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                         XFS_RANDOM_BTREE_CHECK_LBLOCK))) {
 106                 if (bp)
 107                         trace_xfs_btree_corrupt(bp, _RET_IP_);
 108                 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
 109                 return -EFSCORRUPTED;
 110         }
 111         return 0;
 112 }
 113
 114 STATIC int                              /* error (0 or EFSCORRUPTED) */
 115 xfs_btree_check_sblock(
 116         struct xfs_btree_cur    *cur,   /* btree cursor */
 117         struct xfs_btree_block  *block, /* btree short form block pointer */
 118         int                     level,  /* level of the btree block */
 119         struct xfs_buf          *bp)    /* buffer containing block */
 120 {
 121         struct xfs_mount        *mp;    /* file system mount point */
 122         struct xfs_buf          *agbp;  /* buffer for ag. freespace struct */
 123         struct xfs_agf          *agf;   /* ag. freespace structure */
 124         xfs_agblock_t           agflen; /* native ag. freespace length */
 125         int                     sblock_ok = 1; /* block passes checks */
 126         xfs_btnum_t             btnum = cur->bc_btnum;
 127         int                     crc;
 128
 129         mp = cur->bc_mp;
 130         crc = xfs_sb_version_hascrc(&mp->m_sb);
 131         agbp = cur->bc_private.a.agbp;
 132         agf = XFS_BUF_TO_AGF(agbp);
 133         agflen = be32_to_cpu(agf->agf_length);
 134
 135         if (crc) {
 136                 sblock_ok = sblock_ok &&
 137                         uuid_equal(&block->bb_u.s.bb_uuid,
 138                                    &mp->m_sb.sb_meta_uuid) &&
 139                         block->bb_u.s.bb_blkno == cpu_to_be64(
 140                                 bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
 141         }
 142
 143         sblock_ok = sblock_ok &&
 144                 be32_to_cpu(block->bb_magic) == xfs_btree_magic(crc, btnum) &&
 145                 be16_to_cpu(block->bb_level) == level &&
 146                 be16_to_cpu(block->bb_numrecs) <=
 147                         cur->bc_ops->get_maxrecs(cur, level) &&
 148                 (block->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) ||
 149                  be32_to_cpu(block->bb_u.s.bb_leftsib) < agflen) &&
 150                 block->bb_u.s.bb_leftsib &&
 151                 (block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK) ||
 152                  be32_to_cpu(block->bb_u.s.bb_rightsib) < agflen) &&
 153                 block->bb_u.s.bb_rightsib;
 154
 155         if (unlikely(XFS_TEST_ERROR(!sblock_ok, mp,
 156                         XFS_ERRTAG_BTREE_CHECK_SBLOCK,
 157                         XFS_RANDOM_BTREE_CHECK_SBLOCK))) {
 158                 if (bp)
 159                         trace_xfs_btree_corrupt(bp, _RET_IP_);
 160                 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
 161                 return -EFSCORRUPTED;
 162         }
 163         return 0;
 164 }
 165
 166 /*
 167  * Debug routine: check that block header is ok.
 168  */
 169 int
 170 xfs_btree_check_block(
 171         struct xfs_btree_cur    *cur,   /* btree cursor */
 172         struct xfs_btree_block  *block, /* generic btree block pointer */
 173         int                     level,  /* level of the btree block */
 174         struct xfs_buf          *bp)    /* buffer containing block, if any */
 175 {
 176         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
 177                 return xfs_btree_check_lblock(cur, block, level, bp);
 178         else
 179                 return xfs_btree_check_sblock(cur, block, level, bp);
 180 }
 181
 182 /*
 183  * Check that (long) pointer is ok.
 184  */
 185 int                                     /* error (0 or EFSCORRUPTED) */
 186 xfs_btree_check_lptr(
 187         struct xfs_btree_cur    *cur,   /* btree cursor */
 188         xfs_fsblock_t           bno,    /* btree block disk address */
 189         int                     level)  /* btree block level */
 190 {
 191         XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
 192                 level > 0 &&
 193                 bno != NULLFSBLOCK &&
 194                 XFS_FSB_SANITY_CHECK(cur->bc_mp, bno));
 195         return 0;
 196 }
 197
 198 #ifdef DEBUG
 199 /*
 200  * Check that (short) pointer is ok.
 201  */
 202 STATIC int                              /* error (0 or EFSCORRUPTED) */
 203 xfs_btree_check_sptr(
 204         struct xfs_btree_cur    *cur,   /* btree cursor */
 205         xfs_agblock_t           bno,    /* btree block disk address */
 206         int                     level)  /* btree block level */
 207 {
 208         xfs_agblock_t           agblocks = cur->bc_mp->m_sb.sb_agblocks;
 209
 210         XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
 211                 level > 0 &&
 212                 bno != NULLAGBLOCK &&
 213                 bno != 0 &&
 214                 bno < agblocks);
 215         return 0;
 216 }
 217
 218 /*
 219  * Check that block ptr is ok.
 220  */
 221 STATIC int                              /* error (0 or EFSCORRUPTED) */
 222 xfs_btree_check_ptr(
 223         struct xfs_btree_cur    *cur,   /* btree cursor */
 224         union xfs_btree_ptr     *ptr,   /* btree block disk address */
 225         int                     index,  /* offset from ptr to check */
 226         int                     level)  /* btree block level */
 227 {
 228         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 229                 return xfs_btree_check_lptr(cur,
 230                                 be64_to_cpu((&ptr->l)[index]), level);
 231         } else {
 232                 return xfs_btree_check_sptr(cur,
 233                                 be32_to_cpu((&ptr->s)[index]), level);
 234         }
 235 }
 236 #endif
 237
 238 /*
 239  * Calculate CRC on the whole btree block and stuff it into the
 240  * long-form btree header.
 241  *
 242  * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
 243  * it into the buffer so recovery knows what the last modification was that made
 244  * it to disk.
 245  */
 246 void
 247 xfs_btree_lblock_calc_crc(
 248         struct xfs_buf          *bp)
 249 {
 250         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 251         struct xfs_buf_log_item *bip = bp->b_fspriv;
 252
 253         if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
 254                 return;
 255         if (bip)
 256                 block->bb_u.l.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 257         xfs_buf_update_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
 258 }
 259
 260 bool
 261 xfs_btree_lblock_verify_crc(
 262         struct xfs_buf          *bp)
 263 {
 264         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 265         struct xfs_mount        *mp = bp->b_target->bt_mount;
 266
 267         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 268                 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.l.bb_lsn)))
 269                         return false;
 270                 return xfs_buf_verify_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
 271         }
 272
 273         return true;
 274 }
 275
 276 /*
 277  * Calculate CRC on the whole btree block and stuff it into the
 278  * short-form btree header.
 279  *
 280  * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
 281  * it into the buffer so recovery knows what the last modification was that made
 282  * it to disk.
 283  */
 284 void
 285 xfs_btree_sblock_calc_crc(
 286         struct xfs_buf          *bp)
 287 {
 288         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 289         struct xfs_buf_log_item *bip = bp->b_fspriv;
 290
 291         if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
 292                 return;
 293         if (bip)
 294                 block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 295         xfs_buf_update_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
 296 }
 297
 298 bool
 299 xfs_btree_sblock_verify_crc(
 300         struct xfs_buf          *bp)
 301 {
 302         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 303         struct xfs_mount        *mp = bp->b_target->bt_mount;
 304
 305         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 306                 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.s.bb_lsn)))
 307                         return false;
 308                 return xfs_buf_verify_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
 309         }
 310
 311         return true;
 312 }
 313
 314 static int
 315 xfs_btree_free_block(
 316         struct xfs_btree_cur    *cur,
 317         struct xfs_buf          *bp)
 318 {
 319         int                     error;
 320
 321         error = cur->bc_ops->free_block(cur, bp);
 322         if (!error) {
 323                 xfs_trans_binval(cur->bc_tp, bp);
 324                 XFS_BTREE_STATS_INC(cur, free);
 325         }
 326         return error;
 327 }
 328
 329 /*
 330  * Delete the btree cursor.
 331  */
 332 void
 333 xfs_btree_del_cursor(
 334         xfs_btree_cur_t *cur,           /* btree cursor */
 335         int             error)          /* del because of error */
 336 {
 337         int             i;              /* btree level */
 338
 339         /*
 340          * Clear the buffer pointers, and release the buffers.
 341          * If we're doing this in the face of an error, we
 342          * need to make sure to inspect all of the entries
 343          * in the bc_bufs array for buffers to be unlocked.
 344          * This is because some of the btree code works from
 345          * level n down to 0, and if we get an error along
 346          * the way we won't have initialized all the entries
 347          * down to 0.
 348          */
 349         for (i = 0; i < cur->bc_nlevels; i++) {
 350                 if (cur->bc_bufs[i])
 351                         xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
 352                 else if (!error)
 353                         break;
 354         }
 355         /*
 356          * Can't free a bmap cursor without having dealt with the
 357          * allocated indirect blocks' accounting.
 358          */
 359         ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP ||
 360                cur->bc_private.b.allocated == 0);
 361         /*
 362          * Free the cursor.
 363          */
 364         kmem_zone_free(xfs_btree_cur_zone, cur);
 365 }
 366
 367 /*
 368  * Duplicate the btree cursor.
 369  * Allocate a new one, copy the record, re-get the buffers.
 370  */
 371 int                                     /* error */
 372 xfs_btree_dup_cursor(
 373         xfs_btree_cur_t *cur,           /* input cursor */
 374         xfs_btree_cur_t **ncur)         /* output cursor */
 375 {
 376         xfs_buf_t       *bp;            /* btree block's buffer pointer */
 377         int             error;          /* error return value */
 378         int             i;              /* level number of btree block */
 379         xfs_mount_t     *mp;            /* mount structure for filesystem */
 380         xfs_btree_cur_t *new;           /* new cursor value */
 381         xfs_trans_t     *tp;            /* transaction pointer, can be NULL */
 382
 383         tp = cur->bc_tp;
 384         mp = cur->bc_mp;
 385
 386         /*
 387          * Allocate a new cursor like the old one.
 388          */
 389         new = cur->bc_ops->dup_cursor(cur);
 390
 391         /*
 392          * Copy the record currently in the cursor.
 393          */
 394         new->bc_rec = cur->bc_rec;
 395
 396         /*
 397          * For each level current, re-get the buffer and copy the ptr value.
 398          */
 399         for (i = 0; i < new->bc_nlevels; i++) {
 400                 new->bc_ptrs[i] = cur->bc_ptrs[i];
 401                 new->bc_ra[i] = cur->bc_ra[i];
 402                 bp = cur->bc_bufs[i];
 403                 if (bp) {
 404                         error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
 405                                                    XFS_BUF_ADDR(bp), mp->m_bsize,
 406                                                    0, &bp,
 407                                                    cur->bc_ops->buf_ops);
 408                         if (error) {
 409                                 xfs_btree_del_cursor(new, error);
 410                                 *ncur = NULL;
 411                                 return error;
 412                         }
 413                 }
 414                 new->bc_bufs[i] = bp;
 415         }
 416         *ncur = new;
 417         return 0;
 418 }
 419
 420 /*
 421  * XFS btree block layout and addressing:
 422  *
 423  * There are two types of blocks in the btree: leaf and non-leaf blocks.
 424  *
 425  * The leaf record start with a header then followed by records containing
 426  * the values.  A non-leaf block also starts with the same header, and
 427  * then first contains lookup keys followed by an equal number of pointers
 428  * to the btree blocks at the previous level.
 429  *
 430  *              +--------+-------+-------+-------+-------+-------+-------+
 431  * Leaf:        | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
 432  *              +--------+-------+-------+-------+-------+-------+-------+
 433  *
 434  *              +--------+-------+-------+-------+-------+-------+-------+
 435  * Non-Leaf:    | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
 436  *              +--------+-------+-------+-------+-------+-------+-------+
 437  *
 438  * The header is called struct xfs_btree_block for reasons better left unknown
 439  * and comes in different versions for short (32bit) and long (64bit) block
 440  * pointers.  The record and key structures are defined by the btree instances
 441  * and opaque to the btree core.  The block pointers are simple disk endian
 442  * integers, available in a short (32bit) and long (64bit) variant.
 443  *
 444  * The helpers below calculate the offset of a given record, key or pointer
 445  * into a btree block (xfs_btree_*_offset) or return a pointer to the given
 446  * record, key or pointer (xfs_btree_*_addr).  Note that all addressing
 447  * inside the btree block is done using indices starting at one, not zero!
 448  *
 449  * If XFS_BTREE_OVERLAPPING is set, then this btree supports keys containing
 450  * overlapping intervals.  In such a tree, records are still sorted lowest to
 451  * highest and indexed by the smallest key value that refers to the record.
 452  * However, nodes are different: each pointer has two associated keys -- one
 453  * indexing the lowest key available in the block(s) below (the same behavior
 454  * as the key in a regular btree) and another indexing the highest key
 455  * available in the block(s) below.  Because records are /not/ sorted by the
 456  * highest key, all leaf block updates require us to compute the highest key
 457  * that matches any record in the leaf and to recursively update the high keys
 458  * in the nodes going further up in the tree, if necessary.  Nodes look like
 459  * this:
 460  *
 461  *              +--------+-----+-----+-----+-----+-----+-------+-------+-----+
 462  * Non-Leaf:    | header | lo1 | hi1 | lo2 | hi2 | ... | ptr 1 | ptr 2 | ... |
 463  *              +--------+-----+-----+-----+-----+-----+-------+-------+-----+
 464  *
 465  * To perform an interval query on an overlapped tree, perform the usual
 466  * depth-first search and use the low and high keys to decide if we can skip
 467  * that particular node.  If a leaf node is reached, return the records that
 468  * intersect the interval.  Note that an interval query may return numerous
 469  * entries.  For a non-overlapped tree, simply search for the record associated
 470  * with the lowest key and iterate forward until a non-matching record is
 471  * found.  Section 14.3 ("Interval Trees") of _Introduction to Algorithms_ by
 472  * Cormen, Leiserson, Rivest, and Stein (2nd or 3rd ed. only) discuss this in
 473  * more detail.
 474  *
 475  * Why do we care about overlapping intervals?  Let's say you have a bunch of
 476  * reverse mapping records on a reflink filesystem:
 477  *
 478  * 1: +- file A startblock B offset C length D -----------+
 479  * 2:      +- file E startblock F offset G length H --------------+
 480  * 3:      +- file I startblock F offset J length K --+
 481  * 4:                                                        +- file L... --+
 482  *
 483  * Now say we want to map block (B+D) into file A at offset (C+D).  Ideally,
 484  * we'd simply increment the length of record 1.  But how do we find the record
 485  * that ends at (B+D-1) (i.e. record 1)?  A LE lookup of (B+D-1) would return
 486  * record 3 because the keys are ordered first by startblock.  An interval
 487  * query would return records 1 and 2 because they both overlap (B+D-1), and
 488  * from that we can pick out record 1 as the appropriate left neighbor.
 489  *
 490  * In the non-overlapped case you can do a LE lookup and decrement the cursor
 491  * because a record's interval must end before the next record.
 492  */
 493
 494 /*
 495  * Return size of the btree block header for this btree instance.
 496  */
 497 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
 498 {
 499         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 500                 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
 501                         return XFS_BTREE_LBLOCK_CRC_LEN;
 502                 return XFS_BTREE_LBLOCK_LEN;
 503         }
 504         if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
 505                 return XFS_BTREE_SBLOCK_CRC_LEN;
 506         return XFS_BTREE_SBLOCK_LEN;
 507 }
 508
 509 /*
 510  * Return size of btree block pointers for this btree instance.
 511  */
 512 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
 513 {
 514         return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
 515                 sizeof(__be64) : sizeof(__be32);
 516 }
 517
 518 /*
 519  * Calculate offset of the n-th record in a btree block.
 520  */
 521 STATIC size_t
 522 xfs_btree_rec_offset(
 523         struct xfs_btree_cur    *cur,
 524         int                     n)
 525 {
 526         return xfs_btree_block_len(cur) +
 527                 (n - 1) * cur->bc_ops->rec_len;
 528 }
 529
 530 /*
 531  * Calculate offset of the n-th key in a btree block.
 532  */
 533 STATIC size_t
 534 xfs_btree_key_offset(
 535         struct xfs_btree_cur    *cur,
 536         int                     n)
 537 {
 538         return xfs_btree_block_len(cur) +
 539                 (n - 1) * cur->bc_ops->key_len;
 540 }
 541
 542 /*
 543  * Calculate offset of the n-th high key in a btree block.
 544  */
 545 STATIC size_t
 546 xfs_btree_high_key_offset(
 547         struct xfs_btree_cur    *cur,
 548         int                     n)
 549 {
 550         return xfs_btree_block_len(cur) +
 551                 (n - 1) * cur->bc_ops->key_len + (cur->bc_ops->key_len / 2);
 552 }
 553
 554 /*
 555  * Calculate offset of the n-th block pointer in a btree block.
 556  */
 557 STATIC size_t
 558 xfs_btree_ptr_offset(
 559         struct xfs_btree_cur    *cur,
 560         int                     n,
 561         int                     level)
 562 {
 563         return xfs_btree_block_len(cur) +
 564                 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
 565                 (n - 1) * xfs_btree_ptr_len(cur);
 566 }
 567
 568 /*
 569  * Return a pointer to the n-th record in the btree block.
 570  */
 571 STATIC union xfs_btree_rec *
 572 xfs_btree_rec_addr(
 573         struct xfs_btree_cur    *cur,
 574         int                     n,
 575         struct xfs_btree_block  *block)
 576 {
 577         return (union xfs_btree_rec *)
 578                 ((char *)block + xfs_btree_rec_offset(cur, n));
 579 }
 580
 581 /*
 582  * Return a pointer to the n-th key in the btree block.
 583  */
 584 STATIC union xfs_btree_key *
 585 xfs_btree_key_addr(
 586         struct xfs_btree_cur    *cur,
 587         int                     n,
 588         struct xfs_btree_block  *block)
 589 {
 590         return (union xfs_btree_key *)
 591                 ((char *)block + xfs_btree_key_offset(cur, n));
 592 }
 593
 594 /*
 595  * Return a pointer to the n-th high key in the btree block.
 596  */
 597 STATIC union xfs_btree_key *
 598 xfs_btree_high_key_addr(
 599         struct xfs_btree_cur    *cur,
 600         int                     n,
 601         struct xfs_btree_block  *block)
 602 {
 603         return (union xfs_btree_key *)
 604                 ((char *)block + xfs_btree_high_key_offset(cur, n));
 605 }
 606
 607 /*
 608  * Return a pointer to the n-th block pointer in the btree block.
 609  */
 610 STATIC union xfs_btree_ptr *
 611 xfs_btree_ptr_addr(
 612         struct xfs_btree_cur    *cur,
 613         int                     n,
 614         struct xfs_btree_block  *block)
 615 {
 616         int                     level = xfs_btree_get_level(block);
 617
 618         ASSERT(block->bb_level != 0);
 619
 620         return (union xfs_btree_ptr *)
 621                 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
 622 }
 623
 624 /*
 625  * Get the root block which is stored in the inode.
 626  *
 627  * For now this btree implementation assumes the btree root is always
 628  * stored in the if_broot field of an inode fork.
 629  */
 630 STATIC struct xfs_btree_block *
 631 xfs_btree_get_iroot(
 632         struct xfs_btree_cur    *cur)
 633 {
 634         struct xfs_ifork        *ifp;
 635
 636         ifp = XFS_IFORK_PTR(cur->bc_private.b.ip, cur->bc_private.b.whichfork);
 637         return (struct xfs_btree_block *)ifp->if_broot;
 638 }
 639
 640 /*
 641  * Retrieve the block pointer from the cursor at the given level.
 642  * This may be an inode btree root or from a buffer.
 643  */
 644 STATIC struct xfs_btree_block *         /* generic btree block pointer */
 645 xfs_btree_get_block(
 646         struct xfs_btree_cur    *cur,   /* btree cursor */
 647         int                     level,  /* level in btree */
 648         struct xfs_buf          **bpp)  /* buffer containing the block */
 649 {
 650         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
 651             (level == cur->bc_nlevels - 1)) {
 652                 *bpp = NULL;
 653                 return xfs_btree_get_iroot(cur);
 654         }
 655
 656         *bpp = cur->bc_bufs[level];
 657         return XFS_BUF_TO_BLOCK(*bpp);
 658 }
 659
 660 /*
 661  * Get a buffer for the block, return it with no data read.
 662  * Long-form addressing.
 663  */
 664 xfs_buf_t *                             /* buffer for fsbno */
 665 xfs_btree_get_bufl(
 666         xfs_mount_t     *mp,            /* file system mount point */
 667         xfs_trans_t     *tp,            /* transaction pointer */
 668         xfs_fsblock_t   fsbno,          /* file system block number */
 669         uint            lock)           /* lock flags for get_buf */
 670 {
 671         xfs_daddr_t             d;              /* real disk block address */
 672
 673         ASSERT(fsbno != NULLFSBLOCK);
 674         d = XFS_FSB_TO_DADDR(mp, fsbno);
 675         return xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
 676 }
 677
 678 /*
 679  * Get a buffer for the block, return it with no data read.
 680  * Short-form addressing.
 681  */
 682 xfs_buf_t *                             /* buffer for agno/agbno */
 683 xfs_btree_get_bufs(
 684         xfs_mount_t     *mp,            /* file system mount point */
 685         xfs_trans_t     *tp,            /* transaction pointer */
 686         xfs_agnumber_t  agno,           /* allocation group number */
 687         xfs_agblock_t   agbno,          /* allocation group block number */
 688         uint            lock)           /* lock flags for get_buf */
 689 {
 690         xfs_daddr_t             d;              /* real disk block address */
 691
 692         ASSERT(agno != NULLAGNUMBER);
 693         ASSERT(agbno != NULLAGBLOCK);
 694         d = XFS_AGB_TO_DADDR(mp, agno, agbno);
 695         return xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
 696 }
 697
 698 /*
 699  * Check for the cursor referring to the last block at the given level.
 700  */
 701 int                                     /* 1=is last block, 0=not last block */
 702 xfs_btree_islastblock(
 703         xfs_btree_cur_t         *cur,   /* btree cursor */
 704         int                     level)  /* level to check */
 705 {
 706         struct xfs_btree_block  *block; /* generic btree block pointer */
 707         xfs_buf_t               *bp;    /* buffer containing block */
 708
 709         block = xfs_btree_get_block(cur, level, &bp);
 710         xfs_btree_check_block(cur, block, level, bp);
 711         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
 712                 return block->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK);
 713         else
 714                 return block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK);
 715 }
 716
 717 /*
 718  * Change the cursor to point to the first record at the given level.
 719  * Other levels are unaffected.
 720  */
 721 STATIC int                              /* success=1, failure=0 */
 722 xfs_btree_firstrec(
 723         xfs_btree_cur_t         *cur,   /* btree cursor */
 724         int                     level)  /* level to change */
 725 {
 726         struct xfs_btree_block  *block; /* generic btree block pointer */
 727         xfs_buf_t               *bp;    /* buffer containing block */
 728
 729         /*
 730          * Get the block pointer for this level.
 731          */
 732         block = xfs_btree_get_block(cur, level, &bp);
 733         xfs_btree_check_block(cur, block, level, bp);
 734         /*
 735          * It's empty, there is no such record.
 736          */
 737         if (!block->bb_numrecs)
 738                 return 0;
 739         /*
 740          * Set the ptr value to 1, that's the first record/key.
 741          */
 742         cur->bc_ptrs[level] = 1;
 743         return 1;
 744 }
 745
 746 /*
 747  * Change the cursor to point to the last record in the current block
 748  * at the given level.  Other levels are unaffected.
 749  */
 750 STATIC int                              /* success=1, failure=0 */
 751 xfs_btree_lastrec(
 752         xfs_btree_cur_t         *cur,   /* btree cursor */
 753         int                     level)  /* level to change */
 754 {
 755         struct xfs_btree_block  *block; /* generic btree block pointer */
 756         xfs_buf_t               *bp;    /* buffer containing block */
 757
 758         /*
 759          * Get the block pointer for this level.
 760          */
 761         block = xfs_btree_get_block(cur, level, &bp);
 762         xfs_btree_check_block(cur, block, level, bp);
 763         /*
 764          * It's empty, there is no such record.
 765          */
 766         if (!block->bb_numrecs)
 767                 return 0;
 768         /*
 769          * Set the ptr value to numrecs, that's the last record/key.
 770          */
 771         cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
 772         return 1;
 773 }
 774
 775 /*
 776  * Compute first and last byte offsets for the fields given.
 777  * Interprets the offsets table, which contains struct field offsets.
 778  */
 779 void
 780 xfs_btree_offsets(
 781         __int64_t       fields,         /* bitmask of fields */
 782         const short     *offsets,       /* table of field offsets */
 783         int             nbits,          /* number of bits to inspect */
 784         int             *first,         /* output: first byte offset */
 785         int             *last)          /* output: last byte offset */
 786 {
 787         int             i;              /* current bit number */
 788         __int64_t       imask;          /* mask for current bit number */
 789
 790         ASSERT(fields != 0);
 791         /*
 792          * Find the lowest bit, so the first byte offset.
 793          */
 794         for (i = 0, imask = 1LL; ; i++, imask <<= 1) {
 795                 if (imask & fields) {
 796                         *first = offsets[i];
 797                         break;
 798                 }
 799         }
 800         /*
 801          * Find the highest bit, so the last byte offset.
 802          */
 803         for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) {
 804                 if (imask & fields) {
 805                         *last = offsets[i + 1] - 1;
 806                         break;
 807                 }
 808         }
 809 }
 810
 811 /*
 812  * Get a buffer for the block, return it read in.
 813  * Long-form addressing.
 814  */
 815 int
 816 xfs_btree_read_bufl(
 817         struct xfs_mount        *mp,            /* file system mount point */
 818         struct xfs_trans        *tp,            /* transaction pointer */
 819         xfs_fsblock_t           fsbno,          /* file system block number */
 820         uint                    lock,           /* lock flags for read_buf */
 821         struct xfs_buf          **bpp,          /* buffer for fsbno */
 822         int                     refval,         /* ref count value for buffer */
 823         const struct xfs_buf_ops *ops)
 824 {
 825         struct xfs_buf          *bp;            /* return value */
 826         xfs_daddr_t             d;              /* real disk block address */
 827         int                     error;
 828
 829         ASSERT(fsbno != NULLFSBLOCK);
 830         d = XFS_FSB_TO_DADDR(mp, fsbno);
 831         error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
 832                                    mp->m_bsize, lock, &bp, ops);
 833         if (error)
 834                 return error;
 835         if (bp)
 836                 xfs_buf_set_ref(bp, refval);
 837         *bpp = bp;
 838         return 0;
 839 }
 840
 841 /*
 842  * Read-ahead the block, don't wait for it, don't return a buffer.
 843  * Long-form addressing.
 844  */
 845 /* ARGSUSED */
 846 void
 847 xfs_btree_reada_bufl(
 848         struct xfs_mount        *mp,            /* file system mount point */
 849         xfs_fsblock_t           fsbno,          /* file system block number */
 850         xfs_extlen_t            count,          /* count of filesystem blocks */
 851         const struct xfs_buf_ops *ops)
 852 {
 853         xfs_daddr_t             d;
 854
 855         ASSERT(fsbno != NULLFSBLOCK);
 856         d = XFS_FSB_TO_DADDR(mp, fsbno);
 857         xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
 858 }
 859
 860 /*
 861  * Read-ahead the block, don't wait for it, don't return a buffer.
 862  * Short-form addressing.
 863  */
 864 /* ARGSUSED */
 865 void
 866 xfs_btree_reada_bufs(
 867         struct xfs_mount        *mp,            /* file system mount point */
 868         xfs_agnumber_t          agno,           /* allocation group number */
 869         xfs_agblock_t           agbno,          /* allocation group block number */
 870         xfs_extlen_t            count,          /* count of filesystem blocks */
 871         const struct xfs_buf_ops *ops)
 872 {
 873         xfs_daddr_t             d;
 874
 875         ASSERT(agno != NULLAGNUMBER);
 876         ASSERT(agbno != NULLAGBLOCK);
 877         d = XFS_AGB_TO_DADDR(mp, agno, agbno);
 878         xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
 879 }
 880
 881 STATIC int
 882 xfs_btree_readahead_lblock(
 883         struct xfs_btree_cur    *cur,
 884         int                     lr,
 885         struct xfs_btree_block  *block)
 886 {
 887         int                     rval = 0;
 888         xfs_fsblock_t           left = be64_to_cpu(block->bb_u.l.bb_leftsib);
 889         xfs_fsblock_t           right = be64_to_cpu(block->bb_u.l.bb_rightsib);
 890
 891         if ((lr & XFS_BTCUR_LEFTRA) && left != NULLFSBLOCK) {
 892                 xfs_btree_reada_bufl(cur->bc_mp, left, 1,
 893                                      cur->bc_ops->buf_ops);
 894                 rval++;
 895         }
 896
 897         if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLFSBLOCK) {
 898                 xfs_btree_reada_bufl(cur->bc_mp, right, 1,
 899                                      cur->bc_ops->buf_ops);
 900                 rval++;
 901         }
 902
 903         return rval;
 904 }
 905
 906 STATIC int
 907 xfs_btree_readahead_sblock(
 908         struct xfs_btree_cur    *cur,
 909         int                     lr,
 910         struct xfs_btree_block *block)
 911 {
 912         int                     rval = 0;
 913         xfs_agblock_t           left = be32_to_cpu(block->bb_u.s.bb_leftsib);
 914         xfs_agblock_t           right = be32_to_cpu(block->bb_u.s.bb_rightsib);
 915
 916
 917         if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
 918                 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
 919                                      left, 1, cur->bc_ops->buf_ops);
 920                 rval++;
 921         }
 922
 923         if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
 924                 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
 925                                      right, 1, cur->bc_ops->buf_ops);
 926                 rval++;
 927         }
 928
 929         return rval;
 930 }
 931
 932 /*
 933  * Read-ahead btree blocks, at the given level.
 934  * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
 935  */
 936 STATIC int
 937 xfs_btree_readahead(
 938         struct xfs_btree_cur    *cur,           /* btree cursor */
 939         int                     lev,            /* level in btree */
 940         int                     lr)             /* left/right bits */
 941 {
 942         struct xfs_btree_block  *block;
 943
 944         /*
 945          * No readahead needed if we are at the root level and the
 946          * btree root is stored in the inode.
 947          */
 948         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
 949             (lev == cur->bc_nlevels - 1))
 950                 return 0;
 951
 952         if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
 953                 return 0;
 954
 955         cur->bc_ra[lev] |= lr;
 956         block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
 957
 958         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
 959                 return xfs_btree_readahead_lblock(cur, lr, block);
 960         return xfs_btree_readahead_sblock(cur, lr, block);
 961 }
 962
 963 STATIC xfs_daddr_t
 964 xfs_btree_ptr_to_daddr(
 965         struct xfs_btree_cur    *cur,
 966         union xfs_btree_ptr     *ptr)
 967 {
 968         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 969                 ASSERT(ptr->l != cpu_to_be64(NULLFSBLOCK));
 970
 971                 return XFS_FSB_TO_DADDR(cur->bc_mp, be64_to_cpu(ptr->l));
 972         } else {
 973                 ASSERT(cur->bc_private.a.agno != NULLAGNUMBER);
 974                 ASSERT(ptr->s != cpu_to_be32(NULLAGBLOCK));
 975
 976                 return XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_private.a.agno,
 977                                         be32_to_cpu(ptr->s));
 978         }
 979 }
 980
 981 /*
 982  * Readahead @count btree blocks at the given @ptr location.
 983  *
 984  * We don't need to care about long or short form btrees here as we have a
 985  * method of converting the ptr directly to a daddr available to us.
 986  */
 987 STATIC void
 988 xfs_btree_readahead_ptr(
 989         struct xfs_btree_cur    *cur,
 990         union xfs_btree_ptr     *ptr,
 991         xfs_extlen_t            count)
 992 {
 993         xfs_buf_readahead(cur->bc_mp->m_ddev_targp,
 994                           xfs_btree_ptr_to_daddr(cur, ptr),
 995                           cur->bc_mp->m_bsize * count, cur->bc_ops->buf_ops);
 996 }
 997
 998 /*
 999  * Set the buffer for level "lev" in the cursor to bp, releasing
1000  * any previous buffer.
1001  */
1002 STATIC void
1003 xfs_btree_setbuf(
1004         xfs_btree_cur_t         *cur,   /* btree cursor */
1005         int                     lev,    /* level in btree */
1006         xfs_buf_t               *bp)    /* new buffer to set */
1007 {
1008         struct xfs_btree_block  *b;     /* btree block */
1009
1010         if (cur->bc_bufs[lev])
1011                 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
1012         cur->bc_bufs[lev] = bp;
1013         cur->bc_ra[lev] = 0;
1014
1015         b = XFS_BUF_TO_BLOCK(bp);
1016         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1017                 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK))
1018                         cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1019                 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK))
1020                         cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1021         } else {
1022                 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
1023                         cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1024                 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
1025                         cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1026         }
1027 }
1028
1029 STATIC int
1030 xfs_btree_ptr_is_null(
1031         struct xfs_btree_cur    *cur,
1032         union xfs_btree_ptr     *ptr)
1033 {
1034         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1035                 return ptr->l == cpu_to_be64(NULLFSBLOCK);
1036         else
1037                 return ptr->s == cpu_to_be32(NULLAGBLOCK);
1038 }
1039
1040 STATIC void
1041 xfs_btree_set_ptr_null(
1042         struct xfs_btree_cur    *cur,
1043         union xfs_btree_ptr     *ptr)
1044 {
1045         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1046                 ptr->l = cpu_to_be64(NULLFSBLOCK);
1047         else
1048                 ptr->s = cpu_to_be32(NULLAGBLOCK);
1049 }
1050
1051 /*
1052  * Get/set/init sibling pointers
1053  */
1054 STATIC void
1055 xfs_btree_get_sibling(
1056         struct xfs_btree_cur    *cur,
1057         struct xfs_btree_block  *block,
1058         union xfs_btree_ptr     *ptr,
1059         int                     lr)
1060 {
1061         ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1062
1063         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1064                 if (lr == XFS_BB_RIGHTSIB)
1065                         ptr->l = block->bb_u.l.bb_rightsib;
1066                 else
1067                         ptr->l = block->bb_u.l.bb_leftsib;
1068         } else {
1069                 if (lr == XFS_BB_RIGHTSIB)
1070                         ptr->s = block->bb_u.s.bb_rightsib;
1071                 else
1072                         ptr->s = block->bb_u.s.bb_leftsib;
1073         }
1074 }
1075
1076 STATIC void
1077 xfs_btree_set_sibling(
1078         struct xfs_btree_cur    *cur,
1079         struct xfs_btree_block  *block,
1080         union xfs_btree_ptr     *ptr,
1081         int                     lr)
1082 {
1083         ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1084
1085         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1086                 if (lr == XFS_BB_RIGHTSIB)
1087                         block->bb_u.l.bb_rightsib = ptr->l;
1088                 else
1089                         block->bb_u.l.bb_leftsib = ptr->l;
1090         } else {
1091                 if (lr == XFS_BB_RIGHTSIB)
1092                         block->bb_u.s.bb_rightsib = ptr->s;
1093                 else
1094                         block->bb_u.s.bb_leftsib = ptr->s;
1095         }
1096 }
1097
1098 void
1099 xfs_btree_init_block_int(
1100         struct xfs_mount        *mp,
1101         struct xfs_btree_block  *buf,
1102         xfs_daddr_t             blkno,
1103         xfs_btnum_t             btnum,
1104         __u16                   level,
1105         __u16                   numrecs,
1106         __u64                   owner,
1107         unsigned int            flags)
1108 {
1109         int                     crc = xfs_sb_version_hascrc(&mp->m_sb);
1110         __u32                   magic = xfs_btree_magic(crc, btnum);
1111
1112         buf->bb_magic = cpu_to_be32(magic);
1113         buf->bb_level = cpu_to_be16(level);
1114         buf->bb_numrecs = cpu_to_be16(numrecs);
1115
1116         if (flags & XFS_BTREE_LONG_PTRS) {
1117                 buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK);
1118                 buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK);
1119                 if (crc) {
1120                         buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
1121                         buf->bb_u.l.bb_owner = cpu_to_be64(owner);
1122                         uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid);
1123                         buf->bb_u.l.bb_pad = 0;
1124                         buf->bb_u.l.bb_lsn = 0;
1125                 }
1126         } else {
1127                 /* owner is a 32 bit value on short blocks */
1128                 __u32 __owner = (__u32)owner;
1129
1130                 buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1131                 buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1132                 if (crc) {
1133                         buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
1134                         buf->bb_u.s.bb_owner = cpu_to_be32(__owner);
1135                         uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid);
1136                         buf->bb_u.s.bb_lsn = 0;
1137                 }
1138         }
1139 }
1140
1141 void
1142 xfs_btree_init_block(
1143         struct xfs_mount *mp,
1144         struct xfs_buf  *bp,
1145         xfs_btnum_t     btnum,
1146         __u16           level,
1147         __u16           numrecs,
1148         __u64           owner,
1149         unsigned int    flags)
1150 {
1151         xfs_btree_init_block_int(mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1152                                  btnum, level, numrecs, owner, flags);
1153 }
1154
1155 STATIC void
1156 xfs_btree_init_block_cur(
1157         struct xfs_btree_cur    *cur,
1158         struct xfs_buf          *bp,
1159         int                     level,
1160         int                     numrecs)
1161 {
1162         __u64                   owner;
1163
1164         /*
1165          * we can pull the owner from the cursor right now as the different
1166          * owners align directly with the pointer size of the btree. This may
1167          * change in future, but is safe for current users of the generic btree
1168          * code.
1169          */
1170         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1171                 owner = cur->bc_private.b.ip->i_ino;
1172         else
1173                 owner = cur->bc_private.a.agno;
1174
1175         xfs_btree_init_block_int(cur->bc_mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1176                                  cur->bc_btnum, level, numrecs,
1177                                  owner, cur->bc_flags);
1178 }
1179
1180 /*
1181  * Return true if ptr is the last record in the btree and
1182  * we need to track updates to this record.  The decision
1183  * will be further refined in the update_lastrec method.
1184  */
1185 STATIC int
1186 xfs_btree_is_lastrec(
1187         struct xfs_btree_cur    *cur,
1188         struct xfs_btree_block  *block,
1189         int                     level)
1190 {
1191         union xfs_btree_ptr     ptr;
1192
1193         if (level > 0)
1194                 return 0;
1195         if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
1196                 return 0;
1197
1198         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1199         if (!xfs_btree_ptr_is_null(cur, &ptr))
1200                 return 0;
1201         return 1;
1202 }
1203
1204 STATIC void
1205 xfs_btree_buf_to_ptr(
1206         struct xfs_btree_cur    *cur,
1207         struct xfs_buf          *bp,
1208         union xfs_btree_ptr     *ptr)
1209 {
1210         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1211                 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1212                                         XFS_BUF_ADDR(bp)));
1213         else {
1214                 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
1215                                         XFS_BUF_ADDR(bp)));
1216         }
1217 }
1218
1219 STATIC void
1220 xfs_btree_set_refs(
1221         struct xfs_btree_cur    *cur,
1222         struct xfs_buf          *bp)
1223 {
1224         switch (cur->bc_btnum) {
1225         case XFS_BTNUM_BNO:
1226         case XFS_BTNUM_CNT:
1227                 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
1228                 break;
1229         case XFS_BTNUM_INO:
1230         case XFS_BTNUM_FINO:
1231                 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
1232                 break;
1233         case XFS_BTNUM_BMAP:
1234                 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
1235                 break;
1236         case XFS_BTNUM_RMAP:
1237                 xfs_buf_set_ref(bp, XFS_RMAP_BTREE_REF);
1238                 break;
1239         case XFS_BTNUM_REFC:
1240                 xfs_buf_set_ref(bp, XFS_REFC_BTREE_REF);
1241                 break;
1242         default:
1243                 ASSERT(0);
1244         }
1245 }
1246
1247 STATIC int
1248 xfs_btree_get_buf_block(
1249         struct xfs_btree_cur    *cur,
1250         union xfs_btree_ptr     *ptr,
1251         int                     flags,
1252         struct xfs_btree_block  **block,
1253         struct xfs_buf          **bpp)
1254 {
1255         struct xfs_mount        *mp = cur->bc_mp;
1256         xfs_daddr_t             d;
1257
1258         /* need to sort out how callers deal with failures first */
1259         ASSERT(!(flags & XBF_TRYLOCK));
1260
1261         d = xfs_btree_ptr_to_daddr(cur, ptr);
1262         *bpp = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d,
1263                                  mp->m_bsize, flags);
1264
1265         if (!*bpp)
1266                 return -ENOMEM;
1267
1268         (*bpp)->b_ops = cur->bc_ops->buf_ops;
1269         *block = XFS_BUF_TO_BLOCK(*bpp);
1270         return 0;
1271 }
1272
1273 /*
1274  * Read in the buffer at the given ptr and return the buffer and
1275  * the block pointer within the buffer.
1276  */
1277 STATIC int
1278 xfs_btree_read_buf_block(
1279         struct xfs_btree_cur    *cur,
1280         union xfs_btree_ptr     *ptr,
1281         int                     flags,
1282         struct xfs_btree_block  **block,
1283         struct xfs_buf          **bpp)
1284 {
1285         struct xfs_mount        *mp = cur->bc_mp;
1286         xfs_daddr_t             d;
1287         int                     error;
1288
1289         /* need to sort out how callers deal with failures first */
1290         ASSERT(!(flags & XBF_TRYLOCK));
1291
1292         d = xfs_btree_ptr_to_daddr(cur, ptr);
1293         error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
1294                                    mp->m_bsize, flags, bpp,
1295                                    cur->bc_ops->buf_ops);
1296         if (error)
1297                 return error;
1298
1299         xfs_btree_set_refs(cur, *bpp);
1300         *block = XFS_BUF_TO_BLOCK(*bpp);
1301         return 0;
1302 }
1303
1304 /*
1305  * Copy keys from one btree block to another.
1306  */
1307 STATIC void
1308 xfs_btree_copy_keys(
1309         struct xfs_btree_cur    *cur,
1310         union xfs_btree_key     *dst_key,
1311         union xfs_btree_key     *src_key,
1312         int                     numkeys)
1313 {
1314         ASSERT(numkeys >= 0);
1315         memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1316 }
1317
1318 /*
1319  * Copy records from one btree block to another.
1320  */
1321 STATIC void
1322 xfs_btree_copy_recs(
1323         struct xfs_btree_cur    *cur,
1324         union xfs_btree_rec     *dst_rec,
1325         union xfs_btree_rec     *src_rec,
1326         int                     numrecs)
1327 {
1328         ASSERT(numrecs >= 0);
1329         memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1330 }
1331
1332 /*
1333  * Copy block pointers from one btree block to another.
1334  */
1335 STATIC void
1336 xfs_btree_copy_ptrs(
1337         struct xfs_btree_cur    *cur,
1338         union xfs_btree_ptr     *dst_ptr,
1339         union xfs_btree_ptr     *src_ptr,
1340         int                     numptrs)
1341 {
1342         ASSERT(numptrs >= 0);
1343         memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1344 }
1345
1346 /*
1347  * Shift keys one index left/right inside a single btree block.
1348  */
1349 STATIC void
1350 xfs_btree_shift_keys(
1351         struct xfs_btree_cur    *cur,
1352         union xfs_btree_key     *key,
1353         int                     dir,
1354         int                     numkeys)
1355 {
1356         char                    *dst_key;
1357
1358         ASSERT(numkeys >= 0);
1359         ASSERT(dir == 1 || dir == -1);
1360
1361         dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1362         memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1363 }
1364
1365 /*
1366  * Shift records one index left/right inside a single btree block.
1367  */
1368 STATIC void
1369 xfs_btree_shift_recs(
1370         struct xfs_btree_cur    *cur,
1371         union xfs_btree_rec     *rec,
1372         int                     dir,
1373         int                     numrecs)
1374 {
1375         char                    *dst_rec;
1376
1377         ASSERT(numrecs >= 0);
1378         ASSERT(dir == 1 || dir == -1);
1379
1380         dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1381         memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1382 }
1383
1384 /*
1385  * Shift block pointers one index left/right inside a single btree block.
1386  */
1387 STATIC void
1388 xfs_btree_shift_ptrs(
1389         struct xfs_btree_cur    *cur,
1390         union xfs_btree_ptr     *ptr,
1391         int                     dir,
1392         int                     numptrs)
1393 {
1394         char                    *dst_ptr;
1395
1396         ASSERT(numptrs >= 0);
1397         ASSERT(dir == 1 || dir == -1);
1398
1399         dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1400         memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1401 }
1402
1403 /*
1404  * Log key values from the btree block.
1405  */
1406 STATIC void
1407 xfs_btree_log_keys(
1408         struct xfs_btree_cur    *cur,
1409         struct xfs_buf          *bp,
1410         int                     first,
1411         int                     last)
1412 {
1413         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1414         XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1415
1416         if (bp) {
1417                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1418                 xfs_trans_log_buf(cur->bc_tp, bp,
1419                                   xfs_btree_key_offset(cur, first),
1420                                   xfs_btree_key_offset(cur, last + 1) - 1);
1421         } else {
1422                 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1423                                 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1424         }
1425
1426         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1427 }
1428
1429 /*
1430  * Log record values from the btree block.
1431  */
1432 void
1433 xfs_btree_log_recs(
1434         struct xfs_btree_cur    *cur,
1435         struct xfs_buf          *bp,
1436         int                     first,
1437         int                     last)
1438 {
1439         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1440         XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1441
1442         xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1443         xfs_trans_log_buf(cur->bc_tp, bp,
1444                           xfs_btree_rec_offset(cur, first),
1445                           xfs_btree_rec_offset(cur, last + 1) - 1);
1446
1447         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1448 }
1449
1450 /*
1451  * Log block pointer fields from a btree block (nonleaf).
1452  */
1453 STATIC void
1454 xfs_btree_log_ptrs(
1455         struct xfs_btree_cur    *cur,   /* btree cursor */
1456         struct xfs_buf          *bp,    /* buffer containing btree block */
1457         int                     first,  /* index of first pointer to log */
1458         int                     last)   /* index of last pointer to log */
1459 {
1460         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1461         XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1462
1463         if (bp) {
1464                 struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
1465                 int                     level = xfs_btree_get_level(block);
1466
1467                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1468                 xfs_trans_log_buf(cur->bc_tp, bp,
1469                                 xfs_btree_ptr_offset(cur, first, level),
1470                                 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1471         } else {
1472                 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1473                         xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1474         }
1475
1476         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1477 }
1478
1479 /*
1480  * Log fields from a btree block header.
1481  */
1482 void
1483 xfs_btree_log_block(
1484         struct xfs_btree_cur    *cur,   /* btree cursor */
1485         struct xfs_buf          *bp,    /* buffer containing btree block */
1486         int                     fields) /* mask of fields: XFS_BB_... */
1487 {
1488         int                     first;  /* first byte offset logged */
1489         int                     last;   /* last byte offset logged */
1490         static const short      soffsets[] = {  /* table of offsets (short) */
1491                 offsetof(struct xfs_btree_block, bb_magic),
1492                 offsetof(struct xfs_btree_block, bb_level),
1493                 offsetof(struct xfs_btree_block, bb_numrecs),
1494                 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1495                 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1496                 offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1497                 offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1498                 offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1499                 offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1500                 offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1501                 XFS_BTREE_SBLOCK_CRC_LEN
1502         };
1503         static const short      loffsets[] = {  /* table of offsets (long) */
1504                 offsetof(struct xfs_btree_block, bb_magic),
1505                 offsetof(struct xfs_btree_block, bb_level),
1506                 offsetof(struct xfs_btree_block, bb_numrecs),
1507                 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1508                 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1509                 offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1510                 offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1511                 offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1512                 offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1513                 offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1514                 offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1515                 XFS_BTREE_LBLOCK_CRC_LEN
1516         };
1517
1518         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1519         XFS_BTREE_TRACE_ARGBI(cur, bp, fields);
1520
1521         if (bp) {
1522                 int nbits;
1523
1524                 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
1525                         /*
1526                          * We don't log the CRC when updating a btree
1527                          * block but instead recreate it during log
1528                          * recovery.  As the log buffers have checksums
1529                          * of their own this is safe and avoids logging a crc
1530                          * update in a lot of places.
1531                          */
1532                         if (fields == XFS_BB_ALL_BITS)
1533                                 fields = XFS_BB_ALL_BITS_CRC;
1534                         nbits = XFS_BB_NUM_BITS_CRC;
1535                 } else {
1536                         nbits = XFS_BB_NUM_BITS;
1537                 }
1538                 xfs_btree_offsets(fields,
1539                                   (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1540                                         loffsets : soffsets,
1541                                   nbits, &first, &last);
1542                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1543                 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1544         } else {
1545                 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1546                         xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1547         }
1548
1549         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1550 }
1551
1552 /*
1553  * Increment cursor by one record at the level.
1554  * For nonzero levels the leaf-ward information is untouched.
1555  */
1556 int                                             /* error */
1557 xfs_btree_increment(
1558         struct xfs_btree_cur    *cur,
1559         int                     level,
1560         int                     *stat)          /* success/failure */
1561 {
1562         struct xfs_btree_block  *block;
1563         union xfs_btree_ptr     ptr;
1564         struct xfs_buf          *bp;
1565         int                     error;          /* error return value */
1566         int                     lev;
1567
1568         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1569         XFS_BTREE_TRACE_ARGI(cur, level);
1570
1571         ASSERT(level < cur->bc_nlevels);
1572
1573         /* Read-ahead to the right at this level. */
1574         xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1575
1576         /* Get a pointer to the btree block. */
1577         block = xfs_btree_get_block(cur, level, &bp);
1578
1579 #ifdef DEBUG
1580         error = xfs_btree_check_block(cur, block, level, bp);
1581         if (error)
1582                 goto error0;
1583 #endif
1584
1585         /* We're done if we remain in the block after the increment. */
1586         if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
1587                 goto out1;
1588
1589         /* Fail if we just went off the right edge of the tree. */
1590         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1591         if (xfs_btree_ptr_is_null(cur, &ptr))
1592                 goto out0;
1593
1594         XFS_BTREE_STATS_INC(cur, increment);
1595
1596         /*
1597          * March up the tree incrementing pointers.
1598          * Stop when we don't go off the right edge of a block.
1599          */
1600         for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1601                 block = xfs_btree_get_block(cur, lev, &bp);
1602
1603 #ifdef DEBUG
1604                 error = xfs_btree_check_block(cur, block, lev, bp);
1605                 if (error)
1606                         goto error0;
1607 #endif
1608
1609                 if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
1610                         break;
1611
1612                 /* Read-ahead the right block for the next loop. */
1613                 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1614         }
1615
1616         /*
1617          * If we went off the root then we are either seriously
1618          * confused or have the tree root in an inode.
1619          */
1620         if (lev == cur->bc_nlevels) {
1621                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1622                         goto out0;
1623                 ASSERT(0);
1624                 error = -EFSCORRUPTED;
1625                 goto error0;
1626         }
1627         ASSERT(lev < cur->bc_nlevels);
1628
1629         /*
1630          * Now walk back down the tree, fixing up the cursor's buffer
1631          * pointers and key numbers.
1632          */
1633         for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1634                 union xfs_btree_ptr     *ptrp;
1635
1636                 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1637                 --lev;
1638                 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1639                 if (error)
1640                         goto error0;
1641
1642                 xfs_btree_setbuf(cur, lev, bp);
1643                 cur->bc_ptrs[lev] = 1;
1644         }
1645 out1:
1646         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1647         *stat = 1;
1648         return 0;
1649
1650 out0:
1651         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1652         *stat = 0;
1653         return 0;
1654
1655 error0:
1656         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1657         return error;
1658 }
1659
1660 /*
1661  * Decrement cursor by one record at the level.
1662  * For nonzero levels the leaf-ward information is untouched.
1663  */
1664 int                                             /* error */
1665 xfs_btree_decrement(
1666         struct xfs_btree_cur    *cur,
1667         int                     level,
1668         int                     *stat)          /* success/failure */
1669 {
1670         struct xfs_btree_block  *block;
1671         xfs_buf_t               *bp;
1672         int                     error;          /* error return value */
1673         int                     lev;
1674         union xfs_btree_ptr     ptr;
1675
1676         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1677         XFS_BTREE_TRACE_ARGI(cur, level);
1678
1679         ASSERT(level < cur->bc_nlevels);
1680
1681         /* Read-ahead to the left at this level. */
1682         xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1683
1684         /* We're done if we remain in the block after the decrement. */
1685         if (--cur->bc_ptrs[level] > 0)
1686                 goto out1;
1687
1688         /* Get a pointer to the btree block. */
1689         block = xfs_btree_get_block(cur, level, &bp);
1690
1691 #ifdef DEBUG
1692         error = xfs_btree_check_block(cur, block, level, bp);
1693         if (error)
1694                 goto error0;
1695 #endif
1696
1697         /* Fail if we just went off the left edge of the tree. */
1698         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1699         if (xfs_btree_ptr_is_null(cur, &ptr))
1700                 goto out0;
1701
1702         XFS_BTREE_STATS_INC(cur, decrement);
1703
1704         /*
1705          * March up the tree decrementing pointers.
1706          * Stop when we don't go off the left edge of a block.
1707          */
1708         for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1709                 if (--cur->bc_ptrs[lev] > 0)
1710                         break;
1711                 /* Read-ahead the left block for the next loop. */
1712                 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1713         }
1714
1715         /*
1716          * If we went off the root then we are seriously confused.
1717          * or the root of the tree is in an inode.
1718          */
1719         if (lev == cur->bc_nlevels) {
1720                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1721                         goto out0;
1722                 ASSERT(0);
1723                 error = -EFSCORRUPTED;
1724                 goto error0;
1725         }
1726         ASSERT(lev < cur->bc_nlevels);
1727
1728         /*
1729          * Now walk back down the tree, fixing up the cursor's buffer
1730          * pointers and key numbers.
1731          */
1732         for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1733                 union xfs_btree_ptr     *ptrp;
1734
1735                 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1736                 --lev;
1737                 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1738                 if (error)
1739                         goto error0;
1740                 xfs_btree_setbuf(cur, lev, bp);
1741                 cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
1742         }
1743 out1:
1744         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1745         *stat = 1;
1746         return 0;
1747
1748 out0:
1749         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1750         *stat = 0;
1751         return 0;
1752
1753 error0:
1754         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1755         return error;
1756 }
1757
1758 STATIC int
1759 xfs_btree_lookup_get_block(
1760         struct xfs_btree_cur    *cur,   /* btree cursor */
1761         int                     level,  /* level in the btree */
1762         union xfs_btree_ptr     *pp,    /* ptr to btree block */
1763         struct xfs_btree_block  **blkp) /* return btree block */
1764 {
1765         struct xfs_buf          *bp;    /* buffer pointer for btree block */
1766         int                     error = 0;
1767
1768         /* special case the root block if in an inode */
1769         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1770             (level == cur->bc_nlevels - 1)) {
1771                 *blkp = xfs_btree_get_iroot(cur);
1772                 return 0;
1773         }
1774
1775         /*
1776          * If the old buffer at this level for the disk address we are
1777          * looking for re-use it.
1778          *
1779          * Otherwise throw it away and get a new one.
1780          */
1781         bp = cur->bc_bufs[level];
1782         if (bp && XFS_BUF_ADDR(bp) == xfs_btree_ptr_to_daddr(cur, pp)) {
1783                 *blkp = XFS_BUF_TO_BLOCK(bp);
1784                 return 0;
1785         }
1786
1787         error = xfs_btree_read_buf_block(cur, pp, 0, blkp, &bp);
1788         if (error)
1789                 return error;
1790
1791         /* Check the inode owner since the verifiers don't. */
1792         if (xfs_sb_version_hascrc(&cur->bc_mp->m_sb) &&
1793             (cur->bc_flags & XFS_BTREE_LONG_PTRS) &&
1794             be64_to_cpu((*blkp)->bb_u.l.bb_owner) !=
1795                         cur->bc_private.b.ip->i_ino)
1796                 goto out_bad;
1797
1798         /* Did we get the level we were looking for? */
1799         if (be16_to_cpu((*blkp)->bb_level) != level)
1800                 goto out_bad;
1801
1802         /* Check that internal nodes have at least one record. */
1803         if (level != 0 && be16_to_cpu((*blkp)->bb_numrecs) == 0)
1804                 goto out_bad;
1805
1806         xfs_btree_setbuf(cur, level, bp);
1807         return 0;
1808
1809 out_bad:
1810         *blkp = NULL;
1811         xfs_trans_brelse(cur->bc_tp, bp);
1812         return -EFSCORRUPTED;
1813 }
1814
1815 /*
1816  * Get current search key.  For level 0 we don't actually have a key
1817  * structure so we make one up from the record.  For all other levels
1818  * we just return the right key.
1819  */
1820 STATIC union xfs_btree_key *
1821 xfs_lookup_get_search_key(
1822         struct xfs_btree_cur    *cur,
1823         int                     level,
1824         int                     keyno,
1825         struct xfs_btree_block  *block,
1826         union xfs_btree_key     *kp)
1827 {
1828         if (level == 0) {
1829                 cur->bc_ops->init_key_from_rec(kp,
1830                                 xfs_btree_rec_addr(cur, keyno, block));
1831                 return kp;
1832         }
1833
1834         return xfs_btree_key_addr(cur, keyno, block);
1835 }
1836
1837 /*
1838  * Lookup the record.  The cursor is made to point to it, based on dir.
1839  * stat is set to 0 if can't find any such record, 1 for success.
1840  */
1841 int                                     /* error */
1842 xfs_btree_lookup(
1843         struct xfs_btree_cur    *cur,   /* btree cursor */
1844         xfs_lookup_t            dir,    /* <=, ==, or >= */
1845         int                     *stat)  /* success/failure */
1846 {
1847         struct xfs_btree_block  *block; /* current btree block */
1848         __int64_t               diff;   /* difference for the current key */
1849         int                     error;  /* error return value */
1850         int                     keyno;  /* current key number */
1851         int                     level;  /* level in the btree */
1852         union xfs_btree_ptr     *pp;    /* ptr to btree block */
1853         union xfs_btree_ptr     ptr;    /* ptr to btree block */
1854
1855         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1856         XFS_BTREE_TRACE_ARGI(cur, dir);
1857
1858         XFS_BTREE_STATS_INC(cur, lookup);
1859
1860         /* No such thing as a zero-level tree. */
1861         if (cur->bc_nlevels == 0)
1862                 return -EFSCORRUPTED;
1863
1864         block = NULL;
1865         keyno = 0;
1866
1867         /* initialise start pointer from cursor */
1868         cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1869         pp = &ptr;
1870
1871         /*
1872          * Iterate over each level in the btree, starting at the root.
1873          * For each level above the leaves, find the key we need, based
1874          * on the lookup record, then follow the corresponding block
1875          * pointer down to the next level.
1876          */
1877         for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1878                 /* Get the block we need to do the lookup on. */
1879                 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1880                 if (error)
1881                         goto error0;
1882
1883                 if (diff == 0) {
1884                         /*
1885                          * If we already had a key match at a higher level, we
1886                          * know we need to use the first entry in this block.
1887                          */
1888                         keyno = 1;
1889                 } else {
1890                         /* Otherwise search this block. Do a binary search. */
1891
1892                         int     high;   /* high entry number */
1893                         int     low;    /* low entry number */
1894
1895                         /* Set low and high entry numbers, 1-based. */
1896                         low = 1;
1897                         high = xfs_btree_get_numrecs(block);
1898                         if (!high) {
1899                                 /* Block is empty, must be an empty leaf. */
1900                                 ASSERT(level == 0 && cur->bc_nlevels == 1);
1901
1902                                 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1903                                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1904                                 *stat = 0;
1905                                 return 0;
1906                         }
1907
1908                         /* Binary search the block. */
1909                         while (low <= high) {
1910                                 union xfs_btree_key     key;
1911                                 union xfs_btree_key     *kp;
1912
1913                                 XFS_BTREE_STATS_INC(cur, compare);
1914
1915                                 /* keyno is average of low and high. */
1916                                 keyno = (low + high) >> 1;
1917
1918                                 /* Get current search key */
1919                                 kp = xfs_lookup_get_search_key(cur, level,
1920                                                 keyno, block, &key);
1921
1922                                 /*
1923                                  * Compute difference to get next direction:
1924                                  *  - less than, move right
1925                                  *  - greater than, move left
1926                                  *  - equal, we're done
1927                                  */
1928                                 diff = cur->bc_ops->key_diff(cur, kp);
1929                                 if (diff < 0)
1930                                         low = keyno + 1;
1931                                 else if (diff > 0)
1932                                         high = keyno - 1;
1933                                 else
1934                                         break;
1935                         }
1936                 }
1937
1938                 /*
1939                  * If there are more levels, set up for the next level
1940                  * by getting the block number and filling in the cursor.
1941                  */
1942                 if (level > 0) {
1943                         /*
1944                          * If we moved left, need the previous key number,
1945                          * unless there isn't one.
1946                          */
1947                         if (diff > 0 && --keyno < 1)
1948                                 keyno = 1;
1949                         pp = xfs_btree_ptr_addr(cur, keyno, block);
1950
1951 #ifdef DEBUG
1952                         error = xfs_btree_check_ptr(cur, pp, 0, level);
1953                         if (error)
1954                                 goto error0;
1955 #endif
1956                         cur->bc_ptrs[level] = keyno;
1957                 }
1958         }
1959
1960         /* Done with the search. See if we need to adjust the results. */
1961         if (dir != XFS_LOOKUP_LE && diff < 0) {
1962                 keyno++;
1963                 /*
1964                  * If ge search and we went off the end of the block, but it's
1965                  * not the last block, we're in the wrong block.
1966                  */
1967                 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1968                 if (dir == XFS_LOOKUP_GE &&
1969                     keyno > xfs_btree_get_numrecs(block) &&
1970                     !xfs_btree_ptr_is_null(cur, &ptr)) {
1971                         int     i;
1972
1973                         cur->bc_ptrs[0] = keyno;
1974                         error = xfs_btree_increment(cur, 0, &i);
1975                         if (error)
1976                                 goto error0;
1977                         XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1978                         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1979                         *stat = 1;
1980                         return 0;
1981                 }
1982         } else if (dir == XFS_LOOKUP_LE && diff > 0)
1983                 keyno--;
1984         cur->bc_ptrs[0] = keyno;
1985
1986         /* Return if we succeeded or not. */
1987         if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
1988                 *stat = 0;
1989         else if (dir != XFS_LOOKUP_EQ || diff == 0)
1990                 *stat = 1;
1991         else
1992                 *stat = 0;
1993         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1994         return 0;
1995
1996 error0:
1997         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1998         return error;
1999 }
2000
2001 /* Find the high key storage area from a regular key. */
2002 STATIC union xfs_btree_key *
2003 xfs_btree_high_key_from_key(
2004         struct xfs_btree_cur    *cur,
2005         union xfs_btree_key     *key)
2006 {
2007         ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2008         return (union xfs_btree_key *)((char *)key +
2009                         (cur->bc_ops->key_len / 2));
2010 }
2011
2012 /* Determine the low (and high if overlapped) keys of a leaf block */
2013 STATIC void
2014 xfs_btree_get_leaf_keys(
2015         struct xfs_btree_cur    *cur,
2016         struct xfs_btree_block  *block,
2017         union xfs_btree_key     *key)
2018 {
2019         union xfs_btree_key     max_hkey;
2020         union xfs_btree_key     hkey;
2021         union xfs_btree_rec     *rec;
2022         union xfs_btree_key     *high;
2023         int                     n;
2024
2025         rec = xfs_btree_rec_addr(cur, 1, block);
2026         cur->bc_ops->init_key_from_rec(key, rec);
2027
2028         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2029
2030                 cur->bc_ops->init_high_key_from_rec(&max_hkey, rec);
2031                 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2032                         rec = xfs_btree_rec_addr(cur, n, block);
2033                         cur->bc_ops->init_high_key_from_rec(&hkey, rec);
2034                         if (cur->bc_ops->diff_two_keys(cur, &hkey, &max_hkey)
2035                                         > 0)
2036                                 max_hkey = hkey;
2037                 }
2038
2039                 high = xfs_btree_high_key_from_key(cur, key);
2040                 memcpy(high, &max_hkey, cur->bc_ops->key_len / 2);
2041         }
2042 }
2043
2044 /* Determine the low (and high if overlapped) keys of a node block */
2045 STATIC void
2046 xfs_btree_get_node_keys(
2047         struct xfs_btree_cur    *cur,
2048         struct xfs_btree_block  *block,
2049         union xfs_btree_key     *key)
2050 {
2051         union xfs_btree_key     *hkey;
2052         union xfs_btree_key     *max_hkey;
2053         union xfs_btree_key     *high;
2054         int                     n;
2055
2056         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2057                 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2058                                 cur->bc_ops->key_len / 2);
2059
2060                 max_hkey = xfs_btree_high_key_addr(cur, 1, block);
2061                 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2062                         hkey = xfs_btree_high_key_addr(cur, n, block);
2063                         if (cur->bc_ops->diff_two_keys(cur, hkey, max_hkey) > 0)
2064                                 max_hkey = hkey;
2065                 }
2066
2067                 high = xfs_btree_high_key_from_key(cur, key);
2068                 memcpy(high, max_hkey, cur->bc_ops->key_len / 2);
2069         } else {
2070                 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2071                                 cur->bc_ops->key_len);
2072         }
2073 }
2074
2075 /* Derive the keys for any btree block. */
2076 STATIC void
2077 xfs_btree_get_keys(
2078         struct xfs_btree_cur    *cur,
2079         struct xfs_btree_block  *block,
2080         union xfs_btree_key     *key)
2081 {
2082         if (be16_to_cpu(block->bb_level) == 0)
2083                 xfs_btree_get_leaf_keys(cur, block, key);
2084         else
2085                 xfs_btree_get_node_keys(cur, block, key);
2086 }
2087
2088 /*
2089  * Decide if we need to update the parent keys of a btree block.  For
2090  * a standard btree this is only necessary if we're updating the first
2091  * record/key.  For an overlapping btree, we must always update the
2092  * keys because the highest key can be in any of the records or keys
2093  * in the block.
2094  */
2095 static inline bool
2096 xfs_btree_needs_key_update(
2097         struct xfs_btree_cur    *cur,
2098         int                     ptr)
2099 {
2100         return (cur->bc_flags & XFS_BTREE_OVERLAPPING) || ptr == 1;
2101 }
2102
2103 /*
2104  * Update the low and high parent keys of the given level, progressing
2105  * towards the root.  If force_all is false, stop if the keys for a given
2106  * level do not need updating.
2107  */
2108 STATIC int
2109 __xfs_btree_updkeys(
2110         struct xfs_btree_cur    *cur,
2111         int                     level,
2112         struct xfs_btree_block  *block,
2113         struct xfs_buf          *bp0,
2114         bool                    force_all)
2115 {
2116         union xfs_btree_key     key;    /* keys from current level */
2117         union xfs_btree_key     *lkey;  /* keys from the next level up */
2118         union xfs_btree_key     *hkey;
2119         union xfs_btree_key     *nlkey; /* keys from the next level up */
2120         union xfs_btree_key     *nhkey;
2121         struct xfs_buf          *bp;
2122         int                     ptr;
2123
2124         ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2125
2126         /* Exit if there aren't any parent levels to update. */
2127         if (level + 1 >= cur->bc_nlevels)
2128                 return 0;
2129
2130         trace_xfs_btree_updkeys(cur, level, bp0);
2131
2132         lkey = &key;
2133         hkey = xfs_btree_high_key_from_key(cur, lkey);
2134         xfs_btree_get_keys(cur, block, lkey);
2135         for (level++; level < cur->bc_nlevels; level++) {
2136 #ifdef DEBUG
2137                 int             error;
2138 #endif
2139                 block = xfs_btree_get_block(cur, level, &bp);
2140                 trace_xfs_btree_updkeys(cur, level, bp);
2141 #ifdef DEBUG
2142                 error = xfs_btree_check_block(cur, block, level, bp);
2143                 if (error) {
2144                         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2145                         return error;
2146                 }
2147 #endif
2148                 ptr = cur->bc_ptrs[level];
2149                 nlkey = xfs_btree_key_addr(cur, ptr, block);
2150                 nhkey = xfs_btree_high_key_addr(cur, ptr, block);
2151                 if (!force_all &&
2152                     !(cur->bc_ops->diff_two_keys(cur, nlkey, lkey) != 0 ||
2153                       cur->bc_ops->diff_two_keys(cur, nhkey, hkey) != 0))
2154                         break;
2155                 xfs_btree_copy_keys(cur, nlkey, lkey, 1);
2156                 xfs_btree_log_keys(cur, bp, ptr, ptr);
2157                 if (level + 1 >= cur->bc_nlevels)
2158                         break;
2159                 xfs_btree_get_node_keys(cur, block, lkey);
2160         }
2161
2162         return 0;
2163 }
2164
2165 /* Update all the keys from some level in cursor back to the root. */
2166 STATIC int
2167 xfs_btree_updkeys_force(
2168         struct xfs_btree_cur    *cur,
2169         int                     level)
2170 {
2171         struct xfs_buf          *bp;
2172         struct xfs_btree_block  *block;
2173
2174         block = xfs_btree_get_block(cur, level, &bp);
2175         return __xfs_btree_updkeys(cur, level, block, bp, true);
2176 }
2177
2178 /*
2179  * Update the parent keys of the given level, progressing towards the root.
2180  */
2181 STATIC int
2182 xfs_btree_update_keys(
2183         struct xfs_btree_cur    *cur,
2184         int                     level)
2185 {
2186         struct xfs_btree_block  *block;
2187         struct xfs_buf          *bp;
2188         union xfs_btree_key     *kp;
2189         union xfs_btree_key     key;
2190         int                     ptr;
2191
2192         ASSERT(level >= 0);
2193
2194         block = xfs_btree_get_block(cur, level, &bp);
2195         if (cur->bc_flags & XFS_BTREE_OVERLAPPING)
2196                 return __xfs_btree_updkeys(cur, level, block, bp, false);
2197
2198         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2199         XFS_BTREE_TRACE_ARGIK(cur, level, keyp);
2200
2201         /*
2202          * Go up the tree from this level toward the root.
2203          * At each level, update the key value to the value input.
2204          * Stop when we reach a level where the cursor isn't pointing
2205          * at the first entry in the block.
2206          */
2207         xfs_btree_get_keys(cur, block, &key);
2208         for (level++, ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
2209 #ifdef DEBUG
2210                 int             error;
2211 #endif
2212                 block = xfs_btree_get_block(cur, level, &bp);
2213 #ifdef DEBUG
2214                 error = xfs_btree_check_block(cur, block, level, bp);
2215                 if (error) {
2216                         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2217                         return error;
2218                 }
2219 #endif
2220                 ptr = cur->bc_ptrs[level];
2221                 kp = xfs_btree_key_addr(cur, ptr, block);
2222                 xfs_btree_copy_keys(cur, kp, &key, 1);
2223                 xfs_btree_log_keys(cur, bp, ptr, ptr);
2224         }
2225
2226         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2227         return 0;
2228 }
2229
2230 /*
2231  * Update the record referred to by cur to the value in the
2232  * given record. This either works (return 0) or gets an
2233  * EFSCORRUPTED error.
2234  */
2235 int
2236 xfs_btree_update(
2237         struct xfs_btree_cur    *cur,
2238         union xfs_btree_rec     *rec)
2239 {
2240         struct xfs_btree_block  *block;
2241         struct xfs_buf          *bp;
2242         int                     error;
2243         int                     ptr;
2244         union xfs_btree_rec     *rp;
2245
2246         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2247         XFS_BTREE_TRACE_ARGR(cur, rec);
2248
2249         /* Pick up the current block. */
2250         block = xfs_btree_get_block(cur, 0, &bp);
2251
2252 #ifdef DEBUG
2253         error = xfs_btree_check_block(cur, block, 0, bp);
2254         if (error)
2255                 goto error0;
2256 #endif
2257         /* Get the address of the rec to be updated. */
2258         ptr = cur->bc_ptrs[0];
2259         rp = xfs_btree_rec_addr(cur, ptr, block);
2260
2261         /* Fill in the new contents and log them. */
2262         xfs_btree_copy_recs(cur, rp, rec, 1);
2263         xfs_btree_log_recs(cur, bp, ptr, ptr);
2264
2265         /*
2266          * If we are tracking the last record in the tree and
2267          * we are at the far right edge of the tree, update it.
2268          */
2269         if (xfs_btree_is_lastrec(cur, block, 0)) {
2270                 cur->bc_ops->update_lastrec(cur, block, rec,
2271                                             ptr, LASTREC_UPDATE);
2272         }
2273
2274         /* Pass new key value up to our parent. */
2275         if (xfs_btree_needs_key_update(cur, ptr)) {
2276                 error = xfs_btree_update_keys(cur, 0);
2277                 if (error)
2278                         goto error0;
2279         }
2280
2281         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2282         return 0;
2283
2284 error0:
2285         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2286         return error;
2287 }
2288
2289 /*
2290  * Move 1 record left from cur/level if possible.
2291  * Update cur to reflect the new path.
2292  */
2293 STATIC int                                      /* error */
2294 xfs_btree_lshift(
2295         struct xfs_btree_cur    *cur,
2296         int                     level,
2297         int                     *stat)          /* success/failure */
2298 {
2299         struct xfs_buf          *lbp;           /* left buffer pointer */
2300         struct xfs_btree_block  *left;          /* left btree block */
2301         int                     lrecs;          /* left record count */
2302         struct xfs_buf          *rbp;           /* right buffer pointer */
2303         struct xfs_btree_block  *right;         /* right btree block */
2304         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
2305         int                     rrecs;          /* right record count */
2306         union xfs_btree_ptr     lptr;           /* left btree pointer */
2307         union xfs_btree_key     *rkp = NULL;    /* right btree key */
2308         union xfs_btree_ptr     *rpp = NULL;    /* right address pointer */
2309         union xfs_btree_rec     *rrp = NULL;    /* right record pointer */
2310         int                     error;          /* error return value */
2311         int                     i;
2312
2313         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2314         XFS_BTREE_TRACE_ARGI(cur, level);
2315
2316         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2317             level == cur->bc_nlevels - 1)
2318                 goto out0;
2319
2320         /* Set up variables for this block as "right". */
2321         right = xfs_btree_get_block(cur, level, &rbp);
2322
2323 #ifdef DEBUG
2324         error = xfs_btree_check_block(cur, right, level, rbp);
2325         if (error)
2326                 goto error0;
2327 #endif
2328
2329         /* If we've got no left sibling then we can't shift an entry left. */
2330         xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2331         if (xfs_btree_ptr_is_null(cur, &lptr))
2332                 goto out0;
2333
2334         /*
2335          * If the cursor entry is the one that would be moved, don't
2336          * do it... it's too complicated.
2337          */
2338         if (cur->bc_ptrs[level] <= 1)
2339                 goto out0;
2340
2341         /* Set up the left neighbor as "left". */
2342         error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2343         if (error)
2344                 goto error0;
2345
2346         /* If it's full, it can't take another entry. */
2347         lrecs = xfs_btree_get_numrecs(left);
2348         if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2349                 goto out0;
2350
2351         rrecs = xfs_btree_get_numrecs(right);
2352
2353         /*
2354          * We add one entry to the left side and remove one for the right side.
2355          * Account for it here, the changes will be updated on disk and logged
2356          * later.
2357          */
2358         lrecs++;
2359         rrecs--;
2360
2361         XFS_BTREE_STATS_INC(cur, lshift);
2362         XFS_BTREE_STATS_ADD(cur, moves, 1);
2363
2364         /*
2365          * If non-leaf, copy a key and a ptr to the left block.
2366          * Log the changes to the left block.
2367          */
2368         if (level > 0) {
2369                 /* It's a non-leaf.  Move keys and pointers. */
2370                 union xfs_btree_key     *lkp;   /* left btree key */
2371                 union xfs_btree_ptr     *lpp;   /* left address pointer */
2372
2373                 lkp = xfs_btree_key_addr(cur, lrecs, left);
2374                 rkp = xfs_btree_key_addr(cur, 1, right);
2375
2376                 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2377                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2378 #ifdef DEBUG
2379                 error = xfs_btree_check_ptr(cur, rpp, 0, level);
2380                 if (error)
2381                         goto error0;
2382 #endif
2383                 xfs_btree_copy_keys(cur, lkp, rkp, 1);
2384                 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2385
2386                 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2387                 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2388
2389                 ASSERT(cur->bc_ops->keys_inorder(cur,
2390                         xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2391         } else {
2392                 /* It's a leaf.  Move records.  */
2393                 union xfs_btree_rec     *lrp;   /* left record pointer */
2394
2395                 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2396                 rrp = xfs_btree_rec_addr(cur, 1, right);
2397
2398                 xfs_btree_copy_recs(cur, lrp, rrp, 1);
2399                 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2400
2401                 ASSERT(cur->bc_ops->recs_inorder(cur,
2402                         xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2403         }
2404
2405         xfs_btree_set_numrecs(left, lrecs);
2406         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2407
2408         xfs_btree_set_numrecs(right, rrecs);
2409         xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2410
2411         /*
2412          * Slide the contents of right down one entry.
2413          */
2414         XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2415         if (level > 0) {
2416                 /* It's a nonleaf. operate on keys and ptrs */
2417 #ifdef DEBUG
2418                 int                     i;              /* loop index */
2419
2420                 for (i = 0; i < rrecs; i++) {
2421                         error = xfs_btree_check_ptr(cur, rpp, i + 1, level);
2422                         if (error)
2423                                 goto error0;
2424                 }
2425 #endif
2426                 xfs_btree_shift_keys(cur,
2427                                 xfs_btree_key_addr(cur, 2, right),
2428                                 -1, rrecs);
2429                 xfs_btree_shift_ptrs(cur,
2430                                 xfs_btree_ptr_addr(cur, 2, right),
2431                                 -1, rrecs);
2432
2433                 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2434                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2435         } else {
2436                 /* It's a leaf. operate on records */
2437                 xfs_btree_shift_recs(cur,
2438                         xfs_btree_rec_addr(cur, 2, right),
2439                         -1, rrecs);
2440                 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2441         }
2442
2443         /*
2444          * Using a temporary cursor, update the parent key values of the
2445          * block on the left.
2446          */
2447         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2448                 error = xfs_btree_dup_cursor(cur, &tcur);
2449                 if (error)
2450                         goto error0;
2451                 i = xfs_btree_firstrec(tcur, level);
2452                 XFS_WANT_CORRUPTED_GOTO(tcur->bc_mp, i == 1, error0);
2453
2454                 error = xfs_btree_decrement(tcur, level, &i);
2455                 if (error)
2456                         goto error1;
2457
2458                 /* Update the parent high keys of the left block, if needed. */
2459                 error = xfs_btree_update_keys(tcur, level);
2460                 if (error)
2461                         goto error1;
2462
2463                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2464         }
2465
2466         /* Update the parent keys of the right block. */
2467         error = xfs_btree_update_keys(cur, level);
2468         if (error)
2469                 goto error0;
2470
2471         /* Slide the cursor value left one. */
2472         cur->bc_ptrs[level]--;
2473
2474         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2475         *stat = 1;
2476         return 0;
2477
2478 out0:
2479         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2480         *stat = 0;
2481         return 0;
2482
2483 error0:
2484         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2485         return error;
2486
2487 error1:
2488         XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2489         xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2490         return error;
2491 }
2492
2493 /*
2494  * Move 1 record right from cur/level if possible.
2495  * Update cur to reflect the new path.
2496  */
2497 STATIC int                                      /* error */
2498 xfs_btree_rshift(
2499         struct xfs_btree_cur    *cur,
2500         int                     level,
2501         int                     *stat)          /* success/failure */
2502 {
2503         struct xfs_buf          *lbp;           /* left buffer pointer */
2504         struct xfs_btree_block  *left;          /* left btree block */
2505         struct xfs_buf          *rbp;           /* right buffer pointer */
2506         struct xfs_btree_block  *right;         /* right btree block */
2507         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
2508         union xfs_btree_ptr     rptr;           /* right block pointer */
2509         union xfs_btree_key     *rkp;           /* right btree key */
2510         int                     rrecs;          /* right record count */
2511         int                     lrecs;          /* left record count */
2512         int                     error;          /* error return value */
2513         int                     i;              /* loop counter */
2514
2515         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2516         XFS_BTREE_TRACE_ARGI(cur, level);
2517
2518         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2519             (level == cur->bc_nlevels - 1))
2520                 goto out0;
2521
2522         /* Set up variables for this block as "left". */
2523         left = xfs_btree_get_block(cur, level, &lbp);
2524
2525 #ifdef DEBUG
2526         error = xfs_btree_check_block(cur, left, level, lbp);
2527         if (error)
2528                 goto error0;
2529 #endif
2530
2531         /* If we've got no right sibling then we can't shift an entry right. */
2532         xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2533         if (xfs_btree_ptr_is_null(cur, &rptr))
2534                 goto out0;
2535
2536         /*
2537          * If the cursor entry is the one that would be moved, don't
2538          * do it... it's too complicated.
2539          */
2540         lrecs = xfs_btree_get_numrecs(left);
2541         if (cur->bc_ptrs[level] >= lrecs)
2542                 goto out0;
2543
2544         /* Set up the right neighbor as "right". */
2545         error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2546         if (error)
2547                 goto error0;
2548
2549         /* If it's full, it can't take another entry. */
2550         rrecs = xfs_btree_get_numrecs(right);
2551         if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2552                 goto out0;
2553
2554         XFS_BTREE_STATS_INC(cur, rshift);
2555         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2556
2557         /*
2558          * Make a hole at the start of the right neighbor block, then
2559          * copy the last left block entry to the hole.
2560          */
2561         if (level > 0) {
2562                 /* It's a nonleaf. make a hole in the keys and ptrs */
2563                 union xfs_btree_key     *lkp;
2564                 union xfs_btree_ptr     *lpp;
2565                 union xfs_btree_ptr     *rpp;
2566
2567                 lkp = xfs_btree_key_addr(cur, lrecs, left);
2568                 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2569                 rkp = xfs_btree_key_addr(cur, 1, right);
2570                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2571
2572 #ifdef DEBUG
2573                 for (i = rrecs - 1; i >= 0; i--) {
2574                         error = xfs_btree_check_ptr(cur, rpp, i, level);
2575                         if (error)
2576                                 goto error0;
2577                 }
2578 #endif
2579
2580                 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2581                 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2582
2583 #ifdef DEBUG
2584                 error = xfs_btree_check_ptr(cur, lpp, 0, level);
2585                 if (error)
2586                         goto error0;
2587 #endif
2588
2589                 /* Now put the new data in, and log it. */
2590                 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2591                 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2592
2593                 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2594                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2595
2596                 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2597                         xfs_btree_key_addr(cur, 2, right)));
2598         } else {
2599                 /* It's a leaf. make a hole in the records */
2600                 union xfs_btree_rec     *lrp;
2601                 union xfs_btree_rec     *rrp;
2602
2603                 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2604                 rrp = xfs_btree_rec_addr(cur, 1, right);
2605
2606                 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2607
2608                 /* Now put the new data in, and log it. */
2609                 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2610                 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2611         }
2612
2613         /*
2614          * Decrement and log left's numrecs, bump and log right's numrecs.
2615          */
2616         xfs_btree_set_numrecs(left, --lrecs);
2617         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2618
2619         xfs_btree_set_numrecs(right, ++rrecs);
2620         xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2621
2622         /*
2623          * Using a temporary cursor, update the parent key values of the
2624          * block on the right.
2625          */
2626         error = xfs_btree_dup_cursor(cur, &tcur);
2627         if (error)
2628                 goto error0;
2629         i = xfs_btree_lastrec(tcur, level);
2630         XFS_WANT_CORRUPTED_GOTO(tcur->bc_mp, i == 1, error0);
2631
2632         error = xfs_btree_increment(tcur, level, &i);
2633         if (error)
2634                 goto error1;
2635
2636         /* Update the parent high keys of the left block, if needed. */
2637         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2638                 error = xfs_btree_update_keys(cur, level);
2639                 if (error)
2640                         goto error1;
2641         }
2642
2643         /* Update the parent keys of the right block. */
2644         error = xfs_btree_update_keys(tcur, level);
2645         if (error)
2646                 goto error1;
2647
2648         xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2649
2650         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2651         *stat = 1;
2652         return 0;
2653
2654 out0:
2655         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2656         *stat = 0;
2657         return 0;
2658
2659 error0:
2660         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2661         return error;
2662
2663 error1:
2664         XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2665         xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2666         return error;
2667 }
2668
2669 /*
2670  * Split cur/level block in half.
2671  * Return new block number and the key to its first
2672  * record (to be inserted into parent).
2673  */
2674 STATIC int                                      /* error */
2675 __xfs_btree_split(
2676         struct xfs_btree_cur    *cur,
2677         int                     level,
2678         union xfs_btree_ptr     *ptrp,
2679         union xfs_btree_key     *key,
2680         struct xfs_btree_cur    **curp,
2681         int                     *stat)          /* success/failure */
2682 {
2683         union xfs_btree_ptr     lptr;           /* left sibling block ptr */
2684         struct xfs_buf          *lbp;           /* left buffer pointer */
2685         struct xfs_btree_block  *left;          /* left btree block */
2686         union xfs_btree_ptr     rptr;           /* right sibling block ptr */
2687         struct xfs_buf          *rbp;           /* right buffer pointer */
2688         struct xfs_btree_block  *right;         /* right btree block */
2689         union xfs_btree_ptr     rrptr;          /* right-right sibling ptr */
2690         struct xfs_buf          *rrbp;          /* right-right buffer pointer */
2691         struct xfs_btree_block  *rrblock;       /* right-right btree block */
2692         int                     lrecs;
2693         int                     rrecs;
2694         int                     src_index;
2695         int                     error;          /* error return value */
2696 #ifdef DEBUG
2697         int                     i;
2698 #endif
2699
2700         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2701         XFS_BTREE_TRACE_ARGIPK(cur, level, *ptrp, key);
2702
2703         XFS_BTREE_STATS_INC(cur, split);
2704
2705         /* Set up left block (current one). */
2706         left = xfs_btree_get_block(cur, level, &lbp);
2707
2708 #ifdef DEBUG
2709         error = xfs_btree_check_block(cur, left, level, lbp);
2710         if (error)
2711                 goto error0;
2712 #endif
2713
2714         xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2715
2716         /* Allocate the new block. If we can't do it, we're toast. Give up. */
2717         error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, stat);
2718         if (error)
2719                 goto error0;
2720         if (*stat == 0)
2721                 goto out0;
2722         XFS_BTREE_STATS_INC(cur, alloc);
2723
2724         /* Set up the new block as "right". */
2725         error = xfs_btree_get_buf_block(cur, &rptr, 0, &right, &rbp);
2726         if (error)
2727                 goto error0;
2728
2729         /* Fill in the btree header for the new right block. */
2730         xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2731
2732         /*
2733          * Split the entries between the old and the new block evenly.
2734          * Make sure that if there's an odd number of entries now, that
2735          * each new block will have the same number of entries.
2736          */
2737         lrecs = xfs_btree_get_numrecs(left);
2738         rrecs = lrecs / 2;
2739         if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
2740                 rrecs++;
2741         src_index = (lrecs - rrecs + 1);
2742
2743         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2744
2745         /* Adjust numrecs for the later get_*_keys() calls. */
2746         lrecs -= rrecs;
2747         xfs_btree_set_numrecs(left, lrecs);
2748         xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2749
2750         /*
2751          * Copy btree block entries from the left block over to the
2752          * new block, the right. Update the right block and log the
2753          * changes.
2754          */
2755         if (level > 0) {
2756                 /* It's a non-leaf.  Move keys and pointers. */
2757                 union xfs_btree_key     *lkp;   /* left btree key */
2758                 union xfs_btree_ptr     *lpp;   /* left address pointer */
2759                 union xfs_btree_key     *rkp;   /* right btree key */
2760                 union xfs_btree_ptr     *rpp;   /* right address pointer */
2761
2762                 lkp = xfs_btree_key_addr(cur, src_index, left);
2763                 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2764                 rkp = xfs_btree_key_addr(cur, 1, right);
2765                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2766
2767 #ifdef DEBUG
2768                 for (i = src_index; i < rrecs; i++) {
2769                         error = xfs_btree_check_ptr(cur, lpp, i, level);
2770                         if (error)
2771                                 goto error0;
2772                 }
2773 #endif
2774
2775                 /* Copy the keys & pointers to the new block. */
2776                 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2777                 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2778
2779                 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2780                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2781
2782                 /* Stash the keys of the new block for later insertion. */
2783                 xfs_btree_get_node_keys(cur, right, key);
2784         } else {
2785                 /* It's a leaf.  Move records.  */
2786                 union xfs_btree_rec     *lrp;   /* left record pointer */
2787                 union xfs_btree_rec     *rrp;   /* right record pointer */
2788
2789                 lrp = xfs_btree_rec_addr(cur, src_index, left);
2790                 rrp = xfs_btree_rec_addr(cur, 1, right);
2791
2792                 /* Copy records to the new block. */
2793                 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2794                 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2795
2796                 /* Stash the keys of the new block for later insertion. */
2797                 xfs_btree_get_leaf_keys(cur, right, key);
2798         }
2799
2800         /*
2801          * Find the left block number by looking in the buffer.
2802          * Adjust sibling pointers.
2803          */
2804         xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2805         xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2806         xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2807         xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2808
2809         xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2810         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2811
2812         /*
2813          * If there's a block to the new block's right, make that block
2814          * point back to right instead of to left.
2815          */
2816         if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2817                 error = xfs_btree_read_buf_block(cur, &rrptr,
2818                                                         0, &rrblock, &rrbp);
2819                 if (error)
2820                         goto error0;
2821                 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2822                 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2823         }
2824
2825         /* Update the parent high keys of the left block, if needed. */
2826         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2827                 error = xfs_btree_update_keys(cur, level);
2828                 if (error)
2829                         goto error0;
2830         }
2831
2832         /*
2833          * If the cursor is really in the right block, move it there.
2834          * If it's just pointing past the last entry in left, then we'll
2835          * insert there, so don't change anything in that case.
2836          */
2837         if (cur->bc_ptrs[level] > lrecs + 1) {
2838                 xfs_btree_setbuf(cur, level, rbp);
2839                 cur->bc_ptrs[level] -= lrecs;
2840         }
2841         /*
2842          * If there are more levels, we'll need another cursor which refers
2843          * the right block, no matter where this cursor was.
2844          */
2845         if (level + 1 < cur->bc_nlevels) {
2846                 error = xfs_btree_dup_cursor(cur, curp);
2847                 if (error)
2848                         goto error0;
2849                 (*curp)->bc_ptrs[level + 1]++;
2850         }
2851         *ptrp = rptr;
2852         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2853         *stat = 1;
2854         return 0;
2855 out0:
2856         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2857         *stat = 0;
2858         return 0;
2859
2860 error0:
2861         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2862         return error;
2863 }
2864
2865 struct xfs_btree_split_args {
2866         struct xfs_btree_cur    *cur;
2867         int                     level;
2868         union xfs_btree_ptr     *ptrp;
2869         union xfs_btree_key     *key;
2870         struct xfs_btree_cur    **curp;
2871         int                     *stat;          /* success/failure */
2872         int                     result;
2873         bool                    kswapd; /* allocation in kswapd context */
2874         struct completion       *done;
2875         struct work_struct      work;
2876 };
2877
2878 /*
2879  * Stack switching interfaces for allocation
2880  */
2881 static void
2882 xfs_btree_split_worker(
2883         struct work_struct      *work)
2884 {
2885         struct xfs_btree_split_args     *args = container_of(work,
2886                                                 struct xfs_btree_split_args, work);
2887         unsigned long           pflags;
2888         unsigned long           new_pflags = PF_FSTRANS;
2889
2890         /*
2891          * we are in a transaction context here, but may also be doing work
2892          * in kswapd context, and hence we may need to inherit that state
2893          * temporarily to ensure that we don't block waiting for memory reclaim
2894          * in any way.
2895          */
2896         if (args->kswapd)
2897                 new_pflags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
2898
2899         current_set_flags_nested(&pflags, new_pflags);
2900
2901         args->result = __xfs_btree_split(args->cur, args->level, args->ptrp,
2902                                          args->key, args->curp, args->stat);
2903         complete(args->done);
2904
2905         current_restore_flags_nested(&pflags, new_pflags);
2906 }
2907
2908 /*
2909  * BMBT split requests often come in with little stack to work on. Push
2910  * them off to a worker thread so there is lots of stack to use. For the other
2911  * btree types, just call directly to avoid the context switch overhead here.
2912  */
2913 STATIC int                                      /* error */
2914 xfs_btree_split(
2915         struct xfs_btree_cur    *cur,
2916         int                     level,
2917         union xfs_btree_ptr     *ptrp,
2918         union xfs_btree_key     *key,
2919         struct xfs_btree_cur    **curp,
2920         int                     *stat)          /* success/failure */
2921 {
2922         struct xfs_btree_split_args     args;
2923         DECLARE_COMPLETION_ONSTACK(done);
2924
2925         if (cur->bc_btnum != XFS_BTNUM_BMAP)
2926                 return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
2927
2928         args.cur = cur;
2929         args.level = level;
2930         args.ptrp = ptrp;
2931         args.key = key;
2932         args.curp = curp;
2933         args.stat = stat;
2934         args.done = &done;
2935         args.kswapd = current_is_kswapd();
2936         INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker);
2937         queue_work(xfs_alloc_wq, &args.work);
2938         wait_for_completion(&done);
2939         destroy_work_on_stack(&args.work);
2940         return args.result;
2941 }
2942
2943
2944 /*
2945  * Copy the old inode root contents into a real block and make the
2946  * broot point to it.
2947  */
2948 int                                             /* error */
2949 xfs_btree_new_iroot(
2950         struct xfs_btree_cur    *cur,           /* btree cursor */
2951         int                     *logflags,      /* logging flags for inode */
2952         int                     *stat)          /* return status - 0 fail */
2953 {
2954         struct xfs_buf          *cbp;           /* buffer for cblock */
2955         struct xfs_btree_block  *block;         /* btree block */
2956         struct xfs_btree_block  *cblock;        /* child btree block */
2957         union xfs_btree_key     *ckp;           /* child key pointer */
2958         union xfs_btree_ptr     *cpp;           /* child ptr pointer */
2959         union xfs_btree_key     *kp;            /* pointer to btree key */
2960         union xfs_btree_ptr     *pp;            /* pointer to block addr */
2961         union xfs_btree_ptr     nptr;           /* new block addr */
2962         int                     level;          /* btree level */
2963         int                     error;          /* error return code */
2964 #ifdef DEBUG
2965         int                     i;              /* loop counter */
2966 #endif
2967
2968         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2969         XFS_BTREE_STATS_INC(cur, newroot);
2970
2971         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2972
2973         level = cur->bc_nlevels - 1;
2974
2975         block = xfs_btree_get_iroot(cur);
2976         pp = xfs_btree_ptr_addr(cur, 1, block);
2977
2978         /* Allocate the new block. If we can't do it, we're toast. Give up. */
2979         error = cur->bc_ops->alloc_block(cur, pp, &nptr, stat);
2980         if (error)
2981                 goto error0;
2982         if (*stat == 0) {
2983                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2984                 return 0;
2985         }
2986         XFS_BTREE_STATS_INC(cur, alloc);
2987
2988         /* Copy the root into a real block. */
2989         error = xfs_btree_get_buf_block(cur, &nptr, 0, &cblock, &cbp);
2990         if (error)
2991                 goto error0;
2992
2993         /*
2994          * we can't just memcpy() the root in for CRC enabled btree blocks.
2995          * In that case have to also ensure the blkno remains correct
2996          */
2997         memcpy(cblock, block, xfs_btree_block_len(cur));
2998         if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
2999                 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
3000                         cblock->bb_u.l.bb_blkno = cpu_to_be64(cbp->b_bn);
3001                 else
3002                         cblock->bb_u.s.bb_blkno = cpu_to_be64(cbp->b_bn);
3003         }
3004
3005         be16_add_cpu(&block->bb_level, 1);
3006         xfs_btree_set_numrecs(block, 1);
3007         cur->bc_nlevels++;
3008         cur->bc_ptrs[level + 1] = 1;
3009
3010         kp = xfs_btree_key_addr(cur, 1, block);
3011         ckp = xfs_btree_key_addr(cur, 1, cblock);
3012         xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
3013
3014         cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3015 #ifdef DEBUG
3016         for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
3017                 error = xfs_btree_check_ptr(cur, pp, i, level);
3018                 if (error)
3019                         goto error0;
3020         }
3021 #endif
3022         xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
3023
3024 #ifdef DEBUG
3025         error = xfs_btree_check_ptr(cur, &nptr, 0, level);
3026         if (error)
3027                 goto error0;
3028 #endif
3029         xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
3030
3031         xfs_iroot_realloc(cur->bc_private.b.ip,
3032                           1 - xfs_btree_get_numrecs(cblock),
3033                           cur->bc_private.b.whichfork);
3034
3035         xfs_btree_setbuf(cur, level, cbp);
3036
3037         /*
3038          * Do all this logging at the end so that
3039          * the root is at the right level.
3040          */
3041         xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
3042         xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3043         xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3044
3045         *logflags |=
3046                 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork);
3047         *stat = 1;
3048         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3049         return 0;
3050 error0:
3051         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3052         return error;
3053 }
3054
3055 /*
3056  * Allocate a new root block, fill it in.
3057  */
3058 STATIC int                              /* error */
3059 xfs_btree_new_root(
3060         struct xfs_btree_cur    *cur,   /* btree cursor */
3061         int                     *stat)  /* success/failure */
3062 {
3063         struct xfs_btree_block  *block; /* one half of the old root block */
3064         struct xfs_buf          *bp;    /* buffer containing block */
3065         int                     error;  /* error return value */
3066         struct xfs_buf          *lbp;   /* left buffer pointer */
3067         struct xfs_btree_block  *left;  /* left btree block */
3068         struct xfs_buf          *nbp;   /* new (root) buffer */
3069         struct xfs_btree_block  *new;   /* new (root) btree block */
3070         int                     nptr;   /* new value for key index, 1 or 2 */
3071         struct xfs_buf          *rbp;   /* right buffer pointer */
3072         struct xfs_btree_block  *right; /* right btree block */
3073         union xfs_btree_ptr     rptr;
3074         union xfs_btree_ptr     lptr;
3075
3076         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3077         XFS_BTREE_STATS_INC(cur, newroot);
3078
3079         /* initialise our start point from the cursor */
3080         cur->bc_ops->init_ptr_from_cur(cur, &rptr);
3081
3082         /* Allocate the new block. If we can't do it, we're toast. Give up. */
3083         error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, stat);
3084         if (error)
3085                 goto error0;
3086         if (*stat == 0)
3087                 goto out0;
3088         XFS_BTREE_STATS_INC(cur, alloc);
3089
3090         /* Set up the new block. */
3091         error = xfs_btree_get_buf_block(cur, &lptr, 0, &new, &nbp);
3092         if (error)
3093                 goto error0;
3094
3095         /* Set the root in the holding structure  increasing the level by 1. */
3096         cur->bc_ops->set_root(cur, &lptr, 1);
3097
3098         /*
3099          * At the previous root level there are now two blocks: the old root,
3100          * and the new block generated when it was split.  We don't know which
3101          * one the cursor is pointing at, so we set up variables "left" and
3102          * "right" for each case.
3103          */
3104         block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
3105
3106 #ifdef DEBUG
3107         error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
3108         if (error)
3109                 goto error0;
3110 #endif
3111
3112         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3113         if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3114                 /* Our block is left, pick up the right block. */
3115                 lbp = bp;
3116                 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
3117                 left = block;
3118                 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
3119                 if (error)
3120                         goto error0;
3121                 bp = rbp;
3122                 nptr = 1;
3123         } else {
3124                 /* Our block is right, pick up the left block. */
3125                 rbp = bp;
3126                 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
3127                 right = block;
3128                 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
3129                 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3130                 if (error)
3131                         goto error0;
3132                 bp = lbp;
3133                 nptr = 2;
3134         }
3135
3136         /* Fill in the new block's btree header and log it. */
3137         xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
3138         xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
3139         ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
3140                         !xfs_btree_ptr_is_null(cur, &rptr));
3141
3142         /* Fill in the key data in the new root. */
3143         if (xfs_btree_get_level(left) > 0) {
3144                 /*
3145                  * Get the keys for the left block's keys and put them directly
3146                  * in the parent block.  Do the same for the right block.
3147                  */
3148                 xfs_btree_get_node_keys(cur, left,
3149                                 xfs_btree_key_addr(cur, 1, new));
3150                 xfs_btree_get_node_keys(cur, right,
3151                                 xfs_btree_key_addr(cur, 2, new));
3152         } else {
3153                 /*
3154                  * Get the keys for the left block's records and put them
3155                  * directly in the parent block.  Do the same for the right
3156                  * block.
3157                  */
3158                 xfs_btree_get_leaf_keys(cur, left,
3159                         xfs_btree_key_addr(cur, 1, new));
3160                 xfs_btree_get_leaf_keys(cur, right,
3161                         xfs_btree_key_addr(cur, 2, new));
3162         }
3163         xfs_btree_log_keys(cur, nbp, 1, 2);
3164
3165         /* Fill in the pointer data in the new root. */
3166         xfs_btree_copy_ptrs(cur,
3167                 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
3168         xfs_btree_copy_ptrs(cur,
3169                 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
3170         xfs_btree_log_ptrs(cur, nbp, 1, 2);
3171
3172         /* Fix up the cursor. */
3173         xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
3174         cur->bc_ptrs[cur->bc_nlevels] = nptr;
3175         cur->bc_nlevels++;
3176         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3177         *stat = 1;
3178         return 0;
3179 error0:
3180         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3181         return error;
3182 out0:
3183         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3184         *stat = 0;
3185         return 0;
3186 }
3187
3188 STATIC int
3189 xfs_btree_make_block_unfull(
3190         struct xfs_btree_cur    *cur,   /* btree cursor */
3191         int                     level,  /* btree level */
3192         int                     numrecs,/* # of recs in block */
3193         int                     *oindex,/* old tree index */
3194         int                     *index, /* new tree index */
3195         union xfs_btree_ptr     *nptr,  /* new btree ptr */
3196         struct xfs_btree_cur    **ncur, /* new btree cursor */
3197         union xfs_btree_key     *key,   /* key of new block */
3198         int                     *stat)
3199 {
3200         int                     error = 0;
3201
3202         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3203             level == cur->bc_nlevels - 1) {
3204                 struct xfs_inode *ip = cur->bc_private.b.ip;
3205
3206                 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
3207                         /* A root block that can be made bigger. */
3208                         xfs_iroot_realloc(ip, 1, cur->bc_private.b.whichfork);
3209                         *stat = 1;
3210                 } else {
3211                         /* A root block that needs replacing */
3212                         int     logflags = 0;
3213
3214                         error = xfs_btree_new_iroot(cur, &logflags, stat);
3215                         if (error || *stat == 0)
3216                                 return error;
3217
3218                         xfs_trans_log_inode(cur->bc_tp, ip, logflags);
3219                 }
3220
3221                 return 0;
3222         }
3223
3224         /* First, try shifting an entry to the right neighbor. */
3225         error = xfs_btree_rshift(cur, level, stat);
3226         if (error || *stat)
3227                 return error;
3228
3229         /* Next, try shifting an entry to the left neighbor. */
3230         error = xfs_btree_lshift(cur, level, stat);
3231         if (error)
3232                 return error;
3233
3234         if (*stat) {
3235                 *oindex = *index = cur->bc_ptrs[level];
3236                 return 0;
3237         }
3238
3239         /*
3240          * Next, try splitting the current block in half.
3241          *
3242          * If this works we have to re-set our variables because we
3243          * could be in a different block now.
3244          */
3245         error = xfs_btree_split(cur, level, nptr, key, ncur, stat);
3246         if (error || *stat == 0)
3247                 return error;
3248
3249
3250         *index = cur->bc_ptrs[level];
3251         return 0;
3252 }
3253
3254 /*
3255  * Insert one record/level.  Return information to the caller
3256  * allowing the next level up to proceed if necessary.
3257  */
3258 STATIC int
3259 xfs_btree_insrec(
3260         struct xfs_btree_cur    *cur,   /* btree cursor */
3261         int                     level,  /* level to insert record at */
3262         union xfs_btree_ptr     *ptrp,  /* i/o: block number inserted */
3263         union xfs_btree_rec     *rec,   /* record to insert */
3264         union xfs_btree_key     *key,   /* i/o: block key for ptrp */
3265         struct xfs_btree_cur    **curp, /* output: new cursor replacing cur */
3266         int                     *stat)  /* success/failure */
3267 {
3268         struct xfs_btree_block  *block; /* btree block */
3269         struct xfs_buf          *bp;    /* buffer for block */
3270         union xfs_btree_ptr     nptr;   /* new block ptr */
3271         struct xfs_btree_cur    *ncur;  /* new btree cursor */
3272         union xfs_btree_key     nkey;   /* new block key */
3273         union xfs_btree_key     *lkey;
3274         int                     optr;   /* old key/record index */
3275         int                     ptr;    /* key/record index */
3276         int                     numrecs;/* number of records */
3277         int                     error;  /* error return value */
3278 #ifdef DEBUG
3279         int                     i;
3280 #endif
3281         xfs_daddr_t             old_bn;
3282
3283         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3284         XFS_BTREE_TRACE_ARGIPR(cur, level, *ptrp, &rec);
3285
3286         ncur = NULL;
3287         lkey = &nkey;
3288
3289         /*
3290          * If we have an external root pointer, and we've made it to the
3291          * root level, allocate a new root block and we're done.
3292          */
3293         if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3294             (level >= cur->bc_nlevels)) {
3295                 error = xfs_btree_new_root(cur, stat);
3296                 xfs_btree_set_ptr_null(cur, ptrp);
3297
3298                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3299                 return error;
3300         }
3301
3302         /* If we're off the left edge, return failure. */
3303         ptr = cur->bc_ptrs[level];
3304         if (ptr == 0) {
3305                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3306                 *stat = 0;
3307                 return 0;
3308         }
3309
3310         optr = ptr;
3311
3312         XFS_BTREE_STATS_INC(cur, insrec);
3313
3314         /* Get pointers to the btree buffer and block. */
3315         block = xfs_btree_get_block(cur, level, &bp);
3316         old_bn = bp ? bp->b_bn : XFS_BUF_DADDR_NULL;
3317         numrecs = xfs_btree_get_numrecs(block);
3318
3319 #ifdef DEBUG
3320         error = xfs_btree_check_block(cur, block, level, bp);
3321         if (error)
3322                 goto error0;
3323
3324         /* Check that the new entry is being inserted in the right place. */
3325         if (ptr <= numrecs) {
3326                 if (level == 0) {
3327                         ASSERT(cur->bc_ops->recs_inorder(cur, rec,
3328                                 xfs_btree_rec_addr(cur, ptr, block)));
3329                 } else {
3330                         ASSERT(cur->bc_ops->keys_inorder(cur, key,
3331                                 xfs_btree_key_addr(cur, ptr, block)));
3332                 }
3333         }
3334 #endif
3335
3336         /*
3337          * If the block is full, we can't insert the new entry until we
3338          * make the block un-full.
3339          */
3340         xfs_btree_set_ptr_null(cur, &nptr);
3341         if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
3342                 error = xfs_btree_make_block_unfull(cur, level, numrecs,
3343                                         &optr, &ptr, &nptr, &ncur, lkey, stat);
3344                 if (error || *stat == 0)
3345                         goto error0;
3346         }
3347
3348         /*
3349          * The current block may have changed if the block was
3350          * previously full and we have just made space in it.
3351          */
3352         block = xfs_btree_get_block(cur, level, &bp);
3353         numrecs = xfs_btree_get_numrecs(block);
3354
3355 #ifdef DEBUG
3356         error = xfs_btree_check_block(cur, block, level, bp);
3357         if (error)
3358                 return error;
3359 #endif
3360
3361         /*
3362          * At this point we know there's room for our new entry in the block
3363          * we're pointing at.
3364          */
3365         XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
3366
3367         if (level > 0) {
3368                 /* It's a nonleaf. make a hole in the keys and ptrs */
3369                 union xfs_btree_key     *kp;
3370                 union xfs_btree_ptr     *pp;
3371
3372                 kp = xfs_btree_key_addr(cur, ptr, block);
3373                 pp = xfs_btree_ptr_addr(cur, ptr, block);
3374
3375 #ifdef DEBUG
3376                 for (i = numrecs - ptr; i >= 0; i--) {
3377                         error = xfs_btree_check_ptr(cur, pp, i, level);
3378                         if (error)
3379                                 return error;
3380                 }
3381 #endif
3382
3383                 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
3384                 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
3385
3386 #ifdef DEBUG
3387                 error = xfs_btree_check_ptr(cur, ptrp, 0, level);
3388                 if (error)
3389                         goto error0;
3390 #endif
3391
3392                 /* Now put the new data in, bump numrecs and log it. */
3393                 xfs_btree_copy_keys(cur, kp, key, 1);
3394                 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
3395                 numrecs++;
3396                 xfs_btree_set_numrecs(block, numrecs);
3397                 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
3398                 xfs_btree_log_keys(cur, bp, ptr, numrecs);
3399 #ifdef DEBUG
3400                 if (ptr < numrecs) {
3401                         ASSERT(cur->bc_ops->keys_inorder(cur, kp,
3402                                 xfs_btree_key_addr(cur, ptr + 1, block)));
3403                 }
3404 #endif
3405         } else {
3406                 /* It's a leaf. make a hole in the records */
3407                 union xfs_btree_rec             *rp;
3408
3409                 rp = xfs_btree_rec_addr(cur, ptr, block);
3410
3411                 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
3412
3413                 /* Now put the new data in, bump numrecs and log it. */
3414                 xfs_btree_copy_recs(cur, rp, rec, 1);
3415                 xfs_btree_set_numrecs(block, ++numrecs);
3416                 xfs_btree_log_recs(cur, bp, ptr, numrecs);
3417 #ifdef DEBUG
3418                 if (ptr < numrecs) {
3419                         ASSERT(cur->bc_ops->recs_inorder(cur, rp,
3420                                 xfs_btree_rec_addr(cur, ptr + 1, block)));
3421                 }
3422 #endif
3423         }
3424
3425         /* Log the new number of records in the btree header. */
3426         xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3427
3428         /*
3429          * If we just inserted into a new tree block, we have to
3430          * recalculate nkey here because nkey is out of date.
3431          *
3432          * Otherwise we're just updating an existing block (having shoved
3433          * some records into the new tree block), so use the regular key
3434          * update mechanism.
3435          */
3436         if (bp && bp->b_bn != old_bn) {
3437                 xfs_btree_get_keys(cur, block, lkey);
3438         } else if (xfs_btree_needs_key_update(cur, optr)) {
3439                 error = xfs_btree_update_keys(cur, level);
3440                 if (error)
3441                         goto error0;
3442         }
3443
3444         /*
3445          * If we are tracking the last record in the tree and
3446          * we are at the far right edge of the tree, update it.
3447          */
3448         if (xfs_btree_is_lastrec(cur, block, level)) {
3449                 cur->bc_ops->update_lastrec(cur, block, rec,
3450                                             ptr, LASTREC_INSREC);
3451         }
3452
3453         /*
3454          * Return the new block number, if any.
3455          * If there is one, give back a record value and a cursor too.
3456          */
3457         *ptrp = nptr;
3458         if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3459                 xfs_btree_copy_keys(cur, key, lkey, 1);
3460                 *curp = ncur;
3461         }
3462
3463         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3464         *stat = 1;
3465         return 0;
3466
3467 error0:
3468         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3469         return error;
3470 }
3471
3472 /*
3473  * Insert the record at the point referenced by cur.
3474  *
3475  * A multi-level split of the tree on insert will invalidate the original
3476  * cursor.  All callers of this function should assume that the cursor is
3477  * no longer valid and revalidate it.
3478  */
3479 int
3480 xfs_btree_insert(
3481         struct xfs_btree_cur    *cur,
3482         int                     *stat)
3483 {
3484         int                     error;  /* error return value */
3485         int                     i;      /* result value, 0 for failure */
3486         int                     level;  /* current level number in btree */
3487         union xfs_btree_ptr     nptr;   /* new block number (split result) */
3488         struct xfs_btree_cur    *ncur;  /* new cursor (split result) */
3489         struct xfs_btree_cur    *pcur;  /* previous level's cursor */
3490         union xfs_btree_key     bkey;   /* key of block to insert */
3491         union xfs_btree_key     *key;
3492         union xfs_btree_rec     rec;    /* record to insert */
3493
3494         level = 0;
3495         ncur = NULL;
3496         pcur = cur;
3497         key = &bkey;
3498
3499         xfs_btree_set_ptr_null(cur, &nptr);
3500
3501         /* Make a key out of the record data to be inserted, and save it. */
3502         cur->bc_ops->init_rec_from_cur(cur, &rec);
3503         cur->bc_ops->init_key_from_rec(key, &rec);
3504
3505         /*
3506          * Loop going up the tree, starting at the leaf level.
3507          * Stop when we don't get a split block, that must mean that
3508          * the insert is finished with this level.
3509          */
3510         do {
3511                 /*
3512                  * Insert nrec/nptr into this level of the tree.
3513                  * Note if we fail, nptr will be null.
3514                  */
3515                 error = xfs_btree_insrec(pcur, level, &nptr, &rec, key,
3516                                 &ncur, &i);
3517                 if (error) {
3518                         if (pcur != cur)
3519                                 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3520                         goto error0;
3521                 }
3522
3523                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3524                 level++;
3525
3526                 /*
3527                  * See if the cursor we just used is trash.
3528                  * Can't trash the caller's cursor, but otherwise we should
3529                  * if ncur is a new cursor or we're about to be done.
3530                  */
3531                 if (pcur != cur &&
3532                     (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3533                         /* Save the state from the cursor before we trash it */
3534                         if (cur->bc_ops->update_cursor)
3535                                 cur->bc_ops->update_cursor(pcur, cur);
3536                         cur->bc_nlevels = pcur->bc_nlevels;
3537                         xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3538                 }
3539                 /* If we got a new cursor, switch to it. */
3540                 if (ncur) {
3541                         pcur = ncur;
3542                         ncur = NULL;
3543                 }
3544         } while (!xfs_btree_ptr_is_null(cur, &nptr));
3545
3546         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3547         *stat = i;
3548         return 0;
3549 error0:
3550         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3551         return error;
3552 }
3553
3554 /*
3555  * Try to merge a non-leaf block back into the inode root.
3556  *
3557  * Note: the killroot names comes from the fact that we're effectively
3558  * killing the old root block.  But because we can't just delete the
3559  * inode we have to copy the single block it was pointing to into the
3560  * inode.
3561  */
3562 STATIC int
3563 xfs_btree_kill_iroot(
3564         struct xfs_btree_cur    *cur)
3565 {
3566         int                     whichfork = cur->bc_private.b.whichfork;
3567         struct xfs_inode        *ip = cur->bc_private.b.ip;
3568         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
3569         struct xfs_btree_block  *block;
3570         struct xfs_btree_block  *cblock;
3571         union xfs_btree_key     *kp;
3572         union xfs_btree_key     *ckp;
3573         union xfs_btree_ptr     *pp;
3574         union xfs_btree_ptr     *cpp;
3575         struct xfs_buf          *cbp;
3576         int                     level;
3577         int                     index;
3578         int                     numrecs;
3579         int                     error;
3580 #ifdef DEBUG
3581         union xfs_btree_ptr     ptr;
3582         int                     i;
3583 #endif
3584
3585         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3586
3587         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3588         ASSERT(cur->bc_nlevels > 1);
3589
3590         /*
3591          * Don't deal with the root block needs to be a leaf case.
3592          * We're just going to turn the thing back into extents anyway.
3593          */
3594         level = cur->bc_nlevels - 1;
3595         if (level == 1)
3596                 goto out0;
3597
3598         /*
3599          * Give up if the root has multiple children.
3600          */
3601         block = xfs_btree_get_iroot(cur);
3602         if (xfs_btree_get_numrecs(block) != 1)
3603                 goto out0;
3604
3605         cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3606         numrecs = xfs_btree_get_numrecs(cblock);
3607
3608         /*
3609          * Only do this if the next level will fit.
3610          * Then the data must be copied up to the inode,
3611          * instead of freeing the root you free the next level.
3612          */
3613         if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3614                 goto out0;
3615
3616         XFS_BTREE_STATS_INC(cur, killroot);
3617
3618 #ifdef DEBUG
3619         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3620         ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3621         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3622         ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3623 #endif
3624
3625         index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3626         if (index) {
3627                 xfs_iroot_realloc(cur->bc_private.b.ip, index,
3628                                   cur->bc_private.b.whichfork);
3629                 block = ifp->if_broot;
3630         }
3631
3632         be16_add_cpu(&block->bb_numrecs, index);
3633         ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3634
3635         kp = xfs_btree_key_addr(cur, 1, block);
3636         ckp = xfs_btree_key_addr(cur, 1, cblock);
3637         xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3638
3639         pp = xfs_btree_ptr_addr(cur, 1, block);
3640         cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3641 #ifdef DEBUG
3642         for (i = 0; i < numrecs; i++) {
3643                 error = xfs_btree_check_ptr(cur, cpp, i, level - 1);
3644                 if (error) {
3645                         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3646                         return error;
3647                 }
3648         }
3649 #endif
3650         xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3651
3652         error = xfs_btree_free_block(cur, cbp);
3653         if (error) {
3654                 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3655                 return error;
3656         }
3657
3658         cur->bc_bufs[level - 1] = NULL;
3659         be16_add_cpu(&block->bb_level, -1);
3660         xfs_trans_log_inode(cur->bc_tp, ip,
3661                 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork));
3662         cur->bc_nlevels--;
3663 out0:
3664         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3665         return 0;
3666 }
3667
3668 /*
3669  * Kill the current root node, and replace it with it's only child node.
3670  */
3671 STATIC int
3672 xfs_btree_kill_root(
3673         struct xfs_btree_cur    *cur,
3674         struct xfs_buf          *bp,
3675         int                     level,
3676         union xfs_btree_ptr     *newroot)
3677 {
3678         int                     error;
3679
3680         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3681         XFS_BTREE_STATS_INC(cur, killroot);
3682
3683         /*
3684          * Update the root pointer, decreasing the level by 1 and then
3685          * free the old root.
3686          */
3687         cur->bc_ops->set_root(cur, newroot, -1);
3688
3689         error = xfs_btree_free_block(cur, bp);
3690         if (error) {
3691                 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3692                 return error;
3693         }
3694
3695         cur->bc_bufs[level] = NULL;
3696         cur->bc_ra[level] = 0;
3697         cur->bc_nlevels--;
3698
3699         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3700         return 0;
3701 }
3702
3703 STATIC int
3704 xfs_btree_dec_cursor(
3705         struct xfs_btree_cur    *cur,
3706         int                     level,
3707         int                     *stat)
3708 {
3709         int                     error;
3710         int                     i;
3711
3712         if (level > 0) {
3713                 error = xfs_btree_decrement(cur, level, &i);
3714                 if (error)
3715                         return error;
3716         }
3717
3718         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3719         *stat = 1;
3720         return 0;
3721 }
3722
3723 /*
3724  * Single level of the btree record deletion routine.
3725  * Delete record pointed to by cur/level.
3726  * Remove the record from its block then rebalance the tree.
3727  * Return 0 for error, 1 for done, 2 to go on to the next level.
3728  */
3729 STATIC int                                      /* error */
3730 xfs_btree_delrec(
3731         struct xfs_btree_cur    *cur,           /* btree cursor */
3732         int                     level,          /* level removing record from */
3733         int                     *stat)          /* fail/done/go-on */
3734 {
3735         struct xfs_btree_block  *block;         /* btree block */
3736         union xfs_btree_ptr     cptr;           /* current block ptr */
3737         struct xfs_buf          *bp;            /* buffer for block */
3738         int                     error;          /* error return value */
3739         int                     i;              /* loop counter */
3740         union xfs_btree_ptr     lptr;           /* left sibling block ptr */
3741         struct xfs_buf          *lbp;           /* left buffer pointer */
3742         struct xfs_btree_block  *left;          /* left btree block */
3743         int                     lrecs = 0;      /* left record count */
3744         int                     ptr;            /* key/record index */
3745         union xfs_btree_ptr     rptr;           /* right sibling block ptr */
3746         struct xfs_buf          *rbp;           /* right buffer pointer */
3747         struct xfs_btree_block  *right;         /* right btree block */
3748         struct xfs_btree_block  *rrblock;       /* right-right btree block */
3749         struct xfs_buf          *rrbp;          /* right-right buffer pointer */
3750         int                     rrecs = 0;      /* right record count */
3751         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
3752         int                     numrecs;        /* temporary numrec count */
3753
3754         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3755         XFS_BTREE_TRACE_ARGI(cur, level);
3756
3757         tcur = NULL;
3758
3759         /* Get the index of the entry being deleted, check for nothing there. */
3760         ptr = cur->bc_ptrs[level];
3761         if (ptr == 0) {
3762                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3763                 *stat = 0;
3764                 return 0;
3765         }
3766
3767         /* Get the buffer & block containing the record or key/ptr. */
3768         block = xfs_btree_get_block(cur, level, &bp);
3769         numrecs = xfs_btree_get_numrecs(block);
3770
3771 #ifdef DEBUG
3772         error = xfs_btree_check_block(cur, block, level, bp);
3773         if (error)
3774                 goto error0;
3775 #endif
3776
3777         /* Fail if we're off the end of the block. */
3778         if (ptr > numrecs) {
3779                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3780                 *stat = 0;
3781                 return 0;
3782         }
3783
3784         XFS_BTREE_STATS_INC(cur, delrec);
3785         XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3786
3787         /* Excise the entries being deleted. */
3788         if (level > 0) {
3789                 /* It's a nonleaf. operate on keys and ptrs */
3790                 union xfs_btree_key     *lkp;
3791                 union xfs_btree_ptr     *lpp;
3792
3793                 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3794                 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3795
3796 #ifdef DEBUG
3797                 for (i = 0; i < numrecs - ptr; i++) {
3798                         error = xfs_btree_check_ptr(cur, lpp, i, level);
3799                         if (error)
3800                                 goto error0;
3801                 }
3802 #endif
3803
3804                 if (ptr < numrecs) {
3805                         xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3806                         xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3807                         xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3808                         xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3809                 }
3810         } else {
3811                 /* It's a leaf. operate on records */
3812                 if (ptr < numrecs) {
3813                         xfs_btree_shift_recs(cur,
3814                                 xfs_btree_rec_addr(cur, ptr + 1, block),
3815                                 -1, numrecs - ptr);
3816                         xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3817                 }
3818         }
3819
3820         /*
3821          * Decrement and log the number of entries in the block.
3822          */
3823         xfs_btree_set_numrecs(block, --numrecs);
3824         xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3825
3826         /*
3827          * If we are tracking the last record in the tree and
3828          * we are at the far right edge of the tree, update it.
3829          */
3830         if (xfs_btree_is_lastrec(cur, block, level)) {
3831                 cur->bc_ops->update_lastrec(cur, block, NULL,
3832                                             ptr, LASTREC_DELREC);
3833         }
3834
3835         /*
3836          * We're at the root level.  First, shrink the root block in-memory.
3837          * Try to get rid of the next level down.  If we can't then there's
3838          * nothing left to do.
3839          */
3840         if (level == cur->bc_nlevels - 1) {
3841                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3842                         xfs_iroot_realloc(cur->bc_private.b.ip, -1,
3843                                           cur->bc_private.b.whichfork);
3844
3845                         error = xfs_btree_kill_iroot(cur);
3846                         if (error)
3847                                 goto error0;
3848
3849                         error = xfs_btree_dec_cursor(cur, level, stat);
3850                         if (error)
3851                                 goto error0;
3852                         *stat = 1;
3853                         return 0;
3854                 }
3855
3856                 /*
3857                  * If this is the root level, and there's only one entry left,
3858                  * and it's NOT the leaf level, then we can get rid of this
3859                  * level.
3860                  */
3861                 if (numrecs == 1 && level > 0) {
3862                         union xfs_btree_ptr     *pp;
3863                         /*
3864                          * pp is still set to the first pointer in the block.
3865                          * Make it the new root of the btree.
3866                          */
3867                         pp = xfs_btree_ptr_addr(cur, 1, block);
3868                         error = xfs_btree_kill_root(cur, bp, level, pp);
3869                         if (error)
3870                                 goto error0;
3871                 } else if (level > 0) {
3872                         error = xfs_btree_dec_cursor(cur, level, stat);
3873                         if (error)
3874                                 goto error0;
3875                 }
3876                 *stat = 1;
3877                 return 0;
3878         }
3879
3880         /*
3881          * If we deleted the leftmost entry in the block, update the
3882          * key values above us in the tree.
3883          */
3884         if (xfs_btree_needs_key_update(cur, ptr)) {
3885                 error = xfs_btree_update_keys(cur, level);
3886                 if (error)
3887                         goto error0;
3888         }
3889
3890         /*
3891          * If the number of records remaining in the block is at least
3892          * the minimum, we're done.
3893          */
3894         if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3895                 error = xfs_btree_dec_cursor(cur, level, stat);
3896                 if (error)
3897                         goto error0;
3898                 return 0;
3899         }
3900
3901         /*
3902          * Otherwise, we have to move some records around to keep the
3903          * tree balanced.  Look at the left and right sibling blocks to
3904          * see if we can re-balance by moving only one record.
3905          */
3906         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3907         xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3908
3909         if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3910                 /*
3911                  * One child of root, need to get a chance to copy its contents
3912                  * into the root and delete it. Can't go up to next level,
3913                  * there's nothing to delete there.
3914                  */
3915                 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3916                     xfs_btree_ptr_is_null(cur, &lptr) &&
3917                     level == cur->bc_nlevels - 2) {
3918                         error = xfs_btree_kill_iroot(cur);
3919                         if (!error)
3920                                 error = xfs_btree_dec_cursor(cur, level, stat);
3921                         if (error)
3922                                 goto error0;
3923                         return 0;
3924                 }
3925         }
3926
3927         ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3928                !xfs_btree_ptr_is_null(cur, &lptr));
3929
3930         /*
3931          * Duplicate the cursor so our btree manipulations here won't
3932          * disrupt the next level up.
3933          */
3934         error = xfs_btree_dup_cursor(cur, &tcur);
3935         if (error)
3936                 goto error0;
3937
3938         /*
3939          * If there's a right sibling, see if it's ok to shift an entry
3940          * out of it.
3941          */
3942         if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3943                 /*
3944                  * Move the temp cursor to the last entry in the next block.
3945                  * Actually any entry but the first would suffice.
3946                  */
3947                 i = xfs_btree_lastrec(tcur, level);
3948                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3949
3950                 error = xfs_btree_increment(tcur, level, &i);
3951                 if (error)
3952                         goto error0;
3953                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3954
3955                 i = xfs_btree_lastrec(tcur, level);
3956                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3957
3958                 /* Grab a pointer to the block. */
3959                 right = xfs_btree_get_block(tcur, level, &rbp);
3960 #ifdef DEBUG
3961                 error = xfs_btree_check_block(tcur, right, level, rbp);
3962                 if (error)
3963                         goto error0;
3964 #endif
3965                 /* Grab the current block number, for future use. */
3966                 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3967
3968                 /*
3969                  * If right block is full enough so that removing one entry
3970                  * won't make it too empty, and left-shifting an entry out
3971                  * of right to us works, we're done.
3972                  */
3973                 if (xfs_btree_get_numrecs(right) - 1 >=
3974                     cur->bc_ops->get_minrecs(tcur, level)) {
3975                         error = xfs_btree_lshift(tcur, level, &i);
3976                         if (error)
3977                                 goto error0;
3978                         if (i) {
3979                                 ASSERT(xfs_btree_get_numrecs(block) >=
3980                                        cur->bc_ops->get_minrecs(tcur, level));
3981
3982                                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3983                                 tcur = NULL;
3984
3985                                 error = xfs_btree_dec_cursor(cur, level, stat);
3986                                 if (error)
3987                                         goto error0;
3988                                 return 0;
3989                         }
3990                 }
3991
3992                 /*
3993                  * Otherwise, grab the number of records in right for
3994                  * future reference, and fix up the temp cursor to point
3995                  * to our block again (last record).
3996                  */
3997                 rrecs = xfs_btree_get_numrecs(right);
3998                 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3999                         i = xfs_btree_firstrec(tcur, level);
4000                         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
4001
4002                         error = xfs_btree_decrement(tcur, level, &i);
4003                         if (error)
4004                                 goto error0;
4005                         XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
4006                 }
4007         }
4008
4009         /*
4010          * If there's a left sibling, see if it's ok to shift an entry
4011          * out of it.
4012          */
4013         if (!xfs_btree_ptr_is_null(cur, &lptr)) {
4014                 /*
4015                  * Move the temp cursor to the first entry in the
4016                  * previous block.
4017                  */
4018                 i = xfs_btree_firstrec(tcur, level);
4019                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
4020
4021                 error = xfs_btree_decrement(tcur, level, &i);
4022                 if (error)
4023                         goto error0;
4024                 i = xfs_btree_firstrec(tcur, level);
4025                 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
4026
4027                 /* Grab a pointer to the block. */
4028                 left = xfs_btree_get_block(tcur, level, &lbp);
4029 #ifdef DEBUG
4030                 error = xfs_btree_check_block(cur, left, level, lbp);
4031                 if (error)
4032                         goto error0;
4033 #endif
4034                 /* Grab the current block number, for future use. */
4035                 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
4036
4037                 /*
4038                  * If left block is full enough so that removing one entry
4039                  * won't make it too empty, and right-shifting an entry out
4040                  * of left to us works, we're done.
4041                  */
4042                 if (xfs_btree_get_numrecs(left) - 1 >=
4043                     cur->bc_ops->get_minrecs(tcur, level)) {
4044                         error = xfs_btree_rshift(tcur, level, &i);
4045                         if (error)
4046                                 goto error0;
4047                         if (i) {
4048                                 ASSERT(xfs_btree_get_numrecs(block) >=
4049                                        cur->bc_ops->get_minrecs(tcur, level));
4050                                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4051                                 tcur = NULL;
4052                                 if (level == 0)
4053                                         cur->bc_ptrs[0]++;
4054                                 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
4055                                 *stat = 1;
4056                                 return 0;
4057                         }
4058                 }
4059
4060                 /*
4061                  * Otherwise, grab the number of records in right for
4062                  * future reference.
4063                  */
4064                 lrecs = xfs_btree_get_numrecs(left);
4065         }
4066
4067         /* Delete the temp cursor, we're done with it. */
4068         xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4069         tcur = NULL;
4070
4071         /* If here, we need to do a join to keep the tree balanced. */
4072         ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
4073
4074         if (!xfs_btree_ptr_is_null(cur, &lptr) &&
4075             lrecs + xfs_btree_get_numrecs(block) <=
4076                         cur->bc_ops->get_maxrecs(cur, level)) {
4077                 /*
4078                  * Set "right" to be the starting block,
4079                  * "left" to be the left neighbor.
4080                  */
4081                 rptr = cptr;
4082                 right = block;
4083                 rbp = bp;
4084                 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
4085                 if (error)
4086                         goto error0;
4087
4088         /*
4089          * If that won't work, see if we can join with the right neighbor block.
4090          */
4091         } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
4092                    rrecs + xfs_btree_get_numrecs(block) <=
4093                         cur->bc_ops->get_maxrecs(cur, level)) {
4094                 /*
4095                  * Set "left" to be the starting block,
4096                  * "right" to be the right neighbor.
4097                  */
4098                 lptr = cptr;
4099                 left = block;
4100                 lbp = bp;
4101                 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
4102                 if (error)
4103                         goto error0;
4104
4105         /*
4106          * Otherwise, we can't fix the imbalance.
4107          * Just return.  This is probably a logic error, but it's not fatal.
4108          */
4109         } else {
4110                 error = xfs_btree_dec_cursor(cur, level, stat);
4111                 if (error)
4112                         goto error0;
4113                 return 0;
4114         }
4115
4116         rrecs = xfs_btree_get_numrecs(right);
4117         lrecs = xfs_btree_get_numrecs(left);
4118
4119         /*
4120          * We're now going to join "left" and "right" by moving all the stuff
4121          * in "right" to "left" and deleting "right".
4122          */
4123         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
4124         if (level > 0) {
4125                 /* It's a non-leaf.  Move keys and pointers. */
4126                 union xfs_btree_key     *lkp;   /* left btree key */
4127                 union xfs_btree_ptr     *lpp;   /* left address pointer */
4128                 union xfs_btree_key     *rkp;   /* right btree key */
4129                 union xfs_btree_ptr     *rpp;   /* right address pointer */
4130
4131                 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
4132                 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
4133                 rkp = xfs_btree_key_addr(cur, 1, right);
4134                 rpp = xfs_btree_ptr_addr(cur, 1, right);
4135 #ifdef DEBUG
4136                 for (i = 1; i < rrecs; i++) {
4137                         error = xfs_btree_check_ptr(cur, rpp, i, level);
4138                         if (error)
4139                                 goto error0;
4140                 }
4141 #endif
4142                 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
4143                 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
4144
4145                 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
4146                 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
4147         } else {
4148                 /* It's a leaf.  Move records.  */
4149                 union xfs_btree_rec     *lrp;   /* left record pointer */
4150                 union xfs_btree_rec     *rrp;   /* right record pointer */
4151
4152                 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
4153                 rrp = xfs_btree_rec_addr(cur, 1, right);
4154
4155                 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
4156                 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
4157         }
4158
4159         XFS_BTREE_STATS_INC(cur, join);
4160
4161         /*
4162          * Fix up the number of records and right block pointer in the
4163          * surviving block, and log it.
4164          */
4165         xfs_btree_set_numrecs(left, lrecs + rrecs);
4166         xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB),
4167         xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4168         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
4169
4170         /* If there is a right sibling, point it to the remaining block. */
4171         xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4172         if (!xfs_btree_ptr_is_null(cur, &cptr)) {
4173                 error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp);
4174                 if (error)
4175                         goto error0;
4176                 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
4177                 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
4178         }
4179
4180         /* Free the deleted block. */
4181         error = xfs_btree_free_block(cur, rbp);
4182         if (error)
4183                 goto error0;
4184
4185         /*
4186          * If we joined with the left neighbor, set the buffer in the
4187          * cursor to the left block, and fix up the index.
4188          */
4189         if (bp != lbp) {
4190                 cur->bc_bufs[level] = lbp;
4191                 cur->bc_ptrs[level] += lrecs;
4192                 cur->bc_ra[level] = 0;
4193         }
4194         /*
4195          * If we joined with the right neighbor and there's a level above
4196          * us, increment the cursor at that level.
4197          */
4198         else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
4199                    (level + 1 < cur->bc_nlevels)) {
4200                 error = xfs_btree_increment(cur, level + 1, &i);
4201                 if (error)
4202                         goto error0;
4203         }
4204
4205         /*
4206          * Readjust the ptr at this level if it's not a leaf, since it's
4207          * still pointing at the deletion point, which makes the cursor
4208          * inconsistent.  If this makes the ptr 0, the caller fixes it up.
4209          * We can't use decrement because it would change the next level up.
4210          */
4211         if (level > 0)
4212                 cur->bc_ptrs[level]--;
4213
4214         /*
4215          * We combined blocks, so we have to update the parent keys if the
4216          * btree supports overlapped intervals.  However, bc_ptrs[level + 1]
4217          * points to the old block so that the caller knows which record to
4218          * delete.  Therefore, the caller must be savvy enough to call updkeys
4219          * for us if we return stat == 2.  The other exit points from this
4220          * function don't require deletions further up the tree, so they can
4221          * call updkeys directly.
4222          */
4223
4224         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
4225         /* Return value means the next level up has something to do. */
4226         *stat = 2;
4227         return 0;
4228
4229 error0:
4230         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
4231         if (tcur)
4232                 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
4233         return error;
4234 }
4235
4236 /*
4237  * Delete the record pointed to by cur.
4238  * The cursor refers to the place where the record was (could be inserted)
4239  * when the operation returns.
4240  */
4241 int                                     /* error */
4242 xfs_btree_delete(
4243         struct xfs_btree_cur    *cur,
4244         int                     *stat)  /* success/failure */
4245 {
4246         int                     error;  /* error return value */
4247         int                     level;
4248         int                     i;
4249         bool                    joined = false;
4250
4251         XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
4252
4253         /*
4254          * Go up the tree, starting at leaf level.
4255          *
4256          * If 2 is returned then a join was done; go to the next level.
4257          * Otherwise we are done.
4258          */
4259         for (level = 0, i = 2; i == 2; level++) {
4260                 error = xfs_btree_delrec(cur, level, &i);
4261                 if (error)
4262                         goto error0;
4263                 if (i == 2)
4264                         joined = true;
4265         }
4266
4267         /*
4268          * If we combined blocks as part of deleting the record, delrec won't
4269          * have updated the parent high keys so we have to do that here.
4270          */
4271         if (joined && (cur->bc_flags & XFS_BTREE_OVERLAPPING)) {
4272                 error = xfs_btree_updkeys_force(cur, 0);
4273                 if (error)
4274                         goto error0;
4275         }
4276
4277         if (i == 0) {
4278                 for (level = 1; level < cur->bc_nlevels; level++) {
4279                         if (cur->bc_ptrs[level] == 0) {
4280                                 error = xfs_btree_decrement(cur, level, &i);
4281                                 if (error)
4282                                         goto error0;
4283                                 break;
4284                         }
4285                 }
4286         }
4287
4288         XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
4289         *stat = i;
4290         return 0;
4291 error0:
4292         XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
4293         return error;
4294 }
4295
4296 /*
4297  * Get the data from the pointed-to record.
4298  */
4299 int                                     /* error */
4300 xfs_btree_get_rec(
4301         struct xfs_btree_cur    *cur,   /* btree cursor */
4302         union xfs_btree_rec     **recp, /* output: btree record */
4303         int                     *stat)  /* output: success/failure */
4304 {
4305         struct xfs_btree_block  *block; /* btree block */
4306         struct xfs_buf          *bp;    /* buffer pointer */
4307         int                     ptr;    /* record number */
4308 #ifdef DEBUG
4309         int                     error;  /* error return value */
4310 #endif
4311
4312         ptr = cur->bc_ptrs[0];
4313         block = xfs_btree_get_block(cur, 0, &bp);
4314
4315 #ifdef DEBUG
4316         error = xfs_btree_check_block(cur, block, 0, bp);
4317         if (error)
4318                 return error;
4319 #endif
4320
4321         /*
4322          * Off the right end or left end, return failure.
4323          */
4324         if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
4325                 *stat = 0;
4326                 return 0;
4327         }
4328
4329         /*
4330          * Point to the record and extract its data.
4331          */
4332         *recp = xfs_btree_rec_addr(cur, ptr, block);
4333         *stat = 1;
4334         return 0;
4335 }
4336
4337 /* Visit a block in a btree. */
4338 STATIC int
4339 xfs_btree_visit_block(
4340         struct xfs_btree_cur            *cur,
4341         int                             level,
4342         xfs_btree_visit_blocks_fn       fn,
4343         void                            *data)
4344 {
4345         struct xfs_btree_block          *block;
4346         struct xfs_buf                  *bp;
4347         union xfs_btree_ptr             rptr;
4348         int                             error;
4349
4350         /* do right sibling readahead */
4351         xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
4352         block = xfs_btree_get_block(cur, level, &bp);
4353
4354         /* process the block */
4355         error = fn(cur, level, data);
4356         if (error)
4357                 return error;
4358
4359         /* now read rh sibling block for next iteration */
4360         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4361         if (xfs_btree_ptr_is_null(cur, &rptr))
4362                 return -ENOENT;
4363
4364         return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
4365 }
4366
4367
4368 /* Visit every block in a btree. */
4369 int
4370 xfs_btree_visit_blocks(
4371         struct xfs_btree_cur            *cur,
4372         xfs_btree_visit_blocks_fn       fn,
4373         void                            *data)
4374 {
4375         union xfs_btree_ptr             lptr;
4376         int                             level;
4377         struct xfs_btree_block          *block = NULL;
4378         int                             error = 0;
4379
4380         cur->bc_ops->init_ptr_from_cur(cur, &lptr);
4381
4382         /* for each level */
4383         for (level = cur->bc_nlevels - 1; level >= 0; level--) {
4384                 /* grab the left hand block */
4385                 error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
4386                 if (error)
4387                         return error;
4388
4389                 /* readahead the left most block for the next level down */
4390                 if (level > 0) {
4391                         union xfs_btree_ptr     *ptr;
4392
4393                         ptr = xfs_btree_ptr_addr(cur, 1, block);
4394                         xfs_btree_readahead_ptr(cur, ptr, 1);
4395
4396                         /* save for the next iteration of the loop */
4397                         lptr = *ptr;
4398                 }
4399
4400                 /* for each buffer in the level */
4401                 do {
4402                         error = xfs_btree_visit_block(cur, level, fn, data);
4403                 } while (!error);
4404
4405                 if (error != -ENOENT)
4406                         return error;
4407         }
4408
4409         return 0;
4410 }
4411
4412 /*
4413  * Change the owner of a btree.
4414  *
4415  * The mechanism we use here is ordered buffer logging. Because we don't know
4416  * how many buffers were are going to need to modify, we don't really want to
4417  * have to make transaction reservations for the worst case of every buffer in a
4418  * full size btree as that may be more space that we can fit in the log....
4419  *
4420  * We do the btree walk in the most optimal manner possible - we have sibling
4421  * pointers so we can just walk all the blocks on each level from left to right
4422  * in a single pass, and then move to the next level and do the same. We can
4423  * also do readahead on the sibling pointers to get IO moving more quickly,
4424  * though for slow disks this is unlikely to make much difference to performance
4425  * as the amount of CPU work we have to do before moving to the next block is
4426  * relatively small.
4427  *
4428  * For each btree block that we load, modify the owner appropriately, set the
4429  * buffer as an ordered buffer and log it appropriately. We need to ensure that
4430  * we mark the region we change dirty so that if the buffer is relogged in
4431  * a subsequent transaction the changes we make here as an ordered buffer are
4432  * correctly relogged in that transaction.  If we are in recovery context, then
4433  * just queue the modified buffer as delayed write buffer so the transaction
4434  * recovery completion writes the changes to disk.
4435  */
4436 struct xfs_btree_block_change_owner_info {
4437         __uint64_t              new_owner;
4438         struct list_head        *buffer_list;
4439 };
4440
4441 static int
4442 xfs_btree_block_change_owner(
4443         struct xfs_btree_cur    *cur,
4444         int                     level,
4445         void                    *data)
4446 {
4447         struct xfs_btree_block_change_owner_info        *bbcoi = data;
4448         struct xfs_btree_block  *block;
4449         struct xfs_buf          *bp;
4450
4451         /* modify the owner */
4452         block = xfs_btree_get_block(cur, level, &bp);
4453         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
4454                 block->bb_u.l.bb_owner = cpu_to_be64(bbcoi->new_owner);
4455         else
4456                 block->bb_u.s.bb_owner = cpu_to_be32(bbcoi->new_owner);
4457
4458         /*
4459          * If the block is a root block hosted in an inode, we might not have a
4460          * buffer pointer here and we shouldn't attempt to log the change as the
4461          * information is already held in the inode and discarded when the root
4462          * block is formatted into the on-disk inode fork. We still change it,
4463          * though, so everything is consistent in memory.
4464          */
4465         if (bp) {
4466                 if (cur->bc_tp) {
4467                         xfs_trans_ordered_buf(cur->bc_tp, bp);
4468                         xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
4469                 } else {
4470                         xfs_buf_delwri_queue(bp, bbcoi->buffer_list);
4471                 }
4472         } else {
4473                 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
4474                 ASSERT(level == cur->bc_nlevels - 1);
4475         }
4476
4477         return 0;
4478 }
4479
4480 int
4481 xfs_btree_change_owner(
4482         struct xfs_btree_cur    *cur,
4483         __uint64_t              new_owner,
4484         struct list_head        *buffer_list)
4485 {
4486         struct xfs_btree_block_change_owner_info        bbcoi;
4487
4488         bbcoi.new_owner = new_owner;
4489         bbcoi.buffer_list = buffer_list;
4490
4491         return xfs_btree_visit_blocks(cur, xfs_btree_block_change_owner,
4492                         &bbcoi);
4493 }
4494
4495 /**
4496  * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format
4497  *                                    btree block
4498  *
4499  * @bp: buffer containing the btree block
4500  * @max_recs: pointer to the m_*_mxr max records field in the xfs mount
4501  * @pag_max_level: pointer to the per-ag max level field
4502  */
4503 bool
4504 xfs_btree_sblock_v5hdr_verify(
4505         struct xfs_buf          *bp)
4506 {
4507         struct xfs_mount        *mp = bp->b_target->bt_mount;
4508         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4509         struct xfs_perag        *pag = bp->b_pag;
4510
4511         if (!xfs_sb_version_hascrc(&mp->m_sb))
4512                 return false;
4513         if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
4514                 return false;
4515         if (block->bb_u.s.bb_blkno != cpu_to_be64(bp->b_bn))
4516                 return false;
4517         if (pag && be32_to_cpu(block->bb_u.s.bb_owner) != pag->pag_agno)
4518                 return false;
4519         return true;
4520 }
4521
4522 /**
4523  * xfs_btree_sblock_verify() -- verify a short-format btree block
4524  *
4525  * @bp: buffer containing the btree block
4526  * @max_recs: maximum records allowed in this btree node
4527  */
4528 bool
4529 xfs_btree_sblock_verify(
4530         struct xfs_buf          *bp,
4531         unsigned int            max_recs)
4532 {
4533         struct xfs_mount        *mp = bp->b_target->bt_mount;
4534         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4535
4536         /* numrecs verification */
4537         if (be16_to_cpu(block->bb_numrecs) > max_recs)
4538                 return false;
4539
4540         /* sibling pointer verification */
4541         if (!block->bb_u.s.bb_leftsib ||
4542             (be32_to_cpu(block->bb_u.s.bb_leftsib) >= mp->m_sb.sb_agblocks &&
4543              block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK)))
4544                 return false;
4545         if (!block->bb_u.s.bb_rightsib ||
4546             (be32_to_cpu(block->bb_u.s.bb_rightsib) >= mp->m_sb.sb_agblocks &&
4547              block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK)))
4548                 return false;
4549
4550         return true;
4551 }
4552
4553 /*
4554  * Calculate the number of btree levels needed to store a given number of
4555  * records in a short-format btree.
4556  */
4557 uint
4558 xfs_btree_compute_maxlevels(
4559         struct xfs_mount        *mp,
4560         uint                    *limits,
4561         unsigned long           len)
4562 {
4563         uint                    level;
4564         unsigned long           maxblocks;
4565
4566         maxblocks = (len + limits[0] - 1) / limits[0];
4567         for (level = 1; maxblocks > 1; level++)
4568                 maxblocks = (maxblocks + limits[1] - 1) / limits[1];
4569         return level;
4570 }
4571
4572 /*
4573  * Query a regular btree for all records overlapping a given interval.
4574  * Start with a LE lookup of the key of low_rec and return all records
4575  * until we find a record with a key greater than the key of high_rec.
4576  */
4577 STATIC int
4578 xfs_btree_simple_query_range(
4579         struct xfs_btree_cur            *cur,
4580         union xfs_btree_key             *low_key,
4581         union xfs_btree_key             *high_key,
4582         xfs_btree_query_range_fn        fn,
4583         void                            *priv)
4584 {
4585         union xfs_btree_rec             *recp;
4586         union xfs_btree_key             rec_key;
4587         __int64_t                       diff;
4588         int                             stat;
4589         bool                            firstrec = true;
4590         int                             error;
4591
4592         ASSERT(cur->bc_ops->init_high_key_from_rec);
4593         ASSERT(cur->bc_ops->diff_two_keys);
4594
4595         /*
4596          * Find the leftmost record.  The btree cursor must be set
4597          * to the low record used to generate low_key.
4598          */
4599         stat = 0;
4600         error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
4601         if (error)
4602                 goto out;
4603
4604         /* Nothing?  See if there's anything to the right. */
4605         if (!stat) {
4606                 error = xfs_btree_increment(cur, 0, &stat);
4607                 if (error)
4608                         goto out;
4609         }
4610
4611         while (stat) {
4612                 /* Find the record. */
4613                 error = xfs_btree_get_rec(cur, &recp, &stat);
4614                 if (error || !stat)
4615                         break;
4616
4617                 /* Skip if high_key(rec) < low_key. */
4618                 if (firstrec) {
4619                         cur->bc_ops->init_high_key_from_rec(&rec_key, recp);
4620                         firstrec = false;
4621                         diff = cur->bc_ops->diff_two_keys(cur, low_key,
4622                                         &rec_key);
4623                         if (diff > 0)
4624                                 goto advloop;
4625                 }
4626
4627                 /* Stop if high_key < low_key(rec). */
4628                 cur->bc_ops->init_key_from_rec(&rec_key, recp);
4629                 diff = cur->bc_ops->diff_two_keys(cur, &rec_key, high_key);
4630                 if (diff > 0)
4631                         break;
4632
4633                 /* Callback */
4634                 error = fn(cur, recp, priv);
4635                 if (error < 0 || error == XFS_BTREE_QUERY_RANGE_ABORT)
4636                         break;
4637
4638 advloop:
4639                 /* Move on to the next record. */
4640                 error = xfs_btree_increment(cur, 0, &stat);
4641                 if (error)
4642                         break;
4643         }
4644
4645 out:
4646         return error;
4647 }
4648
4649 /*
4650  * Query an overlapped interval btree for all records overlapping a given
4651  * interval.  This function roughly follows the algorithm given in
4652  * "Interval Trees" of _Introduction to Algorithms_, which is section
4653  * 14.3 in the 2nd and 3rd editions.
4654  *
4655  * First, generate keys for the low and high records passed in.
4656  *
4657  * For any leaf node, generate the high and low keys for the record.
4658  * If the record keys overlap with the query low/high keys, pass the
4659  * record to the function iterator.
4660  *
4661  * For any internal node, compare the low and high keys of each
4662  * pointer against the query low/high keys.  If there's an overlap,
4663  * follow the pointer.
4664  *
4665  * As an optimization, we stop scanning a block when we find a low key
4666  * that is greater than the query's high key.
4667  */
4668 STATIC int
4669 xfs_btree_overlapped_query_range(
4670         struct xfs_btree_cur            *cur,
4671         union xfs_btree_key             *low_key,
4672         union xfs_btree_key             *high_key,
4673         xfs_btree_query_range_fn        fn,
4674         void                            *priv)
4675 {
4676         union xfs_btree_ptr             ptr;
4677         union xfs_btree_ptr             *pp;
4678         union xfs_btree_key             rec_key;
4679         union xfs_btree_key             rec_hkey;
4680         union xfs_btree_key             *lkp;
4681         union xfs_btree_key             *hkp;
4682         union xfs_btree_rec             *recp;
4683         struct xfs_btree_block          *block;
4684         __int64_t                       ldiff;
4685         __int64_t                       hdiff;
4686         int                             level;
4687         struct xfs_buf                  *bp;
4688         int                             i;
4689         int                             error;
4690
4691         /* Load the root of the btree. */
4692         level = cur->bc_nlevels - 1;
4693         cur->bc_ops->init_ptr_from_cur(cur, &ptr);
4694         error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
4695         if (error)
4696                 return error;
4697         xfs_btree_get_block(cur, level, &bp);
4698         trace_xfs_btree_overlapped_query_range(cur, level, bp);
4699 #ifdef DEBUG
4700         error = xfs_btree_check_block(cur, block, level, bp);
4701         if (error)
4702                 goto out;
4703 #endif
4704         cur->bc_ptrs[level] = 1;
4705
4706         while (level < cur->bc_nlevels) {
4707                 block = xfs_btree_get_block(cur, level, &bp);
4708
4709                 /* End of node, pop back towards the root. */
4710                 if (cur->bc_ptrs[level] > be16_to_cpu(block->bb_numrecs)) {
4711 pop_up:
4712                         if (level < cur->bc_nlevels - 1)
4713                                 cur->bc_ptrs[level + 1]++;
4714                         level++;
4715                         continue;
4716                 }
4717
4718                 if (level == 0) {
4719                         /* Handle a leaf node. */
4720                         recp = xfs_btree_rec_addr(cur, cur->bc_ptrs[0], block);
4721
4722                         cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
4723                         ldiff = cur->bc_ops->diff_two_keys(cur, &rec_hkey,
4724                                         low_key);
4725
4726                         cur->bc_ops->init_key_from_rec(&rec_key, recp);
4727                         hdiff = cur->bc_ops->diff_two_keys(cur, high_key,
4728                                         &rec_key);
4729
4730                         /*
4731                          * If (record's high key >= query's low key) and
4732                          *    (query's high key >= record's low key), then
4733                          * this record overlaps the query range; callback.
4734                          */
4735                         if (ldiff >= 0 && hdiff >= 0) {
4736                                 error = fn(cur, recp, priv);
4737                                 if (error < 0 ||
4738                                     error == XFS_BTREE_QUERY_RANGE_ABORT)
4739                                         break;
4740                         } else if (hdiff < 0) {
4741                                 /* Record is larger than high key; pop. */
4742                                 goto pop_up;
4743                         }
4744                         cur->bc_ptrs[level]++;
4745                         continue;
4746                 }
4747
4748                 /* Handle an internal node. */
4749                 lkp = xfs_btree_key_addr(cur, cur->bc_ptrs[level], block);
4750                 hkp = xfs_btree_high_key_addr(cur, cur->bc_ptrs[level], block);
4751                 pp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[level], block);
4752
4753                 ldiff = cur->bc_ops->diff_two_keys(cur, hkp, low_key);
4754                 hdiff = cur->bc_ops->diff_two_keys(cur, high_key, lkp);
4755
4756                 /*
4757                  * If (pointer's high key >= query's low key) and
4758                  *    (query's high key >= pointer's low key), then
4759                  * this record overlaps the query range; follow pointer.
4760                  */
4761                 if (ldiff >= 0 && hdiff >= 0) {
4762                         level--;
4763                         error = xfs_btree_lookup_get_block(cur, level, pp,
4764                                         &block);
4765                         if (error)
4766                                 goto out;
4767                         xfs_btree_get_block(cur, level, &bp);
4768                         trace_xfs_btree_overlapped_query_range(cur, level, bp);
4769 #ifdef DEBUG
4770                         error = xfs_btree_check_block(cur, block, level, bp);
4771                         if (error)
4772                                 goto out;
4773 #endif
4774                         cur->bc_ptrs[level] = 1;
4775                         continue;
4776                 } else if (hdiff < 0) {
4777                         /* The low key is larger than the upper range; pop. */
4778                         goto pop_up;
4779                 }
4780                 cur->bc_ptrs[level]++;
4781         }
4782
4783 out:
4784         /*
4785          * If we don't end this function with the cursor pointing at a record
4786          * block, a subsequent non-error cursor deletion will not release
4787          * node-level buffers, causing a buffer leak.  This is quite possible
4788          * with a zero-results range query, so release the buffers if we
4789          * failed to return any results.
4790          */
4791         if (cur->bc_bufs[0] == NULL) {
4792                 for (i = 0; i < cur->bc_nlevels; i++) {
4793                         if (cur->bc_bufs[i]) {
4794                                 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
4795                                 cur->bc_bufs[i] = NULL;
4796                                 cur->bc_ptrs[i] = 0;
4797                                 cur->bc_ra[i] = 0;
4798                         }
4799                 }
4800         }
4801
4802         return error;
4803 }
4804
4805 /*
4806  * Query a btree for all records overlapping a given interval of keys.  The
4807  * supplied function will be called with each record found; return one of the
4808  * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
4809  * code.  This function returns XFS_BTREE_QUERY_RANGE_ABORT, zero, or a
4810  * negative error code.
4811  */
4812 int
4813 xfs_btree_query_range(
4814         struct xfs_btree_cur            *cur,
4815         union xfs_btree_irec            *low_rec,
4816         union xfs_btree_irec            *high_rec,
4817         xfs_btree_query_range_fn        fn,
4818         void                            *priv)
4819 {
4820         union xfs_btree_rec             rec;
4821         union xfs_btree_key             low_key;
4822         union xfs_btree_key             high_key;
4823
4824         /* Find the keys of both ends of the interval. */
4825         cur->bc_rec = *high_rec;
4826         cur->bc_ops->init_rec_from_cur(cur, &rec);
4827         cur->bc_ops->init_key_from_rec(&high_key, &rec);
4828
4829         cur->bc_rec = *low_rec;
4830         cur->bc_ops->init_rec_from_cur(cur, &rec);
4831         cur->bc_ops->init_key_from_rec(&low_key, &rec);
4832
4833         /* Enforce low key < high key. */
4834         if (cur->bc_ops->diff_two_keys(cur, &low_key, &high_key) > 0)
4835                 return -EINVAL;
4836
4837         if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING))
4838                 return xfs_btree_simple_query_range(cur, &low_key,
4839                                 &high_key, fn, priv);
4840         return xfs_btree_overlapped_query_range(cur, &low_key, &high_key,
4841                         fn, priv);
4842 }
4843
4844 /*
4845  * Calculate the number of blocks needed to store a given number of records
4846  * in a short-format (per-AG metadata) btree.
4847  */
4848 xfs_extlen_t
4849 xfs_btree_calc_size(
4850         struct xfs_mount        *mp,
4851         uint                    *limits,
4852         unsigned long long      len)
4853 {
4854         int                     level;
4855         int                     maxrecs;
4856         xfs_extlen_t            rval;
4857
4858         maxrecs = limits[0];
4859         for (level = 0, rval = 0; len > 1; level++) {
4860                 len += maxrecs - 1;
4861                 do_div(len, maxrecs);
4862                 maxrecs = limits[1];
4863                 rval += len;
4864         }
4865         return rval;
4866 }
4867
4868 static int
4869 xfs_btree_count_blocks_helper(
4870         struct xfs_btree_cur    *cur,
4871         int                     level,
4872         void                    *data)
4873 {
4874         xfs_extlen_t            *blocks = data;
4875         (*blocks)++;
4876
4877         return 0;
4878 }
4879
4880 /* Count the blocks in a btree and return the result in *blocks. */
4881 int
4882 xfs_btree_count_blocks(
4883         struct xfs_btree_cur    *cur,
4884         xfs_extlen_t            *blocks)
4885 {
4886         *blocks = 0;
4887         return xfs_btree_visit_blocks(cur, xfs_btree_count_blocks_helper,
4888                         blocks);
4889 }