fs/xfs/libxfs/xfs_btree.c

   1 /*
   2  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_bit.h"
  25 #include "xfs_mount.h"
  26 #include "xfs_defer.h"
  27 #include "xfs_inode.h"
  28 #include "xfs_trans.h"
  29 #include "xfs_inode_item.h"
  30 #include "xfs_buf_item.h"
  31 #include "xfs_btree.h"
  32 #include "xfs_error.h"
  33 #include "xfs_trace.h"
  34 #include "xfs_cksum.h"
  35 #include "xfs_alloc.h"
  36 #include "xfs_log.h"
  37
  38 /*
  39  * Cursor allocation zone.
  40  */
  41 kmem_zone_t     *xfs_btree_cur_zone;
  42
  43 /*
  44  * Btree magic numbers.
  45  */
  46 static const uint32_t xfs_magics[2][XFS_BTNUM_MAX] = {
  47         { XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, 0, XFS_BMAP_MAGIC, XFS_IBT_MAGIC,
  48           XFS_FIBT_MAGIC, 0 },
  49         { XFS_ABTB_CRC_MAGIC, XFS_ABTC_CRC_MAGIC, XFS_RMAP_CRC_MAGIC,
  50           XFS_BMAP_CRC_MAGIC, XFS_IBT_CRC_MAGIC, XFS_FIBT_CRC_MAGIC,
  51           XFS_REFC_CRC_MAGIC }
  52 };
  53
  54 uint32_t
  55 xfs_btree_magic(
  56         int                     crc,
  57         xfs_btnum_t             btnum)
  58 {
  59         uint32_t                magic = xfs_magics[crc][btnum];
  60
  61         /* Ensure we asked for crc for crc-only magics. */
  62         ASSERT(magic != 0);
  63         return magic;
  64 }
  65
  66 STATIC int                              /* error (0 or EFSCORRUPTED) */
  67 xfs_btree_check_lblock(
  68         struct xfs_btree_cur    *cur,   /* btree cursor */
  69         struct xfs_btree_block  *block, /* btree long form block pointer */
  70         int                     level,  /* level of the btree block */
  71         struct xfs_buf          *bp)    /* buffer for block, if any */
  72 {
  73         int                     lblock_ok = 1; /* block passes checks */
  74         struct xfs_mount        *mp;    /* file system mount point */
  75         xfs_btnum_t             btnum = cur->bc_btnum;
  76         int                     crc;
  77
  78         mp = cur->bc_mp;
  79         crc = xfs_sb_version_hascrc(&mp->m_sb);
  80
  81         if (crc) {
  82                 lblock_ok = lblock_ok &&
  83                         uuid_equal(&block->bb_u.l.bb_uuid,
  84                                    &mp->m_sb.sb_meta_uuid) &&
  85                         block->bb_u.l.bb_blkno == cpu_to_be64(
  86                                 bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
  87         }
  88
  89         lblock_ok = lblock_ok &&
  90                 be32_to_cpu(block->bb_magic) == xfs_btree_magic(crc, btnum) &&
  91                 be16_to_cpu(block->bb_level) == level &&
  92                 be16_to_cpu(block->bb_numrecs) <=
  93                         cur->bc_ops->get_maxrecs(cur, level) &&
  94                 block->bb_u.l.bb_leftsib &&
  95                 (block->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK) ||
  96                  XFS_FSB_SANITY_CHECK(mp,
  97                         be64_to_cpu(block->bb_u.l.bb_leftsib))) &&
  98                 block->bb_u.l.bb_rightsib &&
  99                 (block->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK) ||
 100                  XFS_FSB_SANITY_CHECK(mp,
 101                         be64_to_cpu(block->bb_u.l.bb_rightsib)));
 102
 103         if (unlikely(XFS_TEST_ERROR(!lblock_ok, mp,
 104                         XFS_ERRTAG_BTREE_CHECK_LBLOCK))) {
 105                 if (bp)
 106                         trace_xfs_btree_corrupt(bp, _RET_IP_);
 107                 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
 108                 return -EFSCORRUPTED;
 109         }
 110         return 0;
 111 }
 112
 113 STATIC int                              /* error (0 or EFSCORRUPTED) */
 114 xfs_btree_check_sblock(
 115         struct xfs_btree_cur    *cur,   /* btree cursor */
 116         struct xfs_btree_block  *block, /* btree short form block pointer */
 117         int                     level,  /* level of the btree block */
 118         struct xfs_buf          *bp)    /* buffer containing block */
 119 {
 120         struct xfs_mount        *mp;    /* file system mount point */
 121         struct xfs_buf          *agbp;  /* buffer for ag. freespace struct */
 122         struct xfs_agf          *agf;   /* ag. freespace structure */
 123         xfs_agblock_t           agflen; /* native ag. freespace length */
 124         int                     sblock_ok = 1; /* block passes checks */
 125         xfs_btnum_t             btnum = cur->bc_btnum;
 126         int                     crc;
 127
 128         mp = cur->bc_mp;
 129         crc = xfs_sb_version_hascrc(&mp->m_sb);
 130         agbp = cur->bc_private.a.agbp;
 131         agf = XFS_BUF_TO_AGF(agbp);
 132         agflen = be32_to_cpu(agf->agf_length);
 133
 134         if (crc) {
 135                 sblock_ok = sblock_ok &&
 136                         uuid_equal(&block->bb_u.s.bb_uuid,
 137                                    &mp->m_sb.sb_meta_uuid) &&
 138                         block->bb_u.s.bb_blkno == cpu_to_be64(
 139                                 bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
 140         }
 141
 142         sblock_ok = sblock_ok &&
 143                 be32_to_cpu(block->bb_magic) == xfs_btree_magic(crc, btnum) &&
 144                 be16_to_cpu(block->bb_level) == level &&
 145                 be16_to_cpu(block->bb_numrecs) <=
 146                         cur->bc_ops->get_maxrecs(cur, level) &&
 147                 (block->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) ||
 148                  be32_to_cpu(block->bb_u.s.bb_leftsib) < agflen) &&
 149                 block->bb_u.s.bb_leftsib &&
 150                 (block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK) ||
 151                  be32_to_cpu(block->bb_u.s.bb_rightsib) < agflen) &&
 152                 block->bb_u.s.bb_rightsib;
 153
 154         if (unlikely(XFS_TEST_ERROR(!sblock_ok, mp,
 155                         XFS_ERRTAG_BTREE_CHECK_SBLOCK))) {
 156                 if (bp)
 157                         trace_xfs_btree_corrupt(bp, _RET_IP_);
 158                 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
 159                 return -EFSCORRUPTED;
 160         }
 161         return 0;
 162 }
 163
 164 /*
 165  * Debug routine: check that block header is ok.
 166  */
 167 int
 168 xfs_btree_check_block(
 169         struct xfs_btree_cur    *cur,   /* btree cursor */
 170         struct xfs_btree_block  *block, /* generic btree block pointer */
 171         int                     level,  /* level of the btree block */
 172         struct xfs_buf          *bp)    /* buffer containing block, if any */
 173 {
 174         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
 175                 return xfs_btree_check_lblock(cur, block, level, bp);
 176         else
 177                 return xfs_btree_check_sblock(cur, block, level, bp);
 178 }
 179
 180 /*
 181  * Check that (long) pointer is ok.
 182  */
 183 int                                     /* error (0 or EFSCORRUPTED) */
 184 xfs_btree_check_lptr(
 185         struct xfs_btree_cur    *cur,   /* btree cursor */
 186         xfs_fsblock_t           bno,    /* btree block disk address */
 187         int                     level)  /* btree block level */
 188 {
 189         XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
 190                 level > 0 &&
 191                 bno != NULLFSBLOCK &&
 192                 XFS_FSB_SANITY_CHECK(cur->bc_mp, bno));
 193         return 0;
 194 }
 195
 196 #ifdef DEBUG
 197 /*
 198  * Check that (short) pointer is ok.
 199  */
 200 STATIC int                              /* error (0 or EFSCORRUPTED) */
 201 xfs_btree_check_sptr(
 202         struct xfs_btree_cur    *cur,   /* btree cursor */
 203         xfs_agblock_t           bno,    /* btree block disk address */
 204         int                     level)  /* btree block level */
 205 {
 206         xfs_agblock_t           agblocks = cur->bc_mp->m_sb.sb_agblocks;
 207
 208         XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
 209                 level > 0 &&
 210                 bno != NULLAGBLOCK &&
 211                 bno != 0 &&
 212                 bno < agblocks);
 213         return 0;
 214 }
 215
 216 /*
 217  * Check that block ptr is ok.
 218  */
 219 STATIC int                              /* error (0 or EFSCORRUPTED) */
 220 xfs_btree_check_ptr(
 221         struct xfs_btree_cur    *cur,   /* btree cursor */
 222         union xfs_btree_ptr     *ptr,   /* btree block disk address */
 223         int                     index,  /* offset from ptr to check */
 224         int                     level)  /* btree block level */
 225 {
 226         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 227                 return xfs_btree_check_lptr(cur,
 228                                 be64_to_cpu((&ptr->l)[index]), level);
 229         } else {
 230                 return xfs_btree_check_sptr(cur,
 231                                 be32_to_cpu((&ptr->s)[index]), level);
 232         }
 233 }
 234 #endif
 235
 236 /*
 237  * Calculate CRC on the whole btree block and stuff it into the
 238  * long-form btree header.
 239  *
 240  * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
 241  * it into the buffer so recovery knows what the last modification was that made
 242  * it to disk.
 243  */
 244 void
 245 xfs_btree_lblock_calc_crc(
 246         struct xfs_buf          *bp)
 247 {
 248         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 249         struct xfs_buf_log_item *bip = bp->b_fspriv;
 250
 251         if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
 252                 return;
 253         if (bip)
 254                 block->bb_u.l.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 255         xfs_buf_update_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
 256 }
 257
 258 bool
 259 xfs_btree_lblock_verify_crc(
 260         struct xfs_buf          *bp)
 261 {
 262         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 263         struct xfs_mount        *mp = bp->b_target->bt_mount;
 264
 265         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 266                 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.l.bb_lsn)))
 267                         return false;
 268                 return xfs_buf_verify_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
 269         }
 270
 271         return true;
 272 }
 273
 274 /*
 275  * Calculate CRC on the whole btree block and stuff it into the
 276  * short-form btree header.
 277  *
 278  * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
 279  * it into the buffer so recovery knows what the last modification was that made
 280  * it to disk.
 281  */
 282 void
 283 xfs_btree_sblock_calc_crc(
 284         struct xfs_buf          *bp)
 285 {
 286         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 287         struct xfs_buf_log_item *bip = bp->b_fspriv;
 288
 289         if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
 290                 return;
 291         if (bip)
 292                 block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 293         xfs_buf_update_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
 294 }
 295
 296 bool
 297 xfs_btree_sblock_verify_crc(
 298         struct xfs_buf          *bp)
 299 {
 300         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 301         struct xfs_mount        *mp = bp->b_target->bt_mount;
 302
 303         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 304                 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.s.bb_lsn)))
 305                         return false;
 306                 return xfs_buf_verify_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
 307         }
 308
 309         return true;
 310 }
 311
 312 static int
 313 xfs_btree_free_block(
 314         struct xfs_btree_cur    *cur,
 315         struct xfs_buf          *bp)
 316 {
 317         int                     error;
 318
 319         error = cur->bc_ops->free_block(cur, bp);
 320         if (!error) {
 321                 xfs_trans_binval(cur->bc_tp, bp);
 322                 XFS_BTREE_STATS_INC(cur, free);
 323         }
 324         return error;
 325 }
 326
 327 /*
 328  * Delete the btree cursor.
 329  */
 330 void
 331 xfs_btree_del_cursor(
 332         xfs_btree_cur_t *cur,           /* btree cursor */
 333         int             error)          /* del because of error */
 334 {
 335         int             i;              /* btree level */
 336
 337         /*
 338          * Clear the buffer pointers, and release the buffers.
 339          * If we're doing this in the face of an error, we
 340          * need to make sure to inspect all of the entries
 341          * in the bc_bufs array for buffers to be unlocked.
 342          * This is because some of the btree code works from
 343          * level n down to 0, and if we get an error along
 344          * the way we won't have initialized all the entries
 345          * down to 0.
 346          */
 347         for (i = 0; i < cur->bc_nlevels; i++) {
 348                 if (cur->bc_bufs[i])
 349                         xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
 350                 else if (!error)
 351                         break;
 352         }
 353         /*
 354          * Can't free a bmap cursor without having dealt with the
 355          * allocated indirect blocks' accounting.
 356          */
 357         ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP ||
 358                cur->bc_private.b.allocated == 0);
 359         /*
 360          * Free the cursor.
 361          */
 362         kmem_zone_free(xfs_btree_cur_zone, cur);
 363 }
 364
 365 /*
 366  * Duplicate the btree cursor.
 367  * Allocate a new one, copy the record, re-get the buffers.
 368  */
 369 int                                     /* error */
 370 xfs_btree_dup_cursor(
 371         xfs_btree_cur_t *cur,           /* input cursor */
 372         xfs_btree_cur_t **ncur)         /* output cursor */
 373 {
 374         xfs_buf_t       *bp;            /* btree block's buffer pointer */
 375         int             error;          /* error return value */
 376         int             i;              /* level number of btree block */
 377         xfs_mount_t     *mp;            /* mount structure for filesystem */
 378         xfs_btree_cur_t *new;           /* new cursor value */
 379         xfs_trans_t     *tp;            /* transaction pointer, can be NULL */
 380
 381         tp = cur->bc_tp;
 382         mp = cur->bc_mp;
 383
 384         /*
 385          * Allocate a new cursor like the old one.
 386          */
 387         new = cur->bc_ops->dup_cursor(cur);
 388
 389         /*
 390          * Copy the record currently in the cursor.
 391          */
 392         new->bc_rec = cur->bc_rec;
 393
 394         /*
 395          * For each level current, re-get the buffer and copy the ptr value.
 396          */
 397         for (i = 0; i < new->bc_nlevels; i++) {
 398                 new->bc_ptrs[i] = cur->bc_ptrs[i];
 399                 new->bc_ra[i] = cur->bc_ra[i];
 400                 bp = cur->bc_bufs[i];
 401                 if (bp) {
 402                         error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
 403                                                    XFS_BUF_ADDR(bp), mp->m_bsize,
 404                                                    0, &bp,
 405                                                    cur->bc_ops->buf_ops);
 406                         if (error) {
 407                                 xfs_btree_del_cursor(new, error);
 408                                 *ncur = NULL;
 409                                 return error;
 410                         }
 411                 }
 412                 new->bc_bufs[i] = bp;
 413         }
 414         *ncur = new;
 415         return 0;
 416 }
 417
 418 /*
 419  * XFS btree block layout and addressing:
 420  *
 421  * There are two types of blocks in the btree: leaf and non-leaf blocks.
 422  *
 423  * The leaf record start with a header then followed by records containing
 424  * the values.  A non-leaf block also starts with the same header, and
 425  * then first contains lookup keys followed by an equal number of pointers
 426  * to the btree blocks at the previous level.
 427  *
 428  *              +--------+-------+-------+-------+-------+-------+-------+
 429  * Leaf:        | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
 430  *              +--------+-------+-------+-------+-------+-------+-------+
 431  *
 432  *              +--------+-------+-------+-------+-------+-------+-------+
 433  * Non-Leaf:    | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
 434  *              +--------+-------+-------+-------+-------+-------+-------+
 435  *
 436  * The header is called struct xfs_btree_block for reasons better left unknown
 437  * and comes in different versions for short (32bit) and long (64bit) block
 438  * pointers.  The record and key structures are defined by the btree instances
 439  * and opaque to the btree core.  The block pointers are simple disk endian
 440  * integers, available in a short (32bit) and long (64bit) variant.
 441  *
 442  * The helpers below calculate the offset of a given record, key or pointer
 443  * into a btree block (xfs_btree_*_offset) or return a pointer to the given
 444  * record, key or pointer (xfs_btree_*_addr).  Note that all addressing
 445  * inside the btree block is done using indices starting at one, not zero!
 446  *
 447  * If XFS_BTREE_OVERLAPPING is set, then this btree supports keys containing
 448  * overlapping intervals.  In such a tree, records are still sorted lowest to
 449  * highest and indexed by the smallest key value that refers to the record.
 450  * However, nodes are different: each pointer has two associated keys -- one
 451  * indexing the lowest key available in the block(s) below (the same behavior
 452  * as the key in a regular btree) and another indexing the highest key
 453  * available in the block(s) below.  Because records are /not/ sorted by the
 454  * highest key, all leaf block updates require us to compute the highest key
 455  * that matches any record in the leaf and to recursively update the high keys
 456  * in the nodes going further up in the tree, if necessary.  Nodes look like
 457  * this:
 458  *
 459  *              +--------+-----+-----+-----+-----+-----+-------+-------+-----+
 460  * Non-Leaf:    | header | lo1 | hi1 | lo2 | hi2 | ... | ptr 1 | ptr 2 | ... |
 461  *              +--------+-----+-----+-----+-----+-----+-------+-------+-----+
 462  *
 463  * To perform an interval query on an overlapped tree, perform the usual
 464  * depth-first search and use the low and high keys to decide if we can skip
 465  * that particular node.  If a leaf node is reached, return the records that
 466  * intersect the interval.  Note that an interval query may return numerous
 467  * entries.  For a non-overlapped tree, simply search for the record associated
 468  * with the lowest key and iterate forward until a non-matching record is
 469  * found.  Section 14.3 ("Interval Trees") of _Introduction to Algorithms_ by
 470  * Cormen, Leiserson, Rivest, and Stein (2nd or 3rd ed. only) discuss this in
 471  * more detail.
 472  *
 473  * Why do we care about overlapping intervals?  Let's say you have a bunch of
 474  * reverse mapping records on a reflink filesystem:
 475  *
 476  * 1: +- file A startblock B offset C length D -----------+
 477  * 2:      +- file E startblock F offset G length H --------------+
 478  * 3:      +- file I startblock F offset J length K --+
 479  * 4:                                                        +- file L... --+
 480  *
 481  * Now say we want to map block (B+D) into file A at offset (C+D).  Ideally,
 482  * we'd simply increment the length of record 1.  But how do we find the record
 483  * that ends at (B+D-1) (i.e. record 1)?  A LE lookup of (B+D-1) would return
 484  * record 3 because the keys are ordered first by startblock.  An interval
 485  * query would return records 1 and 2 because they both overlap (B+D-1), and
 486  * from that we can pick out record 1 as the appropriate left neighbor.
 487  *
 488  * In the non-overlapped case you can do a LE lookup and decrement the cursor
 489  * because a record's interval must end before the next record.
 490  */
 491
 492 /*
 493  * Return size of the btree block header for this btree instance.
 494  */
 495 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
 496 {
 497         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 498                 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
 499                         return XFS_BTREE_LBLOCK_CRC_LEN;
 500                 return XFS_BTREE_LBLOCK_LEN;
 501         }
 502         if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
 503                 return XFS_BTREE_SBLOCK_CRC_LEN;
 504         return XFS_BTREE_SBLOCK_LEN;
 505 }
 506
 507 /*
 508  * Return size of btree block pointers for this btree instance.
 509  */
 510 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
 511 {
 512         return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
 513                 sizeof(__be64) : sizeof(__be32);
 514 }
 515
 516 /*
 517  * Calculate offset of the n-th record in a btree block.
 518  */
 519 STATIC size_t
 520 xfs_btree_rec_offset(
 521         struct xfs_btree_cur    *cur,
 522         int                     n)
 523 {
 524         return xfs_btree_block_len(cur) +
 525                 (n - 1) * cur->bc_ops->rec_len;
 526 }
 527
 528 /*
 529  * Calculate offset of the n-th key in a btree block.
 530  */
 531 STATIC size_t
 532 xfs_btree_key_offset(
 533         struct xfs_btree_cur    *cur,
 534         int                     n)
 535 {
 536         return xfs_btree_block_len(cur) +
 537                 (n - 1) * cur->bc_ops->key_len;
 538 }
 539
 540 /*
 541  * Calculate offset of the n-th high key in a btree block.
 542  */
 543 STATIC size_t
 544 xfs_btree_high_key_offset(
 545         struct xfs_btree_cur    *cur,
 546         int                     n)
 547 {
 548         return xfs_btree_block_len(cur) +
 549                 (n - 1) * cur->bc_ops->key_len + (cur->bc_ops->key_len / 2);
 550 }
 551
 552 /*
 553  * Calculate offset of the n-th block pointer in a btree block.
 554  */
 555 STATIC size_t
 556 xfs_btree_ptr_offset(
 557         struct xfs_btree_cur    *cur,
 558         int                     n,
 559         int                     level)
 560 {
 561         return xfs_btree_block_len(cur) +
 562                 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
 563                 (n - 1) * xfs_btree_ptr_len(cur);
 564 }
 565
 566 /*
 567  * Return a pointer to the n-th record in the btree block.
 568  */
 569 union xfs_btree_rec *
 570 xfs_btree_rec_addr(
 571         struct xfs_btree_cur    *cur,
 572         int                     n,
 573         struct xfs_btree_block  *block)
 574 {
 575         return (union xfs_btree_rec *)
 576                 ((char *)block + xfs_btree_rec_offset(cur, n));
 577 }
 578
 579 /*
 580  * Return a pointer to the n-th key in the btree block.
 581  */
 582 union xfs_btree_key *
 583 xfs_btree_key_addr(
 584         struct xfs_btree_cur    *cur,
 585         int                     n,
 586         struct xfs_btree_block  *block)
 587 {
 588         return (union xfs_btree_key *)
 589                 ((char *)block + xfs_btree_key_offset(cur, n));
 590 }
 591
 592 /*
 593  * Return a pointer to the n-th high key in the btree block.
 594  */
 595 union xfs_btree_key *
 596 xfs_btree_high_key_addr(
 597         struct xfs_btree_cur    *cur,
 598         int                     n,
 599         struct xfs_btree_block  *block)
 600 {
 601         return (union xfs_btree_key *)
 602                 ((char *)block + xfs_btree_high_key_offset(cur, n));
 603 }
 604
 605 /*
 606  * Return a pointer to the n-th block pointer in the btree block.
 607  */
 608 union xfs_btree_ptr *
 609 xfs_btree_ptr_addr(
 610         struct xfs_btree_cur    *cur,
 611         int                     n,
 612         struct xfs_btree_block  *block)
 613 {
 614         int                     level = xfs_btree_get_level(block);
 615
 616         ASSERT(block->bb_level != 0);
 617
 618         return (union xfs_btree_ptr *)
 619                 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
 620 }
 621
 622 /*
 623  * Get the root block which is stored in the inode.
 624  *
 625  * For now this btree implementation assumes the btree root is always
 626  * stored in the if_broot field of an inode fork.
 627  */
 628 STATIC struct xfs_btree_block *
 629 xfs_btree_get_iroot(
 630         struct xfs_btree_cur    *cur)
 631 {
 632         struct xfs_ifork        *ifp;
 633
 634         ifp = XFS_IFORK_PTR(cur->bc_private.b.ip, cur->bc_private.b.whichfork);
 635         return (struct xfs_btree_block *)ifp->if_broot;
 636 }
 637
 638 /*
 639  * Retrieve the block pointer from the cursor at the given level.
 640  * This may be an inode btree root or from a buffer.
 641  */
 642 struct xfs_btree_block *                /* generic btree block pointer */
 643 xfs_btree_get_block(
 644         struct xfs_btree_cur    *cur,   /* btree cursor */
 645         int                     level,  /* level in btree */
 646         struct xfs_buf          **bpp)  /* buffer containing the block */
 647 {
 648         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
 649             (level == cur->bc_nlevels - 1)) {
 650                 *bpp = NULL;
 651                 return xfs_btree_get_iroot(cur);
 652         }
 653
 654         *bpp = cur->bc_bufs[level];
 655         return XFS_BUF_TO_BLOCK(*bpp);
 656 }
 657
 658 /*
 659  * Get a buffer for the block, return it with no data read.
 660  * Long-form addressing.
 661  */
 662 xfs_buf_t *                             /* buffer for fsbno */
 663 xfs_btree_get_bufl(
 664         xfs_mount_t     *mp,            /* file system mount point */
 665         xfs_trans_t     *tp,            /* transaction pointer */
 666         xfs_fsblock_t   fsbno,          /* file system block number */
 667         uint            lock)           /* lock flags for get_buf */
 668 {
 669         xfs_daddr_t             d;              /* real disk block address */
 670
 671         ASSERT(fsbno != NULLFSBLOCK);
 672         d = XFS_FSB_TO_DADDR(mp, fsbno);
 673         return xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
 674 }
 675
 676 /*
 677  * Get a buffer for the block, return it with no data read.
 678  * Short-form addressing.
 679  */
 680 xfs_buf_t *                             /* buffer for agno/agbno */
 681 xfs_btree_get_bufs(
 682         xfs_mount_t     *mp,            /* file system mount point */
 683         xfs_trans_t     *tp,            /* transaction pointer */
 684         xfs_agnumber_t  agno,           /* allocation group number */
 685         xfs_agblock_t   agbno,          /* allocation group block number */
 686         uint            lock)           /* lock flags for get_buf */
 687 {
 688         xfs_daddr_t             d;              /* real disk block address */
 689
 690         ASSERT(agno != NULLAGNUMBER);
 691         ASSERT(agbno != NULLAGBLOCK);
 692         d = XFS_AGB_TO_DADDR(mp, agno, agbno);
 693         return xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
 694 }
 695
 696 /*
 697  * Check for the cursor referring to the last block at the given level.
 698  */
 699 int                                     /* 1=is last block, 0=not last block */
 700 xfs_btree_islastblock(
 701         xfs_btree_cur_t         *cur,   /* btree cursor */
 702         int                     level)  /* level to check */
 703 {
 704         struct xfs_btree_block  *block; /* generic btree block pointer */
 705         xfs_buf_t               *bp;    /* buffer containing block */
 706
 707         block = xfs_btree_get_block(cur, level, &bp);
 708         xfs_btree_check_block(cur, block, level, bp);
 709         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
 710                 return block->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK);
 711         else
 712                 return block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK);
 713 }
 714
 715 /*
 716  * Change the cursor to point to the first record at the given level.
 717  * Other levels are unaffected.
 718  */
 719 STATIC int                              /* success=1, failure=0 */
 720 xfs_btree_firstrec(
 721         xfs_btree_cur_t         *cur,   /* btree cursor */
 722         int                     level)  /* level to change */
 723 {
 724         struct xfs_btree_block  *block; /* generic btree block pointer */
 725         xfs_buf_t               *bp;    /* buffer containing block */
 726
 727         /*
 728          * Get the block pointer for this level.
 729          */
 730         block = xfs_btree_get_block(cur, level, &bp);
 731         xfs_btree_check_block(cur, block, level, bp);
 732         /*
 733          * It's empty, there is no such record.
 734          */
 735         if (!block->bb_numrecs)
 736                 return 0;
 737         /*
 738          * Set the ptr value to 1, that's the first record/key.
 739          */
 740         cur->bc_ptrs[level] = 1;
 741         return 1;
 742 }
 743
 744 /*
 745  * Change the cursor to point to the last record in the current block
 746  * at the given level.  Other levels are unaffected.
 747  */
 748 STATIC int                              /* success=1, failure=0 */
 749 xfs_btree_lastrec(
 750         xfs_btree_cur_t         *cur,   /* btree cursor */
 751         int                     level)  /* level to change */
 752 {
 753         struct xfs_btree_block  *block; /* generic btree block pointer */
 754         xfs_buf_t               *bp;    /* buffer containing block */
 755
 756         /*
 757          * Get the block pointer for this level.
 758          */
 759         block = xfs_btree_get_block(cur, level, &bp);
 760         xfs_btree_check_block(cur, block, level, bp);
 761         /*
 762          * It's empty, there is no such record.
 763          */
 764         if (!block->bb_numrecs)
 765                 return 0;
 766         /*
 767          * Set the ptr value to numrecs, that's the last record/key.
 768          */
 769         cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
 770         return 1;
 771 }
 772
 773 /*
 774  * Compute first and last byte offsets for the fields given.
 775  * Interprets the offsets table, which contains struct field offsets.
 776  */
 777 void
 778 xfs_btree_offsets(
 779         int64_t         fields,         /* bitmask of fields */
 780         const short     *offsets,       /* table of field offsets */
 781         int             nbits,          /* number of bits to inspect */
 782         int             *first,         /* output: first byte offset */
 783         int             *last)          /* output: last byte offset */
 784 {
 785         int             i;              /* current bit number */
 786         int64_t         imask;          /* mask for current bit number */
 787
 788         ASSERT(fields != 0);
 789         /*
 790          * Find the lowest bit, so the first byte offset.
 791          */
 792         for (i = 0, imask = 1LL; ; i++, imask <<= 1) {
 793                 if (imask & fields) {
 794                         *first = offsets[i];
 795                         break;
 796                 }
 797         }
 798         /*
 799          * Find the highest bit, so the last byte offset.
 800          */
 801         for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) {
 802                 if (imask & fields) {
 803                         *last = offsets[i + 1] - 1;
 804                         break;
 805                 }
 806         }
 807 }
 808
 809 /*
 810  * Get a buffer for the block, return it read in.
 811  * Long-form addressing.
 812  */
 813 int
 814 xfs_btree_read_bufl(
 815         struct xfs_mount        *mp,            /* file system mount point */
 816         struct xfs_trans        *tp,            /* transaction pointer */
 817         xfs_fsblock_t           fsbno,          /* file system block number */
 818         uint                    lock,           /* lock flags for read_buf */
 819         struct xfs_buf          **bpp,          /* buffer for fsbno */
 820         int                     refval,         /* ref count value for buffer */
 821         const struct xfs_buf_ops *ops)
 822 {
 823         struct xfs_buf          *bp;            /* return value */
 824         xfs_daddr_t             d;              /* real disk block address */
 825         int                     error;
 826
 827         if (!XFS_FSB_SANITY_CHECK(mp, fsbno))
 828                 return -EFSCORRUPTED;
 829         d = XFS_FSB_TO_DADDR(mp, fsbno);
 830         error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
 831                                    mp->m_bsize, lock, &bp, ops);
 832         if (error)
 833                 return error;
 834         if (bp)
 835                 xfs_buf_set_ref(bp, refval);
 836         *bpp = bp;
 837         return 0;
 838 }
 839
 840 /*
 841  * Read-ahead the block, don't wait for it, don't return a buffer.
 842  * Long-form addressing.
 843  */
 844 /* ARGSUSED */
 845 void
 846 xfs_btree_reada_bufl(
 847         struct xfs_mount        *mp,            /* file system mount point */
 848         xfs_fsblock_t           fsbno,          /* file system block number */
 849         xfs_extlen_t            count,          /* count of filesystem blocks */
 850         const struct xfs_buf_ops *ops)
 851 {
 852         xfs_daddr_t             d;
 853
 854         ASSERT(fsbno != NULLFSBLOCK);
 855         d = XFS_FSB_TO_DADDR(mp, fsbno);
 856         xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
 857 }
 858
 859 /*
 860  * Read-ahead the block, don't wait for it, don't return a buffer.
 861  * Short-form addressing.
 862  */
 863 /* ARGSUSED */
 864 void
 865 xfs_btree_reada_bufs(
 866         struct xfs_mount        *mp,            /* file system mount point */
 867         xfs_agnumber_t          agno,           /* allocation group number */
 868         xfs_agblock_t           agbno,          /* allocation group block number */
 869         xfs_extlen_t            count,          /* count of filesystem blocks */
 870         const struct xfs_buf_ops *ops)
 871 {
 872         xfs_daddr_t             d;
 873
 874         ASSERT(agno != NULLAGNUMBER);
 875         ASSERT(agbno != NULLAGBLOCK);
 876         d = XFS_AGB_TO_DADDR(mp, agno, agbno);
 877         xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
 878 }
 879
 880 STATIC int
 881 xfs_btree_readahead_lblock(
 882         struct xfs_btree_cur    *cur,
 883         int                     lr,
 884         struct xfs_btree_block  *block)
 885 {
 886         int                     rval = 0;
 887         xfs_fsblock_t           left = be64_to_cpu(block->bb_u.l.bb_leftsib);
 888         xfs_fsblock_t           right = be64_to_cpu(block->bb_u.l.bb_rightsib);
 889
 890         if ((lr & XFS_BTCUR_LEFTRA) && left != NULLFSBLOCK) {
 891                 xfs_btree_reada_bufl(cur->bc_mp, left, 1,
 892                                      cur->bc_ops->buf_ops);
 893                 rval++;
 894         }
 895
 896         if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLFSBLOCK) {
 897                 xfs_btree_reada_bufl(cur->bc_mp, right, 1,
 898                                      cur->bc_ops->buf_ops);
 899                 rval++;
 900         }
 901
 902         return rval;
 903 }
 904
 905 STATIC int
 906 xfs_btree_readahead_sblock(
 907         struct xfs_btree_cur    *cur,
 908         int                     lr,
 909         struct xfs_btree_block *block)
 910 {
 911         int                     rval = 0;
 912         xfs_agblock_t           left = be32_to_cpu(block->bb_u.s.bb_leftsib);
 913         xfs_agblock_t           right = be32_to_cpu(block->bb_u.s.bb_rightsib);
 914
 915
 916         if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
 917                 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
 918                                      left, 1, cur->bc_ops->buf_ops);
 919                 rval++;
 920         }
 921
 922         if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
 923                 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
 924                                      right, 1, cur->bc_ops->buf_ops);
 925                 rval++;
 926         }
 927
 928         return rval;
 929 }
 930
 931 /*
 932  * Read-ahead btree blocks, at the given level.
 933  * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
 934  */
 935 STATIC int
 936 xfs_btree_readahead(
 937         struct xfs_btree_cur    *cur,           /* btree cursor */
 938         int                     lev,            /* level in btree */
 939         int                     lr)             /* left/right bits */
 940 {
 941         struct xfs_btree_block  *block;
 942
 943         /*
 944          * No readahead needed if we are at the root level and the
 945          * btree root is stored in the inode.
 946          */
 947         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
 948             (lev == cur->bc_nlevels - 1))
 949                 return 0;
 950
 951         if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
 952                 return 0;
 953
 954         cur->bc_ra[lev] |= lr;
 955         block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
 956
 957         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
 958                 return xfs_btree_readahead_lblock(cur, lr, block);
 959         return xfs_btree_readahead_sblock(cur, lr, block);
 960 }
 961
 962 STATIC xfs_daddr_t
 963 xfs_btree_ptr_to_daddr(
 964         struct xfs_btree_cur    *cur,
 965         union xfs_btree_ptr     *ptr)
 966 {
 967         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 968                 ASSERT(ptr->l != cpu_to_be64(NULLFSBLOCK));
 969
 970                 return XFS_FSB_TO_DADDR(cur->bc_mp, be64_to_cpu(ptr->l));
 971         } else {
 972                 ASSERT(cur->bc_private.a.agno != NULLAGNUMBER);
 973                 ASSERT(ptr->s != cpu_to_be32(NULLAGBLOCK));
 974
 975                 return XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_private.a.agno,
 976                                         be32_to_cpu(ptr->s));
 977         }
 978 }
 979
 980 /*
 981  * Readahead @count btree blocks at the given @ptr location.
 982  *
 983  * We don't need to care about long or short form btrees here as we have a
 984  * method of converting the ptr directly to a daddr available to us.
 985  */
 986 STATIC void
 987 xfs_btree_readahead_ptr(
 988         struct xfs_btree_cur    *cur,
 989         union xfs_btree_ptr     *ptr,
 990         xfs_extlen_t            count)
 991 {
 992         xfs_buf_readahead(cur->bc_mp->m_ddev_targp,
 993                           xfs_btree_ptr_to_daddr(cur, ptr),
 994                           cur->bc_mp->m_bsize * count, cur->bc_ops->buf_ops);
 995 }
 996
 997 /*
 998  * Set the buffer for level "lev" in the cursor to bp, releasing
 999  * any previous buffer.
1000  */
1001 STATIC void
1002 xfs_btree_setbuf(
1003         xfs_btree_cur_t         *cur,   /* btree cursor */
1004         int                     lev,    /* level in btree */
1005         xfs_buf_t               *bp)    /* new buffer to set */
1006 {
1007         struct xfs_btree_block  *b;     /* btree block */
1008
1009         if (cur->bc_bufs[lev])
1010                 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
1011         cur->bc_bufs[lev] = bp;
1012         cur->bc_ra[lev] = 0;
1013
1014         b = XFS_BUF_TO_BLOCK(bp);
1015         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1016                 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK))
1017                         cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1018                 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK))
1019                         cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1020         } else {
1021                 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
1022                         cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1023                 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
1024                         cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1025         }
1026 }
1027
1028 STATIC int
1029 xfs_btree_ptr_is_null(
1030         struct xfs_btree_cur    *cur,
1031         union xfs_btree_ptr     *ptr)
1032 {
1033         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1034                 return ptr->l == cpu_to_be64(NULLFSBLOCK);
1035         else
1036                 return ptr->s == cpu_to_be32(NULLAGBLOCK);
1037 }
1038
1039 STATIC void
1040 xfs_btree_set_ptr_null(
1041         struct xfs_btree_cur    *cur,
1042         union xfs_btree_ptr     *ptr)
1043 {
1044         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1045                 ptr->l = cpu_to_be64(NULLFSBLOCK);
1046         else
1047                 ptr->s = cpu_to_be32(NULLAGBLOCK);
1048 }
1049
1050 /*
1051  * Get/set/init sibling pointers
1052  */
1053 STATIC void
1054 xfs_btree_get_sibling(
1055         struct xfs_btree_cur    *cur,
1056         struct xfs_btree_block  *block,
1057         union xfs_btree_ptr     *ptr,
1058         int                     lr)
1059 {
1060         ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1061
1062         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1063                 if (lr == XFS_BB_RIGHTSIB)
1064                         ptr->l = block->bb_u.l.bb_rightsib;
1065                 else
1066                         ptr->l = block->bb_u.l.bb_leftsib;
1067         } else {
1068                 if (lr == XFS_BB_RIGHTSIB)
1069                         ptr->s = block->bb_u.s.bb_rightsib;
1070                 else
1071                         ptr->s = block->bb_u.s.bb_leftsib;
1072         }
1073 }
1074
1075 STATIC void
1076 xfs_btree_set_sibling(
1077         struct xfs_btree_cur    *cur,
1078         struct xfs_btree_block  *block,
1079         union xfs_btree_ptr     *ptr,
1080         int                     lr)
1081 {
1082         ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1083
1084         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1085                 if (lr == XFS_BB_RIGHTSIB)
1086                         block->bb_u.l.bb_rightsib = ptr->l;
1087                 else
1088                         block->bb_u.l.bb_leftsib = ptr->l;
1089         } else {
1090                 if (lr == XFS_BB_RIGHTSIB)
1091                         block->bb_u.s.bb_rightsib = ptr->s;
1092                 else
1093                         block->bb_u.s.bb_leftsib = ptr->s;
1094         }
1095 }
1096
1097 void
1098 xfs_btree_init_block_int(
1099         struct xfs_mount        *mp,
1100         struct xfs_btree_block  *buf,
1101         xfs_daddr_t             blkno,
1102         xfs_btnum_t             btnum,
1103         __u16                   level,
1104         __u16                   numrecs,
1105         __u64                   owner,
1106         unsigned int            flags)
1107 {
1108         int                     crc = xfs_sb_version_hascrc(&mp->m_sb);
1109         __u32                   magic = xfs_btree_magic(crc, btnum);
1110
1111         buf->bb_magic = cpu_to_be32(magic);
1112         buf->bb_level = cpu_to_be16(level);
1113         buf->bb_numrecs = cpu_to_be16(numrecs);
1114
1115         if (flags & XFS_BTREE_LONG_PTRS) {
1116                 buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK);
1117                 buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK);
1118                 if (crc) {
1119                         buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
1120                         buf->bb_u.l.bb_owner = cpu_to_be64(owner);
1121                         uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid);
1122                         buf->bb_u.l.bb_pad = 0;
1123                         buf->bb_u.l.bb_lsn = 0;
1124                 }
1125         } else {
1126                 /* owner is a 32 bit value on short blocks */
1127                 __u32 __owner = (__u32)owner;
1128
1129                 buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1130                 buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1131                 if (crc) {
1132                         buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
1133                         buf->bb_u.s.bb_owner = cpu_to_be32(__owner);
1134                         uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid);
1135                         buf->bb_u.s.bb_lsn = 0;
1136                 }
1137         }
1138 }
1139
1140 void
1141 xfs_btree_init_block(
1142         struct xfs_mount *mp,
1143         struct xfs_buf  *bp,
1144         xfs_btnum_t     btnum,
1145         __u16           level,
1146         __u16           numrecs,
1147         __u64           owner,
1148         unsigned int    flags)
1149 {
1150         xfs_btree_init_block_int(mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1151                                  btnum, level, numrecs, owner, flags);
1152 }
1153
1154 STATIC void
1155 xfs_btree_init_block_cur(
1156         struct xfs_btree_cur    *cur,
1157         struct xfs_buf          *bp,
1158         int                     level,
1159         int                     numrecs)
1160 {
1161         __u64                   owner;
1162
1163         /*
1164          * we can pull the owner from the cursor right now as the different
1165          * owners align directly with the pointer size of the btree. This may
1166          * change in future, but is safe for current users of the generic btree
1167          * code.
1168          */
1169         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1170                 owner = cur->bc_private.b.ip->i_ino;
1171         else
1172                 owner = cur->bc_private.a.agno;
1173
1174         xfs_btree_init_block_int(cur->bc_mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1175                                  cur->bc_btnum, level, numrecs,
1176                                  owner, cur->bc_flags);
1177 }
1178
1179 /*
1180  * Return true if ptr is the last record in the btree and
1181  * we need to track updates to this record.  The decision
1182  * will be further refined in the update_lastrec method.
1183  */
1184 STATIC int
1185 xfs_btree_is_lastrec(
1186         struct xfs_btree_cur    *cur,
1187         struct xfs_btree_block  *block,
1188         int                     level)
1189 {
1190         union xfs_btree_ptr     ptr;
1191
1192         if (level > 0)
1193                 return 0;
1194         if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
1195                 return 0;
1196
1197         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1198         if (!xfs_btree_ptr_is_null(cur, &ptr))
1199                 return 0;
1200         return 1;
1201 }
1202
1203 STATIC void
1204 xfs_btree_buf_to_ptr(
1205         struct xfs_btree_cur    *cur,
1206         struct xfs_buf          *bp,
1207         union xfs_btree_ptr     *ptr)
1208 {
1209         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1210                 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1211                                         XFS_BUF_ADDR(bp)));
1212         else {
1213                 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
1214                                         XFS_BUF_ADDR(bp)));
1215         }
1216 }
1217
1218 STATIC void
1219 xfs_btree_set_refs(
1220         struct xfs_btree_cur    *cur,
1221         struct xfs_buf          *bp)
1222 {
1223         switch (cur->bc_btnum) {
1224         case XFS_BTNUM_BNO:
1225         case XFS_BTNUM_CNT:
1226                 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
1227                 break;
1228         case XFS_BTNUM_INO:
1229         case XFS_BTNUM_FINO:
1230                 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
1231                 break;
1232         case XFS_BTNUM_BMAP:
1233                 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
1234                 break;
1235         case XFS_BTNUM_RMAP:
1236                 xfs_buf_set_ref(bp, XFS_RMAP_BTREE_REF);
1237                 break;
1238         case XFS_BTNUM_REFC:
1239                 xfs_buf_set_ref(bp, XFS_REFC_BTREE_REF);
1240                 break;
1241         default:
1242                 ASSERT(0);
1243         }
1244 }
1245
1246 STATIC int
1247 xfs_btree_get_buf_block(
1248         struct xfs_btree_cur    *cur,
1249         union xfs_btree_ptr     *ptr,
1250         int                     flags,
1251         struct xfs_btree_block  **block,
1252         struct xfs_buf          **bpp)
1253 {
1254         struct xfs_mount        *mp = cur->bc_mp;
1255         xfs_daddr_t             d;
1256
1257         /* need to sort out how callers deal with failures first */
1258         ASSERT(!(flags & XBF_TRYLOCK));
1259
1260         d = xfs_btree_ptr_to_daddr(cur, ptr);
1261         *bpp = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d,
1262                                  mp->m_bsize, flags);
1263
1264         if (!*bpp)
1265                 return -ENOMEM;
1266
1267         (*bpp)->b_ops = cur->bc_ops->buf_ops;
1268         *block = XFS_BUF_TO_BLOCK(*bpp);
1269         return 0;
1270 }
1271
1272 /*
1273  * Read in the buffer at the given ptr and return the buffer and
1274  * the block pointer within the buffer.
1275  */
1276 STATIC int
1277 xfs_btree_read_buf_block(
1278         struct xfs_btree_cur    *cur,
1279         union xfs_btree_ptr     *ptr,
1280         int                     flags,
1281         struct xfs_btree_block  **block,
1282         struct xfs_buf          **bpp)
1283 {
1284         struct xfs_mount        *mp = cur->bc_mp;
1285         xfs_daddr_t             d;
1286         int                     error;
1287
1288         /* need to sort out how callers deal with failures first */
1289         ASSERT(!(flags & XBF_TRYLOCK));
1290
1291         d = xfs_btree_ptr_to_daddr(cur, ptr);
1292         error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
1293                                    mp->m_bsize, flags, bpp,
1294                                    cur->bc_ops->buf_ops);
1295         if (error)
1296                 return error;
1297
1298         xfs_btree_set_refs(cur, *bpp);
1299         *block = XFS_BUF_TO_BLOCK(*bpp);
1300         return 0;
1301 }
1302
1303 /*
1304  * Copy keys from one btree block to another.
1305  */
1306 STATIC void
1307 xfs_btree_copy_keys(
1308         struct xfs_btree_cur    *cur,
1309         union xfs_btree_key     *dst_key,
1310         union xfs_btree_key     *src_key,
1311         int                     numkeys)
1312 {
1313         ASSERT(numkeys >= 0);
1314         memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1315 }
1316
1317 /*
1318  * Copy records from one btree block to another.
1319  */
1320 STATIC void
1321 xfs_btree_copy_recs(
1322         struct xfs_btree_cur    *cur,
1323         union xfs_btree_rec     *dst_rec,
1324         union xfs_btree_rec     *src_rec,
1325         int                     numrecs)
1326 {
1327         ASSERT(numrecs >= 0);
1328         memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1329 }
1330
1331 /*
1332  * Copy block pointers from one btree block to another.
1333  */
1334 STATIC void
1335 xfs_btree_copy_ptrs(
1336         struct xfs_btree_cur    *cur,
1337         union xfs_btree_ptr     *dst_ptr,
1338         union xfs_btree_ptr     *src_ptr,
1339         int                     numptrs)
1340 {
1341         ASSERT(numptrs >= 0);
1342         memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1343 }
1344
1345 /*
1346  * Shift keys one index left/right inside a single btree block.
1347  */
1348 STATIC void
1349 xfs_btree_shift_keys(
1350         struct xfs_btree_cur    *cur,
1351         union xfs_btree_key     *key,
1352         int                     dir,
1353         int                     numkeys)
1354 {
1355         char                    *dst_key;
1356
1357         ASSERT(numkeys >= 0);
1358         ASSERT(dir == 1 || dir == -1);
1359
1360         dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1361         memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1362 }
1363
1364 /*
1365  * Shift records one index left/right inside a single btree block.
1366  */
1367 STATIC void
1368 xfs_btree_shift_recs(
1369         struct xfs_btree_cur    *cur,
1370         union xfs_btree_rec     *rec,
1371         int                     dir,
1372         int                     numrecs)
1373 {
1374         char                    *dst_rec;
1375
1376         ASSERT(numrecs >= 0);
1377         ASSERT(dir == 1 || dir == -1);
1378
1379         dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1380         memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1381 }
1382
1383 /*
1384  * Shift block pointers one index left/right inside a single btree block.
1385  */
1386 STATIC void
1387 xfs_btree_shift_ptrs(
1388         struct xfs_btree_cur    *cur,
1389         union xfs_btree_ptr     *ptr,
1390         int                     dir,
1391         int                     numptrs)
1392 {
1393         char                    *dst_ptr;
1394
1395         ASSERT(numptrs >= 0);
1396         ASSERT(dir == 1 || dir == -1);
1397
1398         dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1399         memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1400 }
1401
1402 /*
1403  * Log key values from the btree block.
1404  */
1405 STATIC void
1406 xfs_btree_log_keys(
1407         struct xfs_btree_cur    *cur,
1408         struct xfs_buf          *bp,
1409         int                     first,
1410         int                     last)
1411 {
1412         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1413         XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1414
1415         if (bp) {
1416                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1417                 xfs_trans_log_buf(cur->bc_tp, bp,
1418                                   xfs_btree_key_offset(cur, first),
1419                                   xfs_btree_key_offset(cur, last + 1) - 1);
1420         } else {
1421                 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1422                                 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1423         }
1424
1425         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1426 }
1427
1428 /*
1429  * Log record values from the btree block.
1430  */
1431 void
1432 xfs_btree_log_recs(
1433         struct xfs_btree_cur    *cur,
1434         struct xfs_buf          *bp,
1435         int                     first,
1436         int                     last)
1437 {
1438         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1439         XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1440
1441         xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1442         xfs_trans_log_buf(cur->bc_tp, bp,
1443                           xfs_btree_rec_offset(cur, first),
1444                           xfs_btree_rec_offset(cur, last + 1) - 1);
1445
1446         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1447 }
1448
1449 /*
1450  * Log block pointer fields from a btree block (nonleaf).
1451  */
1452 STATIC void
1453 xfs_btree_log_ptrs(
1454         struct xfs_btree_cur    *cur,   /* btree cursor */
1455         struct xfs_buf          *bp,    /* buffer containing btree block */
1456         int                     first,  /* index of first pointer to log */
1457         int                     last)   /* index of last pointer to log */
1458 {
1459         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1460         XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1461
1462         if (bp) {
1463                 struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
1464                 int                     level = xfs_btree_get_level(block);
1465
1466                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1467                 xfs_trans_log_buf(cur->bc_tp, bp,
1468                                 xfs_btree_ptr_offset(cur, first, level),
1469                                 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1470         } else {
1471                 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1472                         xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1473         }
1474
1475         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1476 }
1477
1478 /*
1479  * Log fields from a btree block header.
1480  */
1481 void
1482 xfs_btree_log_block(
1483         struct xfs_btree_cur    *cur,   /* btree cursor */
1484         struct xfs_buf          *bp,    /* buffer containing btree block */
1485         int                     fields) /* mask of fields: XFS_BB_... */
1486 {
1487         int                     first;  /* first byte offset logged */
1488         int                     last;   /* last byte offset logged */
1489         static const short      soffsets[] = {  /* table of offsets (short) */
1490                 offsetof(struct xfs_btree_block, bb_magic),
1491                 offsetof(struct xfs_btree_block, bb_level),
1492                 offsetof(struct xfs_btree_block, bb_numrecs),
1493                 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1494                 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1495                 offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1496                 offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1497                 offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1498                 offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1499                 offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1500                 XFS_BTREE_SBLOCK_CRC_LEN
1501         };
1502         static const short      loffsets[] = {  /* table of offsets (long) */
1503                 offsetof(struct xfs_btree_block, bb_magic),
1504                 offsetof(struct xfs_btree_block, bb_level),
1505                 offsetof(struct xfs_btree_block, bb_numrecs),
1506                 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1507                 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1508                 offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1509                 offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1510                 offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1511                 offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1512                 offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1513                 offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1514                 XFS_BTREE_LBLOCK_CRC_LEN
1515         };
1516
1517         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1518         XFS_BTREE_TRACE_ARGBI(cur, bp, fields);
1519
1520         if (bp) {
1521                 int nbits;
1522
1523                 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
1524                         /*
1525                          * We don't log the CRC when updating a btree
1526                          * block but instead recreate it during log
1527                          * recovery.  As the log buffers have checksums
1528                          * of their own this is safe and avoids logging a crc
1529                          * update in a lot of places.
1530                          */
1531                         if (fields == XFS_BB_ALL_BITS)
1532                                 fields = XFS_BB_ALL_BITS_CRC;
1533                         nbits = XFS_BB_NUM_BITS_CRC;
1534                 } else {
1535                         nbits = XFS_BB_NUM_BITS;
1536                 }
1537                 xfs_btree_offsets(fields,
1538                                   (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1539                                         loffsets : soffsets,
1540                                   nbits, &first, &last);
1541                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1542                 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1543         } else {
1544                 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1545                         xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1546         }
1547
1548         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1549 }
1550
1551 /*
1552  * Increment cursor by one record at the level.
1553  * For nonzero levels the leaf-ward information is untouched.
1554  */
1555 int                                             /* error */
1556 xfs_btree_increment(
1557         struct xfs_btree_cur    *cur,
1558         int                     level,
1559         int                     *stat)          /* success/failure */
1560 {
1561         struct xfs_btree_block  *block;
1562         union xfs_btree_ptr     ptr;
1563         struct xfs_buf          *bp;
1564         int                     error;          /* error return value */
1565         int                     lev;
1566
1567         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1568         XFS_BTREE_TRACE_ARGI(cur, level);
1569
1570         ASSERT(level < cur->bc_nlevels);
1571
1572         /* Read-ahead to the right at this level. */
1573         xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1574
1575         /* Get a pointer to the btree block. */
1576         block = xfs_btree_get_block(cur, level, &bp);
1577
1578 #ifdef DEBUG
1579         error = xfs_btree_check_block(cur, block, level, bp);
1580         if (error)
1581                 goto error0;
1582 #endif
1583
1584         /* We're done if we remain in the block after the increment. */
1585         if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
1586                 goto out1;
1587
1588         /* Fail if we just went off the right edge of the tree. */
1589         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1590         if (xfs_btree_ptr_is_null(cur, &ptr))
1591                 goto out0;
1592
1593         XFS_BTREE_STATS_INC(cur, increment);
1594
1595         /*
1596          * March up the tree incrementing pointers.
1597          * Stop when we don't go off the right edge of a block.
1598          */
1599         for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1600                 block = xfs_btree_get_block(cur, lev, &bp);
1601
1602 #ifdef DEBUG
1603                 error = xfs_btree_check_block(cur, block, lev, bp);
1604                 if (error)
1605                         goto error0;
1606 #endif
1607
1608                 if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
1609                         break;
1610
1611                 /* Read-ahead the right block for the next loop. */
1612                 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1613         }
1614
1615         /*
1616          * If we went off the root then we are either seriously
1617          * confused or have the tree root in an inode.
1618          */
1619         if (lev == cur->bc_nlevels) {
1620                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1621                         goto out0;
1622                 ASSERT(0);
1623                 error = -EFSCORRUPTED;
1624                 goto error0;
1625         }
1626         ASSERT(lev < cur->bc_nlevels);
1627
1628         /*
1629          * Now walk back down the tree, fixing up the cursor's buffer
1630          * pointers and key numbers.
1631          */
1632         for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1633                 union xfs_btree_ptr     *ptrp;
1634
1635                 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1636                 --lev;
1637                 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1638                 if (error)
1639                         goto error0;
1640
1641                 xfs_btree_setbuf(cur, lev, bp);
1642                 cur->bc_ptrs[lev] = 1;
1643         }
1644 out1:
1645         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1646         *stat = 1;
1647         return 0;
1648
1649 out0:
1650         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1651         *stat = 0;
1652         return 0;
1653
1654 error0:
1655         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1656         return error;
1657 }
1658
1659 /*
1660  * Decrement cursor by one record at the level.
1661  * For nonzero levels the leaf-ward information is untouched.
1662  */
1663 int                                             /* error */
1664 xfs_btree_decrement(
1665         struct xfs_btree_cur    *cur,
1666         int                     level,
1667         int                     *stat)          /* success/failure */
1668 {
1669         struct xfs_btree_block  *block;
1670         xfs_buf_t               *bp;
1671         int                     error;          /* error return value */
1672         int                     lev;
1673         union xfs_btree_ptr     ptr;
1674
1675         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1676         XFS_BTREE_TRACE_ARGI(cur, level);
1677
1678         ASSERT(level < cur->bc_nlevels);
1679
1680         /* Read-ahead to the left at this level. */
1681         xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1682
1683         /* We're done if we remain in the block after the decrement. */
1684         if (--cur->bc_ptrs[level] > 0)
1685                 goto out1;
1686
1687         /* Get a pointer to the btree block. */
1688         block = xfs_btree_get_block(cur, level, &bp);
1689
1690 #ifdef DEBUG
1691         error = xfs_btree_check_block(cur, block, level, bp);
1692         if (error)
1693                 goto error0;
1694 #endif
1695
1696         /* Fail if we just went off the left edge of the tree. */
1697         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1698         if (xfs_btree_ptr_is_null(cur, &ptr))
1699                 goto out0;
1700
1701         XFS_BTREE_STATS_INC(cur, decrement);
1702
1703         /*
1704          * March up the tree decrementing pointers.
1705          * Stop when we don't go off the left edge of a block.
1706          */
1707         for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1708                 if (--cur->bc_ptrs[lev] > 0)
1709                         break;
1710                 /* Read-ahead the left block for the next loop. */
1711                 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1712         }
1713
1714         /*
1715          * If we went off the root then we are seriously confused.
1716          * or the root of the tree is in an inode.
1717          */
1718         if (lev == cur->bc_nlevels) {
1719                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1720                         goto out0;
1721                 ASSERT(0);
1722                 error = -EFSCORRUPTED;
1723                 goto error0;
1724         }
1725         ASSERT(lev < cur->bc_nlevels);
1726
1727         /*
1728          * Now walk back down the tree, fixing up the cursor's buffer
1729          * pointers and key numbers.
1730          */
1731         for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1732                 union xfs_btree_ptr     *ptrp;
1733
1734                 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1735                 --lev;
1736                 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1737                 if (error)
1738                         goto error0;
1739                 xfs_btree_setbuf(cur, lev, bp);
1740                 cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
1741         }
1742 out1:
1743         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1744         *stat = 1;
1745         return 0;
1746
1747 out0:
1748         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1749         *stat = 0;
1750         return 0;
1751
1752 error0:
1753         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1754         return error;
1755 }
1756
1757 int
1758 xfs_btree_lookup_get_block(
1759         struct xfs_btree_cur    *cur,   /* btree cursor */
1760         int                     level,  /* level in the btree */
1761         union xfs_btree_ptr     *pp,    /* ptr to btree block */
1762         struct xfs_btree_block  **blkp) /* return btree block */
1763 {
1764         struct xfs_buf          *bp;    /* buffer pointer for btree block */
1765         int                     error = 0;
1766
1767         /* special case the root block if in an inode */
1768         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1769             (level == cur->bc_nlevels - 1)) {
1770                 *blkp = xfs_btree_get_iroot(cur);
1771                 return 0;
1772         }
1773
1774         /*
1775          * If the old buffer at this level for the disk address we are
1776          * looking for re-use it.
1777          *
1778          * Otherwise throw it away and get a new one.
1779          */
1780         bp = cur->bc_bufs[level];
1781         if (bp && XFS_BUF_ADDR(bp) == xfs_btree_ptr_to_daddr(cur, pp)) {
1782                 *blkp = XFS_BUF_TO_BLOCK(bp);
1783                 return 0;
1784         }
1785
1786         error = xfs_btree_read_buf_block(cur, pp, 0, blkp, &bp);
1787         if (error)
1788                 return error;
1789
1790         /* Check the inode owner since the verifiers don't. */
1791         if (xfs_sb_version_hascrc(&cur->bc_mp->m_sb) &&
1792             (cur->bc_flags & XFS_BTREE_LONG_PTRS) &&
1793             be64_to_cpu((*blkp)->bb_u.l.bb_owner) !=
1794                         cur->bc_private.b.ip->i_ino)
1795                 goto out_bad;
1796
1797         /* Did we get the level we were looking for? */
1798         if (be16_to_cpu((*blkp)->bb_level) != level)
1799                 goto out_bad;
1800
1801         /* Check that internal nodes have at least one record. */
1802         if (level != 0 && be16_to_cpu((*blkp)->bb_numrecs) == 0)
1803                 goto out_bad;
1804
1805         xfs_btree_setbuf(cur, level, bp);
1806         return 0;
1807
1808 out_bad:
1809         *blkp = NULL;
1810         xfs_trans_brelse(cur->bc_tp, bp);
1811         return -EFSCORRUPTED;
1812 }
1813
1814 /*
1815  * Get current search key.  For level 0 we don't actually have a key
1816  * structure so we make one up from the record.  For all other levels
1817  * we just return the right key.
1818  */
1819 STATIC union xfs_btree_key *
1820 xfs_lookup_get_search_key(
1821         struct xfs_btree_cur    *cur,
1822         int                     level,
1823         int                     keyno,
1824         struct xfs_btree_block  *block,
1825         union xfs_btree_key     *kp)
1826 {
1827         if (level == 0) {
1828                 cur->bc_ops->init_key_from_rec(kp,
1829                                 xfs_btree_rec_addr(cur, keyno, block));
1830                 return kp;
1831         }
1832
1833         return xfs_btree_key_addr(cur, keyno, block);
1834 }
1835
1836 /*
1837  * Lookup the record.  The cursor is made to point to it, based on dir.
1838  * stat is set to 0 if can't find any such record, 1 for success.
1839  */
1840 int                                     /* error */
1841 xfs_btree_lookup(
1842         struct xfs_btree_cur    *cur,   /* btree cursor */
1843         xfs_lookup_t            dir,    /* <=, ==, or >= */
1844         int                     *stat)  /* success/failure */
1845 {
1846         struct xfs_btree_block  *block; /* current btree block */
1847         int64_t                 diff;   /* difference for the current key */
1848         int                     error;  /* error return value */
1849         int                     keyno;  /* current key number */
1850         int                     level;  /* level in the btree */
1851         union xfs_btree_ptr     *pp;    /* ptr to btree block */
1852         union xfs_btree_ptr     ptr;    /* ptr to btree block */
1853
1854         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1855         XFS_BTREE_TRACE_ARGI(cur, dir);
1856
1857         XFS_BTREE_STATS_INC(cur, lookup);
1858
1859         /* No such thing as a zero-level tree. */
1860         if (cur->bc_nlevels == 0)
1861                 return -EFSCORRUPTED;
1862
1863         block = NULL;
1864         keyno = 0;
1865
1866         /* initialise start pointer from cursor */
1867         cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1868         pp = &ptr;
1869
1870         /*
1871          * Iterate over each level in the btree, starting at the root.
1872          * For each level above the leaves, find the key we need, based
1873          * on the lookup record, then follow the corresponding block
1874          * pointer down to the next level.
1875          */
1876         for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1877                 /* Get the block we need to do the lookup on. */
1878                 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1879                 if (error)
1880                         goto error0;
1881
1882                 if (diff == 0) {
1883                         /*
1884                          * If we already had a key match at a higher level, we
1885                          * know we need to use the first entry in this block.
1886                          */
1887                         keyno = 1;
1888                 } else {
1889                         /* Otherwise search this block. Do a binary search. */
1890
1891                         int     high;   /* high entry number */
1892                         int     low;    /* low entry number */
1893
1894                         /* Set low and high entry numbers, 1-based. */
1895                         low = 1;
1896                         high = xfs_btree_get_numrecs(block);
1897                         if (!high) {
1898                                 /* Block is empty, must be an empty leaf. */
1899                                 ASSERT(level == 0 && cur->bc_nlevels == 1);
1900
1901                                 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1902                                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1903                                 *stat = 0;
1904                                 return 0;
1905                         }
1906
1907                         /* Binary search the block. */
1908                         while (low <= high) {
1909                                 union xfs_btree_key     key;
1910                                 union xfs_btree_key     *kp;
1911
1912                                 XFS_BTREE_STATS_INC(cur, compare);
1913
1914                                 /* keyno is average of low and high. */
1915                                 keyno = (low + high) >> 1;
1916
1917                                 /* Get current search key */
1918                                 kp = xfs_lookup_get_search_key(cur, level,
1919                                                 keyno, block, &key);
1920
1921                                 /*
1922                                  * Compute difference to get next direction:
1923                                  *  - less than, move right
1924                                  *  - greater than, move left
1925                                  *  - equal, we're done
1926                                  */
1927                                 diff = cur->bc_ops->key_diff(cur, kp);
1928                                 if (diff < 0)
1929                                         low = keyno + 1;
1930                                 else if (diff > 0)
1931                                         high = keyno - 1;
1932                                 else
1933                                         break;
1934                         }
1935                 }
1936
1937                 /*
1938                  * If there are more levels, set up for the next level
1939                  * by getting the block number and filling in the cursor.
1940                  */
1941                 if (level > 0) {
1942                         /*
1943                          * If we moved left, need the previous key number,
1944                          * unless there isn't one.
1945                          */
1946                         if (diff > 0 && --keyno < 1)
1947                                 keyno = 1;
1948                         pp = xfs_btree_ptr_addr(cur, keyno, block);
1949
1950 #ifdef DEBUG
1951                         error = xfs_btree_check_ptr(cur, pp, 0, level);
1952                         if (error)
1953                                 goto error0;
1954 #endif
1955                         cur->bc_ptrs[level] = keyno;
1956                 }
1957         }
1958
1959         /* Done with the search. See if we need to adjust the results. */
1960         if (dir != XFS_LOOKUP_LE && diff < 0) {
1961                 keyno++;
1962                 /*
1963                  * If ge search and we went off the end of the block, but it's
1964                  * not the last block, we're in the wrong block.
1965                  */
1966                 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1967                 if (dir == XFS_LOOKUP_GE &&
1968                     keyno > xfs_btree_get_numrecs(block) &&
1969                     !xfs_btree_ptr_is_null(cur, &ptr)) {
1970                         int     i;
1971
1972                         cur->bc_ptrs[0] = keyno;
1973                         error = xfs_btree_increment(cur, 0, &i);
1974                         if (error)
1975                                 goto error0;
1976                         XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1977                         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1978                         *stat = 1;
1979                         return 0;
1980                 }
1981         } else if (dir == XFS_LOOKUP_LE && diff > 0)
1982                 keyno--;
1983         cur->bc_ptrs[0] = keyno;
1984
1985         /* Return if we succeeded or not. */
1986         if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
1987                 *stat = 0;
1988         else if (dir != XFS_LOOKUP_EQ || diff == 0)
1989                 *stat = 1;
1990         else
1991                 *stat = 0;
1992         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1993         return 0;
1994
1995 error0:
1996         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1997         return error;
1998 }
1999
2000 /* Find the high key storage area from a regular key. */
2001 STATIC union xfs_btree_key *
2002 xfs_btree_high_key_from_key(
2003         struct xfs_btree_cur    *cur,
2004         union xfs_btree_key     *key)
2005 {
2006         ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2007         return (union xfs_btree_key *)((char *)key +
2008                         (cur->bc_ops->key_len / 2));
2009 }
2010
2011 /* Determine the low (and high if overlapped) keys of a leaf block */
2012 STATIC void
2013 xfs_btree_get_leaf_keys(
2014         struct xfs_btree_cur    *cur,
2015         struct xfs_btree_block  *block,
2016         union xfs_btree_key     *key)
2017 {
2018         union xfs_btree_key     max_hkey;
2019         union xfs_btree_key     hkey;
2020         union xfs_btree_rec     *rec;
2021         union xfs_btree_key     *high;
2022         int                     n;
2023
2024         rec = xfs_btree_rec_addr(cur, 1, block);
2025         cur->bc_ops->init_key_from_rec(key, rec);
2026
2027         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2028
2029                 cur->bc_ops->init_high_key_from_rec(&max_hkey, rec);
2030                 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2031                         rec = xfs_btree_rec_addr(cur, n, block);
2032                         cur->bc_ops->init_high_key_from_rec(&hkey, rec);
2033                         if (cur->bc_ops->diff_two_keys(cur, &hkey, &max_hkey)
2034                                         > 0)
2035                                 max_hkey = hkey;
2036                 }
2037
2038                 high = xfs_btree_high_key_from_key(cur, key);
2039                 memcpy(high, &max_hkey, cur->bc_ops->key_len / 2);
2040         }
2041 }
2042
2043 /* Determine the low (and high if overlapped) keys of a node block */
2044 STATIC void
2045 xfs_btree_get_node_keys(
2046         struct xfs_btree_cur    *cur,
2047         struct xfs_btree_block  *block,
2048         union xfs_btree_key     *key)
2049 {
2050         union xfs_btree_key     *hkey;
2051         union xfs_btree_key     *max_hkey;
2052         union xfs_btree_key     *high;
2053         int                     n;
2054
2055         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2056                 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2057                                 cur->bc_ops->key_len / 2);
2058
2059                 max_hkey = xfs_btree_high_key_addr(cur, 1, block);
2060                 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2061                         hkey = xfs_btree_high_key_addr(cur, n, block);
2062                         if (cur->bc_ops->diff_two_keys(cur, hkey, max_hkey) > 0)
2063                                 max_hkey = hkey;
2064                 }
2065
2066                 high = xfs_btree_high_key_from_key(cur, key);
2067                 memcpy(high, max_hkey, cur->bc_ops->key_len / 2);
2068         } else {
2069                 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2070                                 cur->bc_ops->key_len);
2071         }
2072 }
2073
2074 /* Derive the keys for any btree block. */
2075 STATIC void
2076 xfs_btree_get_keys(
2077         struct xfs_btree_cur    *cur,
2078         struct xfs_btree_block  *block,
2079         union xfs_btree_key     *key)
2080 {
2081         if (be16_to_cpu(block->bb_level) == 0)
2082                 xfs_btree_get_leaf_keys(cur, block, key);
2083         else
2084                 xfs_btree_get_node_keys(cur, block, key);
2085 }
2086
2087 /*
2088  * Decide if we need to update the parent keys of a btree block.  For
2089  * a standard btree this is only necessary if we're updating the first
2090  * record/key.  For an overlapping btree, we must always update the
2091  * keys because the highest key can be in any of the records or keys
2092  * in the block.
2093  */
2094 static inline bool
2095 xfs_btree_needs_key_update(
2096         struct xfs_btree_cur    *cur,
2097         int                     ptr)
2098 {
2099         return (cur->bc_flags & XFS_BTREE_OVERLAPPING) || ptr == 1;
2100 }
2101
2102 /*
2103  * Update the low and high parent keys of the given level, progressing
2104  * towards the root.  If force_all is false, stop if the keys for a given
2105  * level do not need updating.
2106  */
2107 STATIC int
2108 __xfs_btree_updkeys(
2109         struct xfs_btree_cur    *cur,
2110         int                     level,
2111         struct xfs_btree_block  *block,
2112         struct xfs_buf          *bp0,
2113         bool                    force_all)
2114 {
2115         union xfs_btree_key     key;    /* keys from current level */
2116         union xfs_btree_key     *lkey;  /* keys from the next level up */
2117         union xfs_btree_key     *hkey;
2118         union xfs_btree_key     *nlkey; /* keys from the next level up */
2119         union xfs_btree_key     *nhkey;
2120         struct xfs_buf          *bp;
2121         int                     ptr;
2122
2123         ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2124
2125         /* Exit if there aren't any parent levels to update. */
2126         if (level + 1 >= cur->bc_nlevels)
2127                 return 0;
2128
2129         trace_xfs_btree_updkeys(cur, level, bp0);
2130
2131         lkey = &key;
2132         hkey = xfs_btree_high_key_from_key(cur, lkey);
2133         xfs_btree_get_keys(cur, block, lkey);
2134         for (level++; level < cur->bc_nlevels; level++) {
2135 #ifdef DEBUG
2136                 int             error;
2137 #endif
2138                 block = xfs_btree_get_block(cur, level, &bp);
2139                 trace_xfs_btree_updkeys(cur, level, bp);
2140 #ifdef DEBUG
2141                 error = xfs_btree_check_block(cur, block, level, bp);
2142                 if (error) {
2143                         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2144                         return error;
2145                 }
2146 #endif
2147                 ptr = cur->bc_ptrs[level];
2148                 nlkey = xfs_btree_key_addr(cur, ptr, block);
2149                 nhkey = xfs_btree_high_key_addr(cur, ptr, block);
2150                 if (!force_all &&
2151                     !(cur->bc_ops->diff_two_keys(cur, nlkey, lkey) != 0 ||
2152                       cur->bc_ops->diff_two_keys(cur, nhkey, hkey) != 0))
2153                         break;
2154                 xfs_btree_copy_keys(cur, nlkey, lkey, 1);
2155                 xfs_btree_log_keys(cur, bp, ptr, ptr);
2156                 if (level + 1 >= cur->bc_nlevels)
2157                         break;
2158                 xfs_btree_get_node_keys(cur, block, lkey);
2159         }
2160
2161         return 0;
2162 }
2163
2164 /* Update all the keys from some level in cursor back to the root. */
2165 STATIC int
2166 xfs_btree_updkeys_force(
2167         struct xfs_btree_cur    *cur,
2168         int                     level)
2169 {
2170         struct xfs_buf          *bp;
2171         struct xfs_btree_block  *block;
2172
2173         block = xfs_btree_get_block(cur, level, &bp);
2174         return __xfs_btree_updkeys(cur, level, block, bp, true);
2175 }
2176
2177 /*
2178  * Update the parent keys of the given level, progressing towards the root.
2179  */
2180 STATIC int
2181 xfs_btree_update_keys(
2182         struct xfs_btree_cur    *cur,
2183         int                     level)
2184 {
2185         struct xfs_btree_block  *block;
2186         struct xfs_buf          *bp;
2187         union xfs_btree_key     *kp;
2188         union xfs_btree_key     key;
2189         int                     ptr;
2190
2191         ASSERT(level >= 0);
2192
2193         block = xfs_btree_get_block(cur, level, &bp);
2194         if (cur->bc_flags & XFS_BTREE_OVERLAPPING)
2195                 return __xfs_btree_updkeys(cur, level, block, bp, false);
2196
2197         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2198         XFS_BTREE_TRACE_ARGIK(cur, level, keyp);
2199
2200         /*
2201          * Go up the tree from this level toward the root.
2202          * At each level, update the key value to the value input.
2203          * Stop when we reach a level where the cursor isn't pointing
2204          * at the first entry in the block.
2205          */
2206         xfs_btree_get_keys(cur, block, &key);
2207         for (level++, ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
2208 #ifdef DEBUG
2209                 int             error;
2210 #endif
2211                 block = xfs_btree_get_block(cur, level, &bp);
2212 #ifdef DEBUG
2213                 error = xfs_btree_check_block(cur, block, level, bp);
2214                 if (error) {
2215                         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2216                         return error;
2217                 }
2218 #endif
2219                 ptr = cur->bc_ptrs[level];
2220                 kp = xfs_btree_key_addr(cur, ptr, block);
2221                 xfs_btree_copy_keys(cur, kp, &key, 1);
2222                 xfs_btree_log_keys(cur, bp, ptr, ptr);
2223         }
2224
2225         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2226         return 0;
2227 }
2228
2229 /*
2230  * Update the record referred to by cur to the value in the
2231  * given record. This either works (return 0) or gets an
2232  * EFSCORRUPTED error.
2233  */
2234 int
2235 xfs_btree_update(
2236         struct xfs_btree_cur    *cur,
2237         union xfs_btree_rec     *rec)
2238 {
2239         struct xfs_btree_block  *block;
2240         struct xfs_buf          *bp;
2241         int                     error;
2242         int                     ptr;
2243         union xfs_btree_rec     *rp;
2244
2245         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2246         XFS_BTREE_TRACE_ARGR(cur, rec);
2247
2248         /* Pick up the current block. */
2249         block = xfs_btree_get_block(cur, 0, &bp);
2250
2251 #ifdef DEBUG
2252         error = xfs_btree_check_block(cur, block, 0, bp);
2253         if (error)
2254                 goto error0;
2255 #endif
2256         /* Get the address of the rec to be updated. */
2257         ptr = cur->bc_ptrs[0];
2258         rp = xfs_btree_rec_addr(cur, ptr, block);
2259
2260         /* Fill in the new contents and log them. */
2261         xfs_btree_copy_recs(cur, rp, rec, 1);
2262         xfs_btree_log_recs(cur, bp, ptr, ptr);
2263
2264         /*
2265          * If we are tracking the last record in the tree and
2266          * we are at the far right edge of the tree, update it.
2267          */
2268         if (xfs_btree_is_lastrec(cur, block, 0)) {
2269                 cur->bc_ops->update_lastrec(cur, block, rec,
2270                                             ptr, LASTREC_UPDATE);
2271         }
2272
2273         /* Pass new key value up to our parent. */
2274         if (xfs_btree_needs_key_update(cur, ptr)) {
2275                 error = xfs_btree_update_keys(cur, 0);
2276                 if (error)
2277                         goto error0;
2278         }
2279
2280         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2281         return 0;
2282
2283 error0:
2284         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2285         return error;
2286 }
2287
2288 /*
2289  * Move 1 record left from cur/level if possible.
2290  * Update cur to reflect the new path.
2291  */
2292 STATIC int                                      /* error */
2293 xfs_btree_lshift(
2294         struct xfs_btree_cur    *cur,
2295         int                     level,
2296         int                     *stat)          /* success/failure */
2297 {
2298         struct xfs_buf          *lbp;           /* left buffer pointer */
2299         struct xfs_btree_block  *left;          /* left btree block */
2300         int                     lrecs;          /* left record count */
2301         struct xfs_buf          *rbp;           /* right buffer pointer */
2302         struct xfs_btree_block  *right;         /* right btree block */
2303         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
2304         int                     rrecs;          /* right record count */
2305         union xfs_btree_ptr     lptr;           /* left btree pointer */
2306         union xfs_btree_key     *rkp = NULL;    /* right btree key */
2307         union xfs_btree_ptr     *rpp = NULL;    /* right address pointer */
2308         union xfs_btree_rec     *rrp = NULL;    /* right record pointer */
2309         int                     error;          /* error return value */
2310         int                     i;
2311
2312         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2313         XFS_BTREE_TRACE_ARGI(cur, level);
2314
2315         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2316             level == cur->bc_nlevels - 1)
2317                 goto out0;
2318
2319         /* Set up variables for this block as "right". */
2320         right = xfs_btree_get_block(cur, level, &rbp);
2321
2322 #ifdef DEBUG
2323         error = xfs_btree_check_block(cur, right, level, rbp);
2324         if (error)
2325                 goto error0;
2326 #endif
2327
2328         /* If we've got no left sibling then we can't shift an entry left. */
2329         xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2330         if (xfs_btree_ptr_is_null(cur, &lptr))
2331                 goto out0;
2332
2333         /*
2334          * If the cursor entry is the one that would be moved, don't
2335          * do it... it's too complicated.
2336          */
2337         if (cur->bc_ptrs[level] <= 1)
2338                 goto out0;
2339
2340         /* Set up the left neighbor as "left". */
2341         error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2342         if (error)
2343                 goto error0;
2344
2345         /* If it's full, it can't take another entry. */
2346         lrecs = xfs_btree_get_numrecs(left);
2347         if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2348                 goto out0;
2349
2350         rrecs = xfs_btree_get_numrecs(right);
2351
2352         /*
2353          * We add one entry to the left side and remove one for the right side.
2354          * Account for it here, the changes will be updated on disk and logged
2355          * later.
2356          */
2357         lrecs++;
2358         rrecs--;
2359
2360         XFS_BTREE_STATS_INC(cur, lshift);
2361         XFS_BTREE_STATS_ADD(cur, moves, 1);
2362
2363         /*
2364          * If non-leaf, copy a key and a ptr to the left block.
2365          * Log the changes to the left block.
2366          */
2367         if (level > 0) {
2368                 /* It's a non-leaf.  Move keys and pointers. */
2369                 union xfs_btree_key     *lkp;   /* left btree key */
2370                 union xfs_btree_ptr     *lpp;   /* left address pointer */
2371
2372                 lkp = xfs_btree_key_addr(cur, lrecs, left);
2373                 rkp = xfs_btree_key_addr(cur, 1, right);
2374
2375                 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2376                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2377 #ifdef DEBUG
2378                 error = xfs_btree_check_ptr(cur, rpp, 0, level);
2379                 if (error)
2380                         goto error0;
2381 #endif
2382                 xfs_btree_copy_keys(cur, lkp, rkp, 1);
2383                 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2384
2385                 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2386                 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2387
2388                 ASSERT(cur->bc_ops->keys_inorder(cur,
2389                         xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2390         } else {
2391                 /* It's a leaf.  Move records.  */
2392                 union xfs_btree_rec     *lrp;   /* left record pointer */
2393
2394                 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2395                 rrp = xfs_btree_rec_addr(cur, 1, right);
2396
2397                 xfs_btree_copy_recs(cur, lrp, rrp, 1);
2398                 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2399
2400                 ASSERT(cur->bc_ops->recs_inorder(cur,
2401                         xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2402         }
2403
2404         xfs_btree_set_numrecs(left, lrecs);
2405         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2406
2407         xfs_btree_set_numrecs(right, rrecs);
2408         xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2409
2410         /*
2411          * Slide the contents of right down one entry.
2412          */
2413         XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2414         if (level > 0) {
2415                 /* It's a nonleaf. operate on keys and ptrs */
2416 #ifdef DEBUG
2417                 int                     i;              /* loop index */
2418
2419                 for (i = 0; i < rrecs; i++) {
2420                         error = xfs_btree_check_ptr(cur, rpp, i + 1, level);
2421                         if (error)
2422                                 goto error0;
2423                 }
2424 #endif
2425                 xfs_btree_shift_keys(cur,
2426                                 xfs_btree_key_addr(cur, 2, right),
2427                                 -1, rrecs);
2428                 xfs_btree_shift_ptrs(cur,
2429                                 xfs_btree_ptr_addr(cur, 2, right),
2430                                 -1, rrecs);
2431
2432                 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2433                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2434         } else {
2435                 /* It's a leaf. operate on records */
2436                 xfs_btree_shift_recs(cur,
2437                         xfs_btree_rec_addr(cur, 2, right),
2438                         -1, rrecs);
2439                 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2440         }
2441
2442         /*
2443          * Using a temporary cursor, update the parent key values of the
2444          * block on the left.
2445          */
2446         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2447                 error = xfs_btree_dup_cursor(cur, &tcur);
2448                 if (error)
2449                         goto error0;
2450                 i = xfs_btree_firstrec(tcur, level);
2451                 XFS_WANT_CORRUPTED_GOTO(tcur->bc_mp, i == 1, error0);
2452
2453                 error = xfs_btree_decrement(tcur, level, &i);
2454                 if (error)
2455                         goto error1;
2456
2457                 /* Update the parent high keys of the left block, if needed. */
2458                 error = xfs_btree_update_keys(tcur, level);
2459                 if (error)
2460                         goto error1;
2461
2462                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2463         }
2464
2465         /* Update the parent keys of the right block. */
2466         error = xfs_btree_update_keys(cur, level);
2467         if (error)
2468                 goto error0;
2469
2470         /* Slide the cursor value left one. */
2471         cur->bc_ptrs[level]--;
2472
2473         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2474         *stat = 1;
2475         return 0;
2476
2477 out0:
2478         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2479         *stat = 0;
2480         return 0;
2481
2482 error0:
2483         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2484         return error;
2485
2486 error1:
2487         XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2488         xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2489         return error;
2490 }
2491
2492 /*
2493  * Move 1 record right from cur/level if possible.
2494  * Update cur to reflect the new path.
2495  */
2496 STATIC int                                      /* error */
2497 xfs_btree_rshift(
2498         struct xfs_btree_cur    *cur,
2499         int                     level,
2500         int                     *stat)          /* success/failure */
2501 {
2502         struct xfs_buf          *lbp;           /* left buffer pointer */
2503         struct xfs_btree_block  *left;          /* left btree block */
2504         struct xfs_buf          *rbp;           /* right buffer pointer */
2505         struct xfs_btree_block  *right;         /* right btree block */
2506         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
2507         union xfs_btree_ptr     rptr;           /* right block pointer */
2508         union xfs_btree_key     *rkp;           /* right btree key */
2509         int                     rrecs;          /* right record count */
2510         int                     lrecs;          /* left record count */
2511         int                     error;          /* error return value */
2512         int                     i;              /* loop counter */
2513
2514         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2515         XFS_BTREE_TRACE_ARGI(cur, level);
2516
2517         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2518             (level == cur->bc_nlevels - 1))
2519                 goto out0;
2520
2521         /* Set up variables for this block as "left". */
2522         left = xfs_btree_get_block(cur, level, &lbp);
2523
2524 #ifdef DEBUG
2525         error = xfs_btree_check_block(cur, left, level, lbp);
2526         if (error)
2527                 goto error0;
2528 #endif
2529
2530         /* If we've got no right sibling then we can't shift an entry right. */
2531         xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2532         if (xfs_btree_ptr_is_null(cur, &rptr))
2533                 goto out0;
2534
2535         /*
2536          * If the cursor entry is the one that would be moved, don't
2537          * do it... it's too complicated.
2538          */
2539         lrecs = xfs_btree_get_numrecs(left);
2540         if (cur->bc_ptrs[level] >= lrecs)
2541                 goto out0;
2542
2543         /* Set up the right neighbor as "right". */
2544         error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2545         if (error)
2546                 goto error0;
2547
2548         /* If it's full, it can't take another entry. */
2549         rrecs = xfs_btree_get_numrecs(right);
2550         if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2551                 goto out0;
2552
2553         XFS_BTREE_STATS_INC(cur, rshift);
2554         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2555
2556         /*
2557          * Make a hole at the start of the right neighbor block, then
2558          * copy the last left block entry to the hole.
2559          */
2560         if (level > 0) {
2561                 /* It's a nonleaf. make a hole in the keys and ptrs */
2562                 union xfs_btree_key     *lkp;
2563                 union xfs_btree_ptr     *lpp;
2564                 union xfs_btree_ptr     *rpp;
2565
2566                 lkp = xfs_btree_key_addr(cur, lrecs, left);
2567                 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2568                 rkp = xfs_btree_key_addr(cur, 1, right);
2569                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2570
2571 #ifdef DEBUG
2572                 for (i = rrecs - 1; i >= 0; i--) {
2573                         error = xfs_btree_check_ptr(cur, rpp, i, level);
2574                         if (error)
2575                                 goto error0;
2576                 }
2577 #endif
2578
2579                 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2580                 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2581
2582 #ifdef DEBUG
2583                 error = xfs_btree_check_ptr(cur, lpp, 0, level);
2584                 if (error)
2585                         goto error0;
2586 #endif
2587
2588                 /* Now put the new data in, and log it. */
2589                 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2590                 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2591
2592                 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2593                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2594
2595                 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2596                         xfs_btree_key_addr(cur, 2, right)));
2597         } else {
2598                 /* It's a leaf. make a hole in the records */
2599                 union xfs_btree_rec     *lrp;
2600                 union xfs_btree_rec     *rrp;
2601
2602                 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2603                 rrp = xfs_btree_rec_addr(cur, 1, right);
2604
2605                 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2606
2607                 /* Now put the new data in, and log it. */
2608                 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2609                 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2610         }
2611
2612         /*
2613          * Decrement and log left's numrecs, bump and log right's numrecs.
2614          */
2615         xfs_btree_set_numrecs(left, --lrecs);
2616         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2617
2618         xfs_btree_set_numrecs(right, ++rrecs);
2619         xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2620
2621         /*
2622          * Using a temporary cursor, update the parent key values of the
2623          * block on the right.
2624          */
2625         error = xfs_btree_dup_cursor(cur, &tcur);
2626         if (error)
2627                 goto error0;
2628         i = xfs_btree_lastrec(tcur, level);
2629         XFS_WANT_CORRUPTED_GOTO(tcur->bc_mp, i == 1, error0);
2630
2631         error = xfs_btree_increment(tcur, level, &i);
2632         if (error)
2633                 goto error1;
2634
2635         /* Update the parent high keys of the left block, if needed. */
2636         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2637                 error = xfs_btree_update_keys(cur, level);
2638                 if (error)
2639                         goto error1;
2640         }
2641
2642         /* Update the parent keys of the right block. */
2643         error = xfs_btree_update_keys(tcur, level);
2644         if (error)
2645                 goto error1;
2646
2647         xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2648
2649         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2650         *stat = 1;
2651         return 0;
2652
2653 out0:
2654         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2655         *stat = 0;
2656         return 0;
2657
2658 error0:
2659         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2660         return error;
2661
2662 error1:
2663         XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2664         xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2665         return error;
2666 }
2667
2668 /*
2669  * Split cur/level block in half.
2670  * Return new block number and the key to its first
2671  * record (to be inserted into parent).
2672  */
2673 STATIC int                                      /* error */
2674 __xfs_btree_split(
2675         struct xfs_btree_cur    *cur,
2676         int                     level,
2677         union xfs_btree_ptr     *ptrp,
2678         union xfs_btree_key     *key,
2679         struct xfs_btree_cur    **curp,
2680         int                     *stat)          /* success/failure */
2681 {
2682         union xfs_btree_ptr     lptr;           /* left sibling block ptr */
2683         struct xfs_buf          *lbp;           /* left buffer pointer */
2684         struct xfs_btree_block  *left;          /* left btree block */
2685         union xfs_btree_ptr     rptr;           /* right sibling block ptr */
2686         struct xfs_buf          *rbp;           /* right buffer pointer */
2687         struct xfs_btree_block  *right;         /* right btree block */
2688         union xfs_btree_ptr     rrptr;          /* right-right sibling ptr */
2689         struct xfs_buf          *rrbp;          /* right-right buffer pointer */
2690         struct xfs_btree_block  *rrblock;       /* right-right btree block */
2691         int                     lrecs;
2692         int                     rrecs;
2693         int                     src_index;
2694         int                     error;          /* error return value */
2695 #ifdef DEBUG
2696         int                     i;
2697 #endif
2698
2699         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2700         XFS_BTREE_TRACE_ARGIPK(cur, level, *ptrp, key);
2701
2702         XFS_BTREE_STATS_INC(cur, split);
2703
2704         /* Set up left block (current one). */
2705         left = xfs_btree_get_block(cur, level, &lbp);
2706
2707 #ifdef DEBUG
2708         error = xfs_btree_check_block(cur, left, level, lbp);
2709         if (error)
2710                 goto error0;
2711 #endif
2712
2713         xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2714
2715         /* Allocate the new block. If we can't do it, we're toast. Give up. */
2716         error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, stat);
2717         if (error)
2718                 goto error0;
2719         if (*stat == 0)
2720                 goto out0;
2721         XFS_BTREE_STATS_INC(cur, alloc);
2722
2723         /* Set up the new block as "right". */
2724         error = xfs_btree_get_buf_block(cur, &rptr, 0, &right, &rbp);
2725         if (error)
2726                 goto error0;
2727
2728         /* Fill in the btree header for the new right block. */
2729         xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2730
2731         /*
2732          * Split the entries between the old and the new block evenly.
2733          * Make sure that if there's an odd number of entries now, that
2734          * each new block will have the same number of entries.
2735          */
2736         lrecs = xfs_btree_get_numrecs(left);
2737         rrecs = lrecs / 2;
2738         if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
2739                 rrecs++;
2740         src_index = (lrecs - rrecs + 1);
2741
2742         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2743
2744         /* Adjust numrecs for the later get_*_keys() calls. */
2745         lrecs -= rrecs;
2746         xfs_btree_set_numrecs(left, lrecs);
2747         xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2748
2749         /*
2750          * Copy btree block entries from the left block over to the
2751          * new block, the right. Update the right block and log the
2752          * changes.
2753          */
2754         if (level > 0) {
2755                 /* It's a non-leaf.  Move keys and pointers. */
2756                 union xfs_btree_key     *lkp;   /* left btree key */
2757                 union xfs_btree_ptr     *lpp;   /* left address pointer */
2758                 union xfs_btree_key     *rkp;   /* right btree key */
2759                 union xfs_btree_ptr     *rpp;   /* right address pointer */
2760
2761                 lkp = xfs_btree_key_addr(cur, src_index, left);
2762                 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2763                 rkp = xfs_btree_key_addr(cur, 1, right);
2764                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2765
2766 #ifdef DEBUG
2767                 for (i = src_index; i < rrecs; i++) {
2768                         error = xfs_btree_check_ptr(cur, lpp, i, level);
2769                         if (error)
2770                                 goto error0;
2771                 }
2772 #endif
2773
2774                 /* Copy the keys & pointers to the new block. */
2775                 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2776                 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2777
2778                 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2779                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2780
2781                 /* Stash the keys of the new block for later insertion. */
2782                 xfs_btree_get_node_keys(cur, right, key);
2783         } else {
2784                 /* It's a leaf.  Move records.  */
2785                 union xfs_btree_rec     *lrp;   /* left record pointer */
2786                 union xfs_btree_rec     *rrp;   /* right record pointer */
2787
2788                 lrp = xfs_btree_rec_addr(cur, src_index, left);
2789                 rrp = xfs_btree_rec_addr(cur, 1, right);
2790
2791                 /* Copy records to the new block. */
2792                 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2793                 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2794
2795                 /* Stash the keys of the new block for later insertion. */
2796                 xfs_btree_get_leaf_keys(cur, right, key);
2797         }
2798
2799         /*
2800          * Find the left block number by looking in the buffer.
2801          * Adjust sibling pointers.
2802          */
2803         xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2804         xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2805         xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2806         xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2807
2808         xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2809         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2810
2811         /*
2812          * If there's a block to the new block's right, make that block
2813          * point back to right instead of to left.
2814          */
2815         if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2816                 error = xfs_btree_read_buf_block(cur, &rrptr,
2817                                                         0, &rrblock, &rrbp);
2818                 if (error)
2819                         goto error0;
2820                 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2821                 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2822         }
2823
2824         /* Update the parent high keys of the left block, if needed. */
2825         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2826                 error = xfs_btree_update_keys(cur, level);
2827                 if (error)
2828                         goto error0;
2829         }
2830
2831         /*
2832          * If the cursor is really in the right block, move it there.
2833          * If it's just pointing past the last entry in left, then we'll
2834          * insert there, so don't change anything in that case.
2835          */
2836         if (cur->bc_ptrs[level] > lrecs + 1) {
2837                 xfs_btree_setbuf(cur, level, rbp);
2838                 cur->bc_ptrs[level] -= lrecs;
2839         }
2840         /*
2841          * If there are more levels, we'll need another cursor which refers
2842          * the right block, no matter where this cursor was.
2843          */
2844         if (level + 1 < cur->bc_nlevels) {
2845                 error = xfs_btree_dup_cursor(cur, curp);
2846                 if (error)
2847                         goto error0;
2848                 (*curp)->bc_ptrs[level + 1]++;
2849         }
2850         *ptrp = rptr;
2851         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2852         *stat = 1;
2853         return 0;
2854 out0:
2855         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2856         *stat = 0;
2857         return 0;
2858
2859 error0:
2860         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2861         return error;
2862 }
2863
2864 struct xfs_btree_split_args {
2865         struct xfs_btree_cur    *cur;
2866         int                     level;
2867         union xfs_btree_ptr     *ptrp;
2868         union xfs_btree_key     *key;
2869         struct xfs_btree_cur    **curp;
2870         int                     *stat;          /* success/failure */
2871         int                     result;
2872         bool                    kswapd; /* allocation in kswapd context */
2873         struct completion       *done;
2874         struct work_struct      work;
2875 };
2876
2877 /*
2878  * Stack switching interfaces for allocation
2879  */
2880 static void
2881 xfs_btree_split_worker(
2882         struct work_struct      *work)
2883 {
2884         struct xfs_btree_split_args     *args = container_of(work,
2885                                                 struct xfs_btree_split_args, work);
2886         unsigned long           pflags;
2887         unsigned long           new_pflags = PF_MEMALLOC_NOFS;
2888
2889         /*
2890          * we are in a transaction context here, but may also be doing work
2891          * in kswapd context, and hence we may need to inherit that state
2892          * temporarily to ensure that we don't block waiting for memory reclaim
2893          * in any way.
2894          */
2895         if (args->kswapd)
2896                 new_pflags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
2897
2898         current_set_flags_nested(&pflags, new_pflags);
2899
2900         args->result = __xfs_btree_split(args->cur, args->level, args->ptrp,
2901                                          args->key, args->curp, args->stat);
2902         complete(args->done);
2903
2904         current_restore_flags_nested(&pflags, new_pflags);
2905 }
2906
2907 /*
2908  * BMBT split requests often come in with little stack to work on. Push
2909  * them off to a worker thread so there is lots of stack to use. For the other
2910  * btree types, just call directly to avoid the context switch overhead here.
2911  */
2912 STATIC int                                      /* error */
2913 xfs_btree_split(
2914         struct xfs_btree_cur    *cur,
2915         int                     level,
2916         union xfs_btree_ptr     *ptrp,
2917         union xfs_btree_key     *key,
2918         struct xfs_btree_cur    **curp,
2919         int                     *stat)          /* success/failure */
2920 {
2921         struct xfs_btree_split_args     args;
2922         DECLARE_COMPLETION_ONSTACK(done);
2923
2924         if (cur->bc_btnum != XFS_BTNUM_BMAP)
2925                 return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
2926
2927         args.cur = cur;
2928         args.level = level;
2929         args.ptrp = ptrp;
2930         args.key = key;
2931         args.curp = curp;
2932         args.stat = stat;
2933         args.done = &done;
2934         args.kswapd = current_is_kswapd();
2935         INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker);
2936         queue_work(xfs_alloc_wq, &args.work);
2937         wait_for_completion(&done);
2938         destroy_work_on_stack(&args.work);
2939         return args.result;
2940 }
2941
2942
2943 /*
2944  * Copy the old inode root contents into a real block and make the
2945  * broot point to it.
2946  */
2947 int                                             /* error */
2948 xfs_btree_new_iroot(
2949         struct xfs_btree_cur    *cur,           /* btree cursor */
2950         int                     *logflags,      /* logging flags for inode */
2951         int                     *stat)          /* return status - 0 fail */
2952 {
2953         struct xfs_buf          *cbp;           /* buffer for cblock */
2954         struct xfs_btree_block  *block;         /* btree block */
2955         struct xfs_btree_block  *cblock;        /* child btree block */
2956         union xfs_btree_key     *ckp;           /* child key pointer */
2957         union xfs_btree_ptr     *cpp;           /* child ptr pointer */
2958         union xfs_btree_key     *kp;            /* pointer to btree key */
2959         union xfs_btree_ptr     *pp;            /* pointer to block addr */
2960         union xfs_btree_ptr     nptr;           /* new block addr */
2961         int                     level;          /* btree level */
2962         int                     error;          /* error return code */
2963 #ifdef DEBUG
2964         int                     i;              /* loop counter */
2965 #endif
2966
2967         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2968         XFS_BTREE_STATS_INC(cur, newroot);
2969
2970         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2971
2972         level = cur->bc_nlevels - 1;
2973
2974         block = xfs_btree_get_iroot(cur);
2975         pp = xfs_btree_ptr_addr(cur, 1, block);
2976
2977         /* Allocate the new block. If we can't do it, we're toast. Give up. */
2978         error = cur->bc_ops->alloc_block(cur, pp, &nptr, stat);
2979         if (error)
2980                 goto error0;
2981         if (*stat == 0) {
2982                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2983                 return 0;
2984         }
2985         XFS_BTREE_STATS_INC(cur, alloc);
2986
2987         /* Copy the root into a real block. */
2988         error = xfs_btree_get_buf_block(cur, &nptr, 0, &cblock, &cbp);
2989         if (error)
2990                 goto error0;
2991
2992         /*
2993          * we can't just memcpy() the root in for CRC enabled btree blocks.
2994          * In that case have to also ensure the blkno remains correct
2995          */
2996         memcpy(cblock, block, xfs_btree_block_len(cur));
2997         if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
2998                 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
2999                         cblock->bb_u.l.bb_blkno = cpu_to_be64(cbp->b_bn);
3000                 else
3001                         cblock->bb_u.s.bb_blkno = cpu_to_be64(cbp->b_bn);
3002         }
3003
3004         be16_add_cpu(&block->bb_level, 1);
3005         xfs_btree_set_numrecs(block, 1);
3006         cur->bc_nlevels++;
3007         cur->bc_ptrs[level + 1] = 1;
3008
3009         kp = xfs_btree_key_addr(cur, 1, block);
3010         ckp = xfs_btree_key_addr(cur, 1, cblock);
3011         xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
3012
3013         cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3014 #ifdef DEBUG
3015         for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
3016                 error = xfs_btree_check_ptr(cur, pp, i, level);
3017                 if (error)
3018                         goto error0;
3019         }
3020 #endif
3021         xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
3022
3023 #ifdef DEBUG
3024         error = xfs_btree_check_ptr(cur, &nptr, 0, level);
3025         if (error)
3026                 goto error0;
3027 #endif
3028         xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
3029
3030         xfs_iroot_realloc(cur->bc_private.b.ip,
3031                           1 - xfs_btree_get_numrecs(cblock),
3032                           cur->bc_private.b.whichfork);
3033
3034         xfs_btree_setbuf(cur, level, cbp);
3035
3036         /*
3037          * Do all this logging at the end so that
3038          * the root is at the right level.
3039          */
3040         xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
3041         xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3042         xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3043
3044         *logflags |=
3045                 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork);
3046         *stat = 1;
3047         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3048         return 0;
3049 error0:
3050         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3051         return error;
3052 }
3053
3054 /*
3055  * Allocate a new root block, fill it in.
3056  */
3057 STATIC int                              /* error */
3058 xfs_btree_new_root(
3059         struct xfs_btree_cur    *cur,   /* btree cursor */
3060         int                     *stat)  /* success/failure */
3061 {
3062         struct xfs_btree_block  *block; /* one half of the old root block */
3063         struct xfs_buf          *bp;    /* buffer containing block */
3064         int                     error;  /* error return value */
3065         struct xfs_buf          *lbp;   /* left buffer pointer */
3066         struct xfs_btree_block  *left;  /* left btree block */
3067         struct xfs_buf          *nbp;   /* new (root) buffer */
3068         struct xfs_btree_block  *new;   /* new (root) btree block */
3069         int                     nptr;   /* new value for key index, 1 or 2 */
3070         struct xfs_buf          *rbp;   /* right buffer pointer */
3071         struct xfs_btree_block  *right; /* right btree block */
3072         union xfs_btree_ptr     rptr;
3073         union xfs_btree_ptr     lptr;
3074
3075         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3076         XFS_BTREE_STATS_INC(cur, newroot);
3077
3078         /* initialise our start point from the cursor */
3079         cur->bc_ops->init_ptr_from_cur(cur, &rptr);
3080
3081         /* Allocate the new block. If we can't do it, we're toast. Give up. */
3082         error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, stat);
3083         if (error)
3084                 goto error0;
3085         if (*stat == 0)
3086                 goto out0;
3087         XFS_BTREE_STATS_INC(cur, alloc);
3088
3089         /* Set up the new block. */
3090         error = xfs_btree_get_buf_block(cur, &lptr, 0, &new, &nbp);
3091         if (error)
3092                 goto error0;
3093
3094         /* Set the root in the holding structure  increasing the level by 1. */
3095         cur->bc_ops->set_root(cur, &lptr, 1);
3096
3097         /*
3098          * At the previous root level there are now two blocks: the old root,
3099          * and the new block generated when it was split.  We don't know which
3100          * one the cursor is pointing at, so we set up variables "left" and
3101          * "right" for each case.
3102          */
3103         block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
3104
3105 #ifdef DEBUG
3106         error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
3107         if (error)
3108                 goto error0;
3109 #endif
3110
3111         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3112         if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3113                 /* Our block is left, pick up the right block. */
3114                 lbp = bp;
3115                 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
3116                 left = block;
3117                 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
3118                 if (error)
3119                         goto error0;
3120                 bp = rbp;
3121                 nptr = 1;
3122         } else {
3123                 /* Our block is right, pick up the left block. */
3124                 rbp = bp;
3125                 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
3126                 right = block;
3127                 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
3128                 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3129                 if (error)
3130                         goto error0;
3131                 bp = lbp;
3132                 nptr = 2;
3133         }
3134
3135         /* Fill in the new block's btree header and log it. */
3136         xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
3137         xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
3138         ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
3139                         !xfs_btree_ptr_is_null(cur, &rptr));
3140
3141         /* Fill in the key data in the new root. */
3142         if (xfs_btree_get_level(left) > 0) {
3143                 /*
3144                  * Get the keys for the left block's keys and put them directly
3145                  * in the parent block.  Do the same for the right block.
3146                  */
3147                 xfs_btree_get_node_keys(cur, left,
3148                                 xfs_btree_key_addr(cur, 1, new));
3149                 xfs_btree_get_node_keys(cur, right,
3150                                 xfs_btree_key_addr(cur, 2, new));
3151         } else {
3152                 /*
3153                  * Get the keys for the left block's records and put them
3154                  * directly in the parent block.  Do the same for the right
3155                  * block.
3156                  */
3157                 xfs_btree_get_leaf_keys(cur, left,
3158                         xfs_btree_key_addr(cur, 1, new));
3159                 xfs_btree_get_leaf_keys(cur, right,
3160                         xfs_btree_key_addr(cur, 2, new));
3161         }
3162         xfs_btree_log_keys(cur, nbp, 1, 2);
3163
3164         /* Fill in the pointer data in the new root. */
3165         xfs_btree_copy_ptrs(cur,
3166                 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
3167         xfs_btree_copy_ptrs(cur,
3168                 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
3169         xfs_btree_log_ptrs(cur, nbp, 1, 2);
3170
3171         /* Fix up the cursor. */
3172         xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
3173         cur->bc_ptrs[cur->bc_nlevels] = nptr;
3174         cur->bc_nlevels++;
3175         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3176         *stat = 1;
3177         return 0;
3178 error0:
3179         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3180         return error;
3181 out0:
3182         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3183         *stat = 0;
3184         return 0;
3185 }
3186
3187 STATIC int
3188 xfs_btree_make_block_unfull(
3189         struct xfs_btree_cur    *cur,   /* btree cursor */
3190         int                     level,  /* btree level */
3191         int                     numrecs,/* # of recs in block */
3192         int                     *oindex,/* old tree index */
3193         int                     *index, /* new tree index */
3194         union xfs_btree_ptr     *nptr,  /* new btree ptr */
3195         struct xfs_btree_cur    **ncur, /* new btree cursor */
3196         union xfs_btree_key     *key,   /* key of new block */
3197         int                     *stat)
3198 {
3199         int                     error = 0;
3200
3201         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3202             level == cur->bc_nlevels - 1) {
3203                 struct xfs_inode *ip = cur->bc_private.b.ip;
3204
3205                 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
3206                         /* A root block that can be made bigger. */
3207                         xfs_iroot_realloc(ip, 1, cur->bc_private.b.whichfork);
3208                         *stat = 1;
3209                 } else {
3210                         /* A root block that needs replacing */
3211                         int     logflags = 0;
3212
3213                         error = xfs_btree_new_iroot(cur, &logflags, stat);
3214                         if (error || *stat == 0)
3215                                 return error;
3216
3217                         xfs_trans_log_inode(cur->bc_tp, ip, logflags);
3218                 }
3219
3220                 return 0;
3221         }
3222
3223         /* First, try shifting an entry to the right neighbor. */
3224         error = xfs_btree_rshift(cur, level, stat);
3225         if (error || *stat)
3226                 return error;
3227
3228         /* Next, try shifting an entry to the left neighbor. */
3229         error = xfs_btree_lshift(cur, level, stat);
3230         if (error)
3231                 return error;
3232
3233         if (*stat) {
3234                 *oindex = *index = cur->bc_ptrs[level];
3235                 return 0;
3236         }
3237
3238         /*
3239          * Next, try splitting the current block in half.
3240          *
3241          * If this works we have to re-set our variables because we
3242          * could be in a different block now.
3243          */
3244         error = xfs_btree_split(cur, level, nptr, key, ncur, stat);
3245         if (error || *stat == 0)
3246                 return error;
3247
3248
3249         *index = cur->bc_ptrs[level];
3250         return 0;
3251 }
3252
3253 /*
3254  * Insert one record/level.  Return information to the caller
3255  * allowing the next level up to proceed if necessary.
3256  */
3257 STATIC int
3258 xfs_btree_insrec(
3259         struct xfs_btree_cur    *cur,   /* btree cursor */
3260         int                     level,  /* level to insert record at */
3261         union xfs_btree_ptr     *ptrp,  /* i/o: block number inserted */
3262         union xfs_btree_rec     *rec,   /* record to insert */
3263         union xfs_btree_key     *key,   /* i/o: block key for ptrp */
3264         struct xfs_btree_cur    **curp, /* output: new cursor replacing cur */
3265         int                     *stat)  /* success/failure */
3266 {
3267         struct xfs_btree_block  *block; /* btree block */
3268         struct xfs_buf          *bp;    /* buffer for block */
3269         union xfs_btree_ptr     nptr;   /* new block ptr */
3270         struct xfs_btree_cur    *ncur;  /* new btree cursor */
3271         union xfs_btree_key     nkey;   /* new block key */
3272         union xfs_btree_key     *lkey;
3273         int                     optr;   /* old key/record index */
3274         int                     ptr;    /* key/record index */
3275         int                     numrecs;/* number of records */
3276         int                     error;  /* error return value */
3277 #ifdef DEBUG
3278         int                     i;
3279 #endif
3280         xfs_daddr_t             old_bn;
3281
3282         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3283         XFS_BTREE_TRACE_ARGIPR(cur, level, *ptrp, &rec);
3284
3285         ncur = NULL;
3286         lkey = &nkey;
3287
3288         /*
3289          * If we have an external root pointer, and we've made it to the
3290          * root level, allocate a new root block and we're done.
3291          */
3292         if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3293             (level >= cur->bc_nlevels)) {
3294                 error = xfs_btree_new_root(cur, stat);
3295                 xfs_btree_set_ptr_null(cur, ptrp);
3296
3297                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3298                 return error;
3299         }
3300
3301         /* If we're off the left edge, return failure. */
3302         ptr = cur->bc_ptrs[level];
3303         if (ptr == 0) {
3304                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3305                 *stat = 0;
3306                 return 0;
3307         }
3308
3309         optr = ptr;
3310
3311         XFS_BTREE_STATS_INC(cur, insrec);
3312
3313         /* Get pointers to the btree buffer and block. */
3314         block = xfs_btree_get_block(cur, level, &bp);
3315         old_bn = bp ? bp->b_bn : XFS_BUF_DADDR_NULL;
3316         numrecs = xfs_btree_get_numrecs(block);
3317
3318 #ifdef DEBUG
3319         error = xfs_btree_check_block(cur, block, level, bp);
3320         if (error)
3321                 goto error0;
3322
3323         /* Check that the new entry is being inserted in the right place. */
3324         if (ptr <= numrecs) {
3325                 if (level == 0) {
3326                         ASSERT(cur->bc_ops->recs_inorder(cur, rec,
3327                                 xfs_btree_rec_addr(cur, ptr, block)));
3328                 } else {
3329                         ASSERT(cur->bc_ops->keys_inorder(cur, key,
3330                                 xfs_btree_key_addr(cur, ptr, block)));
3331                 }
3332         }
3333 #endif
3334
3335         /*
3336          * If the block is full, we can't insert the new entry until we
3337          * make the block un-full.
3338          */
3339         xfs_btree_set_ptr_null(cur, &nptr);
3340         if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
3341                 error = xfs_btree_make_block_unfull(cur, level, numrecs,
3342                                         &optr, &ptr, &nptr, &ncur, lkey, stat);
3343                 if (error || *stat == 0)
3344                         goto error0;
3345         }
3346
3347         /*
3348          * The current block may have changed if the block was
3349          * previously full and we have just made space in it.
3350          */
3351         block = xfs_btree_get_block(cur, level, &bp);
3352         numrecs = xfs_btree_get_numrecs(block);
3353
3354 #ifdef DEBUG
3355         error = xfs_btree_check_block(cur, block, level, bp);
3356         if (error)
3357                 return error;
3358 #endif
3359
3360         /*
3361          * At this point we know there's room for our new entry in the block
3362          * we're pointing at.
3363          */
3364         XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
3365
3366         if (level > 0) {
3367                 /* It's a nonleaf. make a hole in the keys and ptrs */
3368                 union xfs_btree_key     *kp;
3369                 union xfs_btree_ptr     *pp;
3370
3371                 kp = xfs_btree_key_addr(cur, ptr, block);
3372                 pp = xfs_btree_ptr_addr(cur, ptr, block);
3373
3374 #ifdef DEBUG
3375                 for (i = numrecs - ptr; i >= 0; i--) {
3376                         error = xfs_btree_check_ptr(cur, pp, i, level);
3377                         if (error)
3378                                 return error;
3379                 }
3380 #endif
3381
3382                 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
3383                 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
3384
3385 #ifdef DEBUG
3386                 error = xfs_btree_check_ptr(cur, ptrp, 0, level);
3387                 if (error)
3388                         goto error0;
3389 #endif
3390
3391                 /* Now put the new data in, bump numrecs and log it. */
3392                 xfs_btree_copy_keys(cur, kp, key, 1);
3393                 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
3394                 numrecs++;
3395                 xfs_btree_set_numrecs(block, numrecs);
3396                 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
3397                 xfs_btree_log_keys(cur, bp, ptr, numrecs);
3398 #ifdef DEBUG
3399                 if (ptr < numrecs) {
3400                         ASSERT(cur->bc_ops->keys_inorder(cur, kp,
3401                                 xfs_btree_key_addr(cur, ptr + 1, block)));
3402                 }
3403 #endif
3404         } else {
3405                 /* It's a leaf. make a hole in the records */
3406                 union xfs_btree_rec             *rp;
3407
3408                 rp = xfs_btree_rec_addr(cur, ptr, block);
3409
3410                 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
3411
3412                 /* Now put the new data in, bump numrecs and log it. */
3413                 xfs_btree_copy_recs(cur, rp, rec, 1);
3414                 xfs_btree_set_numrecs(block, ++numrecs);
3415                 xfs_btree_log_recs(cur, bp, ptr, numrecs);
3416 #ifdef DEBUG
3417                 if (ptr < numrecs) {
3418                         ASSERT(cur->bc_ops->recs_inorder(cur, rp,
3419                                 xfs_btree_rec_addr(cur, ptr + 1, block)));
3420                 }
3421 #endif
3422         }
3423
3424         /* Log the new number of records in the btree header. */
3425         xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3426
3427         /*
3428          * If we just inserted into a new tree block, we have to
3429          * recalculate nkey here because nkey is out of date.
3430          *
3431          * Otherwise we're just updating an existing block (having shoved
3432          * some records into the new tree block), so use the regular key
3433          * update mechanism.
3434          */
3435         if (bp && bp->b_bn != old_bn) {
3436                 xfs_btree_get_keys(cur, block, lkey);
3437         } else if (xfs_btree_needs_key_update(cur, optr)) {
3438                 error = xfs_btree_update_keys(cur, level);
3439                 if (error)
3440                         goto error0;
3441         }
3442
3443         /*
3444          * If we are tracking the last record in the tree and
3445          * we are at the far right edge of the tree, update it.
3446          */
3447         if (xfs_btree_is_lastrec(cur, block, level)) {
3448                 cur->bc_ops->update_lastrec(cur, block, rec,
3449                                             ptr, LASTREC_INSREC);
3450         }
3451
3452         /*
3453          * Return the new block number, if any.
3454          * If there is one, give back a record value and a cursor too.
3455          */
3456         *ptrp = nptr;
3457         if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3458                 xfs_btree_copy_keys(cur, key, lkey, 1);
3459                 *curp = ncur;
3460         }
3461
3462         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3463         *stat = 1;
3464         return 0;
3465
3466 error0:
3467         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3468         return error;
3469 }
3470
3471 /*
3472  * Insert the record at the point referenced by cur.
3473  *
3474  * A multi-level split of the tree on insert will invalidate the original
3475  * cursor.  All callers of this function should assume that the cursor is
3476  * no longer valid and revalidate it.
3477  */
3478 int
3479 xfs_btree_insert(
3480         struct xfs_btree_cur    *cur,
3481         int                     *stat)
3482 {
3483         int                     error;  /* error return value */
3484         int                     i;      /* result value, 0 for failure */
3485         int                     level;  /* current level number in btree */
3486         union xfs_btree_ptr     nptr;   /* new block number (split result) */
3487         struct xfs_btree_cur    *ncur;  /* new cursor (split result) */
3488         struct xfs_btree_cur    *pcur;  /* previous level's cursor */
3489         union xfs_btree_key     bkey;   /* key of block to insert */
3490         union xfs_btree_key     *key;
3491         union xfs_btree_rec     rec;    /* record to insert */
3492
3493         level = 0;
3494         ncur = NULL;
3495         pcur = cur;
3496         key = &bkey;
3497
3498         xfs_btree_set_ptr_null(cur, &nptr);
3499
3500         /* Make a key out of the record data to be inserted, and save it. */
3501         cur->bc_ops->init_rec_from_cur(cur, &rec);
3502         cur->bc_ops->init_key_from_rec(key, &rec);
3503
3504         /*
3505          * Loop going up the tree, starting at the leaf level.
3506          * Stop when we don't get a split block, that must mean that
3507          * the insert is finished with this level.
3508          */
3509         do {
3510                 /*
3511                  * Insert nrec/nptr into this level of the tree.
3512                  * Note if we fail, nptr will be null.
3513                  */
3514                 error = xfs_btree_insrec(pcur, level, &nptr, &rec, key,
3515                                 &ncur, &i);
3516                 if (error) {
3517                         if (pcur != cur)
3518                                 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3519                         goto error0;
3520                 }
3521
3522                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3523                 level++;
3524
3525                 /*
3526                  * See if the cursor we just used is trash.
3527                  * Can't trash the caller's cursor, but otherwise we should
3528                  * if ncur is a new cursor or we're about to be done.
3529                  */
3530                 if (pcur != cur &&
3531                     (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3532                         /* Save the state from the cursor before we trash it */
3533                         if (cur->bc_ops->update_cursor)
3534                                 cur->bc_ops->update_cursor(pcur, cur);
3535                         cur->bc_nlevels = pcur->bc_nlevels;
3536                         xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3537                 }
3538                 /* If we got a new cursor, switch to it. */
3539                 if (ncur) {
3540                         pcur = ncur;
3541                         ncur = NULL;
3542                 }
3543         } while (!xfs_btree_ptr_is_null(cur, &nptr));
3544
3545         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3546         *stat = i;
3547         return 0;
3548 error0:
3549         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3550         return error;
3551 }
3552
3553 /*
3554  * Try to merge a non-leaf block back into the inode root.
3555  *
3556  * Note: the killroot names comes from the fact that we're effectively
3557  * killing the old root block.  But because we can't just delete the
3558  * inode we have to copy the single block it was pointing to into the
3559  * inode.
3560  */
3561 STATIC int
3562 xfs_btree_kill_iroot(
3563         struct xfs_btree_cur    *cur)
3564 {
3565         int                     whichfork = cur->bc_private.b.whichfork;
3566         struct xfs_inode        *ip = cur->bc_private.b.ip;
3567         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
3568         struct xfs_btree_block  *block;
3569         struct xfs_btree_block  *cblock;
3570         union xfs_btree_key     *kp;
3571         union xfs_btree_key     *ckp;
3572         union xfs_btree_ptr     *pp;
3573         union xfs_btree_ptr     *cpp;
3574         struct xfs_buf          *cbp;
3575         int                     level;
3576         int                     index;
3577         int                     numrecs;
3578         int                     error;
3579 #ifdef DEBUG
3580         union xfs_btree_ptr     ptr;
3581         int                     i;
3582 #endif
3583
3584         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3585
3586         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3587         ASSERT(cur->bc_nlevels > 1);
3588
3589         /*
3590          * Don't deal with the root block needs to be a leaf case.
3591          * We're just going to turn the thing back into extents anyway.
3592          */
3593         level = cur->bc_nlevels - 1;
3594         if (level == 1)
3595                 goto out0;
3596
3597         /*
3598          * Give up if the root has multiple children.
3599          */
3600         block = xfs_btree_get_iroot(cur);
3601         if (xfs_btree_get_numrecs(block) != 1)
3602                 goto out0;
3603
3604         cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3605         numrecs = xfs_btree_get_numrecs(cblock);
3606
3607         /*
3608          * Only do this if the next level will fit.
3609          * Then the data must be copied up to the inode,
3610          * instead of freeing the root you free the next level.
3611          */
3612         if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3613                 goto out0;
3614
3615         XFS_BTREE_STATS_INC(cur, killroot);
3616
3617 #ifdef DEBUG
3618         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3619         ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3620         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3621         ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3622 #endif
3623
3624         index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3625         if (index) {
3626                 xfs_iroot_realloc(cur->bc_private.b.ip, index,
3627                                   cur->bc_private.b.whichfork);
3628                 block = ifp->if_broot;
3629         }
3630
3631         be16_add_cpu(&block->bb_numrecs, index);
3632         ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3633
3634         kp = xfs_btree_key_addr(cur, 1, block);
3635         ckp = xfs_btree_key_addr(cur, 1, cblock);
3636         xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3637
3638         pp = xfs_btree_ptr_addr(cur, 1, block);
3639         cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3640 #ifdef DEBUG
3641         for (i = 0; i < numrecs; i++) {
3642                 error = xfs_btree_check_ptr(cur, cpp, i, level - 1);
3643                 if (error) {
3644                         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3645                         return error;
3646                 }
3647         }
3648 #endif
3649         xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3650
3651         error = xfs_btree_free_block(cur, cbp);
3652         if (error) {
3653                 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3654                 return error;
3655         }
3656
3657         cur->bc_bufs[level - 1] = NULL;
3658         be16_add_cpu(&block->bb_level, -1);
3659         xfs_trans_log_inode(cur->bc_tp, ip,
3660                 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork));
3661         cur->bc_nlevels--;
3662 out0:
3663         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3664         return 0;
3665 }
3666
3667 /*
3668  * Kill the current root node, and replace it with it's only child node.
3669  */
3670 STATIC int
3671 xfs_btree_kill_root(
3672         struct xfs_btree_cur    *cur,
3673         struct xfs_buf          *bp,
3674         int                     level,
3675         union xfs_btree_ptr     *newroot)
3676 {
3677         int                     error;
3678
3679         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3680         XFS_BTREE_STATS_INC(cur, killroot);
3681
3682         /*
3683          * Update the root pointer, decreasing the level by 1 and then
3684          * free the old root.
3685          */
3686         cur->bc_ops->set_root(cur, newroot, -1);
3687
3688         error = xfs_btree_free_block(cur, bp);
3689         if (error) {
3690                 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3691                 return error;
3692         }
3693
3694         cur->bc_bufs[level] = NULL;
3695         cur->bc_ra[level] = 0;
3696         cur->bc_nlevels--;
3697
3698         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3699         return 0;
3700 }
3701
3702 STATIC int
3703 xfs_btree_dec_cursor(
3704         struct xfs_btree_cur    *cur,
3705         int                     level,
3706         int                     *stat)
3707 {
3708         int                     error;
3709         int                     i;
3710
3711         if (level > 0) {
3712                 error = xfs_btree_decrement(cur, level, &i);
3713                 if (error)
3714                         return error;
3715         }
3716
3717         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3718         *stat = 1;
3719         return 0;
3720 }
3721
3722 /*
3723  * Single level of the btree record deletion routine.
3724  * Delete record pointed to by cur/level.
3725  * Remove the record from its block then rebalance the tree.
3726  * Return 0 for error, 1 for done, 2 to go on to the next level.
3727  */
3728 STATIC int                                      /* error */
3729 xfs_btree_delrec(
3730         struct xfs_btree_cur    *cur,           /* btree cursor */
3731         int                     level,          /* level removing record from */
3732         int                     *stat)          /* fail/done/go-on */
3733 {
3734         struct xfs_btree_block  *block;         /* btree block */
3735         union xfs_btree_ptr     cptr;           /* current block ptr */
3736         struct xfs_buf          *bp;            /* buffer for block */
3737         int                     error;          /* error return value */
3738         int                     i;              /* loop counter */
3739         union xfs_btree_ptr     lptr;           /* left sibling block ptr */
3740         struct xfs_buf          *lbp;           /* left buffer pointer */
3741         struct xfs_btree_block  *left;          /* left btree block */
3742         int                     lrecs = 0;      /* left record count */
3743         int                     ptr;            /* key/record index */
3744         union xfs_btree_ptr     rptr;           /* right sibling block ptr */
3745         struct xfs_buf          *rbp;           /* right buffer pointer */
3746         struct xfs_btree_block  *right;         /* right btree block */
3747         struct xfs_btree_block  *rrblock;       /* right-right btree block */
3748         struct xfs_buf          *rrbp;          /* right-right buffer pointer */
3749         int                     rrecs = 0;      /* right record count */
3750         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
3751         int                     numrecs;        /* temporary numrec count */
3752
3753         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3754         XFS_BTREE_TRACE_ARGI(cur, level);
3755
3756         tcur = NULL;
3757
3758         /* Get the index of the entry being deleted, check for nothing there. */
3759         ptr = cur->bc_ptrs[level];
3760         if (ptr == 0) {
3761                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3762                 *stat = 0;
3763                 return 0;
3764         }
3765
3766         /* Get the buffer & block containing the record or key/ptr. */
3767         block = xfs_btree_get_block(cur, level, &bp);
3768         numrecs = xfs_btree_get_numrecs(block);
3769
3770 #ifdef DEBUG
3771         error = xfs_btree_check_block(cur, block, level, bp);
3772         if (error)
3773                 goto error0;
3774 #endif
3775
3776         /* Fail if we're off the end of the block. */
3777         if (ptr > numrecs) {
3778                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3779                 *stat = 0;
3780                 return 0;
3781         }
3782
3783         XFS_BTREE_STATS_INC(cur, delrec);
3784         XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3785
3786         /* Excise the entries being deleted. */
3787         if (level > 0) {
3788                 /* It's a nonleaf. operate on keys and ptrs */
3789                 union xfs_btree_key     *lkp;
3790                 union xfs_btree_ptr     *lpp;
3791
3792                 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3793                 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3794
3795 #ifdef DEBUG
3796                 for (i = 0; i < numrecs - ptr; i++) {
3797                         error = xfs_btree_check_ptr(cur, lpp, i, level);
3798                         if (error)
3799                                 goto error0;
3800                 }
3801 #endif
3802
3803                 if (ptr < numrecs) {
3804                         xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3805                         xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3806                         xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3807                         xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3808                 }
3809         } else {
3810                 /* It's a leaf. operate on records */
3811                 if (ptr < numrecs) {
3812                         xfs_btree_shift_recs(cur,
3813                                 xfs_btree_rec_addr(cur, ptr + 1, block),
3814                                 -1, numrecs - ptr);
3815                         xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3816                 }
3817         }
3818
3819         /*
3820          * Decrement and log the number of entries in the block.
3821          */
3822         xfs_btree_set_numrecs(block, --numrecs);
3823         xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3824
3825         /*
3826          * If we are tracking the last record in the tree and
3827          * we are at the far right edge of the tree, update it.
3828          */
3829         if (xfs_btree_is_lastrec(cur, block, level)) {
3830                 cur->bc_ops->update_lastrec(cur, block, NULL,
3831                                             ptr, LASTREC_DELREC);
3832         }
3833
3834         /*
3835          * We're at the root level.  First, shrink the root block in-memory.
3836          * Try to get rid of the next level down.  If we can't then there's
3837          * nothing left to do.
3838          */
3839         if (level == cur->bc_nlevels - 1) {
3840                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3841                         xfs_iroot_realloc(cur->bc_private.b.ip, -1,
3842                                           cur->bc_private.b.whichfork);
3843
3844                         error = xfs_btree_kill_iroot(cur);
3845                         if (error)
3846                                 goto error0;
3847
3848                         error = xfs_btree_dec_cursor(cur, level, stat);
3849                         if (error)
3850                                 goto error0;
3851                         *stat = 1;
3852                         return 0;
3853                 }
3854
3855                 /*
3856                  * If this is the root level, and there's only one entry left,
3857                  * and it's NOT the leaf level, then we can get rid of this
3858                  * level.
3859                  */
3860                 if (numrecs == 1 && level > 0) {
3861                         union xfs_btree_ptr     *pp;
3862                         /*
3863                          * pp is still set to the first pointer in the block.
3864                          * Make it the new root of the btree.
3865                          */
3866                         pp = xfs_btree_ptr_addr(cur, 1, block);
3867                         error = xfs_btree_kill_root(cur, bp, level, pp);
3868                         if (error)
3869                                 goto error0;
3870                 } else if (level > 0) {
3871                         error = xfs_btree_dec_cursor(cur, level, stat);
3872                         if (error)
3873                                 goto error0;
3874                 }
3875                 *stat = 1;
3876                 return 0;
3877         }
3878
3879         /*
3880          * If we deleted the leftmost entry in the block, update the
3881          * key values above us in the tree.
3882          */
3883         if (xfs_btree_needs_key_update(cur, ptr)) {
3884                 error = xfs_btree_update_keys(cur, level);
3885                 if (error)
3886                         goto error0;
3887         }
3888
3889         /*
3890          * If the number of records remaining in the block is at least
3891          * the minimum, we're done.
3892          */
3893         if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3894                 error = xfs_btree_dec_cursor(cur, level, stat);
3895                 if (error)
3896                         goto error0;
3897                 return 0;
3898         }
3899
3900         /*
3901          * Otherwise, we have to move some records around to keep the
3902          * tree balanced.  Look at the left and right sibling blocks to
3903          * see if we can re-balance by moving only one record.
3904          */
3905         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3906         xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3907
3908         if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3909                 /*
3910                  * One child of root, need to get a chance to copy its contents
3911                  * into the root and delete it. Can't go up to next level,
3912                  * there's nothing to delete there.
3913                  */
3914                 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3915                     xfs_btree_ptr_is_null(cur, &lptr) &&
3916                     level == cur->bc_nlevels - 2) {
3917                         error = xfs_btree_kill_iroot(cur);
3918                         if (!error)
3919                                 error = xfs_btree_dec_cursor(cur, level, stat);
3920                         if (error)
3921                                 goto error0;
3922                         return 0;
3923                 }
3924         }
3925
3926         ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3927                !xfs_btree_ptr_is_null(cur, &lptr));
3928
3929         /*
3930          * Duplicate the cursor so our btree manipulations here won't
3931          * disrupt the next level up.
3932          */
3933         error = xfs_btree_dup_cursor(cur, &tcur);
3934         if (error)
3935                 goto error0;
3936
3937         /*
3938          * If there's a right sibling, see if it's ok to shift an entry
3939          * out of it.
3940          */
3941         if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3942                 /*
3943                  * Move the temp cursor to the last entry in the next block.
3944                  * Actually any entry but the first would suffice.
3945                  */
3946                 i = xfs_btree_lastrec(tcur, level);
3947                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3948
3949                 error = xfs_btree_increment(tcur, level, &i);
3950                 if (error)
3951                         goto error0;
3952                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3953
3954                 i = xfs_btree_lastrec(tcur, level);
3955                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3956
3957                 /* Grab a pointer to the block. */
3958                 right = xfs_btree_get_block(tcur, level, &rbp);
3959 #ifdef DEBUG
3960                 error = xfs_btree_check_block(tcur, right, level, rbp);
3961                 if (error)
3962                         goto error0;
3963 #endif
3964                 /* Grab the current block number, for future use. */
3965                 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3966
3967                 /*
3968                  * If right block is full enough so that removing one entry
3969                  * won't make it too empty, and left-shifting an entry out
3970                  * of right to us works, we're done.
3971                  */
3972                 if (xfs_btree_get_numrecs(right) - 1 >=
3973                     cur->bc_ops->get_minrecs(tcur, level)) {
3974                         error = xfs_btree_lshift(tcur, level, &i);
3975                         if (error)
3976                                 goto error0;
3977                         if (i) {
3978                                 ASSERT(xfs_btree_get_numrecs(block) >=
3979                                        cur->bc_ops->get_minrecs(tcur, level));
3980
3981                                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3982                                 tcur = NULL;
3983
3984                                 error = xfs_btree_dec_cursor(cur, level, stat);
3985                                 if (error)
3986                                         goto error0;
3987                                 return 0;
3988                         }
3989                 }
3990
3991                 /*
3992                  * Otherwise, grab the number of records in right for
3993                  * future reference, and fix up the temp cursor to point
3994                  * to our block again (last record).
3995                  */
3996                 rrecs = xfs_btree_get_numrecs(right);
3997                 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3998                         i = xfs_btree_firstrec(tcur, level);
3999                         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
4000
4001                         error = xfs_btree_decrement(tcur, level, &i);
4002                         if (error)
4003                                 goto error0;
4004                         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
4005                 }
4006         }
4007
4008         /*
4009          * If there's a left sibling, see if it's ok to shift an entry
4010          * out of it.
4011          */
4012         if (!xfs_btree_ptr_is_null(cur, &lptr)) {
4013                 /*
4014                  * Move the temp cursor to the first entry in the
4015                  * previous block.
4016                  */
4017                 i = xfs_btree_firstrec(tcur, level);
4018                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
4019
4020                 error = xfs_btree_decrement(tcur, level, &i);
4021                 if (error)
4022                         goto error0;
4023                 i = xfs_btree_firstrec(tcur, level);
4024                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
4025
4026                 /* Grab a pointer to the block. */
4027                 left = xfs_btree_get_block(tcur, level, &lbp);
4028 #ifdef DEBUG
4029                 error = xfs_btree_check_block(cur, left, level, lbp);
4030                 if (error)
4031                         goto error0;
4032 #endif
4033                 /* Grab the current block number, for future use. */
4034                 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
4035
4036                 /*
4037                  * If left block is full enough so that removing one entry
4038                  * won't make it too empty, and right-shifting an entry out
4039                  * of left to us works, we're done.
4040                  */
4041                 if (xfs_btree_get_numrecs(left) - 1 >=
4042                     cur->bc_ops->get_minrecs(tcur, level)) {
4043                         error = xfs_btree_rshift(tcur, level, &i);
4044                         if (error)
4045                                 goto error0;
4046                         if (i) {
4047                                 ASSERT(xfs_btree_get_numrecs(block) >=
4048                                        cur->bc_ops->get_minrecs(tcur, level));
4049                                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4050                                 tcur = NULL;
4051                                 if (level == 0)
4052                                         cur->bc_ptrs[0]++;
4053                                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
4054                                 *stat = 1;
4055                                 return 0;
4056                         }
4057                 }
4058
4059                 /*
4060                  * Otherwise, grab the number of records in right for
4061                  * future reference.
4062                  */
4063                 lrecs = xfs_btree_get_numrecs(left);
4064         }
4065
4066         /* Delete the temp cursor, we're done with it. */
4067         xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4068         tcur = NULL;
4069
4070         /* If here, we need to do a join to keep the tree balanced. */
4071         ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
4072
4073         if (!xfs_btree_ptr_is_null(cur, &lptr) &&
4074             lrecs + xfs_btree_get_numrecs(block) <=
4075                         cur->bc_ops->get_maxrecs(cur, level)) {
4076                 /*
4077                  * Set "right" to be the starting block,
4078                  * "left" to be the left neighbor.
4079                  */
4080                 rptr = cptr;
4081                 right = block;
4082                 rbp = bp;
4083                 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
4084                 if (error)
4085                         goto error0;
4086
4087         /*
4088          * If that won't work, see if we can join with the right neighbor block.
4089          */
4090         } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
4091                    rrecs + xfs_btree_get_numrecs(block) <=
4092                         cur->bc_ops->get_maxrecs(cur, level)) {
4093                 /*
4094                  * Set "left" to be the starting block,
4095                  * "right" to be the right neighbor.
4096                  */
4097                 lptr = cptr;
4098                 left = block;
4099                 lbp = bp;
4100                 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
4101                 if (error)
4102                         goto error0;
4103
4104         /*
4105          * Otherwise, we can't fix the imbalance.
4106          * Just return.  This is probably a logic error, but it's not fatal.
4107          */
4108         } else {
4109                 error = xfs_btree_dec_cursor(cur, level, stat);
4110                 if (error)
4111                         goto error0;
4112                 return 0;
4113         }
4114
4115         rrecs = xfs_btree_get_numrecs(right);
4116         lrecs = xfs_btree_get_numrecs(left);
4117
4118         /*
4119          * We're now going to join "left" and "right" by moving all the stuff
4120          * in "right" to "left" and deleting "right".
4121          */
4122         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
4123         if (level > 0) {
4124                 /* It's a non-leaf.  Move keys and pointers. */
4125                 union xfs_btree_key     *lkp;   /* left btree key */
4126                 union xfs_btree_ptr     *lpp;   /* left address pointer */
4127                 union xfs_btree_key     *rkp;   /* right btree key */
4128                 union xfs_btree_ptr     *rpp;   /* right address pointer */
4129
4130                 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
4131                 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
4132                 rkp = xfs_btree_key_addr(cur, 1, right);
4133                 rpp = xfs_btree_ptr_addr(cur, 1, right);
4134 #ifdef DEBUG
4135                 for (i = 1; i < rrecs; i++) {
4136                         error = xfs_btree_check_ptr(cur, rpp, i, level);
4137                         if (error)
4138                                 goto error0;
4139                 }
4140 #endif
4141                 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
4142                 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
4143
4144                 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
4145                 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
4146         } else {
4147                 /* It's a leaf.  Move records.  */
4148                 union xfs_btree_rec     *lrp;   /* left record pointer */
4149                 union xfs_btree_rec     *rrp;   /* right record pointer */
4150
4151                 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
4152                 rrp = xfs_btree_rec_addr(cur, 1, right);
4153
4154                 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
4155                 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
4156         }
4157
4158         XFS_BTREE_STATS_INC(cur, join);
4159
4160         /*
4161          * Fix up the number of records and right block pointer in the
4162          * surviving block, and log it.
4163          */
4164         xfs_btree_set_numrecs(left, lrecs + rrecs);
4165         xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB),
4166         xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4167         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
4168
4169         /* If there is a right sibling, point it to the remaining block. */
4170         xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4171         if (!xfs_btree_ptr_is_null(cur, &cptr)) {
4172                 error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp);
4173                 if (error)
4174                         goto error0;
4175                 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
4176                 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
4177         }
4178
4179         /* Free the deleted block. */
4180         error = xfs_btree_free_block(cur, rbp);
4181         if (error)
4182                 goto error0;
4183
4184         /*
4185          * If we joined with the left neighbor, set the buffer in the
4186          * cursor to the left block, and fix up the index.
4187          */
4188         if (bp != lbp) {
4189                 cur->bc_bufs[level] = lbp;
4190                 cur->bc_ptrs[level] += lrecs;
4191                 cur->bc_ra[level] = 0;
4192         }
4193         /*
4194          * If we joined with the right neighbor and there's a level above
4195          * us, increment the cursor at that level.
4196          */
4197         else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
4198                    (level + 1 < cur->bc_nlevels)) {
4199                 error = xfs_btree_increment(cur, level + 1, &i);
4200                 if (error)
4201                         goto error0;
4202         }
4203
4204         /*
4205          * Readjust the ptr at this level if it's not a leaf, since it's
4206          * still pointing at the deletion point, which makes the cursor
4207          * inconsistent.  If this makes the ptr 0, the caller fixes it up.
4208          * We can't use decrement because it would change the next level up.
4209          */
4210         if (level > 0)
4211                 cur->bc_ptrs[level]--;
4212
4213         /*
4214          * We combined blocks, so we have to update the parent keys if the
4215          * btree supports overlapped intervals.  However, bc_ptrs[level + 1]
4216          * points to the old block so that the caller knows which record to
4217          * delete.  Therefore, the caller must be savvy enough to call updkeys
4218          * for us if we return stat == 2.  The other exit points from this
4219          * function don't require deletions further up the tree, so they can
4220          * call updkeys directly.
4221          */
4222
4223         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
4224         /* Return value means the next level up has something to do. */
4225         *stat = 2;
4226         return 0;
4227
4228 error0:
4229         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
4230         if (tcur)
4231                 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
4232         return error;
4233 }
4234
4235 /*
4236  * Delete the record pointed to by cur.
4237  * The cursor refers to the place where the record was (could be inserted)
4238  * when the operation returns.
4239  */
4240 int                                     /* error */
4241 xfs_btree_delete(
4242         struct xfs_btree_cur    *cur,
4243         int                     *stat)  /* success/failure */
4244 {
4245         int                     error;  /* error return value */
4246         int                     level;
4247         int                     i;
4248         bool                    joined = false;
4249
4250         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
4251
4252         /*
4253          * Go up the tree, starting at leaf level.
4254          *
4255          * If 2 is returned then a join was done; go to the next level.
4256          * Otherwise we are done.
4257          */
4258         for (level = 0, i = 2; i == 2; level++) {
4259                 error = xfs_btree_delrec(cur, level, &i);
4260                 if (error)
4261                         goto error0;
4262                 if (i == 2)
4263                         joined = true;
4264         }
4265
4266         /*
4267          * If we combined blocks as part of deleting the record, delrec won't
4268          * have updated the parent high keys so we have to do that here.
4269          */
4270         if (joined && (cur->bc_flags & XFS_BTREE_OVERLAPPING)) {
4271                 error = xfs_btree_updkeys_force(cur, 0);
4272                 if (error)
4273                         goto error0;
4274         }
4275
4276         if (i == 0) {
4277                 for (level = 1; level < cur->bc_nlevels; level++) {
4278                         if (cur->bc_ptrs[level] == 0) {
4279                                 error = xfs_btree_decrement(cur, level, &i);
4280                                 if (error)
4281                                         goto error0;
4282                                 break;
4283                         }
4284                 }
4285         }
4286
4287         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
4288         *stat = i;
4289         return 0;
4290 error0:
4291         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
4292         return error;
4293 }
4294
4295 /*
4296  * Get the data from the pointed-to record.
4297  */
4298 int                                     /* error */
4299 xfs_btree_get_rec(
4300         struct xfs_btree_cur    *cur,   /* btree cursor */
4301         union xfs_btree_rec     **recp, /* output: btree record */
4302         int                     *stat)  /* output: success/failure */
4303 {
4304         struct xfs_btree_block  *block; /* btree block */
4305         struct xfs_buf          *bp;    /* buffer pointer */
4306         int                     ptr;    /* record number */
4307 #ifdef DEBUG
4308         int                     error;  /* error return value */
4309 #endif
4310
4311         ptr = cur->bc_ptrs[0];
4312         block = xfs_btree_get_block(cur, 0, &bp);
4313
4314 #ifdef DEBUG
4315         error = xfs_btree_check_block(cur, block, 0, bp);
4316         if (error)
4317                 return error;
4318 #endif
4319
4320         /*
4321          * Off the right end or left end, return failure.
4322          */
4323         if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
4324                 *stat = 0;
4325                 return 0;
4326         }
4327
4328         /*
4329          * Point to the record and extract its data.
4330          */
4331         *recp = xfs_btree_rec_addr(cur, ptr, block);
4332         *stat = 1;
4333         return 0;
4334 }
4335
4336 /* Visit a block in a btree. */
4337 STATIC int
4338 xfs_btree_visit_block(
4339         struct xfs_btree_cur            *cur,
4340         int                             level,
4341         xfs_btree_visit_blocks_fn       fn,
4342         void                            *data)
4343 {
4344         struct xfs_btree_block          *block;
4345         struct xfs_buf                  *bp;
4346         union xfs_btree_ptr             rptr;
4347         int                             error;
4348
4349         /* do right sibling readahead */
4350         xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
4351         block = xfs_btree_get_block(cur, level, &bp);
4352
4353         /* process the block */
4354         error = fn(cur, level, data);
4355         if (error)
4356                 return error;
4357
4358         /* now read rh sibling block for next iteration */
4359         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4360         if (xfs_btree_ptr_is_null(cur, &rptr))
4361                 return -ENOENT;
4362
4363         return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
4364 }
4365
4366
4367 /* Visit every block in a btree. */
4368 int
4369 xfs_btree_visit_blocks(
4370         struct xfs_btree_cur            *cur,
4371         xfs_btree_visit_blocks_fn       fn,
4372         void                            *data)
4373 {
4374         union xfs_btree_ptr             lptr;
4375         int                             level;
4376         struct xfs_btree_block          *block = NULL;
4377         int                             error = 0;
4378
4379         cur->bc_ops->init_ptr_from_cur(cur, &lptr);
4380
4381         /* for each level */
4382         for (level = cur->bc_nlevels - 1; level >= 0; level--) {
4383                 /* grab the left hand block */
4384                 error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
4385                 if (error)
4386                         return error;
4387
4388                 /* readahead the left most block for the next level down */
4389                 if (level > 0) {
4390                         union xfs_btree_ptr     *ptr;
4391
4392                         ptr = xfs_btree_ptr_addr(cur, 1, block);
4393                         xfs_btree_readahead_ptr(cur, ptr, 1);
4394
4395                         /* save for the next iteration of the loop */
4396                         xfs_btree_copy_ptrs(cur, &lptr, ptr, 1);
4397                 }
4398
4399                 /* for each buffer in the level */
4400                 do {
4401                         error = xfs_btree_visit_block(cur, level, fn, data);
4402                 } while (!error);
4403
4404                 if (error != -ENOENT)
4405                         return error;
4406         }
4407
4408         return 0;
4409 }
4410
4411 /*
4412  * Change the owner of a btree.
4413  *
4414  * The mechanism we use here is ordered buffer logging. Because we don't know
4415  * how many buffers were are going to need to modify, we don't really want to
4416  * have to make transaction reservations for the worst case of every buffer in a
4417  * full size btree as that may be more space that we can fit in the log....
4418  *
4419  * We do the btree walk in the most optimal manner possible - we have sibling
4420  * pointers so we can just walk all the blocks on each level from left to right
4421  * in a single pass, and then move to the next level and do the same. We can
4422  * also do readahead on the sibling pointers to get IO moving more quickly,
4423  * though for slow disks this is unlikely to make much difference to performance
4424  * as the amount of CPU work we have to do before moving to the next block is
4425  * relatively small.
4426  *
4427  * For each btree block that we load, modify the owner appropriately, set the
4428  * buffer as an ordered buffer and log it appropriately. We need to ensure that
4429  * we mark the region we change dirty so that if the buffer is relogged in
4430  * a subsequent transaction the changes we make here as an ordered buffer are
4431  * correctly relogged in that transaction.  If we are in recovery context, then
4432  * just queue the modified buffer as delayed write buffer so the transaction
4433  * recovery completion writes the changes to disk.
4434  */
4435 struct xfs_btree_block_change_owner_info {
4436         uint64_t                new_owner;
4437         struct list_head        *buffer_list;
4438 };
4439
4440 static int
4441 xfs_btree_block_change_owner(
4442         struct xfs_btree_cur    *cur,
4443         int                     level,
4444         void                    *data)
4445 {
4446         struct xfs_btree_block_change_owner_info        *bbcoi = data;
4447         struct xfs_btree_block  *block;
4448         struct xfs_buf          *bp;
4449
4450         /* modify the owner */
4451         block = xfs_btree_get_block(cur, level, &bp);
4452         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
4453                 block->bb_u.l.bb_owner = cpu_to_be64(bbcoi->new_owner);
4454         else
4455                 block->bb_u.s.bb_owner = cpu_to_be32(bbcoi->new_owner);
4456
4457         /*
4458          * If the block is a root block hosted in an inode, we might not have a
4459          * buffer pointer here and we shouldn't attempt to log the change as the
4460          * information is already held in the inode and discarded when the root
4461          * block is formatted into the on-disk inode fork. We still change it,
4462          * though, so everything is consistent in memory.
4463          */
4464         if (bp) {
4465                 if (cur->bc_tp) {
4466                         xfs_trans_ordered_buf(cur->bc_tp, bp);
4467                         xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
4468                 } else {
4469                         xfs_buf_delwri_queue(bp, bbcoi->buffer_list);
4470                 }
4471         } else {
4472                 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
4473                 ASSERT(level == cur->bc_nlevels - 1);
4474         }
4475
4476         return 0;
4477 }
4478
4479 int
4480 xfs_btree_change_owner(
4481         struct xfs_btree_cur    *cur,
4482         uint64_t                new_owner,
4483         struct list_head        *buffer_list)
4484 {
4485         struct xfs_btree_block_change_owner_info        bbcoi;
4486
4487         bbcoi.new_owner = new_owner;
4488         bbcoi.buffer_list = buffer_list;
4489
4490         return xfs_btree_visit_blocks(cur, xfs_btree_block_change_owner,
4491                         &bbcoi);
4492 }
4493
4494 /**
4495  * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format
4496  *                                    btree block
4497  *
4498  * @bp: buffer containing the btree block
4499  * @max_recs: pointer to the m_*_mxr max records field in the xfs mount
4500  * @pag_max_level: pointer to the per-ag max level field
4501  */
4502 bool
4503 xfs_btree_sblock_v5hdr_verify(
4504         struct xfs_buf          *bp)
4505 {
4506         struct xfs_mount        *mp = bp->b_target->bt_mount;
4507         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4508         struct xfs_perag        *pag = bp->b_pag;
4509
4510         if (!xfs_sb_version_hascrc(&mp->m_sb))
4511                 return false;
4512         if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
4513                 return false;
4514         if (block->bb_u.s.bb_blkno != cpu_to_be64(bp->b_bn))
4515                 return false;
4516         if (pag && be32_to_cpu(block->bb_u.s.bb_owner) != pag->pag_agno)
4517                 return false;
4518         return true;
4519 }
4520
4521 /**
4522  * xfs_btree_sblock_verify() -- verify a short-format btree block
4523  *
4524  * @bp: buffer containing the btree block
4525  * @max_recs: maximum records allowed in this btree node
4526  */
4527 bool
4528 xfs_btree_sblock_verify(
4529         struct xfs_buf          *bp,
4530         unsigned int            max_recs)
4531 {
4532         struct xfs_mount        *mp = bp->b_target->bt_mount;
4533         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4534
4535         /* numrecs verification */
4536         if (be16_to_cpu(block->bb_numrecs) > max_recs)
4537                 return false;
4538
4539         /* sibling pointer verification */
4540         if (!block->bb_u.s.bb_leftsib ||
4541             (be32_to_cpu(block->bb_u.s.bb_leftsib) >= mp->m_sb.sb_agblocks &&
4542              block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK)))
4543                 return false;
4544         if (!block->bb_u.s.bb_rightsib ||
4545             (be32_to_cpu(block->bb_u.s.bb_rightsib) >= mp->m_sb.sb_agblocks &&
4546              block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK)))
4547                 return false;
4548
4549         return true;
4550 }
4551
4552 /*
4553  * Calculate the number of btree levels needed to store a given number of
4554  * records in a short-format btree.
4555  */
4556 uint
4557 xfs_btree_compute_maxlevels(
4558         struct xfs_mount        *mp,
4559         uint                    *limits,
4560         unsigned long           len)
4561 {
4562         uint                    level;
4563         unsigned long           maxblocks;
4564
4565         maxblocks = (len + limits[0] - 1) / limits[0];
4566         for (level = 1; maxblocks > 1; level++)
4567                 maxblocks = (maxblocks + limits[1] - 1) / limits[1];
4568         return level;
4569 }
4570
4571 /*
4572  * Query a regular btree for all records overlapping a given interval.
4573  * Start with a LE lookup of the key of low_rec and return all records
4574  * until we find a record with a key greater than the key of high_rec.
4575  */
4576 STATIC int
4577 xfs_btree_simple_query_range(
4578         struct xfs_btree_cur            *cur,
4579         union xfs_btree_key             *low_key,
4580         union xfs_btree_key             *high_key,
4581         xfs_btree_query_range_fn        fn,
4582         void                            *priv)
4583 {
4584         union xfs_btree_rec             *recp;
4585         union xfs_btree_key             rec_key;
4586         int64_t                         diff;
4587         int                             stat;
4588         bool                            firstrec = true;
4589         int                             error;
4590
4591         ASSERT(cur->bc_ops->init_high_key_from_rec);
4592         ASSERT(cur->bc_ops->diff_two_keys);
4593
4594         /*
4595          * Find the leftmost record.  The btree cursor must be set
4596          * to the low record used to generate low_key.
4597          */
4598         stat = 0;
4599         error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
4600         if (error)
4601                 goto out;
4602
4603         /* Nothing?  See if there's anything to the right. */
4604         if (!stat) {
4605                 error = xfs_btree_increment(cur, 0, &stat);
4606                 if (error)
4607                         goto out;
4608         }
4609
4610         while (stat) {
4611                 /* Find the record. */
4612                 error = xfs_btree_get_rec(cur, &recp, &stat);
4613                 if (error || !stat)
4614                         break;
4615
4616                 /* Skip if high_key(rec) < low_key. */
4617                 if (firstrec) {
4618                         cur->bc_ops->init_high_key_from_rec(&rec_key, recp);
4619                         firstrec = false;
4620                         diff = cur->bc_ops->diff_two_keys(cur, low_key,
4621                                         &rec_key);
4622                         if (diff > 0)
4623                                 goto advloop;
4624                 }
4625
4626                 /* Stop if high_key < low_key(rec). */
4627                 cur->bc_ops->init_key_from_rec(&rec_key, recp);
4628                 diff = cur->bc_ops->diff_two_keys(cur, &rec_key, high_key);
4629                 if (diff > 0)
4630                         break;
4631
4632                 /* Callback */
4633                 error = fn(cur, recp, priv);
4634                 if (error < 0 || error == XFS_BTREE_QUERY_RANGE_ABORT)
4635                         break;
4636
4637 advloop:
4638                 /* Move on to the next record. */
4639                 error = xfs_btree_increment(cur, 0, &stat);
4640                 if (error)
4641                         break;
4642         }
4643
4644 out:
4645         return error;
4646 }
4647
4648 /*
4649  * Query an overlapped interval btree for all records overlapping a given
4650  * interval.  This function roughly follows the algorithm given in
4651  * "Interval Trees" of _Introduction to Algorithms_, which is section
4652  * 14.3 in the 2nd and 3rd editions.
4653  *
4654  * First, generate keys for the low and high records passed in.
4655  *
4656  * For any leaf node, generate the high and low keys for the record.
4657  * If the record keys overlap with the query low/high keys, pass the
4658  * record to the function iterator.
4659  *
4660  * For any internal node, compare the low and high keys of each
4661  * pointer against the query low/high keys.  If there's an overlap,
4662  * follow the pointer.
4663  *
4664  * As an optimization, we stop scanning a block when we find a low key
4665  * that is greater than the query's high key.
4666  */
4667 STATIC int
4668 xfs_btree_overlapped_query_range(
4669         struct xfs_btree_cur            *cur,
4670         union xfs_btree_key             *low_key,
4671         union xfs_btree_key             *high_key,
4672         xfs_btree_query_range_fn        fn,
4673         void                            *priv)
4674 {
4675         union xfs_btree_ptr             ptr;
4676         union xfs_btree_ptr             *pp;
4677         union xfs_btree_key             rec_key;
4678         union xfs_btree_key             rec_hkey;
4679         union xfs_btree_key             *lkp;
4680         union xfs_btree_key             *hkp;
4681         union xfs_btree_rec             *recp;
4682         struct xfs_btree_block          *block;
4683         int64_t                         ldiff;
4684         int64_t                         hdiff;
4685         int                             level;
4686         struct xfs_buf                  *bp;
4687         int                             i;
4688         int                             error;
4689
4690         /* Load the root of the btree. */
4691         level = cur->bc_nlevels - 1;
4692         cur->bc_ops->init_ptr_from_cur(cur, &ptr);
4693         error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
4694         if (error)
4695                 return error;
4696         xfs_btree_get_block(cur, level, &bp);
4697         trace_xfs_btree_overlapped_query_range(cur, level, bp);
4698 #ifdef DEBUG
4699         error = xfs_btree_check_block(cur, block, level, bp);
4700         if (error)
4701                 goto out;
4702 #endif
4703         cur->bc_ptrs[level] = 1;
4704
4705         while (level < cur->bc_nlevels) {
4706                 block = xfs_btree_get_block(cur, level, &bp);
4707
4708                 /* End of node, pop back towards the root. */
4709                 if (cur->bc_ptrs[level] > be16_to_cpu(block->bb_numrecs)) {
4710 pop_up:
4711                         if (level < cur->bc_nlevels - 1)
4712                                 cur->bc_ptrs[level + 1]++;
4713                         level++;
4714                         continue;
4715                 }
4716
4717                 if (level == 0) {
4718                         /* Handle a leaf node. */
4719                         recp = xfs_btree_rec_addr(cur, cur->bc_ptrs[0], block);
4720
4721                         cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
4722                         ldiff = cur->bc_ops->diff_two_keys(cur, &rec_hkey,
4723                                         low_key);
4724
4725                         cur->bc_ops->init_key_from_rec(&rec_key, recp);
4726                         hdiff = cur->bc_ops->diff_two_keys(cur, high_key,
4727                                         &rec_key);
4728
4729                         /*
4730                          * If (record's high key >= query's low key) and
4731                          *    (query's high key >= record's low key), then
4732                          * this record overlaps the query range; callback.
4733                          */
4734                         if (ldiff >= 0 && hdiff >= 0) {
4735                                 error = fn(cur, recp, priv);
4736                                 if (error < 0 ||
4737                                     error == XFS_BTREE_QUERY_RANGE_ABORT)
4738                                         break;
4739                         } else if (hdiff < 0) {
4740                                 /* Record is larger than high key; pop. */
4741                                 goto pop_up;
4742                         }
4743                         cur->bc_ptrs[level]++;
4744                         continue;
4745                 }
4746
4747                 /* Handle an internal node. */
4748                 lkp = xfs_btree_key_addr(cur, cur->bc_ptrs[level], block);
4749                 hkp = xfs_btree_high_key_addr(cur, cur->bc_ptrs[level], block);
4750                 pp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[level], block);
4751
4752                 ldiff = cur->bc_ops->diff_two_keys(cur, hkp, low_key);
4753                 hdiff = cur->bc_ops->diff_two_keys(cur, high_key, lkp);
4754
4755                 /*
4756                  * If (pointer's high key >= query's low key) and
4757                  *    (query's high key >= pointer's low key), then
4758                  * this record overlaps the query range; follow pointer.
4759                  */
4760                 if (ldiff >= 0 && hdiff >= 0) {
4761                         level--;
4762                         error = xfs_btree_lookup_get_block(cur, level, pp,
4763                                         &block);
4764                         if (error)
4765                                 goto out;
4766                         xfs_btree_get_block(cur, level, &bp);
4767                         trace_xfs_btree_overlapped_query_range(cur, level, bp);
4768 #ifdef DEBUG
4769                         error = xfs_btree_check_block(cur, block, level, bp);
4770                         if (error)
4771                                 goto out;
4772 #endif
4773                         cur->bc_ptrs[level] = 1;
4774                         continue;
4775                 } else if (hdiff < 0) {
4776                         /* The low key is larger than the upper range; pop. */
4777                         goto pop_up;
4778                 }
4779                 cur->bc_ptrs[level]++;
4780         }
4781
4782 out:
4783         /*
4784          * If we don't end this function with the cursor pointing at a record
4785          * block, a subsequent non-error cursor deletion will not release
4786          * node-level buffers, causing a buffer leak.  This is quite possible
4787          * with a zero-results range query, so release the buffers if we
4788          * failed to return any results.
4789          */
4790         if (cur->bc_bufs[0] == NULL) {
4791                 for (i = 0; i < cur->bc_nlevels; i++) {
4792                         if (cur->bc_bufs[i]) {
4793                                 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
4794                                 cur->bc_bufs[i] = NULL;
4795                                 cur->bc_ptrs[i] = 0;
4796                                 cur->bc_ra[i] = 0;
4797                         }
4798                 }
4799         }
4800
4801         return error;
4802 }
4803
4804 /*
4805  * Query a btree for all records overlapping a given interval of keys.  The
4806  * supplied function will be called with each record found; return one of the
4807  * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
4808  * code.  This function returns XFS_BTREE_QUERY_RANGE_ABORT, zero, or a
4809  * negative error code.
4810  */
4811 int
4812 xfs_btree_query_range(
4813         struct xfs_btree_cur            *cur,
4814         union xfs_btree_irec            *low_rec,
4815         union xfs_btree_irec            *high_rec,
4816         xfs_btree_query_range_fn        fn,
4817         void                            *priv)
4818 {
4819         union xfs_btree_rec             rec;
4820         union xfs_btree_key             low_key;
4821         union xfs_btree_key             high_key;
4822
4823         /* Find the keys of both ends of the interval. */
4824         cur->bc_rec = *high_rec;
4825         cur->bc_ops->init_rec_from_cur(cur, &rec);
4826         cur->bc_ops->init_key_from_rec(&high_key, &rec);
4827
4828         cur->bc_rec = *low_rec;
4829         cur->bc_ops->init_rec_from_cur(cur, &rec);
4830         cur->bc_ops->init_key_from_rec(&low_key, &rec);
4831
4832         /* Enforce low key < high key. */
4833         if (cur->bc_ops->diff_two_keys(cur, &low_key, &high_key) > 0)
4834                 return -EINVAL;
4835
4836         if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING))
4837                 return xfs_btree_simple_query_range(cur, &low_key,
4838                                 &high_key, fn, priv);
4839         return xfs_btree_overlapped_query_range(cur, &low_key, &high_key,
4840                         fn, priv);
4841 }
4842
4843 /* Query a btree for all records. */
4844 int
4845 xfs_btree_query_all(
4846         struct xfs_btree_cur            *cur,
4847         xfs_btree_query_range_fn        fn,
4848         void                            *priv)
4849 {
4850         union xfs_btree_key             low_key;
4851         union xfs_btree_key             high_key;
4852
4853         memset(&cur->bc_rec, 0, sizeof(cur->bc_rec));
4854         memset(&low_key, 0, sizeof(low_key));
4855         memset(&high_key, 0xFF, sizeof(high_key));
4856
4857         return xfs_btree_simple_query_range(cur, &low_key, &high_key, fn, priv);
4858 }
4859
4860 /*
4861  * Calculate the number of blocks needed to store a given number of records
4862  * in a short-format (per-AG metadata) btree.
4863  */
4864 xfs_extlen_t
4865 xfs_btree_calc_size(
4866         struct xfs_mount        *mp,
4867         uint                    *limits,
4868         unsigned long long      len)
4869 {
4870         int                     level;
4871         int                     maxrecs;
4872         xfs_extlen_t            rval;
4873
4874         maxrecs = limits[0];
4875         for (level = 0, rval = 0; len > 1; level++) {
4876                 len += maxrecs - 1;
4877                 do_div(len, maxrecs);
4878                 maxrecs = limits[1];
4879                 rval += len;
4880         }
4881         return rval;
4882 }
4883
4884 static int
4885 xfs_btree_count_blocks_helper(
4886         struct xfs_btree_cur    *cur,
4887         int                     level,
4888         void                    *data)
4889 {
4890         xfs_extlen_t            *blocks = data;
4891         (*blocks)++;
4892
4893         return 0;
4894 }
4895
4896 /* Count the blocks in a btree and return the result in *blocks. */
4897 int
4898 xfs_btree_count_blocks(
4899         struct xfs_btree_cur    *cur,
4900         xfs_extlen_t            *blocks)
4901 {
4902         *blocks = 0;
4903         return xfs_btree_visit_blocks(cur, xfs_btree_count_blocks_helper,
4904                         blocks);
4905 }