linux/fs/xfs/libxfs/xfs_attr_leaf.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) 2000-2005 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_shared.h"
  22#include "xfs_format.h"
  23#include "xfs_log_format.h"
  24#include "xfs_trans_resv.h"
  25#include "xfs_bit.h"
  26#include "xfs_sb.h"
  27#include "xfs_mount.h"
  28#include "xfs_da_format.h"
  29#include "xfs_da_btree.h"
  30#include "xfs_inode.h"
  31#include "xfs_trans.h"
  32#include "xfs_inode_item.h"
  33#include "xfs_bmap_btree.h"
  34#include "xfs_bmap.h"
  35#include "xfs_attr_sf.h"
  36#include "xfs_attr_remote.h"
  37#include "xfs_attr.h"
  38#include "xfs_attr_leaf.h"
  39#include "xfs_error.h"
  40#include "xfs_trace.h"
  41#include "xfs_buf_item.h"
  42#include "xfs_cksum.h"
  43#include "xfs_dir2.h"
  44#include "xfs_log.h"
  45
  46
  47/*
  48 * xfs_attr_leaf.c
  49 *
  50 * Routines to implement leaf blocks of attributes as Btrees of hashed names.
  51 */
  52
  53/*========================================================================
  54 * Function prototypes for the kernel.
  55 *========================================================================*/
  56
  57/*
  58 * Routines used for growing the Btree.
  59 */
  60STATIC int xfs_attr3_leaf_create(struct xfs_da_args *args,
  61                                 xfs_dablk_t which_block, struct xfs_buf **bpp);
  62STATIC int xfs_attr3_leaf_add_work(struct xfs_buf *leaf_buffer,
  63                                   struct xfs_attr3_icleaf_hdr *ichdr,
  64                                   struct xfs_da_args *args, int freemap_index);
  65STATIC void xfs_attr3_leaf_compact(struct xfs_da_args *args,
  66                                   struct xfs_attr3_icleaf_hdr *ichdr,
  67                                   struct xfs_buf *leaf_buffer);
  68STATIC void xfs_attr3_leaf_rebalance(xfs_da_state_t *state,
  69                                                   xfs_da_state_blk_t *blk1,
  70                                                   xfs_da_state_blk_t *blk2);
  71STATIC int xfs_attr3_leaf_figure_balance(xfs_da_state_t *state,
  72                        xfs_da_state_blk_t *leaf_blk_1,
  73                        struct xfs_attr3_icleaf_hdr *ichdr1,
  74                        xfs_da_state_blk_t *leaf_blk_2,
  75                        struct xfs_attr3_icleaf_hdr *ichdr2,
  76                        int *number_entries_in_blk1,
  77                        int *number_usedbytes_in_blk1);
  78
  79/*
  80 * Utility routines.
  81 */
  82STATIC void xfs_attr3_leaf_moveents(struct xfs_da_args *args,
  83                        struct xfs_attr_leafblock *src_leaf,
  84                        struct xfs_attr3_icleaf_hdr *src_ichdr, int src_start,
  85                        struct xfs_attr_leafblock *dst_leaf,
  86                        struct xfs_attr3_icleaf_hdr *dst_ichdr, int dst_start,
  87                        int move_count);
  88STATIC int xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index);
  89
  90/*
  91 * attr3 block 'firstused' conversion helpers.
  92 *
  93 * firstused refers to the offset of the first used byte of the nameval region
  94 * of an attr leaf block. The region starts at the tail of the block and expands
  95 * backwards towards the middle. As such, firstused is initialized to the block
  96 * size for an empty leaf block and is reduced from there.
  97 *
  98 * The attr3 block size is pegged to the fsb size and the maximum fsb is 64k.
  99 * The in-core firstused field is 32-bit and thus supports the maximum fsb size.
 100 * The on-disk field is only 16-bit, however, and overflows at 64k. Since this
 101 * only occurs at exactly 64k, we use zero as a magic on-disk value to represent
 102 * the attr block size. The following helpers manage the conversion between the
 103 * in-core and on-disk formats.
 104 */
 105
 106static void
 107xfs_attr3_leaf_firstused_from_disk(
 108        struct xfs_da_geometry          *geo,
 109        struct xfs_attr3_icleaf_hdr     *to,
 110        struct xfs_attr_leafblock       *from)
 111{
 112        struct xfs_attr3_leaf_hdr       *hdr3;
 113
 114        if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
 115                hdr3 = (struct xfs_attr3_leaf_hdr *) from;
 116                to->firstused = be16_to_cpu(hdr3->firstused);
 117        } else {
 118                to->firstused = be16_to_cpu(from->hdr.firstused);
 119        }
 120
 121        /*
 122         * Convert from the magic fsb size value to actual blocksize. This
 123         * should only occur for empty blocks when the block size overflows
 124         * 16-bits.
 125         */
 126        if (to->firstused == XFS_ATTR3_LEAF_NULLOFF) {
 127                ASSERT(!to->count && !to->usedbytes);
 128                ASSERT(geo->blksize > USHRT_MAX);
 129                to->firstused = geo->blksize;
 130        }
 131}
 132
 133static void
 134xfs_attr3_leaf_firstused_to_disk(
 135        struct xfs_da_geometry          *geo,
 136        struct xfs_attr_leafblock       *to,
 137        struct xfs_attr3_icleaf_hdr     *from)
 138{
 139        struct xfs_attr3_leaf_hdr       *hdr3;
 140        uint32_t                        firstused;
 141
 142        /* magic value should only be seen on disk */
 143        ASSERT(from->firstused != XFS_ATTR3_LEAF_NULLOFF);
 144
 145        /*
 146         * Scale down the 32-bit in-core firstused value to the 16-bit on-disk
 147         * value. This only overflows at the max supported value of 64k. Use the
 148         * magic on-disk value to represent block size in this case.
 149         */
 150        firstused = from->firstused;
 151        if (firstused > USHRT_MAX) {
 152                ASSERT(from->firstused == geo->blksize);
 153                firstused = XFS_ATTR3_LEAF_NULLOFF;
 154        }
 155
 156        if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
 157                hdr3 = (struct xfs_attr3_leaf_hdr *) to;
 158                hdr3->firstused = cpu_to_be16(firstused);
 159        } else {
 160                to->hdr.firstused = cpu_to_be16(firstused);
 161        }
 162}
 163
 164void
 165xfs_attr3_leaf_hdr_from_disk(
 166        struct xfs_da_geometry          *geo,
 167        struct xfs_attr3_icleaf_hdr     *to,
 168        struct xfs_attr_leafblock       *from)
 169{
 170        int     i;
 171
 172        ASSERT(from->hdr.info.magic == cpu_to_be16(XFS_ATTR_LEAF_MAGIC) ||
 173               from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC));
 174
 175        if (from->hdr.info.magic == cpu_to_be16(XFS_ATTR3_LEAF_MAGIC)) {
 176                struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)from;
 177
 178                to->forw = be32_to_cpu(hdr3->info.hdr.forw);
 179                to->back = be32_to_cpu(hdr3->info.hdr.back);
 180                to->magic = be16_to_cpu(hdr3->info.hdr.magic);
 181                to->count = be16_to_cpu(hdr3->count);
 182                to->usedbytes = be16_to_cpu(hdr3->usedbytes);
 183                xfs_attr3_leaf_firstused_from_disk(geo, to, from);
 184                to->holes = hdr3->holes;
 185
 186                for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
 187                        to->freemap[i].base = be16_to_cpu(hdr3->freemap[i].base);
 188                        to->freemap[i].size = be16_to_cpu(hdr3->freemap[i].size);
 189                }
 190                return;
 191        }
 192        to->forw = be32_to_cpu(from->hdr.info.forw);
 193        to->back = be32_to_cpu(from->hdr.info.back);
 194        to->magic = be16_to_cpu(from->hdr.info.magic);
 195        to->count = be16_to_cpu(from->hdr.count);
 196        to->usedbytes = be16_to_cpu(from->hdr.usedbytes);
 197        xfs_attr3_leaf_firstused_from_disk(geo, to, from);
 198        to->holes = from->hdr.holes;
 199
 200        for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
 201                to->freemap[i].base = be16_to_cpu(from->hdr.freemap[i].base);
 202                to->freemap[i].size = be16_to_cpu(from->hdr.freemap[i].size);
 203        }
 204}
 205
 206void
 207xfs_attr3_leaf_hdr_to_disk(
 208        struct xfs_da_geometry          *geo,
 209        struct xfs_attr_leafblock       *to,
 210        struct xfs_attr3_icleaf_hdr     *from)
 211{
 212        int                             i;
 213
 214        ASSERT(from->magic == XFS_ATTR_LEAF_MAGIC ||
 215               from->magic == XFS_ATTR3_LEAF_MAGIC);
 216
 217        if (from->magic == XFS_ATTR3_LEAF_MAGIC) {
 218                struct xfs_attr3_leaf_hdr *hdr3 = (struct xfs_attr3_leaf_hdr *)to;
 219
 220                hdr3->info.hdr.forw = cpu_to_be32(from->forw);
 221                hdr3->info.hdr.back = cpu_to_be32(from->back);
 222                hdr3->info.hdr.magic = cpu_to_be16(from->magic);
 223                hdr3->count = cpu_to_be16(from->count);
 224                hdr3->usedbytes = cpu_to_be16(from->usedbytes);
 225                xfs_attr3_leaf_firstused_to_disk(geo, to, from);
 226                hdr3->holes = from->holes;
 227                hdr3->pad1 = 0;
 228
 229                for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
 230                        hdr3->freemap[i].base = cpu_to_be16(from->freemap[i].base);
 231                        hdr3->freemap[i].size = cpu_to_be16(from->freemap[i].size);
 232                }
 233                return;
 234        }
 235        to->hdr.info.forw = cpu_to_be32(from->forw);
 236        to->hdr.info.back = cpu_to_be32(from->back);
 237        to->hdr.info.magic = cpu_to_be16(from->magic);
 238        to->hdr.count = cpu_to_be16(from->count);
 239        to->hdr.usedbytes = cpu_to_be16(from->usedbytes);
 240        xfs_attr3_leaf_firstused_to_disk(geo, to, from);
 241        to->hdr.holes = from->holes;
 242        to->hdr.pad1 = 0;
 243
 244        for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
 245                to->hdr.freemap[i].base = cpu_to_be16(from->freemap[i].base);
 246                to->hdr.freemap[i].size = cpu_to_be16(from->freemap[i].size);
 247        }
 248}
 249
 250static bool
 251xfs_attr3_leaf_verify(
 252        struct xfs_buf          *bp)
 253{
 254        struct xfs_mount        *mp = bp->b_target->bt_mount;
 255        struct xfs_attr_leafblock *leaf = bp->b_addr;
 256        struct xfs_attr3_icleaf_hdr ichdr;
 257
 258        xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
 259
 260        if (xfs_sb_version_hascrc(&mp->m_sb)) {
 261                struct xfs_da3_node_hdr *hdr3 = bp->b_addr;
 262
 263                if (ichdr.magic != XFS_ATTR3_LEAF_MAGIC)
 264                        return false;
 265
 266                if (!uuid_equal(&hdr3->info.uuid, &mp->m_sb.sb_meta_uuid))
 267                        return false;
 268                if (be64_to_cpu(hdr3->info.blkno) != bp->b_bn)
 269                        return false;
 270                if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->info.lsn)))
 271                        return false;
 272        } else {
 273                if (ichdr.magic != XFS_ATTR_LEAF_MAGIC)
 274                        return false;
 275        }
 276        if (ichdr.count == 0)
 277                return false;
 278
 279        /* XXX: need to range check rest of attr header values */
 280        /* XXX: hash order check? */
 281
 282        return true;
 283}
 284
 285static void
 286xfs_attr3_leaf_write_verify(
 287        struct xfs_buf  *bp)
 288{
 289        struct xfs_mount        *mp = bp->b_target->bt_mount;
 290        struct xfs_buf_log_item *bip = bp->b_fspriv;
 291        struct xfs_attr3_leaf_hdr *hdr3 = bp->b_addr;
 292
 293        if (!xfs_attr3_leaf_verify(bp)) {
 294                xfs_buf_ioerror(bp, -EFSCORRUPTED);
 295                xfs_verifier_error(bp);
 296                return;
 297        }
 298
 299        if (!xfs_sb_version_hascrc(&mp->m_sb))
 300                return;
 301
 302        if (bip)
 303                hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
 304
 305        xfs_buf_update_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF);
 306}
 307
 308/*
 309 * leaf/node format detection on trees is sketchy, so a node read can be done on
 310 * leaf level blocks when detection identifies the tree as a node format tree
 311 * incorrectly. In this case, we need to swap the verifier to match the correct
 312 * format of the block being read.
 313 */
 314static void
 315xfs_attr3_leaf_read_verify(
 316        struct xfs_buf          *bp)
 317{
 318        struct xfs_mount        *mp = bp->b_target->bt_mount;
 319
 320        if (xfs_sb_version_hascrc(&mp->m_sb) &&
 321             !xfs_buf_verify_cksum(bp, XFS_ATTR3_LEAF_CRC_OFF))
 322                xfs_buf_ioerror(bp, -EFSBADCRC);
 323        else if (!xfs_attr3_leaf_verify(bp))
 324                xfs_buf_ioerror(bp, -EFSCORRUPTED);
 325
 326        if (bp->b_error)
 327                xfs_verifier_error(bp);
 328}
 329
 330const struct xfs_buf_ops xfs_attr3_leaf_buf_ops = {
 331        .name = "xfs_attr3_leaf",
 332        .verify_read = xfs_attr3_leaf_read_verify,
 333        .verify_write = xfs_attr3_leaf_write_verify,
 334};
 335
 336int
 337xfs_attr3_leaf_read(
 338        struct xfs_trans        *tp,
 339        struct xfs_inode        *dp,
 340        xfs_dablk_t             bno,
 341        xfs_daddr_t             mappedbno,
 342        struct xfs_buf          **bpp)
 343{
 344        int                     err;
 345
 346        err = xfs_da_read_buf(tp, dp, bno, mappedbno, bpp,
 347                                XFS_ATTR_FORK, &xfs_attr3_leaf_buf_ops);
 348        if (!err && tp)
 349                xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_ATTR_LEAF_BUF);
 350        return err;
 351}
 352
 353/*========================================================================
 354 * Namespace helper routines
 355 *========================================================================*/
 356
 357/*
 358 * If namespace bits don't match return 0.
 359 * If all match then return 1.
 360 */
 361STATIC int
 362xfs_attr_namesp_match(int arg_flags, int ondisk_flags)
 363{
 364        return XFS_ATTR_NSP_ONDISK(ondisk_flags) == XFS_ATTR_NSP_ARGS_TO_ONDISK(arg_flags);
 365}
 366
 367
 368/*========================================================================
 369 * External routines when attribute fork size < XFS_LITINO(mp).
 370 *========================================================================*/
 371
 372/*
 373 * Query whether the requested number of additional bytes of extended
 374 * attribute space will be able to fit inline.
 375 *
 376 * Returns zero if not, else the di_forkoff fork offset to be used in the
 377 * literal area for attribute data once the new bytes have been added.
 378 *
 379 * di_forkoff must be 8 byte aligned, hence is stored as a >>3 value;
 380 * special case for dev/uuid inodes, they have fixed size data forks.
 381 */
 382int
 383xfs_attr_shortform_bytesfit(xfs_inode_t *dp, int bytes)
 384{
 385        int offset;
 386        int minforkoff; /* lower limit on valid forkoff locations */
 387        int maxforkoff; /* upper limit on valid forkoff locations */
 388        int dsize;
 389        xfs_mount_t *mp = dp->i_mount;
 390
 391        /* rounded down */
 392        offset = (XFS_LITINO(mp, dp->i_d.di_version) - bytes) >> 3;
 393
 394        switch (dp->i_d.di_format) {
 395        case XFS_DINODE_FMT_DEV:
 396                minforkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
 397                return (offset >= minforkoff) ? minforkoff : 0;
 398        case XFS_DINODE_FMT_UUID:
 399                minforkoff = roundup(sizeof(uuid_t), 8) >> 3;
 400                return (offset >= minforkoff) ? minforkoff : 0;
 401        }
 402
 403        /*
 404         * If the requested numbers of bytes is smaller or equal to the
 405         * current attribute fork size we can always proceed.
 406         *
 407         * Note that if_bytes in the data fork might actually be larger than
 408         * the current data fork size is due to delalloc extents. In that
 409         * case either the extent count will go down when they are converted
 410         * to real extents, or the delalloc conversion will take care of the
 411         * literal area rebalancing.
 412         */
 413        if (bytes <= XFS_IFORK_ASIZE(dp))
 414                return dp->i_d.di_forkoff;
 415
 416        /*
 417         * For attr2 we can try to move the forkoff if there is space in the
 418         * literal area, but for the old format we are done if there is no
 419         * space in the fixed attribute fork.
 420         */
 421        if (!(mp->m_flags & XFS_MOUNT_ATTR2))
 422                return 0;
 423
 424        dsize = dp->i_df.if_bytes;
 425
 426        switch (dp->i_d.di_format) {
 427        case XFS_DINODE_FMT_EXTENTS:
 428                /*
 429                 * If there is no attr fork and the data fork is extents, 
 430                 * determine if creating the default attr fork will result
 431                 * in the extents form migrating to btree. If so, the
 432                 * minimum offset only needs to be the space required for
 433                 * the btree root.
 434                 */
 435                if (!dp->i_d.di_forkoff && dp->i_df.if_bytes >
 436                    xfs_default_attroffset(dp))
 437                        dsize = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
 438                break;
 439        case XFS_DINODE_FMT_BTREE:
 440                /*
 441                 * If we have a data btree then keep forkoff if we have one,
 442                 * otherwise we are adding a new attr, so then we set
 443                 * minforkoff to where the btree root can finish so we have
 444                 * plenty of room for attrs
 445                 */
 446                if (dp->i_d.di_forkoff) {
 447                        if (offset < dp->i_d.di_forkoff)
 448                                return 0;
 449                        return dp->i_d.di_forkoff;
 450                }
 451                dsize = XFS_BMAP_BROOT_SPACE(mp, dp->i_df.if_broot);
 452                break;
 453        }
 454
 455        /*
 456         * A data fork btree root must have space for at least
 457         * MINDBTPTRS key/ptr pairs if the data fork is small or empty.
 458         */
 459        minforkoff = MAX(dsize, XFS_BMDR_SPACE_CALC(MINDBTPTRS));
 460        minforkoff = roundup(minforkoff, 8) >> 3;
 461
 462        /* attr fork btree root can have at least this many key/ptr pairs */
 463        maxforkoff = XFS_LITINO(mp, dp->i_d.di_version) -
 464                        XFS_BMDR_SPACE_CALC(MINABTPTRS);
 465        maxforkoff = maxforkoff >> 3;   /* rounded down */
 466
 467        if (offset >= maxforkoff)
 468                return maxforkoff;
 469        if (offset >= minforkoff)
 470                return offset;
 471        return 0;
 472}
 473
 474/*
 475 * Switch on the ATTR2 superblock bit (implies also FEATURES2)
 476 */
 477STATIC void
 478xfs_sbversion_add_attr2(xfs_mount_t *mp, xfs_trans_t *tp)
 479{
 480        if ((mp->m_flags & XFS_MOUNT_ATTR2) &&
 481            !(xfs_sb_version_hasattr2(&mp->m_sb))) {
 482                spin_lock(&mp->m_sb_lock);
 483                if (!xfs_sb_version_hasattr2(&mp->m_sb)) {
 484                        xfs_sb_version_addattr2(&mp->m_sb);
 485                        spin_unlock(&mp->m_sb_lock);
 486                        xfs_log_sb(tp);
 487                } else
 488                        spin_unlock(&mp->m_sb_lock);
 489        }
 490}
 491
 492/*
 493 * Create the initial contents of a shortform attribute list.
 494 */
 495void
 496xfs_attr_shortform_create(xfs_da_args_t *args)
 497{
 498        xfs_attr_sf_hdr_t *hdr;
 499        xfs_inode_t *dp;
 500        xfs_ifork_t *ifp;
 501
 502        trace_xfs_attr_sf_create(args);
 503
 504        dp = args->dp;
 505        ASSERT(dp != NULL);
 506        ifp = dp->i_afp;
 507        ASSERT(ifp != NULL);
 508        ASSERT(ifp->if_bytes == 0);
 509        if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) {
 510                ifp->if_flags &= ~XFS_IFEXTENTS;        /* just in case */
 511                dp->i_d.di_aformat = XFS_DINODE_FMT_LOCAL;
 512                ifp->if_flags |= XFS_IFINLINE;
 513        } else {
 514                ASSERT(ifp->if_flags & XFS_IFINLINE);
 515        }
 516        xfs_idata_realloc(dp, sizeof(*hdr), XFS_ATTR_FORK);
 517        hdr = (xfs_attr_sf_hdr_t *)ifp->if_u1.if_data;
 518        hdr->count = 0;
 519        hdr->totsize = cpu_to_be16(sizeof(*hdr));
 520        xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
 521}
 522
 523/*
 524 * Add a name/value pair to the shortform attribute list.
 525 * Overflow from the inode has already been checked for.
 526 */
 527void
 528xfs_attr_shortform_add(xfs_da_args_t *args, int forkoff)
 529{
 530        xfs_attr_shortform_t *sf;
 531        xfs_attr_sf_entry_t *sfe;
 532        int i, offset, size;
 533        xfs_mount_t *mp;
 534        xfs_inode_t *dp;
 535        xfs_ifork_t *ifp;
 536
 537        trace_xfs_attr_sf_add(args);
 538
 539        dp = args->dp;
 540        mp = dp->i_mount;
 541        dp->i_d.di_forkoff = forkoff;
 542
 543        ifp = dp->i_afp;
 544        ASSERT(ifp->if_flags & XFS_IFINLINE);
 545        sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
 546        sfe = &sf->list[0];
 547        for (i = 0; i < sf->hdr.count; sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
 548#ifdef DEBUG
 549                if (sfe->namelen != args->namelen)
 550                        continue;
 551                if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
 552                        continue;
 553                if (!xfs_attr_namesp_match(args->flags, sfe->flags))
 554                        continue;
 555                ASSERT(0);
 556#endif
 557        }
 558
 559        offset = (char *)sfe - (char *)sf;
 560        size = XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
 561        xfs_idata_realloc(dp, size, XFS_ATTR_FORK);
 562        sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
 563        sfe = (xfs_attr_sf_entry_t *)((char *)sf + offset);
 564
 565        sfe->namelen = args->namelen;
 566        sfe->valuelen = args->valuelen;
 567        sfe->flags = XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
 568        memcpy(sfe->nameval, args->name, args->namelen);
 569        memcpy(&sfe->nameval[args->namelen], args->value, args->valuelen);
 570        sf->hdr.count++;
 571        be16_add_cpu(&sf->hdr.totsize, size);
 572        xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_ADATA);
 573
 574        xfs_sbversion_add_attr2(mp, args->trans);
 575}
 576
 577/*
 578 * After the last attribute is removed revert to original inode format,
 579 * making all literal area available to the data fork once more.
 580 */
 581void
 582xfs_attr_fork_remove(
 583        struct xfs_inode        *ip,
 584        struct xfs_trans        *tp)
 585{
 586        xfs_idestroy_fork(ip, XFS_ATTR_FORK);
 587        ip->i_d.di_forkoff = 0;
 588        ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
 589
 590        ASSERT(ip->i_d.di_anextents == 0);
 591        ASSERT(ip->i_afp == NULL);
 592
 593        xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
 594}
 595
 596/*
 597 * Remove an attribute from the shortform attribute list structure.
 598 */
 599int
 600xfs_attr_shortform_remove(xfs_da_args_t *args)
 601{
 602        xfs_attr_shortform_t *sf;
 603        xfs_attr_sf_entry_t *sfe;
 604        int base, size=0, end, totsize, i;
 605        xfs_mount_t *mp;
 606        xfs_inode_t *dp;
 607
 608        trace_xfs_attr_sf_remove(args);
 609
 610        dp = args->dp;
 611        mp = dp->i_mount;
 612        base = sizeof(xfs_attr_sf_hdr_t);
 613        sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
 614        sfe = &sf->list[0];
 615        end = sf->hdr.count;
 616        for (i = 0; i < end; sfe = XFS_ATTR_SF_NEXTENTRY(sfe),
 617                                        base += size, i++) {
 618                size = XFS_ATTR_SF_ENTSIZE(sfe);
 619                if (sfe->namelen != args->namelen)
 620                        continue;
 621                if (memcmp(sfe->nameval, args->name, args->namelen) != 0)
 622                        continue;
 623                if (!xfs_attr_namesp_match(args->flags, sfe->flags))
 624                        continue;
 625                break;
 626        }
 627        if (i == end)
 628                return -ENOATTR;
 629
 630        /*
 631         * Fix up the attribute fork data, covering the hole
 632         */
 633        end = base + size;
 634        totsize = be16_to_cpu(sf->hdr.totsize);
 635        if (end != totsize)
 636                memmove(&((char *)sf)[base], &((char *)sf)[end], totsize - end);
 637        sf->hdr.count--;
 638        be16_add_cpu(&sf->hdr.totsize, -size);
 639
 640        /*
 641         * Fix up the start offset of the attribute fork
 642         */
 643        totsize -= size;
 644        if (totsize == sizeof(xfs_attr_sf_hdr_t) &&
 645            (mp->m_flags & XFS_MOUNT_ATTR2) &&
 646            (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
 647            !(args->op_flags & XFS_DA_OP_ADDNAME)) {
 648                xfs_attr_fork_remove(dp, args->trans);
 649        } else {
 650                xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
 651                dp->i_d.di_forkoff = xfs_attr_shortform_bytesfit(dp, totsize);
 652                ASSERT(dp->i_d.di_forkoff);
 653                ASSERT(totsize > sizeof(xfs_attr_sf_hdr_t) ||
 654                                (args->op_flags & XFS_DA_OP_ADDNAME) ||
 655                                !(mp->m_flags & XFS_MOUNT_ATTR2) ||
 656                                dp->i_d.di_format == XFS_DINODE_FMT_BTREE);
 657                xfs_trans_log_inode(args->trans, dp,
 658                                        XFS_ILOG_CORE | XFS_ILOG_ADATA);
 659        }
 660
 661        xfs_sbversion_add_attr2(mp, args->trans);
 662
 663        return 0;
 664}
 665
 666/*
 667 * Look up a name in a shortform attribute list structure.
 668 */
 669/*ARGSUSED*/
 670int
 671xfs_attr_shortform_lookup(xfs_da_args_t *args)
 672{
 673        xfs_attr_shortform_t *sf;
 674        xfs_attr_sf_entry_t *sfe;
 675        int i;
 676        xfs_ifork_t *ifp;
 677
 678        trace_xfs_attr_sf_lookup(args);
 679
 680        ifp = args->dp->i_afp;
 681        ASSERT(ifp->if_flags & XFS_IFINLINE);
 682        sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
 683        sfe = &sf->list[0];
 684        for (i = 0; i < sf->hdr.count;
 685                                sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
 686                if (sfe->namelen != args->namelen)
 687                        continue;
 688                if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
 689                        continue;
 690                if (!xfs_attr_namesp_match(args->flags, sfe->flags))
 691                        continue;
 692                return -EEXIST;
 693        }
 694        return -ENOATTR;
 695}
 696
 697/*
 698 * Look up a name in a shortform attribute list structure.
 699 */
 700/*ARGSUSED*/
 701int
 702xfs_attr_shortform_getvalue(xfs_da_args_t *args)
 703{
 704        xfs_attr_shortform_t *sf;
 705        xfs_attr_sf_entry_t *sfe;
 706        int i;
 707
 708        ASSERT(args->dp->i_afp->if_flags == XFS_IFINLINE);
 709        sf = (xfs_attr_shortform_t *)args->dp->i_afp->if_u1.if_data;
 710        sfe = &sf->list[0];
 711        for (i = 0; i < sf->hdr.count;
 712                                sfe = XFS_ATTR_SF_NEXTENTRY(sfe), i++) {
 713                if (sfe->namelen != args->namelen)
 714                        continue;
 715                if (memcmp(args->name, sfe->nameval, args->namelen) != 0)
 716                        continue;
 717                if (!xfs_attr_namesp_match(args->flags, sfe->flags))
 718                        continue;
 719                if (args->flags & ATTR_KERNOVAL) {
 720                        args->valuelen = sfe->valuelen;
 721                        return -EEXIST;
 722                }
 723                if (args->valuelen < sfe->valuelen) {
 724                        args->valuelen = sfe->valuelen;
 725                        return -ERANGE;
 726                }
 727                args->valuelen = sfe->valuelen;
 728                memcpy(args->value, &sfe->nameval[args->namelen],
 729                                                    args->valuelen);
 730                return -EEXIST;
 731        }
 732        return -ENOATTR;
 733}
 734
 735/*
 736 * Convert from using the shortform to the leaf.
 737 */
 738int
 739xfs_attr_shortform_to_leaf(xfs_da_args_t *args)
 740{
 741        xfs_inode_t *dp;
 742        xfs_attr_shortform_t *sf;
 743        xfs_attr_sf_entry_t *sfe;
 744        xfs_da_args_t nargs;
 745        char *tmpbuffer;
 746        int error, i, size;
 747        xfs_dablk_t blkno;
 748        struct xfs_buf *bp;
 749        xfs_ifork_t *ifp;
 750
 751        trace_xfs_attr_sf_to_leaf(args);
 752
 753        dp = args->dp;
 754        ifp = dp->i_afp;
 755        sf = (xfs_attr_shortform_t *)ifp->if_u1.if_data;
 756        size = be16_to_cpu(sf->hdr.totsize);
 757        tmpbuffer = kmem_alloc(size, KM_SLEEP);
 758        ASSERT(tmpbuffer != NULL);
 759        memcpy(tmpbuffer, ifp->if_u1.if_data, size);
 760        sf = (xfs_attr_shortform_t *)tmpbuffer;
 761
 762        xfs_idata_realloc(dp, -size, XFS_ATTR_FORK);
 763        xfs_bmap_local_to_extents_empty(dp, XFS_ATTR_FORK);
 764
 765        bp = NULL;
 766        error = xfs_da_grow_inode(args, &blkno);
 767        if (error) {
 768                /*
 769                 * If we hit an IO error middle of the transaction inside
 770                 * grow_inode(), we may have inconsistent data. Bail out.
 771                 */
 772                if (error == -EIO)
 773                        goto out;
 774                xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
 775                memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
 776                goto out;
 777        }
 778
 779        ASSERT(blkno == 0);
 780        error = xfs_attr3_leaf_create(args, blkno, &bp);
 781        if (error) {
 782                error = xfs_da_shrink_inode(args, 0, bp);
 783                bp = NULL;
 784                if (error)
 785                        goto out;
 786                xfs_idata_realloc(dp, size, XFS_ATTR_FORK);     /* try to put */
 787                memcpy(ifp->if_u1.if_data, tmpbuffer, size);    /* it back */
 788                goto out;
 789        }
 790
 791        memset((char *)&nargs, 0, sizeof(nargs));
 792        nargs.dp = dp;
 793        nargs.geo = args->geo;
 794        nargs.firstblock = args->firstblock;
 795        nargs.flist = args->flist;
 796        nargs.total = args->total;
 797        nargs.whichfork = XFS_ATTR_FORK;
 798        nargs.trans = args->trans;
 799        nargs.op_flags = XFS_DA_OP_OKNOENT;
 800
 801        sfe = &sf->list[0];
 802        for (i = 0; i < sf->hdr.count; i++) {
 803                nargs.name = sfe->nameval;
 804                nargs.namelen = sfe->namelen;
 805                nargs.value = &sfe->nameval[nargs.namelen];
 806                nargs.valuelen = sfe->valuelen;
 807                nargs.hashval = xfs_da_hashname(sfe->nameval,
 808                                                sfe->namelen);
 809                nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(sfe->flags);
 810                error = xfs_attr3_leaf_lookup_int(bp, &nargs); /* set a->index */
 811                ASSERT(error == -ENOATTR);
 812                error = xfs_attr3_leaf_add(bp, &nargs);
 813                ASSERT(error != -ENOSPC);
 814                if (error)
 815                        goto out;
 816                sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
 817        }
 818        error = 0;
 819
 820out:
 821        kmem_free(tmpbuffer);
 822        return error;
 823}
 824
 825/*
 826 * Check a leaf attribute block to see if all the entries would fit into
 827 * a shortform attribute list.
 828 */
 829int
 830xfs_attr_shortform_allfit(
 831        struct xfs_buf          *bp,
 832        struct xfs_inode        *dp)
 833{
 834        struct xfs_attr_leafblock *leaf;
 835        struct xfs_attr_leaf_entry *entry;
 836        xfs_attr_leaf_name_local_t *name_loc;
 837        struct xfs_attr3_icleaf_hdr leafhdr;
 838        int                     bytes;
 839        int                     i;
 840        struct xfs_mount        *mp = bp->b_target->bt_mount;
 841
 842        leaf = bp->b_addr;
 843        xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
 844        entry = xfs_attr3_leaf_entryp(leaf);
 845
 846        bytes = sizeof(struct xfs_attr_sf_hdr);
 847        for (i = 0; i < leafhdr.count; entry++, i++) {
 848                if (entry->flags & XFS_ATTR_INCOMPLETE)
 849                        continue;               /* don't copy partial entries */
 850                if (!(entry->flags & XFS_ATTR_LOCAL))
 851                        return 0;
 852                name_loc = xfs_attr3_leaf_name_local(leaf, i);
 853                if (name_loc->namelen >= XFS_ATTR_SF_ENTSIZE_MAX)
 854                        return 0;
 855                if (be16_to_cpu(name_loc->valuelen) >= XFS_ATTR_SF_ENTSIZE_MAX)
 856                        return 0;
 857                bytes += sizeof(struct xfs_attr_sf_entry) - 1
 858                                + name_loc->namelen
 859                                + be16_to_cpu(name_loc->valuelen);
 860        }
 861        if ((dp->i_mount->m_flags & XFS_MOUNT_ATTR2) &&
 862            (dp->i_d.di_format != XFS_DINODE_FMT_BTREE) &&
 863            (bytes == sizeof(struct xfs_attr_sf_hdr)))
 864                return -1;
 865        return xfs_attr_shortform_bytesfit(dp, bytes);
 866}
 867
 868/*
 869 * Convert a leaf attribute list to shortform attribute list
 870 */
 871int
 872xfs_attr3_leaf_to_shortform(
 873        struct xfs_buf          *bp,
 874        struct xfs_da_args      *args,
 875        int                     forkoff)
 876{
 877        struct xfs_attr_leafblock *leaf;
 878        struct xfs_attr3_icleaf_hdr ichdr;
 879        struct xfs_attr_leaf_entry *entry;
 880        struct xfs_attr_leaf_name_local *name_loc;
 881        struct xfs_da_args      nargs;
 882        struct xfs_inode        *dp = args->dp;
 883        char                    *tmpbuffer;
 884        int                     error;
 885        int                     i;
 886
 887        trace_xfs_attr_leaf_to_sf(args);
 888
 889        tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
 890        if (!tmpbuffer)
 891                return -ENOMEM;
 892
 893        memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
 894
 895        leaf = (xfs_attr_leafblock_t *)tmpbuffer;
 896        xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
 897        entry = xfs_attr3_leaf_entryp(leaf);
 898
 899        /* XXX (dgc): buffer is about to be marked stale - why zero it? */
 900        memset(bp->b_addr, 0, args->geo->blksize);
 901
 902        /*
 903         * Clean out the prior contents of the attribute list.
 904         */
 905        error = xfs_da_shrink_inode(args, 0, bp);
 906        if (error)
 907                goto out;
 908
 909        if (forkoff == -1) {
 910                ASSERT(dp->i_mount->m_flags & XFS_MOUNT_ATTR2);
 911                ASSERT(dp->i_d.di_format != XFS_DINODE_FMT_BTREE);
 912                xfs_attr_fork_remove(dp, args->trans);
 913                goto out;
 914        }
 915
 916        xfs_attr_shortform_create(args);
 917
 918        /*
 919         * Copy the attributes
 920         */
 921        memset((char *)&nargs, 0, sizeof(nargs));
 922        nargs.geo = args->geo;
 923        nargs.dp = dp;
 924        nargs.firstblock = args->firstblock;
 925        nargs.flist = args->flist;
 926        nargs.total = args->total;
 927        nargs.whichfork = XFS_ATTR_FORK;
 928        nargs.trans = args->trans;
 929        nargs.op_flags = XFS_DA_OP_OKNOENT;
 930
 931        for (i = 0; i < ichdr.count; entry++, i++) {
 932                if (entry->flags & XFS_ATTR_INCOMPLETE)
 933                        continue;       /* don't copy partial entries */
 934                if (!entry->nameidx)
 935                        continue;
 936                ASSERT(entry->flags & XFS_ATTR_LOCAL);
 937                name_loc = xfs_attr3_leaf_name_local(leaf, i);
 938                nargs.name = name_loc->nameval;
 939                nargs.namelen = name_loc->namelen;
 940                nargs.value = &name_loc->nameval[nargs.namelen];
 941                nargs.valuelen = be16_to_cpu(name_loc->valuelen);
 942                nargs.hashval = be32_to_cpu(entry->hashval);
 943                nargs.flags = XFS_ATTR_NSP_ONDISK_TO_ARGS(entry->flags);
 944                xfs_attr_shortform_add(&nargs, forkoff);
 945        }
 946        error = 0;
 947
 948out:
 949        kmem_free(tmpbuffer);
 950        return error;
 951}
 952
 953/*
 954 * Convert from using a single leaf to a root node and a leaf.
 955 */
 956int
 957xfs_attr3_leaf_to_node(
 958        struct xfs_da_args      *args)
 959{
 960        struct xfs_attr_leafblock *leaf;
 961        struct xfs_attr3_icleaf_hdr icleafhdr;
 962        struct xfs_attr_leaf_entry *entries;
 963        struct xfs_da_node_entry *btree;
 964        struct xfs_da3_icnode_hdr icnodehdr;
 965        struct xfs_da_intnode   *node;
 966        struct xfs_inode        *dp = args->dp;
 967        struct xfs_mount        *mp = dp->i_mount;
 968        struct xfs_buf          *bp1 = NULL;
 969        struct xfs_buf          *bp2 = NULL;
 970        xfs_dablk_t             blkno;
 971        int                     error;
 972
 973        trace_xfs_attr_leaf_to_node(args);
 974
 975        error = xfs_da_grow_inode(args, &blkno);
 976        if (error)
 977                goto out;
 978        error = xfs_attr3_leaf_read(args->trans, dp, 0, -1, &bp1);
 979        if (error)
 980                goto out;
 981
 982        error = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp2, XFS_ATTR_FORK);
 983        if (error)
 984                goto out;
 985
 986        /* copy leaf to new buffer, update identifiers */
 987        xfs_trans_buf_set_type(args->trans, bp2, XFS_BLFT_ATTR_LEAF_BUF);
 988        bp2->b_ops = bp1->b_ops;
 989        memcpy(bp2->b_addr, bp1->b_addr, args->geo->blksize);
 990        if (xfs_sb_version_hascrc(&mp->m_sb)) {
 991                struct xfs_da3_blkinfo *hdr3 = bp2->b_addr;
 992                hdr3->blkno = cpu_to_be64(bp2->b_bn);
 993        }
 994        xfs_trans_log_buf(args->trans, bp2, 0, args->geo->blksize - 1);
 995
 996        /*
 997         * Set up the new root node.
 998         */
 999        error = xfs_da3_node_create(args, 0, 1, &bp1, XFS_ATTR_FORK);
1000        if (error)
1001                goto out;
1002        node = bp1->b_addr;
1003        dp->d_ops->node_hdr_from_disk(&icnodehdr, node);
1004        btree = dp->d_ops->node_tree_p(node);
1005
1006        leaf = bp2->b_addr;
1007        xfs_attr3_leaf_hdr_from_disk(args->geo, &icleafhdr, leaf);
1008        entries = xfs_attr3_leaf_entryp(leaf);
1009
1010        /* both on-disk, don't endian-flip twice */
1011        btree[0].hashval = entries[icleafhdr.count - 1].hashval;
1012        btree[0].before = cpu_to_be32(blkno);
1013        icnodehdr.count = 1;
1014        dp->d_ops->node_hdr_to_disk(node, &icnodehdr);
1015        xfs_trans_log_buf(args->trans, bp1, 0, args->geo->blksize - 1);
1016        error = 0;
1017out:
1018        return error;
1019}
1020
1021/*========================================================================
1022 * Routines used for growing the Btree.
1023 *========================================================================*/
1024
1025/*
1026 * Create the initial contents of a leaf attribute list
1027 * or a leaf in a node attribute list.
1028 */
1029STATIC int
1030xfs_attr3_leaf_create(
1031        struct xfs_da_args      *args,
1032        xfs_dablk_t             blkno,
1033        struct xfs_buf          **bpp)
1034{
1035        struct xfs_attr_leafblock *leaf;
1036        struct xfs_attr3_icleaf_hdr ichdr;
1037        struct xfs_inode        *dp = args->dp;
1038        struct xfs_mount        *mp = dp->i_mount;
1039        struct xfs_buf          *bp;
1040        int                     error;
1041
1042        trace_xfs_attr_leaf_create(args);
1043
1044        error = xfs_da_get_buf(args->trans, args->dp, blkno, -1, &bp,
1045                                            XFS_ATTR_FORK);
1046        if (error)
1047                return error;
1048        bp->b_ops = &xfs_attr3_leaf_buf_ops;
1049        xfs_trans_buf_set_type(args->trans, bp, XFS_BLFT_ATTR_LEAF_BUF);
1050        leaf = bp->b_addr;
1051        memset(leaf, 0, args->geo->blksize);
1052
1053        memset(&ichdr, 0, sizeof(ichdr));
1054        ichdr.firstused = args->geo->blksize;
1055
1056        if (xfs_sb_version_hascrc(&mp->m_sb)) {
1057                struct xfs_da3_blkinfo *hdr3 = bp->b_addr;
1058
1059                ichdr.magic = XFS_ATTR3_LEAF_MAGIC;
1060
1061                hdr3->blkno = cpu_to_be64(bp->b_bn);
1062                hdr3->owner = cpu_to_be64(dp->i_ino);
1063                uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
1064
1065                ichdr.freemap[0].base = sizeof(struct xfs_attr3_leaf_hdr);
1066        } else {
1067                ichdr.magic = XFS_ATTR_LEAF_MAGIC;
1068                ichdr.freemap[0].base = sizeof(struct xfs_attr_leaf_hdr);
1069        }
1070        ichdr.freemap[0].size = ichdr.firstused - ichdr.freemap[0].base;
1071
1072        xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1073        xfs_trans_log_buf(args->trans, bp, 0, args->geo->blksize - 1);
1074
1075        *bpp = bp;
1076        return 0;
1077}
1078
1079/*
1080 * Split the leaf node, rebalance, then add the new entry.
1081 */
1082int
1083xfs_attr3_leaf_split(
1084        struct xfs_da_state     *state,
1085        struct xfs_da_state_blk *oldblk,
1086        struct xfs_da_state_blk *newblk)
1087{
1088        xfs_dablk_t blkno;
1089        int error;
1090
1091        trace_xfs_attr_leaf_split(state->args);
1092
1093        /*
1094         * Allocate space for a new leaf node.
1095         */
1096        ASSERT(oldblk->magic == XFS_ATTR_LEAF_MAGIC);
1097        error = xfs_da_grow_inode(state->args, &blkno);
1098        if (error)
1099                return error;
1100        error = xfs_attr3_leaf_create(state->args, blkno, &newblk->bp);
1101        if (error)
1102                return error;
1103        newblk->blkno = blkno;
1104        newblk->magic = XFS_ATTR_LEAF_MAGIC;
1105
1106        /*
1107         * Rebalance the entries across the two leaves.
1108         * NOTE: rebalance() currently depends on the 2nd block being empty.
1109         */
1110        xfs_attr3_leaf_rebalance(state, oldblk, newblk);
1111        error = xfs_da3_blk_link(state, oldblk, newblk);
1112        if (error)
1113                return error;
1114
1115        /*
1116         * Save info on "old" attribute for "atomic rename" ops, leaf_add()
1117         * modifies the index/blkno/rmtblk/rmtblkcnt fields to show the
1118         * "new" attrs info.  Will need the "old" info to remove it later.
1119         *
1120         * Insert the "new" entry in the correct block.
1121         */
1122        if (state->inleaf) {
1123                trace_xfs_attr_leaf_add_old(state->args);
1124                error = xfs_attr3_leaf_add(oldblk->bp, state->args);
1125        } else {
1126                trace_xfs_attr_leaf_add_new(state->args);
1127                error = xfs_attr3_leaf_add(newblk->bp, state->args);
1128        }
1129
1130        /*
1131         * Update last hashval in each block since we added the name.
1132         */
1133        oldblk->hashval = xfs_attr_leaf_lasthash(oldblk->bp, NULL);
1134        newblk->hashval = xfs_attr_leaf_lasthash(newblk->bp, NULL);
1135        return error;
1136}
1137
1138/*
1139 * Add a name to the leaf attribute list structure.
1140 */
1141int
1142xfs_attr3_leaf_add(
1143        struct xfs_buf          *bp,
1144        struct xfs_da_args      *args)
1145{
1146        struct xfs_attr_leafblock *leaf;
1147        struct xfs_attr3_icleaf_hdr ichdr;
1148        int                     tablesize;
1149        int                     entsize;
1150        int                     sum;
1151        int                     tmp;
1152        int                     i;
1153
1154        trace_xfs_attr_leaf_add(args);
1155
1156        leaf = bp->b_addr;
1157        xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1158        ASSERT(args->index >= 0 && args->index <= ichdr.count);
1159        entsize = xfs_attr_leaf_newentsize(args, NULL);
1160
1161        /*
1162         * Search through freemap for first-fit on new name length.
1163         * (may need to figure in size of entry struct too)
1164         */
1165        tablesize = (ichdr.count + 1) * sizeof(xfs_attr_leaf_entry_t)
1166                                        + xfs_attr3_leaf_hdr_size(leaf);
1167        for (sum = 0, i = XFS_ATTR_LEAF_MAPSIZE - 1; i >= 0; i--) {
1168                if (tablesize > ichdr.firstused) {
1169                        sum += ichdr.freemap[i].size;
1170                        continue;
1171                }
1172                if (!ichdr.freemap[i].size)
1173                        continue;       /* no space in this map */
1174                tmp = entsize;
1175                if (ichdr.freemap[i].base < ichdr.firstused)
1176                        tmp += sizeof(xfs_attr_leaf_entry_t);
1177                if (ichdr.freemap[i].size >= tmp) {
1178                        tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, i);
1179                        goto out_log_hdr;
1180                }
1181                sum += ichdr.freemap[i].size;
1182        }
1183
1184        /*
1185         * If there are no holes in the address space of the block,
1186         * and we don't have enough freespace, then compaction will do us
1187         * no good and we should just give up.
1188         */
1189        if (!ichdr.holes && sum < entsize)
1190                return -ENOSPC;
1191
1192        /*
1193         * Compact the entries to coalesce free space.
1194         * This may change the hdr->count via dropping INCOMPLETE entries.
1195         */
1196        xfs_attr3_leaf_compact(args, &ichdr, bp);
1197
1198        /*
1199         * After compaction, the block is guaranteed to have only one
1200         * free region, in freemap[0].  If it is not big enough, give up.
1201         */
1202        if (ichdr.freemap[0].size < (entsize + sizeof(xfs_attr_leaf_entry_t))) {
1203                tmp = -ENOSPC;
1204                goto out_log_hdr;
1205        }
1206
1207        tmp = xfs_attr3_leaf_add_work(bp, &ichdr, args, 0);
1208
1209out_log_hdr:
1210        xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
1211        xfs_trans_log_buf(args->trans, bp,
1212                XFS_DA_LOGRANGE(leaf, &leaf->hdr,
1213                                xfs_attr3_leaf_hdr_size(leaf)));
1214        return tmp;
1215}
1216
1217/*
1218 * Add a name to a leaf attribute list structure.
1219 */
1220STATIC int
1221xfs_attr3_leaf_add_work(
1222        struct xfs_buf          *bp,
1223        struct xfs_attr3_icleaf_hdr *ichdr,
1224        struct xfs_da_args      *args,
1225        int                     mapindex)
1226{
1227        struct xfs_attr_leafblock *leaf;
1228        struct xfs_attr_leaf_entry *entry;
1229        struct xfs_attr_leaf_name_local *name_loc;
1230        struct xfs_attr_leaf_name_remote *name_rmt;
1231        struct xfs_mount        *mp;
1232        int                     tmp;
1233        int                     i;
1234
1235        trace_xfs_attr_leaf_add_work(args);
1236
1237        leaf = bp->b_addr;
1238        ASSERT(mapindex >= 0 && mapindex < XFS_ATTR_LEAF_MAPSIZE);
1239        ASSERT(args->index >= 0 && args->index <= ichdr->count);
1240
1241        /*
1242         * Force open some space in the entry array and fill it in.
1243         */
1244        entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1245        if (args->index < ichdr->count) {
1246                tmp  = ichdr->count - args->index;
1247                tmp *= sizeof(xfs_attr_leaf_entry_t);
1248                memmove(entry + 1, entry, tmp);
1249                xfs_trans_log_buf(args->trans, bp,
1250                    XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(*entry)));
1251        }
1252        ichdr->count++;
1253
1254        /*
1255         * Allocate space for the new string (at the end of the run).
1256         */
1257        mp = args->trans->t_mountp;
1258        ASSERT(ichdr->freemap[mapindex].base < args->geo->blksize);
1259        ASSERT((ichdr->freemap[mapindex].base & 0x3) == 0);
1260        ASSERT(ichdr->freemap[mapindex].size >=
1261                xfs_attr_leaf_newentsize(args, NULL));
1262        ASSERT(ichdr->freemap[mapindex].size < args->geo->blksize);
1263        ASSERT((ichdr->freemap[mapindex].size & 0x3) == 0);
1264
1265        ichdr->freemap[mapindex].size -= xfs_attr_leaf_newentsize(args, &tmp);
1266
1267        entry->nameidx = cpu_to_be16(ichdr->freemap[mapindex].base +
1268                                     ichdr->freemap[mapindex].size);
1269        entry->hashval = cpu_to_be32(args->hashval);
1270        entry->flags = tmp ? XFS_ATTR_LOCAL : 0;
1271        entry->flags |= XFS_ATTR_NSP_ARGS_TO_ONDISK(args->flags);
1272        if (args->op_flags & XFS_DA_OP_RENAME) {
1273                entry->flags |= XFS_ATTR_INCOMPLETE;
1274                if ((args->blkno2 == args->blkno) &&
1275                    (args->index2 <= args->index)) {
1276                        args->index2++;
1277                }
1278        }
1279        xfs_trans_log_buf(args->trans, bp,
1280                          XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
1281        ASSERT((args->index == 0) ||
1282               (be32_to_cpu(entry->hashval) >= be32_to_cpu((entry-1)->hashval)));
1283        ASSERT((args->index == ichdr->count - 1) ||
1284               (be32_to_cpu(entry->hashval) <= be32_to_cpu((entry+1)->hashval)));
1285
1286        /*
1287         * For "remote" attribute values, simply note that we need to
1288         * allocate space for the "remote" value.  We can't actually
1289         * allocate the extents in this transaction, and we can't decide
1290         * which blocks they should be as we might allocate more blocks
1291         * as part of this transaction (a split operation for example).
1292         */
1293        if (entry->flags & XFS_ATTR_LOCAL) {
1294                name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
1295                name_loc->namelen = args->namelen;
1296                name_loc->valuelen = cpu_to_be16(args->valuelen);
1297                memcpy((char *)name_loc->nameval, args->name, args->namelen);
1298                memcpy((char *)&name_loc->nameval[args->namelen], args->value,
1299                                   be16_to_cpu(name_loc->valuelen));
1300        } else {
1301                name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
1302                name_rmt->namelen = args->namelen;
1303                memcpy((char *)name_rmt->name, args->name, args->namelen);
1304                entry->flags |= XFS_ATTR_INCOMPLETE;
1305                /* just in case */
1306                name_rmt->valuelen = 0;
1307                name_rmt->valueblk = 0;
1308                args->rmtblkno = 1;
1309                args->rmtblkcnt = xfs_attr3_rmt_blocks(mp, args->valuelen);
1310                args->rmtvaluelen = args->valuelen;
1311        }
1312        xfs_trans_log_buf(args->trans, bp,
1313             XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1314                                   xfs_attr_leaf_entsize(leaf, args->index)));
1315
1316        /*
1317         * Update the control info for this leaf node
1318         */
1319        if (be16_to_cpu(entry->nameidx) < ichdr->firstused)
1320                ichdr->firstused = be16_to_cpu(entry->nameidx);
1321
1322        ASSERT(ichdr->firstused >= ichdr->count * sizeof(xfs_attr_leaf_entry_t)
1323                                        + xfs_attr3_leaf_hdr_size(leaf));
1324        tmp = (ichdr->count - 1) * sizeof(xfs_attr_leaf_entry_t)
1325                                        + xfs_attr3_leaf_hdr_size(leaf);
1326
1327        for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1328                if (ichdr->freemap[i].base == tmp) {
1329                        ichdr->freemap[i].base += sizeof(xfs_attr_leaf_entry_t);
1330                        ichdr->freemap[i].size -= sizeof(xfs_attr_leaf_entry_t);
1331                }
1332        }
1333        ichdr->usedbytes += xfs_attr_leaf_entsize(leaf, args->index);
1334        return 0;
1335}
1336
1337/*
1338 * Garbage collect a leaf attribute list block by copying it to a new buffer.
1339 */
1340STATIC void
1341xfs_attr3_leaf_compact(
1342        struct xfs_da_args      *args,
1343        struct xfs_attr3_icleaf_hdr *ichdr_dst,
1344        struct xfs_buf          *bp)
1345{
1346        struct xfs_attr_leafblock *leaf_src;
1347        struct xfs_attr_leafblock *leaf_dst;
1348        struct xfs_attr3_icleaf_hdr ichdr_src;
1349        struct xfs_trans        *trans = args->trans;
1350        char                    *tmpbuffer;
1351
1352        trace_xfs_attr_leaf_compact(args);
1353
1354        tmpbuffer = kmem_alloc(args->geo->blksize, KM_SLEEP);
1355        memcpy(tmpbuffer, bp->b_addr, args->geo->blksize);
1356        memset(bp->b_addr, 0, args->geo->blksize);
1357        leaf_src = (xfs_attr_leafblock_t *)tmpbuffer;
1358        leaf_dst = bp->b_addr;
1359
1360        /*
1361         * Copy the on-disk header back into the destination buffer to ensure
1362         * all the information in the header that is not part of the incore
1363         * header structure is preserved.
1364         */
1365        memcpy(bp->b_addr, tmpbuffer, xfs_attr3_leaf_hdr_size(leaf_src));
1366
1367        /* Initialise the incore headers */
1368        ichdr_src = *ichdr_dst; /* struct copy */
1369        ichdr_dst->firstused = args->geo->blksize;
1370        ichdr_dst->usedbytes = 0;
1371        ichdr_dst->count = 0;
1372        ichdr_dst->holes = 0;
1373        ichdr_dst->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_src);
1374        ichdr_dst->freemap[0].size = ichdr_dst->firstused -
1375                                                ichdr_dst->freemap[0].base;
1376
1377        /* write the header back to initialise the underlying buffer */
1378        xfs_attr3_leaf_hdr_to_disk(args->geo, leaf_dst, ichdr_dst);
1379
1380        /*
1381         * Copy all entry's in the same (sorted) order,
1382         * but allocate name/value pairs packed and in sequence.
1383         */
1384        xfs_attr3_leaf_moveents(args, leaf_src, &ichdr_src, 0,
1385                                leaf_dst, ichdr_dst, 0, ichdr_src.count);
1386        /*
1387         * this logs the entire buffer, but the caller must write the header
1388         * back to the buffer when it is finished modifying it.
1389         */
1390        xfs_trans_log_buf(trans, bp, 0, args->geo->blksize - 1);
1391
1392        kmem_free(tmpbuffer);
1393}
1394
1395/*
1396 * Compare two leaf blocks "order".
1397 * Return 0 unless leaf2 should go before leaf1.
1398 */
1399static int
1400xfs_attr3_leaf_order(
1401        struct xfs_buf  *leaf1_bp,
1402        struct xfs_attr3_icleaf_hdr *leaf1hdr,
1403        struct xfs_buf  *leaf2_bp,
1404        struct xfs_attr3_icleaf_hdr *leaf2hdr)
1405{
1406        struct xfs_attr_leaf_entry *entries1;
1407        struct xfs_attr_leaf_entry *entries2;
1408
1409        entries1 = xfs_attr3_leaf_entryp(leaf1_bp->b_addr);
1410        entries2 = xfs_attr3_leaf_entryp(leaf2_bp->b_addr);
1411        if (leaf1hdr->count > 0 && leaf2hdr->count > 0 &&
1412            ((be32_to_cpu(entries2[0].hashval) <
1413              be32_to_cpu(entries1[0].hashval)) ||
1414             (be32_to_cpu(entries2[leaf2hdr->count - 1].hashval) <
1415              be32_to_cpu(entries1[leaf1hdr->count - 1].hashval)))) {
1416                return 1;
1417        }
1418        return 0;
1419}
1420
1421int
1422xfs_attr_leaf_order(
1423        struct xfs_buf  *leaf1_bp,
1424        struct xfs_buf  *leaf2_bp)
1425{
1426        struct xfs_attr3_icleaf_hdr ichdr1;
1427        struct xfs_attr3_icleaf_hdr ichdr2;
1428        struct xfs_mount *mp = leaf1_bp->b_target->bt_mount;
1429
1430        xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr1, leaf1_bp->b_addr);
1431        xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr2, leaf2_bp->b_addr);
1432        return xfs_attr3_leaf_order(leaf1_bp, &ichdr1, leaf2_bp, &ichdr2);
1433}
1434
1435/*
1436 * Redistribute the attribute list entries between two leaf nodes,
1437 * taking into account the size of the new entry.
1438 *
1439 * NOTE: if new block is empty, then it will get the upper half of the
1440 * old block.  At present, all (one) callers pass in an empty second block.
1441 *
1442 * This code adjusts the args->index/blkno and args->index2/blkno2 fields
1443 * to match what it is doing in splitting the attribute leaf block.  Those
1444 * values are used in "atomic rename" operations on attributes.  Note that
1445 * the "new" and "old" values can end up in different blocks.
1446 */
1447STATIC void
1448xfs_attr3_leaf_rebalance(
1449        struct xfs_da_state     *state,
1450        struct xfs_da_state_blk *blk1,
1451        struct xfs_da_state_blk *blk2)
1452{
1453        struct xfs_da_args      *args;
1454        struct xfs_attr_leafblock *leaf1;
1455        struct xfs_attr_leafblock *leaf2;
1456        struct xfs_attr3_icleaf_hdr ichdr1;
1457        struct xfs_attr3_icleaf_hdr ichdr2;
1458        struct xfs_attr_leaf_entry *entries1;
1459        struct xfs_attr_leaf_entry *entries2;
1460        int                     count;
1461        int                     totallen;
1462        int                     max;
1463        int                     space;
1464        int                     swap;
1465
1466        /*
1467         * Set up environment.
1468         */
1469        ASSERT(blk1->magic == XFS_ATTR_LEAF_MAGIC);
1470        ASSERT(blk2->magic == XFS_ATTR_LEAF_MAGIC);
1471        leaf1 = blk1->bp->b_addr;
1472        leaf2 = blk2->bp->b_addr;
1473        xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr1, leaf1);
1474        xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, leaf2);
1475        ASSERT(ichdr2.count == 0);
1476        args = state->args;
1477
1478        trace_xfs_attr_leaf_rebalance(args);
1479
1480        /*
1481         * Check ordering of blocks, reverse if it makes things simpler.
1482         *
1483         * NOTE: Given that all (current) callers pass in an empty
1484         * second block, this code should never set "swap".
1485         */
1486        swap = 0;
1487        if (xfs_attr3_leaf_order(blk1->bp, &ichdr1, blk2->bp, &ichdr2)) {
1488                struct xfs_da_state_blk *tmp_blk;
1489                struct xfs_attr3_icleaf_hdr tmp_ichdr;
1490
1491                tmp_blk = blk1;
1492                blk1 = blk2;
1493                blk2 = tmp_blk;
1494
1495                /* struct copies to swap them rather than reconverting */
1496                tmp_ichdr = ichdr1;
1497                ichdr1 = ichdr2;
1498                ichdr2 = tmp_ichdr;
1499
1500                leaf1 = blk1->bp->b_addr;
1501                leaf2 = blk2->bp->b_addr;
1502                swap = 1;
1503        }
1504
1505        /*
1506         * Examine entries until we reduce the absolute difference in
1507         * byte usage between the two blocks to a minimum.  Then get
1508         * the direction to copy and the number of elements to move.
1509         *
1510         * "inleaf" is true if the new entry should be inserted into blk1.
1511         * If "swap" is also true, then reverse the sense of "inleaf".
1512         */
1513        state->inleaf = xfs_attr3_leaf_figure_balance(state, blk1, &ichdr1,
1514                                                      blk2, &ichdr2,
1515                                                      &count, &totallen);
1516        if (swap)
1517                state->inleaf = !state->inleaf;
1518
1519        /*
1520         * Move any entries required from leaf to leaf:
1521         */
1522        if (count < ichdr1.count) {
1523                /*
1524                 * Figure the total bytes to be added to the destination leaf.
1525                 */
1526                /* number entries being moved */
1527                count = ichdr1.count - count;
1528                space  = ichdr1.usedbytes - totallen;
1529                space += count * sizeof(xfs_attr_leaf_entry_t);
1530
1531                /*
1532                 * leaf2 is the destination, compact it if it looks tight.
1533                 */
1534                max  = ichdr2.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1535                max -= ichdr2.count * sizeof(xfs_attr_leaf_entry_t);
1536                if (space > max)
1537                        xfs_attr3_leaf_compact(args, &ichdr2, blk2->bp);
1538
1539                /*
1540                 * Move high entries from leaf1 to low end of leaf2.
1541                 */
1542                xfs_attr3_leaf_moveents(args, leaf1, &ichdr1,
1543                                ichdr1.count - count, leaf2, &ichdr2, 0, count);
1544
1545        } else if (count > ichdr1.count) {
1546                /*
1547                 * I assert that since all callers pass in an empty
1548                 * second buffer, this code should never execute.
1549                 */
1550                ASSERT(0);
1551
1552                /*
1553                 * Figure the total bytes to be added to the destination leaf.
1554                 */
1555                /* number entries being moved */
1556                count -= ichdr1.count;
1557                space  = totallen - ichdr1.usedbytes;
1558                space += count * sizeof(xfs_attr_leaf_entry_t);
1559
1560                /*
1561                 * leaf1 is the destination, compact it if it looks tight.
1562                 */
1563                max  = ichdr1.firstused - xfs_attr3_leaf_hdr_size(leaf1);
1564                max -= ichdr1.count * sizeof(xfs_attr_leaf_entry_t);
1565                if (space > max)
1566                        xfs_attr3_leaf_compact(args, &ichdr1, blk1->bp);
1567
1568                /*
1569                 * Move low entries from leaf2 to high end of leaf1.
1570                 */
1571                xfs_attr3_leaf_moveents(args, leaf2, &ichdr2, 0, leaf1, &ichdr1,
1572                                        ichdr1.count, count);
1573        }
1574
1575        xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf1, &ichdr1);
1576        xfs_attr3_leaf_hdr_to_disk(state->args->geo, leaf2, &ichdr2);
1577        xfs_trans_log_buf(args->trans, blk1->bp, 0, args->geo->blksize - 1);
1578        xfs_trans_log_buf(args->trans, blk2->bp, 0, args->geo->blksize - 1);
1579
1580        /*
1581         * Copy out last hashval in each block for B-tree code.
1582         */
1583        entries1 = xfs_attr3_leaf_entryp(leaf1);
1584        entries2 = xfs_attr3_leaf_entryp(leaf2);
1585        blk1->hashval = be32_to_cpu(entries1[ichdr1.count - 1].hashval);
1586        blk2->hashval = be32_to_cpu(entries2[ichdr2.count - 1].hashval);
1587
1588        /*
1589         * Adjust the expected index for insertion.
1590         * NOTE: this code depends on the (current) situation that the
1591         * second block was originally empty.
1592         *
1593         * If the insertion point moved to the 2nd block, we must adjust
1594         * the index.  We must also track the entry just following the
1595         * new entry for use in an "atomic rename" operation, that entry
1596         * is always the "old" entry and the "new" entry is what we are
1597         * inserting.  The index/blkno fields refer to the "old" entry,
1598         * while the index2/blkno2 fields refer to the "new" entry.
1599         */
1600        if (blk1->index > ichdr1.count) {
1601                ASSERT(state->inleaf == 0);
1602                blk2->index = blk1->index - ichdr1.count;
1603                args->index = args->index2 = blk2->index;
1604                args->blkno = args->blkno2 = blk2->blkno;
1605        } else if (blk1->index == ichdr1.count) {
1606                if (state->inleaf) {
1607                        args->index = blk1->index;
1608                        args->blkno = blk1->blkno;
1609                        args->index2 = 0;
1610                        args->blkno2 = blk2->blkno;
1611                } else {
1612                        /*
1613                         * On a double leaf split, the original attr location
1614                         * is already stored in blkno2/index2, so don't
1615                         * overwrite it overwise we corrupt the tree.
1616                         */
1617                        blk2->index = blk1->index - ichdr1.count;
1618                        args->index = blk2->index;
1619                        args->blkno = blk2->blkno;
1620                        if (!state->extravalid) {
1621                                /*
1622                                 * set the new attr location to match the old
1623                                 * one and let the higher level split code
1624                                 * decide where in the leaf to place it.
1625                                 */
1626                                args->index2 = blk2->index;
1627                                args->blkno2 = blk2->blkno;
1628                        }
1629                }
1630        } else {
1631                ASSERT(state->inleaf == 1);
1632                args->index = args->index2 = blk1->index;
1633                args->blkno = args->blkno2 = blk1->blkno;
1634        }
1635}
1636
1637/*
1638 * Examine entries until we reduce the absolute difference in
1639 * byte usage between the two blocks to a minimum.
1640 * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1641 * GROT: there will always be enough room in either block for a new entry.
1642 * GROT: Do a double-split for this case?
1643 */
1644STATIC int
1645xfs_attr3_leaf_figure_balance(
1646        struct xfs_da_state             *state,
1647        struct xfs_da_state_blk         *blk1,
1648        struct xfs_attr3_icleaf_hdr     *ichdr1,
1649        struct xfs_da_state_blk         *blk2,
1650        struct xfs_attr3_icleaf_hdr     *ichdr2,
1651        int                             *countarg,
1652        int                             *usedbytesarg)
1653{
1654        struct xfs_attr_leafblock       *leaf1 = blk1->bp->b_addr;
1655        struct xfs_attr_leafblock       *leaf2 = blk2->bp->b_addr;
1656        struct xfs_attr_leaf_entry      *entry;
1657        int                             count;
1658        int                             max;
1659        int                             index;
1660        int                             totallen = 0;
1661        int                             half;
1662        int                             lastdelta;
1663        int                             foundit = 0;
1664        int                             tmp;
1665
1666        /*
1667         * Examine entries until we reduce the absolute difference in
1668         * byte usage between the two blocks to a minimum.
1669         */
1670        max = ichdr1->count + ichdr2->count;
1671        half = (max + 1) * sizeof(*entry);
1672        half += ichdr1->usedbytes + ichdr2->usedbytes +
1673                        xfs_attr_leaf_newentsize(state->args, NULL);
1674        half /= 2;
1675        lastdelta = state->args->geo->blksize;
1676        entry = xfs_attr3_leaf_entryp(leaf1);
1677        for (count = index = 0; count < max; entry++, index++, count++) {
1678
1679#define XFS_ATTR_ABS(A) (((A) < 0) ? -(A) : (A))
1680                /*
1681                 * The new entry is in the first block, account for it.
1682                 */
1683                if (count == blk1->index) {
1684                        tmp = totallen + sizeof(*entry) +
1685                                xfs_attr_leaf_newentsize(state->args, NULL);
1686                        if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1687                                break;
1688                        lastdelta = XFS_ATTR_ABS(half - tmp);
1689                        totallen = tmp;
1690                        foundit = 1;
1691                }
1692
1693                /*
1694                 * Wrap around into the second block if necessary.
1695                 */
1696                if (count == ichdr1->count) {
1697                        leaf1 = leaf2;
1698                        entry = xfs_attr3_leaf_entryp(leaf1);
1699                        index = 0;
1700                }
1701
1702                /*
1703                 * Figure out if next leaf entry would be too much.
1704                 */
1705                tmp = totallen + sizeof(*entry) + xfs_attr_leaf_entsize(leaf1,
1706                                                                        index);
1707                if (XFS_ATTR_ABS(half - tmp) > lastdelta)
1708                        break;
1709                lastdelta = XFS_ATTR_ABS(half - tmp);
1710                totallen = tmp;
1711#undef XFS_ATTR_ABS
1712        }
1713
1714        /*
1715         * Calculate the number of usedbytes that will end up in lower block.
1716         * If new entry not in lower block, fix up the count.
1717         */
1718        totallen -= count * sizeof(*entry);
1719        if (foundit) {
1720                totallen -= sizeof(*entry) +
1721                                xfs_attr_leaf_newentsize(state->args, NULL);
1722        }
1723
1724        *countarg = count;
1725        *usedbytesarg = totallen;
1726        return foundit;
1727}
1728
1729/*========================================================================
1730 * Routines used for shrinking the Btree.
1731 *========================================================================*/
1732
1733/*
1734 * Check a leaf block and its neighbors to see if the block should be
1735 * collapsed into one or the other neighbor.  Always keep the block
1736 * with the smaller block number.
1737 * If the current block is over 50% full, don't try to join it, return 0.
1738 * If the block is empty, fill in the state structure and return 2.
1739 * If it can be collapsed, fill in the state structure and return 1.
1740 * If nothing can be done, return 0.
1741 *
1742 * GROT: allow for INCOMPLETE entries in calculation.
1743 */
1744int
1745xfs_attr3_leaf_toosmall(
1746        struct xfs_da_state     *state,
1747        int                     *action)
1748{
1749        struct xfs_attr_leafblock *leaf;
1750        struct xfs_da_state_blk *blk;
1751        struct xfs_attr3_icleaf_hdr ichdr;
1752        struct xfs_buf          *bp;
1753        xfs_dablk_t             blkno;
1754        int                     bytes;
1755        int                     forward;
1756        int                     error;
1757        int                     retval;
1758        int                     i;
1759
1760        trace_xfs_attr_leaf_toosmall(state->args);
1761
1762        /*
1763         * Check for the degenerate case of the block being over 50% full.
1764         * If so, it's not worth even looking to see if we might be able
1765         * to coalesce with a sibling.
1766         */
1767        blk = &state->path.blk[ state->path.active-1 ];
1768        leaf = blk->bp->b_addr;
1769        xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr, leaf);
1770        bytes = xfs_attr3_leaf_hdr_size(leaf) +
1771                ichdr.count * sizeof(xfs_attr_leaf_entry_t) +
1772                ichdr.usedbytes;
1773        if (bytes > (state->args->geo->blksize >> 1)) {
1774                *action = 0;    /* blk over 50%, don't try to join */
1775                return 0;
1776        }
1777
1778        /*
1779         * Check for the degenerate case of the block being empty.
1780         * If the block is empty, we'll simply delete it, no need to
1781         * coalesce it with a sibling block.  We choose (arbitrarily)
1782         * to merge with the forward block unless it is NULL.
1783         */
1784        if (ichdr.count == 0) {
1785                /*
1786                 * Make altpath point to the block we want to keep and
1787                 * path point to the block we want to drop (this one).
1788                 */
1789                forward = (ichdr.forw != 0);
1790                memcpy(&state->altpath, &state->path, sizeof(state->path));
1791                error = xfs_da3_path_shift(state, &state->altpath, forward,
1792                                                 0, &retval);
1793                if (error)
1794                        return error;
1795                if (retval) {
1796                        *action = 0;
1797                } else {
1798                        *action = 2;
1799                }
1800                return 0;
1801        }
1802
1803        /*
1804         * Examine each sibling block to see if we can coalesce with
1805         * at least 25% free space to spare.  We need to figure out
1806         * whether to merge with the forward or the backward block.
1807         * We prefer coalescing with the lower numbered sibling so as
1808         * to shrink an attribute list over time.
1809         */
1810        /* start with smaller blk num */
1811        forward = ichdr.forw < ichdr.back;
1812        for (i = 0; i < 2; forward = !forward, i++) {
1813                struct xfs_attr3_icleaf_hdr ichdr2;
1814                if (forward)
1815                        blkno = ichdr.forw;
1816                else
1817                        blkno = ichdr.back;
1818                if (blkno == 0)
1819                        continue;
1820                error = xfs_attr3_leaf_read(state->args->trans, state->args->dp,
1821                                        blkno, -1, &bp);
1822                if (error)
1823                        return error;
1824
1825                xfs_attr3_leaf_hdr_from_disk(state->args->geo, &ichdr2, bp->b_addr);
1826
1827                bytes = state->args->geo->blksize -
1828                        (state->args->geo->blksize >> 2) -
1829                        ichdr.usedbytes - ichdr2.usedbytes -
1830                        ((ichdr.count + ichdr2.count) *
1831                                        sizeof(xfs_attr_leaf_entry_t)) -
1832                        xfs_attr3_leaf_hdr_size(leaf);
1833
1834                xfs_trans_brelse(state->args->trans, bp);
1835                if (bytes >= 0)
1836                        break;  /* fits with at least 25% to spare */
1837        }
1838        if (i >= 2) {
1839                *action = 0;
1840                return 0;
1841        }
1842
1843        /*
1844         * Make altpath point to the block we want to keep (the lower
1845         * numbered block) and path point to the block we want to drop.
1846         */
1847        memcpy(&state->altpath, &state->path, sizeof(state->path));
1848        if (blkno < blk->blkno) {
1849                error = xfs_da3_path_shift(state, &state->altpath, forward,
1850                                                 0, &retval);
1851        } else {
1852                error = xfs_da3_path_shift(state, &state->path, forward,
1853                                                 0, &retval);
1854        }
1855        if (error)
1856                return error;
1857        if (retval) {
1858                *action = 0;
1859        } else {
1860                *action = 1;
1861        }
1862        return 0;
1863}
1864
1865/*
1866 * Remove a name from the leaf attribute list structure.
1867 *
1868 * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1869 * If two leaves are 37% full, when combined they will leave 25% free.
1870 */
1871int
1872xfs_attr3_leaf_remove(
1873        struct xfs_buf          *bp,
1874        struct xfs_da_args      *args)
1875{
1876        struct xfs_attr_leafblock *leaf;
1877        struct xfs_attr3_icleaf_hdr ichdr;
1878        struct xfs_attr_leaf_entry *entry;
1879        int                     before;
1880        int                     after;
1881        int                     smallest;
1882        int                     entsize;
1883        int                     tablesize;
1884        int                     tmp;
1885        int                     i;
1886
1887        trace_xfs_attr_leaf_remove(args);
1888
1889        leaf = bp->b_addr;
1890        xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
1891
1892        ASSERT(ichdr.count > 0 && ichdr.count < args->geo->blksize / 8);
1893        ASSERT(args->index >= 0 && args->index < ichdr.count);
1894        ASSERT(ichdr.firstused >= ichdr.count * sizeof(*entry) +
1895                                        xfs_attr3_leaf_hdr_size(leaf));
1896
1897        entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
1898
1899        ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1900        ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1901
1902        /*
1903         * Scan through free region table:
1904         *    check for adjacency of free'd entry with an existing one,
1905         *    find smallest free region in case we need to replace it,
1906         *    adjust any map that borders the entry table,
1907         */
1908        tablesize = ichdr.count * sizeof(xfs_attr_leaf_entry_t)
1909                                        + xfs_attr3_leaf_hdr_size(leaf);
1910        tmp = ichdr.freemap[0].size;
1911        before = after = -1;
1912        smallest = XFS_ATTR_LEAF_MAPSIZE - 1;
1913        entsize = xfs_attr_leaf_entsize(leaf, args->index);
1914        for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
1915                ASSERT(ichdr.freemap[i].base < args->geo->blksize);
1916                ASSERT(ichdr.freemap[i].size < args->geo->blksize);
1917                if (ichdr.freemap[i].base == tablesize) {
1918                        ichdr.freemap[i].base -= sizeof(xfs_attr_leaf_entry_t);
1919                        ichdr.freemap[i].size += sizeof(xfs_attr_leaf_entry_t);
1920                }
1921
1922                if (ichdr.freemap[i].base + ichdr.freemap[i].size ==
1923                                be16_to_cpu(entry->nameidx)) {
1924                        before = i;
1925                } else if (ichdr.freemap[i].base ==
1926                                (be16_to_cpu(entry->nameidx) + entsize)) {
1927                        after = i;
1928                } else if (ichdr.freemap[i].size < tmp) {
1929                        tmp = ichdr.freemap[i].size;
1930                        smallest = i;
1931                }
1932        }
1933
1934        /*
1935         * Coalesce adjacent freemap regions,
1936         * or replace the smallest region.
1937         */
1938        if ((before >= 0) || (after >= 0)) {
1939                if ((before >= 0) && (after >= 0)) {
1940                        ichdr.freemap[before].size += entsize;
1941                        ichdr.freemap[before].size += ichdr.freemap[after].size;
1942                        ichdr.freemap[after].base = 0;
1943                        ichdr.freemap[after].size = 0;
1944                } else if (before >= 0) {
1945                        ichdr.freemap[before].size += entsize;
1946                } else {
1947                        ichdr.freemap[after].base = be16_to_cpu(entry->nameidx);
1948                        ichdr.freemap[after].size += entsize;
1949                }
1950        } else {
1951                /*
1952                 * Replace smallest region (if it is smaller than free'd entry)
1953                 */
1954                if (ichdr.freemap[smallest].size < entsize) {
1955                        ichdr.freemap[smallest].base = be16_to_cpu(entry->nameidx);
1956                        ichdr.freemap[smallest].size = entsize;
1957                }
1958        }
1959
1960        /*
1961         * Did we remove the first entry?
1962         */
1963        if (be16_to_cpu(entry->nameidx) == ichdr.firstused)
1964                smallest = 1;
1965        else
1966                smallest = 0;
1967
1968        /*
1969         * Compress the remaining entries and zero out the removed stuff.
1970         */
1971        memset(xfs_attr3_leaf_name(leaf, args->index), 0, entsize);
1972        ichdr.usedbytes -= entsize;
1973        xfs_trans_log_buf(args->trans, bp,
1974             XFS_DA_LOGRANGE(leaf, xfs_attr3_leaf_name(leaf, args->index),
1975                                   entsize));
1976
1977        tmp = (ichdr.count - args->index) * sizeof(xfs_attr_leaf_entry_t);
1978        memmove(entry, entry + 1, tmp);
1979        ichdr.count--;
1980        xfs_trans_log_buf(args->trans, bp,
1981            XFS_DA_LOGRANGE(leaf, entry, tmp + sizeof(xfs_attr_leaf_entry_t)));
1982
1983        entry = &xfs_attr3_leaf_entryp(leaf)[ichdr.count];
1984        memset(entry, 0, sizeof(xfs_attr_leaf_entry_t));
1985
1986        /*
1987         * If we removed the first entry, re-find the first used byte
1988         * in the name area.  Note that if the entry was the "firstused",
1989         * then we don't have a "hole" in our block resulting from
1990         * removing the name.
1991         */
1992        if (smallest) {
1993                tmp = args->geo->blksize;
1994                entry = xfs_attr3_leaf_entryp(leaf);
1995                for (i = ichdr.count - 1; i >= 0; entry++, i--) {
1996                        ASSERT(be16_to_cpu(entry->nameidx) >= ichdr.firstused);
1997                        ASSERT(be16_to_cpu(entry->nameidx) < args->geo->blksize);
1998
1999                        if (be16_to_cpu(entry->nameidx) < tmp)
2000                                tmp = be16_to_cpu(entry->nameidx);
2001                }
2002                ichdr.firstused = tmp;
2003                ASSERT(ichdr.firstused != 0);
2004        } else {
2005                ichdr.holes = 1;        /* mark as needing compaction */
2006        }
2007        xfs_attr3_leaf_hdr_to_disk(args->geo, leaf, &ichdr);
2008        xfs_trans_log_buf(args->trans, bp,
2009                          XFS_DA_LOGRANGE(leaf, &leaf->hdr,
2010                                          xfs_attr3_leaf_hdr_size(leaf)));
2011
2012        /*
2013         * Check if leaf is less than 50% full, caller may want to
2014         * "join" the leaf with a sibling if so.
2015         */
2016        tmp = ichdr.usedbytes + xfs_attr3_leaf_hdr_size(leaf) +
2017              ichdr.count * sizeof(xfs_attr_leaf_entry_t);
2018
2019        return tmp < args->geo->magicpct; /* leaf is < 37% full */
2020}
2021
2022/*
2023 * Move all the attribute list entries from drop_leaf into save_leaf.
2024 */
2025void
2026xfs_attr3_leaf_unbalance(
2027        struct xfs_da_state     *state,
2028        struct xfs_da_state_blk *drop_blk,
2029        struct xfs_da_state_blk *save_blk)
2030{
2031        struct xfs_attr_leafblock *drop_leaf = drop_blk->bp->b_addr;
2032        struct xfs_attr_leafblock *save_leaf = save_blk->bp->b_addr;
2033        struct xfs_attr3_icleaf_hdr drophdr;
2034        struct xfs_attr3_icleaf_hdr savehdr;
2035        struct xfs_attr_leaf_entry *entry;
2036
2037        trace_xfs_attr_leaf_unbalance(state->args);
2038
2039        drop_leaf = drop_blk->bp->b_addr;
2040        save_leaf = save_blk->bp->b_addr;
2041        xfs_attr3_leaf_hdr_from_disk(state->args->geo, &drophdr, drop_leaf);
2042        xfs_attr3_leaf_hdr_from_disk(state->args->geo, &savehdr, save_leaf);
2043        entry = xfs_attr3_leaf_entryp(drop_leaf);
2044
2045        /*
2046         * Save last hashval from dying block for later Btree fixup.
2047         */
2048        drop_blk->hashval = be32_to_cpu(entry[drophdr.count - 1].hashval);
2049
2050        /*
2051         * Check if we need a temp buffer, or can we do it in place.
2052         * Note that we don't check "leaf" for holes because we will
2053         * always be dropping it, toosmall() decided that for us already.
2054         */
2055        if (savehdr.holes == 0) {
2056                /*
2057                 * dest leaf has no holes, so we add there.  May need
2058                 * to make some room in the entry array.
2059                 */
2060                if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2061                                         drop_blk->bp, &drophdr)) {
2062                        xfs_attr3_leaf_moveents(state->args,
2063                                                drop_leaf, &drophdr, 0,
2064                                                save_leaf, &savehdr, 0,
2065                                                drophdr.count);
2066                } else {
2067                        xfs_attr3_leaf_moveents(state->args,
2068                                                drop_leaf, &drophdr, 0,
2069                                                save_leaf, &savehdr,
2070                                                savehdr.count, drophdr.count);
2071                }
2072        } else {
2073                /*
2074                 * Destination has holes, so we make a temporary copy
2075                 * of the leaf and add them both to that.
2076                 */
2077                struct xfs_attr_leafblock *tmp_leaf;
2078                struct xfs_attr3_icleaf_hdr tmphdr;
2079
2080                tmp_leaf = kmem_zalloc(state->args->geo->blksize, KM_SLEEP);
2081
2082                /*
2083                 * Copy the header into the temp leaf so that all the stuff
2084                 * not in the incore header is present and gets copied back in
2085                 * once we've moved all the entries.
2086                 */
2087                memcpy(tmp_leaf, save_leaf, xfs_attr3_leaf_hdr_size(save_leaf));
2088
2089                memset(&tmphdr, 0, sizeof(tmphdr));
2090                tmphdr.magic = savehdr.magic;
2091                tmphdr.forw = savehdr.forw;
2092                tmphdr.back = savehdr.back;
2093                tmphdr.firstused = state->args->geo->blksize;
2094
2095                /* write the header to the temp buffer to initialise it */
2096                xfs_attr3_leaf_hdr_to_disk(state->args->geo, tmp_leaf, &tmphdr);
2097
2098                if (xfs_attr3_leaf_order(save_blk->bp, &savehdr,
2099                                         drop_blk->bp, &drophdr)) {
2100                        xfs_attr3_leaf_moveents(state->args,
2101                                                drop_leaf, &drophdr, 0,
2102                                                tmp_leaf, &tmphdr, 0,
2103                                                drophdr.count);
2104                        xfs_attr3_leaf_moveents(state->args,
2105                                                save_leaf, &savehdr, 0,
2106                                                tmp_leaf, &tmphdr, tmphdr.count,
2107                                                savehdr.count);
2108                } else {
2109                        xfs_attr3_leaf_moveents(state->args,
2110                                                save_leaf, &savehdr, 0,
2111                                                tmp_leaf, &tmphdr, 0,
2112                                                savehdr.count);
2113                        xfs_attr3_leaf_moveents(state->args,
2114                                                drop_leaf, &drophdr, 0,
2115                                                tmp_leaf, &tmphdr, tmphdr.count,
2116                                                drophdr.count);
2117                }
2118                memcpy(save_leaf, tmp_leaf, state->args->geo->blksize);
2119                savehdr = tmphdr; /* struct copy */
2120                kmem_free(tmp_leaf);
2121        }
2122
2123        xfs_attr3_leaf_hdr_to_disk(state->args->geo, save_leaf, &savehdr);
2124        xfs_trans_log_buf(state->args->trans, save_blk->bp, 0,
2125                                           state->args->geo->blksize - 1);
2126
2127        /*
2128         * Copy out last hashval in each block for B-tree code.
2129         */
2130        entry = xfs_attr3_leaf_entryp(save_leaf);
2131        save_blk->hashval = be32_to_cpu(entry[savehdr.count - 1].hashval);
2132}
2133
2134/*========================================================================
2135 * Routines used for finding things in the Btree.
2136 *========================================================================*/
2137
2138/*
2139 * Look up a name in a leaf attribute list structure.
2140 * This is the internal routine, it uses the caller's buffer.
2141 *
2142 * Note that duplicate keys are allowed, but only check within the
2143 * current leaf node.  The Btree code must check in adjacent leaf nodes.
2144 *
2145 * Return in args->index the index into the entry[] array of either
2146 * the found entry, or where the entry should have been (insert before
2147 * that entry).
2148 *
2149 * Don't change the args->value unless we find the attribute.
2150 */
2151int
2152xfs_attr3_leaf_lookup_int(
2153        struct xfs_buf          *bp,
2154        struct xfs_da_args      *args)
2155{
2156        struct xfs_attr_leafblock *leaf;
2157        struct xfs_attr3_icleaf_hdr ichdr;
2158        struct xfs_attr_leaf_entry *entry;
2159        struct xfs_attr_leaf_entry *entries;
2160        struct xfs_attr_leaf_name_local *name_loc;
2161        struct xfs_attr_leaf_name_remote *name_rmt;
2162        xfs_dahash_t            hashval;
2163        int                     probe;
2164        int                     span;
2165
2166        trace_xfs_attr_leaf_lookup(args);
2167
2168        leaf = bp->b_addr;
2169        xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2170        entries = xfs_attr3_leaf_entryp(leaf);
2171        ASSERT(ichdr.count < args->geo->blksize / 8);
2172
2173        /*
2174         * Binary search.  (note: small blocks will skip this loop)
2175         */
2176        hashval = args->hashval;
2177        probe = span = ichdr.count / 2;
2178        for (entry = &entries[probe]; span > 4; entry = &entries[probe]) {
2179                span /= 2;
2180                if (be32_to_cpu(entry->hashval) < hashval)
2181                        probe += span;
2182                else if (be32_to_cpu(entry->hashval) > hashval)
2183                        probe -= span;
2184                else
2185                        break;
2186        }
2187        ASSERT(probe >= 0 && (!ichdr.count || probe < ichdr.count));
2188        ASSERT(span <= 4 || be32_to_cpu(entry->hashval) == hashval);
2189
2190        /*
2191         * Since we may have duplicate hashval's, find the first matching
2192         * hashval in the leaf.
2193         */
2194        while (probe > 0 && be32_to_cpu(entry->hashval) >= hashval) {
2195                entry--;
2196                probe--;
2197        }
2198        while (probe < ichdr.count &&
2199               be32_to_cpu(entry->hashval) < hashval) {
2200                entry++;
2201                probe++;
2202        }
2203        if (probe == ichdr.count || be32_to_cpu(entry->hashval) != hashval) {
2204                args->index = probe;
2205                return -ENOATTR;
2206        }
2207
2208        /*
2209         * Duplicate keys may be present, so search all of them for a match.
2210         */
2211        for (; probe < ichdr.count && (be32_to_cpu(entry->hashval) == hashval);
2212                        entry++, probe++) {
2213/*
2214 * GROT: Add code to remove incomplete entries.
2215 */
2216                /*
2217                 * If we are looking for INCOMPLETE entries, show only those.
2218                 * If we are looking for complete entries, show only those.
2219                 */
2220                if ((args->flags & XFS_ATTR_INCOMPLETE) !=
2221                    (entry->flags & XFS_ATTR_INCOMPLETE)) {
2222                        continue;
2223                }
2224                if (entry->flags & XFS_ATTR_LOCAL) {
2225                        name_loc = xfs_attr3_leaf_name_local(leaf, probe);
2226                        if (name_loc->namelen != args->namelen)
2227                                continue;
2228                        if (memcmp(args->name, name_loc->nameval,
2229                                                        args->namelen) != 0)
2230                                continue;
2231                        if (!xfs_attr_namesp_match(args->flags, entry->flags))
2232                                continue;
2233                        args->index = probe;
2234                        return -EEXIST;
2235                } else {
2236                        name_rmt = xfs_attr3_leaf_name_remote(leaf, probe);
2237                        if (name_rmt->namelen != args->namelen)
2238                                continue;
2239                        if (memcmp(args->name, name_rmt->name,
2240                                                        args->namelen) != 0)
2241                                continue;
2242                        if (!xfs_attr_namesp_match(args->flags, entry->flags))
2243                                continue;
2244                        args->index = probe;
2245                        args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2246                        args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2247                        args->rmtblkcnt = xfs_attr3_rmt_blocks(
2248                                                        args->dp->i_mount,
2249                                                        args->rmtvaluelen);
2250                        return -EEXIST;
2251                }
2252        }
2253        args->index = probe;
2254        return -ENOATTR;
2255}
2256
2257/*
2258 * Get the value associated with an attribute name from a leaf attribute
2259 * list structure.
2260 */
2261int
2262xfs_attr3_leaf_getvalue(
2263        struct xfs_buf          *bp,
2264        struct xfs_da_args      *args)
2265{
2266        struct xfs_attr_leafblock *leaf;
2267        struct xfs_attr3_icleaf_hdr ichdr;
2268        struct xfs_attr_leaf_entry *entry;
2269        struct xfs_attr_leaf_name_local *name_loc;
2270        struct xfs_attr_leaf_name_remote *name_rmt;
2271        int                     valuelen;
2272
2273        leaf = bp->b_addr;
2274        xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2275        ASSERT(ichdr.count < args->geo->blksize / 8);
2276        ASSERT(args->index < ichdr.count);
2277
2278        entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2279        if (entry->flags & XFS_ATTR_LOCAL) {
2280                name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2281                ASSERT(name_loc->namelen == args->namelen);
2282                ASSERT(memcmp(args->name, name_loc->nameval, args->namelen) == 0);
2283                valuelen = be16_to_cpu(name_loc->valuelen);
2284                if (args->flags & ATTR_KERNOVAL) {
2285                        args->valuelen = valuelen;
2286                        return 0;
2287                }
2288                if (args->valuelen < valuelen) {
2289                        args->valuelen = valuelen;
2290                        return -ERANGE;
2291                }
2292                args->valuelen = valuelen;
2293                memcpy(args->value, &name_loc->nameval[args->namelen], valuelen);
2294        } else {
2295                name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2296                ASSERT(name_rmt->namelen == args->namelen);
2297                ASSERT(memcmp(args->name, name_rmt->name, args->namelen) == 0);
2298                args->rmtvaluelen = be32_to_cpu(name_rmt->valuelen);
2299                args->rmtblkno = be32_to_cpu(name_rmt->valueblk);
2300                args->rmtblkcnt = xfs_attr3_rmt_blocks(args->dp->i_mount,
2301                                                       args->rmtvaluelen);
2302                if (args->flags & ATTR_KERNOVAL) {
2303                        args->valuelen = args->rmtvaluelen;
2304                        return 0;
2305                }
2306                if (args->valuelen < args->rmtvaluelen) {
2307                        args->valuelen = args->rmtvaluelen;
2308                        return -ERANGE;
2309                }
2310                args->valuelen = args->rmtvaluelen;
2311        }
2312        return 0;
2313}
2314
2315/*========================================================================
2316 * Utility routines.
2317 *========================================================================*/
2318
2319/*
2320 * Move the indicated entries from one leaf to another.
2321 * NOTE: this routine modifies both source and destination leaves.
2322 */
2323/*ARGSUSED*/
2324STATIC void
2325xfs_attr3_leaf_moveents(
2326        struct xfs_da_args              *args,
2327        struct xfs_attr_leafblock       *leaf_s,
2328        struct xfs_attr3_icleaf_hdr     *ichdr_s,
2329        int                             start_s,
2330        struct xfs_attr_leafblock       *leaf_d,
2331        struct xfs_attr3_icleaf_hdr     *ichdr_d,
2332        int                             start_d,
2333        int                             count)
2334{
2335        struct xfs_attr_leaf_entry      *entry_s;
2336        struct xfs_attr_leaf_entry      *entry_d;
2337        int                             desti;
2338        int                             tmp;
2339        int                             i;
2340
2341        /*
2342         * Check for nothing to do.
2343         */
2344        if (count == 0)
2345                return;
2346
2347        /*
2348         * Set up environment.
2349         */
2350        ASSERT(ichdr_s->magic == XFS_ATTR_LEAF_MAGIC ||
2351               ichdr_s->magic == XFS_ATTR3_LEAF_MAGIC);
2352        ASSERT(ichdr_s->magic == ichdr_d->magic);
2353        ASSERT(ichdr_s->count > 0 && ichdr_s->count < args->geo->blksize / 8);
2354        ASSERT(ichdr_s->firstused >= (ichdr_s->count * sizeof(*entry_s))
2355                                        + xfs_attr3_leaf_hdr_size(leaf_s));
2356        ASSERT(ichdr_d->count < args->geo->blksize / 8);
2357        ASSERT(ichdr_d->firstused >= (ichdr_d->count * sizeof(*entry_d))
2358                                        + xfs_attr3_leaf_hdr_size(leaf_d));
2359
2360        ASSERT(start_s < ichdr_s->count);
2361        ASSERT(start_d <= ichdr_d->count);
2362        ASSERT(count <= ichdr_s->count);
2363
2364
2365        /*
2366         * Move the entries in the destination leaf up to make a hole?
2367         */
2368        if (start_d < ichdr_d->count) {
2369                tmp  = ichdr_d->count - start_d;
2370                tmp *= sizeof(xfs_attr_leaf_entry_t);
2371                entry_s = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2372                entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d + count];
2373                memmove(entry_d, entry_s, tmp);
2374        }
2375
2376        /*
2377         * Copy all entry's in the same (sorted) order,
2378         * but allocate attribute info packed and in sequence.
2379         */
2380        entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2381        entry_d = &xfs_attr3_leaf_entryp(leaf_d)[start_d];
2382        desti = start_d;
2383        for (i = 0; i < count; entry_s++, entry_d++, desti++, i++) {
2384                ASSERT(be16_to_cpu(entry_s->nameidx) >= ichdr_s->firstused);
2385                tmp = xfs_attr_leaf_entsize(leaf_s, start_s + i);
2386#ifdef GROT
2387                /*
2388                 * Code to drop INCOMPLETE entries.  Difficult to use as we
2389                 * may also need to change the insertion index.  Code turned
2390                 * off for 6.2, should be revisited later.
2391                 */
2392                if (entry_s->flags & XFS_ATTR_INCOMPLETE) { /* skip partials? */
2393                        memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2394                        ichdr_s->usedbytes -= tmp;
2395                        ichdr_s->count -= 1;
2396                        entry_d--;      /* to compensate for ++ in loop hdr */
2397                        desti--;
2398                        if ((start_s + i) < offset)
2399                                result++;       /* insertion index adjustment */
2400                } else {
2401#endif /* GROT */
2402                        ichdr_d->firstused -= tmp;
2403                        /* both on-disk, don't endian flip twice */
2404                        entry_d->hashval = entry_s->hashval;
2405                        entry_d->nameidx = cpu_to_be16(ichdr_d->firstused);
2406                        entry_d->flags = entry_s->flags;
2407                        ASSERT(be16_to_cpu(entry_d->nameidx) + tmp
2408                                                        <= args->geo->blksize);
2409                        memmove(xfs_attr3_leaf_name(leaf_d, desti),
2410                                xfs_attr3_leaf_name(leaf_s, start_s + i), tmp);
2411                        ASSERT(be16_to_cpu(entry_s->nameidx) + tmp
2412                                                        <= args->geo->blksize);
2413                        memset(xfs_attr3_leaf_name(leaf_s, start_s + i), 0, tmp);
2414                        ichdr_s->usedbytes -= tmp;
2415                        ichdr_d->usedbytes += tmp;
2416                        ichdr_s->count -= 1;
2417                        ichdr_d->count += 1;
2418                        tmp = ichdr_d->count * sizeof(xfs_attr_leaf_entry_t)
2419                                        + xfs_attr3_leaf_hdr_size(leaf_d);
2420                        ASSERT(ichdr_d->firstused >= tmp);
2421#ifdef GROT
2422                }
2423#endif /* GROT */
2424        }
2425
2426        /*
2427         * Zero out the entries we just copied.
2428         */
2429        if (start_s == ichdr_s->count) {
2430                tmp = count * sizeof(xfs_attr_leaf_entry_t);
2431                entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2432                ASSERT(((char *)entry_s + tmp) <=
2433                       ((char *)leaf_s + args->geo->blksize));
2434                memset(entry_s, 0, tmp);
2435        } else {
2436                /*
2437                 * Move the remaining entries down to fill the hole,
2438                 * then zero the entries at the top.
2439                 */
2440                tmp  = (ichdr_s->count - count) * sizeof(xfs_attr_leaf_entry_t);
2441                entry_s = &xfs_attr3_leaf_entryp(leaf_s)[start_s + count];
2442                entry_d = &xfs_attr3_leaf_entryp(leaf_s)[start_s];
2443                memmove(entry_d, entry_s, tmp);
2444
2445                tmp = count * sizeof(xfs_attr_leaf_entry_t);
2446                entry_s = &xfs_attr3_leaf_entryp(leaf_s)[ichdr_s->count];
2447                ASSERT(((char *)entry_s + tmp) <=
2448                       ((char *)leaf_s + args->geo->blksize));
2449                memset(entry_s, 0, tmp);
2450        }
2451
2452        /*
2453         * Fill in the freemap information
2454         */
2455        ichdr_d->freemap[0].base = xfs_attr3_leaf_hdr_size(leaf_d);
2456        ichdr_d->freemap[0].base += ichdr_d->count * sizeof(xfs_attr_leaf_entry_t);
2457        ichdr_d->freemap[0].size = ichdr_d->firstused - ichdr_d->freemap[0].base;
2458        ichdr_d->freemap[1].base = 0;
2459        ichdr_d->freemap[2].base = 0;
2460        ichdr_d->freemap[1].size = 0;
2461        ichdr_d->freemap[2].size = 0;
2462        ichdr_s->holes = 1;     /* leaf may not be compact */
2463}
2464
2465/*
2466 * Pick up the last hashvalue from a leaf block.
2467 */
2468xfs_dahash_t
2469xfs_attr_leaf_lasthash(
2470        struct xfs_buf  *bp,
2471        int             *count)
2472{
2473        struct xfs_attr3_icleaf_hdr ichdr;
2474        struct xfs_attr_leaf_entry *entries;
2475        struct xfs_mount *mp = bp->b_target->bt_mount;
2476
2477        xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, bp->b_addr);
2478        entries = xfs_attr3_leaf_entryp(bp->b_addr);
2479        if (count)
2480                *count = ichdr.count;
2481        if (!ichdr.count)
2482                return 0;
2483        return be32_to_cpu(entries[ichdr.count - 1].hashval);
2484}
2485
2486/*
2487 * Calculate the number of bytes used to store the indicated attribute
2488 * (whether local or remote only calculate bytes in this block).
2489 */
2490STATIC int
2491xfs_attr_leaf_entsize(xfs_attr_leafblock_t *leaf, int index)
2492{
2493        struct xfs_attr_leaf_entry *entries;
2494        xfs_attr_leaf_name_local_t *name_loc;
2495        xfs_attr_leaf_name_remote_t *name_rmt;
2496        int size;
2497
2498        entries = xfs_attr3_leaf_entryp(leaf);
2499        if (entries[index].flags & XFS_ATTR_LOCAL) {
2500                name_loc = xfs_attr3_leaf_name_local(leaf, index);
2501                size = xfs_attr_leaf_entsize_local(name_loc->namelen,
2502                                                   be16_to_cpu(name_loc->valuelen));
2503        } else {
2504                name_rmt = xfs_attr3_leaf_name_remote(leaf, index);
2505                size = xfs_attr_leaf_entsize_remote(name_rmt->namelen);
2506        }
2507        return size;
2508}
2509
2510/*
2511 * Calculate the number of bytes that would be required to store the new
2512 * attribute (whether local or remote only calculate bytes in this block).
2513 * This routine decides as a side effect whether the attribute will be
2514 * a "local" or a "remote" attribute.
2515 */
2516int
2517xfs_attr_leaf_newentsize(
2518        struct xfs_da_args      *args,
2519        int                     *local)
2520{
2521        int                     size;
2522
2523        size = xfs_attr_leaf_entsize_local(args->namelen, args->valuelen);
2524        if (size < xfs_attr_leaf_entsize_local_max(args->geo->blksize)) {
2525                if (local)
2526                        *local = 1;
2527                return size;
2528        }
2529        if (local)
2530                *local = 0;
2531        return xfs_attr_leaf_entsize_remote(args->namelen);
2532}
2533
2534
2535/*========================================================================
2536 * Manage the INCOMPLETE flag in a leaf entry
2537 *========================================================================*/
2538
2539/*
2540 * Clear the INCOMPLETE flag on an entry in a leaf block.
2541 */
2542int
2543xfs_attr3_leaf_clearflag(
2544        struct xfs_da_args      *args)
2545{
2546        struct xfs_attr_leafblock *leaf;
2547        struct xfs_attr_leaf_entry *entry;
2548        struct xfs_attr_leaf_name_remote *name_rmt;
2549        struct xfs_buf          *bp;
2550        int                     error;
2551#ifdef DEBUG
2552        struct xfs_attr3_icleaf_hdr ichdr;
2553        xfs_attr_leaf_name_local_t *name_loc;
2554        int namelen;
2555        char *name;
2556#endif /* DEBUG */
2557
2558        trace_xfs_attr_leaf_clearflag(args);
2559        /*
2560         * Set up the operation.
2561         */
2562        error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2563        if (error)
2564                return error;
2565
2566        leaf = bp->b_addr;
2567        entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2568        ASSERT(entry->flags & XFS_ATTR_INCOMPLETE);
2569
2570#ifdef DEBUG
2571        xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2572        ASSERT(args->index < ichdr.count);
2573        ASSERT(args->index >= 0);
2574
2575        if (entry->flags & XFS_ATTR_LOCAL) {
2576                name_loc = xfs_attr3_leaf_name_local(leaf, args->index);
2577                namelen = name_loc->namelen;
2578                name = (char *)name_loc->nameval;
2579        } else {
2580                name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2581                namelen = name_rmt->namelen;
2582                name = (char *)name_rmt->name;
2583        }
2584        ASSERT(be32_to_cpu(entry->hashval) == args->hashval);
2585        ASSERT(namelen == args->namelen);
2586        ASSERT(memcmp(name, args->name, namelen) == 0);
2587#endif /* DEBUG */
2588
2589        entry->flags &= ~XFS_ATTR_INCOMPLETE;
2590        xfs_trans_log_buf(args->trans, bp,
2591                         XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2592
2593        if (args->rmtblkno) {
2594                ASSERT((entry->flags & XFS_ATTR_LOCAL) == 0);
2595                name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2596                name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2597                name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2598                xfs_trans_log_buf(args->trans, bp,
2599                         XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2600        }
2601
2602        /*
2603         * Commit the flag value change and start the next trans in series.
2604         */
2605        return xfs_trans_roll(&args->trans, args->dp);
2606}
2607
2608/*
2609 * Set the INCOMPLETE flag on an entry in a leaf block.
2610 */
2611int
2612xfs_attr3_leaf_setflag(
2613        struct xfs_da_args      *args)
2614{
2615        struct xfs_attr_leafblock *leaf;
2616        struct xfs_attr_leaf_entry *entry;
2617        struct xfs_attr_leaf_name_remote *name_rmt;
2618        struct xfs_buf          *bp;
2619        int error;
2620#ifdef DEBUG
2621        struct xfs_attr3_icleaf_hdr ichdr;
2622#endif
2623
2624        trace_xfs_attr_leaf_setflag(args);
2625
2626        /*
2627         * Set up the operation.
2628         */
2629        error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
2630        if (error)
2631                return error;
2632
2633        leaf = bp->b_addr;
2634#ifdef DEBUG
2635        xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr, leaf);
2636        ASSERT(args->index < ichdr.count);
2637        ASSERT(args->index >= 0);
2638#endif
2639        entry = &xfs_attr3_leaf_entryp(leaf)[args->index];
2640
2641        ASSERT((entry->flags & XFS_ATTR_INCOMPLETE) == 0);
2642        entry->flags |= XFS_ATTR_INCOMPLETE;
2643        xfs_trans_log_buf(args->trans, bp,
2644                        XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
2645        if ((entry->flags & XFS_ATTR_LOCAL) == 0) {
2646                name_rmt = xfs_attr3_leaf_name_remote(leaf, args->index);
2647                name_rmt->valueblk = 0;
2648                name_rmt->valuelen = 0;
2649                xfs_trans_log_buf(args->trans, bp,
2650                         XFS_DA_LOGRANGE(leaf, name_rmt, sizeof(*name_rmt)));
2651        }
2652
2653        /*
2654         * Commit the flag value change and start the next trans in series.
2655         */
2656        return xfs_trans_roll(&args->trans, args->dp);
2657}
2658
2659/*
2660 * In a single transaction, clear the INCOMPLETE flag on the leaf entry
2661 * given by args->blkno/index and set the INCOMPLETE flag on the leaf
2662 * entry given by args->blkno2/index2.
2663 *
2664 * Note that they could be in different blocks, or in the same block.
2665 */
2666int
2667xfs_attr3_leaf_flipflags(
2668        struct xfs_da_args      *args)
2669{
2670        struct xfs_attr_leafblock *leaf1;
2671        struct xfs_attr_leafblock *leaf2;
2672        struct xfs_attr_leaf_entry *entry1;
2673        struct xfs_attr_leaf_entry *entry2;
2674        struct xfs_attr_leaf_name_remote *name_rmt;
2675        struct xfs_buf          *bp1;
2676        struct xfs_buf          *bp2;
2677        int error;
2678#ifdef DEBUG
2679        struct xfs_attr3_icleaf_hdr ichdr1;
2680        struct xfs_attr3_icleaf_hdr ichdr2;
2681        xfs_attr_leaf_name_local_t *name_loc;
2682        int namelen1, namelen2;
2683        char *name1, *name2;
2684#endif /* DEBUG */
2685
2686        trace_xfs_attr_leaf_flipflags(args);
2687
2688        /*
2689         * Read the block containing the "old" attr
2690         */
2691        error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp1);
2692        if (error)
2693                return error;
2694
2695        /*
2696         * Read the block containing the "new" attr, if it is different
2697         */
2698        if (args->blkno2 != args->blkno) {
2699                error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno2,
2700                                           -1, &bp2);
2701                if (error)
2702                        return error;
2703        } else {
2704                bp2 = bp1;
2705        }
2706
2707        leaf1 = bp1->b_addr;
2708        entry1 = &xfs_attr3_leaf_entryp(leaf1)[args->index];
2709
2710        leaf2 = bp2->b_addr;
2711        entry2 = &xfs_attr3_leaf_entryp(leaf2)[args->index2];
2712
2713#ifdef DEBUG
2714        xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr1, leaf1);
2715        ASSERT(args->index < ichdr1.count);
2716        ASSERT(args->index >= 0);
2717
2718        xfs_attr3_leaf_hdr_from_disk(args->geo, &ichdr2, leaf2);
2719        ASSERT(args->index2 < ichdr2.count);
2720        ASSERT(args->index2 >= 0);
2721
2722        if (entry1->flags & XFS_ATTR_LOCAL) {
2723                name_loc = xfs_attr3_leaf_name_local(leaf1, args->index);
2724                namelen1 = name_loc->namelen;
2725                name1 = (char *)name_loc->nameval;
2726        } else {
2727                name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2728                namelen1 = name_rmt->namelen;
2729                name1 = (char *)name_rmt->name;
2730        }
2731        if (entry2->flags & XFS_ATTR_LOCAL) {
2732                name_loc = xfs_attr3_leaf_name_local(leaf2, args->index2);
2733                namelen2 = name_loc->namelen;
2734                name2 = (char *)name_loc->nameval;
2735        } else {
2736                name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2737                namelen2 = name_rmt->namelen;
2738                name2 = (char *)name_rmt->name;
2739        }
2740        ASSERT(be32_to_cpu(entry1->hashval) == be32_to_cpu(entry2->hashval));
2741        ASSERT(namelen1 == namelen2);
2742        ASSERT(memcmp(name1, name2, namelen1) == 0);
2743#endif /* DEBUG */
2744
2745        ASSERT(entry1->flags & XFS_ATTR_INCOMPLETE);
2746        ASSERT((entry2->flags & XFS_ATTR_INCOMPLETE) == 0);
2747
2748        entry1->flags &= ~XFS_ATTR_INCOMPLETE;
2749        xfs_trans_log_buf(args->trans, bp1,
2750                          XFS_DA_LOGRANGE(leaf1, entry1, sizeof(*entry1)));
2751        if (args->rmtblkno) {
2752                ASSERT((entry1->flags & XFS_ATTR_LOCAL) == 0);
2753                name_rmt = xfs_attr3_leaf_name_remote(leaf1, args->index);
2754                name_rmt->valueblk = cpu_to_be32(args->rmtblkno);
2755                name_rmt->valuelen = cpu_to_be32(args->rmtvaluelen);
2756                xfs_trans_log_buf(args->trans, bp1,
2757                         XFS_DA_LOGRANGE(leaf1, name_rmt, sizeof(*name_rmt)));
2758        }
2759
2760        entry2->flags |= XFS_ATTR_INCOMPLETE;
2761        xfs_trans_log_buf(args->trans, bp2,
2762                          XFS_DA_LOGRANGE(leaf2, entry2, sizeof(*entry2)));
2763        if ((entry2->flags & XFS_ATTR_LOCAL) == 0) {
2764                name_rmt = xfs_attr3_leaf_name_remote(leaf2, args->index2);
2765                name_rmt->valueblk = 0;
2766                name_rmt->valuelen = 0;
2767                xfs_trans_log_buf(args->trans, bp2,
2768                         XFS_DA_LOGRANGE(leaf2, name_rmt, sizeof(*name_rmt)));
2769        }
2770
2771        /*
2772         * Commit the flag value change and start the next trans in series.
2773         */
2774        error = xfs_trans_roll(&args->trans, args->dp);
2775
2776        return error;
2777}
2778