fs/xfs/libxfs/xfs_dir2_node.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   4  * Copyright (c) 2013 Red Hat, Inc.
   5  * All Rights Reserved.
   6  */
   7 #include "xfs.h"
   8 #include "xfs_fs.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_mount.h"
  13 #include "xfs_da_format.h"
  14 #include "xfs_da_btree.h"
  15 #include "xfs_inode.h"
  16 #include "xfs_bmap.h"
  17 #include "xfs_dir2.h"
  18 #include "xfs_dir2_priv.h"
  19 #include "xfs_error.h"
  20 #include "xfs_trace.h"
  21 #include "xfs_trans.h"
  22 #include "xfs_buf_item.h"
  23 #include "xfs_cksum.h"
  24 #include "xfs_log.h"
  25
  26 /*
  27  * Function declarations.
  28  */
  29 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
  30                               int index);
  31 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
  32                                      xfs_da_state_blk_t *blk1,
  33                                      xfs_da_state_blk_t *blk2);
  34 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
  35                                  int index, xfs_da_state_blk_t *dblk,
  36                                  int *rval);
  37 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
  38                                      xfs_da_state_blk_t *fblk);
  39
  40 /*
  41  * Check internal consistency of a leafn block.
  42  */
  43 #ifdef DEBUG
  44 static xfs_failaddr_t
  45 xfs_dir3_leafn_check(
  46         struct xfs_inode        *dp,
  47         struct xfs_buf          *bp)
  48 {
  49         struct xfs_dir2_leaf    *leaf = bp->b_addr;
  50         struct xfs_dir3_icleaf_hdr leafhdr;
  51
  52         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
  53
  54         if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
  55                 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
  56                 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
  57                         return __this_address;
  58         } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
  59                 return __this_address;
  60
  61         return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
  62 }
  63
  64 static inline void
  65 xfs_dir3_leaf_check(
  66         struct xfs_inode        *dp,
  67         struct xfs_buf          *bp)
  68 {
  69         xfs_failaddr_t          fa;
  70
  71         fa = xfs_dir3_leafn_check(dp, bp);
  72         if (!fa)
  73                 return;
  74         xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
  75                         bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
  76                         fa);
  77         ASSERT(0);
  78 }
  79 #else
  80 #define xfs_dir3_leaf_check(dp, bp)
  81 #endif
  82
  83 static xfs_failaddr_t
  84 xfs_dir3_free_verify(
  85         struct xfs_buf          *bp)
  86 {
  87         struct xfs_mount        *mp = bp->b_target->bt_mount;
  88         struct xfs_dir2_free_hdr *hdr = bp->b_addr;
  89
  90         if (!xfs_verify_magic(bp, hdr->magic))
  91                 return __this_address;
  92
  93         if (xfs_sb_version_hascrc(&mp->m_sb)) {
  94                 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
  95
  96                 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
  97                         return __this_address;
  98                 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
  99                         return __this_address;
 100                 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
 101                         return __this_address;
 102         }
 103
 104         /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
 105
 106         return NULL;
 107 }
 108
 109 static void
 110 xfs_dir3_free_read_verify(
 111         struct xfs_buf  *bp)
 112 {
 113         struct xfs_mount        *mp = bp->b_target->bt_mount;
 114         xfs_failaddr_t          fa;
 115
 116         if (xfs_sb_version_hascrc(&mp->m_sb) &&
 117             !xfs_buf_verify_cksum(bp, XFS_DIR3_FREE_CRC_OFF))
 118                 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
 119         else {
 120                 fa = xfs_dir3_free_verify(bp);
 121                 if (fa)
 122                         xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 123         }
 124 }
 125
 126 static void
 127 xfs_dir3_free_write_verify(
 128         struct xfs_buf  *bp)
 129 {
 130         struct xfs_mount        *mp = bp->b_target->bt_mount;
 131         struct xfs_buf_log_item *bip = bp->b_log_item;
 132         struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
 133         xfs_failaddr_t          fa;
 134
 135         fa = xfs_dir3_free_verify(bp);
 136         if (fa) {
 137                 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
 138                 return;
 139         }
 140
 141         if (!xfs_sb_version_hascrc(&mp->m_sb))
 142                 return;
 143
 144         if (bip)
 145                 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
 146
 147         xfs_buf_update_cksum(bp, XFS_DIR3_FREE_CRC_OFF);
 148 }
 149
 150 const struct xfs_buf_ops xfs_dir3_free_buf_ops = {
 151         .name = "xfs_dir3_free",
 152         .magic = { cpu_to_be32(XFS_DIR2_FREE_MAGIC),
 153                    cpu_to_be32(XFS_DIR3_FREE_MAGIC) },
 154         .verify_read = xfs_dir3_free_read_verify,
 155         .verify_write = xfs_dir3_free_write_verify,
 156         .verify_struct = xfs_dir3_free_verify,
 157 };
 158
 159 /* Everything ok in the free block header? */
 160 static xfs_failaddr_t
 161 xfs_dir3_free_header_check(
 162         struct xfs_inode        *dp,
 163         xfs_dablk_t             fbno,
 164         struct xfs_buf          *bp)
 165 {
 166         struct xfs_mount        *mp = dp->i_mount;
 167         unsigned int            firstdb;
 168         int                     maxbests;
 169
 170         maxbests = dp->d_ops->free_max_bests(mp->m_dir_geo);
 171         firstdb = (xfs_dir2_da_to_db(mp->m_dir_geo, fbno) -
 172                    xfs_dir2_byte_to_db(mp->m_dir_geo, XFS_DIR2_FREE_OFFSET)) *
 173                         maxbests;
 174         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 175                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 176
 177                 if (be32_to_cpu(hdr3->firstdb) != firstdb)
 178                         return __this_address;
 179                 if (be32_to_cpu(hdr3->nvalid) > maxbests)
 180                         return __this_address;
 181                 if (be32_to_cpu(hdr3->nvalid) < be32_to_cpu(hdr3->nused))
 182                         return __this_address;
 183         } else {
 184                 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
 185
 186                 if (be32_to_cpu(hdr->firstdb) != firstdb)
 187                         return __this_address;
 188                 if (be32_to_cpu(hdr->nvalid) > maxbests)
 189                         return __this_address;
 190                 if (be32_to_cpu(hdr->nvalid) < be32_to_cpu(hdr->nused))
 191                         return __this_address;
 192         }
 193         return NULL;
 194 }
 195
 196 static int
 197 __xfs_dir3_free_read(
 198         struct xfs_trans        *tp,
 199         struct xfs_inode        *dp,
 200         xfs_dablk_t             fbno,
 201         xfs_daddr_t             mappedbno,
 202         struct xfs_buf          **bpp)
 203 {
 204         xfs_failaddr_t          fa;
 205         int                     err;
 206
 207         err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
 208                                 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
 209         if (err || !*bpp)
 210                 return err;
 211
 212         /* Check things that we can't do in the verifier. */
 213         fa = xfs_dir3_free_header_check(dp, fbno, *bpp);
 214         if (fa) {
 215                 xfs_verifier_error(*bpp, -EFSCORRUPTED, fa);
 216                 xfs_trans_brelse(tp, *bpp);
 217                 return -EFSCORRUPTED;
 218         }
 219
 220         /* try read returns without an error or *bpp if it lands in a hole */
 221         if (tp)
 222                 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
 223
 224         return 0;
 225 }
 226
 227 int
 228 xfs_dir2_free_read(
 229         struct xfs_trans        *tp,
 230         struct xfs_inode        *dp,
 231         xfs_dablk_t             fbno,
 232         struct xfs_buf          **bpp)
 233 {
 234         return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
 235 }
 236
 237 static int
 238 xfs_dir2_free_try_read(
 239         struct xfs_trans        *tp,
 240         struct xfs_inode        *dp,
 241         xfs_dablk_t             fbno,
 242         struct xfs_buf          **bpp)
 243 {
 244         return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
 245 }
 246
 247 static int
 248 xfs_dir3_free_get_buf(
 249         xfs_da_args_t           *args,
 250         xfs_dir2_db_t           fbno,
 251         struct xfs_buf          **bpp)
 252 {
 253         struct xfs_trans        *tp = args->trans;
 254         struct xfs_inode        *dp = args->dp;
 255         struct xfs_mount        *mp = dp->i_mount;
 256         struct xfs_buf          *bp;
 257         int                     error;
 258         struct xfs_dir3_icfree_hdr hdr;
 259
 260         error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, fbno),
 261                                    -1, &bp, XFS_DATA_FORK);
 262         if (error)
 263                 return error;
 264
 265         xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
 266         bp->b_ops = &xfs_dir3_free_buf_ops;
 267
 268         /*
 269          * Initialize the new block to be empty, and remember
 270          * its first slot as our empty slot.
 271          */
 272         memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
 273         memset(&hdr, 0, sizeof(hdr));
 274
 275         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 276                 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
 277
 278                 hdr.magic = XFS_DIR3_FREE_MAGIC;
 279
 280                 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
 281                 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
 282                 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_meta_uuid);
 283         } else
 284                 hdr.magic = XFS_DIR2_FREE_MAGIC;
 285         dp->d_ops->free_hdr_to_disk(bp->b_addr, &hdr);
 286         *bpp = bp;
 287         return 0;
 288 }
 289
 290 /*
 291  * Log entries from a freespace block.
 292  */
 293 STATIC void
 294 xfs_dir2_free_log_bests(
 295         struct xfs_da_args      *args,
 296         struct xfs_buf          *bp,
 297         int                     first,          /* first entry to log */
 298         int                     last)           /* last entry to log */
 299 {
 300         xfs_dir2_free_t         *free;          /* freespace structure */
 301         __be16                  *bests;
 302
 303         free = bp->b_addr;
 304         bests = args->dp->d_ops->free_bests_p(free);
 305         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 306                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 307         xfs_trans_log_buf(args->trans, bp,
 308                 (uint)((char *)&bests[first] - (char *)free),
 309                 (uint)((char *)&bests[last] - (char *)free +
 310                        sizeof(bests[0]) - 1));
 311 }
 312
 313 /*
 314  * Log header from a freespace block.
 315  */
 316 static void
 317 xfs_dir2_free_log_header(
 318         struct xfs_da_args      *args,
 319         struct xfs_buf          *bp)
 320 {
 321 #ifdef DEBUG
 322         xfs_dir2_free_t         *free;          /* freespace structure */
 323
 324         free = bp->b_addr;
 325         ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 326                free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 327 #endif
 328         xfs_trans_log_buf(args->trans, bp, 0,
 329                           args->dp->d_ops->free_hdr_size - 1);
 330 }
 331
 332 /*
 333  * Convert a leaf-format directory to a node-format directory.
 334  * We need to change the magic number of the leaf block, and copy
 335  * the freespace table out of the leaf block into its own block.
 336  */
 337 int                                             /* error */
 338 xfs_dir2_leaf_to_node(
 339         xfs_da_args_t           *args,          /* operation arguments */
 340         struct xfs_buf          *lbp)           /* leaf buffer */
 341 {
 342         xfs_inode_t             *dp;            /* incore directory inode */
 343         int                     error;          /* error return value */
 344         struct xfs_buf          *fbp;           /* freespace buffer */
 345         xfs_dir2_db_t           fdb;            /* freespace block number */
 346         xfs_dir2_free_t         *free;          /* freespace structure */
 347         __be16                  *from;          /* pointer to freespace entry */
 348         int                     i;              /* leaf freespace index */
 349         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 350         xfs_dir2_leaf_tail_t    *ltp;           /* leaf tail structure */
 351         int                     n;              /* count of live freespc ents */
 352         xfs_dir2_data_off_t     off;            /* freespace entry value */
 353         __be16                  *to;            /* pointer to freespace entry */
 354         xfs_trans_t             *tp;            /* transaction pointer */
 355         struct xfs_dir3_icfree_hdr freehdr;
 356
 357         trace_xfs_dir2_leaf_to_node(args);
 358
 359         dp = args->dp;
 360         tp = args->trans;
 361         /*
 362          * Add a freespace block to the directory.
 363          */
 364         if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
 365                 return error;
 366         }
 367         ASSERT(fdb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
 368         /*
 369          * Get the buffer for the new freespace block.
 370          */
 371         error = xfs_dir3_free_get_buf(args, fdb, &fbp);
 372         if (error)
 373                 return error;
 374
 375         free = fbp->b_addr;
 376         dp->d_ops->free_hdr_from_disk(&freehdr, free);
 377         leaf = lbp->b_addr;
 378         ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
 379         if (be32_to_cpu(ltp->bestcount) >
 380                                 (uint)dp->i_d.di_size / args->geo->blksize)
 381                 return -EFSCORRUPTED;
 382
 383         /*
 384          * Copy freespace entries from the leaf block to the new block.
 385          * Count active entries.
 386          */
 387         from = xfs_dir2_leaf_bests_p(ltp);
 388         to = dp->d_ops->free_bests_p(free);
 389         for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
 390                 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
 391                         n++;
 392                 *to = cpu_to_be16(off);
 393         }
 394
 395         /*
 396          * Now initialize the freespace block header.
 397          */
 398         freehdr.nused = n;
 399         freehdr.nvalid = be32_to_cpu(ltp->bestcount);
 400
 401         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
 402         xfs_dir2_free_log_bests(args, fbp, 0, freehdr.nvalid - 1);
 403         xfs_dir2_free_log_header(args, fbp);
 404
 405         /*
 406          * Converting the leaf to a leafnode is just a matter of changing the
 407          * magic number and the ops. Do the change directly to the buffer as
 408          * it's less work (and less code) than decoding the header to host
 409          * format and back again.
 410          */
 411         if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
 412                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
 413         else
 414                 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
 415         lbp->b_ops = &xfs_dir3_leafn_buf_ops;
 416         xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
 417         xfs_dir3_leaf_log_header(args, lbp);
 418         xfs_dir3_leaf_check(dp, lbp);
 419         return 0;
 420 }
 421
 422 /*
 423  * Add a leaf entry to a leaf block in a node-form directory.
 424  * The other work necessary is done from the caller.
 425  */
 426 static int                                      /* error */
 427 xfs_dir2_leafn_add(
 428         struct xfs_buf          *bp,            /* leaf buffer */
 429         struct xfs_da_args      *args,          /* operation arguments */
 430         int                     index)          /* insertion pt for new entry */
 431 {
 432         struct xfs_dir3_icleaf_hdr leafhdr;
 433         struct xfs_inode        *dp = args->dp;
 434         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 435         struct xfs_dir2_leaf_entry *lep;
 436         struct xfs_dir2_leaf_entry *ents;
 437         int                     compact;        /* compacting stale leaves */
 438         int                     highstale = 0;  /* next stale entry */
 439         int                     lfloghigh;      /* high leaf entry logging */
 440         int                     lfloglow;       /* low leaf entry logging */
 441         int                     lowstale = 0;   /* previous stale entry */
 442
 443         trace_xfs_dir2_leafn_add(args, index);
 444
 445         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 446         ents = dp->d_ops->leaf_ents_p(leaf);
 447
 448         /*
 449          * Quick check just to make sure we are not going to index
 450          * into other peoples memory
 451          */
 452         if (index < 0)
 453                 return -EFSCORRUPTED;
 454
 455         /*
 456          * If there are already the maximum number of leaf entries in
 457          * the block, if there are no stale entries it won't fit.
 458          * Caller will do a split.  If there are stale entries we'll do
 459          * a compact.
 460          */
 461
 462         if (leafhdr.count == dp->d_ops->leaf_max_ents(args->geo)) {
 463                 if (!leafhdr.stale)
 464                         return -ENOSPC;
 465                 compact = leafhdr.stale > 1;
 466         } else
 467                 compact = 0;
 468         ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
 469         ASSERT(index == leafhdr.count ||
 470                be32_to_cpu(ents[index].hashval) >= args->hashval);
 471
 472         if (args->op_flags & XFS_DA_OP_JUSTCHECK)
 473                 return 0;
 474
 475         /*
 476          * Compact out all but one stale leaf entry.  Leaves behind
 477          * the entry closest to index.
 478          */
 479         if (compact)
 480                 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
 481                                          &highstale, &lfloglow, &lfloghigh);
 482         else if (leafhdr.stale) {
 483                 /*
 484                  * Set impossible logging indices for this case.
 485                  */
 486                 lfloglow = leafhdr.count;
 487                 lfloghigh = -1;
 488         }
 489
 490         /*
 491          * Insert the new entry, log everything.
 492          */
 493         lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
 494                                        highstale, &lfloglow, &lfloghigh);
 495
 496         lep->hashval = cpu_to_be32(args->hashval);
 497         lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(args->geo,
 498                                 args->blkno, args->index));
 499
 500         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
 501         xfs_dir3_leaf_log_header(args, bp);
 502         xfs_dir3_leaf_log_ents(args, bp, lfloglow, lfloghigh);
 503         xfs_dir3_leaf_check(dp, bp);
 504         return 0;
 505 }
 506
 507 #ifdef DEBUG
 508 static void
 509 xfs_dir2_free_hdr_check(
 510         struct xfs_inode *dp,
 511         struct xfs_buf  *bp,
 512         xfs_dir2_db_t   db)
 513 {
 514         struct xfs_dir3_icfree_hdr hdr;
 515
 516         dp->d_ops->free_hdr_from_disk(&hdr, bp->b_addr);
 517
 518         ASSERT((hdr.firstdb %
 519                 dp->d_ops->free_max_bests(dp->i_mount->m_dir_geo)) == 0);
 520         ASSERT(hdr.firstdb <= db);
 521         ASSERT(db < hdr.firstdb + hdr.nvalid);
 522 }
 523 #else
 524 #define xfs_dir2_free_hdr_check(dp, bp, db)
 525 #endif  /* DEBUG */
 526
 527 /*
 528  * Return the last hash value in the leaf.
 529  * Stale entries are ok.
 530  */
 531 xfs_dahash_t                                    /* hash value */
 532 xfs_dir2_leaf_lasthash(
 533         struct xfs_inode *dp,
 534         struct xfs_buf  *bp,                    /* leaf buffer */
 535         int             *count)                 /* count of entries in leaf */
 536 {
 537         struct xfs_dir2_leaf    *leaf = bp->b_addr;
 538         struct xfs_dir2_leaf_entry *ents;
 539         struct xfs_dir3_icleaf_hdr leafhdr;
 540
 541         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 542
 543         ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
 544                leafhdr.magic == XFS_DIR3_LEAFN_MAGIC ||
 545                leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
 546                leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
 547
 548         if (count)
 549                 *count = leafhdr.count;
 550         if (!leafhdr.count)
 551                 return 0;
 552
 553         ents = dp->d_ops->leaf_ents_p(leaf);
 554         return be32_to_cpu(ents[leafhdr.count - 1].hashval);
 555 }
 556
 557 /*
 558  * Look up a leaf entry for space to add a name in a node-format leaf block.
 559  * The extrablk in state is a freespace block.
 560  */
 561 STATIC int
 562 xfs_dir2_leafn_lookup_for_addname(
 563         struct xfs_buf          *bp,            /* leaf buffer */
 564         xfs_da_args_t           *args,          /* operation arguments */
 565         int                     *indexp,        /* out: leaf entry index */
 566         xfs_da_state_t          *state)         /* state to fill in */
 567 {
 568         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 569         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 570         xfs_dir2_db_t           curfdb = -1;    /* current free block number */
 571         xfs_inode_t             *dp;            /* incore directory inode */
 572         int                     error;          /* error return value */
 573         int                     fi;             /* free entry index */
 574         xfs_dir2_free_t         *free = NULL;   /* free block structure */
 575         int                     index;          /* leaf entry index */
 576         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 577         int                     length;         /* length of new data entry */
 578         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 579         xfs_mount_t             *mp;            /* filesystem mount point */
 580         xfs_dir2_db_t           newdb;          /* new data block number */
 581         xfs_dir2_db_t           newfdb;         /* new free block number */
 582         xfs_trans_t             *tp;            /* transaction pointer */
 583         struct xfs_dir2_leaf_entry *ents;
 584         struct xfs_dir3_icleaf_hdr leafhdr;
 585
 586         dp = args->dp;
 587         tp = args->trans;
 588         mp = dp->i_mount;
 589         leaf = bp->b_addr;
 590         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 591         ents = dp->d_ops->leaf_ents_p(leaf);
 592
 593         xfs_dir3_leaf_check(dp, bp);
 594         ASSERT(leafhdr.count > 0);
 595
 596         /*
 597          * Look up the hash value in the leaf entries.
 598          */
 599         index = xfs_dir2_leaf_search_hash(args, bp);
 600         /*
 601          * Do we have a buffer coming in?
 602          */
 603         if (state->extravalid) {
 604                 /* If so, it's a free block buffer, get the block number. */
 605                 curbp = state->extrablk.bp;
 606                 curfdb = state->extrablk.blkno;
 607                 free = curbp->b_addr;
 608                 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
 609                        free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
 610         }
 611         length = dp->d_ops->data_entsize(args->namelen);
 612         /*
 613          * Loop over leaf entries with the right hash value.
 614          */
 615         for (lep = &ents[index];
 616              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 617              lep++, index++) {
 618                 /*
 619                  * Skip stale leaf entries.
 620                  */
 621                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 622                         continue;
 623                 /*
 624                  * Pull the data block number from the entry.
 625                  */
 626                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 627                                                be32_to_cpu(lep->address));
 628                 /*
 629                  * For addname, we're looking for a place to put the new entry.
 630                  * We want to use a data block with an entry of equal
 631                  * hash value to ours if there is one with room.
 632                  *
 633                  * If this block isn't the data block we already have
 634                  * in hand, take a look at it.
 635                  */
 636                 if (newdb != curdb) {
 637                         __be16 *bests;
 638
 639                         curdb = newdb;
 640                         /*
 641                          * Convert the data block to the free block
 642                          * holding its freespace information.
 643                          */
 644                         newfdb = dp->d_ops->db_to_fdb(args->geo, newdb);
 645                         /*
 646                          * If it's not the one we have in hand, read it in.
 647                          */
 648                         if (newfdb != curfdb) {
 649                                 /*
 650                                  * If we had one before, drop it.
 651                                  */
 652                                 if (curbp)
 653                                         xfs_trans_brelse(tp, curbp);
 654
 655                                 error = xfs_dir2_free_read(tp, dp,
 656                                                 xfs_dir2_db_to_da(args->geo,
 657                                                                   newfdb),
 658                                                 &curbp);
 659                                 if (error)
 660                                         return error;
 661                                 free = curbp->b_addr;
 662
 663                                 xfs_dir2_free_hdr_check(dp, curbp, curdb);
 664                         }
 665                         /*
 666                          * Get the index for our entry.
 667                          */
 668                         fi = dp->d_ops->db_to_fdindex(args->geo, curdb);
 669                         /*
 670                          * If it has room, return it.
 671                          */
 672                         bests = dp->d_ops->free_bests_p(free);
 673                         if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
 674                                 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
 675                                                         XFS_ERRLEVEL_LOW, mp);
 676                                 if (curfdb != newfdb)
 677                                         xfs_trans_brelse(tp, curbp);
 678                                 return -EFSCORRUPTED;
 679                         }
 680                         curfdb = newfdb;
 681                         if (be16_to_cpu(bests[fi]) >= length)
 682                                 goto out;
 683                 }
 684         }
 685         /* Didn't find any space */
 686         fi = -1;
 687 out:
 688         ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 689         if (curbp) {
 690                 /* Giving back a free block. */
 691                 state->extravalid = 1;
 692                 state->extrablk.bp = curbp;
 693                 state->extrablk.index = fi;
 694                 state->extrablk.blkno = curfdb;
 695
 696                 /*
 697                  * Important: this magic number is not in the buffer - it's for
 698                  * buffer type information and therefore only the free/data type
 699                  * matters here, not whether CRCs are enabled or not.
 700                  */
 701                 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
 702         } else {
 703                 state->extravalid = 0;
 704         }
 705         /*
 706          * Return the index, that will be the insertion point.
 707          */
 708         *indexp = index;
 709         return -ENOENT;
 710 }
 711
 712 /*
 713  * Look up a leaf entry in a node-format leaf block.
 714  * The extrablk in state a data block.
 715  */
 716 STATIC int
 717 xfs_dir2_leafn_lookup_for_entry(
 718         struct xfs_buf          *bp,            /* leaf buffer */
 719         xfs_da_args_t           *args,          /* operation arguments */
 720         int                     *indexp,        /* out: leaf entry index */
 721         xfs_da_state_t          *state)         /* state to fill in */
 722 {
 723         struct xfs_buf          *curbp = NULL;  /* current data/free buffer */
 724         xfs_dir2_db_t           curdb = -1;     /* current data block number */
 725         xfs_dir2_data_entry_t   *dep;           /* data block entry */
 726         xfs_inode_t             *dp;            /* incore directory inode */
 727         int                     error;          /* error return value */
 728         int                     index;          /* leaf entry index */
 729         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
 730         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
 731         xfs_mount_t             *mp;            /* filesystem mount point */
 732         xfs_dir2_db_t           newdb;          /* new data block number */
 733         xfs_trans_t             *tp;            /* transaction pointer */
 734         enum xfs_dacmp          cmp;            /* comparison result */
 735         struct xfs_dir2_leaf_entry *ents;
 736         struct xfs_dir3_icleaf_hdr leafhdr;
 737
 738         dp = args->dp;
 739         tp = args->trans;
 740         mp = dp->i_mount;
 741         leaf = bp->b_addr;
 742         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
 743         ents = dp->d_ops->leaf_ents_p(leaf);
 744
 745         xfs_dir3_leaf_check(dp, bp);
 746         ASSERT(leafhdr.count > 0);
 747
 748         /*
 749          * Look up the hash value in the leaf entries.
 750          */
 751         index = xfs_dir2_leaf_search_hash(args, bp);
 752         /*
 753          * Do we have a buffer coming in?
 754          */
 755         if (state->extravalid) {
 756                 curbp = state->extrablk.bp;
 757                 curdb = state->extrablk.blkno;
 758         }
 759         /*
 760          * Loop over leaf entries with the right hash value.
 761          */
 762         for (lep = &ents[index];
 763              index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
 764              lep++, index++) {
 765                 /*
 766                  * Skip stale leaf entries.
 767                  */
 768                 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
 769                         continue;
 770                 /*
 771                  * Pull the data block number from the entry.
 772                  */
 773                 newdb = xfs_dir2_dataptr_to_db(args->geo,
 774                                                be32_to_cpu(lep->address));
 775                 /*
 776                  * Not adding a new entry, so we really want to find
 777                  * the name given to us.
 778                  *
 779                  * If it's a different data block, go get it.
 780                  */
 781                 if (newdb != curdb) {
 782                         /*
 783                          * If we had a block before that we aren't saving
 784                          * for a CI name, drop it
 785                          */
 786                         if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
 787                                                 curdb != state->extrablk.blkno))
 788                                 xfs_trans_brelse(tp, curbp);
 789                         /*
 790                          * If needing the block that is saved with a CI match,
 791                          * use it otherwise read in the new data block.
 792                          */
 793                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 794                                         newdb == state->extrablk.blkno) {
 795                                 ASSERT(state->extravalid);
 796                                 curbp = state->extrablk.bp;
 797                         } else {
 798                                 error = xfs_dir3_data_read(tp, dp,
 799                                                 xfs_dir2_db_to_da(args->geo,
 800                                                                   newdb),
 801                                                 -1, &curbp);
 802                                 if (error)
 803                                         return error;
 804                         }
 805                         xfs_dir3_data_check(dp, curbp);
 806                         curdb = newdb;
 807                 }
 808                 /*
 809                  * Point to the data entry.
 810                  */
 811                 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
 812                         xfs_dir2_dataptr_to_off(args->geo,
 813                                                 be32_to_cpu(lep->address)));
 814                 /*
 815                  * Compare the entry and if it's an exact match, return
 816                  * EEXIST immediately. If it's the first case-insensitive
 817                  * match, store the block & inode number and continue looking.
 818                  */
 819                 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
 820                 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
 821                         /* If there is a CI match block, drop it */
 822                         if (args->cmpresult != XFS_CMP_DIFFERENT &&
 823                                                 curdb != state->extrablk.blkno)
 824                                 xfs_trans_brelse(tp, state->extrablk.bp);
 825                         args->cmpresult = cmp;
 826                         args->inumber = be64_to_cpu(dep->inumber);
 827                         args->filetype = dp->d_ops->data_get_ftype(dep);
 828                         *indexp = index;
 829                         state->extravalid = 1;
 830                         state->extrablk.bp = curbp;
 831                         state->extrablk.blkno = curdb;
 832                         state->extrablk.index = (int)((char *)dep -
 833                                                         (char *)curbp->b_addr);
 834                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 835                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 836                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 837                         if (cmp == XFS_CMP_EXACT)
 838                                 return -EEXIST;
 839                 }
 840         }
 841         ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
 842         if (curbp) {
 843                 if (args->cmpresult == XFS_CMP_DIFFERENT) {
 844                         /* Giving back last used data block. */
 845                         state->extravalid = 1;
 846                         state->extrablk.bp = curbp;
 847                         state->extrablk.index = -1;
 848                         state->extrablk.blkno = curdb;
 849                         state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
 850                         curbp->b_ops = &xfs_dir3_data_buf_ops;
 851                         xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
 852                 } else {
 853                         /* If the curbp is not the CI match block, drop it */
 854                         if (state->extrablk.bp != curbp)
 855                                 xfs_trans_brelse(tp, curbp);
 856                 }
 857         } else {
 858                 state->extravalid = 0;
 859         }
 860         *indexp = index;
 861         return -ENOENT;
 862 }
 863
 864 /*
 865  * Look up a leaf entry in a node-format leaf block.
 866  * If this is an addname then the extrablk in state is a freespace block,
 867  * otherwise it's a data block.
 868  */
 869 int
 870 xfs_dir2_leafn_lookup_int(
 871         struct xfs_buf          *bp,            /* leaf buffer */
 872         xfs_da_args_t           *args,          /* operation arguments */
 873         int                     *indexp,        /* out: leaf entry index */
 874         xfs_da_state_t          *state)         /* state to fill in */
 875 {
 876         if (args->op_flags & XFS_DA_OP_ADDNAME)
 877                 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
 878                                                         state);
 879         return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
 880 }
 881
 882 /*
 883  * Move count leaf entries from source to destination leaf.
 884  * Log entries and headers.  Stale entries are preserved.
 885  */
 886 static void
 887 xfs_dir3_leafn_moveents(
 888         xfs_da_args_t                   *args,  /* operation arguments */
 889         struct xfs_buf                  *bp_s,  /* source */
 890         struct xfs_dir3_icleaf_hdr      *shdr,
 891         struct xfs_dir2_leaf_entry      *sents,
 892         int                             start_s,/* source leaf index */
 893         struct xfs_buf                  *bp_d,  /* destination */
 894         struct xfs_dir3_icleaf_hdr      *dhdr,
 895         struct xfs_dir2_leaf_entry      *dents,
 896         int                             start_d,/* destination leaf index */
 897         int                             count)  /* count of leaves to copy */
 898 {
 899         int                             stale;  /* count stale leaves copied */
 900
 901         trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
 902
 903         /*
 904          * Silently return if nothing to do.
 905          */
 906         if (count == 0)
 907                 return;
 908
 909         /*
 910          * If the destination index is not the end of the current
 911          * destination leaf entries, open up a hole in the destination
 912          * to hold the new entries.
 913          */
 914         if (start_d < dhdr->count) {
 915                 memmove(&dents[start_d + count], &dents[start_d],
 916                         (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
 917                 xfs_dir3_leaf_log_ents(args, bp_d, start_d + count,
 918                                        count + dhdr->count - 1);
 919         }
 920         /*
 921          * If the source has stale leaves, count the ones in the copy range
 922          * so we can update the header correctly.
 923          */
 924         if (shdr->stale) {
 925                 int     i;                      /* temp leaf index */
 926
 927                 for (i = start_s, stale = 0; i < start_s + count; i++) {
 928                         if (sents[i].address ==
 929                                         cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
 930                                 stale++;
 931                 }
 932         } else
 933                 stale = 0;
 934         /*
 935          * Copy the leaf entries from source to destination.
 936          */
 937         memcpy(&dents[start_d], &sents[start_s],
 938                 count * sizeof(xfs_dir2_leaf_entry_t));
 939         xfs_dir3_leaf_log_ents(args, bp_d, start_d, start_d + count - 1);
 940
 941         /*
 942          * If there are source entries after the ones we copied,
 943          * delete the ones we copied by sliding the next ones down.
 944          */
 945         if (start_s + count < shdr->count) {
 946                 memmove(&sents[start_s], &sents[start_s + count],
 947                         count * sizeof(xfs_dir2_leaf_entry_t));
 948                 xfs_dir3_leaf_log_ents(args, bp_s, start_s, start_s + count - 1);
 949         }
 950
 951         /*
 952          * Update the headers and log them.
 953          */
 954         shdr->count -= count;
 955         shdr->stale -= stale;
 956         dhdr->count += count;
 957         dhdr->stale += stale;
 958 }
 959
 960 /*
 961  * Determine the sort order of two leaf blocks.
 962  * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
 963  */
 964 int                                             /* sort order */
 965 xfs_dir2_leafn_order(
 966         struct xfs_inode        *dp,
 967         struct xfs_buf          *leaf1_bp,              /* leaf1 buffer */
 968         struct xfs_buf          *leaf2_bp)              /* leaf2 buffer */
 969 {
 970         struct xfs_dir2_leaf    *leaf1 = leaf1_bp->b_addr;
 971         struct xfs_dir2_leaf    *leaf2 = leaf2_bp->b_addr;
 972         struct xfs_dir2_leaf_entry *ents1;
 973         struct xfs_dir2_leaf_entry *ents2;
 974         struct xfs_dir3_icleaf_hdr hdr1;
 975         struct xfs_dir3_icleaf_hdr hdr2;
 976
 977         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
 978         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
 979         ents1 = dp->d_ops->leaf_ents_p(leaf1);
 980         ents2 = dp->d_ops->leaf_ents_p(leaf2);
 981
 982         if (hdr1.count > 0 && hdr2.count > 0 &&
 983             (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
 984              be32_to_cpu(ents2[hdr2.count - 1].hashval) <
 985                                 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
 986                 return 1;
 987         return 0;
 988 }
 989
 990 /*
 991  * Rebalance leaf entries between two leaf blocks.
 992  * This is actually only called when the second block is new,
 993  * though the code deals with the general case.
 994  * A new entry will be inserted in one of the blocks, and that
 995  * entry is taken into account when balancing.
 996  */
 997 static void
 998 xfs_dir2_leafn_rebalance(
 999         xfs_da_state_t          *state,         /* btree cursor */
1000         xfs_da_state_blk_t      *blk1,          /* first btree block */
1001         xfs_da_state_blk_t      *blk2)          /* second btree block */
1002 {
1003         xfs_da_args_t           *args;          /* operation arguments */
1004         int                     count;          /* count (& direction) leaves */
1005         int                     isleft;         /* new goes in left leaf */
1006         xfs_dir2_leaf_t         *leaf1;         /* first leaf structure */
1007         xfs_dir2_leaf_t         *leaf2;         /* second leaf structure */
1008         int                     mid;            /* midpoint leaf index */
1009 #if defined(DEBUG) || defined(XFS_WARN)
1010         int                     oldstale;       /* old count of stale leaves */
1011 #endif
1012         int                     oldsum;         /* old total leaf count */
1013         int                     swap_blocks;    /* swapped leaf blocks */
1014         struct xfs_dir2_leaf_entry *ents1;
1015         struct xfs_dir2_leaf_entry *ents2;
1016         struct xfs_dir3_icleaf_hdr hdr1;
1017         struct xfs_dir3_icleaf_hdr hdr2;
1018         struct xfs_inode        *dp = state->args->dp;
1019
1020         args = state->args;
1021         /*
1022          * If the block order is wrong, swap the arguments.
1023          */
1024         swap_blocks = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp);
1025         if (swap_blocks)
1026                 swap(blk1, blk2);
1027
1028         leaf1 = blk1->bp->b_addr;
1029         leaf2 = blk2->bp->b_addr;
1030         dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
1031         dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
1032         ents1 = dp->d_ops->leaf_ents_p(leaf1);
1033         ents2 = dp->d_ops->leaf_ents_p(leaf2);
1034
1035         oldsum = hdr1.count + hdr2.count;
1036 #if defined(DEBUG) || defined(XFS_WARN)
1037         oldstale = hdr1.stale + hdr2.stale;
1038 #endif
1039         mid = oldsum >> 1;
1040
1041         /*
1042          * If the old leaf count was odd then the new one will be even,
1043          * so we need to divide the new count evenly.
1044          */
1045         if (oldsum & 1) {
1046                 xfs_dahash_t    midhash;        /* middle entry hash value */
1047
1048                 if (mid >= hdr1.count)
1049                         midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1050                 else
1051                         midhash = be32_to_cpu(ents1[mid].hashval);
1052                 isleft = args->hashval <= midhash;
1053         }
1054         /*
1055          * If the old count is even then the new count is odd, so there's
1056          * no preferred side for the new entry.
1057          * Pick the left one.
1058          */
1059         else
1060                 isleft = 1;
1061         /*
1062          * Calculate moved entry count.  Positive means left-to-right,
1063          * negative means right-to-left.  Then move the entries.
1064          */
1065         count = hdr1.count - mid + (isleft == 0);
1066         if (count > 0)
1067                 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1068                                         hdr1.count - count, blk2->bp,
1069                                         &hdr2, ents2, 0, count);
1070         else if (count < 0)
1071                 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1072                                         blk1->bp, &hdr1, ents1,
1073                                         hdr1.count, count);
1074
1075         ASSERT(hdr1.count + hdr2.count == oldsum);
1076         ASSERT(hdr1.stale + hdr2.stale == oldstale);
1077
1078         /* log the changes made when moving the entries */
1079         dp->d_ops->leaf_hdr_to_disk(leaf1, &hdr1);
1080         dp->d_ops->leaf_hdr_to_disk(leaf2, &hdr2);
1081         xfs_dir3_leaf_log_header(args, blk1->bp);
1082         xfs_dir3_leaf_log_header(args, blk2->bp);
1083
1084         xfs_dir3_leaf_check(dp, blk1->bp);
1085         xfs_dir3_leaf_check(dp, blk2->bp);
1086
1087         /*
1088          * Mark whether we're inserting into the old or new leaf.
1089          */
1090         if (hdr1.count < hdr2.count)
1091                 state->inleaf = swap_blocks;
1092         else if (hdr1.count > hdr2.count)
1093                 state->inleaf = !swap_blocks;
1094         else
1095                 state->inleaf = swap_blocks ^ (blk1->index <= hdr1.count);
1096         /*
1097          * Adjust the expected index for insertion.
1098          */
1099         if (!state->inleaf)
1100                 blk2->index = blk1->index - hdr1.count;
1101
1102         /*
1103          * Finally sanity check just to make sure we are not returning a
1104          * negative index
1105          */
1106         if (blk2->index < 0) {
1107                 state->inleaf = 1;
1108                 blk2->index = 0;
1109                 xfs_alert(dp->i_mount,
1110         "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1111                         __func__, blk1->index);
1112         }
1113 }
1114
1115 static int
1116 xfs_dir3_data_block_free(
1117         xfs_da_args_t           *args,
1118         struct xfs_dir2_data_hdr *hdr,
1119         struct xfs_dir2_free    *free,
1120         xfs_dir2_db_t           fdb,
1121         int                     findex,
1122         struct xfs_buf          *fbp,
1123         int                     longest)
1124 {
1125         int                     logfree = 0;
1126         __be16                  *bests;
1127         struct xfs_dir3_icfree_hdr freehdr;
1128         struct xfs_inode        *dp = args->dp;
1129
1130         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1131         bests = dp->d_ops->free_bests_p(free);
1132         if (hdr) {
1133                 /*
1134                  * Data block is not empty, just set the free entry to the new
1135                  * value.
1136                  */
1137                 bests[findex] = cpu_to_be16(longest);
1138                 xfs_dir2_free_log_bests(args, fbp, findex, findex);
1139                 return 0;
1140         }
1141
1142         /* One less used entry in the free table. */
1143         freehdr.nused--;
1144
1145         /*
1146          * If this was the last entry in the table, we can trim the table size
1147          * back.  There might be other entries at the end referring to
1148          * non-existent data blocks, get those too.
1149          */
1150         if (findex == freehdr.nvalid - 1) {
1151                 int     i;              /* free entry index */
1152
1153                 for (i = findex - 1; i >= 0; i--) {
1154                         if (bests[i] != cpu_to_be16(NULLDATAOFF))
1155                                 break;
1156                 }
1157                 freehdr.nvalid = i + 1;
1158                 logfree = 0;
1159         } else {
1160                 /* Not the last entry, just punch it out.  */
1161                 bests[findex] = cpu_to_be16(NULLDATAOFF);
1162                 logfree = 1;
1163         }
1164
1165         dp->d_ops->free_hdr_to_disk(free, &freehdr);
1166         xfs_dir2_free_log_header(args, fbp);
1167
1168         /*
1169          * If there are no useful entries left in the block, get rid of the
1170          * block if we can.
1171          */
1172         if (!freehdr.nused) {
1173                 int error;
1174
1175                 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1176                 if (error == 0) {
1177                         fbp = NULL;
1178                         logfree = 0;
1179                 } else if (error != -ENOSPC || args->total != 0)
1180                         return error;
1181                 /*
1182                  * It's possible to get ENOSPC if there is no
1183                  * space reservation.  In this case some one
1184                  * else will eventually get rid of this block.
1185                  */
1186         }
1187
1188         /* Log the free entry that changed, unless we got rid of it.  */
1189         if (logfree)
1190                 xfs_dir2_free_log_bests(args, fbp, findex, findex);
1191         return 0;
1192 }
1193
1194 /*
1195  * Remove an entry from a node directory.
1196  * This removes the leaf entry and the data entry,
1197  * and updates the free block if necessary.
1198  */
1199 static int                                      /* error */
1200 xfs_dir2_leafn_remove(
1201         xfs_da_args_t           *args,          /* operation arguments */
1202         struct xfs_buf          *bp,            /* leaf buffer */
1203         int                     index,          /* leaf entry index */
1204         xfs_da_state_blk_t      *dblk,          /* data block */
1205         int                     *rval)          /* resulting block needs join */
1206 {
1207         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1208         xfs_dir2_db_t           db;             /* data block number */
1209         struct xfs_buf          *dbp;           /* data block buffer */
1210         xfs_dir2_data_entry_t   *dep;           /* data block entry */
1211         xfs_inode_t             *dp;            /* incore directory inode */
1212         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1213         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry */
1214         int                     longest;        /* longest data free entry */
1215         int                     off;            /* data block entry offset */
1216         int                     needlog;        /* need to log data header */
1217         int                     needscan;       /* need to rescan data frees */
1218         xfs_trans_t             *tp;            /* transaction pointer */
1219         struct xfs_dir2_data_free *bf;          /* bestfree table */
1220         struct xfs_dir3_icleaf_hdr leafhdr;
1221         struct xfs_dir2_leaf_entry *ents;
1222
1223         trace_xfs_dir2_leafn_remove(args, index);
1224
1225         dp = args->dp;
1226         tp = args->trans;
1227         leaf = bp->b_addr;
1228         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1229         ents = dp->d_ops->leaf_ents_p(leaf);
1230
1231         /*
1232          * Point to the entry we're removing.
1233          */
1234         lep = &ents[index];
1235
1236         /*
1237          * Extract the data block and offset from the entry.
1238          */
1239         db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1240         ASSERT(dblk->blkno == db);
1241         off = xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address));
1242         ASSERT(dblk->index == off);
1243
1244         /*
1245          * Kill the leaf entry by marking it stale.
1246          * Log the leaf block changes.
1247          */
1248         leafhdr.stale++;
1249         dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1250         xfs_dir3_leaf_log_header(args, bp);
1251
1252         lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1253         xfs_dir3_leaf_log_ents(args, bp, index, index);
1254
1255         /*
1256          * Make the data entry free.  Keep track of the longest freespace
1257          * in the data block in case it changes.
1258          */
1259         dbp = dblk->bp;
1260         hdr = dbp->b_addr;
1261         dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1262         bf = dp->d_ops->data_bestfree_p(hdr);
1263         longest = be16_to_cpu(bf[0].length);
1264         needlog = needscan = 0;
1265         xfs_dir2_data_make_free(args, dbp, off,
1266                 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1267         /*
1268          * Rescan the data block freespaces for bestfree.
1269          * Log the data block header if needed.
1270          */
1271         if (needscan)
1272                 xfs_dir2_data_freescan(dp, hdr, &needlog);
1273         if (needlog)
1274                 xfs_dir2_data_log_header(args, dbp);
1275         xfs_dir3_data_check(dp, dbp);
1276         /*
1277          * If the longest data block freespace changes, need to update
1278          * the corresponding freeblock entry.
1279          */
1280         if (longest < be16_to_cpu(bf[0].length)) {
1281                 int             error;          /* error return value */
1282                 struct xfs_buf  *fbp;           /* freeblock buffer */
1283                 xfs_dir2_db_t   fdb;            /* freeblock block number */
1284                 int             findex;         /* index in freeblock entries */
1285                 xfs_dir2_free_t *free;          /* freeblock structure */
1286
1287                 /*
1288                  * Convert the data block number to a free block,
1289                  * read in the free block.
1290                  */
1291                 fdb = dp->d_ops->db_to_fdb(args->geo, db);
1292                 error = xfs_dir2_free_read(tp, dp,
1293                                            xfs_dir2_db_to_da(args->geo, fdb),
1294                                            &fbp);
1295                 if (error)
1296                         return error;
1297                 free = fbp->b_addr;
1298 #ifdef DEBUG
1299         {
1300                 struct xfs_dir3_icfree_hdr freehdr;
1301                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1302                 ASSERT(freehdr.firstdb == dp->d_ops->free_max_bests(args->geo) *
1303                         (fdb - xfs_dir2_byte_to_db(args->geo,
1304                                                    XFS_DIR2_FREE_OFFSET)));
1305         }
1306 #endif
1307                 /*
1308                  * Calculate which entry we need to fix.
1309                  */
1310                 findex = dp->d_ops->db_to_fdindex(args->geo, db);
1311                 longest = be16_to_cpu(bf[0].length);
1312                 /*
1313                  * If the data block is now empty we can get rid of it
1314                  * (usually).
1315                  */
1316                 if (longest == args->geo->blksize -
1317                                dp->d_ops->data_entry_offset) {
1318                         /*
1319                          * Try to punch out the data block.
1320                          */
1321                         error = xfs_dir2_shrink_inode(args, db, dbp);
1322                         if (error == 0) {
1323                                 dblk->bp = NULL;
1324                                 hdr = NULL;
1325                         }
1326                         /*
1327                          * We can get ENOSPC if there's no space reservation.
1328                          * In this case just drop the buffer and some one else
1329                          * will eventually get rid of the empty block.
1330                          */
1331                         else if (!(error == -ENOSPC && args->total == 0))
1332                                 return error;
1333                 }
1334                 /*
1335                  * If we got rid of the data block, we can eliminate that entry
1336                  * in the free block.
1337                  */
1338                 error = xfs_dir3_data_block_free(args, hdr, free,
1339                                                  fdb, findex, fbp, longest);
1340                 if (error)
1341                         return error;
1342         }
1343
1344         xfs_dir3_leaf_check(dp, bp);
1345         /*
1346          * Return indication of whether this leaf block is empty enough
1347          * to justify trying to join it with a neighbor.
1348          */
1349         *rval = (dp->d_ops->leaf_hdr_size +
1350                  (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1351                 args->geo->magicpct;
1352         return 0;
1353 }
1354
1355 /*
1356  * Split the leaf entries in the old block into old and new blocks.
1357  */
1358 int                                             /* error */
1359 xfs_dir2_leafn_split(
1360         xfs_da_state_t          *state,         /* btree cursor */
1361         xfs_da_state_blk_t      *oldblk,        /* original block */
1362         xfs_da_state_blk_t      *newblk)        /* newly created block */
1363 {
1364         xfs_da_args_t           *args;          /* operation arguments */
1365         xfs_dablk_t             blkno;          /* new leaf block number */
1366         int                     error;          /* error return value */
1367         struct xfs_inode        *dp;
1368
1369         /*
1370          * Allocate space for a new leaf node.
1371          */
1372         args = state->args;
1373         dp = args->dp;
1374         ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1375         error = xfs_da_grow_inode(args, &blkno);
1376         if (error) {
1377                 return error;
1378         }
1379         /*
1380          * Initialize the new leaf block.
1381          */
1382         error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(args->geo, blkno),
1383                                       &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1384         if (error)
1385                 return error;
1386
1387         newblk->blkno = blkno;
1388         newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1389         /*
1390          * Rebalance the entries across the two leaves, link the new
1391          * block into the leaves.
1392          */
1393         xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1394         error = xfs_da3_blk_link(state, oldblk, newblk);
1395         if (error) {
1396                 return error;
1397         }
1398         /*
1399          * Insert the new entry in the correct block.
1400          */
1401         if (state->inleaf)
1402                 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1403         else
1404                 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1405         /*
1406          * Update last hashval in each block since we added the name.
1407          */
1408         oldblk->hashval = xfs_dir2_leaf_lasthash(dp, oldblk->bp, NULL);
1409         newblk->hashval = xfs_dir2_leaf_lasthash(dp, newblk->bp, NULL);
1410         xfs_dir3_leaf_check(dp, oldblk->bp);
1411         xfs_dir3_leaf_check(dp, newblk->bp);
1412         return error;
1413 }
1414
1415 /*
1416  * Check a leaf block and its neighbors to see if the block should be
1417  * collapsed into one or the other neighbor.  Always keep the block
1418  * with the smaller block number.
1419  * If the current block is over 50% full, don't try to join it, return 0.
1420  * If the block is empty, fill in the state structure and return 2.
1421  * If it can be collapsed, fill in the state structure and return 1.
1422  * If nothing can be done, return 0.
1423  */
1424 int                                             /* error */
1425 xfs_dir2_leafn_toosmall(
1426         xfs_da_state_t          *state,         /* btree cursor */
1427         int                     *action)        /* resulting action to take */
1428 {
1429         xfs_da_state_blk_t      *blk;           /* leaf block */
1430         xfs_dablk_t             blkno;          /* leaf block number */
1431         struct xfs_buf          *bp;            /* leaf buffer */
1432         int                     bytes;          /* bytes in use */
1433         int                     count;          /* leaf live entry count */
1434         int                     error;          /* error return value */
1435         int                     forward;        /* sibling block direction */
1436         int                     i;              /* sibling counter */
1437         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
1438         int                     rval;           /* result from path_shift */
1439         struct xfs_dir3_icleaf_hdr leafhdr;
1440         struct xfs_dir2_leaf_entry *ents;
1441         struct xfs_inode        *dp = state->args->dp;
1442
1443         /*
1444          * Check for the degenerate case of the block being over 50% full.
1445          * If so, it's not worth even looking to see if we might be able
1446          * to coalesce with a sibling.
1447          */
1448         blk = &state->path.blk[state->path.active - 1];
1449         leaf = blk->bp->b_addr;
1450         dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1451         ents = dp->d_ops->leaf_ents_p(leaf);
1452         xfs_dir3_leaf_check(dp, blk->bp);
1453
1454         count = leafhdr.count - leafhdr.stale;
1455         bytes = dp->d_ops->leaf_hdr_size + count * sizeof(ents[0]);
1456         if (bytes > (state->args->geo->blksize >> 1)) {
1457                 /*
1458                  * Blk over 50%, don't try to join.
1459                  */
1460                 *action = 0;
1461                 return 0;
1462         }
1463         /*
1464          * Check for the degenerate case of the block being empty.
1465          * If the block is empty, we'll simply delete it, no need to
1466          * coalesce it with a sibling block.  We choose (arbitrarily)
1467          * to merge with the forward block unless it is NULL.
1468          */
1469         if (count == 0) {
1470                 /*
1471                  * Make altpath point to the block we want to keep and
1472                  * path point to the block we want to drop (this one).
1473                  */
1474                 forward = (leafhdr.forw != 0);
1475                 memcpy(&state->altpath, &state->path, sizeof(state->path));
1476                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1477                         &rval);
1478                 if (error)
1479                         return error;
1480                 *action = rval ? 2 : 0;
1481                 return 0;
1482         }
1483         /*
1484          * Examine each sibling block to see if we can coalesce with
1485          * at least 25% free space to spare.  We need to figure out
1486          * whether to merge with the forward or the backward block.
1487          * We prefer coalescing with the lower numbered sibling so as
1488          * to shrink a directory over time.
1489          */
1490         forward = leafhdr.forw < leafhdr.back;
1491         for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1492                 struct xfs_dir3_icleaf_hdr hdr2;
1493
1494                 blkno = forward ? leafhdr.forw : leafhdr.back;
1495                 if (blkno == 0)
1496                         continue;
1497                 /*
1498                  * Read the sibling leaf block.
1499                  */
1500                 error = xfs_dir3_leafn_read(state->args->trans, dp,
1501                                             blkno, -1, &bp);
1502                 if (error)
1503                         return error;
1504
1505                 /*
1506                  * Count bytes in the two blocks combined.
1507                  */
1508                 count = leafhdr.count - leafhdr.stale;
1509                 bytes = state->args->geo->blksize -
1510                         (state->args->geo->blksize >> 2);
1511
1512                 leaf = bp->b_addr;
1513                 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf);
1514                 ents = dp->d_ops->leaf_ents_p(leaf);
1515                 count += hdr2.count - hdr2.stale;
1516                 bytes -= count * sizeof(ents[0]);
1517
1518                 /*
1519                  * Fits with at least 25% to spare.
1520                  */
1521                 if (bytes >= 0)
1522                         break;
1523                 xfs_trans_brelse(state->args->trans, bp);
1524         }
1525         /*
1526          * Didn't like either block, give up.
1527          */
1528         if (i >= 2) {
1529                 *action = 0;
1530                 return 0;
1531         }
1532
1533         /*
1534          * Make altpath point to the block we want to keep (the lower
1535          * numbered block) and path point to the block we want to drop.
1536          */
1537         memcpy(&state->altpath, &state->path, sizeof(state->path));
1538         if (blkno < blk->blkno)
1539                 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1540                         &rval);
1541         else
1542                 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1543                         &rval);
1544         if (error) {
1545                 return error;
1546         }
1547         *action = rval ? 0 : 1;
1548         return 0;
1549 }
1550
1551 /*
1552  * Move all the leaf entries from drop_blk to save_blk.
1553  * This is done as part of a join operation.
1554  */
1555 void
1556 xfs_dir2_leafn_unbalance(
1557         xfs_da_state_t          *state,         /* cursor */
1558         xfs_da_state_blk_t      *drop_blk,      /* dead block */
1559         xfs_da_state_blk_t      *save_blk)      /* surviving block */
1560 {
1561         xfs_da_args_t           *args;          /* operation arguments */
1562         xfs_dir2_leaf_t         *drop_leaf;     /* dead leaf structure */
1563         xfs_dir2_leaf_t         *save_leaf;     /* surviving leaf structure */
1564         struct xfs_dir3_icleaf_hdr savehdr;
1565         struct xfs_dir3_icleaf_hdr drophdr;
1566         struct xfs_dir2_leaf_entry *sents;
1567         struct xfs_dir2_leaf_entry *dents;
1568         struct xfs_inode        *dp = state->args->dp;
1569
1570         args = state->args;
1571         ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1572         ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1573         drop_leaf = drop_blk->bp->b_addr;
1574         save_leaf = save_blk->bp->b_addr;
1575
1576         dp->d_ops->leaf_hdr_from_disk(&savehdr, save_leaf);
1577         dp->d_ops->leaf_hdr_from_disk(&drophdr, drop_leaf);
1578         sents = dp->d_ops->leaf_ents_p(save_leaf);
1579         dents = dp->d_ops->leaf_ents_p(drop_leaf);
1580
1581         /*
1582          * If there are any stale leaf entries, take this opportunity
1583          * to purge them.
1584          */
1585         if (drophdr.stale)
1586                 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1587         if (savehdr.stale)
1588                 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1589
1590         /*
1591          * Move the entries from drop to the appropriate end of save.
1592          */
1593         drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1594         if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1595                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1596                                         save_blk->bp, &savehdr, sents, 0,
1597                                         drophdr.count);
1598         else
1599                 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1600                                         save_blk->bp, &savehdr, sents,
1601                                         savehdr.count, drophdr.count);
1602         save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1603
1604         /* log the changes made when moving the entries */
1605         dp->d_ops->leaf_hdr_to_disk(save_leaf, &savehdr);
1606         dp->d_ops->leaf_hdr_to_disk(drop_leaf, &drophdr);
1607         xfs_dir3_leaf_log_header(args, save_blk->bp);
1608         xfs_dir3_leaf_log_header(args, drop_blk->bp);
1609
1610         xfs_dir3_leaf_check(dp, save_blk->bp);
1611         xfs_dir3_leaf_check(dp, drop_blk->bp);
1612 }
1613
1614 /*
1615  * Top-level node form directory addname routine.
1616  */
1617 int                                             /* error */
1618 xfs_dir2_node_addname(
1619         xfs_da_args_t           *args)          /* operation arguments */
1620 {
1621         xfs_da_state_blk_t      *blk;           /* leaf block for insert */
1622         int                     error;          /* error return value */
1623         int                     rval;           /* sub-return value */
1624         xfs_da_state_t          *state;         /* btree cursor */
1625
1626         trace_xfs_dir2_node_addname(args);
1627
1628         /*
1629          * Allocate and initialize the state (btree cursor).
1630          */
1631         state = xfs_da_state_alloc();
1632         state->args = args;
1633         state->mp = args->dp->i_mount;
1634         /*
1635          * Look up the name.  We're not supposed to find it, but
1636          * this gives us the insertion point.
1637          */
1638         error = xfs_da3_node_lookup_int(state, &rval);
1639         if (error)
1640                 rval = error;
1641         if (rval != -ENOENT) {
1642                 goto done;
1643         }
1644         /*
1645          * Add the data entry to a data block.
1646          * Extravalid is set to a freeblock found by lookup.
1647          */
1648         rval = xfs_dir2_node_addname_int(args,
1649                 state->extravalid ? &state->extrablk : NULL);
1650         if (rval) {
1651                 goto done;
1652         }
1653         blk = &state->path.blk[state->path.active - 1];
1654         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1655         /*
1656          * Add the new leaf entry.
1657          */
1658         rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1659         if (rval == 0) {
1660                 /*
1661                  * It worked, fix the hash values up the btree.
1662                  */
1663                 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1664                         xfs_da3_fixhashpath(state, &state->path);
1665         } else {
1666                 /*
1667                  * It didn't work, we need to split the leaf block.
1668                  */
1669                 if (args->total == 0) {
1670                         ASSERT(rval == -ENOSPC);
1671                         goto done;
1672                 }
1673                 /*
1674                  * Split the leaf block and insert the new entry.
1675                  */
1676                 rval = xfs_da3_split(state);
1677         }
1678 done:
1679         xfs_da_state_free(state);
1680         return rval;
1681 }
1682
1683 /*
1684  * Add the data entry for a node-format directory name addition.
1685  * The leaf entry is added in xfs_dir2_leafn_add.
1686  * We may enter with a freespace block that the lookup found.
1687  */
1688 static int                                      /* error */
1689 xfs_dir2_node_addname_int(
1690         xfs_da_args_t           *args,          /* operation arguments */
1691         xfs_da_state_blk_t      *fblk)          /* optional freespace block */
1692 {
1693         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
1694         xfs_dir2_db_t           dbno;           /* data block number */
1695         struct xfs_buf          *dbp;           /* data block buffer */
1696         xfs_dir2_data_entry_t   *dep;           /* data entry pointer */
1697         xfs_inode_t             *dp;            /* incore directory inode */
1698         xfs_dir2_data_unused_t  *dup;           /* data unused entry pointer */
1699         int                     error;          /* error return value */
1700         xfs_dir2_db_t           fbno;           /* freespace block number */
1701         struct xfs_buf          *fbp;           /* freespace buffer */
1702         int                     findex;         /* freespace entry index */
1703         xfs_dir2_free_t         *free=NULL;     /* freespace block structure */
1704         xfs_dir2_db_t           ifbno;          /* initial freespace block no */
1705         xfs_dir2_db_t           lastfbno=0;     /* highest freespace block no */
1706         int                     length;         /* length of the new entry */
1707         int                     logfree;        /* need to log free entry */
1708         xfs_mount_t             *mp;            /* filesystem mount point */
1709         int                     needlog;        /* need to log data header */
1710         int                     needscan;       /* need to rescan data frees */
1711         __be16                  *tagp;          /* data entry tag pointer */
1712         xfs_trans_t             *tp;            /* transaction pointer */
1713         __be16                  *bests;
1714         struct xfs_dir3_icfree_hdr freehdr;
1715         struct xfs_dir2_data_free *bf;
1716         xfs_dir2_data_aoff_t    aoff;
1717
1718         dp = args->dp;
1719         mp = dp->i_mount;
1720         tp = args->trans;
1721         length = dp->d_ops->data_entsize(args->namelen);
1722         /*
1723          * If we came in with a freespace block that means that lookup
1724          * found an entry with our hash value.  This is the freespace
1725          * block for that data entry.
1726          */
1727         if (fblk) {
1728                 fbp = fblk->bp;
1729                 /*
1730                  * Remember initial freespace block number.
1731                  */
1732                 ifbno = fblk->blkno;
1733                 free = fbp->b_addr;
1734                 findex = fblk->index;
1735                 bests = dp->d_ops->free_bests_p(free);
1736                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1737
1738                 /*
1739                  * This means the free entry showed that the data block had
1740                  * space for our entry, so we remembered it.
1741                  * Use that data block.
1742                  */
1743                 if (findex >= 0) {
1744                         ASSERT(findex < freehdr.nvalid);
1745                         ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1746                         ASSERT(be16_to_cpu(bests[findex]) >= length);
1747                         dbno = freehdr.firstdb + findex;
1748                 } else {
1749                         /*
1750                          * The data block looked at didn't have enough room.
1751                          * We'll start at the beginning of the freespace entries.
1752                          */
1753                         dbno = -1;
1754                         findex = 0;
1755                 }
1756         } else {
1757                 /*
1758                  * Didn't come in with a freespace block, so no data block.
1759                  */
1760                 ifbno = dbno = -1;
1761                 fbp = NULL;
1762                 findex = 0;
1763         }
1764
1765         /*
1766          * If we don't have a data block yet, we're going to scan the
1767          * freespace blocks looking for one.  Figure out what the
1768          * highest freespace block number is.
1769          */
1770         if (dbno == -1) {
1771                 xfs_fileoff_t   fo;             /* freespace block number */
1772
1773                 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK)))
1774                         return error;
1775                 lastfbno = xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo);
1776                 fbno = ifbno;
1777         }
1778         /*
1779          * While we haven't identified a data block, search the freeblock
1780          * data for a good data block.  If we find a null freeblock entry,
1781          * indicating a hole in the data blocks, remember that.
1782          */
1783         while (dbno == -1) {
1784                 /*
1785                  * If we don't have a freeblock in hand, get the next one.
1786                  */
1787                 if (fbp == NULL) {
1788                         /*
1789                          * Happens the first time through unless lookup gave
1790                          * us a freespace block to start with.
1791                          */
1792                         if (++fbno == 0)
1793                                 fbno = xfs_dir2_byte_to_db(args->geo,
1794                                                         XFS_DIR2_FREE_OFFSET);
1795                         /*
1796                          * If it's ifbno we already looked at it.
1797                          */
1798                         if (fbno == ifbno)
1799                                 fbno++;
1800                         /*
1801                          * If it's off the end we're done.
1802                          */
1803                         if (fbno >= lastfbno)
1804                                 break;
1805                         /*
1806                          * Read the block.  There can be holes in the
1807                          * freespace blocks, so this might not succeed.
1808                          * This should be really rare, so there's no reason
1809                          * to avoid it.
1810                          */
1811                         error = xfs_dir2_free_try_read(tp, dp,
1812                                         xfs_dir2_db_to_da(args->geo, fbno),
1813                                         &fbp);
1814                         if (error)
1815                                 return error;
1816                         if (!fbp)
1817                                 continue;
1818                         free = fbp->b_addr;
1819                         findex = 0;
1820                 }
1821                 /*
1822                  * Look at the current free entry.  Is it good enough?
1823                  *
1824                  * The bests initialisation should be where the bufer is read in
1825                  * the above branch. But gcc is too stupid to realise that bests
1826                  * and the freehdr are actually initialised if they are placed
1827                  * there, so we have to do it here to avoid warnings. Blech.
1828                  */
1829                 bests = dp->d_ops->free_bests_p(free);
1830                 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1831                 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1832                     be16_to_cpu(bests[findex]) >= length)
1833                         dbno = freehdr.firstdb + findex;
1834                 else {
1835                         /*
1836                          * Are we done with the freeblock?
1837                          */
1838                         if (++findex == freehdr.nvalid) {
1839                                 /*
1840                                  * Drop the block.
1841                                  */
1842                                 xfs_trans_brelse(tp, fbp);
1843                                 fbp = NULL;
1844                                 if (fblk && fblk->bp)
1845                                         fblk->bp = NULL;
1846                         }
1847                 }
1848         }
1849         /*
1850          * If we don't have a data block, we need to allocate one and make
1851          * the freespace entries refer to it.
1852          */
1853         if (unlikely(dbno == -1)) {
1854                 /*
1855                  * Not allowed to allocate, return failure.
1856                  */
1857                 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1858                         return -ENOSPC;
1859
1860                 /*
1861                  * Allocate and initialize the new data block.
1862                  */
1863                 if (unlikely((error = xfs_dir2_grow_inode(args,
1864                                                          XFS_DIR2_DATA_SPACE,
1865                                                          &dbno)) ||
1866                     (error = xfs_dir3_data_init(args, dbno, &dbp))))
1867                         return error;
1868
1869                 /*
1870                  * If (somehow) we have a freespace block, get rid of it.
1871                  */
1872                 if (fbp)
1873                         xfs_trans_brelse(tp, fbp);
1874                 if (fblk && fblk->bp)
1875                         fblk->bp = NULL;
1876
1877                 /*
1878                  * Get the freespace block corresponding to the data block
1879                  * that was just allocated.
1880                  */
1881                 fbno = dp->d_ops->db_to_fdb(args->geo, dbno);
1882                 error = xfs_dir2_free_try_read(tp, dp,
1883                                        xfs_dir2_db_to_da(args->geo, fbno),
1884                                        &fbp);
1885                 if (error)
1886                         return error;
1887
1888                 /*
1889                  * If there wasn't a freespace block, the read will
1890                  * return a NULL fbp.  Allocate and initialize a new one.
1891                  */
1892                 if (!fbp) {
1893                         error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1894                                                     &fbno);
1895                         if (error)
1896                                 return error;
1897
1898                         if (dp->d_ops->db_to_fdb(args->geo, dbno) != fbno) {
1899                                 xfs_alert(mp,
1900 "%s: dir ino %llu needed freesp block %lld for data block %lld, got %lld ifbno %llu lastfbno %d",
1901                                         __func__, (unsigned long long)dp->i_ino,
1902                                         (long long)dp->d_ops->db_to_fdb(
1903                                                                 args->geo, dbno),
1904                                         (long long)dbno, (long long)fbno,
1905                                         (unsigned long long)ifbno, lastfbno);
1906                                 if (fblk) {
1907                                         xfs_alert(mp,
1908                                 " fblk "PTR_FMT" blkno %llu index %d magic 0x%x",
1909                                                 fblk,
1910                                                 (unsigned long long)fblk->blkno,
1911                                                 fblk->index,
1912                                                 fblk->magic);
1913                                 } else {
1914                                         xfs_alert(mp, " ... fblk is NULL");
1915                                 }
1916                                 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1917                                                  XFS_ERRLEVEL_LOW, mp);
1918                                 return -EFSCORRUPTED;
1919                         }
1920
1921                         /*
1922                          * Get a buffer for the new block.
1923                          */
1924                         error = xfs_dir3_free_get_buf(args, fbno, &fbp);
1925                         if (error)
1926                                 return error;
1927                         free = fbp->b_addr;
1928                         bests = dp->d_ops->free_bests_p(free);
1929                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1930
1931                         /*
1932                          * Remember the first slot as our empty slot.
1933                          */
1934                         freehdr.firstdb =
1935                                 (fbno - xfs_dir2_byte_to_db(args->geo,
1936                                                         XFS_DIR2_FREE_OFFSET)) *
1937                                         dp->d_ops->free_max_bests(args->geo);
1938                 } else {
1939                         free = fbp->b_addr;
1940                         bests = dp->d_ops->free_bests_p(free);
1941                         dp->d_ops->free_hdr_from_disk(&freehdr, free);
1942                 }
1943
1944                 /*
1945                  * Set the freespace block index from the data block number.
1946                  */
1947                 findex = dp->d_ops->db_to_fdindex(args->geo, dbno);
1948                 /*
1949                  * If it's after the end of the current entries in the
1950                  * freespace block, extend that table.
1951                  */
1952                 if (findex >= freehdr.nvalid) {
1953                         ASSERT(findex < dp->d_ops->free_max_bests(args->geo));
1954                         freehdr.nvalid = findex + 1;
1955                         /*
1956                          * Tag new entry so nused will go up.
1957                          */
1958                         bests[findex] = cpu_to_be16(NULLDATAOFF);
1959                 }
1960                 /*
1961                  * If this entry was for an empty data block
1962                  * (this should always be true) then update the header.
1963                  */
1964                 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1965                         freehdr.nused++;
1966                         dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
1967                         xfs_dir2_free_log_header(args, fbp);
1968                 }
1969                 /*
1970                  * Update the real value in the table.
1971                  * We haven't allocated the data entry yet so this will
1972                  * change again.
1973                  */
1974                 hdr = dbp->b_addr;
1975                 bf = dp->d_ops->data_bestfree_p(hdr);
1976                 bests[findex] = bf[0].length;
1977                 logfree = 1;
1978         }
1979         /*
1980          * We had a data block so we don't have to make a new one.
1981          */
1982         else {
1983                 /*
1984                  * If just checking, we succeeded.
1985                  */
1986                 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1987                         return 0;
1988
1989                 /*
1990                  * Read the data block in.
1991                  */
1992                 error = xfs_dir3_data_read(tp, dp,
1993                                            xfs_dir2_db_to_da(args->geo, dbno),
1994                                            -1, &dbp);
1995                 if (error)
1996                         return error;
1997                 hdr = dbp->b_addr;
1998                 bf = dp->d_ops->data_bestfree_p(hdr);
1999                 logfree = 0;
2000         }
2001         ASSERT(be16_to_cpu(bf[0].length) >= length);
2002         /*
2003          * Point to the existing unused space.
2004          */
2005         dup = (xfs_dir2_data_unused_t *)
2006               ((char *)hdr + be16_to_cpu(bf[0].offset));
2007         needscan = needlog = 0;
2008         /*
2009          * Mark the first part of the unused space, inuse for us.
2010          */
2011         aoff = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
2012         error = xfs_dir2_data_use_free(args, dbp, dup, aoff, length,
2013                         &needlog, &needscan);
2014         if (error) {
2015                 xfs_trans_brelse(tp, dbp);
2016                 return error;
2017         }
2018         /*
2019          * Fill in the new entry and log it.
2020          */
2021         dep = (xfs_dir2_data_entry_t *)dup;
2022         dep->inumber = cpu_to_be64(args->inumber);
2023         dep->namelen = args->namelen;
2024         memcpy(dep->name, args->name, dep->namelen);
2025         dp->d_ops->data_put_ftype(dep, args->filetype);
2026         tagp = dp->d_ops->data_entry_tag_p(dep);
2027         *tagp = cpu_to_be16((char *)dep - (char *)hdr);
2028         xfs_dir2_data_log_entry(args, dbp, dep);
2029         /*
2030          * Rescan the block for bestfree if needed.
2031          */
2032         if (needscan)
2033                 xfs_dir2_data_freescan(dp, hdr, &needlog);
2034         /*
2035          * Log the data block header if needed.
2036          */
2037         if (needlog)
2038                 xfs_dir2_data_log_header(args, dbp);
2039         /*
2040          * If the freespace entry is now wrong, update it.
2041          */
2042         bests = dp->d_ops->free_bests_p(free); /* gcc is so stupid */
2043         if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
2044                 bests[findex] = bf[0].length;
2045                 logfree = 1;
2046         }
2047         /*
2048          * Log the freespace entry if needed.
2049          */
2050         if (logfree)
2051                 xfs_dir2_free_log_bests(args, fbp, findex, findex);
2052         /*
2053          * Return the data block and offset in args, then drop the data block.
2054          */
2055         args->blkno = (xfs_dablk_t)dbno;
2056         args->index = be16_to_cpu(*tagp);
2057         return 0;
2058 }
2059
2060 /*
2061  * Lookup an entry in a node-format directory.
2062  * All the real work happens in xfs_da3_node_lookup_int.
2063  * The only real output is the inode number of the entry.
2064  */
2065 int                                             /* error */
2066 xfs_dir2_node_lookup(
2067         xfs_da_args_t   *args)                  /* operation arguments */
2068 {
2069         int             error;                  /* error return value */
2070         int             i;                      /* btree level */
2071         int             rval;                   /* operation return value */
2072         xfs_da_state_t  *state;                 /* btree cursor */
2073
2074         trace_xfs_dir2_node_lookup(args);
2075
2076         /*
2077          * Allocate and initialize the btree cursor.
2078          */
2079         state = xfs_da_state_alloc();
2080         state->args = args;
2081         state->mp = args->dp->i_mount;
2082         /*
2083          * Fill in the path to the entry in the cursor.
2084          */
2085         error = xfs_da3_node_lookup_int(state, &rval);
2086         if (error)
2087                 rval = error;
2088         else if (rval == -ENOENT && args->cmpresult == XFS_CMP_CASE) {
2089                 /* If a CI match, dup the actual name and return -EEXIST */
2090                 xfs_dir2_data_entry_t   *dep;
2091
2092                 dep = (xfs_dir2_data_entry_t *)
2093                         ((char *)state->extrablk.bp->b_addr +
2094                                                  state->extrablk.index);
2095                 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2096         }
2097         /*
2098          * Release the btree blocks and leaf block.
2099          */
2100         for (i = 0; i < state->path.active; i++) {
2101                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2102                 state->path.blk[i].bp = NULL;
2103         }
2104         /*
2105          * Release the data block if we have it.
2106          */
2107         if (state->extravalid && state->extrablk.bp) {
2108                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2109                 state->extrablk.bp = NULL;
2110         }
2111         xfs_da_state_free(state);
2112         return rval;
2113 }
2114
2115 /*
2116  * Remove an entry from a node-format directory.
2117  */
2118 int                                             /* error */
2119 xfs_dir2_node_removename(
2120         struct xfs_da_args      *args)          /* operation arguments */
2121 {
2122         struct xfs_da_state_blk *blk;           /* leaf block */
2123         int                     error;          /* error return value */
2124         int                     rval;           /* operation return value */
2125         struct xfs_da_state     *state;         /* btree cursor */
2126
2127         trace_xfs_dir2_node_removename(args);
2128
2129         /*
2130          * Allocate and initialize the btree cursor.
2131          */
2132         state = xfs_da_state_alloc();
2133         state->args = args;
2134         state->mp = args->dp->i_mount;
2135
2136         /* Look up the entry we're deleting, set up the cursor. */
2137         error = xfs_da3_node_lookup_int(state, &rval);
2138         if (error)
2139                 goto out_free;
2140
2141         /* Didn't find it, upper layer screwed up. */
2142         if (rval != -EEXIST) {
2143                 error = rval;
2144                 goto out_free;
2145         }
2146
2147         blk = &state->path.blk[state->path.active - 1];
2148         ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2149         ASSERT(state->extravalid);
2150         /*
2151          * Remove the leaf and data entries.
2152          * Extrablk refers to the data block.
2153          */
2154         error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2155                 &state->extrablk, &rval);
2156         if (error)
2157                 goto out_free;
2158         /*
2159          * Fix the hash values up the btree.
2160          */
2161         xfs_da3_fixhashpath(state, &state->path);
2162         /*
2163          * If we need to join leaf blocks, do it.
2164          */
2165         if (rval && state->path.active > 1)
2166                 error = xfs_da3_join(state);
2167         /*
2168          * If no errors so far, try conversion to leaf format.
2169          */
2170         if (!error)
2171                 error = xfs_dir2_node_to_leaf(state);
2172 out_free:
2173         xfs_da_state_free(state);
2174         return error;
2175 }
2176
2177 /*
2178  * Replace an entry's inode number in a node-format directory.
2179  */
2180 int                                             /* error */
2181 xfs_dir2_node_replace(
2182         xfs_da_args_t           *args)          /* operation arguments */
2183 {
2184         xfs_da_state_blk_t      *blk;           /* leaf block */
2185         xfs_dir2_data_hdr_t     *hdr;           /* data block header */
2186         xfs_dir2_data_entry_t   *dep;           /* data entry changed */
2187         int                     error;          /* error return value */
2188         int                     i;              /* btree level */
2189         xfs_ino_t               inum;           /* new inode number */
2190         int                     ftype;          /* new file type */
2191         xfs_dir2_leaf_t         *leaf;          /* leaf structure */
2192         xfs_dir2_leaf_entry_t   *lep;           /* leaf entry being changed */
2193         int                     rval;           /* internal return value */
2194         xfs_da_state_t          *state;         /* btree cursor */
2195
2196         trace_xfs_dir2_node_replace(args);
2197
2198         /*
2199          * Allocate and initialize the btree cursor.
2200          */
2201         state = xfs_da_state_alloc();
2202         state->args = args;
2203         state->mp = args->dp->i_mount;
2204
2205         /*
2206          * We have to save new inode number and ftype since
2207          * xfs_da3_node_lookup_int() is going to overwrite them
2208          */
2209         inum = args->inumber;
2210         ftype = args->filetype;
2211
2212         /*
2213          * Lookup the entry to change in the btree.
2214          */
2215         error = xfs_da3_node_lookup_int(state, &rval);
2216         if (error) {
2217                 rval = error;
2218         }
2219         /*
2220          * It should be found, since the vnodeops layer has looked it up
2221          * and locked it.  But paranoia is good.
2222          */
2223         if (rval == -EEXIST) {
2224                 struct xfs_dir2_leaf_entry *ents;
2225                 /*
2226                  * Find the leaf entry.
2227                  */
2228                 blk = &state->path.blk[state->path.active - 1];
2229                 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2230                 leaf = blk->bp->b_addr;
2231                 ents = args->dp->d_ops->leaf_ents_p(leaf);
2232                 lep = &ents[blk->index];
2233                 ASSERT(state->extravalid);
2234                 /*
2235                  * Point to the data entry.
2236                  */
2237                 hdr = state->extrablk.bp->b_addr;
2238                 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2239                        hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2240                 dep = (xfs_dir2_data_entry_t *)
2241                       ((char *)hdr +
2242                        xfs_dir2_dataptr_to_off(args->geo,
2243                                                be32_to_cpu(lep->address)));
2244                 ASSERT(inum != be64_to_cpu(dep->inumber));
2245                 /*
2246                  * Fill in the new inode number and log the entry.
2247                  */
2248                 dep->inumber = cpu_to_be64(inum);
2249                 args->dp->d_ops->data_put_ftype(dep, ftype);
2250                 xfs_dir2_data_log_entry(args, state->extrablk.bp, dep);
2251                 rval = 0;
2252         }
2253         /*
2254          * Didn't find it, and we're holding a data block.  Drop it.
2255          */
2256         else if (state->extravalid) {
2257                 xfs_trans_brelse(args->trans, state->extrablk.bp);
2258                 state->extrablk.bp = NULL;
2259         }
2260         /*
2261          * Release all the buffers in the cursor.
2262          */
2263         for (i = 0; i < state->path.active; i++) {
2264                 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2265                 state->path.blk[i].bp = NULL;
2266         }
2267         xfs_da_state_free(state);
2268         return rval;
2269 }
2270
2271 /*
2272  * Trim off a trailing empty freespace block.
2273  * Return (in rvalp) 1 if we did it, 0 if not.
2274  */
2275 int                                             /* error */
2276 xfs_dir2_node_trim_free(
2277         xfs_da_args_t           *args,          /* operation arguments */
2278         xfs_fileoff_t           fo,             /* free block number */
2279         int                     *rvalp)         /* out: did something */
2280 {
2281         struct xfs_buf          *bp;            /* freespace buffer */
2282         xfs_inode_t             *dp;            /* incore directory inode */
2283         int                     error;          /* error return code */
2284         xfs_dir2_free_t         *free;          /* freespace structure */
2285         xfs_trans_t             *tp;            /* transaction pointer */
2286         struct xfs_dir3_icfree_hdr freehdr;
2287
2288         dp = args->dp;
2289         tp = args->trans;
2290
2291         *rvalp = 0;
2292
2293         /*
2294          * Read the freespace block.
2295          */
2296         error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2297         if (error)
2298                 return error;
2299         /*
2300          * There can be holes in freespace.  If fo is a hole, there's
2301          * nothing to do.
2302          */
2303         if (!bp)
2304                 return 0;
2305         free = bp->b_addr;
2306         dp->d_ops->free_hdr_from_disk(&freehdr, free);
2307
2308         /*
2309          * If there are used entries, there's nothing to do.
2310          */
2311         if (freehdr.nused > 0) {
2312                 xfs_trans_brelse(tp, bp);
2313                 return 0;
2314         }
2315         /*
2316          * Blow the block away.
2317          */
2318         error = xfs_dir2_shrink_inode(args,
2319                         xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo), bp);
2320         if (error) {
2321                 /*
2322                  * Can't fail with ENOSPC since that only happens with no
2323                  * space reservation, when breaking up an extent into two
2324                  * pieces.  This is the last block of an extent.
2325                  */
2326                 ASSERT(error != -ENOSPC);
2327                 xfs_trans_brelse(tp, bp);
2328                 return error;
2329         }
2330         /*
2331          * Return that we succeeded.
2332          */
2333         *rvalp = 1;
2334         return 0;
2335 }