git.oblomov.eu Git - linux-2.6/blob - fs/xfs/xfs_dir_leaf.c

   1 /*
   2  * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_log.h"
  22 #include "xfs_inum.h"
  23 #include "xfs_trans.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_dir.h"
  26 #include "xfs_dir2.h"
  27 #include "xfs_dmapi.h"
  28 #include "xfs_mount.h"
  29 #include "xfs_da_btree.h"
  30 #include "xfs_bmap_btree.h"
  31 #include "xfs_alloc_btree.h"
  32 #include "xfs_ialloc_btree.h"
  33 #include "xfs_dir_sf.h"
  34 #include "xfs_dir2_sf.h"
  35 #include "xfs_attr_sf.h"
  36 #include "xfs_dinode.h"
  37 #include "xfs_inode.h"
  38 #include "xfs_inode_item.h"
  39 #include "xfs_alloc.h"
  40 #include "xfs_btree.h"
  41 #include "xfs_bmap.h"
  42 #include "xfs_dir_leaf.h"
  43 #include "xfs_error.h"
  44
  45 /*
  46  * xfs_dir_leaf.c
  47  *
  48  * Routines to implement leaf blocks of directories as Btrees of hashed names.
  49  */
  50
  51 /*========================================================================
  52  * Function prototypes for the kernel.
  53  *========================================================================*/
  54
  55 /*
  56  * Routines used for growing the Btree.
  57  */
  58 STATIC void xfs_dir_leaf_add_work(xfs_dabuf_t *leaf_buffer, xfs_da_args_t *args,
  59                                               int insertion_index,
  60                                               int freemap_index);
  61 STATIC int xfs_dir_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *leaf_buffer,
  62                                             int musthave, int justcheck);
  63 STATIC void xfs_dir_leaf_rebalance(xfs_da_state_t *state,
  64                                                   xfs_da_state_blk_t *blk1,
  65                                                   xfs_da_state_blk_t *blk2);
  66 STATIC int xfs_dir_leaf_figure_balance(xfs_da_state_t *state,
  67                                           xfs_da_state_blk_t *leaf_blk_1,
  68                                           xfs_da_state_blk_t *leaf_blk_2,
  69                                           int *number_entries_in_blk1,
  70                                           int *number_namebytes_in_blk1);
  71
  72 STATIC int xfs_dir_leaf_create(struct xfs_da_args *args,
  73                                 xfs_dablk_t which_block,
  74                                 struct xfs_dabuf **bpp);
  75
  76 /*
  77  * Utility routines.
  78  */
  79 STATIC void xfs_dir_leaf_moveents(xfs_dir_leafblock_t *src_leaf,
  80                                               int src_start,
  81                                               xfs_dir_leafblock_t *dst_leaf,
  82                                               int dst_start, int move_count,
  83                                               xfs_mount_t *mp);
  84
  85
  86 /*========================================================================
  87  * External routines when dirsize < XFS_IFORK_DSIZE(dp).
  88  *========================================================================*/
  89
  90
  91 /*
  92  * Validate a given inode number.
  93  */
  94 int
  95 xfs_dir_ino_validate(xfs_mount_t *mp, xfs_ino_t ino)
  96 {
  97         xfs_agblock_t   agblkno;
  98         xfs_agino_t     agino;
  99         xfs_agnumber_t  agno;
 100         int             ino_ok;
 101         int             ioff;
 102
 103         agno = XFS_INO_TO_AGNO(mp, ino);
 104         agblkno = XFS_INO_TO_AGBNO(mp, ino);
 105         ioff = XFS_INO_TO_OFFSET(mp, ino);
 106         agino = XFS_OFFBNO_TO_AGINO(mp, agblkno, ioff);
 107         ino_ok =
 108                 agno < mp->m_sb.sb_agcount &&
 109                 agblkno < mp->m_sb.sb_agblocks &&
 110                 agblkno != 0 &&
 111                 ioff < (1 << mp->m_sb.sb_inopblog) &&
 112                 XFS_AGINO_TO_INO(mp, agno, agino) == ino;
 113         if (unlikely(XFS_TEST_ERROR(!ino_ok, mp, XFS_ERRTAG_DIR_INO_VALIDATE,
 114                         XFS_RANDOM_DIR_INO_VALIDATE))) {
 115                 xfs_fs_cmn_err(CE_WARN, mp, "Invalid inode number 0x%Lx",
 116                                 (unsigned long long) ino);
 117                 XFS_ERROR_REPORT("xfs_dir_ino_validate", XFS_ERRLEVEL_LOW, mp);
 118                 return XFS_ERROR(EFSCORRUPTED);
 119         }
 120         return 0;
 121 }
 122
 123 /*
 124  * Create the initial contents of a shortform directory.
 125  */
 126 int
 127 xfs_dir_shortform_create(xfs_da_args_t *args, xfs_ino_t parent)
 128 {
 129         xfs_dir_sf_hdr_t *hdr;
 130         xfs_inode_t *dp;
 131
 132         dp = args->dp;
 133         ASSERT(dp != NULL);
 134         ASSERT(dp->i_d.di_size == 0);
 135         if (dp->i_d.di_format == XFS_DINODE_FMT_EXTENTS) {
 136                 dp->i_df.if_flags &= ~XFS_IFEXTENTS;    /* just in case */
 137                 dp->i_d.di_format = XFS_DINODE_FMT_LOCAL;
 138                 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE);
 139                 dp->i_df.if_flags |= XFS_IFINLINE;
 140         }
 141         ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
 142         ASSERT(dp->i_df.if_bytes == 0);
 143         xfs_idata_realloc(dp, sizeof(*hdr), XFS_DATA_FORK);
 144         hdr = (xfs_dir_sf_hdr_t *)dp->i_df.if_u1.if_data;
 145         XFS_DIR_SF_PUT_DIRINO(&parent, &hdr->parent);
 146
 147         hdr->count = 0;
 148         dp->i_d.di_size = sizeof(*hdr);
 149         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_DDATA);
 150         return 0;
 151 }
 152
 153 /*
 154  * Add a name to the shortform directory structure.
 155  * Overflow from the inode has already been checked for.
 156  */
 157 int
 158 xfs_dir_shortform_addname(xfs_da_args_t *args)
 159 {
 160         xfs_dir_shortform_t *sf;
 161         xfs_dir_sf_entry_t *sfe;
 162         int i, offset, size;
 163         xfs_inode_t *dp;
 164
 165         dp = args->dp;
 166         ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
 167         /*
 168          * Catch the case where the conversion from shortform to leaf
 169          * failed part way through.
 170          */
 171         if (dp->i_d.di_size < sizeof(xfs_dir_sf_hdr_t)) {
 172                 ASSERT(XFS_FORCED_SHUTDOWN(dp->i_mount));
 173                 return XFS_ERROR(EIO);
 174         }
 175         ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
 176         ASSERT(dp->i_df.if_u1.if_data != NULL);
 177         sf = (xfs_dir_shortform_t *)dp->i_df.if_u1.if_data;
 178         sfe = &sf->list[0];
 179         for (i = sf->hdr.count-1; i >= 0; i--) {
 180                 if (sfe->namelen == args->namelen &&
 181                     args->name[0] == sfe->name[0] &&
 182                     memcmp(args->name, sfe->name, args->namelen) == 0)
 183                         return XFS_ERROR(EEXIST);
 184                 sfe = XFS_DIR_SF_NEXTENTRY(sfe);
 185         }
 186
 187         offset = (int)((char *)sfe - (char *)sf);
 188         size = XFS_DIR_SF_ENTSIZE_BYNAME(args->namelen);
 189         xfs_idata_realloc(dp, size, XFS_DATA_FORK);
 190         sf = (xfs_dir_shortform_t *)dp->i_df.if_u1.if_data;
 191         sfe = (xfs_dir_sf_entry_t *)((char *)sf + offset);
 192
 193         XFS_DIR_SF_PUT_DIRINO(&args->inumber, &sfe->inumber);
 194         sfe->namelen = args->namelen;
 195         memcpy(sfe->name, args->name, sfe->namelen);
 196         sf->hdr.count++;
 197
 198         dp->i_d.di_size += size;
 199         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_DDATA);
 200
 201         return 0;
 202 }
 203
 204 /*
 205  * Remove a name from the shortform directory structure.
 206  */
 207 int
 208 xfs_dir_shortform_removename(xfs_da_args_t *args)
 209 {
 210         xfs_dir_shortform_t *sf;
 211         xfs_dir_sf_entry_t *sfe;
 212         int base, size = 0, i;
 213         xfs_inode_t *dp;
 214
 215         dp = args->dp;
 216         ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
 217         /*
 218          * Catch the case where the conversion from shortform to leaf
 219          * failed part way through.
 220          */
 221         if (dp->i_d.di_size < sizeof(xfs_dir_sf_hdr_t)) {
 222                 ASSERT(XFS_FORCED_SHUTDOWN(dp->i_mount));
 223                 return XFS_ERROR(EIO);
 224         }
 225         ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
 226         ASSERT(dp->i_df.if_u1.if_data != NULL);
 227         base = sizeof(xfs_dir_sf_hdr_t);
 228         sf = (xfs_dir_shortform_t *)dp->i_df.if_u1.if_data;
 229         sfe = &sf->list[0];
 230         for (i = sf->hdr.count-1; i >= 0; i--) {
 231                 size = XFS_DIR_SF_ENTSIZE_BYENTRY(sfe);
 232                 if (sfe->namelen == args->namelen &&
 233                     sfe->name[0] == args->name[0] &&
 234                     memcmp(sfe->name, args->name, args->namelen) == 0)
 235                         break;
 236                 base += size;
 237                 sfe = XFS_DIR_SF_NEXTENTRY(sfe);
 238         }
 239         if (i < 0) {
 240                 ASSERT(args->oknoent);
 241                 return XFS_ERROR(ENOENT);
 242         }
 243
 244         if ((base + size) != dp->i_d.di_size) {
 245                 memmove(&((char *)sf)[base], &((char *)sf)[base+size],
 246                                               dp->i_d.di_size - (base+size));
 247         }
 248         sf->hdr.count--;
 249
 250         xfs_idata_realloc(dp, -size, XFS_DATA_FORK);
 251         dp->i_d.di_size -= size;
 252         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_DDATA);
 253
 254         return 0;
 255 }
 256
 257 /*
 258  * Look up a name in a shortform directory structure.
 259  */
 260 int
 261 xfs_dir_shortform_lookup(xfs_da_args_t *args)
 262 {
 263         xfs_dir_shortform_t *sf;
 264         xfs_dir_sf_entry_t *sfe;
 265         int i;
 266         xfs_inode_t *dp;
 267
 268         dp = args->dp;
 269         ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
 270         /*
 271          * Catch the case where the conversion from shortform to leaf
 272          * failed part way through.
 273          */
 274         if (dp->i_d.di_size < sizeof(xfs_dir_sf_hdr_t)) {
 275                 ASSERT(XFS_FORCED_SHUTDOWN(dp->i_mount));
 276                 return XFS_ERROR(EIO);
 277         }
 278         ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
 279         ASSERT(dp->i_df.if_u1.if_data != NULL);
 280         sf = (xfs_dir_shortform_t *)dp->i_df.if_u1.if_data;
 281         if (args->namelen == 2 &&
 282             args->name[0] == '.' && args->name[1] == '.') {
 283                 XFS_DIR_SF_GET_DIRINO(&sf->hdr.parent, &args->inumber);
 284                 return(XFS_ERROR(EEXIST));
 285         }
 286         if (args->namelen == 1 && args->name[0] == '.') {
 287                 args->inumber = dp->i_ino;
 288                 return(XFS_ERROR(EEXIST));
 289         }
 290         sfe = &sf->list[0];
 291         for (i = sf->hdr.count-1; i >= 0; i--) {
 292                 if (sfe->namelen == args->namelen &&
 293                     sfe->name[0] == args->name[0] &&
 294                     memcmp(args->name, sfe->name, args->namelen) == 0) {
 295                         XFS_DIR_SF_GET_DIRINO(&sfe->inumber, &args->inumber);
 296                         return(XFS_ERROR(EEXIST));
 297                 }
 298                 sfe = XFS_DIR_SF_NEXTENTRY(sfe);
 299         }
 300         ASSERT(args->oknoent);
 301         return(XFS_ERROR(ENOENT));
 302 }
 303
 304 /*
 305  * Convert from using the shortform to the leaf.
 306  */
 307 int
 308 xfs_dir_shortform_to_leaf(xfs_da_args_t *iargs)
 309 {
 310         xfs_inode_t *dp;
 311         xfs_dir_shortform_t *sf;
 312         xfs_dir_sf_entry_t *sfe;
 313         xfs_da_args_t args;
 314         xfs_ino_t inumber;
 315         char *tmpbuffer;
 316         int retval, i, size;
 317         xfs_dablk_t blkno;
 318         xfs_dabuf_t *bp;
 319
 320         dp = iargs->dp;
 321         /*
 322          * Catch the case where the conversion from shortform to leaf
 323          * failed part way through.
 324          */
 325         if (dp->i_d.di_size < sizeof(xfs_dir_sf_hdr_t)) {
 326                 ASSERT(XFS_FORCED_SHUTDOWN(dp->i_mount));
 327                 return XFS_ERROR(EIO);
 328         }
 329         ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
 330         ASSERT(dp->i_df.if_u1.if_data != NULL);
 331         size = dp->i_df.if_bytes;
 332         tmpbuffer = kmem_alloc(size, KM_SLEEP);
 333         ASSERT(tmpbuffer != NULL);
 334
 335         memcpy(tmpbuffer, dp->i_df.if_u1.if_data, size);
 336
 337         sf = (xfs_dir_shortform_t *)tmpbuffer;
 338         XFS_DIR_SF_GET_DIRINO(&sf->hdr.parent, &inumber);
 339
 340         xfs_idata_realloc(dp, -size, XFS_DATA_FORK);
 341         dp->i_d.di_size = 0;
 342         xfs_trans_log_inode(iargs->trans, dp, XFS_ILOG_CORE);
 343         retval = xfs_da_grow_inode(iargs, &blkno);
 344         if (retval)
 345                 goto out;
 346
 347         ASSERT(blkno == 0);
 348         retval = xfs_dir_leaf_create(iargs, blkno, &bp);
 349         if (retval)
 350                 goto out;
 351         xfs_da_buf_done(bp);
 352
 353         args.name = ".";
 354         args.namelen = 1;
 355         args.hashval = xfs_dir_hash_dot;
 356         args.inumber = dp->i_ino;
 357         args.dp = dp;
 358         args.firstblock = iargs->firstblock;
 359         args.flist = iargs->flist;
 360         args.total = iargs->total;
 361         args.whichfork = XFS_DATA_FORK;
 362         args.trans = iargs->trans;
 363         args.justcheck = 0;
 364         args.addname = args.oknoent = 1;
 365         retval = xfs_dir_leaf_addname(&args);
 366         if (retval)
 367                 goto out;
 368
 369         args.name = "..";
 370         args.namelen = 2;
 371         args.hashval = xfs_dir_hash_dotdot;
 372         args.inumber = inumber;
 373         retval = xfs_dir_leaf_addname(&args);
 374         if (retval)
 375                 goto out;
 376
 377         sfe = &sf->list[0];
 378         for (i = 0; i < sf->hdr.count; i++) {
 379                 args.name = (char *)(sfe->name);
 380                 args.namelen = sfe->namelen;
 381                 args.hashval = xfs_da_hashname((char *)(sfe->name),
 382                                                sfe->namelen);
 383                 XFS_DIR_SF_GET_DIRINO(&sfe->inumber, &args.inumber);
 384                 retval = xfs_dir_leaf_addname(&args);
 385                 if (retval)
 386                         goto out;
 387                 sfe = XFS_DIR_SF_NEXTENTRY(sfe);
 388         }
 389         retval = 0;
 390
 391 out:
 392         kmem_free(tmpbuffer, size);
 393         return retval;
 394 }
 395
 396 STATIC int
 397 xfs_dir_shortform_compare(const void *a, const void *b)
 398 {
 399         xfs_dir_sf_sort_t *sa, *sb;
 400
 401         sa = (xfs_dir_sf_sort_t *)a;
 402         sb = (xfs_dir_sf_sort_t *)b;
 403         if (sa->hash < sb->hash)
 404                 return -1;
 405         else if (sa->hash > sb->hash)
 406                 return 1;
 407         else
 408                 return sa->entno - sb->entno;
 409 }
 410
 411 /*
 412  * Copy out directory entries for getdents(), for shortform directories.
 413  */
 414 /*ARGSUSED*/
 415 int
 416 xfs_dir_shortform_getdents(xfs_inode_t *dp, uio_t *uio, int *eofp,
 417                                        xfs_dirent_t *dbp, xfs_dir_put_t put)
 418 {
 419         xfs_dir_shortform_t *sf;
 420         xfs_dir_sf_entry_t *sfe;
 421         int retval, i, sbsize, nsbuf, lastresid=0, want_entno;
 422         xfs_mount_t *mp;
 423         xfs_dahash_t cookhash, hash;
 424         xfs_dir_put_args_t p;
 425         xfs_dir_sf_sort_t *sbuf, *sbp;
 426
 427         mp = dp->i_mount;
 428         sf = (xfs_dir_shortform_t *)dp->i_df.if_u1.if_data;
 429         cookhash = XFS_DA_COOKIE_HASH(mp, uio->uio_offset);
 430         want_entno = XFS_DA_COOKIE_ENTRY(mp, uio->uio_offset);
 431         nsbuf = sf->hdr.count + 2;
 432         sbsize = (nsbuf + 1) * sizeof(*sbuf);
 433         sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP);
 434
 435         xfs_dir_trace_g_du("sf: start", dp, uio);
 436
 437         /*
 438          * Collect all the entries into the buffer.
 439          * Entry 0 is .
 440          */
 441         sbp->entno = 0;
 442         sbp->seqno = 0;
 443         sbp->hash = xfs_dir_hash_dot;
 444         sbp->ino = dp->i_ino;
 445         sbp->name = ".";
 446         sbp->namelen = 1;
 447         sbp++;
 448
 449         /*
 450          * Entry 1 is ..
 451          */
 452         sbp->entno = 1;
 453         sbp->seqno = 0;
 454         sbp->hash = xfs_dir_hash_dotdot;
 455         sbp->ino = XFS_GET_DIR_INO8(sf->hdr.parent);
 456         sbp->name = "..";
 457         sbp->namelen = 2;
 458         sbp++;
 459
 460         /*
 461          * Scan the directory data for the rest of the entries.
 462          */
 463         for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
 464
 465                 if (unlikely(
 466                     ((char *)sfe < (char *)sf) ||
 467                     ((char *)sfe >= ((char *)sf + dp->i_df.if_bytes)))) {
 468                         xfs_dir_trace_g_du("sf: corrupted", dp, uio);
 469                         XFS_CORRUPTION_ERROR("xfs_dir_shortform_getdents",
 470                                              XFS_ERRLEVEL_LOW, mp, sfe);
 471                         kmem_free(sbuf, sbsize);
 472                         return XFS_ERROR(EFSCORRUPTED);
 473                 }
 474
 475                 sbp->entno = i + 2;
 476                 sbp->seqno = 0;
 477                 sbp->hash = xfs_da_hashname((char *)sfe->name, sfe->namelen);
 478                 sbp->ino = XFS_GET_DIR_INO8(sfe->inumber);
 479                 sbp->name = (char *)sfe->name;
 480                 sbp->namelen = sfe->namelen;
 481                 sfe = XFS_DIR_SF_NEXTENTRY(sfe);
 482                 sbp++;
 483         }
 484
 485         /*
 486          * Sort the entries on hash then entno.
 487          */
 488         xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_dir_shortform_compare);
 489         /*
 490          * Stuff in last entry.
 491          */
 492         sbp->entno = nsbuf;
 493         sbp->hash = XFS_DA_MAXHASH;
 494         sbp->seqno = 0;
 495         /*
 496          * Figure out the sequence numbers in case there's a hash duplicate.
 497          */
 498         for (hash = sbuf->hash, sbp = sbuf + 1;
 499                                 sbp < &sbuf[nsbuf + 1]; sbp++) {
 500                 if (sbp->hash == hash)
 501                         sbp->seqno = sbp[-1].seqno + 1;
 502                 else
 503                         hash = sbp->hash;
 504         }
 505
 506         /*
 507          * Set up put routine.
 508          */
 509         p.dbp = dbp;
 510         p.put = put;
 511         p.uio = uio;
 512
 513         /*
 514          * Find our place.
 515          */
 516         for (sbp = sbuf; sbp < &sbuf[nsbuf + 1]; sbp++) {
 517                 if (sbp->hash > cookhash ||
 518                     (sbp->hash == cookhash && sbp->seqno >= want_entno))
 519                         break;
 520         }
 521
 522         /*
 523          * Did we fail to find anything?  We stop at the last entry,
 524          * the one we put maxhash into.
 525          */
 526         if (sbp == &sbuf[nsbuf]) {
 527                 kmem_free(sbuf, sbsize);
 528                 xfs_dir_trace_g_du("sf: hash beyond end", dp, uio);
 529                 uio->uio_offset = XFS_DA_MAKE_COOKIE(mp, 0, 0, XFS_DA_MAXHASH);
 530                 *eofp = 1;
 531                 return 0;
 532         }
 533
 534         /*
 535          * Loop putting entries into the user buffer.
 536          */
 537         while (sbp < &sbuf[nsbuf]) {
 538                 /*
 539                  * Save the first resid in a run of equal-hashval entries
 540                  * so that we can back them out if they don't all fit.
 541                  */
 542                 if (sbp->seqno == 0 || sbp == sbuf)
 543                         lastresid = uio->uio_resid;
 544                 XFS_PUT_COOKIE(p.cook, mp, 0, sbp[1].seqno, sbp[1].hash);
 545                 p.ino = sbp->ino;
 546 #if XFS_BIG_INUMS
 547                 p.ino += mp->m_inoadd;
 548 #endif
 549                 p.name = sbp->name;
 550                 p.namelen = sbp->namelen;
 551                 retval = p.put(&p);
 552                 if (!p.done) {
 553                         uio->uio_offset =
 554                                 XFS_DA_MAKE_COOKIE(mp, 0, 0, sbp->hash);
 555                         kmem_free(sbuf, sbsize);
 556                         uio->uio_resid = lastresid;
 557                         xfs_dir_trace_g_du("sf: E-O-B", dp, uio);
 558                         return retval;
 559                 }
 560                 sbp++;
 561         }
 562         kmem_free(sbuf, sbsize);
 563         uio->uio_offset = p.cook.o;
 564         *eofp = 1;
 565         xfs_dir_trace_g_du("sf: E-O-F", dp, uio);
 566         return 0;
 567 }
 568
 569 /*
 570  * Look up a name in a shortform directory structure, replace the inode number.
 571  */
 572 int
 573 xfs_dir_shortform_replace(xfs_da_args_t *args)
 574 {
 575         xfs_dir_shortform_t *sf;
 576         xfs_dir_sf_entry_t *sfe;
 577         xfs_inode_t *dp;
 578         int i;
 579
 580         dp = args->dp;
 581         ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
 582         /*
 583          * Catch the case where the conversion from shortform to leaf
 584          * failed part way through.
 585          */
 586         if (dp->i_d.di_size < sizeof(xfs_dir_sf_hdr_t)) {
 587                 ASSERT(XFS_FORCED_SHUTDOWN(dp->i_mount));
 588                 return XFS_ERROR(EIO);
 589         }
 590         ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
 591         ASSERT(dp->i_df.if_u1.if_data != NULL);
 592         sf = (xfs_dir_shortform_t *)dp->i_df.if_u1.if_data;
 593         if (args->namelen == 2 &&
 594             args->name[0] == '.' && args->name[1] == '.') {
 595                 /* XXX - replace assert? */
 596                 XFS_DIR_SF_PUT_DIRINO(&args->inumber, &sf->hdr.parent);
 597                 xfs_trans_log_inode(args->trans, dp, XFS_ILOG_DDATA);
 598                 return 0;
 599         }
 600         ASSERT(args->namelen != 1 || args->name[0] != '.');
 601         sfe = &sf->list[0];
 602         for (i = sf->hdr.count-1; i >= 0; i--) {
 603                 if (sfe->namelen == args->namelen &&
 604                     sfe->name[0] == args->name[0] &&
 605                     memcmp(args->name, sfe->name, args->namelen) == 0) {
 606                         ASSERT(memcmp((char *)&args->inumber,
 607                                 (char *)&sfe->inumber, sizeof(xfs_ino_t)));
 608                         XFS_DIR_SF_PUT_DIRINO(&args->inumber, &sfe->inumber);
 609                         xfs_trans_log_inode(args->trans, dp, XFS_ILOG_DDATA);
 610                         return 0;
 611                 }
 612                 sfe = XFS_DIR_SF_NEXTENTRY(sfe);
 613         }
 614         ASSERT(args->oknoent);
 615         return XFS_ERROR(ENOENT);
 616 }
 617
 618 /*
 619  * Convert a leaf directory to shortform structure
 620  */
 621 int
 622 xfs_dir_leaf_to_shortform(xfs_da_args_t *iargs)
 623 {
 624         xfs_dir_leafblock_t *leaf;
 625         xfs_dir_leaf_hdr_t *hdr;
 626         xfs_dir_leaf_entry_t *entry;
 627         xfs_dir_leaf_name_t *namest;
 628         xfs_da_args_t args;
 629         xfs_inode_t *dp;
 630         xfs_ino_t parent = 0;
 631         char *tmpbuffer;
 632         int retval, i;
 633         xfs_dabuf_t *bp;
 634
 635         dp = iargs->dp;
 636         tmpbuffer = kmem_alloc(XFS_LBSIZE(dp->i_mount), KM_SLEEP);
 637         ASSERT(tmpbuffer != NULL);
 638
 639         retval = xfs_da_read_buf(iargs->trans, iargs->dp, 0, -1, &bp,
 640                                                XFS_DATA_FORK);
 641         if (retval)
 642                 goto out;
 643         ASSERT(bp != NULL);
 644         memcpy(tmpbuffer, bp->data, XFS_LBSIZE(dp->i_mount));
 645         leaf = (xfs_dir_leafblock_t *)tmpbuffer;
 646         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
 647         memset(bp->data, 0, XFS_LBSIZE(dp->i_mount));
 648
 649         /*
 650          * Find and special case the parent inode number
 651          */
 652         hdr = &leaf->hdr;
 653         entry = &leaf->entries[0];
 654         for (i = INT_GET(hdr->count, ARCH_CONVERT)-1; i >= 0; entry++, i--) {
 655                 namest = XFS_DIR_LEAF_NAMESTRUCT(leaf, INT_GET(entry->nameidx, ARCH_CONVERT));
 656                 if ((entry->namelen == 2) &&
 657                     (namest->name[0] == '.') &&
 658                     (namest->name[1] == '.')) {
 659                         XFS_DIR_SF_GET_DIRINO(&namest->inumber, &parent);
 660                         entry->nameidx = 0;
 661                 } else if ((entry->namelen == 1) && (namest->name[0] == '.')) {
 662                         entry->nameidx = 0;
 663                 }
 664         }
 665         retval = xfs_da_shrink_inode(iargs, 0, bp);
 666         if (retval)
 667                 goto out;
 668         retval = xfs_dir_shortform_create(iargs, parent);
 669         if (retval)
 670                 goto out;
 671
 672         /*
 673          * Copy the rest of the filenames
 674          */
 675         entry = &leaf->entries[0];
 676         args.dp = dp;
 677         args.firstblock = iargs->firstblock;
 678         args.flist = iargs->flist;
 679         args.total = iargs->total;
 680         args.whichfork = XFS_DATA_FORK;
 681         args.trans = iargs->trans;
 682         args.justcheck = 0;
 683         args.addname = args.oknoent = 1;
 684         for (i = 0; i < INT_GET(hdr->count, ARCH_CONVERT); entry++, i++) {
 685                 if (!entry->nameidx)
 686                         continue;
 687                 namest = XFS_DIR_LEAF_NAMESTRUCT(leaf, INT_GET(entry->nameidx, ARCH_CONVERT));
 688                 args.name = (char *)(namest->name);
 689                 args.namelen = entry->namelen;
 690                 args.hashval = INT_GET(entry->hashval, ARCH_CONVERT);
 691                 XFS_DIR_SF_GET_DIRINO(&namest->inumber, &args.inumber);
 692                 xfs_dir_shortform_addname(&args);
 693         }
 694
 695 out:
 696         kmem_free(tmpbuffer, XFS_LBSIZE(dp->i_mount));
 697         return retval;
 698 }
 699
 700 /*
 701  * Convert from using a single leaf to a root node and a leaf.
 702  */
 703 int
 704 xfs_dir_leaf_to_node(xfs_da_args_t *args)
 705 {
 706         xfs_dir_leafblock_t *leaf;
 707         xfs_da_intnode_t *node;
 708         xfs_inode_t *dp;
 709         xfs_dabuf_t *bp1, *bp2;
 710         xfs_dablk_t blkno;
 711         int retval;
 712
 713         dp = args->dp;
 714         retval = xfs_da_grow_inode(args, &blkno);
 715         ASSERT(blkno == 1);
 716         if (retval)
 717                 return retval;
 718         retval = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp1,
 719                                               XFS_DATA_FORK);
 720         if (retval)
 721                 return retval;
 722         ASSERT(bp1 != NULL);
 723         retval = xfs_da_get_buf(args->trans, args->dp, 1, -1, &bp2,
 724                                              XFS_DATA_FORK);
 725         if (retval) {
 726                 xfs_da_buf_done(bp1);
 727                 return retval;
 728         }
 729         ASSERT(bp2 != NULL);
 730         memcpy(bp2->data, bp1->data, XFS_LBSIZE(dp->i_mount));
 731         xfs_da_buf_done(bp1);
 732         xfs_da_log_buf(args->trans, bp2, 0, XFS_LBSIZE(dp->i_mount) - 1);
 733
 734         /*
 735          * Set up the new root node.
 736          */
 737         retval = xfs_da_node_create(args, 0, 1, &bp1, XFS_DATA_FORK);
 738         if (retval) {
 739                 xfs_da_buf_done(bp2);
 740                 return retval;
 741         }
 742         node = bp1->data;
 743         leaf = bp2->data;
 744         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
 745         node->btree[0].hashval = cpu_to_be32(
 746                 INT_GET(leaf->entries[
 747                         INT_GET(leaf->hdr.count, ARCH_CONVERT)-1].hashval, ARCH_CONVERT));
 748         xfs_da_buf_done(bp2);
 749         node->btree[0].before = cpu_to_be32(blkno);
 750         node->hdr.count = cpu_to_be16(1);
 751         xfs_da_log_buf(args->trans, bp1,
 752                 XFS_DA_LOGRANGE(node, &node->btree[0], sizeof(node->btree[0])));
 753         xfs_da_buf_done(bp1);
 754
 755         return retval;
 756 }
 757
 758
 759 /*========================================================================
 760  * Routines used for growing the Btree.
 761  *========================================================================*/
 762
 763 /*
 764  * Create the initial contents of a leaf directory
 765  * or a leaf in a node directory.
 766  */
 767 STATIC int
 768 xfs_dir_leaf_create(xfs_da_args_t *args, xfs_dablk_t blkno, xfs_dabuf_t **bpp)
 769 {
 770         xfs_dir_leafblock_t *leaf;
 771         xfs_dir_leaf_hdr_t *hdr;
 772         xfs_inode_t *dp;
 773         xfs_dabuf_t *bp;
 774         int retval;
 775
 776         dp = args->dp;
 777         ASSERT(dp != NULL);
 778         retval = xfs_da_get_buf(args->trans, dp, blkno, -1, &bp, XFS_DATA_FORK);
 779         if (retval)
 780                 return retval;
 781         ASSERT(bp != NULL);
 782         leaf = bp->data;
 783         memset((char *)leaf, 0, XFS_LBSIZE(dp->i_mount));
 784         hdr = &leaf->hdr;
 785         hdr->info.magic = cpu_to_be16(XFS_DIR_LEAF_MAGIC);
 786         INT_SET(hdr->firstused, ARCH_CONVERT, XFS_LBSIZE(dp->i_mount));
 787         if (!hdr->firstused)
 788                 INT_SET(hdr->firstused, ARCH_CONVERT, XFS_LBSIZE(dp->i_mount) - 1);
 789         INT_SET(hdr->freemap[0].base, ARCH_CONVERT, sizeof(xfs_dir_leaf_hdr_t));
 790         INT_SET(hdr->freemap[0].size, ARCH_CONVERT, INT_GET(hdr->firstused, ARCH_CONVERT) - INT_GET(hdr->freemap[0].base, ARCH_CONVERT));
 791
 792         xfs_da_log_buf(args->trans, bp, 0, XFS_LBSIZE(dp->i_mount) - 1);
 793
 794         *bpp = bp;
 795         return 0;
 796 }
 797
 798 /*
 799  * Split the leaf node, rebalance, then add the new entry.
 800  */
 801 int
 802 xfs_dir_leaf_split(xfs_da_state_t *state, xfs_da_state_blk_t *oldblk,
 803                                   xfs_da_state_blk_t *newblk)
 804 {
 805         xfs_dablk_t blkno;
 806         xfs_da_args_t *args;
 807         int error;
 808
 809         /*
 810          * Allocate space for a new leaf node.
 811          */
 812         args = state->args;
 813         ASSERT(args != NULL);
 814         ASSERT(oldblk->magic == XFS_DIR_LEAF_MAGIC);
 815         error = xfs_da_grow_inode(args, &blkno);
 816         if (error)
 817                 return error;
 818         error = xfs_dir_leaf_create(args, blkno, &newblk->bp);
 819         if (error)
 820                 return error;
 821         newblk->blkno = blkno;
 822         newblk->magic = XFS_DIR_LEAF_MAGIC;
 823
 824         /*
 825          * Rebalance the entries across the two leaves.
 826          */
 827         xfs_dir_leaf_rebalance(state, oldblk, newblk);
 828         error = xfs_da_blk_link(state, oldblk, newblk);
 829         if (error)
 830                 return error;
 831
 832         /*
 833          * Insert the new entry in the correct block.
 834          */
 835         if (state->inleaf) {
 836                 error = xfs_dir_leaf_add(oldblk->bp, args, oldblk->index);
 837         } else {
 838                 error = xfs_dir_leaf_add(newblk->bp, args, newblk->index);
 839         }
 840
 841         /*
 842          * Update last hashval in each block since we added the name.
 843          */
 844         oldblk->hashval = xfs_dir_leaf_lasthash(oldblk->bp, NULL);
 845         newblk->hashval = xfs_dir_leaf_lasthash(newblk->bp, NULL);
 846         return error;
 847 }
 848
 849 /*
 850  * Add a name to the leaf directory structure.
 851  *
 852  * Must take into account fragmented leaves and leaves where spacemap has
 853  * lost some freespace information (ie: holes).
 854  */
 855 int
 856 xfs_dir_leaf_add(xfs_dabuf_t *bp, xfs_da_args_t *args, int index)
 857 {
 858         xfs_dir_leafblock_t *leaf;
 859         xfs_dir_leaf_hdr_t *hdr;
 860         xfs_dir_leaf_map_t *map;
 861         int tablesize, entsize, sum, i, tmp, error;
 862
 863         leaf = bp->data;
 864         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
 865         ASSERT((index >= 0) && (index <= INT_GET(leaf->hdr.count, ARCH_CONVERT)));
 866         hdr = &leaf->hdr;
 867         entsize = XFS_DIR_LEAF_ENTSIZE_BYNAME(args->namelen);
 868
 869         /*
 870          * Search through freemap for first-fit on new name length.
 871          * (may need to figure in size of entry struct too)
 872          */
 873         tablesize = (INT_GET(hdr->count, ARCH_CONVERT) + 1) * (uint)sizeof(xfs_dir_leaf_entry_t)
 874                         + (uint)sizeof(xfs_dir_leaf_hdr_t);
 875         map = &hdr->freemap[XFS_DIR_LEAF_MAPSIZE-1];
 876         for (sum = 0, i = XFS_DIR_LEAF_MAPSIZE-1; i >= 0; map--, i--) {
 877                 if (tablesize > INT_GET(hdr->firstused, ARCH_CONVERT)) {
 878                         sum += INT_GET(map->size, ARCH_CONVERT);
 879                         continue;
 880                 }
 881                 if (!map->size)
 882                         continue;       /* no space in this map */
 883                 tmp = entsize;
 884                 if (INT_GET(map->base, ARCH_CONVERT) < INT_GET(hdr->firstused, ARCH_CONVERT))
 885                         tmp += (uint)sizeof(xfs_dir_leaf_entry_t);
 886                 if (INT_GET(map->size, ARCH_CONVERT) >= tmp) {
 887                         if (!args->justcheck)
 888                                 xfs_dir_leaf_add_work(bp, args, index, i);
 889                         return 0;
 890                 }
 891                 sum += INT_GET(map->size, ARCH_CONVERT);
 892         }
 893
 894         /*
 895          * If there are no holes in the address space of the block,
 896          * and we don't have enough freespace, then compaction will do us
 897          * no good and we should just give up.
 898          */
 899         if (!hdr->holes && (sum < entsize))
 900                 return XFS_ERROR(ENOSPC);
 901
 902         /*
 903          * Compact the entries to coalesce free space.
 904          * Pass the justcheck flag so the checking pass can return
 905          * an error, without changing anything, if it won't fit.
 906          */
 907         error = xfs_dir_leaf_compact(args->trans, bp,
 908                         args->total == 0 ?
 909                                 entsize +
 910                                 (uint)sizeof(xfs_dir_leaf_entry_t) : 0,
 911                         args->justcheck);
 912         if (error)
 913                 return error;
 914         /*
 915          * After compaction, the block is guaranteed to have only one
 916          * free region, in freemap[0].  If it is not big enough, give up.
 917          */
 918         if (INT_GET(hdr->freemap[0].size, ARCH_CONVERT) <
 919             (entsize + (uint)sizeof(xfs_dir_leaf_entry_t)))
 920                 return XFS_ERROR(ENOSPC);
 921
 922         if (!args->justcheck)
 923                 xfs_dir_leaf_add_work(bp, args, index, 0);
 924         return 0;
 925 }
 926
 927 /*
 928  * Add a name to a leaf directory structure.
 929  */
 930 STATIC void
 931 xfs_dir_leaf_add_work(xfs_dabuf_t *bp, xfs_da_args_t *args, int index,
 932                       int mapindex)
 933 {
 934         xfs_dir_leafblock_t *leaf;
 935         xfs_dir_leaf_hdr_t *hdr;
 936         xfs_dir_leaf_entry_t *entry;
 937         xfs_dir_leaf_name_t *namest;
 938         xfs_dir_leaf_map_t *map;
 939         /* REFERENCED */
 940         xfs_mount_t *mp;
 941         int tmp, i;
 942
 943         leaf = bp->data;
 944         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
 945         hdr = &leaf->hdr;
 946         ASSERT((mapindex >= 0) && (mapindex < XFS_DIR_LEAF_MAPSIZE));
 947         ASSERT((index >= 0) && (index <= INT_GET(hdr->count, ARCH_CONVERT)));
 948
 949         /*
 950          * Force open some space in the entry array and fill it in.
 951          */
 952         entry = &leaf->entries[index];
 953         if (index < INT_GET(hdr->count, ARCH_CONVERT)) {
 954                 tmp  = INT_GET(hdr->count, ARCH_CONVERT) - index;
 955                 tmp *= (uint)sizeof(xfs_dir_leaf_entry_t);
 956                 memmove(entry + 1, entry, tmp);
 957                 xfs_da_log_buf(args->trans, bp,
 958                     XFS_DA_LOGRANGE(leaf, entry, tmp + (uint)sizeof(*entry)));
 959         }
 960         INT_MOD(hdr->count, ARCH_CONVERT, +1);
 961
 962         /*
 963          * Allocate space for the new string (at the end of the run).
 964          */
 965         map = &hdr->freemap[mapindex];
 966         mp = args->trans->t_mountp;
 967         ASSERT(INT_GET(map->base, ARCH_CONVERT) < XFS_LBSIZE(mp));
 968         ASSERT(INT_GET(map->size, ARCH_CONVERT) >= XFS_DIR_LEAF_ENTSIZE_BYNAME(args->namelen));
 969         ASSERT(INT_GET(map->size, ARCH_CONVERT) < XFS_LBSIZE(mp));
 970         INT_MOD(map->size, ARCH_CONVERT, -(XFS_DIR_LEAF_ENTSIZE_BYNAME(args->namelen)));
 971         INT_SET(entry->nameidx, ARCH_CONVERT, INT_GET(map->base, ARCH_CONVERT) + INT_GET(map->size, ARCH_CONVERT));
 972         INT_SET(entry->hashval, ARCH_CONVERT, args->hashval);
 973         entry->namelen = args->namelen;
 974         xfs_da_log_buf(args->trans, bp,
 975             XFS_DA_LOGRANGE(leaf, entry, sizeof(*entry)));
 976
 977         /*
 978          * Copy the string and inode number into the new space.
 979          */
 980         namest = XFS_DIR_LEAF_NAMESTRUCT(leaf, INT_GET(entry->nameidx, ARCH_CONVERT));
 981         XFS_DIR_SF_PUT_DIRINO(&args->inumber, &namest->inumber);
 982         memcpy(namest->name, args->name, args->namelen);
 983         xfs_da_log_buf(args->trans, bp,
 984             XFS_DA_LOGRANGE(leaf, namest, XFS_DIR_LEAF_ENTSIZE_BYENTRY(entry)));
 985
 986         /*
 987          * Update the control info for this leaf node
 988          */
 989         if (INT_GET(entry->nameidx, ARCH_CONVERT) < INT_GET(hdr->firstused, ARCH_CONVERT))
 990                 INT_COPY(hdr->firstused, entry->nameidx, ARCH_CONVERT);
 991         ASSERT(INT_GET(hdr->firstused, ARCH_CONVERT) >= ((INT_GET(hdr->count, ARCH_CONVERT)*sizeof(*entry))+sizeof(*hdr)));
 992         tmp = (INT_GET(hdr->count, ARCH_CONVERT)-1) * (uint)sizeof(xfs_dir_leaf_entry_t)
 993                         + (uint)sizeof(xfs_dir_leaf_hdr_t);
 994         map = &hdr->freemap[0];
 995         for (i = 0; i < XFS_DIR_LEAF_MAPSIZE; map++, i++) {
 996                 if (INT_GET(map->base, ARCH_CONVERT) == tmp) {
 997                         INT_MOD(map->base, ARCH_CONVERT, (uint)sizeof(xfs_dir_leaf_entry_t));
 998                         INT_MOD(map->size, ARCH_CONVERT, -((uint)sizeof(xfs_dir_leaf_entry_t)));
 999                 }
1000         }
1001         INT_MOD(hdr->namebytes, ARCH_CONVERT, args->namelen);
1002         xfs_da_log_buf(args->trans, bp,
1003                 XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1004 }
1005
1006 /*
1007  * Garbage collect a leaf directory block by copying it to a new buffer.
1008  */
1009 STATIC int
1010 xfs_dir_leaf_compact(xfs_trans_t *trans, xfs_dabuf_t *bp, int musthave,
1011                      int justcheck)
1012 {
1013         xfs_dir_leafblock_t *leaf_s, *leaf_d;
1014         xfs_dir_leaf_hdr_t *hdr_s, *hdr_d;
1015         xfs_mount_t *mp;
1016         char *tmpbuffer;
1017         char *tmpbuffer2=NULL;
1018         int rval;
1019         int lbsize;
1020
1021         mp = trans->t_mountp;
1022         lbsize = XFS_LBSIZE(mp);
1023         tmpbuffer = kmem_alloc(lbsize, KM_SLEEP);
1024         ASSERT(tmpbuffer != NULL);
1025         memcpy(tmpbuffer, bp->data, lbsize);
1026
1027         /*
1028          * Make a second copy in case xfs_dir_leaf_moveents()
1029          * below destroys the original.
1030          */
1031         if (musthave || justcheck) {
1032                 tmpbuffer2 = kmem_alloc(lbsize, KM_SLEEP);
1033                 memcpy(tmpbuffer2, bp->data, lbsize);
1034         }
1035         memset(bp->data, 0, lbsize);
1036
1037         /*
1038          * Copy basic information
1039          */
1040         leaf_s = (xfs_dir_leafblock_t *)tmpbuffer;
1041         leaf_d = bp->data;
1042         hdr_s = &leaf_s->hdr;
1043         hdr_d = &leaf_d->hdr;
1044         hdr_d->info = hdr_s->info;      /* struct copy */
1045         INT_SET(hdr_d->firstused, ARCH_CONVERT, lbsize);
1046         if (!hdr_d->firstused)
1047                 INT_SET(hdr_d->firstused, ARCH_CONVERT, lbsize - 1);
1048         hdr_d->namebytes = 0;
1049         hdr_d->count = 0;
1050         hdr_d->holes = 0;
1051         INT_SET(hdr_d->freemap[0].base, ARCH_CONVERT, sizeof(xfs_dir_leaf_hdr_t));
1052         INT_SET(hdr_d->freemap[0].size, ARCH_CONVERT, INT_GET(hdr_d->firstused, ARCH_CONVERT) - INT_GET(hdr_d->freemap[0].base, ARCH_CONVERT));
1053
1054         /*
1055          * Copy all entry's in the same (sorted) order,
1056          * but allocate filenames packed and in sequence.
1057          * This changes the source (leaf_s) as well.
1058          */
1059         xfs_dir_leaf_moveents(leaf_s, 0, leaf_d, 0, (int)INT_GET(hdr_s->count, ARCH_CONVERT), mp);
1060
1061         if (musthave && INT_GET(hdr_d->freemap[0].size, ARCH_CONVERT) < musthave)
1062                 rval = XFS_ERROR(ENOSPC);
1063         else
1064                 rval = 0;
1065
1066         if (justcheck || rval == ENOSPC) {
1067                 ASSERT(tmpbuffer2);
1068                 memcpy(bp->data, tmpbuffer2, lbsize);
1069         } else {
1070                 xfs_da_log_buf(trans, bp, 0, lbsize - 1);
1071         }
1072
1073         kmem_free(tmpbuffer, lbsize);
1074         if (musthave || justcheck)
1075                 kmem_free(tmpbuffer2, lbsize);
1076         return rval;
1077 }
1078
1079 /*
1080  * Redistribute the directory entries between two leaf nodes,
1081  * taking into account the size of the new entry.
1082  *
1083  * NOTE: if new block is empty, then it will get the upper half of old block.
1084  */
1085 STATIC void
1086 xfs_dir_leaf_rebalance(xfs_da_state_t *state, xfs_da_state_blk_t *blk1,
1087                                       xfs_da_state_blk_t *blk2)
1088 {
1089         xfs_da_state_blk_t *tmp_blk;
1090         xfs_dir_leafblock_t *leaf1, *leaf2;
1091         xfs_dir_leaf_hdr_t *hdr1, *hdr2;
1092         int count, totallen, max, space, swap;
1093
1094         /*
1095          * Set up environment.
1096          */
1097         ASSERT(blk1->magic == XFS_DIR_LEAF_MAGIC);
1098         ASSERT(blk2->magic == XFS_DIR_LEAF_MAGIC);
1099         leaf1 = blk1->bp->data;
1100         leaf2 = blk2->bp->data;
1101         ASSERT(be16_to_cpu(leaf1->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
1102         ASSERT(be16_to_cpu(leaf2->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
1103
1104         /*
1105          * Check ordering of blocks, reverse if it makes things simpler.
1106          */
1107         swap = 0;
1108         if (xfs_dir_leaf_order(blk1->bp, blk2->bp)) {
1109                 tmp_blk = blk1;
1110                 blk1 = blk2;
1111                 blk2 = tmp_blk;
1112                 leaf1 = blk1->bp->data;
1113                 leaf2 = blk2->bp->data;
1114                 swap = 1;
1115         }
1116         hdr1 = &leaf1->hdr;
1117         hdr2 = &leaf2->hdr;
1118
1119         /*
1120          * Examine entries until we reduce the absolute difference in
1121          * byte usage between the two blocks to a minimum.  Then get
1122          * the direction to copy and the number of elements to move.
1123          */
1124         state->inleaf = xfs_dir_leaf_figure_balance(state, blk1, blk2,
1125                                                            &count, &totallen);
1126         if (swap)
1127                 state->inleaf = !state->inleaf;
1128
1129         /*
1130          * Move any entries required from leaf to leaf:
1131          */
1132         if (count < INT_GET(hdr1->count, ARCH_CONVERT)) {
1133                 /*
1134                  * Figure the total bytes to be added to the destination leaf.
1135                  */
1136                 count = INT_GET(hdr1->count, ARCH_CONVERT) - count;     /* number entries being moved */
1137                 space  = INT_GET(hdr1->namebytes, ARCH_CONVERT) - totallen;
1138                 space += count * ((uint)sizeof(xfs_dir_leaf_name_t)-1);
1139                 space += count * (uint)sizeof(xfs_dir_leaf_entry_t);
1140
1141                 /*
1142                  * leaf2 is the destination, compact it if it looks tight.
1143                  */
1144                 max  = INT_GET(hdr2->firstused, ARCH_CONVERT) - (uint)sizeof(xfs_dir_leaf_hdr_t);
1145                 max -= INT_GET(hdr2->count, ARCH_CONVERT) * (uint)sizeof(xfs_dir_leaf_entry_t);
1146                 if (space > max) {
1147                         xfs_dir_leaf_compact(state->args->trans, blk2->bp,
1148                                                                  0, 0);
1149                 }
1150
1151                 /*
1152                  * Move high entries from leaf1 to low end of leaf2.
1153                  */
1154                 xfs_dir_leaf_moveents(leaf1, INT_GET(hdr1->count, ARCH_CONVERT) - count,
1155                                              leaf2, 0, count, state->mp);
1156
1157                 xfs_da_log_buf(state->args->trans, blk1->bp, 0,
1158                                                    state->blocksize-1);
1159                 xfs_da_log_buf(state->args->trans, blk2->bp, 0,
1160                                                    state->blocksize-1);
1161
1162         } else if (count > INT_GET(hdr1->count, ARCH_CONVERT)) {
1163                 /*
1164                  * Figure the total bytes to be added to the destination leaf.
1165                  */
1166                 count -= INT_GET(hdr1->count, ARCH_CONVERT);            /* number entries being moved */
1167                 space  = totallen - INT_GET(hdr1->namebytes, ARCH_CONVERT);
1168                 space += count * ((uint)sizeof(xfs_dir_leaf_name_t)-1);
1169                 space += count * (uint)sizeof(xfs_dir_leaf_entry_t);
1170
1171                 /*
1172                  * leaf1 is the destination, compact it if it looks tight.
1173                  */
1174                 max  = INT_GET(hdr1->firstused, ARCH_CONVERT) - (uint)sizeof(xfs_dir_leaf_hdr_t);
1175                 max -= INT_GET(hdr1->count, ARCH_CONVERT) * (uint)sizeof(xfs_dir_leaf_entry_t);
1176                 if (space > max) {
1177                         xfs_dir_leaf_compact(state->args->trans, blk1->bp,
1178                                                                  0, 0);
1179                 }
1180
1181                 /*
1182                  * Move low entries from leaf2 to high end of leaf1.
1183                  */
1184                 xfs_dir_leaf_moveents(leaf2, 0, leaf1, (int)INT_GET(hdr1->count, ARCH_CONVERT),
1185                                              count, state->mp);
1186
1187                 xfs_da_log_buf(state->args->trans, blk1->bp, 0,
1188                                                    state->blocksize-1);
1189                 xfs_da_log_buf(state->args->trans, blk2->bp, 0,
1190                                                    state->blocksize-1);
1191         }
1192
1193         /*
1194          * Copy out last hashval in each block for B-tree code.
1195          */
1196         blk1->hashval = INT_GET(leaf1->entries[ INT_GET(leaf1->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
1197         blk2->hashval = INT_GET(leaf2->entries[ INT_GET(leaf2->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
1198
1199         /*
1200          * Adjust the expected index for insertion.
1201          * GROT: this doesn't work unless blk2 was originally empty.
1202          */
1203         if (!state->inleaf) {
1204                 blk2->index = blk1->index - INT_GET(leaf1->hdr.count, ARCH_CONVERT);
1205         }
1206 }
1207
1208 /*
1209  * Examine entries until we reduce the absolute difference in
1210  * byte usage between the two blocks to a minimum.
1211  * GROT: Is this really necessary?  With other than a 512 byte blocksize,
1212  * GROT: there will always be enough room in either block for a new entry.
1213  * GROT: Do a double-split for this case?
1214  */
1215 STATIC int
1216 xfs_dir_leaf_figure_balance(xfs_da_state_t *state,
1217                                            xfs_da_state_blk_t *blk1,
1218                                            xfs_da_state_blk_t *blk2,
1219                                            int *countarg, int *namebytesarg)
1220 {
1221         xfs_dir_leafblock_t *leaf1, *leaf2;
1222         xfs_dir_leaf_hdr_t *hdr1, *hdr2;
1223         xfs_dir_leaf_entry_t *entry;
1224         int count, max, totallen, half;
1225         int lastdelta, foundit, tmp;
1226
1227         /*
1228          * Set up environment.
1229          */
1230         leaf1 = blk1->bp->data;
1231         leaf2 = blk2->bp->data;
1232         hdr1 = &leaf1->hdr;
1233         hdr2 = &leaf2->hdr;
1234         foundit = 0;
1235         totallen = 0;
1236
1237         /*
1238          * Examine entries until we reduce the absolute difference in
1239          * byte usage between the two blocks to a minimum.
1240          */
1241         max = INT_GET(hdr1->count, ARCH_CONVERT) + INT_GET(hdr2->count, ARCH_CONVERT);
1242         half  = (max+1) * (uint)(sizeof(*entry)+sizeof(xfs_dir_leaf_entry_t)-1);
1243         half += INT_GET(hdr1->namebytes, ARCH_CONVERT) + INT_GET(hdr2->namebytes, ARCH_CONVERT) + state->args->namelen;
1244         half /= 2;
1245         lastdelta = state->blocksize;
1246         entry = &leaf1->entries[0];
1247         for (count = 0; count < max; entry++, count++) {
1248
1249 #define XFS_DIR_ABS(A)  (((A) < 0) ? -(A) : (A))
1250                 /*
1251                  * The new entry is in the first block, account for it.
1252                  */
1253                 if (count == blk1->index) {
1254                         tmp = totallen + (uint)sizeof(*entry)
1255                                 + XFS_DIR_LEAF_ENTSIZE_BYNAME(state->args->namelen);
1256                         if (XFS_DIR_ABS(half - tmp) > lastdelta)
1257                                 break;
1258                         lastdelta = XFS_DIR_ABS(half - tmp);
1259                         totallen = tmp;
1260                         foundit = 1;
1261                 }
1262
1263                 /*
1264                  * Wrap around into the second block if necessary.
1265                  */
1266                 if (count == INT_GET(hdr1->count, ARCH_CONVERT)) {
1267                         leaf1 = leaf2;
1268                         entry = &leaf1->entries[0];
1269                 }
1270
1271                 /*
1272                  * Figure out if next leaf entry would be too much.
1273                  */
1274                 tmp = totallen + (uint)sizeof(*entry)
1275                                 + XFS_DIR_LEAF_ENTSIZE_BYENTRY(entry);
1276                 if (XFS_DIR_ABS(half - tmp) > lastdelta)
1277                         break;
1278                 lastdelta = XFS_DIR_ABS(half - tmp);
1279                 totallen = tmp;
1280 #undef XFS_DIR_ABS
1281         }
1282
1283         /*
1284          * Calculate the number of namebytes that will end up in lower block.
1285          * If new entry not in lower block, fix up the count.
1286          */
1287         totallen -=
1288                 count * (uint)(sizeof(*entry)+sizeof(xfs_dir_leaf_entry_t)-1);
1289         if (foundit) {
1290                 totallen -= (sizeof(*entry)+sizeof(xfs_dir_leaf_entry_t)-1) +
1291                             state->args->namelen;
1292         }
1293
1294         *countarg = count;
1295         *namebytesarg = totallen;
1296         return foundit;
1297 }
1298
1299 /*========================================================================
1300  * Routines used for shrinking the Btree.
1301  *========================================================================*/
1302
1303 /*
1304  * Check a leaf block and its neighbors to see if the block should be
1305  * collapsed into one or the other neighbor.  Always keep the block
1306  * with the smaller block number.
1307  * If the current block is over 50% full, don't try to join it, return 0.
1308  * If the block is empty, fill in the state structure and return 2.
1309  * If it can be collapsed, fill in the state structure and return 1.
1310  * If nothing can be done, return 0.
1311  */
1312 int
1313 xfs_dir_leaf_toosmall(xfs_da_state_t *state, int *action)
1314 {
1315         xfs_dir_leafblock_t *leaf;
1316         xfs_da_state_blk_t *blk;
1317         xfs_da_blkinfo_t *info;
1318         int count, bytes, forward, error, retval, i;
1319         xfs_dablk_t blkno;
1320         xfs_dabuf_t *bp;
1321
1322         /*
1323          * Check for the degenerate case of the block being over 50% full.
1324          * If so, it's not worth even looking to see if we might be able
1325          * to coalesce with a sibling.
1326          */
1327         blk = &state->path.blk[ state->path.active-1 ];
1328         info = blk->bp->data;
1329         ASSERT(be16_to_cpu(info->magic) == XFS_DIR_LEAF_MAGIC);
1330         leaf = (xfs_dir_leafblock_t *)info;
1331         count = INT_GET(leaf->hdr.count, ARCH_CONVERT);
1332         bytes = (uint)sizeof(xfs_dir_leaf_hdr_t) +
1333                 count * (uint)sizeof(xfs_dir_leaf_entry_t) +
1334                 count * ((uint)sizeof(xfs_dir_leaf_name_t)-1) +
1335                 INT_GET(leaf->hdr.namebytes, ARCH_CONVERT);
1336         if (bytes > (state->blocksize >> 1)) {
1337                 *action = 0;    /* blk over 50%, don't try to join */
1338                 return 0;
1339         }
1340
1341         /*
1342          * Check for the degenerate case of the block being empty.
1343          * If the block is empty, we'll simply delete it, no need to
1344          * coalesce it with a sibling block.  We choose (arbitrarily)
1345          * to merge with the forward block unless it is NULL.
1346          */
1347         if (count == 0) {
1348                 /*
1349                  * Make altpath point to the block we want to keep and
1350                  * path point to the block we want to drop (this one).
1351                  */
1352                 forward = (info->forw != 0);
1353                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1354                 error = xfs_da_path_shift(state, &state->altpath, forward,
1355                                                  0, &retval);
1356                 if (error)
1357                         return error;
1358                 if (retval) {
1359                         *action = 0;
1360                 } else {
1361                         *action = 2;
1362                 }
1363                 return 0;
1364         }
1365
1366         /*
1367          * Examine each sibling block to see if we can coalesce with
1368          * at least 25% free space to spare.  We need to figure out
1369          * whether to merge with the forward or the backward block.
1370          * We prefer coalescing with the lower numbered sibling so as
1371          * to shrink a directory over time.
1372          */
1373         forward = (be32_to_cpu(info->forw) < be32_to_cpu(info->back));  /* start with smaller blk num */
1374         for (i = 0; i < 2; forward = !forward, i++) {
1375                 if (forward)
1376                         blkno = be32_to_cpu(info->forw);
1377                 else
1378                         blkno = be32_to_cpu(info->back);
1379                 if (blkno == 0)
1380                         continue;
1381                 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1382                                                             blkno, -1, &bp,
1383                                                             XFS_DATA_FORK);
1384                 if (error)
1385                         return error;
1386                 ASSERT(bp != NULL);
1387
1388                 leaf = (xfs_dir_leafblock_t *)info;
1389                 count  = INT_GET(leaf->hdr.count, ARCH_CONVERT);
1390                 bytes  = state->blocksize - (state->blocksize>>2);
1391                 bytes -= INT_GET(leaf->hdr.namebytes, ARCH_CONVERT);
1392                 leaf = bp->data;
1393                 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
1394                 count += INT_GET(leaf->hdr.count, ARCH_CONVERT);
1395                 bytes -= INT_GET(leaf->hdr.namebytes, ARCH_CONVERT);
1396                 bytes -= count * ((uint)sizeof(xfs_dir_leaf_name_t) - 1);
1397                 bytes -= count * (uint)sizeof(xfs_dir_leaf_entry_t);
1398                 bytes -= (uint)sizeof(xfs_dir_leaf_hdr_t);
1399                 if (bytes >= 0)
1400                         break;  /* fits with at least 25% to spare */
1401
1402                 xfs_da_brelse(state->args->trans, bp);
1403         }
1404         if (i >= 2) {
1405                 *action = 0;
1406                 return 0;
1407         }
1408         xfs_da_buf_done(bp);
1409
1410         /*
1411          * Make altpath point to the block we want to keep (the lower
1412          * numbered block) and path point to the block we want to drop.
1413          */
1414         memcpy(&state->altpath, &state->path, sizeof(state->path));
1415         if (blkno < blk->blkno) {
1416                 error = xfs_da_path_shift(state, &state->altpath, forward,
1417                                                  0, &retval);
1418         } else {
1419                 error = xfs_da_path_shift(state, &state->path, forward,
1420                                                  0, &retval);
1421         }
1422         if (error)
1423                 return error;
1424         if (retval) {
1425                 *action = 0;
1426         } else {
1427                 *action = 1;
1428         }
1429         return 0;
1430 }
1431
1432 /*
1433  * Remove a name from the leaf directory structure.
1434  *
1435  * Return 1 if leaf is less than 37% full, 0 if >= 37% full.
1436  * If two leaves are 37% full, when combined they will leave 25% free.
1437  */
1438 int
1439 xfs_dir_leaf_remove(xfs_trans_t *trans, xfs_dabuf_t *bp, int index)
1440 {
1441         xfs_dir_leafblock_t *leaf;
1442         xfs_dir_leaf_hdr_t *hdr;
1443         xfs_dir_leaf_map_t *map;
1444         xfs_dir_leaf_entry_t *entry;
1445         xfs_dir_leaf_name_t *namest;
1446         int before, after, smallest, entsize;
1447         int tablesize, tmp, i;
1448         xfs_mount_t *mp;
1449
1450         leaf = bp->data;
1451         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
1452         hdr = &leaf->hdr;
1453         mp = trans->t_mountp;
1454         ASSERT((INT_GET(hdr->count, ARCH_CONVERT) > 0) && (INT_GET(hdr->count, ARCH_CONVERT) < (XFS_LBSIZE(mp)/8)));
1455         ASSERT((index >= 0) && (index < INT_GET(hdr->count, ARCH_CONVERT)));
1456         ASSERT(INT_GET(hdr->firstused, ARCH_CONVERT) >= ((INT_GET(hdr->count, ARCH_CONVERT)*sizeof(*entry))+sizeof(*hdr)));
1457         entry = &leaf->entries[index];
1458         ASSERT(INT_GET(entry->nameidx, ARCH_CONVERT) >= INT_GET(hdr->firstused, ARCH_CONVERT));
1459         ASSERT(INT_GET(entry->nameidx, ARCH_CONVERT) < XFS_LBSIZE(mp));
1460
1461         /*
1462          * Scan through free region table:
1463          *    check for adjacency of free'd entry with an existing one,
1464          *    find smallest free region in case we need to replace it,
1465          *    adjust any map that borders the entry table,
1466          */
1467         tablesize = INT_GET(hdr->count, ARCH_CONVERT) * (uint)sizeof(xfs_dir_leaf_entry_t)
1468                         + (uint)sizeof(xfs_dir_leaf_hdr_t);
1469         map = &hdr->freemap[0];
1470         tmp = INT_GET(map->size, ARCH_CONVERT);
1471         before = after = -1;
1472         smallest = XFS_DIR_LEAF_MAPSIZE - 1;
1473         entsize = XFS_DIR_LEAF_ENTSIZE_BYENTRY(entry);
1474         for (i = 0; i < XFS_DIR_LEAF_MAPSIZE; map++, i++) {
1475                 ASSERT(INT_GET(map->base, ARCH_CONVERT) < XFS_LBSIZE(mp));
1476                 ASSERT(INT_GET(map->size, ARCH_CONVERT) < XFS_LBSIZE(mp));
1477                 if (INT_GET(map->base, ARCH_CONVERT) == tablesize) {
1478                         INT_MOD(map->base, ARCH_CONVERT, -((uint)sizeof(xfs_dir_leaf_entry_t)));
1479                         INT_MOD(map->size, ARCH_CONVERT, (uint)sizeof(xfs_dir_leaf_entry_t));
1480                 }
1481
1482                 if ((INT_GET(map->base, ARCH_CONVERT) + INT_GET(map->size, ARCH_CONVERT)) == INT_GET(entry->nameidx, ARCH_CONVERT)) {
1483                         before = i;
1484                 } else if (INT_GET(map->base, ARCH_CONVERT) == (INT_GET(entry->nameidx, ARCH_CONVERT) + entsize)) {
1485                         after = i;
1486                 } else if (INT_GET(map->size, ARCH_CONVERT) < tmp) {
1487                         tmp = INT_GET(map->size, ARCH_CONVERT);
1488                         smallest = i;
1489                 }
1490         }
1491
1492         /*
1493          * Coalesce adjacent freemap regions,
1494          * or replace the smallest region.
1495          */
1496         if ((before >= 0) || (after >= 0)) {
1497                 if ((before >= 0) && (after >= 0)) {
1498                         map = &hdr->freemap[before];
1499                         INT_MOD(map->size, ARCH_CONVERT, entsize);
1500                         INT_MOD(map->size, ARCH_CONVERT, INT_GET(hdr->freemap[after].size, ARCH_CONVERT));
1501                         hdr->freemap[after].base = 0;
1502                         hdr->freemap[after].size = 0;
1503                 } else if (before >= 0) {
1504                         map = &hdr->freemap[before];
1505                         INT_MOD(map->size, ARCH_CONVERT, entsize);
1506                 } else {
1507                         map = &hdr->freemap[after];
1508                         INT_COPY(map->base, entry->nameidx, ARCH_CONVERT);
1509                         INT_MOD(map->size, ARCH_CONVERT, entsize);
1510                 }
1511         } else {
1512                 /*
1513                  * Replace smallest region (if it is smaller than free'd entry)
1514                  */
1515                 map = &hdr->freemap[smallest];
1516                 if (INT_GET(map->size, ARCH_CONVERT) < entsize) {
1517                         INT_COPY(map->base, entry->nameidx, ARCH_CONVERT);
1518                         INT_SET(map->size, ARCH_CONVERT, entsize);
1519                 }
1520         }
1521
1522         /*
1523          * Did we remove the first entry?
1524          */
1525         if (INT_GET(entry->nameidx, ARCH_CONVERT) == INT_GET(hdr->firstused, ARCH_CONVERT))
1526                 smallest = 1;
1527         else
1528                 smallest = 0;
1529
1530         /*
1531          * Compress the remaining entries and zero out the removed stuff.
1532          */
1533         namest = XFS_DIR_LEAF_NAMESTRUCT(leaf, INT_GET(entry->nameidx, ARCH_CONVERT));
1534         memset((char *)namest, 0, entsize);
1535         xfs_da_log_buf(trans, bp, XFS_DA_LOGRANGE(leaf, namest, entsize));
1536
1537         INT_MOD(hdr->namebytes, ARCH_CONVERT, -(entry->namelen));
1538         tmp = (INT_GET(hdr->count, ARCH_CONVERT) - index) * (uint)sizeof(xfs_dir_leaf_entry_t);
1539         memmove(entry, entry + 1, tmp);
1540         INT_MOD(hdr->count, ARCH_CONVERT, -1);
1541         xfs_da_log_buf(trans, bp,
1542             XFS_DA_LOGRANGE(leaf, entry, tmp + (uint)sizeof(*entry)));
1543         entry = &leaf->entries[INT_GET(hdr->count, ARCH_CONVERT)];
1544         memset((char *)entry, 0, sizeof(xfs_dir_leaf_entry_t));
1545
1546         /*
1547          * If we removed the first entry, re-find the first used byte
1548          * in the name area.  Note that if the entry was the "firstused",
1549          * then we don't have a "hole" in our block resulting from
1550          * removing the name.
1551          */
1552         if (smallest) {
1553                 tmp = XFS_LBSIZE(mp);
1554                 entry = &leaf->entries[0];
1555                 for (i = INT_GET(hdr->count, ARCH_CONVERT)-1; i >= 0; entry++, i--) {
1556                         ASSERT(INT_GET(entry->nameidx, ARCH_CONVERT) >= INT_GET(hdr->firstused, ARCH_CONVERT));
1557                         ASSERT(INT_GET(entry->nameidx, ARCH_CONVERT) < XFS_LBSIZE(mp));
1558                         if (INT_GET(entry->nameidx, ARCH_CONVERT) < tmp)
1559                                 tmp = INT_GET(entry->nameidx, ARCH_CONVERT);
1560                 }
1561                 INT_SET(hdr->firstused, ARCH_CONVERT, tmp);
1562                 if (!hdr->firstused)
1563                         INT_SET(hdr->firstused, ARCH_CONVERT, tmp - 1);
1564         } else {
1565                 hdr->holes = 1;         /* mark as needing compaction */
1566         }
1567
1568         xfs_da_log_buf(trans, bp, XFS_DA_LOGRANGE(leaf, hdr, sizeof(*hdr)));
1569
1570         /*
1571          * Check if leaf is less than 50% full, caller may want to
1572          * "join" the leaf with a sibling if so.
1573          */
1574         tmp  = (uint)sizeof(xfs_dir_leaf_hdr_t);
1575         tmp += INT_GET(leaf->hdr.count, ARCH_CONVERT) * (uint)sizeof(xfs_dir_leaf_entry_t);
1576         tmp += INT_GET(leaf->hdr.count, ARCH_CONVERT) * ((uint)sizeof(xfs_dir_leaf_name_t) - 1);
1577         tmp += INT_GET(leaf->hdr.namebytes, ARCH_CONVERT);
1578         if (tmp < mp->m_dir_magicpct)
1579                 return 1;                       /* leaf is < 37% full */
1580         return 0;
1581 }
1582
1583 /*
1584  * Move all the directory entries from drop_leaf into save_leaf.
1585  */
1586 void
1587 xfs_dir_leaf_unbalance(xfs_da_state_t *state, xfs_da_state_blk_t *drop_blk,
1588                                       xfs_da_state_blk_t *save_blk)
1589 {
1590         xfs_dir_leafblock_t *drop_leaf, *save_leaf, *tmp_leaf;
1591         xfs_dir_leaf_hdr_t *drop_hdr, *save_hdr, *tmp_hdr;
1592         xfs_mount_t *mp;
1593         char *tmpbuffer;
1594
1595         /*
1596          * Set up environment.
1597          */
1598         mp = state->mp;
1599         ASSERT(drop_blk->magic == XFS_DIR_LEAF_MAGIC);
1600         ASSERT(save_blk->magic == XFS_DIR_LEAF_MAGIC);
1601         drop_leaf = drop_blk->bp->data;
1602         save_leaf = save_blk->bp->data;
1603         ASSERT(be16_to_cpu(drop_leaf->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
1604         ASSERT(be16_to_cpu(save_leaf->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
1605         drop_hdr = &drop_leaf->hdr;
1606         save_hdr = &save_leaf->hdr;
1607
1608         /*
1609          * Save last hashval from dying block for later Btree fixup.
1610          */
1611         drop_blk->hashval = INT_GET(drop_leaf->entries[ drop_leaf->hdr.count-1 ].hashval, ARCH_CONVERT);
1612
1613         /*
1614          * Check if we need a temp buffer, or can we do it in place.
1615          * Note that we don't check "leaf" for holes because we will
1616          * always be dropping it, toosmall() decided that for us already.
1617          */
1618         if (save_hdr->holes == 0) {
1619                 /*
1620                  * dest leaf has no holes, so we add there.  May need
1621                  * to make some room in the entry array.
1622                  */
1623                 if (xfs_dir_leaf_order(save_blk->bp, drop_blk->bp)) {
1624                         xfs_dir_leaf_moveents(drop_leaf, 0, save_leaf, 0,
1625                                                  (int)INT_GET(drop_hdr->count, ARCH_CONVERT), mp);
1626                 } else {
1627                         xfs_dir_leaf_moveents(drop_leaf, 0,
1628                                               save_leaf, INT_GET(save_hdr->count, ARCH_CONVERT),
1629                                               (int)INT_GET(drop_hdr->count, ARCH_CONVERT), mp);
1630                 }
1631         } else {
1632                 /*
1633                  * Destination has holes, so we make a temporary copy
1634                  * of the leaf and add them both to that.
1635                  */
1636                 tmpbuffer = kmem_alloc(state->blocksize, KM_SLEEP);
1637                 ASSERT(tmpbuffer != NULL);
1638                 memset(tmpbuffer, 0, state->blocksize);
1639                 tmp_leaf = (xfs_dir_leafblock_t *)tmpbuffer;
1640                 tmp_hdr = &tmp_leaf->hdr;
1641                 tmp_hdr->info = save_hdr->info; /* struct copy */
1642                 tmp_hdr->count = 0;
1643                 INT_SET(tmp_hdr->firstused, ARCH_CONVERT, state->blocksize);
1644                 if (!tmp_hdr->firstused)
1645                         INT_SET(tmp_hdr->firstused, ARCH_CONVERT, state->blocksize - 1);
1646                 tmp_hdr->namebytes = 0;
1647                 if (xfs_dir_leaf_order(save_blk->bp, drop_blk->bp)) {
1648                         xfs_dir_leaf_moveents(drop_leaf, 0, tmp_leaf, 0,
1649                                                  (int)INT_GET(drop_hdr->count, ARCH_CONVERT), mp);
1650                         xfs_dir_leaf_moveents(save_leaf, 0,
1651                                               tmp_leaf, INT_GET(tmp_leaf->hdr.count, ARCH_CONVERT),
1652                                               (int)INT_GET(save_hdr->count, ARCH_CONVERT), mp);
1653                 } else {
1654                         xfs_dir_leaf_moveents(save_leaf, 0, tmp_leaf, 0,
1655                                                  (int)INT_GET(save_hdr->count, ARCH_CONVERT), mp);
1656                         xfs_dir_leaf_moveents(drop_leaf, 0,
1657                                               tmp_leaf, INT_GET(tmp_leaf->hdr.count, ARCH_CONVERT),
1658                                               (int)INT_GET(drop_hdr->count, ARCH_CONVERT), mp);
1659                 }
1660                 memcpy(save_leaf, tmp_leaf, state->blocksize);
1661                 kmem_free(tmpbuffer, state->blocksize);
1662         }
1663
1664         xfs_da_log_buf(state->args->trans, save_blk->bp, 0,
1665                                            state->blocksize - 1);
1666
1667         /*
1668          * Copy out last hashval in each block for B-tree code.
1669          */
1670         save_blk->hashval = INT_GET(save_leaf->entries[ INT_GET(save_leaf->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT);
1671 }
1672
1673 /*========================================================================
1674  * Routines used for finding things in the Btree.
1675  *========================================================================*/
1676
1677 /*
1678  * Look up a name in a leaf directory structure.
1679  * This is the internal routine, it uses the caller's buffer.
1680  *
1681  * Note that duplicate keys are allowed, but only check within the
1682  * current leaf node.  The Btree code must check in adjacent leaf nodes.
1683  *
1684  * Return in *index the index into the entry[] array of either the found
1685  * entry, or where the entry should have been (insert before that entry).
1686  *
1687  * Don't change the args->inumber unless we find the filename.
1688  */
1689 int
1690 xfs_dir_leaf_lookup_int(xfs_dabuf_t *bp, xfs_da_args_t *args, int *index)
1691 {
1692         xfs_dir_leafblock_t *leaf;
1693         xfs_dir_leaf_entry_t *entry;
1694         xfs_dir_leaf_name_t *namest;
1695         int probe, span;
1696         xfs_dahash_t hashval;
1697
1698         leaf = bp->data;
1699         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
1700         ASSERT(INT_GET(leaf->hdr.count, ARCH_CONVERT) < (XFS_LBSIZE(args->dp->i_mount)/8));
1701
1702         /*
1703          * Binary search.  (note: small blocks will skip this loop)
1704          */
1705         hashval = args->hashval;
1706         probe = span = INT_GET(leaf->hdr.count, ARCH_CONVERT) / 2;
1707         for (entry = &leaf->entries[probe]; span > 4;
1708                    entry = &leaf->entries[probe]) {
1709                 span /= 2;
1710                 if (INT_GET(entry->hashval, ARCH_CONVERT) < hashval)
1711                         probe += span;
1712                 else if (INT_GET(entry->hashval, ARCH_CONVERT) > hashval)
1713                         probe -= span;
1714                 else
1715                         break;
1716         }
1717         ASSERT((probe >= 0) && \
1718                ((!leaf->hdr.count) || (probe < INT_GET(leaf->hdr.count, ARCH_CONVERT))));
1719         ASSERT((span <= 4) || (INT_GET(entry->hashval, ARCH_CONVERT) == hashval));
1720
1721         /*
1722          * Since we may have duplicate hashval's, find the first matching
1723          * hashval in the leaf.
1724          */
1725         while ((probe > 0) && (INT_GET(entry->hashval, ARCH_CONVERT) >= hashval)) {
1726                 entry--;
1727                 probe--;
1728         }
1729         while ((probe < INT_GET(leaf->hdr.count, ARCH_CONVERT)) && (INT_GET(entry->hashval, ARCH_CONVERT) < hashval)) {
1730                 entry++;
1731                 probe++;
1732         }
1733         if ((probe == INT_GET(leaf->hdr.count, ARCH_CONVERT)) || (INT_GET(entry->hashval, ARCH_CONVERT) != hashval)) {
1734                 *index = probe;
1735                 ASSERT(args->oknoent);
1736                 return XFS_ERROR(ENOENT);
1737         }
1738
1739         /*
1740          * Duplicate keys may be present, so search all of them for a match.
1741          */
1742         while ((probe < INT_GET(leaf->hdr.count, ARCH_CONVERT)) && (INT_GET(entry->hashval, ARCH_CONVERT) == hashval)) {
1743                 namest = XFS_DIR_LEAF_NAMESTRUCT(leaf, INT_GET(entry->nameidx, ARCH_CONVERT));
1744                 if (entry->namelen == args->namelen &&
1745                     namest->name[0] == args->name[0] &&
1746                     memcmp(args->name, namest->name, args->namelen) == 0) {
1747                         XFS_DIR_SF_GET_DIRINO(&namest->inumber, &args->inumber);
1748                         *index = probe;
1749                         return XFS_ERROR(EEXIST);
1750                 }
1751                 entry++;
1752                 probe++;
1753         }
1754         *index = probe;
1755         ASSERT(probe == INT_GET(leaf->hdr.count, ARCH_CONVERT) || args->oknoent);
1756         return XFS_ERROR(ENOENT);
1757 }
1758
1759 /*========================================================================
1760  * Utility routines.
1761  *========================================================================*/
1762
1763 /*
1764  * Move the indicated entries from one leaf to another.
1765  * NOTE: this routine modifies both source and destination leaves.
1766  */
1767 /* ARGSUSED */
1768 STATIC void
1769 xfs_dir_leaf_moveents(xfs_dir_leafblock_t *leaf_s, int start_s,
1770                       xfs_dir_leafblock_t *leaf_d, int start_d,
1771                       int count, xfs_mount_t *mp)
1772 {
1773         xfs_dir_leaf_hdr_t *hdr_s, *hdr_d;
1774         xfs_dir_leaf_entry_t *entry_s, *entry_d;
1775         int tmp, i;
1776
1777         /*
1778          * Check for nothing to do.
1779          */
1780         if (count == 0)
1781                 return;
1782
1783         /*
1784          * Set up environment.
1785          */
1786         ASSERT(be16_to_cpu(leaf_s->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
1787         ASSERT(be16_to_cpu(leaf_d->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
1788         hdr_s = &leaf_s->hdr;
1789         hdr_d = &leaf_d->hdr;
1790         ASSERT((INT_GET(hdr_s->count, ARCH_CONVERT) > 0) && (INT_GET(hdr_s->count, ARCH_CONVERT) < (XFS_LBSIZE(mp)/8)));
1791         ASSERT(INT_GET(hdr_s->firstused, ARCH_CONVERT) >=
1792                 ((INT_GET(hdr_s->count, ARCH_CONVERT)*sizeof(*entry_s))+sizeof(*hdr_s)));
1793         ASSERT(INT_GET(hdr_d->count, ARCH_CONVERT) < (XFS_LBSIZE(mp)/8));
1794         ASSERT(INT_GET(hdr_d->firstused, ARCH_CONVERT) >=
1795                 ((INT_GET(hdr_d->count, ARCH_CONVERT)*sizeof(*entry_d))+sizeof(*hdr_d)));
1796
1797         ASSERT(start_s < INT_GET(hdr_s->count, ARCH_CONVERT));
1798         ASSERT(start_d <= INT_GET(hdr_d->count, ARCH_CONVERT));
1799         ASSERT(count <= INT_GET(hdr_s->count, ARCH_CONVERT));
1800
1801         /*
1802          * Move the entries in the destination leaf up to make a hole?
1803          */
1804         if (start_d < INT_GET(hdr_d->count, ARCH_CONVERT)) {
1805                 tmp  = INT_GET(hdr_d->count, ARCH_CONVERT) - start_d;
1806                 tmp *= (uint)sizeof(xfs_dir_leaf_entry_t);
1807                 entry_s = &leaf_d->entries[start_d];
1808                 entry_d = &leaf_d->entries[start_d + count];
1809                 memcpy(entry_d, entry_s, tmp);
1810         }
1811
1812         /*
1813          * Copy all entry's in the same (sorted) order,
1814          * but allocate filenames packed and in sequence.
1815          */
1816         entry_s = &leaf_s->entries[start_s];
1817         entry_d = &leaf_d->entries[start_d];
1818         for (i = 0; i < count; entry_s++, entry_d++, i++) {
1819                 ASSERT(INT_GET(entry_s->nameidx, ARCH_CONVERT) >= INT_GET(hdr_s->firstused, ARCH_CONVERT));
1820                 tmp = XFS_DIR_LEAF_ENTSIZE_BYENTRY(entry_s);
1821                 INT_MOD(hdr_d->firstused, ARCH_CONVERT, -(tmp));
1822                 entry_d->hashval = entry_s->hashval; /* INT_: direct copy */
1823                 INT_COPY(entry_d->nameidx, hdr_d->firstused, ARCH_CONVERT);
1824                 entry_d->namelen = entry_s->namelen;
1825                 ASSERT(INT_GET(entry_d->nameidx, ARCH_CONVERT) + tmp <= XFS_LBSIZE(mp));
1826                 memcpy(XFS_DIR_LEAF_NAMESTRUCT(leaf_d, INT_GET(entry_d->nameidx, ARCH_CONVERT)),
1827                        XFS_DIR_LEAF_NAMESTRUCT(leaf_s, INT_GET(entry_s->nameidx, ARCH_CONVERT)), tmp);
1828                 ASSERT(INT_GET(entry_s->nameidx, ARCH_CONVERT) + tmp <= XFS_LBSIZE(mp));
1829                 memset((char *)XFS_DIR_LEAF_NAMESTRUCT(leaf_s, INT_GET(entry_s->nameidx, ARCH_CONVERT)),
1830                       0, tmp);
1831                 INT_MOD(hdr_s->namebytes, ARCH_CONVERT, -(entry_d->namelen));
1832                 INT_MOD(hdr_d->namebytes, ARCH_CONVERT, entry_d->namelen);
1833                 INT_MOD(hdr_s->count, ARCH_CONVERT, -1);
1834                 INT_MOD(hdr_d->count, ARCH_CONVERT, +1);
1835                 tmp  = INT_GET(hdr_d->count, ARCH_CONVERT) * (uint)sizeof(xfs_dir_leaf_entry_t)
1836                                 + (uint)sizeof(xfs_dir_leaf_hdr_t);
1837                 ASSERT(INT_GET(hdr_d->firstused, ARCH_CONVERT) >= tmp);
1838
1839         }
1840
1841         /*
1842          * Zero out the entries we just copied.
1843          */
1844         if (start_s == INT_GET(hdr_s->count, ARCH_CONVERT)) {
1845                 tmp = count * (uint)sizeof(xfs_dir_leaf_entry_t);
1846                 entry_s = &leaf_s->entries[start_s];
1847                 ASSERT((char *)entry_s + tmp <= (char *)leaf_s + XFS_LBSIZE(mp));
1848                 memset((char *)entry_s, 0, tmp);
1849         } else {
1850                 /*
1851                  * Move the remaining entries down to fill the hole,
1852                  * then zero the entries at the top.
1853                  */
1854                 tmp  = INT_GET(hdr_s->count, ARCH_CONVERT) - count;
1855                 tmp *= (uint)sizeof(xfs_dir_leaf_entry_t);
1856                 entry_s = &leaf_s->entries[start_s + count];
1857                 entry_d = &leaf_s->entries[start_s];
1858                 memcpy(entry_d, entry_s, tmp);
1859
1860                 tmp = count * (uint)sizeof(xfs_dir_leaf_entry_t);
1861                 entry_s = &leaf_s->entries[INT_GET(hdr_s->count, ARCH_CONVERT)];
1862                 ASSERT((char *)entry_s + tmp <= (char *)leaf_s + XFS_LBSIZE(mp));
1863                 memset((char *)entry_s, 0, tmp);
1864         }
1865
1866         /*
1867          * Fill in the freemap information
1868          */
1869         INT_SET(hdr_d->freemap[0].base, ARCH_CONVERT, (uint)sizeof(xfs_dir_leaf_hdr_t));
1870         INT_MOD(hdr_d->freemap[0].base, ARCH_CONVERT, INT_GET(hdr_d->count, ARCH_CONVERT) * (uint)sizeof(xfs_dir_leaf_entry_t));
1871         INT_SET(hdr_d->freemap[0].size, ARCH_CONVERT, INT_GET(hdr_d->firstused, ARCH_CONVERT) - INT_GET(hdr_d->freemap[0].base, ARCH_CONVERT));
1872         INT_SET(hdr_d->freemap[1].base, ARCH_CONVERT, (hdr_d->freemap[2].base = 0));
1873         INT_SET(hdr_d->freemap[1].size, ARCH_CONVERT, (hdr_d->freemap[2].size = 0));
1874         hdr_s->holes = 1;       /* leaf may not be compact */
1875 }
1876
1877 /*
1878  * Compare two leaf blocks "order".
1879  */
1880 int
1881 xfs_dir_leaf_order(xfs_dabuf_t *leaf1_bp, xfs_dabuf_t *leaf2_bp)
1882 {
1883         xfs_dir_leafblock_t *leaf1, *leaf2;
1884
1885         leaf1 = leaf1_bp->data;
1886         leaf2 = leaf2_bp->data;
1887         ASSERT((be16_to_cpu(leaf1->hdr.info.magic) == XFS_DIR_LEAF_MAGIC) &&
1888                (be16_to_cpu(leaf2->hdr.info.magic) == XFS_DIR_LEAF_MAGIC));
1889         if ((INT_GET(leaf1->hdr.count, ARCH_CONVERT) > 0) && (INT_GET(leaf2->hdr.count, ARCH_CONVERT) > 0) &&
1890             ((INT_GET(leaf2->entries[ 0 ].hashval, ARCH_CONVERT) <
1891               INT_GET(leaf1->entries[ 0 ].hashval, ARCH_CONVERT)) ||
1892              (INT_GET(leaf2->entries[ INT_GET(leaf2->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT) <
1893               INT_GET(leaf1->entries[ INT_GET(leaf1->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT)))) {
1894                 return 1;
1895         }
1896         return 0;
1897 }
1898
1899 /*
1900  * Pick up the last hashvalue from a leaf block.
1901  */
1902 xfs_dahash_t
1903 xfs_dir_leaf_lasthash(xfs_dabuf_t *bp, int *count)
1904 {
1905         xfs_dir_leafblock_t *leaf;
1906
1907         leaf = bp->data;
1908         ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR_LEAF_MAGIC);
1909         if (count)
1910                 *count = INT_GET(leaf->hdr.count, ARCH_CONVERT);
1911         if (!leaf->hdr.count)
1912                 return(0);
1913         return(INT_GET(leaf->entries[ INT_GET(leaf->hdr.count, ARCH_CONVERT)-1 ].hashval, ARCH_CONVERT));
1914 }
1915
1916 /*
1917  * Copy out directory entries for getdents(), for leaf directories.
1918  */
1919 int
1920 xfs_dir_leaf_getdents_int(
1921         xfs_dabuf_t     *bp,
1922         xfs_inode_t     *dp,
1923         xfs_dablk_t     bno,
1924         uio_t           *uio,
1925         int             *eobp,
1926         xfs_dirent_t    *dbp,
1927         xfs_dir_put_t   put,
1928         xfs_daddr_t             nextda)
1929 {
1930         xfs_dir_leafblock_t     *leaf;
1931         xfs_dir_leaf_entry_t    *entry;
1932         xfs_dir_leaf_name_t     *namest;
1933         int                     entno, want_entno, i, nextentno;
1934         xfs_mount_t             *mp;
1935         xfs_dahash_t            cookhash;
1936         xfs_dahash_t            nexthash = 0;
1937 #if (BITS_PER_LONG == 32)
1938         xfs_dahash_t            lasthash = XFS_DA_MAXHASH;
1939 #endif
1940         xfs_dir_put_args_t      p;
1941
1942         mp = dp->i_mount;
1943         leaf = bp->data;
1944         if (be16_to_cpu(leaf->hdr.info.magic) != XFS_DIR_LEAF_MAGIC) {
1945                 *eobp = 1;
1946                 return XFS_ERROR(ENOENT);       /* XXX wrong code */
1947         }
1948
1949         want_entno = XFS_DA_COOKIE_ENTRY(mp, uio->uio_offset);
1950
1951         cookhash = XFS_DA_COOKIE_HASH(mp, uio->uio_offset);
1952
1953         xfs_dir_trace_g_dul("leaf: start", dp, uio, leaf);
1954
1955         /*
1956          * Re-find our place.
1957          */
1958         for (i = entno = 0, entry = &leaf->entries[0];
1959                      i < INT_GET(leaf->hdr.count, ARCH_CONVERT);
1960                              entry++, i++) {
1961
1962                 namest = XFS_DIR_LEAF_NAMESTRUCT(leaf,
1963                                     INT_GET(entry->nameidx, ARCH_CONVERT));
1964
1965                 if (unlikely(
1966                     ((char *)namest < (char *)leaf) ||
1967                     ((char *)namest >= (char *)leaf + XFS_LBSIZE(mp)))) {
1968                         XFS_CORRUPTION_ERROR("xfs_dir_leaf_getdents_int(1)",
1969                                              XFS_ERRLEVEL_LOW, mp, leaf);
1970                         xfs_dir_trace_g_du("leaf: corrupted", dp, uio);
1971                         return XFS_ERROR(EFSCORRUPTED);
1972                 }
1973                 if (INT_GET(entry->hashval, ARCH_CONVERT) >= cookhash) {
1974                         if (   entno < want_entno
1975                             && INT_GET(entry->hashval, ARCH_CONVERT)
1976                                                         == cookhash) {
1977                                 /*
1978                                  * Trying to get to a particular offset in a
1979                                  * run of equal-hashval entries.
1980                                  */
1981                                 entno++;
1982                         } else if (   want_entno > 0
1983                                    && entno == want_entno
1984                                    && INT_GET(entry->hashval, ARCH_CONVERT)
1985                                                         == cookhash) {
1986                                 break;
1987                         } else {
1988                                 entno = 0;
1989                                 break;
1990                         }
1991                 }
1992         }
1993
1994         if (i == INT_GET(leaf->hdr.count, ARCH_CONVERT)) {
1995                 xfs_dir_trace_g_du("leaf: hash not found", dp, uio);
1996                 if (!leaf->hdr.info.forw)
1997                         uio->uio_offset =
1998                                 XFS_DA_MAKE_COOKIE(mp, 0, 0, XFS_DA_MAXHASH);
1999                 /*
2000                  * Don't set uio_offset if there's another block:
2001                  * the node code will be setting uio_offset anyway.
2002                  */
2003                 *eobp = 0;
2004                 return 0;
2005         }
2006         xfs_dir_trace_g_due("leaf: hash found", dp, uio, entry);
2007
2008         p.dbp = dbp;
2009         p.put = put;
2010         p.uio = uio;
2011
2012         /*
2013          * We're synchronized, start copying entries out to the user.
2014          */
2015         for (; entno >= 0 && i < INT_GET(leaf->hdr.count, ARCH_CONVERT);
2016                              entry++, i++, (entno = nextentno)) {
2017                 int lastresid=0, retval;
2018                 xfs_dircook_t lastoffset;
2019                 xfs_dahash_t thishash;
2020
2021                 /*
2022                  * Check for a damaged directory leaf block and pick up
2023                  * the inode number from this entry.
2024                  */
2025                 namest = XFS_DIR_LEAF_NAMESTRUCT(leaf,
2026                                     INT_GET(entry->nameidx, ARCH_CONVERT));
2027
2028                 if (unlikely(
2029                     ((char *)namest < (char *)leaf) ||
2030                     ((char *)namest >= (char *)leaf + XFS_LBSIZE(mp)))) {
2031                         XFS_CORRUPTION_ERROR("xfs_dir_leaf_getdents_int(2)",
2032                                              XFS_ERRLEVEL_LOW, mp, leaf);
2033                         xfs_dir_trace_g_du("leaf: corrupted", dp, uio);
2034                         return XFS_ERROR(EFSCORRUPTED);
2035                 }
2036
2037                 xfs_dir_trace_g_duc("leaf: middle cookie  ",
2038                                                    dp, uio, p.cook.o);
2039
2040                 if (i < (INT_GET(leaf->hdr.count, ARCH_CONVERT) - 1)) {
2041                         nexthash = INT_GET(entry[1].hashval, ARCH_CONVERT);
2042
2043                         if (nexthash == INT_GET(entry->hashval, ARCH_CONVERT))
2044                                 nextentno = entno + 1;
2045                         else
2046                                 nextentno = 0;
2047                         XFS_PUT_COOKIE(p.cook, mp, bno, nextentno, nexthash);
2048                         xfs_dir_trace_g_duc("leaf: middle cookie  ",
2049                                                    dp, uio, p.cook.o);
2050
2051                 } else if ((thishash = be32_to_cpu(leaf->hdr.info.forw))) {
2052                         xfs_dabuf_t *bp2;
2053                         xfs_dir_leafblock_t *leaf2;
2054
2055                         ASSERT(nextda != -1);
2056
2057                         retval = xfs_da_read_buf(dp->i_transp, dp, thishash,
2058                                                  nextda, &bp2, XFS_DATA_FORK);
2059                         if (retval)
2060                                 return retval;
2061
2062                         ASSERT(bp2 != NULL);
2063
2064                         leaf2 = bp2->data;
2065
2066                         if (unlikely(
2067                                (be16_to_cpu(leaf2->hdr.info.magic)
2068                                                 != XFS_DIR_LEAF_MAGIC)
2069                             || (be32_to_cpu(leaf2->hdr.info.back)
2070                                                 != bno))) {     /* GROT */
2071                                 XFS_CORRUPTION_ERROR("xfs_dir_leaf_getdents_int(3)",
2072                                                      XFS_ERRLEVEL_LOW, mp,
2073                                                      leaf2);
2074                                 xfs_da_brelse(dp->i_transp, bp2);
2075
2076                                 return XFS_ERROR(EFSCORRUPTED);
2077                         }
2078
2079                         nexthash = INT_GET(leaf2->entries[0].hashval,
2080                                                                 ARCH_CONVERT);
2081                         nextentno = -1;
2082                         XFS_PUT_COOKIE(p.cook, mp, thishash, 0, nexthash);
2083                         xfs_da_brelse(dp->i_transp, bp2);
2084                         xfs_dir_trace_g_duc("leaf: next blk cookie",
2085                                                    dp, uio, p.cook.o);
2086                 } else {
2087                         nextentno = -1;
2088                         XFS_PUT_COOKIE(p.cook, mp, 0, 0, XFS_DA_MAXHASH);
2089                 }
2090
2091                 /*
2092                  * Save off the cookie so we can fall back should the
2093                  * 'put' into the outgoing buffer fails.  To handle a run
2094                  * of equal-hashvals, the off_t structure on 64bit
2095                  * builds has entno built into the cookie to ID the
2096                  * entry.  On 32bit builds, we only have space for the
2097                  * hashval so we can't ID specific entries within a group
2098                  * of same hashval entries.   For this, lastoffset is set
2099                  * to the first in the run of equal hashvals so we don't
2100                  * include any entries unless we can include all entries
2101                  * that share the same hashval.  Hopefully the buffer
2102                  * provided is big enough to handle it (see pv763517).
2103                  */
2104 #if (BITS_PER_LONG == 32)
2105                 if ((thishash = INT_GET(entry->hashval, ARCH_CONVERT))
2106                                                                 != lasthash) {
2107                         XFS_PUT_COOKIE(lastoffset, mp, bno, entno, thishash);
2108                         lastresid = uio->uio_resid;
2109                         lasthash = thishash;
2110                 } else {
2111                         xfs_dir_trace_g_duc("leaf: DUP COOKIES, skipped",
2112                                                    dp, uio, p.cook.o);
2113                 }
2114 #else
2115                 thishash = INT_GET(entry->hashval, ARCH_CONVERT);
2116                 XFS_PUT_COOKIE(lastoffset, mp, bno, entno, thishash);
2117                 lastresid = uio->uio_resid;
2118 #endif /* BITS_PER_LONG == 32 */
2119
2120                 /*
2121                  * Put the current entry into the outgoing buffer.  If we fail
2122                  * then restore the UIO to the first entry in the current
2123                  * run of equal-hashval entries (probably one 1 entry long).
2124                  */
2125                 p.ino = XFS_GET_DIR_INO8(namest->inumber);
2126 #if XFS_BIG_INUMS
2127                 p.ino += mp->m_inoadd;
2128 #endif
2129                 p.name = (char *)namest->name;
2130                 p.namelen = entry->namelen;
2131
2132                 retval = p.put(&p);
2133
2134                 if (!p.done) {
2135                         uio->uio_offset = lastoffset.o;
2136                         uio->uio_resid = lastresid;
2137
2138                         *eobp = 1;
2139
2140                         xfs_dir_trace_g_du("leaf: E-O-B", dp, uio);
2141
2142                         return retval;
2143                 }
2144         }
2145
2146         uio->uio_offset = p.cook.o;
2147
2148         *eobp = 0;
2149
2150         xfs_dir_trace_g_du("leaf: E-O-F", dp, uio);
2151
2152         return 0;
2153 }
2154
2155 /*
2156  * Format a dirent64 structure and copy it out the the user's buffer.
2157  */
2158 int
2159 xfs_dir_put_dirent64_direct(xfs_dir_put_args_t *pa)
2160 {
2161         iovec_t *iovp;
2162         int reclen, namelen;
2163         xfs_dirent_t *idbp;
2164         uio_t *uio;
2165
2166         namelen = pa->namelen;
2167         reclen = DIRENTSIZE(namelen);
2168         uio = pa->uio;
2169         if (reclen > uio->uio_resid) {
2170                 pa->done = 0;
2171                 return 0;
2172         }
2173         iovp = uio->uio_iov;
2174         idbp = (xfs_dirent_t *)iovp->iov_base;
2175         iovp->iov_base = (char *)idbp + reclen;
2176         iovp->iov_len -= reclen;
2177         uio->uio_resid -= reclen;
2178         idbp->d_reclen = reclen;
2179         idbp->d_ino = pa->ino;
2180         idbp->d_off = pa->cook.o;
2181         idbp->d_name[namelen] = '\0';
2182         pa->done = 1;
2183         memcpy(idbp->d_name, pa->name, namelen);
2184         return 0;
2185 }
2186
2187 /*
2188  * Format a dirent64 structure and copy it out the the user's buffer.
2189  */
2190 int
2191 xfs_dir_put_dirent64_uio(xfs_dir_put_args_t *pa)
2192 {
2193         int             retval, reclen, namelen;
2194         xfs_dirent_t    *idbp;
2195         uio_t           *uio;
2196
2197         namelen = pa->namelen;
2198         reclen = DIRENTSIZE(namelen);
2199         uio = pa->uio;
2200         if (reclen > uio->uio_resid) {
2201                 pa->done = 0;
2202                 return 0;
2203         }
2204         idbp = pa->dbp;
2205         idbp->d_reclen = reclen;
2206         idbp->d_ino = pa->ino;
2207         idbp->d_off = pa->cook.o;
2208         idbp->d_name[namelen] = '\0';
2209         memcpy(idbp->d_name, pa->name, namelen);
2210         retval = uio_read((caddr_t)idbp, reclen, uio);
2211         pa->done = (retval == 0);
2212         return retval;
2213 }