fs/xfs/libxfs/xfs_dir2_node.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_mount.h"
  25 #include "xfs_da_format.h"
  26 #include "xfs_da_btree.h"
  27 #include "xfs_inode.h"
  28 #include "xfs_bmap.h"
  29 #include "xfs_dir2.h"
  30 #include "xfs_dir2_priv.h"
  31 #include "xfs_error.h"
  32 #include "xfs_trace.h"
  33 #include "xfs_trans.h"
  34 #include "xfs_buf_item.h"
  35 #include "xfs_cksum.h"
  36 #include "xfs_log.h"
  37
  38 /*
  39  * Function declarations.
  40  */
  41 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
  42                               int index);
  43 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  44                                      xfs_da_state_blk_t *blk1,
  45                                      xfs_da_state_blk_t *blk2);
  46 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
  47                                  int index, xfs_da_state_blk_t *dblk,
  48                                  int *rval);
  49 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  50                                      xfs_da_state_blk_t *fblk);
  51
  52 /*
  53  * Check internal consistency of a leafn block.
  54  */
  55 #ifdef DEBUG
  56 #define xfs_dir3_leaf_check(dp, bp) \
  57 do { \
  58         if (!xfs_dir3_leafn_check((dp), (bp))) \
  59                 ASSERT(0); \
  60 } while (0);
  61
  62 static bool
  63 xfs_dir3_leafn_check(
  64         struct xfs_inode        *dp,
  65         struct xfs_buf          *bp)
  66 {
  67         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  68         struct xfs_dir3_icleaf_hdr leafhdr;
  69
  70         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
  71
  72         if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
  73                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  74                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  75                         return false;
  76         } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
  77                 return false;
  78
  79         return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
  80 }
  81 #else
  82 #define xfs_dir3_leaf_check(dp, bp)
  83 #endif
  84
  85 static bool
  86 xfs_dir3_free_verify(
  87         struct xfs_buf          *bp)
  88 {
  89         struct xfs_mount        *mp = bp->b_target->bt_mount;
  90         struct xfs_dir2_free_hdr *hdr = bp->b_addr;
  91
  92         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  93                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
  94
  95                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_FREE_MAGIC))
  96                         return false;
  97                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
  98                         return false;
  99                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 100                         return false;
 101                 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
 102                         return false;
 103         } else {
 104                 if (hdr->magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC))
 105                         return false;
 106         }
 107
 108         /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
 109
 110         return true;
 111 }
 112
 113 static void
 114 xfs_dir3_free_read_verify(
 115         struct xfs_buf  *bp)
 116 {
 117         struct xfs_mount        *mp = bp->b_target->bt_mount;
 118
 119         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 120             !xfs_buf_verify_cksum(bp, XFS_DIR3_FREE_CRC_OFF))
 121                 xfs_buf_ioerror(bp, -EFSBADCRC);
 122         else if (!xfs_dir3_free_verify(bp))
 123                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 124
 125         if (bp->b_error)
 126                 xfs_verifier_error(bp);
 127 }
 128
 129 static void
 130 xfs_dir3_free_write_verify(
 131         struct xfs_buf  *bp)
 132 {
 133         struct xfs_mount        *mp = bp->b_target->bt_mount;
 134         struct xfs_buf_log_item *bip = bp->b_fspriv;
 135         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 136
 137         if (!xfs_dir3_free_verify(bp)) {
 138                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 139                 xfs_verifier_error(bp);
 140                 return;
 141         }
 142
 143         if (!xfs_sb_version_hascrc(&mp->m_sb))
 144                 return;
 145
 146         if (bip)
 147                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 148
 149         xfs_buf_update_cksum(bp, XFS_DIR3_FREE_CRC_OFF);
 150 }
 151
 152 const struct xfs_buf_ops xfs_dir3_free_buf_ops = {
 153         .name = "xfs_dir3_free",
 154         .verify_read = xfs_dir3_free_read_verify,
 155         .verify_write = xfs_dir3_free_write_verify,
 156 };
 157
 158
 159 static int
 160 __xfs_dir3_free_read(
 161         struct xfs_trans        *tp,
 162         struct xfs_inode        *dp,
 163         xfs_dablk_t             fbno,
 164         xfs_daddr_t             mappedbno,
 165         struct xfs_buf          **bpp)
 166 {
 167         int                     err;
 168
 169         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 170                                 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
 171
 172         /* try read returns without an error or *bpp if it lands in a hole */
 173         if (!err && tp && *bpp)
 174                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
 175         return err;
 176 }
 177
 178 int
 179 xfs_dir2_free_read(
 180         struct xfs_trans        *tp,
 181         struct xfs_inode        *dp,
 182         xfs_dablk_t             fbno,
 183         struct xfs_buf          **bpp)
 184 {
 185         return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
 186 }
 187
 188 static int
 189 xfs_dir2_free_try_read(
 190         struct xfs_trans        *tp,
 191         struct xfs_inode        *dp,
 192         xfs_dablk_t             fbno,
 193         struct xfs_buf          **bpp)
 194 {
 195         return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
 196 }
 197
 198 static int
 199 xfs_dir3_free_get_buf(
 200         xfs_da_args_t           *args,
 201         xfs_dir2_db_t           fbno,
 202         struct xfs_buf          **bpp)
 203 {
 204         struct xfs_trans        *tp = args->trans;
 205         struct xfs_inode        *dp = args->dp;
 206         struct xfs_mount        *mp = dp->i_mount;
 207         struct xfs_buf          *bp;
 208         int                     error;
 209         struct xfs_dir3_icfree_hdr hdr;
 210
 211         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, fbno),
 212                                    -1, &bp, XFS_DATA_FORK);
 213         if (error)
 214                 return error;
 215
 216         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
 217         bp->b_ops = &xfs_dir3_free_buf_ops;
 218
 219         /*
 220          * Initialize the new block to be empty, and remember
 221          * its first slot as our empty slot.
 222          */
 223         memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
 224         memset(&hdr, 0, sizeof(hdr));
 225
 226         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 227                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 228
 229                 hdr.magic = XFS_DIR3_FREE_MAGIC;
 230
 231                 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
 232                 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
 233                 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_meta_uuid);
 234         } else
 235                 hdr.magic = XFS_DIR2_FREE_MAGIC;
 236         dp->d_ops->free_hdr_to_disk(bp->b_addr, &hdr);
 237         *bpp = bp;
 238         return 0;
 239 }
 240
 241 /*
 242  * Log entries from a freespace block.
 243  */
 244 STATIC void
 245 xfs_dir2_free_log_bests(
 246         struct xfs_da_args      *args,
 247         struct xfs_buf          *bp,
 248         int                     first,          /* first entry to log */
 249         int                     last)           /* last entry to log */
 250 {
 251         xfs_dir2_free_t         *free;          /* freespace structure */
 252         __be16                  *bests;
 253
 254         free = bp->b_addr;
 255         bests = args->dp->d_ops->free_bests_p(free);
 256         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 257                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 258         xfs_trans_log_buf(args->trans, bp,
 259                 (uint)((char *)&bests[first] - (char *)free),
 260                 (uint)((char *)&bests[last] - (char *)free +
 261                        sizeof(bests[0]) - 1));
 262 }
 263
 264 /*
 265  * Log header from a freespace block.
 266  */
 267 static void
 268 xfs_dir2_free_log_header(
 269         struct xfs_da_args      *args,
 270         struct xfs_buf          *bp)
 271 {
 272 #ifdef DEBUG
 273         xfs_dir2_free_t         *free;          /* freespace structure */
 274
 275         free = bp->b_addr;
 276         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 277                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 278 #endif
 279         xfs_trans_log_buf(args->trans, bp, 0,
 280                           args->dp->d_ops->free_hdr_size - 1);
 281 }
 282
 283 /*
 284  * Convert a leaf-format directory to a node-format directory.
 285  * We need to change the magic number of the leaf block, and copy
 286  * the freespace table out of the leaf block into its own block.
 287  */
 288 int                                             /* error */
 289 xfs_dir2_leaf_to_node(
 290         xfs_da_args_t           *args,          /* operation arguments */
 291         struct xfs_buf          *lbp)           /* leaf buffer */
 292 {
 293         xfs_inode_t             *dp;            /* incore directory inode */
 294         int                     error;          /* error return value */
 295         struct xfs_buf          *fbp;           /* freespace buffer */
 296         xfs_dir2_db_t           fdb;            /* freespace block number */
 297         xfs_dir2_free_t         *free;          /* freespace structure */
 298         __be16                  *from;          /* pointer to freespace entry */
 299         int                     i;              /* leaf freespace index */
 300         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 301         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 302         int                     n;              /* count of live freespc ents */
 303         xfs_dir2_data_off_t     off;            /* freespace entry value */
 304         __be16                  *to;            /* pointer to freespace entry */
 305         xfs_trans_t             *tp;            /* transaction pointer */
 306         struct xfs_dir3_icfree_hdr freehdr;
 307
 308         trace_xfs_dir2_leaf_to_node(args);
 309
 310         dp = args->dp;
 311         tp = args->trans;
 312         /*
 313          * Add a freespace block to the directory.
 314          */
 315         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 316                 return error;
 317         }
 318         ASSERT(fdb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
 319         /*
 320          * Get the buffer for the new freespace block.
 321          */
 322         error = xfs_dir3_free_get_buf(args, fdb, &fbp);
 323         if (error)
 324                 return error;
 325
 326         free = fbp->b_addr;
 327         dp->d_ops->free_hdr_from_disk(&freehdr, free);
 328         leaf = lbp->b_addr;
 329         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 330         ASSERT(be32_to_cpu(ltp->bestcount) <=
 331                                 (uint)dp->i_d.di_size / args->geo->blksize);
 332
 333         /*
 334          * Copy freespace entries from the leaf block to the new block.
 335          * Count active entries.
 336          */
 337         from = xfs_dir2_leaf_bests_p(ltp);
 338         to = dp->d_ops->free_bests_p(free);
 339         for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 340                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 341                         n++;
 342                 *to = cpu_to_be16(off);
 343         }
 344
 345         /*
 346          * Now initialize the freespace block header.
 347          */
 348         freehdr.nused = n;
 349         freehdr.nvalid = be32_to_cpu(ltp->bestcount);
 350
 351         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
 352         xfs_dir2_free_log_bests(args, fbp, 0, freehdr.nvalid - 1);
 353         xfs_dir2_free_log_header(args, fbp);
 354
 355         /*
 356          * Converting the leaf to a leafnode is just a matter of changing the
 357          * magic number and the ops. Do the change directly to the buffer as
 358          * it's less work (and less code) than decoding the header to host
 359          * format and back again.
 360          */
 361         if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
 362                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 363         else
 364                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 365         lbp->b_ops = &xfs_dir3_leafn_buf_ops;
 366         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
 367         xfs_dir3_leaf_log_header(args, lbp);
 368         xfs_dir3_leaf_check(dp, lbp);
 369         return 0;
 370 }
 371
 372 /*
 373  * Add a leaf entry to a leaf block in a node-form directory.
 374  * The other work necessary is done from the caller.
 375  */
 376 static int                                      /* error */
 377 xfs_dir2_leafn_add(
 378         struct xfs_buf          *bp,            /* leaf buffer */
 379         xfs_da_args_t           *args,          /* operation arguments */
 380         int                     index)          /* insertion pt for new entry */
 381 {
 382         int                     compact;        /* compacting stale leaves */
 383         xfs_inode_t             *dp;            /* incore directory inode */
 384         int                     highstale;      /* next stale entry */
 385         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 386         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 387         int                     lfloghigh;      /* high leaf entry logging */
 388         int                     lfloglow;       /* low leaf entry logging */
 389         int                     lowstale;       /* previous stale entry */
 390         struct xfs_dir3_icleaf_hdr leafhdr;
 391         struct xfs_dir2_leaf_entry *ents;
 392
 393         trace_xfs_dir2_leafn_add(args, index);
 394
 395         dp = args->dp;
 396         leaf = bp->b_addr;
 397         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 398         ents = dp->d_ops->leaf_ents_p(leaf);
 399
 400         /*
 401          * Quick check just to make sure we are not going to index
 402          * into other peoples memory
 403          */
 404         if (index < 0)
 405                 return -EFSCORRUPTED;
 406
 407         /*
 408          * If there are already the maximum number of leaf entries in
 409          * the block, if there are no stale entries it won't fit.
 410          * Caller will do a split.  If there are stale entries we'll do
 411          * a compact.
 412          */
 413
 414         if (leafhdr.count == dp->d_ops->leaf_max_ents(args->geo)) {
 415                 if (!leafhdr.stale)
 416                         return -ENOSPC;
 417                 compact = leafhdr.stale > 1;
 418         } else
 419                 compact = 0;
 420         ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
 421         ASSERT(index == leafhdr.count ||
 422                be32_to_cpu(ents[index].hashval) >= args->hashval);
 423
 424         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 425                 return 0;
 426
 427         /*
 428          * Compact out all but one stale leaf entry.  Leaves behind
 429          * the entry closest to index.
 430          */
 431         if (compact)
 432                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 433                                          &highstale, &lfloglow, &lfloghigh);
 434         else if (leafhdr.stale) {
 435                 /*
 436                  * Set impossible logging indices for this case.
 437                  */
 438                 lfloglow = leafhdr.count;
 439                 lfloghigh = -1;
 440         }
 441
 442         /*
 443          * Insert the new entry, log everything.
 444          */
 445         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 446                                        highstale, &lfloglow, &lfloghigh);
 447
 448         lep->hashval = cpu_to_be32(args->hashval);
 449         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(args->geo,
 450                                 args->blkno, args->index));
 451
 452         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
 453         xfs_dir3_leaf_log_header(args, bp);
 454         xfs_dir3_leaf_log_ents(args, bp, lfloglow, lfloghigh);
 455         xfs_dir3_leaf_check(dp, bp);
 456         return 0;
 457 }
 458
 459 #ifdef DEBUG
 460 static void
 461 xfs_dir2_free_hdr_check(
 462         struct xfs_inode *dp,
 463         struct xfs_buf  *bp,
 464         xfs_dir2_db_t   db)
 465 {
 466         struct xfs_dir3_icfree_hdr hdr;
 467
 468         dp->d_ops->free_hdr_from_disk(&hdr, bp->b_addr);
 469
 470         ASSERT((hdr.firstdb %
 471                 dp->d_ops->free_max_bests(dp->i_mount->m_dir_geo)) == 0);
 472         ASSERT(hdr.firstdb <= db);
 473         ASSERT(db < hdr.firstdb + hdr.nvalid);
 474 }
 475 #else
 476 #define xfs_dir2_free_hdr_check(dp, bp, db)
 477 #endif  /* DEBUG */
 478
 479 /*
 480  * Return the last hash value in the leaf.
 481  * Stale entries are ok.
 482  */
 483 xfs_dahash_t                                    /* hash value */
 484 xfs_dir2_leafn_lasthash(
 485         struct xfs_inode *dp,
 486         struct xfs_buf  *bp,                    /* leaf buffer */
 487         int             *count)                 /* count of entries in leaf */
 488 {
 489         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 490         struct xfs_dir2_leaf_entry *ents;
 491         struct xfs_dir3_icleaf_hdr leafhdr;
 492
 493         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 494
 495         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 496                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
 497
 498         if (count)
 499                 *count = leafhdr.count;
 500         if (!leafhdr.count)
 501                 return 0;
 502
 503         ents = dp->d_ops->leaf_ents_p(leaf);
 504         return be32_to_cpu(ents[leafhdr.count - 1].hashval);
 505 }
 506
 507 /*
 508  * Look up a leaf entry for space to add a name in a node-format leaf block.
 509  * The extrablk in state is a freespace block.
 510  */
 511 STATIC int
 512 xfs_dir2_leafn_lookup_for_addname(
 513         struct xfs_buf          *bp,            /* leaf buffer */
 514         xfs_da_args_t           *args,          /* operation arguments */
 515         int                     *indexp,        /* out: leaf entry index */
 516         xfs_da_state_t          *state)         /* state to fill in */
 517 {
 518         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 519         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 520         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 521         xfs_inode_t             *dp;            /* incore directory inode */
 522         int                     error;          /* error return value */
 523         int                     fi;             /* free entry index */
 524         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 525         int                     index;          /* leaf entry index */
 526         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 527         int                     length;         /* length of new data entry */
 528         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 529         xfs_mount_t             *mp;            /* filesystem mount point */
 530         xfs_dir2_db_t           newdb;          /* new data block number */
 531         xfs_dir2_db_t           newfdb;         /* new free block number */
 532         xfs_trans_t             *tp;            /* transaction pointer */
 533         struct xfs_dir2_leaf_entry *ents;
 534         struct xfs_dir3_icleaf_hdr leafhdr;
 535
 536         dp = args->dp;
 537         tp = args->trans;
 538         mp = dp->i_mount;
 539         leaf = bp->b_addr;
 540         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 541         ents = dp->d_ops->leaf_ents_p(leaf);
 542
 543         xfs_dir3_leaf_check(dp, bp);
 544         ASSERT(leafhdr.count > 0);
 545
 546         /*
 547          * Look up the hash value in the leaf entries.
 548          */
 549         index = xfs_dir2_leaf_search_hash(args, bp);
 550         /*
 551          * Do we have a buffer coming in?
 552          */
 553         if (state->extravalid) {
 554                 /* If so, it's a free block buffer, get the block number. */
 555                 curbp = state->extrablk.bp;
 556                 curfdb = state->extrablk.blkno;
 557                 free = curbp->b_addr;
 558                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 559                        free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 560         }
 561         length = dp->d_ops->data_entsize(args->namelen);
 562         /*
 563          * Loop over leaf entries with the right hash value.
 564          */
 565         for (lep = &ents[index];
 566              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 567              lep++, index++) {
 568                 /*
 569                  * Skip stale leaf entries.
 570                  */
 571                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 572                         continue;
 573                 /*
 574                  * Pull the data block number from the entry.
 575                  */
 576                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 577                                                be32_to_cpu(lep->address));
 578                 /*
 579                  * For addname, we're looking for a place to put the new entry.
 580                  * We want to use a data block with an entry of equal
 581                  * hash value to ours if there is one with room.
 582                  *
 583                  * If this block isn't the data block we already have
 584                  * in hand, take a look at it.
 585                  */
 586                 if (newdb != curdb) {
 587                         __be16 *bests;
 588
 589                         curdb = newdb;
 590                         /*
 591                          * Convert the data block to the free block
 592                          * holding its freespace information.
 593                          */
 594                         newfdb = dp->d_ops->db_to_fdb(args->geo, newdb);
 595                         /*
 596                          * If it's not the one we have in hand, read it in.
 597                          */
 598                         if (newfdb != curfdb) {
 599                                 /*
 600                                  * If we had one before, drop it.
 601                                  */
 602                                 if (curbp)
 603                                         xfs_trans_brelse(tp, curbp);
 604
 605                                 error = xfs_dir2_free_read(tp, dp,
 606                                                 xfs_dir2_db_to_da(args->geo,
 607                                                                   newfdb),
 608                                                 &curbp);
 609                                 if (error)
 610                                         return error;
 611                                 free = curbp->b_addr;
 612
 613                                 xfs_dir2_free_hdr_check(dp, curbp, curdb);
 614                         }
 615                         /*
 616                          * Get the index for our entry.
 617                          */
 618                         fi = dp->d_ops->db_to_fdindex(args->geo, curdb);
 619                         /*
 620                          * If it has room, return it.
 621                          */
 622                         bests = dp->d_ops->free_bests_p(free);
 623                         if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
 624                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 625                                                         XFS_ERRLEVEL_LOW, mp);
 626                                 if (curfdb != newfdb)
 627                                         xfs_trans_brelse(tp, curbp);
 628                                 return -EFSCORRUPTED;
 629                         }
 630                         curfdb = newfdb;
 631                         if (be16_to_cpu(bests[fi]) >= length)
 632                                 goto out;
 633                 }
 634         }
 635         /* Didn't find any space */
 636         fi = -1;
 637 out:
 638         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 639         if (curbp) {
 640                 /* Giving back a free block. */
 641                 state->extravalid = 1;
 642                 state->extrablk.bp = curbp;
 643                 state->extrablk.index = fi;
 644                 state->extrablk.blkno = curfdb;
 645
 646                 /*
 647                  * Important: this magic number is not in the buffer - it's for
 648                  * buffer type information and therefore only the free/data type
 649                  * matters here, not whether CRCs are enabled or not.
 650                  */
 651                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 652         } else {
 653                 state->extravalid = 0;
 654         }
 655         /*
 656          * Return the index, that will be the insertion point.
 657          */
 658         *indexp = index;
 659         return -ENOENT;
 660 }
 661
 662 /*
 663  * Look up a leaf entry in a node-format leaf block.
 664  * The extrablk in state a data block.
 665  */
 666 STATIC int
 667 xfs_dir2_leafn_lookup_for_entry(
 668         struct xfs_buf          *bp,            /* leaf buffer */
 669         xfs_da_args_t           *args,          /* operation arguments */
 670         int                     *indexp,        /* out: leaf entry index */
 671         xfs_da_state_t          *state)         /* state to fill in */
 672 {
 673         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 674         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 675         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 676         xfs_inode_t             *dp;            /* incore directory inode */
 677         int                     error;          /* error return value */
 678         int                     index;          /* leaf entry index */
 679         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 680         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 681         xfs_mount_t             *mp;            /* filesystem mount point */
 682         xfs_dir2_db_t           newdb;          /* new data block number */
 683         xfs_trans_t             *tp;            /* transaction pointer */
 684         enum xfs_dacmp          cmp;            /* comparison result */
 685         struct xfs_dir2_leaf_entry *ents;
 686         struct xfs_dir3_icleaf_hdr leafhdr;
 687
 688         dp = args->dp;
 689         tp = args->trans;
 690         mp = dp->i_mount;
 691         leaf = bp->b_addr;
 692         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 693         ents = dp->d_ops->leaf_ents_p(leaf);
 694
 695         xfs_dir3_leaf_check(dp, bp);
 696         ASSERT(leafhdr.count > 0);
 697
 698         /*
 699          * Look up the hash value in the leaf entries.
 700          */
 701         index = xfs_dir2_leaf_search_hash(args, bp);
 702         /*
 703          * Do we have a buffer coming in?
 704          */
 705         if (state->extravalid) {
 706                 curbp = state->extrablk.bp;
 707                 curdb = state->extrablk.blkno;
 708         }
 709         /*
 710          * Loop over leaf entries with the right hash value.
 711          */
 712         for (lep = &ents[index];
 713              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 714              lep++, index++) {
 715                 /*
 716                  * Skip stale leaf entries.
 717                  */
 718                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 719                         continue;
 720                 /*
 721                  * Pull the data block number from the entry.
 722                  */
 723                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 724                                                be32_to_cpu(lep->address));
 725                 /*
 726                  * Not adding a new entry, so we really want to find
 727                  * the name given to us.
 728                  *
 729                  * If it's a different data block, go get it.
 730                  */
 731                 if (newdb != curdb) {
 732                         /*
 733                          * If we had a block before that we aren't saving
 734                          * for a CI name, drop it
 735                          */
 736                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 737                                                 curdb != state->extrablk.blkno))
 738                                 xfs_trans_brelse(tp, curbp);
 739                         /*
 740                          * If needing the block that is saved with a CI match,
 741                          * use it otherwise read in the new data block.
 742                          */
 743                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 744                                         newdb == state->extrablk.blkno) {
 745                                 ASSERT(state->extravalid);
 746                                 curbp = state->extrablk.bp;
 747                         } else {
 748                                 error = xfs_dir3_data_read(tp, dp,
 749                                                 xfs_dir2_db_to_da(args->geo,
 750                                                                   newdb),
 751                                                 -1, &curbp);
 752                                 if (error)
 753                                         return error;
 754                         }
 755                         xfs_dir3_data_check(dp, curbp);
 756                         curdb = newdb;
 757                 }
 758                 /*
 759                  * Point to the data entry.
 760                  */
 761                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 762                         xfs_dir2_dataptr_to_off(args->geo,
 763                                                 be32_to_cpu(lep->address)));
 764                 /*
 765                  * Compare the entry and if it's an exact match, return
 766                  * EEXIST immediately. If it's the first case-insensitive
 767                  * match, store the block & inode number and continue looking.
 768                  */
 769                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 770                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 771                         /* If there is a CI match block, drop it */
 772                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 773                                                 curdb != state->extrablk.blkno)
 774                                 xfs_trans_brelse(tp, state->extrablk.bp);
 775                         args->cmpresult = cmp;
 776                         args->inumber = be64_to_cpu(dep->inumber);
 777                         args->filetype = dp->d_ops->data_get_ftype(dep);
 778                         *indexp = index;
 779                         state->extravalid = 1;
 780                         state->extrablk.bp = curbp;
 781                         state->extrablk.blkno = curdb;
 782                         state->extrablk.index = (int)((char *)dep -
 783                                                         (char *)curbp->b_addr);
 784                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 785                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 786                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 787                         if (cmp == XFS_CMP_EXACT)
 788                                 return -EEXIST;
 789                 }
 790         }
 791         ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
 792         if (curbp) {
 793                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 794                         /* Giving back last used data block. */
 795                         state->extravalid = 1;
 796                         state->extrablk.bp = curbp;
 797                         state->extrablk.index = -1;
 798                         state->extrablk.blkno = curdb;
 799                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 800                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 801                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 802                 } else {
 803                         /* If the curbp is not the CI match block, drop it */
 804                         if (state->extrablk.bp != curbp)
 805                                 xfs_trans_brelse(tp, curbp);
 806                 }
 807         } else {
 808                 state->extravalid = 0;
 809         }
 810         *indexp = index;
 811         return -ENOENT;
 812 }
 813
 814 /*
 815  * Look up a leaf entry in a node-format leaf block.
 816  * If this is an addname then the extrablk in state is a freespace block,
 817  * otherwise it's a data block.
 818  */
 819 int
 820 xfs_dir2_leafn_lookup_int(
 821         struct xfs_buf          *bp,            /* leaf buffer */
 822         xfs_da_args_t           *args,          /* operation arguments */
 823         int                     *indexp,        /* out: leaf entry index */
 824         xfs_da_state_t          *state)         /* state to fill in */
 825 {
 826         if (args->op_flags & XFS_DA_OP_ADDNAME)
 827                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 828                                                         state);
 829         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 830 }
 831
 832 /*
 833  * Move count leaf entries from source to destination leaf.
 834  * Log entries and headers.  Stale entries are preserved.
 835  */
 836 static void
 837 xfs_dir3_leafn_moveents(
 838         xfs_da_args_t                   *args,  /* operation arguments */
 839         struct xfs_buf                  *bp_s,  /* source */
 840         struct xfs_dir3_icleaf_hdr      *shdr,
 841         struct xfs_dir2_leaf_entry      *sents,
 842         int                             start_s,/* source leaf index */
 843         struct xfs_buf                  *bp_d,  /* destination */
 844         struct xfs_dir3_icleaf_hdr      *dhdr,
 845         struct xfs_dir2_leaf_entry      *dents,
 846         int                             start_d,/* destination leaf index */
 847         int                             count)  /* count of leaves to copy */
 848 {
 849         int                             stale;  /* count stale leaves copied */
 850
 851         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 852
 853         /*
 854          * Silently return if nothing to do.
 855          */
 856         if (count == 0)
 857                 return;
 858
 859         /*
 860          * If the destination index is not the end of the current
 861          * destination leaf entries, open up a hole in the destination
 862          * to hold the new entries.
 863          */
 864         if (start_d < dhdr->count) {
 865                 memmove(&dents[start_d + count], &dents[start_d],
 866                         (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
 867                 xfs_dir3_leaf_log_ents(args, bp_d, start_d + count,
 868                                        count + dhdr->count - 1);
 869         }
 870         /*
 871          * If the source has stale leaves, count the ones in the copy range
 872          * so we can update the header correctly.
 873          */
 874         if (shdr->stale) {
 875                 int     i;                      /* temp leaf index */
 876
 877                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 878                         if (sents[i].address ==
 879                                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 880                                 stale++;
 881                 }
 882         } else
 883                 stale = 0;
 884         /*
 885          * Copy the leaf entries from source to destination.
 886          */
 887         memcpy(&dents[start_d], &sents[start_s],
 888                 count * sizeof(xfs_dir2_leaf_entry_t));
 889         xfs_dir3_leaf_log_ents(args, bp_d, start_d, start_d + count - 1);
 890
 891         /*
 892          * If there are source entries after the ones we copied,
 893          * delete the ones we copied by sliding the next ones down.
 894          */
 895         if (start_s + count < shdr->count) {
 896                 memmove(&sents[start_s], &sents[start_s + count],
 897                         count * sizeof(xfs_dir2_leaf_entry_t));
 898                 xfs_dir3_leaf_log_ents(args, bp_s, start_s, start_s + count - 1);
 899         }
 900
 901         /*
 902          * Update the headers and log them.
 903          */
 904         shdr->count -= count;
 905         shdr->stale -= stale;
 906         dhdr->count += count;
 907         dhdr->stale += stale;
 908 }
 909
 910 /*
 911  * Determine the sort order of two leaf blocks.
 912  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 913  */
 914 int                                             /* sort order */
 915 xfs_dir2_leafn_order(
 916         struct xfs_inode        *dp,
 917         struct xfs_buf          *leaf1_bp,              /* leaf1 buffer */
 918         struct xfs_buf          *leaf2_bp)              /* leaf2 buffer */
 919 {
 920         struct xfs_dir2_leaf    *leaf1 = leaf1_bp->b_addr;
 921         struct xfs_dir2_leaf    *leaf2 = leaf2_bp->b_addr;
 922         struct xfs_dir2_leaf_entry *ents1;
 923         struct xfs_dir2_leaf_entry *ents2;
 924         struct xfs_dir3_icleaf_hdr hdr1;
 925         struct xfs_dir3_icleaf_hdr hdr2;
 926
 927         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
 928         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
 929         ents1 = dp->d_ops->leaf_ents_p(leaf1);
 930         ents2 = dp->d_ops->leaf_ents_p(leaf2);
 931
 932         if (hdr1.count > 0 && hdr2.count > 0 &&
 933             (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
 934              be32_to_cpu(ents2[hdr2.count - 1].hashval) <
 935                                 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
 936                 return 1;
 937         return 0;
 938 }
 939
 940 /*
 941  * Rebalance leaf entries between two leaf blocks.
 942  * This is actually only called when the second block is new,
 943  * though the code deals with the general case.
 944  * A new entry will be inserted in one of the blocks, and that
 945  * entry is taken into account when balancing.
 946  */
 947 static void
 948 xfs_dir2_leafn_rebalance(
 949         xfs_da_state_t          *state,         /* btree cursor */
 950         xfs_da_state_blk_t      *blk1,          /* first btree block */
 951         xfs_da_state_blk_t      *blk2)          /* second btree block */
 952 {
 953         xfs_da_args_t           *args;          /* operation arguments */
 954         int                     count;          /* count (& direction) leaves */
 955         int                     isleft;         /* new goes in left leaf */
 956         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 957         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 958         int                     mid;            /* midpoint leaf index */
 959 #if defined(DEBUG) || defined(XFS_WARN)
 960         int                     oldstale;       /* old count of stale leaves */
 961 #endif
 962         int                     oldsum;         /* old total leaf count */
 963         int                     swap;           /* swapped leaf blocks */
 964         struct xfs_dir2_leaf_entry *ents1;
 965         struct xfs_dir2_leaf_entry *ents2;
 966         struct xfs_dir3_icleaf_hdr hdr1;
 967         struct xfs_dir3_icleaf_hdr hdr2;
 968         struct xfs_inode        *dp = state->args->dp;
 969
 970         args = state->args;
 971         /*
 972          * If the block order is wrong, swap the arguments.
 973          */
 974         if ((swap = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp))) {
 975                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
 976
 977                 tmp = blk1;
 978                 blk1 = blk2;
 979                 blk2 = tmp;
 980         }
 981         leaf1 = blk1->bp->b_addr;
 982         leaf2 = blk2->bp->b_addr;
 983         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
 984         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
 985         ents1 = dp->d_ops->leaf_ents_p(leaf1);
 986         ents2 = dp->d_ops->leaf_ents_p(leaf2);
 987
 988         oldsum = hdr1.count + hdr2.count;
 989 #if defined(DEBUG) || defined(XFS_WARN)
 990         oldstale = hdr1.stale + hdr2.stale;
 991 #endif
 992         mid = oldsum >> 1;
 993
 994         /*
 995          * If the old leaf count was odd then the new one will be even,
 996          * so we need to divide the new count evenly.
 997          */
 998         if (oldsum & 1) {
 999                 xfs_dahash_t    midhash;        /* middle entry hash value */
1000
1001                 if (mid >= hdr1.count)
1002                         midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1003                 else
1004                         midhash = be32_to_cpu(ents1[mid].hashval);
1005                 isleft = args->hashval <= midhash;
1006         }
1007         /*
1008          * If the old count is even then the new count is odd, so there's
1009          * no preferred side for the new entry.
1010          * Pick the left one.
1011          */
1012         else
1013                 isleft = 1;
1014         /*
1015          * Calculate moved entry count.  Positive means left-to-right,
1016          * negative means right-to-left.  Then move the entries.
1017          */
1018         count = hdr1.count - mid + (isleft == 0);
1019         if (count > 0)
1020                 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1021                                         hdr1.count - count, blk2->bp,
1022                                         &hdr2, ents2, 0, count);
1023         else if (count < 0)
1024                 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1025                                         blk1->bp, &hdr1, ents1,
1026                                         hdr1.count, count);
1027
1028         ASSERT(hdr1.count + hdr2.count == oldsum);
1029         ASSERT(hdr1.stale + hdr2.stale == oldstale);
1030
1031         /* log the changes made when moving the entries */
1032         dp->d_ops->leaf_hdr_to_disk(leaf1, &hdr1);
1033         dp->d_ops->leaf_hdr_to_disk(leaf2, &hdr2);
1034         xfs_dir3_leaf_log_header(args, blk1->bp);
1035         xfs_dir3_leaf_log_header(args, blk2->bp);
1036
1037         xfs_dir3_leaf_check(dp, blk1->bp);
1038         xfs_dir3_leaf_check(dp, blk2->bp);
1039
1040         /*
1041          * Mark whether we're inserting into the old or new leaf.
1042          */
1043         if (hdr1.count < hdr2.count)
1044                 state->inleaf = swap;
1045         else if (hdr1.count > hdr2.count)
1046                 state->inleaf = !swap;
1047         else
1048                 state->inleaf = swap ^ (blk1->index <= hdr1.count);
1049         /*
1050          * Adjust the expected index for insertion.
1051          */
1052         if (!state->inleaf)
1053                 blk2->index = blk1->index - hdr1.count;
1054
1055         /*
1056          * Finally sanity check just to make sure we are not returning a
1057          * negative index
1058          */
1059         if (blk2->index < 0) {
1060                 state->inleaf = 1;
1061                 blk2->index = 0;
1062                 xfs_alert(dp->i_mount,
1063         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1064                         __func__, blk1->index);
1065         }
1066 }
1067
1068 static int
1069 xfs_dir3_data_block_free(
1070         xfs_da_args_t           *args,
1071         struct xfs_dir2_data_hdr *hdr,
1072         struct xfs_dir2_free    *free,
1073         xfs_dir2_db_t           fdb,
1074         int                     findex,
1075         struct xfs_buf          *fbp,
1076         int                     longest)
1077 {
1078         int                     logfree = 0;
1079         __be16                  *bests;
1080         struct xfs_dir3_icfree_hdr freehdr;
1081         struct xfs_inode        *dp = args->dp;
1082
1083         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1084         bests = dp->d_ops->free_bests_p(free);
1085         if (hdr) {
1086                 /*
1087                  * Data block is not empty, just set the free entry to the new
1088                  * value.
1089                  */
1090                 bests[findex] = cpu_to_be16(longest);
1091                 xfs_dir2_free_log_bests(args, fbp, findex, findex);
1092                 return 0;
1093         }
1094
1095         /* One less used entry in the free table. */
1096         freehdr.nused--;
1097
1098         /*
1099          * If this was the last entry in the table, we can trim the table size
1100          * back.  There might be other entries at the end referring to
1101          * non-existent data blocks, get those too.
1102          */
1103         if (findex == freehdr.nvalid - 1) {
1104                 int     i;              /* free entry index */
1105
1106                 for (i = findex - 1; i >= 0; i--) {
1107                         if (bests[i] != cpu_to_be16(NULLDATAOFF))
1108                                 break;
1109                 }
1110                 freehdr.nvalid = i + 1;
1111                 logfree = 0;
1112         } else {
1113                 /* Not the last entry, just punch it out.  */
1114                 bests[findex] = cpu_to_be16(NULLDATAOFF);
1115                 logfree = 1;
1116         }
1117
1118         dp->d_ops->free_hdr_to_disk(free, &freehdr);
1119         xfs_dir2_free_log_header(args, fbp);
1120
1121         /*
1122          * If there are no useful entries left in the block, get rid of the
1123          * block if we can.
1124          */
1125         if (!freehdr.nused) {
1126                 int error;
1127
1128                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1129                 if (error == 0) {
1130                         fbp = NULL;
1131                         logfree = 0;
1132                 } else if (error != -ENOSPC || args->total != 0)
1133                         return error;
1134                 /*
1135                  * It's possible to get ENOSPC if there is no
1136                  * space reservation.  In this case some one
1137                  * else will eventually get rid of this block.
1138                  */
1139         }
1140
1141         /* Log the free entry that changed, unless we got rid of it.  */
1142         if (logfree)
1143                 xfs_dir2_free_log_bests(args, fbp, findex, findex);
1144         return 0;
1145 }
1146
1147 /*
1148  * Remove an entry from a node directory.
1149  * This removes the leaf entry and the data entry,
1150  * and updates the free block if necessary.
1151  */
1152 static int                                      /* error */
1153 xfs_dir2_leafn_remove(
1154         xfs_da_args_t           *args,          /* operation arguments */
1155         struct xfs_buf          *bp,            /* leaf buffer */
1156         int                     index,          /* leaf entry index */
1157         xfs_da_state_blk_t      *dblk,          /* data block */
1158         int                     *rval)          /* resulting block needs join */
1159 {
1160         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1161         xfs_dir2_db_t           db;             /* data block number */
1162         struct xfs_buf          *dbp;           /* data block buffer */
1163         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1164         xfs_inode_t             *dp;            /* incore directory inode */
1165         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1166         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1167         int                     longest;        /* longest data free entry */
1168         int                     off;            /* data block entry offset */
1169         int                     needlog;        /* need to log data header */
1170         int                     needscan;       /* need to rescan data frees */
1171         xfs_trans_t             *tp;            /* transaction pointer */
1172         struct xfs_dir2_data_free *bf;          /* bestfree table */
1173         struct xfs_dir3_icleaf_hdr leafhdr;
1174         struct xfs_dir2_leaf_entry *ents;
1175
1176         trace_xfs_dir2_leafn_remove(args, index);
1177
1178         dp = args->dp;
1179         tp = args->trans;
1180         leaf = bp->b_addr;
1181         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1182         ents = dp->d_ops->leaf_ents_p(leaf);
1183
1184         /*
1185          * Point to the entry we're removing.
1186          */
1187         lep = &ents[index];
1188
1189         /*
1190          * Extract the data block and offset from the entry.
1191          */
1192         db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1193         ASSERT(dblk->blkno == db);
1194         off = xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address));
1195         ASSERT(dblk->index == off);
1196
1197         /*
1198          * Kill the leaf entry by marking it stale.
1199          * Log the leaf block changes.
1200          */
1201         leafhdr.stale++;
1202         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1203         xfs_dir3_leaf_log_header(args, bp);
1204
1205         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1206         xfs_dir3_leaf_log_ents(args, bp, index, index);
1207
1208         /*
1209          * Make the data entry free.  Keep track of the longest freespace
1210          * in the data block in case it changes.
1211          */
1212         dbp = dblk->bp;
1213         hdr = dbp->b_addr;
1214         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1215         bf = dp->d_ops->data_bestfree_p(hdr);
1216         longest = be16_to_cpu(bf[0].length);
1217         needlog = needscan = 0;
1218         xfs_dir2_data_make_free(args, dbp, off,
1219                 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1220         /*
1221          * Rescan the data block freespaces for bestfree.
1222          * Log the data block header if needed.
1223          */
1224         if (needscan)
1225                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1226         if (needlog)
1227                 xfs_dir2_data_log_header(args, dbp);
1228         xfs_dir3_data_check(dp, dbp);
1229         /*
1230          * If the longest data block freespace changes, need to update
1231          * the corresponding freeblock entry.
1232          */
1233         if (longest < be16_to_cpu(bf[0].length)) {
1234                 int             error;          /* error return value */
1235                 struct xfs_buf  *fbp;           /* freeblock buffer */
1236                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1237                 int             findex;         /* index in freeblock entries */
1238                 xfs_dir2_free_t *free;          /* freeblock structure */
1239
1240                 /*
1241                  * Convert the data block number to a free block,
1242                  * read in the free block.
1243                  */
1244                 fdb = dp->d_ops->db_to_fdb(args->geo, db);
1245                 error = xfs_dir2_free_read(tp, dp,
1246                                            xfs_dir2_db_to_da(args->geo, fdb),
1247                                            &fbp);
1248                 if (error)
1249                         return error;
1250                 free = fbp->b_addr;
1251 #ifdef DEBUG
1252         {
1253                 struct xfs_dir3_icfree_hdr freehdr;
1254                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1255                 ASSERT(freehdr.firstdb == dp->d_ops->free_max_bests(args->geo) *
1256                         (fdb - xfs_dir2_byte_to_db(args->geo,
1257                                                    XFS_DIR2_FREE_OFFSET)));
1258         }
1259 #endif
1260                 /*
1261                  * Calculate which entry we need to fix.
1262                  */
1263                 findex = dp->d_ops->db_to_fdindex(args->geo, db);
1264                 longest = be16_to_cpu(bf[0].length);
1265                 /*
1266                  * If the data block is now empty we can get rid of it
1267                  * (usually).
1268                  */
1269                 if (longest == args->geo->blksize -
1270                                dp->d_ops->data_entry_offset) {
1271                         /*
1272                          * Try to punch out the data block.
1273                          */
1274                         error = xfs_dir2_shrink_inode(args, db, dbp);
1275                         if (error == 0) {
1276                                 dblk->bp = NULL;
1277                                 hdr = NULL;
1278                         }
1279                         /*
1280                          * We can get ENOSPC if there's no space reservation.
1281                          * In this case just drop the buffer and some one else
1282                          * will eventually get rid of the empty block.
1283                          */
1284                         else if (!(error == -ENOSPC && args->total == 0))
1285                                 return error;
1286                 }
1287                 /*
1288                  * If we got rid of the data block, we can eliminate that entry
1289                  * in the free block.
1290                  */
1291                 error = xfs_dir3_data_block_free(args, hdr, free,
1292                                                  fdb, findex, fbp, longest);
1293                 if (error)
1294                         return error;
1295         }
1296
1297         xfs_dir3_leaf_check(dp, bp);
1298         /*
1299          * Return indication of whether this leaf block is empty enough
1300          * to justify trying to join it with a neighbor.
1301          */
1302         *rval = (dp->d_ops->leaf_hdr_size +
1303                  (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1304                 args->geo->magicpct;
1305         return 0;
1306 }
1307
1308 /*
1309  * Split the leaf entries in the old block into old and new blocks.
1310  */
1311 int                                             /* error */
1312 xfs_dir2_leafn_split(
1313         xfs_da_state_t          *state,         /* btree cursor */
1314         xfs_da_state_blk_t      *oldblk,        /* original block */
1315         xfs_da_state_blk_t      *newblk)        /* newly created block */
1316 {
1317         xfs_da_args_t           *args;          /* operation arguments */
1318         xfs_dablk_t             blkno;          /* new leaf block number */
1319         int                     error;          /* error return value */
1320         struct xfs_inode        *dp;
1321
1322         /*
1323          * Allocate space for a new leaf node.
1324          */
1325         args = state->args;
1326         dp = args->dp;
1327         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1328         error = xfs_da_grow_inode(args, &blkno);
1329         if (error) {
1330                 return error;
1331         }
1332         /*
1333          * Initialize the new leaf block.
1334          */
1335         error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(args->geo, blkno),
1336                                       &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1337         if (error)
1338                 return error;
1339
1340         newblk->blkno = blkno;
1341         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1342         /*
1343          * Rebalance the entries across the two leaves, link the new
1344          * block into the leaves.
1345          */
1346         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1347         error = xfs_da3_blk_link(state, oldblk, newblk);
1348         if (error) {
1349                 return error;
1350         }
1351         /*
1352          * Insert the new entry in the correct block.
1353          */
1354         if (state->inleaf)
1355                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1356         else
1357                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1358         /*
1359          * Update last hashval in each block since we added the name.
1360          */
1361         oldblk->hashval = xfs_dir2_leafn_lasthash(dp, oldblk->bp, NULL);
1362         newblk->hashval = xfs_dir2_leafn_lasthash(dp, newblk->bp, NULL);
1363         xfs_dir3_leaf_check(dp, oldblk->bp);
1364         xfs_dir3_leaf_check(dp, newblk->bp);
1365         return error;
1366 }
1367
1368 /*
1369  * Check a leaf block and its neighbors to see if the block should be
1370  * collapsed into one or the other neighbor.  Always keep the block
1371  * with the smaller block number.
1372  * If the current block is over 50% full, don't try to join it, return 0.
1373  * If the block is empty, fill in the state structure and return 2.
1374  * If it can be collapsed, fill in the state structure and return 1.
1375  * If nothing can be done, return 0.
1376  */
1377 int                                             /* error */
1378 xfs_dir2_leafn_toosmall(
1379         xfs_da_state_t          *state,         /* btree cursor */
1380         int                     *action)        /* resulting action to take */
1381 {
1382         xfs_da_state_blk_t      *blk;           /* leaf block */
1383         xfs_dablk_t             blkno;          /* leaf block number */
1384         struct xfs_buf          *bp;            /* leaf buffer */
1385         int                     bytes;          /* bytes in use */
1386         int                     count;          /* leaf live entry count */
1387         int                     error;          /* error return value */
1388         int                     forward;        /* sibling block direction */
1389         int                     i;              /* sibling counter */
1390         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1391         int                     rval;           /* result from path_shift */
1392         struct xfs_dir3_icleaf_hdr leafhdr;
1393         struct xfs_dir2_leaf_entry *ents;
1394         struct xfs_inode        *dp = state->args->dp;
1395
1396         /*
1397          * Check for the degenerate case of the block being over 50% full.
1398          * If so, it's not worth even looking to see if we might be able
1399          * to coalesce with a sibling.
1400          */
1401         blk = &state->path.blk[state->path.active - 1];
1402         leaf = blk->bp->b_addr;
1403         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1404         ents = dp->d_ops->leaf_ents_p(leaf);
1405         xfs_dir3_leaf_check(dp, blk->bp);
1406
1407         count = leafhdr.count - leafhdr.stale;
1408         bytes = dp->d_ops->leaf_hdr_size + count * sizeof(ents[0]);
1409         if (bytes > (state->args->geo->blksize >> 1)) {
1410                 /*
1411                  * Blk over 50%, don't try to join.
1412                  */
1413                 *action = 0;
1414                 return 0;
1415         }
1416         /*
1417          * Check for the degenerate case of the block being empty.
1418          * If the block is empty, we'll simply delete it, no need to
1419          * coalesce it with a sibling block.  We choose (arbitrarily)
1420          * to merge with the forward block unless it is NULL.
1421          */
1422         if (count == 0) {
1423                 /*
1424                  * Make altpath point to the block we want to keep and
1425                  * path point to the block we want to drop (this one).
1426                  */
1427                 forward = (leafhdr.forw != 0);
1428                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1429                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1430                         &rval);
1431                 if (error)
1432                         return error;
1433                 *action = rval ? 2 : 0;
1434                 return 0;
1435         }
1436         /*
1437          * Examine each sibling block to see if we can coalesce with
1438          * at least 25% free space to spare.  We need to figure out
1439          * whether to merge with the forward or the backward block.
1440          * We prefer coalescing with the lower numbered sibling so as
1441          * to shrink a directory over time.
1442          */
1443         forward = leafhdr.forw < leafhdr.back;
1444         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1445                 struct xfs_dir3_icleaf_hdr hdr2;
1446
1447                 blkno = forward ? leafhdr.forw : leafhdr.back;
1448                 if (blkno == 0)
1449                         continue;
1450                 /*
1451                  * Read the sibling leaf block.
1452                  */
1453                 error = xfs_dir3_leafn_read(state->args->trans, dp,
1454                                             blkno, -1, &bp);
1455                 if (error)
1456                         return error;
1457
1458                 /*
1459                  * Count bytes in the two blocks combined.
1460                  */
1461                 count = leafhdr.count - leafhdr.stale;
1462                 bytes = state->args->geo->blksize -
1463                         (state->args->geo->blksize >> 2);
1464
1465                 leaf = bp->b_addr;
1466                 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf);
1467                 ents = dp->d_ops->leaf_ents_p(leaf);
1468                 count += hdr2.count - hdr2.stale;
1469                 bytes -= count * sizeof(ents[0]);
1470
1471                 /*
1472                  * Fits with at least 25% to spare.
1473                  */
1474                 if (bytes >= 0)
1475                         break;
1476                 xfs_trans_brelse(state->args->trans, bp);
1477         }
1478         /*
1479          * Didn't like either block, give up.
1480          */
1481         if (i >= 2) {
1482                 *action = 0;
1483                 return 0;
1484         }
1485
1486         /*
1487          * Make altpath point to the block we want to keep (the lower
1488          * numbered block) and path point to the block we want to drop.
1489          */
1490         memcpy(&state->altpath, &state->path, sizeof(state->path));
1491         if (blkno < blk->blkno)
1492                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1493                         &rval);
1494         else
1495                 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1496                         &rval);
1497         if (error) {
1498                 return error;
1499         }
1500         *action = rval ? 0 : 1;
1501         return 0;
1502 }
1503
1504 /*
1505  * Move all the leaf entries from drop_blk to save_blk.
1506  * This is done as part of a join operation.
1507  */
1508 void
1509 xfs_dir2_leafn_unbalance(
1510         xfs_da_state_t          *state,         /* cursor */
1511         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1512         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1513 {
1514         xfs_da_args_t           *args;          /* operation arguments */
1515         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1516         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1517         struct xfs_dir3_icleaf_hdr savehdr;
1518         struct xfs_dir3_icleaf_hdr drophdr;
1519         struct xfs_dir2_leaf_entry *sents;
1520         struct xfs_dir2_leaf_entry *dents;
1521         struct xfs_inode        *dp = state->args->dp;
1522
1523         args = state->args;
1524         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1525         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1526         drop_leaf = drop_blk->bp->b_addr;
1527         save_leaf = save_blk->bp->b_addr;
1528
1529         dp->d_ops->leaf_hdr_from_disk(&savehdr, save_leaf);
1530         dp->d_ops->leaf_hdr_from_disk(&drophdr, drop_leaf);
1531         sents = dp->d_ops->leaf_ents_p(save_leaf);
1532         dents = dp->d_ops->leaf_ents_p(drop_leaf);
1533
1534         /*
1535          * If there are any stale leaf entries, take this opportunity
1536          * to purge them.
1537          */
1538         if (drophdr.stale)
1539                 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1540         if (savehdr.stale)
1541                 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1542
1543         /*
1544          * Move the entries from drop to the appropriate end of save.
1545          */
1546         drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1547         if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1548                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1549                                         save_blk->bp, &savehdr, sents, 0,
1550                                         drophdr.count);
1551         else
1552                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1553                                         save_blk->bp, &savehdr, sents,
1554                                         savehdr.count, drophdr.count);
1555         save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1556
1557         /* log the changes made when moving the entries */
1558         dp->d_ops->leaf_hdr_to_disk(save_leaf, &savehdr);
1559         dp->d_ops->leaf_hdr_to_disk(drop_leaf, &drophdr);
1560         xfs_dir3_leaf_log_header(args, save_blk->bp);
1561         xfs_dir3_leaf_log_header(args, drop_blk->bp);
1562
1563         xfs_dir3_leaf_check(dp, save_blk->bp);
1564         xfs_dir3_leaf_check(dp, drop_blk->bp);
1565 }
1566
1567 /*
1568  * Top-level node form directory addname routine.
1569  */
1570 int                                             /* error */
1571 xfs_dir2_node_addname(
1572         xfs_da_args_t           *args)          /* operation arguments */
1573 {
1574         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1575         int                     error;          /* error return value */
1576         int                     rval;           /* sub-return value */
1577         xfs_da_state_t          *state;         /* btree cursor */
1578
1579         trace_xfs_dir2_node_addname(args);
1580
1581         /*
1582          * Allocate and initialize the state (btree cursor).
1583          */
1584         state = xfs_da_state_alloc();
1585         state->args = args;
1586         state->mp = args->dp->i_mount;
1587         /*
1588          * Look up the name.  We're not supposed to find it, but
1589          * this gives us the insertion point.
1590          */
1591         error = xfs_da3_node_lookup_int(state, &rval);
1592         if (error)
1593                 rval = error;
1594         if (rval != -ENOENT) {
1595                 goto done;
1596         }
1597         /*
1598          * Add the data entry to a data block.
1599          * Extravalid is set to a freeblock found by lookup.
1600          */
1601         rval = xfs_dir2_node_addname_int(args,
1602                 state->extravalid ? &state->extrablk : NULL);
1603         if (rval) {
1604                 goto done;
1605         }
1606         blk = &state->path.blk[state->path.active - 1];
1607         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1608         /*
1609          * Add the new leaf entry.
1610          */
1611         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1612         if (rval == 0) {
1613                 /*
1614                  * It worked, fix the hash values up the btree.
1615                  */
1616                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1617                         xfs_da3_fixhashpath(state, &state->path);
1618         } else {
1619                 /*
1620                  * It didn't work, we need to split the leaf block.
1621                  */
1622                 if (args->total == 0) {
1623                         ASSERT(rval == -ENOSPC);
1624                         goto done;
1625                 }
1626                 /*
1627                  * Split the leaf block and insert the new entry.
1628                  */
1629                 rval = xfs_da3_split(state);
1630         }
1631 done:
1632         xfs_da_state_free(state);
1633         return rval;
1634 }
1635
1636 /*
1637  * Add the data entry for a node-format directory name addition.
1638  * The leaf entry is added in xfs_dir2_leafn_add.
1639  * We may enter with a freespace block that the lookup found.
1640  */
1641 static int                                      /* error */
1642 xfs_dir2_node_addname_int(
1643         xfs_da_args_t           *args,          /* operation arguments */
1644         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1645 {
1646         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1647         xfs_dir2_db_t           dbno;           /* data block number */
1648         struct xfs_buf          *dbp;           /* data block buffer */
1649         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1650         xfs_inode_t             *dp;            /* incore directory inode */
1651         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1652         int                     error;          /* error return value */
1653         xfs_dir2_db_t           fbno;           /* freespace block number */
1654         struct xfs_buf          *fbp;           /* freespace buffer */
1655         int                     findex;         /* freespace entry index */
1656         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1657         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1658         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1659         int                     length;         /* length of the new entry */
1660         int                     logfree;        /* need to log free entry */
1661         xfs_mount_t             *mp;            /* filesystem mount point */
1662         int                     needlog;        /* need to log data header */
1663         int                     needscan;       /* need to rescan data frees */
1664         __be16                  *tagp;          /* data entry tag pointer */
1665         xfs_trans_t             *tp;            /* transaction pointer */
1666         __be16                  *bests;
1667         struct xfs_dir3_icfree_hdr freehdr;
1668         struct xfs_dir2_data_free *bf;
1669
1670         dp = args->dp;
1671         mp = dp->i_mount;
1672         tp = args->trans;
1673         length = dp->d_ops->data_entsize(args->namelen);
1674         /*
1675          * If we came in with a freespace block that means that lookup
1676          * found an entry with our hash value.  This is the freespace
1677          * block for that data entry.
1678          */
1679         if (fblk) {
1680                 fbp = fblk->bp;
1681                 /*
1682                  * Remember initial freespace block number.
1683                  */
1684                 ifbno = fblk->blkno;
1685                 free = fbp->b_addr;
1686                 findex = fblk->index;
1687                 bests = dp->d_ops->free_bests_p(free);
1688                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1689
1690                 /*
1691                  * This means the free entry showed that the data block had
1692                  * space for our entry, so we remembered it.
1693                  * Use that data block.
1694                  */
1695                 if (findex >= 0) {
1696                         ASSERT(findex < freehdr.nvalid);
1697                         ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1698                         ASSERT(be16_to_cpu(bests[findex]) >= length);
1699                         dbno = freehdr.firstdb + findex;
1700                 } else {
1701                         /*
1702                          * The data block looked at didn't have enough room.
1703                          * We'll start at the beginning of the freespace entries.
1704                          */
1705                         dbno = -1;
1706                         findex = 0;
1707                 }
1708         } else {
1709                 /*
1710                  * Didn't come in with a freespace block, so no data block.
1711                  */
1712                 ifbno = dbno = -1;
1713                 fbp = NULL;
1714                 findex = 0;
1715         }
1716
1717         /*
1718          * If we don't have a data block yet, we're going to scan the
1719          * freespace blocks looking for one.  Figure out what the
1720          * highest freespace block number is.
1721          */
1722         if (dbno == -1) {
1723                 xfs_fileoff_t   fo;             /* freespace block number */
1724
1725                 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK)))
1726                         return error;
1727                 lastfbno = xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo);
1728                 fbno = ifbno;
1729         }
1730         /*
1731          * While we haven't identified a data block, search the freeblock
1732          * data for a good data block.  If we find a null freeblock entry,
1733          * indicating a hole in the data blocks, remember that.
1734          */
1735         while (dbno == -1) {
1736                 /*
1737                  * If we don't have a freeblock in hand, get the next one.
1738                  */
1739                 if (fbp == NULL) {
1740                         /*
1741                          * Happens the first time through unless lookup gave
1742                          * us a freespace block to start with.
1743                          */
1744                         if (++fbno == 0)
1745                                 fbno = xfs_dir2_byte_to_db(args->geo,
1746                                                         XFS_DIR2_FREE_OFFSET);
1747                         /*
1748                          * If it's ifbno we already looked at it.
1749                          */
1750                         if (fbno == ifbno)
1751                                 fbno++;
1752                         /*
1753                          * If it's off the end we're done.
1754                          */
1755                         if (fbno >= lastfbno)
1756                                 break;
1757                         /*
1758                          * Read the block.  There can be holes in the
1759                          * freespace blocks, so this might not succeed.
1760                          * This should be really rare, so there's no reason
1761                          * to avoid it.
1762                          */
1763                         error = xfs_dir2_free_try_read(tp, dp,
1764                                         xfs_dir2_db_to_da(args->geo, fbno),
1765                                         &fbp);
1766                         if (error)
1767                                 return error;
1768                         if (!fbp)
1769                                 continue;
1770                         free = fbp->b_addr;
1771                         findex = 0;
1772                 }
1773                 /*
1774                  * Look at the current free entry.  Is it good enough?
1775                  *
1776                  * The bests initialisation should be where the bufer is read in
1777                  * the above branch. But gcc is too stupid to realise that bests
1778                  * and the freehdr are actually initialised if they are placed
1779                  * there, so we have to do it here to avoid warnings. Blech.
1780                  */
1781                 bests = dp->d_ops->free_bests_p(free);
1782                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1783                 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1784                     be16_to_cpu(bests[findex]) >= length)
1785                         dbno = freehdr.firstdb + findex;
1786                 else {
1787                         /*
1788                          * Are we done with the freeblock?
1789                          */
1790                         if (++findex == freehdr.nvalid) {
1791                                 /*
1792                                  * Drop the block.
1793                                  */
1794                                 xfs_trans_brelse(tp, fbp);
1795                                 fbp = NULL;
1796                                 if (fblk && fblk->bp)
1797                                         fblk->bp = NULL;
1798                         }
1799                 }
1800         }
1801         /*
1802          * If we don't have a data block, we need to allocate one and make
1803          * the freespace entries refer to it.
1804          */
1805         if (unlikely(dbno == -1)) {
1806                 /*
1807                  * Not allowed to allocate, return failure.
1808                  */
1809                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1810                         return -ENOSPC;
1811
1812                 /*
1813                  * Allocate and initialize the new data block.
1814                  */
1815                 if (unlikely((error = xfs_dir2_grow_inode(args,
1816                                                          XFS_DIR2_DATA_SPACE,
1817                                                          &dbno)) ||
1818                     (error = xfs_dir3_data_init(args, dbno, &dbp))))
1819                         return error;
1820
1821                 /*
1822                  * If (somehow) we have a freespace block, get rid of it.
1823                  */
1824                 if (fbp)
1825                         xfs_trans_brelse(tp, fbp);
1826                 if (fblk && fblk->bp)
1827                         fblk->bp = NULL;
1828
1829                 /*
1830                  * Get the freespace block corresponding to the data block
1831                  * that was just allocated.
1832                  */
1833                 fbno = dp->d_ops->db_to_fdb(args->geo, dbno);
1834                 error = xfs_dir2_free_try_read(tp, dp,
1835                                        xfs_dir2_db_to_da(args->geo, fbno),
1836                                        &fbp);
1837                 if (error)
1838                         return error;
1839
1840                 /*
1841                  * If there wasn't a freespace block, the read will
1842                  * return a NULL fbp.  Allocate and initialize a new one.
1843                  */
1844                 if (!fbp) {
1845                         error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1846                                                     &fbno);
1847                         if (error)
1848                                 return error;
1849
1850                         if (dp->d_ops->db_to_fdb(args->geo, dbno) != fbno) {
1851                                 xfs_alert(mp,
1852 "%s: dir ino %llu needed freesp block %lld for data block %lld, got %lld ifbno %llu lastfbno %d",
1853                                         __func__, (unsigned long long)dp->i_ino,
1854                                         (long long)dp->d_ops->db_to_fdb(
1855                                                                 args->geo, dbno),
1856                                         (long long)dbno, (long long)fbno,
1857                                         (unsigned long long)ifbno, lastfbno);
1858                                 if (fblk) {
1859                                         xfs_alert(mp,
1860                                 " fblk 0x%p blkno %llu index %d magic 0x%x",
1861                                                 fblk,
1862                                                 (unsigned long long)fblk->blkno,
1863                                                 fblk->index,
1864                                                 fblk->magic);
1865                                 } else {
1866                                         xfs_alert(mp, " ... fblk is NULL");
1867                                 }
1868                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1869                                                  XFS_ERRLEVEL_LOW, mp);
1870                                 return -EFSCORRUPTED;
1871                         }
1872
1873                         /*
1874                          * Get a buffer for the new block.
1875                          */
1876                         error = xfs_dir3_free_get_buf(args, fbno, &fbp);
1877                         if (error)
1878                                 return error;
1879                         free = fbp->b_addr;
1880                         bests = dp->d_ops->free_bests_p(free);
1881                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1882
1883                         /*
1884                          * Remember the first slot as our empty slot.
1885                          */
1886                         freehdr.firstdb =
1887                                 (fbno - xfs_dir2_byte_to_db(args->geo,
1888                                                         XFS_DIR2_FREE_OFFSET)) *
1889                                         dp->d_ops->free_max_bests(args->geo);
1890                 } else {
1891                         free = fbp->b_addr;
1892                         bests = dp->d_ops->free_bests_p(free);
1893                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1894                 }
1895
1896                 /*
1897                  * Set the freespace block index from the data block number.
1898                  */
1899                 findex = dp->d_ops->db_to_fdindex(args->geo, dbno);
1900                 /*
1901                  * If it's after the end of the current entries in the
1902                  * freespace block, extend that table.
1903                  */
1904                 if (findex >= freehdr.nvalid) {
1905                         ASSERT(findex < dp->d_ops->free_max_bests(args->geo));
1906                         freehdr.nvalid = findex + 1;
1907                         /*
1908                          * Tag new entry so nused will go up.
1909                          */
1910                         bests[findex] = cpu_to_be16(NULLDATAOFF);
1911                 }
1912                 /*
1913                  * If this entry was for an empty data block
1914                  * (this should always be true) then update the header.
1915                  */
1916                 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1917                         freehdr.nused++;
1918                         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
1919                         xfs_dir2_free_log_header(args, fbp);
1920                 }
1921                 /*
1922                  * Update the real value in the table.
1923                  * We haven't allocated the data entry yet so this will
1924                  * change again.
1925                  */
1926                 hdr = dbp->b_addr;
1927                 bf = dp->d_ops->data_bestfree_p(hdr);
1928                 bests[findex] = bf[0].length;
1929                 logfree = 1;
1930         }
1931         /*
1932          * We had a data block so we don't have to make a new one.
1933          */
1934         else {
1935                 /*
1936                  * If just checking, we succeeded.
1937                  */
1938                 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1939                         return 0;
1940
1941                 /*
1942                  * Read the data block in.
1943                  */
1944                 error = xfs_dir3_data_read(tp, dp,
1945                                            xfs_dir2_db_to_da(args->geo, dbno),
1946                                            -1, &dbp);
1947                 if (error)
1948                         return error;
1949                 hdr = dbp->b_addr;
1950                 bf = dp->d_ops->data_bestfree_p(hdr);
1951                 logfree = 0;
1952         }
1953         ASSERT(be16_to_cpu(bf[0].length) >= length);
1954         /*
1955          * Point to the existing unused space.
1956          */
1957         dup = (xfs_dir2_data_unused_t *)
1958               ((char *)hdr + be16_to_cpu(bf[0].offset));
1959         needscan = needlog = 0;
1960         /*
1961          * Mark the first part of the unused space, inuse for us.
1962          */
1963         xfs_dir2_data_use_free(args, dbp, dup,
1964                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
1965                 &needlog, &needscan);
1966         /*
1967          * Fill in the new entry and log it.
1968          */
1969         dep = (xfs_dir2_data_entry_t *)dup;
1970         dep->inumber = cpu_to_be64(args->inumber);
1971         dep->namelen = args->namelen;
1972         memcpy(dep->name, args->name, dep->namelen);
1973         dp->d_ops->data_put_ftype(dep, args->filetype);
1974         tagp = dp->d_ops->data_entry_tag_p(dep);
1975         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1976         xfs_dir2_data_log_entry(args, dbp, dep);
1977         /*
1978          * Rescan the block for bestfree if needed.
1979          */
1980         if (needscan)
1981                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1982         /*
1983          * Log the data block header if needed.
1984          */
1985         if (needlog)
1986                 xfs_dir2_data_log_header(args, dbp);
1987         /*
1988          * If the freespace entry is now wrong, update it.
1989          */
1990         bests = dp->d_ops->free_bests_p(free); /* gcc is so stupid */
1991         if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
1992                 bests[findex] = bf[0].length;
1993                 logfree = 1;
1994         }
1995         /*
1996          * Log the freespace entry if needed.
1997          */
1998         if (logfree)
1999                 xfs_dir2_free_log_bests(args, fbp, findex, findex);
2000         /*
2001          * Return the data block and offset in args, then drop the data block.
2002          */
2003         args->blkno = (xfs_dablk_t)dbno;
2004         args->index = be16_to_cpu(*tagp);
2005         return 0;
2006 }
2007
2008 /*
2009  * Lookup an entry in a node-format directory.
2010  * All the real work happens in xfs_da3_node_lookup_int.
2011  * The only real output is the inode number of the entry.
2012  */
2013 int                                             /* error */
2014 xfs_dir2_node_lookup(
2015         xfs_da_args_t   *args)                  /* operation arguments */
2016 {
2017         int             error;                  /* error return value */
2018         int             i;                      /* btree level */
2019         int             rval;                   /* operation return value */
2020         xfs_da_state_t  *state;                 /* btree cursor */
2021
2022         trace_xfs_dir2_node_lookup(args);
2023
2024         /*
2025          * Allocate and initialize the btree cursor.
2026          */
2027         state = xfs_da_state_alloc();
2028         state->args = args;
2029         state->mp = args->dp->i_mount;
2030         /*
2031          * Fill in the path to the entry in the cursor.
2032          */
2033         error = xfs_da3_node_lookup_int(state, &rval);
2034         if (error)
2035                 rval = error;
2036         else if (rval == -ENOENT && args->cmpresult == XFS_CMP_CASE) {
2037                 /* If a CI match, dup the actual name and return -EEXIST */
2038                 xfs_dir2_data_entry_t   *dep;
2039
2040                 dep = (xfs_dir2_data_entry_t *)
2041                         ((char *)state->extrablk.bp->b_addr +
2042                                                  state->extrablk.index);
2043                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2044         }
2045         /*
2046          * Release the btree blocks and leaf block.
2047          */
2048         for (i = 0; i < state->path.active; i++) {
2049                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2050                 state->path.blk[i].bp = NULL;
2051         }
2052         /*
2053          * Release the data block if we have it.
2054          */
2055         if (state->extravalid && state->extrablk.bp) {
2056                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2057                 state->extrablk.bp = NULL;
2058         }
2059         xfs_da_state_free(state);
2060         return rval;
2061 }
2062
2063 /*
2064  * Remove an entry from a node-format directory.
2065  */
2066 int                                             /* error */
2067 xfs_dir2_node_removename(
2068         struct xfs_da_args      *args)          /* operation arguments */
2069 {
2070         struct xfs_da_state_blk *blk;           /* leaf block */
2071         int                     error;          /* error return value */
2072         int                     rval;           /* operation return value */
2073         struct xfs_da_state     *state;         /* btree cursor */
2074
2075         trace_xfs_dir2_node_removename(args);
2076
2077         /*
2078          * Allocate and initialize the btree cursor.
2079          */
2080         state = xfs_da_state_alloc();
2081         state->args = args;
2082         state->mp = args->dp->i_mount;
2083
2084         /* Look up the entry we're deleting, set up the cursor. */
2085         error = xfs_da3_node_lookup_int(state, &rval);
2086         if (error)
2087                 goto out_free;
2088
2089         /* Didn't find it, upper layer screwed up. */
2090         if (rval != -EEXIST) {
2091                 error = rval;
2092                 goto out_free;
2093         }
2094
2095         blk = &state->path.blk[state->path.active - 1];
2096         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2097         ASSERT(state->extravalid);
2098         /*
2099          * Remove the leaf and data entries.
2100          * Extrablk refers to the data block.
2101          */
2102         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2103                 &state->extrablk, &rval);
2104         if (error)
2105                 goto out_free;
2106         /*
2107          * Fix the hash values up the btree.
2108          */
2109         xfs_da3_fixhashpath(state, &state->path);
2110         /*
2111          * If we need to join leaf blocks, do it.
2112          */
2113         if (rval && state->path.active > 1)
2114                 error = xfs_da3_join(state);
2115         /*
2116          * If no errors so far, try conversion to leaf format.
2117          */
2118         if (!error)
2119                 error = xfs_dir2_node_to_leaf(state);
2120 out_free:
2121         xfs_da_state_free(state);
2122         return error;
2123 }
2124
2125 /*
2126  * Replace an entry's inode number in a node-format directory.
2127  */
2128 int                                             /* error */
2129 xfs_dir2_node_replace(
2130         xfs_da_args_t           *args)          /* operation arguments */
2131 {
2132         xfs_da_state_blk_t      *blk;           /* leaf block */
2133         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
2134         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
2135         int                     error;          /* error return value */
2136         int                     i;              /* btree level */
2137         xfs_ino_t               inum;           /* new inode number */
2138         int                     ftype;          /* new file type */
2139         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
2140         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
2141         int                     rval;           /* internal return value */
2142         xfs_da_state_t          *state;         /* btree cursor */
2143
2144         trace_xfs_dir2_node_replace(args);
2145
2146         /*
2147          * Allocate and initialize the btree cursor.
2148          */
2149         state = xfs_da_state_alloc();
2150         state->args = args;
2151         state->mp = args->dp->i_mount;
2152
2153         /*
2154          * We have to save new inode number and ftype since
2155          * xfs_da3_node_lookup_int() is going to overwrite them
2156          */
2157         inum = args->inumber;
2158         ftype = args->filetype;
2159
2160         /*
2161          * Lookup the entry to change in the btree.
2162          */
2163         error = xfs_da3_node_lookup_int(state, &rval);
2164         if (error) {
2165                 rval = error;
2166         }
2167         /*
2168          * It should be found, since the vnodeops layer has looked it up
2169          * and locked it.  But paranoia is good.
2170          */
2171         if (rval == -EEXIST) {
2172                 struct xfs_dir2_leaf_entry *ents;
2173                 /*
2174                  * Find the leaf entry.
2175                  */
2176                 blk = &state->path.blk[state->path.active - 1];
2177                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2178                 leaf = blk->bp->b_addr;
2179                 ents = args->dp->d_ops->leaf_ents_p(leaf);
2180                 lep = &ents[blk->index];
2181                 ASSERT(state->extravalid);
2182                 /*
2183                  * Point to the data entry.
2184                  */
2185                 hdr = state->extrablk.bp->b_addr;
2186                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2187                        hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2188                 dep = (xfs_dir2_data_entry_t *)
2189                       ((char *)hdr +
2190                        xfs_dir2_dataptr_to_off(args->geo,
2191                                                be32_to_cpu(lep->address)));
2192                 ASSERT(inum != be64_to_cpu(dep->inumber));
2193                 /*
2194                  * Fill in the new inode number and log the entry.
2195                  */
2196                 dep->inumber = cpu_to_be64(inum);
2197                 args->dp->d_ops->data_put_ftype(dep, ftype);
2198                 xfs_dir2_data_log_entry(args, state->extrablk.bp, dep);
2199                 rval = 0;
2200         }
2201         /*
2202          * Didn't find it, and we're holding a data block.  Drop it.
2203          */
2204         else if (state->extravalid) {
2205                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2206                 state->extrablk.bp = NULL;
2207         }
2208         /*
2209          * Release all the buffers in the cursor.
2210          */
2211         for (i = 0; i < state->path.active; i++) {
2212                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2213                 state->path.blk[i].bp = NULL;
2214         }
2215         xfs_da_state_free(state);
2216         return rval;
2217 }
2218
2219 /*
2220  * Trim off a trailing empty freespace block.
2221  * Return (in rvalp) 1 if we did it, 0 if not.
2222  */
2223 int                                             /* error */
2224 xfs_dir2_node_trim_free(
2225         xfs_da_args_t           *args,          /* operation arguments */
2226         xfs_fileoff_t           fo,             /* free block number */
2227         int                     *rvalp)         /* out: did something */
2228 {
2229         struct xfs_buf          *bp;            /* freespace buffer */
2230         xfs_inode_t             *dp;            /* incore directory inode */
2231         int                     error;          /* error return code */
2232         xfs_dir2_free_t         *free;          /* freespace structure */
2233         xfs_trans_t             *tp;            /* transaction pointer */
2234         struct xfs_dir3_icfree_hdr freehdr;
2235
2236         dp = args->dp;
2237         tp = args->trans;
2238         /*
2239          * Read the freespace block.
2240          */
2241         error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2242         if (error)
2243                 return error;
2244         /*
2245          * There can be holes in freespace.  If fo is a hole, there's
2246          * nothing to do.
2247          */
2248         if (!bp)
2249                 return 0;
2250         free = bp->b_addr;
2251         dp->d_ops->free_hdr_from_disk(&freehdr, free);
2252
2253         /*
2254          * If there are used entries, there's nothing to do.
2255          */
2256         if (freehdr.nused > 0) {
2257                 xfs_trans_brelse(tp, bp);
2258                 *rvalp = 0;
2259                 return 0;
2260         }
2261         /*
2262          * Blow the block away.
2263          */
2264         error = xfs_dir2_shrink_inode(args,
2265                         xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo), bp);
2266         if (error) {
2267                 /*
2268                  * Can't fail with ENOSPC since that only happens with no
2269                  * space reservation, when breaking up an extent into two
2270                  * pieces.  This is the last block of an extent.
2271                  */
2272                 ASSERT(error != -ENOSPC);
2273                 xfs_trans_brelse(tp, bp);
2274                 return error;
2275         }
2276         /*
2277          * Return that we succeeded.
2278          */
2279         *rvalp = 1;
2280         return 0;
2281 }