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_shared.h"
  10 #include "xfs_format.h"
  11 #include "xfs_log_format.h"
  12 #include "xfs_trans_resv.h"
  13 #include "xfs_mount.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_bmap.h"
  16 #include "xfs_dir2.h"
  17 #include "xfs_dir2_priv.h"
  18 #include "xfs_error.h"
  19 #include "xfs_trace.h"
  20 #include "xfs_trans.h"
  21 #include "xfs_buf_item.h"
  22 #include "xfs_log.h"
  23
  24 /*
  25  * Function declarations.
  26  */
  27 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
  28                               int index);
  29 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  30                                      xfs_da_state_blk_t *blk1,
  31                                      xfs_da_state_blk_t *blk2);
  32 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
  33                                  int index, xfs_da_state_blk_t *dblk,
  34                                  int *rval);
  35
  36 /*
  37  * Convert data space db to the corresponding free db.
  38  */
  39 static xfs_dir2_db_t
  40 xfs_dir2_db_to_fdb(struct xfs_da_geometry *geo, xfs_dir2_db_t db)
  41 {
  42         return xfs_dir2_byte_to_db(geo, XFS_DIR2_FREE_OFFSET) +
  43                         (db / geo->free_max_bests);
  44 }
  45
  46 /*
  47  * Convert data space db to the corresponding index in a free db.
  48  */
  49 static int
  50 xfs_dir2_db_to_fdindex(struct xfs_da_geometry *geo, xfs_dir2_db_t db)
  51 {
  52         return db % geo->free_max_bests;
  53 }
  54
  55 /*
  56  * Check internal consistency of a leafn block.
  57  */
  58 #ifdef DEBUG
  59 static xfs_failaddr_t
  60 xfs_dir3_leafn_check(
  61         struct xfs_inode        *dp,
  62         struct xfs_buf          *bp)
  63 {
  64         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  65         struct xfs_dir3_icleaf_hdr leafhdr;
  66
  67         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
  68
  69         if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
  70                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  71                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  72                         return __this_address;
  73         } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
  74                 return __this_address;
  75
  76         return xfs_dir3_leaf_check_int(dp->i_mount, &leafhdr, leaf, false);
  77 }
  78
  79 static inline void
  80 xfs_dir3_leaf_check(
  81         struct xfs_inode        *dp,
  82         struct xfs_buf          *bp)
  83 {
  84         xfs_failaddr_t          fa;
  85
  86         fa = xfs_dir3_leafn_check(dp, bp);
  87         if (!fa)
  88                 return;
  89         xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
  90                         bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
  91                         fa);
  92         ASSERT(0);
  93 }
  94 #else
  95 #define xfs_dir3_leaf_check(dp, bp)
  96 #endif
  97
  98 static xfs_failaddr_t
  99 xfs_dir3_free_verify(
 100         struct xfs_buf          *bp)
 101 {
 102         struct xfs_mount        *mp = bp->b_mount;
 103         struct xfs_dir2_free_hdr *hdr = bp->b_addr;
 104
 105         if (!xfs_verify_magic(bp, hdr->magic))
 106                 return __this_address;
 107
 108         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 109                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 110
 111                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
 112                         return __this_address;
 113                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
 114                         return __this_address;
 115                 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
 116                         return __this_address;
 117         }
 118
 119         /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
 120
 121         return NULL;
 122 }
 123
 124 static void
 125 xfs_dir3_free_read_verify(
 126         struct xfs_buf  *bp)
 127 {
 128         struct xfs_mount        *mp = bp->b_mount;
 129         xfs_failaddr_t          fa;
 130
 131         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 132             !xfs_buf_verify_cksum(bp, XFS_DIR3_FREE_CRC_OFF))
 133                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
 134         else {
 135                 fa = xfs_dir3_free_verify(bp);
 136                 if (fa)
 137                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 138         }
 139 }
 140
 141 static void
 142 xfs_dir3_free_write_verify(
 143         struct xfs_buf  *bp)
 144 {
 145         struct xfs_mount        *mp = bp->b_mount;
 146         struct xfs_buf_log_item *bip = bp->b_log_item;
 147         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 148         xfs_failaddr_t          fa;
 149
 150         fa = xfs_dir3_free_verify(bp);
 151         if (fa) {
 152                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 153                 return;
 154         }
 155
 156         if (!xfs_sb_version_hascrc(&mp->m_sb))
 157                 return;
 158
 159         if (bip)
 160                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 161
 162         xfs_buf_update_cksum(bp, XFS_DIR3_FREE_CRC_OFF);
 163 }
 164
 165 const struct xfs_buf_ops xfs_dir3_free_buf_ops = {
 166         .name = "xfs_dir3_free",
 167         .magic = { cpu_to_be32(XFS_DIR2_FREE_MAGIC),
 168                    cpu_to_be32(XFS_DIR3_FREE_MAGIC) },
 169         .verify_read = xfs_dir3_free_read_verify,
 170         .verify_write = xfs_dir3_free_write_verify,
 171         .verify_struct = xfs_dir3_free_verify,
 172 };
 173
 174 /* Everything ok in the free block header? */
 175 static xfs_failaddr_t
 176 xfs_dir3_free_header_check(
 177         struct xfs_inode        *dp,
 178         xfs_dablk_t             fbno,
 179         struct xfs_buf          *bp)
 180 {
 181         struct xfs_mount        *mp = dp->i_mount;
 182         int                     maxbests = mp->m_dir_geo->free_max_bests;
 183         unsigned int            firstdb;
 184
 185         firstdb = (xfs_dir2_da_to_db(mp->m_dir_geo, fbno) -
 186                    xfs_dir2_byte_to_db(mp->m_dir_geo, XFS_DIR2_FREE_OFFSET)) *
 187                         maxbests;
 188         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 189                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 190
 191                 if (be32_to_cpu(hdr3->firstdb) != firstdb)
 192                         return __this_address;
 193                 if (be32_to_cpu(hdr3->nvalid) > maxbests)
 194                         return __this_address;
 195                 if (be32_to_cpu(hdr3->nvalid) < be32_to_cpu(hdr3->nused))
 196                         return __this_address;
 197                 if (be64_to_cpu(hdr3->hdr.owner) != dp->i_ino)
 198                         return __this_address;
 199         } else {
 200                 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
 201
 202                 if (be32_to_cpu(hdr->firstdb) != firstdb)
 203                         return __this_address;
 204                 if (be32_to_cpu(hdr->nvalid) > maxbests)
 205                         return __this_address;
 206                 if (be32_to_cpu(hdr->nvalid) < be32_to_cpu(hdr->nused))
 207                         return __this_address;
 208         }
 209         return NULL;
 210 }
 211
 212 static int
 213 __xfs_dir3_free_read(
 214         struct xfs_trans        *tp,
 215         struct xfs_inode        *dp,
 216         xfs_dablk_t             fbno,
 217         unsigned int            flags,
 218         struct xfs_buf          **bpp)
 219 {
 220         xfs_failaddr_t          fa;
 221         int                     err;
 222
 223         err = xfs_da_read_buf(tp, dp, fbno, flags, bpp, XFS_DATA_FORK,
 224                         &xfs_dir3_free_buf_ops);
 225         if (err || !*bpp)
 226                 return err;
 227
 228         /* Check things that we can't do in the verifier. */
 229         fa = xfs_dir3_free_header_check(dp, fbno, *bpp);
 230         if (fa) {
 231                 __xfs_buf_mark_corrupt(*bpp, fa);
 232                 xfs_trans_brelse(tp, *bpp);
 233                 *bpp = NULL;
 234                 return -EFSCORRUPTED;
 235         }
 236
 237         /* try read returns without an error or *bpp if it lands in a hole */
 238         if (tp)
 239                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
 240
 241         return 0;
 242 }
 243
 244 void
 245 xfs_dir2_free_hdr_from_disk(
 246         struct xfs_mount                *mp,
 247         struct xfs_dir3_icfree_hdr      *to,
 248         struct xfs_dir2_free            *from)
 249 {
 250         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 251                 struct xfs_dir3_free    *from3 = (struct xfs_dir3_free *)from;
 252
 253                 to->magic = be32_to_cpu(from3->hdr.hdr.magic);
 254                 to->firstdb = be32_to_cpu(from3->hdr.firstdb);
 255                 to->nvalid = be32_to_cpu(from3->hdr.nvalid);
 256                 to->nused = be32_to_cpu(from3->hdr.nused);
 257                 to->bests = from3->bests;
 258
 259                 ASSERT(to->magic == XFS_DIR3_FREE_MAGIC);
 260         } else {
 261                 to->magic = be32_to_cpu(from->hdr.magic);
 262                 to->firstdb = be32_to_cpu(from->hdr.firstdb);
 263                 to->nvalid = be32_to_cpu(from->hdr.nvalid);
 264                 to->nused = be32_to_cpu(from->hdr.nused);
 265                 to->bests = from->bests;
 266
 267                 ASSERT(to->magic == XFS_DIR2_FREE_MAGIC);
 268         }
 269 }
 270
 271 static void
 272 xfs_dir2_free_hdr_to_disk(
 273         struct xfs_mount                *mp,
 274         struct xfs_dir2_free            *to,
 275         struct xfs_dir3_icfree_hdr      *from)
 276 {
 277         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 278                 struct xfs_dir3_free    *to3 = (struct xfs_dir3_free *)to;
 279
 280                 ASSERT(from->magic == XFS_DIR3_FREE_MAGIC);
 281
 282                 to3->hdr.hdr.magic = cpu_to_be32(from->magic);
 283                 to3->hdr.firstdb = cpu_to_be32(from->firstdb);
 284                 to3->hdr.nvalid = cpu_to_be32(from->nvalid);
 285                 to3->hdr.nused = cpu_to_be32(from->nused);
 286         } else {
 287                 ASSERT(from->magic == XFS_DIR2_FREE_MAGIC);
 288
 289                 to->hdr.magic = cpu_to_be32(from->magic);
 290                 to->hdr.firstdb = cpu_to_be32(from->firstdb);
 291                 to->hdr.nvalid = cpu_to_be32(from->nvalid);
 292                 to->hdr.nused = cpu_to_be32(from->nused);
 293         }
 294 }
 295
 296 int
 297 xfs_dir2_free_read(
 298         struct xfs_trans        *tp,
 299         struct xfs_inode        *dp,
 300         xfs_dablk_t             fbno,
 301         struct xfs_buf          **bpp)
 302 {
 303         return __xfs_dir3_free_read(tp, dp, fbno, 0, bpp);
 304 }
 305
 306 static int
 307 xfs_dir2_free_try_read(
 308         struct xfs_trans        *tp,
 309         struct xfs_inode        *dp,
 310         xfs_dablk_t             fbno,
 311         struct xfs_buf          **bpp)
 312 {
 313         return __xfs_dir3_free_read(tp, dp, fbno, XFS_DABUF_MAP_HOLE_OK, bpp);
 314 }
 315
 316 static int
 317 xfs_dir3_free_get_buf(
 318         xfs_da_args_t           *args,
 319         xfs_dir2_db_t           fbno,
 320         struct xfs_buf          **bpp)
 321 {
 322         struct xfs_trans        *tp = args->trans;
 323         struct xfs_inode        *dp = args->dp;
 324         struct xfs_mount        *mp = dp->i_mount;
 325         struct xfs_buf          *bp;
 326         int                     error;
 327         struct xfs_dir3_icfree_hdr hdr;
 328
 329         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, fbno),
 330                         &bp, XFS_DATA_FORK);
 331         if (error)
 332                 return error;
 333
 334         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
 335         bp->b_ops = &xfs_dir3_free_buf_ops;
 336
 337         /*
 338          * Initialize the new block to be empty, and remember
 339          * its first slot as our empty slot.
 340          */
 341         memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
 342         memset(&hdr, 0, sizeof(hdr));
 343
 344         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 345                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 346
 347                 hdr.magic = XFS_DIR3_FREE_MAGIC;
 348
 349                 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
 350                 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
 351                 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_meta_uuid);
 352         } else
 353                 hdr.magic = XFS_DIR2_FREE_MAGIC;
 354         xfs_dir2_free_hdr_to_disk(mp, bp->b_addr, &hdr);
 355         *bpp = bp;
 356         return 0;
 357 }
 358
 359 /*
 360  * Log entries from a freespace block.
 361  */
 362 STATIC void
 363 xfs_dir2_free_log_bests(
 364         struct xfs_da_args      *args,
 365         struct xfs_dir3_icfree_hdr *hdr,
 366         struct xfs_buf          *bp,
 367         int                     first,          /* first entry to log */
 368         int                     last)           /* last entry to log */
 369 {
 370         struct xfs_dir2_free    *free = bp->b_addr;
 371
 372         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 373                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 374         xfs_trans_log_buf(args->trans, bp,
 375                           (char *)&hdr->bests[first] - (char *)free,
 376                           (char *)&hdr->bests[last] - (char *)free +
 377                            sizeof(hdr->bests[0]) - 1);
 378 }
 379
 380 /*
 381  * Log header from a freespace block.
 382  */
 383 static void
 384 xfs_dir2_free_log_header(
 385         struct xfs_da_args      *args,
 386         struct xfs_buf          *bp)
 387 {
 388 #ifdef DEBUG
 389         xfs_dir2_free_t         *free;          /* freespace structure */
 390
 391         free = bp->b_addr;
 392         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 393                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 394 #endif
 395         xfs_trans_log_buf(args->trans, bp, 0,
 396                           args->geo->free_hdr_size - 1);
 397 }
 398
 399 /*
 400  * Convert a leaf-format directory to a node-format directory.
 401  * We need to change the magic number of the leaf block, and copy
 402  * the freespace table out of the leaf block into its own block.
 403  */
 404 int                                             /* error */
 405 xfs_dir2_leaf_to_node(
 406         xfs_da_args_t           *args,          /* operation arguments */
 407         struct xfs_buf          *lbp)           /* leaf buffer */
 408 {
 409         xfs_inode_t             *dp;            /* incore directory inode */
 410         int                     error;          /* error return value */
 411         struct xfs_buf          *fbp;           /* freespace buffer */
 412         xfs_dir2_db_t           fdb;            /* freespace block number */
 413         __be16                  *from;          /* pointer to freespace entry */
 414         int                     i;              /* leaf freespace index */
 415         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 416         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 417         int                     n;              /* count of live freespc ents */
 418         xfs_dir2_data_off_t     off;            /* freespace entry value */
 419         xfs_trans_t             *tp;            /* transaction pointer */
 420         struct xfs_dir3_icfree_hdr freehdr;
 421
 422         trace_xfs_dir2_leaf_to_node(args);
 423
 424         dp = args->dp;
 425         tp = args->trans;
 426         /*
 427          * Add a freespace block to the directory.
 428          */
 429         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 430                 return error;
 431         }
 432         ASSERT(fdb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
 433         /*
 434          * Get the buffer for the new freespace block.
 435          */
 436         error = xfs_dir3_free_get_buf(args, fdb, &fbp);
 437         if (error)
 438                 return error;
 439
 440         xfs_dir2_free_hdr_from_disk(dp->i_mount, &freehdr, fbp->b_addr);
 441         leaf = lbp->b_addr;
 442         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 443         if (be32_to_cpu(ltp->bestcount) >
 444                                 (uint)dp->i_disk_size / args->geo->blksize) {
 445                 xfs_buf_mark_corrupt(lbp);
 446                 return -EFSCORRUPTED;
 447         }
 448
 449         /*
 450          * Copy freespace entries from the leaf block to the new block.
 451          * Count active entries.
 452          */
 453         from = xfs_dir2_leaf_bests_p(ltp);
 454         for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++) {
 455                 off = be16_to_cpu(*from);
 456                 if (off != NULLDATAOFF)
 457                         n++;
 458                 freehdr.bests[i] = cpu_to_be16(off);
 459         }
 460
 461         /*
 462          * Now initialize the freespace block header.
 463          */
 464         freehdr.nused = n;
 465         freehdr.nvalid = be32_to_cpu(ltp->bestcount);
 466
 467         xfs_dir2_free_hdr_to_disk(dp->i_mount, fbp->b_addr, &freehdr);
 468         xfs_dir2_free_log_bests(args, &freehdr, fbp, 0, freehdr.nvalid - 1);
 469         xfs_dir2_free_log_header(args, fbp);
 470
 471         /*
 472          * Converting the leaf to a leafnode is just a matter of changing the
 473          * magic number and the ops. Do the change directly to the buffer as
 474          * it's less work (and less code) than decoding the header to host
 475          * format and back again.
 476          */
 477         if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
 478                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 479         else
 480                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 481         lbp->b_ops = &xfs_dir3_leafn_buf_ops;
 482         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
 483         xfs_dir3_leaf_log_header(args, lbp);
 484         xfs_dir3_leaf_check(dp, lbp);
 485         return 0;
 486 }
 487
 488 /*
 489  * Add a leaf entry to a leaf block in a node-form directory.
 490  * The other work necessary is done from the caller.
 491  */
 492 static int                                      /* error */
 493 xfs_dir2_leafn_add(
 494         struct xfs_buf          *bp,            /* leaf buffer */
 495         struct xfs_da_args      *args,          /* operation arguments */
 496         int                     index)          /* insertion pt for new entry */
 497 {
 498         struct xfs_dir3_icleaf_hdr leafhdr;
 499         struct xfs_inode        *dp = args->dp;
 500         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 501         struct xfs_dir2_leaf_entry *lep;
 502         struct xfs_dir2_leaf_entry *ents;
 503         int                     compact;        /* compacting stale leaves */
 504         int                     highstale = 0;  /* next stale entry */
 505         int                     lfloghigh;      /* high leaf entry logging */
 506         int                     lfloglow;       /* low leaf entry logging */
 507         int                     lowstale = 0;   /* previous stale entry */
 508
 509         trace_xfs_dir2_leafn_add(args, index);
 510
 511         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
 512         ents = leafhdr.ents;
 513
 514         /*
 515          * Quick check just to make sure we are not going to index
 516          * into other peoples memory
 517          */
 518         if (index < 0) {
 519                 xfs_buf_mark_corrupt(bp);
 520                 return -EFSCORRUPTED;
 521         }
 522
 523         /*
 524          * If there are already the maximum number of leaf entries in
 525          * the block, if there are no stale entries it won't fit.
 526          * Caller will do a split.  If there are stale entries we'll do
 527          * a compact.
 528          */
 529
 530         if (leafhdr.count == args->geo->leaf_max_ents) {
 531                 if (!leafhdr.stale)
 532                         return -ENOSPC;
 533                 compact = leafhdr.stale > 1;
 534         } else
 535                 compact = 0;
 536         ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
 537         ASSERT(index == leafhdr.count ||
 538                be32_to_cpu(ents[index].hashval) >= args->hashval);
 539
 540         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 541                 return 0;
 542
 543         /*
 544          * Compact out all but one stale leaf entry.  Leaves behind
 545          * the entry closest to index.
 546          */
 547         if (compact)
 548                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 549                                          &highstale, &lfloglow, &lfloghigh);
 550         else if (leafhdr.stale) {
 551                 /*
 552                  * Set impossible logging indices for this case.
 553                  */
 554                 lfloglow = leafhdr.count;
 555                 lfloghigh = -1;
 556         }
 557
 558         /*
 559          * Insert the new entry, log everything.
 560          */
 561         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 562                                        highstale, &lfloglow, &lfloghigh);
 563
 564         lep->hashval = cpu_to_be32(args->hashval);
 565         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(args->geo,
 566                                 args->blkno, args->index));
 567
 568         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
 569         xfs_dir3_leaf_log_header(args, bp);
 570         xfs_dir3_leaf_log_ents(args, &leafhdr, bp, lfloglow, lfloghigh);
 571         xfs_dir3_leaf_check(dp, bp);
 572         return 0;
 573 }
 574
 575 #ifdef DEBUG
 576 static void
 577 xfs_dir2_free_hdr_check(
 578         struct xfs_inode *dp,
 579         struct xfs_buf  *bp,
 580         xfs_dir2_db_t   db)
 581 {
 582         struct xfs_dir3_icfree_hdr hdr;
 583
 584         xfs_dir2_free_hdr_from_disk(dp->i_mount, &hdr, bp->b_addr);
 585
 586         ASSERT((hdr.firstdb % dp->i_mount->m_dir_geo->free_max_bests) == 0);
 587         ASSERT(hdr.firstdb <= db);
 588         ASSERT(db < hdr.firstdb + hdr.nvalid);
 589 }
 590 #else
 591 #define xfs_dir2_free_hdr_check(dp, bp, db)
 592 #endif  /* DEBUG */
 593
 594 /*
 595  * Return the last hash value in the leaf.
 596  * Stale entries are ok.
 597  */
 598 xfs_dahash_t                                    /* hash value */
 599 xfs_dir2_leaf_lasthash(
 600         struct xfs_inode *dp,
 601         struct xfs_buf  *bp,                    /* leaf buffer */
 602         int             *count)                 /* count of entries in leaf */
 603 {
 604         struct xfs_dir3_icleaf_hdr leafhdr;
 605
 606         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, bp->b_addr);
 607
 608         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 609                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC ||
 610                leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
 611                leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
 612
 613         if (count)
 614                 *count = leafhdr.count;
 615         if (!leafhdr.count)
 616                 return 0;
 617         return be32_to_cpu(leafhdr.ents[leafhdr.count - 1].hashval);
 618 }
 619
 620 /*
 621  * Look up a leaf entry for space to add a name in a node-format leaf block.
 622  * The extrablk in state is a freespace block.
 623  */
 624 STATIC int
 625 xfs_dir2_leafn_lookup_for_addname(
 626         struct xfs_buf          *bp,            /* leaf buffer */
 627         xfs_da_args_t           *args,          /* operation arguments */
 628         int                     *indexp,        /* out: leaf entry index */
 629         xfs_da_state_t          *state)         /* state to fill in */
 630 {
 631         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 632         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 633         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 634         xfs_inode_t             *dp;            /* incore directory inode */
 635         int                     error;          /* error return value */
 636         int                     fi;             /* free entry index */
 637         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 638         int                     index;          /* leaf entry index */
 639         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 640         int                     length;         /* length of new data entry */
 641         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 642         xfs_mount_t             *mp;            /* filesystem mount point */
 643         xfs_dir2_db_t           newdb;          /* new data block number */
 644         xfs_dir2_db_t           newfdb;         /* new free block number */
 645         xfs_trans_t             *tp;            /* transaction pointer */
 646         struct xfs_dir3_icleaf_hdr leafhdr;
 647
 648         dp = args->dp;
 649         tp = args->trans;
 650         mp = dp->i_mount;
 651         leaf = bp->b_addr;
 652         xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, leaf);
 653
 654         xfs_dir3_leaf_check(dp, bp);
 655         ASSERT(leafhdr.count > 0);
 656
 657         /*
 658          * Look up the hash value in the leaf entries.
 659          */
 660         index = xfs_dir2_leaf_search_hash(args, bp);
 661         /*
 662          * Do we have a buffer coming in?
 663          */
 664         if (state->extravalid) {
 665                 /* If so, it's a free block buffer, get the block number. */
 666                 curbp = state->extrablk.bp;
 667                 curfdb = state->extrablk.blkno;
 668                 free = curbp->b_addr;
 669                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 670                        free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 671         }
 672         length = xfs_dir2_data_entsize(mp, args->namelen);
 673         /*
 674          * Loop over leaf entries with the right hash value.
 675          */
 676         for (lep = &leafhdr.ents[index];
 677              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 678              lep++, index++) {
 679                 /*
 680                  * Skip stale leaf entries.
 681                  */
 682                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 683                         continue;
 684                 /*
 685                  * Pull the data block number from the entry.
 686                  */
 687                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 688                                                be32_to_cpu(lep->address));
 689                 /*
 690                  * For addname, we're looking for a place to put the new entry.
 691                  * We want to use a data block with an entry of equal
 692                  * hash value to ours if there is one with room.
 693                  *
 694                  * If this block isn't the data block we already have
 695                  * in hand, take a look at it.
 696                  */
 697                 if (newdb != curdb) {
 698                         struct xfs_dir3_icfree_hdr freehdr;
 699
 700                         curdb = newdb;
 701                         /*
 702                          * Convert the data block to the free block
 703                          * holding its freespace information.
 704                          */
 705                         newfdb = xfs_dir2_db_to_fdb(args->geo, newdb);
 706                         /*
 707                          * If it's not the one we have in hand, read it in.
 708                          */
 709                         if (newfdb != curfdb) {
 710                                 /*
 711                                  * If we had one before, drop it.
 712                                  */
 713                                 if (curbp)
 714                                         xfs_trans_brelse(tp, curbp);
 715
 716                                 error = xfs_dir2_free_read(tp, dp,
 717                                                 xfs_dir2_db_to_da(args->geo,
 718                                                                   newfdb),
 719                                                 &curbp);
 720                                 if (error)
 721                                         return error;
 722                                 free = curbp->b_addr;
 723
 724                                 xfs_dir2_free_hdr_check(dp, curbp, curdb);
 725                         }
 726                         /*
 727                          * Get the index for our entry.
 728                          */
 729                         fi = xfs_dir2_db_to_fdindex(args->geo, curdb);
 730                         /*
 731                          * If it has room, return it.
 732                          */
 733                         xfs_dir2_free_hdr_from_disk(mp, &freehdr, free);
 734                         if (XFS_IS_CORRUPT(mp,
 735                                            freehdr.bests[fi] ==
 736                                            cpu_to_be16(NULLDATAOFF))) {
 737                                 if (curfdb != newfdb)
 738                                         xfs_trans_brelse(tp, curbp);
 739                                 return -EFSCORRUPTED;
 740                         }
 741                         curfdb = newfdb;
 742                         if (be16_to_cpu(freehdr.bests[fi]) >= length)
 743                                 goto out;
 744                 }
 745         }
 746         /* Didn't find any space */
 747         fi = -1;
 748 out:
 749         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 750         if (curbp) {
 751                 /* Giving back a free block. */
 752                 state->extravalid = 1;
 753                 state->extrablk.bp = curbp;
 754                 state->extrablk.index = fi;
 755                 state->extrablk.blkno = curfdb;
 756
 757                 /*
 758                  * Important: this magic number is not in the buffer - it's for
 759                  * buffer type information and therefore only the free/data type
 760                  * matters here, not whether CRCs are enabled or not.
 761                  */
 762                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 763         } else {
 764                 state->extravalid = 0;
 765         }
 766         /*
 767          * Return the index, that will be the insertion point.
 768          */
 769         *indexp = index;
 770         return -ENOENT;
 771 }
 772
 773 /*
 774  * Look up a leaf entry in a node-format leaf block.
 775  * The extrablk in state a data block.
 776  */
 777 STATIC int
 778 xfs_dir2_leafn_lookup_for_entry(
 779         struct xfs_buf          *bp,            /* leaf buffer */
 780         xfs_da_args_t           *args,          /* operation arguments */
 781         int                     *indexp,        /* out: leaf entry index */
 782         xfs_da_state_t          *state)         /* state to fill in */
 783 {
 784         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 785         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 786         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 787         xfs_inode_t             *dp;            /* incore directory inode */
 788         int                     error;          /* error return value */
 789         int                     index;          /* leaf entry index */
 790         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 791         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 792         xfs_mount_t             *mp;            /* filesystem mount point */
 793         xfs_dir2_db_t           newdb;          /* new data block number */
 794         xfs_trans_t             *tp;            /* transaction pointer */
 795         enum xfs_dacmp          cmp;            /* comparison result */
 796         struct xfs_dir3_icleaf_hdr leafhdr;
 797
 798         dp = args->dp;
 799         tp = args->trans;
 800         mp = dp->i_mount;
 801         leaf = bp->b_addr;
 802         xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, leaf);
 803
 804         xfs_dir3_leaf_check(dp, bp);
 805         if (leafhdr.count <= 0) {
 806                 xfs_buf_mark_corrupt(bp);
 807                 return -EFSCORRUPTED;
 808         }
 809
 810         /*
 811          * Look up the hash value in the leaf entries.
 812          */
 813         index = xfs_dir2_leaf_search_hash(args, bp);
 814         /*
 815          * Do we have a buffer coming in?
 816          */
 817         if (state->extravalid) {
 818                 curbp = state->extrablk.bp;
 819                 curdb = state->extrablk.blkno;
 820         }
 821         /*
 822          * Loop over leaf entries with the right hash value.
 823          */
 824         for (lep = &leafhdr.ents[index];
 825              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 826              lep++, index++) {
 827                 /*
 828                  * Skip stale leaf entries.
 829                  */
 830                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 831                         continue;
 832                 /*
 833                  * Pull the data block number from the entry.
 834                  */
 835                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 836                                                be32_to_cpu(lep->address));
 837                 /*
 838                  * Not adding a new entry, so we really want to find
 839                  * the name given to us.
 840                  *
 841                  * If it's a different data block, go get it.
 842                  */
 843                 if (newdb != curdb) {
 844                         /*
 845                          * If we had a block before that we aren't saving
 846                          * for a CI name, drop it
 847                          */
 848                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 849                                                 curdb != state->extrablk.blkno))
 850                                 xfs_trans_brelse(tp, curbp);
 851                         /*
 852                          * If needing the block that is saved with a CI match,
 853                          * use it otherwise read in the new data block.
 854                          */
 855                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 856                                         newdb == state->extrablk.blkno) {
 857                                 ASSERT(state->extravalid);
 858                                 curbp = state->extrablk.bp;
 859                         } else {
 860                                 error = xfs_dir3_data_read(tp, dp,
 861                                                 xfs_dir2_db_to_da(args->geo,
 862                                                                   newdb),
 863                                                 0, &curbp);
 864                                 if (error)
 865                                         return error;
 866                         }
 867                         xfs_dir3_data_check(dp, curbp);
 868                         curdb = newdb;
 869                 }
 870                 /*
 871                  * Point to the data entry.
 872                  */
 873                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 874                         xfs_dir2_dataptr_to_off(args->geo,
 875                                                 be32_to_cpu(lep->address)));
 876                 /*
 877                  * Compare the entry and if it's an exact match, return
 878                  * EEXIST immediately. If it's the first case-insensitive
 879                  * match, store the block & inode number and continue looking.
 880                  */
 881                 cmp = xfs_dir2_compname(args, dep->name, dep->namelen);
 882                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 883                         /* If there is a CI match block, drop it */
 884                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 885                                                 curdb != state->extrablk.blkno)
 886                                 xfs_trans_brelse(tp, state->extrablk.bp);
 887                         args->cmpresult = cmp;
 888                         args->inumber = be64_to_cpu(dep->inumber);
 889                         args->filetype = xfs_dir2_data_get_ftype(mp, dep);
 890                         *indexp = index;
 891                         state->extravalid = 1;
 892                         state->extrablk.bp = curbp;
 893                         state->extrablk.blkno = curdb;
 894                         state->extrablk.index = (int)((char *)dep -
 895                                                         (char *)curbp->b_addr);
 896                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 897                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 898                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 899                         if (cmp == XFS_CMP_EXACT)
 900                                 return -EEXIST;
 901                 }
 902         }
 903         ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
 904         if (curbp) {
 905                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 906                         /* Giving back last used data block. */
 907                         state->extravalid = 1;
 908                         state->extrablk.bp = curbp;
 909                         state->extrablk.index = -1;
 910                         state->extrablk.blkno = curdb;
 911                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 912                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 913                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 914                 } else {
 915                         /* If the curbp is not the CI match block, drop it */
 916                         if (state->extrablk.bp != curbp)
 917                                 xfs_trans_brelse(tp, curbp);
 918                 }
 919         } else {
 920                 state->extravalid = 0;
 921         }
 922         *indexp = index;
 923         return -ENOENT;
 924 }
 925
 926 /*
 927  * Look up a leaf entry in a node-format leaf block.
 928  * If this is an addname then the extrablk in state is a freespace block,
 929  * otherwise it's a data block.
 930  */
 931 int
 932 xfs_dir2_leafn_lookup_int(
 933         struct xfs_buf          *bp,            /* leaf buffer */
 934         xfs_da_args_t           *args,          /* operation arguments */
 935         int                     *indexp,        /* out: leaf entry index */
 936         xfs_da_state_t          *state)         /* state to fill in */
 937 {
 938         if (args->op_flags & XFS_DA_OP_ADDNAME)
 939                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 940                                                         state);
 941         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 942 }
 943
 944 /*
 945  * Move count leaf entries from source to destination leaf.
 946  * Log entries and headers.  Stale entries are preserved.
 947  */
 948 static void
 949 xfs_dir3_leafn_moveents(
 950         xfs_da_args_t                   *args,  /* operation arguments */
 951         struct xfs_buf                  *bp_s,  /* source */
 952         struct xfs_dir3_icleaf_hdr      *shdr,
 953         struct xfs_dir2_leaf_entry      *sents,
 954         int                             start_s,/* source leaf index */
 955         struct xfs_buf                  *bp_d,  /* destination */
 956         struct xfs_dir3_icleaf_hdr      *dhdr,
 957         struct xfs_dir2_leaf_entry      *dents,
 958         int                             start_d,/* destination leaf index */
 959         int                             count)  /* count of leaves to copy */
 960 {
 961         int                             stale;  /* count stale leaves copied */
 962
 963         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 964
 965         /*
 966          * Silently return if nothing to do.
 967          */
 968         if (count == 0)
 969                 return;
 970
 971         /*
 972          * If the destination index is not the end of the current
 973          * destination leaf entries, open up a hole in the destination
 974          * to hold the new entries.
 975          */
 976         if (start_d < dhdr->count) {
 977                 memmove(&dents[start_d + count], &dents[start_d],
 978                         (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
 979                 xfs_dir3_leaf_log_ents(args, dhdr, bp_d, start_d + count,
 980                                        count + dhdr->count - 1);
 981         }
 982         /*
 983          * If the source has stale leaves, count the ones in the copy range
 984          * so we can update the header correctly.
 985          */
 986         if (shdr->stale) {
 987                 int     i;                      /* temp leaf index */
 988
 989                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 990                         if (sents[i].address ==
 991                                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 992                                 stale++;
 993                 }
 994         } else
 995                 stale = 0;
 996         /*
 997          * Copy the leaf entries from source to destination.
 998          */
 999         memcpy(&dents[start_d], &sents[start_s],
1000                 count * sizeof(xfs_dir2_leaf_entry_t));
1001         xfs_dir3_leaf_log_ents(args, dhdr, bp_d, start_d, start_d + count - 1);
1002
1003         /*
1004          * If there are source entries after the ones we copied,
1005          * delete the ones we copied by sliding the next ones down.
1006          */
1007         if (start_s + count < shdr->count) {
1008                 memmove(&sents[start_s], &sents[start_s + count],
1009                         count * sizeof(xfs_dir2_leaf_entry_t));
1010                 xfs_dir3_leaf_log_ents(args, shdr, bp_s, start_s,
1011                                        start_s + count - 1);
1012         }
1013
1014         /*
1015          * Update the headers and log them.
1016          */
1017         shdr->count -= count;
1018         shdr->stale -= stale;
1019         dhdr->count += count;
1020         dhdr->stale += stale;
1021 }
1022
1023 /*
1024  * Determine the sort order of two leaf blocks.
1025  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
1026  */
1027 int                                             /* sort order */
1028 xfs_dir2_leafn_order(
1029         struct xfs_inode        *dp,
1030         struct xfs_buf          *leaf1_bp,              /* leaf1 buffer */
1031         struct xfs_buf          *leaf2_bp)              /* leaf2 buffer */
1032 {
1033         struct xfs_dir2_leaf    *leaf1 = leaf1_bp->b_addr;
1034         struct xfs_dir2_leaf    *leaf2 = leaf2_bp->b_addr;
1035         struct xfs_dir2_leaf_entry *ents1;
1036         struct xfs_dir2_leaf_entry *ents2;
1037         struct xfs_dir3_icleaf_hdr hdr1;
1038         struct xfs_dir3_icleaf_hdr hdr2;
1039
1040         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &hdr1, leaf1);
1041         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &hdr2, leaf2);
1042         ents1 = hdr1.ents;
1043         ents2 = hdr2.ents;
1044
1045         if (hdr1.count > 0 && hdr2.count > 0 &&
1046             (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
1047              be32_to_cpu(ents2[hdr2.count - 1].hashval) <
1048                                 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
1049                 return 1;
1050         return 0;
1051 }
1052
1053 /*
1054  * Rebalance leaf entries between two leaf blocks.
1055  * This is actually only called when the second block is new,
1056  * though the code deals with the general case.
1057  * A new entry will be inserted in one of the blocks, and that
1058  * entry is taken into account when balancing.
1059  */
1060 static void
1061 xfs_dir2_leafn_rebalance(
1062         xfs_da_state_t          *state,         /* btree cursor */
1063         xfs_da_state_blk_t      *blk1,          /* first btree block */
1064         xfs_da_state_blk_t      *blk2)          /* second btree block */
1065 {
1066         xfs_da_args_t           *args;          /* operation arguments */
1067         int                     count;          /* count (& direction) leaves */
1068         int                     isleft;         /* new goes in left leaf */
1069         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
1070         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
1071         int                     mid;            /* midpoint leaf index */
1072 #if defined(DEBUG) || defined(XFS_WARN)
1073         int                     oldstale;       /* old count of stale leaves */
1074 #endif
1075         int                     oldsum;         /* old total leaf count */
1076         int                     swap_blocks;    /* swapped leaf blocks */
1077         struct xfs_dir2_leaf_entry *ents1;
1078         struct xfs_dir2_leaf_entry *ents2;
1079         struct xfs_dir3_icleaf_hdr hdr1;
1080         struct xfs_dir3_icleaf_hdr hdr2;
1081         struct xfs_inode        *dp = state->args->dp;
1082
1083         args = state->args;
1084         /*
1085          * If the block order is wrong, swap the arguments.
1086          */
1087         swap_blocks = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp);
1088         if (swap_blocks)
1089                 swap(blk1, blk2);
1090
1091         leaf1 = blk1->bp->b_addr;
1092         leaf2 = blk2->bp->b_addr;
1093         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &hdr1, leaf1);
1094         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &hdr2, leaf2);
1095         ents1 = hdr1.ents;
1096         ents2 = hdr2.ents;
1097
1098         oldsum = hdr1.count + hdr2.count;
1099 #if defined(DEBUG) || defined(XFS_WARN)
1100         oldstale = hdr1.stale + hdr2.stale;
1101 #endif
1102         mid = oldsum >> 1;
1103
1104         /*
1105          * If the old leaf count was odd then the new one will be even,
1106          * so we need to divide the new count evenly.
1107          */
1108         if (oldsum & 1) {
1109                 xfs_dahash_t    midhash;        /* middle entry hash value */
1110
1111                 if (mid >= hdr1.count)
1112                         midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1113                 else
1114                         midhash = be32_to_cpu(ents1[mid].hashval);
1115                 isleft = args->hashval <= midhash;
1116         }
1117         /*
1118          * If the old count is even then the new count is odd, so there's
1119          * no preferred side for the new entry.
1120          * Pick the left one.
1121          */
1122         else
1123                 isleft = 1;
1124         /*
1125          * Calculate moved entry count.  Positive means left-to-right,
1126          * negative means right-to-left.  Then move the entries.
1127          */
1128         count = hdr1.count - mid + (isleft == 0);
1129         if (count > 0)
1130                 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1131                                         hdr1.count - count, blk2->bp,
1132                                         &hdr2, ents2, 0, count);
1133         else if (count < 0)
1134                 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1135                                         blk1->bp, &hdr1, ents1,
1136                                         hdr1.count, count);
1137
1138         ASSERT(hdr1.count + hdr2.count == oldsum);
1139         ASSERT(hdr1.stale + hdr2.stale == oldstale);
1140
1141         /* log the changes made when moving the entries */
1142         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf1, &hdr1);
1143         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf2, &hdr2);
1144         xfs_dir3_leaf_log_header(args, blk1->bp);
1145         xfs_dir3_leaf_log_header(args, blk2->bp);
1146
1147         xfs_dir3_leaf_check(dp, blk1->bp);
1148         xfs_dir3_leaf_check(dp, blk2->bp);
1149
1150         /*
1151          * Mark whether we're inserting into the old or new leaf.
1152          */
1153         if (hdr1.count < hdr2.count)
1154                 state->inleaf = swap_blocks;
1155         else if (hdr1.count > hdr2.count)
1156                 state->inleaf = !swap_blocks;
1157         else
1158                 state->inleaf = swap_blocks ^ (blk1->index <= hdr1.count);
1159         /*
1160          * Adjust the expected index for insertion.
1161          */
1162         if (!state->inleaf)
1163                 blk2->index = blk1->index - hdr1.count;
1164
1165         /*
1166          * Finally sanity check just to make sure we are not returning a
1167          * negative index
1168          */
1169         if (blk2->index < 0) {
1170                 state->inleaf = 1;
1171                 blk2->index = 0;
1172                 xfs_alert(dp->i_mount,
1173         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1174                         __func__, blk1->index);
1175         }
1176 }
1177
1178 static int
1179 xfs_dir3_data_block_free(
1180         xfs_da_args_t           *args,
1181         struct xfs_dir2_data_hdr *hdr,
1182         struct xfs_dir2_free    *free,
1183         xfs_dir2_db_t           fdb,
1184         int                     findex,
1185         struct xfs_buf          *fbp,
1186         int                     longest)
1187 {
1188         int                     logfree = 0;
1189         struct xfs_dir3_icfree_hdr freehdr;
1190         struct xfs_inode        *dp = args->dp;
1191
1192         xfs_dir2_free_hdr_from_disk(dp->i_mount, &freehdr, free);
1193         if (hdr) {
1194                 /*
1195                  * Data block is not empty, just set the free entry to the new
1196                  * value.
1197                  */
1198                 freehdr.bests[findex] = cpu_to_be16(longest);
1199                 xfs_dir2_free_log_bests(args, &freehdr, fbp, findex, findex);
1200                 return 0;
1201         }
1202
1203         /* One less used entry in the free table. */
1204         freehdr.nused--;
1205
1206         /*
1207          * If this was the last entry in the table, we can trim the table size
1208          * back.  There might be other entries at the end referring to
1209          * non-existent data blocks, get those too.
1210          */
1211         if (findex == freehdr.nvalid - 1) {
1212                 int     i;              /* free entry index */
1213
1214                 for (i = findex - 1; i >= 0; i--) {
1215                         if (freehdr.bests[i] != cpu_to_be16(NULLDATAOFF))
1216                                 break;
1217                 }
1218                 freehdr.nvalid = i + 1;
1219                 logfree = 0;
1220         } else {
1221                 /* Not the last entry, just punch it out.  */
1222                 freehdr.bests[findex] = cpu_to_be16(NULLDATAOFF);
1223                 logfree = 1;
1224         }
1225
1226         xfs_dir2_free_hdr_to_disk(dp->i_mount, free, &freehdr);
1227         xfs_dir2_free_log_header(args, fbp);
1228
1229         /*
1230          * If there are no useful entries left in the block, get rid of the
1231          * block if we can.
1232          */
1233         if (!freehdr.nused) {
1234                 int error;
1235
1236                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1237                 if (error == 0) {
1238                         fbp = NULL;
1239                         logfree = 0;
1240                 } else if (error != -ENOSPC || args->total != 0)
1241                         return error;
1242                 /*
1243                  * It's possible to get ENOSPC if there is no
1244                  * space reservation.  In this case some one
1245                  * else will eventually get rid of this block.
1246                  */
1247         }
1248
1249         /* Log the free entry that changed, unless we got rid of it.  */
1250         if (logfree)
1251                 xfs_dir2_free_log_bests(args, &freehdr, fbp, findex, findex);
1252         return 0;
1253 }
1254
1255 /*
1256  * Remove an entry from a node directory.
1257  * This removes the leaf entry and the data entry,
1258  * and updates the free block if necessary.
1259  */
1260 static int                                      /* error */
1261 xfs_dir2_leafn_remove(
1262         xfs_da_args_t           *args,          /* operation arguments */
1263         struct xfs_buf          *bp,            /* leaf buffer */
1264         int                     index,          /* leaf entry index */
1265         xfs_da_state_blk_t      *dblk,          /* data block */
1266         int                     *rval)          /* resulting block needs join */
1267 {
1268         struct xfs_da_geometry  *geo = args->geo;
1269         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1270         xfs_dir2_db_t           db;             /* data block number */
1271         struct xfs_buf          *dbp;           /* data block buffer */
1272         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1273         xfs_inode_t             *dp;            /* incore directory inode */
1274         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1275         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1276         int                     longest;        /* longest data free entry */
1277         int                     off;            /* data block entry offset */
1278         int                     needlog;        /* need to log data header */
1279         int                     needscan;       /* need to rescan data frees */
1280         xfs_trans_t             *tp;            /* transaction pointer */
1281         struct xfs_dir2_data_free *bf;          /* bestfree table */
1282         struct xfs_dir3_icleaf_hdr leafhdr;
1283
1284         trace_xfs_dir2_leafn_remove(args, index);
1285
1286         dp = args->dp;
1287         tp = args->trans;
1288         leaf = bp->b_addr;
1289         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
1290
1291         /*
1292          * Point to the entry we're removing.
1293          */
1294         lep = &leafhdr.ents[index];
1295
1296         /*
1297          * Extract the data block and offset from the entry.
1298          */
1299         db = xfs_dir2_dataptr_to_db(geo, be32_to_cpu(lep->address));
1300         ASSERT(dblk->blkno == db);
1301         off = xfs_dir2_dataptr_to_off(geo, be32_to_cpu(lep->address));
1302         ASSERT(dblk->index == off);
1303
1304         /*
1305          * Kill the leaf entry by marking it stale.
1306          * Log the leaf block changes.
1307          */
1308         leafhdr.stale++;
1309         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
1310         xfs_dir3_leaf_log_header(args, bp);
1311
1312         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1313         xfs_dir3_leaf_log_ents(args, &leafhdr, bp, index, index);
1314
1315         /*
1316          * Make the data entry free.  Keep track of the longest freespace
1317          * in the data block in case it changes.
1318          */
1319         dbp = dblk->bp;
1320         hdr = dbp->b_addr;
1321         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1322         bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1323         longest = be16_to_cpu(bf[0].length);
1324         needlog = needscan = 0;
1325         xfs_dir2_data_make_free(args, dbp, off,
1326                 xfs_dir2_data_entsize(dp->i_mount, dep->namelen), &needlog,
1327                 &needscan);
1328         /*
1329          * Rescan the data block freespaces for bestfree.
1330          * Log the data block header if needed.
1331          */
1332         if (needscan)
1333                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
1334         if (needlog)
1335                 xfs_dir2_data_log_header(args, dbp);
1336         xfs_dir3_data_check(dp, dbp);
1337         /*
1338          * If the longest data block freespace changes, need to update
1339          * the corresponding freeblock entry.
1340          */
1341         if (longest < be16_to_cpu(bf[0].length)) {
1342                 int             error;          /* error return value */
1343                 struct xfs_buf  *fbp;           /* freeblock buffer */
1344                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1345                 int             findex;         /* index in freeblock entries */
1346                 xfs_dir2_free_t *free;          /* freeblock structure */
1347
1348                 /*
1349                  * Convert the data block number to a free block,
1350                  * read in the free block.
1351                  */
1352                 fdb = xfs_dir2_db_to_fdb(geo, db);
1353                 error = xfs_dir2_free_read(tp, dp, xfs_dir2_db_to_da(geo, fdb),
1354                                            &fbp);
1355                 if (error)
1356                         return error;
1357                 free = fbp->b_addr;
1358 #ifdef DEBUG
1359         {
1360                 struct xfs_dir3_icfree_hdr freehdr;
1361
1362                 xfs_dir2_free_hdr_from_disk(dp->i_mount, &freehdr, free);
1363                 ASSERT(freehdr.firstdb == geo->free_max_bests *
1364                         (fdb - xfs_dir2_byte_to_db(geo, XFS_DIR2_FREE_OFFSET)));
1365         }
1366 #endif
1367                 /*
1368                  * Calculate which entry we need to fix.
1369                  */
1370                 findex = xfs_dir2_db_to_fdindex(geo, db);
1371                 longest = be16_to_cpu(bf[0].length);
1372                 /*
1373                  * If the data block is now empty we can get rid of it
1374                  * (usually).
1375                  */
1376                 if (longest == geo->blksize - geo->data_entry_offset) {
1377                         /*
1378                          * Try to punch out the data block.
1379                          */
1380                         error = xfs_dir2_shrink_inode(args, db, dbp);
1381                         if (error == 0) {
1382                                 dblk->bp = NULL;
1383                                 hdr = NULL;
1384                         }
1385                         /*
1386                          * We can get ENOSPC if there's no space reservation.
1387                          * In this case just drop the buffer and some one else
1388                          * will eventually get rid of the empty block.
1389                          */
1390                         else if (!(error == -ENOSPC && args->total == 0))
1391                                 return error;
1392                 }
1393                 /*
1394                  * If we got rid of the data block, we can eliminate that entry
1395                  * in the free block.
1396                  */
1397                 error = xfs_dir3_data_block_free(args, hdr, free,
1398                                                  fdb, findex, fbp, longest);
1399                 if (error)
1400                         return error;
1401         }
1402
1403         xfs_dir3_leaf_check(dp, bp);
1404         /*
1405          * Return indication of whether this leaf block is empty enough
1406          * to justify trying to join it with a neighbor.
1407          */
1408         *rval = (geo->leaf_hdr_size +
1409                  (uint)sizeof(leafhdr.ents) * (leafhdr.count - leafhdr.stale)) <
1410                 geo->magicpct;
1411         return 0;
1412 }
1413
1414 /*
1415  * Split the leaf entries in the old block into old and new blocks.
1416  */
1417 int                                             /* error */
1418 xfs_dir2_leafn_split(
1419         xfs_da_state_t          *state,         /* btree cursor */
1420         xfs_da_state_blk_t      *oldblk,        /* original block */
1421         xfs_da_state_blk_t      *newblk)        /* newly created block */
1422 {
1423         xfs_da_args_t           *args;          /* operation arguments */
1424         xfs_dablk_t             blkno;          /* new leaf block number */
1425         int                     error;          /* error return value */
1426         struct xfs_inode        *dp;
1427
1428         /*
1429          * Allocate space for a new leaf node.
1430          */
1431         args = state->args;
1432         dp = args->dp;
1433         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1434         error = xfs_da_grow_inode(args, &blkno);
1435         if (error) {
1436                 return error;
1437         }
1438         /*
1439          * Initialize the new leaf block.
1440          */
1441         error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(args->geo, blkno),
1442                                       &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1443         if (error)
1444                 return error;
1445
1446         newblk->blkno = blkno;
1447         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1448         /*
1449          * Rebalance the entries across the two leaves, link the new
1450          * block into the leaves.
1451          */
1452         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1453         error = xfs_da3_blk_link(state, oldblk, newblk);
1454         if (error) {
1455                 return error;
1456         }
1457         /*
1458          * Insert the new entry in the correct block.
1459          */
1460         if (state->inleaf)
1461                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1462         else
1463                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1464         /*
1465          * Update last hashval in each block since we added the name.
1466          */
1467         oldblk->hashval = xfs_dir2_leaf_lasthash(dp, oldblk->bp, NULL);
1468         newblk->hashval = xfs_dir2_leaf_lasthash(dp, newblk->bp, NULL);
1469         xfs_dir3_leaf_check(dp, oldblk->bp);
1470         xfs_dir3_leaf_check(dp, newblk->bp);
1471         return error;
1472 }
1473
1474 /*
1475  * Check a leaf block and its neighbors to see if the block should be
1476  * collapsed into one or the other neighbor.  Always keep the block
1477  * with the smaller block number.
1478  * If the current block is over 50% full, don't try to join it, return 0.
1479  * If the block is empty, fill in the state structure and return 2.
1480  * If it can be collapsed, fill in the state structure and return 1.
1481  * If nothing can be done, return 0.
1482  */
1483 int                                             /* error */
1484 xfs_dir2_leafn_toosmall(
1485         xfs_da_state_t          *state,         /* btree cursor */
1486         int                     *action)        /* resulting action to take */
1487 {
1488         xfs_da_state_blk_t      *blk;           /* leaf block */
1489         xfs_dablk_t             blkno;          /* leaf block number */
1490         struct xfs_buf          *bp;            /* leaf buffer */
1491         int                     bytes;          /* bytes in use */
1492         int                     count;          /* leaf live entry count */
1493         int                     error;          /* error return value */
1494         int                     forward;        /* sibling block direction */
1495         int                     i;              /* sibling counter */
1496         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1497         int                     rval;           /* result from path_shift */
1498         struct xfs_dir3_icleaf_hdr leafhdr;
1499         struct xfs_dir2_leaf_entry *ents;
1500         struct xfs_inode        *dp = state->args->dp;
1501
1502         /*
1503          * Check for the degenerate case of the block being over 50% full.
1504          * If so, it's not worth even looking to see if we might be able
1505          * to coalesce with a sibling.
1506          */
1507         blk = &state->path.blk[state->path.active - 1];
1508         leaf = blk->bp->b_addr;
1509         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
1510         ents = leafhdr.ents;
1511         xfs_dir3_leaf_check(dp, blk->bp);
1512
1513         count = leafhdr.count - leafhdr.stale;
1514         bytes = state->args->geo->leaf_hdr_size + count * sizeof(ents[0]);
1515         if (bytes > (state->args->geo->blksize >> 1)) {
1516                 /*
1517                  * Blk over 50%, don't try to join.
1518                  */
1519                 *action = 0;
1520                 return 0;
1521         }
1522         /*
1523          * Check for the degenerate case of the block being empty.
1524          * If the block is empty, we'll simply delete it, no need to
1525          * coalesce it with a sibling block.  We choose (arbitrarily)
1526          * to merge with the forward block unless it is NULL.
1527          */
1528         if (count == 0) {
1529                 /*
1530                  * Make altpath point to the block we want to keep and
1531                  * path point to the block we want to drop (this one).
1532                  */
1533                 forward = (leafhdr.forw != 0);
1534                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1535                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1536                         &rval);
1537                 if (error)
1538                         return error;
1539                 *action = rval ? 2 : 0;
1540                 return 0;
1541         }
1542         /*
1543          * Examine each sibling block to see if we can coalesce with
1544          * at least 25% free space to spare.  We need to figure out
1545          * whether to merge with the forward or the backward block.
1546          * We prefer coalescing with the lower numbered sibling so as
1547          * to shrink a directory over time.
1548          */
1549         forward = leafhdr.forw < leafhdr.back;
1550         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1551                 struct xfs_dir3_icleaf_hdr hdr2;
1552
1553                 blkno = forward ? leafhdr.forw : leafhdr.back;
1554                 if (blkno == 0)
1555                         continue;
1556                 /*
1557                  * Read the sibling leaf block.
1558                  */
1559                 error = xfs_dir3_leafn_read(state->args->trans, dp, blkno, &bp);
1560                 if (error)
1561                         return error;
1562
1563                 /*
1564                  * Count bytes in the two blocks combined.
1565                  */
1566                 count = leafhdr.count - leafhdr.stale;
1567                 bytes = state->args->geo->blksize -
1568                         (state->args->geo->blksize >> 2);
1569
1570                 leaf = bp->b_addr;
1571                 xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &hdr2, leaf);
1572                 ents = hdr2.ents;
1573                 count += hdr2.count - hdr2.stale;
1574                 bytes -= count * sizeof(ents[0]);
1575
1576                 /*
1577                  * Fits with at least 25% to spare.
1578                  */
1579                 if (bytes >= 0)
1580                         break;
1581                 xfs_trans_brelse(state->args->trans, bp);
1582         }
1583         /*
1584          * Didn't like either block, give up.
1585          */
1586         if (i >= 2) {
1587                 *action = 0;
1588                 return 0;
1589         }
1590
1591         /*
1592          * Make altpath point to the block we want to keep (the lower
1593          * numbered block) and path point to the block we want to drop.
1594          */
1595         memcpy(&state->altpath, &state->path, sizeof(state->path));
1596         if (blkno < blk->blkno)
1597                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1598                         &rval);
1599         else
1600                 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1601                         &rval);
1602         if (error) {
1603                 return error;
1604         }
1605         *action = rval ? 0 : 1;
1606         return 0;
1607 }
1608
1609 /*
1610  * Move all the leaf entries from drop_blk to save_blk.
1611  * This is done as part of a join operation.
1612  */
1613 void
1614 xfs_dir2_leafn_unbalance(
1615         xfs_da_state_t          *state,         /* cursor */
1616         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1617         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1618 {
1619         xfs_da_args_t           *args;          /* operation arguments */
1620         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1621         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1622         struct xfs_dir3_icleaf_hdr savehdr;
1623         struct xfs_dir3_icleaf_hdr drophdr;
1624         struct xfs_dir2_leaf_entry *sents;
1625         struct xfs_dir2_leaf_entry *dents;
1626         struct xfs_inode        *dp = state->args->dp;
1627
1628         args = state->args;
1629         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1630         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1631         drop_leaf = drop_blk->bp->b_addr;
1632         save_leaf = save_blk->bp->b_addr;
1633
1634         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &savehdr, save_leaf);
1635         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &drophdr, drop_leaf);
1636         sents = savehdr.ents;
1637         dents = drophdr.ents;
1638
1639         /*
1640          * If there are any stale leaf entries, take this opportunity
1641          * to purge them.
1642          */
1643         if (drophdr.stale)
1644                 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1645         if (savehdr.stale)
1646                 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1647
1648         /*
1649          * Move the entries from drop to the appropriate end of save.
1650          */
1651         drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1652         if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1653                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1654                                         save_blk->bp, &savehdr, sents, 0,
1655                                         drophdr.count);
1656         else
1657                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1658                                         save_blk->bp, &savehdr, sents,
1659                                         savehdr.count, drophdr.count);
1660         save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1661
1662         /* log the changes made when moving the entries */
1663         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, save_leaf, &savehdr);
1664         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, drop_leaf, &drophdr);
1665         xfs_dir3_leaf_log_header(args, save_blk->bp);
1666         xfs_dir3_leaf_log_header(args, drop_blk->bp);
1667
1668         xfs_dir3_leaf_check(dp, save_blk->bp);
1669         xfs_dir3_leaf_check(dp, drop_blk->bp);
1670 }
1671
1672 /*
1673  * Add a new data block to the directory at the free space index that the caller
1674  * has specified.
1675  */
1676 static int
1677 xfs_dir2_node_add_datablk(
1678         struct xfs_da_args      *args,
1679         struct xfs_da_state_blk *fblk,
1680         xfs_dir2_db_t           *dbno,
1681         struct xfs_buf          **dbpp,
1682         struct xfs_buf          **fbpp,
1683         struct xfs_dir3_icfree_hdr *hdr,
1684         int                     *findex)
1685 {
1686         struct xfs_inode        *dp = args->dp;
1687         struct xfs_trans        *tp = args->trans;
1688         struct xfs_mount        *mp = dp->i_mount;
1689         struct xfs_dir2_data_free *bf;
1690         xfs_dir2_db_t           fbno;
1691         struct xfs_buf          *fbp;
1692         struct xfs_buf          *dbp;
1693         int                     error;
1694
1695         /* Not allowed to allocate, return failure. */
1696         if (args->total == 0)
1697                 return -ENOSPC;
1698
1699         /* Allocate and initialize the new data block.  */
1700         error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, dbno);
1701         if (error)
1702                 return error;
1703         error = xfs_dir3_data_init(args, *dbno, &dbp);
1704         if (error)
1705                 return error;
1706
1707         /*
1708          * Get the freespace block corresponding to the data block
1709          * that was just allocated.
1710          */
1711         fbno = xfs_dir2_db_to_fdb(args->geo, *dbno);
1712         error = xfs_dir2_free_try_read(tp, dp,
1713                                xfs_dir2_db_to_da(args->geo, fbno), &fbp);
1714         if (error)
1715                 return error;
1716
1717         /*
1718          * If there wasn't a freespace block, the read will
1719          * return a NULL fbp.  Allocate and initialize a new one.
1720          */
1721         if (!fbp) {
1722                 error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fbno);
1723                 if (error)
1724                         return error;
1725
1726                 if (XFS_IS_CORRUPT(mp,
1727                                    xfs_dir2_db_to_fdb(args->geo, *dbno) !=
1728                                    fbno)) {
1729                         xfs_alert(mp,
1730 "%s: dir ino %llu needed freesp block %lld for data block %lld, got %lld",
1731                                 __func__, (unsigned long long)dp->i_ino,
1732                                 (long long)xfs_dir2_db_to_fdb(args->geo, *dbno),
1733                                 (long long)*dbno, (long long)fbno);
1734                         if (fblk) {
1735                                 xfs_alert(mp,
1736                         " fblk "PTR_FMT" blkno %llu index %d magic 0x%x",
1737                                         fblk, (unsigned long long)fblk->blkno,
1738                                         fblk->index, fblk->magic);
1739                         } else {
1740                                 xfs_alert(mp, " ... fblk is NULL");
1741                         }
1742                         return -EFSCORRUPTED;
1743                 }
1744
1745                 /* Get a buffer for the new block. */
1746                 error = xfs_dir3_free_get_buf(args, fbno, &fbp);
1747                 if (error)
1748                         return error;
1749                 xfs_dir2_free_hdr_from_disk(mp, hdr, fbp->b_addr);
1750
1751                 /* Remember the first slot as our empty slot. */
1752                 hdr->firstdb = (fbno - xfs_dir2_byte_to_db(args->geo,
1753                                                         XFS_DIR2_FREE_OFFSET)) *
1754                                 args->geo->free_max_bests;
1755         } else {
1756                 xfs_dir2_free_hdr_from_disk(mp, hdr, fbp->b_addr);
1757         }
1758
1759         /* Set the freespace block index from the data block number. */
1760         *findex = xfs_dir2_db_to_fdindex(args->geo, *dbno);
1761
1762         /* Extend the freespace table if the new data block is off the end. */
1763         if (*findex >= hdr->nvalid) {
1764                 ASSERT(*findex < args->geo->free_max_bests);
1765                 hdr->nvalid = *findex + 1;
1766                 hdr->bests[*findex] = cpu_to_be16(NULLDATAOFF);
1767         }
1768
1769         /*
1770          * If this entry was for an empty data block (this should always be
1771          * true) then update the header.
1772          */
1773         if (hdr->bests[*findex] == cpu_to_be16(NULLDATAOFF)) {
1774                 hdr->nused++;
1775                 xfs_dir2_free_hdr_to_disk(mp, fbp->b_addr, hdr);
1776                 xfs_dir2_free_log_header(args, fbp);
1777         }
1778
1779         /* Update the freespace value for the new block in the table. */
1780         bf = xfs_dir2_data_bestfree_p(mp, dbp->b_addr);
1781         hdr->bests[*findex] = bf[0].length;
1782
1783         *dbpp = dbp;
1784         *fbpp = fbp;
1785         return 0;
1786 }
1787
1788 static int
1789 xfs_dir2_node_find_freeblk(
1790         struct xfs_da_args      *args,
1791         struct xfs_da_state_blk *fblk,
1792         xfs_dir2_db_t           *dbnop,
1793         struct xfs_buf          **fbpp,
1794         struct xfs_dir3_icfree_hdr *hdr,
1795         int                     *findexp,
1796         int                     length)
1797 {
1798         struct xfs_inode        *dp = args->dp;
1799         struct xfs_trans        *tp = args->trans;
1800         struct xfs_buf          *fbp = NULL;
1801         xfs_dir2_db_t           firstfbno;
1802         xfs_dir2_db_t           lastfbno;
1803         xfs_dir2_db_t           ifbno = -1;
1804         xfs_dir2_db_t           dbno = -1;
1805         xfs_dir2_db_t           fbno;
1806         xfs_fileoff_t           fo;
1807         int                     findex = 0;
1808         int                     error;
1809
1810         /*
1811          * If we came in with a freespace block that means that lookup
1812          * found an entry with our hash value.  This is the freespace
1813          * block for that data entry.
1814          */
1815         if (fblk) {
1816                 fbp = fblk->bp;
1817                 findex = fblk->index;
1818                 xfs_dir2_free_hdr_from_disk(dp->i_mount, hdr, fbp->b_addr);
1819                 if (findex >= 0) {
1820                         /* caller already found the freespace for us. */
1821                         ASSERT(findex < hdr->nvalid);
1822                         ASSERT(be16_to_cpu(hdr->bests[findex]) != NULLDATAOFF);
1823                         ASSERT(be16_to_cpu(hdr->bests[findex]) >= length);
1824                         dbno = hdr->firstdb + findex;
1825                         goto found_block;
1826                 }
1827
1828                 /*
1829                  * The data block looked at didn't have enough room.
1830                  * We'll start at the beginning of the freespace entries.
1831                  */
1832                 ifbno = fblk->blkno;
1833                 xfs_trans_brelse(tp, fbp);
1834                 fbp = NULL;
1835                 fblk->bp = NULL;
1836         }
1837
1838         /*
1839          * If we don't have a data block yet, we're going to scan the freespace
1840          * data for a data block with enough free space in it.
1841          */
1842         error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK);
1843         if (error)
1844                 return error;
1845         lastfbno = xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo);
1846         firstfbno = xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET);
1847
1848         for (fbno = lastfbno - 1; fbno >= firstfbno; fbno--) {
1849                 /* If it's ifbno we already looked at it. */
1850                 if (fbno == ifbno)
1851                         continue;
1852
1853                 /*
1854                  * Read the block.  There can be holes in the freespace blocks,
1855                  * so this might not succeed.  This should be really rare, so
1856                  * there's no reason to avoid it.
1857                  */
1858                 error = xfs_dir2_free_try_read(tp, dp,
1859                                 xfs_dir2_db_to_da(args->geo, fbno),
1860                                 &fbp);
1861                 if (error)
1862                         return error;
1863                 if (!fbp)
1864                         continue;
1865
1866                 xfs_dir2_free_hdr_from_disk(dp->i_mount, hdr, fbp->b_addr);
1867
1868                 /* Scan the free entry array for a large enough free space. */
1869                 for (findex = hdr->nvalid - 1; findex >= 0; findex--) {
1870                         if (be16_to_cpu(hdr->bests[findex]) != NULLDATAOFF &&
1871                             be16_to_cpu(hdr->bests[findex]) >= length) {
1872                                 dbno = hdr->firstdb + findex;
1873                                 goto found_block;
1874                         }
1875                 }
1876
1877                 /* Didn't find free space, go on to next free block */
1878                 xfs_trans_brelse(tp, fbp);
1879         }
1880
1881 found_block:
1882         *dbnop = dbno;
1883         *fbpp = fbp;
1884         *findexp = findex;
1885         return 0;
1886 }
1887
1888 /*
1889  * Add the data entry for a node-format directory name addition.
1890  * The leaf entry is added in xfs_dir2_leafn_add.
1891  * We may enter with a freespace block that the lookup found.
1892  */
1893 static int
1894 xfs_dir2_node_addname_int(
1895         struct xfs_da_args      *args,          /* operation arguments */
1896         struct xfs_da_state_blk *fblk)          /* optional freespace block */
1897 {
1898         struct xfs_dir2_data_unused *dup;       /* data unused entry pointer */
1899         struct xfs_dir2_data_entry *dep;        /* data entry pointer */
1900         struct xfs_dir2_data_hdr *hdr;          /* data block header */
1901         struct xfs_dir2_data_free *bf;
1902         struct xfs_trans        *tp = args->trans;
1903         struct xfs_inode        *dp = args->dp;
1904         struct xfs_dir3_icfree_hdr freehdr;
1905         struct xfs_buf          *dbp;           /* data block buffer */
1906         struct xfs_buf          *fbp;           /* freespace buffer */
1907         xfs_dir2_data_aoff_t    aoff;
1908         xfs_dir2_db_t           dbno;           /* data block number */
1909         int                     error;          /* error return value */
1910         int                     findex;         /* freespace entry index */
1911         int                     length;         /* length of the new entry */
1912         int                     logfree = 0;    /* need to log free entry */
1913         int                     needlog = 0;    /* need to log data header */
1914         int                     needscan = 0;   /* need to rescan data frees */
1915         __be16                  *tagp;          /* data entry tag pointer */
1916
1917         length = xfs_dir2_data_entsize(dp->i_mount, args->namelen);
1918         error = xfs_dir2_node_find_freeblk(args, fblk, &dbno, &fbp, &freehdr,
1919                                            &findex, length);
1920         if (error)
1921                 return error;
1922
1923         /*
1924          * Now we know if we must allocate blocks, so if we are checking whether
1925          * we can insert without allocation then we can return now.
1926          */
1927         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
1928                 if (dbno == -1)
1929                         return -ENOSPC;
1930                 return 0;
1931         }
1932
1933         /*
1934          * If we don't have a data block, we need to allocate one and make
1935          * the freespace entries refer to it.
1936          */
1937         if (dbno == -1) {
1938                 /* we're going to have to log the free block index later */
1939                 logfree = 1;
1940                 error = xfs_dir2_node_add_datablk(args, fblk, &dbno, &dbp, &fbp,
1941                                                   &freehdr, &findex);
1942         } else {
1943                 /* Read the data block in. */
1944                 error = xfs_dir3_data_read(tp, dp,
1945                                            xfs_dir2_db_to_da(args->geo, dbno),
1946                                            0, &dbp);
1947         }
1948         if (error)
1949                 return error;
1950
1951         /* setup for data block up now */
1952         hdr = dbp->b_addr;
1953         bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1954         ASSERT(be16_to_cpu(bf[0].length) >= length);
1955
1956         /* Point to the existing unused space. */
1957         dup = (xfs_dir2_data_unused_t *)
1958               ((char *)hdr + be16_to_cpu(bf[0].offset));
1959
1960         /* Mark the first part of the unused space, inuse for us. */
1961         aoff = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
1962         error = xfs_dir2_data_use_free(args, dbp, dup, aoff, length,
1963                         &needlog, &needscan);
1964         if (error) {
1965                 xfs_trans_brelse(tp, dbp);
1966                 return error;
1967         }
1968
1969         /* Fill in the new entry and log it. */
1970         dep = (xfs_dir2_data_entry_t *)dup;
1971         dep->inumber = cpu_to_be64(args->inumber);
1972         dep->namelen = args->namelen;
1973         memcpy(dep->name, args->name, dep->namelen);
1974         xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
1975         tagp = xfs_dir2_data_entry_tag_p(dp->i_mount, dep);
1976         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1977         xfs_dir2_data_log_entry(args, dbp, dep);
1978
1979         /* Rescan the freespace and log the data block if needed. */
1980         if (needscan)
1981                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
1982         if (needlog)
1983                 xfs_dir2_data_log_header(args, dbp);
1984
1985         /* If the freespace block entry is now wrong, update it. */
1986         if (freehdr.bests[findex] != bf[0].length) {
1987                 freehdr.bests[findex] = bf[0].length;
1988                 logfree = 1;
1989         }
1990
1991         /* Log the freespace entry if needed. */
1992         if (logfree)
1993                 xfs_dir2_free_log_bests(args, &freehdr, fbp, findex, findex);
1994
1995         /* Return the data block and offset in args. */
1996         args->blkno = (xfs_dablk_t)dbno;
1997         args->index = be16_to_cpu(*tagp);
1998         return 0;
1999 }
2000
2001 /*
2002  * Top-level node form directory addname routine.
2003  */
2004 int                                             /* error */
2005 xfs_dir2_node_addname(
2006         xfs_da_args_t           *args)          /* operation arguments */
2007 {
2008         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
2009         int                     error;          /* error return value */
2010         int                     rval;           /* sub-return value */
2011         xfs_da_state_t          *state;         /* btree cursor */
2012
2013         trace_xfs_dir2_node_addname(args);
2014
2015         /*
2016          * Allocate and initialize the state (btree cursor).
2017          */
2018         state = xfs_da_state_alloc(args);
2019         /*
2020          * Look up the name.  We're not supposed to find it, but
2021          * this gives us the insertion point.
2022          */
2023         error = xfs_da3_node_lookup_int(state, &rval);
2024         if (error)
2025                 rval = error;
2026         if (rval != -ENOENT) {
2027                 goto done;
2028         }
2029         /*
2030          * Add the data entry to a data block.
2031          * Extravalid is set to a freeblock found by lookup.
2032          */
2033         rval = xfs_dir2_node_addname_int(args,
2034                 state->extravalid ? &state->extrablk : NULL);
2035         if (rval) {
2036                 goto done;
2037         }
2038         blk = &state->path.blk[state->path.active - 1];
2039         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2040         /*
2041          * Add the new leaf entry.
2042          */
2043         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
2044         if (rval == 0) {
2045                 /*
2046                  * It worked, fix the hash values up the btree.
2047                  */
2048                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
2049                         xfs_da3_fixhashpath(state, &state->path);
2050         } else {
2051                 /*
2052                  * It didn't work, we need to split the leaf block.
2053                  */
2054                 if (args->total == 0) {
2055                         ASSERT(rval == -ENOSPC);
2056                         goto done;
2057                 }
2058                 /*
2059                  * Split the leaf block and insert the new entry.
2060                  */
2061                 rval = xfs_da3_split(state);
2062         }
2063 done:
2064         xfs_da_state_free(state);
2065         return rval;
2066 }
2067
2068 /*
2069  * Lookup an entry in a node-format directory.
2070  * All the real work happens in xfs_da3_node_lookup_int.
2071  * The only real output is the inode number of the entry.
2072  */
2073 int                                             /* error */
2074 xfs_dir2_node_lookup(
2075         xfs_da_args_t   *args)                  /* operation arguments */
2076 {
2077         int             error;                  /* error return value */
2078         int             i;                      /* btree level */
2079         int             rval;                   /* operation return value */
2080         xfs_da_state_t  *state;                 /* btree cursor */
2081
2082         trace_xfs_dir2_node_lookup(args);
2083
2084         /*
2085          * Allocate and initialize the btree cursor.
2086          */
2087         state = xfs_da_state_alloc(args);
2088
2089         /*
2090          * Fill in the path to the entry in the cursor.
2091          */
2092         error = xfs_da3_node_lookup_int(state, &rval);
2093         if (error)
2094                 rval = error;
2095         else if (rval == -ENOENT && args->cmpresult == XFS_CMP_CASE) {
2096                 /* If a CI match, dup the actual name and return -EEXIST */
2097                 xfs_dir2_data_entry_t   *dep;
2098
2099                 dep = (xfs_dir2_data_entry_t *)
2100                         ((char *)state->extrablk.bp->b_addr +
2101                                                  state->extrablk.index);
2102                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2103         }
2104         /*
2105          * Release the btree blocks and leaf block.
2106          */
2107         for (i = 0; i < state->path.active; i++) {
2108                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2109                 state->path.blk[i].bp = NULL;
2110         }
2111         /*
2112          * Release the data block if we have it.
2113          */
2114         if (state->extravalid && state->extrablk.bp) {
2115                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2116                 state->extrablk.bp = NULL;
2117         }
2118         xfs_da_state_free(state);
2119         return rval;
2120 }
2121
2122 /*
2123  * Remove an entry from a node-format directory.
2124  */
2125 int                                             /* error */
2126 xfs_dir2_node_removename(
2127         struct xfs_da_args      *args)          /* operation arguments */
2128 {
2129         struct xfs_da_state_blk *blk;           /* leaf block */
2130         int                     error;          /* error return value */
2131         int                     rval;           /* operation return value */
2132         struct xfs_da_state     *state;         /* btree cursor */
2133
2134         trace_xfs_dir2_node_removename(args);
2135
2136         /*
2137          * Allocate and initialize the btree cursor.
2138          */
2139         state = xfs_da_state_alloc(args);
2140
2141         /* Look up the entry we're deleting, set up the cursor. */
2142         error = xfs_da3_node_lookup_int(state, &rval);
2143         if (error)
2144                 goto out_free;
2145
2146         /* Didn't find it, upper layer screwed up. */
2147         if (rval != -EEXIST) {
2148                 error = rval;
2149                 goto out_free;
2150         }
2151
2152         blk = &state->path.blk[state->path.active - 1];
2153         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2154         ASSERT(state->extravalid);
2155         /*
2156          * Remove the leaf and data entries.
2157          * Extrablk refers to the data block.
2158          */
2159         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2160                 &state->extrablk, &rval);
2161         if (error)
2162                 goto out_free;
2163         /*
2164          * Fix the hash values up the btree.
2165          */
2166         xfs_da3_fixhashpath(state, &state->path);
2167         /*
2168          * If we need to join leaf blocks, do it.
2169          */
2170         if (rval && state->path.active > 1)
2171                 error = xfs_da3_join(state);
2172         /*
2173          * If no errors so far, try conversion to leaf format.
2174          */
2175         if (!error)
2176                 error = xfs_dir2_node_to_leaf(state);
2177 out_free:
2178         xfs_da_state_free(state);
2179         return error;
2180 }
2181
2182 /*
2183  * Replace an entry's inode number in a node-format directory.
2184  */
2185 int                                             /* error */
2186 xfs_dir2_node_replace(
2187         xfs_da_args_t           *args)          /* operation arguments */
2188 {
2189         xfs_da_state_blk_t      *blk;           /* leaf block */
2190         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
2191         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
2192         int                     error;          /* error return value */
2193         int                     i;              /* btree level */
2194         xfs_ino_t               inum;           /* new inode number */
2195         int                     ftype;          /* new file type */
2196         int                     rval;           /* internal return value */
2197         xfs_da_state_t          *state;         /* btree cursor */
2198
2199         trace_xfs_dir2_node_replace(args);
2200
2201         /*
2202          * Allocate and initialize the btree cursor.
2203          */
2204         state = xfs_da_state_alloc(args);
2205
2206         /*
2207          * We have to save new inode number and ftype since
2208          * xfs_da3_node_lookup_int() is going to overwrite them
2209          */
2210         inum = args->inumber;
2211         ftype = args->filetype;
2212
2213         /*
2214          * Lookup the entry to change in the btree.
2215          */
2216         error = xfs_da3_node_lookup_int(state, &rval);
2217         if (error) {
2218                 rval = error;
2219         }
2220         /*
2221          * It should be found, since the vnodeops layer has looked it up
2222          * and locked it.  But paranoia is good.
2223          */
2224         if (rval == -EEXIST) {
2225                 struct xfs_dir3_icleaf_hdr      leafhdr;
2226
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                 ASSERT(state->extravalid);
2233
2234                 xfs_dir2_leaf_hdr_from_disk(state->mp, &leafhdr,
2235                                             blk->bp->b_addr);
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(leafhdr.ents[blk->index].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                 xfs_dir2_data_put_ftype(state->mp, 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         xfs_dir2_free_hdr_from_disk(dp->i_mount, &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 }