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