fs/xfs/xfs_dir2_leaf.c

   1 /*
   2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_inode.h"
  30 #include "xfs_bmap.h"
  31 #include "xfs_dir2.h"
  32 #include "xfs_dir2_priv.h"
  33 #include "xfs_error.h"
  34 #include "xfs_trace.h"
  35 #include "xfs_trans.h"
  36 #include "xfs_buf_item.h"
  37 #include "xfs_cksum.h"
  38
  39 /*
  40  * Local function declarations.
  41  */
  42 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
  43                                     int *indexp, struct xfs_buf **dbpp);
  44 static void xfs_dir3_leaf_log_bests(struct xfs_trans *tp, struct xfs_buf *bp,
  45                                     int first, int last);
  46 static void xfs_dir3_leaf_log_tail(struct xfs_trans *tp, struct xfs_buf *bp);
  47
  48 /*
  49  * Check the internal consistency of a leaf1 block.
  50  * Pop an assert if something is wrong.
  51  */
  52 #ifdef DEBUG
  53 #define xfs_dir3_leaf_check(mp, bp) \
  54 do { \
  55         if (!xfs_dir3_leaf1_check((mp), (bp))) \
  56                 ASSERT(0); \
  57 } while (0);
  58
  59 STATIC bool
  60 xfs_dir3_leaf1_check(
  61         struct xfs_mount        *mp,
  62         struct xfs_buf          *bp)
  63 {
  64         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  65         struct xfs_dir3_icleaf_hdr leafhdr;
  66
  67         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
  68
  69         if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
  70                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  71                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  72                         return false;
  73         } else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
  74                 return false;
  75
  76         return xfs_dir3_leaf_check_int(mp, &leafhdr, leaf);
  77 }
  78 #else
  79 #define xfs_dir3_leaf_check(mp, bp)
  80 #endif
  81
  82 void
  83 xfs_dir3_leaf_hdr_from_disk(
  84         struct xfs_dir3_icleaf_hdr      *to,
  85         struct xfs_dir2_leaf            *from)
  86 {
  87         if (from->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
  88             from->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC)) {
  89                 to->forw = be32_to_cpu(from->hdr.info.forw);
  90                 to->back = be32_to_cpu(from->hdr.info.back);
  91                 to->magic = be16_to_cpu(from->hdr.info.magic);
  92                 to->count = be16_to_cpu(from->hdr.count);
  93                 to->stale = be16_to_cpu(from->hdr.stale);
  94         } else {
  95                 struct xfs_dir3_leaf_hdr *hdr3 = (struct xfs_dir3_leaf_hdr *)from;
  96
  97                 to->forw = be32_to_cpu(hdr3->info.hdr.forw);
  98                 to->back = be32_to_cpu(hdr3->info.hdr.back);
  99                 to->magic = be16_to_cpu(hdr3->info.hdr.magic);
 100                 to->count = be16_to_cpu(hdr3->count);
 101                 to->stale = be16_to_cpu(hdr3->stale);
 102         }
 103
 104         ASSERT(to->magic == XFS_DIR2_LEAF1_MAGIC ||
 105                to->magic == XFS_DIR3_LEAF1_MAGIC ||
 106                to->magic == XFS_DIR2_LEAFN_MAGIC ||
 107                to->magic == XFS_DIR3_LEAFN_MAGIC);
 108 }
 109
 110 void
 111 xfs_dir3_leaf_hdr_to_disk(
 112         struct xfs_dir2_leaf            *to,
 113         struct xfs_dir3_icleaf_hdr      *from)
 114 {
 115         ASSERT(from->magic == XFS_DIR2_LEAF1_MAGIC ||
 116                from->magic == XFS_DIR3_LEAF1_MAGIC ||
 117                from->magic == XFS_DIR2_LEAFN_MAGIC ||
 118                from->magic == XFS_DIR3_LEAFN_MAGIC);
 119
 120         if (from->magic == XFS_DIR2_LEAF1_MAGIC ||
 121             from->magic == XFS_DIR2_LEAFN_MAGIC) {
 122                 to->hdr.info.forw = cpu_to_be32(from->forw);
 123                 to->hdr.info.back = cpu_to_be32(from->back);
 124                 to->hdr.info.magic = cpu_to_be16(from->magic);
 125                 to->hdr.count = cpu_to_be16(from->count);
 126                 to->hdr.stale = cpu_to_be16(from->stale);
 127         } else {
 128                 struct xfs_dir3_leaf_hdr *hdr3 = (struct xfs_dir3_leaf_hdr *)to;
 129
 130                 hdr3->info.hdr.forw = cpu_to_be32(from->forw);
 131                 hdr3->info.hdr.back = cpu_to_be32(from->back);
 132                 hdr3->info.hdr.magic = cpu_to_be16(from->magic);
 133                 hdr3->count = cpu_to_be16(from->count);
 134                 hdr3->stale = cpu_to_be16(from->stale);
 135         }
 136 }
 137
 138 bool
 139 xfs_dir3_leaf_check_int(
 140         struct xfs_mount        *mp,
 141         struct xfs_dir3_icleaf_hdr *hdr,
 142         struct xfs_dir2_leaf    *leaf)
 143 {
 144         struct xfs_dir2_leaf_entry *ents;
 145         xfs_dir2_leaf_tail_t    *ltp;
 146         int                     stale;
 147         int                     i;
 148
 149         ents = xfs_dir3_leaf_ents_p(leaf);
 150         ltp = xfs_dir2_leaf_tail_p(mp, 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 di_size.
 156          */
 157         if (hdr->count > xfs_dir3_max_leaf_ents(mp, leaf))
 158                 return false;
 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 *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
 164                 return false;
 165
 166         /* Check hash value order, count stale entries.  */
 167         for (i = stale = 0; i < hdr->count; i++) {
 168                 if (i + 1 < hdr->count) {
 169                         if (be32_to_cpu(ents[i].hashval) >
 170                                         be32_to_cpu(ents[i + 1].hashval))
 171                                 return false;
 172                 }
 173                 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 174                         stale++;
 175         }
 176         if (hdr->stale != stale)
 177                 return false;
 178         return true;
 179 }
 180
 181 /*
 182  * We verify the magic numbers before decoding the leaf header so that on debug
 183  * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
 184  * to incorrect magic numbers.
 185  */
 186 static bool
 187 xfs_dir3_leaf_verify(
 188         struct xfs_buf          *bp,
 189         __uint16_t              magic)
 190 {
 191         struct xfs_mount        *mp = bp->b_target->bt_mount;
 192         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 193         struct xfs_dir3_icleaf_hdr leafhdr;
 194
 195         ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
 196
 197         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 198                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 199                 __uint16_t              magic3;
 200
 201                 magic3 = (magic == XFS_DIR2_LEAF1_MAGIC) ? XFS_DIR3_LEAF1_MAGIC
 202                                                          : XFS_DIR3_LEAFN_MAGIC;
 203
 204                 if (leaf3->info.hdr.magic != cpu_to_be16(magic3))
 205                         return false;
 206                 if (!uuid_equal(&leaf3->info.uuid, &mp->m_sb.sb_uuid))
 207                         return false;
 208                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
 209                         return false;
 210         } else {
 211                 if (leaf->hdr.info.magic != cpu_to_be16(magic))
 212                         return false;
 213         }
 214
 215         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 216         return xfs_dir3_leaf_check_int(mp, &leafhdr, leaf);
 217 }
 218
 219 static void
 220 __read_verify(
 221         struct xfs_buf  *bp,
 222         __uint16_t      magic)
 223 {
 224         struct xfs_mount        *mp = bp->b_target->bt_mount;
 225
 226         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
 227              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
 228                                           XFS_DIR3_LEAF_CRC_OFF)) ||
 229             !xfs_dir3_leaf_verify(bp, magic)) {
 230                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 231                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 232         }
 233 }
 234
 235 static void
 236 __write_verify(
 237         struct xfs_buf  *bp,
 238         __uint16_t      magic)
 239 {
 240         struct xfs_mount        *mp = bp->b_target->bt_mount;
 241         struct xfs_buf_log_item *bip = bp->b_fspriv;
 242         struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
 243
 244         if (!xfs_dir3_leaf_verify(bp, magic)) {
 245                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 246                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 247                 return;
 248         }
 249
 250         if (!xfs_sb_version_hascrc(&mp->m_sb))
 251                 return;
 252
 253         if (bip)
 254                 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 255
 256         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_LEAF_CRC_OFF);
 257 }
 258
 259 static void
 260 xfs_dir3_leaf1_read_verify(
 261         struct xfs_buf  *bp)
 262 {
 263         __read_verify(bp, XFS_DIR2_LEAF1_MAGIC);
 264 }
 265
 266 static void
 267 xfs_dir3_leaf1_write_verify(
 268         struct xfs_buf  *bp)
 269 {
 270         __write_verify(bp, XFS_DIR2_LEAF1_MAGIC);
 271 }
 272
 273 static void
 274 xfs_dir3_leafn_read_verify(
 275         struct xfs_buf  *bp)
 276 {
 277         __read_verify(bp, XFS_DIR2_LEAFN_MAGIC);
 278 }
 279
 280 static void
 281 xfs_dir3_leafn_write_verify(
 282         struct xfs_buf  *bp)
 283 {
 284         __write_verify(bp, XFS_DIR2_LEAFN_MAGIC);
 285 }
 286
 287 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
 288         .verify_read = xfs_dir3_leaf1_read_verify,
 289         .verify_write = xfs_dir3_leaf1_write_verify,
 290 };
 291
 292 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
 293         .verify_read = xfs_dir3_leafn_read_verify,
 294         .verify_write = xfs_dir3_leafn_write_verify,
 295 };
 296
 297 static int
 298 xfs_dir3_leaf_read(
 299         struct xfs_trans        *tp,
 300         struct xfs_inode        *dp,
 301         xfs_dablk_t             fbno,
 302         xfs_daddr_t             mappedbno,
 303         struct xfs_buf          **bpp)
 304 {
 305         int                     err;
 306
 307         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 308                                 XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
 309         if (!err && tp)
 310                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
 311         return err;
 312 }
 313
 314 int
 315 xfs_dir3_leafn_read(
 316         struct xfs_trans        *tp,
 317         struct xfs_inode        *dp,
 318         xfs_dablk_t             fbno,
 319         xfs_daddr_t             mappedbno,
 320         struct xfs_buf          **bpp)
 321 {
 322         int                     err;
 323
 324         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 325                                 XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
 326         if (!err && tp)
 327                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
 328         return err;
 329 }
 330
 331 /*
 332  * Initialize a new leaf block, leaf1 or leafn magic accepted.
 333  */
 334 static void
 335 xfs_dir3_leaf_init(
 336         struct xfs_mount        *mp,
 337         struct xfs_trans        *tp,
 338         struct xfs_buf          *bp,
 339         xfs_ino_t               owner,
 340         __uint16_t              type)
 341 {
 342         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 343
 344         ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
 345
 346         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 347                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
 348
 349                 memset(leaf3, 0, sizeof(*leaf3));
 350
 351                 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
 352                                          ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
 353                                          : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 354                 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
 355                 leaf3->info.owner = cpu_to_be64(owner);
 356                 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_uuid);
 357         } else {
 358                 memset(leaf, 0, sizeof(*leaf));
 359                 leaf->hdr.info.magic = cpu_to_be16(type);
 360         }
 361
 362         /*
 363          * If it's a leaf-format directory initialize the tail.
 364          * Caller is responsible for initialising the bests table.
 365          */
 366         if (type == XFS_DIR2_LEAF1_MAGIC) {
 367                 struct xfs_dir2_leaf_tail *ltp;
 368
 369                 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 370                 ltp->bestcount = 0;
 371                 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
 372                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
 373         } else {
 374                 bp->b_ops = &xfs_dir3_leafn_buf_ops;
 375                 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
 376         }
 377 }
 378
 379 int
 380 xfs_dir3_leaf_get_buf(
 381         xfs_da_args_t           *args,
 382         xfs_dir2_db_t           bno,
 383         struct xfs_buf          **bpp,
 384         __uint16_t              magic)
 385 {
 386         struct xfs_inode        *dp = args->dp;
 387         struct xfs_trans        *tp = args->trans;
 388         struct xfs_mount        *mp = dp->i_mount;
 389         struct xfs_buf          *bp;
 390         int                     error;
 391
 392         ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
 393         ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
 394                bno < XFS_DIR2_FREE_FIRSTDB(mp));
 395
 396         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
 397                                XFS_DATA_FORK);
 398         if (error)
 399                 return error;
 400
 401         xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
 402         xfs_dir3_leaf_log_header(tp, bp);
 403         if (magic == XFS_DIR2_LEAF1_MAGIC)
 404                 xfs_dir3_leaf_log_tail(tp, bp);
 405         *bpp = bp;
 406         return 0;
 407 }
 408
 409 /*
 410  * Convert a block form directory to a leaf form directory.
 411  */
 412 int                                             /* error */
 413 xfs_dir2_block_to_leaf(
 414         xfs_da_args_t           *args,          /* operation arguments */
 415         struct xfs_buf          *dbp)           /* input block's buffer */
 416 {
 417         __be16                  *bestsp;        /* leaf's bestsp entries */
 418         xfs_dablk_t             blkno;          /* leaf block's bno */
 419         xfs_dir2_data_hdr_t     *hdr;           /* block header */
 420         xfs_dir2_leaf_entry_t   *blp;           /* block's leaf entries */
 421         xfs_dir2_block_tail_t   *btp;           /* block's tail */
 422         xfs_inode_t             *dp;            /* incore directory inode */
 423         int                     error;          /* error return code */
 424         struct xfs_buf          *lbp;           /* leaf block's buffer */
 425         xfs_dir2_db_t           ldb;            /* leaf block's bno */
 426         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 427         xfs_dir2_leaf_tail_t    *ltp;           /* leaf's tail */
 428         xfs_mount_t             *mp;            /* filesystem mount point */
 429         int                     needlog;        /* need to log block header */
 430         int                     needscan;       /* need to rescan bestfree */
 431         xfs_trans_t             *tp;            /* transaction pointer */
 432         struct xfs_dir2_data_free *bf;
 433         struct xfs_dir2_leaf_entry *ents;
 434         struct xfs_dir3_icleaf_hdr leafhdr;
 435
 436         trace_xfs_dir2_block_to_leaf(args);
 437
 438         dp = args->dp;
 439         mp = dp->i_mount;
 440         tp = args->trans;
 441         /*
 442          * Add the leaf block to the inode.
 443          * This interface will only put blocks in the leaf/node range.
 444          * Since that's empty now, we'll get the root (block 0 in range).
 445          */
 446         if ((error = xfs_da_grow_inode(args, &blkno))) {
 447                 return error;
 448         }
 449         ldb = xfs_dir2_da_to_db(mp, blkno);
 450         ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
 451         /*
 452          * Initialize the leaf block, get a buffer for it.
 453          */
 454         error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
 455         if (error)
 456                 return error;
 457
 458         leaf = lbp->b_addr;
 459         hdr = dbp->b_addr;
 460         xfs_dir3_data_check(dp, dbp);
 461         btp = xfs_dir2_block_tail_p(mp, hdr);
 462         blp = xfs_dir2_block_leaf_p(btp);
 463         bf = xfs_dir3_data_bestfree_p(hdr);
 464         ents = xfs_dir3_leaf_ents_p(leaf);
 465
 466         /*
 467          * Set the counts in the leaf header.
 468          */
 469         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 470         leafhdr.count = be32_to_cpu(btp->count);
 471         leafhdr.stale = be32_to_cpu(btp->stale);
 472         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
 473         xfs_dir3_leaf_log_header(tp, lbp);
 474
 475         /*
 476          * Could compact these but I think we always do the conversion
 477          * after squeezing out stale entries.
 478          */
 479         memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
 480         xfs_dir3_leaf_log_ents(tp, lbp, 0, leafhdr.count - 1);
 481         needscan = 0;
 482         needlog = 1;
 483         /*
 484          * Make the space formerly occupied by the leaf entries and block
 485          * tail be free.
 486          */
 487         xfs_dir2_data_make_free(tp, dbp,
 488                 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
 489                 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
 490                                        (char *)blp),
 491                 &needlog, &needscan);
 492         /*
 493          * Fix up the block header, make it a data block.
 494          */
 495         dbp->b_ops = &xfs_dir3_data_buf_ops;
 496         xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
 497         if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
 498                 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 499         else
 500                 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 501
 502         if (needscan)
 503                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 504         /*
 505          * Set up leaf tail and bests table.
 506          */
 507         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 508         ltp->bestcount = cpu_to_be32(1);
 509         bestsp = xfs_dir2_leaf_bests_p(ltp);
 510         bestsp[0] =  bf[0].length;
 511         /*
 512          * Log the data header and leaf bests table.
 513          */
 514         if (needlog)
 515                 xfs_dir2_data_log_header(tp, dbp);
 516         xfs_dir3_leaf_check(mp, lbp);
 517         xfs_dir3_data_check(dp, dbp);
 518         xfs_dir3_leaf_log_bests(tp, lbp, 0, 0);
 519         return 0;
 520 }
 521
 522 STATIC void
 523 xfs_dir3_leaf_find_stale(
 524         struct xfs_dir3_icleaf_hdr *leafhdr,
 525         struct xfs_dir2_leaf_entry *ents,
 526         int                     index,
 527         int                     *lowstale,
 528         int                     *highstale)
 529 {
 530         /*
 531          * Find the first stale entry before our index, if any.
 532          */
 533         for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
 534                 if (ents[*lowstale].address ==
 535                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 536                         break;
 537         }
 538
 539         /*
 540          * Find the first stale entry at or after our index, if any.
 541          * Stop if the result would require moving more entries than using
 542          * lowstale.
 543          */
 544         for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
 545                 if (ents[*highstale].address ==
 546                     cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 547                         break;
 548                 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
 549                         break;
 550         }
 551 }
 552
 553 struct xfs_dir2_leaf_entry *
 554 xfs_dir3_leaf_find_entry(
 555         struct xfs_dir3_icleaf_hdr *leafhdr,
 556         struct xfs_dir2_leaf_entry *ents,
 557         int                     index,          /* leaf table position */
 558         int                     compact,        /* need to compact leaves */
 559         int                     lowstale,       /* index of prev stale leaf */
 560         int                     highstale,      /* index of next stale leaf */
 561         int                     *lfloglow,      /* low leaf logging index */
 562         int                     *lfloghigh)     /* high leaf logging index */
 563 {
 564         if (!leafhdr->stale) {
 565                 xfs_dir2_leaf_entry_t   *lep;   /* leaf entry table pointer */
 566
 567                 /*
 568                  * Now we need to make room to insert the leaf entry.
 569                  *
 570                  * If there are no stale entries, just insert a hole at index.
 571                  */
 572                 lep = &ents[index];
 573                 if (index < leafhdr->count)
 574                         memmove(lep + 1, lep,
 575                                 (leafhdr->count - index) * sizeof(*lep));
 576
 577                 /*
 578                  * Record low and high logging indices for the leaf.
 579                  */
 580                 *lfloglow = index;
 581                 *lfloghigh = leafhdr->count++;
 582                 return lep;
 583         }
 584
 585         /*
 586          * There are stale entries.
 587          *
 588          * We will use one of them for the new entry.  It's probably not at
 589          * the right location, so we'll have to shift some up or down first.
 590          *
 591          * If we didn't compact before, we need to find the nearest stale
 592          * entries before and after our insertion point.
 593          */
 594         if (compact == 0)
 595                 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
 596                                          &lowstale, &highstale);
 597
 598         /*
 599          * If the low one is better, use it.
 600          */
 601         if (lowstale >= 0 &&
 602             (highstale == leafhdr->count ||
 603              index - lowstale - 1 < highstale - index)) {
 604                 ASSERT(index - lowstale - 1 >= 0);
 605                 ASSERT(ents[lowstale].address ==
 606                        cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 607
 608                 /*
 609                  * Copy entries up to cover the stale entry and make room
 610                  * for the new entry.
 611                  */
 612                 if (index - lowstale - 1 > 0) {
 613                         memmove(&ents[lowstale], &ents[lowstale + 1],
 614                                 (index - lowstale - 1) *
 615                                         sizeof(xfs_dir2_leaf_entry_t));
 616                 }
 617                 *lfloglow = MIN(lowstale, *lfloglow);
 618                 *lfloghigh = MAX(index - 1, *lfloghigh);
 619                 leafhdr->stale--;
 620                 return &ents[index - 1];
 621         }
 622
 623         /*
 624          * The high one is better, so use that one.
 625          */
 626         ASSERT(highstale - index >= 0);
 627         ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
 628
 629         /*
 630          * Copy entries down to cover the stale entry and make room for the
 631          * new entry.
 632          */
 633         if (highstale - index > 0) {
 634                 memmove(&ents[index + 1], &ents[index],
 635                         (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
 636         }
 637         *lfloglow = MIN(index, *lfloglow);
 638         *lfloghigh = MAX(highstale, *lfloghigh);
 639         leafhdr->stale--;
 640         return &ents[index];
 641 }
 642
 643 /*
 644  * Add an entry to a leaf form directory.
 645  */
 646 int                                             /* error */
 647 xfs_dir2_leaf_addname(
 648         xfs_da_args_t           *args)          /* operation arguments */
 649 {
 650         __be16                  *bestsp;        /* freespace table in leaf */
 651         int                     compact;        /* need to compact leaves */
 652         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 653         struct xfs_buf          *dbp;           /* data block buffer */
 654         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 655         xfs_inode_t             *dp;            /* incore directory inode */
 656         xfs_dir2_data_unused_t  *dup;           /* data unused entry */
 657         int                     error;          /* error return value */
 658         int                     grown;          /* allocated new data block */
 659         int                     highstale;      /* index of next stale leaf */
 660         int                     i;              /* temporary, index */
 661         int                     index;          /* leaf table position */
 662         struct xfs_buf          *lbp;           /* leaf's buffer */
 663         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 664         int                     length;         /* length of new entry */
 665         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry table pointer */
 666         int                     lfloglow;       /* low leaf logging index */
 667         int                     lfloghigh;      /* high leaf logging index */
 668         int                     lowstale;       /* index of prev stale leaf */
 669         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail pointer */
 670         xfs_mount_t             *mp;            /* filesystem mount point */
 671         int                     needbytes;      /* leaf block bytes needed */
 672         int                     needlog;        /* need to log data header */
 673         int                     needscan;       /* need to rescan data free */
 674         __be16                  *tagp;          /* end of data entry */
 675         xfs_trans_t             *tp;            /* transaction pointer */
 676         xfs_dir2_db_t           use_block;      /* data block number */
 677         struct xfs_dir2_data_free *bf;          /* bestfree table */
 678         struct xfs_dir2_leaf_entry *ents;
 679         struct xfs_dir3_icleaf_hdr leafhdr;
 680
 681         trace_xfs_dir2_leaf_addname(args);
 682
 683         dp = args->dp;
 684         tp = args->trans;
 685         mp = dp->i_mount;
 686
 687         error = xfs_dir3_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
 688         if (error)
 689                 return error;
 690
 691         /*
 692          * Look up the entry by hash value and name.
 693          * We know it's not there, our caller has already done a lookup.
 694          * So the index is of the entry to insert in front of.
 695          * But if there are dup hash values the index is of the first of those.
 696          */
 697         index = xfs_dir2_leaf_search_hash(args, lbp);
 698         leaf = lbp->b_addr;
 699         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 700         ents = xfs_dir3_leaf_ents_p(leaf);
 701         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 702         bestsp = xfs_dir2_leaf_bests_p(ltp);
 703         length = xfs_dir3_data_entsize(mp, args->namelen);
 704
 705         /*
 706          * See if there are any entries with the same hash value
 707          * and space in their block for the new entry.
 708          * This is good because it puts multiple same-hash value entries
 709          * in a data block, improving the lookup of those entries.
 710          */
 711         for (use_block = -1, lep = &ents[index];
 712              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 713              index++, lep++) {
 714                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 715                         continue;
 716                 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 717                 ASSERT(i < be32_to_cpu(ltp->bestcount));
 718                 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
 719                 if (be16_to_cpu(bestsp[i]) >= length) {
 720                         use_block = i;
 721                         break;
 722                 }
 723         }
 724         /*
 725          * Didn't find a block yet, linear search all the data blocks.
 726          */
 727         if (use_block == -1) {
 728                 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
 729                         /*
 730                          * Remember a block we see that's missing.
 731                          */
 732                         if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
 733                             use_block == -1)
 734                                 use_block = i;
 735                         else if (be16_to_cpu(bestsp[i]) >= length) {
 736                                 use_block = i;
 737                                 break;
 738                         }
 739                 }
 740         }
 741         /*
 742          * How many bytes do we need in the leaf block?
 743          */
 744         needbytes = 0;
 745         if (!leafhdr.stale)
 746                 needbytes += sizeof(xfs_dir2_leaf_entry_t);
 747         if (use_block == -1)
 748                 needbytes += sizeof(xfs_dir2_data_off_t);
 749
 750         /*
 751          * Now kill use_block if it refers to a missing block, so we
 752          * can use it as an indication of allocation needed.
 753          */
 754         if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
 755                 use_block = -1;
 756         /*
 757          * If we don't have enough free bytes but we can make enough
 758          * by compacting out stale entries, we'll do that.
 759          */
 760         if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
 761             leafhdr.stale > 1)
 762                 compact = 1;
 763
 764         /*
 765          * Otherwise if we don't have enough free bytes we need to
 766          * convert to node form.
 767          */
 768         else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
 769                 /*
 770                  * Just checking or no space reservation, give up.
 771                  */
 772                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
 773                                                         args->total == 0) {
 774                         xfs_trans_brelse(tp, lbp);
 775                         return XFS_ERROR(ENOSPC);
 776                 }
 777                 /*
 778                  * Convert to node form.
 779                  */
 780                 error = xfs_dir2_leaf_to_node(args, lbp);
 781                 if (error)
 782                         return error;
 783                 /*
 784                  * Then add the new entry.
 785                  */
 786                 return xfs_dir2_node_addname(args);
 787         }
 788         /*
 789          * Otherwise it will fit without compaction.
 790          */
 791         else
 792                 compact = 0;
 793         /*
 794          * If just checking, then it will fit unless we needed to allocate
 795          * a new data block.
 796          */
 797         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
 798                 xfs_trans_brelse(tp, lbp);
 799                 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
 800         }
 801         /*
 802          * If no allocations are allowed, return now before we've
 803          * changed anything.
 804          */
 805         if (args->total == 0 && use_block == -1) {
 806                 xfs_trans_brelse(tp, lbp);
 807                 return XFS_ERROR(ENOSPC);
 808         }
 809         /*
 810          * Need to compact the leaf entries, removing stale ones.
 811          * Leave one stale entry behind - the one closest to our
 812          * insertion index - and we'll shift that one to our insertion
 813          * point later.
 814          */
 815         if (compact) {
 816                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 817                         &highstale, &lfloglow, &lfloghigh);
 818         }
 819         /*
 820          * There are stale entries, so we'll need log-low and log-high
 821          * impossibly bad values later.
 822          */
 823         else if (leafhdr.stale) {
 824                 lfloglow = leafhdr.count;
 825                 lfloghigh = -1;
 826         }
 827         /*
 828          * If there was no data block space found, we need to allocate
 829          * a new one.
 830          */
 831         if (use_block == -1) {
 832                 /*
 833                  * Add the new data block.
 834                  */
 835                 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
 836                                 &use_block))) {
 837                         xfs_trans_brelse(tp, lbp);
 838                         return error;
 839                 }
 840                 /*
 841                  * Initialize the block.
 842                  */
 843                 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
 844                         xfs_trans_brelse(tp, lbp);
 845                         return error;
 846                 }
 847                 /*
 848                  * If we're adding a new data block on the end we need to
 849                  * extend the bests table.  Copy it up one entry.
 850                  */
 851                 if (use_block >= be32_to_cpu(ltp->bestcount)) {
 852                         bestsp--;
 853                         memmove(&bestsp[0], &bestsp[1],
 854                                 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
 855                         be32_add_cpu(&ltp->bestcount, 1);
 856                         xfs_dir3_leaf_log_tail(tp, lbp);
 857                         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
 858                 }
 859                 /*
 860                  * If we're filling in a previously empty block just log it.
 861                  */
 862                 else
 863                         xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
 864                 hdr = dbp->b_addr;
 865                 bf = xfs_dir3_data_bestfree_p(hdr);
 866                 bestsp[use_block] = bf[0].length;
 867                 grown = 1;
 868         } else {
 869                 /*
 870                  * Already had space in some data block.
 871                  * Just read that one in.
 872                  */
 873                 error = xfs_dir3_data_read(tp, dp,
 874                                            xfs_dir2_db_to_da(mp, use_block),
 875                                            -1, &dbp);
 876                 if (error) {
 877                         xfs_trans_brelse(tp, lbp);
 878                         return error;
 879                 }
 880                 hdr = dbp->b_addr;
 881                 bf = xfs_dir3_data_bestfree_p(hdr);
 882                 grown = 0;
 883         }
 884         /*
 885          * Point to the biggest freespace in our data block.
 886          */
 887         dup = (xfs_dir2_data_unused_t *)
 888               ((char *)hdr + be16_to_cpu(bf[0].offset));
 889         ASSERT(be16_to_cpu(dup->length) >= length);
 890         needscan = needlog = 0;
 891         /*
 892          * Mark the initial part of our freespace in use for the new entry.
 893          */
 894         xfs_dir2_data_use_free(tp, dbp, dup,
 895                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
 896                 &needlog, &needscan);
 897         /*
 898          * Initialize our new entry (at last).
 899          */
 900         dep = (xfs_dir2_data_entry_t *)dup;
 901         dep->inumber = cpu_to_be64(args->inumber);
 902         dep->namelen = args->namelen;
 903         memcpy(dep->name, args->name, dep->namelen);
 904         xfs_dir3_dirent_put_ftype(mp, dep, args->filetype);
 905         tagp = xfs_dir3_data_entry_tag_p(mp, dep);
 906         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
 907         /*
 908          * Need to scan fix up the bestfree table.
 909          */
 910         if (needscan)
 911                 xfs_dir2_data_freescan(mp, hdr, &needlog);
 912         /*
 913          * Need to log the data block's header.
 914          */
 915         if (needlog)
 916                 xfs_dir2_data_log_header(tp, dbp);
 917         xfs_dir2_data_log_entry(tp, dbp, dep);
 918         /*
 919          * If the bests table needs to be changed, do it.
 920          * Log the change unless we've already done that.
 921          */
 922         if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
 923                 bestsp[use_block] = bf[0].length;
 924                 if (!grown)
 925                         xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
 926         }
 927
 928         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 929                                        highstale, &lfloglow, &lfloghigh);
 930
 931         /*
 932          * Fill in the new leaf entry.
 933          */
 934         lep->hashval = cpu_to_be32(args->hashval);
 935         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
 936                                 be16_to_cpu(*tagp)));
 937         /*
 938          * Log the leaf fields and give up the buffers.
 939          */
 940         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
 941         xfs_dir3_leaf_log_header(tp, lbp);
 942         xfs_dir3_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
 943         xfs_dir3_leaf_check(mp, lbp);
 944         xfs_dir3_data_check(dp, dbp);
 945         return 0;
 946 }
 947
 948 /*
 949  * Compact out any stale entries in the leaf.
 950  * Log the header and changed leaf entries, if any.
 951  */
 952 void
 953 xfs_dir3_leaf_compact(
 954         xfs_da_args_t   *args,          /* operation arguments */
 955         struct xfs_dir3_icleaf_hdr *leafhdr,
 956         struct xfs_buf  *bp)            /* leaf buffer */
 957 {
 958         int             from;           /* source leaf index */
 959         xfs_dir2_leaf_t *leaf;          /* leaf structure */
 960         int             loglow;         /* first leaf entry to log */
 961         int             to;             /* target leaf index */
 962         struct xfs_dir2_leaf_entry *ents;
 963
 964         leaf = bp->b_addr;
 965         if (!leafhdr->stale)
 966                 return;
 967
 968         /*
 969          * Compress out the stale entries in place.
 970          */
 971         ents = xfs_dir3_leaf_ents_p(leaf);
 972         for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
 973                 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 974                         continue;
 975                 /*
 976                  * Only actually copy the entries that are different.
 977                  */
 978                 if (from > to) {
 979                         if (loglow == -1)
 980                                 loglow = to;
 981                         ents[to] = ents[from];
 982                 }
 983                 to++;
 984         }
 985         /*
 986          * Update and log the header, log the leaf entries.
 987          */
 988         ASSERT(leafhdr->stale == from - to);
 989         leafhdr->count -= leafhdr->stale;
 990         leafhdr->stale = 0;
 991
 992         xfs_dir3_leaf_hdr_to_disk(leaf, leafhdr);
 993         xfs_dir3_leaf_log_header(args->trans, bp);
 994         if (loglow != -1)
 995                 xfs_dir3_leaf_log_ents(args->trans, bp, loglow, to - 1);
 996 }
 997
 998 /*
 999  * Compact the leaf entries, removing stale ones.
1000  * Leave one stale entry behind - the one closest to our
1001  * insertion index - and the caller will shift that one to our insertion
1002  * point later.
1003  * Return new insertion index, where the remaining stale entry is,
1004  * and leaf logging indices.
1005  */
1006 void
1007 xfs_dir3_leaf_compact_x1(
1008         struct xfs_dir3_icleaf_hdr *leafhdr,
1009         struct xfs_dir2_leaf_entry *ents,
1010         int             *indexp,        /* insertion index */
1011         int             *lowstalep,     /* out: stale entry before us */
1012         int             *highstalep,    /* out: stale entry after us */
1013         int             *lowlogp,       /* out: low log index */
1014         int             *highlogp)      /* out: high log index */
1015 {
1016         int             from;           /* source copy index */
1017         int             highstale;      /* stale entry at/after index */
1018         int             index;          /* insertion index */
1019         int             keepstale;      /* source index of kept stale */
1020         int             lowstale;       /* stale entry before index */
1021         int             newindex=0;     /* new insertion index */
1022         int             to;             /* destination copy index */
1023
1024         ASSERT(leafhdr->stale > 1);
1025         index = *indexp;
1026
1027         xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
1028
1029         /*
1030          * Pick the better of lowstale and highstale.
1031          */
1032         if (lowstale >= 0 &&
1033             (highstale == leafhdr->count ||
1034              index - lowstale <= highstale - index))
1035                 keepstale = lowstale;
1036         else
1037                 keepstale = highstale;
1038         /*
1039          * Copy the entries in place, removing all the stale entries
1040          * except keepstale.
1041          */
1042         for (from = to = 0; from < leafhdr->count; from++) {
1043                 /*
1044                  * Notice the new value of index.
1045                  */
1046                 if (index == from)
1047                         newindex = to;
1048                 if (from != keepstale &&
1049                     ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1050                         if (from == to)
1051                                 *lowlogp = to;
1052                         continue;
1053                 }
1054                 /*
1055                  * Record the new keepstale value for the insertion.
1056                  */
1057                 if (from == keepstale)
1058                         lowstale = highstale = to;
1059                 /*
1060                  * Copy only the entries that have moved.
1061                  */
1062                 if (from > to)
1063                         ents[to] = ents[from];
1064                 to++;
1065         }
1066         ASSERT(from > to);
1067         /*
1068          * If the insertion point was past the last entry,
1069          * set the new insertion point accordingly.
1070          */
1071         if (index == from)
1072                 newindex = to;
1073         *indexp = newindex;
1074         /*
1075          * Adjust the leaf header values.
1076          */
1077         leafhdr->count -= from - to;
1078         leafhdr->stale = 1;
1079         /*
1080          * Remember the low/high stale value only in the "right"
1081          * direction.
1082          */
1083         if (lowstale >= newindex)
1084                 lowstale = -1;
1085         else
1086                 highstale = leafhdr->count;
1087         *highlogp = leafhdr->count - 1;
1088         *lowstalep = lowstale;
1089         *highstalep = highstale;
1090 }
1091
1092 /*
1093  * Log the bests entries indicated from a leaf1 block.
1094  */
1095 static void
1096 xfs_dir3_leaf_log_bests(
1097         xfs_trans_t             *tp,            /* transaction pointer */
1098         struct xfs_buf          *bp,            /* leaf buffer */
1099         int                     first,          /* first entry to log */
1100         int                     last)           /* last entry to log */
1101 {
1102         __be16                  *firstb;        /* pointer to first entry */
1103         __be16                  *lastb;         /* pointer to last entry */
1104         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1105         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1106
1107         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1108                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1109
1110         ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1111         firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1112         lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1113         xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1114                 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1115 }
1116
1117 /*
1118  * Log the leaf entries indicated from a leaf1 or leafn block.
1119  */
1120 void
1121 xfs_dir3_leaf_log_ents(
1122         xfs_trans_t             *tp,            /* transaction pointer */
1123         struct xfs_buf          *bp,            /* leaf buffer */
1124         int                     first,          /* first entry to log */
1125         int                     last)           /* last entry to log */
1126 {
1127         xfs_dir2_leaf_entry_t   *firstlep;      /* pointer to first entry */
1128         xfs_dir2_leaf_entry_t   *lastlep;       /* pointer to last entry */
1129         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1130         struct xfs_dir2_leaf_entry *ents;
1131
1132         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1133                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1134                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1135                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1136
1137         ents = xfs_dir3_leaf_ents_p(leaf);
1138         firstlep = &ents[first];
1139         lastlep = &ents[last];
1140         xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1141                 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1142 }
1143
1144 /*
1145  * Log the header of the leaf1 or leafn block.
1146  */
1147 void
1148 xfs_dir3_leaf_log_header(
1149         struct xfs_trans        *tp,
1150         struct xfs_buf          *bp)
1151 {
1152         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1153
1154         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1155                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1156                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1157                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1158
1159         xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1160                           xfs_dir3_leaf_hdr_size(leaf) - 1);
1161 }
1162
1163 /*
1164  * Log the tail of the leaf1 block.
1165  */
1166 STATIC void
1167 xfs_dir3_leaf_log_tail(
1168         struct xfs_trans        *tp,
1169         struct xfs_buf          *bp)
1170 {
1171         struct xfs_dir2_leaf    *leaf = bp->b_addr;
1172         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1173         struct xfs_mount        *mp = tp->t_mountp;
1174
1175         ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1176                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1177                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1178                leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1179
1180         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1181         xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1182                 (uint)(mp->m_dirblksize - 1));
1183 }
1184
1185 /*
1186  * Look up the entry referred to by args in the leaf format directory.
1187  * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1188  * is also used by the node-format code.
1189  */
1190 int
1191 xfs_dir2_leaf_lookup(
1192         xfs_da_args_t           *args)          /* operation arguments */
1193 {
1194         struct xfs_buf          *dbp;           /* data block buffer */
1195         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1196         xfs_inode_t             *dp;            /* incore directory inode */
1197         int                     error;          /* error return code */
1198         int                     index;          /* found entry index */
1199         struct xfs_buf          *lbp;           /* leaf buffer */
1200         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1201         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1202         xfs_trans_t             *tp;            /* transaction pointer */
1203         struct xfs_dir2_leaf_entry *ents;
1204
1205         trace_xfs_dir2_leaf_lookup(args);
1206
1207         /*
1208          * Look up name in the leaf block, returning both buffers and index.
1209          */
1210         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1211                 return error;
1212         }
1213         tp = args->trans;
1214         dp = args->dp;
1215         xfs_dir3_leaf_check(dp->i_mount, lbp);
1216         leaf = lbp->b_addr;
1217         ents = xfs_dir3_leaf_ents_p(leaf);
1218         /*
1219          * Get to the leaf entry and contained data entry address.
1220          */
1221         lep = &ents[index];
1222
1223         /*
1224          * Point to the data entry.
1225          */
1226         dep = (xfs_dir2_data_entry_t *)
1227               ((char *)dbp->b_addr +
1228                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1229         /*
1230          * Return the found inode number & CI name if appropriate
1231          */
1232         args->inumber = be64_to_cpu(dep->inumber);
1233         args->filetype = xfs_dir3_dirent_get_ftype(dp->i_mount, dep);
1234         error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1235         xfs_trans_brelse(tp, dbp);
1236         xfs_trans_brelse(tp, lbp);
1237         return XFS_ERROR(error);
1238 }
1239
1240 /*
1241  * Look up name/hash in the leaf block.
1242  * Fill in indexp with the found index, and dbpp with the data buffer.
1243  * If not found dbpp will be NULL, and ENOENT comes back.
1244  * lbpp will always be filled in with the leaf buffer unless there's an error.
1245  */
1246 static int                                      /* error */
1247 xfs_dir2_leaf_lookup_int(
1248         xfs_da_args_t           *args,          /* operation arguments */
1249         struct xfs_buf          **lbpp,         /* out: leaf buffer */
1250         int                     *indexp,        /* out: index in leaf block */
1251         struct xfs_buf          **dbpp)         /* out: data buffer */
1252 {
1253         xfs_dir2_db_t           curdb = -1;     /* current data block number */
1254         struct xfs_buf          *dbp = NULL;    /* data buffer */
1255         xfs_dir2_data_entry_t   *dep;           /* data entry */
1256         xfs_inode_t             *dp;            /* incore directory inode */
1257         int                     error;          /* error return code */
1258         int                     index;          /* index in leaf block */
1259         struct xfs_buf          *lbp;           /* leaf buffer */
1260         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1261         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1262         xfs_mount_t             *mp;            /* filesystem mount point */
1263         xfs_dir2_db_t           newdb;          /* new data block number */
1264         xfs_trans_t             *tp;            /* transaction pointer */
1265         xfs_dir2_db_t           cidb = -1;      /* case match data block no. */
1266         enum xfs_dacmp          cmp;            /* name compare result */
1267         struct xfs_dir2_leaf_entry *ents;
1268         struct xfs_dir3_icleaf_hdr leafhdr;
1269
1270         dp = args->dp;
1271         tp = args->trans;
1272         mp = dp->i_mount;
1273
1274         error = xfs_dir3_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
1275         if (error)
1276                 return error;
1277
1278         *lbpp = lbp;
1279         leaf = lbp->b_addr;
1280         xfs_dir3_leaf_check(mp, lbp);
1281         ents = xfs_dir3_leaf_ents_p(leaf);
1282         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1283
1284         /*
1285          * Look for the first leaf entry with our hash value.
1286          */
1287         index = xfs_dir2_leaf_search_hash(args, lbp);
1288         /*
1289          * Loop over all the entries with the right hash value
1290          * looking to match the name.
1291          */
1292         for (lep = &ents[index];
1293              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1294              lep++, index++) {
1295                 /*
1296                  * Skip over stale leaf entries.
1297                  */
1298                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1299                         continue;
1300                 /*
1301                  * Get the new data block number.
1302                  */
1303                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1304                 /*
1305                  * If it's not the same as the old data block number,
1306                  * need to pitch the old one and read the new one.
1307                  */
1308                 if (newdb != curdb) {
1309                         if (dbp)
1310                                 xfs_trans_brelse(tp, dbp);
1311                         error = xfs_dir3_data_read(tp, dp,
1312                                                    xfs_dir2_db_to_da(mp, newdb),
1313                                                    -1, &dbp);
1314                         if (error) {
1315                                 xfs_trans_brelse(tp, lbp);
1316                                 return error;
1317                         }
1318                         curdb = newdb;
1319                 }
1320                 /*
1321                  * Point to the data entry.
1322                  */
1323                 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1324                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1325                 /*
1326                  * Compare name and if it's an exact match, return the index
1327                  * and buffer. If it's the first case-insensitive match, store
1328                  * the index and buffer and continue looking for an exact match.
1329                  */
1330                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1331                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1332                         args->cmpresult = cmp;
1333                         *indexp = index;
1334                         /* case exact match: return the current buffer. */
1335                         if (cmp == XFS_CMP_EXACT) {
1336                                 *dbpp = dbp;
1337                                 return 0;
1338                         }
1339                         cidb = curdb;
1340                 }
1341         }
1342         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1343         /*
1344          * Here, we can only be doing a lookup (not a rename or remove).
1345          * If a case-insensitive match was found earlier, re-read the
1346          * appropriate data block if required and return it.
1347          */
1348         if (args->cmpresult == XFS_CMP_CASE) {
1349                 ASSERT(cidb != -1);
1350                 if (cidb != curdb) {
1351                         xfs_trans_brelse(tp, dbp);
1352                         error = xfs_dir3_data_read(tp, dp,
1353                                                    xfs_dir2_db_to_da(mp, cidb),
1354                                                    -1, &dbp);
1355                         if (error) {
1356                                 xfs_trans_brelse(tp, lbp);
1357                                 return error;
1358                         }
1359                 }
1360                 *dbpp = dbp;
1361                 return 0;
1362         }
1363         /*
1364          * No match found, return ENOENT.
1365          */
1366         ASSERT(cidb == -1);
1367         if (dbp)
1368                 xfs_trans_brelse(tp, dbp);
1369         xfs_trans_brelse(tp, lbp);
1370         return XFS_ERROR(ENOENT);
1371 }
1372
1373 /*
1374  * Remove an entry from a leaf format directory.
1375  */
1376 int                                             /* error */
1377 xfs_dir2_leaf_removename(
1378         xfs_da_args_t           *args)          /* operation arguments */
1379 {
1380         __be16                  *bestsp;        /* leaf block best freespace */
1381         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1382         xfs_dir2_db_t           db;             /* data block number */
1383         struct xfs_buf          *dbp;           /* data block buffer */
1384         xfs_dir2_data_entry_t   *dep;           /* data entry structure */
1385         xfs_inode_t             *dp;            /* incore directory inode */
1386         int                     error;          /* error return code */
1387         xfs_dir2_db_t           i;              /* temporary data block # */
1388         int                     index;          /* index into leaf entries */
1389         struct xfs_buf          *lbp;           /* leaf buffer */
1390         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1391         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1392         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1393         xfs_mount_t             *mp;            /* filesystem mount point */
1394         int                     needlog;        /* need to log data header */
1395         int                     needscan;       /* need to rescan data frees */
1396         xfs_dir2_data_off_t     oldbest;        /* old value of best free */
1397         xfs_trans_t             *tp;            /* transaction pointer */
1398         struct xfs_dir2_data_free *bf;          /* bestfree table */
1399         struct xfs_dir2_leaf_entry *ents;
1400         struct xfs_dir3_icleaf_hdr leafhdr;
1401
1402         trace_xfs_dir2_leaf_removename(args);
1403
1404         /*
1405          * Lookup the leaf entry, get the leaf and data blocks read in.
1406          */
1407         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1408                 return error;
1409         }
1410         dp = args->dp;
1411         tp = args->trans;
1412         mp = dp->i_mount;
1413         leaf = lbp->b_addr;
1414         hdr = dbp->b_addr;
1415         xfs_dir3_data_check(dp, dbp);
1416         bf = xfs_dir3_data_bestfree_p(hdr);
1417         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1418         ents = xfs_dir3_leaf_ents_p(leaf);
1419         /*
1420          * Point to the leaf entry, use that to point to the data entry.
1421          */
1422         lep = &ents[index];
1423         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1424         dep = (xfs_dir2_data_entry_t *)
1425               ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1426         needscan = needlog = 0;
1427         oldbest = be16_to_cpu(bf[0].length);
1428         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1429         bestsp = xfs_dir2_leaf_bests_p(ltp);
1430         ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1431         /*
1432          * Mark the former data entry unused.
1433          */
1434         xfs_dir2_data_make_free(tp, dbp,
1435                 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1436                 xfs_dir3_data_entsize(mp, dep->namelen), &needlog, &needscan);
1437         /*
1438          * We just mark the leaf entry stale by putting a null in it.
1439          */
1440         leafhdr.stale++;
1441         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
1442         xfs_dir3_leaf_log_header(tp, lbp);
1443
1444         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1445         xfs_dir3_leaf_log_ents(tp, lbp, index, index);
1446
1447         /*
1448          * Scan the freespace in the data block again if necessary,
1449          * log the data block header if necessary.
1450          */
1451         if (needscan)
1452                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1453         if (needlog)
1454                 xfs_dir2_data_log_header(tp, dbp);
1455         /*
1456          * If the longest freespace in the data block has changed,
1457          * put the new value in the bests table and log that.
1458          */
1459         if (be16_to_cpu(bf[0].length) != oldbest) {
1460                 bestsp[db] = bf[0].length;
1461                 xfs_dir3_leaf_log_bests(tp, lbp, db, db);
1462         }
1463         xfs_dir3_data_check(dp, dbp);
1464         /*
1465          * If the data block is now empty then get rid of the data block.
1466          */
1467         if (be16_to_cpu(bf[0].length) ==
1468                         mp->m_dirblksize - xfs_dir3_data_entry_offset(hdr)) {
1469                 ASSERT(db != mp->m_dirdatablk);
1470                 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1471                         /*
1472                          * Nope, can't get rid of it because it caused
1473                          * allocation of a bmap btree block to do so.
1474                          * Just go on, returning success, leaving the
1475                          * empty block in place.
1476                          */
1477                         if (error == ENOSPC && args->total == 0)
1478                                 error = 0;
1479                         xfs_dir3_leaf_check(mp, lbp);
1480                         return error;
1481                 }
1482                 dbp = NULL;
1483                 /*
1484                  * If this is the last data block then compact the
1485                  * bests table by getting rid of entries.
1486                  */
1487                 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1488                         /*
1489                          * Look for the last active entry (i).
1490                          */
1491                         for (i = db - 1; i > 0; i--) {
1492                                 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1493                                         break;
1494                         }
1495                         /*
1496                          * Copy the table down so inactive entries at the
1497                          * end are removed.
1498                          */
1499                         memmove(&bestsp[db - i], bestsp,
1500                                 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1501                         be32_add_cpu(&ltp->bestcount, -(db - i));
1502                         xfs_dir3_leaf_log_tail(tp, lbp);
1503                         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1504                 } else
1505                         bestsp[db] = cpu_to_be16(NULLDATAOFF);
1506         }
1507         /*
1508          * If the data block was not the first one, drop it.
1509          */
1510         else if (db != mp->m_dirdatablk)
1511                 dbp = NULL;
1512
1513         xfs_dir3_leaf_check(mp, lbp);
1514         /*
1515          * See if we can convert to block form.
1516          */
1517         return xfs_dir2_leaf_to_block(args, lbp, dbp);
1518 }
1519
1520 /*
1521  * Replace the inode number in a leaf format directory entry.
1522  */
1523 int                                             /* error */
1524 xfs_dir2_leaf_replace(
1525         xfs_da_args_t           *args)          /* operation arguments */
1526 {
1527         struct xfs_buf          *dbp;           /* data block buffer */
1528         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1529         xfs_inode_t             *dp;            /* incore directory inode */
1530         int                     error;          /* error return code */
1531         int                     index;          /* index of leaf entry */
1532         struct xfs_buf          *lbp;           /* leaf buffer */
1533         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1534         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1535         xfs_trans_t             *tp;            /* transaction pointer */
1536         struct xfs_dir2_leaf_entry *ents;
1537
1538         trace_xfs_dir2_leaf_replace(args);
1539
1540         /*
1541          * Look up the entry.
1542          */
1543         if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1544                 return error;
1545         }
1546         dp = args->dp;
1547         leaf = lbp->b_addr;
1548         ents = xfs_dir3_leaf_ents_p(leaf);
1549         /*
1550          * Point to the leaf entry, get data address from it.
1551          */
1552         lep = &ents[index];
1553         /*
1554          * Point to the data entry.
1555          */
1556         dep = (xfs_dir2_data_entry_t *)
1557               ((char *)dbp->b_addr +
1558                xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1559         ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1560         /*
1561          * Put the new inode number in, log it.
1562          */
1563         dep->inumber = cpu_to_be64(args->inumber);
1564         xfs_dir3_dirent_put_ftype(dp->i_mount, dep, args->filetype);
1565         tp = args->trans;
1566         xfs_dir2_data_log_entry(tp, dbp, dep);
1567         xfs_dir3_leaf_check(dp->i_mount, lbp);
1568         xfs_trans_brelse(tp, lbp);
1569         return 0;
1570 }
1571
1572 /*
1573  * Return index in the leaf block (lbp) which is either the first
1574  * one with this hash value, or if there are none, the insert point
1575  * for that hash value.
1576  */
1577 int                                             /* index value */
1578 xfs_dir2_leaf_search_hash(
1579         xfs_da_args_t           *args,          /* operation arguments */
1580         struct xfs_buf          *lbp)           /* leaf buffer */
1581 {
1582         xfs_dahash_t            hash=0;         /* hash from this entry */
1583         xfs_dahash_t            hashwant;       /* hash value looking for */
1584         int                     high;           /* high leaf index */
1585         int                     low;            /* low leaf index */
1586         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1587         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1588         int                     mid=0;          /* current leaf index */
1589         struct xfs_dir2_leaf_entry *ents;
1590         struct xfs_dir3_icleaf_hdr leafhdr;
1591
1592         leaf = lbp->b_addr;
1593         ents = xfs_dir3_leaf_ents_p(leaf);
1594         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1595
1596         /*
1597          * Note, the table cannot be empty, so we have to go through the loop.
1598          * Binary search the leaf entries looking for our hash value.
1599          */
1600         for (lep = ents, low = 0, high = leafhdr.count - 1,
1601                 hashwant = args->hashval;
1602              low <= high; ) {
1603                 mid = (low + high) >> 1;
1604                 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1605                         break;
1606                 if (hash < hashwant)
1607                         low = mid + 1;
1608                 else
1609                         high = mid - 1;
1610         }
1611         /*
1612          * Found one, back up through all the equal hash values.
1613          */
1614         if (hash == hashwant) {
1615                 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1616                         mid--;
1617                 }
1618         }
1619         /*
1620          * Need to point to an entry higher than ours.
1621          */
1622         else if (hash < hashwant)
1623                 mid++;
1624         return mid;
1625 }
1626
1627 /*
1628  * Trim off a trailing data block.  We know it's empty since the leaf
1629  * freespace table says so.
1630  */
1631 int                                             /* error */
1632 xfs_dir2_leaf_trim_data(
1633         xfs_da_args_t           *args,          /* operation arguments */
1634         struct xfs_buf          *lbp,           /* leaf buffer */
1635         xfs_dir2_db_t           db)             /* data block number */
1636 {
1637         __be16                  *bestsp;        /* leaf bests table */
1638         struct xfs_buf          *dbp;           /* data block buffer */
1639         xfs_inode_t             *dp;            /* incore directory inode */
1640         int                     error;          /* error return value */
1641         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1642         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
1643         xfs_mount_t             *mp;            /* filesystem mount point */
1644         xfs_trans_t             *tp;            /* transaction pointer */
1645
1646         dp = args->dp;
1647         mp = dp->i_mount;
1648         tp = args->trans;
1649         /*
1650          * Read the offending data block.  We need its buffer.
1651          */
1652         error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp);
1653         if (error)
1654                 return error;
1655
1656         leaf = lbp->b_addr;
1657         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1658
1659 #ifdef DEBUG
1660 {
1661         struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1662         struct xfs_dir2_data_free *bf = xfs_dir3_data_bestfree_p(hdr);
1663
1664         ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1665                hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1666         ASSERT(be16_to_cpu(bf[0].length) ==
1667                mp->m_dirblksize - xfs_dir3_data_entry_offset(hdr));
1668         ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1669 }
1670 #endif
1671
1672         /*
1673          * Get rid of the data block.
1674          */
1675         if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1676                 ASSERT(error != ENOSPC);
1677                 xfs_trans_brelse(tp, dbp);
1678                 return error;
1679         }
1680         /*
1681          * Eliminate the last bests entry from the table.
1682          */
1683         bestsp = xfs_dir2_leaf_bests_p(ltp);
1684         be32_add_cpu(&ltp->bestcount, -1);
1685         memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1686         xfs_dir3_leaf_log_tail(tp, lbp);
1687         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1688         return 0;
1689 }
1690
1691 static inline size_t
1692 xfs_dir3_leaf_size(
1693         struct xfs_dir3_icleaf_hdr      *hdr,
1694         int                             counts)
1695 {
1696         int     entries;
1697         int     hdrsize;
1698
1699         entries = hdr->count - hdr->stale;
1700         if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1701             hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1702                 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1703         else
1704                 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1705
1706         return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1707                        + counts * sizeof(xfs_dir2_data_off_t)
1708                        + sizeof(xfs_dir2_leaf_tail_t);
1709 }
1710
1711 /*
1712  * Convert node form directory to leaf form directory.
1713  * The root of the node form dir needs to already be a LEAFN block.
1714  * Just return if we can't do anything.
1715  */
1716 int                                             /* error */
1717 xfs_dir2_node_to_leaf(
1718         xfs_da_state_t          *state)         /* directory operation state */
1719 {
1720         xfs_da_args_t           *args;          /* operation arguments */
1721         xfs_inode_t             *dp;            /* incore directory inode */
1722         int                     error;          /* error return code */
1723         struct xfs_buf          *fbp;           /* buffer for freespace block */
1724         xfs_fileoff_t           fo;             /* freespace file offset */
1725         xfs_dir2_free_t         *free;          /* freespace structure */
1726         struct xfs_buf          *lbp;           /* buffer for leaf block */
1727         xfs_dir2_leaf_tail_t    *ltp;           /* tail of leaf structure */
1728         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1729         xfs_mount_t             *mp;            /* filesystem mount point */
1730         int                     rval;           /* successful free trim? */
1731         xfs_trans_t             *tp;            /* transaction pointer */
1732         struct xfs_dir3_icleaf_hdr leafhdr;
1733         struct xfs_dir3_icfree_hdr freehdr;
1734
1735         /*
1736          * There's more than a leaf level in the btree, so there must
1737          * be multiple leafn blocks.  Give up.
1738          */
1739         if (state->path.active > 1)
1740                 return 0;
1741         args = state->args;
1742
1743         trace_xfs_dir2_node_to_leaf(args);
1744
1745         mp = state->mp;
1746         dp = args->dp;
1747         tp = args->trans;
1748         /*
1749          * Get the last offset in the file.
1750          */
1751         if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1752                 return error;
1753         }
1754         fo -= mp->m_dirblkfsbs;
1755         /*
1756          * If there are freespace blocks other than the first one,
1757          * take this opportunity to remove trailing empty freespace blocks
1758          * that may have been left behind during no-space-reservation
1759          * operations.
1760          */
1761         while (fo > mp->m_dirfreeblk) {
1762                 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1763                         return error;
1764                 }
1765                 if (rval)
1766                         fo -= mp->m_dirblkfsbs;
1767                 else
1768                         return 0;
1769         }
1770         /*
1771          * Now find the block just before the freespace block.
1772          */
1773         if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1774                 return error;
1775         }
1776         /*
1777          * If it's not the single leaf block, give up.
1778          */
1779         if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1780                 return 0;
1781         lbp = state->path.blk[0].bp;
1782         leaf = lbp->b_addr;
1783         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1784
1785         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1786                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1787
1788         /*
1789          * Read the freespace block.
1790          */
1791         error = xfs_dir2_free_read(tp, dp,  mp->m_dirfreeblk, &fbp);
1792         if (error)
1793                 return error;
1794         free = fbp->b_addr;
1795         xfs_dir3_free_hdr_from_disk(&freehdr, free);
1796
1797         ASSERT(!freehdr.firstdb);
1798
1799         /*
1800          * Now see if the leafn and free data will fit in a leaf1.
1801          * If not, release the buffer and give up.
1802          */
1803         if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > mp->m_dirblksize) {
1804                 xfs_trans_brelse(tp, fbp);
1805                 return 0;
1806         }
1807
1808         /*
1809          * If the leaf has any stale entries in it, compress them out.
1810          */
1811         if (leafhdr.stale)
1812                 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1813
1814         lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1815         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1816         leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1817                                         ? XFS_DIR2_LEAF1_MAGIC
1818                                         : XFS_DIR3_LEAF1_MAGIC;
1819
1820         /*
1821          * Set up the leaf tail from the freespace block.
1822          */
1823         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1824         ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1825
1826         /*
1827          * Set up the leaf bests table.
1828          */
1829         memcpy(xfs_dir2_leaf_bests_p(ltp), xfs_dir3_free_bests_p(mp, free),
1830                 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1831
1832         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
1833         xfs_dir3_leaf_log_header(tp, lbp);
1834         xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1835         xfs_dir3_leaf_log_tail(tp, lbp);
1836         xfs_dir3_leaf_check(mp, lbp);
1837
1838         /*
1839          * Get rid of the freespace block.
1840          */
1841         error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1842         if (error) {
1843                 /*
1844                  * This can't fail here because it can only happen when
1845                  * punching out the middle of an extent, and this is an
1846                  * isolated block.
1847                  */
1848                 ASSERT(error != ENOSPC);
1849                 return error;
1850         }
1851         fbp = NULL;
1852         /*
1853          * Now see if we can convert the single-leaf directory
1854          * down to a block form directory.
1855          * This routine always kills the dabuf for the leaf, so
1856          * eliminate it from the path.
1857          */
1858         error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1859         state->path.blk[0].bp = NULL;
1860         return error;
1861 }