fs/xfs/xfs_dir2_node.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_inode.h"
  30 #include "xfs_bmap.h"
  31 #include "xfs_dir2.h"
  32 #include "xfs_dir2_priv.h"
  33 #include "xfs_error.h"
  34 #include "xfs_trace.h"
  35 #include "xfs_trans.h"
  36 #include "xfs_buf_item.h"
  37 #include "xfs_cksum.h"
  38
  39 /*
  40  * Function declarations.
  41  */
  42 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
  43                               int index);
  44 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  45                                      xfs_da_state_blk_t *blk1,
  46                                      xfs_da_state_blk_t *blk2);
  47 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
  48                                  int index, xfs_da_state_blk_t *dblk,
  49                                  int *rval);
  50 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  51                                      xfs_da_state_blk_t *fblk);
  52
  53 /*
  54  * Check internal consistency of a leafn block.
  55  */
  56 #ifdef DEBUG
  57 #define xfs_dir3_leaf_check(dp, bp) \
  58 do { \
  59         if (!xfs_dir3_leafn_check((dp), (bp))) \
  60                 ASSERT(0); \
  61 } while (0);
  62
  63 static bool
  64 xfs_dir3_leafn_check(
  65         struct xfs_inode        *dp,
  66         struct xfs_buf          *bp)
  67 {
  68         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  69         struct xfs_dir3_icleaf_hdr leafhdr;
  70
  71         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
  72
  73         if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
  74                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  75                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  76                         return false;
  77         } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
  78                 return false;
  79
  80         return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
  81 }
  82 #else
  83 #define xfs_dir3_leaf_check(dp, bp)
  84 #endif
  85
  86 static bool
  87 xfs_dir3_free_verify(
  88         struct xfs_buf          *bp)
  89 {
  90         struct xfs_mount        *mp = bp->b_target->bt_mount;
  91         struct xfs_dir2_free_hdr *hdr = bp->b_addr;
  92
  93         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  94                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
  95
  96                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_FREE_MAGIC))
  97                         return false;
  98                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
  99                         return false;
 100                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 101                         return false;
 102         } else {
 103                 if (hdr->magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC))
 104                         return false;
 105         }
 106
 107         /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
 108
 109         return true;
 110 }
 111
 112 static void
 113 xfs_dir3_free_read_verify(
 114         struct xfs_buf  *bp)
 115 {
 116         struct xfs_mount        *mp = bp->b_target->bt_mount;
 117
 118         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 119             !xfs_buf_verify_cksum(bp, XFS_DIR3_FREE_CRC_OFF))
 120                 xfs_buf_ioerror(bp, EFSBADCRC);
 121         else if (!xfs_dir3_free_verify(bp))
 122                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 123
 124         if (bp->b_error)
 125                 xfs_verifier_error(bp);
 126 }
 127
 128 static void
 129 xfs_dir3_free_write_verify(
 130         struct xfs_buf  *bp)
 131 {
 132         struct xfs_mount        *mp = bp->b_target->bt_mount;
 133         struct xfs_buf_log_item *bip = bp->b_fspriv;
 134         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 135
 136         if (!xfs_dir3_free_verify(bp)) {
 137                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 138                 xfs_verifier_error(bp);
 139                 return;
 140         }
 141
 142         if (!xfs_sb_version_hascrc(&mp->m_sb))
 143                 return;
 144
 145         if (bip)
 146                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 147
 148         xfs_buf_update_cksum(bp, XFS_DIR3_FREE_CRC_OFF);
 149 }
 150
 151 const struct xfs_buf_ops xfs_dir3_free_buf_ops = {
 152         .verify_read = xfs_dir3_free_read_verify,
 153         .verify_write = xfs_dir3_free_write_verify,
 154 };
 155
 156
 157 static int
 158 __xfs_dir3_free_read(
 159         struct xfs_trans        *tp,
 160         struct xfs_inode        *dp,
 161         xfs_dablk_t             fbno,
 162         xfs_daddr_t             mappedbno,
 163         struct xfs_buf          **bpp)
 164 {
 165         int                     err;
 166
 167         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 168                                 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
 169
 170         /* try read returns without an error or *bpp if it lands in a hole */
 171         if (!err && tp && *bpp)
 172                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
 173         return err;
 174 }
 175
 176 int
 177 xfs_dir2_free_read(
 178         struct xfs_trans        *tp,
 179         struct xfs_inode        *dp,
 180         xfs_dablk_t             fbno,
 181         struct xfs_buf          **bpp)
 182 {
 183         return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
 184 }
 185
 186 static int
 187 xfs_dir2_free_try_read(
 188         struct xfs_trans        *tp,
 189         struct xfs_inode        *dp,
 190         xfs_dablk_t             fbno,
 191         struct xfs_buf          **bpp)
 192 {
 193         return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
 194 }
 195
 196 static int
 197 xfs_dir3_free_get_buf(
 198         struct xfs_trans        *tp,
 199         struct xfs_inode        *dp,
 200         xfs_dir2_db_t           fbno,
 201         struct xfs_buf          **bpp)
 202 {
 203         struct xfs_mount        *mp = dp->i_mount;
 204         struct xfs_buf          *bp;
 205         int                     error;
 206         struct xfs_dir3_icfree_hdr hdr;
 207
 208         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, fbno),
 209                                    -1, &bp, XFS_DATA_FORK);
 210         if (error)
 211                 return error;
 212
 213         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
 214         bp->b_ops = &xfs_dir3_free_buf_ops;
 215
 216         /*
 217          * Initialize the new block to be empty, and remember
 218          * its first slot as our empty slot.
 219          */
 220         memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
 221         memset(&hdr, 0, sizeof(hdr));
 222
 223         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 224                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 225
 226                 hdr.magic = XFS_DIR3_FREE_MAGIC;
 227
 228                 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
 229                 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
 230                 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_uuid);
 231         } else
 232                 hdr.magic = XFS_DIR2_FREE_MAGIC;
 233         dp->d_ops->free_hdr_to_disk(bp->b_addr, &hdr);
 234         *bpp = bp;
 235         return 0;
 236 }
 237
 238 /*
 239  * Log entries from a freespace block.
 240  */
 241 STATIC void
 242 xfs_dir2_free_log_bests(
 243         struct xfs_trans        *tp,
 244         struct xfs_inode        *dp,
 245         struct xfs_buf          *bp,
 246         int                     first,          /* first entry to log */
 247         int                     last)           /* last entry to log */
 248 {
 249         xfs_dir2_free_t         *free;          /* freespace structure */
 250         __be16                  *bests;
 251
 252         free = bp->b_addr;
 253         bests = dp->d_ops->free_bests_p(free);
 254         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 255                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 256         xfs_trans_log_buf(tp, bp,
 257                 (uint)((char *)&bests[first] - (char *)free),
 258                 (uint)((char *)&bests[last] - (char *)free +
 259                        sizeof(bests[0]) - 1));
 260 }
 261
 262 /*
 263  * Log header from a freespace block.
 264  */
 265 static void
 266 xfs_dir2_free_log_header(
 267         struct xfs_trans        *tp,
 268         struct xfs_inode        *dp,
 269         struct xfs_buf          *bp)
 270 {
 271 #ifdef DEBUG
 272         xfs_dir2_free_t         *free;          /* freespace structure */
 273
 274         free = bp->b_addr;
 275         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 276                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 277 #endif
 278         xfs_trans_log_buf(tp, bp, 0, dp->d_ops->free_hdr_size - 1);
 279 }
 280
 281 /*
 282  * Convert a leaf-format directory to a node-format directory.
 283  * We need to change the magic number of the leaf block, and copy
 284  * the freespace table out of the leaf block into its own block.
 285  */
 286 int                                             /* error */
 287 xfs_dir2_leaf_to_node(
 288         xfs_da_args_t           *args,          /* operation arguments */
 289         struct xfs_buf          *lbp)           /* leaf buffer */
 290 {
 291         xfs_inode_t             *dp;            /* incore directory inode */
 292         int                     error;          /* error return value */
 293         struct xfs_buf          *fbp;           /* freespace buffer */
 294         xfs_dir2_db_t           fdb;            /* freespace block number */
 295         xfs_dir2_free_t         *free;          /* freespace structure */
 296         __be16                  *from;          /* pointer to freespace entry */
 297         int                     i;              /* leaf freespace index */
 298         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 299         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 300         xfs_mount_t             *mp;            /* filesystem mount point */
 301         int                     n;              /* count of live freespc ents */
 302         xfs_dir2_data_off_t     off;            /* freespace entry value */
 303         __be16                  *to;            /* pointer to freespace entry */
 304         xfs_trans_t             *tp;            /* transaction pointer */
 305         struct xfs_dir3_icfree_hdr freehdr;
 306
 307         trace_xfs_dir2_leaf_to_node(args);
 308
 309         dp = args->dp;
 310         mp = dp->i_mount;
 311         tp = args->trans;
 312         /*
 313          * Add a freespace block to the directory.
 314          */
 315         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 316                 return error;
 317         }
 318         ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
 319         /*
 320          * Get the buffer for the new freespace block.
 321          */
 322         error = xfs_dir3_free_get_buf(tp, dp, fdb, &fbp);
 323         if (error)
 324                 return error;
 325
 326         free = fbp->b_addr;
 327         dp->d_ops->free_hdr_from_disk(&freehdr, free);
 328         leaf = lbp->b_addr;
 329         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 330         ASSERT(be32_to_cpu(ltp->bestcount) <=
 331                                 (uint)dp->i_d.di_size / mp->m_dirblksize);
 332
 333         /*
 334          * Copy freespace entries from the leaf block to the new block.
 335          * Count active entries.
 336          */
 337         from = xfs_dir2_leaf_bests_p(ltp);
 338         to = dp->d_ops->free_bests_p(free);
 339         for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 340                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 341                         n++;
 342                 *to = cpu_to_be16(off);
 343         }
 344
 345         /*
 346          * Now initialize the freespace block header.
 347          */
 348         freehdr.nused = n;
 349         freehdr.nvalid = be32_to_cpu(ltp->bestcount);
 350
 351         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
 352         xfs_dir2_free_log_bests(tp, dp, fbp, 0, freehdr.nvalid - 1);
 353         xfs_dir2_free_log_header(tp, dp, fbp);
 354
 355         /*
 356          * Converting the leaf to a leafnode is just a matter of changing the
 357          * magic number and the ops. Do the change directly to the buffer as
 358          * it's less work (and less code) than decoding the header to host
 359          * format and back again.
 360          */
 361         if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
 362                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 363         else
 364                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 365         lbp->b_ops = &xfs_dir3_leafn_buf_ops;
 366         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
 367         xfs_dir3_leaf_log_header(tp, dp, lbp);
 368         xfs_dir3_leaf_check(dp, lbp);
 369         return 0;
 370 }
 371
 372 /*
 373  * Add a leaf entry to a leaf block in a node-form directory.
 374  * The other work necessary is done from the caller.
 375  */
 376 static int                                      /* error */
 377 xfs_dir2_leafn_add(
 378         struct xfs_buf          *bp,            /* leaf buffer */
 379         xfs_da_args_t           *args,          /* operation arguments */
 380         int                     index)          /* insertion pt for new entry */
 381 {
 382         int                     compact;        /* compacting stale leaves */
 383         xfs_inode_t             *dp;            /* incore directory inode */
 384         int                     highstale;      /* next stale entry */
 385         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 386         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 387         int                     lfloghigh;      /* high leaf entry logging */
 388         int                     lfloglow;       /* low leaf entry logging */
 389         int                     lowstale;       /* previous stale entry */
 390         xfs_mount_t             *mp;            /* filesystem mount point */
 391         xfs_trans_t             *tp;            /* transaction pointer */
 392         struct xfs_dir3_icleaf_hdr leafhdr;
 393         struct xfs_dir2_leaf_entry *ents;
 394
 395         trace_xfs_dir2_leafn_add(args, index);
 396
 397         dp = args->dp;
 398         mp = dp->i_mount;
 399         tp = args->trans;
 400         leaf = bp->b_addr;
 401         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 402         ents = dp->d_ops->leaf_ents_p(leaf);
 403
 404         /*
 405          * Quick check just to make sure we are not going to index
 406          * into other peoples memory
 407          */
 408         if (index < 0)
 409                 return XFS_ERROR(EFSCORRUPTED);
 410
 411         /*
 412          * If there are already the maximum number of leaf entries in
 413          * the block, if there are no stale entries it won't fit.
 414          * Caller will do a split.  If there are stale entries we'll do
 415          * a compact.
 416          */
 417
 418         if (leafhdr.count == dp->d_ops->leaf_max_ents(mp)) {
 419                 if (!leafhdr.stale)
 420                         return XFS_ERROR(ENOSPC);
 421                 compact = leafhdr.stale > 1;
 422         } else
 423                 compact = 0;
 424         ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
 425         ASSERT(index == leafhdr.count ||
 426                be32_to_cpu(ents[index].hashval) >= args->hashval);
 427
 428         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 429                 return 0;
 430
 431         /*
 432          * Compact out all but one stale leaf entry.  Leaves behind
 433          * the entry closest to index.
 434          */
 435         if (compact)
 436                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 437                                          &highstale, &lfloglow, &lfloghigh);
 438         else if (leafhdr.stale) {
 439                 /*
 440                  * Set impossible logging indices for this case.
 441                  */
 442                 lfloglow = leafhdr.count;
 443                 lfloghigh = -1;
 444         }
 445
 446         /*
 447          * Insert the new entry, log everything.
 448          */
 449         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 450                                        highstale, &lfloglow, &lfloghigh);
 451
 452         lep->hashval = cpu_to_be32(args->hashval);
 453         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
 454                                 args->blkno, args->index));
 455
 456         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
 457         xfs_dir3_leaf_log_header(tp, dp, bp);
 458         xfs_dir3_leaf_log_ents(tp, dp, bp, lfloglow, lfloghigh);
 459         xfs_dir3_leaf_check(dp, bp);
 460         return 0;
 461 }
 462
 463 #ifdef DEBUG
 464 static void
 465 xfs_dir2_free_hdr_check(
 466         struct xfs_inode *dp,
 467         struct xfs_buf  *bp,
 468         xfs_dir2_db_t   db)
 469 {
 470         struct xfs_dir3_icfree_hdr hdr;
 471
 472         dp->d_ops->free_hdr_from_disk(&hdr, bp->b_addr);
 473
 474         ASSERT((hdr.firstdb % dp->d_ops->free_max_bests(dp->i_mount)) == 0);
 475         ASSERT(hdr.firstdb <= db);
 476         ASSERT(db < hdr.firstdb + hdr.nvalid);
 477 }
 478 #else
 479 #define xfs_dir2_free_hdr_check(dp, bp, db)
 480 #endif  /* DEBUG */
 481
 482 /*
 483  * Return the last hash value in the leaf.
 484  * Stale entries are ok.
 485  */
 486 xfs_dahash_t                                    /* hash value */
 487 xfs_dir2_leafn_lasthash(
 488         struct xfs_inode *dp,
 489         struct xfs_buf  *bp,                    /* leaf buffer */
 490         int             *count)                 /* count of entries in leaf */
 491 {
 492         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 493         struct xfs_dir2_leaf_entry *ents;
 494         struct xfs_dir3_icleaf_hdr leafhdr;
 495
 496         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 497
 498         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 499                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
 500
 501         if (count)
 502                 *count = leafhdr.count;
 503         if (!leafhdr.count)
 504                 return 0;
 505
 506         ents = dp->d_ops->leaf_ents_p(leaf);
 507         return be32_to_cpu(ents[leafhdr.count - 1].hashval);
 508 }
 509
 510 /*
 511  * Look up a leaf entry for space to add a name in a node-format leaf block.
 512  * The extrablk in state is a freespace block.
 513  */
 514 STATIC int
 515 xfs_dir2_leafn_lookup_for_addname(
 516         struct xfs_buf          *bp,            /* leaf buffer */
 517         xfs_da_args_t           *args,          /* operation arguments */
 518         int                     *indexp,        /* out: leaf entry index */
 519         xfs_da_state_t          *state)         /* state to fill in */
 520 {
 521         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 522         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 523         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 524         xfs_inode_t             *dp;            /* incore directory inode */
 525         int                     error;          /* error return value */
 526         int                     fi;             /* free entry index */
 527         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 528         int                     index;          /* leaf entry index */
 529         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 530         int                     length;         /* length of new data entry */
 531         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 532         xfs_mount_t             *mp;            /* filesystem mount point */
 533         xfs_dir2_db_t           newdb;          /* new data block number */
 534         xfs_dir2_db_t           newfdb;         /* new free block number */
 535         xfs_trans_t             *tp;            /* transaction pointer */
 536         struct xfs_dir2_leaf_entry *ents;
 537         struct xfs_dir3_icleaf_hdr leafhdr;
 538
 539         dp = args->dp;
 540         tp = args->trans;
 541         mp = dp->i_mount;
 542         leaf = bp->b_addr;
 543         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 544         ents = dp->d_ops->leaf_ents_p(leaf);
 545
 546         xfs_dir3_leaf_check(dp, bp);
 547         ASSERT(leafhdr.count > 0);
 548
 549         /*
 550          * Look up the hash value in the leaf entries.
 551          */
 552         index = xfs_dir2_leaf_search_hash(args, bp);
 553         /*
 554          * Do we have a buffer coming in?
 555          */
 556         if (state->extravalid) {
 557                 /* If so, it's a free block buffer, get the block number. */
 558                 curbp = state->extrablk.bp;
 559                 curfdb = state->extrablk.blkno;
 560                 free = curbp->b_addr;
 561                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 562                        free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 563         }
 564         length = dp->d_ops->data_entsize(args->namelen);
 565         /*
 566          * Loop over leaf entries with the right hash value.
 567          */
 568         for (lep = &ents[index];
 569              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 570              lep++, index++) {
 571                 /*
 572                  * Skip stale leaf entries.
 573                  */
 574                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 575                         continue;
 576                 /*
 577                  * Pull the data block number from the entry.
 578                  */
 579                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 580                 /*
 581                  * For addname, we're looking for a place to put the new entry.
 582                  * We want to use a data block with an entry of equal
 583                  * hash value to ours if there is one with room.
 584                  *
 585                  * If this block isn't the data block we already have
 586                  * in hand, take a look at it.
 587                  */
 588                 if (newdb != curdb) {
 589                         __be16 *bests;
 590
 591                         curdb = newdb;
 592                         /*
 593                          * Convert the data block to the free block
 594                          * holding its freespace information.
 595                          */
 596                         newfdb = dp->d_ops->db_to_fdb(mp, newdb);
 597                         /*
 598                          * If it's not the one we have in hand, read it in.
 599                          */
 600                         if (newfdb != curfdb) {
 601                                 /*
 602                                  * If we had one before, drop it.
 603                                  */
 604                                 if (curbp)
 605                                         xfs_trans_brelse(tp, curbp);
 606
 607                                 error = xfs_dir2_free_read(tp, dp,
 608                                                 xfs_dir2_db_to_da(mp, newfdb),
 609                                                 &curbp);
 610                                 if (error)
 611                                         return error;
 612                                 free = curbp->b_addr;
 613
 614                                 xfs_dir2_free_hdr_check(dp, curbp, curdb);
 615                         }
 616                         /*
 617                          * Get the index for our entry.
 618                          */
 619                         fi = dp->d_ops->db_to_fdindex(mp, curdb);
 620                         /*
 621                          * If it has room, return it.
 622                          */
 623                         bests = dp->d_ops->free_bests_p(free);
 624                         if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
 625                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 626                                                         XFS_ERRLEVEL_LOW, mp);
 627                                 if (curfdb != newfdb)
 628                                         xfs_trans_brelse(tp, curbp);
 629                                 return XFS_ERROR(EFSCORRUPTED);
 630                         }
 631                         curfdb = newfdb;
 632                         if (be16_to_cpu(bests[fi]) >= length)
 633                                 goto out;
 634                 }
 635         }
 636         /* Didn't find any space */
 637         fi = -1;
 638 out:
 639         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 640         if (curbp) {
 641                 /* Giving back a free block. */
 642                 state->extravalid = 1;
 643                 state->extrablk.bp = curbp;
 644                 state->extrablk.index = fi;
 645                 state->extrablk.blkno = curfdb;
 646
 647                 /*
 648                  * Important: this magic number is not in the buffer - it's for
 649                  * buffer type information and therefore only the free/data type
 650                  * matters here, not whether CRCs are enabled or not.
 651                  */
 652                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 653         } else {
 654                 state->extravalid = 0;
 655         }
 656         /*
 657          * Return the index, that will be the insertion point.
 658          */
 659         *indexp = index;
 660         return XFS_ERROR(ENOENT);
 661 }
 662
 663 /*
 664  * Look up a leaf entry in a node-format leaf block.
 665  * The extrablk in state a data block.
 666  */
 667 STATIC int
 668 xfs_dir2_leafn_lookup_for_entry(
 669         struct xfs_buf          *bp,            /* leaf buffer */
 670         xfs_da_args_t           *args,          /* operation arguments */
 671         int                     *indexp,        /* out: leaf entry index */
 672         xfs_da_state_t          *state)         /* state to fill in */
 673 {
 674         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 675         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 676         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 677         xfs_inode_t             *dp;            /* incore directory inode */
 678         int                     error;          /* error return value */
 679         int                     index;          /* leaf entry index */
 680         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 681         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 682         xfs_mount_t             *mp;            /* filesystem mount point */
 683         xfs_dir2_db_t           newdb;          /* new data block number */
 684         xfs_trans_t             *tp;            /* transaction pointer */
 685         enum xfs_dacmp          cmp;            /* comparison result */
 686         struct xfs_dir2_leaf_entry *ents;
 687         struct xfs_dir3_icleaf_hdr leafhdr;
 688
 689         dp = args->dp;
 690         tp = args->trans;
 691         mp = dp->i_mount;
 692         leaf = bp->b_addr;
 693         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 694         ents = dp->d_ops->leaf_ents_p(leaf);
 695
 696         xfs_dir3_leaf_check(dp, bp);
 697         ASSERT(leafhdr.count > 0);
 698
 699         /*
 700          * Look up the hash value in the leaf entries.
 701          */
 702         index = xfs_dir2_leaf_search_hash(args, bp);
 703         /*
 704          * Do we have a buffer coming in?
 705          */
 706         if (state->extravalid) {
 707                 curbp = state->extrablk.bp;
 708                 curdb = state->extrablk.blkno;
 709         }
 710         /*
 711          * Loop over leaf entries with the right hash value.
 712          */
 713         for (lep = &ents[index];
 714              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 715              lep++, index++) {
 716                 /*
 717                  * Skip stale leaf entries.
 718                  */
 719                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 720                         continue;
 721                 /*
 722                  * Pull the data block number from the entry.
 723                  */
 724                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 725                 /*
 726                  * Not adding a new entry, so we really want to find
 727                  * the name given to us.
 728                  *
 729                  * If it's a different data block, go get it.
 730                  */
 731                 if (newdb != curdb) {
 732                         /*
 733                          * If we had a block before that we aren't saving
 734                          * for a CI name, drop it
 735                          */
 736                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 737                                                 curdb != state->extrablk.blkno))
 738                                 xfs_trans_brelse(tp, curbp);
 739                         /*
 740                          * If needing the block that is saved with a CI match,
 741                          * use it otherwise read in the new data block.
 742                          */
 743                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 744                                         newdb == state->extrablk.blkno) {
 745                                 ASSERT(state->extravalid);
 746                                 curbp = state->extrablk.bp;
 747                         } else {
 748                                 error = xfs_dir3_data_read(tp, dp,
 749                                                 xfs_dir2_db_to_da(mp, newdb),
 750                                                 -1, &curbp);
 751                                 if (error)
 752                                         return error;
 753                         }
 754                         xfs_dir3_data_check(dp, curbp);
 755                         curdb = newdb;
 756                 }
 757                 /*
 758                  * Point to the data entry.
 759                  */
 760                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 761                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
 762                 /*
 763                  * Compare the entry and if it's an exact match, return
 764                  * EEXIST immediately. If it's the first case-insensitive
 765                  * match, store the block & inode number and continue looking.
 766                  */
 767                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 768                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 769                         /* If there is a CI match block, drop it */
 770                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 771                                                 curdb != state->extrablk.blkno)
 772                                 xfs_trans_brelse(tp, state->extrablk.bp);
 773                         args->cmpresult = cmp;
 774                         args->inumber = be64_to_cpu(dep->inumber);
 775                         args->filetype = dp->d_ops->data_get_ftype(dep);
 776                         *indexp = index;
 777                         state->extravalid = 1;
 778                         state->extrablk.bp = curbp;
 779                         state->extrablk.blkno = curdb;
 780                         state->extrablk.index = (int)((char *)dep -
 781                                                         (char *)curbp->b_addr);
 782                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 783                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 784                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 785                         if (cmp == XFS_CMP_EXACT)
 786                                 return XFS_ERROR(EEXIST);
 787                 }
 788         }
 789         ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
 790         if (curbp) {
 791                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 792                         /* Giving back last used data block. */
 793                         state->extravalid = 1;
 794                         state->extrablk.bp = curbp;
 795                         state->extrablk.index = -1;
 796                         state->extrablk.blkno = curdb;
 797                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 798                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 799                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 800                 } else {
 801                         /* If the curbp is not the CI match block, drop it */
 802                         if (state->extrablk.bp != curbp)
 803                                 xfs_trans_brelse(tp, curbp);
 804                 }
 805         } else {
 806                 state->extravalid = 0;
 807         }
 808         *indexp = index;
 809         return XFS_ERROR(ENOENT);
 810 }
 811
 812 /*
 813  * Look up a leaf entry in a node-format leaf block.
 814  * If this is an addname then the extrablk in state is a freespace block,
 815  * otherwise it's a data block.
 816  */
 817 int
 818 xfs_dir2_leafn_lookup_int(
 819         struct xfs_buf          *bp,            /* leaf buffer */
 820         xfs_da_args_t           *args,          /* operation arguments */
 821         int                     *indexp,        /* out: leaf entry index */
 822         xfs_da_state_t          *state)         /* state to fill in */
 823 {
 824         if (args->op_flags & XFS_DA_OP_ADDNAME)
 825                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 826                                                         state);
 827         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 828 }
 829
 830 /*
 831  * Move count leaf entries from source to destination leaf.
 832  * Log entries and headers.  Stale entries are preserved.
 833  */
 834 static void
 835 xfs_dir3_leafn_moveents(
 836         xfs_da_args_t                   *args,  /* operation arguments */
 837         struct xfs_buf                  *bp_s,  /* source */
 838         struct xfs_dir3_icleaf_hdr      *shdr,
 839         struct xfs_dir2_leaf_entry      *sents,
 840         int                             start_s,/* source leaf index */
 841         struct xfs_buf                  *bp_d,  /* destination */
 842         struct xfs_dir3_icleaf_hdr      *dhdr,
 843         struct xfs_dir2_leaf_entry      *dents,
 844         int                             start_d,/* destination leaf index */
 845         int                             count)  /* count of leaves to copy */
 846 {
 847         struct xfs_trans                *tp = args->trans;
 848         int                             stale;  /* count stale leaves copied */
 849
 850         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 851
 852         /*
 853          * Silently return if nothing to do.
 854          */
 855         if (count == 0)
 856                 return;
 857
 858         /*
 859          * If the destination index is not the end of the current
 860          * destination leaf entries, open up a hole in the destination
 861          * to hold the new entries.
 862          */
 863         if (start_d < dhdr->count) {
 864                 memmove(&dents[start_d + count], &dents[start_d],
 865                         (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
 866                 xfs_dir3_leaf_log_ents(tp, args->dp, bp_d, start_d + count,
 867                                        count + dhdr->count - 1);
 868         }
 869         /*
 870          * If the source has stale leaves, count the ones in the copy range
 871          * so we can update the header correctly.
 872          */
 873         if (shdr->stale) {
 874                 int     i;                      /* temp leaf index */
 875
 876                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 877                         if (sents[i].address ==
 878                                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 879                                 stale++;
 880                 }
 881         } else
 882                 stale = 0;
 883         /*
 884          * Copy the leaf entries from source to destination.
 885          */
 886         memcpy(&dents[start_d], &sents[start_s],
 887                 count * sizeof(xfs_dir2_leaf_entry_t));
 888         xfs_dir3_leaf_log_ents(tp, args->dp, bp_d,
 889                                start_d, start_d + count - 1);
 890
 891         /*
 892          * If there are source entries after the ones we copied,
 893          * delete the ones we copied by sliding the next ones down.
 894          */
 895         if (start_s + count < shdr->count) {
 896                 memmove(&sents[start_s], &sents[start_s + count],
 897                         count * sizeof(xfs_dir2_leaf_entry_t));
 898                 xfs_dir3_leaf_log_ents(tp, args->dp, bp_s,
 899                                        start_s, start_s + count - 1);
 900         }
 901
 902         /*
 903          * Update the headers and log them.
 904          */
 905         shdr->count -= count;
 906         shdr->stale -= stale;
 907         dhdr->count += count;
 908         dhdr->stale += stale;
 909 }
 910
 911 /*
 912  * Determine the sort order of two leaf blocks.
 913  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 914  */
 915 int                                             /* sort order */
 916 xfs_dir2_leafn_order(
 917         struct xfs_inode        *dp,
 918         struct xfs_buf          *leaf1_bp,              /* leaf1 buffer */
 919         struct xfs_buf          *leaf2_bp)              /* leaf2 buffer */
 920 {
 921         struct xfs_dir2_leaf    *leaf1 = leaf1_bp->b_addr;
 922         struct xfs_dir2_leaf    *leaf2 = leaf2_bp->b_addr;
 923         struct xfs_dir2_leaf_entry *ents1;
 924         struct xfs_dir2_leaf_entry *ents2;
 925         struct xfs_dir3_icleaf_hdr hdr1;
 926         struct xfs_dir3_icleaf_hdr hdr2;
 927
 928         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
 929         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
 930         ents1 = dp->d_ops->leaf_ents_p(leaf1);
 931         ents2 = dp->d_ops->leaf_ents_p(leaf2);
 932
 933         if (hdr1.count > 0 && hdr2.count > 0 &&
 934             (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
 935              be32_to_cpu(ents2[hdr2.count - 1].hashval) <
 936                                 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
 937                 return 1;
 938         return 0;
 939 }
 940
 941 /*
 942  * Rebalance leaf entries between two leaf blocks.
 943  * This is actually only called when the second block is new,
 944  * though the code deals with the general case.
 945  * A new entry will be inserted in one of the blocks, and that
 946  * entry is taken into account when balancing.
 947  */
 948 static void
 949 xfs_dir2_leafn_rebalance(
 950         xfs_da_state_t          *state,         /* btree cursor */
 951         xfs_da_state_blk_t      *blk1,          /* first btree block */
 952         xfs_da_state_blk_t      *blk2)          /* second btree block */
 953 {
 954         xfs_da_args_t           *args;          /* operation arguments */
 955         int                     count;          /* count (& direction) leaves */
 956         int                     isleft;         /* new goes in left leaf */
 957         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 958         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 959         int                     mid;            /* midpoint leaf index */
 960 #if defined(DEBUG) || defined(XFS_WARN)
 961         int                     oldstale;       /* old count of stale leaves */
 962 #endif
 963         int                     oldsum;         /* old total leaf count */
 964         int                     swap;           /* swapped leaf blocks */
 965         struct xfs_dir2_leaf_entry *ents1;
 966         struct xfs_dir2_leaf_entry *ents2;
 967         struct xfs_dir3_icleaf_hdr hdr1;
 968         struct xfs_dir3_icleaf_hdr hdr2;
 969         struct xfs_inode        *dp = state->args->dp;
 970
 971         args = state->args;
 972         /*
 973          * If the block order is wrong, swap the arguments.
 974          */
 975         if ((swap = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp))) {
 976                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
 977
 978                 tmp = blk1;
 979                 blk1 = blk2;
 980                 blk2 = tmp;
 981         }
 982         leaf1 = blk1->bp->b_addr;
 983         leaf2 = blk2->bp->b_addr;
 984         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
 985         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
 986         ents1 = dp->d_ops->leaf_ents_p(leaf1);
 987         ents2 = dp->d_ops->leaf_ents_p(leaf2);
 988
 989         oldsum = hdr1.count + hdr2.count;
 990 #if defined(DEBUG) || defined(XFS_WARN)
 991         oldstale = hdr1.stale + hdr2.stale;
 992 #endif
 993         mid = oldsum >> 1;
 994
 995         /*
 996          * If the old leaf count was odd then the new one will be even,
 997          * so we need to divide the new count evenly.
 998          */
 999         if (oldsum & 1) {
1000                 xfs_dahash_t    midhash;        /* middle entry hash value */
1001
1002                 if (mid >= hdr1.count)
1003                         midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1004                 else
1005                         midhash = be32_to_cpu(ents1[mid].hashval);
1006                 isleft = args->hashval <= midhash;
1007         }
1008         /*
1009          * If the old count is even then the new count is odd, so there's
1010          * no preferred side for the new entry.
1011          * Pick the left one.
1012          */
1013         else
1014                 isleft = 1;
1015         /*
1016          * Calculate moved entry count.  Positive means left-to-right,
1017          * negative means right-to-left.  Then move the entries.
1018          */
1019         count = hdr1.count - mid + (isleft == 0);
1020         if (count > 0)
1021                 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1022                                         hdr1.count - count, blk2->bp,
1023                                         &hdr2, ents2, 0, count);
1024         else if (count < 0)
1025                 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1026                                         blk1->bp, &hdr1, ents1,
1027                                         hdr1.count, count);
1028
1029         ASSERT(hdr1.count + hdr2.count == oldsum);
1030         ASSERT(hdr1.stale + hdr2.stale == oldstale);
1031
1032         /* log the changes made when moving the entries */
1033         dp->d_ops->leaf_hdr_to_disk(leaf1, &hdr1);
1034         dp->d_ops->leaf_hdr_to_disk(leaf2, &hdr2);
1035         xfs_dir3_leaf_log_header(args->trans, dp, blk1->bp);
1036         xfs_dir3_leaf_log_header(args->trans, dp, blk2->bp);
1037
1038         xfs_dir3_leaf_check(dp, blk1->bp);
1039         xfs_dir3_leaf_check(dp, blk2->bp);
1040
1041         /*
1042          * Mark whether we're inserting into the old or new leaf.
1043          */
1044         if (hdr1.count < hdr2.count)
1045                 state->inleaf = swap;
1046         else if (hdr1.count > hdr2.count)
1047                 state->inleaf = !swap;
1048         else
1049                 state->inleaf = swap ^ (blk1->index <= hdr1.count);
1050         /*
1051          * Adjust the expected index for insertion.
1052          */
1053         if (!state->inleaf)
1054                 blk2->index = blk1->index - hdr1.count;
1055
1056         /*
1057          * Finally sanity check just to make sure we are not returning a
1058          * negative index
1059          */
1060         if (blk2->index < 0) {
1061                 state->inleaf = 1;
1062                 blk2->index = 0;
1063                 xfs_alert(dp->i_mount,
1064         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1065                         __func__, blk1->index);
1066         }
1067 }
1068
1069 static int
1070 xfs_dir3_data_block_free(
1071         xfs_da_args_t           *args,
1072         struct xfs_dir2_data_hdr *hdr,
1073         struct xfs_dir2_free    *free,
1074         xfs_dir2_db_t           fdb,
1075         int                     findex,
1076         struct xfs_buf          *fbp,
1077         int                     longest)
1078 {
1079         struct xfs_trans        *tp = args->trans;
1080         int                     logfree = 0;
1081         __be16                  *bests;
1082         struct xfs_dir3_icfree_hdr freehdr;
1083         struct xfs_inode        *dp = args->dp;
1084
1085         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1086         bests = dp->d_ops->free_bests_p(free);
1087         if (hdr) {
1088                 /*
1089                  * Data block is not empty, just set the free entry to the new
1090                  * value.
1091                  */
1092                 bests[findex] = cpu_to_be16(longest);
1093                 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
1094                 return 0;
1095         }
1096
1097         /* One less used entry in the free table. */
1098         freehdr.nused--;
1099
1100         /*
1101          * If this was the last entry in the table, we can trim the table size
1102          * back.  There might be other entries at the end referring to
1103          * non-existent data blocks, get those too.
1104          */
1105         if (findex == freehdr.nvalid - 1) {
1106                 int     i;              /* free entry index */
1107
1108                 for (i = findex - 1; i >= 0; i--) {
1109                         if (bests[i] != cpu_to_be16(NULLDATAOFF))
1110                                 break;
1111                 }
1112                 freehdr.nvalid = i + 1;
1113                 logfree = 0;
1114         } else {
1115                 /* Not the last entry, just punch it out.  */
1116                 bests[findex] = cpu_to_be16(NULLDATAOFF);
1117                 logfree = 1;
1118         }
1119
1120         dp->d_ops->free_hdr_to_disk(free, &freehdr);
1121         xfs_dir2_free_log_header(tp, dp, fbp);
1122
1123         /*
1124          * If there are no useful entries left in the block, get rid of the
1125          * block if we can.
1126          */
1127         if (!freehdr.nused) {
1128                 int error;
1129
1130                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1131                 if (error == 0) {
1132                         fbp = NULL;
1133                         logfree = 0;
1134                 } else if (error != ENOSPC || args->total != 0)
1135                         return error;
1136                 /*
1137                  * It's possible to get ENOSPC if there is no
1138                  * space reservation.  In this case some one
1139                  * else will eventually get rid of this block.
1140                  */
1141         }
1142
1143         /* Log the free entry that changed, unless we got rid of it.  */
1144         if (logfree)
1145                 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
1146         return 0;
1147 }
1148
1149 /*
1150  * Remove an entry from a node directory.
1151  * This removes the leaf entry and the data entry,
1152  * and updates the free block if necessary.
1153  */
1154 static int                                      /* error */
1155 xfs_dir2_leafn_remove(
1156         xfs_da_args_t           *args,          /* operation arguments */
1157         struct xfs_buf          *bp,            /* leaf buffer */
1158         int                     index,          /* leaf entry index */
1159         xfs_da_state_blk_t      *dblk,          /* data block */
1160         int                     *rval)          /* resulting block needs join */
1161 {
1162         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1163         xfs_dir2_db_t           db;             /* data block number */
1164         struct xfs_buf          *dbp;           /* data block buffer */
1165         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1166         xfs_inode_t             *dp;            /* incore directory inode */
1167         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1168         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1169         int                     longest;        /* longest data free entry */
1170         int                     off;            /* data block entry offset */
1171         xfs_mount_t             *mp;            /* filesystem mount point */
1172         int                     needlog;        /* need to log data header */
1173         int                     needscan;       /* need to rescan data frees */
1174         xfs_trans_t             *tp;            /* transaction pointer */
1175         struct xfs_dir2_data_free *bf;          /* bestfree table */
1176         struct xfs_dir3_icleaf_hdr leafhdr;
1177         struct xfs_dir2_leaf_entry *ents;
1178
1179         trace_xfs_dir2_leafn_remove(args, index);
1180
1181         dp = args->dp;
1182         tp = args->trans;
1183         mp = dp->i_mount;
1184         leaf = bp->b_addr;
1185         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1186         ents = dp->d_ops->leaf_ents_p(leaf);
1187
1188         /*
1189          * Point to the entry we're removing.
1190          */
1191         lep = &ents[index];
1192
1193         /*
1194          * Extract the data block and offset from the entry.
1195          */
1196         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1197         ASSERT(dblk->blkno == db);
1198         off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
1199         ASSERT(dblk->index == off);
1200
1201         /*
1202          * Kill the leaf entry by marking it stale.
1203          * Log the leaf block changes.
1204          */
1205         leafhdr.stale++;
1206         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1207         xfs_dir3_leaf_log_header(tp, dp, bp);
1208
1209         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1210         xfs_dir3_leaf_log_ents(tp, dp, bp, index, index);
1211
1212         /*
1213          * Make the data entry free.  Keep track of the longest freespace
1214          * in the data block in case it changes.
1215          */
1216         dbp = dblk->bp;
1217         hdr = dbp->b_addr;
1218         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1219         bf = dp->d_ops->data_bestfree_p(hdr);
1220         longest = be16_to_cpu(bf[0].length);
1221         needlog = needscan = 0;
1222         xfs_dir2_data_make_free(tp, dp, dbp, off,
1223                 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1224         /*
1225          * Rescan the data block freespaces for bestfree.
1226          * Log the data block header if needed.
1227          */
1228         if (needscan)
1229                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1230         if (needlog)
1231                 xfs_dir2_data_log_header(tp, dp, dbp);
1232         xfs_dir3_data_check(dp, dbp);
1233         /*
1234          * If the longest data block freespace changes, need to update
1235          * the corresponding freeblock entry.
1236          */
1237         if (longest < be16_to_cpu(bf[0].length)) {
1238                 int             error;          /* error return value */
1239                 struct xfs_buf  *fbp;           /* freeblock buffer */
1240                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1241                 int             findex;         /* index in freeblock entries */
1242                 xfs_dir2_free_t *free;          /* freeblock structure */
1243
1244                 /*
1245                  * Convert the data block number to a free block,
1246                  * read in the free block.
1247                  */
1248                 fdb = dp->d_ops->db_to_fdb(mp, db);
1249                 error = xfs_dir2_free_read(tp, dp, xfs_dir2_db_to_da(mp, fdb),
1250                                            &fbp);
1251                 if (error)
1252                         return error;
1253                 free = fbp->b_addr;
1254 #ifdef DEBUG
1255         {
1256                 struct xfs_dir3_icfree_hdr freehdr;
1257                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1258                 ASSERT(freehdr.firstdb == dp->d_ops->free_max_bests(mp) *
1259                                           (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
1260         }
1261 #endif
1262                 /*
1263                  * Calculate which entry we need to fix.
1264                  */
1265                 findex = dp->d_ops->db_to_fdindex(mp, db);
1266                 longest = be16_to_cpu(bf[0].length);
1267                 /*
1268                  * If the data block is now empty we can get rid of it
1269                  * (usually).
1270                  */
1271                 if (longest == mp->m_dirblksize -
1272                                dp->d_ops->data_entry_offset) {
1273                         /*
1274                          * Try to punch out the data block.
1275                          */
1276                         error = xfs_dir2_shrink_inode(args, db, dbp);
1277                         if (error == 0) {
1278                                 dblk->bp = NULL;
1279                                 hdr = NULL;
1280                         }
1281                         /*
1282                          * We can get ENOSPC if there's no space reservation.
1283                          * In this case just drop the buffer and some one else
1284                          * will eventually get rid of the empty block.
1285                          */
1286                         else if (!(error == ENOSPC && args->total == 0))
1287                                 return error;
1288                 }
1289                 /*
1290                  * If we got rid of the data block, we can eliminate that entry
1291                  * in the free block.
1292                  */
1293                 error = xfs_dir3_data_block_free(args, hdr, free,
1294                                                  fdb, findex, fbp, longest);
1295                 if (error)
1296                         return error;
1297         }
1298
1299         xfs_dir3_leaf_check(dp, bp);
1300         /*
1301          * Return indication of whether this leaf block is empty enough
1302          * to justify trying to join it with a neighbor.
1303          */
1304         *rval = (dp->d_ops->leaf_hdr_size +
1305                  (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1306                 mp->m_dir_magicpct;
1307         return 0;
1308 }
1309
1310 /*
1311  * Split the leaf entries in the old block into old and new blocks.
1312  */
1313 int                                             /* error */
1314 xfs_dir2_leafn_split(
1315         xfs_da_state_t          *state,         /* btree cursor */
1316         xfs_da_state_blk_t      *oldblk,        /* original block */
1317         xfs_da_state_blk_t      *newblk)        /* newly created block */
1318 {
1319         xfs_da_args_t           *args;          /* operation arguments */
1320         xfs_dablk_t             blkno;          /* new leaf block number */
1321         int                     error;          /* error return value */
1322         xfs_mount_t             *mp;            /* filesystem mount point */
1323         struct xfs_inode        *dp;
1324
1325         /*
1326          * Allocate space for a new leaf node.
1327          */
1328         args = state->args;
1329         dp = args->dp;
1330         mp = dp->i_mount;
1331         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1332         error = xfs_da_grow_inode(args, &blkno);
1333         if (error) {
1334                 return error;
1335         }
1336         /*
1337          * Initialize the new leaf block.
1338          */
1339         error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(mp, blkno),
1340                                       &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1341         if (error)
1342                 return error;
1343
1344         newblk->blkno = blkno;
1345         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1346         /*
1347          * Rebalance the entries across the two leaves, link the new
1348          * block into the leaves.
1349          */
1350         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1351         error = xfs_da3_blk_link(state, oldblk, newblk);
1352         if (error) {
1353                 return error;
1354         }
1355         /*
1356          * Insert the new entry in the correct block.
1357          */
1358         if (state->inleaf)
1359                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1360         else
1361                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1362         /*
1363          * Update last hashval in each block since we added the name.
1364          */
1365         oldblk->hashval = xfs_dir2_leafn_lasthash(dp, oldblk->bp, NULL);
1366         newblk->hashval = xfs_dir2_leafn_lasthash(dp, newblk->bp, NULL);
1367         xfs_dir3_leaf_check(dp, oldblk->bp);
1368         xfs_dir3_leaf_check(dp, newblk->bp);
1369         return error;
1370 }
1371
1372 /*
1373  * Check a leaf block and its neighbors to see if the block should be
1374  * collapsed into one or the other neighbor.  Always keep the block
1375  * with the smaller block number.
1376  * If the current block is over 50% full, don't try to join it, return 0.
1377  * If the block is empty, fill in the state structure and return 2.
1378  * If it can be collapsed, fill in the state structure and return 1.
1379  * If nothing can be done, return 0.
1380  */
1381 int                                             /* error */
1382 xfs_dir2_leafn_toosmall(
1383         xfs_da_state_t          *state,         /* btree cursor */
1384         int                     *action)        /* resulting action to take */
1385 {
1386         xfs_da_state_blk_t      *blk;           /* leaf block */
1387         xfs_dablk_t             blkno;          /* leaf block number */
1388         struct xfs_buf          *bp;            /* leaf buffer */
1389         int                     bytes;          /* bytes in use */
1390         int                     count;          /* leaf live entry count */
1391         int                     error;          /* error return value */
1392         int                     forward;        /* sibling block direction */
1393         int                     i;              /* sibling counter */
1394         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1395         int                     rval;           /* result from path_shift */
1396         struct xfs_dir3_icleaf_hdr leafhdr;
1397         struct xfs_dir2_leaf_entry *ents;
1398         struct xfs_inode        *dp = state->args->dp;
1399
1400         /*
1401          * Check for the degenerate case of the block being over 50% full.
1402          * If so, it's not worth even looking to see if we might be able
1403          * to coalesce with a sibling.
1404          */
1405         blk = &state->path.blk[state->path.active - 1];
1406         leaf = blk->bp->b_addr;
1407         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1408         ents = dp->d_ops->leaf_ents_p(leaf);
1409         xfs_dir3_leaf_check(dp, blk->bp);
1410
1411         count = leafhdr.count - leafhdr.stale;
1412         bytes = dp->d_ops->leaf_hdr_size + count * sizeof(ents[0]);
1413         if (bytes > (state->blocksize >> 1)) {
1414                 /*
1415                  * Blk over 50%, don't try to join.
1416                  */
1417                 *action = 0;
1418                 return 0;
1419         }
1420         /*
1421          * Check for the degenerate case of the block being empty.
1422          * If the block is empty, we'll simply delete it, no need to
1423          * coalesce it with a sibling block.  We choose (arbitrarily)
1424          * to merge with the forward block unless it is NULL.
1425          */
1426         if (count == 0) {
1427                 /*
1428                  * Make altpath point to the block we want to keep and
1429                  * path point to the block we want to drop (this one).
1430                  */
1431                 forward = (leafhdr.forw != 0);
1432                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1433                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1434                         &rval);
1435                 if (error)
1436                         return error;
1437                 *action = rval ? 2 : 0;
1438                 return 0;
1439         }
1440         /*
1441          * Examine each sibling block to see if we can coalesce with
1442          * at least 25% free space to spare.  We need to figure out
1443          * whether to merge with the forward or the backward block.
1444          * We prefer coalescing with the lower numbered sibling so as
1445          * to shrink a directory over time.
1446          */
1447         forward = leafhdr.forw < leafhdr.back;
1448         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1449                 struct xfs_dir3_icleaf_hdr hdr2;
1450
1451                 blkno = forward ? leafhdr.forw : leafhdr.back;
1452                 if (blkno == 0)
1453                         continue;
1454                 /*
1455                  * Read the sibling leaf block.
1456                  */
1457                 error = xfs_dir3_leafn_read(state->args->trans, dp,
1458                                             blkno, -1, &bp);
1459                 if (error)
1460                         return error;
1461
1462                 /*
1463                  * Count bytes in the two blocks combined.
1464                  */
1465                 count = leafhdr.count - leafhdr.stale;
1466                 bytes = state->blocksize - (state->blocksize >> 2);
1467
1468                 leaf = bp->b_addr;
1469                 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf);
1470                 ents = dp->d_ops->leaf_ents_p(leaf);
1471                 count += hdr2.count - hdr2.stale;
1472                 bytes -= count * sizeof(ents[0]);
1473
1474                 /*
1475                  * Fits with at least 25% to spare.
1476                  */
1477                 if (bytes >= 0)
1478                         break;
1479                 xfs_trans_brelse(state->args->trans, bp);
1480         }
1481         /*
1482          * Didn't like either block, give up.
1483          */
1484         if (i >= 2) {
1485                 *action = 0;
1486                 return 0;
1487         }
1488
1489         /*
1490          * Make altpath point to the block we want to keep (the lower
1491          * numbered block) and path point to the block we want to drop.
1492          */
1493         memcpy(&state->altpath, &state->path, sizeof(state->path));
1494         if (blkno < blk->blkno)
1495                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1496                         &rval);
1497         else
1498                 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1499                         &rval);
1500         if (error) {
1501                 return error;
1502         }
1503         *action = rval ? 0 : 1;
1504         return 0;
1505 }
1506
1507 /*
1508  * Move all the leaf entries from drop_blk to save_blk.
1509  * This is done as part of a join operation.
1510  */
1511 void
1512 xfs_dir2_leafn_unbalance(
1513         xfs_da_state_t          *state,         /* cursor */
1514         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1515         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1516 {
1517         xfs_da_args_t           *args;          /* operation arguments */
1518         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1519         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1520         struct xfs_dir3_icleaf_hdr savehdr;
1521         struct xfs_dir3_icleaf_hdr drophdr;
1522         struct xfs_dir2_leaf_entry *sents;
1523         struct xfs_dir2_leaf_entry *dents;
1524         struct xfs_inode        *dp = state->args->dp;
1525
1526         args = state->args;
1527         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1528         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1529         drop_leaf = drop_blk->bp->b_addr;
1530         save_leaf = save_blk->bp->b_addr;
1531
1532         dp->d_ops->leaf_hdr_from_disk(&savehdr, save_leaf);
1533         dp->d_ops->leaf_hdr_from_disk(&drophdr, drop_leaf);
1534         sents = dp->d_ops->leaf_ents_p(save_leaf);
1535         dents = dp->d_ops->leaf_ents_p(drop_leaf);
1536
1537         /*
1538          * If there are any stale leaf entries, take this opportunity
1539          * to purge them.
1540          */
1541         if (drophdr.stale)
1542                 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1543         if (savehdr.stale)
1544                 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1545
1546         /*
1547          * Move the entries from drop to the appropriate end of save.
1548          */
1549         drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1550         if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1551                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1552                                         save_blk->bp, &savehdr, sents, 0,
1553                                         drophdr.count);
1554         else
1555                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1556                                         save_blk->bp, &savehdr, sents,
1557                                         savehdr.count, drophdr.count);
1558         save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1559
1560         /* log the changes made when moving the entries */
1561         dp->d_ops->leaf_hdr_to_disk(save_leaf, &savehdr);
1562         dp->d_ops->leaf_hdr_to_disk(drop_leaf, &drophdr);
1563         xfs_dir3_leaf_log_header(args->trans, dp, save_blk->bp);
1564         xfs_dir3_leaf_log_header(args->trans, dp, drop_blk->bp);
1565
1566         xfs_dir3_leaf_check(dp, save_blk->bp);
1567         xfs_dir3_leaf_check(dp, drop_blk->bp);
1568 }
1569
1570 /*
1571  * Top-level node form directory addname routine.
1572  */
1573 int                                             /* error */
1574 xfs_dir2_node_addname(
1575         xfs_da_args_t           *args)          /* operation arguments */
1576 {
1577         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1578         int                     error;          /* error return value */
1579         int                     rval;           /* sub-return value */
1580         xfs_da_state_t          *state;         /* btree cursor */
1581
1582         trace_xfs_dir2_node_addname(args);
1583
1584         /*
1585          * Allocate and initialize the state (btree cursor).
1586          */
1587         state = xfs_da_state_alloc();
1588         state->args = args;
1589         state->mp = args->dp->i_mount;
1590         state->blocksize = state->mp->m_dirblksize;
1591         state->node_ents = state->mp->m_dir_node_ents;
1592         /*
1593          * Look up the name.  We're not supposed to find it, but
1594          * this gives us the insertion point.
1595          */
1596         error = xfs_da3_node_lookup_int(state, &rval);
1597         if (error)
1598                 rval = error;
1599         if (rval != ENOENT) {
1600                 goto done;
1601         }
1602         /*
1603          * Add the data entry to a data block.
1604          * Extravalid is set to a freeblock found by lookup.
1605          */
1606         rval = xfs_dir2_node_addname_int(args,
1607                 state->extravalid ? &state->extrablk : NULL);
1608         if (rval) {
1609                 goto done;
1610         }
1611         blk = &state->path.blk[state->path.active - 1];
1612         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1613         /*
1614          * Add the new leaf entry.
1615          */
1616         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1617         if (rval == 0) {
1618                 /*
1619                  * It worked, fix the hash values up the btree.
1620                  */
1621                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1622                         xfs_da3_fixhashpath(state, &state->path);
1623         } else {
1624                 /*
1625                  * It didn't work, we need to split the leaf block.
1626                  */
1627                 if (args->total == 0) {
1628                         ASSERT(rval == ENOSPC);
1629                         goto done;
1630                 }
1631                 /*
1632                  * Split the leaf block and insert the new entry.
1633                  */
1634                 rval = xfs_da3_split(state);
1635         }
1636 done:
1637         xfs_da_state_free(state);
1638         return rval;
1639 }
1640
1641 /*
1642  * Add the data entry for a node-format directory name addition.
1643  * The leaf entry is added in xfs_dir2_leafn_add.
1644  * We may enter with a freespace block that the lookup found.
1645  */
1646 static int                                      /* error */
1647 xfs_dir2_node_addname_int(
1648         xfs_da_args_t           *args,          /* operation arguments */
1649         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1650 {
1651         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1652         xfs_dir2_db_t           dbno;           /* data block number */
1653         struct xfs_buf          *dbp;           /* data block buffer */
1654         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1655         xfs_inode_t             *dp;            /* incore directory inode */
1656         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1657         int                     error;          /* error return value */
1658         xfs_dir2_db_t           fbno;           /* freespace block number */
1659         struct xfs_buf          *fbp;           /* freespace buffer */
1660         int                     findex;         /* freespace entry index */
1661         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1662         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1663         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1664         int                     length;         /* length of the new entry */
1665         int                     logfree;        /* need to log free entry */
1666         xfs_mount_t             *mp;            /* filesystem mount point */
1667         int                     needlog;        /* need to log data header */
1668         int                     needscan;       /* need to rescan data frees */
1669         __be16                  *tagp;          /* data entry tag pointer */
1670         xfs_trans_t             *tp;            /* transaction pointer */
1671         __be16                  *bests;
1672         struct xfs_dir3_icfree_hdr freehdr;
1673         struct xfs_dir2_data_free *bf;
1674
1675         dp = args->dp;
1676         mp = dp->i_mount;
1677         tp = args->trans;
1678         length = dp->d_ops->data_entsize(args->namelen);
1679         /*
1680          * If we came in with a freespace block that means that lookup
1681          * found an entry with our hash value.  This is the freespace
1682          * block for that data entry.
1683          */
1684         if (fblk) {
1685                 fbp = fblk->bp;
1686                 /*
1687                  * Remember initial freespace block number.
1688                  */
1689                 ifbno = fblk->blkno;
1690                 free = fbp->b_addr;
1691                 findex = fblk->index;
1692                 bests = dp->d_ops->free_bests_p(free);
1693                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1694
1695                 /*
1696                  * This means the free entry showed that the data block had
1697                  * space for our entry, so we remembered it.
1698                  * Use that data block.
1699                  */
1700                 if (findex >= 0) {
1701                         ASSERT(findex < freehdr.nvalid);
1702                         ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1703                         ASSERT(be16_to_cpu(bests[findex]) >= length);
1704                         dbno = freehdr.firstdb + findex;
1705                 } else {
1706                         /*
1707                          * The data block looked at didn't have enough room.
1708                          * We'll start at the beginning of the freespace entries.
1709                          */
1710                         dbno = -1;
1711                         findex = 0;
1712                 }
1713         } else {
1714                 /*
1715                  * Didn't come in with a freespace block, so no data block.
1716                  */
1717                 ifbno = dbno = -1;
1718                 fbp = NULL;
1719                 findex = 0;
1720         }
1721
1722         /*
1723          * If we don't have a data block yet, we're going to scan the
1724          * freespace blocks looking for one.  Figure out what the
1725          * highest freespace block number is.
1726          */
1727         if (dbno == -1) {
1728                 xfs_fileoff_t   fo;             /* freespace block number */
1729
1730                 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1731                         return error;
1732                 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1733                 fbno = ifbno;
1734         }
1735         /*
1736          * While we haven't identified a data block, search the freeblock
1737          * data for a good data block.  If we find a null freeblock entry,
1738          * indicating a hole in the data blocks, remember that.
1739          */
1740         while (dbno == -1) {
1741                 /*
1742                  * If we don't have a freeblock in hand, get the next one.
1743                  */
1744                 if (fbp == NULL) {
1745                         /*
1746                          * Happens the first time through unless lookup gave
1747                          * us a freespace block to start with.
1748                          */
1749                         if (++fbno == 0)
1750                                 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1751                         /*
1752                          * If it's ifbno we already looked at it.
1753                          */
1754                         if (fbno == ifbno)
1755                                 fbno++;
1756                         /*
1757                          * If it's off the end we're done.
1758                          */
1759                         if (fbno >= lastfbno)
1760                                 break;
1761                         /*
1762                          * Read the block.  There can be holes in the
1763                          * freespace blocks, so this might not succeed.
1764                          * This should be really rare, so there's no reason
1765                          * to avoid it.
1766                          */
1767                         error = xfs_dir2_free_try_read(tp, dp,
1768                                                 xfs_dir2_db_to_da(mp, fbno),
1769                                                 &fbp);
1770                         if (error)
1771                                 return error;
1772                         if (!fbp)
1773                                 continue;
1774                         free = fbp->b_addr;
1775                         findex = 0;
1776                 }
1777                 /*
1778                  * Look at the current free entry.  Is it good enough?
1779                  *
1780                  * The bests initialisation should be where the bufer is read in
1781                  * the above branch. But gcc is too stupid to realise that bests
1782                  * and the freehdr are actually initialised if they are placed
1783                  * there, so we have to do it here to avoid warnings. Blech.
1784                  */
1785                 bests = dp->d_ops->free_bests_p(free);
1786                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1787                 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1788                     be16_to_cpu(bests[findex]) >= length)
1789                         dbno = freehdr.firstdb + findex;
1790                 else {
1791                         /*
1792                          * Are we done with the freeblock?
1793                          */
1794                         if (++findex == freehdr.nvalid) {
1795                                 /*
1796                                  * Drop the block.
1797                                  */
1798                                 xfs_trans_brelse(tp, fbp);
1799                                 fbp = NULL;
1800                                 if (fblk && fblk->bp)
1801                                         fblk->bp = NULL;
1802                         }
1803                 }
1804         }
1805         /*
1806          * If we don't have a data block, we need to allocate one and make
1807          * the freespace entries refer to it.
1808          */
1809         if (unlikely(dbno == -1)) {
1810                 /*
1811                  * Not allowed to allocate, return failure.
1812                  */
1813                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1814                         return XFS_ERROR(ENOSPC);
1815
1816                 /*
1817                  * Allocate and initialize the new data block.
1818                  */
1819                 if (unlikely((error = xfs_dir2_grow_inode(args,
1820                                                          XFS_DIR2_DATA_SPACE,
1821                                                          &dbno)) ||
1822                     (error = xfs_dir3_data_init(args, dbno, &dbp))))
1823                         return error;
1824
1825                 /*
1826                  * If (somehow) we have a freespace block, get rid of it.
1827                  */
1828                 if (fbp)
1829                         xfs_trans_brelse(tp, fbp);
1830                 if (fblk && fblk->bp)
1831                         fblk->bp = NULL;
1832
1833                 /*
1834                  * Get the freespace block corresponding to the data block
1835                  * that was just allocated.
1836                  */
1837                 fbno = dp->d_ops->db_to_fdb(mp, dbno);
1838                 error = xfs_dir2_free_try_read(tp, dp,
1839                                                xfs_dir2_db_to_da(mp, fbno),
1840                                                &fbp);
1841                 if (error)
1842                         return error;
1843
1844                 /*
1845                  * If there wasn't a freespace block, the read will
1846                  * return a NULL fbp.  Allocate and initialize a new one.
1847                  */
1848                 if (!fbp) {
1849                         error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1850                                                     &fbno);
1851                         if (error)
1852                                 return error;
1853
1854                         if (unlikely(dp->d_ops->db_to_fdb(mp, dbno) != fbno)) {
1855                                 xfs_alert(mp,
1856                         "%s: dir ino %llu needed freesp block %lld for\n"
1857                         "  data block %lld, got %lld ifbno %llu lastfbno %d",
1858                                         __func__, (unsigned long long)dp->i_ino,
1859                                         (long long)dp->d_ops->db_to_fdb(mp, dbno),
1860                                         (long long)dbno, (long long)fbno,
1861                                         (unsigned long long)ifbno, lastfbno);
1862                                 if (fblk) {
1863                                         xfs_alert(mp,
1864                                 " fblk 0x%p blkno %llu index %d magic 0x%x",
1865                                                 fblk,
1866                                                 (unsigned long long)fblk->blkno,
1867                                                 fblk->index,
1868                                                 fblk->magic);
1869                                 } else {
1870                                         xfs_alert(mp, " ... fblk is NULL");
1871                                 }
1872                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1873                                                  XFS_ERRLEVEL_LOW, mp);
1874                                 return XFS_ERROR(EFSCORRUPTED);
1875                         }
1876
1877                         /*
1878                          * Get a buffer for the new block.
1879                          */
1880                         error = xfs_dir3_free_get_buf(tp, dp, fbno, &fbp);
1881                         if (error)
1882                                 return error;
1883                         free = fbp->b_addr;
1884                         bests = dp->d_ops->free_bests_p(free);
1885                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1886
1887                         /*
1888                          * Remember the first slot as our empty slot.
1889                          */
1890                         freehdr.firstdb = (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1891                                         dp->d_ops->free_max_bests(mp);
1892                 } else {
1893                         free = fbp->b_addr;
1894                         bests = dp->d_ops->free_bests_p(free);
1895                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1896                 }
1897
1898                 /*
1899                  * Set the freespace block index from the data block number.
1900                  */
1901                 findex = dp->d_ops->db_to_fdindex(mp, dbno);
1902                 /*
1903                  * If it's after the end of the current entries in the
1904                  * freespace block, extend that table.
1905                  */
1906                 if (findex >= freehdr.nvalid) {
1907                         ASSERT(findex < dp->d_ops->free_max_bests(mp));
1908                         freehdr.nvalid = findex + 1;
1909                         /*
1910                          * Tag new entry so nused will go up.
1911                          */
1912                         bests[findex] = cpu_to_be16(NULLDATAOFF);
1913                 }
1914                 /*
1915                  * If this entry was for an empty data block
1916                  * (this should always be true) then update the header.
1917                  */
1918                 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1919                         freehdr.nused++;
1920                         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
1921                         xfs_dir2_free_log_header(tp, dp, fbp);
1922                 }
1923                 /*
1924                  * Update the real value in the table.
1925                  * We haven't allocated the data entry yet so this will
1926                  * change again.
1927                  */
1928                 hdr = dbp->b_addr;
1929                 bf = dp->d_ops->data_bestfree_p(hdr);
1930                 bests[findex] = bf[0].length;
1931                 logfree = 1;
1932         }
1933         /*
1934          * We had a data block so we don't have to make a new one.
1935          */
1936         else {
1937                 /*
1938                  * If just checking, we succeeded.
1939                  */
1940                 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1941                         return 0;
1942
1943                 /*
1944                  * Read the data block in.
1945                  */
1946                 error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1947                                            -1, &dbp);
1948                 if (error)
1949                         return error;
1950                 hdr = dbp->b_addr;
1951                 bf = dp->d_ops->data_bestfree_p(hdr);
1952                 logfree = 0;
1953         }
1954         ASSERT(be16_to_cpu(bf[0].length) >= length);
1955         /*
1956          * Point to the existing unused space.
1957          */
1958         dup = (xfs_dir2_data_unused_t *)
1959               ((char *)hdr + be16_to_cpu(bf[0].offset));
1960         needscan = needlog = 0;
1961         /*
1962          * Mark the first part of the unused space, inuse for us.
1963          */
1964         xfs_dir2_data_use_free(tp, dp, dbp, dup,
1965                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
1966                 &needlog, &needscan);
1967         /*
1968          * Fill in the new entry and log it.
1969          */
1970         dep = (xfs_dir2_data_entry_t *)dup;
1971         dep->inumber = cpu_to_be64(args->inumber);
1972         dep->namelen = args->namelen;
1973         memcpy(dep->name, args->name, dep->namelen);
1974         dp->d_ops->data_put_ftype(dep, args->filetype);
1975         tagp = dp->d_ops->data_entry_tag_p(dep);
1976         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1977         xfs_dir2_data_log_entry(tp, dp, dbp, dep);
1978         /*
1979          * Rescan the block for bestfree if needed.
1980          */
1981         if (needscan)
1982                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1983         /*
1984          * Log the data block header if needed.
1985          */
1986         if (needlog)
1987                 xfs_dir2_data_log_header(tp, dp, dbp);
1988         /*
1989          * If the freespace entry is now wrong, update it.
1990          */
1991         bests = dp->d_ops->free_bests_p(free); /* gcc is so stupid */
1992         if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
1993                 bests[findex] = bf[0].length;
1994                 logfree = 1;
1995         }
1996         /*
1997          * Log the freespace entry if needed.
1998          */
1999         if (logfree)
2000                 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
2001         /*
2002          * Return the data block and offset in args, then drop the data block.
2003          */
2004         args->blkno = (xfs_dablk_t)dbno;
2005         args->index = be16_to_cpu(*tagp);
2006         return 0;
2007 }
2008
2009 /*
2010  * Lookup an entry in a node-format directory.
2011  * All the real work happens in xfs_da3_node_lookup_int.
2012  * The only real output is the inode number of the entry.
2013  */
2014 int                                             /* error */
2015 xfs_dir2_node_lookup(
2016         xfs_da_args_t   *args)                  /* operation arguments */
2017 {
2018         int             error;                  /* error return value */
2019         int             i;                      /* btree level */
2020         int             rval;                   /* operation return value */
2021         xfs_da_state_t  *state;                 /* btree cursor */
2022
2023         trace_xfs_dir2_node_lookup(args);
2024
2025         /*
2026          * Allocate and initialize the btree cursor.
2027          */
2028         state = xfs_da_state_alloc();
2029         state->args = args;
2030         state->mp = args->dp->i_mount;
2031         state->blocksize = state->mp->m_dirblksize;
2032         state->node_ents = state->mp->m_dir_node_ents;
2033         /*
2034          * Fill in the path to the entry in the cursor.
2035          */
2036         error = xfs_da3_node_lookup_int(state, &rval);
2037         if (error)
2038                 rval = error;
2039         else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
2040                 /* If a CI match, dup the actual name and return EEXIST */
2041                 xfs_dir2_data_entry_t   *dep;
2042
2043                 dep = (xfs_dir2_data_entry_t *)
2044                         ((char *)state->extrablk.bp->b_addr +
2045                                                  state->extrablk.index);
2046                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2047         }
2048         /*
2049          * Release the btree blocks and leaf block.
2050          */
2051         for (i = 0; i < state->path.active; i++) {
2052                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2053                 state->path.blk[i].bp = NULL;
2054         }
2055         /*
2056          * Release the data block if we have it.
2057          */
2058         if (state->extravalid && state->extrablk.bp) {
2059                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2060                 state->extrablk.bp = NULL;
2061         }
2062         xfs_da_state_free(state);
2063         return rval;
2064 }
2065
2066 /*
2067  * Remove an entry from a node-format directory.
2068  */
2069 int                                             /* error */
2070 xfs_dir2_node_removename(
2071         struct xfs_da_args      *args)          /* operation arguments */
2072 {
2073         struct xfs_da_state_blk *blk;           /* leaf block */
2074         int                     error;          /* error return value */
2075         int                     rval;           /* operation return value */
2076         struct xfs_da_state     *state;         /* btree cursor */
2077
2078         trace_xfs_dir2_node_removename(args);
2079
2080         /*
2081          * Allocate and initialize the btree cursor.
2082          */
2083         state = xfs_da_state_alloc();
2084         state->args = args;
2085         state->mp = args->dp->i_mount;
2086         state->blocksize = state->mp->m_dirblksize;
2087         state->node_ents = state->mp->m_dir_node_ents;
2088
2089         /* Look up the entry we're deleting, set up the cursor. */
2090         error = xfs_da3_node_lookup_int(state, &rval);
2091         if (error)
2092                 goto out_free;
2093
2094         /* Didn't find it, upper layer screwed up. */
2095         if (rval != EEXIST) {
2096                 error = rval;
2097                 goto out_free;
2098         }
2099
2100         blk = &state->path.blk[state->path.active - 1];
2101         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2102         ASSERT(state->extravalid);
2103         /*
2104          * Remove the leaf and data entries.
2105          * Extrablk refers to the data block.
2106          */
2107         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2108                 &state->extrablk, &rval);
2109         if (error)
2110                 goto out_free;
2111         /*
2112          * Fix the hash values up the btree.
2113          */
2114         xfs_da3_fixhashpath(state, &state->path);
2115         /*
2116          * If we need to join leaf blocks, do it.
2117          */
2118         if (rval && state->path.active > 1)
2119                 error = xfs_da3_join(state);
2120         /*
2121          * If no errors so far, try conversion to leaf format.
2122          */
2123         if (!error)
2124                 error = xfs_dir2_node_to_leaf(state);
2125 out_free:
2126         xfs_da_state_free(state);
2127         return error;
2128 }
2129
2130 /*
2131  * Replace an entry's inode number in a node-format directory.
2132  */
2133 int                                             /* error */
2134 xfs_dir2_node_replace(
2135         xfs_da_args_t           *args)          /* operation arguments */
2136 {
2137         xfs_da_state_blk_t      *blk;           /* leaf block */
2138         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
2139         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
2140         int                     error;          /* error return value */
2141         int                     i;              /* btree level */
2142         xfs_ino_t               inum;           /* new inode number */
2143         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
2144         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
2145         int                     rval;           /* internal return value */
2146         xfs_da_state_t          *state;         /* btree cursor */
2147
2148         trace_xfs_dir2_node_replace(args);
2149
2150         /*
2151          * Allocate and initialize the btree cursor.
2152          */
2153         state = xfs_da_state_alloc();
2154         state->args = args;
2155         state->mp = args->dp->i_mount;
2156         state->blocksize = state->mp->m_dirblksize;
2157         state->node_ents = state->mp->m_dir_node_ents;
2158         inum = args->inumber;
2159         /*
2160          * Lookup the entry to change in the btree.
2161          */
2162         error = xfs_da3_node_lookup_int(state, &rval);
2163         if (error) {
2164                 rval = error;
2165         }
2166         /*
2167          * It should be found, since the vnodeops layer has looked it up
2168          * and locked it.  But paranoia is good.
2169          */
2170         if (rval == EEXIST) {
2171                 struct xfs_dir2_leaf_entry *ents;
2172                 /*
2173                  * Find the leaf entry.
2174                  */
2175                 blk = &state->path.blk[state->path.active - 1];
2176                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2177                 leaf = blk->bp->b_addr;
2178                 ents = args->dp->d_ops->leaf_ents_p(leaf);
2179                 lep = &ents[blk->index];
2180                 ASSERT(state->extravalid);
2181                 /*
2182                  * Point to the data entry.
2183                  */
2184                 hdr = state->extrablk.bp->b_addr;
2185                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2186                        hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2187                 dep = (xfs_dir2_data_entry_t *)
2188                       ((char *)hdr +
2189                        xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
2190                 ASSERT(inum != be64_to_cpu(dep->inumber));
2191                 /*
2192                  * Fill in the new inode number and log the entry.
2193                  */
2194                 dep->inumber = cpu_to_be64(inum);
2195                 args->dp->d_ops->data_put_ftype(dep, args->filetype);
2196                 xfs_dir2_data_log_entry(args->trans, args->dp,
2197                                         state->extrablk.bp, dep);
2198                 rval = 0;
2199         }
2200         /*
2201          * Didn't find it, and we're holding a data block.  Drop it.
2202          */
2203         else if (state->extravalid) {
2204                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2205                 state->extrablk.bp = NULL;
2206         }
2207         /*
2208          * Release all the buffers in the cursor.
2209          */
2210         for (i = 0; i < state->path.active; i++) {
2211                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2212                 state->path.blk[i].bp = NULL;
2213         }
2214         xfs_da_state_free(state);
2215         return rval;
2216 }
2217
2218 /*
2219  * Trim off a trailing empty freespace block.
2220  * Return (in rvalp) 1 if we did it, 0 if not.
2221  */
2222 int                                             /* error */
2223 xfs_dir2_node_trim_free(
2224         xfs_da_args_t           *args,          /* operation arguments */
2225         xfs_fileoff_t           fo,             /* free block number */
2226         int                     *rvalp)         /* out: did something */
2227 {
2228         struct xfs_buf          *bp;            /* freespace buffer */
2229         xfs_inode_t             *dp;            /* incore directory inode */
2230         int                     error;          /* error return code */
2231         xfs_dir2_free_t         *free;          /* freespace structure */
2232         xfs_mount_t             *mp;            /* filesystem mount point */
2233         xfs_trans_t             *tp;            /* transaction pointer */
2234         struct xfs_dir3_icfree_hdr freehdr;
2235
2236         dp = args->dp;
2237         mp = dp->i_mount;
2238         tp = args->trans;
2239         /*
2240          * Read the freespace block.
2241          */
2242         error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2243         if (error)
2244                 return error;
2245         /*
2246          * There can be holes in freespace.  If fo is a hole, there's
2247          * nothing to do.
2248          */
2249         if (!bp)
2250                 return 0;
2251         free = bp->b_addr;
2252         dp->d_ops->free_hdr_from_disk(&freehdr, free);
2253
2254         /*
2255          * If there are used entries, there's nothing to do.
2256          */
2257         if (freehdr.nused > 0) {
2258                 xfs_trans_brelse(tp, bp);
2259                 *rvalp = 0;
2260                 return 0;
2261         }
2262         /*
2263          * Blow the block away.
2264          */
2265         if ((error =
2266             xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
2267                     bp))) {
2268                 /*
2269                  * Can't fail with ENOSPC since that only happens with no
2270                  * space reservation, when breaking up an extent into two
2271                  * pieces.  This is the last block of an extent.
2272                  */
2273                 ASSERT(error != ENOSPC);
2274                 xfs_trans_brelse(tp, bp);
2275                 return error;
2276         }
2277         /*
2278          * Return that we succeeded.
2279          */
2280         *rvalp = 1;
2281         return 0;
2282 }