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, false);
 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         bool                            expensive_checking)
 144 {
 145         struct xfs_da_geometry          *geo = mp->m_dir_geo;
 146         xfs_dir2_leaf_tail_t            *ltp;
 147         int                             stale;
 148         int                             i;
 149
 150         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
 151
 152         /*
 153          * XXX (dgc): This value is not restrictive enough.
 154          * Should factor in the size of the bests table as well.
 155          * We can deduce a value for that from i_disk_size.
 156          */
 157         if (hdr->count > geo->leaf_max_ents)
 158                 return __this_address;
 159
 160         /* Leaves and bests don't overlap in leaf format. */
 161         if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
 162              hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
 163             (char *)&hdr->ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
 164                 return __this_address;
 165
 166         if (!expensive_checking)
 167                 return NULL;
 168
 169         /* Check hash value order, count stale entries.  */
 170         for (i = stale = 0; i < hdr->count; i++) {
 171                 if (i + 1 < hdr->count) {
 172                         if (be32_to_cpu(hdr->ents[i].hashval) >
 173                                         be32_to_cpu(hdr->ents[i + 1].hashval))
 174                                 return __this_address;
 175                 }
 176                 if (hdr->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 177                         stale++;
 178         }
 179         if (hdr->stale != stale)
 180                 return __this_address;
 181         return NULL;
 182 }
 183
 184 /*
 185  * We verify the magic numbers before decoding the leaf header so that on debug
 186  * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
 187  * to incorrect magic numbers.
 188  */
 189 static xfs_failaddr_t
 190 xfs_dir3_leaf_verify(
 191         struct xfs_buf                  *bp)
 192 {
 193         struct xfs_mount                *mp = bp->b_mount;
 194         struct xfs_dir3_icleaf_hdr      leafhdr;
 195         xfs_failaddr_t                  fa;
 196
 197         fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
 198         if (fa)
 199                 return fa;
 200
 201         xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, bp->b_addr);
 202         return xfs_dir3_leaf_check_int(mp, &leafhdr, bp->b_addr, true);
 203 }
 204
 205 static void
 206 xfs_dir3_leaf_read_verify(
 207         struct xfs_buf  *bp)
 208 {
 209         struct xfs_mount        *mp = bp->b_mount;
 210         xfs_failaddr_t          fa;
 211
 212         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 213              !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
 214                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
 215         else {
 216                 fa = xfs_dir3_leaf_verify(bp);
 217                 if (fa)
 218                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 219         }
 220 }
 221
 222 static void
 223 xfs_dir3_leaf_write_verify(
 224         struct xfs_buf  *bp)
 225 {
 226         struct xfs_mount        *mp = bp->b_mount;
 227         struct xfs_buf_log_item *bip = bp->b_log_item;
 228         struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
 229         xfs_failaddr_t          fa;
 230
 231         fa = xfs_dir3_leaf_verify(bp);
 232         if (fa) {
 233                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 234                 return;
 235         }
 236
 237         if (!xfs_sb_version_hascrc(&mp->m_sb))
 238                 return;
 239
 240         if (bip)
 241                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 242
 243         xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
 244 }
 245
 246 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
 247         .name = "xfs_dir3_leaf1",
 248         .magic16 = { cpu_to_be16(XFS_DIR2_LEAF1_MAGIC),
 249                      cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) },
 250         .verify_read = xfs_dir3_leaf_read_verify,
 251         .verify_write = xfs_dir3_leaf_write_verify,
 252         .verify_struct = xfs_dir3_leaf_verify,
 253 };
 254
 255 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
 256         .name = "xfs_dir3_leafn",
 257         .magic16 = { cpu_to_be16(XFS_DIR2_LEAFN_MAGIC),
 258                      cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) },
 259         .verify_read = xfs_dir3_leaf_read_verify,
 260         .verify_write = xfs_dir3_leaf_write_verify,
 261         .verify_struct = xfs_dir3_leaf_verify,
 262 };
 263
 264 int
 265 xfs_dir3_leaf_read(
 266         struct xfs_trans        *tp,
 267         struct xfs_inode        *dp,
 268         xfs_dablk_t             fbno,
 269         struct xfs_buf          **bpp)
 270 {
 271         int                     err;
 272
 273         err = xfs_da_read_buf(tp, dp, fbno, 0, bpp, XFS_DATA_FORK,
 274                         &xfs_dir3_leaf1_buf_ops);
 275         if (!err && tp && *bpp)
 276                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
 277         return err;
 278 }
 279
 280 int
 281 xfs_dir3_leafn_read(
 282         struct xfs_trans        *tp,
 283         struct xfs_inode        *dp,
 284         xfs_dablk_t             fbno,
 285         struct xfs_buf          **bpp)
 286 {
 287         int                     err;
 288
 289         err = xfs_da_read_buf(tp, dp, fbno, 0, bpp, XFS_DATA_FORK,
 290                         &xfs_dir3_leafn_buf_ops);
 291         if (!err && tp && *bpp)
 292                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
 293         return err;
 294 }
 295
 296 /*
 297  * Initialize a new leaf block, leaf1 or leafn magic accepted.
 298  */
 299 static void
 300 xfs_dir3_leaf_init(
 301         struct xfs_mount        *mp,
 302         struct xfs_trans        *tp,
 303         struct xfs_buf          *bp,
 304         xfs_ino_t               owner,
 305         uint16_t                type)
 306 {
 307         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 308
 309         ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
 310
 311         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 312                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 313
 314                 memset(leaf3, 0, sizeof(*leaf3));
 315
 316                 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
 317                                          ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
 318                                          : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 319                 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
 320                 leaf3->info.owner = cpu_to_be64(owner);
 321                 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid);
 322         } else {
 323                 memset(leaf, 0, sizeof(*leaf));
 324                 leaf->hdr.info.magic = cpu_to_be16(type);
 325         }
 326
 327         /*
 328          * If it's a leaf-format directory initialize the tail.
 329          * Caller is responsible for initialising the bests table.
 330          */
 331         if (type == XFS_DIR2_LEAF1_MAGIC) {
 332                 struct xfs_dir2_leaf_tail *ltp;
 333
 334                 ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
 335                 ltp->bestcount = 0;
 336                 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
 337                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
 338         } else {
 339                 bp->b_ops = &xfs_dir3_leafn_buf_ops;
 340                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
 341         }
 342 }
 343
 344 int
 345 xfs_dir3_leaf_get_buf(
 346         xfs_da_args_t           *args,
 347         xfs_dir2_db_t           bno,
 348         struct xfs_buf          **bpp,
 349         uint16_t                magic)
 350 {
 351         struct xfs_inode        *dp = args->dp;
 352         struct xfs_trans        *tp = args->trans;
 353         struct xfs_mount        *mp = dp->i_mount;
 354         struct xfs_buf          *bp;
 355         int                     error;
 356
 357         ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
 358         ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
 359                bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
 360
 361         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
 362                                &bp, XFS_DATA_FORK);
 363         if (error)
 364                 return error;
 365
 366         xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
 367         xfs_dir3_leaf_log_header(args, bp);
 368         if (magic == XFS_DIR2_LEAF1_MAGIC)
 369                 xfs_dir3_leaf_log_tail(args, bp);
 370         *bpp = bp;
 371         return 0;
 372 }
 373
 374 /*
 375  * Convert a block form directory to a leaf form directory.
 376  */
 377 int                                             /* error */
 378 xfs_dir2_block_to_leaf(
 379         xfs_da_args_t           *args,          /* operation arguments */
 380         struct xfs_buf          *dbp)           /* input block's buffer */
 381 {
 382         __be16                  *bestsp;        /* leaf's bestsp entries */
 383         xfs_dablk_t             blkno;          /* leaf block's bno */
 384         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 385         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
 386         xfs_dir2_block_tail_t   *btp;           /* block's tail */
 387         xfs_inode_t             *dp;            /* incore directory inode */
 388         int                     error;          /* error return code */
 389         struct xfs_buf          *lbp;           /* leaf block's buffer */
 390         xfs_dir2_db_t           ldb;            /* leaf block's bno */
 391         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 392         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
 393         int                     needlog;        /* need to log block header */
 394         int                     needscan;       /* need to rescan bestfree */
 395         xfs_trans_t             *tp;            /* transaction pointer */
 396         struct xfs_dir2_data_free *bf;
 397         struct xfs_dir3_icleaf_hdr leafhdr;
 398
 399         trace_xfs_dir2_block_to_leaf(args);
 400
 401         dp = args->dp;
 402         tp = args->trans;
 403         /*
 404          * Add the leaf block to the inode.
 405          * This interface will only put blocks in the leaf/node range.
 406          * Since that's empty now, we'll get the root (block 0 in range).
 407          */
 408         if ((error = xfs_da_grow_inode(args, &blkno))) {
 409                 return error;
 410         }
 411         ldb = xfs_dir2_da_to_db(args->geo, blkno);
 412         ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
 413         /*
 414          * Initialize the leaf block, get a buffer for it.
 415          */
 416         error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
 417         if (error)
 418                 return error;
 419
 420         leaf = lbp->b_addr;
 421         hdr = dbp->b_addr;
 422         xfs_dir3_data_check(dp, dbp);
 423         btp = xfs_dir2_block_tail_p(args->geo, hdr);
 424         blp = xfs_dir2_block_leaf_p(btp);
 425         bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
 426
 427         /*
 428          * Set the counts in the leaf header.
 429          */
 430         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
 431         leafhdr.count = be32_to_cpu(btp->count);
 432         leafhdr.stale = be32_to_cpu(btp->stale);
 433         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
 434         xfs_dir3_leaf_log_header(args, lbp);
 435
 436         /*
 437          * Could compact these but I think we always do the conversion
 438          * after squeezing out stale entries.
 439          */
 440         memcpy(leafhdr.ents, blp,
 441                 be32_to_cpu(btp->count) * sizeof(struct xfs_dir2_leaf_entry));
 442         xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, 0, leafhdr.count - 1);
 443         needscan = 0;
 444         needlog = 1;
 445         /*
 446          * Make the space formerly occupied by the leaf entries and block
 447          * tail be free.
 448          */
 449         xfs_dir2_data_make_free(args, dbp,
 450                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 451                 (xfs_dir2_data_aoff_t)((char *)hdr + args->geo->blksize -
 452                                        (char *)blp),
 453                 &needlog, &needscan);
 454         /*
 455          * Fix up the block header, make it a data block.
 456          */
 457         dbp->b_ops = &xfs_dir3_data_buf_ops;
 458         xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
 459         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
 460                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 461         else
 462                 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 463
 464         if (needscan)
 465                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
 466         /*
 467          * Set up leaf tail and bests table.
 468          */
 469         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 470         ltp->bestcount = cpu_to_be32(1);
 471         bestsp = xfs_dir2_leaf_bests_p(ltp);
 472         bestsp[0] =  bf[0].length;
 473         /*
 474          * Log the data header and leaf bests table.
 475          */
 476         if (needlog)
 477                 xfs_dir2_data_log_header(args, dbp);
 478         xfs_dir3_leaf_check(dp, lbp);
 479         xfs_dir3_data_check(dp, dbp);
 480         xfs_dir3_leaf_log_bests(args, lbp, 0, 0);
 481         return 0;
 482 }
 483
 484 STATIC void
 485 xfs_dir3_leaf_find_stale(
 486         struct xfs_dir3_icleaf_hdr *leafhdr,
 487         struct xfs_dir2_leaf_entry *ents,
 488         int                     index,
 489         int                     *lowstale,
 490         int                     *highstale)
 491 {
 492         /*
 493          * Find the first stale entry before our index, if any.
 494          */
 495         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
 496                 if (ents[*lowstale].address ==
 497                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 498                         break;
 499         }
 500
 501         /*
 502          * Find the first stale entry at or after our index, if any.
 503          * Stop if the result would require moving more entries than using
 504          * lowstale.
 505          */
 506         for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
 507                 if (ents[*highstale].address ==
 508                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 509                         break;
 510                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
 511                         break;
 512         }
 513 }
 514
 515 struct xfs_dir2_leaf_entry *
 516 xfs_dir3_leaf_find_entry(
 517         struct xfs_dir3_icleaf_hdr *leafhdr,
 518         struct xfs_dir2_leaf_entry *ents,
 519         int                     index,          /* leaf table position */
 520         int                     compact,        /* need to compact leaves */
 521         int                     lowstale,       /* index of prev stale leaf */
 522         int                     highstale,      /* index of next stale leaf */
 523         int                     *lfloglow,      /* low leaf logging index */
 524         int                     *lfloghigh)     /* high leaf logging index */
 525 {
 526         if (!leafhdr->stale) {
 527                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
 528
 529                 /*
 530                  * Now we need to make room to insert the leaf entry.
 531                  *
 532                  * If there are no stale entries, just insert a hole at index.
 533                  */
 534                 lep = &ents[index];
 535                 if (index < leafhdr->count)
 536                         memmove(lep + 1, lep,
 537                                 (leafhdr->count - index) * sizeof(*lep));
 538
 539                 /*
 540                  * Record low and high logging indices for the leaf.
 541                  */
 542                 *lfloglow = index;
 543                 *lfloghigh = leafhdr->count++;
 544                 return lep;
 545         }
 546
 547         /*
 548          * There are stale entries.
 549          *
 550          * We will use one of them for the new entry.  It's probably not at
 551          * the right location, so we'll have to shift some up or down first.
 552          *
 553          * If we didn't compact before, we need to find the nearest stale
 554          * entries before and after our insertion point.
 555          */
 556         if (compact == 0)
 557                 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
 558                                          &lowstale, &highstale);
 559
 560         /*
 561          * If the low one is better, use it.
 562          */
 563         if (lowstale >= 0 &&
 564             (highstale == leafhdr->count ||
 565              index - lowstale - 1 < highstale - index)) {
 566                 ASSERT(index - lowstale - 1 >= 0);
 567                 ASSERT(ents[lowstale].address ==
 568                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 569
 570                 /*
 571                  * Copy entries up to cover the stale entry and make room
 572                  * for the new entry.
 573                  */
 574                 if (index - lowstale - 1 > 0) {
 575                         memmove(&ents[lowstale], &ents[lowstale + 1],
 576                                 (index - lowstale - 1) *
 577                                         sizeof(xfs_dir2_leaf_entry_t));
 578                 }
 579                 *lfloglow = min(lowstale, *lfloglow);
 580                 *lfloghigh = max(index - 1, *lfloghigh);
 581                 leafhdr->stale--;
 582                 return &ents[index - 1];
 583         }
 584
 585         /*
 586          * The high one is better, so use that one.
 587          */
 588         ASSERT(highstale - index >= 0);
 589         ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 590
 591         /*
 592          * Copy entries down to cover the stale entry and make room for the
 593          * new entry.
 594          */
 595         if (highstale - index > 0) {
 596                 memmove(&ents[index + 1], &ents[index],
 597                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
 598         }
 599         *lfloglow = min(index, *lfloglow);
 600         *lfloghigh = max(highstale, *lfloghigh);
 601         leafhdr->stale--;
 602         return &ents[index];
 603 }
 604
 605 /*
 606  * Add an entry to a leaf form directory.
 607  */
 608 int                                             /* error */
 609 xfs_dir2_leaf_addname(
 610         struct xfs_da_args      *args)          /* operation arguments */
 611 {
 612         struct xfs_dir3_icleaf_hdr leafhdr;
 613         struct xfs_trans        *tp = args->trans;
 614         __be16                  *bestsp;        /* freespace table in leaf */
 615         __be16                  *tagp;          /* end of data entry */
 616         struct xfs_buf          *dbp;           /* data block buffer */
 617         struct xfs_buf          *lbp;           /* leaf's buffer */
 618         struct xfs_dir2_leaf    *leaf;          /* leaf structure */
 619         struct xfs_inode        *dp = args->dp; /* incore directory inode */
 620         struct xfs_dir2_data_hdr *hdr;          /* data block header */
 621         struct xfs_dir2_data_entry *dep;        /* data block entry */
 622         struct xfs_dir2_leaf_entry *lep;        /* leaf entry table pointer */
 623         struct xfs_dir2_leaf_entry *ents;
 624         struct xfs_dir2_data_unused *dup;       /* data unused entry */
 625         struct xfs_dir2_leaf_tail *ltp;         /* leaf tail pointer */
 626         struct xfs_dir2_data_free *bf;          /* bestfree table */
 627         int                     compact;        /* need to compact leaves */
 628         int                     error;          /* error return value */
 629         int                     grown;          /* allocated new data block */
 630         int                     highstale = 0;  /* index of next stale leaf */
 631         int                     i;              /* temporary, index */
 632         int                     index;          /* leaf table position */
 633         int                     length;         /* length of new entry */
 634         int                     lfloglow;       /* low leaf logging index */
 635         int                     lfloghigh;      /* high leaf logging index */
 636         int                     lowstale = 0;   /* index of prev stale leaf */
 637         int                     needbytes;      /* leaf block bytes needed */
 638         int                     needlog;        /* need to log data header */
 639         int                     needscan;       /* need to rescan data free */
 640         xfs_dir2_db_t           use_block;      /* data block number */
 641
 642         trace_xfs_dir2_leaf_addname(args);
 643
 644         error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, &lbp);
 645         if (error)
 646                 return error;
 647
 648         /*
 649          * Look up the entry by hash value and name.
 650          * We know it's not there, our caller has already done a lookup.
 651          * So the index is of the entry to insert in front of.
 652          * But if there are dup hash values the index is of the first of those.
 653          */
 654         index = xfs_dir2_leaf_search_hash(args, lbp);
 655         leaf = lbp->b_addr;
 656         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 657         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
 658         ents = leafhdr.ents;
 659         bestsp = xfs_dir2_leaf_bests_p(ltp);
 660         length = xfs_dir2_data_entsize(dp->i_mount, args->namelen);
 661
 662         /*
 663          * See if there are any entries with the same hash value
 664          * and space in their block for the new entry.
 665          * This is good because it puts multiple same-hash value entries
 666          * in a data block, improving the lookup of those entries.
 667          */
 668         for (use_block = -1, lep = &ents[index];
 669              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 670              index++, lep++) {
 671                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 672                         continue;
 673                 i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
 674                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 675                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
 676                 if (be16_to_cpu(bestsp[i]) >= length) {
 677                         use_block = i;
 678                         break;
 679                 }
 680         }
 681         /*
 682          * Didn't find a block yet, linear search all the data blocks.
 683          */
 684         if (use_block == -1) {
 685                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 686                         /*
 687                          * Remember a block we see that's missing.
 688                          */
 689                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
 690                             use_block == -1)
 691                                 use_block = i;
 692                         else if (be16_to_cpu(bestsp[i]) >= length) {
 693                                 use_block = i;
 694                                 break;
 695                         }
 696                 }
 697         }
 698         /*
 699          * How many bytes do we need in the leaf block?
 700          */
 701         needbytes = 0;
 702         if (!leafhdr.stale)
 703                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
 704         if (use_block == -1)
 705                 needbytes += sizeof(xfs_dir2_data_off_t);
 706
 707         /*
 708          * Now kill use_block if it refers to a missing block, so we
 709          * can use it as an indication of allocation needed.
 710          */
 711         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
 712                 use_block = -1;
 713         /*
 714          * If we don't have enough free bytes but we can make enough
 715          * by compacting out stale entries, we'll do that.
 716          */
 717         if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
 718             leafhdr.stale > 1)
 719                 compact = 1;
 720
 721         /*
 722          * Otherwise if we don't have enough free bytes we need to
 723          * convert to node form.
 724          */
 725         else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
 726                 /*
 727                  * Just checking or no space reservation, give up.
 728                  */
 729                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 730                                                         args->total == 0) {
 731                         xfs_trans_brelse(tp, lbp);
 732                         return -ENOSPC;
 733                 }
 734                 /*
 735                  * Convert to node form.
 736                  */
 737                 error = xfs_dir2_leaf_to_node(args, lbp);
 738                 if (error)
 739                         return error;
 740                 /*
 741                  * Then add the new entry.
 742                  */
 743                 return xfs_dir2_node_addname(args);
 744         }
 745         /*
 746          * Otherwise it will fit without compaction.
 747          */
 748         else
 749                 compact = 0;
 750         /*
 751          * If just checking, then it will fit unless we needed to allocate
 752          * a new data block.
 753          */
 754         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 755                 xfs_trans_brelse(tp, lbp);
 756                 return use_block == -1 ? -ENOSPC : 0;
 757         }
 758         /*
 759          * If no allocations are allowed, return now before we've
 760          * changed anything.
 761          */
 762         if (args->total == 0 && use_block == -1) {
 763                 xfs_trans_brelse(tp, lbp);
 764                 return -ENOSPC;
 765         }
 766         /*
 767          * Need to compact the leaf entries, removing stale ones.
 768          * Leave one stale entry behind - the one closest to our
 769          * insertion index - and we'll shift that one to our insertion
 770          * point later.
 771          */
 772         if (compact) {
 773                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 774                         &highstale, &lfloglow, &lfloghigh);
 775         }
 776         /*
 777          * There are stale entries, so we'll need log-low and log-high
 778          * impossibly bad values later.
 779          */
 780         else if (leafhdr.stale) {
 781                 lfloglow = leafhdr.count;
 782                 lfloghigh = -1;
 783         }
 784         /*
 785          * If there was no data block space found, we need to allocate
 786          * a new one.
 787          */
 788         if (use_block == -1) {
 789                 /*
 790                  * Add the new data block.
 791                  */
 792                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 793                                 &use_block))) {
 794                         xfs_trans_brelse(tp, lbp);
 795                         return error;
 796                 }
 797                 /*
 798                  * Initialize the block.
 799                  */
 800                 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
 801                         xfs_trans_brelse(tp, lbp);
 802                         return error;
 803                 }
 804                 /*
 805                  * If we're adding a new data block on the end we need to
 806                  * extend the bests table.  Copy it up one entry.
 807                  */
 808                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
 809                         bestsp--;
 810                         memmove(&bestsp[0], &bestsp[1],
 811                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 812                         be32_add_cpu(&ltp->bestcount, 1);
 813                         xfs_dir3_leaf_log_tail(args, lbp);
 814                         xfs_dir3_leaf_log_bests(args, lbp, 0,
 815                                                 be32_to_cpu(ltp->bestcount) - 1);
 816                 }
 817                 /*
 818                  * If we're filling in a previously empty block just log it.
 819                  */
 820                 else
 821                         xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
 822                 hdr = dbp->b_addr;
 823                 bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
 824                 bestsp[use_block] = bf[0].length;
 825                 grown = 1;
 826         } else {
 827                 /*
 828                  * Already had space in some data block.
 829                  * Just read that one in.
 830                  */
 831                 error = xfs_dir3_data_read(tp, dp,
 832                                    xfs_dir2_db_to_da(args->geo, use_block),
 833                                    0, &dbp);
 834                 if (error) {
 835                         xfs_trans_brelse(tp, lbp);
 836                         return error;
 837                 }
 838                 hdr = dbp->b_addr;
 839                 bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
 840                 grown = 0;
 841         }
 842         /*
 843          * Point to the biggest freespace in our data block.
 844          */
 845         dup = (xfs_dir2_data_unused_t *)
 846               ((char *)hdr + be16_to_cpu(bf[0].offset));
 847         needscan = needlog = 0;
 848         /*
 849          * Mark the initial part of our freespace in use for the new entry.
 850          */
 851         error = xfs_dir2_data_use_free(args, dbp, dup,
 852                         (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
 853                         length, &needlog, &needscan);
 854         if (error) {
 855                 xfs_trans_brelse(tp, lbp);
 856                 return error;
 857         }
 858         /*
 859          * Initialize our new entry (at last).
 860          */
 861         dep = (xfs_dir2_data_entry_t *)dup;
 862         dep->inumber = cpu_to_be64(args->inumber);
 863         dep->namelen = args->namelen;
 864         memcpy(dep->name, args->name, dep->namelen);
 865         xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
 866         tagp = xfs_dir2_data_entry_tag_p(dp->i_mount, dep);
 867         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 868         /*
 869          * Need to scan fix up the bestfree table.
 870          */
 871         if (needscan)
 872                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
 873         /*
 874          * Need to log the data block's header.
 875          */
 876         if (needlog)
 877                 xfs_dir2_data_log_header(args, dbp);
 878         xfs_dir2_data_log_entry(args, dbp, dep);
 879         /*
 880          * If the bests table needs to be changed, do it.
 881          * Log the change unless we've already done that.
 882          */
 883         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
 884                 bestsp[use_block] = bf[0].length;
 885                 if (!grown)
 886                         xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
 887         }
 888
 889         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 890                                        highstale, &lfloglow, &lfloghigh);
 891
 892         /*
 893          * Fill in the new leaf entry.
 894          */
 895         lep->hashval = cpu_to_be32(args->hashval);
 896         lep->address = cpu_to_be32(
 897                                 xfs_dir2_db_off_to_dataptr(args->geo, use_block,
 898                                 be16_to_cpu(*tagp)));
 899         /*
 900          * Log the leaf fields and give up the buffers.
 901          */
 902         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
 903         xfs_dir3_leaf_log_header(args, lbp);
 904         xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, lfloglow, lfloghigh);
 905         xfs_dir3_leaf_check(dp, lbp);
 906         xfs_dir3_data_check(dp, dbp);
 907         return 0;
 908 }
 909
 910 /*
 911  * Compact out any stale entries in the leaf.
 912  * Log the header and changed leaf entries, if any.
 913  */
 914 void
 915 xfs_dir3_leaf_compact(
 916         xfs_da_args_t   *args,          /* operation arguments */
 917         struct xfs_dir3_icleaf_hdr *leafhdr,
 918         struct xfs_buf  *bp)            /* leaf buffer */
 919 {
 920         int             from;           /* source leaf index */
 921         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 922         int             loglow;         /* first leaf entry to log */
 923         int             to;             /* target leaf index */
 924         struct xfs_inode *dp = args->dp;
 925
 926         leaf = bp->b_addr;
 927         if (!leafhdr->stale)
 928                 return;
 929
 930         /*
 931          * Compress out the stale entries in place.
 932          */
 933         for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
 934                 if (leafhdr->ents[from].address ==
 935                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 936                         continue;
 937                 /*
 938                  * Only actually copy the entries that are different.
 939                  */
 940                 if (from > to) {
 941                         if (loglow == -1)
 942                                 loglow = to;
 943                         leafhdr->ents[to] = leafhdr->ents[from];
 944                 }
 945                 to++;
 946         }
 947         /*
 948          * Update and log the header, log the leaf entries.
 949          */
 950         ASSERT(leafhdr->stale == from - to);
 951         leafhdr->count -= leafhdr->stale;
 952         leafhdr->stale = 0;
 953
 954         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, leafhdr);
 955         xfs_dir3_leaf_log_header(args, bp);
 956         if (loglow != -1)
 957                 xfs_dir3_leaf_log_ents(args, leafhdr, bp, loglow, to - 1);
 958 }
 959
 960 /*
 961  * Compact the leaf entries, removing stale ones.
 962  * Leave one stale entry behind - the one closest to our
 963  * insertion index - and the caller will shift that one to our insertion
 964  * point later.
 965  * Return new insertion index, where the remaining stale entry is,
 966  * and leaf logging indices.
 967  */
 968 void
 969 xfs_dir3_leaf_compact_x1(
 970         struct xfs_dir3_icleaf_hdr *leafhdr,
 971         struct xfs_dir2_leaf_entry *ents,
 972         int             *indexp,        /* insertion index */
 973         int             *lowstalep,     /* out: stale entry before us */
 974         int             *highstalep,    /* out: stale entry after us */
 975         int             *lowlogp,       /* out: low log index */
 976         int             *highlogp)      /* out: high log index */
 977 {
 978         int             from;           /* source copy index */
 979         int             highstale;      /* stale entry at/after index */
 980         int             index;          /* insertion index */
 981         int             keepstale;      /* source index of kept stale */
 982         int             lowstale;       /* stale entry before index */
 983         int             newindex=0;     /* new insertion index */
 984         int             to;             /* destination copy index */
 985
 986         ASSERT(leafhdr->stale > 1);
 987         index = *indexp;
 988
 989         xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
 990
 991         /*
 992          * Pick the better of lowstale and highstale.
 993          */
 994         if (lowstale >= 0 &&
 995             (highstale == leafhdr->count ||
 996              index - lowstale <= highstale - index))
 997                 keepstale = lowstale;
 998         else
 999                 keepstale = highstale;
1000         /*
1001          * Copy the entries in place, removing all the stale entries
1002          * except keepstale.
1003          */
1004         for (from = to = 0; from < leafhdr->count; from++) {
1005                 /*
1006                  * Notice the new value of index.
1007                  */
1008                 if (index == from)
1009                         newindex = to;
1010                 if (from != keepstale &&
1011                     ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1012                         if (from == to)
1013                                 *lowlogp = to;
1014                         continue;
1015                 }
1016                 /*
1017                  * Record the new keepstale value for the insertion.
1018                  */
1019                 if (from == keepstale)
1020                         lowstale = highstale = to;
1021                 /*
1022                  * Copy only the entries that have moved.
1023                  */
1024                 if (from > to)
1025                         ents[to] = ents[from];
1026                 to++;
1027         }
1028         ASSERT(from > to);
1029         /*
1030          * If the insertion point was past the last entry,
1031          * set the new insertion point accordingly.
1032          */
1033         if (index == from)
1034                 newindex = to;
1035         *indexp = newindex;
1036         /*
1037          * Adjust the leaf header values.
1038          */
1039         leafhdr->count -= from - to;
1040         leafhdr->stale = 1;
1041         /*
1042          * Remember the low/high stale value only in the "right"
1043          * direction.
1044          */
1045         if (lowstale >= newindex)
1046                 lowstale = -1;
1047         else
1048                 highstale = leafhdr->count;
1049         *highlogp = leafhdr->count - 1;
1050         *lowstalep = lowstale;
1051         *highstalep = highstale;
1052 }
1053
1054 /*
1055  * Log the bests entries indicated from a leaf1 block.
1056  */
1057 static void
1058 xfs_dir3_leaf_log_bests(
1059         struct xfs_da_args      *args,
1060         struct xfs_buf          *bp,            /* leaf buffer */
1061         int                     first,          /* first entry to log */
1062         int                     last)           /* last entry to log */
1063 {
1064         __be16                  *firstb;        /* pointer to first entry */
1065         __be16                  *lastb;         /* pointer to last entry */
1066         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1067         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1068
1069         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1070                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1071
1072         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1073         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1074         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1075         xfs_trans_log_buf(args->trans, bp,
1076                 (uint)((char *)firstb - (char *)leaf),
1077                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1078 }
1079
1080 /*
1081  * Log the leaf entries indicated from a leaf1 or leafn block.
1082  */
1083 void
1084 xfs_dir3_leaf_log_ents(
1085         struct xfs_da_args      *args,
1086         struct xfs_dir3_icleaf_hdr *hdr,
1087         struct xfs_buf          *bp,
1088         int                     first,
1089         int                     last)
1090 {
1091         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1092         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1093         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1094
1095         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1096                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1097                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1098                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1099
1100         firstlep = &hdr->ents[first];
1101         lastlep = &hdr->ents[last];
1102         xfs_trans_log_buf(args->trans, bp,
1103                 (uint)((char *)firstlep - (char *)leaf),
1104                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1105 }
1106
1107 /*
1108  * Log the header of the leaf1 or leafn block.
1109  */
1110 void
1111 xfs_dir3_leaf_log_header(
1112         struct xfs_da_args      *args,
1113         struct xfs_buf          *bp)
1114 {
1115         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1116
1117         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1118                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1119                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1120                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1121
1122         xfs_trans_log_buf(args->trans, bp,
1123                           (uint)((char *)&leaf->hdr - (char *)leaf),
1124                           args->geo->leaf_hdr_size - 1);
1125 }
1126
1127 /*
1128  * Log the tail of the leaf1 block.
1129  */
1130 STATIC void
1131 xfs_dir3_leaf_log_tail(
1132         struct xfs_da_args      *args,
1133         struct xfs_buf          *bp)
1134 {
1135         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1136         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1137
1138         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1139                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1140                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1141                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1142
1143         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1144         xfs_trans_log_buf(args->trans, bp, (uint)((char *)ltp - (char *)leaf),
1145                 (uint)(args->geo->blksize - 1));
1146 }
1147
1148 /*
1149  * Look up the entry referred to by args in the leaf format directory.
1150  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1151  * is also used by the node-format code.
1152  */
1153 int
1154 xfs_dir2_leaf_lookup(
1155         xfs_da_args_t           *args)          /* operation arguments */
1156 {
1157         struct xfs_buf          *dbp;           /* data block buffer */
1158         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1159         xfs_inode_t             *dp;            /* incore directory inode */
1160         int                     error;          /* error return code */
1161         int                     index;          /* found entry index */
1162         struct xfs_buf          *lbp;           /* leaf buffer */
1163         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1164         xfs_trans_t             *tp;            /* transaction pointer */
1165         struct xfs_dir3_icleaf_hdr leafhdr;
1166
1167         trace_xfs_dir2_leaf_lookup(args);
1168
1169         /*
1170          * Look up name in the leaf block, returning both buffers and index.
1171          */
1172         error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1173         if (error)
1174                 return error;
1175
1176         tp = args->trans;
1177         dp = args->dp;
1178         xfs_dir3_leaf_check(dp, lbp);
1179
1180         /*
1181          * Get to the leaf entry and contained data entry address.
1182          */
1183         lep = &leafhdr.ents[index];
1184
1185         /*
1186          * Point to the data entry.
1187          */
1188         dep = (xfs_dir2_data_entry_t *)
1189               ((char *)dbp->b_addr +
1190                xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1191         /*
1192          * Return the found inode number & CI name if appropriate
1193          */
1194         args->inumber = be64_to_cpu(dep->inumber);
1195         args->filetype = xfs_dir2_data_get_ftype(dp->i_mount, dep);
1196         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1197         xfs_trans_brelse(tp, dbp);
1198         xfs_trans_brelse(tp, lbp);
1199         return error;
1200 }
1201
1202 /*
1203  * Look up name/hash in the leaf block.
1204  * Fill in indexp with the found index, and dbpp with the data buffer.
1205  * If not found dbpp will be NULL, and ENOENT comes back.
1206  * lbpp will always be filled in with the leaf buffer unless there's an error.
1207  */
1208 static int                                      /* error */
1209 xfs_dir2_leaf_lookup_int(
1210         xfs_da_args_t           *args,          /* operation arguments */
1211         struct xfs_buf          **lbpp,         /* out: leaf buffer */
1212         int                     *indexp,        /* out: index in leaf block */
1213         struct xfs_buf          **dbpp,         /* out: data buffer */
1214         struct xfs_dir3_icleaf_hdr *leafhdr)
1215 {
1216         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1217         struct xfs_buf          *dbp = NULL;    /* data buffer */
1218         xfs_dir2_data_entry_t   *dep;           /* data entry */
1219         xfs_inode_t             *dp;            /* incore directory inode */
1220         int                     error;          /* error return code */
1221         int                     index;          /* index in leaf block */
1222         struct xfs_buf          *lbp;           /* leaf buffer */
1223         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1224         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1225         xfs_mount_t             *mp;            /* filesystem mount point */
1226         xfs_dir2_db_t           newdb;          /* new data block number */
1227         xfs_trans_t             *tp;            /* transaction pointer */
1228         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1229         enum xfs_dacmp          cmp;            /* name compare result */
1230
1231         dp = args->dp;
1232         tp = args->trans;
1233         mp = dp->i_mount;
1234
1235         error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, &lbp);
1236         if (error)
1237                 return error;
1238
1239         *lbpp = lbp;
1240         leaf = lbp->b_addr;
1241         xfs_dir3_leaf_check(dp, lbp);
1242         xfs_dir2_leaf_hdr_from_disk(mp, leafhdr, leaf);
1243
1244         /*
1245          * Look for the first leaf entry with our hash value.
1246          */
1247         index = xfs_dir2_leaf_search_hash(args, lbp);
1248         /*
1249          * Loop over all the entries with the right hash value
1250          * looking to match the name.
1251          */
1252         for (lep = &leafhdr->ents[index];
1253              index < leafhdr->count &&
1254                         be32_to_cpu(lep->hashval) == args->hashval;
1255              lep++, index++) {
1256                 /*
1257                  * Skip over stale leaf entries.
1258                  */
1259                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1260                         continue;
1261                 /*
1262                  * Get the new data block number.
1263                  */
1264                 newdb = xfs_dir2_dataptr_to_db(args->geo,
1265                                                be32_to_cpu(lep->address));
1266                 /*
1267                  * If it's not the same as the old data block number,
1268                  * need to pitch the old one and read the new one.
1269                  */
1270                 if (newdb != curdb) {
1271                         if (dbp)
1272                                 xfs_trans_brelse(tp, dbp);
1273                         error = xfs_dir3_data_read(tp, dp,
1274                                            xfs_dir2_db_to_da(args->geo, newdb),
1275                                            0, &dbp);
1276                         if (error) {
1277                                 xfs_trans_brelse(tp, lbp);
1278                                 return error;
1279                         }
1280                         curdb = newdb;
1281                 }
1282                 /*
1283                  * Point to the data entry.
1284                  */
1285                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1286                         xfs_dir2_dataptr_to_off(args->geo,
1287                                                 be32_to_cpu(lep->address)));
1288                 /*
1289                  * Compare name and if it's an exact match, return the index
1290                  * and buffer. If it's the first case-insensitive match, store
1291                  * the index and buffer and continue looking for an exact match.
1292                  */
1293                 cmp = xfs_dir2_compname(args, dep->name, dep->namelen);
1294                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1295                         args->cmpresult = cmp;
1296                         *indexp = index;
1297                         /* case exact match: return the current buffer. */
1298                         if (cmp == XFS_CMP_EXACT) {
1299                                 *dbpp = dbp;
1300                                 return 0;
1301                         }
1302                         cidb = curdb;
1303                 }
1304         }
1305         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1306         /*
1307          * Here, we can only be doing a lookup (not a rename or remove).
1308          * If a case-insensitive match was found earlier, re-read the
1309          * appropriate data block if required and return it.
1310          */
1311         if (args->cmpresult == XFS_CMP_CASE) {
1312                 ASSERT(cidb != -1);
1313                 if (cidb != curdb) {
1314                         xfs_trans_brelse(tp, dbp);
1315                         error = xfs_dir3_data_read(tp, dp,
1316                                            xfs_dir2_db_to_da(args->geo, cidb),
1317                                            0, &dbp);
1318                         if (error) {
1319                                 xfs_trans_brelse(tp, lbp);
1320                                 return error;
1321                         }
1322                 }
1323                 *dbpp = dbp;
1324                 return 0;
1325         }
1326         /*
1327          * No match found, return -ENOENT.
1328          */
1329         ASSERT(cidb == -1);
1330         if (dbp)
1331                 xfs_trans_brelse(tp, dbp);
1332         xfs_trans_brelse(tp, lbp);
1333         return -ENOENT;
1334 }
1335
1336 /*
1337  * Remove an entry from a leaf format directory.
1338  */
1339 int                                             /* error */
1340 xfs_dir2_leaf_removename(
1341         xfs_da_args_t           *args)          /* operation arguments */
1342 {
1343         struct xfs_da_geometry  *geo = args->geo;
1344         __be16                  *bestsp;        /* leaf block best freespace */
1345         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1346         xfs_dir2_db_t           db;             /* data block number */
1347         struct xfs_buf          *dbp;           /* data block buffer */
1348         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1349         xfs_inode_t             *dp;            /* incore directory inode */
1350         int                     error;          /* error return code */
1351         xfs_dir2_db_t           i;              /* temporary data block # */
1352         int                     index;          /* index into leaf entries */
1353         struct xfs_buf          *lbp;           /* leaf buffer */
1354         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1355         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1356         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1357         int                     needlog;        /* need to log data header */
1358         int                     needscan;       /* need to rescan data frees */
1359         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1360         struct xfs_dir2_data_free *bf;          /* bestfree table */
1361         struct xfs_dir3_icleaf_hdr leafhdr;
1362
1363         trace_xfs_dir2_leaf_removename(args);
1364
1365         /*
1366          * Lookup the leaf entry, get the leaf and data blocks read in.
1367          */
1368         error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1369         if (error)
1370                 return error;
1371
1372         dp = args->dp;
1373         leaf = lbp->b_addr;
1374         hdr = dbp->b_addr;
1375         xfs_dir3_data_check(dp, dbp);
1376         bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1377
1378         /*
1379          * Point to the leaf entry, use that to point to the data entry.
1380          */
1381         lep = &leafhdr.ents[index];
1382         db = xfs_dir2_dataptr_to_db(geo, be32_to_cpu(lep->address));
1383         dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1384                 xfs_dir2_dataptr_to_off(geo, be32_to_cpu(lep->address)));
1385         needscan = needlog = 0;
1386         oldbest = be16_to_cpu(bf[0].length);
1387         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
1388         bestsp = xfs_dir2_leaf_bests_p(ltp);
1389         if (be16_to_cpu(bestsp[db]) != oldbest) {
1390                 xfs_buf_mark_corrupt(lbp);
1391                 return -EFSCORRUPTED;
1392         }
1393         /*
1394          * Mark the former data entry unused.
1395          */
1396         xfs_dir2_data_make_free(args, dbp,
1397                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1398                 xfs_dir2_data_entsize(dp->i_mount, dep->namelen), &needlog,
1399                 &needscan);
1400         /*
1401          * We just mark the leaf entry stale by putting a null in it.
1402          */
1403         leafhdr.stale++;
1404         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
1405         xfs_dir3_leaf_log_header(args, lbp);
1406
1407         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1408         xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, index, index);
1409
1410         /*
1411          * Scan the freespace in the data block again if necessary,
1412          * log the data block header if necessary.
1413          */
1414         if (needscan)
1415                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
1416         if (needlog)
1417                 xfs_dir2_data_log_header(args, dbp);
1418         /*
1419          * If the longest freespace in the data block has changed,
1420          * put the new value in the bests table and log that.
1421          */
1422         if (be16_to_cpu(bf[0].length) != oldbest) {
1423                 bestsp[db] = bf[0].length;
1424                 xfs_dir3_leaf_log_bests(args, lbp, db, db);
1425         }
1426         xfs_dir3_data_check(dp, dbp);
1427         /*
1428          * If the data block is now empty then get rid of the data block.
1429          */
1430         if (be16_to_cpu(bf[0].length) ==
1431             geo->blksize - geo->data_entry_offset) {
1432                 ASSERT(db != geo->datablk);
1433                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1434                         /*
1435                          * Nope, can't get rid of it because it caused
1436                          * allocation of a bmap btree block to do so.
1437                          * Just go on, returning success, leaving the
1438                          * empty block in place.
1439                          */
1440                         if (error == -ENOSPC && args->total == 0)
1441                                 error = 0;
1442                         xfs_dir3_leaf_check(dp, lbp);
1443                         return error;
1444                 }
1445                 dbp = NULL;
1446                 /*
1447                  * If this is the last data block then compact the
1448                  * bests table by getting rid of entries.
1449                  */
1450                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1451                         /*
1452                          * Look for the last active entry (i).
1453                          */
1454                         for (i = db - 1; i > 0; i--) {
1455                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1456                                         break;
1457                         }
1458                         /*
1459                          * Copy the table down so inactive entries at the
1460                          * end are removed.
1461                          */
1462                         memmove(&bestsp[db - i], bestsp,
1463                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1464                         be32_add_cpu(&ltp->bestcount, -(db - i));
1465                         xfs_dir3_leaf_log_tail(args, lbp);
1466                         xfs_dir3_leaf_log_bests(args, lbp, 0,
1467                                                 be32_to_cpu(ltp->bestcount) - 1);
1468                 } else
1469                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1470         }
1471         /*
1472          * If the data block was not the first one, drop it.
1473          */
1474         else if (db != geo->datablk)
1475                 dbp = NULL;
1476
1477         xfs_dir3_leaf_check(dp, lbp);
1478         /*
1479          * See if we can convert to block form.
1480          */
1481         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1482 }
1483
1484 /*
1485  * Replace the inode number in a leaf format directory entry.
1486  */
1487 int                                             /* error */
1488 xfs_dir2_leaf_replace(
1489         xfs_da_args_t           *args)          /* operation arguments */
1490 {
1491         struct xfs_buf          *dbp;           /* data block buffer */
1492         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1493         xfs_inode_t             *dp;            /* incore directory inode */
1494         int                     error;          /* error return code */
1495         int                     index;          /* index of leaf entry */
1496         struct xfs_buf          *lbp;           /* leaf buffer */
1497         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1498         xfs_trans_t             *tp;            /* transaction pointer */
1499         struct xfs_dir3_icleaf_hdr leafhdr;
1500
1501         trace_xfs_dir2_leaf_replace(args);
1502
1503         /*
1504          * Look up the entry.
1505          */
1506         error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1507         if (error)
1508                 return error;
1509
1510         dp = args->dp;
1511         /*
1512          * Point to the leaf entry, get data address from it.
1513          */
1514         lep = &leafhdr.ents[index];
1515         /*
1516          * Point to the data entry.
1517          */
1518         dep = (xfs_dir2_data_entry_t *)
1519               ((char *)dbp->b_addr +
1520                xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1521         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1522         /*
1523          * Put the new inode number in, log it.
1524          */
1525         dep->inumber = cpu_to_be64(args->inumber);
1526         xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
1527         tp = args->trans;
1528         xfs_dir2_data_log_entry(args, dbp, dep);
1529         xfs_dir3_leaf_check(dp, lbp);
1530         xfs_trans_brelse(tp, lbp);
1531         return 0;
1532 }
1533
1534 /*
1535  * Return index in the leaf block (lbp) which is either the first
1536  * one with this hash value, or if there are none, the insert point
1537  * for that hash value.
1538  */
1539 int                                             /* index value */
1540 xfs_dir2_leaf_search_hash(
1541         xfs_da_args_t           *args,          /* operation arguments */
1542         struct xfs_buf          *lbp)           /* leaf buffer */
1543 {
1544         xfs_dahash_t            hash=0;         /* hash from this entry */
1545         xfs_dahash_t            hashwant;       /* hash value looking for */
1546         int                     high;           /* high leaf index */
1547         int                     low;            /* low leaf index */
1548         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1549         int                     mid=0;          /* current leaf index */
1550         struct xfs_dir3_icleaf_hdr leafhdr;
1551
1552         xfs_dir2_leaf_hdr_from_disk(args->dp->i_mount, &leafhdr, lbp->b_addr);
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 = leafhdr.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         struct xfs_da_geometry  *geo = args->geo;
1596         __be16                  *bestsp;        /* leaf bests table */
1597         struct xfs_buf          *dbp;           /* data block buffer */
1598         xfs_inode_t             *dp;            /* incore directory inode */
1599         int                     error;          /* error return value */
1600         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1601         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1602         xfs_trans_t             *tp;            /* transaction pointer */
1603
1604         dp = args->dp;
1605         tp = args->trans;
1606         /*
1607          * Read the offending data block.  We need its buffer.
1608          */
1609         error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(geo, db), 0, &dbp);
1610         if (error)
1611                 return error;
1612
1613         leaf = lbp->b_addr;
1614         ltp = xfs_dir2_leaf_tail_p(geo, leaf);
1615
1616 #ifdef DEBUG
1617 {
1618         struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1619         struct xfs_dir2_data_free *bf =
1620                 xfs_dir2_data_bestfree_p(dp->i_mount, 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                geo->blksize - geo->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(args, lbp);
1645         xfs_dir3_leaf_log_bests(args, 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         struct xfs_buf          *lbp;           /* buffer for leaf block */
1684         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1685         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1686         xfs_mount_t             *mp;            /* filesystem mount point */
1687         int                     rval;           /* successful free trim? */
1688         xfs_trans_t             *tp;            /* transaction pointer */
1689         struct xfs_dir3_icleaf_hdr leafhdr;
1690         struct xfs_dir3_icfree_hdr freehdr;
1691
1692         /*
1693          * There's more than a leaf level in the btree, so there must
1694          * be multiple leafn blocks.  Give up.
1695          */
1696         if (state->path.active > 1)
1697                 return 0;
1698         args = state->args;
1699
1700         trace_xfs_dir2_node_to_leaf(args);
1701
1702         mp = state->mp;
1703         dp = args->dp;
1704         tp = args->trans;
1705         /*
1706          * Get the last offset in the file.
1707          */
1708         if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1709                 return error;
1710         }
1711         fo -= args->geo->fsbcount;
1712         /*
1713          * If there are freespace blocks other than the first one,
1714          * take this opportunity to remove trailing empty freespace blocks
1715          * that may have been left behind during no-space-reservation
1716          * operations.
1717          */
1718         while (fo > args->geo->freeblk) {
1719                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1720                         return error;
1721                 }
1722                 if (rval)
1723                         fo -= args->geo->fsbcount;
1724                 else
1725                         return 0;
1726         }
1727         /*
1728          * Now find the block just before the freespace block.
1729          */
1730         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1731                 return error;
1732         }
1733         /*
1734          * If it's not the single leaf block, give up.
1735          */
1736         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1737                 return 0;
1738         lbp = state->path.blk[0].bp;
1739         leaf = lbp->b_addr;
1740         xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, leaf);
1741
1742         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1743                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1744
1745         /*
1746          * Read the freespace block.
1747          */
1748         error = xfs_dir2_free_read(tp, dp,  args->geo->freeblk, &fbp);
1749         if (error)
1750                 return error;
1751         xfs_dir2_free_hdr_from_disk(mp, &freehdr, fbp->b_addr);
1752
1753         ASSERT(!freehdr.firstdb);
1754
1755         /*
1756          * Now see if the leafn and free data will fit in a leaf1.
1757          * If not, release the buffer and give up.
1758          */
1759         if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > args->geo->blksize) {
1760                 xfs_trans_brelse(tp, fbp);
1761                 return 0;
1762         }
1763
1764         /*
1765          * If the leaf has any stale entries in it, compress them out.
1766          */
1767         if (leafhdr.stale)
1768                 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1769
1770         lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1771         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1772         leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1773                                         ? XFS_DIR2_LEAF1_MAGIC
1774                                         : XFS_DIR3_LEAF1_MAGIC;
1775
1776         /*
1777          * Set up the leaf tail from the freespace block.
1778          */
1779         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1780         ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1781
1782         /*
1783          * Set up the leaf bests table.
1784          */
1785         memcpy(xfs_dir2_leaf_bests_p(ltp), freehdr.bests,
1786                 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1787
1788         xfs_dir2_leaf_hdr_to_disk(mp, leaf, &leafhdr);
1789         xfs_dir3_leaf_log_header(args, lbp);
1790         xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1791         xfs_dir3_leaf_log_tail(args, lbp);
1792         xfs_dir3_leaf_check(dp, lbp);
1793
1794         /*
1795          * Get rid of the freespace block.
1796          */
1797         error = xfs_dir2_shrink_inode(args,
1798                         xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET),
1799                         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 }