Pileus Git - ~andy/linux/blob - fs/xfs/xfs_dir2_leaf.c

   1 /*
   2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_types.h"
  22 #include "xfs_bit.h"
  23 #include "xfs_log.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_mount.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_bmap_btree.h"
  30 #include "xfs_dinode.h"
  31 #include "xfs_inode.h"
  32 #include "xfs_bmap.h"
  33 #include "xfs_dir2_format.h"
  34 #include "xfs_dir2.h"
  35 #include "xfs_dir2_priv.h"
  36 #include "xfs_error.h"
  37 #include "xfs_trace.h"
  38 #include "xfs_buf_item.h"
  39 #include "xfs_cksum.h"
  40
  41 /*
  42  * Local function declarations.
  43  */
  44 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
  45                                     int *indexp, struct xfs_buf **dbpp);
  46 static void xfs_dir3_leaf_log_bests(struct xfs_trans *tp, struct xfs_buf *bp,
  47                                     int first, int last);
  48 static void xfs_dir3_leaf_log_tail(struct xfs_trans *tp, struct xfs_buf *bp);
  49
  50 /*
  51  * Check the internal consistency of a leaf1 block.
  52  * Pop an assert if something is wrong.
  53  */
  54 #ifdef DEBUG
  55 #define xfs_dir3_leaf_check(mp, bp) \
  56 do { \
  57         if (!xfs_dir3_leaf1_check((mp), (bp))) \
  58                 ASSERT(0); \
  59 } while (0);
  60
  61 STATIC bool
  62 xfs_dir3_leaf1_check(
  63         struct xfs_mount        *mp,
  64         struct xfs_buf          *bp)
  65 {
  66         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  67         struct xfs_dir3_icleaf_hdr leafhdr;
  68
  69         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
  70
  71         if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
  72                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  73                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  74                         return false;
  75         } else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
  76                 return false;
  77
  78         return xfs_dir3_leaf_check_int(mp, &leafhdr, leaf);
  79 }
  80 #else
  81 #define xfs_dir3_leaf_check(mp, bp)
  82 #endif
  83
  84 void
  85 xfs_dir3_leaf_hdr_from_disk(
  86         struct xfs_dir3_icleaf_hdr      *to,
  87         struct xfs_dir2_leaf            *from)
  88 {
  89         if (from->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
  90             from->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) {
  91                 to->forw = be32_to_cpu(from->hdr.info.forw);
  92                 to->back = be32_to_cpu(from->hdr.info.back);
  93                 to->magic = be16_to_cpu(from->hdr.info.magic);
  94                 to->count = be16_to_cpu(from->hdr.count);
  95                 to->stale = be16_to_cpu(from->hdr.stale);
  96         } else {
  97                 struct xfs_dir3_leaf_hdr *hdr3 = (struct xfs_dir3_leaf_hdr *)from;
  98
  99                 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
 100                 to->back = be32_to_cpu(hdr3->info.hdr.back);
 101                 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
 102                 to->count = be16_to_cpu(hdr3->count);
 103                 to->stale = be16_to_cpu(hdr3->stale);
 104         }
 105
 106         ASSERT(to->magic == XFS_DIR2_LEAF1_MAGIC ||
 107                to->magic == XFS_DIR3_LEAF1_MAGIC ||
 108                to->magic == XFS_DIR2_LEAFN_MAGIC ||
 109                to->magic == XFS_DIR3_LEAFN_MAGIC);
 110 }
 111
 112 void
 113 xfs_dir3_leaf_hdr_to_disk(
 114         struct xfs_dir2_leaf            *to,
 115         struct xfs_dir3_icleaf_hdr      *from)
 116 {
 117         ASSERT(from->magic == XFS_DIR2_LEAF1_MAGIC ||
 118                from->magic == XFS_DIR3_LEAF1_MAGIC ||
 119                from->magic == XFS_DIR2_LEAFN_MAGIC ||
 120                from->magic == XFS_DIR3_LEAFN_MAGIC);
 121
 122         if (from->magic == XFS_DIR2_LEAF1_MAGIC ||
 123             from->magic == XFS_DIR2_LEAFN_MAGIC) {
 124                 to->hdr.info.forw = cpu_to_be32(from->forw);
 125                 to->hdr.info.back = cpu_to_be32(from->back);
 126                 to->hdr.info.magic = cpu_to_be16(from->magic);
 127                 to->hdr.count = cpu_to_be16(from->count);
 128                 to->hdr.stale = cpu_to_be16(from->stale);
 129         } else {
 130                 struct xfs_dir3_leaf_hdr *hdr3 = (struct xfs_dir3_leaf_hdr *)to;
 131
 132                 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
 133                 hdr3->info.hdr.back = cpu_to_be32(from->back);
 134                 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
 135                 hdr3->count = cpu_to_be16(from->count);
 136                 hdr3->stale = cpu_to_be16(from->stale);
 137         }
 138 }
 139
 140 bool
 141 xfs_dir3_leaf_check_int(
 142         struct xfs_mount        *mp,
 143         struct xfs_dir3_icleaf_hdr *hdr,
 144         struct xfs_dir2_leaf    *leaf)
 145 {
 146         struct xfs_dir2_leaf_entry *ents;
 147         xfs_dir2_leaf_tail_t    *ltp;
 148         int                     stale;
 149         int                     i;
 150
 151         ents = xfs_dir3_leaf_ents_p(leaf);
 152         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 153
 154         /*
 155          * XXX (dgc): This value is not restrictive enough.
 156          * Should factor in the size of the bests table as well.
 157          * We can deduce a value for that from di_size.
 158          */
 159         if (hdr->count > xfs_dir3_max_leaf_ents(mp, leaf))
 160                 return false;
 161
 162         /* Leaves and bests don't overlap in leaf format. */
 163         if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
 164              hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
 165             (char *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
 166                 return false;
 167
 168         /* Check hash value order, count stale entries.  */
 169         for (i = stale = 0; i < hdr->count; i++) {
 170                 if (i + 1 < hdr->count) {
 171                         if (be32_to_cpu(ents[i].hashval) >
 172                                         be32_to_cpu(ents[i + 1].hashval))
 173                                 return false;
 174                 }
 175                 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 176                         stale++;
 177         }
 178         if (hdr->stale != stale)
 179                 return false;
 180         return true;
 181 }
 182
 183 static bool
 184 xfs_dir3_leaf_verify(
 185         struct xfs_buf          *bp,
 186         __uint16_t              magic)
 187 {
 188         struct xfs_mount        *mp = bp->b_target->bt_mount;
 189         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 190         struct xfs_dir3_icleaf_hdr leafhdr;
 191
 192         ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
 193
 194         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 195         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 196                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 197
 198                 if ((magic == XFS_DIR2_LEAF1_MAGIC &&
 199                      leafhdr.magic != XFS_DIR3_LEAF1_MAGIC) ||
 200                     (magic == XFS_DIR2_LEAFN_MAGIC &&
 201                      leafhdr.magic != XFS_DIR3_LEAFN_MAGIC))
 202                         return false;
 203
 204                 if (!uuid_equal(&leaf3->info.uuid, &mp->m_sb.sb_uuid))
 205                         return false;
 206                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
 207                         return false;
 208         } else {
 209                 if (leafhdr.magic != magic)
 210                         return false;
 211         }
 212         return xfs_dir3_leaf_check_int(mp, &leafhdr, leaf);
 213 }
 214
 215 static void
 216 __read_verify(
 217         struct xfs_buf  *bp,
 218         __uint16_t      magic)
 219 {
 220         struct xfs_mount        *mp = bp->b_target->bt_mount;
 221
 222         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
 223              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
 224                                           XFS_DIR3_LEAF_CRC_OFF)) ||
 225             !xfs_dir3_leaf_verify(bp, magic)) {
 226                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 227                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 228         }
 229 }
 230
 231 static void
 232 __write_verify(
 233         struct xfs_buf  *bp,
 234         __uint16_t      magic)
 235 {
 236         struct xfs_mount        *mp = bp->b_target->bt_mount;
 237         struct xfs_buf_log_item *bip = bp->b_fspriv;
 238         struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
 239
 240         if (!xfs_dir3_leaf_verify(bp, magic)) {
 241                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 242                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 243                 return;
 244         }
 245
 246         if (!xfs_sb_version_hascrc(&mp->m_sb))
 247                 return;
 248
 249         if (bip)
 250                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 251
 252         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_LEAF_CRC_OFF);
 253 }
 254
 255 static void
 256 xfs_dir3_leaf1_read_verify(
 257         struct xfs_buf  *bp)
 258 {
 259         __read_verify(bp, XFS_DIR2_LEAF1_MAGIC);
 260 }
 261
 262 static void
 263 xfs_dir3_leaf1_write_verify(
 264         struct xfs_buf  *bp)
 265 {
 266         __write_verify(bp, XFS_DIR2_LEAF1_MAGIC);
 267 }
 268
 269 static void
 270 xfs_dir3_leafn_read_verify(
 271         struct xfs_buf  *bp)
 272 {
 273         __read_verify(bp, XFS_DIR2_LEAFN_MAGIC);
 274 }
 275
 276 static void
 277 xfs_dir3_leafn_write_verify(
 278         struct xfs_buf  *bp)
 279 {
 280         __write_verify(bp, XFS_DIR2_LEAFN_MAGIC);
 281 }
 282
 283 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
 284         .verify_read = xfs_dir3_leaf1_read_verify,
 285         .verify_write = xfs_dir3_leaf1_write_verify,
 286 };
 287
 288 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
 289         .verify_read = xfs_dir3_leafn_read_verify,
 290         .verify_write = xfs_dir3_leafn_write_verify,
 291 };
 292
 293 static int
 294 xfs_dir3_leaf_read(
 295         struct xfs_trans        *tp,
 296         struct xfs_inode        *dp,
 297         xfs_dablk_t             fbno,
 298         xfs_daddr_t             mappedbno,
 299         struct xfs_buf          **bpp)
 300 {
 301         int                     err;
 302
 303         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 304                                 XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
 305         if (!err && tp)
 306                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
 307         return err;
 308 }
 309
 310 int
 311 xfs_dir3_leafn_read(
 312         struct xfs_trans        *tp,
 313         struct xfs_inode        *dp,
 314         xfs_dablk_t             fbno,
 315         xfs_daddr_t             mappedbno,
 316         struct xfs_buf          **bpp)
 317 {
 318         int                     err;
 319
 320         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 321                                 XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
 322         if (!err && tp)
 323                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
 324         return err;
 325 }
 326
 327 /*
 328  * Initialize a new leaf block, leaf1 or leafn magic accepted.
 329  */
 330 static void
 331 xfs_dir3_leaf_init(
 332         struct xfs_mount        *mp,
 333         struct xfs_trans        *tp,
 334         struct xfs_buf          *bp,
 335         xfs_ino_t               owner,
 336         __uint16_t              type)
 337 {
 338         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 339
 340         ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
 341
 342         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 343                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 344
 345                 memset(leaf3, 0, sizeof(*leaf3));
 346
 347                 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
 348                                          ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
 349                                          : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 350                 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
 351                 leaf3->info.owner = cpu_to_be64(owner);
 352                 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_uuid);
 353         } else {
 354                 memset(leaf, 0, sizeof(*leaf));
 355                 leaf->hdr.info.magic = cpu_to_be16(type);
 356         }
 357
 358         /*
 359          * If it's a leaf-format directory initialize the tail.
 360          * Caller is responsible for initialising the bests table.
 361          */
 362         if (type == XFS_DIR2_LEAF1_MAGIC) {
 363                 struct xfs_dir2_leaf_tail *ltp;
 364
 365                 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 366                 ltp->bestcount = 0;
 367                 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
 368                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
 369         } else {
 370                 bp->b_ops = &xfs_dir3_leafn_buf_ops;
 371                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
 372         }
 373 }
 374
 375 int
 376 xfs_dir3_leaf_get_buf(
 377         xfs_da_args_t           *args,
 378         xfs_dir2_db_t           bno,
 379         struct xfs_buf          **bpp,
 380         __uint16_t              magic)
 381 {
 382         struct xfs_inode        *dp = args->dp;
 383         struct xfs_trans        *tp = args->trans;
 384         struct xfs_mount        *mp = dp->i_mount;
 385         struct xfs_buf          *bp;
 386         int                     error;
 387
 388         ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
 389         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
 390                bno < XFS_DIR2_FREE_FIRSTDB(mp));
 391
 392         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
 393                                XFS_DATA_FORK);
 394         if (error)
 395                 return error;
 396
 397         xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
 398         xfs_dir3_leaf_log_header(tp, bp);
 399         if (magic == XFS_DIR2_LEAF1_MAGIC)
 400                 xfs_dir3_leaf_log_tail(tp, bp);
 401         *bpp = bp;
 402         return 0;
 403 }
 404
 405 /*
 406  * Convert a block form directory to a leaf form directory.
 407  */
 408 int                                             /* error */
 409 xfs_dir2_block_to_leaf(
 410         xfs_da_args_t           *args,          /* operation arguments */
 411         struct xfs_buf          *dbp)           /* input block's buffer */
 412 {
 413         __be16                  *bestsp;        /* leaf's bestsp entries */
 414         xfs_dablk_t             blkno;          /* leaf block's bno */
 415         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 416         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
 417         xfs_dir2_block_tail_t   *btp;           /* block's tail */
 418         xfs_inode_t             *dp;            /* incore directory inode */
 419         int                     error;          /* error return code */
 420         struct xfs_buf          *lbp;           /* leaf block's buffer */
 421         xfs_dir2_db_t           ldb;            /* leaf block's bno */
 422         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 423         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
 424         xfs_mount_t             *mp;            /* filesystem mount point */
 425         int                     needlog;        /* need to log block header */
 426         int                     needscan;       /* need to rescan bestfree */
 427         xfs_trans_t             *tp;            /* transaction pointer */
 428         struct xfs_dir2_data_free *bf;
 429         struct xfs_dir2_leaf_entry *ents;
 430         struct xfs_dir3_icleaf_hdr leafhdr;
 431
 432         trace_xfs_dir2_block_to_leaf(args);
 433
 434         dp = args->dp;
 435         mp = dp->i_mount;
 436         tp = args->trans;
 437         /*
 438          * Add the leaf block to the inode.
 439          * This interface will only put blocks in the leaf/node range.
 440          * Since that's empty now, we'll get the root (block 0 in range).
 441          */
 442         if ((error = xfs_da_grow_inode(args, &blkno))) {
 443                 return error;
 444         }
 445         ldb = xfs_dir2_da_to_db(mp, blkno);
 446         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
 447         /*
 448          * Initialize the leaf block, get a buffer for it.
 449          */
 450         error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
 451         if (error)
 452                 return error;
 453
 454         leaf = lbp->b_addr;
 455         hdr = dbp->b_addr;
 456         xfs_dir3_data_check(dp, dbp);
 457         btp = xfs_dir2_block_tail_p(mp, hdr);
 458         blp = xfs_dir2_block_leaf_p(btp);
 459         bf = xfs_dir3_data_bestfree_p(hdr);
 460         ents = xfs_dir3_leaf_ents_p(leaf);
 461
 462         /*
 463          * Set the counts in the leaf header.
 464          */
 465         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 466         leafhdr.count = be32_to_cpu(btp->count);
 467         leafhdr.stale = be32_to_cpu(btp->stale);
 468         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
 469         xfs_dir3_leaf_log_header(tp, lbp);
 470
 471         /*
 472          * Could compact these but I think we always do the conversion
 473          * after squeezing out stale entries.
 474          */
 475         memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
 476         xfs_dir3_leaf_log_ents(tp, lbp, 0, leafhdr.count - 1);
 477         needscan = 0;
 478         needlog = 1;
 479         /*
 480          * Make the space formerly occupied by the leaf entries and block
 481          * tail be free.
 482          */
 483         xfs_dir2_data_make_free(tp, dbp,
 484                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 485                 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
 486                                        (char *)blp),
 487                 &needlog, &needscan);
 488         /*
 489          * Fix up the block header, make it a data block.
 490          */
 491         dbp->b_ops = &xfs_dir3_data_buf_ops;
 492         xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
 493         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
 494                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 495         else
 496                 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 497
 498         if (needscan)
 499                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 500         /*
 501          * Set up leaf tail and bests table.
 502          */
 503         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 504         ltp->bestcount = cpu_to_be32(1);
 505         bestsp = xfs_dir2_leaf_bests_p(ltp);
 506         bestsp[0] =  bf[0].length;
 507         /*
 508          * Log the data header and leaf bests table.
 509          */
 510         if (needlog)
 511                 xfs_dir2_data_log_header(tp, dbp);
 512         xfs_dir3_leaf_check(mp, lbp);
 513         xfs_dir3_data_check(dp, dbp);
 514         xfs_dir3_leaf_log_bests(tp, lbp, 0, 0);
 515         return 0;
 516 }
 517
 518 STATIC void
 519 xfs_dir3_leaf_find_stale(
 520         struct xfs_dir3_icleaf_hdr *leafhdr,
 521         struct xfs_dir2_leaf_entry *ents,
 522         int                     index,
 523         int                     *lowstale,
 524         int                     *highstale)
 525 {
 526         /*
 527          * Find the first stale entry before our index, if any.
 528          */
 529         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
 530                 if (ents[*lowstale].address ==
 531                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 532                         break;
 533         }
 534
 535         /*
 536          * Find the first stale entry at or after our index, if any.
 537          * Stop if the result would require moving more entries than using
 538          * lowstale.
 539          */
 540         for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
 541                 if (ents[*highstale].address ==
 542                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 543                         break;
 544                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
 545                         break;
 546         }
 547 }
 548
 549 struct xfs_dir2_leaf_entry *
 550 xfs_dir3_leaf_find_entry(
 551         struct xfs_dir3_icleaf_hdr *leafhdr,
 552         struct xfs_dir2_leaf_entry *ents,
 553         int                     index,          /* leaf table position */
 554         int                     compact,        /* need to compact leaves */
 555         int                     lowstale,       /* index of prev stale leaf */
 556         int                     highstale,      /* index of next stale leaf */
 557         int                     *lfloglow,      /* low leaf logging index */
 558         int                     *lfloghigh)     /* high leaf logging index */
 559 {
 560         if (!leafhdr->stale) {
 561                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
 562
 563                 /*
 564                  * Now we need to make room to insert the leaf entry.
 565                  *
 566                  * If there are no stale entries, just insert a hole at index.
 567                  */
 568                 lep = &ents[index];
 569                 if (index < leafhdr->count)
 570                         memmove(lep + 1, lep,
 571                                 (leafhdr->count - index) * sizeof(*lep));
 572
 573                 /*
 574                  * Record low and high logging indices for the leaf.
 575                  */
 576                 *lfloglow = index;
 577                 *lfloghigh = leafhdr->count++;
 578                 return lep;
 579         }
 580
 581         /*
 582          * There are stale entries.
 583          *
 584          * We will use one of them for the new entry.  It's probably not at
 585          * the right location, so we'll have to shift some up or down first.
 586          *
 587          * If we didn't compact before, we need to find the nearest stale
 588          * entries before and after our insertion point.
 589          */
 590         if (compact == 0)
 591                 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
 592                                          &lowstale, &highstale);
 593
 594         /*
 595          * If the low one is better, use it.
 596          */
 597         if (lowstale >= 0 &&
 598             (highstale == leafhdr->count ||
 599              index - lowstale - 1 < highstale - index)) {
 600                 ASSERT(index - lowstale - 1 >= 0);
 601                 ASSERT(ents[lowstale].address ==
 602                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 603
 604                 /*
 605                  * Copy entries up to cover the stale entry and make room
 606                  * for the new entry.
 607                  */
 608                 if (index - lowstale - 1 > 0) {
 609                         memmove(&ents[lowstale], &ents[lowstale + 1],
 610                                 (index - lowstale - 1) *
 611                                         sizeof(xfs_dir2_leaf_entry_t));
 612                 }
 613                 *lfloglow = MIN(lowstale, *lfloglow);
 614                 *lfloghigh = MAX(index - 1, *lfloghigh);
 615                 leafhdr->stale--;
 616                 return &ents[index - 1];
 617         }
 618
 619         /*
 620          * The high one is better, so use that one.
 621          */
 622         ASSERT(highstale - index >= 0);
 623         ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 624
 625         /*
 626          * Copy entries down to cover the stale entry and make room for the
 627          * new entry.
 628          */
 629         if (highstale - index > 0) {
 630                 memmove(&ents[index + 1], &ents[index],
 631                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
 632         }
 633         *lfloglow = MIN(index, *lfloglow);
 634         *lfloghigh = MAX(highstale, *lfloghigh);
 635         leafhdr->stale--;
 636         return &ents[index];
 637 }
 638
 639 /*
 640  * Add an entry to a leaf form directory.
 641  */
 642 int                                             /* error */
 643 xfs_dir2_leaf_addname(
 644         xfs_da_args_t           *args)          /* operation arguments */
 645 {
 646         __be16                  *bestsp;        /* freespace table in leaf */
 647         int                     compact;        /* need to compact leaves */
 648         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 649         struct xfs_buf          *dbp;           /* data block buffer */
 650         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 651         xfs_inode_t             *dp;            /* incore directory inode */
 652         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
 653         int                     error;          /* error return value */
 654         int                     grown;          /* allocated new data block */
 655         int                     highstale;      /* index of next stale leaf */
 656         int                     i;              /* temporary, index */
 657         int                     index;          /* leaf table position */
 658         struct xfs_buf          *lbp;           /* leaf's buffer */
 659         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 660         int                     length;         /* length of new entry */
 661         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
 662         int                     lfloglow;       /* low leaf logging index */
 663         int                     lfloghigh;      /* high leaf logging index */
 664         int                     lowstale;       /* index of prev stale leaf */
 665         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 666         xfs_mount_t             *mp;            /* filesystem mount point */
 667         int                     needbytes;      /* leaf block bytes needed */
 668         int                     needlog;        /* need to log data header */
 669         int                     needscan;       /* need to rescan data free */
 670         __be16                  *tagp;          /* end of data entry */
 671         xfs_trans_t             *tp;            /* transaction pointer */
 672         xfs_dir2_db_t           use_block;      /* data block number */
 673         struct xfs_dir2_data_free *bf;          /* bestfree table */
 674         struct xfs_dir2_leaf_entry *ents;
 675         struct xfs_dir3_icleaf_hdr leafhdr;
 676
 677         trace_xfs_dir2_leaf_addname(args);
 678
 679         dp = args->dp;
 680         tp = args->trans;
 681         mp = dp->i_mount;
 682
 683         error = xfs_dir3_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
 684         if (error)
 685                 return error;
 686
 687         /*
 688          * Look up the entry by hash value and name.
 689          * We know it's not there, our caller has already done a lookup.
 690          * So the index is of the entry to insert in front of.
 691          * But if there are dup hash values the index is of the first of those.
 692          */
 693         index = xfs_dir2_leaf_search_hash(args, lbp);
 694         leaf = lbp->b_addr;
 695         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 696         ents = xfs_dir3_leaf_ents_p(leaf);
 697         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 698         bestsp = xfs_dir2_leaf_bests_p(ltp);
 699         length = xfs_dir3_data_entsize(mp, args->namelen);
 700
 701         /*
 702          * See if there are any entries with the same hash value
 703          * and space in their block for the new entry.
 704          * This is good because it puts multiple same-hash value entries
 705          * in a data block, improving the lookup of those entries.
 706          */
 707         for (use_block = -1, lep = &ents[index];
 708              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 709              index++, lep++) {
 710                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 711                         continue;
 712                 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 713                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 714                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
 715                 if (be16_to_cpu(bestsp[i]) >= length) {
 716                         use_block = i;
 717                         break;
 718                 }
 719         }
 720         /*
 721          * Didn't find a block yet, linear search all the data blocks.
 722          */
 723         if (use_block == -1) {
 724                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 725                         /*
 726                          * Remember a block we see that's missing.
 727                          */
 728                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
 729                             use_block == -1)
 730                                 use_block = i;
 731                         else if (be16_to_cpu(bestsp[i]) >= length) {
 732                                 use_block = i;
 733                                 break;
 734                         }
 735                 }
 736         }
 737         /*
 738          * How many bytes do we need in the leaf block?
 739          */
 740         needbytes = 0;
 741         if (!leafhdr.stale)
 742                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
 743         if (use_block == -1)
 744                 needbytes += sizeof(xfs_dir2_data_off_t);
 745
 746         /*
 747          * Now kill use_block if it refers to a missing block, so we
 748          * can use it as an indication of allocation needed.
 749          */
 750         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
 751                 use_block = -1;
 752         /*
 753          * If we don't have enough free bytes but we can make enough
 754          * by compacting out stale entries, we'll do that.
 755          */
 756         if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
 757             leafhdr.stale > 1)
 758                 compact = 1;
 759
 760         /*
 761          * Otherwise if we don't have enough free bytes we need to
 762          * convert to node form.
 763          */
 764         else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
 765                 /*
 766                  * Just checking or no space reservation, give up.
 767                  */
 768                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 769                                                         args->total == 0) {
 770                         xfs_trans_brelse(tp, lbp);
 771                         return XFS_ERROR(ENOSPC);
 772                 }
 773                 /*
 774                  * Convert to node form.
 775                  */
 776                 error = xfs_dir2_leaf_to_node(args, lbp);
 777                 if (error)
 778                         return error;
 779                 /*
 780                  * Then add the new entry.
 781                  */
 782                 return xfs_dir2_node_addname(args);
 783         }
 784         /*
 785          * Otherwise it will fit without compaction.
 786          */
 787         else
 788                 compact = 0;
 789         /*
 790          * If just checking, then it will fit unless we needed to allocate
 791          * a new data block.
 792          */
 793         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 794                 xfs_trans_brelse(tp, lbp);
 795                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
 796         }
 797         /*
 798          * If no allocations are allowed, return now before we've
 799          * changed anything.
 800          */
 801         if (args->total == 0 && use_block == -1) {
 802                 xfs_trans_brelse(tp, lbp);
 803                 return XFS_ERROR(ENOSPC);
 804         }
 805         /*
 806          * Need to compact the leaf entries, removing stale ones.
 807          * Leave one stale entry behind - the one closest to our
 808          * insertion index - and we'll shift that one to our insertion
 809          * point later.
 810          */
 811         if (compact) {
 812                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 813                         &highstale, &lfloglow, &lfloghigh);
 814         }
 815         /*
 816          * There are stale entries, so we'll need log-low and log-high
 817          * impossibly bad values later.
 818          */
 819         else if (leafhdr.stale) {
 820                 lfloglow = leafhdr.count;
 821                 lfloghigh = -1;
 822         }
 823         /*
 824          * If there was no data block space found, we need to allocate
 825          * a new one.
 826          */
 827         if (use_block == -1) {
 828                 /*
 829                  * Add the new data block.
 830                  */
 831                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 832                                 &use_block))) {
 833                         xfs_trans_brelse(tp, lbp);
 834                         return error;
 835                 }
 836                 /*
 837                  * Initialize the block.
 838                  */
 839                 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
 840                         xfs_trans_brelse(tp, lbp);
 841                         return error;
 842                 }
 843                 /*
 844                  * If we're adding a new data block on the end we need to
 845                  * extend the bests table.  Copy it up one entry.
 846                  */
 847                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
 848                         bestsp--;
 849                         memmove(&bestsp[0], &bestsp[1],
 850                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 851                         be32_add_cpu(&ltp->bestcount, 1);
 852                         xfs_dir3_leaf_log_tail(tp, lbp);
 853                         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
 854                 }
 855                 /*
 856                  * If we're filling in a previously empty block just log it.
 857                  */
 858                 else
 859                         xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
 860                 hdr = dbp->b_addr;
 861                 bf = xfs_dir3_data_bestfree_p(hdr);
 862                 bestsp[use_block] = bf[0].length;
 863                 grown = 1;
 864         } else {
 865                 /*
 866                  * Already had space in some data block.
 867                  * Just read that one in.
 868                  */
 869                 error = xfs_dir3_data_read(tp, dp,
 870                                            xfs_dir2_db_to_da(mp, use_block),
 871                                            -1, &dbp);
 872                 if (error) {
 873                         xfs_trans_brelse(tp, lbp);
 874                         return error;
 875                 }
 876                 hdr = dbp->b_addr;
 877                 bf = xfs_dir3_data_bestfree_p(hdr);
 878                 grown = 0;
 879         }
 880         /*
 881          * Point to the biggest freespace in our data block.
 882          */
 883         dup = (xfs_dir2_data_unused_t *)
 884               ((char *)hdr + be16_to_cpu(bf[0].offset));
 885         ASSERT(be16_to_cpu(dup->length) >= length);
 886         needscan = needlog = 0;
 887         /*
 888          * Mark the initial part of our freespace in use for the new entry.
 889          */
 890         xfs_dir2_data_use_free(tp, dbp, dup,
 891                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
 892                 &needlog, &needscan);
 893         /*
 894          * Initialize our new entry (at last).
 895          */
 896         dep = (xfs_dir2_data_entry_t *)dup;
 897         dep->inumber = cpu_to_be64(args->inumber);
 898         dep->namelen = args->namelen;
 899         memcpy(dep->name, args->name, dep->namelen);
 900         xfs_dir3_dirent_put_ftype(mp, dep, args->filetype);
 901         tagp = xfs_dir3_data_entry_tag_p(mp, dep);
 902         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 903         /*
 904          * Need to scan fix up the bestfree table.
 905          */
 906         if (needscan)
 907                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 908         /*
 909          * Need to log the data block's header.
 910          */
 911         if (needlog)
 912                 xfs_dir2_data_log_header(tp, dbp);
 913         xfs_dir2_data_log_entry(tp, dbp, dep);
 914         /*
 915          * If the bests table needs to be changed, do it.
 916          * Log the change unless we've already done that.
 917          */
 918         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
 919                 bestsp[use_block] = bf[0].length;
 920                 if (!grown)
 921                         xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
 922         }
 923
 924         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 925                                        highstale, &lfloglow, &lfloghigh);
 926
 927         /*
 928          * Fill in the new leaf entry.
 929          */
 930         lep->hashval = cpu_to_be32(args->hashval);
 931         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
 932                                 be16_to_cpu(*tagp)));
 933         /*
 934          * Log the leaf fields and give up the buffers.
 935          */
 936         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
 937         xfs_dir3_leaf_log_header(tp, lbp);
 938         xfs_dir3_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
 939         xfs_dir3_leaf_check(mp, lbp);
 940         xfs_dir3_data_check(dp, dbp);
 941         return 0;
 942 }
 943
 944 /*
 945  * Compact out any stale entries in the leaf.
 946  * Log the header and changed leaf entries, if any.
 947  */
 948 void
 949 xfs_dir3_leaf_compact(
 950         xfs_da_args_t   *args,          /* operation arguments */
 951         struct xfs_dir3_icleaf_hdr *leafhdr,
 952         struct xfs_buf  *bp)            /* leaf buffer */
 953 {
 954         int             from;           /* source leaf index */
 955         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 956         int             loglow;         /* first leaf entry to log */
 957         int             to;             /* target leaf index */
 958         struct xfs_dir2_leaf_entry *ents;
 959
 960         leaf = bp->b_addr;
 961         if (!leafhdr->stale)
 962                 return;
 963
 964         /*
 965          * Compress out the stale entries in place.
 966          */
 967         ents = xfs_dir3_leaf_ents_p(leaf);
 968         for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
 969                 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 970                         continue;
 971                 /*
 972                  * Only actually copy the entries that are different.
 973                  */
 974                 if (from > to) {
 975                         if (loglow == -1)
 976                                 loglow = to;
 977                         ents[to] = ents[from];
 978                 }
 979                 to++;
 980         }
 981         /*
 982          * Update and log the header, log the leaf entries.
 983          */
 984         ASSERT(leafhdr->stale == from - to);
 985         leafhdr->count -= leafhdr->stale;
 986         leafhdr->stale = 0;
 987
 988         xfs_dir3_leaf_hdr_to_disk(leaf, leafhdr);
 989         xfs_dir3_leaf_log_header(args->trans, bp);
 990         if (loglow != -1)
 991                 xfs_dir3_leaf_log_ents(args->trans, bp, loglow, to - 1);
 992 }
 993
 994 /*
 995  * Compact the leaf entries, removing stale ones.
 996  * Leave one stale entry behind - the one closest to our
 997  * insertion index - and the caller will shift that one to our insertion
 998  * point later.
 999  * Return new insertion index, where the remaining stale entry is,
1000  * and leaf logging indices.
1001  */
1002 void
1003 xfs_dir3_leaf_compact_x1(
1004         struct xfs_dir3_icleaf_hdr *leafhdr,
1005         struct xfs_dir2_leaf_entry *ents,
1006         int             *indexp,        /* insertion index */
1007         int             *lowstalep,     /* out: stale entry before us */
1008         int             *highstalep,    /* out: stale entry after us */
1009         int             *lowlogp,       /* out: low log index */
1010         int             *highlogp)      /* out: high log index */
1011 {
1012         int             from;           /* source copy index */
1013         int             highstale;      /* stale entry at/after index */
1014         int             index;          /* insertion index */
1015         int             keepstale;      /* source index of kept stale */
1016         int             lowstale;       /* stale entry before index */
1017         int             newindex=0;     /* new insertion index */
1018         int             to;             /* destination copy index */
1019
1020         ASSERT(leafhdr->stale > 1);
1021         index = *indexp;
1022
1023         xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
1024
1025         /*
1026          * Pick the better of lowstale and highstale.
1027          */
1028         if (lowstale >= 0 &&
1029             (highstale == leafhdr->count ||
1030              index - lowstale <= highstale - index))
1031                 keepstale = lowstale;
1032         else
1033                 keepstale = highstale;
1034         /*
1035          * Copy the entries in place, removing all the stale entries
1036          * except keepstale.
1037          */
1038         for (from = to = 0; from < leafhdr->count; from++) {
1039                 /*
1040                  * Notice the new value of index.
1041                  */
1042                 if (index == from)
1043                         newindex = to;
1044                 if (from != keepstale &&
1045                     ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1046                         if (from == to)
1047                                 *lowlogp = to;
1048                         continue;
1049                 }
1050                 /*
1051                  * Record the new keepstale value for the insertion.
1052                  */
1053                 if (from == keepstale)
1054                         lowstale = highstale = to;
1055                 /*
1056                  * Copy only the entries that have moved.
1057                  */
1058                 if (from > to)
1059                         ents[to] = ents[from];
1060                 to++;
1061         }
1062         ASSERT(from > to);
1063         /*
1064          * If the insertion point was past the last entry,
1065          * set the new insertion point accordingly.
1066          */
1067         if (index == from)
1068                 newindex = to;
1069         *indexp = newindex;
1070         /*
1071          * Adjust the leaf header values.
1072          */
1073         leafhdr->count -= from - to;
1074         leafhdr->stale = 1;
1075         /*
1076          * Remember the low/high stale value only in the "right"
1077          * direction.
1078          */
1079         if (lowstale >= newindex)
1080                 lowstale = -1;
1081         else
1082                 highstale = leafhdr->count;
1083         *highlogp = leafhdr->count - 1;
1084         *lowstalep = lowstale;
1085         *highstalep = highstale;
1086 }
1087
1088 /*
1089  * Log the bests entries indicated from a leaf1 block.
1090  */
1091 static void
1092 xfs_dir3_leaf_log_bests(
1093         xfs_trans_t             *tp,            /* transaction pointer */
1094         struct xfs_buf          *bp,            /* leaf buffer */
1095         int                     first,          /* first entry to log */
1096         int                     last)           /* last entry to log */
1097 {
1098         __be16                  *firstb;        /* pointer to first entry */
1099         __be16                  *lastb;         /* pointer to last entry */
1100         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1101         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1102
1103         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1104                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1105
1106         ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1107         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1108         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1109         xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1110                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1111 }
1112
1113 /*
1114  * Log the leaf entries indicated from a leaf1 or leafn block.
1115  */
1116 void
1117 xfs_dir3_leaf_log_ents(
1118         xfs_trans_t             *tp,            /* transaction pointer */
1119         struct xfs_buf          *bp,            /* leaf buffer */
1120         int                     first,          /* first entry to log */
1121         int                     last)           /* last entry to log */
1122 {
1123         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1124         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1125         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1126         struct xfs_dir2_leaf_entry *ents;
1127
1128         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1129                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1130                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1131                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1132
1133         ents = xfs_dir3_leaf_ents_p(leaf);
1134         firstlep = &ents[first];
1135         lastlep = &ents[last];
1136         xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1137                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1138 }
1139
1140 /*
1141  * Log the header of the leaf1 or leafn block.
1142  */
1143 void
1144 xfs_dir3_leaf_log_header(
1145         struct xfs_trans        *tp,
1146         struct xfs_buf          *bp)
1147 {
1148         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1149
1150         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1151                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1152                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1153                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1154
1155         xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1156                           xfs_dir3_leaf_hdr_size(leaf) - 1);
1157 }
1158
1159 /*
1160  * Log the tail of the leaf1 block.
1161  */
1162 STATIC void
1163 xfs_dir3_leaf_log_tail(
1164         struct xfs_trans        *tp,
1165         struct xfs_buf          *bp)
1166 {
1167         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1168         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1169         struct xfs_mount        *mp = tp->t_mountp;
1170
1171         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1172                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1173                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1174                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1175
1176         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1177         xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1178                 (uint)(mp->m_dirblksize - 1));
1179 }
1180
1181 /*
1182  * Look up the entry referred to by args in the leaf format directory.
1183  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1184  * is also used by the node-format code.
1185  */
1186 int
1187 xfs_dir2_leaf_lookup(
1188         xfs_da_args_t           *args)          /* operation arguments */
1189 {
1190         struct xfs_buf          *dbp;           /* data block buffer */
1191         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1192         xfs_inode_t             *dp;            /* incore directory inode */
1193         int                     error;          /* error return code */
1194         int                     index;          /* found entry index */
1195         struct xfs_buf          *lbp;           /* leaf buffer */
1196         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1197         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1198         xfs_trans_t             *tp;            /* transaction pointer */
1199         struct xfs_dir2_leaf_entry *ents;
1200
1201         trace_xfs_dir2_leaf_lookup(args);
1202
1203         /*
1204          * Look up name in the leaf block, returning both buffers and index.
1205          */
1206         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1207                 return error;
1208         }
1209         tp = args->trans;
1210         dp = args->dp;
1211         xfs_dir3_leaf_check(dp->i_mount, lbp);
1212         leaf = lbp->b_addr;
1213         ents = xfs_dir3_leaf_ents_p(leaf);
1214         /*
1215          * Get to the leaf entry and contained data entry address.
1216          */
1217         lep = &ents[index];
1218
1219         /*
1220          * Point to the data entry.
1221          */
1222         dep = (xfs_dir2_data_entry_t *)
1223               ((char *)dbp->b_addr +
1224                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1225         /*
1226          * Return the found inode number & CI name if appropriate
1227          */
1228         args->inumber = be64_to_cpu(dep->inumber);
1229         args->filetype = xfs_dir3_dirent_get_ftype(dp->i_mount, dep);
1230         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1231         xfs_trans_brelse(tp, dbp);
1232         xfs_trans_brelse(tp, lbp);
1233         return XFS_ERROR(error);
1234 }
1235
1236 /*
1237  * Look up name/hash in the leaf block.
1238  * Fill in indexp with the found index, and dbpp with the data buffer.
1239  * If not found dbpp will be NULL, and ENOENT comes back.
1240  * lbpp will always be filled in with the leaf buffer unless there's an error.
1241  */
1242 static int                                      /* error */
1243 xfs_dir2_leaf_lookup_int(
1244         xfs_da_args_t           *args,          /* operation arguments */
1245         struct xfs_buf          **lbpp,         /* out: leaf buffer */
1246         int                     *indexp,        /* out: index in leaf block */
1247         struct xfs_buf          **dbpp)         /* out: data buffer */
1248 {
1249         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1250         struct xfs_buf          *dbp = NULL;    /* data buffer */
1251         xfs_dir2_data_entry_t   *dep;           /* data entry */
1252         xfs_inode_t             *dp;            /* incore directory inode */
1253         int                     error;          /* error return code */
1254         int                     index;          /* index in leaf block */
1255         struct xfs_buf          *lbp;           /* leaf buffer */
1256         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1257         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1258         xfs_mount_t             *mp;            /* filesystem mount point */
1259         xfs_dir2_db_t           newdb;          /* new data block number */
1260         xfs_trans_t             *tp;            /* transaction pointer */
1261         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1262         enum xfs_dacmp          cmp;            /* name compare result */
1263         struct xfs_dir2_leaf_entry *ents;
1264         struct xfs_dir3_icleaf_hdr leafhdr;
1265
1266         dp = args->dp;
1267         tp = args->trans;
1268         mp = dp->i_mount;
1269
1270         error = xfs_dir3_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
1271         if (error)
1272                 return error;
1273
1274         *lbpp = lbp;
1275         leaf = lbp->b_addr;
1276         xfs_dir3_leaf_check(mp, lbp);
1277         ents = xfs_dir3_leaf_ents_p(leaf);
1278         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1279
1280         /*
1281          * Look for the first leaf entry with our hash value.
1282          */
1283         index = xfs_dir2_leaf_search_hash(args, lbp);
1284         /*
1285          * Loop over all the entries with the right hash value
1286          * looking to match the name.
1287          */
1288         for (lep = &ents[index];
1289              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1290              lep++, index++) {
1291                 /*
1292                  * Skip over stale leaf entries.
1293                  */
1294                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1295                         continue;
1296                 /*
1297                  * Get the new data block number.
1298                  */
1299                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1300                 /*
1301                  * If it's not the same as the old data block number,
1302                  * need to pitch the old one and read the new one.
1303                  */
1304                 if (newdb != curdb) {
1305                         if (dbp)
1306                                 xfs_trans_brelse(tp, dbp);
1307                         error = xfs_dir3_data_read(tp, dp,
1308                                                    xfs_dir2_db_to_da(mp, newdb),
1309                                                    -1, &dbp);
1310                         if (error) {
1311                                 xfs_trans_brelse(tp, lbp);
1312                                 return error;
1313                         }
1314                         curdb = newdb;
1315                 }
1316                 /*
1317                  * Point to the data entry.
1318                  */
1319                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1320                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1321                 /*
1322                  * Compare name and if it's an exact match, return the index
1323                  * and buffer. If it's the first case-insensitive match, store
1324                  * the index and buffer and continue looking for an exact match.
1325                  */
1326                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1327                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1328                         args->cmpresult = cmp;
1329                         *indexp = index;
1330                         /* case exact match: return the current buffer. */
1331                         if (cmp == XFS_CMP_EXACT) {
1332                                 *dbpp = dbp;
1333                                 return 0;
1334                         }
1335                         cidb = curdb;
1336                 }
1337         }
1338         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1339         /*
1340          * Here, we can only be doing a lookup (not a rename or remove).
1341          * If a case-insensitive match was found earlier, re-read the
1342          * appropriate data block if required and return it.
1343          */
1344         if (args->cmpresult == XFS_CMP_CASE) {
1345                 ASSERT(cidb != -1);
1346                 if (cidb != curdb) {
1347                         xfs_trans_brelse(tp, dbp);
1348                         error = xfs_dir3_data_read(tp, dp,
1349                                                    xfs_dir2_db_to_da(mp, cidb),
1350                                                    -1, &dbp);
1351                         if (error) {
1352                                 xfs_trans_brelse(tp, lbp);
1353                                 return error;
1354                         }
1355                 }
1356                 *dbpp = dbp;
1357                 return 0;
1358         }
1359         /*
1360          * No match found, return ENOENT.
1361          */
1362         ASSERT(cidb == -1);
1363         if (dbp)
1364                 xfs_trans_brelse(tp, dbp);
1365         xfs_trans_brelse(tp, lbp);
1366         return XFS_ERROR(ENOENT);
1367 }
1368
1369 /*
1370  * Remove an entry from a leaf format directory.
1371  */
1372 int                                             /* error */
1373 xfs_dir2_leaf_removename(
1374         xfs_da_args_t           *args)          /* operation arguments */
1375 {
1376         __be16                  *bestsp;        /* leaf block best freespace */
1377         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1378         xfs_dir2_db_t           db;             /* data block number */
1379         struct xfs_buf          *dbp;           /* data block buffer */
1380         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1381         xfs_inode_t             *dp;            /* incore directory inode */
1382         int                     error;          /* error return code */
1383         xfs_dir2_db_t           i;              /* temporary data block # */
1384         int                     index;          /* index into leaf entries */
1385         struct xfs_buf          *lbp;           /* leaf buffer */
1386         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1387         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1388         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1389         xfs_mount_t             *mp;            /* filesystem mount point */
1390         int                     needlog;        /* need to log data header */
1391         int                     needscan;       /* need to rescan data frees */
1392         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1393         xfs_trans_t             *tp;            /* transaction pointer */
1394         struct xfs_dir2_data_free *bf;          /* bestfree table */
1395         struct xfs_dir2_leaf_entry *ents;
1396         struct xfs_dir3_icleaf_hdr leafhdr;
1397
1398         trace_xfs_dir2_leaf_removename(args);
1399
1400         /*
1401          * Lookup the leaf entry, get the leaf and data blocks read in.
1402          */
1403         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1404                 return error;
1405         }
1406         dp = args->dp;
1407         tp = args->trans;
1408         mp = dp->i_mount;
1409         leaf = lbp->b_addr;
1410         hdr = dbp->b_addr;
1411         xfs_dir3_data_check(dp, dbp);
1412         bf = xfs_dir3_data_bestfree_p(hdr);
1413         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1414         ents = xfs_dir3_leaf_ents_p(leaf);
1415         /*
1416          * Point to the leaf entry, use that to point to the data entry.
1417          */
1418         lep = &ents[index];
1419         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1420         dep = (xfs_dir2_data_entry_t *)
1421               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1422         needscan = needlog = 0;
1423         oldbest = be16_to_cpu(bf[0].length);
1424         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1425         bestsp = xfs_dir2_leaf_bests_p(ltp);
1426         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1427         /*
1428          * Mark the former data entry unused.
1429          */
1430         xfs_dir2_data_make_free(tp, dbp,
1431                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1432                 xfs_dir3_data_entsize(mp, dep->namelen), &needlog, &needscan);
1433         /*
1434          * We just mark the leaf entry stale by putting a null in it.
1435          */
1436         leafhdr.stale++;
1437         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
1438         xfs_dir3_leaf_log_header(tp, lbp);
1439
1440         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1441         xfs_dir3_leaf_log_ents(tp, lbp, index, index);
1442
1443         /*
1444          * Scan the freespace in the data block again if necessary,
1445          * log the data block header if necessary.
1446          */
1447         if (needscan)
1448                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1449         if (needlog)
1450                 xfs_dir2_data_log_header(tp, dbp);
1451         /*
1452          * If the longest freespace in the data block has changed,
1453          * put the new value in the bests table and log that.
1454          */
1455         if (be16_to_cpu(bf[0].length) != oldbest) {
1456                 bestsp[db] = bf[0].length;
1457                 xfs_dir3_leaf_log_bests(tp, lbp, db, db);
1458         }
1459         xfs_dir3_data_check(dp, dbp);
1460         /*
1461          * If the data block is now empty then get rid of the data block.
1462          */
1463         if (be16_to_cpu(bf[0].length) ==
1464                         mp->m_dirblksize - xfs_dir3_data_entry_offset(hdr)) {
1465                 ASSERT(db != mp->m_dirdatablk);
1466                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1467                         /*
1468                          * Nope, can't get rid of it because it caused
1469                          * allocation of a bmap btree block to do so.
1470                          * Just go on, returning success, leaving the
1471                          * empty block in place.
1472                          */
1473                         if (error == ENOSPC && args->total == 0)
1474                                 error = 0;
1475                         xfs_dir3_leaf_check(mp, lbp);
1476                         return error;
1477                 }
1478                 dbp = NULL;
1479                 /*
1480                  * If this is the last data block then compact the
1481                  * bests table by getting rid of entries.
1482                  */
1483                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1484                         /*
1485                          * Look for the last active entry (i).
1486                          */
1487                         for (i = db - 1; i > 0; i--) {
1488                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1489                                         break;
1490                         }
1491                         /*
1492                          * Copy the table down so inactive entries at the
1493                          * end are removed.
1494                          */
1495                         memmove(&bestsp[db - i], bestsp,
1496                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1497                         be32_add_cpu(&ltp->bestcount, -(db - i));
1498                         xfs_dir3_leaf_log_tail(tp, lbp);
1499                         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1500                 } else
1501                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1502         }
1503         /*
1504          * If the data block was not the first one, drop it.
1505          */
1506         else if (db != mp->m_dirdatablk)
1507                 dbp = NULL;
1508
1509         xfs_dir3_leaf_check(mp, lbp);
1510         /*
1511          * See if we can convert to block form.
1512          */
1513         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1514 }
1515
1516 /*
1517  * Replace the inode number in a leaf format directory entry.
1518  */
1519 int                                             /* error */
1520 xfs_dir2_leaf_replace(
1521         xfs_da_args_t           *args)          /* operation arguments */
1522 {
1523         struct xfs_buf          *dbp;           /* data block buffer */
1524         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1525         xfs_inode_t             *dp;            /* incore directory inode */
1526         int                     error;          /* error return code */
1527         int                     index;          /* index of leaf entry */
1528         struct xfs_buf          *lbp;           /* leaf buffer */
1529         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1530         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1531         xfs_trans_t             *tp;            /* transaction pointer */
1532         struct xfs_dir2_leaf_entry *ents;
1533
1534         trace_xfs_dir2_leaf_replace(args);
1535
1536         /*
1537          * Look up the entry.
1538          */
1539         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1540                 return error;
1541         }
1542         dp = args->dp;
1543         leaf = lbp->b_addr;
1544         ents = xfs_dir3_leaf_ents_p(leaf);
1545         /*
1546          * Point to the leaf entry, get data address from it.
1547          */
1548         lep = &ents[index];
1549         /*
1550          * Point to the data entry.
1551          */
1552         dep = (xfs_dir2_data_entry_t *)
1553               ((char *)dbp->b_addr +
1554                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1555         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1556         /*
1557          * Put the new inode number in, log it.
1558          */
1559         dep->inumber = cpu_to_be64(args->inumber);
1560         xfs_dir3_dirent_put_ftype(dp->i_mount, dep, args->filetype);
1561         tp = args->trans;
1562         xfs_dir2_data_log_entry(tp, dbp, dep);
1563         xfs_dir3_leaf_check(dp->i_mount, lbp);
1564         xfs_trans_brelse(tp, lbp);
1565         return 0;
1566 }
1567
1568 /*
1569  * Return index in the leaf block (lbp) which is either the first
1570  * one with this hash value, or if there are none, the insert point
1571  * for that hash value.
1572  */
1573 int                                             /* index value */
1574 xfs_dir2_leaf_search_hash(
1575         xfs_da_args_t           *args,          /* operation arguments */
1576         struct xfs_buf          *lbp)           /* leaf buffer */
1577 {
1578         xfs_dahash_t            hash=0;         /* hash from this entry */
1579         xfs_dahash_t            hashwant;       /* hash value looking for */
1580         int                     high;           /* high leaf index */
1581         int                     low;            /* low leaf index */
1582         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1583         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1584         int                     mid=0;          /* current leaf index */
1585         struct xfs_dir2_leaf_entry *ents;
1586         struct xfs_dir3_icleaf_hdr leafhdr;
1587
1588         leaf = lbp->b_addr;
1589         ents = xfs_dir3_leaf_ents_p(leaf);
1590         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1591
1592         /*
1593          * Note, the table cannot be empty, so we have to go through the loop.
1594          * Binary search the leaf entries looking for our hash value.
1595          */
1596         for (lep = ents, low = 0, high = leafhdr.count - 1,
1597                 hashwant = args->hashval;
1598              low <= high; ) {
1599                 mid = (low + high) >> 1;
1600                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1601                         break;
1602                 if (hash < hashwant)
1603                         low = mid + 1;
1604                 else
1605                         high = mid - 1;
1606         }
1607         /*
1608          * Found one, back up through all the equal hash values.
1609          */
1610         if (hash == hashwant) {
1611                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1612                         mid--;
1613                 }
1614         }
1615         /*
1616          * Need to point to an entry higher than ours.
1617          */
1618         else if (hash < hashwant)
1619                 mid++;
1620         return mid;
1621 }
1622
1623 /*
1624  * Trim off a trailing data block.  We know it's empty since the leaf
1625  * freespace table says so.
1626  */
1627 int                                             /* error */
1628 xfs_dir2_leaf_trim_data(
1629         xfs_da_args_t           *args,          /* operation arguments */
1630         struct xfs_buf          *lbp,           /* leaf buffer */
1631         xfs_dir2_db_t           db)             /* data block number */
1632 {
1633         __be16                  *bestsp;        /* leaf bests table */
1634         struct xfs_buf          *dbp;           /* data block buffer */
1635         xfs_inode_t             *dp;            /* incore directory inode */
1636         int                     error;          /* error return value */
1637         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1638         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1639         xfs_mount_t             *mp;            /* filesystem mount point */
1640         xfs_trans_t             *tp;            /* transaction pointer */
1641
1642         dp = args->dp;
1643         mp = dp->i_mount;
1644         tp = args->trans;
1645         /*
1646          * Read the offending data block.  We need its buffer.
1647          */
1648         error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp);
1649         if (error)
1650                 return error;
1651
1652         leaf = lbp->b_addr;
1653         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1654
1655 #ifdef DEBUG
1656 {
1657         struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1658         struct xfs_dir2_data_free *bf = xfs_dir3_data_bestfree_p(hdr);
1659
1660         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1661                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1662         ASSERT(be16_to_cpu(bf[0].length) ==
1663                mp->m_dirblksize - xfs_dir3_data_entry_offset(hdr));
1664         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1665 }
1666 #endif
1667
1668         /*
1669          * Get rid of the data block.
1670          */
1671         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1672                 ASSERT(error != ENOSPC);
1673                 xfs_trans_brelse(tp, dbp);
1674                 return error;
1675         }
1676         /*
1677          * Eliminate the last bests entry from the table.
1678          */
1679         bestsp = xfs_dir2_leaf_bests_p(ltp);
1680         be32_add_cpu(&ltp->bestcount, -1);
1681         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1682         xfs_dir3_leaf_log_tail(tp, lbp);
1683         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1684         return 0;
1685 }
1686
1687 static inline size_t
1688 xfs_dir3_leaf_size(
1689         struct xfs_dir3_icleaf_hdr      *hdr,
1690         int                             counts)
1691 {
1692         int     entries;
1693         int     hdrsize;
1694
1695         entries = hdr->count - hdr->stale;
1696         if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1697             hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1698                 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1699         else
1700                 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1701
1702         return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1703                        + counts * sizeof(xfs_dir2_data_off_t)
1704                        + sizeof(xfs_dir2_leaf_tail_t);
1705 }
1706
1707 /*
1708  * Convert node form directory to leaf form directory.
1709  * The root of the node form dir needs to already be a LEAFN block.
1710  * Just return if we can't do anything.
1711  */
1712 int                                             /* error */
1713 xfs_dir2_node_to_leaf(
1714         xfs_da_state_t          *state)         /* directory operation state */
1715 {
1716         xfs_da_args_t           *args;          /* operation arguments */
1717         xfs_inode_t             *dp;            /* incore directory inode */
1718         int                     error;          /* error return code */
1719         struct xfs_buf          *fbp;           /* buffer for freespace block */
1720         xfs_fileoff_t           fo;             /* freespace file offset */
1721         xfs_dir2_free_t         *free;          /* freespace structure */
1722         struct xfs_buf          *lbp;           /* buffer for leaf block */
1723         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1724         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1725         xfs_mount_t             *mp;            /* filesystem mount point */
1726         int                     rval;           /* successful free trim? */
1727         xfs_trans_t             *tp;            /* transaction pointer */
1728         struct xfs_dir3_icleaf_hdr leafhdr;
1729         struct xfs_dir3_icfree_hdr freehdr;
1730
1731         /*
1732          * There's more than a leaf level in the btree, so there must
1733          * be multiple leafn blocks.  Give up.
1734          */
1735         if (state->path.active > 1)
1736                 return 0;
1737         args = state->args;
1738
1739         trace_xfs_dir2_node_to_leaf(args);
1740
1741         mp = state->mp;
1742         dp = args->dp;
1743         tp = args->trans;
1744         /*
1745          * Get the last offset in the file.
1746          */
1747         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1748                 return error;
1749         }
1750         fo -= mp->m_dirblkfsbs;
1751         /*
1752          * If there are freespace blocks other than the first one,
1753          * take this opportunity to remove trailing empty freespace blocks
1754          * that may have been left behind during no-space-reservation
1755          * operations.
1756          */
1757         while (fo > mp->m_dirfreeblk) {
1758                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1759                         return error;
1760                 }
1761                 if (rval)
1762                         fo -= mp->m_dirblkfsbs;
1763                 else
1764                         return 0;
1765         }
1766         /*
1767          * Now find the block just before the freespace block.
1768          */
1769         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1770                 return error;
1771         }
1772         /*
1773          * If it's not the single leaf block, give up.
1774          */
1775         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1776                 return 0;
1777         lbp = state->path.blk[0].bp;
1778         leaf = lbp->b_addr;
1779         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1780
1781         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1782                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1783
1784         /*
1785          * Read the freespace block.
1786          */
1787         error = xfs_dir2_free_read(tp, dp,  mp->m_dirfreeblk, &fbp);
1788         if (error)
1789                 return error;
1790         free = fbp->b_addr;
1791         xfs_dir3_free_hdr_from_disk(&freehdr, free);
1792
1793         ASSERT(!freehdr.firstdb);
1794
1795         /*
1796          * Now see if the leafn and free data will fit in a leaf1.
1797          * If not, release the buffer and give up.
1798          */
1799         if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > mp->m_dirblksize) {
1800                 xfs_trans_brelse(tp, fbp);
1801                 return 0;
1802         }
1803
1804         /*
1805          * If the leaf has any stale entries in it, compress them out.
1806          */
1807         if (leafhdr.stale)
1808                 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1809
1810         lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1811         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1812         leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1813                                         ? XFS_DIR2_LEAF1_MAGIC
1814                                         : XFS_DIR3_LEAF1_MAGIC;
1815
1816         /*
1817          * Set up the leaf tail from the freespace block.
1818          */
1819         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1820         ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1821
1822         /*
1823          * Set up the leaf bests table.
1824          */
1825         memcpy(xfs_dir2_leaf_bests_p(ltp), xfs_dir3_free_bests_p(mp, free),
1826                 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1827
1828         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
1829         xfs_dir3_leaf_log_header(tp, lbp);
1830         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1831         xfs_dir3_leaf_log_tail(tp, lbp);
1832         xfs_dir3_leaf_check(mp, lbp);
1833
1834         /*
1835          * Get rid of the freespace block.
1836          */
1837         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1838         if (error) {
1839                 /*
1840                  * This can't fail here because it can only happen when
1841                  * punching out the middle of an extent, and this is an
1842                  * isolated block.
1843                  */
1844                 ASSERT(error != ENOSPC);
1845                 return error;
1846         }
1847         fbp = NULL;
1848         /*
1849          * Now see if we can convert the single-leaf directory
1850          * down to a block form directory.
1851          * This routine always kills the dabuf for the leaf, so
1852          * eliminate it from the path.
1853          */
1854         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1855         state->path.blk[0].bp = NULL;
1856         return error;
1857 }