kernel/fs/xfs/libxfs/xfs_dir2_node.c

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