fs/xfs/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_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_inode.h"
  30 #include "xfs_bmap.h"
  31 #include "xfs_dir2.h"
  32 #include "xfs_dir2_priv.h"
  33 #include "xfs_error.h"
  34 #include "xfs_trace.h"
  35 #include "xfs_trans.h"
  36 #include "xfs_buf_item.h"
  37 #include "xfs_cksum.h"
  38
  39 /*
  40  * Function declarations.
  41  */
  42 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
  43                               int index);
  44 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  45                                      xfs_da_state_blk_t *blk1,
  46                                      xfs_da_state_blk_t *blk2);
  47 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
  48                                  int index, xfs_da_state_blk_t *dblk,
  49                                  int *rval);
  50 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  51                                      xfs_da_state_blk_t *fblk);
  52
  53 /*
  54  * Check internal consistency of a leafn block.
  55  */
  56 #ifdef DEBUG
  57 #define xfs_dir3_leaf_check(dp, bp) \
  58 do { \
  59         if (!xfs_dir3_leafn_check((dp), (bp))) \
  60                 ASSERT(0); \
  61 } while (0);
  62
  63 static bool
  64 xfs_dir3_leafn_check(
  65         struct xfs_inode        *dp,
  66         struct xfs_buf          *bp)
  67 {
  68         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  69         struct xfs_dir3_icleaf_hdr leafhdr;
  70
  71         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
  72
  73         if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
  74                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  75                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  76                         return false;
  77         } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
  78                 return false;
  79
  80         return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
  81 }
  82 #else
  83 #define xfs_dir3_leaf_check(dp, bp)
  84 #endif
  85
  86 static bool
  87 xfs_dir3_free_verify(
  88         struct xfs_buf          *bp)
  89 {
  90         struct xfs_mount        *mp = bp->b_target->bt_mount;
  91         struct xfs_dir2_free_hdr *hdr = bp->b_addr;
  92
  93         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  94                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
  95
  96                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_FREE_MAGIC))
  97                         return false;
  98                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
  99                         return false;
 100                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 101                         return false;
 102         } else {
 103                 if (hdr->magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC))
 104                         return false;
 105         }
 106
 107         /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
 108
 109         return true;
 110 }
 111
 112 static void
 113 xfs_dir3_free_read_verify(
 114         struct xfs_buf  *bp)
 115 {
 116         struct xfs_mount        *mp = bp->b_target->bt_mount;
 117
 118         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 119             !xfs_buf_verify_cksum(bp, XFS_DIR3_FREE_CRC_OFF))
 120                 xfs_buf_ioerror(bp, EFSBADCRC);
 121         else if (!xfs_dir3_free_verify(bp))
 122                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 123
 124         if (bp->b_error)
 125                 xfs_verifier_error(bp);
 126 }
 127
 128 static void
 129 xfs_dir3_free_write_verify(
 130         struct xfs_buf  *bp)
 131 {
 132         struct xfs_mount        *mp = bp->b_target->bt_mount;
 133         struct xfs_buf_log_item *bip = bp->b_fspriv;
 134         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 135
 136         if (!xfs_dir3_free_verify(bp)) {
 137                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 138                 xfs_verifier_error(bp);
 139                 return;
 140         }
 141
 142         if (!xfs_sb_version_hascrc(&mp->m_sb))
 143                 return;
 144
 145         if (bip)
 146                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 147
 148         xfs_buf_update_cksum(bp, XFS_DIR3_FREE_CRC_OFF);
 149 }
 150
 151 const struct xfs_buf_ops xfs_dir3_free_buf_ops = {
 152         .verify_read = xfs_dir3_free_read_verify,
 153         .verify_write = xfs_dir3_free_write_verify,
 154 };
 155
 156
 157 static int
 158 __xfs_dir3_free_read(
 159         struct xfs_trans        *tp,
 160         struct xfs_inode        *dp,
 161         xfs_dablk_t             fbno,
 162         xfs_daddr_t             mappedbno,
 163         struct xfs_buf          **bpp)
 164 {
 165         int                     err;
 166
 167         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 168                                 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
 169
 170         /* try read returns without an error or *bpp if it lands in a hole */
 171         if (!err && tp && *bpp)
 172                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
 173         return err;
 174 }
 175
 176 int
 177 xfs_dir2_free_read(
 178         struct xfs_trans        *tp,
 179         struct xfs_inode        *dp,
 180         xfs_dablk_t             fbno,
 181         struct xfs_buf          **bpp)
 182 {
 183         return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
 184 }
 185
 186 static int
 187 xfs_dir2_free_try_read(
 188         struct xfs_trans        *tp,
 189         struct xfs_inode        *dp,
 190         xfs_dablk_t             fbno,
 191         struct xfs_buf          **bpp)
 192 {
 193         return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
 194 }
 195
 196 static int
 197 xfs_dir3_free_get_buf(
 198         xfs_da_args_t           *args,
 199         xfs_dir2_db_t           fbno,
 200         struct xfs_buf          **bpp)
 201 {
 202         struct xfs_trans        *tp = args->trans;
 203         struct xfs_inode        *dp = args->dp;
 204         struct xfs_mount        *mp = dp->i_mount;
 205         struct xfs_buf          *bp;
 206         int                     error;
 207         struct xfs_dir3_icfree_hdr hdr;
 208
 209         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, fbno),
 210                                    -1, &bp, XFS_DATA_FORK);
 211         if (error)
 212                 return error;
 213
 214         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
 215         bp->b_ops = &xfs_dir3_free_buf_ops;
 216
 217         /*
 218          * Initialize the new block to be empty, and remember
 219          * its first slot as our empty slot.
 220          */
 221         memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
 222         memset(&hdr, 0, sizeof(hdr));
 223
 224         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 225                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 226
 227                 hdr.magic = XFS_DIR3_FREE_MAGIC;
 228
 229                 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
 230                 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
 231                 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_uuid);
 232         } else
 233                 hdr.magic = XFS_DIR2_FREE_MAGIC;
 234         dp->d_ops->free_hdr_to_disk(bp->b_addr, &hdr);
 235         *bpp = bp;
 236         return 0;
 237 }
 238
 239 /*
 240  * Log entries from a freespace block.
 241  */
 242 STATIC void
 243 xfs_dir2_free_log_bests(
 244         struct xfs_trans        *tp,
 245         struct xfs_inode        *dp,
 246         struct xfs_buf          *bp,
 247         int                     first,          /* first entry to log */
 248         int                     last)           /* last entry to log */
 249 {
 250         xfs_dir2_free_t         *free;          /* freespace structure */
 251         __be16                  *bests;
 252
 253         free = bp->b_addr;
 254         bests = dp->d_ops->free_bests_p(free);
 255         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 256                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 257         xfs_trans_log_buf(tp, bp,
 258                 (uint)((char *)&bests[first] - (char *)free),
 259                 (uint)((char *)&bests[last] - (char *)free +
 260                        sizeof(bests[0]) - 1));
 261 }
 262
 263 /*
 264  * Log header from a freespace block.
 265  */
 266 static void
 267 xfs_dir2_free_log_header(
 268         struct xfs_trans        *tp,
 269         struct xfs_inode        *dp,
 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(tp, bp, 0, dp->d_ops->free_hdr_size - 1);
 280 }
 281
 282 /*
 283  * Convert a leaf-format directory to a node-format directory.
 284  * We need to change the magic number of the leaf block, and copy
 285  * the freespace table out of the leaf block into its own block.
 286  */
 287 int                                             /* error */
 288 xfs_dir2_leaf_to_node(
 289         xfs_da_args_t           *args,          /* operation arguments */
 290         struct xfs_buf          *lbp)           /* leaf buffer */
 291 {
 292         xfs_inode_t             *dp;            /* incore directory inode */
 293         int                     error;          /* error return value */
 294         struct xfs_buf          *fbp;           /* freespace buffer */
 295         xfs_dir2_db_t           fdb;            /* freespace block number */
 296         xfs_dir2_free_t         *free;          /* freespace structure */
 297         __be16                  *from;          /* pointer to freespace entry */
 298         int                     i;              /* leaf freespace index */
 299         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 300         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 301         xfs_mount_t             *mp;            /* filesystem mount point */
 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         mp = dp->i_mount;
 312         tp = args->trans;
 313         /*
 314          * Add a freespace block to the directory.
 315          */
 316         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 317                 return error;
 318         }
 319         ASSERT(fdb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
 320         /*
 321          * Get the buffer for the new freespace block.
 322          */
 323         error = xfs_dir3_free_get_buf(args, fdb, &fbp);
 324         if (error)
 325                 return error;
 326
 327         free = fbp->b_addr;
 328         dp->d_ops->free_hdr_from_disk(&freehdr, free);
 329         leaf = lbp->b_addr;
 330         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 331         ASSERT(be32_to_cpu(ltp->bestcount) <=
 332                                 (uint)dp->i_d.di_size / args->geo->blksize);
 333
 334         /*
 335          * Copy freespace entries from the leaf block to the new block.
 336          * Count active entries.
 337          */
 338         from = xfs_dir2_leaf_bests_p(ltp);
 339         to = dp->d_ops->free_bests_p(free);
 340         for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 341                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 342                         n++;
 343                 *to = cpu_to_be16(off);
 344         }
 345
 346         /*
 347          * Now initialize the freespace block header.
 348          */
 349         freehdr.nused = n;
 350         freehdr.nvalid = be32_to_cpu(ltp->bestcount);
 351
 352         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
 353         xfs_dir2_free_log_bests(tp, dp, fbp, 0, freehdr.nvalid - 1);
 354         xfs_dir2_free_log_header(tp, dp, fbp);
 355
 356         /*
 357          * Converting the leaf to a leafnode is just a matter of changing the
 358          * magic number and the ops. Do the change directly to the buffer as
 359          * it's less work (and less code) than decoding the header to host
 360          * format and back again.
 361          */
 362         if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
 363                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 364         else
 365                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 366         lbp->b_ops = &xfs_dir3_leafn_buf_ops;
 367         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
 368         xfs_dir3_leaf_log_header(tp, dp, lbp);
 369         xfs_dir3_leaf_check(dp, lbp);
 370         return 0;
 371 }
 372
 373 /*
 374  * Add a leaf entry to a leaf block in a node-form directory.
 375  * The other work necessary is done from the caller.
 376  */
 377 static int                                      /* error */
 378 xfs_dir2_leafn_add(
 379         struct xfs_buf          *bp,            /* leaf buffer */
 380         xfs_da_args_t           *args,          /* operation arguments */
 381         int                     index)          /* insertion pt for new entry */
 382 {
 383         int                     compact;        /* compacting stale leaves */
 384         xfs_inode_t             *dp;            /* incore directory inode */
 385         int                     highstale;      /* next stale entry */
 386         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 387         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 388         int                     lfloghigh;      /* high leaf entry logging */
 389         int                     lfloglow;       /* low leaf entry logging */
 390         int                     lowstale;       /* previous stale entry */
 391         xfs_mount_t             *mp;            /* filesystem mount point */
 392         xfs_trans_t             *tp;            /* transaction pointer */
 393         struct xfs_dir3_icleaf_hdr leafhdr;
 394         struct xfs_dir2_leaf_entry *ents;
 395
 396         trace_xfs_dir2_leafn_add(args, index);
 397
 398         dp = args->dp;
 399         mp = dp->i_mount;
 400         tp = args->trans;
 401         leaf = bp->b_addr;
 402         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 403         ents = dp->d_ops->leaf_ents_p(leaf);
 404
 405         /*
 406          * Quick check just to make sure we are not going to index
 407          * into other peoples memory
 408          */
 409         if (index < 0)
 410                 return XFS_ERROR(EFSCORRUPTED);
 411
 412         /*
 413          * If there are already the maximum number of leaf entries in
 414          * the block, if there are no stale entries it won't fit.
 415          * Caller will do a split.  If there are stale entries we'll do
 416          * a compact.
 417          */
 418
 419         if (leafhdr.count == dp->d_ops->leaf_max_ents(args->geo)) {
 420                 if (!leafhdr.stale)
 421                         return XFS_ERROR(ENOSPC);
 422                 compact = leafhdr.stale > 1;
 423         } else
 424                 compact = 0;
 425         ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
 426         ASSERT(index == leafhdr.count ||
 427                be32_to_cpu(ents[index].hashval) >= args->hashval);
 428
 429         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 430                 return 0;
 431
 432         /*
 433          * Compact out all but one stale leaf entry.  Leaves behind
 434          * the entry closest to index.
 435          */
 436         if (compact)
 437                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 438                                          &highstale, &lfloglow, &lfloghigh);
 439         else if (leafhdr.stale) {
 440                 /*
 441                  * Set impossible logging indices for this case.
 442                  */
 443                 lfloglow = leafhdr.count;
 444                 lfloghigh = -1;
 445         }
 446
 447         /*
 448          * Insert the new entry, log everything.
 449          */
 450         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 451                                        highstale, &lfloglow, &lfloghigh);
 452
 453         lep->hashval = cpu_to_be32(args->hashval);
 454         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(args->geo,
 455                                 args->blkno, args->index));
 456
 457         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
 458         xfs_dir3_leaf_log_header(tp, dp, bp);
 459         xfs_dir3_leaf_log_ents(tp, dp, bp, lfloglow, lfloghigh);
 460         xfs_dir3_leaf_check(dp, bp);
 461         return 0;
 462 }
 463
 464 #ifdef DEBUG
 465 static void
 466 xfs_dir2_free_hdr_check(
 467         struct xfs_inode *dp,
 468         struct xfs_buf  *bp,
 469         xfs_dir2_db_t   db)
 470 {
 471         struct xfs_dir3_icfree_hdr hdr;
 472
 473         dp->d_ops->free_hdr_from_disk(&hdr, bp->b_addr);
 474
 475         ASSERT((hdr.firstdb %
 476                 dp->d_ops->free_max_bests(dp->i_mount->m_dir_geo)) == 0);
 477         ASSERT(hdr.firstdb <= db);
 478         ASSERT(db < hdr.firstdb + hdr.nvalid);
 479 }
 480 #else
 481 #define xfs_dir2_free_hdr_check(dp, bp, db)
 482 #endif  /* DEBUG */
 483
 484 /*
 485  * Return the last hash value in the leaf.
 486  * Stale entries are ok.
 487  */
 488 xfs_dahash_t                                    /* hash value */
 489 xfs_dir2_leafn_lasthash(
 490         struct xfs_inode *dp,
 491         struct xfs_buf  *bp,                    /* leaf buffer */
 492         int             *count)                 /* count of entries in leaf */
 493 {
 494         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 495         struct xfs_dir2_leaf_entry *ents;
 496         struct xfs_dir3_icleaf_hdr leafhdr;
 497
 498         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 499
 500         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 501                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
 502
 503         if (count)
 504                 *count = leafhdr.count;
 505         if (!leafhdr.count)
 506                 return 0;
 507
 508         ents = dp->d_ops->leaf_ents_p(leaf);
 509         return be32_to_cpu(ents[leafhdr.count - 1].hashval);
 510 }
 511
 512 /*
 513  * Look up a leaf entry for space to add a name in a node-format leaf block.
 514  * The extrablk in state is a freespace block.
 515  */
 516 STATIC int
 517 xfs_dir2_leafn_lookup_for_addname(
 518         struct xfs_buf          *bp,            /* leaf buffer */
 519         xfs_da_args_t           *args,          /* operation arguments */
 520         int                     *indexp,        /* out: leaf entry index */
 521         xfs_da_state_t          *state)         /* state to fill in */
 522 {
 523         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 524         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 525         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 526         xfs_inode_t             *dp;            /* incore directory inode */
 527         int                     error;          /* error return value */
 528         int                     fi;             /* free entry index */
 529         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 530         int                     index;          /* leaf entry index */
 531         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 532         int                     length;         /* length of new data entry */
 533         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 534         xfs_mount_t             *mp;            /* filesystem mount point */
 535         xfs_dir2_db_t           newdb;          /* new data block number */
 536         xfs_dir2_db_t           newfdb;         /* new free block number */
 537         xfs_trans_t             *tp;            /* transaction pointer */
 538         struct xfs_dir2_leaf_entry *ents;
 539         struct xfs_dir3_icleaf_hdr leafhdr;
 540
 541         dp = args->dp;
 542         tp = args->trans;
 543         mp = dp->i_mount;
 544         leaf = bp->b_addr;
 545         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 546         ents = dp->d_ops->leaf_ents_p(leaf);
 547
 548         xfs_dir3_leaf_check(dp, bp);
 549         ASSERT(leafhdr.count > 0);
 550
 551         /*
 552          * Look up the hash value in the leaf entries.
 553          */
 554         index = xfs_dir2_leaf_search_hash(args, bp);
 555         /*
 556          * Do we have a buffer coming in?
 557          */
 558         if (state->extravalid) {
 559                 /* If so, it's a free block buffer, get the block number. */
 560                 curbp = state->extrablk.bp;
 561                 curfdb = state->extrablk.blkno;
 562                 free = curbp->b_addr;
 563                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 564                        free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 565         }
 566         length = dp->d_ops->data_entsize(args->namelen);
 567         /*
 568          * Loop over leaf entries with the right hash value.
 569          */
 570         for (lep = &ents[index];
 571              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 572              lep++, index++) {
 573                 /*
 574                  * Skip stale leaf entries.
 575                  */
 576                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 577                         continue;
 578                 /*
 579                  * Pull the data block number from the entry.
 580                  */
 581                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 582                                                be32_to_cpu(lep->address));
 583                 /*
 584                  * For addname, we're looking for a place to put the new entry.
 585                  * We want to use a data block with an entry of equal
 586                  * hash value to ours if there is one with room.
 587                  *
 588                  * If this block isn't the data block we already have
 589                  * in hand, take a look at it.
 590                  */
 591                 if (newdb != curdb) {
 592                         __be16 *bests;
 593
 594                         curdb = newdb;
 595                         /*
 596                          * Convert the data block to the free block
 597                          * holding its freespace information.
 598                          */
 599                         newfdb = dp->d_ops->db_to_fdb(args->geo, newdb);
 600                         /*
 601                          * If it's not the one we have in hand, read it in.
 602                          */
 603                         if (newfdb != curfdb) {
 604                                 /*
 605                                  * If we had one before, drop it.
 606                                  */
 607                                 if (curbp)
 608                                         xfs_trans_brelse(tp, curbp);
 609
 610                                 error = xfs_dir2_free_read(tp, dp,
 611                                                 xfs_dir2_db_to_da(args->geo,
 612                                                                   newfdb),
 613                                                 &curbp);
 614                                 if (error)
 615                                         return error;
 616                                 free = curbp->b_addr;
 617
 618                                 xfs_dir2_free_hdr_check(dp, curbp, curdb);
 619                         }
 620                         /*
 621                          * Get the index for our entry.
 622                          */
 623                         fi = dp->d_ops->db_to_fdindex(args->geo, curdb);
 624                         /*
 625                          * If it has room, return it.
 626                          */
 627                         bests = dp->d_ops->free_bests_p(free);
 628                         if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
 629                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 630                                                         XFS_ERRLEVEL_LOW, mp);
 631                                 if (curfdb != newfdb)
 632                                         xfs_trans_brelse(tp, curbp);
 633                                 return XFS_ERROR(EFSCORRUPTED);
 634                         }
 635                         curfdb = newfdb;
 636                         if (be16_to_cpu(bests[fi]) >= length)
 637                                 goto out;
 638                 }
 639         }
 640         /* Didn't find any space */
 641         fi = -1;
 642 out:
 643         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 644         if (curbp) {
 645                 /* Giving back a free block. */
 646                 state->extravalid = 1;
 647                 state->extrablk.bp = curbp;
 648                 state->extrablk.index = fi;
 649                 state->extrablk.blkno = curfdb;
 650
 651                 /*
 652                  * Important: this magic number is not in the buffer - it's for
 653                  * buffer type information and therefore only the free/data type
 654                  * matters here, not whether CRCs are enabled or not.
 655                  */
 656                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 657         } else {
 658                 state->extravalid = 0;
 659         }
 660         /*
 661          * Return the index, that will be the insertion point.
 662          */
 663         *indexp = index;
 664         return XFS_ERROR(ENOENT);
 665 }
 666
 667 /*
 668  * Look up a leaf entry in a node-format leaf block.
 669  * The extrablk in state a data block.
 670  */
 671 STATIC int
 672 xfs_dir2_leafn_lookup_for_entry(
 673         struct xfs_buf          *bp,            /* leaf buffer */
 674         xfs_da_args_t           *args,          /* operation arguments */
 675         int                     *indexp,        /* out: leaf entry index */
 676         xfs_da_state_t          *state)         /* state to fill in */
 677 {
 678         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 679         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 680         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 681         xfs_inode_t             *dp;            /* incore directory inode */
 682         int                     error;          /* error return value */
 683         int                     index;          /* leaf entry index */
 684         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 685         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 686         xfs_mount_t             *mp;            /* filesystem mount point */
 687         xfs_dir2_db_t           newdb;          /* new data block number */
 688         xfs_trans_t             *tp;            /* transaction pointer */
 689         enum xfs_dacmp          cmp;            /* comparison result */
 690         struct xfs_dir2_leaf_entry *ents;
 691         struct xfs_dir3_icleaf_hdr leafhdr;
 692
 693         dp = args->dp;
 694         tp = args->trans;
 695         mp = dp->i_mount;
 696         leaf = bp->b_addr;
 697         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 698         ents = dp->d_ops->leaf_ents_p(leaf);
 699
 700         xfs_dir3_leaf_check(dp, bp);
 701         ASSERT(leafhdr.count > 0);
 702
 703         /*
 704          * Look up the hash value in the leaf entries.
 705          */
 706         index = xfs_dir2_leaf_search_hash(args, bp);
 707         /*
 708          * Do we have a buffer coming in?
 709          */
 710         if (state->extravalid) {
 711                 curbp = state->extrablk.bp;
 712                 curdb = state->extrablk.blkno;
 713         }
 714         /*
 715          * Loop over leaf entries with the right hash value.
 716          */
 717         for (lep = &ents[index];
 718              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 719              lep++, index++) {
 720                 /*
 721                  * Skip stale leaf entries.
 722                  */
 723                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 724                         continue;
 725                 /*
 726                  * Pull the data block number from the entry.
 727                  */
 728                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 729                                                be32_to_cpu(lep->address));
 730                 /*
 731                  * Not adding a new entry, so we really want to find
 732                  * the name given to us.
 733                  *
 734                  * If it's a different data block, go get it.
 735                  */
 736                 if (newdb != curdb) {
 737                         /*
 738                          * If we had a block before that we aren't saving
 739                          * for a CI name, drop it
 740                          */
 741                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 742                                                 curdb != state->extrablk.blkno))
 743                                 xfs_trans_brelse(tp, curbp);
 744                         /*
 745                          * If needing the block that is saved with a CI match,
 746                          * use it otherwise read in the new data block.
 747                          */
 748                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 749                                         newdb == state->extrablk.blkno) {
 750                                 ASSERT(state->extravalid);
 751                                 curbp = state->extrablk.bp;
 752                         } else {
 753                                 error = xfs_dir3_data_read(tp, dp,
 754                                                 xfs_dir2_db_to_da(args->geo,
 755                                                                   newdb),
 756                                                 -1, &curbp);
 757                                 if (error)
 758                                         return error;
 759                         }
 760                         xfs_dir3_data_check(dp, curbp);
 761                         curdb = newdb;
 762                 }
 763                 /*
 764                  * Point to the data entry.
 765                  */
 766                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 767                         xfs_dir2_dataptr_to_off(args->geo,
 768                                                 be32_to_cpu(lep->address)));
 769                 /*
 770                  * Compare the entry and if it's an exact match, return
 771                  * EEXIST immediately. If it's the first case-insensitive
 772                  * match, store the block & inode number and continue looking.
 773                  */
 774                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 775                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 776                         /* If there is a CI match block, drop it */
 777                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 778                                                 curdb != state->extrablk.blkno)
 779                                 xfs_trans_brelse(tp, state->extrablk.bp);
 780                         args->cmpresult = cmp;
 781                         args->inumber = be64_to_cpu(dep->inumber);
 782                         args->filetype = dp->d_ops->data_get_ftype(dep);
 783                         *indexp = index;
 784                         state->extravalid = 1;
 785                         state->extrablk.bp = curbp;
 786                         state->extrablk.blkno = curdb;
 787                         state->extrablk.index = (int)((char *)dep -
 788                                                         (char *)curbp->b_addr);
 789                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 790                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 791                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 792                         if (cmp == XFS_CMP_EXACT)
 793                                 return XFS_ERROR(EEXIST);
 794                 }
 795         }
 796         ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
 797         if (curbp) {
 798                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 799                         /* Giving back last used data block. */
 800                         state->extravalid = 1;
 801                         state->extrablk.bp = curbp;
 802                         state->extrablk.index = -1;
 803                         state->extrablk.blkno = curdb;
 804                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 805                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 806                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 807                 } else {
 808                         /* If the curbp is not the CI match block, drop it */
 809                         if (state->extrablk.bp != curbp)
 810                                 xfs_trans_brelse(tp, curbp);
 811                 }
 812         } else {
 813                 state->extravalid = 0;
 814         }
 815         *indexp = index;
 816         return XFS_ERROR(ENOENT);
 817 }
 818
 819 /*
 820  * Look up a leaf entry in a node-format leaf block.
 821  * If this is an addname then the extrablk in state is a freespace block,
 822  * otherwise it's a data block.
 823  */
 824 int
 825 xfs_dir2_leafn_lookup_int(
 826         struct xfs_buf          *bp,            /* leaf buffer */
 827         xfs_da_args_t           *args,          /* operation arguments */
 828         int                     *indexp,        /* out: leaf entry index */
 829         xfs_da_state_t          *state)         /* state to fill in */
 830 {
 831         if (args->op_flags & XFS_DA_OP_ADDNAME)
 832                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 833                                                         state);
 834         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 835 }
 836
 837 /*
 838  * Move count leaf entries from source to destination leaf.
 839  * Log entries and headers.  Stale entries are preserved.
 840  */
 841 static void
 842 xfs_dir3_leafn_moveents(
 843         xfs_da_args_t                   *args,  /* operation arguments */
 844         struct xfs_buf                  *bp_s,  /* source */
 845         struct xfs_dir3_icleaf_hdr      *shdr,
 846         struct xfs_dir2_leaf_entry      *sents,
 847         int                             start_s,/* source leaf index */
 848         struct xfs_buf                  *bp_d,  /* destination */
 849         struct xfs_dir3_icleaf_hdr      *dhdr,
 850         struct xfs_dir2_leaf_entry      *dents,
 851         int                             start_d,/* destination leaf index */
 852         int                             count)  /* count of leaves to copy */
 853 {
 854         struct xfs_trans                *tp = args->trans;
 855         int                             stale;  /* count stale leaves copied */
 856
 857         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 858
 859         /*
 860          * Silently return if nothing to do.
 861          */
 862         if (count == 0)
 863                 return;
 864
 865         /*
 866          * If the destination index is not the end of the current
 867          * destination leaf entries, open up a hole in the destination
 868          * to hold the new entries.
 869          */
 870         if (start_d < dhdr->count) {
 871                 memmove(&dents[start_d + count], &dents[start_d],
 872                         (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
 873                 xfs_dir3_leaf_log_ents(tp, args->dp, bp_d, start_d + count,
 874                                        count + dhdr->count - 1);
 875         }
 876         /*
 877          * If the source has stale leaves, count the ones in the copy range
 878          * so we can update the header correctly.
 879          */
 880         if (shdr->stale) {
 881                 int     i;                      /* temp leaf index */
 882
 883                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 884                         if (sents[i].address ==
 885                                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 886                                 stale++;
 887                 }
 888         } else
 889                 stale = 0;
 890         /*
 891          * Copy the leaf entries from source to destination.
 892          */
 893         memcpy(&dents[start_d], &sents[start_s],
 894                 count * sizeof(xfs_dir2_leaf_entry_t));
 895         xfs_dir3_leaf_log_ents(tp, args->dp, bp_d,
 896                                start_d, start_d + count - 1);
 897
 898         /*
 899          * If there are source entries after the ones we copied,
 900          * delete the ones we copied by sliding the next ones down.
 901          */
 902         if (start_s + count < shdr->count) {
 903                 memmove(&sents[start_s], &sents[start_s + count],
 904                         count * sizeof(xfs_dir2_leaf_entry_t));
 905                 xfs_dir3_leaf_log_ents(tp, args->dp, bp_s,
 906                                        start_s, start_s + count - 1);
 907         }
 908
 909         /*
 910          * Update the headers and log them.
 911          */
 912         shdr->count -= count;
 913         shdr->stale -= stale;
 914         dhdr->count += count;
 915         dhdr->stale += stale;
 916 }
 917
 918 /*
 919  * Determine the sort order of two leaf blocks.
 920  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 921  */
 922 int                                             /* sort order */
 923 xfs_dir2_leafn_order(
 924         struct xfs_inode        *dp,
 925         struct xfs_buf          *leaf1_bp,              /* leaf1 buffer */
 926         struct xfs_buf          *leaf2_bp)              /* leaf2 buffer */
 927 {
 928         struct xfs_dir2_leaf    *leaf1 = leaf1_bp->b_addr;
 929         struct xfs_dir2_leaf    *leaf2 = leaf2_bp->b_addr;
 930         struct xfs_dir2_leaf_entry *ents1;
 931         struct xfs_dir2_leaf_entry *ents2;
 932         struct xfs_dir3_icleaf_hdr hdr1;
 933         struct xfs_dir3_icleaf_hdr hdr2;
 934
 935         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
 936         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
 937         ents1 = dp->d_ops->leaf_ents_p(leaf1);
 938         ents2 = dp->d_ops->leaf_ents_p(leaf2);
 939
 940         if (hdr1.count > 0 && hdr2.count > 0 &&
 941             (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
 942              be32_to_cpu(ents2[hdr2.count - 1].hashval) <
 943                                 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
 944                 return 1;
 945         return 0;
 946 }
 947
 948 /*
 949  * Rebalance leaf entries between two leaf blocks.
 950  * This is actually only called when the second block is new,
 951  * though the code deals with the general case.
 952  * A new entry will be inserted in one of the blocks, and that
 953  * entry is taken into account when balancing.
 954  */
 955 static void
 956 xfs_dir2_leafn_rebalance(
 957         xfs_da_state_t          *state,         /* btree cursor */
 958         xfs_da_state_blk_t      *blk1,          /* first btree block */
 959         xfs_da_state_blk_t      *blk2)          /* second btree block */
 960 {
 961         xfs_da_args_t           *args;          /* operation arguments */
 962         int                     count;          /* count (& direction) leaves */
 963         int                     isleft;         /* new goes in left leaf */
 964         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 965         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 966         int                     mid;            /* midpoint leaf index */
 967 #if defined(DEBUG) || defined(XFS_WARN)
 968         int                     oldstale;       /* old count of stale leaves */
 969 #endif
 970         int                     oldsum;         /* old total leaf count */
 971         int                     swap;           /* swapped leaf blocks */
 972         struct xfs_dir2_leaf_entry *ents1;
 973         struct xfs_dir2_leaf_entry *ents2;
 974         struct xfs_dir3_icleaf_hdr hdr1;
 975         struct xfs_dir3_icleaf_hdr hdr2;
 976         struct xfs_inode        *dp = state->args->dp;
 977
 978         args = state->args;
 979         /*
 980          * If the block order is wrong, swap the arguments.
 981          */
 982         if ((swap = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp))) {
 983                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
 984
 985                 tmp = blk1;
 986                 blk1 = blk2;
 987                 blk2 = tmp;
 988         }
 989         leaf1 = blk1->bp->b_addr;
 990         leaf2 = blk2->bp->b_addr;
 991         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
 992         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
 993         ents1 = dp->d_ops->leaf_ents_p(leaf1);
 994         ents2 = dp->d_ops->leaf_ents_p(leaf2);
 995
 996         oldsum = hdr1.count + hdr2.count;
 997 #if defined(DEBUG) || defined(XFS_WARN)
 998         oldstale = hdr1.stale + hdr2.stale;
 999 #endif
1000         mid = oldsum >> 1;
1001
1002         /*
1003          * If the old leaf count was odd then the new one will be even,
1004          * so we need to divide the new count evenly.
1005          */
1006         if (oldsum & 1) {
1007                 xfs_dahash_t    midhash;        /* middle entry hash value */
1008
1009                 if (mid >= hdr1.count)
1010                         midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1011                 else
1012                         midhash = be32_to_cpu(ents1[mid].hashval);
1013                 isleft = args->hashval <= midhash;
1014         }
1015         /*
1016          * If the old count is even then the new count is odd, so there's
1017          * no preferred side for the new entry.
1018          * Pick the left one.
1019          */
1020         else
1021                 isleft = 1;
1022         /*
1023          * Calculate moved entry count.  Positive means left-to-right,
1024          * negative means right-to-left.  Then move the entries.
1025          */
1026         count = hdr1.count - mid + (isleft == 0);
1027         if (count > 0)
1028                 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1029                                         hdr1.count - count, blk2->bp,
1030                                         &hdr2, ents2, 0, count);
1031         else if (count < 0)
1032                 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1033                                         blk1->bp, &hdr1, ents1,
1034                                         hdr1.count, count);
1035
1036         ASSERT(hdr1.count + hdr2.count == oldsum);
1037         ASSERT(hdr1.stale + hdr2.stale == oldstale);
1038
1039         /* log the changes made when moving the entries */
1040         dp->d_ops->leaf_hdr_to_disk(leaf1, &hdr1);
1041         dp->d_ops->leaf_hdr_to_disk(leaf2, &hdr2);
1042         xfs_dir3_leaf_log_header(args->trans, dp, blk1->bp);
1043         xfs_dir3_leaf_log_header(args->trans, dp, blk2->bp);
1044
1045         xfs_dir3_leaf_check(dp, blk1->bp);
1046         xfs_dir3_leaf_check(dp, blk2->bp);
1047
1048         /*
1049          * Mark whether we're inserting into the old or new leaf.
1050          */
1051         if (hdr1.count < hdr2.count)
1052                 state->inleaf = swap;
1053         else if (hdr1.count > hdr2.count)
1054                 state->inleaf = !swap;
1055         else
1056                 state->inleaf = swap ^ (blk1->index <= hdr1.count);
1057         /*
1058          * Adjust the expected index for insertion.
1059          */
1060         if (!state->inleaf)
1061                 blk2->index = blk1->index - hdr1.count;
1062
1063         /*
1064          * Finally sanity check just to make sure we are not returning a
1065          * negative index
1066          */
1067         if (blk2->index < 0) {
1068                 state->inleaf = 1;
1069                 blk2->index = 0;
1070                 xfs_alert(dp->i_mount,
1071         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1072                         __func__, blk1->index);
1073         }
1074 }
1075
1076 static int
1077 xfs_dir3_data_block_free(
1078         xfs_da_args_t           *args,
1079         struct xfs_dir2_data_hdr *hdr,
1080         struct xfs_dir2_free    *free,
1081         xfs_dir2_db_t           fdb,
1082         int                     findex,
1083         struct xfs_buf          *fbp,
1084         int                     longest)
1085 {
1086         struct xfs_trans        *tp = args->trans;
1087         int                     logfree = 0;
1088         __be16                  *bests;
1089         struct xfs_dir3_icfree_hdr freehdr;
1090         struct xfs_inode        *dp = args->dp;
1091
1092         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1093         bests = dp->d_ops->free_bests_p(free);
1094         if (hdr) {
1095                 /*
1096                  * Data block is not empty, just set the free entry to the new
1097                  * value.
1098                  */
1099                 bests[findex] = cpu_to_be16(longest);
1100                 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
1101                 return 0;
1102         }
1103
1104         /* One less used entry in the free table. */
1105         freehdr.nused--;
1106
1107         /*
1108          * If this was the last entry in the table, we can trim the table size
1109          * back.  There might be other entries at the end referring to
1110          * non-existent data blocks, get those too.
1111          */
1112         if (findex == freehdr.nvalid - 1) {
1113                 int     i;              /* free entry index */
1114
1115                 for (i = findex - 1; i >= 0; i--) {
1116                         if (bests[i] != cpu_to_be16(NULLDATAOFF))
1117                                 break;
1118                 }
1119                 freehdr.nvalid = i + 1;
1120                 logfree = 0;
1121         } else {
1122                 /* Not the last entry, just punch it out.  */
1123                 bests[findex] = cpu_to_be16(NULLDATAOFF);
1124                 logfree = 1;
1125         }
1126
1127         dp->d_ops->free_hdr_to_disk(free, &freehdr);
1128         xfs_dir2_free_log_header(tp, dp, fbp);
1129
1130         /*
1131          * If there are no useful entries left in the block, get rid of the
1132          * block if we can.
1133          */
1134         if (!freehdr.nused) {
1135                 int error;
1136
1137                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1138                 if (error == 0) {
1139                         fbp = NULL;
1140                         logfree = 0;
1141                 } else if (error != ENOSPC || args->total != 0)
1142                         return error;
1143                 /*
1144                  * It's possible to get ENOSPC if there is no
1145                  * space reservation.  In this case some one
1146                  * else will eventually get rid of this block.
1147                  */
1148         }
1149
1150         /* Log the free entry that changed, unless we got rid of it.  */
1151         if (logfree)
1152                 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
1153         return 0;
1154 }
1155
1156 /*
1157  * Remove an entry from a node directory.
1158  * This removes the leaf entry and the data entry,
1159  * and updates the free block if necessary.
1160  */
1161 static int                                      /* error */
1162 xfs_dir2_leafn_remove(
1163         xfs_da_args_t           *args,          /* operation arguments */
1164         struct xfs_buf          *bp,            /* leaf buffer */
1165         int                     index,          /* leaf entry index */
1166         xfs_da_state_blk_t      *dblk,          /* data block */
1167         int                     *rval)          /* resulting block needs join */
1168 {
1169         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1170         xfs_dir2_db_t           db;             /* data block number */
1171         struct xfs_buf          *dbp;           /* data block buffer */
1172         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1173         xfs_inode_t             *dp;            /* incore directory inode */
1174         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1175         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1176         int                     longest;        /* longest data free entry */
1177         int                     off;            /* data block entry offset */
1178         xfs_mount_t             *mp;            /* filesystem mount point */
1179         int                     needlog;        /* need to log data header */
1180         int                     needscan;       /* need to rescan data frees */
1181         xfs_trans_t             *tp;            /* transaction pointer */
1182         struct xfs_dir2_data_free *bf;          /* bestfree table */
1183         struct xfs_dir3_icleaf_hdr leafhdr;
1184         struct xfs_dir2_leaf_entry *ents;
1185
1186         trace_xfs_dir2_leafn_remove(args, index);
1187
1188         dp = args->dp;
1189         tp = args->trans;
1190         mp = dp->i_mount;
1191         leaf = bp->b_addr;
1192         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1193         ents = dp->d_ops->leaf_ents_p(leaf);
1194
1195         /*
1196          * Point to the entry we're removing.
1197          */
1198         lep = &ents[index];
1199
1200         /*
1201          * Extract the data block and offset from the entry.
1202          */
1203         db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1204         ASSERT(dblk->blkno == db);
1205         off = xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address));
1206         ASSERT(dblk->index == off);
1207
1208         /*
1209          * Kill the leaf entry by marking it stale.
1210          * Log the leaf block changes.
1211          */
1212         leafhdr.stale++;
1213         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1214         xfs_dir3_leaf_log_header(tp, dp, bp);
1215
1216         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1217         xfs_dir3_leaf_log_ents(tp, dp, bp, index, index);
1218
1219         /*
1220          * Make the data entry free.  Keep track of the longest freespace
1221          * in the data block in case it changes.
1222          */
1223         dbp = dblk->bp;
1224         hdr = dbp->b_addr;
1225         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1226         bf = dp->d_ops->data_bestfree_p(hdr);
1227         longest = be16_to_cpu(bf[0].length);
1228         needlog = needscan = 0;
1229         xfs_dir2_data_make_free(tp, dp, dbp, off,
1230                 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1231         /*
1232          * Rescan the data block freespaces for bestfree.
1233          * Log the data block header if needed.
1234          */
1235         if (needscan)
1236                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1237         if (needlog)
1238                 xfs_dir2_data_log_header(tp, dp, dbp);
1239         xfs_dir3_data_check(dp, dbp);
1240         /*
1241          * If the longest data block freespace changes, need to update
1242          * the corresponding freeblock entry.
1243          */
1244         if (longest < be16_to_cpu(bf[0].length)) {
1245                 int             error;          /* error return value */
1246                 struct xfs_buf  *fbp;           /* freeblock buffer */
1247                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1248                 int             findex;         /* index in freeblock entries */
1249                 xfs_dir2_free_t *free;          /* freeblock structure */
1250
1251                 /*
1252                  * Convert the data block number to a free block,
1253                  * read in the free block.
1254                  */
1255                 fdb = dp->d_ops->db_to_fdb(args->geo, db);
1256                 error = xfs_dir2_free_read(tp, dp,
1257                                            xfs_dir2_db_to_da(args->geo, fdb),
1258                                            &fbp);
1259                 if (error)
1260                         return error;
1261                 free = fbp->b_addr;
1262 #ifdef DEBUG
1263         {
1264                 struct xfs_dir3_icfree_hdr freehdr;
1265                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1266                 ASSERT(freehdr.firstdb == dp->d_ops->free_max_bests(args->geo) *
1267                         (fdb - xfs_dir2_byte_to_db(args->geo,
1268                                                    XFS_DIR2_FREE_OFFSET)));
1269         }
1270 #endif
1271                 /*
1272                  * Calculate which entry we need to fix.
1273                  */
1274                 findex = dp->d_ops->db_to_fdindex(args->geo, db);
1275                 longest = be16_to_cpu(bf[0].length);
1276                 /*
1277                  * If the data block is now empty we can get rid of it
1278                  * (usually).
1279                  */
1280                 if (longest == args->geo->blksize -
1281                                dp->d_ops->data_entry_offset) {
1282                         /*
1283                          * Try to punch out the data block.
1284                          */
1285                         error = xfs_dir2_shrink_inode(args, db, dbp);
1286                         if (error == 0) {
1287                                 dblk->bp = NULL;
1288                                 hdr = NULL;
1289                         }
1290                         /*
1291                          * We can get ENOSPC if there's no space reservation.
1292                          * In this case just drop the buffer and some one else
1293                          * will eventually get rid of the empty block.
1294                          */
1295                         else if (!(error == ENOSPC && args->total == 0))
1296                                 return error;
1297                 }
1298                 /*
1299                  * If we got rid of the data block, we can eliminate that entry
1300                  * in the free block.
1301                  */
1302                 error = xfs_dir3_data_block_free(args, hdr, free,
1303                                                  fdb, findex, fbp, longest);
1304                 if (error)
1305                         return error;
1306         }
1307
1308         xfs_dir3_leaf_check(dp, bp);
1309         /*
1310          * Return indication of whether this leaf block is empty enough
1311          * to justify trying to join it with a neighbor.
1312          */
1313         *rval = (dp->d_ops->leaf_hdr_size +
1314                  (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1315                 mp->m_dir_magicpct;
1316         return 0;
1317 }
1318
1319 /*
1320  * Split the leaf entries in the old block into old and new blocks.
1321  */
1322 int                                             /* error */
1323 xfs_dir2_leafn_split(
1324         xfs_da_state_t          *state,         /* btree cursor */
1325         xfs_da_state_blk_t      *oldblk,        /* original block */
1326         xfs_da_state_blk_t      *newblk)        /* newly created block */
1327 {
1328         xfs_da_args_t           *args;          /* operation arguments */
1329         xfs_dablk_t             blkno;          /* new leaf block number */
1330         int                     error;          /* error return value */
1331         xfs_mount_t             *mp;            /* filesystem mount point */
1332         struct xfs_inode        *dp;
1333
1334         /*
1335          * Allocate space for a new leaf node.
1336          */
1337         args = state->args;
1338         dp = args->dp;
1339         mp = dp->i_mount;
1340         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1341         error = xfs_da_grow_inode(args, &blkno);
1342         if (error) {
1343                 return error;
1344         }
1345         /*
1346          * Initialize the new leaf block.
1347          */
1348         error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(args->geo, blkno),
1349                                       &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1350         if (error)
1351                 return error;
1352
1353         newblk->blkno = blkno;
1354         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1355         /*
1356          * Rebalance the entries across the two leaves, link the new
1357          * block into the leaves.
1358          */
1359         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1360         error = xfs_da3_blk_link(state, oldblk, newblk);
1361         if (error) {
1362                 return error;
1363         }
1364         /*
1365          * Insert the new entry in the correct block.
1366          */
1367         if (state->inleaf)
1368                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1369         else
1370                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1371         /*
1372          * Update last hashval in each block since we added the name.
1373          */
1374         oldblk->hashval = xfs_dir2_leafn_lasthash(dp, oldblk->bp, NULL);
1375         newblk->hashval = xfs_dir2_leafn_lasthash(dp, newblk->bp, NULL);
1376         xfs_dir3_leaf_check(dp, oldblk->bp);
1377         xfs_dir3_leaf_check(dp, newblk->bp);
1378         return error;
1379 }
1380
1381 /*
1382  * Check a leaf block and its neighbors to see if the block should be
1383  * collapsed into one or the other neighbor.  Always keep the block
1384  * with the smaller block number.
1385  * If the current block is over 50% full, don't try to join it, return 0.
1386  * If the block is empty, fill in the state structure and return 2.
1387  * If it can be collapsed, fill in the state structure and return 1.
1388  * If nothing can be done, return 0.
1389  */
1390 int                                             /* error */
1391 xfs_dir2_leafn_toosmall(
1392         xfs_da_state_t          *state,         /* btree cursor */
1393         int                     *action)        /* resulting action to take */
1394 {
1395         xfs_da_state_blk_t      *blk;           /* leaf block */
1396         xfs_dablk_t             blkno;          /* leaf block number */
1397         struct xfs_buf          *bp;            /* leaf buffer */
1398         int                     bytes;          /* bytes in use */
1399         int                     count;          /* leaf live entry count */
1400         int                     error;          /* error return value */
1401         int                     forward;        /* sibling block direction */
1402         int                     i;              /* sibling counter */
1403         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1404         int                     rval;           /* result from path_shift */
1405         struct xfs_dir3_icleaf_hdr leafhdr;
1406         struct xfs_dir2_leaf_entry *ents;
1407         struct xfs_inode        *dp = state->args->dp;
1408
1409         /*
1410          * Check for the degenerate case of the block being over 50% full.
1411          * If so, it's not worth even looking to see if we might be able
1412          * to coalesce with a sibling.
1413          */
1414         blk = &state->path.blk[state->path.active - 1];
1415         leaf = blk->bp->b_addr;
1416         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1417         ents = dp->d_ops->leaf_ents_p(leaf);
1418         xfs_dir3_leaf_check(dp, blk->bp);
1419
1420         count = leafhdr.count - leafhdr.stale;
1421         bytes = dp->d_ops->leaf_hdr_size + count * sizeof(ents[0]);
1422         if (bytes > (state->blocksize >> 1)) {
1423                 /*
1424                  * Blk over 50%, don't try to join.
1425                  */
1426                 *action = 0;
1427                 return 0;
1428         }
1429         /*
1430          * Check for the degenerate case of the block being empty.
1431          * If the block is empty, we'll simply delete it, no need to
1432          * coalesce it with a sibling block.  We choose (arbitrarily)
1433          * to merge with the forward block unless it is NULL.
1434          */
1435         if (count == 0) {
1436                 /*
1437                  * Make altpath point to the block we want to keep and
1438                  * path point to the block we want to drop (this one).
1439                  */
1440                 forward = (leafhdr.forw != 0);
1441                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1442                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1443                         &rval);
1444                 if (error)
1445                         return error;
1446                 *action = rval ? 2 : 0;
1447                 return 0;
1448         }
1449         /*
1450          * Examine each sibling block to see if we can coalesce with
1451          * at least 25% free space to spare.  We need to figure out
1452          * whether to merge with the forward or the backward block.
1453          * We prefer coalescing with the lower numbered sibling so as
1454          * to shrink a directory over time.
1455          */
1456         forward = leafhdr.forw < leafhdr.back;
1457         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1458                 struct xfs_dir3_icleaf_hdr hdr2;
1459
1460                 blkno = forward ? leafhdr.forw : leafhdr.back;
1461                 if (blkno == 0)
1462                         continue;
1463                 /*
1464                  * Read the sibling leaf block.
1465                  */
1466                 error = xfs_dir3_leafn_read(state->args->trans, dp,
1467                                             blkno, -1, &bp);
1468                 if (error)
1469                         return error;
1470
1471                 /*
1472                  * Count bytes in the two blocks combined.
1473                  */
1474                 count = leafhdr.count - leafhdr.stale;
1475                 bytes = state->blocksize - (state->blocksize >> 2);
1476
1477                 leaf = bp->b_addr;
1478                 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf);
1479                 ents = dp->d_ops->leaf_ents_p(leaf);
1480                 count += hdr2.count - hdr2.stale;
1481                 bytes -= count * sizeof(ents[0]);
1482
1483                 /*
1484                  * Fits with at least 25% to spare.
1485                  */
1486                 if (bytes >= 0)
1487                         break;
1488                 xfs_trans_brelse(state->args->trans, bp);
1489         }
1490         /*
1491          * Didn't like either block, give up.
1492          */
1493         if (i >= 2) {
1494                 *action = 0;
1495                 return 0;
1496         }
1497
1498         /*
1499          * Make altpath point to the block we want to keep (the lower
1500          * numbered block) and path point to the block we want to drop.
1501          */
1502         memcpy(&state->altpath, &state->path, sizeof(state->path));
1503         if (blkno < blk->blkno)
1504                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1505                         &rval);
1506         else
1507                 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1508                         &rval);
1509         if (error) {
1510                 return error;
1511         }
1512         *action = rval ? 0 : 1;
1513         return 0;
1514 }
1515
1516 /*
1517  * Move all the leaf entries from drop_blk to save_blk.
1518  * This is done as part of a join operation.
1519  */
1520 void
1521 xfs_dir2_leafn_unbalance(
1522         xfs_da_state_t          *state,         /* cursor */
1523         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1524         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1525 {
1526         xfs_da_args_t           *args;          /* operation arguments */
1527         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1528         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1529         struct xfs_dir3_icleaf_hdr savehdr;
1530         struct xfs_dir3_icleaf_hdr drophdr;
1531         struct xfs_dir2_leaf_entry *sents;
1532         struct xfs_dir2_leaf_entry *dents;
1533         struct xfs_inode        *dp = state->args->dp;
1534
1535         args = state->args;
1536         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1537         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1538         drop_leaf = drop_blk->bp->b_addr;
1539         save_leaf = save_blk->bp->b_addr;
1540
1541         dp->d_ops->leaf_hdr_from_disk(&savehdr, save_leaf);
1542         dp->d_ops->leaf_hdr_from_disk(&drophdr, drop_leaf);
1543         sents = dp->d_ops->leaf_ents_p(save_leaf);
1544         dents = dp->d_ops->leaf_ents_p(drop_leaf);
1545
1546         /*
1547          * If there are any stale leaf entries, take this opportunity
1548          * to purge them.
1549          */
1550         if (drophdr.stale)
1551                 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1552         if (savehdr.stale)
1553                 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1554
1555         /*
1556          * Move the entries from drop to the appropriate end of save.
1557          */
1558         drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1559         if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1560                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1561                                         save_blk->bp, &savehdr, sents, 0,
1562                                         drophdr.count);
1563         else
1564                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1565                                         save_blk->bp, &savehdr, sents,
1566                                         savehdr.count, drophdr.count);
1567         save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1568
1569         /* log the changes made when moving the entries */
1570         dp->d_ops->leaf_hdr_to_disk(save_leaf, &savehdr);
1571         dp->d_ops->leaf_hdr_to_disk(drop_leaf, &drophdr);
1572         xfs_dir3_leaf_log_header(args->trans, dp, save_blk->bp);
1573         xfs_dir3_leaf_log_header(args->trans, dp, drop_blk->bp);
1574
1575         xfs_dir3_leaf_check(dp, save_blk->bp);
1576         xfs_dir3_leaf_check(dp, drop_blk->bp);
1577 }
1578
1579 /*
1580  * Top-level node form directory addname routine.
1581  */
1582 int                                             /* error */
1583 xfs_dir2_node_addname(
1584         xfs_da_args_t           *args)          /* operation arguments */
1585 {
1586         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1587         int                     error;          /* error return value */
1588         int                     rval;           /* sub-return value */
1589         xfs_da_state_t          *state;         /* btree cursor */
1590
1591         trace_xfs_dir2_node_addname(args);
1592
1593         /*
1594          * Allocate and initialize the state (btree cursor).
1595          */
1596         state = xfs_da_state_alloc();
1597         state->args = args;
1598         state->mp = args->dp->i_mount;
1599         state->blocksize = state->args->geo->blksize;
1600         state->node_ents = state->mp->m_dir_node_ents;
1601         /*
1602          * Look up the name.  We're not supposed to find it, but
1603          * this gives us the insertion point.
1604          */
1605         error = xfs_da3_node_lookup_int(state, &rval);
1606         if (error)
1607                 rval = error;
1608         if (rval != ENOENT) {
1609                 goto done;
1610         }
1611         /*
1612          * Add the data entry to a data block.
1613          * Extravalid is set to a freeblock found by lookup.
1614          */
1615         rval = xfs_dir2_node_addname_int(args,
1616                 state->extravalid ? &state->extrablk : NULL);
1617         if (rval) {
1618                 goto done;
1619         }
1620         blk = &state->path.blk[state->path.active - 1];
1621         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1622         /*
1623          * Add the new leaf entry.
1624          */
1625         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1626         if (rval == 0) {
1627                 /*
1628                  * It worked, fix the hash values up the btree.
1629                  */
1630                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1631                         xfs_da3_fixhashpath(state, &state->path);
1632         } else {
1633                 /*
1634                  * It didn't work, we need to split the leaf block.
1635                  */
1636                 if (args->total == 0) {
1637                         ASSERT(rval == ENOSPC);
1638                         goto done;
1639                 }
1640                 /*
1641                  * Split the leaf block and insert the new entry.
1642                  */
1643                 rval = xfs_da3_split(state);
1644         }
1645 done:
1646         xfs_da_state_free(state);
1647         return rval;
1648 }
1649
1650 /*
1651  * Add the data entry for a node-format directory name addition.
1652  * The leaf entry is added in xfs_dir2_leafn_add.
1653  * We may enter with a freespace block that the lookup found.
1654  */
1655 static int                                      /* error */
1656 xfs_dir2_node_addname_int(
1657         xfs_da_args_t           *args,          /* operation arguments */
1658         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1659 {
1660         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1661         xfs_dir2_db_t           dbno;           /* data block number */
1662         struct xfs_buf          *dbp;           /* data block buffer */
1663         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1664         xfs_inode_t             *dp;            /* incore directory inode */
1665         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1666         int                     error;          /* error return value */
1667         xfs_dir2_db_t           fbno;           /* freespace block number */
1668         struct xfs_buf          *fbp;           /* freespace buffer */
1669         int                     findex;         /* freespace entry index */
1670         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1671         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1672         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1673         int                     length;         /* length of the new entry */
1674         int                     logfree;        /* need to log free entry */
1675         xfs_mount_t             *mp;            /* filesystem mount point */
1676         int                     needlog;        /* need to log data header */
1677         int                     needscan;       /* need to rescan data frees */
1678         __be16                  *tagp;          /* data entry tag pointer */
1679         xfs_trans_t             *tp;            /* transaction pointer */
1680         __be16                  *bests;
1681         struct xfs_dir3_icfree_hdr freehdr;
1682         struct xfs_dir2_data_free *bf;
1683
1684         dp = args->dp;
1685         mp = dp->i_mount;
1686         tp = args->trans;
1687         length = dp->d_ops->data_entsize(args->namelen);
1688         /*
1689          * If we came in with a freespace block that means that lookup
1690          * found an entry with our hash value.  This is the freespace
1691          * block for that data entry.
1692          */
1693         if (fblk) {
1694                 fbp = fblk->bp;
1695                 /*
1696                  * Remember initial freespace block number.
1697                  */
1698                 ifbno = fblk->blkno;
1699                 free = fbp->b_addr;
1700                 findex = fblk->index;
1701                 bests = dp->d_ops->free_bests_p(free);
1702                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1703
1704                 /*
1705                  * This means the free entry showed that the data block had
1706                  * space for our entry, so we remembered it.
1707                  * Use that data block.
1708                  */
1709                 if (findex >= 0) {
1710                         ASSERT(findex < freehdr.nvalid);
1711                         ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1712                         ASSERT(be16_to_cpu(bests[findex]) >= length);
1713                         dbno = freehdr.firstdb + findex;
1714                 } else {
1715                         /*
1716                          * The data block looked at didn't have enough room.
1717                          * We'll start at the beginning of the freespace entries.
1718                          */
1719                         dbno = -1;
1720                         findex = 0;
1721                 }
1722         } else {
1723                 /*
1724                  * Didn't come in with a freespace block, so no data block.
1725                  */
1726                 ifbno = dbno = -1;
1727                 fbp = NULL;
1728                 findex = 0;
1729         }
1730
1731         /*
1732          * If we don't have a data block yet, we're going to scan the
1733          * freespace blocks looking for one.  Figure out what the
1734          * highest freespace block number is.
1735          */
1736         if (dbno == -1) {
1737                 xfs_fileoff_t   fo;             /* freespace block number */
1738
1739                 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK)))
1740                         return error;
1741                 lastfbno = xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo);
1742                 fbno = ifbno;
1743         }
1744         /*
1745          * While we haven't identified a data block, search the freeblock
1746          * data for a good data block.  If we find a null freeblock entry,
1747          * indicating a hole in the data blocks, remember that.
1748          */
1749         while (dbno == -1) {
1750                 /*
1751                  * If we don't have a freeblock in hand, get the next one.
1752                  */
1753                 if (fbp == NULL) {
1754                         /*
1755                          * Happens the first time through unless lookup gave
1756                          * us a freespace block to start with.
1757                          */
1758                         if (++fbno == 0)
1759                                 fbno = xfs_dir2_byte_to_db(args->geo,
1760                                                         XFS_DIR2_FREE_OFFSET);
1761                         /*
1762                          * If it's ifbno we already looked at it.
1763                          */
1764                         if (fbno == ifbno)
1765                                 fbno++;
1766                         /*
1767                          * If it's off the end we're done.
1768                          */
1769                         if (fbno >= lastfbno)
1770                                 break;
1771                         /*
1772                          * Read the block.  There can be holes in the
1773                          * freespace blocks, so this might not succeed.
1774                          * This should be really rare, so there's no reason
1775                          * to avoid it.
1776                          */
1777                         error = xfs_dir2_free_try_read(tp, dp,
1778                                         xfs_dir2_db_to_da(args->geo, fbno),
1779                                         &fbp);
1780                         if (error)
1781                                 return error;
1782                         if (!fbp)
1783                                 continue;
1784                         free = fbp->b_addr;
1785                         findex = 0;
1786                 }
1787                 /*
1788                  * Look at the current free entry.  Is it good enough?
1789                  *
1790                  * The bests initialisation should be where the bufer is read in
1791                  * the above branch. But gcc is too stupid to realise that bests
1792                  * and the freehdr are actually initialised if they are placed
1793                  * there, so we have to do it here to avoid warnings. Blech.
1794                  */
1795                 bests = dp->d_ops->free_bests_p(free);
1796                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1797                 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1798                     be16_to_cpu(bests[findex]) >= length)
1799                         dbno = freehdr.firstdb + findex;
1800                 else {
1801                         /*
1802                          * Are we done with the freeblock?
1803                          */
1804                         if (++findex == freehdr.nvalid) {
1805                                 /*
1806                                  * Drop the block.
1807                                  */
1808                                 xfs_trans_brelse(tp, fbp);
1809                                 fbp = NULL;
1810                                 if (fblk && fblk->bp)
1811                                         fblk->bp = NULL;
1812                         }
1813                 }
1814         }
1815         /*
1816          * If we don't have a data block, we need to allocate one and make
1817          * the freespace entries refer to it.
1818          */
1819         if (unlikely(dbno == -1)) {
1820                 /*
1821                  * Not allowed to allocate, return failure.
1822                  */
1823                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1824                         return XFS_ERROR(ENOSPC);
1825
1826                 /*
1827                  * Allocate and initialize the new data block.
1828                  */
1829                 if (unlikely((error = xfs_dir2_grow_inode(args,
1830                                                          XFS_DIR2_DATA_SPACE,
1831                                                          &dbno)) ||
1832                     (error = xfs_dir3_data_init(args, dbno, &dbp))))
1833                         return error;
1834
1835                 /*
1836                  * If (somehow) we have a freespace block, get rid of it.
1837                  */
1838                 if (fbp)
1839                         xfs_trans_brelse(tp, fbp);
1840                 if (fblk && fblk->bp)
1841                         fblk->bp = NULL;
1842
1843                 /*
1844                  * Get the freespace block corresponding to the data block
1845                  * that was just allocated.
1846                  */
1847                 fbno = dp->d_ops->db_to_fdb(args->geo, dbno);
1848                 error = xfs_dir2_free_try_read(tp, dp,
1849                                        xfs_dir2_db_to_da(args->geo, fbno),
1850                                        &fbp);
1851                 if (error)
1852                         return error;
1853
1854                 /*
1855                  * If there wasn't a freespace block, the read will
1856                  * return a NULL fbp.  Allocate and initialize a new one.
1857                  */
1858                 if (!fbp) {
1859                         error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1860                                                     &fbno);
1861                         if (error)
1862                                 return error;
1863
1864                         if (dp->d_ops->db_to_fdb(args->geo, dbno) != fbno) {
1865                                 xfs_alert(mp,
1866                         "%s: dir ino %llu needed freesp block %lld for\n"
1867                         "  data block %lld, got %lld ifbno %llu lastfbno %d",
1868                                         __func__, (unsigned long long)dp->i_ino,
1869                                         (long long)dp->d_ops->db_to_fdb(
1870                                                                 args->geo, dbno),
1871                                         (long long)dbno, (long long)fbno,
1872                                         (unsigned long long)ifbno, lastfbno);
1873                                 if (fblk) {
1874                                         xfs_alert(mp,
1875                                 " fblk 0x%p blkno %llu index %d magic 0x%x",
1876                                                 fblk,
1877                                                 (unsigned long long)fblk->blkno,
1878                                                 fblk->index,
1879                                                 fblk->magic);
1880                                 } else {
1881                                         xfs_alert(mp, " ... fblk is NULL");
1882                                 }
1883                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1884                                                  XFS_ERRLEVEL_LOW, mp);
1885                                 return XFS_ERROR(EFSCORRUPTED);
1886                         }
1887
1888                         /*
1889                          * Get a buffer for the new block.
1890                          */
1891                         error = xfs_dir3_free_get_buf(args, fbno, &fbp);
1892                         if (error)
1893                                 return error;
1894                         free = fbp->b_addr;
1895                         bests = dp->d_ops->free_bests_p(free);
1896                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1897
1898                         /*
1899                          * Remember the first slot as our empty slot.
1900                          */
1901                         freehdr.firstdb =
1902                                 (fbno - xfs_dir2_byte_to_db(args->geo,
1903                                                         XFS_DIR2_FREE_OFFSET)) *
1904                                         dp->d_ops->free_max_bests(args->geo);
1905                 } else {
1906                         free = fbp->b_addr;
1907                         bests = dp->d_ops->free_bests_p(free);
1908                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1909                 }
1910
1911                 /*
1912                  * Set the freespace block index from the data block number.
1913                  */
1914                 findex = dp->d_ops->db_to_fdindex(args->geo, dbno);
1915                 /*
1916                  * If it's after the end of the current entries in the
1917                  * freespace block, extend that table.
1918                  */
1919                 if (findex >= freehdr.nvalid) {
1920                         ASSERT(findex < dp->d_ops->free_max_bests(args->geo));
1921                         freehdr.nvalid = findex + 1;
1922                         /*
1923                          * Tag new entry so nused will go up.
1924                          */
1925                         bests[findex] = cpu_to_be16(NULLDATAOFF);
1926                 }
1927                 /*
1928                  * If this entry was for an empty data block
1929                  * (this should always be true) then update the header.
1930                  */
1931                 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1932                         freehdr.nused++;
1933                         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
1934                         xfs_dir2_free_log_header(tp, dp, fbp);
1935                 }
1936                 /*
1937                  * Update the real value in the table.
1938                  * We haven't allocated the data entry yet so this will
1939                  * change again.
1940                  */
1941                 hdr = dbp->b_addr;
1942                 bf = dp->d_ops->data_bestfree_p(hdr);
1943                 bests[findex] = bf[0].length;
1944                 logfree = 1;
1945         }
1946         /*
1947          * We had a data block so we don't have to make a new one.
1948          */
1949         else {
1950                 /*
1951                  * If just checking, we succeeded.
1952                  */
1953                 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1954                         return 0;
1955
1956                 /*
1957                  * Read the data block in.
1958                  */
1959                 error = xfs_dir3_data_read(tp, dp,
1960                                            xfs_dir2_db_to_da(args->geo, dbno),
1961                                            -1, &dbp);
1962                 if (error)
1963                         return error;
1964                 hdr = dbp->b_addr;
1965                 bf = dp->d_ops->data_bestfree_p(hdr);
1966                 logfree = 0;
1967         }
1968         ASSERT(be16_to_cpu(bf[0].length) >= length);
1969         /*
1970          * Point to the existing unused space.
1971          */
1972         dup = (xfs_dir2_data_unused_t *)
1973               ((char *)hdr + be16_to_cpu(bf[0].offset));
1974         needscan = needlog = 0;
1975         /*
1976          * Mark the first part of the unused space, inuse for us.
1977          */
1978         xfs_dir2_data_use_free(tp, dp, dbp, dup,
1979                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
1980                 &needlog, &needscan);
1981         /*
1982          * Fill in the new entry and log it.
1983          */
1984         dep = (xfs_dir2_data_entry_t *)dup;
1985         dep->inumber = cpu_to_be64(args->inumber);
1986         dep->namelen = args->namelen;
1987         memcpy(dep->name, args->name, dep->namelen);
1988         dp->d_ops->data_put_ftype(dep, args->filetype);
1989         tagp = dp->d_ops->data_entry_tag_p(dep);
1990         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1991         xfs_dir2_data_log_entry(tp, dp, dbp, dep);
1992         /*
1993          * Rescan the block for bestfree if needed.
1994          */
1995         if (needscan)
1996                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1997         /*
1998          * Log the data block header if needed.
1999          */
2000         if (needlog)
2001                 xfs_dir2_data_log_header(tp, dp, dbp);
2002         /*
2003          * If the freespace entry is now wrong, update it.
2004          */
2005         bests = dp->d_ops->free_bests_p(free); /* gcc is so stupid */
2006         if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
2007                 bests[findex] = bf[0].length;
2008                 logfree = 1;
2009         }
2010         /*
2011          * Log the freespace entry if needed.
2012          */
2013         if (logfree)
2014                 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
2015         /*
2016          * Return the data block and offset in args, then drop the data block.
2017          */
2018         args->blkno = (xfs_dablk_t)dbno;
2019         args->index = be16_to_cpu(*tagp);
2020         return 0;
2021 }
2022
2023 /*
2024  * Lookup an entry in a node-format directory.
2025  * All the real work happens in xfs_da3_node_lookup_int.
2026  * The only real output is the inode number of the entry.
2027  */
2028 int                                             /* error */
2029 xfs_dir2_node_lookup(
2030         xfs_da_args_t   *args)                  /* operation arguments */
2031 {
2032         int             error;                  /* error return value */
2033         int             i;                      /* btree level */
2034         int             rval;                   /* operation return value */
2035         xfs_da_state_t  *state;                 /* btree cursor */
2036
2037         trace_xfs_dir2_node_lookup(args);
2038
2039         /*
2040          * Allocate and initialize the btree cursor.
2041          */
2042         state = xfs_da_state_alloc();
2043         state->args = args;
2044         state->mp = args->dp->i_mount;
2045         state->blocksize = args->geo->blksize;
2046         state->node_ents = state->mp->m_dir_node_ents;
2047         /*
2048          * Fill in the path to the entry in the cursor.
2049          */
2050         error = xfs_da3_node_lookup_int(state, &rval);
2051         if (error)
2052                 rval = error;
2053         else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
2054                 /* If a CI match, dup the actual name and return EEXIST */
2055                 xfs_dir2_data_entry_t   *dep;
2056
2057                 dep = (xfs_dir2_data_entry_t *)
2058                         ((char *)state->extrablk.bp->b_addr +
2059                                                  state->extrablk.index);
2060                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2061         }
2062         /*
2063          * Release the btree blocks and leaf block.
2064          */
2065         for (i = 0; i < state->path.active; i++) {
2066                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2067                 state->path.blk[i].bp = NULL;
2068         }
2069         /*
2070          * Release the data block if we have it.
2071          */
2072         if (state->extravalid && state->extrablk.bp) {
2073                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2074                 state->extrablk.bp = NULL;
2075         }
2076         xfs_da_state_free(state);
2077         return rval;
2078 }
2079
2080 /*
2081  * Remove an entry from a node-format directory.
2082  */
2083 int                                             /* error */
2084 xfs_dir2_node_removename(
2085         struct xfs_da_args      *args)          /* operation arguments */
2086 {
2087         struct xfs_da_state_blk *blk;           /* leaf block */
2088         int                     error;          /* error return value */
2089         int                     rval;           /* operation return value */
2090         struct xfs_da_state     *state;         /* btree cursor */
2091
2092         trace_xfs_dir2_node_removename(args);
2093
2094         /*
2095          * Allocate and initialize the btree cursor.
2096          */
2097         state = xfs_da_state_alloc();
2098         state->args = args;
2099         state->mp = args->dp->i_mount;
2100         state->blocksize = args->geo->blksize;
2101         state->node_ents = state->mp->m_dir_node_ents;
2102
2103         /* Look up the entry we're deleting, set up the cursor. */
2104         error = xfs_da3_node_lookup_int(state, &rval);
2105         if (error)
2106                 goto out_free;
2107
2108         /* Didn't find it, upper layer screwed up. */
2109         if (rval != EEXIST) {
2110                 error = rval;
2111                 goto out_free;
2112         }
2113
2114         blk = &state->path.blk[state->path.active - 1];
2115         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2116         ASSERT(state->extravalid);
2117         /*
2118          * Remove the leaf and data entries.
2119          * Extrablk refers to the data block.
2120          */
2121         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2122                 &state->extrablk, &rval);
2123         if (error)
2124                 goto out_free;
2125         /*
2126          * Fix the hash values up the btree.
2127          */
2128         xfs_da3_fixhashpath(state, &state->path);
2129         /*
2130          * If we need to join leaf blocks, do it.
2131          */
2132         if (rval && state->path.active > 1)
2133                 error = xfs_da3_join(state);
2134         /*
2135          * If no errors so far, try conversion to leaf format.
2136          */
2137         if (!error)
2138                 error = xfs_dir2_node_to_leaf(state);
2139 out_free:
2140         xfs_da_state_free(state);
2141         return error;
2142 }
2143
2144 /*
2145  * Replace an entry's inode number in a node-format directory.
2146  */
2147 int                                             /* error */
2148 xfs_dir2_node_replace(
2149         xfs_da_args_t           *args)          /* operation arguments */
2150 {
2151         xfs_da_state_blk_t      *blk;           /* leaf block */
2152         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
2153         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
2154         int                     error;          /* error return value */
2155         int                     i;              /* btree level */
2156         xfs_ino_t               inum;           /* new inode number */
2157         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
2158         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
2159         int                     rval;           /* internal return value */
2160         xfs_da_state_t          *state;         /* btree cursor */
2161
2162         trace_xfs_dir2_node_replace(args);
2163
2164         /*
2165          * Allocate and initialize the btree cursor.
2166          */
2167         state = xfs_da_state_alloc();
2168         state->args = args;
2169         state->mp = args->dp->i_mount;
2170         state->blocksize = args->geo->blksize;
2171         state->node_ents = state->mp->m_dir_node_ents;
2172         inum = args->inumber;
2173         /*
2174          * Lookup the entry to change in the btree.
2175          */
2176         error = xfs_da3_node_lookup_int(state, &rval);
2177         if (error) {
2178                 rval = error;
2179         }
2180         /*
2181          * It should be found, since the vnodeops layer has looked it up
2182          * and locked it.  But paranoia is good.
2183          */
2184         if (rval == EEXIST) {
2185                 struct xfs_dir2_leaf_entry *ents;
2186                 /*
2187                  * Find the leaf entry.
2188                  */
2189                 blk = &state->path.blk[state->path.active - 1];
2190                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2191                 leaf = blk->bp->b_addr;
2192                 ents = args->dp->d_ops->leaf_ents_p(leaf);
2193                 lep = &ents[blk->index];
2194                 ASSERT(state->extravalid);
2195                 /*
2196                  * Point to the data entry.
2197                  */
2198                 hdr = state->extrablk.bp->b_addr;
2199                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2200                        hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2201                 dep = (xfs_dir2_data_entry_t *)
2202                       ((char *)hdr +
2203                        xfs_dir2_dataptr_to_off(args->geo,
2204                                                be32_to_cpu(lep->address)));
2205                 ASSERT(inum != be64_to_cpu(dep->inumber));
2206                 /*
2207                  * Fill in the new inode number and log the entry.
2208                  */
2209                 dep->inumber = cpu_to_be64(inum);
2210                 args->dp->d_ops->data_put_ftype(dep, args->filetype);
2211                 xfs_dir2_data_log_entry(args->trans, args->dp,
2212                                         state->extrablk.bp, dep);
2213                 rval = 0;
2214         }
2215         /*
2216          * Didn't find it, and we're holding a data block.  Drop it.
2217          */
2218         else if (state->extravalid) {
2219                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2220                 state->extrablk.bp = NULL;
2221         }
2222         /*
2223          * Release all the buffers in the cursor.
2224          */
2225         for (i = 0; i < state->path.active; i++) {
2226                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2227                 state->path.blk[i].bp = NULL;
2228         }
2229         xfs_da_state_free(state);
2230         return rval;
2231 }
2232
2233 /*
2234  * Trim off a trailing empty freespace block.
2235  * Return (in rvalp) 1 if we did it, 0 if not.
2236  */
2237 int                                             /* error */
2238 xfs_dir2_node_trim_free(
2239         xfs_da_args_t           *args,          /* operation arguments */
2240         xfs_fileoff_t           fo,             /* free block number */
2241         int                     *rvalp)         /* out: did something */
2242 {
2243         struct xfs_buf          *bp;            /* freespace buffer */
2244         xfs_inode_t             *dp;            /* incore directory inode */
2245         int                     error;          /* error return code */
2246         xfs_dir2_free_t         *free;          /* freespace structure */
2247         xfs_mount_t             *mp;            /* filesystem mount point */
2248         xfs_trans_t             *tp;            /* transaction pointer */
2249         struct xfs_dir3_icfree_hdr freehdr;
2250
2251         dp = args->dp;
2252         mp = dp->i_mount;
2253         tp = args->trans;
2254         /*
2255          * Read the freespace block.
2256          */
2257         error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2258         if (error)
2259                 return error;
2260         /*
2261          * There can be holes in freespace.  If fo is a hole, there's
2262          * nothing to do.
2263          */
2264         if (!bp)
2265                 return 0;
2266         free = bp->b_addr;
2267         dp->d_ops->free_hdr_from_disk(&freehdr, free);
2268
2269         /*
2270          * If there are used entries, there's nothing to do.
2271          */
2272         if (freehdr.nused > 0) {
2273                 xfs_trans_brelse(tp, bp);
2274                 *rvalp = 0;
2275                 return 0;
2276         }
2277         /*
2278          * Blow the block away.
2279          */
2280         error = xfs_dir2_shrink_inode(args,
2281                         xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo), bp);
2282         if (error) {
2283                 /*
2284                  * Can't fail with ENOSPC since that only happens with no
2285                  * space reservation, when breaking up an extent into two
2286                  * pieces.  This is the last block of an extent.
2287                  */
2288                 ASSERT(error != ENOSPC);
2289                 xfs_trans_brelse(tp, bp);
2290                 return error;
2291         }
2292         /*
2293          * Return that we succeeded.
2294          */
2295         *rvalp = 1;
2296         return 0;
2297 }