kernel/fs/xfs/libxfs/xfs_dir2_node.c

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