             i;
 148
 149         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 150
 151         /*
 152          * XXX (dgc): This value is not restrictive enough.
 153          * Should factor in the size of the bests table as well.
 154          * We can deduce a value for that from di_size.
 155          */
 156         if (hdr->count > geo->leaf_max_ents)
 157                 return __this_address;
 158
 159         /* Leaves and bests don't overlap in leaf format. */
 160         if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
 161              hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
 162             (char *)&hdr->ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
 163                 return __this_address;
 164
 165         /* Check hash value order, count stale entries.  */
 166         for (i = stale = 0; i < hdr->count; i++) {
 167                 if (i + 1 < hdr->count) {
 168                         if (be32_to_cpu(hdr->ents[i].hashval) >
 169                                         be32_to_cpu(hdr->ents[i + 1].hashval))
 170                                 return __this_address;
 171                 }
 172                 if (hdr->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 173                         stale++;
 174         }
 175         if (hdr->stale != stale)
 176                 return __this_address;
 177         return NULL;
 178 }
 179
 180 /*
 181  * We verify the magic numbers before decoding the leaf header so that on debug
 182  * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
 183  * to incorrect magic numbers.
 184  */
 185 static xfs_failaddr_t
 186 xfs_dir3_leaf_verify(
 187         struct xfs_buf                  *bp)
 188 {
 189         struct xfs_mount                *mp = bp->b_mount;
 190         struct xfs_dir3_icleaf_hdr      leafhdr;
 191         xfs_failaddr_t                  fa;
 192
 193         fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
 194         if (fa)
 195                 return fa;
 196
 197         xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, bp->b_addr);
 198         return xfs_dir3_leaf_check_int(mp, &leafhdr, bp->b_addr);
 199 }
 200
 201 static void
 202 xfs_dir3_leaf_read_verify(
 203         struct xfs_buf  *bp)
 204 {
 205         struct xfs_mount        *mp = bp->b_mount;
 206         xfs_failaddr_t          fa;
 207
 208         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 209              !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
 210                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
 211         else {
 212                 fa = xfs_dir3_leaf_verify(bp);
 213                 if (fa)
 214                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 215         }
 216 }
 217
 218 static void
 219 xfs_dir3_leaf_write_verify(
 220         struct xfs_buf  *bp)
 221 {
 222         struct xfs_mount        *mp = bp->b_mount;
 223         struct xfs_buf_log_item *bip = bp->b_log_item;
 224         struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
 225         xfs_failaddr_t          fa;
 226
 227         fa = xfs_dir3_leaf_verify(bp);
 228         if (fa) {
 229                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 230                 return;
 231         }
 232
 233         if (!xfs_sb_version_hascrc(&mp->m_sb))
 234                 return;
 235
 236         if (bip)
 237                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 238
 239         xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
 240 }
 241
 242 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
 243         .name = "xfs_dir3_leaf1",
 244         .magic16 = { cpu_to_be16(XFS_DIR2_LEAF1_MAGIC),
 245                      cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) },
 246         .verify_read = xfs_dir3_leaf_read_verify,
 247         .verify_write = xfs_dir3_leaf_write_verify,
 248         .verify_struct = xfs_dir3_leaf_verify,
 249 };
 250
 251 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
 252         .name = "xfs_dir3_leafn",
 253         .magic16 = { cpu_to_be16(XFS_DIR2_LEAFN_MAGIC),
 254                      cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) },
 255         .verify_read = xfs_dir3_leaf_read_verify,
 256         .verify_write = xfs_dir3_leaf_write_verify,
 257         .verify_struct = xfs_dir3_leaf_verify,
 258 };
 259
 260 int
 261 xfs_dir3_leaf_read(
 262         struct xfs_trans        *tp,
 263         struct xfs_inode        *dp,
 264         xfs_dablk_t             fbno,
 265         struct xfs_buf          **bpp)
 266 {
 267         int                     err;
 268
 269         err = xfs_da_read_buf(tp, dp, fbno, 0, bpp, XFS_DATA_FORK,
 270                         &xfs_dir3_leaf1_buf_ops);
 271         if (!err && tp && *bpp)
 272                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
 273         return err;
 274 }
 275
 276 int
 277 xfs_dir3_leafn_read(
 278         struct xfs_trans        *tp,
 279         struct xfs_inode        *dp,
 280         xfs_dablk_t             fbno,
 281         struct xfs_buf          **bpp)
 282 {
 283         int                     err;
 284
 285         err = xfs_da_read_buf(tp, dp, fbno, 0, bpp, XFS_DATA_FORK,
 286                         &xfs_dir3_leafn_buf_ops);
 287         if (!err && tp && *bpp)
 288                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
 289         return err;
 290 }
 291
 292 /*
 293  * Initialize a new leaf block, leaf1 or leafn magic accepted.
 294  */
 295 static void
 296 xfs_dir3_leaf_init(
 297         struct xfs_mount        *mp,
 298         struct xfs_trans        *tp,
 299         struct xfs_buf          *bp,
 300         xfs_ino_t               owner,
 301         uint16_t                type)
 302 {
 303         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 304
 305         ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
 306
 307         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 308                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 309
 310                 memset(leaf3, 0, sizeof(*leaf3));
 311
 312                 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
 313                                          ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
 314                                          : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 315                 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
 316                 leaf3->info.owner = cpu_to_be64(owner);
 317                 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid);
 318         } else {
 319                 memset(leaf, 0, sizeof(*leaf));
 320                 leaf->hdr.info.magic = cpu_to_be16(type);
 321         }
 322
 323         /*
 324          * If it's a leaf-format directory initialize the tail.
 325          * Caller is responsible for initialising the bests table.
 326          */
 327         if (type == XFS_DIR2_LEAF1_MAGIC) {
 328                 struct xfs_dir2_leaf_tail *ltp;
 329
 330                 ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
 331                 ltp->bestcount = 0;
 332                 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
 333                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
 334         } else {
 335                 bp->b_ops = &xfs_dir3_leafn_buf_ops;
 336                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
 337         }
 338 }
 339
 340 int
 341 xfs_dir3_leaf_get_buf(
 342         xfs_da_args_t           *args,
 343         xfs_dir2_db_t           bno,
 344         struct xfs_buf          **bpp,
 345         uint16_t                magic)
 346 {
 347         struct xfs_inode        *dp = args->dp;
 348         struct xfs_trans        *tp = args->trans;
 349         struct xfs_mount        *mp = dp->i_mount;
 350         struct xfs_buf          *bp;
 351         int                     error;
 352
 353         ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
 354         ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
 355                bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
 356
 357         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
 358                                &bp, XFS_DATA_FORK);
 359         if (error)
 360                 return error;
 361
 362         xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
 363         xfs_dir3_leaf_log_header(args, bp);
 364         if (magic == XFS_DIR2_LEAF1_MAGIC)
 365                 xfs_dir3_leaf_log_tail(args, bp);
 366         *bpp = bp;
 367         return 0;
 368 }
 369
 370 /*
 371  * Convert a block form directory to a leaf form directory.
 372  */
 373 int                                             /* error */
 374 xfs_dir2_block_to_leaf(
 375         xfs_da_args_t           *args,          /* operation arguments */
 376         struct xfs_buf          *dbp)           /* input block's buffer */
 377 {
 378         __be16                  *bestsp;        /* leaf's bestsp entries */
 379         xfs_dablk_t             blkno;          /* leaf block's bno */
 380         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 381         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
 382         xfs_dir2_block_tail_t   *btp;           /* block's tail */
 383         xfs_inode_t             *dp;            /* incore directory inode */
 384         int                     error;          /* error return code */
 385         struct xfs_buf          *lbp;           /* leaf block's buffer */
 386         xfs_dir2_db_t           ldb;            /* leaf block's bno */
 387         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 388         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
 389         int                     needlog;        /* need to log block header */
 390         int                     needscan;       /* need to rescan bestfree */
 391         xfs_trans_t             *tp;            /* transaction pointer */
 392         struct xfs_dir2_data_free *bf;
 393         struct xfs_dir3_icleaf_hdr leafhdr;
 394
 395         trace_xfs_dir2_block_to_leaf(args);
 396
 397         dp = args->dp;
 398         tp = args->trans;
 399         /*
 400          * Add the leaf block to the inode.
 401          * This interface will only put blocks in the leaf/node range.
 402          * Since that's empty now, we'll get the root (block 0 in range).
 403          */
 404         if ((error = xfs_da_grow_inode(args, &blkno))) {
 405                 return error;
 406         }
 407         ldb = xfs_dir2_da_to_db(args->geo, blkno);
 408         ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
 409         /*
 410          * Initialize the leaf block, get a buffer for it.
 411          */
 412         error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
 413         if (error)
 414                 return error;
 415
 416         leaf = lbp->b_addr;
 417         hdr = dbp->b_addr;
 418         xfs_dir3_data_check(dp, dbp);
 419         btp = xfs_dir2_block_tail_p(args->geo, hdr);
 420         blp = xfs_dir2_block_leaf_p(btp);
 421         bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
 422
 423         /*
 424          * Set the counts in the leaf header.
 425          */
 426         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
 427         leafhdr.count = be32_to_cpu(btp->count);
 428         leafhdr.stale = be32_to_cpu(btp->stale);
 429         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
 430         xfs_dir3_leaf_log_header(args, lbp);
 431
 432         /*
 433          * Could compact these but I think we always do the conversion
 434          * after squeezing out stale entries.
 435          */
 436         memcpy(leafhdr.ents, blp,
 437                 be32_to_cpu(btp->count) * sizeof(struct xfs_dir2_leaf_entry));
 438         xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, 0, leafhdr.count - 1);
 439         needscan = 0;
 440         needlog = 1;
 441         /*
 442          * Make the space formerly occupied by the leaf entries and block
 443          * tail be free.
 444          */
 445         xfs_dir2_data_make_free(args, dbp,
 446                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 447                 (xfs_dir2_data_aoff_t)((char *)hdr + args->geo->blksize -
 448                                        (char *)blp),
 449                 &needlog, &needscan);
 450         /*
 451          * Fix up the block header, make it a data block.
 452          */
 453         dbp->b_ops = &xfs_dir3_data_buf_ops;
 454         xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
 455         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
 456                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 457         else
 458                 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 459
 460         if (needscan)
 461                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
 462         /*
 463          * Set up leaf tail and bests table.
 464          */
 465         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 466         ltp->bestcount = cpu_to_be32(1);
 467         bestsp = xfs_dir2_leaf_bests_p(ltp);
 468         bestsp[0] =  bf[0].length;
 469         /*
 470          * Log the data header and leaf bests table.
 471          */
 472         if (needlog)
 473                 xfs_dir2_data_log_header(args, dbp);
 474         xfs_dir3_leaf_check(dp, lbp);
 475         xfs_dir3_data_check(dp, dbp);
 476         xfs_dir3_leaf_log_bests(args, lbp, 0, 0);
 477         return 0;
 478 }
 479
 480 STATIC void
 481 xfs_dir3_leaf_find_stale(
 482         struct xfs_dir3_icleaf_hdr *leafhdr,
 483         struct xfs_dir2_leaf_entry *ents,
 484         int                     index,
 485         int                     *lowstale,
 486         int                     *highstale)
 487 {
 488         /*
 489          * Find the first stale entry before our index, if any.
 490          */
 491         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
 492                 if (ents[*lowstale].address ==
 493                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 494                         break;
 495         }
 496
 497         /*
 498          * Find the first stale entry at or after our index, if any.
 499          * Stop if the result would require moving more entries than using
 500          * lowstale.
 501          */
 502         for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
 503                 if (ents[*highstale].address ==
 504                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 505                         break;
 506                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
 507                         break;
 508         }
 509 }
 510
 511 struct xfs_dir2_leaf_entry *
 512 xfs_dir3_leaf_find_entry(
 513         struct xfs_dir3_icleaf_hdr *leafhdr,
 514         struct xfs_dir2_leaf_entry *ents,
 515         int                     index,          /* leaf table position */
 516         int                     compact,        /* need to compact leaves */
 517         int                     lowstale,       /* index of prev stale leaf */
 518         int                     highstale,      /* index of next stale leaf */
 519         int                     *lfloglow,      /* low leaf logging index */
 520         int                     *lfloghigh)     /* high leaf logging index */
 521 {
 522         if (!leafhdr->stale) {
 523                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
 524
 525                 /*
 526                  * Now we need to make room to insert the leaf entry.
 527                  *
 528                  * If there are no stale entries, just insert a hole at index.
 529                  */
 530                 lep = &ents[index];
 531                 if (index < leafhdr->count)
 532                         memmove(lep + 1, lep,
 533                                 (leafhdr->count - index) * sizeof(*lep));
 534
 535                 /*
 536                  * Record low and high logging indices for the leaf.
 537                  */
 538                 *lfloglow = index;
 539                 *lfloghigh = leafhdr->count++;
 540                 return lep;
 541         }
 542
 543         /*
 544          * There are stale entries.
 545          *
 546          * We will use one of them for the new entry.  It's probably not at
 547          * the right location, so we'll have to shift some up or down first.
 548          *
 549          * If we didn't compact before, we need to find the nearest stale
 550          * entries before and after our insertion point.
 551          */
 552         if (compact == 0)
 553                 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
 554                                          &lowstale, &highstale);
 555
 556         /*
 557          * If the low one is better, use it.
 558          */
 559         if (lowstale >= 0 &&
 560             (highstale == leafhdr->count ||
 561              index - lowstale - 1 < highstale - index)) {
 562                 ASSERT(index - lowstale - 1 >= 0);
 563                 ASSERT(ents[lowstale].address ==
 564                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 565
 566                 /*
 567                  * Copy entries up to cover the stale entry and make room
 568                  * for the new entry.
 569                  */
 570                 if (index - lowstale - 1 > 0) {
 571                         memmove(&ents[lowstale], &ents[lowstale + 1],
 572                                 (index - lowstale - 1) *
 573                                         sizeof(xfs_dir2_leaf_entry_t));
 574                 }
 575                 *lfloglow = min(lowstale, *lfloglow);
 576                 *lfloghigh = max(index - 1, *lfloghigh);
 577                 leafhdr->stale--;
 578                 return &ents[index - 1];
 579         }
 580
 581         /*
 582          * The high one is better, so use that one.
 583          */
 584         ASSERT(highstale - index >= 0);
 585         ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 586
 587         /*
 588          * Copy entries down to cover the stale entry and make room for the
 589          * new entry.
 590          */
 591         if (highstale - index > 0) {
 592                 memmove(&ents[index + 1], &ents[index],
 593                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
 594         }
 595         *lfloglow = min(index, *lfloglow);
 596         *lfloghigh = max(highstale, *lfloghigh);
 597         leafhdr->stale--;
 598         return &ents[index];
 599 }
 600
 601 /*
 602  * Add an entry to a leaf form directory.
 603  */
 604 int                                             /* error */
 605 xfs_dir2_leaf_addname(
 606         struct xfs_da_args      *args)          /* operation arguments */
 607 {
 608         struct xfs_dir3_icleaf_hdr leafhdr;
 609         struct xfs_trans        *tp = args->trans;
 610         __be16                  *bestsp;        /* freespace table in leaf */
 611         __be16                  *tagp;          /* end of data entry */
 612         struct xfs_buf          *dbp;           /* data block buffer */
 613         struct xfs_buf          *lbp;           /* leaf's buffer */
 614         struct xfs_dir2_leaf    *leaf;          /* leaf structure */
 615         struct xfs_inode        *dp = args->dp; /* incore directory inode */
 616         struct xfs_dir2_data_hdr *hdr;          /* data block header */
 617         struct xfs_dir2_data_entry *dep;        /* data block entry */
 618         struct xfs_dir2_leaf_entry *lep;        /* leaf entry table pointer */
 619         struct xfs_dir2_leaf_entry *ents;
 620         struct xfs_dir2_data_unused *dup;       /* data unused entry */
 621         struct xfs_dir2_leaf_tail *ltp;         /* leaf tail pointer */
 622         struct xfs_dir2_data_free *bf;          /* bestfree table */
 623         int                     compact;        /* need to compact leaves */
 624         int                     error;          /* error return value */
 625         int                     grown;          /* allocated new data block */
 626         int                     highstale = 0;  /* index of next stale leaf */
 627         int                     i;              /* temporary, index */
 628         int                     index;          /* leaf table position */
 629         int                     length;         /* length of new entry */
 630         int                     lfloglow;       /* low leaf logging index */
 631         int                     lfloghigh;      /* high leaf logging index */
 632         int                     lowstale = 0;   /* index of prev stale leaf */
 633         int                     needbytes;      /* leaf block bytes needed */
 634         int                     needlog;        /* need to log data header */
 635         int                     needscan;       /* need to rescan data free */
 636         xfs_dir2_db_t           use_block;      /* data block number */
 637
 638         trace_xfs_dir2_leaf_addname(args);
 639
 640         error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, &lbp);
 641         if (error)
 642                 return error;
 643
 644         /*
 645          * Look up the entry by hash value and name.
 646          * We know it's not there, our caller has already done a lookup.
 647          * So the index is of the entry to insert in front of.
 648          * But if there are dup hash values the index is of the first of those.
 649          */
 650         index = xfs_dir2_leaf_search_hash(args, lbp);
 651         leaf = lbp->b_addr;
 652         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 653         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
 654         ents = leafhdr.ents;
 655         bestsp = xfs_dir2_leaf_bests_p(ltp);
 656         length = xfs_dir2_data_entsize(dp->i_mount, args->namelen);
 657
 658         /*
 659          * See if there are any entries with the same hash value
 660          * and space in their block for the new entry.
 661          * This is good because it puts multiple same-hash value entries
 662          * in a data block, improving the lookup of those entries.
 663          */
 664         for (use_block = -1, lep = &ents[index];
 665              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 666              index++, lep++) {
 667                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 668                         continue;
 669                 i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
 670                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 671                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
 672                 if (be16_to_cpu(bestsp[i]) >= length) {
 673                         use_block = i;
 674                         break;
 675                 }
 676         }
 677         /*
 678          * Didn't find a block yet, linear search all the data blocks.
 679          */
 680         if (use_block == -1) {
 681                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 682                         /*
 683                          * Remember a block we see that's missing.
 684                          */
 685                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
 686                             use_block == -1)
 687                                 use_block = i;
 688                         else if (be16_to_cpu(bestsp[i]) >= length) {
 689                                 use_block = i;
 690                                 break;
 691                         }
 692                 }
 693         }
 694         /*
 695          * How many bytes do we need in the leaf block?
 696          */
 697         needbytes = 0;
 698         if (!leafhdr.stale)
 699                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
 700         if (use_block == -1)
 701                 needbytes += sizeof(xfs_dir2_data_off_t);
 702
 703         /*
 704          * Now kill use_block if it refers to a missing block, so we
 705          * can use it as an indication of allocation needed.
 706          */
 707         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
 708                 use_block = -1;
 709         /*
 710          * If we don't have enough free bytes but we can make enough
 711          * by compacting out stale entries, we'll do that.
 712          */
 713         if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
 714             leafhdr.stale > 1)
 715                 compact = 1;
 716
 717         /*
 718          * Otherwise if we don't have enough free bytes we need to
 719          * convert to node form.
 720          */
 721         else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
 722                 /*
 723                  * Just checking or no space reservation, give up.
 724                  */
 725                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 726                                                         args->total == 0) {
 727                         xfs_trans_brelse(tp, lbp);
 728                         return -ENOSPC;
 729                 }
 730                 /*
 731                  * Convert to node form.
 732                  */
 733                 error = xfs_dir2_leaf_to_node(args, lbp);
 734                 if (error)
 735                         return error;
 736                 /*
 737                  * Then add the new entry.
 738                  */
 739                 return xfs_dir2_node_addname(args);
 740         }
 741         /*
 742          * Otherwise it will fit without compaction.
 743          */
 744         else
 745                 compact = 0;
 746         /*
 747          * If just checking, then it will fit unless we needed to allocate
 748          * a new data block.
 749          */
 750         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 751                 xfs_trans_brelse(tp, lbp);
 752                 return use_block == -1 ? -ENOSPC : 0;
 753         }
 754         /*
 755          * If no allocations are allowed, return now before we've
 756          * changed anything.
 757          */
 758         if (args->total == 0 && use_block == -1) {
 759                 xfs_trans_brelse(tp, lbp);
 760                 return -ENOSPC;
 761         }
 762         /*
 763          * Need to compact the leaf entries, removing stale ones.
 764          * Leave one stale entry behind - the one closest to our
 765          * insertion index - and we'll shift that one to our insertion
 766          * point later.
 767          */
 768         if (compact) {
 769                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 770                         &highstale, &lfloglow, &lfloghigh);
 771         }
 772         /*
 773          * There are stale entries, so we'll need log-low and log-high
 774          * impossibly bad values later.
 775          */
 776         else if (leafhdr.stale) {
 777                 lfloglow = leafhdr.count;
 778                 lfloghigh = -1;
 779         }
 780         /*
 781          * If there was no data block space found, we need to allocate
 782          * a new one.
 783          */
 784         if (use_block == -1) {
 785                 /*
 786                  * Add the new data block.
 787                  */
 788                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 789                                 &use_block))) {
 790                         xfs_trans_brelse(tp, lbp);
 791                         return error;
 792                 }
 793                 /*
 794                  * Initialize the block.
 795                  */
 796                 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
 797                         xfs_trans_brelse(tp, lbp);
 798                         return error;
 799                 }
 800                 /*
 801                  * If we're adding a new data block on the end we need to
 802                  * extend the bests table.  Copy it up one entry.
 803                  */
 804                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
 805                         bestsp--;
 806                         memmove(&bestsp[0], &bestsp[1],
 807                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 808                         be32_add_cpu(&ltp->bestcount, 1);
 809                         xfs_dir3_leaf_log_tail(args, lbp);
 810                         xfs_dir3_leaf_log_bests(args, lbp, 0,
 811                                                 be32_to_cpu(ltp->bestcount) - 1);
 812                 }
 813                 /*
 814                  * If we're filling in a previously empty block just log it.
 815                  */
 816                 else
 817                         xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
 818                 hdr = dbp->b_addr;
 819                 bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
 820                 bestsp[use_block] = bf[0].length;
 821                 grown = 1;
 822         } else {
 823                 /*
 824                  * Already had space in some data block.
 825                  * Just read that one in.
 826                  */
 827                 error = xfs_dir3_data_read(tp, dp,
 828                                    xfs_dir2_db_to_da(args->geo, use_block),
 829                                    0, &dbp);
 830                 if (error) {
 831                         xfs_trans_brelse(tp, lbp);
 832                         return error;
 833                 }
 834                 hdr = dbp->b_addr;
 835                 bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
 836                 grown = 0;
 837         }
 838         /*
 839          * Point to the biggest freespace in our data block.
 840          */
 841         dup = (xfs_dir2_data_unused_t *)
 842               ((char *)hdr + be16_to_cpu(bf[0].offset));
 843         needscan = needlog = 0;
 844         /*
 845          * Mark the initial part of our freespace in use for the new entry.
 846          */
 847         error = xfs_dir2_data_use_free(args, dbp, dup,
 848                         (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
 849                         length, &needlog, &needscan);
 850         if (error) {
 851                 xfs_trans_brelse(tp, lbp);
 852                 return error;
 853         }
 854         /*
 855          * Initialize our new entry (at last).
 856          */
 857         dep = (xfs_dir2_data_entry_t *)dup;
 858         dep->inumber = cpu_to_be64(args->inumber);
 859         dep->namelen = args->namelen;
 860         memcpy(dep->name, args->name, dep->namelen);
 861         xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
 862         tagp = xfs_dir2_data_entry_tag_p(dp->i_mount, dep);
 863         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 864         /*
 865          * Need to scan fix up the bestfree table.
 866          */
 867         if (needscan)
 868                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
 869         /*
 870          * Need to log the data block's header.
 871          */
 872         if (needlog)
 873                 xfs_dir2_data_log_header(args, dbp);
 874         xfs_dir2_data_log_entry(args, dbp, dep);
 875         /*
 876          * If the bests table needs to be changed, do it.
 877          * Log the change unless we've already done that.
 878          */
 879         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
 880                 bestsp[use_block] = bf[0].length;
 881                 if (!grown)
 882                         xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
 883         }
 884
 885         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 886                                        highstale, &lfloglow, &lfloghigh);
 887
 888         /*
 889          * Fill in the new leaf entry.
 890          */
 891         lep->hashval = cpu_to_be32(args->hashval);
 892         lep->address = cpu_to_be32(
 893                                 xfs_dir2_db_off_to_dataptr(args->geo, use_block,
 894                                 be16_to_cpu(*tagp)));
 895         /*
 896          * Log the leaf fields and give up the buffers.
 897          */
 898         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
 899         xfs_dir3_leaf_log_header(args, lbp);
 900         xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, lfloglow, lfloghigh);
 901         xfs_dir3_leaf_check(dp, lbp);
 902         xfs_dir3_data_check(dp, dbp);
 903         return 0;
 904 }
 905
 906 /*
 907  * Compact out any stale entries in the leaf.
 908  * Log the header and changed leaf entries, if any.
 909  */
 910 void
 911 xfs_dir3_leaf_compact(
 912         xfs_da_args_t   *args,          /* operation arguments */
 913         struct xfs_dir3_icleaf_hdr *leafhdr,
 914         struct xfs_buf  *bp)            /* leaf buffer */
 915 {
 916         int             from;           /* source leaf index */
 917         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 918         int             loglow;         /* first leaf entry to log */
 919         int             to;             /* target leaf index */
 920         struct xfs_inode *dp = args->dp;
 921
 922         leaf = bp->b_addr;
 923         if (!leafhdr->stale)
 924                 return;
 925
 926         /*
 927          * Compress out the stale entries in place.
 928          */
 929         for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
 930                 if (leafhdr->ents[from].address ==
 931                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 932                         continue;
 933                 /*
 934                  * Only actually copy the entries that are different.
 935                  */
 936                 if (from > to) {
 937                         if (loglow == -1)
 938                                 loglow = to;
 939                         leafhdr->ents[to] = leafhdr->ents[from];
 940                 }
 941                 to++;
 942         }
 943         /*
 944          * Update and log the header, log the leaf entries.
 945          */
 946         ASSERT(leafhdr->stale == from - to);
 947         leafhdr->count -= leafhdr->stale;
 948         leafhdr->stale = 0;
 949
 950         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, leafhdr);
 951         xfs_dir3_leaf_log_header(args, bp);
 952         if (loglow != -1)
 953                 xfs_dir3_leaf_log_ents(args, leafhdr, bp, loglow, to - 1);
 954 }
 955
 956 /*
 957  * Compact the leaf entries, removing stale ones.
 958  * Leave one stale entry behind - the one closest to our
 959  * insertion index - and the caller will shift that one to our insertion
 960  * point later.
 961  * Return new insertion index, where the remaining stale entry is,
 962  * and leaf logging indices.
 963  */
 964 void
 965 xfs_dir3_leaf_compact_x1(
 966         struct xfs_dir3_icleaf_hdr *leafhdr,
 967         struct xfs_dir2_leaf_entry *ents,
 968         int             *indexp,        /* insertion index */
 969         int             *lowstalep,     /* out: stale entry before us */
 970         int             *highstalep,    /* out: stale entry after us */
 971         int             *lowlogp,       /* out: low log index */
 972         int             *highlogp)      /* out: high log index */
 973 {
 974         int             from;           /* source copy index */
 975         int             highstale;      /* stale entry at/after index */
 976         int             index;          /* insertion index */
 977         int             keepstale;      /* source index of kept stale */
 978         int             lowstale;       /* stale entry before index */
 979         int             newindex=0;     /* new insertion index */
 980         int             to;             /* destination copy index */
 981
 982         ASSERT(leafhdr->stale > 1);
 983         index = *indexp;
 984
 985         xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
 986
 987         /*
 988          * Pick the better of lowstale and highstale.
 989          */
 990         if (lowstale >= 0 &&
 991             (highstale == leafhdr->count ||
 992              index - lowstale <= highstale - index))
 993                 keepstale = lowstale;
 994         else
 995                 keepstale = highstale;
 996         /*
 997          * Copy the entries in place, removing all the stale entries
 998          * except keepstale.
 999          */
1000         for (from = to = 0; from < leafhdr->count; from++) {
1001                 /*
1002                  * Notice the new value of index.
1003                  */
1004                 if (index == from)
1005                         newindex = to;
1006                 if (from != keepstale &&
1007                     ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1008                         if (from == to)
1009                                 *lowlogp = to;
1010                         continue;
1011                 }
1012                 /*
1013                  * Record the new keepstale value for the insertion.
1014                  */
1015                 if (from == keepstale)
1016                         lowstale = highstale = to;
1017                 /*
1018                  * Copy only the entries that have moved.
1019                  */
1020                 if (from > to)
1021                         ents[to] = ents[from];
1022                 to++;
1023         }
1024         ASSERT(from > to);
1025         /*
1026          * If the insertion point was past the last entry,
1027          * set the new insertion point accordingly.
1028          */
1029         if (index == from)
1030                 newindex = to;
1031         *indexp = newindex;
1032         /*
1033          * Adjust the leaf header values.
1034          */
1035         leafhdr->count -= from - to;
1036         leafhdr->stale = 1;
1037         /*
1038          * Remember the low/high stale value only in the "right"
1039          * direction.
1040          */
1041         if (lowstale >= newindex)
1042                 lowstale = -1;
1043         else
1044                 highstale = leafhdr->count;
1045         *highlogp = leafhdr->count - 1;
1046         *lowstalep = lowstale;
1047         *highstalep = highstale;
1048 }
1049
1050 /*
1051  * Log the bests entries indicated from a leaf1 block.
1052  */
1053 static void
1054 xfs_dir3_leaf_log_bests(
1055         struct xfs_da_args      *args,
1056         struct xfs_buf          *bp,            /* leaf buffer */
1057         int                     first,          /* first entry to log */
1058         int                     last)           /* last entry to log */
1059 {
1060         __be16                  *firstb;        /* pointer to first entry */
1061         __be16                  *lastb;         /* pointer to last entry */
1062         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1063         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1064
1065         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1066                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1067
1068         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1069         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1070         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1071         xfs_trans_log_buf(args->trans, bp,
1072                 (uint)((char *)firstb - (char *)leaf),
1073                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1074 }
1075
1076 /*
1077  * Log the leaf entries indicated from a leaf1 or leafn block.
1078  */
1079 void
1080 xfs_dir3_leaf_log_ents(
1081         struct xfs_da_args      *args,
1082         struct xfs_dir3_icleaf_hdr *hdr,
1083         struct xfs_buf          *bp,
1084         int                     first,
1085         int                     last)
1086 {
1087         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1088         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1089         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1090
1091         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1092                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1093                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1094                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1095
1096         firstlep = &hdr->ents[first];
1097         lastlep = &hdr->ents[last];
1098         xfs_trans_log_buf(args->trans, bp,
1099                 (uint)((char *)firstlep - (char *)leaf),
1100                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1101 }
1102
1103 /*
1104  * Log the header of the leaf1 or leafn block.
1105  */
1106 void
1107 xfs_dir3_leaf_log_header(
1108         struct xfs_da_args      *args,
1109         struct xfs_buf          *bp)
1110 {
1111         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1112
1113         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1114                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1115                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1116                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1117
1118         xfs_trans_log_buf(args->trans, bp,
1119                           (uint)((char *)&leaf->hdr - (char *)leaf),
1120                           args->geo->leaf_hdr_size - 1);
1121 }
1122
1123 /*
1124  * Log the tail of the leaf1 block.
1125  */
1126 STATIC void
1127 xfs_dir3_leaf_log_tail(
1128         struct xfs_da_args      *args,
1129         struct xfs_buf          *bp)
1130 {
1131         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1132         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1133
1134         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1135                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1136                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1137                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1138
1139         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1140         xfs_trans_log_buf(args->trans, bp, (uint)((char *)ltp - (char *)leaf),
1141                 (uint)(args->geo->blksize - 1));
1142 }
1143
1144 /*
1145  * Look up the entry referred to by args in the leaf format directory.
1146  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1147  * is also used by the node-format code.
1148  */
1149 int
1150 xfs_dir2_leaf_lookup(
1151         xfs_da_args_t           *args)          /* operation arguments */
1152 {
1153         struct xfs_buf          *dbp;           /* data block buffer */
1154         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1155         xfs_inode_t             *dp;            /* incore directory inode */
1156         int                     error;          /* error return code */
1157         int                     index;          /* found entry index */
1158         struct xfs_buf          *lbp;           /* leaf buffer */
1159         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1160         xfs_trans_t             *tp;            /* transaction pointer */
1161         struct xfs_dir3_icleaf_hdr leafhdr;
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         error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1169         if (error)
1170                 return error;
1171
1172         tp = args->trans;
1173         dp = args->dp;
1174         xfs_dir3_leaf_check(dp, lbp);
1175
1176         /*
1177          * Get to the leaf entry and contained data entry address.
1178          */
1179         lep = &leafhdr.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(args->geo, 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 = xfs_dir2_data_get_ftype(dp->i_mount, 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 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         struct xfs_dir3_icleaf_hdr *leafhdr)
1211 {
1212         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1213         struct xfs_buf          *dbp = NULL;    /* data buffer */
1214         xfs_dir2_data_entry_t   *dep;           /* data entry */
1215         xfs_inode_t             *dp;            /* incore directory inode */
1216         int                     error;          /* error return code */
1217         int                     index;          /* index in leaf block */
1218         struct xfs_buf          *lbp;           /* leaf buffer */
1219         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1220         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1221         xfs_mount_t             *mp;            /* filesystem mount point */
1222         xfs_dir2_db_t           newdb;          /* new data block number */
1223         xfs_trans_t             *tp;            /* transaction pointer */
1224         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1225         enum xfs_dacmp          cmp;            /* name compare result */
1226
1227         dp = args->dp;
1228         tp = args->trans;
1229         mp = dp->i_mount;
1230
1231         error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, &lbp);
1232         if (error)
1233                 return error;
1234
1235         *lbpp = lbp;
1236         leaf = lbp->b_addr;
1237         xfs_dir3_leaf_check(dp, lbp);
1238         xfs_dir2_leaf_hdr_from_disk(mp, leafhdr, leaf);
1239
1240         /*
1241          * Look for the first leaf entry with our hash value.
1242          */
1243         index = xfs_dir2_leaf_search_hash(args, lbp);
1244         /*
1245          * Loop over all the entries with the right hash value
1246          * looking to match the name.
1247          */
1248         for (lep = &leafhdr->ents[index];
1249              index < leafhdr->count &&
1250                         be32_to_cpu(lep->hashval) == args->hashval;
1251              lep++, index++) {
1252                 /*
1253                  * Skip over stale leaf entries.
1254                  */
1255                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1256                         continue;
1257                 /*
1258                  * Get the new data block number.
1259                  */
1260                 newdb = xfs_dir2_dataptr_to_db(args->geo,
1261                                                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(args->geo, newdb),
1271                                            0, &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(args->geo,
1283                                                 be32_to_cpu(lep->address)));
1284                 /*
1285                  * Compare name and if it's an exact match, return the index
1286                  * and buffer. If it's the first case-insensitive match, store
1287                  * the index and buffer and continue looking for an exact match.
1288                  */
1289                 cmp = xfs_dir2_compname(args, dep->name, dep->namelen);
1290                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1291                         args->cmpresult = cmp;
1292                         *indexp = index;
1293                         /* case exact match: return the current buffer. */
1294                         if (cmp == XFS_CMP_EXACT) {
1295                                 *dbpp = dbp;
1296                                 return 0;
1297                         }
1298                         cidb = curdb;
1299                 }
1300         }
1301         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1302         /*
1303          * Here, we can only be doing a lookup (not a rename or remove).
1304          * If a case-insensitive match was found earlier, re-read the
1305          * appropriate data block if required and return it.
1306          */
1307         if (args->cmpresult == XFS_CMP_CASE) {
1308                 ASSERT(cidb != -1);
1309                 if (cidb != curdb) {
1310                         xfs_trans_brelse(tp, dbp);
1311                         error = xfs_dir3_data_read(tp, dp,
1312                                            xfs_dir2_db_to_da(args->geo, cidb),
1313                                            0, &dbp);
1314                         if (error) {
1315                                 xfs_trans_brelse(tp, lbp);
1316                                 return error;
1317                         }
1318                 }
1319                 *dbpp = dbp;
1320                 return 0;
1321         }
1322         /*
1323          * No match found, return -ENOENT.
1324          */
1325         ASSERT(cidb == -1);
1326         if (dbp)
1327                 xfs_trans_brelse(tp, dbp);
1328         xfs_trans_brelse(tp, lbp);
1329         return -ENOENT;
1330 }
1331
1332 /*
1333  * Remove an entry from a leaf format directory.
1334  */
1335 int                                             /* error */
1336 xfs_dir2_leaf_removename(
1337         xfs_da_args_t           *args)          /* operation arguments */
1338 {
1339         struct xfs_da_geometry  *geo = args->geo;
1340         __be16                  *bestsp;        /* leaf block best freespace */
1341         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1342         xfs_dir2_db_t           db;             /* data block number */
1343         struct xfs_buf          *dbp;           /* data block buffer */
1344         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1345         xfs_inode_t             *dp;            /* incore directory inode */
1346         int                     error;          /* error return code */
1347         xfs_dir2_db_t           i;              /* temporary data block # */
1348         int                     index;          /* index into leaf entries */
1349         struct xfs_buf          *lbp;           /* leaf buffer */
1350         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1351         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1352         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1353         int                     needlog;        /* need to log data header */
1354         int                     needscan;       /* need to rescan data frees */
1355         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1356         struct xfs_dir2_data_free *bf;          /* bestfree table */
1357         struct xfs_dir3_icleaf_hdr leafhdr;
1358
1359         trace_xfs_dir2_leaf_removename(args);
1360
1361         /*
1362          * Lookup the leaf entry, get the leaf and data blocks read in.
1363          */
1364         error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1365         if (error)
1366                 return error;
1367
1368         dp = args->dp;
1369         leaf = lbp->b_addr;
1370         hdr = dbp->b_addr;
1371         xfs_dir3_data_check(dp, dbp);
1372         bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1373
1374         /*
1375          * Point to the leaf entry, use that to point to the data entry.
1376          */
1377         lep = &leafhdr.ents[index];
1378         db = xfs_dir2_dataptr_to_db(geo, be32_to_cpu(lep->address));
1379         dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1380                 xfs_dir2_dataptr_to_off(geo, be32_to_cpu(lep->address)));
1381         needscan = needlog = 0;
1382         oldbest = be16_to_cpu(bf[0].length);
1383         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
1384         bestsp = xfs_dir2_leaf_bests_p(ltp);
1385         if (be16_to_cpu(bestsp[db]) != oldbest) {
1386                 xfs_buf_mark_corrupt(lbp);
1387                 return -EFSCORRUPTED;
1388         }
1389         /*
1390          * Mark the former data entry unused.
1391          */
1392         xfs_dir2_data_make_free(args, dbp,
1393                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1394                 xfs_dir2_data_entsize(dp->i_mount, dep->namelen), &needlog,
1395                 &needscan);
1396         /*
1397          * We just mark the leaf entry stale by putting a null in it.
1398          */
1399         leafhdr.stale++;
1400         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
1401         xfs_dir3_leaf_log_header(args, lbp);
1402
1403         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1404         xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, index, index);
1405
1406         /*
1407          * Scan the freespace in the data block again if necessary,
1408          * log the data block header if necessary.
1409          */
1410         if (needscan)
1411                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
1412         if (needlog)
1413                 xfs_dir2_data_log_header(args, dbp);
1414         /*
1415          * If the longest freespace in the data block has changed,
1416          * put the new value in the bests table and log that.
1417          */
1418         if (be16_to_cpu(bf[0].length) != oldbest) {
1419                 bestsp[db] = bf[0].length;
1420                 xfs_dir3_leaf_log_bests(args, lbp, db, db);
1421         }
1422         xfs_dir3_data_check(dp, dbp);
1423         /*
1424          * If the data block is now empty then get rid of the data block.
1425          */
1426         if (be16_to_cpu(bf[0].length) ==
1427             geo->blksize - geo->data_entry_offset) {
1428                 ASSERT(db != geo->datablk);
1429                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1430                         /*
1431                          * Nope, can't get rid of it because it caused
1432                          * allocation of a bmap btree block to do so.
1433                          * Just go on, returning success, leaving the
1434                          * empty block in place.
1435                          */
1436                         if (error == -ENOSPC && args->total == 0)
1437                                 error = 0;
1438                         xfs_dir3_leaf_check(dp, lbp);
1439                         return error;
1440                 }
1441                 dbp = NULL;
1442                 /*
1443                  * If this is the last data block then compact the
1444                  * bests table by getting rid of entries.
1445                  */
1446                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1447                         /*
1448                          * Look for the last active entry (i).
1449                          */
1450                         for (i = db - 1; i > 0; i--) {
1451                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1452                                         break;
1453                         }
1454                         /*
1455                          * Copy the table down so inactive entries at the
1456                          * end are removed.
1457                          */
1458                         memmove(&bestsp[db - i], bestsp,
1459                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1460                         be32_add_cpu(&ltp->bestcount, -(db - i));
1461                         xfs_dir3_leaf_log_tail(args, lbp);
1462                         xfs_dir3_leaf_log_bests(args, lbp, 0,
1463                                                 be32_to_cpu(ltp->bestcount) - 1);
1464                 } else
1465                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1466         }
1467         /*
1468          * If the data block was not the first one, drop it.
1469          */
1470         else if (db != geo->datablk)
1471                 dbp = NULL;
1472
1473         xfs_dir3_leaf_check(dp, lbp);
1474         /*
1475          * See if we can convert to block form.
1476          */
1477         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1478 }
1479
1480 /*
1481  * Replace the inode number in a leaf format directory entry.
1482  */
1483 int                                             /* error */
1484 xfs_dir2_leaf_replace(
1485         xfs_da_args_t           *args)          /* operation arguments */
1486 {
1487         struct xfs_buf          *dbp;           /* data block buffer */
1488         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1489         xfs_inode_t             *dp;            /* incore directory inode */
1490         int                     error;          /* error return code */
1491         int                     index;          /* index of leaf entry */
1492         struct xfs_buf          *lbp;           /* leaf buffer */
1493         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1494         xfs_trans_t             *tp;            /* transaction pointer */
1495         struct xfs_dir3_icleaf_hdr leafhdr;
1496
1497         trace_xfs_dir2_leaf_replace(args);
1498
1499         /*
1500          * Look up the entry.
1501          */
1502         error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1503         if (error)
1504                 return error;
1505
1506         dp = args->dp;
1507         /*
1508          * Point to the leaf entry, get data address from it.
1509          */
1510         lep = &leafhdr.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(args->geo, 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         xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
1523         tp = args->trans;
1524         xfs_dir2_data_log_entry(args, 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_entry_t   *lep;           /* leaf entry */
1545         int                     mid=0;          /* current leaf index */
1546         struct xfs_dir3_icleaf_hdr leafhdr;
1547
1548         xfs_dir2_leaf_hdr_from_disk(args->dp->i_mount, &leafhdr, lbp->b_addr);
1549
1550         /*
1551          * Note, the table cannot be empty, so we have to go through the loop.
1552          * Binary search the leaf entries looking for our hash value.
1553          */
1554         for (lep = leafhdr.ents, low = 0, high = leafhdr.count - 1,
1555                 hashwant = args->hashval;
1556              low <= high; ) {
1557                 mid = (low + high) >> 1;
1558                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1559                         break;
1560                 if (hash < hashwant)
1561                         low = mid + 1;
1562                 else
1563                         high = mid - 1;
1564         }
1565         /*
1566          * Found one, back up through all the equal hash values.
1567          */
1568         if (hash == hashwant) {
1569                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1570                         mid--;
1571                 }
1572         }
1573         /*
1574          * Need to point to an entry higher than ours.
1575          */
1576         else if (hash < hashwant)
1577                 mid++;
1578         return mid;
1579 }
1580
1581 /*
1582  * Trim off a trailing data block.  We know it's empty since the leaf
1583  * freespace table says so.
1584  */
1585 int                                             /* error */
1586 xfs_dir2_leaf_trim_data(
1587         xfs_da_args_t           *args,          /* operation arguments */
1588         struct xfs_buf          *lbp,           /* leaf buffer */
1589         xfs_dir2_db_t           db)             /* data block number */
1590 {
1591         struct xfs_da_geometry  *geo = args->geo;
1592         __be16                  *bestsp;        /* leaf bests table */
1593         struct xfs_buf          *dbp;           /* data block buffer */
1594         xfs_inode_t             *dp;            /* incore directory inode */
1595         int                     error;          /* error return value */
1596         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1597         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1598         xfs_trans_t             *tp;            /* transaction pointer */
1599
1600         dp = args->dp;
1601         tp = args->trans;
1602         /*
1603          * Read the offending data block.  We need its buffer.
1604          */
1605         error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(geo, db), 0, &dbp);
1606         if (error)
1607                 return error;
1608
1609         leaf = lbp->b_addr;
1610         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
1611
1612 #ifdef DEBUG
1613 {
1614         struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1615         struct xfs_dir2_data_free *bf =
1616                 xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1617
1618         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1619                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1620         ASSERT(be16_to_cpu(bf[0].length) ==
1621                geo->blksize - geo->data_entry_offset);
1622         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1623 }
1624 #endif
1625
1626         /*
1627          * Get rid of the data block.
1628          */
1629         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1630                 ASSERT(error != -ENOSPC);
1631                 xfs_trans_brelse(tp, dbp);
1632                 return error;
1633         }
1634         /*
1635          * Eliminate the last bests entry from the table.
1636          */
1637         bestsp = xfs_dir2_leaf_bests_p(ltp);
1638         be32_add_cpu(&ltp->bestcount, -1);
1639         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1640         xfs_dir3_leaf_log_tail(args, lbp);
1641         xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1642         return 0;
1643 }
1644
1645 static inline size_t
1646 xfs_dir3_leaf_size(
1647         struct xfs_dir3_icleaf_hdr      *hdr,
1648         int                             counts)
1649 {
1650         int     entries;
1651         int     hdrsize;
1652
1653         entries = hdr->count - hdr->stale;
1654         if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1655             hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1656                 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1657         else
1658                 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1659
1660         return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1661                        + counts * sizeof(xfs_dir2_data_off_t)
1662                        + sizeof(xfs_dir2_leaf_tail_t);
1663 }
1664
1665 /*
1666  * Convert node form directory to leaf form directory.
1667  * The root of the node form dir needs to already be a LEAFN block.
1668  * Just return if we can't do anything.
1669  */
1670 int                                             /* error */
1671 xfs_dir2_node_to_leaf(
1672         xfs_da_state_t          *state)         /* directory operation state */
1673 {
1674         xfs_da_args_t           *args;          /* operation arguments */
1675         xfs_inode_t             *dp;            /* incore directory inode */
1676         int                     error;          /* error return code */
1677         struct xfs_buf          *fbp;           /* buffer for freespace block */
1678         xfs_fileoff_t           fo;             /* freespace file offset */
1679         struct xfs_buf          *lbp;           /* buffer for leaf block */
1680         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1681         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1682         xfs_mount_t             *mp;            /* filesystem mount point */
1683         int                     rval;           /* successful free trim? */
1684         xfs_trans_t             *tp;            /* transaction pointer */
1685         struct xfs_dir3_icleaf_hdr leafhdr;
1686         struct xfs_dir3_icfree_hdr freehdr;
1687
1688         /*
1689          * There's more than a leaf level in the btree, so there must
1690          * be multiple leafn blocks.  Give up.
1691          */
1692         if (state->path.active > 1)
1693                 return 0;
1694         args = state->args;
1695
1696         trace_xfs_dir2_node_to_leaf(args);
1697
1698         mp = state->mp;
1699         dp = args->dp;
1700         tp = args->trans;
1701         /*
1702          * Get the last offset in the file.
1703          */
1704         if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1705                 return error;
1706         }
1707         fo -= args->geo->fsbcount;
1708         /*
1709          * If there are freespace blocks other than the first one,
1710          * take this opportunity to remove trailing empty freespace blocks
1711          * that may have been left behind during no-space-reservation
1712          * operations.
1713          */
1714         while (fo > args->geo->freeblk) {
1715                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1716                         return error;
1717                 }
1718                 if (rval)
1719                         fo -= args->geo->fsbcount;
1720                 else
1721                         return 0;
1722         }
1723         /*
1724          * Now find the block just before the freespace block.
1725          */
1726         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1727                 return error;
1728         }
1729         /*
1730          * If it's not the single leaf block, give up.
1731          */
1732         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1733                 return 0;
1734         lbp = state->path.blk[0].bp;
1735         leaf = lbp->b_addr;
1736         xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, leaf);
1737
1738         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1739                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1740
1741         /*
1742          * Read the freespace block.
1743          */
1744         error = xfs_dir2_free_read(tp, dp,  args->geo->freeblk, &fbp);
1745         if (error)
1746                 return error;
1747         xfs_dir2_free_hdr_from_disk(mp, &freehdr, fbp->b_addr);
1748
1749         ASSERT(!freehdr.firstdb);
1750
1751         /*
1752          * Now see if the leafn and free data will fit in a leaf1.
1753          * If not, release the buffer and give up.
1754          */
1755         if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > args->geo->blksize) {
1756                 xfs_trans_brelse(tp, fbp);
1757                 return 0;
1758         }
1759
1760         /*
1761          * If the leaf has any stale entries in it, compress them out.
1762          */
1763         if (leafhdr.stale)
1764                 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1765
1766         lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1767         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1768         leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1769                                         ? XFS_DIR2_LEAF1_MAGIC
1770                                         : XFS_DIR3_LEAF1_MAGIC;
1771
1772         /*
1773          * Set up the leaf tail from the freespace block.
1774          */
1775         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1776         ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1777
1778         /*
1779          * Set up the leaf bests table.
1780          */
1781         memcpy(xfs_dir2_leaf_bests_p(ltp), freehdr.bests,
1782                 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1783
1784         xfs_dir2_leaf_hdr_to_disk(mp, leaf, &leafhdr);
1785         xfs_dir3_leaf_log_header(args, lbp);
1786         xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1787         xfs_dir3_leaf_log_tail(args, lbp);
1788         xfs_dir3_leaf_check(dp, lbp);
1789
1790         /*
1791          * Get rid of the freespace block.
1792          */
1793         error = xfs_dir2_shrink_inode(args,
1794                         xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET),
1795                         fbp);
1796         if (error) {
1797                 /*
1798                  * This can't fail here because it can only happen when
1799                  * punching out the middle of an extent, and this is an
1800                  * isolated block.
1801                  */
1802                 ASSERT(error != -ENOSPC);
1803                 return error;
1804         }
1805         fbp = NULL;
1806         /*
1807          * Now see if we can convert the single-leaf directory
1808          * down to a block form directory.
1809          * This routine always kills the dabuf for the leaf, so
1810          * eliminate it from the path.
1811          */
1812         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1813         state->path.blk[0].bp = NULL;
1814         return error;
1815 }