fs/xfs/libxfs/xfs_dir2_node.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * Copyright (c) 2013 Red Hat, Inc.
   4  * All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License as
   8  * published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it would be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write the Free Software Foundation,
  17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  18  */
  19 #include "xfs.h"
  20 #include "xfs_fs.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_mount.h"
  25 #include "xfs_da_format.h"
  26 #include "xfs_da_btree.h"
  27 #include "xfs_inode.h"
  28 #include "xfs_bmap.h"
  29 #include "xfs_dir2.h"
  30 #include "xfs_dir2_priv.h"
  31 #include "xfs_error.h"
  32 #include "xfs_trace.h"
  33 #include "xfs_trans.h"
  34 #include "xfs_buf_item.h"
  35 #include "xfs_cksum.h"
  36 #include "xfs_log.h"
  37
  38 /*
  39  * Function declarations.
  40  */
  41 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
  42                               int index);
  43 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  44                                      xfs_da_state_blk_t *blk1,
  45                                      xfs_da_state_blk_t *blk2);
  46 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
  47                                  int index, xfs_da_state_blk_t *dblk,
  48                                  int *rval);
  49 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  50                                      xfs_da_state_blk_t *fblk);
  51
  52 /*
  53  * Check internal consistency of a leafn block.
  54  */
  55 #ifdef DEBUG
  56 #define xfs_dir3_leaf_check(dp, bp) \
  57 do { \
  58         if (!xfs_dir3_leafn_check((dp), (bp))) \
  59                 ASSERT(0); \
  60 } while (0);
  61
  62 static bool
  63 xfs_dir3_leafn_check(
  64         struct xfs_inode        *dp,
  65         struct xfs_buf          *bp)
  66 {
  67         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  68         struct xfs_dir3_icleaf_hdr leafhdr;
  69
  70         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
  71
  72         if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
  73                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  74                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  75                         return false;
  76         } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
  77                 return false;
  78
  79         return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
  80 }
  81 #else
  82 #define xfs_dir3_leaf_check(dp, bp)
  83 #endif
  84
  85 static bool
  86 xfs_dir3_free_verify(
  87         struct xfs_buf          *bp)
  88 {
  89         struct xfs_mount        *mp = bp->b_target->bt_mount;
  90         struct xfs_dir2_free_hdr *hdr = bp->b_addr;
  91
  92         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  93                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
  94
  95                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_FREE_MAGIC))
  96                         return false;
  97                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
  98                         return false;
  99                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 100                         return false;
 101                 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
 102                         return false;
 103         } else {
 104                 if (hdr->magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC))
 105                         return false;
 106         }
 107
 108         /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
 109
 110         return true;
 111 }
 112
 113 static void
 114 xfs_dir3_free_read_verify(
 115         struct xfs_buf  *bp)
 116 {
 117         struct xfs_mount        *mp = bp->b_target->bt_mount;
 118
 119         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 120             !xfs_buf_verify_cksum(bp, XFS_DIR3_FREE_CRC_OFF))
 121                 xfs_buf_ioerror(bp, -EFSBADCRC);
 122         else if (!xfs_dir3_free_verify(bp))
 123                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 124
 125         if (bp->b_error)
 126                 xfs_verifier_error(bp);
 127 }
 128
 129 static void
 130 xfs_dir3_free_write_verify(
 131         struct xfs_buf  *bp)
 132 {
 133         struct xfs_mount        *mp = bp->b_target->bt_mount;
 134         struct xfs_buf_log_item *bip = bp->b_fspriv;
 135         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 136
 137         if (!xfs_dir3_free_verify(bp)) {
 138                 xfs_buf_ioerror(bp, -EFSCORRUPTED);
 139                 xfs_verifier_error(bp);
 140                 return;
 141         }
 142
 143         if (!xfs_sb_version_hascrc(&mp->m_sb))
 144                 return;
 145
 146         if (bip)
 147                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 148
 149         xfs_buf_update_cksum(bp, XFS_DIR3_FREE_CRC_OFF);
 150 }
 151
 152 const struct xfs_buf_ops xfs_dir3_free_buf_ops = {
 153         .name = "xfs_dir3_free",
 154         .verify_read = xfs_dir3_free_read_verify,
 155         .verify_write = xfs_dir3_free_write_verify,
 156 };
 157
 158 /* Everything ok in the free block header? */
 159 static bool
 160 xfs_dir3_free_header_check(
 161         struct xfs_inode        *dp,
 162         xfs_dablk_t             fbno,
 163         struct xfs_buf          *bp)
 164 {
 165         struct xfs_mount        *mp = dp->i_mount;
 166         unsigned int            firstdb;
 167         int                     maxbests;
 168
 169         maxbests = dp->d_ops->free_max_bests(mp->m_dir_geo);
 170         firstdb = (xfs_dir2_da_to_db(mp->m_dir_geo, fbno) -
 171                    xfs_dir2_byte_to_db(mp->m_dir_geo, XFS_DIR2_FREE_OFFSET)) *
 172                         maxbests;
 173         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 174                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 175
 176                 if (be32_to_cpu(hdr3->firstdb) != firstdb)
 177                         return false;
 178                 if (be32_to_cpu(hdr3->nvalid) > maxbests)
 179                         return false;
 180                 if (be32_to_cpu(hdr3->nvalid) < be32_to_cpu(hdr3->nused))
 181                         return false;
 182         } else {
 183                 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
 184
 185                 if (be32_to_cpu(hdr->firstdb) != firstdb)
 186                         return false;
 187                 if (be32_to_cpu(hdr->nvalid) > maxbests)
 188                         return false;
 189                 if (be32_to_cpu(hdr->nvalid) < be32_to_cpu(hdr->nused))
 190                         return false;
 191         }
 192         return true;
 193 }
 194
 195 static int
 196 __xfs_dir3_free_read(
 197         struct xfs_trans        *tp,
 198         struct xfs_inode        *dp,
 199         xfs_dablk_t             fbno,
 200         xfs_daddr_t             mappedbno,
 201         struct xfs_buf          **bpp)
 202 {
 203         int                     err;
 204
 205         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 206                                 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
 207         if (err || !*bpp)
 208                 return err;
 209
 210         /* Check things that we can't do in the verifier. */
 211         if (!xfs_dir3_free_header_check(dp, fbno, *bpp)) {
 212                 xfs_buf_ioerror(*bpp, -EFSCORRUPTED);
 213                 xfs_verifier_error(*bpp);
 214                 xfs_trans_brelse(tp, *bpp);
 215                 return -EFSCORRUPTED;
 216         }
 217
 218         /* try read returns without an error or *bpp if it lands in a hole */
 219         if (tp)
 220                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
 221
 222         return 0;
 223 }
 224
 225 int
 226 xfs_dir2_free_read(
 227         struct xfs_trans        *tp,
 228         struct xfs_inode        *dp,
 229         xfs_dablk_t             fbno,
 230         struct xfs_buf          **bpp)
 231 {
 232         return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
 233 }
 234
 235 static int
 236 xfs_dir2_free_try_read(
 237         struct xfs_trans        *tp,
 238         struct xfs_inode        *dp,
 239         xfs_dablk_t             fbno,
 240         struct xfs_buf          **bpp)
 241 {
 242         return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
 243 }
 244
 245 static int
 246 xfs_dir3_free_get_buf(
 247         xfs_da_args_t           *args,
 248         xfs_dir2_db_t           fbno,
 249         struct xfs_buf          **bpp)
 250 {
 251         struct xfs_trans        *tp = args->trans;
 252         struct xfs_inode        *dp = args->dp;
 253         struct xfs_mount        *mp = dp->i_mount;
 254         struct xfs_buf          *bp;
 255         int                     error;
 256         struct xfs_dir3_icfree_hdr hdr;
 257
 258         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, fbno),
 259                                    -1, &bp, XFS_DATA_FORK);
 260         if (error)
 261                 return error;
 262
 263         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
 264         bp->b_ops = &xfs_dir3_free_buf_ops;
 265
 266         /*
 267          * Initialize the new block to be empty, and remember
 268          * its first slot as our empty slot.
 269          */
 270         memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
 271         memset(&hdr, 0, sizeof(hdr));
 272
 273         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 274                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 275
 276                 hdr.magic = XFS_DIR3_FREE_MAGIC;
 277
 278                 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
 279                 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
 280                 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_meta_uuid);
 281         } else
 282                 hdr.magic = XFS_DIR2_FREE_MAGIC;
 283         dp->d_ops->free_hdr_to_disk(bp->b_addr, &hdr);
 284         *bpp = bp;
 285         return 0;
 286 }
 287
 288 /*
 289  * Log entries from a freespace block.
 290  */
 291 STATIC void
 292 xfs_dir2_free_log_bests(
 293         struct xfs_da_args      *args,
 294         struct xfs_buf          *bp,
 295         int                     first,          /* first entry to log */
 296         int                     last)           /* last entry to log */
 297 {
 298         xfs_dir2_free_t         *free;          /* freespace structure */
 299         __be16                  *bests;
 300
 301         free = bp->b_addr;
 302         bests = args->dp->d_ops->free_bests_p(free);
 303         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 304                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 305         xfs_trans_log_buf(args->trans, bp,
 306                 (uint)((char *)&bests[first] - (char *)free),
 307                 (uint)((char *)&bests[last] - (char *)free +
 308                        sizeof(bests[0]) - 1));
 309 }
 310
 311 /*
 312  * Log header from a freespace block.
 313  */
 314 static void
 315 xfs_dir2_free_log_header(
 316         struct xfs_da_args      *args,
 317         struct xfs_buf          *bp)
 318 {
 319 #ifdef DEBUG
 320         xfs_dir2_free_t         *free;          /* freespace structure */
 321
 322         free = bp->b_addr;
 323         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 324                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 325 #endif
 326         xfs_trans_log_buf(args->trans, bp, 0,
 327                           args->dp->d_ops->free_hdr_size - 1);
 328 }
 329
 330 /*
 331  * Convert a leaf-format directory to a node-format directory.
 332  * We need to change the magic number of the leaf block, and copy
 333  * the freespace table out of the leaf block into its own block.
 334  */
 335 int                                             /* error */
 336 xfs_dir2_leaf_to_node(
 337         xfs_da_args_t           *args,          /* operation arguments */
 338         struct xfs_buf          *lbp)           /* leaf buffer */
 339 {
 340         xfs_inode_t             *dp;            /* incore directory inode */
 341         int                     error;          /* error return value */
 342         struct xfs_buf          *fbp;           /* freespace buffer */
 343         xfs_dir2_db_t           fdb;            /* freespace block number */
 344         xfs_dir2_free_t         *free;          /* freespace structure */
 345         __be16                  *from;          /* pointer to freespace entry */
 346         int                     i;              /* leaf freespace index */
 347         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 348         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 349         int                     n;              /* count of live freespc ents */
 350         xfs_dir2_data_off_t     off;            /* freespace entry value */
 351         __be16                  *to;            /* pointer to freespace entry */
 352         xfs_trans_t             *tp;            /* transaction pointer */
 353         struct xfs_dir3_icfree_hdr freehdr;
 354
 355         trace_xfs_dir2_leaf_to_node(args);
 356
 357         dp = args->dp;
 358         tp = args->trans;
 359         /*
 360          * Add a freespace block to the directory.
 361          */
 362         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 363                 return error;
 364         }
 365         ASSERT(fdb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
 366         /*
 367          * Get the buffer for the new freespace block.
 368          */
 369         error = xfs_dir3_free_get_buf(args, fdb, &fbp);
 370         if (error)
 371                 return error;
 372
 373         free = fbp->b_addr;
 374         dp->d_ops->free_hdr_from_disk(&freehdr, free);
 375         leaf = lbp->b_addr;
 376         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 377         ASSERT(be32_to_cpu(ltp->bestcount) <=
 378                                 (uint)dp->i_d.di_size / args->geo->blksize);
 379
 380         /*
 381          * Copy freespace entries from the leaf block to the new block.
 382          * Count active entries.
 383          */
 384         from = xfs_dir2_leaf_bests_p(ltp);
 385         to = dp->d_ops->free_bests_p(free);
 386         for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 387                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 388                         n++;
 389                 *to = cpu_to_be16(off);
 390         }
 391
 392         /*
 393          * Now initialize the freespace block header.
 394          */
 395         freehdr.nused = n;
 396         freehdr.nvalid = be32_to_cpu(ltp->bestcount);
 397
 398         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
 399         xfs_dir2_free_log_bests(args, fbp, 0, freehdr.nvalid - 1);
 400         xfs_dir2_free_log_header(args, fbp);
 401
 402         /*
 403          * Converting the leaf to a leafnode is just a matter of changing the
 404          * magic number and the ops. Do the change directly to the buffer as
 405          * it's less work (and less code) than decoding the header to host
 406          * format and back again.
 407          */
 408         if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
 409                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 410         else
 411                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 412         lbp->b_ops = &xfs_dir3_leafn_buf_ops;
 413         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
 414         xfs_dir3_leaf_log_header(args, lbp);
 415         xfs_dir3_leaf_check(dp, lbp);
 416         return 0;
 417 }
 418
 419 /*
 420  * Add a leaf entry to a leaf block in a node-form directory.
 421  * The other work necessary is done from the caller.
 422  */
 423 static int                                      /* error */
 424 xfs_dir2_leafn_add(
 425         struct xfs_buf          *bp,            /* leaf buffer */
 426         xfs_da_args_t           *args,          /* operation arguments */
 427         int                     index)          /* insertion pt for new entry */
 428 {
 429         int                     compact;        /* compacting stale leaves */
 430         xfs_inode_t             *dp;            /* incore directory inode */
 431         int                     highstale;      /* next stale entry */
 432         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 433         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 434         int                     lfloghigh;      /* high leaf entry logging */
 435         int                     lfloglow;       /* low leaf entry logging */
 436         int                     lowstale;       /* previous stale entry */
 437         struct xfs_dir3_icleaf_hdr leafhdr;
 438         struct xfs_dir2_leaf_entry *ents;
 439
 440         trace_xfs_dir2_leafn_add(args, index);
 441
 442         dp = args->dp;
 443         leaf = bp->b_addr;
 444         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 445         ents = dp->d_ops->leaf_ents_p(leaf);
 446
 447         /*
 448          * Quick check just to make sure we are not going to index
 449          * into other peoples memory
 450          */
 451         if (index < 0)
 452                 return -EFSCORRUPTED;
 453
 454         /*
 455          * If there are already the maximum number of leaf entries in
 456          * the block, if there are no stale entries it won't fit.
 457          * Caller will do a split.  If there are stale entries we'll do
 458          * a compact.
 459          */
 460
 461         if (leafhdr.count == dp->d_ops->leaf_max_ents(args->geo)) {
 462                 if (!leafhdr.stale)
 463                         return -ENOSPC;
 464                 compact = leafhdr.stale > 1;
 465         } else
 466                 compact = 0;
 467         ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
 468         ASSERT(index == leafhdr.count ||
 469                be32_to_cpu(ents[index].hashval) >= args->hashval);
 470
 471         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 472                 return 0;
 473
 474         /*
 475          * Compact out all but one stale leaf entry.  Leaves behind
 476          * the entry closest to index.
 477          */
 478         if (compact)
 479                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 480                                          &highstale, &lfloglow, &lfloghigh);
 481         else if (leafhdr.stale) {
 482                 /*
 483                  * Set impossible logging indices for this case.
 484                  */
 485                 lfloglow = leafhdr.count;
 486                 lfloghigh = -1;
 487         }
 488
 489         /*
 490          * Insert the new entry, log everything.
 491          */
 492         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 493                                        highstale, &lfloglow, &lfloghigh);
 494
 495         lep->hashval = cpu_to_be32(args->hashval);
 496         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(args->geo,
 497                                 args->blkno, args->index));
 498
 499         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
 500         xfs_dir3_leaf_log_header(args, bp);
 501         xfs_dir3_leaf_log_ents(args, bp, lfloglow, lfloghigh);
 502         xfs_dir3_leaf_check(dp, bp);
 503         return 0;
 504 }
 505
 506 #ifdef DEBUG
 507 static void
 508 xfs_dir2_free_hdr_check(
 509         struct xfs_inode *dp,
 510         struct xfs_buf  *bp,
 511         xfs_dir2_db_t   db)
 512 {
 513         struct xfs_dir3_icfree_hdr hdr;
 514
 515         dp->d_ops->free_hdr_from_disk(&hdr, bp->b_addr);
 516
 517         ASSERT((hdr.firstdb %
 518                 dp->d_ops->free_max_bests(dp->i_mount->m_dir_geo)) == 0);
 519         ASSERT(hdr.firstdb <= db);
 520         ASSERT(db < hdr.firstdb + hdr.nvalid);
 521 }
 522 #else
 523 #define xfs_dir2_free_hdr_check(dp, bp, 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_leaf_lasthash(
 532         struct xfs_inode *dp,
 533         struct xfs_buf  *bp,                    /* leaf buffer */
 534         int             *count)                 /* count of entries in leaf */
 535 {
 536         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 537         struct xfs_dir2_leaf_entry *ents;
 538         struct xfs_dir3_icleaf_hdr leafhdr;
 539
 540         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 541
 542         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 543                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC ||
 544                leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
 545                leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
 546
 547         if (count)
 548                 *count = leafhdr.count;
 549         if (!leafhdr.count)
 550                 return 0;
 551
 552         ents = dp->d_ops->leaf_ents_p(leaf);
 553         return be32_to_cpu(ents[leafhdr.count - 1].hashval);
 554 }
 555
 556 /*
 557  * Look up a leaf entry for space to add a name in a node-format leaf block.
 558  * The extrablk in state is a freespace block.
 559  */
 560 STATIC int
 561 xfs_dir2_leafn_lookup_for_addname(
 562         struct xfs_buf          *bp,            /* leaf buffer */
 563         xfs_da_args_t           *args,          /* operation arguments */
 564         int                     *indexp,        /* out: leaf entry index */
 565         xfs_da_state_t          *state)         /* state to fill in */
 566 {
 567         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 568         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 569         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 570         xfs_inode_t             *dp;            /* incore directory inode */
 571         int                     error;          /* error return value */
 572         int                     fi;             /* free entry index */
 573         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 574         int                     index;          /* leaf entry index */
 575         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 576         int                     length;         /* length of new data entry */
 577         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 578         xfs_mount_t             *mp;            /* filesystem mount point */
 579         xfs_dir2_db_t           newdb;          /* new data block number */
 580         xfs_dir2_db_t           newfdb;         /* new free block number */
 581         xfs_trans_t             *tp;            /* transaction pointer */
 582         struct xfs_dir2_leaf_entry *ents;
 583         struct xfs_dir3_icleaf_hdr leafhdr;
 584
 585         dp = args->dp;
 586         tp = args->trans;
 587         mp = dp->i_mount;
 588         leaf = bp->b_addr;
 589         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 590         ents = dp->d_ops->leaf_ents_p(leaf);
 591
 592         xfs_dir3_leaf_check(dp, bp);
 593         ASSERT(leafhdr.count > 0);
 594
 595         /*
 596          * Look up the hash value in the leaf entries.
 597          */
 598         index = xfs_dir2_leaf_search_hash(args, bp);
 599         /*
 600          * Do we have a buffer coming in?
 601          */
 602         if (state->extravalid) {
 603                 /* If so, it's a free block buffer, get the block number. */
 604                 curbp = state->extrablk.bp;
 605                 curfdb = state->extrablk.blkno;
 606                 free = curbp->b_addr;
 607                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 608                        free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 609         }
 610         length = dp->d_ops->data_entsize(args->namelen);
 611         /*
 612          * Loop over leaf entries with the right hash value.
 613          */
 614         for (lep = &ents[index];
 615              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 616              lep++, index++) {
 617                 /*
 618                  * Skip stale leaf entries.
 619                  */
 620                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 621                         continue;
 622                 /*
 623                  * Pull the data block number from the entry.
 624                  */
 625                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 626                                                be32_to_cpu(lep->address));
 627                 /*
 628                  * For addname, we're looking for a place to put the new entry.
 629                  * We want to use a data block with an entry of equal
 630                  * hash value to ours if there is one with room.
 631                  *
 632                  * If this block isn't the data block we already have
 633                  * in hand, take a look at it.
 634                  */
 635                 if (newdb != curdb) {
 636                         __be16 *bests;
 637
 638                         curdb = newdb;
 639                         /*
 640                          * Convert the data block to the free block
 641                          * holding its freespace information.
 642                          */
 643                         newfdb = dp->d_ops->db_to_fdb(args->geo, newdb);
 644                         /*
 645                          * If it's not the one we have in hand, read it in.
 646                          */
 647                         if (newfdb != curfdb) {
 648                                 /*
 649                                  * If we had one before, drop it.
 650                                  */
 651                                 if (curbp)
 652                                         xfs_trans_brelse(tp, curbp);
 653
 654                                 error = xfs_dir2_free_read(tp, dp,
 655                                                 xfs_dir2_db_to_da(args->geo,
 656                                                                   newfdb),
 657                                                 &curbp);
 658                                 if (error)
 659                                         return error;
 660                                 free = curbp->b_addr;
 661
 662                                 xfs_dir2_free_hdr_check(dp, curbp, curdb);
 663                         }
 664                         /*
 665                          * Get the index for our entry.
 666                          */
 667                         fi = dp->d_ops->db_to_fdindex(args->geo, curdb);
 668                         /*
 669                          * If it has room, return it.
 670                          */
 671                         bests = dp->d_ops->free_bests_p(free);
 672                         if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
 673                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 674                                                         XFS_ERRLEVEL_LOW, mp);
 675                                 if (curfdb != newfdb)
 676                                         xfs_trans_brelse(tp, curbp);
 677                                 return -EFSCORRUPTED;
 678                         }
 679                         curfdb = newfdb;
 680                         if (be16_to_cpu(bests[fi]) >= length)
 681                                 goto out;
 682                 }
 683         }
 684         /* Didn't find any space */
 685         fi = -1;
 686 out:
 687         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 688         if (curbp) {
 689                 /* Giving back a free block. */
 690                 state->extravalid = 1;
 691                 state->extrablk.bp = curbp;
 692                 state->extrablk.index = fi;
 693                 state->extrablk.blkno = curfdb;
 694
 695                 /*
 696                  * Important: this magic number is not in the buffer - it's for
 697                  * buffer type information and therefore only the free/data type
 698                  * matters here, not whether CRCs are enabled or not.
 699                  */
 700                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 701         } else {
 702                 state->extravalid = 0;
 703         }
 704         /*
 705          * Return the index, that will be the insertion point.
 706          */
 707         *indexp = index;
 708         return -ENOENT;
 709 }
 710
 711 /*
 712  * Look up a leaf entry in a node-format leaf block.
 713  * The extrablk in state a data block.
 714  */
 715 STATIC int
 716 xfs_dir2_leafn_lookup_for_entry(
 717         struct xfs_buf          *bp,            /* leaf buffer */
 718         xfs_da_args_t           *args,          /* operation arguments */
 719         int                     *indexp,        /* out: leaf entry index */
 720         xfs_da_state_t          *state)         /* state to fill in */
 721 {
 722         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 723         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 724         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 725         xfs_inode_t             *dp;            /* incore directory inode */
 726         int                     error;          /* error return value */
 727         int                     index;          /* leaf entry index */
 728         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 729         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 730         xfs_mount_t             *mp;            /* filesystem mount point */
 731         xfs_dir2_db_t           newdb;          /* new data block number */
 732         xfs_trans_t             *tp;            /* transaction pointer */
 733         enum xfs_dacmp          cmp;            /* comparison result */
 734         struct xfs_dir2_leaf_entry *ents;
 735         struct xfs_dir3_icleaf_hdr leafhdr;
 736
 737         dp = args->dp;
 738         tp = args->trans;
 739         mp = dp->i_mount;
 740         leaf = bp->b_addr;
 741         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 742         ents = dp->d_ops->leaf_ents_p(leaf);
 743
 744         xfs_dir3_leaf_check(dp, bp);
 745         ASSERT(leafhdr.count > 0);
 746
 747         /*
 748          * Look up the hash value in the leaf entries.
 749          */
 750         index = xfs_dir2_leaf_search_hash(args, bp);
 751         /*
 752          * Do we have a buffer coming in?
 753          */
 754         if (state->extravalid) {
 755                 curbp = state->extrablk.bp;
 756                 curdb = state->extrablk.blkno;
 757         }
 758         /*
 759          * Loop over leaf entries with the right hash value.
 760          */
 761         for (lep = &ents[index];
 762              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 763              lep++, index++) {
 764                 /*
 765                  * Skip stale leaf entries.
 766                  */
 767                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 768                         continue;
 769                 /*
 770                  * Pull the data block number from the entry.
 771                  */
 772                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 773                                                be32_to_cpu(lep->address));
 774                 /*
 775                  * Not adding a new entry, so we really want to find
 776                  * the name given to us.
 777                  *
 778                  * If it's a different data block, go get it.
 779                  */
 780                 if (newdb != curdb) {
 781                         /*
 782                          * If we had a block before that we aren't saving
 783                          * for a CI name, drop it
 784                          */
 785                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 786                                                 curdb != state->extrablk.blkno))
 787                                 xfs_trans_brelse(tp, curbp);
 788                         /*
 789                          * If needing the block that is saved with a CI match,
 790                          * use it otherwise read in the new data block.
 791                          */
 792                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 793                                         newdb == state->extrablk.blkno) {
 794                                 ASSERT(state->extravalid);
 795                                 curbp = state->extrablk.bp;
 796                         } else {
 797                                 error = xfs_dir3_data_read(tp, dp,
 798                                                 xfs_dir2_db_to_da(args->geo,
 799                                                                   newdb),
 800                                                 -1, &curbp);
 801                                 if (error)
 802                                         return error;
 803                         }
 804                         xfs_dir3_data_check(dp, curbp);
 805                         curdb = newdb;
 806                 }
 807                 /*
 808                  * Point to the data entry.
 809                  */
 810                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 811                         xfs_dir2_dataptr_to_off(args->geo,
 812                                                 be32_to_cpu(lep->address)));
 813                 /*
 814                  * Compare the entry and if it's an exact match, return
 815                  * EEXIST immediately. If it's the first case-insensitive
 816                  * match, store the block & inode number and continue looking.
 817                  */
 818                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 819                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 820                         /* If there is a CI match block, drop it */
 821                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 822                                                 curdb != state->extrablk.blkno)
 823                                 xfs_trans_brelse(tp, state->extrablk.bp);
 824                         args->cmpresult = cmp;
 825                         args->inumber = be64_to_cpu(dep->inumber);
 826                         args->filetype = dp->d_ops->data_get_ftype(dep);
 827                         *indexp = index;
 828                         state->extravalid = 1;
 829                         state->extrablk.bp = curbp;
 830                         state->extrablk.blkno = curdb;
 831                         state->extrablk.index = (int)((char *)dep -
 832                                                         (char *)curbp->b_addr);
 833                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 834                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 835                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 836                         if (cmp == XFS_CMP_EXACT)
 837                                 return -EEXIST;
 838                 }
 839         }
 840         ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
 841         if (curbp) {
 842                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 843                         /* Giving back last used data block. */
 844                         state->extravalid = 1;
 845                         state->extrablk.bp = curbp;
 846                         state->extrablk.index = -1;
 847                         state->extrablk.blkno = curdb;
 848                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 849                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 850                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 851                 } else {
 852                         /* If the curbp is not the CI match block, drop it */
 853                         if (state->extrablk.bp != curbp)
 854                                 xfs_trans_brelse(tp, curbp);
 855                 }
 856         } else {
 857                 state->extravalid = 0;
 858         }
 859         *indexp = index;
 860         return -ENOENT;
 861 }
 862
 863 /*
 864  * Look up a leaf entry in a node-format leaf block.
 865  * If this is an addname then the extrablk in state is a freespace block,
 866  * otherwise it's a data block.
 867  */
 868 int
 869 xfs_dir2_leafn_lookup_int(
 870         struct xfs_buf          *bp,            /* leaf buffer */
 871         xfs_da_args_t           *args,          /* operation arguments */
 872         int                     *indexp,        /* out: leaf entry index */
 873         xfs_da_state_t          *state)         /* state to fill in */
 874 {
 875         if (args->op_flags & XFS_DA_OP_ADDNAME)
 876                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 877                                                         state);
 878         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 879 }
 880
 881 /*
 882  * Move count leaf entries from source to destination leaf.
 883  * Log entries and headers.  Stale entries are preserved.
 884  */
 885 static void
 886 xfs_dir3_leafn_moveents(
 887         xfs_da_args_t                   *args,  /* operation arguments */
 888         struct xfs_buf                  *bp_s,  /* source */
 889         struct xfs_dir3_icleaf_hdr      *shdr,
 890         struct xfs_dir2_leaf_entry      *sents,
 891         int                             start_s,/* source leaf index */
 892         struct xfs_buf                  *bp_d,  /* destination */
 893         struct xfs_dir3_icleaf_hdr      *dhdr,
 894         struct xfs_dir2_leaf_entry      *dents,
 895         int                             start_d,/* destination leaf index */
 896         int                             count)  /* count of leaves to copy */
 897 {
 898         int                             stale;  /* count stale leaves copied */
 899
 900         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 901
 902         /*
 903          * Silently return if nothing to do.
 904          */
 905         if (count == 0)
 906                 return;
 907
 908         /*
 909          * If the destination index is not the end of the current
 910          * destination leaf entries, open up a hole in the destination
 911          * to hold the new entries.
 912          */
 913         if (start_d < dhdr->count) {
 914                 memmove(&dents[start_d + count], &dents[start_d],
 915                         (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
 916                 xfs_dir3_leaf_log_ents(args, bp_d, start_d + count,
 917                                        count + dhdr->count - 1);
 918         }
 919         /*
 920          * If the source has stale leaves, count the ones in the copy range
 921          * so we can update the header correctly.
 922          */
 923         if (shdr->stale) {
 924                 int     i;                      /* temp leaf index */
 925
 926                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 927                         if (sents[i].address ==
 928                                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 929                                 stale++;
 930                 }
 931         } else
 932                 stale = 0;
 933         /*
 934          * Copy the leaf entries from source to destination.
 935          */
 936         memcpy(&dents[start_d], &sents[start_s],
 937                 count * sizeof(xfs_dir2_leaf_entry_t));
 938         xfs_dir3_leaf_log_ents(args, bp_d, start_d, start_d + count - 1);
 939
 940         /*
 941          * If there are source entries after the ones we copied,
 942          * delete the ones we copied by sliding the next ones down.
 943          */
 944         if (start_s + count < shdr->count) {
 945                 memmove(&sents[start_s], &sents[start_s + count],
 946                         count * sizeof(xfs_dir2_leaf_entry_t));
 947                 xfs_dir3_leaf_log_ents(args, bp_s, start_s, start_s + count - 1);
 948         }
 949
 950         /*
 951          * Update the headers and log them.
 952          */
 953         shdr->count -= count;
 954         shdr->stale -= stale;
 955         dhdr->count += count;
 956         dhdr->stale += stale;
 957 }
 958
 959 /*
 960  * Determine the sort order of two leaf blocks.
 961  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 962  */
 963 int                                             /* sort order */
 964 xfs_dir2_leafn_order(
 965         struct xfs_inode        *dp,
 966         struct xfs_buf          *leaf1_bp,              /* leaf1 buffer */
 967         struct xfs_buf          *leaf2_bp)              /* leaf2 buffer */
 968 {
 969         struct xfs_dir2_leaf    *leaf1 = leaf1_bp->b_addr;
 970         struct xfs_dir2_leaf    *leaf2 = leaf2_bp->b_addr;
 971         struct xfs_dir2_leaf_entry *ents1;
 972         struct xfs_dir2_leaf_entry *ents2;
 973         struct xfs_dir3_icleaf_hdr hdr1;
 974         struct xfs_dir3_icleaf_hdr hdr2;
 975
 976         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
 977         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
 978         ents1 = dp->d_ops->leaf_ents_p(leaf1);
 979         ents2 = dp->d_ops->leaf_ents_p(leaf2);
 980
 981         if (hdr1.count > 0 && hdr2.count > 0 &&
 982             (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
 983              be32_to_cpu(ents2[hdr2.count - 1].hashval) <
 984                                 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
 985                 return 1;
 986         return 0;
 987 }
 988
 989 /*
 990  * Rebalance leaf entries between two leaf blocks.
 991  * This is actually only called when the second block is new,
 992  * though the code deals with the general case.
 993  * A new entry will be inserted in one of the blocks, and that
 994  * entry is taken into account when balancing.
 995  */
 996 static void
 997 xfs_dir2_leafn_rebalance(
 998         xfs_da_state_t          *state,         /* btree cursor */
 999         xfs_da_state_blk_t      *blk1,          /* first btree block */
1000         xfs_da_state_blk_t      *blk2)          /* second btree block */
1001 {
1002         xfs_da_args_t           *args;          /* operation arguments */
1003         int                     count;          /* count (& direction) leaves */
1004         int                     isleft;         /* new goes in left leaf */
1005         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
1006         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
1007         int                     mid;            /* midpoint leaf index */
1008 #if defined(DEBUG) || defined(XFS_WARN)
1009         int                     oldstale;       /* old count of stale leaves */
1010 #endif
1011         int                     oldsum;         /* old total leaf count */
1012         int                     swap;           /* swapped leaf blocks */
1013         struct xfs_dir2_leaf_entry *ents1;
1014         struct xfs_dir2_leaf_entry *ents2;
1015         struct xfs_dir3_icleaf_hdr hdr1;
1016         struct xfs_dir3_icleaf_hdr hdr2;
1017         struct xfs_inode        *dp = state->args->dp;
1018
1019         args = state->args;
1020         /*
1021          * If the block order is wrong, swap the arguments.
1022          */
1023         if ((swap = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp))) {
1024                 xfs_da_state_blk_t      *tmp;   /* temp for block swap */
1025
1026                 tmp = blk1;
1027                 blk1 = blk2;
1028                 blk2 = tmp;
1029         }
1030         leaf1 = blk1->bp->b_addr;
1031         leaf2 = blk2->bp->b_addr;
1032         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
1033         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
1034         ents1 = dp->d_ops->leaf_ents_p(leaf1);
1035         ents2 = dp->d_ops->leaf_ents_p(leaf2);
1036
1037         oldsum = hdr1.count + hdr2.count;
1038 #if defined(DEBUG) || defined(XFS_WARN)
1039         oldstale = hdr1.stale + hdr2.stale;
1040 #endif
1041         mid = oldsum >> 1;
1042
1043         /*
1044          * If the old leaf count was odd then the new one will be even,
1045          * so we need to divide the new count evenly.
1046          */
1047         if (oldsum & 1) {
1048                 xfs_dahash_t    midhash;        /* middle entry hash value */
1049
1050                 if (mid >= hdr1.count)
1051                         midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1052                 else
1053                         midhash = be32_to_cpu(ents1[mid].hashval);
1054                 isleft = args->hashval <= midhash;
1055         }
1056         /*
1057          * If the old count is even then the new count is odd, so there's
1058          * no preferred side for the new entry.
1059          * Pick the left one.
1060          */
1061         else
1062                 isleft = 1;
1063         /*
1064          * Calculate moved entry count.  Positive means left-to-right,
1065          * negative means right-to-left.  Then move the entries.
1066          */
1067         count = hdr1.count - mid + (isleft == 0);
1068         if (count > 0)
1069                 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1070                                         hdr1.count - count, blk2->bp,
1071                                         &hdr2, ents2, 0, count);
1072         else if (count < 0)
1073                 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1074                                         blk1->bp, &hdr1, ents1,
1075                                         hdr1.count, count);
1076
1077         ASSERT(hdr1.count + hdr2.count == oldsum);
1078         ASSERT(hdr1.stale + hdr2.stale == oldstale);
1079
1080         /* log the changes made when moving the entries */
1081         dp->d_ops->leaf_hdr_to_disk(leaf1, &hdr1);
1082         dp->d_ops->leaf_hdr_to_disk(leaf2, &hdr2);
1083         xfs_dir3_leaf_log_header(args, blk1->bp);
1084         xfs_dir3_leaf_log_header(args, blk2->bp);
1085
1086         xfs_dir3_leaf_check(dp, blk1->bp);
1087         xfs_dir3_leaf_check(dp, blk2->bp);
1088
1089         /*
1090          * Mark whether we're inserting into the old or new leaf.
1091          */
1092         if (hdr1.count < hdr2.count)
1093                 state->inleaf = swap;
1094         else if (hdr1.count > hdr2.count)
1095                 state->inleaf = !swap;
1096         else
1097                 state->inleaf = swap ^ (blk1->index <= hdr1.count);
1098         /*
1099          * Adjust the expected index for insertion.
1100          */
1101         if (!state->inleaf)
1102                 blk2->index = blk1->index - hdr1.count;
1103
1104         /*
1105          * Finally sanity check just to make sure we are not returning a
1106          * negative index
1107          */
1108         if (blk2->index < 0) {
1109                 state->inleaf = 1;
1110                 blk2->index = 0;
1111                 xfs_alert(dp->i_mount,
1112         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1113                         __func__, blk1->index);
1114         }
1115 }
1116
1117 static int
1118 xfs_dir3_data_block_free(
1119         xfs_da_args_t           *args,
1120         struct xfs_dir2_data_hdr *hdr,
1121         struct xfs_dir2_free    *free,
1122         xfs_dir2_db_t           fdb,
1123         int                     findex,
1124         struct xfs_buf          *fbp,
1125         int                     longest)
1126 {
1127         int                     logfree = 0;
1128         __be16                  *bests;
1129         struct xfs_dir3_icfree_hdr freehdr;
1130         struct xfs_inode        *dp = args->dp;
1131
1132         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1133         bests = dp->d_ops->free_bests_p(free);
1134         if (hdr) {
1135                 /*
1136                  * Data block is not empty, just set the free entry to the new
1137                  * value.
1138                  */
1139                 bests[findex] = cpu_to_be16(longest);
1140                 xfs_dir2_free_log_bests(args, fbp, findex, findex);
1141                 return 0;
1142         }
1143
1144         /* One less used entry in the free table. */
1145         freehdr.nused--;
1146
1147         /*
1148          * If this was the last entry in the table, we can trim the table size
1149          * back.  There might be other entries at the end referring to
1150          * non-existent data blocks, get those too.
1151          */
1152         if (findex == freehdr.nvalid - 1) {
1153                 int     i;              /* free entry index */
1154
1155                 for (i = findex - 1; i >= 0; i--) {
1156                         if (bests[i] != cpu_to_be16(NULLDATAOFF))
1157                                 break;
1158                 }
1159                 freehdr.nvalid = i + 1;
1160                 logfree = 0;
1161         } else {
1162                 /* Not the last entry, just punch it out.  */
1163                 bests[findex] = cpu_to_be16(NULLDATAOFF);
1164                 logfree = 1;
1165         }
1166
1167         dp->d_ops->free_hdr_to_disk(free, &freehdr);
1168         xfs_dir2_free_log_header(args, fbp);
1169
1170         /*
1171          * If there are no useful entries left in the block, get rid of the
1172          * block if we can.
1173          */
1174         if (!freehdr.nused) {
1175                 int error;
1176
1177                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1178                 if (error == 0) {
1179                         fbp = NULL;
1180                         logfree = 0;
1181                 } else if (error != -ENOSPC || args->total != 0)
1182                         return error;
1183                 /*
1184                  * It's possible to get ENOSPC if there is no
1185                  * space reservation.  In this case some one
1186                  * else will eventually get rid of this block.
1187                  */
1188         }
1189
1190         /* Log the free entry that changed, unless we got rid of it.  */
1191         if (logfree)
1192                 xfs_dir2_free_log_bests(args, fbp, findex, findex);
1193         return 0;
1194 }
1195
1196 /*
1197  * Remove an entry from a node directory.
1198  * This removes the leaf entry and the data entry,
1199  * and updates the free block if necessary.
1200  */
1201 static int                                      /* error */
1202 xfs_dir2_leafn_remove(
1203         xfs_da_args_t           *args,          /* operation arguments */
1204         struct xfs_buf          *bp,            /* leaf buffer */
1205         int                     index,          /* leaf entry index */
1206         xfs_da_state_blk_t      *dblk,          /* data block */
1207         int                     *rval)          /* resulting block needs join */
1208 {
1209         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1210         xfs_dir2_db_t           db;             /* data block number */
1211         struct xfs_buf          *dbp;           /* data block buffer */
1212         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1213         xfs_inode_t             *dp;            /* incore directory inode */
1214         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1215         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1216         int                     longest;        /* longest data free entry */
1217         int                     off;            /* data block entry offset */
1218         int                     needlog;        /* need to log data header */
1219         int                     needscan;       /* need to rescan data frees */
1220         xfs_trans_t             *tp;            /* transaction pointer */
1221         struct xfs_dir2_data_free *bf;          /* bestfree table */
1222         struct xfs_dir3_icleaf_hdr leafhdr;
1223         struct xfs_dir2_leaf_entry *ents;
1224
1225         trace_xfs_dir2_leafn_remove(args, index);
1226
1227         dp = args->dp;
1228         tp = args->trans;
1229         leaf = bp->b_addr;
1230         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1231         ents = dp->d_ops->leaf_ents_p(leaf);
1232
1233         /*
1234          * Point to the entry we're removing.
1235          */
1236         lep = &ents[index];
1237
1238         /*
1239          * Extract the data block and offset from the entry.
1240          */
1241         db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1242         ASSERT(dblk->blkno == db);
1243         off = xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address));
1244         ASSERT(dblk->index == off);
1245
1246         /*
1247          * Kill the leaf entry by marking it stale.
1248          * Log the leaf block changes.
1249          */
1250         leafhdr.stale++;
1251         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1252         xfs_dir3_leaf_log_header(args, bp);
1253
1254         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1255         xfs_dir3_leaf_log_ents(args, bp, index, index);
1256
1257         /*
1258          * Make the data entry free.  Keep track of the longest freespace
1259          * in the data block in case it changes.
1260          */
1261         dbp = dblk->bp;
1262         hdr = dbp->b_addr;
1263         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1264         bf = dp->d_ops->data_bestfree_p(hdr);
1265         longest = be16_to_cpu(bf[0].length);
1266         needlog = needscan = 0;
1267         xfs_dir2_data_make_free(args, dbp, off,
1268                 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1269         /*
1270          * Rescan the data block freespaces for bestfree.
1271          * Log the data block header if needed.
1272          */
1273         if (needscan)
1274                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1275         if (needlog)
1276                 xfs_dir2_data_log_header(args, dbp);
1277         xfs_dir3_data_check(dp, dbp);
1278         /*
1279          * If the longest data block freespace changes, need to update
1280          * the corresponding freeblock entry.
1281          */
1282         if (longest < be16_to_cpu(bf[0].length)) {
1283                 int             error;          /* error return value */
1284                 struct xfs_buf  *fbp;           /* freeblock buffer */
1285                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1286                 int             findex;         /* index in freeblock entries */
1287                 xfs_dir2_free_t *free;          /* freeblock structure */
1288
1289                 /*
1290                  * Convert the data block number to a free block,
1291                  * read in the free block.
1292                  */
1293                 fdb = dp->d_ops->db_to_fdb(args->geo, db);
1294                 error = xfs_dir2_free_read(tp, dp,
1295                                            xfs_dir2_db_to_da(args->geo, fdb),
1296                                            &fbp);
1297                 if (error)
1298                         return error;
1299                 free = fbp->b_addr;
1300 #ifdef DEBUG
1301         {
1302                 struct xfs_dir3_icfree_hdr freehdr;
1303                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1304                 ASSERT(freehdr.firstdb == dp->d_ops->free_max_bests(args->geo) *
1305                         (fdb - xfs_dir2_byte_to_db(args->geo,
1306                                                    XFS_DIR2_FREE_OFFSET)));
1307         }
1308 #endif
1309                 /*
1310                  * Calculate which entry we need to fix.
1311                  */
1312                 findex = dp->d_ops->db_to_fdindex(args->geo, db);
1313                 longest = be16_to_cpu(bf[0].length);
1314                 /*
1315                  * If the data block is now empty we can get rid of it
1316                  * (usually).
1317                  */
1318                 if (longest == args->geo->blksize -
1319                                dp->d_ops->data_entry_offset) {
1320                         /*
1321                          * Try to punch out the data block.
1322                          */
1323                         error = xfs_dir2_shrink_inode(args, db, dbp);
1324                         if (error == 0) {
1325                                 dblk->bp = NULL;
1326                                 hdr = NULL;
1327                         }
1328                         /*
1329                          * We can get ENOSPC if there's no space reservation.
1330                          * In this case just drop the buffer and some one else
1331                          * will eventually get rid of the empty block.
1332                          */
1333                         else if (!(error == -ENOSPC && args->total == 0))
1334                                 return error;
1335                 }
1336                 /*
1337                  * If we got rid of the data block, we can eliminate that entry
1338                  * in the free block.
1339                  */
1340                 error = xfs_dir3_data_block_free(args, hdr, free,
1341                                                  fdb, findex, fbp, longest);
1342                 if (error)
1343                         return error;
1344         }
1345
1346         xfs_dir3_leaf_check(dp, bp);
1347         /*
1348          * Return indication of whether this leaf block is empty enough
1349          * to justify trying to join it with a neighbor.
1350          */
1351         *rval = (dp->d_ops->leaf_hdr_size +
1352                  (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1353                 args->geo->magicpct;
1354         return 0;
1355 }
1356
1357 /*
1358  * Split the leaf entries in the old block into old and new blocks.
1359  */
1360 int                                             /* error */
1361 xfs_dir2_leafn_split(
1362         xfs_da_state_t          *state,         /* btree cursor */
1363         xfs_da_state_blk_t      *oldblk,        /* original block */
1364         xfs_da_state_blk_t      *newblk)        /* newly created block */
1365 {
1366         xfs_da_args_t           *args;          /* operation arguments */
1367         xfs_dablk_t             blkno;          /* new leaf block number */
1368         int                     error;          /* error return value */
1369         struct xfs_inode        *dp;
1370
1371         /*
1372          * Allocate space for a new leaf node.
1373          */
1374         args = state->args;
1375         dp = args->dp;
1376         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1377         error = xfs_da_grow_inode(args, &blkno);
1378         if (error) {
1379                 return error;
1380         }
1381         /*
1382          * Initialize the new leaf block.
1383          */
1384         error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(args->geo, blkno),
1385                                       &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1386         if (error)
1387                 return error;
1388
1389         newblk->blkno = blkno;
1390         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1391         /*
1392          * Rebalance the entries across the two leaves, link the new
1393          * block into the leaves.
1394          */
1395         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1396         error = xfs_da3_blk_link(state, oldblk, newblk);
1397         if (error) {
1398                 return error;
1399         }
1400         /*
1401          * Insert the new entry in the correct block.
1402          */
1403         if (state->inleaf)
1404                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1405         else
1406                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1407         /*
1408          * Update last hashval in each block since we added the name.
1409          */
1410         oldblk->hashval = xfs_dir2_leaf_lasthash(dp, oldblk->bp, NULL);
1411         newblk->hashval = xfs_dir2_leaf_lasthash(dp, newblk->bp, NULL);
1412         xfs_dir3_leaf_check(dp, oldblk->bp);
1413         xfs_dir3_leaf_check(dp, newblk->bp);
1414         return error;
1415 }
1416
1417 /*
1418  * Check a leaf block and its neighbors to see if the block should be
1419  * collapsed into one or the other neighbor.  Always keep the block
1420  * with the smaller block number.
1421  * If the current block is over 50% full, don't try to join it, return 0.
1422  * If the block is empty, fill in the state structure and return 2.
1423  * If it can be collapsed, fill in the state structure and return 1.
1424  * If nothing can be done, return 0.
1425  */
1426 int                                             /* error */
1427 xfs_dir2_leafn_toosmall(
1428         xfs_da_state_t          *state,         /* btree cursor */
1429         int                     *action)        /* resulting action to take */
1430 {
1431         xfs_da_state_blk_t      *blk;           /* leaf block */
1432         xfs_dablk_t             blkno;          /* leaf block number */
1433         struct xfs_buf          *bp;            /* leaf buffer */
1434         int                     bytes;          /* bytes in use */
1435         int                     count;          /* leaf live entry count */
1436         int                     error;          /* error return value */
1437         int                     forward;        /* sibling block direction */
1438         int                     i;              /* sibling counter */
1439         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1440         int                     rval;           /* result from path_shift */
1441         struct xfs_dir3_icleaf_hdr leafhdr;
1442         struct xfs_dir2_leaf_entry *ents;
1443         struct xfs_inode        *dp = state->args->dp;
1444
1445         /*
1446          * Check for the degenerate case of the block being over 50% full.
1447          * If so, it's not worth even looking to see if we might be able
1448          * to coalesce with a sibling.
1449          */
1450         blk = &state->path.blk[state->path.active - 1];
1451         leaf = blk->bp->b_addr;
1452         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1453         ents = dp->d_ops->leaf_ents_p(leaf);
1454         xfs_dir3_leaf_check(dp, blk->bp);
1455
1456         count = leafhdr.count - leafhdr.stale;
1457         bytes = dp->d_ops->leaf_hdr_size + count * sizeof(ents[0]);
1458         if (bytes > (state->args->geo->blksize >> 1)) {
1459                 /*
1460                  * Blk over 50%, don't try to join.
1461                  */
1462                 *action = 0;
1463                 return 0;
1464         }
1465         /*
1466          * Check for the degenerate case of the block being empty.
1467          * If the block is empty, we'll simply delete it, no need to
1468          * coalesce it with a sibling block.  We choose (arbitrarily)
1469          * to merge with the forward block unless it is NULL.
1470          */
1471         if (count == 0) {
1472                 /*
1473                  * Make altpath point to the block we want to keep and
1474                  * path point to the block we want to drop (this one).
1475                  */
1476                 forward = (leafhdr.forw != 0);
1477                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1478                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1479                         &rval);
1480                 if (error)
1481                         return error;
1482                 *action = rval ? 2 : 0;
1483                 return 0;
1484         }
1485         /*
1486          * Examine each sibling block to see if we can coalesce with
1487          * at least 25% free space to spare.  We need to figure out
1488          * whether to merge with the forward or the backward block.
1489          * We prefer coalescing with the lower numbered sibling so as
1490          * to shrink a directory over time.
1491          */
1492         forward = leafhdr.forw < leafhdr.back;
1493         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1494                 struct xfs_dir3_icleaf_hdr hdr2;
1495
1496                 blkno = forward ? leafhdr.forw : leafhdr.back;
1497                 if (blkno == 0)
1498                         continue;
1499                 /*
1500                  * Read the sibling leaf block.
1501                  */
1502                 error = xfs_dir3_leafn_read(state->args->trans, dp,
1503                                             blkno, -1, &bp);
1504                 if (error)
1505                         return error;
1506
1507                 /*
1508                  * Count bytes in the two blocks combined.
1509                  */
1510                 count = leafhdr.count - leafhdr.stale;
1511                 bytes = state->args->geo->blksize -
1512                         (state->args->geo->blksize >> 2);
1513
1514                 leaf = bp->b_addr;
1515                 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf);
1516                 ents = dp->d_ops->leaf_ents_p(leaf);
1517                 count += hdr2.count - hdr2.stale;
1518                 bytes -= count * sizeof(ents[0]);
1519
1520                 /*
1521                  * Fits with at least 25% to spare.
1522                  */
1523                 if (bytes >= 0)
1524                         break;
1525                 xfs_trans_brelse(state->args->trans, bp);
1526         }
1527         /*
1528          * Didn't like either block, give up.
1529          */
1530         if (i >= 2) {
1531                 *action = 0;
1532                 return 0;
1533         }
1534
1535         /*
1536          * Make altpath point to the block we want to keep (the lower
1537          * numbered block) and path point to the block we want to drop.
1538          */
1539         memcpy(&state->altpath, &state->path, sizeof(state->path));
1540         if (blkno < blk->blkno)
1541                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1542                         &rval);
1543         else
1544                 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1545                         &rval);
1546         if (error) {
1547                 return error;
1548         }
1549         *action = rval ? 0 : 1;
1550         return 0;
1551 }
1552
1553 /*
1554  * Move all the leaf entries from drop_blk to save_blk.
1555  * This is done as part of a join operation.
1556  */
1557 void
1558 xfs_dir2_leafn_unbalance(
1559         xfs_da_state_t          *state,         /* cursor */
1560         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1561         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1562 {
1563         xfs_da_args_t           *args;          /* operation arguments */
1564         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1565         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1566         struct xfs_dir3_icleaf_hdr savehdr;
1567         struct xfs_dir3_icleaf_hdr drophdr;
1568         struct xfs_dir2_leaf_entry *sents;
1569         struct xfs_dir2_leaf_entry *dents;
1570         struct xfs_inode        *dp = state->args->dp;
1571
1572         args = state->args;
1573         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1574         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1575         drop_leaf = drop_blk->bp->b_addr;
1576         save_leaf = save_blk->bp->b_addr;
1577
1578         dp->d_ops->leaf_hdr_from_disk(&savehdr, save_leaf);
1579         dp->d_ops->leaf_hdr_from_disk(&drophdr, drop_leaf);
1580         sents = dp->d_ops->leaf_ents_p(save_leaf);
1581         dents = dp->d_ops->leaf_ents_p(drop_leaf);
1582
1583         /*
1584          * If there are any stale leaf entries, take this opportunity
1585          * to purge them.
1586          */
1587         if (drophdr.stale)
1588                 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1589         if (savehdr.stale)
1590                 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1591
1592         /*
1593          * Move the entries from drop to the appropriate end of save.
1594          */
1595         drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1596         if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1597                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1598                                         save_blk->bp, &savehdr, sents, 0,
1599                                         drophdr.count);
1600         else
1601                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1602                                         save_blk->bp, &savehdr, sents,
1603                                         savehdr.count, drophdr.count);
1604         save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1605
1606         /* log the changes made when moving the entries */
1607         dp->d_ops->leaf_hdr_to_disk(save_leaf, &savehdr);
1608         dp->d_ops->leaf_hdr_to_disk(drop_leaf, &drophdr);
1609         xfs_dir3_leaf_log_header(args, save_blk->bp);
1610         xfs_dir3_leaf_log_header(args, drop_blk->bp);
1611
1612         xfs_dir3_leaf_check(dp, save_blk->bp);
1613         xfs_dir3_leaf_check(dp, drop_blk->bp);
1614 }
1615
1616 /*
1617  * Top-level node form directory addname routine.
1618  */
1619 int                                             /* error */
1620 xfs_dir2_node_addname(
1621         xfs_da_args_t           *args)          /* operation arguments */
1622 {
1623         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1624         int                     error;          /* error return value */
1625         int                     rval;           /* sub-return value */
1626         xfs_da_state_t          *state;         /* btree cursor */
1627
1628         trace_xfs_dir2_node_addname(args);
1629
1630         /*
1631          * Allocate and initialize the state (btree cursor).
1632          */
1633         state = xfs_da_state_alloc();
1634         state->args = args;
1635         state->mp = args->dp->i_mount;
1636         /*
1637          * Look up the name.  We're not supposed to find it, but
1638          * this gives us the insertion point.
1639          */
1640         error = xfs_da3_node_lookup_int(state, &rval);
1641         if (error)
1642                 rval = error;
1643         if (rval != -ENOENT) {
1644                 goto done;
1645         }
1646         /*
1647          * Add the data entry to a data block.
1648          * Extravalid is set to a freeblock found by lookup.
1649          */
1650         rval = xfs_dir2_node_addname_int(args,
1651                 state->extravalid ? &state->extrablk : NULL);
1652         if (rval) {
1653                 goto done;
1654         }
1655         blk = &state->path.blk[state->path.active - 1];
1656         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1657         /*
1658          * Add the new leaf entry.
1659          */
1660         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1661         if (rval == 0) {
1662                 /*
1663                  * It worked, fix the hash values up the btree.
1664                  */
1665                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1666                         xfs_da3_fixhashpath(state, &state->path);
1667         } else {
1668                 /*
1669                  * It didn't work, we need to split the leaf block.
1670                  */
1671                 if (args->total == 0) {
1672                         ASSERT(rval == -ENOSPC);
1673                         goto done;
1674                 }
1675                 /*
1676                  * Split the leaf block and insert the new entry.
1677                  */
1678                 rval = xfs_da3_split(state);
1679         }
1680 done:
1681         xfs_da_state_free(state);
1682         return rval;
1683 }
1684
1685 /*
1686  * Add the data entry for a node-format directory name addition.
1687  * The leaf entry is added in xfs_dir2_leafn_add.
1688  * We may enter with a freespace block that the lookup found.
1689  */
1690 static int                                      /* error */
1691 xfs_dir2_node_addname_int(
1692         xfs_da_args_t           *args,          /* operation arguments */
1693         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1694 {
1695         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1696         xfs_dir2_db_t           dbno;           /* data block number */
1697         struct xfs_buf          *dbp;           /* data block buffer */
1698         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1699         xfs_inode_t             *dp;            /* incore directory inode */
1700         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1701         int                     error;          /* error return value */
1702         xfs_dir2_db_t           fbno;           /* freespace block number */
1703         struct xfs_buf          *fbp;           /* freespace buffer */
1704         int                     findex;         /* freespace entry index */
1705         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1706         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1707         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1708         int                     length;         /* length of the new entry */
1709         int                     logfree;        /* need to log free entry */
1710         xfs_mount_t             *mp;            /* filesystem mount point */
1711         int                     needlog;        /* need to log data header */
1712         int                     needscan;       /* need to rescan data frees */
1713         __be16                  *tagp;          /* data entry tag pointer */
1714         xfs_trans_t             *tp;            /* transaction pointer */
1715         __be16                  *bests;
1716         struct xfs_dir3_icfree_hdr freehdr;
1717         struct xfs_dir2_data_free *bf;
1718
1719         dp = args->dp;
1720         mp = dp->i_mount;
1721         tp = args->trans;
1722         length = dp->d_ops->data_entsize(args->namelen);
1723         /*
1724          * If we came in with a freespace block that means that lookup
1725          * found an entry with our hash value.  This is the freespace
1726          * block for that data entry.
1727          */
1728         if (fblk) {
1729                 fbp = fblk->bp;
1730                 /*
1731                  * Remember initial freespace block number.
1732                  */
1733                 ifbno = fblk->blkno;
1734                 free = fbp->b_addr;
1735                 findex = fblk->index;
1736                 bests = dp->d_ops->free_bests_p(free);
1737                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1738
1739                 /*
1740                  * This means the free entry showed that the data block had
1741                  * space for our entry, so we remembered it.
1742                  * Use that data block.
1743                  */
1744                 if (findex >= 0) {
1745                         ASSERT(findex < freehdr.nvalid);
1746                         ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1747                         ASSERT(be16_to_cpu(bests[findex]) >= length);
1748                         dbno = freehdr.firstdb + findex;
1749                 } else {
1750                         /*
1751                          * The data block looked at didn't have enough room.
1752                          * We'll start at the beginning of the freespace entries.
1753                          */
1754                         dbno = -1;
1755                         findex = 0;
1756                 }
1757         } else {
1758                 /*
1759                  * Didn't come in with a freespace block, so no data block.
1760                  */
1761                 ifbno = dbno = -1;
1762                 fbp = NULL;
1763                 findex = 0;
1764         }
1765
1766         /*
1767          * If we don't have a data block yet, we're going to scan the
1768          * freespace blocks looking for one.  Figure out what the
1769          * highest freespace block number is.
1770          */
1771         if (dbno == -1) {
1772                 xfs_fileoff_t   fo;             /* freespace block number */
1773
1774                 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK)))
1775                         return error;
1776                 lastfbno = xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo);
1777                 fbno = ifbno;
1778         }
1779         /*
1780          * While we haven't identified a data block, search the freeblock
1781          * data for a good data block.  If we find a null freeblock entry,
1782          * indicating a hole in the data blocks, remember that.
1783          */
1784         while (dbno == -1) {
1785                 /*
1786                  * If we don't have a freeblock in hand, get the next one.
1787                  */
1788                 if (fbp == NULL) {
1789                         /*
1790                          * Happens the first time through unless lookup gave
1791                          * us a freespace block to start with.
1792                          */
1793                         if (++fbno == 0)
1794                                 fbno = xfs_dir2_byte_to_db(args->geo,
1795                                                         XFS_DIR2_FREE_OFFSET);
1796                         /*
1797                          * If it's ifbno we already looked at it.
1798                          */
1799                         if (fbno == ifbno)
1800                                 fbno++;
1801                         /*
1802                          * If it's off the end we're done.
1803                          */
1804                         if (fbno >= lastfbno)
1805                                 break;
1806                         /*
1807                          * Read the block.  There can be holes in the
1808                          * freespace blocks, so this might not succeed.
1809                          * This should be really rare, so there's no reason
1810                          * to avoid it.
1811                          */
1812                         error = xfs_dir2_free_try_read(tp, dp,
1813                                         xfs_dir2_db_to_da(args->geo, fbno),
1814                                         &fbp);
1815                         if (error)
1816                                 return error;
1817                         if (!fbp)
1818                                 continue;
1819                         free = fbp->b_addr;
1820                         findex = 0;
1821                 }
1822                 /*
1823                  * Look at the current free entry.  Is it good enough?
1824                  *
1825                  * The bests initialisation should be where the bufer is read in
1826                  * the above branch. But gcc is too stupid to realise that bests
1827                  * and the freehdr are actually initialised if they are placed
1828                  * there, so we have to do it here to avoid warnings. Blech.
1829                  */
1830                 bests = dp->d_ops->free_bests_p(free);
1831                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1832                 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1833                     be16_to_cpu(bests[findex]) >= length)
1834                         dbno = freehdr.firstdb + findex;
1835                 else {
1836                         /*
1837                          * Are we done with the freeblock?
1838                          */
1839                         if (++findex == freehdr.nvalid) {
1840                                 /*
1841                                  * Drop the block.
1842                                  */
1843                                 xfs_trans_brelse(tp, fbp);
1844                                 fbp = NULL;
1845                                 if (fblk && fblk->bp)
1846                                         fblk->bp = NULL;
1847                         }
1848                 }
1849         }
1850         /*
1851          * If we don't have a data block, we need to allocate one and make
1852          * the freespace entries refer to it.
1853          */
1854         if (unlikely(dbno == -1)) {
1855                 /*
1856                  * Not allowed to allocate, return failure.
1857                  */
1858                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1859                         return -ENOSPC;
1860
1861                 /*
1862                  * Allocate and initialize the new data block.
1863                  */
1864                 if (unlikely((error = xfs_dir2_grow_inode(args,
1865                                                          XFS_DIR2_DATA_SPACE,
1866                                                          &dbno)) ||
1867                     (error = xfs_dir3_data_init(args, dbno, &dbp))))
1868                         return error;
1869
1870                 /*
1871                  * If (somehow) we have a freespace block, get rid of it.
1872                  */
1873                 if (fbp)
1874                         xfs_trans_brelse(tp, fbp);
1875                 if (fblk && fblk->bp)
1876                         fblk->bp = NULL;
1877
1878                 /*
1879                  * Get the freespace block corresponding to the data block
1880                  * that was just allocated.
1881                  */
1882                 fbno = dp->d_ops->db_to_fdb(args->geo, dbno);
1883                 error = xfs_dir2_free_try_read(tp, dp,
1884                                        xfs_dir2_db_to_da(args->geo, fbno),
1885                                        &fbp);
1886                 if (error)
1887                         return error;
1888
1889                 /*
1890                  * If there wasn't a freespace block, the read will
1891                  * return a NULL fbp.  Allocate and initialize a new one.
1892                  */
1893                 if (!fbp) {
1894                         error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1895                                                     &fbno);
1896                         if (error)
1897                                 return error;
1898
1899                         if (dp->d_ops->db_to_fdb(args->geo, dbno) != fbno) {
1900                                 xfs_alert(mp,
1901 "%s: dir ino %llu needed freesp block %lld for data block %lld, got %lld ifbno %llu lastfbno %d",
1902                                         __func__, (unsigned long long)dp->i_ino,
1903                                         (long long)dp->d_ops->db_to_fdb(
1904                                                                 args->geo, dbno),
1905                                         (long long)dbno, (long long)fbno,
1906                                         (unsigned long long)ifbno, lastfbno);
1907                                 if (fblk) {
1908                                         xfs_alert(mp,
1909                                 " fblk 0x%p blkno %llu index %d magic 0x%x",
1910                                                 fblk,
1911                                                 (unsigned long long)fblk->blkno,
1912                                                 fblk->index,
1913                                                 fblk->magic);
1914                                 } else {
1915                                         xfs_alert(mp, " ... fblk is NULL");
1916                                 }
1917                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1918                                                  XFS_ERRLEVEL_LOW, mp);
1919                                 return -EFSCORRUPTED;
1920                         }
1921
1922                         /*
1923                          * Get a buffer for the new block.
1924                          */
1925                         error = xfs_dir3_free_get_buf(args, fbno, &fbp);
1926                         if (error)
1927                                 return error;
1928                         free = fbp->b_addr;
1929                         bests = dp->d_ops->free_bests_p(free);
1930                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1931
1932                         /*
1933                          * Remember the first slot as our empty slot.
1934                          */
1935                         freehdr.firstdb =
1936                                 (fbno - xfs_dir2_byte_to_db(args->geo,
1937                                                         XFS_DIR2_FREE_OFFSET)) *
1938                                         dp->d_ops->free_max_bests(args->geo);
1939                 } else {
1940                         free = fbp->b_addr;
1941                         bests = dp->d_ops->free_bests_p(free);
1942                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1943                 }
1944
1945                 /*
1946                  * Set the freespace block index from the data block number.
1947                  */
1948                 findex = dp->d_ops->db_to_fdindex(args->geo, dbno);
1949                 /*
1950                  * If it's after the end of the current entries in the
1951                  * freespace block, extend that table.
1952                  */
1953                 if (findex >= freehdr.nvalid) {
1954                         ASSERT(findex < dp->d_ops->free_max_bests(args->geo));
1955                         freehdr.nvalid = findex + 1;
1956                         /*
1957                          * Tag new entry so nused will go up.
1958                          */
1959                         bests[findex] = cpu_to_be16(NULLDATAOFF);
1960                 }
1961                 /*
1962                  * If this entry was for an empty data block
1963                  * (this should always be true) then update the header.
1964                  */
1965                 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1966                         freehdr.nused++;
1967                         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
1968                         xfs_dir2_free_log_header(args, fbp);
1969                 }
1970                 /*
1971                  * Update the real value in the table.
1972                  * We haven't allocated the data entry yet so this will
1973                  * change again.
1974                  */
1975                 hdr = dbp->b_addr;
1976                 bf = dp->d_ops->data_bestfree_p(hdr);
1977                 bests[findex] = bf[0].length;
1978                 logfree = 1;
1979         }
1980         /*
1981          * We had a data block so we don't have to make a new one.
1982          */
1983         else {
1984                 /*
1985                  * If just checking, we succeeded.
1986                  */
1987                 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1988                         return 0;
1989
1990                 /*
1991                  * Read the data block in.
1992                  */
1993                 error = xfs_dir3_data_read(tp, dp,
1994                                            xfs_dir2_db_to_da(args->geo, dbno),
1995                                            -1, &dbp);
1996                 if (error)
1997                         return error;
1998                 hdr = dbp->b_addr;
1999                 bf = dp->d_ops->data_bestfree_p(hdr);
2000                 logfree = 0;
2001         }
2002         ASSERT(be16_to_cpu(bf[0].length) >= length);
2003         /*
2004          * Point to the existing unused space.
2005          */
2006         dup = (xfs_dir2_data_unused_t *)
2007               ((char *)hdr + be16_to_cpu(bf[0].offset));
2008         needscan = needlog = 0;
2009         /*
2010          * Mark the first part of the unused space, inuse for us.
2011          */
2012         xfs_dir2_data_use_free(args, dbp, dup,
2013                 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
2014                 &needlog, &needscan);
2015         /*
2016          * Fill in the new entry and log it.
2017          */
2018         dep = (xfs_dir2_data_entry_t *)dup;
2019         dep->inumber = cpu_to_be64(args->inumber);
2020         dep->namelen = args->namelen;
2021         memcpy(dep->name, args->name, dep->namelen);
2022         dp->d_ops->data_put_ftype(dep, args->filetype);
2023         tagp = dp->d_ops->data_entry_tag_p(dep);
2024         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
2025         xfs_dir2_data_log_entry(args, dbp, dep);
2026         /*
2027          * Rescan the block for bestfree if needed.
2028          */
2029         if (needscan)
2030                 xfs_dir2_data_freescan(dp, hdr, &needlog);
2031         /*
2032          * Log the data block header if needed.
2033          */
2034         if (needlog)
2035                 xfs_dir2_data_log_header(args, dbp);
2036         /*
2037          * If the freespace entry is now wrong, update it.
2038          */
2039         bests = dp->d_ops->free_bests_p(free); /* gcc is so stupid */
2040         if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
2041                 bests[findex] = bf[0].length;
2042                 logfree = 1;
2043         }
2044         /*
2045          * Log the freespace entry if needed.
2046          */
2047         if (logfree)
2048                 xfs_dir2_free_log_bests(args, fbp, findex, findex);
2049         /*
2050          * Return the data block and offset in args, then drop the data block.
2051          */
2052         args->blkno = (xfs_dablk_t)dbno;
2053         args->index = be16_to_cpu(*tagp);
2054         return 0;
2055 }
2056
2057 /*
2058  * Lookup an entry in a node-format directory.
2059  * All the real work happens in xfs_da3_node_lookup_int.
2060  * The only real output is the inode number of the entry.
2061  */
2062 int                                             /* error */
2063 xfs_dir2_node_lookup(
2064         xfs_da_args_t   *args)                  /* operation arguments */
2065 {
2066         int             error;                  /* error return value */
2067         int             i;                      /* btree level */
2068         int             rval;                   /* operation return value */
2069         xfs_da_state_t  *state;                 /* btree cursor */
2070
2071         trace_xfs_dir2_node_lookup(args);
2072
2073         /*
2074          * Allocate and initialize the btree cursor.
2075          */
2076         state = xfs_da_state_alloc();
2077         state->args = args;
2078         state->mp = args->dp->i_mount;
2079         /*
2080          * Fill in the path to the entry in the cursor.
2081          */
2082         error = xfs_da3_node_lookup_int(state, &rval);
2083         if (error)
2084                 rval = error;
2085         else if (rval == -ENOENT && args->cmpresult == XFS_CMP_CASE) {
2086                 /* If a CI match, dup the actual name and return -EEXIST */
2087                 xfs_dir2_data_entry_t   *dep;
2088
2089                 dep = (xfs_dir2_data_entry_t *)
2090                         ((char *)state->extrablk.bp->b_addr +
2091                                                  state->extrablk.index);
2092                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2093         }
2094         /*
2095          * Release the btree blocks and leaf block.
2096          */
2097         for (i = 0; i < state->path.active; i++) {
2098                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2099                 state->path.blk[i].bp = NULL;
2100         }
2101         /*
2102          * Release the data block if we have it.
2103          */
2104         if (state->extravalid && state->extrablk.bp) {
2105                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2106                 state->extrablk.bp = NULL;
2107         }
2108         xfs_da_state_free(state);
2109         return rval;
2110 }
2111
2112 /*
2113  * Remove an entry from a node-format directory.
2114  */
2115 int                                             /* error */
2116 xfs_dir2_node_removename(
2117         struct xfs_da_args      *args)          /* operation arguments */
2118 {
2119         struct xfs_da_state_blk *blk;           /* leaf block */
2120         int                     error;          /* error return value */
2121         int                     rval;           /* operation return value */
2122         struct xfs_da_state     *state;         /* btree cursor */
2123
2124         trace_xfs_dir2_node_removename(args);
2125
2126         /*
2127          * Allocate and initialize the btree cursor.
2128          */
2129         state = xfs_da_state_alloc();
2130         state->args = args;
2131         state->mp = args->dp->i_mount;
2132
2133         /* Look up the entry we're deleting, set up the cursor. */
2134         error = xfs_da3_node_lookup_int(state, &rval);
2135         if (error)
2136                 goto out_free;
2137
2138         /* Didn't find it, upper layer screwed up. */
2139         if (rval != -EEXIST) {
2140                 error = rval;
2141                 goto out_free;
2142         }
2143
2144         blk = &state->path.blk[state->path.active - 1];
2145         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2146         ASSERT(state->extravalid);
2147         /*
2148          * Remove the leaf and data entries.
2149          * Extrablk refers to the data block.
2150          */
2151         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2152                 &state->extrablk, &rval);
2153         if (error)
2154                 goto out_free;
2155         /*
2156          * Fix the hash values up the btree.
2157          */
2158         xfs_da3_fixhashpath(state, &state->path);
2159         /*
2160          * If we need to join leaf blocks, do it.
2161          */
2162         if (rval && state->path.active > 1)
2163                 error = xfs_da3_join(state);
2164         /*
2165          * If no errors so far, try conversion to leaf format.
2166          */
2167         if (!error)
2168                 error = xfs_dir2_node_to_leaf(state);
2169 out_free:
2170         xfs_da_state_free(state);
2171         return error;
2172 }
2173
2174 /*
2175  * Replace an entry's inode number in a node-format directory.
2176  */
2177 int                                             /* error */
2178 xfs_dir2_node_replace(
2179         xfs_da_args_t           *args)          /* operation arguments */
2180 {
2181         xfs_da_state_blk_t      *blk;           /* leaf block */
2182         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
2183         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
2184         int                     error;          /* error return value */
2185         int                     i;              /* btree level */
2186         xfs_ino_t               inum;           /* new inode number */
2187         int                     ftype;          /* new file type */
2188         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
2189         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
2190         int                     rval;           /* internal return value */
2191         xfs_da_state_t          *state;         /* btree cursor */
2192
2193         trace_xfs_dir2_node_replace(args);
2194
2195         /*
2196          * Allocate and initialize the btree cursor.
2197          */
2198         state = xfs_da_state_alloc();
2199         state->args = args;
2200         state->mp = args->dp->i_mount;
2201
2202         /*
2203          * We have to save new inode number and ftype since
2204          * xfs_da3_node_lookup_int() is going to overwrite them
2205          */
2206         inum = args->inumber;
2207         ftype = args->filetype;
2208
2209         /*
2210          * Lookup the entry to change in the btree.
2211          */
2212         error = xfs_da3_node_lookup_int(state, &rval);
2213         if (error) {
2214                 rval = error;
2215         }
2216         /*
2217          * It should be found, since the vnodeops layer has looked it up
2218          * and locked it.  But paranoia is good.
2219          */
2220         if (rval == -EEXIST) {
2221                 struct xfs_dir2_leaf_entry *ents;
2222                 /*
2223                  * Find the leaf entry.
2224                  */
2225                 blk = &state->path.blk[state->path.active - 1];
2226                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2227                 leaf = blk->bp->b_addr;
2228                 ents = args->dp->d_ops->leaf_ents_p(leaf);
2229                 lep = &ents[blk->index];
2230                 ASSERT(state->extravalid);
2231                 /*
2232                  * Point to the data entry.
2233                  */
2234                 hdr = state->extrablk.bp->b_addr;
2235                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2236                        hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2237                 dep = (xfs_dir2_data_entry_t *)
2238                       ((char *)hdr +
2239                        xfs_dir2_dataptr_to_off(args->geo,
2240                                                be32_to_cpu(lep->address)));
2241                 ASSERT(inum != be64_to_cpu(dep->inumber));
2242                 /*
2243                  * Fill in the new inode number and log the entry.
2244                  */
2245                 dep->inumber = cpu_to_be64(inum);
2246                 args->dp->d_ops->data_put_ftype(dep, ftype);
2247                 xfs_dir2_data_log_entry(args, state->extrablk.bp, dep);
2248                 rval = 0;
2249         }
2250         /*
2251          * Didn't find it, and we're holding a data block.  Drop it.
2252          */
2253         else if (state->extravalid) {
2254                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2255                 state->extrablk.bp = NULL;
2256         }
2257         /*
2258          * Release all the buffers in the cursor.
2259          */
2260         for (i = 0; i < state->path.active; i++) {
2261                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2262                 state->path.blk[i].bp = NULL;
2263         }
2264         xfs_da_state_free(state);
2265         return rval;
2266 }
2267
2268 /*
2269  * Trim off a trailing empty freespace block.
2270  * Return (in rvalp) 1 if we did it, 0 if not.
2271  */
2272 int                                             /* error */
2273 xfs_dir2_node_trim_free(
2274         xfs_da_args_t           *args,          /* operation arguments */
2275         xfs_fileoff_t           fo,             /* free block number */
2276         int                     *rvalp)         /* out: did something */
2277 {
2278         struct xfs_buf          *bp;            /* freespace buffer */
2279         xfs_inode_t             *dp;            /* incore directory inode */
2280         int                     error;          /* error return code */
2281         xfs_dir2_free_t         *free;          /* freespace structure */
2282         xfs_trans_t             *tp;            /* transaction pointer */
2283         struct xfs_dir3_icfree_hdr freehdr;
2284
2285         dp = args->dp;
2286         tp = args->trans;
2287
2288         *rvalp = 0;
2289
2290         /*
2291          * Read the freespace block.
2292          */
2293         error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2294         if (error)
2295                 return error;
2296         /*
2297          * There can be holes in freespace.  If fo is a hole, there's
2298          * nothing to do.
2299          */
2300         if (!bp)
2301                 return 0;
2302         free = bp->b_addr;
2303         dp->d_ops->free_hdr_from_disk(&freehdr, free);
2304
2305         /*
2306          * If there are used entries, there's nothing to do.
2307          */
2308         if (freehdr.nused > 0) {
2309                 xfs_trans_brelse(tp, bp);
2310                 return 0;
2311         }
2312         /*
2313          * Blow the block away.
2314          */
2315         error = xfs_dir2_shrink_inode(args,
2316                         xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo), bp);
2317         if (error) {
2318                 /*
2319                  * Can't fail with ENOSPC since that only happens with no
2320                  * space reservation, when breaking up an extent into two
2321                  * pieces.  This is the last block of an extent.
2322                  */
2323                 ASSERT(error != -ENOSPC);
2324                 xfs_trans_brelse(tp, bp);
2325                 return error;
2326         }
2327         /*
2328          * Return that we succeeded.
2329          */
2330         *rvalp = 1;
2331         return 0;
2332 }