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);
  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         } else {
 198                 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
 199
 200                 if (be32_to_cpu(hdr->firstdb) != firstdb)
 201                         return __this_address;
 202                 if (be32_to_cpu(hdr->nvalid) > maxbests)
 203                         return __this_address;
 204                 if (be32_to_cpu(hdr->nvalid) < be32_to_cpu(hdr->nused))
 205                         return __this_address;
 206         }
 207         return NULL;
 208 }
 209
 210 static int
 211 __xfs_dir3_free_read(
 212         struct xfs_trans        *tp,
 213         struct xfs_inode        *dp,
 214         xfs_dablk_t             fbno,
 215         xfs_daddr_t             mappedbno,
 216         struct xfs_buf          **bpp)
 217 {
 218         xfs_failaddr_t          fa;
 219         int                     err;
 220
 221         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 222                                 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
 223         if (err || !*bpp)
 224                 return err;
 225
 226         /* Check things that we can't do in the verifier. */
 227         fa = xfs_dir3_free_header_check(dp, fbno, *bpp);
 228         if (fa) {
 229                 xfs_verifier_error(*bpp, -EFSCORRUPTED, fa);
 230                 xfs_trans_brelse(tp, *bpp);
 231                 return -EFSCORRUPTED;
 232         }
 233
 234         /* try read returns without an error or *bpp if it lands in a hole */
 235         if (tp)
 236                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
 237
 238         return 0;
 239 }
 240
 241 void
 242 xfs_dir2_free_hdr_from_disk(
 243         struct xfs_mount                *mp,
 244         struct xfs_dir3_icfree_hdr      *to,
 245         struct xfs_dir2_free            *from)
 246 {
 247         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 248                 struct xfs_dir3_free    *from3 = (struct xfs_dir3_free *)from;
 249
 250                 to->magic = be32_to_cpu(from3->hdr.hdr.magic);
 251                 to->firstdb = be32_to_cpu(from3->hdr.firstdb);
 252                 to->nvalid = be32_to_cpu(from3->hdr.nvalid);
 253                 to->nused = be32_to_cpu(from3->hdr.nused);
 254                 to->bests = from3->bests;
 255
 256                 ASSERT(to->magic == XFS_DIR3_FREE_MAGIC);
 257         } else {
 258                 to->magic = be32_to_cpu(from->hdr.magic);
 259                 to->firstdb = be32_to_cpu(from->hdr.firstdb);
 260                 to->nvalid = be32_to_cpu(from->hdr.nvalid);
 261                 to->nused = be32_to_cpu(from->hdr.nused);
 262                 to->bests = from->bests;
 263
 264                 ASSERT(to->magic == XFS_DIR2_FREE_MAGIC);
 265         }
 266 }
 267
 268 static void
 269 xfs_dir2_free_hdr_to_disk(
 270         struct xfs_mount                *mp,
 271         struct xfs_dir2_free            *to,
 272         struct xfs_dir3_icfree_hdr      *from)
 273 {
 274         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 275                 struct xfs_dir3_free    *to3 = (struct xfs_dir3_free *)to;
 276
 277                 ASSERT(from->magic == XFS_DIR3_FREE_MAGIC);
 278
 279                 to3->hdr.hdr.magic = cpu_to_be32(from->magic);
 280                 to3->hdr.firstdb = cpu_to_be32(from->firstdb);
 281                 to3->hdr.nvalid = cpu_to_be32(from->nvalid);
 282                 to3->hdr.nused = cpu_to_be32(from->nused);
 283         } else {
 284                 ASSERT(from->magic == XFS_DIR2_FREE_MAGIC);
 285
 286                 to->hdr.magic = cpu_to_be32(from->magic);
 287                 to->hdr.firstdb = cpu_to_be32(from->firstdb);
 288                 to->hdr.nvalid = cpu_to_be32(from->nvalid);
 289                 to->hdr.nused = cpu_to_be32(from->nused);
 290         }
 291 }
 292
 293 int
 294 xfs_dir2_free_read(
 295         struct xfs_trans        *tp,
 296         struct xfs_inode        *dp,
 297         xfs_dablk_t             fbno,
 298         struct xfs_buf          **bpp)
 299 {
 300         return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
 301 }
 302
 303 static int
 304 xfs_dir2_free_try_read(
 305         struct xfs_trans        *tp,
 306         struct xfs_inode        *dp,
 307         xfs_dablk_t             fbno,
 308         struct xfs_buf          **bpp)
 309 {
 310         return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
 311 }
 312
 313 static int
 314 xfs_dir3_free_get_buf(
 315         xfs_da_args_t           *args,
 316         xfs_dir2_db_t           fbno,
 317         struct xfs_buf          **bpp)
 318 {
 319         struct xfs_trans        *tp = args->trans;
 320         struct xfs_inode        *dp = args->dp;
 321         struct xfs_mount        *mp = dp->i_mount;
 322         struct xfs_buf          *bp;
 323         int                     error;
 324         struct xfs_dir3_icfree_hdr hdr;
 325
 326         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, fbno),
 327                                    -1, &bp, XFS_DATA_FORK);
 328         if (error)
 329                 return error;
 330
 331         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
 332         bp->b_ops = &xfs_dir3_free_buf_ops;
 333
 334         /*
 335          * Initialize the new block to be empty, and remember
 336          * its first slot as our empty slot.
 337          */
 338         memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
 339         memset(&hdr, 0, sizeof(hdr));
 340
 341         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 342                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 343
 344                 hdr.magic = XFS_DIR3_FREE_MAGIC;
 345
 346                 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
 347                 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
 348                 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_meta_uuid);
 349         } else
 350                 hdr.magic = XFS_DIR2_FREE_MAGIC;
 351         xfs_dir2_free_hdr_to_disk(mp, bp->b_addr, &hdr);
 352         *bpp = bp;
 353         return 0;
 354 }
 355
 356 /*
 357  * Log entries from a freespace block.
 358  */
 359 STATIC void
 360 xfs_dir2_free_log_bests(
 361         struct xfs_da_args      *args,
 362         struct xfs_dir3_icfree_hdr *hdr,
 363         struct xfs_buf          *bp,
 364         int                     first,          /* first entry to log */
 365         int                     last)           /* last entry to log */
 366 {
 367         struct xfs_dir2_free    *free = bp->b_addr;
 368
 369         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 370                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 371         xfs_trans_log_buf(args->trans, bp,
 372                           (char *)&hdr->bests[first] - (char *)free,
 373                           (char *)&hdr->bests[last] - (char *)free +
 374                            sizeof(hdr->bests[0]) - 1);
 375 }
 376
 377 /*
 378  * Log header from a freespace block.
 379  */
 380 static void
 381 xfs_dir2_free_log_header(
 382         struct xfs_da_args      *args,
 383         struct xfs_buf          *bp)
 384 {
 385 #ifdef DEBUG
 386         xfs_dir2_free_t         *free;          /* freespace structure */
 387
 388         free = bp->b_addr;
 389         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 390                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 391 #endif
 392         xfs_trans_log_buf(args->trans, bp, 0,
 393                           args->geo->free_hdr_size - 1);
 394 }
 395
 396 /*
 397  * Convert a leaf-format directory to a node-format directory.
 398  * We need to change the magic number of the leaf block, and copy
 399  * the freespace table out of the leaf block into its own block.
 400  */
 401 int                                             /* error */
 402 xfs_dir2_leaf_to_node(
 403         xfs_da_args_t           *args,          /* operation arguments */
 404         struct xfs_buf          *lbp)           /* leaf buffer */
 405 {
 406         xfs_inode_t             *dp;            /* incore directory inode */
 407         int                     error;          /* error return value */
 408         struct xfs_buf          *fbp;           /* freespace buffer */
 409         xfs_dir2_db_t           fdb;            /* freespace block number */
 410         __be16                  *from;          /* pointer to freespace entry */
 411         int                     i;              /* leaf freespace index */
 412         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 413         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 414         int                     n;              /* count of live freespc ents */
 415         xfs_dir2_data_off_t     off;            /* freespace entry value */
 416         xfs_trans_t             *tp;            /* transaction pointer */
 417         struct xfs_dir3_icfree_hdr freehdr;
 418
 419         trace_xfs_dir2_leaf_to_node(args);
 420
 421         dp = args->dp;
 422         tp = args->trans;
 423         /*
 424          * Add a freespace block to the directory.
 425          */
 426         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 427                 return error;
 428         }
 429         ASSERT(fdb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
 430         /*
 431          * Get the buffer for the new freespace block.
 432          */
 433         error = xfs_dir3_free_get_buf(args, fdb, &fbp);
 434         if (error)
 435                 return error;
 436
 437         xfs_dir2_free_hdr_from_disk(dp->i_mount, &freehdr, fbp->b_addr);
 438         leaf = lbp->b_addr;
 439         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 440         if (be32_to_cpu(ltp->bestcount) >
 441                                 (uint)dp->i_d.di_size / args->geo->blksize) {
 442                 xfs_buf_corruption_error(lbp);
 443                 return -EFSCORRUPTED;
 444         }
 445
 446         /*
 447          * Copy freespace entries from the leaf block to the new block.
 448          * Count active entries.
 449          */
 450         from = xfs_dir2_leaf_bests_p(ltp);
 451         for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++) {
 452                 off = be16_to_cpu(*from);
 453                 if (off != NULLDATAOFF)
 454                         n++;
 455                 freehdr.bests[i] = cpu_to_be16(off);
 456         }
 457
 458         /*
 459          * Now initialize the freespace block header.
 460          */
 461         freehdr.nused = n;
 462         freehdr.nvalid = be32_to_cpu(ltp->bestcount);
 463
 464         xfs_dir2_free_hdr_to_disk(dp->i_mount, fbp->b_addr, &freehdr);
 465         xfs_dir2_free_log_bests(args, &freehdr, fbp, 0, freehdr.nvalid - 1);
 466         xfs_dir2_free_log_header(args, fbp);
 467
 468         /*
 469          * Converting the leaf to a leafnode is just a matter of changing the
 470          * magic number and the ops. Do the change directly to the buffer as
 471          * it's less work (and less code) than decoding the header to host
 472          * format and back again.
 473          */
 474         if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
 475                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 476         else
 477                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 478         lbp->b_ops = &xfs_dir3_leafn_buf_ops;
 479         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
 480         xfs_dir3_leaf_log_header(args, lbp);
 481         xfs_dir3_leaf_check(dp, lbp);
 482         return 0;
 483 }
 484
 485 /*
 486  * Add a leaf entry to a leaf block in a node-form directory.
 487  * The other work necessary is done from the caller.
 488  */
 489 static int                                      /* error */
 490 xfs_dir2_leafn_add(
 491         struct xfs_buf          *bp,            /* leaf buffer */
 492         struct xfs_da_args      *args,          /* operation arguments */
 493         int                     index)          /* insertion pt for new entry */
 494 {
 495         struct xfs_dir3_icleaf_hdr leafhdr;
 496         struct xfs_inode        *dp = args->dp;
 497         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 498         struct xfs_dir2_leaf_entry *lep;
 499         struct xfs_dir2_leaf_entry *ents;
 500         int                     compact;        /* compacting stale leaves */
 501         int                     highstale = 0;  /* next stale entry */
 502         int                     lfloghigh;      /* high leaf entry logging */
 503         int                     lfloglow;       /* low leaf entry logging */
 504         int                     lowstale = 0;   /* previous stale entry */
 505
 506         trace_xfs_dir2_leafn_add(args, index);
 507
 508         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
 509         ents = leafhdr.ents;
 510
 511         /*
 512          * Quick check just to make sure we are not going to index
 513          * into other peoples memory
 514          */
 515         if (index < 0) {
 516                 xfs_buf_corruption_error(bp);
 517                 return -EFSCORRUPTED;
 518         }
 519
 520         /*
 521          * If there are already the maximum number of leaf entries in
 522          * the block, if there are no stale entries it won't fit.
 523          * Caller will do a split.  If there are stale entries we'll do
 524          * a compact.
 525          */
 526
 527         if (leafhdr.count == args->geo->leaf_max_ents) {
 528                 if (!leafhdr.stale)
 529                         return -ENOSPC;
 530                 compact = leafhdr.stale > 1;
 531         } else
 532                 compact = 0;
 533         ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
 534         ASSERT(index == leafhdr.count ||
 535                be32_to_cpu(ents[index].hashval) >= args->hashval);
 536
 537         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 538                 return 0;
 539
 540         /*
 541          * Compact out all but one stale leaf entry.  Leaves behind
 542          * the entry closest to index.
 543          */
 544         if (compact)
 545                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 546                                          &highstale, &lfloglow, &lfloghigh);
 547         else if (leafhdr.stale) {
 548                 /*
 549                  * Set impossible logging indices for this case.
 550                  */
 551                 lfloglow = leafhdr.count;
 552                 lfloghigh = -1;
 553         }
 554
 555         /*
 556          * Insert the new entry, log everything.
 557          */
 558         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 559                                        highstale, &lfloglow, &lfloghigh);
 560
 561         lep->hashval = cpu_to_be32(args->hashval);
 562         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(args->geo,
 563                                 args->blkno, args->index));
 564
 565         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
 566         xfs_dir3_leaf_log_header(args, bp);
 567         xfs_dir3_leaf_log_ents(args, &leafhdr, bp, lfloglow, lfloghigh);
 568         xfs_dir3_leaf_check(dp, bp);
 569         return 0;
 570 }
 571
 572 #ifdef DEBUG
 573 static void
 574 xfs_dir2_free_hdr_check(
 575         struct xfs_inode *dp,
 576         struct xfs_buf  *bp,
 577         xfs_dir2_db_t   db)
 578 {
 579         struct xfs_dir3_icfree_hdr hdr;
 580
 581         xfs_dir2_free_hdr_from_disk(dp->i_mount, &hdr, bp->b_addr);
 582
 583         ASSERT((hdr.firstdb % dp->i_mount->m_dir_geo->free_max_bests) == 0);
 584         ASSERT(hdr.firstdb <= db);
 585         ASSERT(db < hdr.firstdb + hdr.nvalid);
 586 }
 587 #else
 588 #define xfs_dir2_free_hdr_check(dp, bp, db)
 589 #endif  /* DEBUG */
 590
 591 /*
 592  * Return the last hash value in the leaf.
 593  * Stale entries are ok.
 594  */
 595 xfs_dahash_t                                    /* hash value */
 596 xfs_dir2_leaf_lasthash(
 597         struct xfs_inode *dp,
 598         struct xfs_buf  *bp,                    /* leaf buffer */
 599         int             *count)                 /* count of entries in leaf */
 600 {
 601         struct xfs_dir3_icleaf_hdr leafhdr;
 602
 603         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, bp->b_addr);
 604
 605         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 606                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC ||
 607                leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
 608                leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
 609
 610         if (count)
 611                 *count = leafhdr.count;
 612         if (!leafhdr.count)
 613                 return 0;
 614         return be32_to_cpu(leafhdr.ents[leafhdr.count - 1].hashval);
 615 }
 616
 617 /*
 618  * Look up a leaf entry for space to add a name in a node-format leaf block.
 619  * The extrablk in state is a freespace block.
 620  */
 621 STATIC int
 622 xfs_dir2_leafn_lookup_for_addname(
 623         struct xfs_buf          *bp,            /* leaf buffer */
 624         xfs_da_args_t           *args,          /* operation arguments */
 625         int                     *indexp,        /* out: leaf entry index */
 626         xfs_da_state_t          *state)         /* state to fill in */
 627 {
 628         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 629         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 630         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 631         xfs_inode_t             *dp;            /* incore directory inode */
 632         int                     error;          /* error return value */
 633         int                     fi;             /* free entry index */
 634         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 635         int                     index;          /* leaf entry index */
 636         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 637         int                     length;         /* length of new data entry */
 638         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 639         xfs_mount_t             *mp;            /* filesystem mount point */
 640         xfs_dir2_db_t           newdb;          /* new data block number */
 641         xfs_dir2_db_t           newfdb;         /* new free block number */
 642         xfs_trans_t             *tp;            /* transaction pointer */
 643         struct xfs_dir3_icleaf_hdr leafhdr;
 644
 645         dp = args->dp;
 646         tp = args->trans;
 647         mp = dp->i_mount;
 648         leaf = bp->b_addr;
 649         xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, leaf);
 650
 651         xfs_dir3_leaf_check(dp, bp);
 652         ASSERT(leafhdr.count > 0);
 653
 654         /*
 655          * Look up the hash value in the leaf entries.
 656          */
 657         index = xfs_dir2_leaf_search_hash(args, bp);
 658         /*
 659          * Do we have a buffer coming in?
 660          */
 661         if (state->extravalid) {
 662                 /* If so, it's a free block buffer, get the block number. */
 663                 curbp = state->extrablk.bp;
 664                 curfdb = state->extrablk.blkno;
 665                 free = curbp->b_addr;
 666                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 667                        free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 668         }
 669         length = xfs_dir2_data_entsize(mp, args->namelen);
 670         /*
 671          * Loop over leaf entries with the right hash value.
 672          */
 673         for (lep = &leafhdr.ents[index];
 674              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 675              lep++, index++) {
 676                 /*
 677                  * Skip stale leaf entries.
 678                  */
 679                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 680                         continue;
 681                 /*
 682                  * Pull the data block number from the entry.
 683                  */
 684                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 685                                                be32_to_cpu(lep->address));
 686                 /*
 687                  * For addname, we're looking for a place to put the new entry.
 688                  * We want to use a data block with an entry of equal
 689                  * hash value to ours if there is one with room.
 690                  *
 691                  * If this block isn't the data block we already have
 692                  * in hand, take a look at it.
 693                  */
 694                 if (newdb != curdb) {
 695                         struct xfs_dir3_icfree_hdr freehdr;
 696
 697                         curdb = newdb;
 698                         /*
 699                          * Convert the data block to the free block
 700                          * holding its freespace information.
 701                          */
 702                         newfdb = xfs_dir2_db_to_fdb(args->geo, newdb);
 703                         /*
 704                          * If it's not the one we have in hand, read it in.
 705                          */
 706                         if (newfdb != curfdb) {
 707                                 /*
 708                                  * If we had one before, drop it.
 709                                  */
 710                                 if (curbp)
 711                                         xfs_trans_brelse(tp, curbp);
 712
 713                                 error = xfs_dir2_free_read(tp, dp,
 714                                                 xfs_dir2_db_to_da(args->geo,
 715                                                                   newfdb),
 716                                                 &curbp);
 717                                 if (error)
 718                                         return error;
 719                                 free = curbp->b_addr;
 720
 721                                 xfs_dir2_free_hdr_check(dp, curbp, curdb);
 722                         }
 723                         /*
 724                          * Get the index for our entry.
 725                          */
 726                         fi = xfs_dir2_db_to_fdindex(args->geo, curdb);
 727                         /*
 728                          * If it has room, return it.
 729                          */
 730                         xfs_dir2_free_hdr_from_disk(mp, &freehdr, free);
 731                         if (XFS_IS_CORRUPT(mp,
 732                                            freehdr.bests[fi] ==
 733                                            cpu_to_be16(NULLDATAOFF))) {
 734                                 if (curfdb != newfdb)
 735                                         xfs_trans_brelse(tp, curbp);
 736                                 return -EFSCORRUPTED;
 737                         }
 738                         curfdb = newfdb;
 739                         if (be16_to_cpu(freehdr.bests[fi]) >= length)
 740                                 goto out;
 741                 }
 742         }
 743         /* Didn't find any space */
 744         fi = -1;
 745 out:
 746         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 747         if (curbp) {
 748                 /* Giving back a free block. */
 749                 state->extravalid = 1;
 750                 state->extrablk.bp = curbp;
 751                 state->extrablk.index = fi;
 752                 state->extrablk.blkno = curfdb;
 753
 754                 /*
 755                  * Important: this magic number is not in the buffer - it's for
 756                  * buffer type information and therefore only the free/data type
 757                  * matters here, not whether CRCs are enabled or not.
 758                  */
 759                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 760         } else {
 761                 state->extravalid = 0;
 762         }
 763         /*
 764          * Return the index, that will be the insertion point.
 765          */
 766         *indexp = index;
 767         return -ENOENT;
 768 }
 769
 770 /*
 771  * Look up a leaf entry in a node-format leaf block.
 772  * The extrablk in state a data block.
 773  */
 774 STATIC int
 775 xfs_dir2_leafn_lookup_for_entry(
 776         struct xfs_buf          *bp,            /* leaf buffer */
 777         xfs_da_args_t           *args,          /* operation arguments */
 778         int                     *indexp,        /* out: leaf entry index */
 779         xfs_da_state_t          *state)         /* state to fill in */
 780 {
 781         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 782         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 783         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 784         xfs_inode_t             *dp;            /* incore directory inode */
 785         int                     error;          /* error return value */
 786         int                     index;          /* leaf entry index */
 787         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 788         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 789         xfs_mount_t             *mp;            /* filesystem mount point */
 790         xfs_dir2_db_t           newdb;          /* new data block number */
 791         xfs_trans_t             *tp;            /* transaction pointer */
 792         enum xfs_dacmp          cmp;            /* comparison result */
 793         struct xfs_dir3_icleaf_hdr leafhdr;
 794
 795         dp = args->dp;
 796         tp = args->trans;
 797         mp = dp->i_mount;
 798         leaf = bp->b_addr;
 799         xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, leaf);
 800
 801         xfs_dir3_leaf_check(dp, bp);
 802         if (leafhdr.count <= 0) {
 803                 xfs_buf_corruption_error(bp);
 804                 return -EFSCORRUPTED;
 805         }
 806
 807         /*
 808          * Look up the hash value in the leaf entries.
 809          */
 810         index = xfs_dir2_leaf_search_hash(args, bp);
 811         /*
 812          * Do we have a buffer coming in?
 813          */
 814         if (state->extravalid) {
 815                 curbp = state->extrablk.bp;
 816                 curdb = state->extrablk.blkno;
 817         }
 818         /*
 819          * Loop over leaf entries with the right hash value.
 820          */
 821         for (lep = &leafhdr.ents[index];
 822              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 823              lep++, index++) {
 824                 /*
 825                  * Skip stale leaf entries.
 826                  */
 827                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 828                         continue;
 829                 /*
 830                  * Pull the data block number from the entry.
 831                  */
 832                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 833                                                be32_to_cpu(lep->address));
 834                 /*
 835                  * Not adding a new entry, so we really want to find
 836                  * the name given to us.
 837                  *
 838                  * If it's a different data block, go get it.
 839                  */
 840                 if (newdb != curdb) {
 841                         /*
 842                          * If we had a block before that we aren't saving
 843                          * for a CI name, drop it
 844                          */
 845                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 846                                                 curdb != state->extrablk.blkno))
 847                                 xfs_trans_brelse(tp, curbp);
 848                         /*
 849                          * If needing the block that is saved with a CI match,
 850                          * use it otherwise read in the new data block.
 851                          */
 852                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 853                                         newdb == state->extrablk.blkno) {
 854                                 ASSERT(state->extravalid);
 855                                 curbp = state->extrablk.bp;
 856                         } else {
 857                                 error = xfs_dir3_data_read(tp, dp,
 858                                                 xfs_dir2_db_to_da(args->geo,
 859                                                                   newdb),
 860                                                 -1, &curbp);
 861                                 if (error)
 862                                         return error;
 863                         }
 864                         xfs_dir3_data_check(dp, curbp);
 865                         curdb = newdb;
 866                 }
 867                 /*
 868                  * Point to the data entry.
 869                  */
 870                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 871                         xfs_dir2_dataptr_to_off(args->geo,
 872                                                 be32_to_cpu(lep->address)));
 873                 /*
 874                  * Compare the entry and if it's an exact match, return
 875                  * EEXIST immediately. If it's the first case-insensitive
 876                  * match, store the block & inode number and continue looking.
 877                  */
 878                 cmp = xfs_dir2_compname(args, dep->name, dep->namelen);
 879                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 880                         /* If there is a CI match block, drop it */
 881                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 882                                                 curdb != state->extrablk.blkno)
 883                                 xfs_trans_brelse(tp, state->extrablk.bp);
 884                         args->cmpresult = cmp;
 885                         args->inumber = be64_to_cpu(dep->inumber);
 886                         args->filetype = xfs_dir2_data_get_ftype(mp, dep);
 887                         *indexp = index;
 888                         state->extravalid = 1;
 889                         state->extrablk.bp = curbp;
 890                         state->extrablk.blkno = curdb;
 891                         state->extrablk.index = (int)((char *)dep -
 892                                                         (char *)curbp->b_addr);
 893                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 894                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 895                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 896                         if (cmp == XFS_CMP_EXACT)
 897                                 return -EEXIST;
 898                 }
 899         }
 900         ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
 901         if (curbp) {
 902                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 903                         /* Giving back last used data block. */
 904                         state->extravalid = 1;
 905                         state->extrablk.bp = curbp;
 906                         state->extrablk.index = -1;
 907                         state->extrablk.blkno = curdb;
 908                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 909                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 910                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 911                 } else {
 912                         /* If the curbp is not the CI match block, drop it */
 913                         if (state->extrablk.bp != curbp)
 914                                 xfs_trans_brelse(tp, curbp);
 915                 }
 916         } else {
 917                 state->extravalid = 0;
 918         }
 919         *indexp = index;
 920         return -ENOENT;
 921 }
 922
 923 /*
 924  * Look up a leaf entry in a node-format leaf block.
 925  * If this is an addname then the extrablk in state is a freespace block,
 926  * otherwise it's a data block.
 927  */
 928 int
 929 xfs_dir2_leafn_lookup_int(
 930         struct xfs_buf          *bp,            /* leaf buffer */
 931         xfs_da_args_t           *args,          /* operation arguments */
 932         int                     *indexp,        /* out: leaf entry index */
 933         xfs_da_state_t          *state)         /* state to fill in */
 934 {
 935         if (args->op_flags & XFS_DA_OP_ADDNAME)
 936                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 937                                                         state);
 938         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 939 }
 940
 941 /*
 942  * Move count leaf entries from source to destination leaf.
 943  * Log entries and headers.  Stale entries are preserved.
 944  */
 945 static void
 946 xfs_dir3_leafn_moveents(
 947         xfs_da_args_t                   *args,  /* operation arguments */
 948         struct xfs_buf                  *bp_s,  /* source */
 949         struct xfs_dir3_icleaf_hdr      *shdr,
 950         struct xfs_dir2_leaf_entry      *sents,
 951         int                             start_s,/* source leaf index */
 952         struct xfs_buf                  *bp_d,  /* destination */
 953         struct xfs_dir3_icleaf_hdr      *dhdr,
 954         struct xfs_dir2_leaf_entry      *dents,
 955         int                             start_d,/* destination leaf index */
 956         int                             count)  /* count of leaves to copy */
 957 {
 958         int                             stale;  /* count stale leaves copied */
 959
 960         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 961
 962         /*
 963          * Silently return if nothing to do.
 964          */
 965         if (count == 0)
 966                 return;
 967
 968         /*
 969          * If the destination index is not the end of the current
 970          * destination leaf entries, open up a hole in the destination
 971          * to hold the new entries.
 972          */
 973         if (start_d < dhdr->count) {
 974                 memmove(&dents[start_d + count], &dents[start_d],
 975                         (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
 976                 xfs_dir3_leaf_log_ents(args, dhdr, bp_d, start_d + count,
 977                                        count + dhdr->count - 1);
 978         }
 979         /*
 980          * If the source has stale leaves, count the ones in the copy range
 981          * so we can update the header correctly.
 982          */
 983         if (shdr->stale) {
 984                 int     i;                      /* temp leaf index */
 985
 986                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 987                         if (sents[i].address ==
 988                                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 989                                 stale++;
 990                 }
 991         } else
 992                 stale = 0;
 993         /*
 994          * Copy the leaf entries from source to destination.
 995          */
 996         memcpy(&dents[start_d], &sents[start_s],
 997                 count * sizeof(xfs_dir2_leaf_entry_t));
 998         xfs_dir3_leaf_log_ents(args, dhdr, bp_d, start_d, start_d + count - 1);
 999
1000         /*
1001          * If there are source entries after the ones we copied,
1002          * delete the ones we copied by sliding the next ones down.
1003          */
1004         if (start_s + count < shdr->count) {
1005                 memmove(&sents[start_s], &sents[start_s + count],
1006                         count * sizeof(xfs_dir2_leaf_entry_t));
1007                 xfs_dir3_leaf_log_ents(args, shdr, bp_s, start_s,
1008                                        start_s + count - 1);
1009         }
1010
1011         /*
1012          * Update the headers and log them.
1013          */
1014         shdr->count -= count;
1015         shdr->stale -= stale;
1016         dhdr->count += count;
1017         dhdr->stale += stale;
1018 }
1019
1020 /*
1021  * Determine the sort order of two leaf blocks.
1022  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
1023  */
1024 int                                             /* sort order */
1025 xfs_dir2_leafn_order(
1026         struct xfs_inode        *dp,
1027         struct xfs_buf          *leaf1_bp,              /* leaf1 buffer */
1028         struct xfs_buf          *leaf2_bp)              /* leaf2 buffer */
1029 {
1030         struct xfs_dir2_leaf    *leaf1 = leaf1_bp->b_addr;
1031         struct xfs_dir2_leaf    *leaf2 = leaf2_bp->b_addr;
1032         struct xfs_dir2_leaf_entry *ents1;
1033         struct xfs_dir2_leaf_entry *ents2;
1034         struct xfs_dir3_icleaf_hdr hdr1;
1035         struct xfs_dir3_icleaf_hdr hdr2;
1036
1037         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &hdr1, leaf1);
1038         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &hdr2, leaf2);
1039         ents1 = hdr1.ents;
1040         ents2 = hdr2.ents;
1041
1042         if (hdr1.count > 0 && hdr2.count > 0 &&
1043             (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
1044              be32_to_cpu(ents2[hdr2.count - 1].hashval) <
1045                                 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
1046                 return 1;
1047         return 0;
1048 }
1049
1050 /*
1051  * Rebalance leaf entries between two leaf blocks.
1052  * This is actually only called when the second block is new,
1053  * though the code deals with the general case.
1054  * A new entry will be inserted in one of the blocks, and that
1055  * entry is taken into account when balancing.
1056  */
1057 static void
1058 xfs_dir2_leafn_rebalance(
1059         xfs_da_state_t          *state,         /* btree cursor */
1060         xfs_da_state_blk_t      *blk1,          /* first btree block */
1061         xfs_da_state_blk_t      *blk2)          /* second btree block */
1062 {
1063         xfs_da_args_t           *args;          /* operation arguments */
1064         int                     count;          /* count (& direction) leaves */
1065         int                     isleft;         /* new goes in left leaf */
1066         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
1067         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
1068         int                     mid;            /* midpoint leaf index */
1069 #if defined(DEBUG) || defined(XFS_WARN)
1070         int                     oldstale;       /* old count of stale leaves */
1071 #endif
1072         int                     oldsum;         /* old total leaf count */
1073         int                     swap_blocks;    /* swapped leaf blocks */
1074         struct xfs_dir2_leaf_entry *ents1;
1075         struct xfs_dir2_leaf_entry *ents2;
1076         struct xfs_dir3_icleaf_hdr hdr1;
1077         struct xfs_dir3_icleaf_hdr hdr2;
1078         struct xfs_inode        *dp = state->args->dp;
1079
1080         args = state->args;
1081         /*
1082          * If the block order is wrong, swap the arguments.
1083          */
1084         swap_blocks = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp);
1085         if (swap_blocks)
1086                 swap(blk1, blk2);
1087
1088         leaf1 = blk1->bp->b_addr;
1089         leaf2 = blk2->bp->b_addr;
1090         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &hdr1, leaf1);
1091         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &hdr2, leaf2);
1092         ents1 = hdr1.ents;
1093         ents2 = hdr2.ents;
1094
1095         oldsum = hdr1.count + hdr2.count;
1096 #if defined(DEBUG) || defined(XFS_WARN)
1097         oldstale = hdr1.stale + hdr2.stale;
1098 #endif
1099         mid = oldsum >> 1;
1100
1101         /*
1102          * If the old leaf count was odd then the new one will be even,
1103          * so we need to divide the new count evenly.
1104          */
1105         if (oldsum & 1) {
1106                 xfs_dahash_t    midhash;        /* middle entry hash value */
1107
1108                 if (mid >= hdr1.count)
1109                         midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1110                 else
1111                         midhash = be32_to_cpu(ents1[mid].hashval);
1112                 isleft = args->hashval <= midhash;
1113         }
1114         /*
1115          * If the old count is even then the new count is odd, so there's
1116          * no preferred side for the new entry.
1117          * Pick the left one.
1118          */
1119         else
1120                 isleft = 1;
1121         /*
1122          * Calculate moved entry count.  Positive means left-to-right,
1123          * negative means right-to-left.  Then move the entries.
1124          */
1125         count = hdr1.count - mid + (isleft == 0);
1126         if (count > 0)
1127                 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1128                                         hdr1.count - count, blk2->bp,
1129                                         &hdr2, ents2, 0, count);
1130         else if (count < 0)
1131                 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1132                                         blk1->bp, &hdr1, ents1,
1133                                         hdr1.count, count);
1134
1135         ASSERT(hdr1.count + hdr2.count == oldsum);
1136         ASSERT(hdr1.stale + hdr2.stale == oldstale);
1137
1138         /* log the changes made when moving the entries */
1139         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf1, &hdr1);
1140         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf2, &hdr2);
1141         xfs_dir3_leaf_log_header(args, blk1->bp);
1142         xfs_dir3_leaf_log_header(args, blk2->bp);
1143
1144         xfs_dir3_leaf_check(dp, blk1->bp);
1145         xfs_dir3_leaf_check(dp, blk2->bp);
1146
1147         /*
1148          * Mark whether we're inserting into the old or new leaf.
1149          */
1150         if (hdr1.count < hdr2.count)
1151                 state->inleaf = swap_blocks;
1152         else if (hdr1.count > hdr2.count)
1153                 state->inleaf = !swap_blocks;
1154         else
1155                 state->inleaf = swap_blocks ^ (blk1->index <= hdr1.count);
1156         /*
1157          * Adjust the expected index for insertion.
1158          */
1159         if (!state->inleaf)
1160                 blk2->index = blk1->index - hdr1.count;
1161
1162         /*
1163          * Finally sanity check just to make sure we are not returning a
1164          * negative index
1165          */
1166         if (blk2->index < 0) {
1167                 state->inleaf = 1;
1168                 blk2->index = 0;
1169                 xfs_alert(dp->i_mount,
1170         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1171                         __func__, blk1->index);
1172         }
1173 }
1174
1175 static int
1176 xfs_dir3_data_block_free(
1177         xfs_da_args_t           *args,
1178         struct xfs_dir2_data_hdr *hdr,
1179         struct xfs_dir2_free    *free,
1180         xfs_dir2_db_t           fdb,
1181         int                     findex,
1182         struct xfs_buf          *fbp,
1183         int                     longest)
1184 {
1185         int                     logfree = 0;
1186         struct xfs_dir3_icfree_hdr freehdr;
1187         struct xfs_inode        *dp = args->dp;
1188
1189         xfs_dir2_free_hdr_from_disk(dp->i_mount, &freehdr, free);
1190         if (hdr) {
1191                 /*
1192                  * Data block is not empty, just set the free entry to the new
1193                  * value.
1194                  */
1195                 freehdr.bests[findex] = cpu_to_be16(longest);
1196                 xfs_dir2_free_log_bests(args, &freehdr, fbp, findex, findex);
1197                 return 0;
1198         }
1199
1200         /* One less used entry in the free table. */
1201         freehdr.nused--;
1202
1203         /*
1204          * If this was the last entry in the table, we can trim the table size
1205          * back.  There might be other entries at the end referring to
1206          * non-existent data blocks, get those too.
1207          */
1208         if (findex == freehdr.nvalid - 1) {
1209                 int     i;              /* free entry index */
1210
1211                 for (i = findex - 1; i >= 0; i--) {
1212                         if (freehdr.bests[i] != cpu_to_be16(NULLDATAOFF))
1213                                 break;
1214                 }
1215                 freehdr.nvalid = i + 1;
1216                 logfree = 0;
1217         } else {
1218                 /* Not the last entry, just punch it out.  */
1219                 freehdr.bests[findex] = cpu_to_be16(NULLDATAOFF);
1220                 logfree = 1;
1221         }
1222
1223         xfs_dir2_free_hdr_to_disk(dp->i_mount, free, &freehdr);
1224         xfs_dir2_free_log_header(args, fbp);
1225
1226         /*
1227          * If there are no useful entries left in the block, get rid of the
1228          * block if we can.
1229          */
1230         if (!freehdr.nused) {
1231                 int error;
1232
1233                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1234                 if (error == 0) {
1235                         fbp = NULL;
1236                         logfree = 0;
1237                 } else if (error != -ENOSPC || args->total != 0)
1238                         return error;
1239                 /*
1240                  * It's possible to get ENOSPC if there is no
1241                  * space reservation.  In this case some one
1242                  * else will eventually get rid of this block.
1243                  */
1244         }
1245
1246         /* Log the free entry that changed, unless we got rid of it.  */
1247         if (logfree)
1248                 xfs_dir2_free_log_bests(args, &freehdr, fbp, findex, findex);
1249         return 0;
1250 }
1251
1252 /*
1253  * Remove an entry from a node directory.
1254  * This removes the leaf entry and the data entry,
1255  * and updates the free block if necessary.
1256  */
1257 static int                                      /* error */
1258 xfs_dir2_leafn_remove(
1259         xfs_da_args_t           *args,          /* operation arguments */
1260         struct xfs_buf          *bp,            /* leaf buffer */
1261         int                     index,          /* leaf entry index */
1262         xfs_da_state_blk_t      *dblk,          /* data block */
1263         int                     *rval)          /* resulting block needs join */
1264 {
1265         struct xfs_da_geometry  *geo = args->geo;
1266         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1267         xfs_dir2_db_t           db;             /* data block number */
1268         struct xfs_buf          *dbp;           /* data block buffer */
1269         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1270         xfs_inode_t             *dp;            /* incore directory inode */
1271         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1272         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1273         int                     longest;        /* longest data free entry */
1274         int                     off;            /* data block entry offset */
1275         int                     needlog;        /* need to log data header */
1276         int                     needscan;       /* need to rescan data frees */
1277         xfs_trans_t             *tp;            /* transaction pointer */
1278         struct xfs_dir2_data_free *bf;          /* bestfree table */
1279         struct xfs_dir3_icleaf_hdr leafhdr;
1280
1281         trace_xfs_dir2_leafn_remove(args, index);
1282
1283         dp = args->dp;
1284         tp = args->trans;
1285         leaf = bp->b_addr;
1286         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
1287
1288         /*
1289          * Point to the entry we're removing.
1290          */
1291         lep = &leafhdr.ents[index];
1292
1293         /*
1294          * Extract the data block and offset from the entry.
1295          */
1296         db = xfs_dir2_dataptr_to_db(geo, be32_to_cpu(lep->address));
1297         ASSERT(dblk->blkno == db);
1298         off = xfs_dir2_dataptr_to_off(geo, be32_to_cpu(lep->address));
1299         ASSERT(dblk->index == off);
1300
1301         /*
1302          * Kill the leaf entry by marking it stale.
1303          * Log the leaf block changes.
1304          */
1305         leafhdr.stale++;
1306         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
1307         xfs_dir3_leaf_log_header(args, bp);
1308
1309         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1310         xfs_dir3_leaf_log_ents(args, &leafhdr, bp, index, index);
1311
1312         /*
1313          * Make the data entry free.  Keep track of the longest freespace
1314          * in the data block in case it changes.
1315          */
1316         dbp = dblk->bp;
1317         hdr = dbp->b_addr;
1318         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1319         bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1320         longest = be16_to_cpu(bf[0].length);
1321         needlog = needscan = 0;
1322         xfs_dir2_data_make_free(args, dbp, off,
1323                 xfs_dir2_data_entsize(dp->i_mount, dep->namelen), &needlog,
1324                 &needscan);
1325         /*
1326          * Rescan the data block freespaces for bestfree.
1327          * Log the data block header if needed.
1328          */
1329         if (needscan)
1330                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
1331         if (needlog)
1332                 xfs_dir2_data_log_header(args, dbp);
1333         xfs_dir3_data_check(dp, dbp);
1334         /*
1335          * If the longest data block freespace changes, need to update
1336          * the corresponding freeblock entry.
1337          */
1338         if (longest < be16_to_cpu(bf[0].length)) {
1339                 int             error;          /* error return value */
1340                 struct xfs_buf  *fbp;           /* freeblock buffer */
1341                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1342                 int             findex;         /* index in freeblock entries */
1343                 xfs_dir2_free_t *free;          /* freeblock structure */
1344
1345                 /*
1346                  * Convert the data block number to a free block,
1347                  * read in the free block.
1348                  */
1349                 fdb = xfs_dir2_db_to_fdb(geo, db);
1350                 error = xfs_dir2_free_read(tp, dp, xfs_dir2_db_to_da(geo, fdb),
1351                                            &fbp);
1352                 if (error)
1353                         return error;
1354                 free = fbp->b_addr;
1355 #ifdef DEBUG
1356         {
1357                 struct xfs_dir3_icfree_hdr freehdr;
1358
1359                 xfs_dir2_free_hdr_from_disk(dp->i_mount, &freehdr, free);
1360                 ASSERT(freehdr.firstdb == geo->free_max_bests *
1361                         (fdb - xfs_dir2_byte_to_db(geo, XFS_DIR2_FREE_OFFSET)));
1362         }
1363 #endif
1364                 /*
1365                  * Calculate which entry we need to fix.
1366                  */
1367                 findex = xfs_dir2_db_to_fdindex(geo, db);
1368                 longest = be16_to_cpu(bf[0].length);
1369                 /*
1370                  * If the data block is now empty we can get rid of it
1371                  * (usually).
1372                  */
1373                 if (longest == geo->blksize - geo->data_entry_offset) {
1374                         /*
1375                          * Try to punch out the data block.
1376                          */
1377                         error = xfs_dir2_shrink_inode(args, db, dbp);
1378                         if (error == 0) {
1379                                 dblk->bp = NULL;
1380                                 hdr = NULL;
1381                         }
1382                         /*
1383                          * We can get ENOSPC if there's no space reservation.
1384                          * In this case just drop the buffer and some one else
1385                          * will eventually get rid of the empty block.
1386                          */
1387                         else if (!(error == -ENOSPC && args->total == 0))
1388                                 return error;
1389                 }
1390                 /*
1391                  * If we got rid of the data block, we can eliminate that entry
1392                  * in the free block.
1393                  */
1394                 error = xfs_dir3_data_block_free(args, hdr, free,
1395                                                  fdb, findex, fbp, longest);
1396                 if (error)
1397                         return error;
1398         }
1399
1400         xfs_dir3_leaf_check(dp, bp);
1401         /*
1402          * Return indication of whether this leaf block is empty enough
1403          * to justify trying to join it with a neighbor.
1404          */
1405         *rval = (geo->leaf_hdr_size +
1406                  (uint)sizeof(leafhdr.ents) * (leafhdr.count - leafhdr.stale)) <
1407                 geo->magicpct;
1408         return 0;
1409 }
1410
1411 /*
1412  * Split the leaf entries in the old block into old and new blocks.
1413  */
1414 int                                             /* error */
1415 xfs_dir2_leafn_split(
1416         xfs_da_state_t          *state,         /* btree cursor */
1417         xfs_da_state_blk_t      *oldblk,        /* original block */
1418         xfs_da_state_blk_t      *newblk)        /* newly created block */
1419 {
1420         xfs_da_args_t           *args;          /* operation arguments */
1421         xfs_dablk_t             blkno;          /* new leaf block number */
1422         int                     error;          /* error return value */
1423         struct xfs_inode        *dp;
1424
1425         /*
1426          * Allocate space for a new leaf node.
1427          */
1428         args = state->args;
1429         dp = args->dp;
1430         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1431         error = xfs_da_grow_inode(args, &blkno);
1432         if (error) {
1433                 return error;
1434         }
1435         /*
1436          * Initialize the new leaf block.
1437          */
1438         error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(args->geo, blkno),
1439                                       &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1440         if (error)
1441                 return error;
1442
1443         newblk->blkno = blkno;
1444         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1445         /*
1446          * Rebalance the entries across the two leaves, link the new
1447          * block into the leaves.
1448          */
1449         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1450         error = xfs_da3_blk_link(state, oldblk, newblk);
1451         if (error) {
1452                 return error;
1453         }
1454         /*
1455          * Insert the new entry in the correct block.
1456          */
1457         if (state->inleaf)
1458                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1459         else
1460                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1461         /*
1462          * Update last hashval in each block since we added the name.
1463          */
1464         oldblk->hashval = xfs_dir2_leaf_lasthash(dp, oldblk->bp, NULL);
1465         newblk->hashval = xfs_dir2_leaf_lasthash(dp, newblk->bp, NULL);
1466         xfs_dir3_leaf_check(dp, oldblk->bp);
1467         xfs_dir3_leaf_check(dp, newblk->bp);
1468         return error;
1469 }
1470
1471 /*
1472  * Check a leaf block and its neighbors to see if the block should be
1473  * collapsed into one or the other neighbor.  Always keep the block
1474  * with the smaller block number.
1475  * If the current block is over 50% full, don't try to join it, return 0.
1476  * If the block is empty, fill in the state structure and return 2.
1477  * If it can be collapsed, fill in the state structure and return 1.
1478  * If nothing can be done, return 0.
1479  */
1480 int                                             /* error */
1481 xfs_dir2_leafn_toosmall(
1482         xfs_da_state_t          *state,         /* btree cursor */
1483         int                     *action)        /* resulting action to take */
1484 {
1485         xfs_da_state_blk_t      *blk;           /* leaf block */
1486         xfs_dablk_t             blkno;          /* leaf block number */
1487         struct xfs_buf          *bp;            /* leaf buffer */
1488         int                     bytes;          /* bytes in use */
1489         int                     count;          /* leaf live entry count */
1490         int                     error;          /* error return value */
1491         int                     forward;        /* sibling block direction */
1492         int                     i;              /* sibling counter */
1493         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1494         int                     rval;           /* result from path_shift */
1495         struct xfs_dir3_icleaf_hdr leafhdr;
1496         struct xfs_dir2_leaf_entry *ents;
1497         struct xfs_inode        *dp = state->args->dp;
1498
1499         /*
1500          * Check for the degenerate case of the block being over 50% full.
1501          * If so, it's not worth even looking to see if we might be able
1502          * to coalesce with a sibling.
1503          */
1504         blk = &state->path.blk[state->path.active - 1];
1505         leaf = blk->bp->b_addr;
1506         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
1507         ents = leafhdr.ents;
1508         xfs_dir3_leaf_check(dp, blk->bp);
1509
1510         count = leafhdr.count - leafhdr.stale;
1511         bytes = state->args->geo->leaf_hdr_size + count * sizeof(ents[0]);
1512         if (bytes > (state->args->geo->blksize >> 1)) {
1513                 /*
1514                  * Blk over 50%, don't try to join.
1515                  */
1516                 *action = 0;
1517                 return 0;
1518         }
1519         /*
1520          * Check for the degenerate case of the block being empty.
1521          * If the block is empty, we'll simply delete it, no need to
1522          * coalesce it with a sibling block.  We choose (arbitrarily)
1523          * to merge with the forward block unless it is NULL.
1524          */
1525         if (count == 0) {
1526                 /*
1527                  * Make altpath point to the block we want to keep and
1528                  * path point to the block we want to drop (this one).
1529                  */
1530                 forward = (leafhdr.forw != 0);
1531                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1532                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1533                         &rval);
1534                 if (error)
1535                         return error;
1536                 *action = rval ? 2 : 0;
1537                 return 0;
1538         }
1539         /*
1540          * Examine each sibling block to see if we can coalesce with
1541          * at least 25% free space to spare.  We need to figure out
1542          * whether to merge with the forward or the backward block.
1543          * We prefer coalescing with the lower numbered sibling so as
1544          * to shrink a directory over time.
1545          */
1546         forward = leafhdr.forw < leafhdr.back;
1547         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1548                 struct xfs_dir3_icleaf_hdr hdr2;
1549
1550                 blkno = forward ? leafhdr.forw : leafhdr.back;
1551                 if (blkno == 0)
1552                         continue;
1553                 /*
1554                  * Read the sibling leaf block.
1555                  */
1556                 error = xfs_dir3_leafn_read(state->args->trans, dp,
1557                                             blkno, -1, &bp);
1558                 if (error)
1559                         return error;
1560
1561                 /*
1562                  * Count bytes in the two blocks combined.
1563                  */
1564                 count = leafhdr.count - leafhdr.stale;
1565                 bytes = state->args->geo->blksize -
1566                         (state->args->geo->blksize >> 2);
1567
1568                 leaf = bp->b_addr;
1569                 xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &hdr2, leaf);
1570                 ents = hdr2.ents;
1571                 count += hdr2.count - hdr2.stale;
1572                 bytes -= count * sizeof(ents[0]);
1573
1574                 /*
1575                  * Fits with at least 25% to spare.
1576                  */
1577                 if (bytes >= 0)
1578                         break;
1579                 xfs_trans_brelse(state->args->trans, bp);
1580         }
1581         /*
1582          * Didn't like either block, give up.
1583          */
1584         if (i >= 2) {
1585                 *action = 0;
1586                 return 0;
1587         }
1588
1589         /*
1590          * Make altpath point to the block we want to keep (the lower
1591          * numbered block) and path point to the block we want to drop.
1592          */
1593         memcpy(&state->altpath, &state->path, sizeof(state->path));
1594         if (blkno < blk->blkno)
1595                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1596                         &rval);
1597         else
1598                 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1599                         &rval);
1600         if (error) {
1601                 return error;
1602         }
1603         *action = rval ? 0 : 1;
1604         return 0;
1605 }
1606
1607 /*
1608  * Move all the leaf entries from drop_blk to save_blk.
1609  * This is done as part of a join operation.
1610  */
1611 void
1612 xfs_dir2_leafn_unbalance(
1613         xfs_da_state_t          *state,         /* cursor */
1614         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1615         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1616 {
1617         xfs_da_args_t           *args;          /* operation arguments */
1618         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1619         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1620         struct xfs_dir3_icleaf_hdr savehdr;
1621         struct xfs_dir3_icleaf_hdr drophdr;
1622         struct xfs_dir2_leaf_entry *sents;
1623         struct xfs_dir2_leaf_entry *dents;
1624         struct xfs_inode        *dp = state->args->dp;
1625
1626         args = state->args;
1627         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1628         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1629         drop_leaf = drop_blk->bp->b_addr;
1630         save_leaf = save_blk->bp->b_addr;
1631
1632         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &savehdr, save_leaf);
1633         xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &drophdr, drop_leaf);
1634         sents = savehdr.ents;
1635         dents = drophdr.ents;
1636
1637         /*
1638          * If there are any stale leaf entries, take this opportunity
1639          * to purge them.
1640          */
1641         if (drophdr.stale)
1642                 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1643         if (savehdr.stale)
1644                 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1645
1646         /*
1647          * Move the entries from drop to the appropriate end of save.
1648          */
1649         drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1650         if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1651                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1652                                         save_blk->bp, &savehdr, sents, 0,
1653                                         drophdr.count);
1654         else
1655                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1656                                         save_blk->bp, &savehdr, sents,
1657                                         savehdr.count, drophdr.count);
1658         save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1659
1660         /* log the changes made when moving the entries */
1661         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, save_leaf, &savehdr);
1662         xfs_dir2_leaf_hdr_to_disk(dp->i_mount, drop_leaf, &drophdr);
1663         xfs_dir3_leaf_log_header(args, save_blk->bp);
1664         xfs_dir3_leaf_log_header(args, drop_blk->bp);
1665
1666         xfs_dir3_leaf_check(dp, save_blk->bp);
1667         xfs_dir3_leaf_check(dp, drop_blk->bp);
1668 }
1669
1670 /*
1671  * Add a new data block to the directory at the free space index that the caller
1672  * has specified.
1673  */
1674 static int
1675 xfs_dir2_node_add_datablk(
1676         struct xfs_da_args      *args,
1677         struct xfs_da_state_blk *fblk,
1678         xfs_dir2_db_t           *dbno,
1679         struct xfs_buf          **dbpp,
1680         struct xfs_buf          **fbpp,
1681         struct xfs_dir3_icfree_hdr *hdr,
1682         int                     *findex)
1683 {
1684         struct xfs_inode        *dp = args->dp;
1685         struct xfs_trans        *tp = args->trans;
1686         struct xfs_mount        *mp = dp->i_mount;
1687         struct xfs_dir2_data_free *bf;
1688         xfs_dir2_db_t           fbno;
1689         struct xfs_buf          *fbp;
1690         struct xfs_buf          *dbp;
1691         int                     error;
1692
1693         /* Not allowed to allocate, return failure. */
1694         if (args->total == 0)
1695                 return -ENOSPC;
1696
1697         /* Allocate and initialize the new data block.  */
1698         error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, dbno);
1699         if (error)
1700                 return error;
1701         error = xfs_dir3_data_init(args, *dbno, &dbp);
1702         if (error)
1703                 return error;
1704
1705         /*
1706          * Get the freespace block corresponding to the data block
1707          * that was just allocated.
1708          */
1709         fbno = xfs_dir2_db_to_fdb(args->geo, *dbno);
1710         error = xfs_dir2_free_try_read(tp, dp,
1711                                xfs_dir2_db_to_da(args->geo, fbno), &fbp);
1712         if (error)
1713                 return error;
1714
1715         /*
1716          * If there wasn't a freespace block, the read will
1717          * return a NULL fbp.  Allocate and initialize a new one.
1718          */
1719         if (!fbp) {
1720                 error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fbno);
1721                 if (error)
1722                         return error;
1723
1724                 if (XFS_IS_CORRUPT(mp,
1725                                    xfs_dir2_db_to_fdb(args->geo, *dbno) !=
1726                                    fbno)) {
1727                         xfs_alert(mp,
1728 "%s: dir ino %llu needed freesp block %lld for data block %lld, got %lld",
1729                                 __func__, (unsigned long long)dp->i_ino,
1730                                 (long long)xfs_dir2_db_to_fdb(args->geo, *dbno),
1731                                 (long long)*dbno, (long long)fbno);
1732                         if (fblk) {
1733                                 xfs_alert(mp,
1734                         " fblk "PTR_FMT" blkno %llu index %d magic 0x%x",
1735                                         fblk, (unsigned long long)fblk->blkno,
1736                                         fblk->index, fblk->magic);
1737                         } else {
1738                                 xfs_alert(mp, " ... fblk is NULL");
1739                         }
1740                         return -EFSCORRUPTED;
1741                 }
1742
1743                 /* Get a buffer for the new block. */
1744                 error = xfs_dir3_free_get_buf(args, fbno, &fbp);
1745                 if (error)
1746                         return error;
1747                 xfs_dir2_free_hdr_from_disk(mp, hdr, fbp->b_addr);
1748
1749                 /* Remember the first slot as our empty slot. */
1750                 hdr->firstdb = (fbno - xfs_dir2_byte_to_db(args->geo,
1751                                                         XFS_DIR2_FREE_OFFSET)) *
1752                                 args->geo->free_max_bests;
1753         } else {
1754                 xfs_dir2_free_hdr_from_disk(mp, hdr, fbp->b_addr);
1755         }
1756
1757         /* Set the freespace block index from the data block number. */
1758         *findex = xfs_dir2_db_to_fdindex(args->geo, *dbno);
1759
1760         /* Extend the freespace table if the new data block is off the end. */
1761         if (*findex >= hdr->nvalid) {
1762                 ASSERT(*findex < args->geo->free_max_bests);
1763                 hdr->nvalid = *findex + 1;
1764                 hdr->bests[*findex] = cpu_to_be16(NULLDATAOFF);
1765         }
1766
1767         /*
1768          * If this entry was for an empty data block (this should always be
1769          * true) then update the header.
1770          */
1771         if (hdr->bests[*findex] == cpu_to_be16(NULLDATAOFF)) {
1772                 hdr->nused++;
1773                 xfs_dir2_free_hdr_to_disk(mp, fbp->b_addr, hdr);
1774                 xfs_dir2_free_log_header(args, fbp);
1775         }
1776
1777         /* Update the freespace value for the new block in the table. */
1778         bf = xfs_dir2_data_bestfree_p(mp, dbp->b_addr);
1779         hdr->bests[*findex] = bf[0].length;
1780
1781         *dbpp = dbp;
1782         *fbpp = fbp;
1783         return 0;
1784 }
1785
1786 static int
1787 xfs_dir2_node_find_freeblk(
1788         struct xfs_da_args      *args,
1789         struct xfs_da_state_blk *fblk,
1790         xfs_dir2_db_t           *dbnop,
1791         struct xfs_buf          **fbpp,
1792         struct xfs_dir3_icfree_hdr *hdr,
1793         int                     *findexp,
1794         int                     length)
1795 {
1796         struct xfs_inode        *dp = args->dp;
1797         struct xfs_trans        *tp = args->trans;
1798         struct xfs_buf          *fbp = NULL;
1799         xfs_dir2_db_t           firstfbno;
1800         xfs_dir2_db_t           lastfbno;
1801         xfs_dir2_db_t           ifbno = -1;
1802         xfs_dir2_db_t           dbno = -1;
1803         xfs_dir2_db_t           fbno;
1804         xfs_fileoff_t           fo;
1805         int                     findex = 0;
1806         int                     error;
1807
1808         /*
1809          * If we came in with a freespace block that means that lookup
1810          * found an entry with our hash value.  This is the freespace
1811          * block for that data entry.
1812          */
1813         if (fblk) {
1814                 fbp = fblk->bp;
1815                 findex = fblk->index;
1816                 xfs_dir2_free_hdr_from_disk(dp->i_mount, hdr, fbp->b_addr);
1817                 if (findex >= 0) {
1818                         /* caller already found the freespace for us. */
1819                         ASSERT(findex < hdr->nvalid);
1820                         ASSERT(be16_to_cpu(hdr->bests[findex]) != NULLDATAOFF);
1821                         ASSERT(be16_to_cpu(hdr->bests[findex]) >= length);
1822                         dbno = hdr->firstdb + findex;
1823                         goto found_block;
1824                 }
1825
1826                 /*
1827                  * The data block looked at didn't have enough room.
1828                  * We'll start at the beginning of the freespace entries.
1829                  */
1830                 ifbno = fblk->blkno;
1831                 xfs_trans_brelse(tp, fbp);
1832                 fbp = NULL;
1833                 fblk->bp = NULL;
1834         }
1835
1836         /*
1837          * If we don't have a data block yet, we're going to scan the freespace
1838          * data for a data block with enough free space in it.
1839          */
1840         error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK);
1841         if (error)
1842                 return error;
1843         lastfbno = xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo);
1844         firstfbno = xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET);
1845
1846         for (fbno = lastfbno - 1; fbno >= firstfbno; fbno--) {
1847                 /* If it's ifbno we already looked at it. */
1848                 if (fbno == ifbno)
1849                         continue;
1850
1851                 /*
1852                  * Read the block.  There can be holes in the freespace blocks,
1853                  * so this might not succeed.  This should be really rare, so
1854                  * there's no reason to avoid it.
1855                  */
1856                 error = xfs_dir2_free_try_read(tp, dp,
1857                                 xfs_dir2_db_to_da(args->geo, fbno),
1858                                 &fbp);
1859                 if (error)
1860                         return error;
1861                 if (!fbp)
1862                         continue;
1863
1864                 xfs_dir2_free_hdr_from_disk(dp->i_mount, hdr, fbp->b_addr);
1865
1866                 /* Scan the free entry array for a large enough free space. */
1867                 for (findex = hdr->nvalid - 1; findex >= 0; findex--) {
1868                         if (be16_to_cpu(hdr->bests[findex]) != NULLDATAOFF &&
1869                             be16_to_cpu(hdr->bests[findex]) >= length) {
1870                                 dbno = hdr->firstdb + findex;
1871                                 goto found_block;
1872                         }
1873                 }
1874
1875                 /* Didn't find free space, go on to next free block */
1876                 xfs_trans_brelse(tp, fbp);
1877         }
1878
1879 found_block:
1880         *dbnop = dbno;
1881         *fbpp = fbp;
1882         *findexp = findex;
1883         return 0;
1884 }
1885
1886 /*
1887  * Add the data entry for a node-format directory name addition.
1888  * The leaf entry is added in xfs_dir2_leafn_add.
1889  * We may enter with a freespace block that the lookup found.
1890  */
1891 static int
1892 xfs_dir2_node_addname_int(
1893         struct xfs_da_args      *args,          /* operation arguments */
1894         struct xfs_da_state_blk *fblk)          /* optional freespace block */
1895 {
1896         struct xfs_dir2_data_unused *dup;       /* data unused entry pointer */
1897         struct xfs_dir2_data_entry *dep;        /* data entry pointer */
1898         struct xfs_dir2_data_hdr *hdr;          /* data block header */
1899         struct xfs_dir2_data_free *bf;
1900         struct xfs_trans        *tp = args->trans;
1901         struct xfs_inode        *dp = args->dp;
1902         struct xfs_dir3_icfree_hdr freehdr;
1903         struct xfs_buf          *dbp;           /* data block buffer */
1904         struct xfs_buf          *fbp;           /* freespace buffer */
1905         xfs_dir2_data_aoff_t    aoff;
1906         xfs_dir2_db_t           dbno;           /* data block number */
1907         int                     error;          /* error return value */
1908         int                     findex;         /* freespace entry index */
1909         int                     length;         /* length of the new entry */
1910         int                     logfree = 0;    /* need to log free entry */
1911         int                     needlog = 0;    /* need to log data header */
1912         int                     needscan = 0;   /* need to rescan data frees */
1913         __be16                  *tagp;          /* data entry tag pointer */
1914
1915         length = xfs_dir2_data_entsize(dp->i_mount, args->namelen);
1916         error = xfs_dir2_node_find_freeblk(args, fblk, &dbno, &fbp, &freehdr,
1917                                            &findex, length);
1918         if (error)
1919                 return error;
1920
1921         /*
1922          * Now we know if we must allocate blocks, so if we are checking whether
1923          * we can insert without allocation then we can return now.
1924          */
1925         if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
1926                 if (dbno == -1)
1927                         return -ENOSPC;
1928                 return 0;
1929         }
1930
1931         /*
1932          * If we don't have a data block, we need to allocate one and make
1933          * the freespace entries refer to it.
1934          */
1935         if (dbno == -1) {
1936                 /* we're going to have to log the free block index later */
1937                 logfree = 1;
1938                 error = xfs_dir2_node_add_datablk(args, fblk, &dbno, &dbp, &fbp,
1939                                                   &freehdr, &findex);
1940         } else {
1941                 /* Read the data block in. */
1942                 error = xfs_dir3_data_read(tp, dp,
1943                                            xfs_dir2_db_to_da(args->geo, dbno),
1944                                            -1, &dbp);
1945         }
1946         if (error)
1947                 return error;
1948
1949         /* setup for data block up now */
1950         hdr = dbp->b_addr;
1951         bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1952         ASSERT(be16_to_cpu(bf[0].length) >= length);
1953
1954         /* Point to the existing unused space. */
1955         dup = (xfs_dir2_data_unused_t *)
1956               ((char *)hdr + be16_to_cpu(bf[0].offset));
1957
1958         /* Mark the first part of the unused space, inuse for us. */
1959         aoff = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
1960         error = xfs_dir2_data_use_free(args, dbp, dup, aoff, length,
1961                         &needlog, &needscan);
1962         if (error) {
1963                 xfs_trans_brelse(tp, dbp);
1964                 return error;
1965         }
1966
1967         /* Fill in the new entry and log it. */
1968         dep = (xfs_dir2_data_entry_t *)dup;
1969         dep->inumber = cpu_to_be64(args->inumber);
1970         dep->namelen = args->namelen;
1971         memcpy(dep->name, args->name, dep->namelen);
1972         xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
1973         tagp = xfs_dir2_data_entry_tag_p(dp->i_mount, dep);
1974         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1975         xfs_dir2_data_log_entry(args, dbp, dep);
1976
1977         /* Rescan the freespace and log the data block if needed. */
1978         if (needscan)
1979                 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
1980         if (needlog)
1981                 xfs_dir2_data_log_header(args, dbp);
1982
1983         /* If the freespace block entry is now wrong, update it. */
1984         if (freehdr.bests[findex] != bf[0].length) {
1985                 freehdr.bests[findex] = bf[0].length;
1986                 logfree = 1;
1987         }
1988
1989         /* Log the freespace entry if needed. */
1990         if (logfree)
1991                 xfs_dir2_free_log_bests(args, &freehdr, fbp, findex, findex);
1992
1993         /* Return the data block and offset in args. */
1994         args->blkno = (xfs_dablk_t)dbno;
1995         args->index = be16_to_cpu(*tagp);
1996         return 0;
1997 }
1998
1999 /*
2000  * Top-level node form directory addname routine.
2001  */
2002 int                                             /* error */
2003 xfs_dir2_node_addname(
2004         xfs_da_args_t           *args)          /* operation arguments */
2005 {
2006         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
2007         int                     error;          /* error return value */
2008         int                     rval;           /* sub-return value */
2009         xfs_da_state_t          *state;         /* btree cursor */
2010
2011         trace_xfs_dir2_node_addname(args);
2012
2013         /*
2014          * Allocate and initialize the state (btree cursor).
2015          */
2016         state = xfs_da_state_alloc();
2017         state->args = args;
2018         state->mp = args->dp->i_mount;
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();
2088         state->args = args;
2089         state->mp = args->dp->i_mount;
2090         /*
2091          * Fill in the path to the entry in the cursor.
2092          */
2093         error = xfs_da3_node_lookup_int(state, &rval);
2094         if (error)
2095                 rval = error;
2096         else if (rval == -ENOENT && args->cmpresult == XFS_CMP_CASE) {
2097                 /* If a CI match, dup the actual name and return -EEXIST */
2098                 xfs_dir2_data_entry_t   *dep;
2099
2100                 dep = (xfs_dir2_data_entry_t *)
2101                         ((char *)state->extrablk.bp->b_addr +
2102                                                  state->extrablk.index);
2103                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2104         }
2105         /*
2106          * Release the btree blocks and leaf block.
2107          */
2108         for (i = 0; i < state->path.active; i++) {
2109                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2110                 state->path.blk[i].bp = NULL;
2111         }
2112         /*
2113          * Release the data block if we have it.
2114          */
2115         if (state->extravalid && state->extrablk.bp) {
2116                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2117                 state->extrablk.bp = NULL;
2118         }
2119         xfs_da_state_free(state);
2120         return rval;
2121 }
2122
2123 /*
2124  * Remove an entry from a node-format directory.
2125  */
2126 int                                             /* error */
2127 xfs_dir2_node_removename(
2128         struct xfs_da_args      *args)          /* operation arguments */
2129 {
2130         struct xfs_da_state_blk *blk;           /* leaf block */
2131         int                     error;          /* error return value */
2132         int                     rval;           /* operation return value */
2133         struct xfs_da_state     *state;         /* btree cursor */
2134
2135         trace_xfs_dir2_node_removename(args);
2136
2137         /*
2138          * Allocate and initialize the btree cursor.
2139          */
2140         state = xfs_da_state_alloc();
2141         state->args = args;
2142         state->mp = args->dp->i_mount;
2143
2144         /* Look up the entry we're deleting, set up the cursor. */
2145         error = xfs_da3_node_lookup_int(state, &rval);
2146         if (error)
2147                 goto out_free;
2148
2149         /* Didn't find it, upper layer screwed up. */
2150         if (rval != -EEXIST) {
2151                 error = rval;
2152                 goto out_free;
2153         }
2154
2155         blk = &state->path.blk[state->path.active - 1];
2156         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2157         ASSERT(state->extravalid);
2158         /*
2159          * Remove the leaf and data entries.
2160          * Extrablk refers to the data block.
2161          */
2162         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2163                 &state->extrablk, &rval);
2164         if (error)
2165                 goto out_free;
2166         /*
2167          * Fix the hash values up the btree.
2168          */
2169         xfs_da3_fixhashpath(state, &state->path);
2170         /*
2171          * If we need to join leaf blocks, do it.
2172          */
2173         if (rval && state->path.active > 1)
2174                 error = xfs_da3_join(state);
2175         /*
2176          * If no errors so far, try conversion to leaf format.
2177          */
2178         if (!error)
2179                 error = xfs_dir2_node_to_leaf(state);
2180 out_free:
2181         xfs_da_state_free(state);
2182         return error;
2183 }
2184
2185 /*
2186  * Replace an entry's inode number in a node-format directory.
2187  */
2188 int                                             /* error */
2189 xfs_dir2_node_replace(
2190         xfs_da_args_t           *args)          /* operation arguments */
2191 {
2192         xfs_da_state_blk_t      *blk;           /* leaf block */
2193         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
2194         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
2195         int                     error;          /* error return value */
2196         int                     i;              /* btree level */
2197         xfs_ino_t               inum;           /* new inode number */
2198         int                     ftype;          /* new file type */
2199         int                     rval;           /* internal return value */
2200         xfs_da_state_t          *state;         /* btree cursor */
2201
2202         trace_xfs_dir2_node_replace(args);
2203
2204         /*
2205          * Allocate and initialize the btree cursor.
2206          */
2207         state = xfs_da_state_alloc();
2208         state->args = args;
2209         state->mp = args->dp->i_mount;
2210
2211         /*
2212          * We have to save new inode number and ftype since
2213          * xfs_da3_node_lookup_int() is going to overwrite them
2214          */
2215         inum = args->inumber;
2216         ftype = args->filetype;
2217
2218         /*
2219          * Lookup the entry to change in the btree.
2220          */
2221         error = xfs_da3_node_lookup_int(state, &rval);
2222         if (error) {
2223                 rval = error;
2224         }
2225         /*
2226          * It should be found, since the vnodeops layer has looked it up
2227          * and locked it.  But paranoia is good.
2228          */
2229         if (rval == -EEXIST) {
2230                 struct xfs_dir3_icleaf_hdr      leafhdr;
2231
2232                 /*
2233                  * Find the leaf entry.
2234                  */
2235                 blk = &state->path.blk[state->path.active - 1];
2236                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2237                 ASSERT(state->extravalid);
2238
2239                 xfs_dir2_leaf_hdr_from_disk(state->mp, &leafhdr,
2240                                             blk->bp->b_addr);
2241                 /*
2242                  * Point to the data entry.
2243                  */
2244                 hdr = state->extrablk.bp->b_addr;
2245                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2246                        hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2247                 dep = (xfs_dir2_data_entry_t *)
2248                       ((char *)hdr +
2249                        xfs_dir2_dataptr_to_off(args->geo,
2250                                 be32_to_cpu(leafhdr.ents[blk->index].address)));
2251                 ASSERT(inum != be64_to_cpu(dep->inumber));
2252                 /*
2253                  * Fill in the new inode number and log the entry.
2254                  */
2255                 dep->inumber = cpu_to_be64(inum);
2256                 xfs_dir2_data_put_ftype(state->mp, dep, ftype);
2257                 xfs_dir2_data_log_entry(args, state->extrablk.bp, dep);
2258                 rval = 0;
2259         }
2260         /*
2261          * Didn't find it, and we're holding a data block.  Drop it.
2262          */
2263         else if (state->extravalid) {
2264                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2265                 state->extrablk.bp = NULL;
2266         }
2267         /*
2268          * Release all the buffers in the cursor.
2269          */
2270         for (i = 0; i < state->path.active; i++) {
2271                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2272                 state->path.blk[i].bp = NULL;
2273         }
2274         xfs_da_state_free(state);
2275         return rval;
2276 }
2277
2278 /*
2279  * Trim off a trailing empty freespace block.
2280  * Return (in rvalp) 1 if we did it, 0 if not.
2281  */
2282 int                                             /* error */
2283 xfs_dir2_node_trim_free(
2284         xfs_da_args_t           *args,          /* operation arguments */
2285         xfs_fileoff_t           fo,             /* free block number */
2286         int                     *rvalp)         /* out: did something */
2287 {
2288         struct xfs_buf          *bp;            /* freespace buffer */
2289         xfs_inode_t             *dp;            /* incore directory inode */
2290         int                     error;          /* error return code */
2291         xfs_dir2_free_t         *free;          /* freespace structure */
2292         xfs_trans_t             *tp;            /* transaction pointer */
2293         struct xfs_dir3_icfree_hdr freehdr;
2294
2295         dp = args->dp;
2296         tp = args->trans;
2297
2298         *rvalp = 0;
2299
2300         /*
2301          * Read the freespace block.
2302          */
2303         error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2304         if (error)
2305                 return error;
2306         /*
2307          * There can be holes in freespace.  If fo is a hole, there's
2308          * nothing to do.
2309          */
2310         if (!bp)
2311                 return 0;
2312         free = bp->b_addr;
2313         xfs_dir2_free_hdr_from_disk(dp->i_mount, &freehdr, free);
2314
2315         /*
2316          * If there are used entries, there's nothing to do.
2317          */
2318         if (freehdr.nused > 0) {
2319                 xfs_trans_brelse(tp, bp);
2320                 return 0;
2321         }
2322         /*
2323          * Blow the block away.
2324          */
2325         error = xfs_dir2_shrink_inode(args,
2326                         xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo), bp);
2327         if (error) {
2328                 /*
2329                  * Can't fail with ENOSPC since that only happens with no
2330                  * space reservation, when breaking up an extent into two
2331                  * pieces.  This is the last block of an extent.
2332                  */
2333                 ASSERT(error != -ENOSPC);
2334                 xfs_trans_brelse(tp, bp);
2335                 return error;
2336         }
2337         /*
2338          * Return that we succeeded.
2339          */
2340         *rvalp = 1;
2341         return 0;
2342 }