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