fs/xfs/xfs_dir2_node.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_inode.h"
  30 #include "xfs_bmap.h"
  31 #include "xfs_dir2.h"
  32 #include "xfs_dir2_priv.h"
  33 #include "xfs_error.h"
  34 #include "xfs_trace.h"
  35 #include "xfs_trans.h"
  36 #include "xfs_buf_item.h"
  37 #include "xfs_cksum.h"
  38
  39 /*
  40  * Function declarations.
  41  */
  42 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
  43                               int index);
  44 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  45                                      xfs_da_state_blk_t *blk1,
  46                                      xfs_da_state_blk_t *blk2);
  47 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
  48                                  int index, xfs_da_state_blk_t *dblk,
  49                                  int *rval);
  50 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  51                                      xfs_da_state_blk_t *fblk);
  52
  53 /*
  54  * Check internal consistency of a leafn block.
  55  */
  56 #ifdef DEBUG
  57 #define xfs_dir3_leaf_check(dp, bp) \
  58 do { \
  59         if (!xfs_dir3_leafn_check((dp), (bp))) \
  60                 ASSERT(0); \
  61 } while (0);
  62
  63 static bool
  64 xfs_dir3_leafn_check(
  65         struct xfs_inode        *dp,
  66         struct xfs_buf          *bp)
  67 {
  68         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  69         struct xfs_dir3_icleaf_hdr leafhdr;
  70
  71         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
  72
  73         if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
  74                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  75                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  76                         return false;
  77         } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
  78                 return false;
  79
  80         return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
  81 }
  82 #else
  83 #define xfs_dir3_leaf_check(dp, bp)
  84 #endif
  85
  86 static bool
  87 xfs_dir3_free_verify(
  88         struct xfs_buf          *bp)
  89 {
  90         struct xfs_mount        *mp = bp->b_target->bt_mount;
  91         struct xfs_dir2_free_hdr *hdr = bp->b_addr;
  92
  93         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  94                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
  95
  96                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_FREE_MAGIC))
  97                         return false;
  98                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
  99                         return false;
 100                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 101                         return false;
 102         } else {
 103                 if (hdr->magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC))
 104                         return false;
 105         }
 106
 107         /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
 108
 109         return true;
 110 }
 111
 112 static void
 113 xfs_dir3_free_read_verify(
 114         struct xfs_buf  *bp)
 115 {
 116         struct xfs_mount        *mp = bp->b_target->bt_mount;
 117
 118         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 119             !xfs_buf_verify_cksum(bp, XFS_DIR3_FREE_CRC_OFF))
 120                 xfs_buf_ioerror(bp, EFSBADCRC);
 121         else if (!xfs_dir3_free_verify(bp))
 122                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 123
 124         if (bp->b_error)
 125                 xfs_verifier_error(bp);
 126 }
 127
 128 static void
 129 xfs_dir3_free_write_verify(
 130         struct xfs_buf  *bp)
 131 {
 132         struct xfs_mount        *mp = bp->b_target->bt_mount;
 133         struct xfs_buf_log_item *bip = bp->b_fspriv;
 134         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 135
 136         if (!xfs_dir3_free_verify(bp)) {
 137                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 138                 xfs_verifier_error(bp);
 139                 return;
 140         }
 141
 142         if (!xfs_sb_version_hascrc(&mp->m_sb))
 143                 return;
 144
 145         if (bip)
 146                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 147
 148         xfs_buf_update_cksum(bp, XFS_DIR3_FREE_CRC_OFF);
 149 }
 150
 151 const struct xfs_buf_ops xfs_dir3_free_buf_ops = {
 152         .verify_read = xfs_dir3_free_read_verify,
 153         .verify_write = xfs_dir3_free_write_verify,
 154 };
 155
 156
 157 static int
 158 __xfs_dir3_free_read(
 159         struct xfs_trans        *tp,
 160         struct xfs_inode        *dp,
 161         xfs_dablk_t             fbno,
 162         xfs_daddr_t             mappedbno,
 163         struct xfs_buf          **bpp)
 164 {
 165         int                     err;
 166
 167         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 168                                 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
 169
 170         /* try read returns without an error or *bpp if it lands in a hole */
 171         if (!err && tp && *bpp)
 172                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
 173         return err;
 174 }
 175
 176 int
 177 xfs_dir2_free_read(
 178         struct xfs_trans        *tp,
 179         struct xfs_inode        *dp,
 180         xfs_dablk_t             fbno,
 181         struct xfs_buf          **bpp)
 182 {
 183         return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
 184 }
 185
 186 static int
 187 xfs_dir2_free_try_read(
 188         struct xfs_trans        *tp,
 189         struct xfs_inode        *dp,
 190         xfs_dablk_t             fbno,
 191         struct xfs_buf          **bpp)
 192 {
 193         return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
 194 }
 195
 196 static int
 197 xfs_dir3_free_get_buf(
 198         xfs_da_args_t           *args,
 199         xfs_dir2_db_t           fbno,
 200         struct xfs_buf          **bpp)
 201 {
 202         struct xfs_trans        *tp = args->trans;
 203         struct xfs_inode        *dp = args->dp;
 204         struct xfs_mount        *mp = dp->i_mount;
 205         struct xfs_buf          *bp;
 206         int                     error;
 207         struct xfs_dir3_icfree_hdr hdr;
 208
 209         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, fbno),
 210                                    -1, &bp, XFS_DATA_FORK);
 211         if (error)
 212                 return error;
 213
 214         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
 215         bp->b_ops = &xfs_dir3_free_buf_ops;
 216
 217         /*
 218          * Initialize the new block to be empty, and remember
 219          * its first slot as our empty slot.
 220          */
 221         memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
 222         memset(&hdr, 0, sizeof(hdr));
 223
 224         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 225                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 226
 227                 hdr.magic = XFS_DIR3_FREE_MAGIC;
 228
 229                 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
 230                 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
 231                 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_uuid);
 232         } else
 233                 hdr.magic = XFS_DIR2_FREE_MAGIC;
 234         dp->d_ops->free_hdr_to_disk(bp->b_addr, &hdr);
 235         *bpp = bp;
 236         return 0;
 237 }
 238
 239 /*
 240  * Log entries from a freespace block.
 241  */
 242 STATIC void
 243 xfs_dir2_free_log_bests(
 244         struct xfs_trans        *tp,
 245         struct xfs_inode        *dp,
 246         struct xfs_buf          *bp,
 247         int                     first,          /* first entry to log */
 248         int                     last)           /* last entry to log */
 249 {
 250         xfs_dir2_free_t         *free;          /* freespace structure */
 251         __be16                  *bests;
 252
 253         free = bp->b_addr;
 254         bests = dp->d_ops->free_bests_p(free);
 255         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 256                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 257         xfs_trans_log_buf(tp, bp,
 258                 (uint)((char *)&bests[first] - (char *)free),
 259                 (uint)((char *)&bests[last] - (char *)free +
 260                        sizeof(bests[0]) - 1));
 261 }
 262
 263 /*
 264  * Log header from a freespace block.
 265  */
 266 static void
 267 xfs_dir2_free_log_header(
 268         struct xfs_trans        *tp,
 269         struct xfs_inode        *dp,
 270         struct xfs_buf          *bp)
 271 {
 272 #ifdef DEBUG
 273         xfs_dir2_free_t         *free;          /* freespace structure */
 274
 275         free = bp->b_addr;
 276         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 277                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 278 #endif
 279         xfs_trans_log_buf(tp, bp, 0, dp->d_ops->free_hdr_size - 1);
 280 }
 281
 282 /*
 283  * Convert a leaf-format directory to a node-format directory.
 284  * We need to change the magic number of the leaf block, and copy
 285  * the freespace table out of the leaf block into its own block.
 286  */
 287 int                                             /* error */
 288 xfs_dir2_leaf_to_node(
 289         xfs_da_args_t           *args,          /* operation arguments */
 290         struct xfs_buf          *lbp)           /* leaf buffer */
 291 {
 292         xfs_inode_t             *dp;            /* incore directory inode */
 293         int                     error;          /* error return value */
 294         struct xfs_buf          *fbp;           /* freespace buffer */
 295         xfs_dir2_db_t           fdb;            /* freespace block number */
 296         xfs_dir2_free_t         *free;          /* freespace structure */
 297         __be16                  *from;          /* pointer to freespace entry */
 298         int                     i;              /* leaf freespace index */
 299         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 300         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 301         xfs_mount_t             *mp;            /* filesystem mount point */
 302         int                     n;              /* count of live freespc ents */
 303         xfs_dir2_data_off_t     off;            /* freespace entry value */
 304         __be16                  *to;            /* pointer to freespace entry */
 305         xfs_trans_t             *tp;            /* transaction pointer */
 306         struct xfs_dir3_icfree_hdr freehdr;
 307
 308         trace_xfs_dir2_leaf_to_node(args);
 309
 310         dp = args->dp;
 311         mp = dp->i_mount;
 312         tp = args->trans;
 313         /*
 314          * Add a freespace block to the directory.
 315          */
 316         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 317                 return error;
 318         }
 319         ASSERT(fdb == xfs_dir2_byte_to_db(mp, XFS_DIR2_FREE_OFFSET));
 320         /*
 321          * Get the buffer for the new freespace block.
 322          */
 323         error = xfs_dir3_free_get_buf(args, fdb, &fbp);
 324         if (error)
 325                 return error;
 326
 327         free = fbp->b_addr;
 328         dp->d_ops->free_hdr_from_disk(&freehdr, free);
 329         leaf = lbp->b_addr;
 330         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 331         ASSERT(be32_to_cpu(ltp->bestcount) <=
 332                                 (uint)dp->i_d.di_size / mp->m_dirblksize);
 333
 334         /*
 335          * Copy freespace entries from the leaf block to the new block.
 336          * Count active entries.
 337          */
 338         from = xfs_dir2_leaf_bests_p(ltp);
 339         to = dp->d_ops->free_bests_p(free);
 340         for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 341                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 342                         n++;
 343                 *to = cpu_to_be16(off);
 344         }
 345
 346         /*
 347          * Now initialize the freespace block header.
 348          */
 349         freehdr.nused = n;
 350         freehdr.nvalid = be32_to_cpu(ltp->bestcount);
 351
 352         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
 353         xfs_dir2_free_log_bests(tp, dp, fbp, 0, freehdr.nvalid - 1);
 354         xfs_dir2_free_log_header(tp, dp, fbp);
 355
 356         /*
 357          * Converting the leaf to a leafnode is just a matter of changing the
 358          * magic number and the ops. Do the change directly to the buffer as
 359          * it's less work (and less code) than decoding the header to host
 360          * format and back again.
 361          */
 362         if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
 363                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 364         else
 365                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 366         lbp->b_ops = &xfs_dir3_leafn_buf_ops;
 367         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
 368         xfs_dir3_leaf_log_header(tp, dp, lbp);
 369         xfs_dir3_leaf_check(dp, lbp);
 370         return 0;
 371 }
 372
 373 /*
 374  * Add a leaf entry to a leaf block in a node-form directory.
 375  * The other work necessary is done from the caller.
 376  */
 377 static int                                      /* error */
 378 xfs_dir2_leafn_add(
 379         struct xfs_buf          *bp,            /* leaf buffer */
 380         xfs_da_args_t           *args,          /* operation arguments */
 381         int                     index)          /* insertion pt for new entry */
 382 {
 383         int                     compact;        /* compacting stale leaves */
 384         xfs_inode_t             *dp;            /* incore directory inode */
 385         int                     highstale;      /* next stale entry */
 386         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 387         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 388         int                     lfloghigh;      /* high leaf entry logging */
 389         int                     lfloglow;       /* low leaf entry logging */
 390         int                     lowstale;       /* previous stale entry */
 391         xfs_mount_t             *mp;            /* filesystem mount point */
 392         xfs_trans_t             *tp;            /* transaction pointer */
 393         struct xfs_dir3_icleaf_hdr leafhdr;
 394         struct xfs_dir2_leaf_entry *ents;
 395
 396         trace_xfs_dir2_leafn_add(args, index);
 397
 398         dp = args->dp;
 399         mp = dp->i_mount;
 400         tp = args->trans;
 401         leaf = bp->b_addr;
 402         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 403         ents = dp->d_ops->leaf_ents_p(leaf);
 404
 405         /*
 406          * Quick check just to make sure we are not going to index
 407          * into other peoples memory
 408          */
 409         if (index < 0)
 410                 return XFS_ERROR(EFSCORRUPTED);
 411
 412         /*
 413          * If there are already the maximum number of leaf entries in
 414          * the block, if there are no stale entries it won't fit.
 415          * Caller will do a split.  If there are stale entries we'll do
 416          * a compact.
 417          */
 418
 419         if (leafhdr.count == dp->d_ops->leaf_max_ents(mp)) {
 420                 if (!leafhdr.stale)
 421                         return XFS_ERROR(ENOSPC);
 422                 compact = leafhdr.stale > 1;
 423         } else
 424                 compact = 0;
 425         ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
 426         ASSERT(index == leafhdr.count ||
 427                be32_to_cpu(ents[index].hashval) >= args->hashval);
 428
 429         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 430                 return 0;
 431
 432         /*
 433          * Compact out all but one stale leaf entry.  Leaves behind
 434          * the entry closest to index.
 435          */
 436         if (compact)
 437                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 438                                          &highstale, &lfloglow, &lfloghigh);
 439         else if (leafhdr.stale) {
 440                 /*
 441                  * Set impossible logging indices for this case.
 442                  */
 443                 lfloglow = leafhdr.count;
 444                 lfloghigh = -1;
 445         }
 446
 447         /*
 448          * Insert the new entry, log everything.
 449          */
 450         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 451                                        highstale, &lfloglow, &lfloghigh);
 452
 453         lep->hashval = cpu_to_be32(args->hashval);
 454         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
 455                                 args->blkno, args->index));
 456
 457         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
 458         xfs_dir3_leaf_log_header(tp, dp, bp);
 459         xfs_dir3_leaf_log_ents(tp, dp, bp, lfloglow, lfloghigh);
 460         xfs_dir3_leaf_check(dp, bp);
 461         return 0;
 462 }
 463
 464 #ifdef DEBUG
 465 static void
 466 xfs_dir2_free_hdr_check(
 467         struct xfs_inode *dp,
 468         struct xfs_buf  *bp,
 469         xfs_dir2_db_t   db)
 470 {
 471         struct xfs_dir3_icfree_hdr hdr;
 472
 473         dp->d_ops->free_hdr_from_disk(&hdr, bp->b_addr);
 474
 475         ASSERT((hdr.firstdb % dp->d_ops->free_max_bests(dp->i_mount)) == 0);
 476         ASSERT(hdr.firstdb <= db);
 477         ASSERT(db < hdr.firstdb + hdr.nvalid);
 478 }
 479 #else
 480 #define xfs_dir2_free_hdr_check(dp, bp, db)
 481 #endif  /* DEBUG */
 482
 483 /*
 484  * Return the last hash value in the leaf.
 485  * Stale entries are ok.
 486  */
 487 xfs_dahash_t                                    /* hash value */
 488 xfs_dir2_leafn_lasthash(
 489         struct xfs_inode *dp,
 490         struct xfs_buf  *bp,                    /* leaf buffer */
 491         int             *count)                 /* count of entries in leaf */
 492 {
 493         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 494         struct xfs_dir2_leaf_entry *ents;
 495         struct xfs_dir3_icleaf_hdr leafhdr;
 496
 497         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 498
 499         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 500                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
 501
 502         if (count)
 503                 *count = leafhdr.count;
 504         if (!leafhdr.count)
 505                 return 0;
 506
 507         ents = dp->d_ops->leaf_ents_p(leaf);
 508         return be32_to_cpu(ents[leafhdr.count - 1].hashval);
 509 }
 510
 511 /*
 512  * Look up a leaf entry for space to add a name in a node-format leaf block.
 513  * The extrablk in state is a freespace block.
 514  */
 515 STATIC int
 516 xfs_dir2_leafn_lookup_for_addname(
 517         struct xfs_buf          *bp,            /* leaf buffer */
 518         xfs_da_args_t           *args,          /* operation arguments */
 519         int                     *indexp,        /* out: leaf entry index */
 520         xfs_da_state_t          *state)         /* state to fill in */
 521 {
 522         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 523         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 524         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 525         xfs_inode_t             *dp;            /* incore directory inode */
 526         int                     error;          /* error return value */
 527         int                     fi;             /* free entry index */
 528         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 529         int                     index;          /* leaf entry index */
 530         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 531         int                     length;         /* length of new data entry */
 532         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 533         xfs_mount_t             *mp;            /* filesystem mount point */
 534         xfs_dir2_db_t           newdb;          /* new data block number */
 535         xfs_dir2_db_t           newfdb;         /* new free block number */
 536         xfs_trans_t             *tp;            /* transaction pointer */
 537         struct xfs_dir2_leaf_entry *ents;
 538         struct xfs_dir3_icleaf_hdr leafhdr;
 539
 540         dp = args->dp;
 541         tp = args->trans;
 542         mp = dp->i_mount;
 543         leaf = bp->b_addr;
 544         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 545         ents = dp->d_ops->leaf_ents_p(leaf);
 546
 547         xfs_dir3_leaf_check(dp, bp);
 548         ASSERT(leafhdr.count > 0);
 549
 550         /*
 551          * Look up the hash value in the leaf entries.
 552          */
 553         index = xfs_dir2_leaf_search_hash(args, bp);
 554         /*
 555          * Do we have a buffer coming in?
 556          */
 557         if (state->extravalid) {
 558                 /* If so, it's a free block buffer, get the block number. */
 559                 curbp = state->extrablk.bp;
 560                 curfdb = state->extrablk.blkno;
 561                 free = curbp->b_addr;
 562                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 563                        free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 564         }
 565         length = dp->d_ops->data_entsize(args->namelen);
 566         /*
 567          * Loop over leaf entries with the right hash value.
 568          */
 569         for (lep = &ents[index];
 570              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 571              lep++, index++) {
 572                 /*
 573                  * Skip stale leaf entries.
 574                  */
 575                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 576                         continue;
 577                 /*
 578                  * Pull the data block number from the entry.
 579                  */
 580                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 581                 /*
 582                  * For addname, we're looking for a place to put the new entry.
 583                  * We want to use a data block with an entry of equal
 584                  * hash value to ours if there is one with room.
 585                  *
 586                  * If this block isn't the data block we already have
 587                  * in hand, take a look at it.
 588                  */
 589                 if (newdb != curdb) {
 590                         __be16 *bests;
 591
 592                         curdb = newdb;
 593                         /*
 594                          * Convert the data block to the free block
 595                          * holding its freespace information.
 596                          */
 597                         newfdb = dp->d_ops->db_to_fdb(mp, newdb);
 598                         /*
 599                          * If it's not the one we have in hand, read it in.
 600                          */
 601                         if (newfdb != curfdb) {
 602                                 /*
 603                                  * If we had one before, drop it.
 604                                  */
 605                                 if (curbp)
 606                                         xfs_trans_brelse(tp, curbp);
 607
 608                                 error = xfs_dir2_free_read(tp, dp,
 609                                                 xfs_dir2_db_to_da(args->geo,
 610                                                                   newfdb),
 611                                                 &curbp);
 612                                 if (error)
 613                                         return error;
 614                                 free = curbp->b_addr;
 615
 616                                 xfs_dir2_free_hdr_check(dp, curbp, curdb);
 617                         }
 618                         /*
 619                          * Get the index for our entry.
 620                          */
 621                         fi = dp->d_ops->db_to_fdindex(mp, curdb);
 622                         /*
 623                          * If it has room, return it.
 624                          */
 625                         bests = dp->d_ops->free_bests_p(free);
 626                         if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
 627                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 628                                                         XFS_ERRLEVEL_LOW, mp);
 629                                 if (curfdb != newfdb)
 630                                         xfs_trans_brelse(tp, curbp);
 631                                 return XFS_ERROR(EFSCORRUPTED);
 632                         }
 633                         curfdb = newfdb;
 634                         if (be16_to_cpu(bests[fi]) >= length)
 635                                 goto out;
 636                 }
 637         }
 638         /* Didn't find any space */
 639         fi = -1;
 640 out:
 641         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 642         if (curbp) {
 643                 /* Giving back a free block. */
 644                 state->extravalid = 1;
 645                 state->extrablk.bp = curbp;
 646                 state->extrablk.index = fi;
 647                 state->extrablk.blkno = curfdb;
 648
 649                 /*
 650                  * Important: this magic number is not in the buffer - it's for
 651                  * buffer type information and therefore only the free/data type
 652                  * matters here, not whether CRCs are enabled or not.
 653                  */
 654                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 655         } else {
 656                 state->extravalid = 0;
 657         }
 658         /*
 659          * Return the index, that will be the insertion point.
 660          */
 661         *indexp = index;
 662         return XFS_ERROR(ENOENT);
 663 }
 664
 665 /*
 666  * Look up a leaf entry in a node-format leaf block.
 667  * The extrablk in state a data block.
 668  */
 669 STATIC int
 670 xfs_dir2_leafn_lookup_for_entry(
 671         struct xfs_buf          *bp,            /* leaf buffer */
 672         xfs_da_args_t           *args,          /* operation arguments */
 673         int                     *indexp,        /* out: leaf entry index */
 674         xfs_da_state_t          *state)         /* state to fill in */
 675 {
 676         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 677         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 678         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 679         xfs_inode_t             *dp;            /* incore directory inode */
 680         int                     error;          /* error return value */
 681         int                     index;          /* leaf entry index */
 682         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 683         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 684         xfs_mount_t             *mp;            /* filesystem mount point */
 685         xfs_dir2_db_t           newdb;          /* new data block number */
 686         xfs_trans_t             *tp;            /* transaction pointer */
 687         enum xfs_dacmp          cmp;            /* comparison result */
 688         struct xfs_dir2_leaf_entry *ents;
 689         struct xfs_dir3_icleaf_hdr leafhdr;
 690
 691         dp = args->dp;
 692         tp = args->trans;
 693         mp = dp->i_mount;
 694         leaf = bp->b_addr;
 695         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 696         ents = dp->d_ops->leaf_ents_p(leaf);
 697
 698         xfs_dir3_leaf_check(dp, bp);
 699         ASSERT(leafhdr.count > 0);
 700
 701         /*
 702          * Look up the hash value in the leaf entries.
 703          */
 704         index = xfs_dir2_leaf_search_hash(args, bp);
 705         /*
 706          * Do we have a buffer coming in?
 707          */
 708         if (state->extravalid) {
 709                 curbp = state->extrablk.bp;
 710                 curdb = state->extrablk.blkno;
 711         }
 712         /*
 713          * Loop over leaf entries with the right hash value.
 714          */
 715         for (lep = &ents[index];
 716              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 717              lep++, index++) {
 718                 /*
 719                  * Skip stale leaf entries.
 720                  */
 721                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 722                         continue;
 723                 /*
 724                  * Pull the data block number from the entry.
 725                  */
 726                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 727                 /*
 728                  * Not adding a new entry, so we really want to find
 729                  * the name given to us.
 730                  *
 731                  * If it's a different data block, go get it.
 732                  */
 733                 if (newdb != curdb) {
 734                         /*
 735                          * If we had a block before that we aren't saving
 736                          * for a CI name, drop it
 737                          */
 738                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 739                                                 curdb != state->extrablk.blkno))
 740                                 xfs_trans_brelse(tp, curbp);
 741                         /*
 742                          * If needing the block that is saved with a CI match,
 743                          * use it otherwise read in the new data block.
 744                          */
 745                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 746                                         newdb == state->extrablk.blkno) {
 747                                 ASSERT(state->extravalid);
 748                                 curbp = state->extrablk.bp;
 749                         } else {
 750                                 error = xfs_dir3_data_read(tp, dp,
 751                                                 xfs_dir2_db_to_da(args->geo,
 752                                                                   newdb),
 753                                                 -1, &curbp);
 754                                 if (error)
 755                                         return error;
 756                         }
 757                         xfs_dir3_data_check(dp, curbp);
 758                         curdb = newdb;
 759                 }
 760                 /*
 761                  * Point to the data entry.
 762                  */
 763                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 764                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
 765                 /*
 766                  * Compare the entry and if it's an exact match, return
 767                  * EEXIST immediately. If it's the first case-insensitive
 768                  * match, store the block & inode number and continue looking.
 769                  */
 770                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 771                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 772                         /* If there is a CI match block, drop it */
 773                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 774                                                 curdb != state->extrablk.blkno)
 775                                 xfs_trans_brelse(tp, state->extrablk.bp);
 776                         args->cmpresult = cmp;
 777                         args->inumber = be64_to_cpu(dep->inumber);
 778                         args->filetype = dp->d_ops->data_get_ftype(dep);
 779                         *indexp = index;
 780                         state->extravalid = 1;
 781                         state->extrablk.bp = curbp;
 782                         state->extrablk.blkno = curdb;
 783                         state->extrablk.index = (int)((char *)dep -
 784                                                         (char *)curbp->b_addr);
 785                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 786                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 787                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 788                         if (cmp == XFS_CMP_EXACT)
 789                                 return XFS_ERROR(EEXIST);
 790                 }
 791         }
 792         ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
 793         if (curbp) {
 794                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 795                         /* Giving back last used data block. */
 796                         state->extravalid = 1;
 797                         state->extrablk.bp = curbp;
 798                         state->extrablk.index = -1;
 799                         state->extrablk.blkno = curdb;
 800                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 801                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 802                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 803                 } else {
 804                         /* If the curbp is not the CI match block, drop it */
 805                         if (state->extrablk.bp != curbp)
 806                                 xfs_trans_brelse(tp, curbp);
 807                 }
 808         } else {
 809                 state->extravalid = 0;
 810         }
 811         *indexp = index;
 812         return XFS_ERROR(ENOENT);
 813 }
 814
 815 /*
 816  * Look up a leaf entry in a node-format leaf block.
 817  * If this is an addname then the extrablk in state is a freespace block,
 818  * otherwise it's a data block.
 819  */
 820 int
 821 xfs_dir2_leafn_lookup_int(
 822         struct xfs_buf          *bp,            /* leaf buffer */
 823         xfs_da_args_t           *args,          /* operation arguments */
 824         int                     *indexp,        /* out: leaf entry index */
 825         xfs_da_state_t          *state)         /* state to fill in */
 826 {
 827         if (args->op_flags & XFS_DA_OP_ADDNAME)
 828                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 829                                                         state);
 830         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 831 }
 832
 833 /*
 834  * Move count leaf entries from source to destination leaf.
 835  * Log entries and headers.  Stale entries are preserved.
 836  */
 837 static void
 838 xfs_dir3_leafn_moveents(
 839         xfs_da_args_t                   *args,  /* operation arguments */
 840         struct xfs_buf                  *bp_s,  /* source */
 841         struct xfs_dir3_icleaf_hdr      *shdr,
 842         struct xfs_dir2_leaf_entry      *sents,
 843         int                             start_s,/* source leaf index */
 844         struct xfs_buf                  *bp_d,  /* destination */
 845         struct xfs_dir3_icleaf_hdr      *dhdr,
 846         struct xfs_dir2_leaf_entry      *dents,
 847         int                             start_d,/* destination leaf index */
 848         int                             count)  /* count of leaves to copy */
 849 {
 850         struct xfs_trans                *tp = args->trans;
 851         int                             stale;  /* count stale leaves copied */
 852
 853         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 854
 855         /*
 856          * Silently return if nothing to do.
 857          */
 858         if (count == 0)
 859                 return;
 860
 861         /*
 862          * If the destination index is not the end of the current
 863          * destination leaf entries, open up a hole in the destination
 864          * to hold the new entries.
 865          */
 866         if (start_d < dhdr->count) {
 867                 memmove(&dents[start_d + count], &dents[start_d],
 868                         (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
 869                 xfs_dir3_leaf_log_ents(tp, args->dp, bp_d, start_d + count,
 870                                        count + dhdr->count - 1);
 871         }
 872         /*
 873          * If the source has stale leaves, count the ones in the copy range
 874          * so we can update the header correctly.
 875          */
 876         if (shdr->stale) {
 877                 int     i;                      /* temp leaf index */
 878
 879                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 880                         if (sents[i].address ==
 881                                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 882                                 stale++;
 883                 }
 884         } else
 885                 stale = 0;
 886         /*
 887          * Copy the leaf entries from source to destination.
 888          */
 889         memcpy(&dents[start_d], &sents[start_s],
 890                 count * sizeof(xfs_dir2_leaf_entry_t));
 891         xfs_dir3_leaf_log_ents(tp, args->dp, bp_d,
 892                                start_d, start_d + count - 1);
 893
 894         /*
 895          * If there are source entries after the ones we copied,
 896          * delete the ones we copied by sliding the next ones down.
 897          */
 898         if (start_s + count < shdr->count) {
 899                 memmove(&sents[start_s], &sents[start_s + count],
 900                         count * sizeof(xfs_dir2_leaf_entry_t));
 901                 xfs_dir3_leaf_log_ents(tp, args->dp, bp_s,
 902                                        start_s, start_s + count - 1);
 903         }
 904
 905         /*
 906          * Update the headers and log them.
 907          */
 908         shdr->count -= count;
 909         shdr->stale -= stale;
 910         dhdr->count += count;
 911         dhdr->stale += stale;
 912 }
 913
 914 /*
 915  * Determine the sort order of two leaf blocks.
 916  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 917  */
 918 int                                             /* sort order */
 919 xfs_dir2_leafn_order(
 920         struct xfs_inode        *dp,
 921         struct xfs_buf          *leaf1_bp,              /* leaf1 buffer */
 922         struct xfs_buf          *leaf2_bp)              /* leaf2 buffer */
 923 {
 924         struct xfs_dir2_leaf    *leaf1 = leaf1_bp->b_addr;
 925         struct xfs_dir2_leaf    *leaf2 = leaf2_bp->b_addr;
 926         struct xfs_dir2_leaf_entry *ents1;
 927         struct xfs_dir2_leaf_entry *ents2;
 928         struct xfs_dir3_icleaf_hdr hdr1;
 929         struct xfs_dir3_icleaf_hdr hdr2;
 930
 931         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
 932         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
 933         ents1 = dp->d_ops->leaf_ents_p(leaf1);
 934         ents2 = dp->d_ops->leaf_ents_p(leaf2);
 935
 936         if (hdr1.count > 0 && hdr2.count > 0 &&
 937             (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
 938              be32_to_cpu(ents2[hdr2.count - 1].hashval) <
 939                                 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
 940                 return 1;
 941         return 0;
 942 }
 943
 944 /*
 945  * Rebalance leaf entries between two leaf blocks.
 946  * This is actually only called when the second block is new,
 947  * though the code deals with the general case.
 948  * A new entry will be inserted in one of the blocks, and that
 949  * entry is taken into account when balancing.
 950  */
 951 static void
 952 xfs_dir2_leafn_rebalance(
 953         xfs_da_state_t          *state,         /* btree cursor */
 954         xfs_da_state_blk_t      *blk1,          /* first btree block */
 955         xfs_da_state_blk_t      *blk2)          /* second btree block */
 956 {
 957         xfs_da_args_t           *args;          /* operation arguments */
 958         int                     count;          /* count (& direction) leaves */
 959         int                     isleft;         /* new goes in left leaf */
 960         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 961         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 962         int                     mid;            /* midpoint leaf index */
 963 #if defined(DEBUG) || defined(XFS_WARN)
 964         int                     oldstale;       /* old count of stale leaves */
 965 #endif
 966         int                     oldsum;         /* old total leaf count */
 967         int                     swap;           /* swapped leaf blocks */
 968         struct xfs_dir2_leaf_entry *ents1;
 969         struct xfs_dir2_leaf_entry *ents2;
 970         struct xfs_dir3_icleaf_hdr hdr1;
 971         struct xfs_dir3_icleaf_hdr hdr2;
 972         struct xfs_inode        *dp = state->args->dp;
 973
 974         args = state->args;
 975         /*
 976          * If the block order is wrong, swap the arguments.
 977          */
 978         if ((swap = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp))) {
 979                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
 980
 981                 tmp = blk1;
 982                 blk1 = blk2;
 983                 blk2 = tmp;
 984         }
 985         leaf1 = blk1->bp->b_addr;
 986         leaf2 = blk2->bp->b_addr;
 987         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
 988         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
 989         ents1 = dp->d_ops->leaf_ents_p(leaf1);
 990         ents2 = dp->d_ops->leaf_ents_p(leaf2);
 991
 992         oldsum = hdr1.count + hdr2.count;
 993 #if defined(DEBUG) || defined(XFS_WARN)
 994         oldstale = hdr1.stale + hdr2.stale;
 995 #endif
 996         mid = oldsum >> 1;
 997
 998         /*
 999          * If the old leaf count was odd then the new one will be even,
1000          * so we need to divide the new count evenly.
1001          */
1002         if (oldsum & 1) {
1003                 xfs_dahash_t    midhash;        /* middle entry hash value */
1004
1005                 if (mid >= hdr1.count)
1006                         midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1007                 else
1008                         midhash = be32_to_cpu(ents1[mid].hashval);
1009                 isleft = args->hashval <= midhash;
1010         }
1011         /*
1012          * If the old count is even then the new count is odd, so there's
1013          * no preferred side for the new entry.
1014          * Pick the left one.
1015          */
1016         else
1017                 isleft = 1;
1018         /*
1019          * Calculate moved entry count.  Positive means left-to-right,
1020          * negative means right-to-left.  Then move the entries.
1021          */
1022         count = hdr1.count - mid + (isleft == 0);
1023         if (count > 0)
1024                 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1025                                         hdr1.count - count, blk2->bp,
1026                                         &hdr2, ents2, 0, count);
1027         else if (count < 0)
1028                 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1029                                         blk1->bp, &hdr1, ents1,
1030                                         hdr1.count, count);
1031
1032         ASSERT(hdr1.count + hdr2.count == oldsum);
1033         ASSERT(hdr1.stale + hdr2.stale == oldstale);
1034
1035         /* log the changes made when moving the entries */
1036         dp->d_ops->leaf_hdr_to_disk(leaf1, &hdr1);
1037         dp->d_ops->leaf_hdr_to_disk(leaf2, &hdr2);
1038         xfs_dir3_leaf_log_header(args->trans, dp, blk1->bp);
1039         xfs_dir3_leaf_log_header(args->trans, dp, blk2->bp);
1040
1041         xfs_dir3_leaf_check(dp, blk1->bp);
1042         xfs_dir3_leaf_check(dp, blk2->bp);
1043
1044         /*
1045          * Mark whether we're inserting into the old or new leaf.
1046          */
1047         if (hdr1.count < hdr2.count)
1048                 state->inleaf = swap;
1049         else if (hdr1.count > hdr2.count)
1050                 state->inleaf = !swap;
1051         else
1052                 state->inleaf = swap ^ (blk1->index <= hdr1.count);
1053         /*
1054          * Adjust the expected index for insertion.
1055          */
1056         if (!state->inleaf)
1057                 blk2->index = blk1->index - hdr1.count;
1058
1059         /*
1060          * Finally sanity check just to make sure we are not returning a
1061          * negative index
1062          */
1063         if (blk2->index < 0) {
1064                 state->inleaf = 1;
1065                 blk2->index = 0;
1066                 xfs_alert(dp->i_mount,
1067         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1068                         __func__, blk1->index);
1069         }
1070 }
1071
1072 static int
1073 xfs_dir3_data_block_free(
1074         xfs_da_args_t           *args,
1075         struct xfs_dir2_data_hdr *hdr,
1076         struct xfs_dir2_free    *free,
1077         xfs_dir2_db_t           fdb,
1078         int                     findex,
1079         struct xfs_buf          *fbp,
1080         int                     longest)
1081 {
1082         struct xfs_trans        *tp = args->trans;
1083         int                     logfree = 0;
1084         __be16                  *bests;
1085         struct xfs_dir3_icfree_hdr freehdr;
1086         struct xfs_inode        *dp = args->dp;
1087
1088         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1089         bests = dp->d_ops->free_bests_p(free);
1090         if (hdr) {
1091                 /*
1092                  * Data block is not empty, just set the free entry to the new
1093                  * value.
1094                  */
1095                 bests[findex] = cpu_to_be16(longest);
1096                 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
1097                 return 0;
1098         }
1099
1100         /* One less used entry in the free table. */
1101         freehdr.nused--;
1102
1103         /*
1104          * If this was the last entry in the table, we can trim the table size
1105          * back.  There might be other entries at the end referring to
1106          * non-existent data blocks, get those too.
1107          */
1108         if (findex == freehdr.nvalid - 1) {
1109                 int     i;              /* free entry index */
1110
1111                 for (i = findex - 1; i >= 0; i--) {
1112                         if (bests[i] != cpu_to_be16(NULLDATAOFF))
1113                                 break;
1114                 }
1115                 freehdr.nvalid = i + 1;
1116                 logfree = 0;
1117         } else {
1118                 /* Not the last entry, just punch it out.  */
1119                 bests[findex] = cpu_to_be16(NULLDATAOFF);
1120                 logfree = 1;
1121         }
1122
1123         dp->d_ops->free_hdr_to_disk(free, &freehdr);
1124         xfs_dir2_free_log_header(tp, dp, fbp);
1125
1126         /*
1127          * If there are no useful entries left in the block, get rid of the
1128          * block if we can.
1129          */
1130         if (!freehdr.nused) {
1131                 int error;
1132
1133                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1134                 if (error == 0) {
1135                         fbp = NULL;
1136                         logfree = 0;
1137                 } else if (error != ENOSPC || args->total != 0)
1138                         return error;
1139                 /*
1140                  * It's possible to get ENOSPC if there is no
1141                  * space reservation.  In this case some one
1142                  * else will eventually get rid of this block.
1143                  */
1144         }
1145
1146         /* Log the free entry that changed, unless we got rid of it.  */
1147         if (logfree)
1148                 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
1149         return 0;
1150 }
1151
1152 /*
1153  * Remove an entry from a node directory.
1154  * This removes the leaf entry and the data entry,
1155  * and updates the free block if necessary.
1156  */
1157 static int                                      /* error */
1158 xfs_dir2_leafn_remove(
1159         xfs_da_args_t           *args,          /* operation arguments */
1160         struct xfs_buf          *bp,            /* leaf buffer */
1161         int                     index,          /* leaf entry index */
1162         xfs_da_state_blk_t      *dblk,          /* data block */
1163         int                     *rval)          /* resulting block needs join */
1164 {
1165         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1166         xfs_dir2_db_t           db;             /* data block number */
1167         struct xfs_buf          *dbp;           /* data block buffer */
1168         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1169         xfs_inode_t             *dp;            /* incore directory inode */
1170         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1171         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1172         int                     longest;        /* longest data free entry */
1173         int                     off;            /* data block entry offset */
1174         xfs_mount_t             *mp;            /* filesystem mount point */
1175         int                     needlog;        /* need to log data header */
1176         int                     needscan;       /* need to rescan data frees */
1177         xfs_trans_t             *tp;            /* transaction pointer */
1178         struct xfs_dir2_data_free *bf;          /* bestfree table */
1179         struct xfs_dir3_icleaf_hdr leafhdr;
1180         struct xfs_dir2_leaf_entry *ents;
1181
1182         trace_xfs_dir2_leafn_remove(args, index);
1183
1184         dp = args->dp;
1185         tp = args->trans;
1186         mp = dp->i_mount;
1187         leaf = bp->b_addr;
1188         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1189         ents = dp->d_ops->leaf_ents_p(leaf);
1190
1191         /*
1192          * Point to the entry we're removing.
1193          */
1194         lep = &ents[index];
1195
1196         /*
1197          * Extract the data block and offset from the entry.
1198          */
1199         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1200         ASSERT(dblk->blkno == db);
1201         off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
1202         ASSERT(dblk->index == off);
1203
1204         /*
1205          * Kill the leaf entry by marking it stale.
1206          * Log the leaf block changes.
1207          */
1208         leafhdr.stale++;
1209         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1210         xfs_dir3_leaf_log_header(tp, dp, bp);
1211
1212         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1213         xfs_dir3_leaf_log_ents(tp, dp, bp, index, index);
1214
1215         /*
1216          * Make the data entry free.  Keep track of the longest freespace
1217          * in the data block in case it changes.
1218          */
1219         dbp = dblk->bp;
1220         hdr = dbp->b_addr;
1221         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1222         bf = dp->d_ops->data_bestfree_p(hdr);
1223         longest = be16_to_cpu(bf[0].length);
1224         needlog = needscan = 0;
1225         xfs_dir2_data_make_free(tp, dp, dbp, off,
1226                 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1227         /*
1228          * Rescan the data block freespaces for bestfree.
1229          * Log the data block header if needed.
1230          */
1231         if (needscan)
1232                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1233         if (needlog)
1234                 xfs_dir2_data_log_header(tp, dp, dbp);
1235         xfs_dir3_data_check(dp, dbp);
1236         /*
1237          * If the longest data block freespace changes, need to update
1238          * the corresponding freeblock entry.
1239          */
1240         if (longest < be16_to_cpu(bf[0].length)) {
1241                 int             error;          /* error return value */
1242                 struct xfs_buf  *fbp;           /* freeblock buffer */
1243                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1244                 int             findex;         /* index in freeblock entries */
1245                 xfs_dir2_free_t *free;          /* freeblock structure */
1246
1247                 /*
1248                  * Convert the data block number to a free block,
1249                  * read in the free block.
1250                  */
1251                 fdb = dp->d_ops->db_to_fdb(mp, db);
1252                 error = xfs_dir2_free_read(tp, dp,
1253                                            xfs_dir2_db_to_da(args->geo, fdb),
1254                                            &fbp);
1255                 if (error)
1256                         return error;
1257                 free = fbp->b_addr;
1258 #ifdef DEBUG
1259         {
1260                 struct xfs_dir3_icfree_hdr freehdr;
1261                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1262                 ASSERT(freehdr.firstdb == dp->d_ops->free_max_bests(mp) *
1263                           (fdb - xfs_dir2_byte_to_db(mp, XFS_DIR2_FREE_OFFSET)));
1264         }
1265 #endif
1266                 /*
1267                  * Calculate which entry we need to fix.
1268                  */
1269                 findex = dp->d_ops->db_to_fdindex(mp, db);
1270                 longest = be16_to_cpu(bf[0].length);
1271                 /*
1272                  * If the data block is now empty we can get rid of it
1273                  * (usually).
1274                  */
1275                 if (longest == mp->m_dirblksize -
1276                                dp->d_ops->data_entry_offset) {
1277                         /*
1278                          * Try to punch out the data block.
1279                          */
1280                         error = xfs_dir2_shrink_inode(args, db, dbp);
1281                         if (error == 0) {
1282                                 dblk->bp = NULL;
1283                                 hdr = NULL;
1284                         }
1285                         /*
1286                          * We can get ENOSPC if there's no space reservation.
1287                          * In this case just drop the buffer and some one else
1288                          * will eventually get rid of the empty block.
1289                          */
1290                         else if (!(error == ENOSPC && args->total == 0))
1291                                 return error;
1292                 }
1293                 /*
1294                  * If we got rid of the data block, we can eliminate that entry
1295                  * in the free block.
1296                  */
1297                 error = xfs_dir3_data_block_free(args, hdr, free,
1298                                                  fdb, findex, fbp, longest);
1299                 if (error)
1300                         return error;
1301         }
1302
1303         xfs_dir3_leaf_check(dp, bp);
1304         /*
1305          * Return indication of whether this leaf block is empty enough
1306          * to justify trying to join it with a neighbor.
1307          */
1308         *rval = (dp->d_ops->leaf_hdr_size +
1309                  (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1310                 mp->m_dir_magicpct;
1311         return 0;
1312 }
1313
1314 /*
1315  * Split the leaf entries in the old block into old and new blocks.
1316  */
1317 int                                             /* error */
1318 xfs_dir2_leafn_split(
1319         xfs_da_state_t          *state,         /* btree cursor */
1320         xfs_da_state_blk_t      *oldblk,        /* original block */
1321         xfs_da_state_blk_t      *newblk)        /* newly created block */
1322 {
1323         xfs_da_args_t           *args;          /* operation arguments */
1324         xfs_dablk_t             blkno;          /* new leaf block number */
1325         int                     error;          /* error return value */
1326         xfs_mount_t             *mp;            /* filesystem mount point */
1327         struct xfs_inode        *dp;
1328
1329         /*
1330          * Allocate space for a new leaf node.
1331          */
1332         args = state->args;
1333         dp = args->dp;
1334         mp = dp->i_mount;
1335         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1336         error = xfs_da_grow_inode(args, &blkno);
1337         if (error) {
1338                 return error;
1339         }
1340         /*
1341          * Initialize the new leaf block.
1342          */
1343         error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(args->geo, blkno),
1344                                       &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1345         if (error)
1346                 return error;
1347
1348         newblk->blkno = blkno;
1349         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1350         /*
1351          * Rebalance the entries across the two leaves, link the new
1352          * block into the leaves.
1353          */
1354         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1355         error = xfs_da3_blk_link(state, oldblk, newblk);
1356         if (error) {
1357                 return error;
1358         }
1359         /*
1360          * Insert the new entry in the correct block.
1361          */
1362         if (state->inleaf)
1363                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1364         else
1365                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1366         /*
1367          * Update last hashval in each block since we added the name.
1368          */
1369         oldblk->hashval = xfs_dir2_leafn_lasthash(dp, oldblk->bp, NULL);
1370         newblk->hashval = xfs_dir2_leafn_lasthash(dp, newblk->bp, NULL);
1371         xfs_dir3_leaf_check(dp, oldblk->bp);
1372         xfs_dir3_leaf_check(dp, newblk->bp);
1373         return error;
1374 }
1375
1376 /*
1377  * Check a leaf block and its neighbors to see if the block should be
1378  * collapsed into one or the other neighbor.  Always keep the block
1379  * with the smaller block number.
1380  * If the current block is over 50% full, don't try to join it, return 0.
1381  * If the block is empty, fill in the state structure and return 2.
1382  * If it can be collapsed, fill in the state structure and return 1.
1383  * If nothing can be done, return 0.
1384  */
1385 int                                             /* error */
1386 xfs_dir2_leafn_toosmall(
1387         xfs_da_state_t          *state,         /* btree cursor */
1388         int                     *action)        /* resulting action to take */
1389 {
1390         xfs_da_state_blk_t      *blk;           /* leaf block */
1391         xfs_dablk_t             blkno;          /* leaf block number */
1392         struct xfs_buf          *bp;            /* leaf buffer */
1393         int                     bytes;          /* bytes in use */
1394         int                     count;          /* leaf live entry count */
1395         int                     error;          /* error return value */
1396         int                     forward;        /* sibling block direction */
1397         int                     i;              /* sibling counter */
1398         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1399         int                     rval;           /* result from path_shift */
1400         struct xfs_dir3_icleaf_hdr leafhdr;
1401         struct xfs_dir2_leaf_entry *ents;
1402         struct xfs_inode        *dp = state->args->dp;
1403
1404         /*
1405          * Check for the degenerate case of the block being over 50% full.
1406          * If so, it's not worth even looking to see if we might be able
1407          * to coalesce with a sibling.
1408          */
1409         blk = &state->path.blk[state->path.active - 1];
1410         leaf = blk->bp->b_addr;
1411         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1412         ents = dp->d_ops->leaf_ents_p(leaf);
1413         xfs_dir3_leaf_check(dp, blk->bp);
1414
1415         count = leafhdr.count - leafhdr.stale;
1416         bytes = dp->d_ops->leaf_hdr_size + count * sizeof(ents[0]);
1417         if (bytes > (state->blocksize >> 1)) {
1418                 /*
1419                  * Blk over 50%, don't try to join.
1420                  */
1421                 *action = 0;
1422                 return 0;
1423         }
1424         /*
1425          * Check for the degenerate case of the block being empty.
1426          * If the block is empty, we'll simply delete it, no need to
1427          * coalesce it with a sibling block.  We choose (arbitrarily)
1428          * to merge with the forward block unless it is NULL.
1429          */
1430         if (count == 0) {
1431                 /*
1432                  * Make altpath point to the block we want to keep and
1433                  * path point to the block we want to drop (this one).
1434                  */
1435                 forward = (leafhdr.forw != 0);
1436                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1437                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1438                         &rval);
1439                 if (error)
1440                         return error;
1441                 *action = rval ? 2 : 0;
1442                 return 0;
1443         }
1444         /*
1445          * Examine each sibling block to see if we can coalesce with
1446          * at least 25% free space to spare.  We need to figure out
1447          * whether to merge with the forward or the backward block.
1448          * We prefer coalescing with the lower numbered sibling so as
1449          * to shrink a directory over time.
1450          */
1451         forward = leafhdr.forw < leafhdr.back;
1452         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1453                 struct xfs_dir3_icleaf_hdr hdr2;
1454
1455                 blkno = forward ? leafhdr.forw : leafhdr.back;
1456                 if (blkno == 0)
1457                         continue;
1458                 /*
1459                  * Read the sibling leaf block.
1460                  */
1461                 error = xfs_dir3_leafn_read(state->args->trans, dp,
1462                                             blkno, -1, &bp);
1463                 if (error)
1464                         return error;
1465
1466                 /*
1467                  * Count bytes in the two blocks combined.
1468                  */
1469                 count = leafhdr.count - leafhdr.stale;
1470                 bytes = state->blocksize - (state->blocksize >> 2);
1471
1472                 leaf = bp->b_addr;
1473                 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf);
1474                 ents = dp->d_ops->leaf_ents_p(leaf);
1475                 count += hdr2.count - hdr2.stale;
1476                 bytes -= count * sizeof(ents[0]);
1477
1478                 /*
1479                  * Fits with at least 25% to spare.
1480                  */
1481                 if (bytes >= 0)
1482                         break;
1483                 xfs_trans_brelse(state->args->trans, bp);
1484         }
1485         /*
1486          * Didn't like either block, give up.
1487          */
1488         if (i >= 2) {
1489                 *action = 0;
1490                 return 0;
1491         }
1492
1493         /*
1494          * Make altpath point to the block we want to keep (the lower
1495          * numbered block) and path point to the block we want to drop.
1496          */
1497         memcpy(&state->altpath, &state->path, sizeof(state->path));
1498         if (blkno < blk->blkno)
1499                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1500                         &rval);
1501         else
1502                 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1503                         &rval);
1504         if (error) {
1505                 return error;
1506         }
1507         *action = rval ? 0 : 1;
1508         return 0;
1509 }
1510
1511 /*
1512  * Move all the leaf entries from drop_blk to save_blk.
1513  * This is done as part of a join operation.
1514  */
1515 void
1516 xfs_dir2_leafn_unbalance(
1517         xfs_da_state_t          *state,         /* cursor */
1518         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1519         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1520 {
1521         xfs_da_args_t           *args;          /* operation arguments */
1522         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1523         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1524         struct xfs_dir3_icleaf_hdr savehdr;
1525         struct xfs_dir3_icleaf_hdr drophdr;
1526         struct xfs_dir2_leaf_entry *sents;
1527         struct xfs_dir2_leaf_entry *dents;
1528         struct xfs_inode        *dp = state->args->dp;
1529
1530         args = state->args;
1531         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1532         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1533         drop_leaf = drop_blk->bp->b_addr;
1534         save_leaf = save_blk->bp->b_addr;
1535
1536         dp->d_ops->leaf_hdr_from_disk(&savehdr, save_leaf);
1537         dp->d_ops->leaf_hdr_from_disk(&drophdr, drop_leaf);
1538         sents = dp->d_ops->leaf_ents_p(save_leaf);
1539         dents = dp->d_ops->leaf_ents_p(drop_leaf);
1540
1541         /*
1542          * If there are any stale leaf entries, take this opportunity
1543          * to purge them.
1544          */
1545         if (drophdr.stale)
1546                 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1547         if (savehdr.stale)
1548                 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1549
1550         /*
1551          * Move the entries from drop to the appropriate end of save.
1552          */
1553         drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1554         if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1555                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1556                                         save_blk->bp, &savehdr, sents, 0,
1557                                         drophdr.count);
1558         else
1559                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1560                                         save_blk->bp, &savehdr, sents,
1561                                         savehdr.count, drophdr.count);
1562         save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1563
1564         /* log the changes made when moving the entries */
1565         dp->d_ops->leaf_hdr_to_disk(save_leaf, &savehdr);
1566         dp->d_ops->leaf_hdr_to_disk(drop_leaf, &drophdr);
1567         xfs_dir3_leaf_log_header(args->trans, dp, save_blk->bp);
1568         xfs_dir3_leaf_log_header(args->trans, dp, drop_blk->bp);
1569
1570         xfs_dir3_leaf_check(dp, save_blk->bp);
1571         xfs_dir3_leaf_check(dp, drop_blk->bp);
1572 }
1573
1574 /*
1575  * Top-level node form directory addname routine.
1576  */
1577 int                                             /* error */
1578 xfs_dir2_node_addname(
1579         xfs_da_args_t           *args)          /* operation arguments */
1580 {
1581         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1582         int                     error;          /* error return value */
1583         int                     rval;           /* sub-return value */
1584         xfs_da_state_t          *state;         /* btree cursor */
1585
1586         trace_xfs_dir2_node_addname(args);
1587
1588         /*
1589          * Allocate and initialize the state (btree cursor).
1590          */
1591         state = xfs_da_state_alloc();
1592         state->args = args;
1593         state->mp = args->dp->i_mount;
1594         state->blocksize = state->mp->m_dirblksize;
1595         state->node_ents = state->mp->m_dir_node_ents;
1596         /*
1597          * Look up the name.  We're not supposed to find it, but
1598          * this gives us the insertion point.
1599          */
1600         error = xfs_da3_node_lookup_int(state, &rval);
1601         if (error)
1602                 rval = error;
1603         if (rval != ENOENT) {
1604                 goto done;
1605         }
1606         /*
1607          * Add the data entry to a data block.
1608          * Extravalid is set to a freeblock found by lookup.
1609          */
1610         rval = xfs_dir2_node_addname_int(args,
1611                 state->extravalid ? &state->extrablk : NULL);
1612         if (rval) {
1613                 goto done;
1614         }
1615         blk = &state->path.blk[state->path.active - 1];
1616         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1617         /*
1618          * Add the new leaf entry.
1619          */
1620         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1621         if (rval == 0) {
1622                 /*
1623                  * It worked, fix the hash values up the btree.
1624                  */
1625                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1626                         xfs_da3_fixhashpath(state, &state->path);
1627         } else {
1628                 /*
1629                  * It didn't work, we need to split the leaf block.
1630                  */
1631                 if (args->total == 0) {
1632                         ASSERT(rval == ENOSPC);
1633                         goto done;
1634                 }
1635                 /*
1636                  * Split the leaf block and insert the new entry.
1637                  */
1638                 rval = xfs_da3_split(state);
1639         }
1640 done:
1641         xfs_da_state_free(state);
1642         return rval;
1643 }
1644
1645 /*
1646  * Add the data entry for a node-format directory name addition.
1647  * The leaf entry is added in xfs_dir2_leafn_add.
1648  * We may enter with a freespace block that the lookup found.
1649  */
1650 static int                                      /* error */
1651 xfs_dir2_node_addname_int(
1652         xfs_da_args_t           *args,          /* operation arguments */
1653         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1654 {
1655         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1656         xfs_dir2_db_t           dbno;           /* data block number */
1657         struct xfs_buf          *dbp;           /* data block buffer */
1658         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1659         xfs_inode_t             *dp;            /* incore directory inode */
1660         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1661         int                     error;          /* error return value */
1662         xfs_dir2_db_t           fbno;           /* freespace block number */
1663         struct xfs_buf          *fbp;           /* freespace buffer */
1664         int                     findex;         /* freespace entry index */
1665         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1666         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1667         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1668         int                     length;         /* length of the new entry */
1669         int                     logfree;        /* need to log free entry */
1670         xfs_mount_t             *mp;            /* filesystem mount point */
1671         int                     needlog;        /* need to log data header */
1672         int                     needscan;       /* need to rescan data frees */
1673         __be16                  *tagp;          /* data entry tag pointer */
1674         xfs_trans_t             *tp;            /* transaction pointer */
1675         __be16                  *bests;
1676         struct xfs_dir3_icfree_hdr freehdr;
1677         struct xfs_dir2_data_free *bf;
1678
1679         dp = args->dp;
1680         mp = dp->i_mount;
1681         tp = args->trans;
1682         length = dp->d_ops->data_entsize(args->namelen);
1683         /*
1684          * If we came in with a freespace block that means that lookup
1685          * found an entry with our hash value.  This is the freespace
1686          * block for that data entry.
1687          */
1688         if (fblk) {
1689                 fbp = fblk->bp;
1690                 /*
1691                  * Remember initial freespace block number.
1692                  */
1693                 ifbno = fblk->blkno;
1694                 free = fbp->b_addr;
1695                 findex = fblk->index;
1696                 bests = dp->d_ops->free_bests_p(free);
1697                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1698
1699                 /*
1700                  * This means the free entry showed that the data block had
1701                  * space for our entry, so we remembered it.
1702                  * Use that data block.
1703                  */
1704                 if (findex >= 0) {
1705                         ASSERT(findex < freehdr.nvalid);
1706                         ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1707                         ASSERT(be16_to_cpu(bests[findex]) >= length);
1708                         dbno = freehdr.firstdb + findex;
1709                 } else {
1710                         /*
1711                          * The data block looked at didn't have enough room.
1712                          * We'll start at the beginning of the freespace entries.
1713                          */
1714                         dbno = -1;
1715                         findex = 0;
1716                 }
1717         } else {
1718                 /*
1719                  * Didn't come in with a freespace block, so no data block.
1720                  */
1721                 ifbno = dbno = -1;
1722                 fbp = NULL;
1723                 findex = 0;
1724         }
1725
1726         /*
1727          * If we don't have a data block yet, we're going to scan the
1728          * freespace blocks looking for one.  Figure out what the
1729          * highest freespace block number is.
1730          */
1731         if (dbno == -1) {
1732                 xfs_fileoff_t   fo;             /* freespace block number */
1733
1734                 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK)))
1735                         return error;
1736                 lastfbno = xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo);
1737                 fbno = ifbno;
1738         }
1739         /*
1740          * While we haven't identified a data block, search the freeblock
1741          * data for a good data block.  If we find a null freeblock entry,
1742          * indicating a hole in the data blocks, remember that.
1743          */
1744         while (dbno == -1) {
1745                 /*
1746                  * If we don't have a freeblock in hand, get the next one.
1747                  */
1748                 if (fbp == NULL) {
1749                         /*
1750                          * Happens the first time through unless lookup gave
1751                          * us a freespace block to start with.
1752                          */
1753                         if (++fbno == 0)
1754                                 fbno = xfs_dir2_byte_to_db(mp,
1755                                                         XFS_DIR2_FREE_OFFSET);
1756                         /*
1757                          * If it's ifbno we already looked at it.
1758                          */
1759                         if (fbno == ifbno)
1760                                 fbno++;
1761                         /*
1762                          * If it's off the end we're done.
1763                          */
1764                         if (fbno >= lastfbno)
1765                                 break;
1766                         /*
1767                          * Read the block.  There can be holes in the
1768                          * freespace blocks, so this might not succeed.
1769                          * This should be really rare, so there's no reason
1770                          * to avoid it.
1771                          */
1772                         error = xfs_dir2_free_try_read(tp, dp,
1773                                         xfs_dir2_db_to_da(args->geo, fbno),
1774                                         &fbp);
1775                         if (error)
1776                                 return error;
1777                         if (!fbp)
1778                                 continue;
1779                         free = fbp->b_addr;
1780                         findex = 0;
1781                 }
1782                 /*
1783                  * Look at the current free entry.  Is it good enough?
1784                  *
1785                  * The bests initialisation should be where the bufer is read in
1786                  * the above branch. But gcc is too stupid to realise that bests
1787                  * and the freehdr are actually initialised if they are placed
1788                  * there, so we have to do it here to avoid warnings. Blech.
1789                  */
1790                 bests = dp->d_ops->free_bests_p(free);
1791                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1792                 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1793                     be16_to_cpu(bests[findex]) >= length)
1794                         dbno = freehdr.firstdb + findex;
1795                 else {
1796                         /*
1797                          * Are we done with the freeblock?
1798                          */
1799                         if (++findex == freehdr.nvalid) {
1800                                 /*
1801                                  * Drop the block.
1802                                  */
1803                                 xfs_trans_brelse(tp, fbp);
1804                                 fbp = NULL;
1805                                 if (fblk && fblk->bp)
1806                                         fblk->bp = NULL;
1807                         }
1808                 }
1809         }
1810         /*
1811          * If we don't have a data block, we need to allocate one and make
1812          * the freespace entries refer to it.
1813          */
1814         if (unlikely(dbno == -1)) {
1815                 /*
1816                  * Not allowed to allocate, return failure.
1817                  */
1818                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1819                         return XFS_ERROR(ENOSPC);
1820
1821                 /*
1822                  * Allocate and initialize the new data block.
1823                  */
1824                 if (unlikely((error = xfs_dir2_grow_inode(args,
1825                                                          XFS_DIR2_DATA_SPACE,
1826                                                          &dbno)) ||
1827                     (error = xfs_dir3_data_init(args, dbno, &dbp))))
1828                         return error;
1829
1830                 /*
1831                  * If (somehow) we have a freespace block, get rid of it.
1832                  */
1833                 if (fbp)
1834                         xfs_trans_brelse(tp, fbp);
1835                 if (fblk && fblk->bp)
1836                         fblk->bp = NULL;
1837
1838                 /*
1839                  * Get the freespace block corresponding to the data block
1840                  * that was just allocated.
1841                  */
1842                 fbno = dp->d_ops->db_to_fdb(mp, dbno);
1843                 error = xfs_dir2_free_try_read(tp, dp,
1844                                        xfs_dir2_db_to_da(args->geo, fbno),
1845                                        &fbp);
1846                 if (error)
1847                         return error;
1848
1849                 /*
1850                  * If there wasn't a freespace block, the read will
1851                  * return a NULL fbp.  Allocate and initialize a new one.
1852                  */
1853                 if (!fbp) {
1854                         error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1855                                                     &fbno);
1856                         if (error)
1857                                 return error;
1858
1859                         if (unlikely(dp->d_ops->db_to_fdb(mp, dbno) != fbno)) {
1860                                 xfs_alert(mp,
1861                         "%s: dir ino %llu needed freesp block %lld for\n"
1862                         "  data block %lld, got %lld ifbno %llu lastfbno %d",
1863                                         __func__, (unsigned long long)dp->i_ino,
1864                                         (long long)dp->d_ops->db_to_fdb(mp, dbno),
1865                                         (long long)dbno, (long long)fbno,
1866                                         (unsigned long long)ifbno, lastfbno);
1867                                 if (fblk) {
1868                                         xfs_alert(mp,
1869                                 " fblk 0x%p blkno %llu index %d magic 0x%x",
1870                                                 fblk,
1871                                                 (unsigned long long)fblk->blkno,
1872                                                 fblk->index,
1873                                                 fblk->magic);
1874                                 } else {
1875                                         xfs_alert(mp, " ... fblk is NULL");
1876                                 }
1877                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1878                                                  XFS_ERRLEVEL_LOW, mp);
1879                                 return XFS_ERROR(EFSCORRUPTED);
1880                         }
1881
1882                         /*
1883                          * Get a buffer for the new block.
1884                          */
1885                         error = xfs_dir3_free_get_buf(args, fbno, &fbp);
1886                         if (error)
1887                                 return error;
1888                         free = fbp->b_addr;
1889                         bests = dp->d_ops->free_bests_p(free);
1890                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1891
1892                         /*
1893                          * Remember the first slot as our empty slot.
1894                          */
1895                         freehdr.firstdb =
1896                                 (fbno - xfs_dir2_byte_to_db(mp,
1897                                                         XFS_DIR2_FREE_OFFSET)) *
1898                                         dp->d_ops->free_max_bests(mp);
1899                 } else {
1900                         free = fbp->b_addr;
1901                         bests = dp->d_ops->free_bests_p(free);
1902                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1903                 }
1904
1905                 /*
1906                  * Set the freespace block index from the data block number.
1907                  */
1908                 findex = dp->d_ops->db_to_fdindex(mp, dbno);
1909                 /*
1910                  * If it's after the end of the current entries in the
1911                  * freespace block, extend that table.
1912                  */
1913                 if (findex >= freehdr.nvalid) {
1914                         ASSERT(findex < dp->d_ops->free_max_bests(mp));
1915                         freehdr.nvalid = findex + 1;
1916                         /*
1917                          * Tag new entry so nused will go up.
1918                          */
1919                         bests[findex] = cpu_to_be16(NULLDATAOFF);
1920                 }
1921                 /*
1922                  * If this entry was for an empty data block
1923                  * (this should always be true) then update the header.
1924                  */
1925                 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1926                         freehdr.nused++;
1927                         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
1928                         xfs_dir2_free_log_header(tp, dp, fbp);
1929                 }
1930                 /*
1931                  * Update the real value in the table.
1932                  * We haven't allocated the data entry yet so this will
1933                  * change again.
1934                  */
1935                 hdr = dbp->b_addr;
1936                 bf = dp->d_ops->data_bestfree_p(hdr);
1937                 bests[findex] = bf[0].length;
1938                 logfree = 1;
1939         }
1940         /*
1941          * We had a data block so we don't have to make a new one.
1942          */
1943         else {
1944                 /*
1945                  * If just checking, we succeeded.
1946                  */
1947                 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1948                         return 0;
1949
1950                 /*
1951                  * Read the data block in.
1952                  */
1953                 error = xfs_dir3_data_read(tp, dp,
1954                                            xfs_dir2_db_to_da(args->geo, dbno),
1955                                            -1, &dbp);
1956                 if (error)
1957                         return error;
1958                 hdr = dbp->b_addr;
1959                 bf = dp->d_ops->data_bestfree_p(hdr);
1960                 logfree = 0;
1961         }
1962         ASSERT(be16_to_cpu(bf[0].length) >= length);
1963         /*
1964          * Point to the existing unused space.
1965          */
1966         dup = (xfs_dir2_data_unused_t *)
1967               ((char *)hdr + be16_to_cpu(bf[0].offset));
1968         needscan = needlog = 0;
1969         /*
1970          * Mark the first part of the unused space, inuse for us.
1971          */
1972         xfs_dir2_data_use_free(tp, dp, dbp, dup,
1973                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
1974                 &needlog, &needscan);
1975         /*
1976          * Fill in the new entry and log it.
1977          */
1978         dep = (xfs_dir2_data_entry_t *)dup;
1979         dep->inumber = cpu_to_be64(args->inumber);
1980         dep->namelen = args->namelen;
1981         memcpy(dep->name, args->name, dep->namelen);
1982         dp->d_ops->data_put_ftype(dep, args->filetype);
1983         tagp = dp->d_ops->data_entry_tag_p(dep);
1984         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1985         xfs_dir2_data_log_entry(tp, dp, dbp, dep);
1986         /*
1987          * Rescan the block for bestfree if needed.
1988          */
1989         if (needscan)
1990                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1991         /*
1992          * Log the data block header if needed.
1993          */
1994         if (needlog)
1995                 xfs_dir2_data_log_header(tp, dp, dbp);
1996         /*
1997          * If the freespace entry is now wrong, update it.
1998          */
1999         bests = dp->d_ops->free_bests_p(free); /* gcc is so stupid */
2000         if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
2001                 bests[findex] = bf[0].length;
2002                 logfree = 1;
2003         }
2004         /*
2005          * Log the freespace entry if needed.
2006          */
2007         if (logfree)
2008                 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
2009         /*
2010          * Return the data block and offset in args, then drop the data block.
2011          */
2012         args->blkno = (xfs_dablk_t)dbno;
2013         args->index = be16_to_cpu(*tagp);
2014         return 0;
2015 }
2016
2017 /*
2018  * Lookup an entry in a node-format directory.
2019  * All the real work happens in xfs_da3_node_lookup_int.
2020  * The only real output is the inode number of the entry.
2021  */
2022 int                                             /* error */
2023 xfs_dir2_node_lookup(
2024         xfs_da_args_t   *args)                  /* operation arguments */
2025 {
2026         int             error;                  /* error return value */
2027         int             i;                      /* btree level */
2028         int             rval;                   /* operation return value */
2029         xfs_da_state_t  *state;                 /* btree cursor */
2030
2031         trace_xfs_dir2_node_lookup(args);
2032
2033         /*
2034          * Allocate and initialize the btree cursor.
2035          */
2036         state = xfs_da_state_alloc();
2037         state->args = args;
2038         state->mp = args->dp->i_mount;
2039         state->blocksize = state->mp->m_dirblksize;
2040         state->node_ents = state->mp->m_dir_node_ents;
2041         /*
2042          * Fill in the path to the entry in the cursor.
2043          */
2044         error = xfs_da3_node_lookup_int(state, &rval);
2045         if (error)
2046                 rval = error;
2047         else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
2048                 /* If a CI match, dup the actual name and return EEXIST */
2049                 xfs_dir2_data_entry_t   *dep;
2050
2051                 dep = (xfs_dir2_data_entry_t *)
2052                         ((char *)state->extrablk.bp->b_addr +
2053                                                  state->extrablk.index);
2054                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2055         }
2056         /*
2057          * Release the btree blocks and leaf block.
2058          */
2059         for (i = 0; i < state->path.active; i++) {
2060                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2061                 state->path.blk[i].bp = NULL;
2062         }
2063         /*
2064          * Release the data block if we have it.
2065          */
2066         if (state->extravalid && state->extrablk.bp) {
2067                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2068                 state->extrablk.bp = NULL;
2069         }
2070         xfs_da_state_free(state);
2071         return rval;
2072 }
2073
2074 /*
2075  * Remove an entry from a node-format directory.
2076  */
2077 int                                             /* error */
2078 xfs_dir2_node_removename(
2079         struct xfs_da_args      *args)          /* operation arguments */
2080 {
2081         struct xfs_da_state_blk *blk;           /* leaf block */
2082         int                     error;          /* error return value */
2083         int                     rval;           /* operation return value */
2084         struct xfs_da_state     *state;         /* btree cursor */
2085
2086         trace_xfs_dir2_node_removename(args);
2087
2088         /*
2089          * Allocate and initialize the btree cursor.
2090          */
2091         state = xfs_da_state_alloc();
2092         state->args = args;
2093         state->mp = args->dp->i_mount;
2094         state->blocksize = state->mp->m_dirblksize;
2095         state->node_ents = state->mp->m_dir_node_ents;
2096
2097         /* Look up the entry we're deleting, set up the cursor. */
2098         error = xfs_da3_node_lookup_int(state, &rval);
2099         if (error)
2100                 goto out_free;
2101
2102         /* Didn't find it, upper layer screwed up. */
2103         if (rval != EEXIST) {
2104                 error = rval;
2105                 goto out_free;
2106         }
2107
2108         blk = &state->path.blk[state->path.active - 1];
2109         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2110         ASSERT(state->extravalid);
2111         /*
2112          * Remove the leaf and data entries.
2113          * Extrablk refers to the data block.
2114          */
2115         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2116                 &state->extrablk, &rval);
2117         if (error)
2118                 goto out_free;
2119         /*
2120          * Fix the hash values up the btree.
2121          */
2122         xfs_da3_fixhashpath(state, &state->path);
2123         /*
2124          * If we need to join leaf blocks, do it.
2125          */
2126         if (rval && state->path.active > 1)
2127                 error = xfs_da3_join(state);
2128         /*
2129          * If no errors so far, try conversion to leaf format.
2130          */
2131         if (!error)
2132                 error = xfs_dir2_node_to_leaf(state);
2133 out_free:
2134         xfs_da_state_free(state);
2135         return error;
2136 }
2137
2138 /*
2139  * Replace an entry's inode number in a node-format directory.
2140  */
2141 int                                             /* error */
2142 xfs_dir2_node_replace(
2143         xfs_da_args_t           *args)          /* operation arguments */
2144 {
2145         xfs_da_state_blk_t      *blk;           /* leaf block */
2146         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
2147         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
2148         int                     error;          /* error return value */
2149         int                     i;              /* btree level */
2150         xfs_ino_t               inum;           /* new inode number */
2151         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
2152         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
2153         int                     rval;           /* internal return value */
2154         xfs_da_state_t          *state;         /* btree cursor */
2155
2156         trace_xfs_dir2_node_replace(args);
2157
2158         /*
2159          * Allocate and initialize the btree cursor.
2160          */
2161         state = xfs_da_state_alloc();
2162         state->args = args;
2163         state->mp = args->dp->i_mount;
2164         state->blocksize = state->mp->m_dirblksize;
2165         state->node_ents = state->mp->m_dir_node_ents;
2166         inum = args->inumber;
2167         /*
2168          * Lookup the entry to change in the btree.
2169          */
2170         error = xfs_da3_node_lookup_int(state, &rval);
2171         if (error) {
2172                 rval = error;
2173         }
2174         /*
2175          * It should be found, since the vnodeops layer has looked it up
2176          * and locked it.  But paranoia is good.
2177          */
2178         if (rval == EEXIST) {
2179                 struct xfs_dir2_leaf_entry *ents;
2180                 /*
2181                  * Find the leaf entry.
2182                  */
2183                 blk = &state->path.blk[state->path.active - 1];
2184                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2185                 leaf = blk->bp->b_addr;
2186                 ents = args->dp->d_ops->leaf_ents_p(leaf);
2187                 lep = &ents[blk->index];
2188                 ASSERT(state->extravalid);
2189                 /*
2190                  * Point to the data entry.
2191                  */
2192                 hdr = state->extrablk.bp->b_addr;
2193                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2194                        hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2195                 dep = (xfs_dir2_data_entry_t *)
2196                       ((char *)hdr +
2197                        xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
2198                 ASSERT(inum != be64_to_cpu(dep->inumber));
2199                 /*
2200                  * Fill in the new inode number and log the entry.
2201                  */
2202                 dep->inumber = cpu_to_be64(inum);
2203                 args->dp->d_ops->data_put_ftype(dep, args->filetype);
2204                 xfs_dir2_data_log_entry(args->trans, args->dp,
2205                                         state->extrablk.bp, dep);
2206                 rval = 0;
2207         }
2208         /*
2209          * Didn't find it, and we're holding a data block.  Drop it.
2210          */
2211         else if (state->extravalid) {
2212                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2213                 state->extrablk.bp = NULL;
2214         }
2215         /*
2216          * Release all the buffers in the cursor.
2217          */
2218         for (i = 0; i < state->path.active; i++) {
2219                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2220                 state->path.blk[i].bp = NULL;
2221         }
2222         xfs_da_state_free(state);
2223         return rval;
2224 }
2225
2226 /*
2227  * Trim off a trailing empty freespace block.
2228  * Return (in rvalp) 1 if we did it, 0 if not.
2229  */
2230 int                                             /* error */
2231 xfs_dir2_node_trim_free(
2232         xfs_da_args_t           *args,          /* operation arguments */
2233         xfs_fileoff_t           fo,             /* free block number */
2234         int                     *rvalp)         /* out: did something */
2235 {
2236         struct xfs_buf          *bp;            /* freespace buffer */
2237         xfs_inode_t             *dp;            /* incore directory inode */
2238         int                     error;          /* error return code */
2239         xfs_dir2_free_t         *free;          /* freespace structure */
2240         xfs_mount_t             *mp;            /* filesystem mount point */
2241         xfs_trans_t             *tp;            /* transaction pointer */
2242         struct xfs_dir3_icfree_hdr freehdr;
2243
2244         dp = args->dp;
2245         mp = dp->i_mount;
2246         tp = args->trans;
2247         /*
2248          * Read the freespace block.
2249          */
2250         error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2251         if (error)
2252                 return error;
2253         /*
2254          * There can be holes in freespace.  If fo is a hole, there's
2255          * nothing to do.
2256          */
2257         if (!bp)
2258                 return 0;
2259         free = bp->b_addr;
2260         dp->d_ops->free_hdr_from_disk(&freehdr, free);
2261
2262         /*
2263          * If there are used entries, there's nothing to do.
2264          */
2265         if (freehdr.nused > 0) {
2266                 xfs_trans_brelse(tp, bp);
2267                 *rvalp = 0;
2268                 return 0;
2269         }
2270         /*
2271          * Blow the block away.
2272          */
2273         error = xfs_dir2_shrink_inode(args,
2274                         xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo), bp);
2275         if (error) {
2276                 /*
2277                  * Can't fail with ENOSPC since that only happens with no
2278                  * space reservation, when breaking up an extent into two
2279                  * pieces.  This is the last block of an extent.
2280                  */
2281                 ASSERT(error != ENOSPC);
2282                 xfs_trans_brelse(tp, bp);
2283                 return error;
2284         }
2285         /*
2286          * Return that we succeeded.
2287          */
2288         *rvalp = 1;
2289         return 0;
2290 }