fs/xfs/libxfs/xfs_dir2_node.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   4  * Copyright (c) 2013 Red Hat, Inc.
   5  * All Rights Reserved.
   6  */
   7 #include "xfs.h"
   8 #include "xfs_fs.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_mount.h"
  13 #include "xfs_da_format.h"
  14 #include "xfs_da_btree.h"
  15 #include "xfs_inode.h"
  16 #include "xfs_bmap.h"
  17 #include "xfs_dir2.h"
  18 #include "xfs_dir2_priv.h"
  19 #include "xfs_error.h"
  20 #include "xfs_trace.h"
  21 #include "xfs_trans.h"
  22 #include "xfs_buf_item.h"
  23 #include "xfs_cksum.h"
  24 #include "xfs_log.h"
  25
  26 /*
  27  * Function declarations.
  28  */
  29 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
  30                               int index);
  31 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  32                                      xfs_da_state_blk_t *blk1,
  33                                      xfs_da_state_blk_t *blk2);
  34 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
  35                                  int index, xfs_da_state_blk_t *dblk,
  36                                  int *rval);
  37 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  38                                      xfs_da_state_blk_t *fblk);
  39
  40 /*
  41  * Check internal consistency of a leafn block.
  42  */
  43 #ifdef DEBUG
  44 static xfs_failaddr_t
  45 xfs_dir3_leafn_check(
  46         struct xfs_inode        *dp,
  47         struct xfs_buf          *bp)
  48 {
  49         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  50         struct xfs_dir3_icleaf_hdr leafhdr;
  51
  52         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
  53
  54         if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
  55                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  56                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  57                         return __this_address;
  58         } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
  59                 return __this_address;
  60
  61         return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
  62 }
  63
  64 static inline void
  65 xfs_dir3_leaf_check(
  66         struct xfs_inode        *dp,
  67         struct xfs_buf          *bp)
  68 {
  69         xfs_failaddr_t          fa;
  70
  71         fa = xfs_dir3_leafn_check(dp, bp);
  72         if (!fa)
  73                 return;
  74         xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
  75                         bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
  76                         fa);
  77         ASSERT(0);
  78 }
  79 #else
  80 #define xfs_dir3_leaf_check(dp, bp)
  81 #endif
  82
  83 static xfs_failaddr_t
  84 xfs_dir3_free_verify(
  85         struct xfs_buf          *bp)
  86 {
  87         struct xfs_mount        *mp = bp->b_target->bt_mount;
  88         struct xfs_dir2_free_hdr *hdr = bp->b_addr;
  89
  90         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  91                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
  92
  93                 if (hdr3->magic != cpu_to_be32(XFS_DIR3_FREE_MAGIC))
  94                         return __this_address;
  95                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
  96                         return __this_address;
  97                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
  98                         return __this_address;
  99                 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
 100                         return __this_address;
 101         } else {
 102                 if (hdr->magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC))
 103                         return __this_address;
 104         }
 105
 106         /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
 107
 108         return NULL;
 109 }
 110
 111 static void
 112 xfs_dir3_free_read_verify(
 113         struct xfs_buf  *bp)
 114 {
 115         struct xfs_mount        *mp = bp->b_target->bt_mount;
 116         xfs_failaddr_t          fa;
 117
 118         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 119             !xfs_buf_verify_cksum(bp, XFS_DIR3_FREE_CRC_OFF))
 120                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
 121         else {
 122                 fa = xfs_dir3_free_verify(bp);
 123                 if (fa)
 124                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 125         }
 126 }
 127
 128 static void
 129 xfs_dir3_free_write_verify(
 130         struct xfs_buf  *bp)
 131 {
 132         struct xfs_mount        *mp = bp->b_target->bt_mount;
 133         struct xfs_buf_log_item *bip = bp->b_log_item;
 134         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 135         xfs_failaddr_t          fa;
 136
 137         fa = xfs_dir3_free_verify(bp);
 138         if (fa) {
 139                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 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         .verify_struct = xfs_dir3_free_verify,
 157 };
 158
 159 /* Everything ok in the free block header? */
 160 static xfs_failaddr_t
 161 xfs_dir3_free_header_check(
 162         struct xfs_inode        *dp,
 163         xfs_dablk_t             fbno,
 164         struct xfs_buf          *bp)
 165 {
 166         struct xfs_mount        *mp = dp->i_mount;
 167         unsigned int            firstdb;
 168         int                     maxbests;
 169
 170         maxbests = dp->d_ops->free_max_bests(mp->m_dir_geo);
 171         firstdb = (xfs_dir2_da_to_db(mp->m_dir_geo, fbno) -
 172                    xfs_dir2_byte_to_db(mp->m_dir_geo, XFS_DIR2_FREE_OFFSET)) *
 173                         maxbests;
 174         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 175                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 176
 177                 if (be32_to_cpu(hdr3->firstdb) != firstdb)
 178                         return __this_address;
 179                 if (be32_to_cpu(hdr3->nvalid) > maxbests)
 180                         return __this_address;
 181                 if (be32_to_cpu(hdr3->nvalid) < be32_to_cpu(hdr3->nused))
 182                         return __this_address;
 183         } else {
 184                 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
 185
 186                 if (be32_to_cpu(hdr->firstdb) != firstdb)
 187                         return __this_address;
 188                 if (be32_to_cpu(hdr->nvalid) > maxbests)
 189                         return __this_address;
 190                 if (be32_to_cpu(hdr->nvalid) < be32_to_cpu(hdr->nused))
 191                         return __this_address;
 192         }
 193         return NULL;
 194 }
 195
 196 static int
 197 __xfs_dir3_free_read(
 198         struct xfs_trans        *tp,
 199         struct xfs_inode        *dp,
 200         xfs_dablk_t             fbno,
 201         xfs_daddr_t             mappedbno,
 202         struct xfs_buf          **bpp)
 203 {
 204         xfs_failaddr_t          fa;
 205         int                     err;
 206
 207         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 208                                 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
 209         if (err || !*bpp)
 210                 return err;
 211
 212         /* Check things that we can't do in the verifier. */
 213         fa = xfs_dir3_free_header_check(dp, fbno, *bpp);
 214         if (fa) {
 215                 xfs_verifier_error(*bpp, -EFSCORRUPTED, fa);
 216                 xfs_trans_brelse(tp, *bpp);
 217                 *bpp = NULL;
 218                 return -EFSCORRUPTED;
 219         }
 220
 221         /* try read returns without an error or *bpp if it lands in a hole */
 222         if (tp)
 223                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
 224
 225         return 0;
 226 }
 227
 228 int
 229 xfs_dir2_free_read(
 230         struct xfs_trans        *tp,
 231         struct xfs_inode        *dp,
 232         xfs_dablk_t             fbno,
 233         struct xfs_buf          **bpp)
 234 {
 235         return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
 236 }
 237
 238 static int
 239 xfs_dir2_free_try_read(
 240         struct xfs_trans        *tp,
 241         struct xfs_inode        *dp,
 242         xfs_dablk_t             fbno,
 243         struct xfs_buf          **bpp)
 244 {
 245         return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
 246 }
 247
 248 static int
 249 xfs_dir3_free_get_buf(
 250         xfs_da_args_t           *args,
 251         xfs_dir2_db_t           fbno,
 252         struct xfs_buf          **bpp)
 253 {
 254         struct xfs_trans        *tp = args->trans;
 255         struct xfs_inode        *dp = args->dp;
 256         struct xfs_mount        *mp = dp->i_mount;
 257         struct xfs_buf          *bp;
 258         int                     error;
 259         struct xfs_dir3_icfree_hdr hdr;
 260
 261         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, fbno),
 262                                    -1, &bp, XFS_DATA_FORK);
 263         if (error)
 264                 return error;
 265
 266         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
 267         bp->b_ops = &xfs_dir3_free_buf_ops;
 268
 269         /*
 270          * Initialize the new block to be empty, and remember
 271          * its first slot as our empty slot.
 272          */
 273         memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
 274         memset(&hdr, 0, sizeof(hdr));
 275
 276         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 277                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 278
 279                 hdr.magic = XFS_DIR3_FREE_MAGIC;
 280
 281                 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
 282                 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
 283                 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_meta_uuid);
 284         } else
 285                 hdr.magic = XFS_DIR2_FREE_MAGIC;
 286         dp->d_ops->free_hdr_to_disk(bp->b_addr, &hdr);
 287         *bpp = bp;
 288         return 0;
 289 }
 290
 291 /*
 292  * Log entries from a freespace block.
 293  */
 294 STATIC void
 295 xfs_dir2_free_log_bests(
 296         struct xfs_da_args      *args,
 297         struct xfs_buf          *bp,
 298         int                     first,          /* first entry to log */
 299         int                     last)           /* last entry to log */
 300 {
 301         xfs_dir2_free_t         *free;          /* freespace structure */
 302         __be16                  *bests;
 303
 304         free = bp->b_addr;
 305         bests = args->dp->d_ops->free_bests_p(free);
 306         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 307                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 308         xfs_trans_log_buf(args->trans, bp,
 309                 (uint)((char *)&bests[first] - (char *)free),
 310                 (uint)((char *)&bests[last] - (char *)free +
 311                        sizeof(bests[0]) - 1));
 312 }
 313
 314 /*
 315  * Log header from a freespace block.
 316  */
 317 static void
 318 xfs_dir2_free_log_header(
 319         struct xfs_da_args      *args,
 320         struct xfs_buf          *bp)
 321 {
 322 #ifdef DEBUG
 323         xfs_dir2_free_t         *free;          /* freespace structure */
 324
 325         free = bp->b_addr;
 326         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 327                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 328 #endif
 329         xfs_trans_log_buf(args->trans, bp, 0,
 330                           args->dp->d_ops->free_hdr_size - 1);
 331 }
 332
 333 /*
 334  * Convert a leaf-format directory to a node-format directory.
 335  * We need to change the magic number of the leaf block, and copy
 336  * the freespace table out of the leaf block into its own block.
 337  */
 338 int                                             /* error */
 339 xfs_dir2_leaf_to_node(
 340         xfs_da_args_t           *args,          /* operation arguments */
 341         struct xfs_buf          *lbp)           /* leaf buffer */
 342 {
 343         xfs_inode_t             *dp;            /* incore directory inode */
 344         int                     error;          /* error return value */
 345         struct xfs_buf          *fbp;           /* freespace buffer */
 346         xfs_dir2_db_t           fdb;            /* freespace block number */
 347         xfs_dir2_free_t         *free;          /* freespace structure */
 348         __be16                  *from;          /* pointer to freespace entry */
 349         int                     i;              /* leaf freespace index */
 350         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 351         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 352         int                     n;              /* count of live freespc ents */
 353         xfs_dir2_data_off_t     off;            /* freespace entry value */
 354         __be16                  *to;            /* pointer to freespace entry */
 355         xfs_trans_t             *tp;            /* transaction pointer */
 356         struct xfs_dir3_icfree_hdr freehdr;
 357
 358         trace_xfs_dir2_leaf_to_node(args);
 359
 360         dp = args->dp;
 361         tp = args->trans;
 362         /*
 363          * Add a freespace block to the directory.
 364          */
 365         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 366                 return error;
 367         }
 368         ASSERT(fdb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
 369         /*
 370          * Get the buffer for the new freespace block.
 371          */
 372         error = xfs_dir3_free_get_buf(args, fdb, &fbp);
 373         if (error)
 374                 return error;
 375
 376         free = fbp->b_addr;
 377         dp->d_ops->free_hdr_from_disk(&freehdr, free);
 378         leaf = lbp->b_addr;
 379         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 380         if (be32_to_cpu(ltp->bestcount) >
 381                                 (uint)dp->i_d.di_size / args->geo->blksize)
 382                 return -EFSCORRUPTED;
 383
 384         /*
 385          * Copy freespace entries from the leaf block to the new block.
 386          * Count active entries.
 387          */
 388         from = xfs_dir2_leaf_bests_p(ltp);
 389         to = dp->d_ops->free_bests_p(free);
 390         for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 391                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 392                         n++;
 393                 *to = cpu_to_be16(off);
 394         }
 395
 396         /*
 397          * Now initialize the freespace block header.
 398          */
 399         freehdr.nused = n;
 400         freehdr.nvalid = be32_to_cpu(ltp->bestcount);
 401
 402         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
 403         xfs_dir2_free_log_bests(args, fbp, 0, freehdr.nvalid - 1);
 404         xfs_dir2_free_log_header(args, fbp);
 405
 406         /*
 407          * Converting the leaf to a leafnode is just a matter of changing the
 408          * magic number and the ops. Do the change directly to the buffer as
 409          * it's less work (and less code) than decoding the header to host
 410          * format and back again.
 411          */
 412         if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
 413                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 414         else
 415                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 416         lbp->b_ops = &xfs_dir3_leafn_buf_ops;
 417         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
 418         xfs_dir3_leaf_log_header(args, lbp);
 419         xfs_dir3_leaf_check(dp, lbp);
 420         return 0;
 421 }
 422
 423 /*
 424  * Add a leaf entry to a leaf block in a node-form directory.
 425  * The other work necessary is done from the caller.
 426  */
 427 static int                                      /* error */
 428 xfs_dir2_leafn_add(
 429         struct xfs_buf          *bp,            /* leaf buffer */
 430         xfs_da_args_t           *args,          /* operation arguments */
 431         int                     index)          /* insertion pt for new entry */
 432 {
 433         int                     compact;        /* compacting stale leaves */
 434         xfs_inode_t             *dp;            /* incore directory inode */
 435         int                     highstale;      /* next stale entry */
 436         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 437         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 438         int                     lfloghigh;      /* high leaf entry logging */
 439         int                     lfloglow;       /* low leaf entry logging */
 440         int                     lowstale;       /* previous stale entry */
 441         struct xfs_dir3_icleaf_hdr leafhdr;
 442         struct xfs_dir2_leaf_entry *ents;
 443
 444         trace_xfs_dir2_leafn_add(args, index);
 445
 446         dp = args->dp;
 447         leaf = bp->b_addr;
 448         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 449         ents = dp->d_ops->leaf_ents_p(leaf);
 450
 451         /*
 452          * Quick check just to make sure we are not going to index
 453          * into other peoples memory
 454          */
 455         if (index < 0)
 456                 return -EFSCORRUPTED;
 457
 458         /*
 459          * If there are already the maximum number of leaf entries in
 460          * the block, if there are no stale entries it won't fit.
 461          * Caller will do a split.  If there are stale entries we'll do
 462          * a compact.
 463          */
 464
 465         if (leafhdr.count == dp->d_ops->leaf_max_ents(args->geo)) {
 466                 if (!leafhdr.stale)
 467                         return -ENOSPC;
 468                 compact = leafhdr.stale > 1;
 469         } else
 470                 compact = 0;
 471         ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
 472         ASSERT(index == leafhdr.count ||
 473                be32_to_cpu(ents[index].hashval) >= args->hashval);
 474
 475         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 476                 return 0;
 477
 478         /*
 479          * Compact out all but one stale leaf entry.  Leaves behind
 480          * the entry closest to index.
 481          */
 482         if (compact)
 483                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 484                                          &highstale, &lfloglow, &lfloghigh);
 485         else if (leafhdr.stale) {
 486                 /*
 487                  * Set impossible logging indices for this case.
 488                  */
 489                 lfloglow = leafhdr.count;
 490                 lfloghigh = -1;
 491         }
 492
 493         /*
 494          * Insert the new entry, log everything.
 495          */
 496         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 497                                        highstale, &lfloglow, &lfloghigh);
 498
 499         lep->hashval = cpu_to_be32(args->hashval);
 500         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(args->geo,
 501                                 args->blkno, args->index));
 502
 503         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
 504         xfs_dir3_leaf_log_header(args, bp);
 505         xfs_dir3_leaf_log_ents(args, bp, lfloglow, lfloghigh);
 506         xfs_dir3_leaf_check(dp, bp);
 507         return 0;
 508 }
 509
 510 #ifdef DEBUG
 511 static void
 512 xfs_dir2_free_hdr_check(
 513         struct xfs_inode *dp,
 514         struct xfs_buf  *bp,
 515         xfs_dir2_db_t   db)
 516 {
 517         struct xfs_dir3_icfree_hdr hdr;
 518
 519         dp->d_ops->free_hdr_from_disk(&hdr, bp->b_addr);
 520
 521         ASSERT((hdr.firstdb %
 522                 dp->d_ops->free_max_bests(dp->i_mount->m_dir_geo)) == 0);
 523         ASSERT(hdr.firstdb <= db);
 524         ASSERT(db < hdr.firstdb + hdr.nvalid);
 525 }
 526 #else
 527 #define xfs_dir2_free_hdr_check(dp, bp, db)
 528 #endif  /* DEBUG */
 529
 530 /*
 531  * Return the last hash value in the leaf.
 532  * Stale entries are ok.
 533  */
 534 xfs_dahash_t                                    /* hash value */
 535 xfs_dir2_leaf_lasthash(
 536         struct xfs_inode *dp,
 537         struct xfs_buf  *bp,                    /* leaf buffer */
 538         int             *count)                 /* count of entries in leaf */
 539 {
 540         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 541         struct xfs_dir2_leaf_entry *ents;
 542         struct xfs_dir3_icleaf_hdr leafhdr;
 543
 544         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 545
 546         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 547                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC ||
 548                leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
 549                leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
 550
 551         if (count)
 552                 *count = leafhdr.count;
 553         if (!leafhdr.count)
 554                 return 0;
 555
 556         ents = dp->d_ops->leaf_ents_p(leaf);
 557         return be32_to_cpu(ents[leafhdr.count - 1].hashval);
 558 }
 559
 560 /*
 561  * Look up a leaf entry for space to add a name in a node-format leaf block.
 562  * The extrablk in state is a freespace block.
 563  */
 564 STATIC int
 565 xfs_dir2_leafn_lookup_for_addname(
 566         struct xfs_buf          *bp,            /* leaf buffer */
 567         xfs_da_args_t           *args,          /* operation arguments */
 568         int                     *indexp,        /* out: leaf entry index */
 569         xfs_da_state_t          *state)         /* state to fill in */
 570 {
 571         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 572         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 573         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 574         xfs_inode_t             *dp;            /* incore directory inode */
 575         int                     error;          /* error return value */
 576         int                     fi;             /* free entry index */
 577         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 578         int                     index;          /* leaf entry index */
 579         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 580         int                     length;         /* length of new data entry */
 581         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 582         xfs_mount_t             *mp;            /* filesystem mount point */
 583         xfs_dir2_db_t           newdb;          /* new data block number */
 584         xfs_dir2_db_t           newfdb;         /* new free block number */
 585         xfs_trans_t             *tp;            /* transaction pointer */
 586         struct xfs_dir2_leaf_entry *ents;
 587         struct xfs_dir3_icleaf_hdr leafhdr;
 588
 589         dp = args->dp;
 590         tp = args->trans;
 591         mp = dp->i_mount;
 592         leaf = bp->b_addr;
 593         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 594         ents = dp->d_ops->leaf_ents_p(leaf);
 595
 596         xfs_dir3_leaf_check(dp, bp);
 597         ASSERT(leafhdr.count > 0);
 598
 599         /*
 600          * Look up the hash value in the leaf entries.
 601          */
 602         index = xfs_dir2_leaf_search_hash(args, bp);
 603         /*
 604          * Do we have a buffer coming in?
 605          */
 606         if (state->extravalid) {
 607                 /* If so, it's a free block buffer, get the block number. */
 608                 curbp = state->extrablk.bp;
 609                 curfdb = state->extrablk.blkno;
 610                 free = curbp->b_addr;
 611                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 612                        free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 613         }
 614         length = dp->d_ops->data_entsize(args->namelen);
 615         /*
 616          * Loop over leaf entries with the right hash value.
 617          */
 618         for (lep = &ents[index];
 619              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 620              lep++, index++) {
 621                 /*
 622                  * Skip stale leaf entries.
 623                  */
 624                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 625                         continue;
 626                 /*
 627                  * Pull the data block number from the entry.
 628                  */
 629                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 630                                                be32_to_cpu(lep->address));
 631                 /*
 632                  * For addname, we're looking for a place to put the new entry.
 633                  * We want to use a data block with an entry of equal
 634                  * hash value to ours if there is one with room.
 635                  *
 636                  * If this block isn't the data block we already have
 637                  * in hand, take a look at it.
 638                  */
 639                 if (newdb != curdb) {
 640                         __be16 *bests;
 641
 642                         curdb = newdb;
 643                         /*
 644                          * Convert the data block to the free block
 645                          * holding its freespace information.
 646                          */
 647                         newfdb = dp->d_ops->db_to_fdb(args->geo, newdb);
 648                         /*
 649                          * If it's not the one we have in hand, read it in.
 650                          */
 651                         if (newfdb != curfdb) {
 652                                 /*
 653                                  * If we had one before, drop it.
 654                                  */
 655                                 if (curbp)
 656                                         xfs_trans_brelse(tp, curbp);
 657
 658                                 error = xfs_dir2_free_read(tp, dp,
 659                                                 xfs_dir2_db_to_da(args->geo,
 660                                                                   newfdb),
 661                                                 &curbp);
 662                                 if (error)
 663                                         return error;
 664                                 free = curbp->b_addr;
 665
 666                                 xfs_dir2_free_hdr_check(dp, curbp, curdb);
 667                         }
 668                         /*
 669                          * Get the index for our entry.
 670                          */
 671                         fi = dp->d_ops->db_to_fdindex(args->geo, curdb);
 672                         /*
 673                          * If it has room, return it.
 674                          */
 675                         bests = dp->d_ops->free_bests_p(free);
 676                         if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
 677                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 678                                                         XFS_ERRLEVEL_LOW, mp);
 679                                 if (curfdb != newfdb)
 680                                         xfs_trans_brelse(tp, curbp);
 681                                 return -EFSCORRUPTED;
 682                         }
 683                         curfdb = newfdb;
 684                         if (be16_to_cpu(bests[fi]) >= length)
 685                                 goto out;
 686                 }
 687         }
 688         /* Didn't find any space */
 689         fi = -1;
 690 out:
 691         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 692         if (curbp) {
 693                 /* Giving back a free block. */
 694                 state->extravalid = 1;
 695                 state->extrablk.bp = curbp;
 696                 state->extrablk.index = fi;
 697                 state->extrablk.blkno = curfdb;
 698
 699                 /*
 700                  * Important: this magic number is not in the buffer - it's for
 701                  * buffer type information and therefore only the free/data type
 702                  * matters here, not whether CRCs are enabled or not.
 703                  */
 704                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 705         } else {
 706                 state->extravalid = 0;
 707         }
 708         /*
 709          * Return the index, that will be the insertion point.
 710          */
 711         *indexp = index;
 712         return -ENOENT;
 713 }
 714
 715 /*
 716  * Look up a leaf entry in a node-format leaf block.
 717  * The extrablk in state a data block.
 718  */
 719 STATIC int
 720 xfs_dir2_leafn_lookup_for_entry(
 721         struct xfs_buf          *bp,            /* leaf buffer */
 722         xfs_da_args_t           *args,          /* operation arguments */
 723         int                     *indexp,        /* out: leaf entry index */
 724         xfs_da_state_t          *state)         /* state to fill in */
 725 {
 726         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 727         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 728         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 729         xfs_inode_t             *dp;            /* incore directory inode */
 730         int                     error;          /* error return value */
 731         int                     index;          /* leaf entry index */
 732         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 733         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 734         xfs_mount_t             *mp;            /* filesystem mount point */
 735         xfs_dir2_db_t           newdb;          /* new data block number */
 736         xfs_trans_t             *tp;            /* transaction pointer */
 737         enum xfs_dacmp          cmp;            /* comparison result */
 738         struct xfs_dir2_leaf_entry *ents;
 739         struct xfs_dir3_icleaf_hdr leafhdr;
 740
 741         dp = args->dp;
 742         tp = args->trans;
 743         mp = dp->i_mount;
 744         leaf = bp->b_addr;
 745         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 746         ents = dp->d_ops->leaf_ents_p(leaf);
 747
 748         xfs_dir3_leaf_check(dp, bp);
 749         ASSERT(leafhdr.count > 0);
 750
 751         /*
 752          * Look up the hash value in the leaf entries.
 753          */
 754         index = xfs_dir2_leaf_search_hash(args, bp);
 755         /*
 756          * Do we have a buffer coming in?
 757          */
 758         if (state->extravalid) {
 759                 curbp = state->extrablk.bp;
 760                 curdb = state->extrablk.blkno;
 761         }
 762         /*
 763          * Loop over leaf entries with the right hash value.
 764          */
 765         for (lep = &ents[index];
 766              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 767              lep++, index++) {
 768                 /*
 769                  * Skip stale leaf entries.
 770                  */
 771                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 772                         continue;
 773                 /*
 774                  * Pull the data block number from the entry.
 775                  */
 776                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 777                                                be32_to_cpu(lep->address));
 778                 /*
 779                  * Not adding a new entry, so we really want to find
 780                  * the name given to us.
 781                  *
 782                  * If it's a different data block, go get it.
 783                  */
 784                 if (newdb != curdb) {
 785                         /*
 786                          * If we had a block before that we aren't saving
 787                          * for a CI name, drop it
 788                          */
 789                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 790                                                 curdb != state->extrablk.blkno))
 791                                 xfs_trans_brelse(tp, curbp);
 792                         /*
 793                          * If needing the block that is saved with a CI match,
 794                          * use it otherwise read in the new data block.
 795                          */
 796                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 797                                         newdb == state->extrablk.blkno) {
 798                                 ASSERT(state->extravalid);
 799                                 curbp = state->extrablk.bp;
 800                         } else {
 801                                 error = xfs_dir3_data_read(tp, dp,
 802                                                 xfs_dir2_db_to_da(args->geo,
 803                                                                   newdb),
 804                                                 -1, &curbp);
 805                                 if (error)
 806                                         return error;
 807                         }
 808                         xfs_dir3_data_check(dp, curbp);
 809                         curdb = newdb;
 810                 }
 811                 /*
 812                  * Point to the data entry.
 813                  */
 814                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 815                         xfs_dir2_dataptr_to_off(args->geo,
 816                                                 be32_to_cpu(lep->address)));
 817                 /*
 818                  * Compare the entry and if it's an exact match, return
 819                  * EEXIST immediately. If it's the first case-insensitive
 820                  * match, store the block & inode number and continue looking.
 821                  */
 822                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 823                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 824                         /* If there is a CI match block, drop it */
 825                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 826                                                 curdb != state->extrablk.blkno)
 827                                 xfs_trans_brelse(tp, state->extrablk.bp);
 828                         args->cmpresult = cmp;
 829                         args->inumber = be64_to_cpu(dep->inumber);
 830                         args->filetype = dp->d_ops->data_get_ftype(dep);
 831                         *indexp = index;
 832                         state->extravalid = 1;
 833                         state->extrablk.bp = curbp;
 834                         state->extrablk.blkno = curdb;
 835                         state->extrablk.index = (int)((char *)dep -
 836                                                         (char *)curbp->b_addr);
 837                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 838                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 839                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 840                         if (cmp == XFS_CMP_EXACT)
 841                                 return -EEXIST;
 842                 }
 843         }
 844         ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
 845         if (curbp) {
 846                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 847                         /* Giving back last used data block. */
 848                         state->extravalid = 1;
 849                         state->extrablk.bp = curbp;
 850                         state->extrablk.index = -1;
 851                         state->extrablk.blkno = curdb;
 852                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 853                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 854                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 855                 } else {
 856                         /* If the curbp is not the CI match block, drop it */
 857                         if (state->extrablk.bp != curbp)
 858                                 xfs_trans_brelse(tp, curbp);
 859                 }
 860         } else {
 861                 state->extravalid = 0;
 862         }
 863         *indexp = index;
 864         return -ENOENT;
 865 }
 866
 867 /*
 868  * Look up a leaf entry in a node-format leaf block.
 869  * If this is an addname then the extrablk in state is a freespace block,
 870  * otherwise it's a data block.
 871  */
 872 int
 873 xfs_dir2_leafn_lookup_int(
 874         struct xfs_buf          *bp,            /* leaf buffer */
 875         xfs_da_args_t           *args,          /* operation arguments */
 876         int                     *indexp,        /* out: leaf entry index */
 877         xfs_da_state_t          *state)         /* state to fill in */
 878 {
 879         if (args->op_flags & XFS_DA_OP_ADDNAME)
 880                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 881                                                         state);
 882         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 883 }
 884
 885 /*
 886  * Move count leaf entries from source to destination leaf.
 887  * Log entries and headers.  Stale entries are preserved.
 888  */
 889 static void
 890 xfs_dir3_leafn_moveents(
 891         xfs_da_args_t                   *args,  /* operation arguments */
 892         struct xfs_buf                  *bp_s,  /* source */
 893         struct xfs_dir3_icleaf_hdr      *shdr,
 894         struct xfs_dir2_leaf_entry      *sents,
 895         int                             start_s,/* source leaf index */
 896         struct xfs_buf                  *bp_d,  /* destination */
 897         struct xfs_dir3_icleaf_hdr      *dhdr,
 898         struct xfs_dir2_leaf_entry      *dents,
 899         int                             start_d,/* destination leaf index */
 900         int                             count)  /* count of leaves to copy */
 901 {
 902         int                             stale;  /* count stale leaves copied */
 903
 904         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 905
 906         /*
 907          * Silently return if nothing to do.
 908          */
 909         if (count == 0)
 910                 return;
 911
 912         /*
 913          * If the destination index is not the end of the current
 914          * destination leaf entries, open up a hole in the destination
 915          * to hold the new entries.
 916          */
 917         if (start_d < dhdr->count) {
 918                 memmove(&dents[start_d + count], &dents[start_d],
 919                         (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
 920                 xfs_dir3_leaf_log_ents(args, bp_d, start_d + count,
 921                                        count + dhdr->count - 1);
 922         }
 923         /*
 924          * If the source has stale leaves, count the ones in the copy range
 925          * so we can update the header correctly.
 926          */
 927         if (shdr->stale) {
 928                 int     i;                      /* temp leaf index */
 929
 930                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 931                         if (sents[i].address ==
 932                                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 933                                 stale++;
 934                 }
 935         } else
 936                 stale = 0;
 937         /*
 938          * Copy the leaf entries from source to destination.
 939          */
 940         memcpy(&dents[start_d], &sents[start_s],
 941                 count * sizeof(xfs_dir2_leaf_entry_t));
 942         xfs_dir3_leaf_log_ents(args, bp_d, start_d, start_d + count - 1);
 943
 944         /*
 945          * If there are source entries after the ones we copied,
 946          * delete the ones we copied by sliding the next ones down.
 947          */
 948         if (start_s + count < shdr->count) {
 949                 memmove(&sents[start_s], &sents[start_s + count],
 950                         count * sizeof(xfs_dir2_leaf_entry_t));
 951                 xfs_dir3_leaf_log_ents(args, bp_s, start_s, start_s + count - 1);
 952         }
 953
 954         /*
 955          * Update the headers and log them.
 956          */
 957         shdr->count -= count;
 958         shdr->stale -= stale;
 959         dhdr->count += count;
 960         dhdr->stale += stale;
 961 }
 962
 963 /*
 964  * Determine the sort order of two leaf blocks.
 965  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 966  */
 967 int                                             /* sort order */
 968 xfs_dir2_leafn_order(
 969         struct xfs_inode        *dp,
 970         struct xfs_buf          *leaf1_bp,              /* leaf1 buffer */
 971         struct xfs_buf          *leaf2_bp)              /* leaf2 buffer */
 972 {
 973         struct xfs_dir2_leaf    *leaf1 = leaf1_bp->b_addr;
 974         struct xfs_dir2_leaf    *leaf2 = leaf2_bp->b_addr;
 975         struct xfs_dir2_leaf_entry *ents1;
 976         struct xfs_dir2_leaf_entry *ents2;
 977         struct xfs_dir3_icleaf_hdr hdr1;
 978         struct xfs_dir3_icleaf_hdr hdr2;
 979
 980         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
 981         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
 982         ents1 = dp->d_ops->leaf_ents_p(leaf1);
 983         ents2 = dp->d_ops->leaf_ents_p(leaf2);
 984
 985         if (hdr1.count > 0 && hdr2.count > 0 &&
 986             (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
 987              be32_to_cpu(ents2[hdr2.count - 1].hashval) <
 988                                 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
 989                 return 1;
 990         return 0;
 991 }
 992
 993 /*
 994  * Rebalance leaf entries between two leaf blocks.
 995  * This is actually only called when the second block is new,
 996  * though the code deals with the general case.
 997  * A new entry will be inserted in one of the blocks, and that
 998  * entry is taken into account when balancing.
 999  */
1000 static void
1001 xfs_dir2_leafn_rebalance(
1002         xfs_da_state_t          *state,         /* btree cursor */
1003         xfs_da_state_blk_t      *blk1,          /* first btree block */
1004         xfs_da_state_blk_t      *blk2)          /* second btree block */
1005 {
1006         xfs_da_args_t           *args;          /* operation arguments */
1007         int                     count;          /* count (& direction) leaves */
1008         int                     isleft;         /* new goes in left leaf */
1009         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
1010         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
1011         int                     mid;            /* midpoint leaf index */
1012 #if defined(DEBUG) || defined(XFS_WARN)
1013         int                     oldstale;       /* old count of stale leaves */
1014 #endif
1015         int                     oldsum;         /* old total leaf count */
1016         int                     swap_blocks;    /* swapped leaf blocks */
1017         struct xfs_dir2_leaf_entry *ents1;
1018         struct xfs_dir2_leaf_entry *ents2;
1019         struct xfs_dir3_icleaf_hdr hdr1;
1020         struct xfs_dir3_icleaf_hdr hdr2;
1021         struct xfs_inode        *dp = state->args->dp;
1022
1023         args = state->args;
1024         /*
1025          * If the block order is wrong, swap the arguments.
1026          */
1027         swap_blocks = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp);
1028         if (swap_blocks)
1029                 swap(blk1, blk2);
1030
1031         leaf1 = blk1->bp->b_addr;
1032         leaf2 = blk2->bp->b_addr;
1033         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
1034         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
1035         ents1 = dp->d_ops->leaf_ents_p(leaf1);
1036         ents2 = dp->d_ops->leaf_ents_p(leaf2);
1037
1038         oldsum = hdr1.count + hdr2.count;
1039 #if defined(DEBUG) || defined(XFS_WARN)
1040         oldstale = hdr1.stale + hdr2.stale;
1041 #endif
1042         mid = oldsum >> 1;
1043
1044         /*
1045          * If the old leaf count was odd then the new one will be even,
1046          * so we need to divide the new count evenly.
1047          */
1048         if (oldsum & 1) {
1049                 xfs_dahash_t    midhash;        /* middle entry hash value */
1050
1051                 if (mid >= hdr1.count)
1052                         midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1053                 else
1054                         midhash = be32_to_cpu(ents1[mid].hashval);
1055                 isleft = args->hashval <= midhash;
1056         }
1057         /*
1058          * If the old count is even then the new count is odd, so there's
1059          * no preferred side for the new entry.
1060          * Pick the left one.
1061          */
1062         else
1063                 isleft = 1;
1064         /*
1065          * Calculate moved entry count.  Positive means left-to-right,
1066          * negative means right-to-left.  Then move the entries.
1067          */
1068         count = hdr1.count - mid + (isleft == 0);
1069         if (count > 0)
1070                 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1071                                         hdr1.count - count, blk2->bp,
1072                                         &hdr2, ents2, 0, count);
1073         else if (count < 0)
1074                 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1075                                         blk1->bp, &hdr1, ents1,
1076                                         hdr1.count, count);
1077
1078         ASSERT(hdr1.count + hdr2.count == oldsum);
1079         ASSERT(hdr1.stale + hdr2.stale == oldstale);
1080
1081         /* log the changes made when moving the entries */
1082         dp->d_ops->leaf_hdr_to_disk(leaf1, &hdr1);
1083         dp->d_ops->leaf_hdr_to_disk(leaf2, &hdr2);
1084         xfs_dir3_leaf_log_header(args, blk1->bp);
1085         xfs_dir3_leaf_log_header(args, blk2->bp);
1086
1087         xfs_dir3_leaf_check(dp, blk1->bp);
1088         xfs_dir3_leaf_check(dp, blk2->bp);
1089
1090         /*
1091          * Mark whether we're inserting into the old or new leaf.
1092          */
1093         if (hdr1.count < hdr2.count)
1094                 state->inleaf = swap_blocks;
1095         else if (hdr1.count > hdr2.count)
1096                 state->inleaf = !swap_blocks;
1097         else
1098                 state->inleaf = swap_blocks ^ (blk1->index <= hdr1.count);
1099         /*
1100          * Adjust the expected index for insertion.
1101          */
1102         if (!state->inleaf)
1103                 blk2->index = blk1->index - hdr1.count;
1104
1105         /*
1106          * Finally sanity check just to make sure we are not returning a
1107          * negative index
1108          */
1109         if (blk2->index < 0) {
1110                 state->inleaf = 1;
1111                 blk2->index = 0;
1112                 xfs_alert(dp->i_mount,
1113         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1114                         __func__, blk1->index);
1115         }
1116 }
1117
1118 static int
1119 xfs_dir3_data_block_free(
1120         xfs_da_args_t           *args,
1121         struct xfs_dir2_data_hdr *hdr,
1122         struct xfs_dir2_free    *free,
1123         xfs_dir2_db_t           fdb,
1124         int                     findex,
1125         struct xfs_buf          *fbp,
1126         int                     longest)
1127 {
1128         int                     logfree = 0;
1129         __be16                  *bests;
1130         struct xfs_dir3_icfree_hdr freehdr;
1131         struct xfs_inode        *dp = args->dp;
1132
1133         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1134         bests = dp->d_ops->free_bests_p(free);
1135         if (hdr) {
1136                 /*
1137                  * Data block is not empty, just set the free entry to the new
1138                  * value.
1139                  */
1140                 bests[findex] = cpu_to_be16(longest);
1141                 xfs_dir2_free_log_bests(args, fbp, findex, findex);
1142                 return 0;
1143         }
1144
1145         /* One less used entry in the free table. */
1146         freehdr.nused--;
1147
1148         /*
1149          * If this was the last entry in the table, we can trim the table size
1150          * back.  There might be other entries at the end referring to
1151          * non-existent data blocks, get those too.
1152          */
1153         if (findex == freehdr.nvalid - 1) {
1154                 int     i;              /* free entry index */
1155
1156                 for (i = findex - 1; i >= 0; i--) {
1157                         if (bests[i] != cpu_to_be16(NULLDATAOFF))
1158                                 break;
1159                 }
1160                 freehdr.nvalid = i + 1;
1161                 logfree = 0;
1162         } else {
1163                 /* Not the last entry, just punch it out.  */
1164                 bests[findex] = cpu_to_be16(NULLDATAOFF);
1165                 logfree = 1;
1166         }
1167
1168         dp->d_ops->free_hdr_to_disk(free, &freehdr);
1169         xfs_dir2_free_log_header(args, fbp);
1170
1171         /*
1172          * If there are no useful entries left in the block, get rid of the
1173          * block if we can.
1174          */
1175         if (!freehdr.nused) {
1176                 int error;
1177
1178                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1179                 if (error == 0) {
1180                         fbp = NULL;
1181                         logfree = 0;
1182                 } else if (error != -ENOSPC || args->total != 0)
1183                         return error;
1184                 /*
1185                  * It's possible to get ENOSPC if there is no
1186                  * space reservation.  In this case some one
1187                  * else will eventually get rid of this block.
1188                  */
1189         }
1190
1191         /* Log the free entry that changed, unless we got rid of it.  */
1192         if (logfree)
1193                 xfs_dir2_free_log_bests(args, fbp, findex, findex);
1194         return 0;
1195 }
1196
1197 /*
1198  * Remove an entry from a node directory.
1199  * This removes the leaf entry and the data entry,
1200  * and updates the free block if necessary.
1201  */
1202 static int                                      /* error */
1203 xfs_dir2_leafn_remove(
1204         xfs_da_args_t           *args,          /* operation arguments */
1205         struct xfs_buf          *bp,            /* leaf buffer */
1206         int                     index,          /* leaf entry index */
1207         xfs_da_state_blk_t      *dblk,          /* data block */
1208         int                     *rval)          /* resulting block needs join */
1209 {
1210         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1211         xfs_dir2_db_t           db;             /* data block number */
1212         struct xfs_buf          *dbp;           /* data block buffer */
1213         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1214         xfs_inode_t             *dp;            /* incore directory inode */
1215         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1216         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1217         int                     longest;        /* longest data free entry */
1218         int                     off;            /* data block entry offset */
1219         int                     needlog;        /* need to log data header */
1220         int                     needscan;       /* need to rescan data frees */
1221         xfs_trans_t             *tp;            /* transaction pointer */
1222         struct xfs_dir2_data_free *bf;          /* bestfree table */
1223         struct xfs_dir3_icleaf_hdr leafhdr;
1224         struct xfs_dir2_leaf_entry *ents;
1225
1226         trace_xfs_dir2_leafn_remove(args, index);
1227
1228         dp = args->dp;
1229         tp = args->trans;
1230         leaf = bp->b_addr;
1231         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1232         ents = dp->d_ops->leaf_ents_p(leaf);
1233
1234         /*
1235          * Point to the entry we're removing.
1236          */
1237         lep = &ents[index];
1238
1239         /*
1240          * Extract the data block and offset from the entry.
1241          */
1242         db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1243         ASSERT(dblk->blkno == db);
1244         off = xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address));
1245         ASSERT(dblk->index == off);
1246
1247         /*
1248          * Kill the leaf entry by marking it stale.
1249          * Log the leaf block changes.
1250          */
1251         leafhdr.stale++;
1252         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1253         xfs_dir3_leaf_log_header(args, bp);
1254
1255         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1256         xfs_dir3_leaf_log_ents(args, bp, index, index);
1257
1258         /*
1259          * Make the data entry free.  Keep track of the longest freespace
1260          * in the data block in case it changes.
1261          */
1262         dbp = dblk->bp;
1263         hdr = dbp->b_addr;
1264         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1265         bf = dp->d_ops->data_bestfree_p(hdr);
1266         longest = be16_to_cpu(bf[0].length);
1267         needlog = needscan = 0;
1268         xfs_dir2_data_make_free(args, dbp, off,
1269                 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1270         /*
1271          * Rescan the data block freespaces for bestfree.
1272          * Log the data block header if needed.
1273          */
1274         if (needscan)
1275                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1276         if (needlog)
1277                 xfs_dir2_data_log_header(args, dbp);
1278         xfs_dir3_data_check(dp, dbp);
1279         /*
1280          * If the longest data block freespace changes, need to update
1281          * the corresponding freeblock entry.
1282          */
1283         if (longest < be16_to_cpu(bf[0].length)) {
1284                 int             error;          /* error return value */
1285                 struct xfs_buf  *fbp;           /* freeblock buffer */
1286                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1287                 int             findex;         /* index in freeblock entries */
1288                 xfs_dir2_free_t *free;          /* freeblock structure */
1289
1290                 /*
1291                  * Convert the data block number to a free block,
1292                  * read in the free block.
1293                  */
1294                 fdb = dp->d_ops->db_to_fdb(args->geo, db);
1295                 error = xfs_dir2_free_read(tp, dp,
1296                                            xfs_dir2_db_to_da(args->geo, fdb),
1297                                            &fbp);
1298                 if (error)
1299                         return error;
1300                 free = fbp->b_addr;
1301 #ifdef DEBUG
1302         {
1303                 struct xfs_dir3_icfree_hdr freehdr;
1304                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1305                 ASSERT(freehdr.firstdb == dp->d_ops->free_max_bests(args->geo) *
1306                         (fdb - xfs_dir2_byte_to_db(args->geo,
1307                                                    XFS_DIR2_FREE_OFFSET)));
1308         }
1309 #endif
1310                 /*
1311                  * Calculate which entry we need to fix.
1312                  */
1313                 findex = dp->d_ops->db_to_fdindex(args->geo, db);
1314                 longest = be16_to_cpu(bf[0].length);
1315                 /*
1316                  * If the data block is now empty we can get rid of it
1317                  * (usually).
1318                  */
1319                 if (longest == args->geo->blksize -
1320                                dp->d_ops->data_entry_offset) {
1321                         /*
1322                          * Try to punch out the data block.
1323                          */
1324                         error = xfs_dir2_shrink_inode(args, db, dbp);
1325                         if (error == 0) {
1326                                 dblk->bp = NULL;
1327                                 hdr = NULL;
1328                         }
1329                         /*
1330                          * We can get ENOSPC if there's no space reservation.
1331                          * In this case just drop the buffer and some one else
1332                          * will eventually get rid of the empty block.
1333                          */
1334                         else if (!(error == -ENOSPC && args->total == 0))
1335                                 return error;
1336                 }
1337                 /*
1338                  * If we got rid of the data block, we can eliminate that entry
1339                  * in the free block.
1340                  */
1341                 error = xfs_dir3_data_block_free(args, hdr, free,
1342                                                  fdb, findex, fbp, longest);
1343                 if (error)
1344                         return error;
1345         }
1346
1347         xfs_dir3_leaf_check(dp, bp);
1348         /*
1349          * Return indication of whether this leaf block is empty enough
1350          * to justify trying to join it with a neighbor.
1351          */
1352         *rval = (dp->d_ops->leaf_hdr_size +
1353                  (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1354                 args->geo->magicpct;
1355         return 0;
1356 }
1357
1358 /*
1359  * Split the leaf entries in the old block into old and new blocks.
1360  */
1361 int                                             /* error */
1362 xfs_dir2_leafn_split(
1363         xfs_da_state_t          *state,         /* btree cursor */
1364         xfs_da_state_blk_t      *oldblk,        /* original block */
1365         xfs_da_state_blk_t      *newblk)        /* newly created block */
1366 {
1367         xfs_da_args_t           *args;          /* operation arguments */
1368         xfs_dablk_t             blkno;          /* new leaf block number */
1369         int                     error;          /* error return value */
1370         struct xfs_inode        *dp;
1371
1372         /*
1373          * Allocate space for a new leaf node.
1374          */
1375         args = state->args;
1376         dp = args->dp;
1377         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1378         error = xfs_da_grow_inode(args, &blkno);
1379         if (error) {
1380                 return error;
1381         }
1382         /*
1383          * Initialize the new leaf block.
1384          */
1385         error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(args->geo, blkno),
1386                                       &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1387         if (error)
1388                 return error;
1389
1390         newblk->blkno = blkno;
1391         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1392         /*
1393          * Rebalance the entries across the two leaves, link the new
1394          * block into the leaves.
1395          */
1396         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1397         error = xfs_da3_blk_link(state, oldblk, newblk);
1398         if (error) {
1399                 return error;
1400         }
1401         /*
1402          * Insert the new entry in the correct block.
1403          */
1404         if (state->inleaf)
1405                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1406         else
1407                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1408         /*
1409          * Update last hashval in each block since we added the name.
1410          */
1411         oldblk->hashval = xfs_dir2_leaf_lasthash(dp, oldblk->bp, NULL);
1412         newblk->hashval = xfs_dir2_leaf_lasthash(dp, newblk->bp, NULL);
1413         xfs_dir3_leaf_check(dp, oldblk->bp);
1414         xfs_dir3_leaf_check(dp, newblk->bp);
1415         return error;
1416 }
1417
1418 /*
1419  * Check a leaf block and its neighbors to see if the block should be
1420  * collapsed into one or the other neighbor.  Always keep the block
1421  * with the smaller block number.
1422  * If the current block is over 50% full, don't try to join it, return 0.
1423  * If the block is empty, fill in the state structure and return 2.
1424  * If it can be collapsed, fill in the state structure and return 1.
1425  * If nothing can be done, return 0.
1426  */
1427 int                                             /* error */
1428 xfs_dir2_leafn_toosmall(
1429         xfs_da_state_t          *state,         /* btree cursor */
1430         int                     *action)        /* resulting action to take */
1431 {
1432         xfs_da_state_blk_t      *blk;           /* leaf block */
1433         xfs_dablk_t             blkno;          /* leaf block number */
1434         struct xfs_buf          *bp;            /* leaf buffer */
1435         int                     bytes;          /* bytes in use */
1436         int                     count;          /* leaf live entry count */
1437         int                     error;          /* error return value */
1438         int                     forward;        /* sibling block direction */
1439         int                     i;              /* sibling counter */
1440         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1441         int                     rval;           /* result from path_shift */
1442         struct xfs_dir3_icleaf_hdr leafhdr;
1443         struct xfs_dir2_leaf_entry *ents;
1444         struct xfs_inode        *dp = state->args->dp;
1445
1446         /*
1447          * Check for the degenerate case of the block being over 50% full.
1448          * If so, it's not worth even looking to see if we might be able
1449          * to coalesce with a sibling.
1450          */
1451         blk = &state->path.blk[state->path.active - 1];
1452         leaf = blk->bp->b_addr;
1453         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1454         ents = dp->d_ops->leaf_ents_p(leaf);
1455         xfs_dir3_leaf_check(dp, blk->bp);
1456
1457         count = leafhdr.count - leafhdr.stale;
1458         bytes = dp->d_ops->leaf_hdr_size + count * sizeof(ents[0]);
1459         if (bytes > (state->args->geo->blksize >> 1)) {
1460                 /*
1461                  * Blk over 50%, don't try to join.
1462                  */
1463                 *action = 0;
1464                 return 0;
1465         }
1466         /*
1467          * Check for the degenerate case of the block being empty.
1468          * If the block is empty, we'll simply delete it, no need to
1469          * coalesce it with a sibling block.  We choose (arbitrarily)
1470          * to merge with the forward block unless it is NULL.
1471          */
1472         if (count == 0) {
1473                 /*
1474                  * Make altpath point to the block we want to keep and
1475                  * path point to the block we want to drop (this one).
1476                  */
1477                 forward = (leafhdr.forw != 0);
1478                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1479                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1480                         &rval);
1481                 if (error)
1482                         return error;
1483                 *action = rval ? 2 : 0;
1484                 return 0;
1485         }
1486         /*
1487          * Examine each sibling block to see if we can coalesce with
1488          * at least 25% free space to spare.  We need to figure out
1489          * whether to merge with the forward or the backward block.
1490          * We prefer coalescing with the lower numbered sibling so as
1491          * to shrink a directory over time.
1492          */
1493         forward = leafhdr.forw < leafhdr.back;
1494         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1495                 struct xfs_dir3_icleaf_hdr hdr2;
1496
1497                 blkno = forward ? leafhdr.forw : leafhdr.back;
1498                 if (blkno == 0)
1499                         continue;
1500                 /*
1501                  * Read the sibling leaf block.
1502                  */
1503                 error = xfs_dir3_leafn_read(state->args->trans, dp,
1504                                             blkno, -1, &bp);
1505                 if (error)
1506                         return error;
1507
1508                 /*
1509                  * Count bytes in the two blocks combined.
1510                  */
1511                 count = leafhdr.count - leafhdr.stale;
1512                 bytes = state->args->geo->blksize -
1513                         (state->args->geo->blksize >> 2);
1514
1515                 leaf = bp->b_addr;
1516                 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf);
1517                 ents = dp->d_ops->leaf_ents_p(leaf);
1518                 count += hdr2.count - hdr2.stale;
1519                 bytes -= count * sizeof(ents[0]);
1520
1521                 /*
1522                  * Fits with at least 25% to spare.
1523                  */
1524                 if (bytes >= 0)
1525                         break;
1526                 xfs_trans_brelse(state->args->trans, bp);
1527         }
1528         /*
1529          * Didn't like either block, give up.
1530          */
1531         if (i >= 2) {
1532                 *action = 0;
1533                 return 0;
1534         }
1535
1536         /*
1537          * Make altpath point to the block we want to keep (the lower
1538          * numbered block) and path point to the block we want to drop.
1539          */
1540         memcpy(&state->altpath, &state->path, sizeof(state->path));
1541         if (blkno < blk->blkno)
1542                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1543                         &rval);
1544         else
1545                 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1546                         &rval);
1547         if (error) {
1548                 return error;
1549         }
1550         *action = rval ? 0 : 1;
1551         return 0;
1552 }
1553
1554 /*
1555  * Move all the leaf entries from drop_blk to save_blk.
1556  * This is done as part of a join operation.
1557  */
1558 void
1559 xfs_dir2_leafn_unbalance(
1560         xfs_da_state_t          *state,         /* cursor */
1561         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1562         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1563 {
1564         xfs_da_args_t           *args;          /* operation arguments */
1565         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1566         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1567         struct xfs_dir3_icleaf_hdr savehdr;
1568         struct xfs_dir3_icleaf_hdr drophdr;
1569         struct xfs_dir2_leaf_entry *sents;
1570         struct xfs_dir2_leaf_entry *dents;
1571         struct xfs_inode        *dp = state->args->dp;
1572
1573         args = state->args;
1574         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1575         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1576         drop_leaf = drop_blk->bp->b_addr;
1577         save_leaf = save_blk->bp->b_addr;
1578
1579         dp->d_ops->leaf_hdr_from_disk(&savehdr, save_leaf);
1580         dp->d_ops->leaf_hdr_from_disk(&drophdr, drop_leaf);
1581         sents = dp->d_ops->leaf_ents_p(save_leaf);
1582         dents = dp->d_ops->leaf_ents_p(drop_leaf);
1583
1584         /*
1585          * If there are any stale leaf entries, take this opportunity
1586          * to purge them.
1587          */
1588         if (drophdr.stale)
1589                 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1590         if (savehdr.stale)
1591                 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1592
1593         /*
1594          * Move the entries from drop to the appropriate end of save.
1595          */
1596         drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1597         if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1598                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1599                                         save_blk->bp, &savehdr, sents, 0,
1600                                         drophdr.count);
1601         else
1602                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1603                                         save_blk->bp, &savehdr, sents,
1604                                         savehdr.count, drophdr.count);
1605         save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1606
1607         /* log the changes made when moving the entries */
1608         dp->d_ops->leaf_hdr_to_disk(save_leaf, &savehdr);
1609         dp->d_ops->leaf_hdr_to_disk(drop_leaf, &drophdr);
1610         xfs_dir3_leaf_log_header(args, save_blk->bp);
1611         xfs_dir3_leaf_log_header(args, drop_blk->bp);
1612
1613         xfs_dir3_leaf_check(dp, save_blk->bp);
1614         xfs_dir3_leaf_check(dp, drop_blk->bp);
1615 }
1616
1617 /*
1618  * Top-level node form directory addname routine.
1619  */
1620 int                                             /* error */
1621 xfs_dir2_node_addname(
1622         xfs_da_args_t           *args)          /* operation arguments */
1623 {
1624         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1625         int                     error;          /* error return value */
1626         int                     rval;           /* sub-return value */
1627         xfs_da_state_t          *state;         /* btree cursor */
1628
1629         trace_xfs_dir2_node_addname(args);
1630
1631         /*
1632          * Allocate and initialize the state (btree cursor).
1633          */
1634         state = xfs_da_state_alloc();
1635         state->args = args;
1636         state->mp = args->dp->i_mount;
1637         /*
1638          * Look up the name.  We're not supposed to find it, but
1639          * this gives us the insertion point.
1640          */
1641         error = xfs_da3_node_lookup_int(state, &rval);
1642         if (error)
1643                 rval = error;
1644         if (rval != -ENOENT) {
1645                 goto done;
1646         }
1647         /*
1648          * Add the data entry to a data block.
1649          * Extravalid is set to a freeblock found by lookup.
1650          */
1651         rval = xfs_dir2_node_addname_int(args,
1652                 state->extravalid ? &state->extrablk : NULL);
1653         if (rval) {
1654                 goto done;
1655         }
1656         blk = &state->path.blk[state->path.active - 1];
1657         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1658         /*
1659          * Add the new leaf entry.
1660          */
1661         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1662         if (rval == 0) {
1663                 /*
1664                  * It worked, fix the hash values up the btree.
1665                  */
1666                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1667                         xfs_da3_fixhashpath(state, &state->path);
1668         } else {
1669                 /*
1670                  * It didn't work, we need to split the leaf block.
1671                  */
1672                 if (args->total == 0) {
1673                         ASSERT(rval == -ENOSPC);
1674                         goto done;
1675                 }
1676                 /*
1677                  * Split the leaf block and insert the new entry.
1678                  */
1679                 rval = xfs_da3_split(state);
1680         }
1681 done:
1682         xfs_da_state_free(state);
1683         return rval;
1684 }
1685
1686 /*
1687  * Add the data entry for a node-format directory name addition.
1688  * The leaf entry is added in xfs_dir2_leafn_add.
1689  * We may enter with a freespace block that the lookup found.
1690  */
1691 static int                                      /* error */
1692 xfs_dir2_node_addname_int(
1693         xfs_da_args_t           *args,          /* operation arguments */
1694         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1695 {
1696         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1697         xfs_dir2_db_t           dbno;           /* data block number */
1698         struct xfs_buf          *dbp;           /* data block buffer */
1699         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1700         xfs_inode_t             *dp;            /* incore directory inode */
1701         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1702         int                     error;          /* error return value */
1703         xfs_dir2_db_t           fbno;           /* freespace block number */
1704         struct xfs_buf          *fbp;           /* freespace buffer */
1705         int                     findex;         /* freespace entry index */
1706         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1707         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1708         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1709         int                     length;         /* length of the new entry */
1710         int                     logfree;        /* need to log free entry */
1711         xfs_mount_t             *mp;            /* filesystem mount point */
1712         int                     needlog;        /* need to log data header */
1713         int                     needscan;       /* need to rescan data frees */
1714         __be16                  *tagp;          /* data entry tag pointer */
1715         xfs_trans_t             *tp;            /* transaction pointer */
1716         __be16                  *bests;
1717         struct xfs_dir3_icfree_hdr freehdr;
1718         struct xfs_dir2_data_free *bf;
1719         xfs_dir2_data_aoff_t    aoff;
1720
1721         dp = args->dp;
1722         mp = dp->i_mount;
1723         tp = args->trans;
1724         length = dp->d_ops->data_entsize(args->namelen);
1725         /*
1726          * If we came in with a freespace block that means that lookup
1727          * found an entry with our hash value.  This is the freespace
1728          * block for that data entry.
1729          */
1730         if (fblk) {
1731                 fbp = fblk->bp;
1732                 /*
1733                  * Remember initial freespace block number.
1734                  */
1735                 ifbno = fblk->blkno;
1736                 free = fbp->b_addr;
1737                 findex = fblk->index;
1738                 bests = dp->d_ops->free_bests_p(free);
1739                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1740
1741                 /*
1742                  * This means the free entry showed that the data block had
1743                  * space for our entry, so we remembered it.
1744                  * Use that data block.
1745                  */
1746                 if (findex >= 0) {
1747                         ASSERT(findex < freehdr.nvalid);
1748                         ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1749                         ASSERT(be16_to_cpu(bests[findex]) >= length);
1750                         dbno = freehdr.firstdb + findex;
1751                 } else {
1752                         /*
1753                          * The data block looked at didn't have enough room.
1754                          * We'll start at the beginning of the freespace entries.
1755                          */
1756                         dbno = -1;
1757                         findex = 0;
1758                 }
1759         } else {
1760                 /*
1761                  * Didn't come in with a freespace block, so no data block.
1762                  */
1763                 ifbno = dbno = -1;
1764                 fbp = NULL;
1765                 findex = 0;
1766         }
1767
1768         /*
1769          * If we don't have a data block yet, we're going to scan the
1770          * freespace blocks looking for one.  Figure out what the
1771          * highest freespace block number is.
1772          */
1773         if (dbno == -1) {
1774                 xfs_fileoff_t   fo;             /* freespace block number */
1775
1776                 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK)))
1777                         return error;
1778                 lastfbno = xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo);
1779                 fbno = ifbno;
1780         }
1781         /*
1782          * While we haven't identified a data block, search the freeblock
1783          * data for a good data block.  If we find a null freeblock entry,
1784          * indicating a hole in the data blocks, remember that.
1785          */
1786         while (dbno == -1) {
1787                 /*
1788                  * If we don't have a freeblock in hand, get the next one.
1789                  */
1790                 if (fbp == NULL) {
1791                         /*
1792                          * Happens the first time through unless lookup gave
1793                          * us a freespace block to start with.
1794                          */
1795                         if (++fbno == 0)
1796                                 fbno = xfs_dir2_byte_to_db(args->geo,
1797                                                         XFS_DIR2_FREE_OFFSET);
1798                         /*
1799                          * If it's ifbno we already looked at it.
1800                          */
1801                         if (fbno == ifbno)
1802                                 fbno++;
1803                         /*
1804                          * If it's off the end we're done.
1805                          */
1806                         if (fbno >= lastfbno)
1807                                 break;
1808                         /*
1809                          * Read the block.  There can be holes in the
1810                          * freespace blocks, so this might not succeed.
1811                          * This should be really rare, so there's no reason
1812                          * to avoid it.
1813                          */
1814                         error = xfs_dir2_free_try_read(tp, dp,
1815                                         xfs_dir2_db_to_da(args->geo, fbno),
1816                                         &fbp);
1817                         if (error)
1818                                 return error;
1819                         if (!fbp)
1820                                 continue;
1821                         free = fbp->b_addr;
1822                         findex = 0;
1823                 }
1824                 /*
1825                  * Look at the current free entry.  Is it good enough?
1826                  *
1827                  * The bests initialisation should be where the bufer is read in
1828                  * the above branch. But gcc is too stupid to realise that bests
1829                  * and the freehdr are actually initialised if they are placed
1830                  * there, so we have to do it here to avoid warnings. Blech.
1831                  */
1832                 bests = dp->d_ops->free_bests_p(free);
1833                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1834                 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1835                     be16_to_cpu(bests[findex]) >= length)
1836                         dbno = freehdr.firstdb + findex;
1837                 else {
1838                         /*
1839                          * Are we done with the freeblock?
1840                          */
1841                         if (++findex == freehdr.nvalid) {
1842                                 /*
1843                                  * Drop the block.
1844                                  */
1845                                 xfs_trans_brelse(tp, fbp);
1846                                 fbp = NULL;
1847                                 if (fblk && fblk->bp)
1848                                         fblk->bp = NULL;
1849                         }
1850                 }
1851         }
1852         /*
1853          * If we don't have a data block, we need to allocate one and make
1854          * the freespace entries refer to it.
1855          */
1856         if (unlikely(dbno == -1)) {
1857                 /*
1858                  * Not allowed to allocate, return failure.
1859                  */
1860                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1861                         return -ENOSPC;
1862
1863                 /*
1864                  * Allocate and initialize the new data block.
1865                  */
1866                 if (unlikely((error = xfs_dir2_grow_inode(args,
1867                                                          XFS_DIR2_DATA_SPACE,
1868                                                          &dbno)) ||
1869                     (error = xfs_dir3_data_init(args, dbno, &dbp))))
1870                         return error;
1871
1872                 /*
1873                  * If (somehow) we have a freespace block, get rid of it.
1874                  */
1875                 if (fbp)
1876                         xfs_trans_brelse(tp, fbp);
1877                 if (fblk && fblk->bp)
1878                         fblk->bp = NULL;
1879
1880                 /*
1881                  * Get the freespace block corresponding to the data block
1882                  * that was just allocated.
1883                  */
1884                 fbno = dp->d_ops->db_to_fdb(args->geo, dbno);
1885                 error = xfs_dir2_free_try_read(tp, dp,
1886                                        xfs_dir2_db_to_da(args->geo, fbno),
1887                                        &fbp);
1888                 if (error)
1889                         return error;
1890
1891                 /*
1892                  * If there wasn't a freespace block, the read will
1893                  * return a NULL fbp.  Allocate and initialize a new one.
1894                  */
1895                 if (!fbp) {
1896                         error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1897                                                     &fbno);
1898                         if (error)
1899                                 return error;
1900
1901                         if (dp->d_ops->db_to_fdb(args->geo, dbno) != fbno) {
1902                                 xfs_alert(mp,
1903 "%s: dir ino %llu needed freesp block %lld for data block %lld, got %lld ifbno %llu lastfbno %d",
1904                                         __func__, (unsigned long long)dp->i_ino,
1905                                         (long long)dp->d_ops->db_to_fdb(
1906                                                                 args->geo, dbno),
1907                                         (long long)dbno, (long long)fbno,
1908                                         (unsigned long long)ifbno, lastfbno);
1909                                 if (fblk) {
1910                                         xfs_alert(mp,
1911                                 " fblk "PTR_FMT" blkno %llu index %d magic 0x%x",
1912                                                 fblk,
1913                                                 (unsigned long long)fblk->blkno,
1914                                                 fblk->index,
1915                                                 fblk->magic);
1916                                 } else {
1917                                         xfs_alert(mp, " ... fblk is NULL");
1918                                 }
1919                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1920                                                  XFS_ERRLEVEL_LOW, mp);
1921                                 return -EFSCORRUPTED;
1922                         }
1923
1924                         /*
1925                          * Get a buffer for the new block.
1926                          */
1927                         error = xfs_dir3_free_get_buf(args, fbno, &fbp);
1928                         if (error)
1929                                 return error;
1930                         free = fbp->b_addr;
1931                         bests = dp->d_ops->free_bests_p(free);
1932                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1933
1934                         /*
1935                          * Remember the first slot as our empty slot.
1936                          */
1937                         freehdr.firstdb =
1938                                 (fbno - xfs_dir2_byte_to_db(args->geo,
1939                                                         XFS_DIR2_FREE_OFFSET)) *
1940                                         dp->d_ops->free_max_bests(args->geo);
1941                 } else {
1942                         free = fbp->b_addr;
1943                         bests = dp->d_ops->free_bests_p(free);
1944                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1945                 }
1946
1947                 /*
1948                  * Set the freespace block index from the data block number.
1949                  */
1950                 findex = dp->d_ops->db_to_fdindex(args->geo, dbno);
1951                 /*
1952                  * If it's after the end of the current entries in the
1953                  * freespace block, extend that table.
1954                  */
1955                 if (findex >= freehdr.nvalid) {
1956                         ASSERT(findex < dp->d_ops->free_max_bests(args->geo));
1957                         freehdr.nvalid = findex + 1;
1958                         /*
1959                          * Tag new entry so nused will go up.
1960                          */
1961                         bests[findex] = cpu_to_be16(NULLDATAOFF);
1962                 }
1963                 /*
1964                  * If this entry was for an empty data block
1965                  * (this should always be true) then update the header.
1966                  */
1967                 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1968                         freehdr.nused++;
1969                         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
1970                         xfs_dir2_free_log_header(args, fbp);
1971                 }
1972                 /*
1973                  * Update the real value in the table.
1974                  * We haven't allocated the data entry yet so this will
1975                  * change again.
1976                  */
1977                 hdr = dbp->b_addr;
1978                 bf = dp->d_ops->data_bestfree_p(hdr);
1979                 bests[findex] = bf[0].length;
1980                 logfree = 1;
1981         }
1982         /*
1983          * We had a data block so we don't have to make a new one.
1984          */
1985         else {
1986                 /*
1987                  * If just checking, we succeeded.
1988                  */
1989                 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1990                         return 0;
1991
1992                 /*
1993                  * Read the data block in.
1994                  */
1995                 error = xfs_dir3_data_read(tp, dp,
1996                                            xfs_dir2_db_to_da(args->geo, dbno),
1997                                            -1, &dbp);
1998                 if (error)
1999                         return error;
2000                 hdr = dbp->b_addr;
2001                 bf = dp->d_ops->data_bestfree_p(hdr);
2002                 logfree = 0;
2003         }
2004         ASSERT(be16_to_cpu(bf[0].length) >= length);
2005         /*
2006          * Point to the existing unused space.
2007          */
2008         dup = (xfs_dir2_data_unused_t *)
2009               ((char *)hdr + be16_to_cpu(bf[0].offset));
2010         needscan = needlog = 0;
2011         /*
2012          * Mark the first part of the unused space, inuse for us.
2013          */
2014         aoff = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
2015         error = xfs_dir2_data_use_free(args, dbp, dup, aoff, length,
2016                         &needlog, &needscan);
2017         if (error) {
2018                 xfs_trans_brelse(tp, dbp);
2019                 return error;
2020         }
2021         /*
2022          * Fill in the new entry and log it.
2023          */
2024         dep = (xfs_dir2_data_entry_t *)dup;
2025         dep->inumber = cpu_to_be64(args->inumber);
2026         dep->namelen = args->namelen;
2027         memcpy(dep->name, args->name, dep->namelen);
2028         dp->d_ops->data_put_ftype(dep, args->filetype);
2029         tagp = dp->d_ops->data_entry_tag_p(dep);
2030         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
2031         xfs_dir2_data_log_entry(args, dbp, dep);
2032         /*
2033          * Rescan the block for bestfree if needed.
2034          */
2035         if (needscan)
2036                 xfs_dir2_data_freescan(dp, hdr, &needlog);
2037         /*
2038          * Log the data block header if needed.
2039          */
2040         if (needlog)
2041                 xfs_dir2_data_log_header(args, dbp);
2042         /*
2043          * If the freespace entry is now wrong, update it.
2044          */
2045         bests = dp->d_ops->free_bests_p(free); /* gcc is so stupid */
2046         if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
2047                 bests[findex] = bf[0].length;
2048                 logfree = 1;
2049         }
2050         /*
2051          * Log the freespace entry if needed.
2052          */
2053         if (logfree)
2054                 xfs_dir2_free_log_bests(args, fbp, findex, findex);
2055         /*
2056          * Return the data block and offset in args, then drop the data block.
2057          */
2058         args->blkno = (xfs_dablk_t)dbno;
2059         args->index = be16_to_cpu(*tagp);
2060         return 0;
2061 }
2062
2063 /*
2064  * Lookup an entry in a node-format directory.
2065  * All the real work happens in xfs_da3_node_lookup_int.
2066  * The only real output is the inode number of the entry.
2067  */
2068 int                                             /* error */
2069 xfs_dir2_node_lookup(
2070         xfs_da_args_t   *args)                  /* operation arguments */
2071 {
2072         int             error;                  /* error return value */
2073         int             i;                      /* btree level */
2074         int             rval;                   /* operation return value */
2075         xfs_da_state_t  *state;                 /* btree cursor */
2076
2077         trace_xfs_dir2_node_lookup(args);
2078
2079         /*
2080          * Allocate and initialize the btree cursor.
2081          */
2082         state = xfs_da_state_alloc();
2083         state->args = args;
2084         state->mp = args->dp->i_mount;
2085         /*
2086          * Fill in the path to the entry in the cursor.
2087          */
2088         error = xfs_da3_node_lookup_int(state, &rval);
2089         if (error)
2090                 rval = error;
2091         else if (rval == -ENOENT && args->cmpresult == XFS_CMP_CASE) {
2092                 /* If a CI match, dup the actual name and return -EEXIST */
2093                 xfs_dir2_data_entry_t   *dep;
2094
2095                 dep = (xfs_dir2_data_entry_t *)
2096                         ((char *)state->extrablk.bp->b_addr +
2097                                                  state->extrablk.index);
2098                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2099         }
2100         /*
2101          * Release the btree blocks and leaf block.
2102          */
2103         for (i = 0; i < state->path.active; i++) {
2104                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2105                 state->path.blk[i].bp = NULL;
2106         }
2107         /*
2108          * Release the data block if we have it.
2109          */
2110         if (state->extravalid && state->extrablk.bp) {
2111                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2112                 state->extrablk.bp = NULL;
2113         }
2114         xfs_da_state_free(state);
2115         return rval;
2116 }
2117
2118 /*
2119  * Remove an entry from a node-format directory.
2120  */
2121 int                                             /* error */
2122 xfs_dir2_node_removename(
2123         struct xfs_da_args      *args)          /* operation arguments */
2124 {
2125         struct xfs_da_state_blk *blk;           /* leaf block */
2126         int                     error;          /* error return value */
2127         int                     rval;           /* operation return value */
2128         struct xfs_da_state     *state;         /* btree cursor */
2129
2130         trace_xfs_dir2_node_removename(args);
2131
2132         /*
2133          * Allocate and initialize the btree cursor.
2134          */
2135         state = xfs_da_state_alloc();
2136         state->args = args;
2137         state->mp = args->dp->i_mount;
2138
2139         /* Look up the entry we're deleting, set up the cursor. */
2140         error = xfs_da3_node_lookup_int(state, &rval);
2141         if (error)
2142                 goto out_free;
2143
2144         /* Didn't find it, upper layer screwed up. */
2145         if (rval != -EEXIST) {
2146                 error = rval;
2147                 goto out_free;
2148         }
2149
2150         blk = &state->path.blk[state->path.active - 1];
2151         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2152         ASSERT(state->extravalid);
2153         /*
2154          * Remove the leaf and data entries.
2155          * Extrablk refers to the data block.
2156          */
2157         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2158                 &state->extrablk, &rval);
2159         if (error)
2160                 goto out_free;
2161         /*
2162          * Fix the hash values up the btree.
2163          */
2164         xfs_da3_fixhashpath(state, &state->path);
2165         /*
2166          * If we need to join leaf blocks, do it.
2167          */
2168         if (rval && state->path.active > 1)
2169                 error = xfs_da3_join(state);
2170         /*
2171          * If no errors so far, try conversion to leaf format.
2172          */
2173         if (!error)
2174                 error = xfs_dir2_node_to_leaf(state);
2175 out_free:
2176         xfs_da_state_free(state);
2177         return error;
2178 }
2179
2180 /*
2181  * Replace an entry's inode number in a node-format directory.
2182  */
2183 int                                             /* error */
2184 xfs_dir2_node_replace(
2185         xfs_da_args_t           *args)          /* operation arguments */
2186 {
2187         xfs_da_state_blk_t      *blk;           /* leaf block */
2188         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
2189         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
2190         int                     error;          /* error return value */
2191         int                     i;              /* btree level */
2192         xfs_ino_t               inum;           /* new inode number */
2193         int                     ftype;          /* new file type */
2194         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
2195         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
2196         int                     rval;           /* internal return value */
2197         xfs_da_state_t          *state;         /* btree cursor */
2198
2199         trace_xfs_dir2_node_replace(args);
2200
2201         /*
2202          * Allocate and initialize the btree cursor.
2203          */
2204         state = xfs_da_state_alloc();
2205         state->args = args;
2206         state->mp = args->dp->i_mount;
2207
2208         /*
2209          * We have to save new inode number and ftype since
2210          * xfs_da3_node_lookup_int() is going to overwrite them
2211          */
2212         inum = args->inumber;
2213         ftype = args->filetype;
2214
2215         /*
2216          * Lookup the entry to change in the btree.
2217          */
2218         error = xfs_da3_node_lookup_int(state, &rval);
2219         if (error) {
2220                 rval = error;
2221         }
2222         /*
2223          * It should be found, since the vnodeops layer has looked it up
2224          * and locked it.  But paranoia is good.
2225          */
2226         if (rval == -EEXIST) {
2227                 struct xfs_dir2_leaf_entry *ents;
2228                 /*
2229                  * Find the leaf entry.
2230                  */
2231                 blk = &state->path.blk[state->path.active - 1];
2232                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2233                 leaf = blk->bp->b_addr;
2234                 ents = args->dp->d_ops->leaf_ents_p(leaf);
2235                 lep = &ents[blk->index];
2236                 ASSERT(state->extravalid);
2237                 /*
2238                  * Point to the data entry.
2239                  */
2240                 hdr = state->extrablk.bp->b_addr;
2241                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2242                        hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2243                 dep = (xfs_dir2_data_entry_t *)
2244                       ((char *)hdr +
2245                        xfs_dir2_dataptr_to_off(args->geo,
2246                                                be32_to_cpu(lep->address)));
2247                 ASSERT(inum != be64_to_cpu(dep->inumber));
2248                 /*
2249                  * Fill in the new inode number and log the entry.
2250                  */
2251                 dep->inumber = cpu_to_be64(inum);
2252                 args->dp->d_ops->data_put_ftype(dep, ftype);
2253                 xfs_dir2_data_log_entry(args, state->extrablk.bp, dep);
2254                 rval = 0;
2255         }
2256         /*
2257          * Didn't find it, and we're holding a data block.  Drop it.
2258          */
2259         else if (state->extravalid) {
2260                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2261                 state->extrablk.bp = NULL;
2262         }
2263         /*
2264          * Release all the buffers in the cursor.
2265          */
2266         for (i = 0; i < state->path.active; i++) {
2267                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2268                 state->path.blk[i].bp = NULL;
2269         }
2270         xfs_da_state_free(state);
2271         return rval;
2272 }
2273
2274 /*
2275  * Trim off a trailing empty freespace block.
2276  * Return (in rvalp) 1 if we did it, 0 if not.
2277  */
2278 int                                             /* error */
2279 xfs_dir2_node_trim_free(
2280         xfs_da_args_t           *args,          /* operation arguments */
2281         xfs_fileoff_t           fo,             /* free block number */
2282         int                     *rvalp)         /* out: did something */
2283 {
2284         struct xfs_buf          *bp;            /* freespace buffer */
2285         xfs_inode_t             *dp;            /* incore directory inode */
2286         int                     error;          /* error return code */
2287         xfs_dir2_free_t         *free;          /* freespace structure */
2288         xfs_trans_t             *tp;            /* transaction pointer */
2289         struct xfs_dir3_icfree_hdr freehdr;
2290
2291         dp = args->dp;
2292         tp = args->trans;
2293
2294         *rvalp = 0;
2295
2296         /*
2297          * Read the freespace block.
2298          */
2299         error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2300         if (error)
2301                 return error;
2302         /*
2303          * There can be holes in freespace.  If fo is a hole, there's
2304          * nothing to do.
2305          */
2306         if (!bp)
2307                 return 0;
2308         free = bp->b_addr;
2309         dp->d_ops->free_hdr_from_disk(&freehdr, free);
2310
2311         /*
2312          * If there are used entries, there's nothing to do.
2313          */
2314         if (freehdr.nused > 0) {
2315                 xfs_trans_brelse(tp, bp);
2316                 return 0;
2317         }
2318         /*
2319          * Blow the block away.
2320          */
2321         error = xfs_dir2_shrink_inode(args,
2322                         xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo), bp);
2323         if (error) {
2324                 /*
2325                  * Can't fail with ENOSPC since that only happens with no
2326                  * space reservation, when breaking up an extent into two
2327                  * pieces.  This is the last block of an extent.
2328                  */
2329                 ASSERT(error != -ENOSPC);
2330                 xfs_trans_brelse(tp, bp);
2331                 return error;
2332         }
2333         /*
2334          * Return that we succeeded.
2335          */
2336         *rvalp = 1;
2337         return 0;
2338 }