fs/xfs/xfs_dir2_node.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_inode.h"
  30 #include "xfs_bmap.h"
  31 #include "xfs_dir2.h"
  32 #include "xfs_dir2_priv.h"
  33 #include "xfs_error.h"
  34 #include "xfs_trace.h"
  35 #include "xfs_trans.h"
  36 #include "xfs_buf_item.h"
  37 #include "xfs_cksum.h"
  38
  39 /*
  40  * Function declarations.
  41  */
  42 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
  43                               int index);
  44 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  45                                      xfs_da_state_blk_t *blk1,
  46                                      xfs_da_state_blk_t *blk2);
  47 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
  48                                  int index, xfs_da_state_blk_t *dblk,
  49                                  int *rval);
  50 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  51                                      xfs_da_state_blk_t *fblk);
  52
  53 /*
  54  * Check internal consistency of a leafn block.
  55  */
  56 #ifdef DEBUG
  57 #define xfs_dir3_leaf_check(mp, bp) \
  58 do { \
  59         if (!xfs_dir3_leafn_check((mp), (bp))) \
  60                 ASSERT(0); \
  61 } while (0);
  62
  63 static bool
  64 xfs_dir3_leafn_check(
  65         struct xfs_mount        *mp,
  66         struct xfs_buf          *bp)
  67 {
  68         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  69         struct xfs_dir3_icleaf_hdr leafhdr;
  70
  71         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
  72
  73         if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
  74                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  75                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  76                         return false;
  77         } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
  78                 return false;
  79
  80         return xfs_dir3_leaf_check_int(mp, &leafhdr, leaf);
  81 }
  82 #else
  83 #define xfs_dir3_leaf_check(mp, bp)
  84 #endif
  85
  86 static bool
  87 xfs_dir3_free_verify(
  88         struct xfs_buf          *bp)
  89 {
  90         struct xfs_mount        *mp = bp->b_target->bt_mount;
  91         struct xfs_dir2_free_hdr *hdr = bp->b_addr;
  92
  93         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  94                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
  95
  96                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_FREE_MAGIC))
  97                         return false;
  98                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
  99                         return false;
 100                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 101                         return false;
 102         } else {
 103                 if (hdr->magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC))
 104                         return false;
 105         }
 106
 107         /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
 108
 109         return true;
 110 }
 111
 112 static void
 113 xfs_dir3_free_read_verify(
 114         struct xfs_buf  *bp)
 115 {
 116         struct xfs_mount        *mp = bp->b_target->bt_mount;
 117
 118         if ((xfs_sb_version_hascrc(&mp->m_sb) &&
 119              !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
 120                                           XFS_DIR3_FREE_CRC_OFF)) ||
 121             !xfs_dir3_free_verify(bp)) {
 122                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 123                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 124         }
 125 }
 126
 127 static void
 128 xfs_dir3_free_write_verify(
 129         struct xfs_buf  *bp)
 130 {
 131         struct xfs_mount        *mp = bp->b_target->bt_mount;
 132         struct xfs_buf_log_item *bip = bp->b_fspriv;
 133         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 134
 135         if (!xfs_dir3_free_verify(bp)) {
 136                 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
 137                 xfs_buf_ioerror(bp, EFSCORRUPTED);
 138                 return;
 139         }
 140
 141         if (!xfs_sb_version_hascrc(&mp->m_sb))
 142                 return;
 143
 144         if (bip)
 145                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 146
 147         xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_FREE_CRC_OFF);
 148 }
 149
 150 const struct xfs_buf_ops xfs_dir3_free_buf_ops = {
 151         .verify_read = xfs_dir3_free_read_verify,
 152         .verify_write = xfs_dir3_free_write_verify,
 153 };
 154
 155
 156 static int
 157 __xfs_dir3_free_read(
 158         struct xfs_trans        *tp,
 159         struct xfs_inode        *dp,
 160         xfs_dablk_t             fbno,
 161         xfs_daddr_t             mappedbno,
 162         struct xfs_buf          **bpp)
 163 {
 164         int                     err;
 165
 166         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 167                                 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
 168
 169         /* try read returns without an error or *bpp if it lands in a hole */
 170         if (!err && tp && *bpp)
 171                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
 172         return err;
 173 }
 174
 175 int
 176 xfs_dir2_free_read(
 177         struct xfs_trans        *tp,
 178         struct xfs_inode        *dp,
 179         xfs_dablk_t             fbno,
 180         struct xfs_buf          **bpp)
 181 {
 182         return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
 183 }
 184
 185 static int
 186 xfs_dir2_free_try_read(
 187         struct xfs_trans        *tp,
 188         struct xfs_inode        *dp,
 189         xfs_dablk_t             fbno,
 190         struct xfs_buf          **bpp)
 191 {
 192         return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
 193 }
 194
 195
 196 void
 197 xfs_dir3_free_hdr_from_disk(
 198         struct xfs_dir3_icfree_hdr      *to,
 199         struct xfs_dir2_free            *from)
 200 {
 201         if (from->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC)) {
 202                 to->magic = be32_to_cpu(from->hdr.magic);
 203                 to->firstdb = be32_to_cpu(from->hdr.firstdb);
 204                 to->nvalid = be32_to_cpu(from->hdr.nvalid);
 205                 to->nused = be32_to_cpu(from->hdr.nused);
 206         } else {
 207                 struct xfs_dir3_free_hdr *hdr3 = (struct xfs_dir3_free_hdr *)from;
 208
 209                 to->magic = be32_to_cpu(hdr3->hdr.magic);
 210                 to->firstdb = be32_to_cpu(hdr3->firstdb);
 211                 to->nvalid = be32_to_cpu(hdr3->nvalid);
 212                 to->nused = be32_to_cpu(hdr3->nused);
 213         }
 214
 215         ASSERT(to->magic == XFS_DIR2_FREE_MAGIC ||
 216                to->magic == XFS_DIR3_FREE_MAGIC);
 217 }
 218
 219 static void
 220 xfs_dir3_free_hdr_to_disk(
 221         struct xfs_dir2_free            *to,
 222         struct xfs_dir3_icfree_hdr      *from)
 223 {
 224         ASSERT(from->magic == XFS_DIR2_FREE_MAGIC ||
 225                from->magic == XFS_DIR3_FREE_MAGIC);
 226
 227         if (from->magic == XFS_DIR2_FREE_MAGIC) {
 228                 to->hdr.magic = cpu_to_be32(from->magic);
 229                 to->hdr.firstdb = cpu_to_be32(from->firstdb);
 230                 to->hdr.nvalid = cpu_to_be32(from->nvalid);
 231                 to->hdr.nused = cpu_to_be32(from->nused);
 232         } else {
 233                 struct xfs_dir3_free_hdr *hdr3 = (struct xfs_dir3_free_hdr *)to;
 234
 235                 hdr3->hdr.magic = cpu_to_be32(from->magic);
 236                 hdr3->firstdb = cpu_to_be32(from->firstdb);
 237                 hdr3->nvalid = cpu_to_be32(from->nvalid);
 238                 hdr3->nused = cpu_to_be32(from->nused);
 239         }
 240 }
 241
 242 static int
 243 xfs_dir3_free_get_buf(
 244         struct xfs_trans        *tp,
 245         struct xfs_inode        *dp,
 246         xfs_dir2_db_t           fbno,
 247         struct xfs_buf          **bpp)
 248 {
 249         struct xfs_mount        *mp = dp->i_mount;
 250         struct xfs_buf          *bp;
 251         int                     error;
 252         struct xfs_dir3_icfree_hdr hdr;
 253
 254         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, fbno),
 255                                    -1, &bp, XFS_DATA_FORK);
 256         if (error)
 257                 return error;
 258
 259         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
 260         bp->b_ops = &xfs_dir3_free_buf_ops;
 261
 262         /*
 263          * Initialize the new block to be empty, and remember
 264          * its first slot as our empty slot.
 265          */
 266         memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
 267         memset(&hdr, 0, sizeof(hdr));
 268
 269         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 270                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 271
 272                 hdr.magic = XFS_DIR3_FREE_MAGIC;
 273
 274                 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
 275                 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
 276                 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_uuid);
 277         } else
 278                 hdr.magic = XFS_DIR2_FREE_MAGIC;
 279         xfs_dir3_free_hdr_to_disk(bp->b_addr, &hdr);
 280         *bpp = bp;
 281         return 0;
 282 }
 283
 284 /*
 285  * Log entries from a freespace block.
 286  */
 287 STATIC void
 288 xfs_dir2_free_log_bests(
 289         struct xfs_trans        *tp,
 290         struct xfs_buf          *bp,
 291         int                     first,          /* first entry to log */
 292         int                     last)           /* last entry to log */
 293 {
 294         xfs_dir2_free_t         *free;          /* freespace structure */
 295         __be16                  *bests;
 296
 297         free = bp->b_addr;
 298         bests = xfs_dir3_free_bests_p(tp->t_mountp, free);
 299         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 300                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 301         xfs_trans_log_buf(tp, bp,
 302                 (uint)((char *)&bests[first] - (char *)free),
 303                 (uint)((char *)&bests[last] - (char *)free +
 304                        sizeof(bests[0]) - 1));
 305 }
 306
 307 /*
 308  * Log header from a freespace block.
 309  */
 310 static void
 311 xfs_dir2_free_log_header(
 312         struct xfs_trans        *tp,
 313         struct xfs_buf          *bp)
 314 {
 315 #ifdef DEBUG
 316         xfs_dir2_free_t         *free;          /* freespace structure */
 317
 318         free = bp->b_addr;
 319         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 320                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 321 #endif
 322         xfs_trans_log_buf(tp, bp, 0, xfs_dir3_free_hdr_size(tp->t_mountp) - 1);
 323 }
 324
 325 /*
 326  * Convert a leaf-format directory to a node-format directory.
 327  * We need to change the magic number of the leaf block, and copy
 328  * the freespace table out of the leaf block into its own block.
 329  */
 330 int                                             /* error */
 331 xfs_dir2_leaf_to_node(
 332         xfs_da_args_t           *args,          /* operation arguments */
 333         struct xfs_buf          *lbp)           /* leaf buffer */
 334 {
 335         xfs_inode_t             *dp;            /* incore directory inode */
 336         int                     error;          /* error return value */
 337         struct xfs_buf          *fbp;           /* freespace buffer */
 338         xfs_dir2_db_t           fdb;            /* freespace block number */
 339         xfs_dir2_free_t         *free;          /* freespace structure */
 340         __be16                  *from;          /* pointer to freespace entry */
 341         int                     i;              /* leaf freespace index */
 342         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 343         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 344         xfs_mount_t             *mp;            /* filesystem mount point */
 345         int                     n;              /* count of live freespc ents */
 346         xfs_dir2_data_off_t     off;            /* freespace entry value */
 347         __be16                  *to;            /* pointer to freespace entry */
 348         xfs_trans_t             *tp;            /* transaction pointer */
 349         struct xfs_dir3_icfree_hdr freehdr;
 350
 351         trace_xfs_dir2_leaf_to_node(args);
 352
 353         dp = args->dp;
 354         mp = dp->i_mount;
 355         tp = args->trans;
 356         /*
 357          * Add a freespace block to the directory.
 358          */
 359         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 360                 return error;
 361         }
 362         ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
 363         /*
 364          * Get the buffer for the new freespace block.
 365          */
 366         error = xfs_dir3_free_get_buf(tp, dp, fdb, &fbp);
 367         if (error)
 368                 return error;
 369
 370         free = fbp->b_addr;
 371         xfs_dir3_free_hdr_from_disk(&freehdr, free);
 372         leaf = lbp->b_addr;
 373         ltp = xfs_dir2_leaf_tail_p(mp, leaf);
 374         ASSERT(be32_to_cpu(ltp->bestcount) <=
 375                                 (uint)dp->i_d.di_size / mp->m_dirblksize);
 376
 377         /*
 378          * Copy freespace entries from the leaf block to the new block.
 379          * Count active entries.
 380          */
 381         from = xfs_dir2_leaf_bests_p(ltp);
 382         to = xfs_dir3_free_bests_p(mp, free);
 383         for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 384                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 385                         n++;
 386                 *to = cpu_to_be16(off);
 387         }
 388
 389         /*
 390          * Now initialize the freespace block header.
 391          */
 392         freehdr.nused = n;
 393         freehdr.nvalid = be32_to_cpu(ltp->bestcount);
 394
 395         xfs_dir3_free_hdr_to_disk(fbp->b_addr, &freehdr);
 396         xfs_dir2_free_log_bests(tp, fbp, 0, freehdr.nvalid - 1);
 397         xfs_dir2_free_log_header(tp, fbp);
 398
 399         /*
 400          * Converting the leaf to a leafnode is just a matter of changing the
 401          * magic number and the ops. Do the change directly to the buffer as
 402          * it's less work (and less code) than decoding the header to host
 403          * format and back again.
 404          */
 405         if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
 406                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 407         else
 408                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 409         lbp->b_ops = &xfs_dir3_leafn_buf_ops;
 410         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
 411         xfs_dir3_leaf_log_header(tp, lbp);
 412         xfs_dir3_leaf_check(mp, lbp);
 413         return 0;
 414 }
 415
 416 /*
 417  * Add a leaf entry to a leaf block in a node-form directory.
 418  * The other work necessary is done from the caller.
 419  */
 420 static int                                      /* error */
 421 xfs_dir2_leafn_add(
 422         struct xfs_buf          *bp,            /* leaf buffer */
 423         xfs_da_args_t           *args,          /* operation arguments */
 424         int                     index)          /* insertion pt for new entry */
 425 {
 426         int                     compact;        /* compacting stale leaves */
 427         xfs_inode_t             *dp;            /* incore directory inode */
 428         int                     highstale;      /* next stale entry */
 429         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 430         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 431         int                     lfloghigh;      /* high leaf entry logging */
 432         int                     lfloglow;       /* low leaf entry logging */
 433         int                     lowstale;       /* previous stale entry */
 434         xfs_mount_t             *mp;            /* filesystem mount point */
 435         xfs_trans_t             *tp;            /* transaction pointer */
 436         struct xfs_dir3_icleaf_hdr leafhdr;
 437         struct xfs_dir2_leaf_entry *ents;
 438
 439         trace_xfs_dir2_leafn_add(args, index);
 440
 441         dp = args->dp;
 442         mp = dp->i_mount;
 443         tp = args->trans;
 444         leaf = bp->b_addr;
 445         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 446         ents = xfs_dir3_leaf_ents_p(leaf);
 447
 448         /*
 449          * Quick check just to make sure we are not going to index
 450          * into other peoples memory
 451          */
 452         if (index < 0)
 453                 return XFS_ERROR(EFSCORRUPTED);
 454
 455         /*
 456          * If there are already the maximum number of leaf entries in
 457          * the block, if there are no stale entries it won't fit.
 458          * Caller will do a split.  If there are stale entries we'll do
 459          * a compact.
 460          */
 461
 462         if (leafhdr.count == xfs_dir3_max_leaf_ents(mp, leaf)) {
 463                 if (!leafhdr.stale)
 464                         return XFS_ERROR(ENOSPC);
 465                 compact = leafhdr.stale > 1;
 466         } else
 467                 compact = 0;
 468         ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
 469         ASSERT(index == leafhdr.count ||
 470                be32_to_cpu(ents[index].hashval) >= args->hashval);
 471
 472         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 473                 return 0;
 474
 475         /*
 476          * Compact out all but one stale leaf entry.  Leaves behind
 477          * the entry closest to index.
 478          */
 479         if (compact)
 480                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 481                                          &highstale, &lfloglow, &lfloghigh);
 482         else if (leafhdr.stale) {
 483                 /*
 484                  * Set impossible logging indices for this case.
 485                  */
 486                 lfloglow = leafhdr.count;
 487                 lfloghigh = -1;
 488         }
 489
 490         /*
 491          * Insert the new entry, log everything.
 492          */
 493         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 494                                        highstale, &lfloglow, &lfloghigh);
 495
 496         lep->hashval = cpu_to_be32(args->hashval);
 497         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
 498                                 args->blkno, args->index));
 499
 500         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
 501         xfs_dir3_leaf_log_header(tp, bp);
 502         xfs_dir3_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
 503         xfs_dir3_leaf_check(mp, bp);
 504         return 0;
 505 }
 506
 507 #ifdef DEBUG
 508 static void
 509 xfs_dir2_free_hdr_check(
 510         struct xfs_mount *mp,
 511         struct xfs_buf  *bp,
 512         xfs_dir2_db_t   db)
 513 {
 514         struct xfs_dir3_icfree_hdr hdr;
 515
 516         xfs_dir3_free_hdr_from_disk(&hdr, bp->b_addr);
 517
 518         ASSERT((hdr.firstdb % xfs_dir3_free_max_bests(mp)) == 0);
 519         ASSERT(hdr.firstdb <= db);
 520         ASSERT(db < hdr.firstdb + hdr.nvalid);
 521 }
 522 #else
 523 #define xfs_dir2_free_hdr_check(mp, dp, db)
 524 #endif  /* DEBUG */
 525
 526 /*
 527  * Return the last hash value in the leaf.
 528  * Stale entries are ok.
 529  */
 530 xfs_dahash_t                                    /* hash value */
 531 xfs_dir2_leafn_lasthash(
 532         struct xfs_buf  *bp,                    /* leaf buffer */
 533         int             *count)                 /* count of entries in leaf */
 534 {
 535         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 536         struct xfs_dir2_leaf_entry *ents;
 537         struct xfs_dir3_icleaf_hdr leafhdr;
 538
 539         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 540
 541         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 542                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
 543
 544         if (count)
 545                 *count = leafhdr.count;
 546         if (!leafhdr.count)
 547                 return 0;
 548
 549         ents = xfs_dir3_leaf_ents_p(leaf);
 550         return be32_to_cpu(ents[leafhdr.count - 1].hashval);
 551 }
 552
 553 /*
 554  * Look up a leaf entry for space to add a name in a node-format leaf block.
 555  * The extrablk in state is a freespace block.
 556  */
 557 STATIC int
 558 xfs_dir2_leafn_lookup_for_addname(
 559         struct xfs_buf          *bp,            /* leaf buffer */
 560         xfs_da_args_t           *args,          /* operation arguments */
 561         int                     *indexp,        /* out: leaf entry index */
 562         xfs_da_state_t          *state)         /* state to fill in */
 563 {
 564         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 565         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 566         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 567         xfs_inode_t             *dp;            /* incore directory inode */
 568         int                     error;          /* error return value */
 569         int                     fi;             /* free entry index */
 570         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 571         int                     index;          /* leaf entry index */
 572         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 573         int                     length;         /* length of new data entry */
 574         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 575         xfs_mount_t             *mp;            /* filesystem mount point */
 576         xfs_dir2_db_t           newdb;          /* new data block number */
 577         xfs_dir2_db_t           newfdb;         /* new free block number */
 578         xfs_trans_t             *tp;            /* transaction pointer */
 579         struct xfs_dir2_leaf_entry *ents;
 580         struct xfs_dir3_icleaf_hdr leafhdr;
 581
 582         dp = args->dp;
 583         tp = args->trans;
 584         mp = dp->i_mount;
 585         leaf = bp->b_addr;
 586         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 587         ents = xfs_dir3_leaf_ents_p(leaf);
 588
 589         xfs_dir3_leaf_check(mp, bp);
 590         ASSERT(leafhdr.count > 0);
 591
 592         /*
 593          * Look up the hash value in the leaf entries.
 594          */
 595         index = xfs_dir2_leaf_search_hash(args, bp);
 596         /*
 597          * Do we have a buffer coming in?
 598          */
 599         if (state->extravalid) {
 600                 /* If so, it's a free block buffer, get the block number. */
 601                 curbp = state->extrablk.bp;
 602                 curfdb = state->extrablk.blkno;
 603                 free = curbp->b_addr;
 604                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 605                        free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 606         }
 607         length = xfs_dir3_data_entsize(mp, args->namelen);
 608         /*
 609          * Loop over leaf entries with the right hash value.
 610          */
 611         for (lep = &ents[index];
 612              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 613              lep++, index++) {
 614                 /*
 615                  * Skip stale leaf entries.
 616                  */
 617                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 618                         continue;
 619                 /*
 620                  * Pull the data block number from the entry.
 621                  */
 622                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 623                 /*
 624                  * For addname, we're looking for a place to put the new entry.
 625                  * We want to use a data block with an entry of equal
 626                  * hash value to ours if there is one with room.
 627                  *
 628                  * If this block isn't the data block we already have
 629                  * in hand, take a look at it.
 630                  */
 631                 if (newdb != curdb) {
 632                         __be16 *bests;
 633
 634                         curdb = newdb;
 635                         /*
 636                          * Convert the data block to the free block
 637                          * holding its freespace information.
 638                          */
 639                         newfdb = xfs_dir2_db_to_fdb(mp, newdb);
 640                         /*
 641                          * If it's not the one we have in hand, read it in.
 642                          */
 643                         if (newfdb != curfdb) {
 644                                 /*
 645                                  * If we had one before, drop it.
 646                                  */
 647                                 if (curbp)
 648                                         xfs_trans_brelse(tp, curbp);
 649
 650                                 error = xfs_dir2_free_read(tp, dp,
 651                                                 xfs_dir2_db_to_da(mp, newfdb),
 652                                                 &curbp);
 653                                 if (error)
 654                                         return error;
 655                                 free = curbp->b_addr;
 656
 657                                 xfs_dir2_free_hdr_check(mp, curbp, curdb);
 658                         }
 659                         /*
 660                          * Get the index for our entry.
 661                          */
 662                         fi = xfs_dir2_db_to_fdindex(mp, curdb);
 663                         /*
 664                          * If it has room, return it.
 665                          */
 666                         bests = xfs_dir3_free_bests_p(mp, free);
 667                         if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
 668                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 669                                                         XFS_ERRLEVEL_LOW, mp);
 670                                 if (curfdb != newfdb)
 671                                         xfs_trans_brelse(tp, curbp);
 672                                 return XFS_ERROR(EFSCORRUPTED);
 673                         }
 674                         curfdb = newfdb;
 675                         if (be16_to_cpu(bests[fi]) >= length)
 676                                 goto out;
 677                 }
 678         }
 679         /* Didn't find any space */
 680         fi = -1;
 681 out:
 682         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 683         if (curbp) {
 684                 /* Giving back a free block. */
 685                 state->extravalid = 1;
 686                 state->extrablk.bp = curbp;
 687                 state->extrablk.index = fi;
 688                 state->extrablk.blkno = curfdb;
 689
 690                 /*
 691                  * Important: this magic number is not in the buffer - it's for
 692                  * buffer type information and therefore only the free/data type
 693                  * matters here, not whether CRCs are enabled or not.
 694                  */
 695                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 696         } else {
 697                 state->extravalid = 0;
 698         }
 699         /*
 700          * Return the index, that will be the insertion point.
 701          */
 702         *indexp = index;
 703         return XFS_ERROR(ENOENT);
 704 }
 705
 706 /*
 707  * Look up a leaf entry in a node-format leaf block.
 708  * The extrablk in state a data block.
 709  */
 710 STATIC int
 711 xfs_dir2_leafn_lookup_for_entry(
 712         struct xfs_buf          *bp,            /* leaf buffer */
 713         xfs_da_args_t           *args,          /* operation arguments */
 714         int                     *indexp,        /* out: leaf entry index */
 715         xfs_da_state_t          *state)         /* state to fill in */
 716 {
 717         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 718         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 719         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 720         xfs_inode_t             *dp;            /* incore directory inode */
 721         int                     error;          /* error return value */
 722         int                     index;          /* leaf entry index */
 723         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 724         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 725         xfs_mount_t             *mp;            /* filesystem mount point */
 726         xfs_dir2_db_t           newdb;          /* new data block number */
 727         xfs_trans_t             *tp;            /* transaction pointer */
 728         enum xfs_dacmp          cmp;            /* comparison result */
 729         struct xfs_dir2_leaf_entry *ents;
 730         struct xfs_dir3_icleaf_hdr leafhdr;
 731
 732         dp = args->dp;
 733         tp = args->trans;
 734         mp = dp->i_mount;
 735         leaf = bp->b_addr;
 736         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
 737         ents = xfs_dir3_leaf_ents_p(leaf);
 738
 739         xfs_dir3_leaf_check(mp, bp);
 740         ASSERT(leafhdr.count > 0);
 741
 742         /*
 743          * Look up the hash value in the leaf entries.
 744          */
 745         index = xfs_dir2_leaf_search_hash(args, bp);
 746         /*
 747          * Do we have a buffer coming in?
 748          */
 749         if (state->extravalid) {
 750                 curbp = state->extrablk.bp;
 751                 curdb = state->extrablk.blkno;
 752         }
 753         /*
 754          * Loop over leaf entries with the right hash value.
 755          */
 756         for (lep = &ents[index];
 757              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 758              lep++, index++) {
 759                 /*
 760                  * Skip stale leaf entries.
 761                  */
 762                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 763                         continue;
 764                 /*
 765                  * Pull the data block number from the entry.
 766                  */
 767                 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
 768                 /*
 769                  * Not adding a new entry, so we really want to find
 770                  * the name given to us.
 771                  *
 772                  * If it's a different data block, go get it.
 773                  */
 774                 if (newdb != curdb) {
 775                         /*
 776                          * If we had a block before that we aren't saving
 777                          * for a CI name, drop it
 778                          */
 779                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 780                                                 curdb != state->extrablk.blkno))
 781                                 xfs_trans_brelse(tp, curbp);
 782                         /*
 783                          * If needing the block that is saved with a CI match,
 784                          * use it otherwise read in the new data block.
 785                          */
 786                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 787                                         newdb == state->extrablk.blkno) {
 788                                 ASSERT(state->extravalid);
 789                                 curbp = state->extrablk.bp;
 790                         } else {
 791                                 error = xfs_dir3_data_read(tp, dp,
 792                                                 xfs_dir2_db_to_da(mp, newdb),
 793                                                 -1, &curbp);
 794                                 if (error)
 795                                         return error;
 796                         }
 797                         xfs_dir3_data_check(dp, curbp);
 798                         curdb = newdb;
 799                 }
 800                 /*
 801                  * Point to the data entry.
 802                  */
 803                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 804                         xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
 805                 /*
 806                  * Compare the entry and if it's an exact match, return
 807                  * EEXIST immediately. If it's the first case-insensitive
 808                  * match, store the block & inode number and continue looking.
 809                  */
 810                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 811                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 812                         /* If there is a CI match block, drop it */
 813                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 814                                                 curdb != state->extrablk.blkno)
 815                                 xfs_trans_brelse(tp, state->extrablk.bp);
 816                         args->cmpresult = cmp;
 817                         args->inumber = be64_to_cpu(dep->inumber);
 818                         args->filetype = xfs_dir3_dirent_get_ftype(mp, dep);
 819                         *indexp = index;
 820                         state->extravalid = 1;
 821                         state->extrablk.bp = curbp;
 822                         state->extrablk.blkno = curdb;
 823                         state->extrablk.index = (int)((char *)dep -
 824                                                         (char *)curbp->b_addr);
 825                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 826                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 827                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 828                         if (cmp == XFS_CMP_EXACT)
 829                                 return XFS_ERROR(EEXIST);
 830                 }
 831         }
 832         ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
 833         if (curbp) {
 834                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 835                         /* Giving back last used data block. */
 836                         state->extravalid = 1;
 837                         state->extrablk.bp = curbp;
 838                         state->extrablk.index = -1;
 839                         state->extrablk.blkno = curdb;
 840                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 841                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 842                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 843                 } else {
 844                         /* If the curbp is not the CI match block, drop it */
 845                         if (state->extrablk.bp != curbp)
 846                                 xfs_trans_brelse(tp, curbp);
 847                 }
 848         } else {
 849                 state->extravalid = 0;
 850         }
 851         *indexp = index;
 852         return XFS_ERROR(ENOENT);
 853 }
 854
 855 /*
 856  * Look up a leaf entry in a node-format leaf block.
 857  * If this is an addname then the extrablk in state is a freespace block,
 858  * otherwise it's a data block.
 859  */
 860 int
 861 xfs_dir2_leafn_lookup_int(
 862         struct xfs_buf          *bp,            /* leaf buffer */
 863         xfs_da_args_t           *args,          /* operation arguments */
 864         int                     *indexp,        /* out: leaf entry index */
 865         xfs_da_state_t          *state)         /* state to fill in */
 866 {
 867         if (args->op_flags & XFS_DA_OP_ADDNAME)
 868                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 869                                                         state);
 870         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 871 }
 872
 873 /*
 874  * Move count leaf entries from source to destination leaf.
 875  * Log entries and headers.  Stale entries are preserved.
 876  */
 877 static void
 878 xfs_dir3_leafn_moveents(
 879         xfs_da_args_t                   *args,  /* operation arguments */
 880         struct xfs_buf                  *bp_s,  /* source */
 881         struct xfs_dir3_icleaf_hdr      *shdr,
 882         struct xfs_dir2_leaf_entry      *sents,
 883         int                             start_s,/* source leaf index */
 884         struct xfs_buf                  *bp_d,  /* destination */
 885         struct xfs_dir3_icleaf_hdr      *dhdr,
 886         struct xfs_dir2_leaf_entry      *dents,
 887         int                             start_d,/* destination leaf index */
 888         int                             count)  /* count of leaves to copy */
 889 {
 890         struct xfs_trans                *tp = args->trans;
 891         int                             stale;  /* count stale leaves copied */
 892
 893         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 894
 895         /*
 896          * Silently return if nothing to do.
 897          */
 898         if (count == 0)
 899                 return;
 900
 901         /*
 902          * If the destination index is not the end of the current
 903          * destination leaf entries, open up a hole in the destination
 904          * to hold the new entries.
 905          */
 906         if (start_d < dhdr->count) {
 907                 memmove(&dents[start_d + count], &dents[start_d],
 908                         (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
 909                 xfs_dir3_leaf_log_ents(tp, bp_d, start_d + count,
 910                                        count + dhdr->count - 1);
 911         }
 912         /*
 913          * If the source has stale leaves, count the ones in the copy range
 914          * so we can update the header correctly.
 915          */
 916         if (shdr->stale) {
 917                 int     i;                      /* temp leaf index */
 918
 919                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 920                         if (sents[i].address ==
 921                                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 922                                 stale++;
 923                 }
 924         } else
 925                 stale = 0;
 926         /*
 927          * Copy the leaf entries from source to destination.
 928          */
 929         memcpy(&dents[start_d], &sents[start_s],
 930                 count * sizeof(xfs_dir2_leaf_entry_t));
 931         xfs_dir3_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
 932
 933         /*
 934          * If there are source entries after the ones we copied,
 935          * delete the ones we copied by sliding the next ones down.
 936          */
 937         if (start_s + count < shdr->count) {
 938                 memmove(&sents[start_s], &sents[start_s + count],
 939                         count * sizeof(xfs_dir2_leaf_entry_t));
 940                 xfs_dir3_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
 941         }
 942
 943         /*
 944          * Update the headers and log them.
 945          */
 946         shdr->count -= count;
 947         shdr->stale -= stale;
 948         dhdr->count += count;
 949         dhdr->stale += stale;
 950 }
 951
 952 /*
 953  * Determine the sort order of two leaf blocks.
 954  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 955  */
 956 int                                             /* sort order */
 957 xfs_dir2_leafn_order(
 958         struct xfs_buf          *leaf1_bp,              /* leaf1 buffer */
 959         struct xfs_buf          *leaf2_bp)              /* leaf2 buffer */
 960 {
 961         struct xfs_dir2_leaf    *leaf1 = leaf1_bp->b_addr;
 962         struct xfs_dir2_leaf    *leaf2 = leaf2_bp->b_addr;
 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
 968         xfs_dir3_leaf_hdr_from_disk(&hdr1, leaf1);
 969         xfs_dir3_leaf_hdr_from_disk(&hdr2, leaf2);
 970         ents1 = xfs_dir3_leaf_ents_p(leaf1);
 971         ents2 = xfs_dir3_leaf_ents_p(leaf2);
 972
 973         if (hdr1.count > 0 && hdr2.count > 0 &&
 974             (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
 975              be32_to_cpu(ents2[hdr2.count - 1].hashval) <
 976                                 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
 977                 return 1;
 978         return 0;
 979 }
 980
 981 /*
 982  * Rebalance leaf entries between two leaf blocks.
 983  * This is actually only called when the second block is new,
 984  * though the code deals with the general case.
 985  * A new entry will be inserted in one of the blocks, and that
 986  * entry is taken into account when balancing.
 987  */
 988 static void
 989 xfs_dir2_leafn_rebalance(
 990         xfs_da_state_t          *state,         /* btree cursor */
 991         xfs_da_state_blk_t      *blk1,          /* first btree block */
 992         xfs_da_state_blk_t      *blk2)          /* second btree block */
 993 {
 994         xfs_da_args_t           *args;          /* operation arguments */
 995         int                     count;          /* count (& direction) leaves */
 996         int                     isleft;         /* new goes in left leaf */
 997         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
 998         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
 999         int                     mid;            /* midpoint leaf index */
1000 #if defined(DEBUG) || defined(XFS_WARN)
1001         int                     oldstale;       /* old count of stale leaves */
1002 #endif
1003         int                     oldsum;         /* old total leaf count */
1004         int                     swap;           /* swapped leaf blocks */
1005         struct xfs_dir2_leaf_entry *ents1;
1006         struct xfs_dir2_leaf_entry *ents2;
1007         struct xfs_dir3_icleaf_hdr hdr1;
1008         struct xfs_dir3_icleaf_hdr hdr2;
1009
1010         args = state->args;
1011         /*
1012          * If the block order is wrong, swap the arguments.
1013          */
1014         if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
1015                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
1016
1017                 tmp = blk1;
1018                 blk1 = blk2;
1019                 blk2 = tmp;
1020         }
1021         leaf1 = blk1->bp->b_addr;
1022         leaf2 = blk2->bp->b_addr;
1023         xfs_dir3_leaf_hdr_from_disk(&hdr1, leaf1);
1024         xfs_dir3_leaf_hdr_from_disk(&hdr2, leaf2);
1025         ents1 = xfs_dir3_leaf_ents_p(leaf1);
1026         ents2 = xfs_dir3_leaf_ents_p(leaf2);
1027
1028         oldsum = hdr1.count + hdr2.count;
1029 #if defined(DEBUG) || defined(XFS_WARN)
1030         oldstale = hdr1.stale + hdr2.stale;
1031 #endif
1032         mid = oldsum >> 1;
1033
1034         /*
1035          * If the old leaf count was odd then the new one will be even,
1036          * so we need to divide the new count evenly.
1037          */
1038         if (oldsum & 1) {
1039                 xfs_dahash_t    midhash;        /* middle entry hash value */
1040
1041                 if (mid >= hdr1.count)
1042                         midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1043                 else
1044                         midhash = be32_to_cpu(ents1[mid].hashval);
1045                 isleft = args->hashval <= midhash;
1046         }
1047         /*
1048          * If the old count is even then the new count is odd, so there's
1049          * no preferred side for the new entry.
1050          * Pick the left one.
1051          */
1052         else
1053                 isleft = 1;
1054         /*
1055          * Calculate moved entry count.  Positive means left-to-right,
1056          * negative means right-to-left.  Then move the entries.
1057          */
1058         count = hdr1.count - mid + (isleft == 0);
1059         if (count > 0)
1060                 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1061                                         hdr1.count - count, blk2->bp,
1062                                         &hdr2, ents2, 0, count);
1063         else if (count < 0)
1064                 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1065                                         blk1->bp, &hdr1, ents1,
1066                                         hdr1.count, count);
1067
1068         ASSERT(hdr1.count + hdr2.count == oldsum);
1069         ASSERT(hdr1.stale + hdr2.stale == oldstale);
1070
1071         /* log the changes made when moving the entries */
1072         xfs_dir3_leaf_hdr_to_disk(leaf1, &hdr1);
1073         xfs_dir3_leaf_hdr_to_disk(leaf2, &hdr2);
1074         xfs_dir3_leaf_log_header(args->trans, blk1->bp);
1075         xfs_dir3_leaf_log_header(args->trans, blk2->bp);
1076
1077         xfs_dir3_leaf_check(args->dp->i_mount, blk1->bp);
1078         xfs_dir3_leaf_check(args->dp->i_mount, blk2->bp);
1079
1080         /*
1081          * Mark whether we're inserting into the old or new leaf.
1082          */
1083         if (hdr1.count < hdr2.count)
1084                 state->inleaf = swap;
1085         else if (hdr1.count > hdr2.count)
1086                 state->inleaf = !swap;
1087         else
1088                 state->inleaf = swap ^ (blk1->index <= hdr1.count);
1089         /*
1090          * Adjust the expected index for insertion.
1091          */
1092         if (!state->inleaf)
1093                 blk2->index = blk1->index - hdr1.count;
1094
1095         /*
1096          * Finally sanity check just to make sure we are not returning a
1097          * negative index
1098          */
1099         if(blk2->index < 0) {
1100                 state->inleaf = 1;
1101                 blk2->index = 0;
1102                 xfs_alert(args->dp->i_mount,
1103         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1104                         __func__, blk1->index);
1105         }
1106 }
1107
1108 static int
1109 xfs_dir3_data_block_free(
1110         xfs_da_args_t           *args,
1111         struct xfs_dir2_data_hdr *hdr,
1112         struct xfs_dir2_free    *free,
1113         xfs_dir2_db_t           fdb,
1114         int                     findex,
1115         struct xfs_buf          *fbp,
1116         int                     longest)
1117 {
1118         struct xfs_trans        *tp = args->trans;
1119         int                     logfree = 0;
1120         __be16                  *bests;
1121         struct xfs_dir3_icfree_hdr freehdr;
1122
1123         xfs_dir3_free_hdr_from_disk(&freehdr, free);
1124
1125         bests = xfs_dir3_free_bests_p(tp->t_mountp, free);
1126         if (hdr) {
1127                 /*
1128                  * Data block is not empty, just set the free entry to the new
1129                  * value.
1130                  */
1131                 bests[findex] = cpu_to_be16(longest);
1132                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1133                 return 0;
1134         }
1135
1136         /* One less used entry in the free table. */
1137         freehdr.nused--;
1138
1139         /*
1140          * If this was the last entry in the table, we can trim the table size
1141          * back.  There might be other entries at the end referring to
1142          * non-existent data blocks, get those too.
1143          */
1144         if (findex == freehdr.nvalid - 1) {
1145                 int     i;              /* free entry index */
1146
1147                 for (i = findex - 1; i >= 0; i--) {
1148                         if (bests[i] != cpu_to_be16(NULLDATAOFF))
1149                                 break;
1150                 }
1151                 freehdr.nvalid = i + 1;
1152                 logfree = 0;
1153         } else {
1154                 /* Not the last entry, just punch it out.  */
1155                 bests[findex] = cpu_to_be16(NULLDATAOFF);
1156                 logfree = 1;
1157         }
1158
1159         xfs_dir3_free_hdr_to_disk(free, &freehdr);
1160         xfs_dir2_free_log_header(tp, fbp);
1161
1162         /*
1163          * If there are no useful entries left in the block, get rid of the
1164          * block if we can.
1165          */
1166         if (!freehdr.nused) {
1167                 int error;
1168
1169                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1170                 if (error == 0) {
1171                         fbp = NULL;
1172                         logfree = 0;
1173                 } else if (error != ENOSPC || args->total != 0)
1174                         return error;
1175                 /*
1176                  * It's possible to get ENOSPC if there is no
1177                  * space reservation.  In this case some one
1178                  * else will eventually get rid of this block.
1179                  */
1180         }
1181
1182         /* Log the free entry that changed, unless we got rid of it.  */
1183         if (logfree)
1184                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1185         return 0;
1186 }
1187
1188 /*
1189  * Remove an entry from a node directory.
1190  * This removes the leaf entry and the data entry,
1191  * and updates the free block if necessary.
1192  */
1193 static int                                      /* error */
1194 xfs_dir2_leafn_remove(
1195         xfs_da_args_t           *args,          /* operation arguments */
1196         struct xfs_buf          *bp,            /* leaf buffer */
1197         int                     index,          /* leaf entry index */
1198         xfs_da_state_blk_t      *dblk,          /* data block */
1199         int                     *rval)          /* resulting block needs join */
1200 {
1201         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1202         xfs_dir2_db_t           db;             /* data block number */
1203         struct xfs_buf          *dbp;           /* data block buffer */
1204         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1205         xfs_inode_t             *dp;            /* incore directory inode */
1206         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1207         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1208         int                     longest;        /* longest data free entry */
1209         int                     off;            /* data block entry offset */
1210         xfs_mount_t             *mp;            /* filesystem mount point */
1211         int                     needlog;        /* need to log data header */
1212         int                     needscan;       /* need to rescan data frees */
1213         xfs_trans_t             *tp;            /* transaction pointer */
1214         struct xfs_dir2_data_free *bf;          /* bestfree table */
1215         struct xfs_dir3_icleaf_hdr leafhdr;
1216         struct xfs_dir2_leaf_entry *ents;
1217
1218         trace_xfs_dir2_leafn_remove(args, index);
1219
1220         dp = args->dp;
1221         tp = args->trans;
1222         mp = dp->i_mount;
1223         leaf = bp->b_addr;
1224         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1225         ents = xfs_dir3_leaf_ents_p(leaf);
1226
1227         /*
1228          * Point to the entry we're removing.
1229          */
1230         lep = &ents[index];
1231
1232         /*
1233          * Extract the data block and offset from the entry.
1234          */
1235         db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1236         ASSERT(dblk->blkno == db);
1237         off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
1238         ASSERT(dblk->index == off);
1239
1240         /*
1241          * Kill the leaf entry by marking it stale.
1242          * Log the leaf block changes.
1243          */
1244         leafhdr.stale++;
1245         xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
1246         xfs_dir3_leaf_log_header(tp, bp);
1247
1248         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1249         xfs_dir3_leaf_log_ents(tp, bp, index, index);
1250
1251         /*
1252          * Make the data entry free.  Keep track of the longest freespace
1253          * in the data block in case it changes.
1254          */
1255         dbp = dblk->bp;
1256         hdr = dbp->b_addr;
1257         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1258         bf = xfs_dir3_data_bestfree_p(hdr);
1259         longest = be16_to_cpu(bf[0].length);
1260         needlog = needscan = 0;
1261         xfs_dir2_data_make_free(tp, dbp, off,
1262                 xfs_dir3_data_entsize(mp, dep->namelen), &needlog, &needscan);
1263         /*
1264          * Rescan the data block freespaces for bestfree.
1265          * Log the data block header if needed.
1266          */
1267         if (needscan)
1268                 xfs_dir2_data_freescan(mp, hdr, &needlog);
1269         if (needlog)
1270                 xfs_dir2_data_log_header(tp, dbp);
1271         xfs_dir3_data_check(dp, dbp);
1272         /*
1273          * If the longest data block freespace changes, need to update
1274          * the corresponding freeblock entry.
1275          */
1276         if (longest < be16_to_cpu(bf[0].length)) {
1277                 int             error;          /* error return value */
1278                 struct xfs_buf  *fbp;           /* freeblock buffer */
1279                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1280                 int             findex;         /* index in freeblock entries */
1281                 xfs_dir2_free_t *free;          /* freeblock structure */
1282
1283                 /*
1284                  * Convert the data block number to a free block,
1285                  * read in the free block.
1286                  */
1287                 fdb = xfs_dir2_db_to_fdb(mp, db);
1288                 error = xfs_dir2_free_read(tp, dp, xfs_dir2_db_to_da(mp, fdb),
1289                                            &fbp);
1290                 if (error)
1291                         return error;
1292                 free = fbp->b_addr;
1293 #ifdef DEBUG
1294         {
1295                 struct xfs_dir3_icfree_hdr freehdr;
1296                 xfs_dir3_free_hdr_from_disk(&freehdr, free);
1297                 ASSERT(freehdr.firstdb == xfs_dir3_free_max_bests(mp) *
1298                                           (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
1299         }
1300 #endif
1301                 /*
1302                  * Calculate which entry we need to fix.
1303                  */
1304                 findex = xfs_dir2_db_to_fdindex(mp, db);
1305                 longest = be16_to_cpu(bf[0].length);
1306                 /*
1307                  * If the data block is now empty we can get rid of it
1308                  * (usually).
1309                  */
1310                 if (longest == mp->m_dirblksize -
1311                                xfs_dir3_data_entry_offset(hdr)) {
1312                         /*
1313                          * Try to punch out the data block.
1314                          */
1315                         error = xfs_dir2_shrink_inode(args, db, dbp);
1316                         if (error == 0) {
1317                                 dblk->bp = NULL;
1318                                 hdr = NULL;
1319                         }
1320                         /*
1321                          * We can get ENOSPC if there's no space reservation.
1322                          * In this case just drop the buffer and some one else
1323                          * will eventually get rid of the empty block.
1324                          */
1325                         else if (!(error == ENOSPC && args->total == 0))
1326                                 return error;
1327                 }
1328                 /*
1329                  * If we got rid of the data block, we can eliminate that entry
1330                  * in the free block.
1331                  */
1332                 error = xfs_dir3_data_block_free(args, hdr, free,
1333                                                  fdb, findex, fbp, longest);
1334                 if (error)
1335                         return error;
1336         }
1337
1338         xfs_dir3_leaf_check(mp, bp);
1339         /*
1340          * Return indication of whether this leaf block is empty enough
1341          * to justify trying to join it with a neighbor.
1342          */
1343         *rval = (xfs_dir3_leaf_hdr_size(leaf) +
1344                  (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1345                 mp->m_dir_magicpct;
1346         return 0;
1347 }
1348
1349 /*
1350  * Split the leaf entries in the old block into old and new blocks.
1351  */
1352 int                                             /* error */
1353 xfs_dir2_leafn_split(
1354         xfs_da_state_t          *state,         /* btree cursor */
1355         xfs_da_state_blk_t      *oldblk,        /* original block */
1356         xfs_da_state_blk_t      *newblk)        /* newly created block */
1357 {
1358         xfs_da_args_t           *args;          /* operation arguments */
1359         xfs_dablk_t             blkno;          /* new leaf block number */
1360         int                     error;          /* error return value */
1361         xfs_mount_t             *mp;            /* filesystem mount point */
1362
1363         /*
1364          * Allocate space for a new leaf node.
1365          */
1366         args = state->args;
1367         mp = args->dp->i_mount;
1368         ASSERT(args != NULL);
1369         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1370         error = xfs_da_grow_inode(args, &blkno);
1371         if (error) {
1372                 return error;
1373         }
1374         /*
1375          * Initialize the new leaf block.
1376          */
1377         error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(mp, blkno),
1378                                       &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1379         if (error)
1380                 return error;
1381
1382         newblk->blkno = blkno;
1383         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1384         /*
1385          * Rebalance the entries across the two leaves, link the new
1386          * block into the leaves.
1387          */
1388         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1389         error = xfs_da3_blk_link(state, oldblk, newblk);
1390         if (error) {
1391                 return error;
1392         }
1393         /*
1394          * Insert the new entry in the correct block.
1395          */
1396         if (state->inleaf)
1397                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1398         else
1399                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1400         /*
1401          * Update last hashval in each block since we added the name.
1402          */
1403         oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1404         newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1405         xfs_dir3_leaf_check(mp, oldblk->bp);
1406         xfs_dir3_leaf_check(mp, newblk->bp);
1407         return error;
1408 }
1409
1410 /*
1411  * Check a leaf block and its neighbors to see if the block should be
1412  * collapsed into one or the other neighbor.  Always keep the block
1413  * with the smaller block number.
1414  * If the current block is over 50% full, don't try to join it, return 0.
1415  * If the block is empty, fill in the state structure and return 2.
1416  * If it can be collapsed, fill in the state structure and return 1.
1417  * If nothing can be done, return 0.
1418  */
1419 int                                             /* error */
1420 xfs_dir2_leafn_toosmall(
1421         xfs_da_state_t          *state,         /* btree cursor */
1422         int                     *action)        /* resulting action to take */
1423 {
1424         xfs_da_state_blk_t      *blk;           /* leaf block */
1425         xfs_dablk_t             blkno;          /* leaf block number */
1426         struct xfs_buf          *bp;            /* leaf buffer */
1427         int                     bytes;          /* bytes in use */
1428         int                     count;          /* leaf live entry count */
1429         int                     error;          /* error return value */
1430         int                     forward;        /* sibling block direction */
1431         int                     i;              /* sibling counter */
1432         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1433         int                     rval;           /* result from path_shift */
1434         struct xfs_dir3_icleaf_hdr leafhdr;
1435         struct xfs_dir2_leaf_entry *ents;
1436
1437         /*
1438          * Check for the degenerate case of the block being over 50% full.
1439          * If so, it's not worth even looking to see if we might be able
1440          * to coalesce with a sibling.
1441          */
1442         blk = &state->path.blk[state->path.active - 1];
1443         leaf = blk->bp->b_addr;
1444         xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1445         ents = xfs_dir3_leaf_ents_p(leaf);
1446         xfs_dir3_leaf_check(state->args->dp->i_mount, blk->bp);
1447
1448         count = leafhdr.count - leafhdr.stale;
1449         bytes = xfs_dir3_leaf_hdr_size(leaf) + count * sizeof(ents[0]);
1450         if (bytes > (state->blocksize >> 1)) {
1451                 /*
1452                  * Blk over 50%, don't try to join.
1453                  */
1454                 *action = 0;
1455                 return 0;
1456         }
1457         /*
1458          * Check for the degenerate case of the block being empty.
1459          * If the block is empty, we'll simply delete it, no need to
1460          * coalesce it with a sibling block.  We choose (arbitrarily)
1461          * to merge with the forward block unless it is NULL.
1462          */
1463         if (count == 0) {
1464                 /*
1465                  * Make altpath point to the block we want to keep and
1466                  * path point to the block we want to drop (this one).
1467                  */
1468                 forward = (leafhdr.forw != 0);
1469                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1470                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1471                         &rval);
1472                 if (error)
1473                         return error;
1474                 *action = rval ? 2 : 0;
1475                 return 0;
1476         }
1477         /*
1478          * Examine each sibling block to see if we can coalesce with
1479          * at least 25% free space to spare.  We need to figure out
1480          * whether to merge with the forward or the backward block.
1481          * We prefer coalescing with the lower numbered sibling so as
1482          * to shrink a directory over time.
1483          */
1484         forward = leafhdr.forw < leafhdr.back;
1485         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1486                 struct xfs_dir3_icleaf_hdr hdr2;
1487
1488                 blkno = forward ? leafhdr.forw : leafhdr.back;
1489                 if (blkno == 0)
1490                         continue;
1491                 /*
1492                  * Read the sibling leaf block.
1493                  */
1494                 error = xfs_dir3_leafn_read(state->args->trans, state->args->dp,
1495                                             blkno, -1, &bp);
1496                 if (error)
1497                         return error;
1498
1499                 /*
1500                  * Count bytes in the two blocks combined.
1501                  */
1502                 count = leafhdr.count - leafhdr.stale;
1503                 bytes = state->blocksize - (state->blocksize >> 2);
1504
1505                 leaf = bp->b_addr;
1506                 xfs_dir3_leaf_hdr_from_disk(&hdr2, leaf);
1507                 ents = xfs_dir3_leaf_ents_p(leaf);
1508                 count += hdr2.count - hdr2.stale;
1509                 bytes -= count * sizeof(ents[0]);
1510
1511                 /*
1512                  * Fits with at least 25% to spare.
1513                  */
1514                 if (bytes >= 0)
1515                         break;
1516                 xfs_trans_brelse(state->args->trans, bp);
1517         }
1518         /*
1519          * Didn't like either block, give up.
1520          */
1521         if (i >= 2) {
1522                 *action = 0;
1523                 return 0;
1524         }
1525
1526         /*
1527          * Make altpath point to the block we want to keep (the lower
1528          * numbered block) and path point to the block we want to drop.
1529          */
1530         memcpy(&state->altpath, &state->path, sizeof(state->path));
1531         if (blkno < blk->blkno)
1532                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1533                         &rval);
1534         else
1535                 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1536                         &rval);
1537         if (error) {
1538                 return error;
1539         }
1540         *action = rval ? 0 : 1;
1541         return 0;
1542 }
1543
1544 /*
1545  * Move all the leaf entries from drop_blk to save_blk.
1546  * This is done as part of a join operation.
1547  */
1548 void
1549 xfs_dir2_leafn_unbalance(
1550         xfs_da_state_t          *state,         /* cursor */
1551         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1552         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1553 {
1554         xfs_da_args_t           *args;          /* operation arguments */
1555         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1556         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1557         struct xfs_dir3_icleaf_hdr savehdr;
1558         struct xfs_dir3_icleaf_hdr drophdr;
1559         struct xfs_dir2_leaf_entry *sents;
1560         struct xfs_dir2_leaf_entry *dents;
1561
1562         args = state->args;
1563         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1564         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1565         drop_leaf = drop_blk->bp->b_addr;
1566         save_leaf = save_blk->bp->b_addr;
1567
1568         xfs_dir3_leaf_hdr_from_disk(&savehdr, save_leaf);
1569         xfs_dir3_leaf_hdr_from_disk(&drophdr, drop_leaf);
1570         sents = xfs_dir3_leaf_ents_p(save_leaf);
1571         dents = xfs_dir3_leaf_ents_p(drop_leaf);
1572
1573         /*
1574          * If there are any stale leaf entries, take this opportunity
1575          * to purge them.
1576          */
1577         if (drophdr.stale)
1578                 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1579         if (savehdr.stale)
1580                 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1581
1582         /*
1583          * Move the entries from drop to the appropriate end of save.
1584          */
1585         drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1586         if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1587                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1588                                         save_blk->bp, &savehdr, sents, 0,
1589                                         drophdr.count);
1590         else
1591                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1592                                         save_blk->bp, &savehdr, sents,
1593                                         savehdr.count, drophdr.count);
1594         save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1595
1596         /* log the changes made when moving the entries */
1597         xfs_dir3_leaf_hdr_to_disk(save_leaf, &savehdr);
1598         xfs_dir3_leaf_hdr_to_disk(drop_leaf, &drophdr);
1599         xfs_dir3_leaf_log_header(args->trans, save_blk->bp);
1600         xfs_dir3_leaf_log_header(args->trans, drop_blk->bp);
1601
1602         xfs_dir3_leaf_check(args->dp->i_mount, save_blk->bp);
1603         xfs_dir3_leaf_check(args->dp->i_mount, drop_blk->bp);
1604 }
1605
1606 /*
1607  * Top-level node form directory addname routine.
1608  */
1609 int                                             /* error */
1610 xfs_dir2_node_addname(
1611         xfs_da_args_t           *args)          /* operation arguments */
1612 {
1613         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1614         int                     error;          /* error return value */
1615         int                     rval;           /* sub-return value */
1616         xfs_da_state_t          *state;         /* btree cursor */
1617
1618         trace_xfs_dir2_node_addname(args);
1619
1620         /*
1621          * Allocate and initialize the state (btree cursor).
1622          */
1623         state = xfs_da_state_alloc();
1624         state->args = args;
1625         state->mp = args->dp->i_mount;
1626         state->blocksize = state->mp->m_dirblksize;
1627         state->node_ents = state->mp->m_dir_node_ents;
1628         /*
1629          * Look up the name.  We're not supposed to find it, but
1630          * this gives us the insertion point.
1631          */
1632         error = xfs_da3_node_lookup_int(state, &rval);
1633         if (error)
1634                 rval = error;
1635         if (rval != ENOENT) {
1636                 goto done;
1637         }
1638         /*
1639          * Add the data entry to a data block.
1640          * Extravalid is set to a freeblock found by lookup.
1641          */
1642         rval = xfs_dir2_node_addname_int(args,
1643                 state->extravalid ? &state->extrablk : NULL);
1644         if (rval) {
1645                 goto done;
1646         }
1647         blk = &state->path.blk[state->path.active - 1];
1648         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1649         /*
1650          * Add the new leaf entry.
1651          */
1652         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1653         if (rval == 0) {
1654                 /*
1655                  * It worked, fix the hash values up the btree.
1656                  */
1657                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1658                         xfs_da3_fixhashpath(state, &state->path);
1659         } else {
1660                 /*
1661                  * It didn't work, we need to split the leaf block.
1662                  */
1663                 if (args->total == 0) {
1664                         ASSERT(rval == ENOSPC);
1665                         goto done;
1666                 }
1667                 /*
1668                  * Split the leaf block and insert the new entry.
1669                  */
1670                 rval = xfs_da3_split(state);
1671         }
1672 done:
1673         xfs_da_state_free(state);
1674         return rval;
1675 }
1676
1677 /*
1678  * Add the data entry for a node-format directory name addition.
1679  * The leaf entry is added in xfs_dir2_leafn_add.
1680  * We may enter with a freespace block that the lookup found.
1681  */
1682 static int                                      /* error */
1683 xfs_dir2_node_addname_int(
1684         xfs_da_args_t           *args,          /* operation arguments */
1685         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1686 {
1687         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1688         xfs_dir2_db_t           dbno;           /* data block number */
1689         struct xfs_buf          *dbp;           /* data block buffer */
1690         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1691         xfs_inode_t             *dp;            /* incore directory inode */
1692         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1693         int                     error;          /* error return value */
1694         xfs_dir2_db_t           fbno;           /* freespace block number */
1695         struct xfs_buf          *fbp;           /* freespace buffer */
1696         int                     findex;         /* freespace entry index */
1697         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1698         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1699         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1700         int                     length;         /* length of the new entry */
1701         int                     logfree;        /* need to log free entry */
1702         xfs_mount_t             *mp;            /* filesystem mount point */
1703         int                     needlog;        /* need to log data header */
1704         int                     needscan;       /* need to rescan data frees */
1705         __be16                  *tagp;          /* data entry tag pointer */
1706         xfs_trans_t             *tp;            /* transaction pointer */
1707         __be16                  *bests;
1708         struct xfs_dir3_icfree_hdr freehdr;
1709         struct xfs_dir2_data_free *bf;
1710
1711         dp = args->dp;
1712         mp = dp->i_mount;
1713         tp = args->trans;
1714         length = xfs_dir3_data_entsize(mp, args->namelen);
1715         /*
1716          * If we came in with a freespace block that means that lookup
1717          * found an entry with our hash value.  This is the freespace
1718          * block for that data entry.
1719          */
1720         if (fblk) {
1721                 fbp = fblk->bp;
1722                 /*
1723                  * Remember initial freespace block number.
1724                  */
1725                 ifbno = fblk->blkno;
1726                 free = fbp->b_addr;
1727                 findex = fblk->index;
1728                 bests = xfs_dir3_free_bests_p(mp, free);
1729                 xfs_dir3_free_hdr_from_disk(&freehdr, free);
1730
1731                 /*
1732                  * This means the free entry showed that the data block had
1733                  * space for our entry, so we remembered it.
1734                  * Use that data block.
1735                  */
1736                 if (findex >= 0) {
1737                         ASSERT(findex < freehdr.nvalid);
1738                         ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1739                         ASSERT(be16_to_cpu(bests[findex]) >= length);
1740                         dbno = freehdr.firstdb + findex;
1741                 } else {
1742                         /*
1743                          * The data block looked at didn't have enough room.
1744                          * We'll start at the beginning of the freespace entries.
1745                          */
1746                         dbno = -1;
1747                         findex = 0;
1748                 }
1749         } else {
1750                 /*
1751                  * Didn't come in with a freespace block, so no data block.
1752                  */
1753                 ifbno = dbno = -1;
1754                 fbp = NULL;
1755                 findex = 0;
1756         }
1757
1758         /*
1759          * If we don't have a data block yet, we're going to scan the
1760          * freespace blocks looking for one.  Figure out what the
1761          * highest freespace block number is.
1762          */
1763         if (dbno == -1) {
1764                 xfs_fileoff_t   fo;             /* freespace block number */
1765
1766                 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1767                         return error;
1768                 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1769                 fbno = ifbno;
1770         }
1771         /*
1772          * While we haven't identified a data block, search the freeblock
1773          * data for a good data block.  If we find a null freeblock entry,
1774          * indicating a hole in the data blocks, remember that.
1775          */
1776         while (dbno == -1) {
1777                 /*
1778                  * If we don't have a freeblock in hand, get the next one.
1779                  */
1780                 if (fbp == NULL) {
1781                         /*
1782                          * Happens the first time through unless lookup gave
1783                          * us a freespace block to start with.
1784                          */
1785                         if (++fbno == 0)
1786                                 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1787                         /*
1788                          * If it's ifbno we already looked at it.
1789                          */
1790                         if (fbno == ifbno)
1791                                 fbno++;
1792                         /*
1793                          * If it's off the end we're done.
1794                          */
1795                         if (fbno >= lastfbno)
1796                                 break;
1797                         /*
1798                          * Read the block.  There can be holes in the
1799                          * freespace blocks, so this might not succeed.
1800                          * This should be really rare, so there's no reason
1801                          * to avoid it.
1802                          */
1803                         error = xfs_dir2_free_try_read(tp, dp,
1804                                                 xfs_dir2_db_to_da(mp, fbno),
1805                                                 &fbp);
1806                         if (error)
1807                                 return error;
1808                         if (!fbp)
1809                                 continue;
1810                         free = fbp->b_addr;
1811                         findex = 0;
1812                 }
1813                 /*
1814                  * Look at the current free entry.  Is it good enough?
1815                  *
1816                  * The bests initialisation should be where the bufer is read in
1817                  * the above branch. But gcc is too stupid to realise that bests
1818                  * and the freehdr are actually initialised if they are placed
1819                  * there, so we have to do it here to avoid warnings. Blech.
1820                  */
1821                 bests = xfs_dir3_free_bests_p(mp, free);
1822                 xfs_dir3_free_hdr_from_disk(&freehdr, free);
1823                 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1824                     be16_to_cpu(bests[findex]) >= length)
1825                         dbno = freehdr.firstdb + findex;
1826                 else {
1827                         /*
1828                          * Are we done with the freeblock?
1829                          */
1830                         if (++findex == freehdr.nvalid) {
1831                                 /*
1832                                  * Drop the block.
1833                                  */
1834                                 xfs_trans_brelse(tp, fbp);
1835                                 fbp = NULL;
1836                                 if (fblk && fblk->bp)
1837                                         fblk->bp = NULL;
1838                         }
1839                 }
1840         }
1841         /*
1842          * If we don't have a data block, we need to allocate one and make
1843          * the freespace entries refer to it.
1844          */
1845         if (unlikely(dbno == -1)) {
1846                 /*
1847                  * Not allowed to allocate, return failure.
1848                  */
1849                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1850                         return XFS_ERROR(ENOSPC);
1851
1852                 /*
1853                  * Allocate and initialize the new data block.
1854                  */
1855                 if (unlikely((error = xfs_dir2_grow_inode(args,
1856                                                          XFS_DIR2_DATA_SPACE,
1857                                                          &dbno)) ||
1858                     (error = xfs_dir3_data_init(args, dbno, &dbp))))
1859                         return error;
1860
1861                 /*
1862                  * If (somehow) we have a freespace block, get rid of it.
1863                  */
1864                 if (fbp)
1865                         xfs_trans_brelse(tp, fbp);
1866                 if (fblk && fblk->bp)
1867                         fblk->bp = NULL;
1868
1869                 /*
1870                  * Get the freespace block corresponding to the data block
1871                  * that was just allocated.
1872                  */
1873                 fbno = xfs_dir2_db_to_fdb(mp, dbno);
1874                 error = xfs_dir2_free_try_read(tp, dp,
1875                                                xfs_dir2_db_to_da(mp, fbno),
1876                                                &fbp);
1877                 if (error)
1878                         return error;
1879
1880                 /*
1881                  * If there wasn't a freespace block, the read will
1882                  * return a NULL fbp.  Allocate and initialize a new one.
1883                  */
1884                 if (!fbp) {
1885                         error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1886                                                     &fbno);
1887                         if (error)
1888                                 return error;
1889
1890                         if (unlikely(xfs_dir2_db_to_fdb(mp, dbno) != fbno)) {
1891                                 xfs_alert(mp,
1892                         "%s: dir ino %llu needed freesp block %lld for\n"
1893                         "  data block %lld, got %lld ifbno %llu lastfbno %d",
1894                                         __func__, (unsigned long long)dp->i_ino,
1895                                         (long long)xfs_dir2_db_to_fdb(mp, dbno),
1896                                         (long long)dbno, (long long)fbno,
1897                                         (unsigned long long)ifbno, lastfbno);
1898                                 if (fblk) {
1899                                         xfs_alert(mp,
1900                                 " fblk 0x%p blkno %llu index %d magic 0x%x",
1901                                                 fblk,
1902                                                 (unsigned long long)fblk->blkno,
1903                                                 fblk->index,
1904                                                 fblk->magic);
1905                                 } else {
1906                                         xfs_alert(mp, " ... fblk is NULL");
1907                                 }
1908                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1909                                                  XFS_ERRLEVEL_LOW, mp);
1910                                 return XFS_ERROR(EFSCORRUPTED);
1911                         }
1912
1913                         /*
1914                          * Get a buffer for the new block.
1915                          */
1916                         error = xfs_dir3_free_get_buf(tp, dp, fbno, &fbp);
1917                         if (error)
1918                                 return error;
1919                         free = fbp->b_addr;
1920                         bests = xfs_dir3_free_bests_p(mp, free);
1921                         xfs_dir3_free_hdr_from_disk(&freehdr, free);
1922
1923                         /*
1924                          * Remember the first slot as our empty slot.
1925                          */
1926                         freehdr.firstdb = (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1927                                         xfs_dir3_free_max_bests(mp);
1928                 } else {
1929                         free = fbp->b_addr;
1930                         bests = xfs_dir3_free_bests_p(mp, free);
1931                         xfs_dir3_free_hdr_from_disk(&freehdr, free);
1932                 }
1933
1934                 /*
1935                  * Set the freespace block index from the data block number.
1936                  */
1937                 findex = xfs_dir2_db_to_fdindex(mp, dbno);
1938                 /*
1939                  * If it's after the end of the current entries in the
1940                  * freespace block, extend that table.
1941                  */
1942                 if (findex >= freehdr.nvalid) {
1943                         ASSERT(findex < xfs_dir3_free_max_bests(mp));
1944                         freehdr.nvalid = findex + 1;
1945                         /*
1946                          * Tag new entry so nused will go up.
1947                          */
1948                         bests[findex] = cpu_to_be16(NULLDATAOFF);
1949                 }
1950                 /*
1951                  * If this entry was for an empty data block
1952                  * (this should always be true) then update the header.
1953                  */
1954                 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1955                         freehdr.nused++;
1956                         xfs_dir3_free_hdr_to_disk(fbp->b_addr, &freehdr);
1957                         xfs_dir2_free_log_header(tp, fbp);
1958                 }
1959                 /*
1960                  * Update the real value in the table.
1961                  * We haven't allocated the data entry yet so this will
1962                  * change again.
1963                  */
1964                 hdr = dbp->b_addr;
1965                 bf = xfs_dir3_data_bestfree_p(hdr);
1966                 bests[findex] = bf[0].length;
1967                 logfree = 1;
1968         }
1969         /*
1970          * We had a data block so we don't have to make a new one.
1971          */
1972         else {
1973                 /*
1974                  * If just checking, we succeeded.
1975                  */
1976                 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1977                         return 0;
1978
1979                 /*
1980                  * Read the data block in.
1981                  */
1982                 error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1983                                            -1, &dbp);
1984                 if (error)
1985                         return error;
1986                 hdr = dbp->b_addr;
1987                 bf = xfs_dir3_data_bestfree_p(hdr);
1988                 logfree = 0;
1989         }
1990         ASSERT(be16_to_cpu(bf[0].length) >= length);
1991         /*
1992          * Point to the existing unused space.
1993          */
1994         dup = (xfs_dir2_data_unused_t *)
1995               ((char *)hdr + be16_to_cpu(bf[0].offset));
1996         needscan = needlog = 0;
1997         /*
1998          * Mark the first part of the unused space, inuse for us.
1999          */
2000         xfs_dir2_data_use_free(tp, dbp, dup,
2001                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
2002                 &needlog, &needscan);
2003         /*
2004          * Fill in the new entry and log it.
2005          */
2006         dep = (xfs_dir2_data_entry_t *)dup;
2007         dep->inumber = cpu_to_be64(args->inumber);
2008         dep->namelen = args->namelen;
2009         memcpy(dep->name, args->name, dep->namelen);
2010         xfs_dir3_dirent_put_ftype(mp, dep, args->filetype);
2011         tagp = xfs_dir3_data_entry_tag_p(mp, dep);
2012         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
2013         xfs_dir2_data_log_entry(tp, dbp, dep);
2014         /*
2015          * Rescan the block for bestfree if needed.
2016          */
2017         if (needscan)
2018                 xfs_dir2_data_freescan(mp, hdr, &needlog);
2019         /*
2020          * Log the data block header if needed.
2021          */
2022         if (needlog)
2023                 xfs_dir2_data_log_header(tp, dbp);
2024         /*
2025          * If the freespace entry is now wrong, update it.
2026          */
2027         bests = xfs_dir3_free_bests_p(mp, free); /* gcc is so stupid */
2028         if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
2029                 bests[findex] = bf[0].length;
2030                 logfree = 1;
2031         }
2032         /*
2033          * Log the freespace entry if needed.
2034          */
2035         if (logfree)
2036                 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
2037         /*
2038          * Return the data block and offset in args, then drop the data block.
2039          */
2040         args->blkno = (xfs_dablk_t)dbno;
2041         args->index = be16_to_cpu(*tagp);
2042         return 0;
2043 }
2044
2045 /*
2046  * Lookup an entry in a node-format directory.
2047  * All the real work happens in xfs_da3_node_lookup_int.
2048  * The only real output is the inode number of the entry.
2049  */
2050 int                                             /* error */
2051 xfs_dir2_node_lookup(
2052         xfs_da_args_t   *args)                  /* operation arguments */
2053 {
2054         int             error;                  /* error return value */
2055         int             i;                      /* btree level */
2056         int             rval;                   /* operation return value */
2057         xfs_da_state_t  *state;                 /* btree cursor */
2058
2059         trace_xfs_dir2_node_lookup(args);
2060
2061         /*
2062          * Allocate and initialize the btree cursor.
2063          */
2064         state = xfs_da_state_alloc();
2065         state->args = args;
2066         state->mp = args->dp->i_mount;
2067         state->blocksize = state->mp->m_dirblksize;
2068         state->node_ents = state->mp->m_dir_node_ents;
2069         /*
2070          * Fill in the path to the entry in the cursor.
2071          */
2072         error = xfs_da3_node_lookup_int(state, &rval);
2073         if (error)
2074                 rval = error;
2075         else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
2076                 /* If a CI match, dup the actual name and return EEXIST */
2077                 xfs_dir2_data_entry_t   *dep;
2078
2079                 dep = (xfs_dir2_data_entry_t *)
2080                         ((char *)state->extrablk.bp->b_addr +
2081                                                  state->extrablk.index);
2082                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2083         }
2084         /*
2085          * Release the btree blocks and leaf block.
2086          */
2087         for (i = 0; i < state->path.active; i++) {
2088                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2089                 state->path.blk[i].bp = NULL;
2090         }
2091         /*
2092          * Release the data block if we have it.
2093          */
2094         if (state->extravalid && state->extrablk.bp) {
2095                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2096                 state->extrablk.bp = NULL;
2097         }
2098         xfs_da_state_free(state);
2099         return rval;
2100 }
2101
2102 /*
2103  * Remove an entry from a node-format directory.
2104  */
2105 int                                             /* error */
2106 xfs_dir2_node_removename(
2107         xfs_da_args_t           *args)          /* operation arguments */
2108 {
2109         xfs_da_state_blk_t      *blk;           /* leaf block */
2110         int                     error;          /* error return value */
2111         int                     rval;           /* operation return value */
2112         xfs_da_state_t          *state;         /* btree cursor */
2113
2114         trace_xfs_dir2_node_removename(args);
2115
2116         /*
2117          * Allocate and initialize the btree cursor.
2118          */
2119         state = xfs_da_state_alloc();
2120         state->args = args;
2121         state->mp = args->dp->i_mount;
2122         state->blocksize = state->mp->m_dirblksize;
2123         state->node_ents = state->mp->m_dir_node_ents;
2124         /*
2125          * Look up the entry we're deleting, set up the cursor.
2126          */
2127         error = xfs_da3_node_lookup_int(state, &rval);
2128         if (error)
2129                 rval = error;
2130         /*
2131          * Didn't find it, upper layer screwed up.
2132          */
2133         if (rval != EEXIST) {
2134                 xfs_da_state_free(state);
2135                 return rval;
2136         }
2137         blk = &state->path.blk[state->path.active - 1];
2138         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2139         ASSERT(state->extravalid);
2140         /*
2141          * Remove the leaf and data entries.
2142          * Extrablk refers to the data block.
2143          */
2144         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2145                 &state->extrablk, &rval);
2146         if (error)
2147                 return error;
2148         /*
2149          * Fix the hash values up the btree.
2150          */
2151         xfs_da3_fixhashpath(state, &state->path);
2152         /*
2153          * If we need to join leaf blocks, do it.
2154          */
2155         if (rval && state->path.active > 1)
2156                 error = xfs_da3_join(state);
2157         /*
2158          * If no errors so far, try conversion to leaf format.
2159          */
2160         if (!error)
2161                 error = xfs_dir2_node_to_leaf(state);
2162         xfs_da_state_free(state);
2163         return error;
2164 }
2165
2166 /*
2167  * Replace an entry's inode number in a node-format directory.
2168  */
2169 int                                             /* error */
2170 xfs_dir2_node_replace(
2171         xfs_da_args_t           *args)          /* operation arguments */
2172 {
2173         xfs_da_state_blk_t      *blk;           /* leaf block */
2174         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
2175         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
2176         int                     error;          /* error return value */
2177         int                     i;              /* btree level */
2178         xfs_ino_t               inum;           /* new inode number */
2179         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
2180         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
2181         int                     rval;           /* internal return value */
2182         xfs_da_state_t          *state;         /* btree cursor */
2183
2184         trace_xfs_dir2_node_replace(args);
2185
2186         /*
2187          * Allocate and initialize the btree cursor.
2188          */
2189         state = xfs_da_state_alloc();
2190         state->args = args;
2191         state->mp = args->dp->i_mount;
2192         state->blocksize = state->mp->m_dirblksize;
2193         state->node_ents = state->mp->m_dir_node_ents;
2194         inum = args->inumber;
2195         /*
2196          * Lookup the entry to change in the btree.
2197          */
2198         error = xfs_da3_node_lookup_int(state, &rval);
2199         if (error) {
2200                 rval = error;
2201         }
2202         /*
2203          * It should be found, since the vnodeops layer has looked it up
2204          * and locked it.  But paranoia is good.
2205          */
2206         if (rval == EEXIST) {
2207                 struct xfs_dir2_leaf_entry *ents;
2208                 /*
2209                  * Find the leaf entry.
2210                  */
2211                 blk = &state->path.blk[state->path.active - 1];
2212                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2213                 leaf = blk->bp->b_addr;
2214                 ents = xfs_dir3_leaf_ents_p(leaf);
2215                 lep = &ents[blk->index];
2216                 ASSERT(state->extravalid);
2217                 /*
2218                  * Point to the data entry.
2219                  */
2220                 hdr = state->extrablk.bp->b_addr;
2221                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2222                        hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2223                 dep = (xfs_dir2_data_entry_t *)
2224                       ((char *)hdr +
2225                        xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
2226                 ASSERT(inum != be64_to_cpu(dep->inumber));
2227                 /*
2228                  * Fill in the new inode number and log the entry.
2229                  */
2230                 dep->inumber = cpu_to_be64(inum);
2231                 xfs_dir3_dirent_put_ftype(state->mp, dep, args->filetype);
2232                 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
2233                 rval = 0;
2234         }
2235         /*
2236          * Didn't find it, and we're holding a data block.  Drop it.
2237          */
2238         else if (state->extravalid) {
2239                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2240                 state->extrablk.bp = NULL;
2241         }
2242         /*
2243          * Release all the buffers in the cursor.
2244          */
2245         for (i = 0; i < state->path.active; i++) {
2246                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2247                 state->path.blk[i].bp = NULL;
2248         }
2249         xfs_da_state_free(state);
2250         return rval;
2251 }
2252
2253 /*
2254  * Trim off a trailing empty freespace block.
2255  * Return (in rvalp) 1 if we did it, 0 if not.
2256  */
2257 int                                             /* error */
2258 xfs_dir2_node_trim_free(
2259         xfs_da_args_t           *args,          /* operation arguments */
2260         xfs_fileoff_t           fo,             /* free block number */
2261         int                     *rvalp)         /* out: did something */
2262 {
2263         struct xfs_buf          *bp;            /* freespace buffer */
2264         xfs_inode_t             *dp;            /* incore directory inode */
2265         int                     error;          /* error return code */
2266         xfs_dir2_free_t         *free;          /* freespace structure */
2267         xfs_mount_t             *mp;            /* filesystem mount point */
2268         xfs_trans_t             *tp;            /* transaction pointer */
2269         struct xfs_dir3_icfree_hdr freehdr;
2270
2271         dp = args->dp;
2272         mp = dp->i_mount;
2273         tp = args->trans;
2274         /*
2275          * Read the freespace block.
2276          */
2277         error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2278         if (error)
2279                 return error;
2280         /*
2281          * There can be holes in freespace.  If fo is a hole, there's
2282          * nothing to do.
2283          */
2284         if (!bp)
2285                 return 0;
2286         free = bp->b_addr;
2287         xfs_dir3_free_hdr_from_disk(&freehdr, free);
2288
2289         /*
2290          * If there are used entries, there's nothing to do.
2291          */
2292         if (freehdr.nused > 0) {
2293                 xfs_trans_brelse(tp, bp);
2294                 *rvalp = 0;
2295                 return 0;
2296         }
2297         /*
2298          * Blow the block away.
2299          */
2300         if ((error =
2301             xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
2302                     bp))) {
2303                 /*
2304                  * Can't fail with ENOSPC since that only happens with no
2305                  * space reservation, when breaking up an extent into two
2306                  * pieces.  This is the last block of an extent.
2307                  */
2308                 ASSERT(error != ENOSPC);
2309                 xfs_trans_brelse(tp, bp);
2310                 return error;
2311         }
2312         /*
2313          * Return that we succeeded.
2314          */
2315         *rvalp = 1;
2316         return 0;
2317 }