linux/fs/xfs/xfs_dir2_data.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2000-2002,2005 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_dir2.h"
  31#include "xfs_dir2_priv.h"
  32#include "xfs_error.h"
  33#include "xfs_trans.h"
  34#include "xfs_buf_item.h"
  35#include "xfs_cksum.h"
  36
  37/*
  38 * Check the consistency of the data block.
  39 * The input can also be a block-format directory.
  40 * Return 0 is the buffer is good, otherwise an error.
  41 */
  42int
  43__xfs_dir3_data_check(
  44        struct xfs_inode        *dp,            /* incore inode pointer */
  45        struct xfs_buf          *bp)            /* data block's buffer */
  46{
  47        xfs_dir2_dataptr_t      addr;           /* addr for leaf lookup */
  48        xfs_dir2_data_free_t    *bf;            /* bestfree table */
  49        xfs_dir2_block_tail_t   *btp=NULL;      /* block tail */
  50        int                     count;          /* count of entries found */
  51        xfs_dir2_data_hdr_t     *hdr;           /* data block header */
  52        xfs_dir2_data_entry_t   *dep;           /* data entry */
  53        xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
  54        xfs_dir2_data_unused_t  *dup;           /* unused entry */
  55        char                    *endp;          /* end of useful data */
  56        int                     freeseen;       /* mask of bestfrees seen */
  57        xfs_dahash_t            hash;           /* hash of current name */
  58        int                     i;              /* leaf index */
  59        int                     lastfree;       /* last entry was unused */
  60        xfs_dir2_leaf_entry_t   *lep=NULL;      /* block leaf entries */
  61        xfs_mount_t             *mp;            /* filesystem mount point */
  62        char                    *p;             /* current data position */
  63        int                     stale;          /* count of stale leaves */
  64        struct xfs_name         name;
  65        const struct xfs_dir_ops *ops;
  66
  67        mp = bp->b_target->bt_mount;
  68
  69        /*
  70         * We can be passed a null dp here from a verifier, so we need to go the
  71         * hard way to get them.
  72         */
  73        ops = xfs_dir_get_ops(mp, dp);
  74
  75        hdr = bp->b_addr;
  76        p = (char *)ops->data_entry_p(hdr);
  77
  78        switch (hdr->magic) {
  79        case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
  80        case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
  81                btp = xfs_dir2_block_tail_p(mp, hdr);
  82                lep = xfs_dir2_block_leaf_p(btp);
  83                endp = (char *)lep;
  84
  85                /*
  86                 * The number of leaf entries is limited by the size of the
  87                 * block and the amount of space used by the data entries.
  88                 * We don't know how much space is used by the data entries yet,
  89                 * so just ensure that the count falls somewhere inside the
  90                 * block right now.
  91                 */
  92                XFS_WANT_CORRUPTED_RETURN(be32_to_cpu(btp->count) <
  93                        ((char *)btp - p) / sizeof(struct xfs_dir2_leaf_entry));
  94                break;
  95        case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
  96        case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
  97                endp = (char *)hdr + mp->m_dirblksize;
  98                break;
  99        default:
 100                XFS_ERROR_REPORT("Bad Magic", XFS_ERRLEVEL_LOW, mp);
 101                return EFSCORRUPTED;
 102        }
 103
 104        /*
 105         * Account for zero bestfree entries.
 106         */
 107        bf = ops->data_bestfree_p(hdr);
 108        count = lastfree = freeseen = 0;
 109        if (!bf[0].length) {
 110                XFS_WANT_CORRUPTED_RETURN(!bf[0].offset);
 111                freeseen |= 1 << 0;
 112        }
 113        if (!bf[1].length) {
 114                XFS_WANT_CORRUPTED_RETURN(!bf[1].offset);
 115                freeseen |= 1 << 1;
 116        }
 117        if (!bf[2].length) {
 118                XFS_WANT_CORRUPTED_RETURN(!bf[2].offset);
 119                freeseen |= 1 << 2;
 120        }
 121
 122        XFS_WANT_CORRUPTED_RETURN(be16_to_cpu(bf[0].length) >=
 123                                                be16_to_cpu(bf[1].length));
 124        XFS_WANT_CORRUPTED_RETURN(be16_to_cpu(bf[1].length) >=
 125                                                be16_to_cpu(bf[2].length));
 126        /*
 127         * Loop over the data/unused entries.
 128         */
 129        while (p < endp) {
 130                dup = (xfs_dir2_data_unused_t *)p;
 131                /*
 132                 * If it's unused, look for the space in the bestfree table.
 133                 * If we find it, account for that, else make sure it
 134                 * doesn't need to be there.
 135                 */
 136                if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 137                        XFS_WANT_CORRUPTED_RETURN(lastfree == 0);
 138                        XFS_WANT_CORRUPTED_RETURN(
 139                                be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) ==
 140                                               (char *)dup - (char *)hdr);
 141                        dfp = xfs_dir2_data_freefind(hdr, bf, dup);
 142                        if (dfp) {
 143                                i = (int)(dfp - bf);
 144                                XFS_WANT_CORRUPTED_RETURN(
 145                                        (freeseen & (1 << i)) == 0);
 146                                freeseen |= 1 << i;
 147                        } else {
 148                                XFS_WANT_CORRUPTED_RETURN(
 149                                        be16_to_cpu(dup->length) <=
 150                                                be16_to_cpu(bf[2].length));
 151                        }
 152                        p += be16_to_cpu(dup->length);
 153                        lastfree = 1;
 154                        continue;
 155                }
 156                /*
 157                 * It's a real entry.  Validate the fields.
 158                 * If this is a block directory then make sure it's
 159                 * in the leaf section of the block.
 160                 * The linear search is crude but this is DEBUG code.
 161                 */
 162                dep = (xfs_dir2_data_entry_t *)p;
 163                XFS_WANT_CORRUPTED_RETURN(dep->namelen != 0);
 164                XFS_WANT_CORRUPTED_RETURN(
 165                        !xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber)));
 166                XFS_WANT_CORRUPTED_RETURN(
 167                        be16_to_cpu(*ops->data_entry_tag_p(dep)) ==
 168                                               (char *)dep - (char *)hdr);
 169                XFS_WANT_CORRUPTED_RETURN(
 170                                ops->data_get_ftype(dep) < XFS_DIR3_FT_MAX);
 171                count++;
 172                lastfree = 0;
 173                if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 174                    hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 175                        addr = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
 176                                (xfs_dir2_data_aoff_t)
 177                                ((char *)dep - (char *)hdr));
 178                        name.name = dep->name;
 179                        name.len = dep->namelen;
 180                        hash = mp->m_dirnameops->hashname(&name);
 181                        for (i = 0; i < be32_to_cpu(btp->count); i++) {
 182                                if (be32_to_cpu(lep[i].address) == addr &&
 183                                    be32_to_cpu(lep[i].hashval) == hash)
 184                                        break;
 185                        }
 186                        XFS_WANT_CORRUPTED_RETURN(i < be32_to_cpu(btp->count));
 187                }
 188                p += ops->data_entsize(dep->namelen);
 189        }
 190        /*
 191         * Need to have seen all the entries and all the bestfree slots.
 192         */
 193        XFS_WANT_CORRUPTED_RETURN(freeseen == 7);
 194        if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 195            hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 196                for (i = stale = 0; i < be32_to_cpu(btp->count); i++) {
 197                        if (lep[i].address ==
 198                            cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 199                                stale++;
 200                        if (i > 0)
 201                                XFS_WANT_CORRUPTED_RETURN(
 202                                        be32_to_cpu(lep[i].hashval) >=
 203                                                be32_to_cpu(lep[i - 1].hashval));
 204                }
 205                XFS_WANT_CORRUPTED_RETURN(count ==
 206                        be32_to_cpu(btp->count) - be32_to_cpu(btp->stale));
 207                XFS_WANT_CORRUPTED_RETURN(stale == be32_to_cpu(btp->stale));
 208        }
 209        return 0;
 210}
 211
 212static bool
 213xfs_dir3_data_verify(
 214        struct xfs_buf          *bp)
 215{
 216        struct xfs_mount        *mp = bp->b_target->bt_mount;
 217        struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 218
 219        if (xfs_sb_version_hascrc(&mp->m_sb)) {
 220                if (hdr3->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 221                        return false;
 222                if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
 223                        return false;
 224                if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 225                        return false;
 226        } else {
 227                if (hdr3->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC))
 228                        return false;
 229        }
 230        if (__xfs_dir3_data_check(NULL, bp))
 231                return false;
 232        return true;
 233}
 234
 235/*
 236 * Readahead of the first block of the directory when it is opened is completely
 237 * oblivious to the format of the directory. Hence we can either get a block
 238 * format buffer or a data format buffer on readahead.
 239 */
 240static void
 241xfs_dir3_data_reada_verify(
 242        struct xfs_buf          *bp)
 243{
 244        struct xfs_mount        *mp = bp->b_target->bt_mount;
 245        struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 246
 247        switch (hdr->magic) {
 248        case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
 249        case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
 250                bp->b_ops = &xfs_dir3_block_buf_ops;
 251                bp->b_ops->verify_read(bp);
 252                return;
 253        case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
 254        case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
 255                xfs_dir3_data_verify(bp);
 256                return;
 257        default:
 258                XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, hdr);
 259                xfs_buf_ioerror(bp, EFSCORRUPTED);
 260                break;
 261        }
 262}
 263
 264static void
 265xfs_dir3_data_read_verify(
 266        struct xfs_buf  *bp)
 267{
 268        struct xfs_mount        *mp = bp->b_target->bt_mount;
 269
 270        if ((xfs_sb_version_hascrc(&mp->m_sb) &&
 271             !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
 272                                          XFS_DIR3_DATA_CRC_OFF)) ||
 273            !xfs_dir3_data_verify(bp)) {
 274                XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 275                xfs_buf_ioerror(bp, EFSCORRUPTED);
 276        }
 277}
 278
 279static void
 280xfs_dir3_data_write_verify(
 281        struct xfs_buf  *bp)
 282{
 283        struct xfs_mount        *mp = bp->b_target->bt_mount;
 284        struct xfs_buf_log_item *bip = bp->b_fspriv;
 285        struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 286
 287        if (!xfs_dir3_data_verify(bp)) {
 288                XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 289                xfs_buf_ioerror(bp, EFSCORRUPTED);
 290                return;
 291        }
 292
 293        if (!xfs_sb_version_hascrc(&mp->m_sb))
 294                return;
 295
 296        if (bip)
 297                hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 298
 299        xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_DATA_CRC_OFF);
 300}
 301
 302const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
 303        .verify_read = xfs_dir3_data_read_verify,
 304        .verify_write = xfs_dir3_data_write_verify,
 305};
 306
 307static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
 308        .verify_read = xfs_dir3_data_reada_verify,
 309        .verify_write = xfs_dir3_data_write_verify,
 310};
 311
 312
 313int
 314xfs_dir3_data_read(
 315        struct xfs_trans        *tp,
 316        struct xfs_inode        *dp,
 317        xfs_dablk_t             bno,
 318        xfs_daddr_t             mapped_bno,
 319        struct xfs_buf          **bpp)
 320{
 321        int                     err;
 322
 323        err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
 324                                XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
 325        if (!err && tp)
 326                xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
 327        return err;
 328}
 329
 330int
 331xfs_dir3_data_readahead(
 332        struct xfs_trans        *tp,
 333        struct xfs_inode        *dp,
 334        xfs_dablk_t             bno,
 335        xfs_daddr_t             mapped_bno)
 336{
 337        return xfs_da_reada_buf(tp, dp, bno, mapped_bno,
 338                                XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
 339}
 340
 341/*
 342 * Given a data block and an unused entry from that block,
 343 * return the bestfree entry if any that corresponds to it.
 344 */
 345xfs_dir2_data_free_t *
 346xfs_dir2_data_freefind(
 347        struct xfs_dir2_data_hdr *hdr,          /* data block header */
 348        struct xfs_dir2_data_free *bf,          /* bestfree table pointer */
 349        struct xfs_dir2_data_unused *dup)       /* unused space */
 350{
 351        xfs_dir2_data_free_t    *dfp;           /* bestfree entry */
 352        xfs_dir2_data_aoff_t    off;            /* offset value needed */
 353#ifdef DEBUG
 354        int                     matched;        /* matched the value */
 355        int                     seenzero;       /* saw a 0 bestfree entry */
 356#endif
 357
 358        off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
 359
 360#ifdef DEBUG
 361        /*
 362         * Validate some consistency in the bestfree table.
 363         * Check order, non-overlapping entries, and if we find the
 364         * one we're looking for it has to be exact.
 365         */
 366        ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 367               hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 368               hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 369               hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 370        for (dfp = &bf[0], seenzero = matched = 0;
 371             dfp < &bf[XFS_DIR2_DATA_FD_COUNT];
 372             dfp++) {
 373                if (!dfp->offset) {
 374                        ASSERT(!dfp->length);
 375                        seenzero = 1;
 376                        continue;
 377                }
 378                ASSERT(seenzero == 0);
 379                if (be16_to_cpu(dfp->offset) == off) {
 380                        matched = 1;
 381                        ASSERT(dfp->length == dup->length);
 382                } else if (off < be16_to_cpu(dfp->offset))
 383                        ASSERT(off + be16_to_cpu(dup->length) <= be16_to_cpu(dfp->offset));
 384                else
 385                        ASSERT(be16_to_cpu(dfp->offset) + be16_to_cpu(dfp->length) <= off);
 386                ASSERT(matched || be16_to_cpu(dfp->length) >= be16_to_cpu(dup->length));
 387                if (dfp > &bf[0])
 388                        ASSERT(be16_to_cpu(dfp[-1].length) >= be16_to_cpu(dfp[0].length));
 389        }
 390#endif
 391        /*
 392         * If this is smaller than the smallest bestfree entry,
 393         * it can't be there since they're sorted.
 394         */
 395        if (be16_to_cpu(dup->length) <
 396            be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
 397                return NULL;
 398        /*
 399         * Look at the three bestfree entries for our guy.
 400         */
 401        for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
 402                if (!dfp->offset)
 403                        return NULL;
 404                if (be16_to_cpu(dfp->offset) == off)
 405                        return dfp;
 406        }
 407        /*
 408         * Didn't find it.  This only happens if there are duplicate lengths.
 409         */
 410        return NULL;
 411}
 412
 413/*
 414 * Insert an unused-space entry into the bestfree table.
 415 */
 416xfs_dir2_data_free_t *                          /* entry inserted */
 417xfs_dir2_data_freeinsert(
 418        struct xfs_dir2_data_hdr *hdr,          /* data block pointer */
 419        struct xfs_dir2_data_free *dfp,         /* bestfree table pointer */
 420        struct xfs_dir2_data_unused *dup,       /* unused space */
 421        int                     *loghead)       /* log the data header (out) */
 422{
 423        xfs_dir2_data_free_t    new;            /* new bestfree entry */
 424
 425        ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 426               hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 427               hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 428               hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 429
 430        new.length = dup->length;
 431        new.offset = cpu_to_be16((char *)dup - (char *)hdr);
 432
 433        /*
 434         * Insert at position 0, 1, or 2; or not at all.
 435         */
 436        if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
 437                dfp[2] = dfp[1];
 438                dfp[1] = dfp[0];
 439                dfp[0] = new;
 440                *loghead = 1;
 441                return &dfp[0];
 442        }
 443        if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
 444                dfp[2] = dfp[1];
 445                dfp[1] = new;
 446                *loghead = 1;
 447                return &dfp[1];
 448        }
 449        if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
 450                dfp[2] = new;
 451                *loghead = 1;
 452                return &dfp[2];
 453        }
 454        return NULL;
 455}
 456
 457/*
 458 * Remove a bestfree entry from the table.
 459 */
 460STATIC void
 461xfs_dir2_data_freeremove(
 462        struct xfs_dir2_data_hdr *hdr,          /* data block header */
 463        struct xfs_dir2_data_free *bf,          /* bestfree table pointer */
 464        struct xfs_dir2_data_free *dfp,         /* bestfree entry pointer */
 465        int                     *loghead)       /* out: log data header */
 466{
 467
 468        ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 469               hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 470               hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 471               hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 472
 473        /*
 474         * It's the first entry, slide the next 2 up.
 475         */
 476        if (dfp == &bf[0]) {
 477                bf[0] = bf[1];
 478                bf[1] = bf[2];
 479        }
 480        /*
 481         * It's the second entry, slide the 3rd entry up.
 482         */
 483        else if (dfp == &bf[1])
 484                bf[1] = bf[2];
 485        /*
 486         * Must be the last entry.
 487         */
 488        else
 489                ASSERT(dfp == &bf[2]);
 490        /*
 491         * Clear the 3rd entry, must be zero now.
 492         */
 493        bf[2].length = 0;
 494        bf[2].offset = 0;
 495        *loghead = 1;
 496}
 497
 498/*
 499 * Given a data block, reconstruct its bestfree map.
 500 */
 501void
 502xfs_dir2_data_freescan(
 503        struct xfs_inode        *dp,
 504        struct xfs_dir2_data_hdr *hdr,
 505        int                     *loghead)
 506{
 507        xfs_dir2_block_tail_t   *btp;           /* block tail */
 508        xfs_dir2_data_entry_t   *dep;           /* active data entry */
 509        xfs_dir2_data_unused_t  *dup;           /* unused data entry */
 510        struct xfs_dir2_data_free *bf;
 511        char                    *endp;          /* end of block's data */
 512        char                    *p;             /* current entry pointer */
 513
 514        ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 515               hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 516               hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 517               hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 518
 519        /*
 520         * Start by clearing the table.
 521         */
 522        bf = dp->d_ops->data_bestfree_p(hdr);
 523        memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
 524        *loghead = 1;
 525        /*
 526         * Set up pointers.
 527         */
 528        p = (char *)dp->d_ops->data_entry_p(hdr);
 529        if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 530            hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
 531                btp = xfs_dir2_block_tail_p(dp->i_mount, hdr);
 532                endp = (char *)xfs_dir2_block_leaf_p(btp);
 533        } else
 534                endp = (char *)hdr + dp->i_mount->m_dirblksize;
 535        /*
 536         * Loop over the block's entries.
 537         */
 538        while (p < endp) {
 539                dup = (xfs_dir2_data_unused_t *)p;
 540                /*
 541                 * If it's a free entry, insert it.
 542                 */
 543                if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 544                        ASSERT((char *)dup - (char *)hdr ==
 545                               be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 546                        xfs_dir2_data_freeinsert(hdr, bf, dup, loghead);
 547                        p += be16_to_cpu(dup->length);
 548                }
 549                /*
 550                 * For active entries, check their tags and skip them.
 551                 */
 552                else {
 553                        dep = (xfs_dir2_data_entry_t *)p;
 554                        ASSERT((char *)dep - (char *)hdr ==
 555                               be16_to_cpu(*dp->d_ops->data_entry_tag_p(dep)));
 556                        p += dp->d_ops->data_entsize(dep->namelen);
 557                }
 558        }
 559}
 560
 561/*
 562 * Initialize a data block at the given block number in the directory.
 563 * Give back the buffer for the created block.
 564 */
 565int                                             /* error */
 566xfs_dir3_data_init(
 567        xfs_da_args_t           *args,          /* directory operation args */
 568        xfs_dir2_db_t           blkno,          /* logical dir block number */
 569        struct xfs_buf          **bpp)          /* output block buffer */
 570{
 571        struct xfs_buf          *bp;            /* block buffer */
 572        xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 573        xfs_inode_t             *dp;            /* incore directory inode */
 574        xfs_dir2_data_unused_t  *dup;           /* unused entry pointer */
 575        struct xfs_dir2_data_free *bf;
 576        int                     error;          /* error return value */
 577        int                     i;              /* bestfree index */
 578        xfs_mount_t             *mp;            /* filesystem mount point */
 579        xfs_trans_t             *tp;            /* transaction pointer */
 580        int                     t;              /* temp */
 581
 582        dp = args->dp;
 583        mp = dp->i_mount;
 584        tp = args->trans;
 585        /*
 586         * Get the buffer set up for the block.
 587         */
 588        error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, blkno), -1, &bp,
 589                XFS_DATA_FORK);
 590        if (error)
 591                return error;
 592        bp->b_ops = &xfs_dir3_data_buf_ops;
 593        xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
 594
 595        /*
 596         * Initialize the header.
 597         */
 598        hdr = bp->b_addr;
 599        if (xfs_sb_version_hascrc(&mp->m_sb)) {
 600                struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 601
 602                memset(hdr3, 0, sizeof(*hdr3));
 603                hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
 604                hdr3->blkno = cpu_to_be64(bp->b_bn);
 605                hdr3->owner = cpu_to_be64(dp->i_ino);
 606                uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
 607
 608        } else
 609                hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
 610
 611        bf = dp->d_ops->data_bestfree_p(hdr);
 612        bf[0].offset = cpu_to_be16(dp->d_ops->data_entry_offset);
 613        for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
 614                bf[i].length = 0;
 615                bf[i].offset = 0;
 616        }
 617
 618        /*
 619         * Set up an unused entry for the block's body.
 620         */
 621        dup = dp->d_ops->data_unused_p(hdr);
 622        dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 623
 624        t = mp->m_dirblksize - (uint)dp->d_ops->data_entry_offset;
 625        bf[0].length = cpu_to_be16(t);
 626        dup->length = cpu_to_be16(t);
 627        *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
 628        /*
 629         * Log it and return it.
 630         */
 631        xfs_dir2_data_log_header(tp, dp, bp);
 632        xfs_dir2_data_log_unused(tp, bp, dup);
 633        *bpp = bp;
 634        return 0;
 635}
 636
 637/*
 638 * Log an active data entry from the block.
 639 */
 640void
 641xfs_dir2_data_log_entry(
 642        struct xfs_trans        *tp,
 643        struct xfs_inode        *dp,
 644        struct xfs_buf          *bp,
 645        xfs_dir2_data_entry_t   *dep)           /* data entry pointer */
 646{
 647        struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 648
 649        ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 650               hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 651               hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 652               hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 653
 654        xfs_trans_log_buf(tp, bp, (uint)((char *)dep - (char *)hdr),
 655                (uint)((char *)(dp->d_ops->data_entry_tag_p(dep) + 1) -
 656                       (char *)hdr - 1));
 657}
 658
 659/*
 660 * Log a data block header.
 661 */
 662void
 663xfs_dir2_data_log_header(
 664        struct xfs_trans        *tp,
 665        struct xfs_inode        *dp,
 666        struct xfs_buf          *bp)
 667{
 668#ifdef DEBUG
 669        struct xfs_dir2_data_hdr *hdr = bp->b_addr;
 670
 671        ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 672               hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 673               hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 674               hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 675#endif
 676
 677        xfs_trans_log_buf(tp, bp, 0, dp->d_ops->data_entry_offset - 1);
 678}
 679
 680/*
 681 * Log a data unused entry.
 682 */
 683void
 684xfs_dir2_data_log_unused(
 685        struct xfs_trans        *tp,
 686        struct xfs_buf          *bp,
 687        xfs_dir2_data_unused_t  *dup)           /* data unused pointer */
 688{
 689        xfs_dir2_data_hdr_t     *hdr = bp->b_addr;
 690
 691        ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 692               hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 693               hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 694               hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 695
 696        /*
 697         * Log the first part of the unused entry.
 698         */
 699        xfs_trans_log_buf(tp, bp, (uint)((char *)dup - (char *)hdr),
 700                (uint)((char *)&dup->length + sizeof(dup->length) -
 701                       1 - (char *)hdr));
 702        /*
 703         * Log the end (tag) of the unused entry.
 704         */
 705        xfs_trans_log_buf(tp, bp,
 706                (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
 707                (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
 708                       sizeof(xfs_dir2_data_off_t) - 1));
 709}
 710
 711/*
 712 * Make a byte range in the data block unused.
 713 * Its current contents are unimportant.
 714 */
 715void
 716xfs_dir2_data_make_free(
 717        struct xfs_trans        *tp,
 718        struct xfs_inode        *dp,
 719        struct xfs_buf          *bp,
 720        xfs_dir2_data_aoff_t    offset,         /* starting byte offset */
 721        xfs_dir2_data_aoff_t    len,            /* length in bytes */
 722        int                     *needlogp,      /* out: log header */
 723        int                     *needscanp)     /* out: regen bestfree */
 724{
 725        xfs_dir2_data_hdr_t     *hdr;           /* data block pointer */
 726        xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 727        char                    *endptr;        /* end of data area */
 728        xfs_mount_t             *mp;            /* filesystem mount point */
 729        int                     needscan;       /* need to regen bestfree */
 730        xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 731        xfs_dir2_data_unused_t  *postdup;       /* unused entry after us */
 732        xfs_dir2_data_unused_t  *prevdup;       /* unused entry before us */
 733        struct xfs_dir2_data_free *bf;
 734
 735        mp = tp->t_mountp;
 736        hdr = bp->b_addr;
 737
 738        /*
 739         * Figure out where the end of the data area is.
 740         */
 741        if (hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 742            hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC))
 743                endptr = (char *)hdr + mp->m_dirblksize;
 744        else {
 745                xfs_dir2_block_tail_t   *btp;   /* block tail */
 746
 747                ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 748                        hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 749                btp = xfs_dir2_block_tail_p(mp, hdr);
 750                endptr = (char *)xfs_dir2_block_leaf_p(btp);
 751        }
 752        /*
 753         * If this isn't the start of the block, then back up to
 754         * the previous entry and see if it's free.
 755         */
 756        if (offset > dp->d_ops->data_entry_offset) {
 757                __be16                  *tagp;  /* tag just before us */
 758
 759                tagp = (__be16 *)((char *)hdr + offset) - 1;
 760                prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
 761                if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 762                        prevdup = NULL;
 763        } else
 764                prevdup = NULL;
 765        /*
 766         * If this isn't the end of the block, see if the entry after
 767         * us is free.
 768         */
 769        if ((char *)hdr + offset + len < endptr) {
 770                postdup =
 771                        (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 772                if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
 773                        postdup = NULL;
 774        } else
 775                postdup = NULL;
 776        ASSERT(*needscanp == 0);
 777        needscan = 0;
 778        /*
 779         * Previous and following entries are both free,
 780         * merge everything into a single free entry.
 781         */
 782        bf = dp->d_ops->data_bestfree_p(hdr);
 783        if (prevdup && postdup) {
 784                xfs_dir2_data_free_t    *dfp2;  /* another bestfree pointer */
 785
 786                /*
 787                 * See if prevdup and/or postdup are in bestfree table.
 788                 */
 789                dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
 790                dfp2 = xfs_dir2_data_freefind(hdr, bf, postdup);
 791                /*
 792                 * We need a rescan unless there are exactly 2 free entries
 793                 * namely our two.  Then we know what's happening, otherwise
 794                 * since the third bestfree is there, there might be more
 795                 * entries.
 796                 */
 797                needscan = (bf[2].length != 0);
 798                /*
 799                 * Fix up the new big freespace.
 800                 */
 801                be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
 802                *xfs_dir2_data_unused_tag_p(prevdup) =
 803                        cpu_to_be16((char *)prevdup - (char *)hdr);
 804                xfs_dir2_data_log_unused(tp, bp, prevdup);
 805                if (!needscan) {
 806                        /*
 807                         * Has to be the case that entries 0 and 1 are
 808                         * dfp and dfp2 (don't know which is which), and
 809                         * entry 2 is empty.
 810                         * Remove entry 1 first then entry 0.
 811                         */
 812                        ASSERT(dfp && dfp2);
 813                        if (dfp == &bf[1]) {
 814                                dfp = &bf[0];
 815                                ASSERT(dfp2 == dfp);
 816                                dfp2 = &bf[1];
 817                        }
 818                        xfs_dir2_data_freeremove(hdr, bf, dfp2, needlogp);
 819                        xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 820                        /*
 821                         * Now insert the new entry.
 822                         */
 823                        dfp = xfs_dir2_data_freeinsert(hdr, bf, prevdup,
 824                                                       needlogp);
 825                        ASSERT(dfp == &bf[0]);
 826                        ASSERT(dfp->length == prevdup->length);
 827                        ASSERT(!dfp[1].length);
 828                        ASSERT(!dfp[2].length);
 829                }
 830        }
 831        /*
 832         * The entry before us is free, merge with it.
 833         */
 834        else if (prevdup) {
 835                dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
 836                be16_add_cpu(&prevdup->length, len);
 837                *xfs_dir2_data_unused_tag_p(prevdup) =
 838                        cpu_to_be16((char *)prevdup - (char *)hdr);
 839                xfs_dir2_data_log_unused(tp, bp, prevdup);
 840                /*
 841                 * If the previous entry was in the table, the new entry
 842                 * is longer, so it will be in the table too.  Remove
 843                 * the old one and add the new one.
 844                 */
 845                if (dfp) {
 846                        xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 847                        xfs_dir2_data_freeinsert(hdr, bf, prevdup, needlogp);
 848                }
 849                /*
 850                 * Otherwise we need a scan if the new entry is big enough.
 851                 */
 852                else {
 853                        needscan = be16_to_cpu(prevdup->length) >
 854                                   be16_to_cpu(bf[2].length);
 855                }
 856        }
 857        /*
 858         * The following entry is free, merge with it.
 859         */
 860        else if (postdup) {
 861                dfp = xfs_dir2_data_freefind(hdr, bf, postdup);
 862                newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 863                newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 864                newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
 865                *xfs_dir2_data_unused_tag_p(newdup) =
 866                        cpu_to_be16((char *)newdup - (char *)hdr);
 867                xfs_dir2_data_log_unused(tp, bp, newdup);
 868                /*
 869                 * If the following entry was in the table, the new entry
 870                 * is longer, so it will be in the table too.  Remove
 871                 * the old one and add the new one.
 872                 */
 873                if (dfp) {
 874                        xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 875                        xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
 876                }
 877                /*
 878                 * Otherwise we need a scan if the new entry is big enough.
 879                 */
 880                else {
 881                        needscan = be16_to_cpu(newdup->length) >
 882                                   be16_to_cpu(bf[2].length);
 883                }
 884        }
 885        /*
 886         * Neither neighbor is free.  Make a new entry.
 887         */
 888        else {
 889                newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
 890                newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 891                newdup->length = cpu_to_be16(len);
 892                *xfs_dir2_data_unused_tag_p(newdup) =
 893                        cpu_to_be16((char *)newdup - (char *)hdr);
 894                xfs_dir2_data_log_unused(tp, bp, newdup);
 895                xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
 896        }
 897        *needscanp = needscan;
 898}
 899
 900/*
 901 * Take a byte range out of an existing unused space and make it un-free.
 902 */
 903void
 904xfs_dir2_data_use_free(
 905        struct xfs_trans        *tp,
 906        struct xfs_inode        *dp,
 907        struct xfs_buf          *bp,
 908        xfs_dir2_data_unused_t  *dup,           /* unused entry */
 909        xfs_dir2_data_aoff_t    offset,         /* starting offset to use */
 910        xfs_dir2_data_aoff_t    len,            /* length to use */
 911        int                     *needlogp,      /* out: need to log header */
 912        int                     *needscanp)     /* out: need regen bestfree */
 913{
 914        xfs_dir2_data_hdr_t     *hdr;           /* data block header */
 915        xfs_dir2_data_free_t    *dfp;           /* bestfree pointer */
 916        int                     matchback;      /* matches end of freespace */
 917        int                     matchfront;     /* matches start of freespace */
 918        int                     needscan;       /* need to regen bestfree */
 919        xfs_dir2_data_unused_t  *newdup;        /* new unused entry */
 920        xfs_dir2_data_unused_t  *newdup2;       /* another new unused entry */
 921        int                     oldlen;         /* old unused entry's length */
 922        struct xfs_dir2_data_free *bf;
 923
 924        hdr = bp->b_addr;
 925        ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
 926               hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
 927               hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
 928               hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
 929        ASSERT(be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG);
 930        ASSERT(offset >= (char *)dup - (char *)hdr);
 931        ASSERT(offset + len <= (char *)dup + be16_to_cpu(dup->length) - (char *)hdr);
 932        ASSERT((char *)dup - (char *)hdr == be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
 933        /*
 934         * Look up the entry in the bestfree table.
 935         */
 936        oldlen = be16_to_cpu(dup->length);
 937        bf = dp->d_ops->data_bestfree_p(hdr);
 938        dfp = xfs_dir2_data_freefind(hdr, bf, dup);
 939        ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
 940        /*
 941         * Check for alignment with front and back of the entry.
 942         */
 943        matchfront = (char *)dup - (char *)hdr == offset;
 944        matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
 945        ASSERT(*needscanp == 0);
 946        needscan = 0;
 947        /*
 948         * If we matched it exactly we just need to get rid of it from
 949         * the bestfree table.
 950         */
 951        if (matchfront && matchback) {
 952                if (dfp) {
 953                        needscan = (bf[2].offset != 0);
 954                        if (!needscan)
 955                                xfs_dir2_data_freeremove(hdr, bf, dfp,
 956                                                         needlogp);
 957                }
 958        }
 959        /*
 960         * We match the first part of the entry.
 961         * Make a new entry with the remaining freespace.
 962         */
 963        else if (matchfront) {
 964                newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
 965                newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
 966                newdup->length = cpu_to_be16(oldlen - len);
 967                *xfs_dir2_data_unused_tag_p(newdup) =
 968                        cpu_to_be16((char *)newdup - (char *)hdr);
 969                xfs_dir2_data_log_unused(tp, bp, newdup);
 970                /*
 971                 * If it was in the table, remove it and add the new one.
 972                 */
 973                if (dfp) {
 974                        xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
 975                        dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
 976                                                       needlogp);
 977                        ASSERT(dfp != NULL);
 978                        ASSERT(dfp->length == newdup->length);
 979                        ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
 980                        /*
 981                         * If we got inserted at the last slot,
 982                         * that means we don't know if there was a better
 983                         * choice for the last slot, or not.  Rescan.
 984                         */
 985                        needscan = dfp == &bf[2];
 986                }
 987        }
 988        /*
 989         * We match the last part of the entry.
 990         * Trim the allocated space off the tail of the entry.
 991         */
 992        else if (matchback) {
 993                newdup = dup;
 994                newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
 995                *xfs_dir2_data_unused_tag_p(newdup) =
 996                        cpu_to_be16((char *)newdup - (char *)hdr);
 997                xfs_dir2_data_log_unused(tp, bp, newdup);
 998                /*
 999                 * If it was in the table, remove it and add the new one.
1000                 */
1001                if (dfp) {
1002                        xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
1003                        dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
1004                                                       needlogp);
1005                        ASSERT(dfp != NULL);
1006                        ASSERT(dfp->length == newdup->length);
1007                        ASSERT(be16_to_cpu(dfp->offset) == (char *)newdup - (char *)hdr);
1008                        /*
1009                         * If we got inserted at the last slot,
1010                         * that means we don't know if there was a better
1011                         * choice for the last slot, or not.  Rescan.
1012                         */
1013                        needscan = dfp == &bf[2];
1014                }
1015        }
1016        /*
1017         * Poking out the middle of an entry.
1018         * Make two new entries.
1019         */
1020        else {
1021                newdup = dup;
1022                newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1023                *xfs_dir2_data_unused_tag_p(newdup) =
1024                        cpu_to_be16((char *)newdup - (char *)hdr);
1025                xfs_dir2_data_log_unused(tp, bp, newdup);
1026                newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1027                newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1028                newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1029                *xfs_dir2_data_unused_tag_p(newdup2) =
1030                        cpu_to_be16((char *)newdup2 - (char *)hdr);
1031                xfs_dir2_data_log_unused(tp, bp, newdup2);
1032                /*
1033                 * If the old entry was in the table, we need to scan
1034                 * if the 3rd entry was valid, since these entries
1035                 * are smaller than the old one.
1036                 * If we don't need to scan that means there were 1 or 2
1037                 * entries in the table, and removing the old and adding
1038                 * the 2 new will work.
1039                 */
1040                if (dfp) {
1041                        needscan = (bf[2].length != 0);
1042                        if (!needscan) {
1043                                xfs_dir2_data_freeremove(hdr, bf, dfp,
1044                                                         needlogp);
1045                                xfs_dir2_data_freeinsert(hdr, bf, newdup,
1046                                                         needlogp);
1047                                xfs_dir2_data_freeinsert(hdr, bf, newdup2,
1048                                                         needlogp);
1049                        }
1050                }
1051        }
1052        *needscanp = needscan;
1053}
1054