fs/xfs/libxfs/xfs_btree.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   4  * All Rights Reserved.
   5  */
   6 #include "xfs.h"
   7 #include "xfs_fs.h"
   8 #include "xfs_shared.h"
   9 #include "xfs_format.h"
  10 #include "xfs_log_format.h"
  11 #include "xfs_trans_resv.h"
  12 #include "xfs_bit.h"
  13 #include "xfs_mount.h"
  14 #include "xfs_inode.h"
  15 #include "xfs_trans.h"
  16 #include "xfs_buf_item.h"
  17 #include "xfs_btree.h"
  18 #include "xfs_errortag.h"
  19 #include "xfs_error.h"
  20 #include "xfs_trace.h"
  21 #include "xfs_alloc.h"
  22 #include "xfs_log.h"
  23 #include "xfs_btree_staging.h"
  24
  25 /*
  26  * Cursor allocation zone.
  27  */
  28 kmem_zone_t     *xfs_btree_cur_zone;
  29
  30 /*
  31  * Btree magic numbers.
  32  */
  33 static const uint32_t xfs_magics[2][XFS_BTNUM_MAX] = {
  34         { XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, 0, XFS_BMAP_MAGIC, XFS_IBT_MAGIC,
  35           XFS_FIBT_MAGIC, 0 },
  36         { XFS_ABTB_CRC_MAGIC, XFS_ABTC_CRC_MAGIC, XFS_RMAP_CRC_MAGIC,
  37           XFS_BMAP_CRC_MAGIC, XFS_IBT_CRC_MAGIC, XFS_FIBT_CRC_MAGIC,
  38           XFS_REFC_CRC_MAGIC }
  39 };
  40
  41 uint32_t
  42 xfs_btree_magic(
  43         int                     crc,
  44         xfs_btnum_t             btnum)
  45 {
  46         uint32_t                magic = xfs_magics[crc][btnum];
  47
  48         /* Ensure we asked for crc for crc-only magics. */
  49         ASSERT(magic != 0);
  50         return magic;
  51 }
  52
  53 /*
  54  * Check a long btree block header.  Return the address of the failing check,
  55  * or NULL if everything is ok.
  56  */
  57 xfs_failaddr_t
  58 __xfs_btree_check_lblock(
  59         struct xfs_btree_cur    *cur,
  60         struct xfs_btree_block  *block,
  61         int                     level,
  62         struct xfs_buf          *bp)
  63 {
  64         struct xfs_mount        *mp = cur->bc_mp;
  65         xfs_btnum_t             btnum = cur->bc_btnum;
  66         int                     crc = xfs_sb_version_hascrc(&mp->m_sb);
  67
  68         if (crc) {
  69                 if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
  70                         return __this_address;
  71                 if (block->bb_u.l.bb_blkno !=
  72                     cpu_to_be64(bp ? bp->b_bn : XFS_BUF_DADDR_NULL))
  73                         return __this_address;
  74                 if (block->bb_u.l.bb_pad != cpu_to_be32(0))
  75                         return __this_address;
  76         }
  77
  78         if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum))
  79                 return __this_address;
  80         if (be16_to_cpu(block->bb_level) != level)
  81                 return __this_address;
  82         if (be16_to_cpu(block->bb_numrecs) >
  83             cur->bc_ops->get_maxrecs(cur, level))
  84                 return __this_address;
  85         if (block->bb_u.l.bb_leftsib != cpu_to_be64(NULLFSBLOCK) &&
  86             !xfs_btree_check_lptr(cur, be64_to_cpu(block->bb_u.l.bb_leftsib),
  87                         level + 1))
  88                 return __this_address;
  89         if (block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK) &&
  90             !xfs_btree_check_lptr(cur, be64_to_cpu(block->bb_u.l.bb_rightsib),
  91                         level + 1))
  92                 return __this_address;
  93
  94         return NULL;
  95 }
  96
  97 /* Check a long btree block header. */
  98 static int
  99 xfs_btree_check_lblock(
 100         struct xfs_btree_cur    *cur,
 101         struct xfs_btree_block  *block,
 102         int                     level,
 103         struct xfs_buf          *bp)
 104 {
 105         struct xfs_mount        *mp = cur->bc_mp;
 106         xfs_failaddr_t          fa;
 107
 108         fa = __xfs_btree_check_lblock(cur, block, level, bp);
 109         if (XFS_IS_CORRUPT(mp, fa != NULL) ||
 110             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BTREE_CHECK_LBLOCK)) {
 111                 if (bp)
 112                         trace_xfs_btree_corrupt(bp, _RET_IP_);
 113                 return -EFSCORRUPTED;
 114         }
 115         return 0;
 116 }
 117
 118 /*
 119  * Check a short btree block header.  Return the address of the failing check,
 120  * or NULL if everything is ok.
 121  */
 122 xfs_failaddr_t
 123 __xfs_btree_check_sblock(
 124         struct xfs_btree_cur    *cur,
 125         struct xfs_btree_block  *block,
 126         int                     level,
 127         struct xfs_buf          *bp)
 128 {
 129         struct xfs_mount        *mp = cur->bc_mp;
 130         xfs_btnum_t             btnum = cur->bc_btnum;
 131         int                     crc = xfs_sb_version_hascrc(&mp->m_sb);
 132
 133         if (crc) {
 134                 if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
 135                         return __this_address;
 136                 if (block->bb_u.s.bb_blkno !=
 137                     cpu_to_be64(bp ? bp->b_bn : XFS_BUF_DADDR_NULL))
 138                         return __this_address;
 139         }
 140
 141         if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum))
 142                 return __this_address;
 143         if (be16_to_cpu(block->bb_level) != level)
 144                 return __this_address;
 145         if (be16_to_cpu(block->bb_numrecs) >
 146             cur->bc_ops->get_maxrecs(cur, level))
 147                 return __this_address;
 148         if (block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK) &&
 149             !xfs_btree_check_sptr(cur, be32_to_cpu(block->bb_u.s.bb_leftsib),
 150                         level + 1))
 151                 return __this_address;
 152         if (block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK) &&
 153             !xfs_btree_check_sptr(cur, be32_to_cpu(block->bb_u.s.bb_rightsib),
 154                         level + 1))
 155                 return __this_address;
 156
 157         return NULL;
 158 }
 159
 160 /* Check a short btree block header. */
 161 STATIC int
 162 xfs_btree_check_sblock(
 163         struct xfs_btree_cur    *cur,
 164         struct xfs_btree_block  *block,
 165         int                     level,
 166         struct xfs_buf          *bp)
 167 {
 168         struct xfs_mount        *mp = cur->bc_mp;
 169         xfs_failaddr_t          fa;
 170
 171         fa = __xfs_btree_check_sblock(cur, block, level, bp);
 172         if (XFS_IS_CORRUPT(mp, fa != NULL) ||
 173             XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BTREE_CHECK_SBLOCK)) {
 174                 if (bp)
 175                         trace_xfs_btree_corrupt(bp, _RET_IP_);
 176                 return -EFSCORRUPTED;
 177         }
 178         return 0;
 179 }
 180
 181 /*
 182  * Debug routine: check that block header is ok.
 183  */
 184 int
 185 xfs_btree_check_block(
 186         struct xfs_btree_cur    *cur,   /* btree cursor */
 187         struct xfs_btree_block  *block, /* generic btree block pointer */
 188         int                     level,  /* level of the btree block */
 189         struct xfs_buf          *bp)    /* buffer containing block, if any */
 190 {
 191         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
 192                 return xfs_btree_check_lblock(cur, block, level, bp);
 193         else
 194                 return xfs_btree_check_sblock(cur, block, level, bp);
 195 }
 196
 197 /* Check that this long pointer is valid and points within the fs. */
 198 bool
 199 xfs_btree_check_lptr(
 200         struct xfs_btree_cur    *cur,
 201         xfs_fsblock_t           fsbno,
 202         int                     level)
 203 {
 204         if (level <= 0)
 205                 return false;
 206         return xfs_verify_fsbno(cur->bc_mp, fsbno);
 207 }
 208
 209 /* Check that this short pointer is valid and points within the AG. */
 210 bool
 211 xfs_btree_check_sptr(
 212         struct xfs_btree_cur    *cur,
 213         xfs_agblock_t           agbno,
 214         int                     level)
 215 {
 216         if (level <= 0)
 217                 return false;
 218         return xfs_verify_agbno(cur->bc_mp, cur->bc_ag.agno, agbno);
 219 }
 220
 221 /*
 222  * Check that a given (indexed) btree pointer at a certain level of a
 223  * btree is valid and doesn't point past where it should.
 224  */
 225 static int
 226 xfs_btree_check_ptr(
 227         struct xfs_btree_cur    *cur,
 228         union xfs_btree_ptr     *ptr,
 229         int                     index,
 230         int                     level)
 231 {
 232         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 233                 if (xfs_btree_check_lptr(cur, be64_to_cpu((&ptr->l)[index]),
 234                                 level))
 235                         return 0;
 236                 xfs_err(cur->bc_mp,
 237 "Inode %llu fork %d: Corrupt btree %d pointer at level %d index %d.",
 238                                 cur->bc_ino.ip->i_ino,
 239                                 cur->bc_ino.whichfork, cur->bc_btnum,
 240                                 level, index);
 241         } else {
 242                 if (xfs_btree_check_sptr(cur, be32_to_cpu((&ptr->s)[index]),
 243                                 level))
 244                         return 0;
 245                 xfs_err(cur->bc_mp,
 246 "AG %u: Corrupt btree %d pointer at level %d index %d.",
 247                                 cur->bc_ag.agno, cur->bc_btnum,
 248                                 level, index);
 249         }
 250
 251         return -EFSCORRUPTED;
 252 }
 253
 254 #ifdef DEBUG
 255 # define xfs_btree_debug_check_ptr      xfs_btree_check_ptr
 256 #else
 257 # define xfs_btree_debug_check_ptr(...) (0)
 258 #endif
 259
 260 /*
 261  * Calculate CRC on the whole btree block and stuff it into the
 262  * long-form btree header.
 263  *
 264  * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
 265  * it into the buffer so recovery knows what the last modification was that made
 266  * it to disk.
 267  */
 268 void
 269 xfs_btree_lblock_calc_crc(
 270         struct xfs_buf          *bp)
 271 {
 272         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 273         struct xfs_buf_log_item *bip = bp->b_log_item;
 274
 275         if (!xfs_sb_version_hascrc(&bp->b_mount->m_sb))
 276                 return;
 277         if (bip)
 278                 block->bb_u.l.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 279         xfs_buf_update_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
 280 }
 281
 282 bool
 283 xfs_btree_lblock_verify_crc(
 284         struct xfs_buf          *bp)
 285 {
 286         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 287         struct xfs_mount        *mp = bp->b_mount;
 288
 289         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 290                 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.l.bb_lsn)))
 291                         return false;
 292                 return xfs_buf_verify_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
 293         }
 294
 295         return true;
 296 }
 297
 298 /*
 299  * Calculate CRC on the whole btree block and stuff it into the
 300  * short-form btree header.
 301  *
 302  * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
 303  * it into the buffer so recovery knows what the last modification was that made
 304  * it to disk.
 305  */
 306 void
 307 xfs_btree_sblock_calc_crc(
 308         struct xfs_buf          *bp)
 309 {
 310         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 311         struct xfs_buf_log_item *bip = bp->b_log_item;
 312
 313         if (!xfs_sb_version_hascrc(&bp->b_mount->m_sb))
 314                 return;
 315         if (bip)
 316                 block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
 317         xfs_buf_update_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
 318 }
 319
 320 bool
 321 xfs_btree_sblock_verify_crc(
 322         struct xfs_buf          *bp)
 323 {
 324         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 325         struct xfs_mount        *mp = bp->b_mount;
 326
 327         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 328                 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.s.bb_lsn)))
 329                         return false;
 330                 return xfs_buf_verify_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
 331         }
 332
 333         return true;
 334 }
 335
 336 static int
 337 xfs_btree_free_block(
 338         struct xfs_btree_cur    *cur,
 339         struct xfs_buf          *bp)
 340 {
 341         int                     error;
 342
 343         error = cur->bc_ops->free_block(cur, bp);
 344         if (!error) {
 345                 xfs_trans_binval(cur->bc_tp, bp);
 346                 XFS_BTREE_STATS_INC(cur, free);
 347         }
 348         return error;
 349 }
 350
 351 /*
 352  * Delete the btree cursor.
 353  */
 354 void
 355 xfs_btree_del_cursor(
 356         xfs_btree_cur_t *cur,           /* btree cursor */
 357         int             error)          /* del because of error */
 358 {
 359         int             i;              /* btree level */
 360
 361         /*
 362          * Clear the buffer pointers, and release the buffers.
 363          * If we're doing this in the face of an error, we
 364          * need to make sure to inspect all of the entries
 365          * in the bc_bufs array for buffers to be unlocked.
 366          * This is because some of the btree code works from
 367          * level n down to 0, and if we get an error along
 368          * the way we won't have initialized all the entries
 369          * down to 0.
 370          */
 371         for (i = 0; i < cur->bc_nlevels; i++) {
 372                 if (cur->bc_bufs[i])
 373                         xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
 374                 else if (!error)
 375                         break;
 376         }
 377         /*
 378          * Can't free a bmap cursor without having dealt with the
 379          * allocated indirect blocks' accounting.
 380          */
 381         ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP ||
 382                cur->bc_ino.allocated == 0);
 383         /*
 384          * Free the cursor.
 385          */
 386         if (unlikely(cur->bc_flags & XFS_BTREE_STAGING))
 387                 kmem_free((void *)cur->bc_ops);
 388         kmem_cache_free(xfs_btree_cur_zone, cur);
 389 }
 390
 391 /*
 392  * Duplicate the btree cursor.
 393  * Allocate a new one, copy the record, re-get the buffers.
 394  */
 395 int                                     /* error */
 396 xfs_btree_dup_cursor(
 397         xfs_btree_cur_t *cur,           /* input cursor */
 398         xfs_btree_cur_t **ncur)         /* output cursor */
 399 {
 400         struct xfs_buf  *bp;            /* btree block's buffer pointer */
 401         int             error;          /* error return value */
 402         int             i;              /* level number of btree block */
 403         xfs_mount_t     *mp;            /* mount structure for filesystem */
 404         xfs_btree_cur_t *new;           /* new cursor value */
 405         xfs_trans_t     *tp;            /* transaction pointer, can be NULL */
 406
 407         tp = cur->bc_tp;
 408         mp = cur->bc_mp;
 409
 410         /*
 411          * Allocate a new cursor like the old one.
 412          */
 413         new = cur->bc_ops->dup_cursor(cur);
 414
 415         /*
 416          * Copy the record currently in the cursor.
 417          */
 418         new->bc_rec = cur->bc_rec;
 419
 420         /*
 421          * For each level current, re-get the buffer and copy the ptr value.
 422          */
 423         for (i = 0; i < new->bc_nlevels; i++) {
 424                 new->bc_ptrs[i] = cur->bc_ptrs[i];
 425                 new->bc_ra[i] = cur->bc_ra[i];
 426                 bp = cur->bc_bufs[i];
 427                 if (bp) {
 428                         error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
 429                                                    XFS_BUF_ADDR(bp), mp->m_bsize,
 430                                                    0, &bp,
 431                                                    cur->bc_ops->buf_ops);
 432                         if (error) {
 433                                 xfs_btree_del_cursor(new, error);
 434                                 *ncur = NULL;
 435                                 return error;
 436                         }
 437                 }
 438                 new->bc_bufs[i] = bp;
 439         }
 440         *ncur = new;
 441         return 0;
 442 }
 443
 444 /*
 445  * XFS btree block layout and addressing:
 446  *
 447  * There are two types of blocks in the btree: leaf and non-leaf blocks.
 448  *
 449  * The leaf record start with a header then followed by records containing
 450  * the values.  A non-leaf block also starts with the same header, and
 451  * then first contains lookup keys followed by an equal number of pointers
 452  * to the btree blocks at the previous level.
 453  *
 454  *              +--------+-------+-------+-------+-------+-------+-------+
 455  * Leaf:        | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
 456  *              +--------+-------+-------+-------+-------+-------+-------+
 457  *
 458  *              +--------+-------+-------+-------+-------+-------+-------+
 459  * Non-Leaf:    | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
 460  *              +--------+-------+-------+-------+-------+-------+-------+
 461  *
 462  * The header is called struct xfs_btree_block for reasons better left unknown
 463  * and comes in different versions for short (32bit) and long (64bit) block
 464  * pointers.  The record and key structures are defined by the btree instances
 465  * and opaque to the btree core.  The block pointers are simple disk endian
 466  * integers, available in a short (32bit) and long (64bit) variant.
 467  *
 468  * The helpers below calculate the offset of a given record, key or pointer
 469  * into a btree block (xfs_btree_*_offset) or return a pointer to the given
 470  * record, key or pointer (xfs_btree_*_addr).  Note that all addressing
 471  * inside the btree block is done using indices starting at one, not zero!
 472  *
 473  * If XFS_BTREE_OVERLAPPING is set, then this btree supports keys containing
 474  * overlapping intervals.  In such a tree, records are still sorted lowest to
 475  * highest and indexed by the smallest key value that refers to the record.
 476  * However, nodes are different: each pointer has two associated keys -- one
 477  * indexing the lowest key available in the block(s) below (the same behavior
 478  * as the key in a regular btree) and another indexing the highest key
 479  * available in the block(s) below.  Because records are /not/ sorted by the
 480  * highest key, all leaf block updates require us to compute the highest key
 481  * that matches any record in the leaf and to recursively update the high keys
 482  * in the nodes going further up in the tree, if necessary.  Nodes look like
 483  * this:
 484  *
 485  *              +--------+-----+-----+-----+-----+-----+-------+-------+-----+
 486  * Non-Leaf:    | header | lo1 | hi1 | lo2 | hi2 | ... | ptr 1 | ptr 2 | ... |
 487  *              +--------+-----+-----+-----+-----+-----+-------+-------+-----+
 488  *
 489  * To perform an interval query on an overlapped tree, perform the usual
 490  * depth-first search and use the low and high keys to decide if we can skip
 491  * that particular node.  If a leaf node is reached, return the records that
 492  * intersect the interval.  Note that an interval query may return numerous
 493  * entries.  For a non-overlapped tree, simply search for the record associated
 494  * with the lowest key and iterate forward until a non-matching record is
 495  * found.  Section 14.3 ("Interval Trees") of _Introduction to Algorithms_ by
 496  * Cormen, Leiserson, Rivest, and Stein (2nd or 3rd ed. only) discuss this in
 497  * more detail.
 498  *
 499  * Why do we care about overlapping intervals?  Let's say you have a bunch of
 500  * reverse mapping records on a reflink filesystem:
 501  *
 502  * 1: +- file A startblock B offset C length D -----------+
 503  * 2:      +- file E startblock F offset G length H --------------+
 504  * 3:      +- file I startblock F offset J length K --+
 505  * 4:                                                        +- file L... --+
 506  *
 507  * Now say we want to map block (B+D) into file A at offset (C+D).  Ideally,
 508  * we'd simply increment the length of record 1.  But how do we find the record
 509  * that ends at (B+D-1) (i.e. record 1)?  A LE lookup of (B+D-1) would return
 510  * record 3 because the keys are ordered first by startblock.  An interval
 511  * query would return records 1 and 2 because they both overlap (B+D-1), and
 512  * from that we can pick out record 1 as the appropriate left neighbor.
 513  *
 514  * In the non-overlapped case you can do a LE lookup and decrement the cursor
 515  * because a record's interval must end before the next record.
 516  */
 517
 518 /*
 519  * Return size of the btree block header for this btree instance.
 520  */
 521 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
 522 {
 523         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 524                 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
 525                         return XFS_BTREE_LBLOCK_CRC_LEN;
 526                 return XFS_BTREE_LBLOCK_LEN;
 527         }
 528         if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
 529                 return XFS_BTREE_SBLOCK_CRC_LEN;
 530         return XFS_BTREE_SBLOCK_LEN;
 531 }
 532
 533 /*
 534  * Return size of btree block pointers for this btree instance.
 535  */
 536 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
 537 {
 538         return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
 539                 sizeof(__be64) : sizeof(__be32);
 540 }
 541
 542 /*
 543  * Calculate offset of the n-th record in a btree block.
 544  */
 545 STATIC size_t
 546 xfs_btree_rec_offset(
 547         struct xfs_btree_cur    *cur,
 548         int                     n)
 549 {
 550         return xfs_btree_block_len(cur) +
 551                 (n - 1) * cur->bc_ops->rec_len;
 552 }
 553
 554 /*
 555  * Calculate offset of the n-th key in a btree block.
 556  */
 557 STATIC size_t
 558 xfs_btree_key_offset(
 559         struct xfs_btree_cur    *cur,
 560         int                     n)
 561 {
 562         return xfs_btree_block_len(cur) +
 563                 (n - 1) * cur->bc_ops->key_len;
 564 }
 565
 566 /*
 567  * Calculate offset of the n-th high key in a btree block.
 568  */
 569 STATIC size_t
 570 xfs_btree_high_key_offset(
 571         struct xfs_btree_cur    *cur,
 572         int                     n)
 573 {
 574         return xfs_btree_block_len(cur) +
 575                 (n - 1) * cur->bc_ops->key_len + (cur->bc_ops->key_len / 2);
 576 }
 577
 578 /*
 579  * Calculate offset of the n-th block pointer in a btree block.
 580  */
 581 STATIC size_t
 582 xfs_btree_ptr_offset(
 583         struct xfs_btree_cur    *cur,
 584         int                     n,
 585         int                     level)
 586 {
 587         return xfs_btree_block_len(cur) +
 588                 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
 589                 (n - 1) * xfs_btree_ptr_len(cur);
 590 }
 591
 592 /*
 593  * Return a pointer to the n-th record in the btree block.
 594  */
 595 union xfs_btree_rec *
 596 xfs_btree_rec_addr(
 597         struct xfs_btree_cur    *cur,
 598         int                     n,
 599         struct xfs_btree_block  *block)
 600 {
 601         return (union xfs_btree_rec *)
 602                 ((char *)block + xfs_btree_rec_offset(cur, n));
 603 }
 604
 605 /*
 606  * Return a pointer to the n-th key in the btree block.
 607  */
 608 union xfs_btree_key *
 609 xfs_btree_key_addr(
 610         struct xfs_btree_cur    *cur,
 611         int                     n,
 612         struct xfs_btree_block  *block)
 613 {
 614         return (union xfs_btree_key *)
 615                 ((char *)block + xfs_btree_key_offset(cur, n));
 616 }
 617
 618 /*
 619  * Return a pointer to the n-th high key in the btree block.
 620  */
 621 union xfs_btree_key *
 622 xfs_btree_high_key_addr(
 623         struct xfs_btree_cur    *cur,
 624         int                     n,
 625         struct xfs_btree_block  *block)
 626 {
 627         return (union xfs_btree_key *)
 628                 ((char *)block + xfs_btree_high_key_offset(cur, n));
 629 }
 630
 631 /*
 632  * Return a pointer to the n-th block pointer in the btree block.
 633  */
 634 union xfs_btree_ptr *
 635 xfs_btree_ptr_addr(
 636         struct xfs_btree_cur    *cur,
 637         int                     n,
 638         struct xfs_btree_block  *block)
 639 {
 640         int                     level = xfs_btree_get_level(block);
 641
 642         ASSERT(block->bb_level != 0);
 643
 644         return (union xfs_btree_ptr *)
 645                 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
 646 }
 647
 648 struct xfs_ifork *
 649 xfs_btree_ifork_ptr(
 650         struct xfs_btree_cur    *cur)
 651 {
 652         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
 653
 654         if (cur->bc_flags & XFS_BTREE_STAGING)
 655                 return cur->bc_ino.ifake->if_fork;
 656         return XFS_IFORK_PTR(cur->bc_ino.ip, cur->bc_ino.whichfork);
 657 }
 658
 659 /*
 660  * Get the root block which is stored in the inode.
 661  *
 662  * For now this btree implementation assumes the btree root is always
 663  * stored in the if_broot field of an inode fork.
 664  */
 665 STATIC struct xfs_btree_block *
 666 xfs_btree_get_iroot(
 667         struct xfs_btree_cur    *cur)
 668 {
 669         struct xfs_ifork        *ifp = xfs_btree_ifork_ptr(cur);
 670
 671         return (struct xfs_btree_block *)ifp->if_broot;
 672 }
 673
 674 /*
 675  * Retrieve the block pointer from the cursor at the given level.
 676  * This may be an inode btree root or from a buffer.
 677  */
 678 struct xfs_btree_block *                /* generic btree block pointer */
 679 xfs_btree_get_block(
 680         struct xfs_btree_cur    *cur,   /* btree cursor */
 681         int                     level,  /* level in btree */
 682         struct xfs_buf          **bpp)  /* buffer containing the block */
 683 {
 684         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
 685             (level == cur->bc_nlevels - 1)) {
 686                 *bpp = NULL;
 687                 return xfs_btree_get_iroot(cur);
 688         }
 689
 690         *bpp = cur->bc_bufs[level];
 691         return XFS_BUF_TO_BLOCK(*bpp);
 692 }
 693
 694 /*
 695  * Change the cursor to point to the first record at the given level.
 696  * Other levels are unaffected.
 697  */
 698 STATIC int                              /* success=1, failure=0 */
 699 xfs_btree_firstrec(
 700         xfs_btree_cur_t         *cur,   /* btree cursor */
 701         int                     level)  /* level to change */
 702 {
 703         struct xfs_btree_block  *block; /* generic btree block pointer */
 704         struct xfs_buf          *bp;    /* buffer containing block */
 705
 706         /*
 707          * Get the block pointer for this level.
 708          */
 709         block = xfs_btree_get_block(cur, level, &bp);
 710         if (xfs_btree_check_block(cur, block, level, bp))
 711                 return 0;
 712         /*
 713          * It's empty, there is no such record.
 714          */
 715         if (!block->bb_numrecs)
 716                 return 0;
 717         /*
 718          * Set the ptr value to 1, that's the first record/key.
 719          */
 720         cur->bc_ptrs[level] = 1;
 721         return 1;
 722 }
 723
 724 /*
 725  * Change the cursor to point to the last record in the current block
 726  * at the given level.  Other levels are unaffected.
 727  */
 728 STATIC int                              /* success=1, failure=0 */
 729 xfs_btree_lastrec(
 730         xfs_btree_cur_t         *cur,   /* btree cursor */
 731         int                     level)  /* level to change */
 732 {
 733         struct xfs_btree_block  *block; /* generic btree block pointer */
 734         struct xfs_buf          *bp;    /* buffer containing block */
 735
 736         /*
 737          * Get the block pointer for this level.
 738          */
 739         block = xfs_btree_get_block(cur, level, &bp);
 740         if (xfs_btree_check_block(cur, block, level, bp))
 741                 return 0;
 742         /*
 743          * It's empty, there is no such record.
 744          */
 745         if (!block->bb_numrecs)
 746                 return 0;
 747         /*
 748          * Set the ptr value to numrecs, that's the last record/key.
 749          */
 750         cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
 751         return 1;
 752 }
 753
 754 /*
 755  * Compute first and last byte offsets for the fields given.
 756  * Interprets the offsets table, which contains struct field offsets.
 757  */
 758 void
 759 xfs_btree_offsets(
 760         int64_t         fields,         /* bitmask of fields */
 761         const short     *offsets,       /* table of field offsets */
 762         int             nbits,          /* number of bits to inspect */
 763         int             *first,         /* output: first byte offset */
 764         int             *last)          /* output: last byte offset */
 765 {
 766         int             i;              /* current bit number */
 767         int64_t         imask;          /* mask for current bit number */
 768
 769         ASSERT(fields != 0);
 770         /*
 771          * Find the lowest bit, so the first byte offset.
 772          */
 773         for (i = 0, imask = 1LL; ; i++, imask <<= 1) {
 774                 if (imask & fields) {
 775                         *first = offsets[i];
 776                         break;
 777                 }
 778         }
 779         /*
 780          * Find the highest bit, so the last byte offset.
 781          */
 782         for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) {
 783                 if (imask & fields) {
 784                         *last = offsets[i + 1] - 1;
 785                         break;
 786                 }
 787         }
 788 }
 789
 790 /*
 791  * Get a buffer for the block, return it read in.
 792  * Long-form addressing.
 793  */
 794 int
 795 xfs_btree_read_bufl(
 796         struct xfs_mount        *mp,            /* file system mount point */
 797         struct xfs_trans        *tp,            /* transaction pointer */
 798         xfs_fsblock_t           fsbno,          /* file system block number */
 799         struct xfs_buf          **bpp,          /* buffer for fsbno */
 800         int                     refval,         /* ref count value for buffer */
 801         const struct xfs_buf_ops *ops)
 802 {
 803         struct xfs_buf          *bp;            /* return value */
 804         xfs_daddr_t             d;              /* real disk block address */
 805         int                     error;
 806
 807         if (!xfs_verify_fsbno(mp, fsbno))
 808                 return -EFSCORRUPTED;
 809         d = XFS_FSB_TO_DADDR(mp, fsbno);
 810         error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
 811                                    mp->m_bsize, 0, &bp, ops);
 812         if (error)
 813                 return error;
 814         if (bp)
 815                 xfs_buf_set_ref(bp, refval);
 816         *bpp = bp;
 817         return 0;
 818 }
 819
 820 /*
 821  * Read-ahead the block, don't wait for it, don't return a buffer.
 822  * Long-form addressing.
 823  */
 824 /* ARGSUSED */
 825 void
 826 xfs_btree_reada_bufl(
 827         struct xfs_mount        *mp,            /* file system mount point */
 828         xfs_fsblock_t           fsbno,          /* file system block number */
 829         xfs_extlen_t            count,          /* count of filesystem blocks */
 830         const struct xfs_buf_ops *ops)
 831 {
 832         xfs_daddr_t             d;
 833
 834         ASSERT(fsbno != NULLFSBLOCK);
 835         d = XFS_FSB_TO_DADDR(mp, fsbno);
 836         xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
 837 }
 838
 839 /*
 840  * Read-ahead the block, don't wait for it, don't return a buffer.
 841  * Short-form addressing.
 842  */
 843 /* ARGSUSED */
 844 void
 845 xfs_btree_reada_bufs(
 846         struct xfs_mount        *mp,            /* file system mount point */
 847         xfs_agnumber_t          agno,           /* allocation group number */
 848         xfs_agblock_t           agbno,          /* allocation group block number */
 849         xfs_extlen_t            count,          /* count of filesystem blocks */
 850         const struct xfs_buf_ops *ops)
 851 {
 852         xfs_daddr_t             d;
 853
 854         ASSERT(agno != NULLAGNUMBER);
 855         ASSERT(agbno != NULLAGBLOCK);
 856         d = XFS_AGB_TO_DADDR(mp, agno, agbno);
 857         xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
 858 }
 859
 860 STATIC int
 861 xfs_btree_readahead_lblock(
 862         struct xfs_btree_cur    *cur,
 863         int                     lr,
 864         struct xfs_btree_block  *block)
 865 {
 866         int                     rval = 0;
 867         xfs_fsblock_t           left = be64_to_cpu(block->bb_u.l.bb_leftsib);
 868         xfs_fsblock_t           right = be64_to_cpu(block->bb_u.l.bb_rightsib);
 869
 870         if ((lr & XFS_BTCUR_LEFTRA) && left != NULLFSBLOCK) {
 871                 xfs_btree_reada_bufl(cur->bc_mp, left, 1,
 872                                      cur->bc_ops->buf_ops);
 873                 rval++;
 874         }
 875
 876         if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLFSBLOCK) {
 877                 xfs_btree_reada_bufl(cur->bc_mp, right, 1,
 878                                      cur->bc_ops->buf_ops);
 879                 rval++;
 880         }
 881
 882         return rval;
 883 }
 884
 885 STATIC int
 886 xfs_btree_readahead_sblock(
 887         struct xfs_btree_cur    *cur,
 888         int                     lr,
 889         struct xfs_btree_block *block)
 890 {
 891         int                     rval = 0;
 892         xfs_agblock_t           left = be32_to_cpu(block->bb_u.s.bb_leftsib);
 893         xfs_agblock_t           right = be32_to_cpu(block->bb_u.s.bb_rightsib);
 894
 895
 896         if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
 897                 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_ag.agno,
 898                                      left, 1, cur->bc_ops->buf_ops);
 899                 rval++;
 900         }
 901
 902         if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
 903                 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_ag.agno,
 904                                      right, 1, cur->bc_ops->buf_ops);
 905                 rval++;
 906         }
 907
 908         return rval;
 909 }
 910
 911 /*
 912  * Read-ahead btree blocks, at the given level.
 913  * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
 914  */
 915 STATIC int
 916 xfs_btree_readahead(
 917         struct xfs_btree_cur    *cur,           /* btree cursor */
 918         int                     lev,            /* level in btree */
 919         int                     lr)             /* left/right bits */
 920 {
 921         struct xfs_btree_block  *block;
 922
 923         /*
 924          * No readahead needed if we are at the root level and the
 925          * btree root is stored in the inode.
 926          */
 927         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
 928             (lev == cur->bc_nlevels - 1))
 929                 return 0;
 930
 931         if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
 932                 return 0;
 933
 934         cur->bc_ra[lev] |= lr;
 935         block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
 936
 937         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
 938                 return xfs_btree_readahead_lblock(cur, lr, block);
 939         return xfs_btree_readahead_sblock(cur, lr, block);
 940 }
 941
 942 STATIC int
 943 xfs_btree_ptr_to_daddr(
 944         struct xfs_btree_cur    *cur,
 945         union xfs_btree_ptr     *ptr,
 946         xfs_daddr_t             *daddr)
 947 {
 948         xfs_fsblock_t           fsbno;
 949         xfs_agblock_t           agbno;
 950         int                     error;
 951
 952         error = xfs_btree_check_ptr(cur, ptr, 0, 1);
 953         if (error)
 954                 return error;
 955
 956         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
 957                 fsbno = be64_to_cpu(ptr->l);
 958                 *daddr = XFS_FSB_TO_DADDR(cur->bc_mp, fsbno);
 959         } else {
 960                 agbno = be32_to_cpu(ptr->s);
 961                 *daddr = XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_ag.agno,
 962                                 agbno);
 963         }
 964
 965         return 0;
 966 }
 967
 968 /*
 969  * Readahead @count btree blocks at the given @ptr location.
 970  *
 971  * We don't need to care about long or short form btrees here as we have a
 972  * method of converting the ptr directly to a daddr available to us.
 973  */
 974 STATIC void
 975 xfs_btree_readahead_ptr(
 976         struct xfs_btree_cur    *cur,
 977         union xfs_btree_ptr     *ptr,
 978         xfs_extlen_t            count)
 979 {
 980         xfs_daddr_t             daddr;
 981
 982         if (xfs_btree_ptr_to_daddr(cur, ptr, &daddr))
 983                 return;
 984         xfs_buf_readahead(cur->bc_mp->m_ddev_targp, daddr,
 985                           cur->bc_mp->m_bsize * count, cur->bc_ops->buf_ops);
 986 }
 987
 988 /*
 989  * Set the buffer for level "lev" in the cursor to bp, releasing
 990  * any previous buffer.
 991  */
 992 STATIC void
 993 xfs_btree_setbuf(
 994         xfs_btree_cur_t         *cur,   /* btree cursor */
 995         int                     lev,    /* level in btree */
 996         struct xfs_buf          *bp)    /* new buffer to set */
 997 {
 998         struct xfs_btree_block  *b;     /* btree block */
 999
1000         if (cur->bc_bufs[lev])
1001                 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
1002         cur->bc_bufs[lev] = bp;
1003         cur->bc_ra[lev] = 0;
1004
1005         b = XFS_BUF_TO_BLOCK(bp);
1006         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1007                 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK))
1008                         cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1009                 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK))
1010                         cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1011         } else {
1012                 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
1013                         cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1014                 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
1015                         cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1016         }
1017 }
1018
1019 bool
1020 xfs_btree_ptr_is_null(
1021         struct xfs_btree_cur    *cur,
1022         union xfs_btree_ptr     *ptr)
1023 {
1024         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1025                 return ptr->l == cpu_to_be64(NULLFSBLOCK);
1026         else
1027                 return ptr->s == cpu_to_be32(NULLAGBLOCK);
1028 }
1029
1030 void
1031 xfs_btree_set_ptr_null(
1032         struct xfs_btree_cur    *cur,
1033         union xfs_btree_ptr     *ptr)
1034 {
1035         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1036                 ptr->l = cpu_to_be64(NULLFSBLOCK);
1037         else
1038                 ptr->s = cpu_to_be32(NULLAGBLOCK);
1039 }
1040
1041 /*
1042  * Get/set/init sibling pointers
1043  */
1044 void
1045 xfs_btree_get_sibling(
1046         struct xfs_btree_cur    *cur,
1047         struct xfs_btree_block  *block,
1048         union xfs_btree_ptr     *ptr,
1049         int                     lr)
1050 {
1051         ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1052
1053         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1054                 if (lr == XFS_BB_RIGHTSIB)
1055                         ptr->l = block->bb_u.l.bb_rightsib;
1056                 else
1057                         ptr->l = block->bb_u.l.bb_leftsib;
1058         } else {
1059                 if (lr == XFS_BB_RIGHTSIB)
1060                         ptr->s = block->bb_u.s.bb_rightsib;
1061                 else
1062                         ptr->s = block->bb_u.s.bb_leftsib;
1063         }
1064 }
1065
1066 void
1067 xfs_btree_set_sibling(
1068         struct xfs_btree_cur    *cur,
1069         struct xfs_btree_block  *block,
1070         union xfs_btree_ptr     *ptr,
1071         int                     lr)
1072 {
1073         ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1074
1075         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1076                 if (lr == XFS_BB_RIGHTSIB)
1077                         block->bb_u.l.bb_rightsib = ptr->l;
1078                 else
1079                         block->bb_u.l.bb_leftsib = ptr->l;
1080         } else {
1081                 if (lr == XFS_BB_RIGHTSIB)
1082                         block->bb_u.s.bb_rightsib = ptr->s;
1083                 else
1084                         block->bb_u.s.bb_leftsib = ptr->s;
1085         }
1086 }
1087
1088 void
1089 xfs_btree_init_block_int(
1090         struct xfs_mount        *mp,
1091         struct xfs_btree_block  *buf,
1092         xfs_daddr_t             blkno,
1093         xfs_btnum_t             btnum,
1094         __u16                   level,
1095         __u16                   numrecs,
1096         __u64                   owner,
1097         unsigned int            flags)
1098 {
1099         int                     crc = xfs_sb_version_hascrc(&mp->m_sb);
1100         __u32                   magic = xfs_btree_magic(crc, btnum);
1101
1102         buf->bb_magic = cpu_to_be32(magic);
1103         buf->bb_level = cpu_to_be16(level);
1104         buf->bb_numrecs = cpu_to_be16(numrecs);
1105
1106         if (flags & XFS_BTREE_LONG_PTRS) {
1107                 buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK);
1108                 buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK);
1109                 if (crc) {
1110                         buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
1111                         buf->bb_u.l.bb_owner = cpu_to_be64(owner);
1112                         uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid);
1113                         buf->bb_u.l.bb_pad = 0;
1114                         buf->bb_u.l.bb_lsn = 0;
1115                 }
1116         } else {
1117                 /* owner is a 32 bit value on short blocks */
1118                 __u32 __owner = (__u32)owner;
1119
1120                 buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1121                 buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1122                 if (crc) {
1123                         buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
1124                         buf->bb_u.s.bb_owner = cpu_to_be32(__owner);
1125                         uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid);
1126                         buf->bb_u.s.bb_lsn = 0;
1127                 }
1128         }
1129 }
1130
1131 void
1132 xfs_btree_init_block(
1133         struct xfs_mount *mp,
1134         struct xfs_buf  *bp,
1135         xfs_btnum_t     btnum,
1136         __u16           level,
1137         __u16           numrecs,
1138         __u64           owner)
1139 {
1140         xfs_btree_init_block_int(mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1141                                  btnum, level, numrecs, owner, 0);
1142 }
1143
1144 void
1145 xfs_btree_init_block_cur(
1146         struct xfs_btree_cur    *cur,
1147         struct xfs_buf          *bp,
1148         int                     level,
1149         int                     numrecs)
1150 {
1151         __u64                   owner;
1152
1153         /*
1154          * we can pull the owner from the cursor right now as the different
1155          * owners align directly with the pointer size of the btree. This may
1156          * change in future, but is safe for current users of the generic btree
1157          * code.
1158          */
1159         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1160                 owner = cur->bc_ino.ip->i_ino;
1161         else
1162                 owner = cur->bc_ag.agno;
1163
1164         xfs_btree_init_block_int(cur->bc_mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1165                                  cur->bc_btnum, level, numrecs,
1166                                  owner, cur->bc_flags);
1167 }
1168
1169 /*
1170  * Return true if ptr is the last record in the btree and
1171  * we need to track updates to this record.  The decision
1172  * will be further refined in the update_lastrec method.
1173  */
1174 STATIC int
1175 xfs_btree_is_lastrec(
1176         struct xfs_btree_cur    *cur,
1177         struct xfs_btree_block  *block,
1178         int                     level)
1179 {
1180         union xfs_btree_ptr     ptr;
1181
1182         if (level > 0)
1183                 return 0;
1184         if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
1185                 return 0;
1186
1187         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1188         if (!xfs_btree_ptr_is_null(cur, &ptr))
1189                 return 0;
1190         return 1;
1191 }
1192
1193 STATIC void
1194 xfs_btree_buf_to_ptr(
1195         struct xfs_btree_cur    *cur,
1196         struct xfs_buf          *bp,
1197         union xfs_btree_ptr     *ptr)
1198 {
1199         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1200                 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1201                                         XFS_BUF_ADDR(bp)));
1202         else {
1203                 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
1204                                         XFS_BUF_ADDR(bp)));
1205         }
1206 }
1207
1208 STATIC void
1209 xfs_btree_set_refs(
1210         struct xfs_btree_cur    *cur,
1211         struct xfs_buf          *bp)
1212 {
1213         switch (cur->bc_btnum) {
1214         case XFS_BTNUM_BNO:
1215         case XFS_BTNUM_CNT:
1216                 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
1217                 break;
1218         case XFS_BTNUM_INO:
1219         case XFS_BTNUM_FINO:
1220                 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
1221                 break;
1222         case XFS_BTNUM_BMAP:
1223                 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
1224                 break;
1225         case XFS_BTNUM_RMAP:
1226                 xfs_buf_set_ref(bp, XFS_RMAP_BTREE_REF);
1227                 break;
1228         case XFS_BTNUM_REFC:
1229                 xfs_buf_set_ref(bp, XFS_REFC_BTREE_REF);
1230                 break;
1231         default:
1232                 ASSERT(0);
1233         }
1234 }
1235
1236 int
1237 xfs_btree_get_buf_block(
1238         struct xfs_btree_cur    *cur,
1239         union xfs_btree_ptr     *ptr,
1240         struct xfs_btree_block  **block,
1241         struct xfs_buf          **bpp)
1242 {
1243         struct xfs_mount        *mp = cur->bc_mp;
1244         xfs_daddr_t             d;
1245         int                     error;
1246
1247         error = xfs_btree_ptr_to_daddr(cur, ptr, &d);
1248         if (error)
1249                 return error;
1250         error = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d, mp->m_bsize,
1251                         0, bpp);
1252         if (error)
1253                 return error;
1254
1255         (*bpp)->b_ops = cur->bc_ops->buf_ops;
1256         *block = XFS_BUF_TO_BLOCK(*bpp);
1257         return 0;
1258 }
1259
1260 /*
1261  * Read in the buffer at the given ptr and return the buffer and
1262  * the block pointer within the buffer.
1263  */
1264 STATIC int
1265 xfs_btree_read_buf_block(
1266         struct xfs_btree_cur    *cur,
1267         union xfs_btree_ptr     *ptr,
1268         int                     flags,
1269         struct xfs_btree_block  **block,
1270         struct xfs_buf          **bpp)
1271 {
1272         struct xfs_mount        *mp = cur->bc_mp;
1273         xfs_daddr_t             d;
1274         int                     error;
1275
1276         /* need to sort out how callers deal with failures first */
1277         ASSERT(!(flags & XBF_TRYLOCK));
1278
1279         error = xfs_btree_ptr_to_daddr(cur, ptr, &d);
1280         if (error)
1281                 return error;
1282         error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
1283                                    mp->m_bsize, flags, bpp,
1284                                    cur->bc_ops->buf_ops);
1285         if (error)
1286                 return error;
1287
1288         xfs_btree_set_refs(cur, *bpp);
1289         *block = XFS_BUF_TO_BLOCK(*bpp);
1290         return 0;
1291 }
1292
1293 /*
1294  * Copy keys from one btree block to another.
1295  */
1296 void
1297 xfs_btree_copy_keys(
1298         struct xfs_btree_cur    *cur,
1299         union xfs_btree_key     *dst_key,
1300         union xfs_btree_key     *src_key,
1301         int                     numkeys)
1302 {
1303         ASSERT(numkeys >= 0);
1304         memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1305 }
1306
1307 /*
1308  * Copy records from one btree block to another.
1309  */
1310 STATIC void
1311 xfs_btree_copy_recs(
1312         struct xfs_btree_cur    *cur,
1313         union xfs_btree_rec     *dst_rec,
1314         union xfs_btree_rec     *src_rec,
1315         int                     numrecs)
1316 {
1317         ASSERT(numrecs >= 0);
1318         memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1319 }
1320
1321 /*
1322  * Copy block pointers from one btree block to another.
1323  */
1324 void
1325 xfs_btree_copy_ptrs(
1326         struct xfs_btree_cur    *cur,
1327         union xfs_btree_ptr     *dst_ptr,
1328         const union xfs_btree_ptr *src_ptr,
1329         int                     numptrs)
1330 {
1331         ASSERT(numptrs >= 0);
1332         memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1333 }
1334
1335 /*
1336  * Shift keys one index left/right inside a single btree block.
1337  */
1338 STATIC void
1339 xfs_btree_shift_keys(
1340         struct xfs_btree_cur    *cur,
1341         union xfs_btree_key     *key,
1342         int                     dir,
1343         int                     numkeys)
1344 {
1345         char                    *dst_key;
1346
1347         ASSERT(numkeys >= 0);
1348         ASSERT(dir == 1 || dir == -1);
1349
1350         dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1351         memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1352 }
1353
1354 /*
1355  * Shift records one index left/right inside a single btree block.
1356  */
1357 STATIC void
1358 xfs_btree_shift_recs(
1359         struct xfs_btree_cur    *cur,
1360         union xfs_btree_rec     *rec,
1361         int                     dir,
1362         int                     numrecs)
1363 {
1364         char                    *dst_rec;
1365
1366         ASSERT(numrecs >= 0);
1367         ASSERT(dir == 1 || dir == -1);
1368
1369         dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1370         memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1371 }
1372
1373 /*
1374  * Shift block pointers one index left/right inside a single btree block.
1375  */
1376 STATIC void
1377 xfs_btree_shift_ptrs(
1378         struct xfs_btree_cur    *cur,
1379         union xfs_btree_ptr     *ptr,
1380         int                     dir,
1381         int                     numptrs)
1382 {
1383         char                    *dst_ptr;
1384
1385         ASSERT(numptrs >= 0);
1386         ASSERT(dir == 1 || dir == -1);
1387
1388         dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1389         memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1390 }
1391
1392 /*
1393  * Log key values from the btree block.
1394  */
1395 STATIC void
1396 xfs_btree_log_keys(
1397         struct xfs_btree_cur    *cur,
1398         struct xfs_buf          *bp,
1399         int                     first,
1400         int                     last)
1401 {
1402
1403         if (bp) {
1404                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1405                 xfs_trans_log_buf(cur->bc_tp, bp,
1406                                   xfs_btree_key_offset(cur, first),
1407                                   xfs_btree_key_offset(cur, last + 1) - 1);
1408         } else {
1409                 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1410                                 xfs_ilog_fbroot(cur->bc_ino.whichfork));
1411         }
1412 }
1413
1414 /*
1415  * Log record values from the btree block.
1416  */
1417 void
1418 xfs_btree_log_recs(
1419         struct xfs_btree_cur    *cur,
1420         struct xfs_buf          *bp,
1421         int                     first,
1422         int                     last)
1423 {
1424
1425         xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1426         xfs_trans_log_buf(cur->bc_tp, bp,
1427                           xfs_btree_rec_offset(cur, first),
1428                           xfs_btree_rec_offset(cur, last + 1) - 1);
1429
1430 }
1431
1432 /*
1433  * Log block pointer fields from a btree block (nonleaf).
1434  */
1435 STATIC void
1436 xfs_btree_log_ptrs(
1437         struct xfs_btree_cur    *cur,   /* btree cursor */
1438         struct xfs_buf          *bp,    /* buffer containing btree block */
1439         int                     first,  /* index of first pointer to log */
1440         int                     last)   /* index of last pointer to log */
1441 {
1442
1443         if (bp) {
1444                 struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
1445                 int                     level = xfs_btree_get_level(block);
1446
1447                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1448                 xfs_trans_log_buf(cur->bc_tp, bp,
1449                                 xfs_btree_ptr_offset(cur, first, level),
1450                                 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1451         } else {
1452                 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1453                         xfs_ilog_fbroot(cur->bc_ino.whichfork));
1454         }
1455
1456 }
1457
1458 /*
1459  * Log fields from a btree block header.
1460  */
1461 void
1462 xfs_btree_log_block(
1463         struct xfs_btree_cur    *cur,   /* btree cursor */
1464         struct xfs_buf          *bp,    /* buffer containing btree block */
1465         int                     fields) /* mask of fields: XFS_BB_... */
1466 {
1467         int                     first;  /* first byte offset logged */
1468         int                     last;   /* last byte offset logged */
1469         static const short      soffsets[] = {  /* table of offsets (short) */
1470                 offsetof(struct xfs_btree_block, bb_magic),
1471                 offsetof(struct xfs_btree_block, bb_level),
1472                 offsetof(struct xfs_btree_block, bb_numrecs),
1473                 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1474                 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1475                 offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1476                 offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1477                 offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1478                 offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1479                 offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1480                 XFS_BTREE_SBLOCK_CRC_LEN
1481         };
1482         static const short      loffsets[] = {  /* table of offsets (long) */
1483                 offsetof(struct xfs_btree_block, bb_magic),
1484                 offsetof(struct xfs_btree_block, bb_level),
1485                 offsetof(struct xfs_btree_block, bb_numrecs),
1486                 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1487                 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1488                 offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1489                 offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1490                 offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1491                 offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1492                 offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1493                 offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1494                 XFS_BTREE_LBLOCK_CRC_LEN
1495         };
1496
1497         if (bp) {
1498                 int nbits;
1499
1500                 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
1501                         /*
1502                          * We don't log the CRC when updating a btree
1503                          * block but instead recreate it during log
1504                          * recovery.  As the log buffers have checksums
1505                          * of their own this is safe and avoids logging a crc
1506                          * update in a lot of places.
1507                          */
1508                         if (fields == XFS_BB_ALL_BITS)
1509                                 fields = XFS_BB_ALL_BITS_CRC;
1510                         nbits = XFS_BB_NUM_BITS_CRC;
1511                 } else {
1512                         nbits = XFS_BB_NUM_BITS;
1513                 }
1514                 xfs_btree_offsets(fields,
1515                                   (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1516                                         loffsets : soffsets,
1517                                   nbits, &first, &last);
1518                 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1519                 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1520         } else {
1521                 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1522                         xfs_ilog_fbroot(cur->bc_ino.whichfork));
1523         }
1524 }
1525
1526 /*
1527  * Increment cursor by one record at the level.
1528  * For nonzero levels the leaf-ward information is untouched.
1529  */
1530 int                                             /* error */
1531 xfs_btree_increment(
1532         struct xfs_btree_cur    *cur,
1533         int                     level,
1534         int                     *stat)          /* success/failure */
1535 {
1536         struct xfs_btree_block  *block;
1537         union xfs_btree_ptr     ptr;
1538         struct xfs_buf          *bp;
1539         int                     error;          /* error return value */
1540         int                     lev;
1541
1542         ASSERT(level < cur->bc_nlevels);
1543
1544         /* Read-ahead to the right at this level. */
1545         xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1546
1547         /* Get a pointer to the btree block. */
1548         block = xfs_btree_get_block(cur, level, &bp);
1549
1550 #ifdef DEBUG
1551         error = xfs_btree_check_block(cur, block, level, bp);
1552         if (error)
1553                 goto error0;
1554 #endif
1555
1556         /* We're done if we remain in the block after the increment. */
1557         if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
1558                 goto out1;
1559
1560         /* Fail if we just went off the right edge of the tree. */
1561         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1562         if (xfs_btree_ptr_is_null(cur, &ptr))
1563                 goto out0;
1564
1565         XFS_BTREE_STATS_INC(cur, increment);
1566
1567         /*
1568          * March up the tree incrementing pointers.
1569          * Stop when we don't go off the right edge of a block.
1570          */
1571         for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1572                 block = xfs_btree_get_block(cur, lev, &bp);
1573
1574 #ifdef DEBUG
1575                 error = xfs_btree_check_block(cur, block, lev, bp);
1576                 if (error)
1577                         goto error0;
1578 #endif
1579
1580                 if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
1581                         break;
1582
1583                 /* Read-ahead the right block for the next loop. */
1584                 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1585         }
1586
1587         /*
1588          * If we went off the root then we are either seriously
1589          * confused or have the tree root in an inode.
1590          */
1591         if (lev == cur->bc_nlevels) {
1592                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1593                         goto out0;
1594                 ASSERT(0);
1595                 error = -EFSCORRUPTED;
1596                 goto error0;
1597         }
1598         ASSERT(lev < cur->bc_nlevels);
1599
1600         /*
1601          * Now walk back down the tree, fixing up the cursor's buffer
1602          * pointers and key numbers.
1603          */
1604         for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1605                 union xfs_btree_ptr     *ptrp;
1606
1607                 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1608                 --lev;
1609                 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1610                 if (error)
1611                         goto error0;
1612
1613                 xfs_btree_setbuf(cur, lev, bp);
1614                 cur->bc_ptrs[lev] = 1;
1615         }
1616 out1:
1617         *stat = 1;
1618         return 0;
1619
1620 out0:
1621         *stat = 0;
1622         return 0;
1623
1624 error0:
1625         return error;
1626 }
1627
1628 /*
1629  * Decrement cursor by one record at the level.
1630  * For nonzero levels the leaf-ward information is untouched.
1631  */
1632 int                                             /* error */
1633 xfs_btree_decrement(
1634         struct xfs_btree_cur    *cur,
1635         int                     level,
1636         int                     *stat)          /* success/failure */
1637 {
1638         struct xfs_btree_block  *block;
1639         struct xfs_buf          *bp;
1640         int                     error;          /* error return value */
1641         int                     lev;
1642         union xfs_btree_ptr     ptr;
1643
1644         ASSERT(level < cur->bc_nlevels);
1645
1646         /* Read-ahead to the left at this level. */
1647         xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1648
1649         /* We're done if we remain in the block after the decrement. */
1650         if (--cur->bc_ptrs[level] > 0)
1651                 goto out1;
1652
1653         /* Get a pointer to the btree block. */
1654         block = xfs_btree_get_block(cur, level, &bp);
1655
1656 #ifdef DEBUG
1657         error = xfs_btree_check_block(cur, block, level, bp);
1658         if (error)
1659                 goto error0;
1660 #endif
1661
1662         /* Fail if we just went off the left edge of the tree. */
1663         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1664         if (xfs_btree_ptr_is_null(cur, &ptr))
1665                 goto out0;
1666
1667         XFS_BTREE_STATS_INC(cur, decrement);
1668
1669         /*
1670          * March up the tree decrementing pointers.
1671          * Stop when we don't go off the left edge of a block.
1672          */
1673         for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1674                 if (--cur->bc_ptrs[lev] > 0)
1675                         break;
1676                 /* Read-ahead the left block for the next loop. */
1677                 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1678         }
1679
1680         /*
1681          * If we went off the root then we are seriously confused.
1682          * or the root of the tree is in an inode.
1683          */
1684         if (lev == cur->bc_nlevels) {
1685                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1686                         goto out0;
1687                 ASSERT(0);
1688                 error = -EFSCORRUPTED;
1689                 goto error0;
1690         }
1691         ASSERT(lev < cur->bc_nlevels);
1692
1693         /*
1694          * Now walk back down the tree, fixing up the cursor's buffer
1695          * pointers and key numbers.
1696          */
1697         for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1698                 union xfs_btree_ptr     *ptrp;
1699
1700                 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1701                 --lev;
1702                 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1703                 if (error)
1704                         goto error0;
1705                 xfs_btree_setbuf(cur, lev, bp);
1706                 cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
1707         }
1708 out1:
1709         *stat = 1;
1710         return 0;
1711
1712 out0:
1713         *stat = 0;
1714         return 0;
1715
1716 error0:
1717         return error;
1718 }
1719
1720 int
1721 xfs_btree_lookup_get_block(
1722         struct xfs_btree_cur    *cur,   /* btree cursor */
1723         int                     level,  /* level in the btree */
1724         union xfs_btree_ptr     *pp,    /* ptr to btree block */
1725         struct xfs_btree_block  **blkp) /* return btree block */
1726 {
1727         struct xfs_buf          *bp;    /* buffer pointer for btree block */
1728         xfs_daddr_t             daddr;
1729         int                     error = 0;
1730
1731         /* special case the root block if in an inode */
1732         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1733             (level == cur->bc_nlevels - 1)) {
1734                 *blkp = xfs_btree_get_iroot(cur);
1735                 return 0;
1736         }
1737
1738         /*
1739          * If the old buffer at this level for the disk address we are
1740          * looking for re-use it.
1741          *
1742          * Otherwise throw it away and get a new one.
1743          */
1744         bp = cur->bc_bufs[level];
1745         error = xfs_btree_ptr_to_daddr(cur, pp, &daddr);
1746         if (error)
1747                 return error;
1748         if (bp && XFS_BUF_ADDR(bp) == daddr) {
1749                 *blkp = XFS_BUF_TO_BLOCK(bp);
1750                 return 0;
1751         }
1752
1753         error = xfs_btree_read_buf_block(cur, pp, 0, blkp, &bp);
1754         if (error)
1755                 return error;
1756
1757         /* Check the inode owner since the verifiers don't. */
1758         if (xfs_sb_version_hascrc(&cur->bc_mp->m_sb) &&
1759             !(cur->bc_ino.flags & XFS_BTCUR_BMBT_INVALID_OWNER) &&
1760             (cur->bc_flags & XFS_BTREE_LONG_PTRS) &&
1761             be64_to_cpu((*blkp)->bb_u.l.bb_owner) !=
1762                         cur->bc_ino.ip->i_ino)
1763                 goto out_bad;
1764
1765         /* Did we get the level we were looking for? */
1766         if (be16_to_cpu((*blkp)->bb_level) != level)
1767                 goto out_bad;
1768
1769         /* Check that internal nodes have at least one record. */
1770         if (level != 0 && be16_to_cpu((*blkp)->bb_numrecs) == 0)
1771                 goto out_bad;
1772
1773         xfs_btree_setbuf(cur, level, bp);
1774         return 0;
1775
1776 out_bad:
1777         *blkp = NULL;
1778         xfs_buf_mark_corrupt(bp);
1779         xfs_trans_brelse(cur->bc_tp, bp);
1780         return -EFSCORRUPTED;
1781 }
1782
1783 /*
1784  * Get current search key.  For level 0 we don't actually have a key
1785  * structure so we make one up from the record.  For all other levels
1786  * we just return the right key.
1787  */
1788 STATIC union xfs_btree_key *
1789 xfs_lookup_get_search_key(
1790         struct xfs_btree_cur    *cur,
1791         int                     level,
1792         int                     keyno,
1793         struct xfs_btree_block  *block,
1794         union xfs_btree_key     *kp)
1795 {
1796         if (level == 0) {
1797                 cur->bc_ops->init_key_from_rec(kp,
1798                                 xfs_btree_rec_addr(cur, keyno, block));
1799                 return kp;
1800         }
1801
1802         return xfs_btree_key_addr(cur, keyno, block);
1803 }
1804
1805 /*
1806  * Lookup the record.  The cursor is made to point to it, based on dir.
1807  * stat is set to 0 if can't find any such record, 1 for success.
1808  */
1809 int                                     /* error */
1810 xfs_btree_lookup(
1811         struct xfs_btree_cur    *cur,   /* btree cursor */
1812         xfs_lookup_t            dir,    /* <=, ==, or >= */
1813         int                     *stat)  /* success/failure */
1814 {
1815         struct xfs_btree_block  *block; /* current btree block */
1816         int64_t                 diff;   /* difference for the current key */
1817         int                     error;  /* error return value */
1818         int                     keyno;  /* current key number */
1819         int                     level;  /* level in the btree */
1820         union xfs_btree_ptr     *pp;    /* ptr to btree block */
1821         union xfs_btree_ptr     ptr;    /* ptr to btree block */
1822
1823         XFS_BTREE_STATS_INC(cur, lookup);
1824
1825         /* No such thing as a zero-level tree. */
1826         if (XFS_IS_CORRUPT(cur->bc_mp, cur->bc_nlevels == 0))
1827                 return -EFSCORRUPTED;
1828
1829         block = NULL;
1830         keyno = 0;
1831
1832         /* initialise start pointer from cursor */
1833         cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1834         pp = &ptr;
1835
1836         /*
1837          * Iterate over each level in the btree, starting at the root.
1838          * For each level above the leaves, find the key we need, based
1839          * on the lookup record, then follow the corresponding block
1840          * pointer down to the next level.
1841          */
1842         for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1843                 /* Get the block we need to do the lookup on. */
1844                 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1845                 if (error)
1846                         goto error0;
1847
1848                 if (diff == 0) {
1849                         /*
1850                          * If we already had a key match at a higher level, we
1851                          * know we need to use the first entry in this block.
1852                          */
1853                         keyno = 1;
1854                 } else {
1855                         /* Otherwise search this block. Do a binary search. */
1856
1857                         int     high;   /* high entry number */
1858                         int     low;    /* low entry number */
1859
1860                         /* Set low and high entry numbers, 1-based. */
1861                         low = 1;
1862                         high = xfs_btree_get_numrecs(block);
1863                         if (!high) {
1864                                 /* Block is empty, must be an empty leaf. */
1865                                 if (level != 0 || cur->bc_nlevels != 1) {
1866                                         XFS_CORRUPTION_ERROR(__func__,
1867                                                         XFS_ERRLEVEL_LOW,
1868                                                         cur->bc_mp, block,
1869                                                         sizeof(*block));
1870                                         return -EFSCORRUPTED;
1871                                 }
1872
1873                                 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1874                                 *stat = 0;
1875                                 return 0;
1876                         }
1877
1878                         /* Binary search the block. */
1879                         while (low <= high) {
1880                                 union xfs_btree_key     key;
1881                                 union xfs_btree_key     *kp;
1882
1883                                 XFS_BTREE_STATS_INC(cur, compare);
1884
1885                                 /* keyno is average of low and high. */
1886                                 keyno = (low + high) >> 1;
1887
1888                                 /* Get current search key */
1889                                 kp = xfs_lookup_get_search_key(cur, level,
1890                                                 keyno, block, &key);
1891
1892                                 /*
1893                                  * Compute difference to get next direction:
1894                                  *  - less than, move right
1895                                  *  - greater than, move left
1896                                  *  - equal, we're done
1897                                  */
1898                                 diff = cur->bc_ops->key_diff(cur, kp);
1899                                 if (diff < 0)
1900                                         low = keyno + 1;
1901                                 else if (diff > 0)
1902                                         high = keyno - 1;
1903                                 else
1904                                         break;
1905                         }
1906                 }
1907
1908                 /*
1909                  * If there are more levels, set up for the next level
1910                  * by getting the block number and filling in the cursor.
1911                  */
1912                 if (level > 0) {
1913                         /*
1914                          * If we moved left, need the previous key number,
1915                          * unless there isn't one.
1916                          */
1917                         if (diff > 0 && --keyno < 1)
1918                                 keyno = 1;
1919                         pp = xfs_btree_ptr_addr(cur, keyno, block);
1920
1921                         error = xfs_btree_debug_check_ptr(cur, pp, 0, level);
1922                         if (error)
1923                                 goto error0;
1924
1925                         cur->bc_ptrs[level] = keyno;
1926                 }
1927         }
1928
1929         /* Done with the search. See if we need to adjust the results. */
1930         if (dir != XFS_LOOKUP_LE && diff < 0) {
1931                 keyno++;
1932                 /*
1933                  * If ge search and we went off the end of the block, but it's
1934                  * not the last block, we're in the wrong block.
1935                  */
1936                 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1937                 if (dir == XFS_LOOKUP_GE &&
1938                     keyno > xfs_btree_get_numrecs(block) &&
1939                     !xfs_btree_ptr_is_null(cur, &ptr)) {
1940                         int     i;
1941
1942                         cur->bc_ptrs[0] = keyno;
1943                         error = xfs_btree_increment(cur, 0, &i);
1944                         if (error)
1945                                 goto error0;
1946                         if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1947                                 return -EFSCORRUPTED;
1948                         *stat = 1;
1949                         return 0;
1950                 }
1951         } else if (dir == XFS_LOOKUP_LE && diff > 0)
1952                 keyno--;
1953         cur->bc_ptrs[0] = keyno;
1954
1955         /* Return if we succeeded or not. */
1956         if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
1957                 *stat = 0;
1958         else if (dir != XFS_LOOKUP_EQ || diff == 0)
1959                 *stat = 1;
1960         else
1961                 *stat = 0;
1962         return 0;
1963
1964 error0:
1965         return error;
1966 }
1967
1968 /* Find the high key storage area from a regular key. */
1969 union xfs_btree_key *
1970 xfs_btree_high_key_from_key(
1971         struct xfs_btree_cur    *cur,
1972         union xfs_btree_key     *key)
1973 {
1974         ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
1975         return (union xfs_btree_key *)((char *)key +
1976                         (cur->bc_ops->key_len / 2));
1977 }
1978
1979 /* Determine the low (and high if overlapped) keys of a leaf block */
1980 STATIC void
1981 xfs_btree_get_leaf_keys(
1982         struct xfs_btree_cur    *cur,
1983         struct xfs_btree_block  *block,
1984         union xfs_btree_key     *key)
1985 {
1986         union xfs_btree_key     max_hkey;
1987         union xfs_btree_key     hkey;
1988         union xfs_btree_rec     *rec;
1989         union xfs_btree_key     *high;
1990         int                     n;
1991
1992         rec = xfs_btree_rec_addr(cur, 1, block);
1993         cur->bc_ops->init_key_from_rec(key, rec);
1994
1995         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
1996
1997                 cur->bc_ops->init_high_key_from_rec(&max_hkey, rec);
1998                 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
1999                         rec = xfs_btree_rec_addr(cur, n, block);
2000                         cur->bc_ops->init_high_key_from_rec(&hkey, rec);
2001                         if (cur->bc_ops->diff_two_keys(cur, &hkey, &max_hkey)
2002                                         > 0)
2003                                 max_hkey = hkey;
2004                 }
2005
2006                 high = xfs_btree_high_key_from_key(cur, key);
2007                 memcpy(high, &max_hkey, cur->bc_ops->key_len / 2);
2008         }
2009 }
2010
2011 /* Determine the low (and high if overlapped) keys of a node block */
2012 STATIC void
2013 xfs_btree_get_node_keys(
2014         struct xfs_btree_cur    *cur,
2015         struct xfs_btree_block  *block,
2016         union xfs_btree_key     *key)
2017 {
2018         union xfs_btree_key     *hkey;
2019         union xfs_btree_key     *max_hkey;
2020         union xfs_btree_key     *high;
2021         int                     n;
2022
2023         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2024                 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2025                                 cur->bc_ops->key_len / 2);
2026
2027                 max_hkey = xfs_btree_high_key_addr(cur, 1, block);
2028                 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2029                         hkey = xfs_btree_high_key_addr(cur, n, block);
2030                         if (cur->bc_ops->diff_two_keys(cur, hkey, max_hkey) > 0)
2031                                 max_hkey = hkey;
2032                 }
2033
2034                 high = xfs_btree_high_key_from_key(cur, key);
2035                 memcpy(high, max_hkey, cur->bc_ops->key_len / 2);
2036         } else {
2037                 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2038                                 cur->bc_ops->key_len);
2039         }
2040 }
2041
2042 /* Derive the keys for any btree block. */
2043 void
2044 xfs_btree_get_keys(
2045         struct xfs_btree_cur    *cur,
2046         struct xfs_btree_block  *block,
2047         union xfs_btree_key     *key)
2048 {
2049         if (be16_to_cpu(block->bb_level) == 0)
2050                 xfs_btree_get_leaf_keys(cur, block, key);
2051         else
2052                 xfs_btree_get_node_keys(cur, block, key);
2053 }
2054
2055 /*
2056  * Decide if we need to update the parent keys of a btree block.  For
2057  * a standard btree this is only necessary if we're updating the first
2058  * record/key.  For an overlapping btree, we must always update the
2059  * keys because the highest key can be in any of the records or keys
2060  * in the block.
2061  */
2062 static inline bool
2063 xfs_btree_needs_key_update(
2064         struct xfs_btree_cur    *cur,
2065         int                     ptr)
2066 {
2067         return (cur->bc_flags & XFS_BTREE_OVERLAPPING) || ptr == 1;
2068 }
2069
2070 /*
2071  * Update the low and high parent keys of the given level, progressing
2072  * towards the root.  If force_all is false, stop if the keys for a given
2073  * level do not need updating.
2074  */
2075 STATIC int
2076 __xfs_btree_updkeys(
2077         struct xfs_btree_cur    *cur,
2078         int                     level,
2079         struct xfs_btree_block  *block,
2080         struct xfs_buf          *bp0,
2081         bool                    force_all)
2082 {
2083         union xfs_btree_key     key;    /* keys from current level */
2084         union xfs_btree_key     *lkey;  /* keys from the next level up */
2085         union xfs_btree_key     *hkey;
2086         union xfs_btree_key     *nlkey; /* keys from the next level up */
2087         union xfs_btree_key     *nhkey;
2088         struct xfs_buf          *bp;
2089         int                     ptr;
2090
2091         ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2092
2093         /* Exit if there aren't any parent levels to update. */
2094         if (level + 1 >= cur->bc_nlevels)
2095                 return 0;
2096
2097         trace_xfs_btree_updkeys(cur, level, bp0);
2098
2099         lkey = &key;
2100         hkey = xfs_btree_high_key_from_key(cur, lkey);
2101         xfs_btree_get_keys(cur, block, lkey);
2102         for (level++; level < cur->bc_nlevels; level++) {
2103 #ifdef DEBUG
2104                 int             error;
2105 #endif
2106                 block = xfs_btree_get_block(cur, level, &bp);
2107                 trace_xfs_btree_updkeys(cur, level, bp);
2108 #ifdef DEBUG
2109                 error = xfs_btree_check_block(cur, block, level, bp);
2110                 if (error)
2111                         return error;
2112 #endif
2113                 ptr = cur->bc_ptrs[level];
2114                 nlkey = xfs_btree_key_addr(cur, ptr, block);
2115                 nhkey = xfs_btree_high_key_addr(cur, ptr, block);
2116                 if (!force_all &&
2117                     !(cur->bc_ops->diff_two_keys(cur, nlkey, lkey) != 0 ||
2118                       cur->bc_ops->diff_two_keys(cur, nhkey, hkey) != 0))
2119                         break;
2120                 xfs_btree_copy_keys(cur, nlkey, lkey, 1);
2121                 xfs_btree_log_keys(cur, bp, ptr, ptr);
2122                 if (level + 1 >= cur->bc_nlevels)
2123                         break;
2124                 xfs_btree_get_node_keys(cur, block, lkey);
2125         }
2126
2127         return 0;
2128 }
2129
2130 /* Update all the keys from some level in cursor back to the root. */
2131 STATIC int
2132 xfs_btree_updkeys_force(
2133         struct xfs_btree_cur    *cur,
2134         int                     level)
2135 {
2136         struct xfs_buf          *bp;
2137         struct xfs_btree_block  *block;
2138
2139         block = xfs_btree_get_block(cur, level, &bp);
2140         return __xfs_btree_updkeys(cur, level, block, bp, true);
2141 }
2142
2143 /*
2144  * Update the parent keys of the given level, progressing towards the root.
2145  */
2146 STATIC int
2147 xfs_btree_update_keys(
2148         struct xfs_btree_cur    *cur,
2149         int                     level)
2150 {
2151         struct xfs_btree_block  *block;
2152         struct xfs_buf          *bp;
2153         union xfs_btree_key     *kp;
2154         union xfs_btree_key     key;
2155         int                     ptr;
2156
2157         ASSERT(level >= 0);
2158
2159         block = xfs_btree_get_block(cur, level, &bp);
2160         if (cur->bc_flags & XFS_BTREE_OVERLAPPING)
2161                 return __xfs_btree_updkeys(cur, level, block, bp, false);
2162
2163         /*
2164          * Go up the tree from this level toward the root.
2165          * At each level, update the key value to the value input.
2166          * Stop when we reach a level where the cursor isn't pointing
2167          * at the first entry in the block.
2168          */
2169         xfs_btree_get_keys(cur, block, &key);
2170         for (level++, ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
2171 #ifdef DEBUG
2172                 int             error;
2173 #endif
2174                 block = xfs_btree_get_block(cur, level, &bp);
2175 #ifdef DEBUG
2176                 error = xfs_btree_check_block(cur, block, level, bp);
2177                 if (error)
2178                         return error;
2179 #endif
2180                 ptr = cur->bc_ptrs[level];
2181                 kp = xfs_btree_key_addr(cur, ptr, block);
2182                 xfs_btree_copy_keys(cur, kp, &key, 1);
2183                 xfs_btree_log_keys(cur, bp, ptr, ptr);
2184         }
2185
2186         return 0;
2187 }
2188
2189 /*
2190  * Update the record referred to by cur to the value in the
2191  * given record. This either works (return 0) or gets an
2192  * EFSCORRUPTED error.
2193  */
2194 int
2195 xfs_btree_update(
2196         struct xfs_btree_cur    *cur,
2197         union xfs_btree_rec     *rec)
2198 {
2199         struct xfs_btree_block  *block;
2200         struct xfs_buf          *bp;
2201         int                     error;
2202         int                     ptr;
2203         union xfs_btree_rec     *rp;
2204
2205         /* Pick up the current block. */
2206         block = xfs_btree_get_block(cur, 0, &bp);
2207
2208 #ifdef DEBUG
2209         error = xfs_btree_check_block(cur, block, 0, bp);
2210         if (error)
2211                 goto error0;
2212 #endif
2213         /* Get the address of the rec to be updated. */
2214         ptr = cur->bc_ptrs[0];
2215         rp = xfs_btree_rec_addr(cur, ptr, block);
2216
2217         /* Fill in the new contents and log them. */
2218         xfs_btree_copy_recs(cur, rp, rec, 1);
2219         xfs_btree_log_recs(cur, bp, ptr, ptr);
2220
2221         /*
2222          * If we are tracking the last record in the tree and
2223          * we are at the far right edge of the tree, update it.
2224          */
2225         if (xfs_btree_is_lastrec(cur, block, 0)) {
2226                 cur->bc_ops->update_lastrec(cur, block, rec,
2227                                             ptr, LASTREC_UPDATE);
2228         }
2229
2230         /* Pass new key value up to our parent. */
2231         if (xfs_btree_needs_key_update(cur, ptr)) {
2232                 error = xfs_btree_update_keys(cur, 0);
2233                 if (error)
2234                         goto error0;
2235         }
2236
2237         return 0;
2238
2239 error0:
2240         return error;
2241 }
2242
2243 /*
2244  * Move 1 record left from cur/level if possible.
2245  * Update cur to reflect the new path.
2246  */
2247 STATIC int                                      /* error */
2248 xfs_btree_lshift(
2249         struct xfs_btree_cur    *cur,
2250         int                     level,
2251         int                     *stat)          /* success/failure */
2252 {
2253         struct xfs_buf          *lbp;           /* left buffer pointer */
2254         struct xfs_btree_block  *left;          /* left btree block */
2255         int                     lrecs;          /* left record count */
2256         struct xfs_buf          *rbp;           /* right buffer pointer */
2257         struct xfs_btree_block  *right;         /* right btree block */
2258         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
2259         int                     rrecs;          /* right record count */
2260         union xfs_btree_ptr     lptr;           /* left btree pointer */
2261         union xfs_btree_key     *rkp = NULL;    /* right btree key */
2262         union xfs_btree_ptr     *rpp = NULL;    /* right address pointer */
2263         union xfs_btree_rec     *rrp = NULL;    /* right record pointer */
2264         int                     error;          /* error return value */
2265         int                     i;
2266
2267         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2268             level == cur->bc_nlevels - 1)
2269                 goto out0;
2270
2271         /* Set up variables for this block as "right". */
2272         right = xfs_btree_get_block(cur, level, &rbp);
2273
2274 #ifdef DEBUG
2275         error = xfs_btree_check_block(cur, right, level, rbp);
2276         if (error)
2277                 goto error0;
2278 #endif
2279
2280         /* If we've got no left sibling then we can't shift an entry left. */
2281         xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2282         if (xfs_btree_ptr_is_null(cur, &lptr))
2283                 goto out0;
2284
2285         /*
2286          * If the cursor entry is the one that would be moved, don't
2287          * do it... it's too complicated.
2288          */
2289         if (cur->bc_ptrs[level] <= 1)
2290                 goto out0;
2291
2292         /* Set up the left neighbor as "left". */
2293         error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2294         if (error)
2295                 goto error0;
2296
2297         /* If it's full, it can't take another entry. */
2298         lrecs = xfs_btree_get_numrecs(left);
2299         if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2300                 goto out0;
2301
2302         rrecs = xfs_btree_get_numrecs(right);
2303
2304         /*
2305          * We add one entry to the left side and remove one for the right side.
2306          * Account for it here, the changes will be updated on disk and logged
2307          * later.
2308          */
2309         lrecs++;
2310         rrecs--;
2311
2312         XFS_BTREE_STATS_INC(cur, lshift);
2313         XFS_BTREE_STATS_ADD(cur, moves, 1);
2314
2315         /*
2316          * If non-leaf, copy a key and a ptr to the left block.
2317          * Log the changes to the left block.
2318          */
2319         if (level > 0) {
2320                 /* It's a non-leaf.  Move keys and pointers. */
2321                 union xfs_btree_key     *lkp;   /* left btree key */
2322                 union xfs_btree_ptr     *lpp;   /* left address pointer */
2323
2324                 lkp = xfs_btree_key_addr(cur, lrecs, left);
2325                 rkp = xfs_btree_key_addr(cur, 1, right);
2326
2327                 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2328                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2329
2330                 error = xfs_btree_debug_check_ptr(cur, rpp, 0, level);
2331                 if (error)
2332                         goto error0;
2333
2334                 xfs_btree_copy_keys(cur, lkp, rkp, 1);
2335                 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2336
2337                 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2338                 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2339
2340                 ASSERT(cur->bc_ops->keys_inorder(cur,
2341                         xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2342         } else {
2343                 /* It's a leaf.  Move records.  */
2344                 union xfs_btree_rec     *lrp;   /* left record pointer */
2345
2346                 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2347                 rrp = xfs_btree_rec_addr(cur, 1, right);
2348
2349                 xfs_btree_copy_recs(cur, lrp, rrp, 1);
2350                 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2351
2352                 ASSERT(cur->bc_ops->recs_inorder(cur,
2353                         xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2354         }
2355
2356         xfs_btree_set_numrecs(left, lrecs);
2357         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2358
2359         xfs_btree_set_numrecs(right, rrecs);
2360         xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2361
2362         /*
2363          * Slide the contents of right down one entry.
2364          */
2365         XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2366         if (level > 0) {
2367                 /* It's a nonleaf. operate on keys and ptrs */
2368                 for (i = 0; i < rrecs; i++) {
2369                         error = xfs_btree_debug_check_ptr(cur, rpp, i + 1, level);
2370                         if (error)
2371                                 goto error0;
2372                 }
2373
2374                 xfs_btree_shift_keys(cur,
2375                                 xfs_btree_key_addr(cur, 2, right),
2376                                 -1, rrecs);
2377                 xfs_btree_shift_ptrs(cur,
2378                                 xfs_btree_ptr_addr(cur, 2, right),
2379                                 -1, rrecs);
2380
2381                 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2382                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2383         } else {
2384                 /* It's a leaf. operate on records */
2385                 xfs_btree_shift_recs(cur,
2386                         xfs_btree_rec_addr(cur, 2, right),
2387                         -1, rrecs);
2388                 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2389         }
2390
2391         /*
2392          * Using a temporary cursor, update the parent key values of the
2393          * block on the left.
2394          */
2395         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2396                 error = xfs_btree_dup_cursor(cur, &tcur);
2397                 if (error)
2398                         goto error0;
2399                 i = xfs_btree_firstrec(tcur, level);
2400                 if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2401                         error = -EFSCORRUPTED;
2402                         goto error0;
2403                 }
2404
2405                 error = xfs_btree_decrement(tcur, level, &i);
2406                 if (error)
2407                         goto error1;
2408
2409                 /* Update the parent high keys of the left block, if needed. */
2410                 error = xfs_btree_update_keys(tcur, level);
2411                 if (error)
2412                         goto error1;
2413
2414                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2415         }
2416
2417         /* Update the parent keys of the right block. */
2418         error = xfs_btree_update_keys(cur, level);
2419         if (error)
2420                 goto error0;
2421
2422         /* Slide the cursor value left one. */
2423         cur->bc_ptrs[level]--;
2424
2425         *stat = 1;
2426         return 0;
2427
2428 out0:
2429         *stat = 0;
2430         return 0;
2431
2432 error0:
2433         return error;
2434
2435 error1:
2436         xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2437         return error;
2438 }
2439
2440 /*
2441  * Move 1 record right from cur/level if possible.
2442  * Update cur to reflect the new path.
2443  */
2444 STATIC int                                      /* error */
2445 xfs_btree_rshift(
2446         struct xfs_btree_cur    *cur,
2447         int                     level,
2448         int                     *stat)          /* success/failure */
2449 {
2450         struct xfs_buf          *lbp;           /* left buffer pointer */
2451         struct xfs_btree_block  *left;          /* left btree block */
2452         struct xfs_buf          *rbp;           /* right buffer pointer */
2453         struct xfs_btree_block  *right;         /* right btree block */
2454         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
2455         union xfs_btree_ptr     rptr;           /* right block pointer */
2456         union xfs_btree_key     *rkp;           /* right btree key */
2457         int                     rrecs;          /* right record count */
2458         int                     lrecs;          /* left record count */
2459         int                     error;          /* error return value */
2460         int                     i;              /* loop counter */
2461
2462         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2463             (level == cur->bc_nlevels - 1))
2464                 goto out0;
2465
2466         /* Set up variables for this block as "left". */
2467         left = xfs_btree_get_block(cur, level, &lbp);
2468
2469 #ifdef DEBUG
2470         error = xfs_btree_check_block(cur, left, level, lbp);
2471         if (error)
2472                 goto error0;
2473 #endif
2474
2475         /* If we've got no right sibling then we can't shift an entry right. */
2476         xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2477         if (xfs_btree_ptr_is_null(cur, &rptr))
2478                 goto out0;
2479
2480         /*
2481          * If the cursor entry is the one that would be moved, don't
2482          * do it... it's too complicated.
2483          */
2484         lrecs = xfs_btree_get_numrecs(left);
2485         if (cur->bc_ptrs[level] >= lrecs)
2486                 goto out0;
2487
2488         /* Set up the right neighbor as "right". */
2489         error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2490         if (error)
2491                 goto error0;
2492
2493         /* If it's full, it can't take another entry. */
2494         rrecs = xfs_btree_get_numrecs(right);
2495         if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2496                 goto out0;
2497
2498         XFS_BTREE_STATS_INC(cur, rshift);
2499         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2500
2501         /*
2502          * Make a hole at the start of the right neighbor block, then
2503          * copy the last left block entry to the hole.
2504          */
2505         if (level > 0) {
2506                 /* It's a nonleaf. make a hole in the keys and ptrs */
2507                 union xfs_btree_key     *lkp;
2508                 union xfs_btree_ptr     *lpp;
2509                 union xfs_btree_ptr     *rpp;
2510
2511                 lkp = xfs_btree_key_addr(cur, lrecs, left);
2512                 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2513                 rkp = xfs_btree_key_addr(cur, 1, right);
2514                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2515
2516                 for (i = rrecs - 1; i >= 0; i--) {
2517                         error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
2518                         if (error)
2519                                 goto error0;
2520                 }
2521
2522                 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2523                 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2524
2525                 error = xfs_btree_debug_check_ptr(cur, lpp, 0, level);
2526                 if (error)
2527                         goto error0;
2528
2529                 /* Now put the new data in, and log it. */
2530                 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2531                 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2532
2533                 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2534                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2535
2536                 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2537                         xfs_btree_key_addr(cur, 2, right)));
2538         } else {
2539                 /* It's a leaf. make a hole in the records */
2540                 union xfs_btree_rec     *lrp;
2541                 union xfs_btree_rec     *rrp;
2542
2543                 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2544                 rrp = xfs_btree_rec_addr(cur, 1, right);
2545
2546                 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2547
2548                 /* Now put the new data in, and log it. */
2549                 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2550                 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2551         }
2552
2553         /*
2554          * Decrement and log left's numrecs, bump and log right's numrecs.
2555          */
2556         xfs_btree_set_numrecs(left, --lrecs);
2557         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2558
2559         xfs_btree_set_numrecs(right, ++rrecs);
2560         xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2561
2562         /*
2563          * Using a temporary cursor, update the parent key values of the
2564          * block on the right.
2565          */
2566         error = xfs_btree_dup_cursor(cur, &tcur);
2567         if (error)
2568                 goto error0;
2569         i = xfs_btree_lastrec(tcur, level);
2570         if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2571                 error = -EFSCORRUPTED;
2572                 goto error0;
2573         }
2574
2575         error = xfs_btree_increment(tcur, level, &i);
2576         if (error)
2577                 goto error1;
2578
2579         /* Update the parent high keys of the left block, if needed. */
2580         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2581                 error = xfs_btree_update_keys(cur, level);
2582                 if (error)
2583                         goto error1;
2584         }
2585
2586         /* Update the parent keys of the right block. */
2587         error = xfs_btree_update_keys(tcur, level);
2588         if (error)
2589                 goto error1;
2590
2591         xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2592
2593         *stat = 1;
2594         return 0;
2595
2596 out0:
2597         *stat = 0;
2598         return 0;
2599
2600 error0:
2601         return error;
2602
2603 error1:
2604         xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2605         return error;
2606 }
2607
2608 /*
2609  * Split cur/level block in half.
2610  * Return new block number and the key to its first
2611  * record (to be inserted into parent).
2612  */
2613 STATIC int                                      /* error */
2614 __xfs_btree_split(
2615         struct xfs_btree_cur    *cur,
2616         int                     level,
2617         union xfs_btree_ptr     *ptrp,
2618         union xfs_btree_key     *key,
2619         struct xfs_btree_cur    **curp,
2620         int                     *stat)          /* success/failure */
2621 {
2622         union xfs_btree_ptr     lptr;           /* left sibling block ptr */
2623         struct xfs_buf          *lbp;           /* left buffer pointer */
2624         struct xfs_btree_block  *left;          /* left btree block */
2625         union xfs_btree_ptr     rptr;           /* right sibling block ptr */
2626         struct xfs_buf          *rbp;           /* right buffer pointer */
2627         struct xfs_btree_block  *right;         /* right btree block */
2628         union xfs_btree_ptr     rrptr;          /* right-right sibling ptr */
2629         struct xfs_buf          *rrbp;          /* right-right buffer pointer */
2630         struct xfs_btree_block  *rrblock;       /* right-right btree block */
2631         int                     lrecs;
2632         int                     rrecs;
2633         int                     src_index;
2634         int                     error;          /* error return value */
2635         int                     i;
2636
2637         XFS_BTREE_STATS_INC(cur, split);
2638
2639         /* Set up left block (current one). */
2640         left = xfs_btree_get_block(cur, level, &lbp);
2641
2642 #ifdef DEBUG
2643         error = xfs_btree_check_block(cur, left, level, lbp);
2644         if (error)
2645                 goto error0;
2646 #endif
2647
2648         xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2649
2650         /* Allocate the new block. If we can't do it, we're toast. Give up. */
2651         error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, stat);
2652         if (error)
2653                 goto error0;
2654         if (*stat == 0)
2655                 goto out0;
2656         XFS_BTREE_STATS_INC(cur, alloc);
2657
2658         /* Set up the new block as "right". */
2659         error = xfs_btree_get_buf_block(cur, &rptr, &right, &rbp);
2660         if (error)
2661                 goto error0;
2662
2663         /* Fill in the btree header for the new right block. */
2664         xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2665
2666         /*
2667          * Split the entries between the old and the new block evenly.
2668          * Make sure that if there's an odd number of entries now, that
2669          * each new block will have the same number of entries.
2670          */
2671         lrecs = xfs_btree_get_numrecs(left);
2672         rrecs = lrecs / 2;
2673         if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
2674                 rrecs++;
2675         src_index = (lrecs - rrecs + 1);
2676
2677         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2678
2679         /* Adjust numrecs for the later get_*_keys() calls. */
2680         lrecs -= rrecs;
2681         xfs_btree_set_numrecs(left, lrecs);
2682         xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2683
2684         /*
2685          * Copy btree block entries from the left block over to the
2686          * new block, the right. Update the right block and log the
2687          * changes.
2688          */
2689         if (level > 0) {
2690                 /* It's a non-leaf.  Move keys and pointers. */
2691                 union xfs_btree_key     *lkp;   /* left btree key */
2692                 union xfs_btree_ptr     *lpp;   /* left address pointer */
2693                 union xfs_btree_key     *rkp;   /* right btree key */
2694                 union xfs_btree_ptr     *rpp;   /* right address pointer */
2695
2696                 lkp = xfs_btree_key_addr(cur, src_index, left);
2697                 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2698                 rkp = xfs_btree_key_addr(cur, 1, right);
2699                 rpp = xfs_btree_ptr_addr(cur, 1, right);
2700
2701                 for (i = src_index; i < rrecs; i++) {
2702                         error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
2703                         if (error)
2704                                 goto error0;
2705                 }
2706
2707                 /* Copy the keys & pointers to the new block. */
2708                 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2709                 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2710
2711                 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2712                 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2713
2714                 /* Stash the keys of the new block for later insertion. */
2715                 xfs_btree_get_node_keys(cur, right, key);
2716         } else {
2717                 /* It's a leaf.  Move records.  */
2718                 union xfs_btree_rec     *lrp;   /* left record pointer */
2719                 union xfs_btree_rec     *rrp;   /* right record pointer */
2720
2721                 lrp = xfs_btree_rec_addr(cur, src_index, left);
2722                 rrp = xfs_btree_rec_addr(cur, 1, right);
2723
2724                 /* Copy records to the new block. */
2725                 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2726                 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2727
2728                 /* Stash the keys of the new block for later insertion. */
2729                 xfs_btree_get_leaf_keys(cur, right, key);
2730         }
2731
2732         /*
2733          * Find the left block number by looking in the buffer.
2734          * Adjust sibling pointers.
2735          */
2736         xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2737         xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2738         xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2739         xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2740
2741         xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2742         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2743
2744         /*
2745          * If there's a block to the new block's right, make that block
2746          * point back to right instead of to left.
2747          */
2748         if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2749                 error = xfs_btree_read_buf_block(cur, &rrptr,
2750                                                         0, &rrblock, &rrbp);
2751                 if (error)
2752                         goto error0;
2753                 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2754                 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2755         }
2756
2757         /* Update the parent high keys of the left block, if needed. */
2758         if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2759                 error = xfs_btree_update_keys(cur, level);
2760                 if (error)
2761                         goto error0;
2762         }
2763
2764         /*
2765          * If the cursor is really in the right block, move it there.
2766          * If it's just pointing past the last entry in left, then we'll
2767          * insert there, so don't change anything in that case.
2768          */
2769         if (cur->bc_ptrs[level] > lrecs + 1) {
2770                 xfs_btree_setbuf(cur, level, rbp);
2771                 cur->bc_ptrs[level] -= lrecs;
2772         }
2773         /*
2774          * If there are more levels, we'll need another cursor which refers
2775          * the right block, no matter where this cursor was.
2776          */
2777         if (level + 1 < cur->bc_nlevels) {
2778                 error = xfs_btree_dup_cursor(cur, curp);
2779                 if (error)
2780                         goto error0;
2781                 (*curp)->bc_ptrs[level + 1]++;
2782         }
2783         *ptrp = rptr;
2784         *stat = 1;
2785         return 0;
2786 out0:
2787         *stat = 0;
2788         return 0;
2789
2790 error0:
2791         return error;
2792 }
2793
2794 struct xfs_btree_split_args {
2795         struct xfs_btree_cur    *cur;
2796         int                     level;
2797         union xfs_btree_ptr     *ptrp;
2798         union xfs_btree_key     *key;
2799         struct xfs_btree_cur    **curp;
2800         int                     *stat;          /* success/failure */
2801         int                     result;
2802         bool                    kswapd; /* allocation in kswapd context */
2803         struct completion       *done;
2804         struct work_struct      work;
2805 };
2806
2807 /*
2808  * Stack switching interfaces for allocation
2809  */
2810 static void
2811 xfs_btree_split_worker(
2812         struct work_struct      *work)
2813 {
2814         struct xfs_btree_split_args     *args = container_of(work,
2815                                                 struct xfs_btree_split_args, work);
2816         unsigned long           pflags;
2817         unsigned long           new_pflags = PF_MEMALLOC_NOFS;
2818
2819         /*
2820          * we are in a transaction context here, but may also be doing work
2821          * in kswapd context, and hence we may need to inherit that state
2822          * temporarily to ensure that we don't block waiting for memory reclaim
2823          * in any way.
2824          */
2825         if (args->kswapd)
2826                 new_pflags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
2827
2828         current_set_flags_nested(&pflags, new_pflags);
2829
2830         args->result = __xfs_btree_split(args->cur, args->level, args->ptrp,
2831                                          args->key, args->curp, args->stat);
2832         complete(args->done);
2833
2834         current_restore_flags_nested(&pflags, new_pflags);
2835 }
2836
2837 /*
2838  * BMBT split requests often come in with little stack to work on. Push
2839  * them off to a worker thread so there is lots of stack to use. For the other
2840  * btree types, just call directly to avoid the context switch overhead here.
2841  */
2842 STATIC int                                      /* error */
2843 xfs_btree_split(
2844         struct xfs_btree_cur    *cur,
2845         int                     level,
2846         union xfs_btree_ptr     *ptrp,
2847         union xfs_btree_key     *key,
2848         struct xfs_btree_cur    **curp,
2849         int                     *stat)          /* success/failure */
2850 {
2851         struct xfs_btree_split_args     args;
2852         DECLARE_COMPLETION_ONSTACK(done);
2853
2854         if (cur->bc_btnum != XFS_BTNUM_BMAP)
2855                 return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
2856
2857         args.cur = cur;
2858         args.level = level;
2859         args.ptrp = ptrp;
2860         args.key = key;
2861         args.curp = curp;
2862         args.stat = stat;
2863         args.done = &done;
2864         args.kswapd = current_is_kswapd();
2865         INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker);
2866         queue_work(xfs_alloc_wq, &args.work);
2867         wait_for_completion(&done);
2868         destroy_work_on_stack(&args.work);
2869         return args.result;
2870 }
2871
2872
2873 /*
2874  * Copy the old inode root contents into a real block and make the
2875  * broot point to it.
2876  */
2877 int                                             /* error */
2878 xfs_btree_new_iroot(
2879         struct xfs_btree_cur    *cur,           /* btree cursor */
2880         int                     *logflags,      /* logging flags for inode */
2881         int                     *stat)          /* return status - 0 fail */
2882 {
2883         struct xfs_buf          *cbp;           /* buffer for cblock */
2884         struct xfs_btree_block  *block;         /* btree block */
2885         struct xfs_btree_block  *cblock;        /* child btree block */
2886         union xfs_btree_key     *ckp;           /* child key pointer */
2887         union xfs_btree_ptr     *cpp;           /* child ptr pointer */
2888         union xfs_btree_key     *kp;            /* pointer to btree key */
2889         union xfs_btree_ptr     *pp;            /* pointer to block addr */
2890         union xfs_btree_ptr     nptr;           /* new block addr */
2891         int                     level;          /* btree level */
2892         int                     error;          /* error return code */
2893         int                     i;              /* loop counter */
2894
2895         XFS_BTREE_STATS_INC(cur, newroot);
2896
2897         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2898
2899         level = cur->bc_nlevels - 1;
2900
2901         block = xfs_btree_get_iroot(cur);
2902         pp = xfs_btree_ptr_addr(cur, 1, block);
2903
2904         /* Allocate the new block. If we can't do it, we're toast. Give up. */
2905         error = cur->bc_ops->alloc_block(cur, pp, &nptr, stat);
2906         if (error)
2907                 goto error0;
2908         if (*stat == 0)
2909                 return 0;
2910
2911         XFS_BTREE_STATS_INC(cur, alloc);
2912
2913         /* Copy the root into a real block. */
2914         error = xfs_btree_get_buf_block(cur, &nptr, &cblock, &cbp);
2915         if (error)
2916                 goto error0;
2917
2918         /*
2919          * we can't just memcpy() the root in for CRC enabled btree blocks.
2920          * In that case have to also ensure the blkno remains correct
2921          */
2922         memcpy(cblock, block, xfs_btree_block_len(cur));
2923         if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
2924                 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
2925                         cblock->bb_u.l.bb_blkno = cpu_to_be64(cbp->b_bn);
2926                 else
2927                         cblock->bb_u.s.bb_blkno = cpu_to_be64(cbp->b_bn);
2928         }
2929
2930         be16_add_cpu(&block->bb_level, 1);
2931         xfs_btree_set_numrecs(block, 1);
2932         cur->bc_nlevels++;
2933         cur->bc_ptrs[level + 1] = 1;
2934
2935         kp = xfs_btree_key_addr(cur, 1, block);
2936         ckp = xfs_btree_key_addr(cur, 1, cblock);
2937         xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
2938
2939         cpp = xfs_btree_ptr_addr(cur, 1, cblock);
2940         for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
2941                 error = xfs_btree_debug_check_ptr(cur, pp, i, level);
2942                 if (error)
2943                         goto error0;
2944         }
2945
2946         xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
2947
2948         error = xfs_btree_debug_check_ptr(cur, &nptr, 0, level);
2949         if (error)
2950                 goto error0;
2951
2952         xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
2953
2954         xfs_iroot_realloc(cur->bc_ino.ip,
2955                           1 - xfs_btree_get_numrecs(cblock),
2956                           cur->bc_ino.whichfork);
2957
2958         xfs_btree_setbuf(cur, level, cbp);
2959
2960         /*
2961          * Do all this logging at the end so that
2962          * the root is at the right level.
2963          */
2964         xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
2965         xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2966         xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2967
2968         *logflags |=
2969                 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork);
2970         *stat = 1;
2971         return 0;
2972 error0:
2973         return error;
2974 }
2975
2976 /*
2977  * Allocate a new root block, fill it in.
2978  */
2979 STATIC int                              /* error */
2980 xfs_btree_new_root(
2981         struct xfs_btree_cur    *cur,   /* btree cursor */
2982         int                     *stat)  /* success/failure */
2983 {
2984         struct xfs_btree_block  *block; /* one half of the old root block */
2985         struct xfs_buf          *bp;    /* buffer containing block */
2986         int                     error;  /* error return value */
2987         struct xfs_buf          *lbp;   /* left buffer pointer */
2988         struct xfs_btree_block  *left;  /* left btree block */
2989         struct xfs_buf          *nbp;   /* new (root) buffer */
2990         struct xfs_btree_block  *new;   /* new (root) btree block */
2991         int                     nptr;   /* new value for key index, 1 or 2 */
2992         struct xfs_buf          *rbp;   /* right buffer pointer */
2993         struct xfs_btree_block  *right; /* right btree block */
2994         union xfs_btree_ptr     rptr;
2995         union xfs_btree_ptr     lptr;
2996
2997         XFS_BTREE_STATS_INC(cur, newroot);
2998
2999         /* initialise our start point from the cursor */
3000         cur->bc_ops->init_ptr_from_cur(cur, &rptr);
3001
3002         /* Allocate the new block. If we can't do it, we're toast. Give up. */
3003         error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, stat);
3004         if (error)
3005                 goto error0;
3006         if (*stat == 0)
3007                 goto out0;
3008         XFS_BTREE_STATS_INC(cur, alloc);
3009
3010         /* Set up the new block. */
3011         error = xfs_btree_get_buf_block(cur, &lptr, &new, &nbp);
3012         if (error)
3013                 goto error0;
3014
3015         /* Set the root in the holding structure  increasing the level by 1. */
3016         cur->bc_ops->set_root(cur, &lptr, 1);
3017
3018         /*
3019          * At the previous root level there are now two blocks: the old root,
3020          * and the new block generated when it was split.  We don't know which
3021          * one the cursor is pointing at, so we set up variables "left" and
3022          * "right" for each case.
3023          */
3024         block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
3025
3026 #ifdef DEBUG
3027         error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
3028         if (error)
3029                 goto error0;
3030 #endif
3031
3032         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3033         if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3034                 /* Our block is left, pick up the right block. */
3035                 lbp = bp;
3036                 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
3037                 left = block;
3038                 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
3039                 if (error)
3040                         goto error0;
3041                 bp = rbp;
3042                 nptr = 1;
3043         } else {
3044                 /* Our block is right, pick up the left block. */
3045                 rbp = bp;
3046                 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
3047                 right = block;
3048                 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
3049                 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3050                 if (error)
3051                         goto error0;
3052                 bp = lbp;
3053                 nptr = 2;
3054         }
3055
3056         /* Fill in the new block's btree header and log it. */
3057         xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
3058         xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
3059         ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
3060                         !xfs_btree_ptr_is_null(cur, &rptr));
3061
3062         /* Fill in the key data in the new root. */
3063         if (xfs_btree_get_level(left) > 0) {
3064                 /*
3065                  * Get the keys for the left block's keys and put them directly
3066                  * in the parent block.  Do the same for the right block.
3067                  */
3068                 xfs_btree_get_node_keys(cur, left,
3069                                 xfs_btree_key_addr(cur, 1, new));
3070                 xfs_btree_get_node_keys(cur, right,
3071                                 xfs_btree_key_addr(cur, 2, new));
3072         } else {
3073                 /*
3074                  * Get the keys for the left block's records and put them
3075                  * directly in the parent block.  Do the same for the right
3076                  * block.
3077                  */
3078                 xfs_btree_get_leaf_keys(cur, left,
3079                         xfs_btree_key_addr(cur, 1, new));
3080                 xfs_btree_get_leaf_keys(cur, right,
3081                         xfs_btree_key_addr(cur, 2, new));
3082         }
3083         xfs_btree_log_keys(cur, nbp, 1, 2);
3084
3085         /* Fill in the pointer data in the new root. */
3086         xfs_btree_copy_ptrs(cur,
3087                 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
3088         xfs_btree_copy_ptrs(cur,
3089                 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
3090         xfs_btree_log_ptrs(cur, nbp, 1, 2);
3091
3092         /* Fix up the cursor. */
3093         xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
3094         cur->bc_ptrs[cur->bc_nlevels] = nptr;
3095         cur->bc_nlevels++;
3096         *stat = 1;
3097         return 0;
3098 error0:
3099         return error;
3100 out0:
3101         *stat = 0;
3102         return 0;
3103 }
3104
3105 STATIC int
3106 xfs_btree_make_block_unfull(
3107         struct xfs_btree_cur    *cur,   /* btree cursor */
3108         int                     level,  /* btree level */
3109         int                     numrecs,/* # of recs in block */
3110         int                     *oindex,/* old tree index */
3111         int                     *index, /* new tree index */
3112         union xfs_btree_ptr     *nptr,  /* new btree ptr */
3113         struct xfs_btree_cur    **ncur, /* new btree cursor */
3114         union xfs_btree_key     *key,   /* key of new block */
3115         int                     *stat)
3116 {
3117         int                     error = 0;
3118
3119         if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3120             level == cur->bc_nlevels - 1) {
3121                 struct xfs_inode *ip = cur->bc_ino.ip;
3122
3123                 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
3124                         /* A root block that can be made bigger. */
3125                         xfs_iroot_realloc(ip, 1, cur->bc_ino.whichfork);
3126                         *stat = 1;
3127                 } else {
3128                         /* A root block that needs replacing */
3129                         int     logflags = 0;
3130
3131                         error = xfs_btree_new_iroot(cur, &logflags, stat);
3132                         if (error || *stat == 0)
3133                                 return error;
3134
3135                         xfs_trans_log_inode(cur->bc_tp, ip, logflags);
3136                 }
3137
3138                 return 0;
3139         }
3140
3141         /* First, try shifting an entry to the right neighbor. */
3142         error = xfs_btree_rshift(cur, level, stat);
3143         if (error || *stat)
3144                 return error;
3145
3146         /* Next, try shifting an entry to the left neighbor. */
3147         error = xfs_btree_lshift(cur, level, stat);
3148         if (error)
3149                 return error;
3150
3151         if (*stat) {
3152                 *oindex = *index = cur->bc_ptrs[level];
3153                 return 0;
3154         }
3155
3156         /*
3157          * Next, try splitting the current block in half.
3158          *
3159          * If this works we have to re-set our variables because we
3160          * could be in a different block now.
3161          */
3162         error = xfs_btree_split(cur, level, nptr, key, ncur, stat);
3163         if (error || *stat == 0)
3164                 return error;
3165
3166
3167         *index = cur->bc_ptrs[level];
3168         return 0;
3169 }
3170
3171 /*
3172  * Insert one record/level.  Return information to the caller
3173  * allowing the next level up to proceed if necessary.
3174  */
3175 STATIC int
3176 xfs_btree_insrec(
3177         struct xfs_btree_cur    *cur,   /* btree cursor */
3178         int                     level,  /* level to insert record at */
3179         union xfs_btree_ptr     *ptrp,  /* i/o: block number inserted */
3180         union xfs_btree_rec     *rec,   /* record to insert */
3181         union xfs_btree_key     *key,   /* i/o: block key for ptrp */
3182         struct xfs_btree_cur    **curp, /* output: new cursor replacing cur */
3183         int                     *stat)  /* success/failure */
3184 {
3185         struct xfs_btree_block  *block; /* btree block */
3186         struct xfs_buf          *bp;    /* buffer for block */
3187         union xfs_btree_ptr     nptr;   /* new block ptr */
3188         struct xfs_btree_cur    *ncur;  /* new btree cursor */
3189         union xfs_btree_key     nkey;   /* new block key */
3190         union xfs_btree_key     *lkey;
3191         int                     optr;   /* old key/record index */
3192         int                     ptr;    /* key/record index */
3193         int                     numrecs;/* number of records */
3194         int                     error;  /* error return value */
3195         int                     i;
3196         xfs_daddr_t             old_bn;
3197
3198         ncur = NULL;
3199         lkey = &nkey;
3200
3201         /*
3202          * If we have an external root pointer, and we've made it to the
3203          * root level, allocate a new root block and we're done.
3204          */
3205         if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3206             (level >= cur->bc_nlevels)) {
3207                 error = xfs_btree_new_root(cur, stat);
3208                 xfs_btree_set_ptr_null(cur, ptrp);
3209
3210                 return error;
3211         }
3212
3213         /* If we're off the left edge, return failure. */
3214         ptr = cur->bc_ptrs[level];
3215         if (ptr == 0) {
3216                 *stat = 0;
3217                 return 0;
3218         }
3219
3220         optr = ptr;
3221
3222         XFS_BTREE_STATS_INC(cur, insrec);
3223
3224         /* Get pointers to the btree buffer and block. */
3225         block = xfs_btree_get_block(cur, level, &bp);
3226         old_bn = bp ? bp->b_bn : XFS_BUF_DADDR_NULL;
3227         numrecs = xfs_btree_get_numrecs(block);
3228
3229 #ifdef DEBUG
3230         error = xfs_btree_check_block(cur, block, level, bp);
3231         if (error)
3232                 goto error0;
3233
3234         /* Check that the new entry is being inserted in the right place. */
3235         if (ptr <= numrecs) {
3236                 if (level == 0) {
3237                         ASSERT(cur->bc_ops->recs_inorder(cur, rec,
3238                                 xfs_btree_rec_addr(cur, ptr, block)));
3239                 } else {
3240                         ASSERT(cur->bc_ops->keys_inorder(cur, key,
3241                                 xfs_btree_key_addr(cur, ptr, block)));
3242                 }
3243         }
3244 #endif
3245
3246         /*
3247          * If the block is full, we can't insert the new entry until we
3248          * make the block un-full.
3249          */
3250         xfs_btree_set_ptr_null(cur, &nptr);
3251         if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
3252                 error = xfs_btree_make_block_unfull(cur, level, numrecs,
3253                                         &optr, &ptr, &nptr, &ncur, lkey, stat);
3254                 if (error || *stat == 0)
3255                         goto error0;
3256         }
3257
3258         /*
3259          * The current block may have changed if the block was
3260          * previously full and we have just made space in it.
3261          */
3262         block = xfs_btree_get_block(cur, level, &bp);
3263         numrecs = xfs_btree_get_numrecs(block);
3264
3265 #ifdef DEBUG
3266         error = xfs_btree_check_block(cur, block, level, bp);
3267         if (error)
3268                 return error;
3269 #endif
3270
3271         /*
3272          * At this point we know there's room for our new entry in the block
3273          * we're pointing at.
3274          */
3275         XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
3276
3277         if (level > 0) {
3278                 /* It's a nonleaf. make a hole in the keys and ptrs */
3279                 union xfs_btree_key     *kp;
3280                 union xfs_btree_ptr     *pp;
3281
3282                 kp = xfs_btree_key_addr(cur, ptr, block);
3283                 pp = xfs_btree_ptr_addr(cur, ptr, block);
3284
3285                 for (i = numrecs - ptr; i >= 0; i--) {
3286                         error = xfs_btree_debug_check_ptr(cur, pp, i, level);
3287                         if (error)
3288                                 return error;
3289                 }
3290
3291                 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
3292                 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
3293
3294                 error = xfs_btree_debug_check_ptr(cur, ptrp, 0, level);
3295                 if (error)
3296                         goto error0;
3297
3298                 /* Now put the new data in, bump numrecs and log it. */
3299                 xfs_btree_copy_keys(cur, kp, key, 1);
3300                 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
3301                 numrecs++;
3302                 xfs_btree_set_numrecs(block, numrecs);
3303                 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
3304                 xfs_btree_log_keys(cur, bp, ptr, numrecs);
3305 #ifdef DEBUG
3306                 if (ptr < numrecs) {
3307                         ASSERT(cur->bc_ops->keys_inorder(cur, kp,
3308                                 xfs_btree_key_addr(cur, ptr + 1, block)));
3309                 }
3310 #endif
3311         } else {
3312                 /* It's a leaf. make a hole in the records */
3313                 union xfs_btree_rec             *rp;
3314
3315                 rp = xfs_btree_rec_addr(cur, ptr, block);
3316
3317                 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
3318
3319                 /* Now put the new data in, bump numrecs and log it. */
3320                 xfs_btree_copy_recs(cur, rp, rec, 1);
3321                 xfs_btree_set_numrecs(block, ++numrecs);
3322                 xfs_btree_log_recs(cur, bp, ptr, numrecs);
3323 #ifdef DEBUG
3324                 if (ptr < numrecs) {
3325                         ASSERT(cur->bc_ops->recs_inorder(cur, rp,
3326                                 xfs_btree_rec_addr(cur, ptr + 1, block)));
3327                 }
3328 #endif
3329         }
3330
3331         /* Log the new number of records in the btree header. */
3332         xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3333
3334         /*
3335          * If we just inserted into a new tree block, we have to
3336          * recalculate nkey here because nkey is out of date.
3337          *
3338          * Otherwise we're just updating an existing block (having shoved
3339          * some records into the new tree block), so use the regular key
3340          * update mechanism.
3341          */
3342         if (bp && bp->b_bn != old_bn) {
3343                 xfs_btree_get_keys(cur, block, lkey);
3344         } else if (xfs_btree_needs_key_update(cur, optr)) {
3345                 error = xfs_btree_update_keys(cur, level);
3346                 if (error)
3347                         goto error0;
3348         }
3349
3350         /*
3351          * If we are tracking the last record in the tree and
3352          * we are at the far right edge of the tree, update it.
3353          */
3354         if (xfs_btree_is_lastrec(cur, block, level)) {
3355                 cur->bc_ops->update_lastrec(cur, block, rec,
3356                                             ptr, LASTREC_INSREC);
3357         }
3358
3359         /*
3360          * Return the new block number, if any.
3361          * If there is one, give back a record value and a cursor too.
3362          */
3363         *ptrp = nptr;
3364         if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3365                 xfs_btree_copy_keys(cur, key, lkey, 1);
3366                 *curp = ncur;
3367         }
3368
3369         *stat = 1;
3370         return 0;
3371
3372 error0:
3373         return error;
3374 }
3375
3376 /*
3377  * Insert the record at the point referenced by cur.
3378  *
3379  * A multi-level split of the tree on insert will invalidate the original
3380  * cursor.  All callers of this function should assume that the cursor is
3381  * no longer valid and revalidate it.
3382  */
3383 int
3384 xfs_btree_insert(
3385         struct xfs_btree_cur    *cur,
3386         int                     *stat)
3387 {
3388         int                     error;  /* error return value */
3389         int                     i;      /* result value, 0 for failure */
3390         int                     level;  /* current level number in btree */
3391         union xfs_btree_ptr     nptr;   /* new block number (split result) */
3392         struct xfs_btree_cur    *ncur;  /* new cursor (split result) */
3393         struct xfs_btree_cur    *pcur;  /* previous level's cursor */
3394         union xfs_btree_key     bkey;   /* key of block to insert */
3395         union xfs_btree_key     *key;
3396         union xfs_btree_rec     rec;    /* record to insert */
3397
3398         level = 0;
3399         ncur = NULL;
3400         pcur = cur;
3401         key = &bkey;
3402
3403         xfs_btree_set_ptr_null(cur, &nptr);
3404
3405         /* Make a key out of the record data to be inserted, and save it. */
3406         cur->bc_ops->init_rec_from_cur(cur, &rec);
3407         cur->bc_ops->init_key_from_rec(key, &rec);
3408
3409         /*
3410          * Loop going up the tree, starting at the leaf level.
3411          * Stop when we don't get a split block, that must mean that
3412          * the insert is finished with this level.
3413          */
3414         do {
3415                 /*
3416                  * Insert nrec/nptr into this level of the tree.
3417                  * Note if we fail, nptr will be null.
3418                  */
3419                 error = xfs_btree_insrec(pcur, level, &nptr, &rec, key,
3420                                 &ncur, &i);
3421                 if (error) {
3422                         if (pcur != cur)
3423                                 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3424                         goto error0;
3425                 }
3426
3427                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3428                         error = -EFSCORRUPTED;
3429                         goto error0;
3430                 }
3431                 level++;
3432
3433                 /*
3434                  * See if the cursor we just used is trash.
3435                  * Can't trash the caller's cursor, but otherwise we should
3436                  * if ncur is a new cursor or we're about to be done.
3437                  */
3438                 if (pcur != cur &&
3439                     (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3440                         /* Save the state from the cursor before we trash it */
3441                         if (cur->bc_ops->update_cursor)
3442                                 cur->bc_ops->update_cursor(pcur, cur);
3443                         cur->bc_nlevels = pcur->bc_nlevels;
3444                         xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3445                 }
3446                 /* If we got a new cursor, switch to it. */
3447                 if (ncur) {
3448                         pcur = ncur;
3449                         ncur = NULL;
3450                 }
3451         } while (!xfs_btree_ptr_is_null(cur, &nptr));
3452
3453         *stat = i;
3454         return 0;
3455 error0:
3456         return error;
3457 }
3458
3459 /*
3460  * Try to merge a non-leaf block back into the inode root.
3461  *
3462  * Note: the killroot names comes from the fact that we're effectively
3463  * killing the old root block.  But because we can't just delete the
3464  * inode we have to copy the single block it was pointing to into the
3465  * inode.
3466  */
3467 STATIC int
3468 xfs_btree_kill_iroot(
3469         struct xfs_btree_cur    *cur)
3470 {
3471         int                     whichfork = cur->bc_ino.whichfork;
3472         struct xfs_inode        *ip = cur->bc_ino.ip;
3473         struct xfs_ifork        *ifp = XFS_IFORK_PTR(ip, whichfork);
3474         struct xfs_btree_block  *block;
3475         struct xfs_btree_block  *cblock;
3476         union xfs_btree_key     *kp;
3477         union xfs_btree_key     *ckp;
3478         union xfs_btree_ptr     *pp;
3479         union xfs_btree_ptr     *cpp;
3480         struct xfs_buf          *cbp;
3481         int                     level;
3482         int                     index;
3483         int                     numrecs;
3484         int                     error;
3485 #ifdef DEBUG
3486         union xfs_btree_ptr     ptr;
3487 #endif
3488         int                     i;
3489
3490         ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3491         ASSERT(cur->bc_nlevels > 1);
3492
3493         /*
3494          * Don't deal with the root block needs to be a leaf case.
3495          * We're just going to turn the thing back into extents anyway.
3496          */
3497         level = cur->bc_nlevels - 1;
3498         if (level == 1)
3499                 goto out0;
3500
3501         /*
3502          * Give up if the root has multiple children.
3503          */
3504         block = xfs_btree_get_iroot(cur);
3505         if (xfs_btree_get_numrecs(block) != 1)
3506                 goto out0;
3507
3508         cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3509         numrecs = xfs_btree_get_numrecs(cblock);
3510
3511         /*
3512          * Only do this if the next level will fit.
3513          * Then the data must be copied up to the inode,
3514          * instead of freeing the root you free the next level.
3515          */
3516         if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3517                 goto out0;
3518
3519         XFS_BTREE_STATS_INC(cur, killroot);
3520
3521 #ifdef DEBUG
3522         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3523         ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3524         xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3525         ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3526 #endif
3527
3528         index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3529         if (index) {
3530                 xfs_iroot_realloc(cur->bc_ino.ip, index,
3531                                   cur->bc_ino.whichfork);
3532                 block = ifp->if_broot;
3533         }
3534
3535         be16_add_cpu(&block->bb_numrecs, index);
3536         ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3537
3538         kp = xfs_btree_key_addr(cur, 1, block);
3539         ckp = xfs_btree_key_addr(cur, 1, cblock);
3540         xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3541
3542         pp = xfs_btree_ptr_addr(cur, 1, block);
3543         cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3544
3545         for (i = 0; i < numrecs; i++) {
3546                 error = xfs_btree_debug_check_ptr(cur, cpp, i, level - 1);
3547                 if (error)
3548                         return error;
3549         }
3550
3551         xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3552
3553         error = xfs_btree_free_block(cur, cbp);
3554         if (error)
3555                 return error;
3556
3557         cur->bc_bufs[level - 1] = NULL;
3558         be16_add_cpu(&block->bb_level, -1);
3559         xfs_trans_log_inode(cur->bc_tp, ip,
3560                 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork));
3561         cur->bc_nlevels--;
3562 out0:
3563         return 0;
3564 }
3565
3566 /*
3567  * Kill the current root node, and replace it with it's only child node.
3568  */
3569 STATIC int
3570 xfs_btree_kill_root(
3571         struct xfs_btree_cur    *cur,
3572         struct xfs_buf          *bp,
3573         int                     level,
3574         union xfs_btree_ptr     *newroot)
3575 {
3576         int                     error;
3577
3578         XFS_BTREE_STATS_INC(cur, killroot);
3579
3580         /*
3581          * Update the root pointer, decreasing the level by 1 and then
3582          * free the old root.
3583          */
3584         cur->bc_ops->set_root(cur, newroot, -1);
3585
3586         error = xfs_btree_free_block(cur, bp);
3587         if (error)
3588                 return error;
3589
3590         cur->bc_bufs[level] = NULL;
3591         cur->bc_ra[level] = 0;
3592         cur->bc_nlevels--;
3593
3594         return 0;
3595 }
3596
3597 STATIC int
3598 xfs_btree_dec_cursor(
3599         struct xfs_btree_cur    *cur,
3600         int                     level,
3601         int                     *stat)
3602 {
3603         int                     error;
3604         int                     i;
3605
3606         if (level > 0) {
3607                 error = xfs_btree_decrement(cur, level, &i);
3608                 if (error)
3609                         return error;
3610         }
3611
3612         *stat = 1;
3613         return 0;
3614 }
3615
3616 /*
3617  * Single level of the btree record deletion routine.
3618  * Delete record pointed to by cur/level.
3619  * Remove the record from its block then rebalance the tree.
3620  * Return 0 for error, 1 for done, 2 to go on to the next level.
3621  */
3622 STATIC int                                      /* error */
3623 xfs_btree_delrec(
3624         struct xfs_btree_cur    *cur,           /* btree cursor */
3625         int                     level,          /* level removing record from */
3626         int                     *stat)          /* fail/done/go-on */
3627 {
3628         struct xfs_btree_block  *block;         /* btree block */
3629         union xfs_btree_ptr     cptr;           /* current block ptr */
3630         struct xfs_buf          *bp;            /* buffer for block */
3631         int                     error;          /* error return value */
3632         int                     i;              /* loop counter */
3633         union xfs_btree_ptr     lptr;           /* left sibling block ptr */
3634         struct xfs_buf          *lbp;           /* left buffer pointer */
3635         struct xfs_btree_block  *left;          /* left btree block */
3636         int                     lrecs = 0;      /* left record count */
3637         int                     ptr;            /* key/record index */
3638         union xfs_btree_ptr     rptr;           /* right sibling block ptr */
3639         struct xfs_buf          *rbp;           /* right buffer pointer */
3640         struct xfs_btree_block  *right;         /* right btree block */
3641         struct xfs_btree_block  *rrblock;       /* right-right btree block */
3642         struct xfs_buf          *rrbp;          /* right-right buffer pointer */
3643         int                     rrecs = 0;      /* right record count */
3644         struct xfs_btree_cur    *tcur;          /* temporary btree cursor */
3645         int                     numrecs;        /* temporary numrec count */
3646
3647         tcur = NULL;
3648
3649         /* Get the index of the entry being deleted, check for nothing there. */
3650         ptr = cur->bc_ptrs[level];
3651         if (ptr == 0) {
3652                 *stat = 0;
3653                 return 0;
3654         }
3655
3656         /* Get the buffer & block containing the record or key/ptr. */
3657         block = xfs_btree_get_block(cur, level, &bp);
3658         numrecs = xfs_btree_get_numrecs(block);
3659
3660 #ifdef DEBUG
3661         error = xfs_btree_check_block(cur, block, level, bp);
3662         if (error)
3663                 goto error0;
3664 #endif
3665
3666         /* Fail if we're off the end of the block. */
3667         if (ptr > numrecs) {
3668                 *stat = 0;
3669                 return 0;
3670         }
3671
3672         XFS_BTREE_STATS_INC(cur, delrec);
3673         XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3674
3675         /* Excise the entries being deleted. */
3676         if (level > 0) {
3677                 /* It's a nonleaf. operate on keys and ptrs */
3678                 union xfs_btree_key     *lkp;
3679                 union xfs_btree_ptr     *lpp;
3680
3681                 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3682                 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3683
3684                 for (i = 0; i < numrecs - ptr; i++) {
3685                         error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
3686                         if (error)
3687                                 goto error0;
3688                 }
3689
3690                 if (ptr < numrecs) {
3691                         xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3692                         xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3693                         xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3694                         xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3695                 }
3696         } else {
3697                 /* It's a leaf. operate on records */
3698                 if (ptr < numrecs) {
3699                         xfs_btree_shift_recs(cur,
3700                                 xfs_btree_rec_addr(cur, ptr + 1, block),
3701                                 -1, numrecs - ptr);
3702                         xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3703                 }
3704         }
3705
3706         /*
3707          * Decrement and log the number of entries in the block.
3708          */
3709         xfs_btree_set_numrecs(block, --numrecs);
3710         xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3711
3712         /*
3713          * If we are tracking the last record in the tree and
3714          * we are at the far right edge of the tree, update it.
3715          */
3716         if (xfs_btree_is_lastrec(cur, block, level)) {
3717                 cur->bc_ops->update_lastrec(cur, block, NULL,
3718                                             ptr, LASTREC_DELREC);
3719         }
3720
3721         /*
3722          * We're at the root level.  First, shrink the root block in-memory.
3723          * Try to get rid of the next level down.  If we can't then there's
3724          * nothing left to do.
3725          */
3726         if (level == cur->bc_nlevels - 1) {
3727                 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3728                         xfs_iroot_realloc(cur->bc_ino.ip, -1,
3729                                           cur->bc_ino.whichfork);
3730
3731                         error = xfs_btree_kill_iroot(cur);
3732                         if (error)
3733                                 goto error0;
3734
3735                         error = xfs_btree_dec_cursor(cur, level, stat);
3736                         if (error)
3737                                 goto error0;
3738                         *stat = 1;
3739                         return 0;
3740                 }
3741
3742                 /*
3743                  * If this is the root level, and there's only one entry left,
3744                  * and it's NOT the leaf level, then we can get rid of this
3745                  * level.
3746                  */
3747                 if (numrecs == 1 && level > 0) {
3748                         union xfs_btree_ptr     *pp;
3749                         /*
3750                          * pp is still set to the first pointer in the block.
3751                          * Make it the new root of the btree.
3752                          */
3753                         pp = xfs_btree_ptr_addr(cur, 1, block);
3754                         error = xfs_btree_kill_root(cur, bp, level, pp);
3755                         if (error)
3756                                 goto error0;
3757                 } else if (level > 0) {
3758                         error = xfs_btree_dec_cursor(cur, level, stat);
3759                         if (error)
3760                                 goto error0;
3761                 }
3762                 *stat = 1;
3763                 return 0;
3764         }
3765
3766         /*
3767          * If we deleted the leftmost entry in the block, update the
3768          * key values above us in the tree.
3769          */
3770         if (xfs_btree_needs_key_update(cur, ptr)) {
3771                 error = xfs_btree_update_keys(cur, level);
3772                 if (error)
3773                         goto error0;
3774         }
3775
3776         /*
3777          * If the number of records remaining in the block is at least
3778          * the minimum, we're done.
3779          */
3780         if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3781                 error = xfs_btree_dec_cursor(cur, level, stat);
3782                 if (error)
3783                         goto error0;
3784                 return 0;
3785         }
3786
3787         /*
3788          * Otherwise, we have to move some records around to keep the
3789          * tree balanced.  Look at the left and right sibling blocks to
3790          * see if we can re-balance by moving only one record.
3791          */
3792         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3793         xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3794
3795         if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3796                 /*
3797                  * One child of root, need to get a chance to copy its contents
3798                  * into the root and delete it. Can't go up to next level,
3799                  * there's nothing to delete there.
3800                  */
3801                 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3802                     xfs_btree_ptr_is_null(cur, &lptr) &&
3803                     level == cur->bc_nlevels - 2) {
3804                         error = xfs_btree_kill_iroot(cur);
3805                         if (!error)
3806                                 error = xfs_btree_dec_cursor(cur, level, stat);
3807                         if (error)
3808                                 goto error0;
3809                         return 0;
3810                 }
3811         }
3812
3813         ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3814                !xfs_btree_ptr_is_null(cur, &lptr));
3815
3816         /*
3817          * Duplicate the cursor so our btree manipulations here won't
3818          * disrupt the next level up.
3819          */
3820         error = xfs_btree_dup_cursor(cur, &tcur);
3821         if (error)
3822                 goto error0;
3823
3824         /*
3825          * If there's a right sibling, see if it's ok to shift an entry
3826          * out of it.
3827          */
3828         if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3829                 /*
3830                  * Move the temp cursor to the last entry in the next block.
3831                  * Actually any entry but the first would suffice.
3832                  */
3833                 i = xfs_btree_lastrec(tcur, level);
3834                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3835                         error = -EFSCORRUPTED;
3836                         goto error0;
3837                 }
3838
3839                 error = xfs_btree_increment(tcur, level, &i);
3840                 if (error)
3841                         goto error0;
3842                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3843                         error = -EFSCORRUPTED;
3844                         goto error0;
3845                 }
3846
3847                 i = xfs_btree_lastrec(tcur, level);
3848                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3849                         error = -EFSCORRUPTED;
3850                         goto error0;
3851                 }
3852
3853                 /* Grab a pointer to the block. */
3854                 right = xfs_btree_get_block(tcur, level, &rbp);
3855 #ifdef DEBUG
3856                 error = xfs_btree_check_block(tcur, right, level, rbp);
3857                 if (error)
3858                         goto error0;
3859 #endif
3860                 /* Grab the current block number, for future use. */
3861                 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3862
3863                 /*
3864                  * If right block is full enough so that removing one entry
3865                  * won't make it too empty, and left-shifting an entry out
3866                  * of right to us works, we're done.
3867                  */
3868                 if (xfs_btree_get_numrecs(right) - 1 >=
3869                     cur->bc_ops->get_minrecs(tcur, level)) {
3870                         error = xfs_btree_lshift(tcur, level, &i);
3871                         if (error)
3872                                 goto error0;
3873                         if (i) {
3874                                 ASSERT(xfs_btree_get_numrecs(block) >=
3875                                        cur->bc_ops->get_minrecs(tcur, level));
3876
3877                                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3878                                 tcur = NULL;
3879
3880                                 error = xfs_btree_dec_cursor(cur, level, stat);
3881                                 if (error)
3882                                         goto error0;
3883                                 return 0;
3884                         }
3885                 }
3886
3887                 /*
3888                  * Otherwise, grab the number of records in right for
3889                  * future reference, and fix up the temp cursor to point
3890                  * to our block again (last record).
3891                  */
3892                 rrecs = xfs_btree_get_numrecs(right);
3893                 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3894                         i = xfs_btree_firstrec(tcur, level);
3895                         if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3896                                 error = -EFSCORRUPTED;
3897                                 goto error0;
3898                         }
3899
3900                         error = xfs_btree_decrement(tcur, level, &i);
3901                         if (error)
3902                                 goto error0;
3903                         if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3904                                 error = -EFSCORRUPTED;
3905                                 goto error0;
3906                         }
3907                 }
3908         }
3909
3910         /*
3911          * If there's a left sibling, see if it's ok to shift an entry
3912          * out of it.
3913          */
3914         if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3915                 /*
3916                  * Move the temp cursor to the first entry in the
3917                  * previous block.
3918                  */
3919                 i = xfs_btree_firstrec(tcur, level);
3920                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3921                         error = -EFSCORRUPTED;
3922                         goto error0;
3923                 }
3924
3925                 error = xfs_btree_decrement(tcur, level, &i);
3926                 if (error)
3927                         goto error0;
3928                 i = xfs_btree_firstrec(tcur, level);
3929                 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3930                         error = -EFSCORRUPTED;
3931                         goto error0;
3932                 }
3933
3934                 /* Grab a pointer to the block. */
3935                 left = xfs_btree_get_block(tcur, level, &lbp);
3936 #ifdef DEBUG
3937                 error = xfs_btree_check_block(cur, left, level, lbp);
3938                 if (error)
3939                         goto error0;
3940 #endif
3941                 /* Grab the current block number, for future use. */
3942                 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
3943
3944                 /*
3945                  * If left block is full enough so that removing one entry
3946                  * won't make it too empty, and right-shifting an entry out
3947                  * of left to us works, we're done.
3948                  */
3949                 if (xfs_btree_get_numrecs(left) - 1 >=
3950                     cur->bc_ops->get_minrecs(tcur, level)) {
3951                         error = xfs_btree_rshift(tcur, level, &i);
3952                         if (error)
3953                                 goto error0;
3954                         if (i) {
3955                                 ASSERT(xfs_btree_get_numrecs(block) >=
3956                                        cur->bc_ops->get_minrecs(tcur, level));
3957                                 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3958                                 tcur = NULL;
3959                                 if (level == 0)
3960                                         cur->bc_ptrs[0]++;
3961
3962                                 *stat = 1;
3963                                 return 0;
3964                         }
3965                 }
3966
3967                 /*
3968                  * Otherwise, grab the number of records in right for
3969                  * future reference.
3970                  */
3971                 lrecs = xfs_btree_get_numrecs(left);
3972         }
3973
3974         /* Delete the temp cursor, we're done with it. */
3975         xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3976         tcur = NULL;
3977
3978         /* If here, we need to do a join to keep the tree balanced. */
3979         ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
3980
3981         if (!xfs_btree_ptr_is_null(cur, &lptr) &&
3982             lrecs + xfs_btree_get_numrecs(block) <=
3983                         cur->bc_ops->get_maxrecs(cur, level)) {
3984                 /*
3985                  * Set "right" to be the starting block,
3986                  * "left" to be the left neighbor.
3987                  */
3988                 rptr = cptr;
3989                 right = block;
3990                 rbp = bp;
3991                 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3992                 if (error)
3993                         goto error0;
3994
3995         /*
3996          * If that won't work, see if we can join with the right neighbor block.
3997          */
3998         } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
3999                    rrecs + xfs_btree_get_numrecs(block) <=
4000                         cur->bc_ops->get_maxrecs(cur, level)) {
4001                 /*
4002                  * Set "left" to be the starting block,
4003                  * "right" to be the right neighbor.
4004                  */
4005                 lptr = cptr;
4006                 left = block;
4007                 lbp = bp;
4008                 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
4009                 if (error)
4010                         goto error0;
4011
4012         /*
4013          * Otherwise, we can't fix the imbalance.
4014          * Just return.  This is probably a logic error, but it's not fatal.
4015          */
4016         } else {
4017                 error = xfs_btree_dec_cursor(cur, level, stat);
4018                 if (error)
4019                         goto error0;
4020                 return 0;
4021         }
4022
4023         rrecs = xfs_btree_get_numrecs(right);
4024         lrecs = xfs_btree_get_numrecs(left);
4025
4026         /*
4027          * We're now going to join "left" and "right" by moving all the stuff
4028          * in "right" to "left" and deleting "right".
4029          */
4030         XFS_BTREE_STATS_ADD(cur, moves, rrecs);
4031         if (level > 0) {
4032                 /* It's a non-leaf.  Move keys and pointers. */
4033                 union xfs_btree_key     *lkp;   /* left btree key */
4034                 union xfs_btree_ptr     *lpp;   /* left address pointer */
4035                 union xfs_btree_key     *rkp;   /* right btree key */
4036                 union xfs_btree_ptr     *rpp;   /* right address pointer */
4037
4038                 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
4039                 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
4040                 rkp = xfs_btree_key_addr(cur, 1, right);
4041                 rpp = xfs_btree_ptr_addr(cur, 1, right);
4042
4043                 for (i = 1; i < rrecs; i++) {
4044                         error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
4045                         if (error)
4046                                 goto error0;
4047                 }
4048
4049                 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
4050                 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
4051
4052                 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
4053                 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
4054         } else {
4055                 /* It's a leaf.  Move records.  */
4056                 union xfs_btree_rec     *lrp;   /* left record pointer */
4057                 union xfs_btree_rec     *rrp;   /* right record pointer */
4058
4059                 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
4060                 rrp = xfs_btree_rec_addr(cur, 1, right);
4061
4062                 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
4063                 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
4064         }
4065
4066         XFS_BTREE_STATS_INC(cur, join);
4067
4068         /*
4069          * Fix up the number of records and right block pointer in the
4070          * surviving block, and log it.
4071          */
4072         xfs_btree_set_numrecs(left, lrecs + rrecs);
4073         xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB);
4074         xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4075         xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
4076
4077         /* If there is a right sibling, point it to the remaining block. */
4078         xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4079         if (!xfs_btree_ptr_is_null(cur, &cptr)) {
4080                 error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp);
4081                 if (error)
4082                         goto error0;
4083                 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
4084                 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
4085         }
4086
4087         /* Free the deleted block. */
4088         error = xfs_btree_free_block(cur, rbp);
4089         if (error)
4090                 goto error0;
4091
4092         /*
4093          * If we joined with the left neighbor, set the buffer in the
4094          * cursor to the left block, and fix up the index.
4095          */
4096         if (bp != lbp) {
4097                 cur->bc_bufs[level] = lbp;
4098                 cur->bc_ptrs[level] += lrecs;
4099                 cur->bc_ra[level] = 0;
4100         }
4101         /*
4102          * If we joined with the right neighbor and there's a level above
4103          * us, increment the cursor at that level.
4104          */
4105         else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
4106                    (level + 1 < cur->bc_nlevels)) {
4107                 error = xfs_btree_increment(cur, level + 1, &i);
4108                 if (error)
4109                         goto error0;
4110         }
4111
4112         /*
4113          * Readjust the ptr at this level if it's not a leaf, since it's
4114          * still pointing at the deletion point, which makes the cursor
4115          * inconsistent.  If this makes the ptr 0, the caller fixes it up.
4116          * We can't use decrement because it would change the next level up.
4117          */
4118         if (level > 0)
4119                 cur->bc_ptrs[level]--;
4120
4121         /*
4122          * We combined blocks, so we have to update the parent keys if the
4123          * btree supports overlapped intervals.  However, bc_ptrs[level + 1]
4124          * points to the old block so that the caller knows which record to
4125          * delete.  Therefore, the caller must be savvy enough to call updkeys
4126          * for us if we return stat == 2.  The other exit points from this
4127          * function don't require deletions further up the tree, so they can
4128          * call updkeys directly.
4129          */
4130
4131         /* Return value means the next level up has something to do. */
4132         *stat = 2;
4133         return 0;
4134
4135 error0:
4136         if (tcur)
4137                 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
4138         return error;
4139 }
4140
4141 /*
4142  * Delete the record pointed to by cur.
4143  * The cursor refers to the place where the record was (could be inserted)
4144  * when the operation returns.
4145  */
4146 int                                     /* error */
4147 xfs_btree_delete(
4148         struct xfs_btree_cur    *cur,
4149         int                     *stat)  /* success/failure */
4150 {
4151         int                     error;  /* error return value */
4152         int                     level;
4153         int                     i;
4154         bool                    joined = false;
4155
4156         /*
4157          * Go up the tree, starting at leaf level.
4158          *
4159          * If 2 is returned then a join was done; go to the next level.
4160          * Otherwise we are done.
4161          */
4162         for (level = 0, i = 2; i == 2; level++) {
4163                 error = xfs_btree_delrec(cur, level, &i);
4164                 if (error)
4165                         goto error0;
4166                 if (i == 2)
4167                         joined = true;
4168         }
4169
4170         /*
4171          * If we combined blocks as part of deleting the record, delrec won't
4172          * have updated the parent high keys so we have to do that here.
4173          */
4174         if (joined && (cur->bc_flags & XFS_BTREE_OVERLAPPING)) {
4175                 error = xfs_btree_updkeys_force(cur, 0);
4176                 if (error)
4177                         goto error0;
4178         }
4179
4180         if (i == 0) {
4181                 for (level = 1; level < cur->bc_nlevels; level++) {
4182                         if (cur->bc_ptrs[level] == 0) {
4183                                 error = xfs_btree_decrement(cur, level, &i);
4184                                 if (error)
4185                                         goto error0;
4186                                 break;
4187                         }
4188                 }
4189         }
4190
4191         *stat = i;
4192         return 0;
4193 error0:
4194         return error;
4195 }
4196
4197 /*
4198  * Get the data from the pointed-to record.
4199  */
4200 int                                     /* error */
4201 xfs_btree_get_rec(
4202         struct xfs_btree_cur    *cur,   /* btree cursor */
4203         union xfs_btree_rec     **recp, /* output: btree record */
4204         int                     *stat)  /* output: success/failure */
4205 {
4206         struct xfs_btree_block  *block; /* btree block */
4207         struct xfs_buf          *bp;    /* buffer pointer */
4208         int                     ptr;    /* record number */
4209 #ifdef DEBUG
4210         int                     error;  /* error return value */
4211 #endif
4212
4213         ptr = cur->bc_ptrs[0];
4214         block = xfs_btree_get_block(cur, 0, &bp);
4215
4216 #ifdef DEBUG
4217         error = xfs_btree_check_block(cur, block, 0, bp);
4218         if (error)
4219                 return error;
4220 #endif
4221
4222         /*
4223          * Off the right end or left end, return failure.
4224          */
4225         if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
4226                 *stat = 0;
4227                 return 0;
4228         }
4229
4230         /*
4231          * Point to the record and extract its data.
4232          */
4233         *recp = xfs_btree_rec_addr(cur, ptr, block);
4234         *stat = 1;
4235         return 0;
4236 }
4237
4238 /* Visit a block in a btree. */
4239 STATIC int
4240 xfs_btree_visit_block(
4241         struct xfs_btree_cur            *cur,
4242         int                             level,
4243         xfs_btree_visit_blocks_fn       fn,
4244         void                            *data)
4245 {
4246         struct xfs_btree_block          *block;
4247         struct xfs_buf                  *bp;
4248         union xfs_btree_ptr             rptr;
4249         int                             error;
4250
4251         /* do right sibling readahead */
4252         xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
4253         block = xfs_btree_get_block(cur, level, &bp);
4254
4255         /* process the block */
4256         error = fn(cur, level, data);
4257         if (error)
4258                 return error;
4259
4260         /* now read rh sibling block for next iteration */
4261         xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4262         if (xfs_btree_ptr_is_null(cur, &rptr))
4263                 return -ENOENT;
4264
4265         return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
4266 }
4267
4268
4269 /* Visit every block in a btree. */
4270 int
4271 xfs_btree_visit_blocks(
4272         struct xfs_btree_cur            *cur,
4273         xfs_btree_visit_blocks_fn       fn,
4274         unsigned int                    flags,
4275         void                            *data)
4276 {
4277         union xfs_btree_ptr             lptr;
4278         int                             level;
4279         struct xfs_btree_block          *block = NULL;
4280         int                             error = 0;
4281
4282         cur->bc_ops->init_ptr_from_cur(cur, &lptr);
4283
4284         /* for each level */
4285         for (level = cur->bc_nlevels - 1; level >= 0; level--) {
4286                 /* grab the left hand block */
4287                 error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
4288                 if (error)
4289                         return error;
4290
4291                 /* readahead the left most block for the next level down */
4292                 if (level > 0) {
4293                         union xfs_btree_ptr     *ptr;
4294
4295                         ptr = xfs_btree_ptr_addr(cur, 1, block);
4296                         xfs_btree_readahead_ptr(cur, ptr, 1);
4297
4298                         /* save for the next iteration of the loop */
4299                         xfs_btree_copy_ptrs(cur, &lptr, ptr, 1);
4300
4301                         if (!(flags & XFS_BTREE_VISIT_LEAVES))
4302                                 continue;
4303                 } else if (!(flags & XFS_BTREE_VISIT_RECORDS)) {
4304                         continue;
4305                 }
4306
4307                 /* for each buffer in the level */
4308                 do {
4309                         error = xfs_btree_visit_block(cur, level, fn, data);
4310                 } while (!error);
4311
4312                 if (error != -ENOENT)
4313                         return error;
4314         }
4315
4316         return 0;
4317 }
4318
4319 /*
4320  * Change the owner of a btree.
4321  *
4322  * The mechanism we use here is ordered buffer logging. Because we don't know
4323  * how many buffers were are going to need to modify, we don't really want to
4324  * have to make transaction reservations for the worst case of every buffer in a
4325  * full size btree as that may be more space that we can fit in the log....
4326  *
4327  * We do the btree walk in the most optimal manner possible - we have sibling
4328  * pointers so we can just walk all the blocks on each level from left to right
4329  * in a single pass, and then move to the next level and do the same. We can
4330  * also do readahead on the sibling pointers to get IO moving more quickly,
4331  * though for slow disks this is unlikely to make much difference to performance
4332  * as the amount of CPU work we have to do before moving to the next block is
4333  * relatively small.
4334  *
4335  * For each btree block that we load, modify the owner appropriately, set the
4336  * buffer as an ordered buffer and log it appropriately. We need to ensure that
4337  * we mark the region we change dirty so that if the buffer is relogged in
4338  * a subsequent transaction the changes we make here as an ordered buffer are
4339  * correctly relogged in that transaction.  If we are in recovery context, then
4340  * just queue the modified buffer as delayed write buffer so the transaction
4341  * recovery completion writes the changes to disk.
4342  */
4343 struct xfs_btree_block_change_owner_info {
4344         uint64_t                new_owner;
4345         struct list_head        *buffer_list;
4346 };
4347
4348 static int
4349 xfs_btree_block_change_owner(
4350         struct xfs_btree_cur    *cur,
4351         int                     level,
4352         void                    *data)
4353 {
4354         struct xfs_btree_block_change_owner_info        *bbcoi = data;
4355         struct xfs_btree_block  *block;
4356         struct xfs_buf          *bp;
4357
4358         /* modify the owner */
4359         block = xfs_btree_get_block(cur, level, &bp);
4360         if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
4361                 if (block->bb_u.l.bb_owner == cpu_to_be64(bbcoi->new_owner))
4362                         return 0;
4363                 block->bb_u.l.bb_owner = cpu_to_be64(bbcoi->new_owner);
4364         } else {
4365                 if (block->bb_u.s.bb_owner == cpu_to_be32(bbcoi->new_owner))
4366                         return 0;
4367                 block->bb_u.s.bb_owner = cpu_to_be32(bbcoi->new_owner);
4368         }
4369
4370         /*
4371          * If the block is a root block hosted in an inode, we might not have a
4372          * buffer pointer here and we shouldn't attempt to log the change as the
4373          * information is already held in the inode and discarded when the root
4374          * block is formatted into the on-disk inode fork. We still change it,
4375          * though, so everything is consistent in memory.
4376          */
4377         if (!bp) {
4378                 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
4379                 ASSERT(level == cur->bc_nlevels - 1);
4380                 return 0;
4381         }
4382
4383         if (cur->bc_tp) {
4384                 if (!xfs_trans_ordered_buf(cur->bc_tp, bp)) {
4385                         xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
4386                         return -EAGAIN;
4387                 }
4388         } else {
4389                 xfs_buf_delwri_queue(bp, bbcoi->buffer_list);
4390         }
4391
4392         return 0;
4393 }
4394
4395 int
4396 xfs_btree_change_owner(
4397         struct xfs_btree_cur    *cur,
4398         uint64_t                new_owner,
4399         struct list_head        *buffer_list)
4400 {
4401         struct xfs_btree_block_change_owner_info        bbcoi;
4402
4403         bbcoi.new_owner = new_owner;
4404         bbcoi.buffer_list = buffer_list;
4405
4406         return xfs_btree_visit_blocks(cur, xfs_btree_block_change_owner,
4407                         XFS_BTREE_VISIT_ALL, &bbcoi);
4408 }
4409
4410 /* Verify the v5 fields of a long-format btree block. */
4411 xfs_failaddr_t
4412 xfs_btree_lblock_v5hdr_verify(
4413         struct xfs_buf          *bp,
4414         uint64_t                owner)
4415 {
4416         struct xfs_mount        *mp = bp->b_mount;
4417         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4418
4419         if (!xfs_sb_version_hascrc(&mp->m_sb))
4420                 return __this_address;
4421         if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
4422                 return __this_address;
4423         if (block->bb_u.l.bb_blkno != cpu_to_be64(bp->b_bn))
4424                 return __this_address;
4425         if (owner != XFS_RMAP_OWN_UNKNOWN &&
4426             be64_to_cpu(block->bb_u.l.bb_owner) != owner)
4427                 return __this_address;
4428         return NULL;
4429 }
4430
4431 /* Verify a long-format btree block. */
4432 xfs_failaddr_t
4433 xfs_btree_lblock_verify(
4434         struct xfs_buf          *bp,
4435         unsigned int            max_recs)
4436 {
4437         struct xfs_mount        *mp = bp->b_mount;
4438         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4439
4440         /* numrecs verification */
4441         if (be16_to_cpu(block->bb_numrecs) > max_recs)
4442                 return __this_address;
4443
4444         /* sibling pointer verification */
4445         if (block->bb_u.l.bb_leftsib != cpu_to_be64(NULLFSBLOCK) &&
4446             !xfs_verify_fsbno(mp, be64_to_cpu(block->bb_u.l.bb_leftsib)))
4447                 return __this_address;
4448         if (block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK) &&
4449             !xfs_verify_fsbno(mp, be64_to_cpu(block->bb_u.l.bb_rightsib)))
4450                 return __this_address;
4451
4452         return NULL;
4453 }
4454
4455 /**
4456  * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format
4457  *                                    btree block
4458  *
4459  * @bp: buffer containing the btree block
4460  */
4461 xfs_failaddr_t
4462 xfs_btree_sblock_v5hdr_verify(
4463         struct xfs_buf          *bp)
4464 {
4465         struct xfs_mount        *mp = bp->b_mount;
4466         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4467         struct xfs_perag        *pag = bp->b_pag;
4468
4469         if (!xfs_sb_version_hascrc(&mp->m_sb))
4470                 return __this_address;
4471         if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
4472                 return __this_address;
4473         if (block->bb_u.s.bb_blkno != cpu_to_be64(bp->b_bn))
4474                 return __this_address;
4475         if (pag && be32_to_cpu(block->bb_u.s.bb_owner) != pag->pag_agno)
4476                 return __this_address;
4477         return NULL;
4478 }
4479
4480 /**
4481  * xfs_btree_sblock_verify() -- verify a short-format btree block
4482  *
4483  * @bp: buffer containing the btree block
4484  * @max_recs: maximum records allowed in this btree node
4485  */
4486 xfs_failaddr_t
4487 xfs_btree_sblock_verify(
4488         struct xfs_buf          *bp,
4489         unsigned int            max_recs)
4490 {
4491         struct xfs_mount        *mp = bp->b_mount;
4492         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
4493         xfs_agblock_t           agno;
4494
4495         /* numrecs verification */
4496         if (be16_to_cpu(block->bb_numrecs) > max_recs)
4497                 return __this_address;
4498
4499         /* sibling pointer verification */
4500         agno = xfs_daddr_to_agno(mp, XFS_BUF_ADDR(bp));
4501         if (block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK) &&
4502             !xfs_verify_agbno(mp, agno, be32_to_cpu(block->bb_u.s.bb_leftsib)))
4503                 return __this_address;
4504         if (block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK) &&
4505             !xfs_verify_agbno(mp, agno, be32_to_cpu(block->bb_u.s.bb_rightsib)))
4506                 return __this_address;
4507
4508         return NULL;
4509 }
4510
4511 /*
4512  * Calculate the number of btree levels needed to store a given number of
4513  * records in a short-format btree.
4514  */
4515 uint
4516 xfs_btree_compute_maxlevels(
4517         uint                    *limits,
4518         unsigned long           len)
4519 {
4520         uint                    level;
4521         unsigned long           maxblocks;
4522
4523         maxblocks = (len + limits[0] - 1) / limits[0];
4524         for (level = 1; maxblocks > 1; level++)
4525                 maxblocks = (maxblocks + limits[1] - 1) / limits[1];
4526         return level;
4527 }
4528
4529 /*
4530  * Query a regular btree for all records overlapping a given interval.
4531  * Start with a LE lookup of the key of low_rec and return all records
4532  * until we find a record with a key greater than the key of high_rec.
4533  */
4534 STATIC int
4535 xfs_btree_simple_query_range(
4536         struct xfs_btree_cur            *cur,
4537         union xfs_btree_key             *low_key,
4538         union xfs_btree_key             *high_key,
4539         xfs_btree_query_range_fn        fn,
4540         void                            *priv)
4541 {
4542         union xfs_btree_rec             *recp;
4543         union xfs_btree_key             rec_key;
4544         int64_t                         diff;
4545         int                             stat;
4546         bool                            firstrec = true;
4547         int                             error;
4548
4549         ASSERT(cur->bc_ops->init_high_key_from_rec);
4550         ASSERT(cur->bc_ops->diff_two_keys);
4551
4552         /*
4553          * Find the leftmost record.  The btree cursor must be set
4554          * to the low record used to generate low_key.
4555          */
4556         stat = 0;
4557         error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
4558         if (error)
4559                 goto out;
4560
4561         /* Nothing?  See if there's anything to the right. */
4562         if (!stat) {
4563                 error = xfs_btree_increment(cur, 0, &stat);
4564                 if (error)
4565                         goto out;
4566         }
4567
4568         while (stat) {
4569                 /* Find the record. */
4570                 error = xfs_btree_get_rec(cur, &recp, &stat);
4571                 if (error || !stat)
4572                         break;
4573
4574                 /* Skip if high_key(rec) < low_key. */
4575                 if (firstrec) {
4576                         cur->bc_ops->init_high_key_from_rec(&rec_key, recp);
4577                         firstrec = false;
4578                         diff = cur->bc_ops->diff_two_keys(cur, low_key,
4579                                         &rec_key);
4580                         if (diff > 0)
4581                                 goto advloop;
4582                 }
4583
4584                 /* Stop if high_key < low_key(rec). */
4585                 cur->bc_ops->init_key_from_rec(&rec_key, recp);
4586                 diff = cur->bc_ops->diff_two_keys(cur, &rec_key, high_key);
4587                 if (diff > 0)
4588                         break;
4589
4590                 /* Callback */
4591                 error = fn(cur, recp, priv);
4592                 if (error)
4593                         break;
4594
4595 advloop:
4596                 /* Move on to the next record. */
4597                 error = xfs_btree_increment(cur, 0, &stat);
4598                 if (error)
4599                         break;
4600         }
4601
4602 out:
4603         return error;
4604 }
4605
4606 /*
4607  * Query an overlapped interval btree for all records overlapping a given
4608  * interval.  This function roughly follows the algorithm given in
4609  * "Interval Trees" of _Introduction to Algorithms_, which is section
4610  * 14.3 in the 2nd and 3rd editions.
4611  *
4612  * First, generate keys for the low and high records passed in.
4613  *
4614  * For any leaf node, generate the high and low keys for the record.
4615  * If the record keys overlap with the query low/high keys, pass the
4616  * record to the function iterator.
4617  *
4618  * For any internal node, compare the low and high keys of each
4619  * pointer against the query low/high keys.  If there's an overlap,
4620  * follow the pointer.
4621  *
4622  * As an optimization, we stop scanning a block when we find a low key
4623  * that is greater than the query's high key.
4624  */
4625 STATIC int
4626 xfs_btree_overlapped_query_range(
4627         struct xfs_btree_cur            *cur,
4628         union xfs_btree_key             *low_key,
4629         union xfs_btree_key             *high_key,
4630         xfs_btree_query_range_fn        fn,
4631         void                            *priv)
4632 {
4633         union xfs_btree_ptr             ptr;
4634         union xfs_btree_ptr             *pp;
4635         union xfs_btree_key             rec_key;
4636         union xfs_btree_key             rec_hkey;
4637         union xfs_btree_key             *lkp;
4638         union xfs_btree_key             *hkp;
4639         union xfs_btree_rec             *recp;
4640         struct xfs_btree_block          *block;
4641         int64_t                         ldiff;
4642         int64_t                         hdiff;
4643         int                             level;
4644         struct xfs_buf                  *bp;
4645         int                             i;
4646         int                             error;
4647
4648         /* Load the root of the btree. */
4649         level = cur->bc_nlevels - 1;
4650         cur->bc_ops->init_ptr_from_cur(cur, &ptr);
4651         error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
4652         if (error)
4653                 return error;
4654         xfs_btree_get_block(cur, level, &bp);
4655         trace_xfs_btree_overlapped_query_range(cur, level, bp);
4656 #ifdef DEBUG
4657         error = xfs_btree_check_block(cur, block, level, bp);
4658         if (error)
4659                 goto out;
4660 #endif
4661         cur->bc_ptrs[level] = 1;
4662
4663         while (level < cur->bc_nlevels) {
4664                 block = xfs_btree_get_block(cur, level, &bp);
4665
4666                 /* End of node, pop back towards the root. */
4667                 if (cur->bc_ptrs[level] > be16_to_cpu(block->bb_numrecs)) {
4668 pop_up:
4669                         if (level < cur->bc_nlevels - 1)
4670                                 cur->bc_ptrs[level + 1]++;
4671                         level++;
4672                         continue;
4673                 }
4674
4675                 if (level == 0) {
4676                         /* Handle a leaf node. */
4677                         recp = xfs_btree_rec_addr(cur, cur->bc_ptrs[0], block);
4678
4679                         cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
4680                         ldiff = cur->bc_ops->diff_two_keys(cur, &rec_hkey,
4681                                         low_key);
4682
4683                         cur->bc_ops->init_key_from_rec(&rec_key, recp);
4684                         hdiff = cur->bc_ops->diff_two_keys(cur, high_key,
4685                                         &rec_key);
4686
4687                         /*
4688                          * If (record's high key >= query's low key) and
4689                          *    (query's high key >= record's low key), then
4690                          * this record overlaps the query range; callback.
4691                          */
4692                         if (ldiff >= 0 && hdiff >= 0) {
4693                                 error = fn(cur, recp, priv);
4694                                 if (error)
4695                                         break;
4696                         } else if (hdiff < 0) {
4697                                 /* Record is larger than high key; pop. */
4698                                 goto pop_up;
4699                         }
4700                         cur->bc_ptrs[level]++;
4701                         continue;
4702                 }
4703
4704                 /* Handle an internal node. */
4705                 lkp = xfs_btree_key_addr(cur, cur->bc_ptrs[level], block);
4706                 hkp = xfs_btree_high_key_addr(cur, cur->bc_ptrs[level], block);
4707                 pp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[level], block);
4708
4709                 ldiff = cur->bc_ops->diff_two_keys(cur, hkp, low_key);
4710                 hdiff = cur->bc_ops->diff_two_keys(cur, high_key, lkp);
4711
4712                 /*
4713                  * If (pointer's high key >= query's low key) and
4714                  *    (query's high key >= pointer's low key), then
4715                  * this record overlaps the query range; follow pointer.
4716                  */
4717                 if (ldiff >= 0 && hdiff >= 0) {
4718                         level--;
4719                         error = xfs_btree_lookup_get_block(cur, level, pp,
4720                                         &block);
4721                         if (error)
4722                                 goto out;
4723                         xfs_btree_get_block(cur, level, &bp);
4724                         trace_xfs_btree_overlapped_query_range(cur, level, bp);
4725 #ifdef DEBUG
4726                         error = xfs_btree_check_block(cur, block, level, bp);
4727                         if (error)
4728                                 goto out;
4729 #endif
4730                         cur->bc_ptrs[level] = 1;
4731                         continue;
4732                 } else if (hdiff < 0) {
4733                         /* The low key is larger than the upper range; pop. */
4734                         goto pop_up;
4735                 }
4736                 cur->bc_ptrs[level]++;
4737         }
4738
4739 out:
4740         /*
4741          * If we don't end this function with the cursor pointing at a record
4742          * block, a subsequent non-error cursor deletion will not release
4743          * node-level buffers, causing a buffer leak.  This is quite possible
4744          * with a zero-results range query, so release the buffers if we
4745          * failed to return any results.
4746          */
4747         if (cur->bc_bufs[0] == NULL) {
4748                 for (i = 0; i < cur->bc_nlevels; i++) {
4749                         if (cur->bc_bufs[i]) {
4750                                 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
4751                                 cur->bc_bufs[i] = NULL;
4752                                 cur->bc_ptrs[i] = 0;
4753                                 cur->bc_ra[i] = 0;
4754                         }
4755                 }
4756         }
4757
4758         return error;
4759 }
4760
4761 /*
4762  * Query a btree for all records overlapping a given interval of keys.  The
4763  * supplied function will be called with each record found; return one of the
4764  * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
4765  * code.  This function returns -ECANCELED, zero, or a negative error code.
4766  */
4767 int
4768 xfs_btree_query_range(
4769         struct xfs_btree_cur            *cur,
4770         union xfs_btree_irec            *low_rec,
4771         union xfs_btree_irec            *high_rec,
4772         xfs_btree_query_range_fn        fn,
4773         void                            *priv)
4774 {
4775         union xfs_btree_rec             rec;
4776         union xfs_btree_key             low_key;
4777         union xfs_btree_key             high_key;
4778
4779         /* Find the keys of both ends of the interval. */
4780         cur->bc_rec = *high_rec;
4781         cur->bc_ops->init_rec_from_cur(cur, &rec);
4782         cur->bc_ops->init_key_from_rec(&high_key, &rec);
4783
4784         cur->bc_rec = *low_rec;
4785         cur->bc_ops->init_rec_from_cur(cur, &rec);
4786         cur->bc_ops->init_key_from_rec(&low_key, &rec);
4787
4788         /* Enforce low key < high key. */
4789         if (cur->bc_ops->diff_two_keys(cur, &low_key, &high_key) > 0)
4790                 return -EINVAL;
4791
4792         if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING))
4793                 return xfs_btree_simple_query_range(cur, &low_key,
4794                                 &high_key, fn, priv);
4795         return xfs_btree_overlapped_query_range(cur, &low_key, &high_key,
4796                         fn, priv);
4797 }
4798
4799 /* Query a btree for all records. */
4800 int
4801 xfs_btree_query_all(
4802         struct xfs_btree_cur            *cur,
4803         xfs_btree_query_range_fn        fn,
4804         void                            *priv)
4805 {
4806         union xfs_btree_key             low_key;
4807         union xfs_btree_key             high_key;
4808
4809         memset(&cur->bc_rec, 0, sizeof(cur->bc_rec));
4810         memset(&low_key, 0, sizeof(low_key));
4811         memset(&high_key, 0xFF, sizeof(high_key));
4812
4813         return xfs_btree_simple_query_range(cur, &low_key, &high_key, fn, priv);
4814 }
4815
4816 /*
4817  * Calculate the number of blocks needed to store a given number of records
4818  * in a short-format (per-AG metadata) btree.
4819  */
4820 unsigned long long
4821 xfs_btree_calc_size(
4822         uint                    *limits,
4823         unsigned long long      len)
4824 {
4825         int                     level;
4826         int                     maxrecs;
4827         unsigned long long      rval;
4828
4829         maxrecs = limits[0];
4830         for (level = 0, rval = 0; len > 1; level++) {
4831                 len += maxrecs - 1;
4832                 do_div(len, maxrecs);
4833                 maxrecs = limits[1];
4834                 rval += len;
4835         }
4836         return rval;
4837 }
4838
4839 static int
4840 xfs_btree_count_blocks_helper(
4841         struct xfs_btree_cur    *cur,
4842         int                     level,
4843         void                    *data)
4844 {
4845         xfs_extlen_t            *blocks = data;
4846         (*blocks)++;
4847
4848         return 0;
4849 }
4850
4851 /* Count the blocks in a btree and return the result in *blocks. */
4852 int
4853 xfs_btree_count_blocks(
4854         struct xfs_btree_cur    *cur,
4855         xfs_extlen_t            *blocks)
4856 {
4857         *blocks = 0;
4858         return xfs_btree_visit_blocks(cur, xfs_btree_count_blocks_helper,
4859                         XFS_BTREE_VISIT_ALL, blocks);
4860 }
4861
4862 /* Compare two btree pointers. */
4863 int64_t
4864 xfs_btree_diff_two_ptrs(
4865         struct xfs_btree_cur            *cur,
4866         const union xfs_btree_ptr       *a,
4867         const union xfs_btree_ptr       *b)
4868 {
4869         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
4870                 return (int64_t)be64_to_cpu(a->l) - be64_to_cpu(b->l);
4871         return (int64_t)be32_to_cpu(a->s) - be32_to_cpu(b->s);
4872 }
4873
4874 /* If there's an extent, we're done. */
4875 STATIC int
4876 xfs_btree_has_record_helper(
4877         struct xfs_btree_cur            *cur,
4878         union xfs_btree_rec             *rec,
4879         void                            *priv)
4880 {
4881         return -ECANCELED;
4882 }
4883
4884 /* Is there a record covering a given range of keys? */
4885 int
4886 xfs_btree_has_record(
4887         struct xfs_btree_cur    *cur,
4888         union xfs_btree_irec    *low,
4889         union xfs_btree_irec    *high,
4890         bool                    *exists)
4891 {
4892         int                     error;
4893
4894         error = xfs_btree_query_range(cur, low, high,
4895                         &xfs_btree_has_record_helper, NULL);
4896         if (error == -ECANCELED) {
4897                 *exists = true;
4898                 return 0;
4899         }
4900         *exists = false;
4901         return error;
4902 }
4903
4904 /* Are there more records in this btree? */
4905 bool
4906 xfs_btree_has_more_records(
4907         struct xfs_btree_cur    *cur)
4908 {
4909         struct xfs_btree_block  *block;
4910         struct xfs_buf          *bp;
4911
4912         block = xfs_btree_get_block(cur, 0, &bp);
4913
4914         /* There are still records in this block. */
4915         if (cur->bc_ptrs[0] < xfs_btree_get_numrecs(block))
4916                 return true;
4917
4918         /* There are more record blocks. */
4919         if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
4920                 return block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK);
4921         else
4922                 return block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK);
4923 }