2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
23 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_alloc_btree.h"
29 #include "xfs_ialloc_btree.h"
30 #include "xfs_dinode.h"
31 #include "xfs_inode.h"
32 #include "xfs_inode_item.h"
33 #include "xfs_btree.h"
34 #include "xfs_error.h"
35 #include "xfs_trace.h"
38 * Cursor allocation zone.
40 kmem_zone_t *xfs_btree_cur_zone;
43 * Btree magic numbers.
45 const __uint32_t xfs_magics[XFS_BTNUM_MAX] = {
46 XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, XFS_BMAP_MAGIC, XFS_IBT_MAGIC
50 STATIC int /* error (0 or EFSCORRUPTED) */
51 xfs_btree_check_lblock(
52 struct xfs_btree_cur *cur, /* btree cursor */
53 struct xfs_btree_block *block, /* btree long form block pointer */
54 int level, /* level of the btree block */
55 struct xfs_buf *bp) /* buffer for block, if any */
57 int lblock_ok; /* block passes checks */
58 struct xfs_mount *mp; /* file system mount point */
62 be32_to_cpu(block->bb_magic) == xfs_magics[cur->bc_btnum] &&
63 be16_to_cpu(block->bb_level) == level &&
64 be16_to_cpu(block->bb_numrecs) <=
65 cur->bc_ops->get_maxrecs(cur, level) &&
66 block->bb_u.l.bb_leftsib &&
67 (block->bb_u.l.bb_leftsib == cpu_to_be64(NULLDFSBNO) ||
68 XFS_FSB_SANITY_CHECK(mp,
69 be64_to_cpu(block->bb_u.l.bb_leftsib))) &&
70 block->bb_u.l.bb_rightsib &&
71 (block->bb_u.l.bb_rightsib == cpu_to_be64(NULLDFSBNO) ||
72 XFS_FSB_SANITY_CHECK(mp,
73 be64_to_cpu(block->bb_u.l.bb_rightsib)));
74 if (unlikely(XFS_TEST_ERROR(!lblock_ok, mp,
75 XFS_ERRTAG_BTREE_CHECK_LBLOCK,
76 XFS_RANDOM_BTREE_CHECK_LBLOCK))) {
78 trace_xfs_btree_corrupt(bp, _RET_IP_);
79 XFS_ERROR_REPORT("xfs_btree_check_lblock", XFS_ERRLEVEL_LOW,
81 return XFS_ERROR(EFSCORRUPTED);
86 STATIC int /* error (0 or EFSCORRUPTED) */
87 xfs_btree_check_sblock(
88 struct xfs_btree_cur *cur, /* btree cursor */
89 struct xfs_btree_block *block, /* btree short form block pointer */
90 int level, /* level of the btree block */
91 struct xfs_buf *bp) /* buffer containing block */
93 struct xfs_buf *agbp; /* buffer for ag. freespace struct */
94 struct xfs_agf *agf; /* ag. freespace structure */
95 xfs_agblock_t agflen; /* native ag. freespace length */
96 int sblock_ok; /* block passes checks */
98 agbp = cur->bc_private.a.agbp;
99 agf = XFS_BUF_TO_AGF(agbp);
100 agflen = be32_to_cpu(agf->agf_length);
102 be32_to_cpu(block->bb_magic) == xfs_magics[cur->bc_btnum] &&
103 be16_to_cpu(block->bb_level) == level &&
104 be16_to_cpu(block->bb_numrecs) <=
105 cur->bc_ops->get_maxrecs(cur, level) &&
106 (block->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) ||
107 be32_to_cpu(block->bb_u.s.bb_leftsib) < agflen) &&
108 block->bb_u.s.bb_leftsib &&
109 (block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK) ||
110 be32_to_cpu(block->bb_u.s.bb_rightsib) < agflen) &&
111 block->bb_u.s.bb_rightsib;
112 if (unlikely(XFS_TEST_ERROR(!sblock_ok, cur->bc_mp,
113 XFS_ERRTAG_BTREE_CHECK_SBLOCK,
114 XFS_RANDOM_BTREE_CHECK_SBLOCK))) {
116 trace_xfs_btree_corrupt(bp, _RET_IP_);
117 XFS_CORRUPTION_ERROR("xfs_btree_check_sblock",
118 XFS_ERRLEVEL_LOW, cur->bc_mp, block);
119 return XFS_ERROR(EFSCORRUPTED);
125 * Debug routine: check that block header is ok.
128 xfs_btree_check_block(
129 struct xfs_btree_cur *cur, /* btree cursor */
130 struct xfs_btree_block *block, /* generic btree block pointer */
131 int level, /* level of the btree block */
132 struct xfs_buf *bp) /* buffer containing block, if any */
134 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
135 return xfs_btree_check_lblock(cur, block, level, bp);
137 return xfs_btree_check_sblock(cur, block, level, bp);
141 * Check that (long) pointer is ok.
143 int /* error (0 or EFSCORRUPTED) */
144 xfs_btree_check_lptr(
145 struct xfs_btree_cur *cur, /* btree cursor */
146 xfs_dfsbno_t bno, /* btree block disk address */
147 int level) /* btree block level */
149 XFS_WANT_CORRUPTED_RETURN(
152 XFS_FSB_SANITY_CHECK(cur->bc_mp, bno));
158 * Check that (short) pointer is ok.
160 STATIC int /* error (0 or EFSCORRUPTED) */
161 xfs_btree_check_sptr(
162 struct xfs_btree_cur *cur, /* btree cursor */
163 xfs_agblock_t bno, /* btree block disk address */
164 int level) /* btree block level */
166 xfs_agblock_t agblocks = cur->bc_mp->m_sb.sb_agblocks;
168 XFS_WANT_CORRUPTED_RETURN(
170 bno != NULLAGBLOCK &&
177 * Check that block ptr is ok.
179 STATIC int /* error (0 or EFSCORRUPTED) */
181 struct xfs_btree_cur *cur, /* btree cursor */
182 union xfs_btree_ptr *ptr, /* btree block disk address */
183 int index, /* offset from ptr to check */
184 int level) /* btree block level */
186 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
187 return xfs_btree_check_lptr(cur,
188 be64_to_cpu((&ptr->l)[index]), level);
190 return xfs_btree_check_sptr(cur,
191 be32_to_cpu((&ptr->s)[index]), level);
197 * Delete the btree cursor.
200 xfs_btree_del_cursor(
201 xfs_btree_cur_t *cur, /* btree cursor */
202 int error) /* del because of error */
204 int i; /* btree level */
207 * Clear the buffer pointers, and release the buffers.
208 * If we're doing this in the face of an error, we
209 * need to make sure to inspect all of the entries
210 * in the bc_bufs array for buffers to be unlocked.
211 * This is because some of the btree code works from
212 * level n down to 0, and if we get an error along
213 * the way we won't have initialized all the entries
216 for (i = 0; i < cur->bc_nlevels; i++) {
218 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
223 * Can't free a bmap cursor without having dealt with the
224 * allocated indirect blocks' accounting.
226 ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP ||
227 cur->bc_private.b.allocated == 0);
231 kmem_zone_free(xfs_btree_cur_zone, cur);
235 * Duplicate the btree cursor.
236 * Allocate a new one, copy the record, re-get the buffers.
239 xfs_btree_dup_cursor(
240 xfs_btree_cur_t *cur, /* input cursor */
241 xfs_btree_cur_t **ncur) /* output cursor */
243 xfs_buf_t *bp; /* btree block's buffer pointer */
244 int error; /* error return value */
245 int i; /* level number of btree block */
246 xfs_mount_t *mp; /* mount structure for filesystem */
247 xfs_btree_cur_t *new; /* new cursor value */
248 xfs_trans_t *tp; /* transaction pointer, can be NULL */
254 * Allocate a new cursor like the old one.
256 new = cur->bc_ops->dup_cursor(cur);
259 * Copy the record currently in the cursor.
261 new->bc_rec = cur->bc_rec;
264 * For each level current, re-get the buffer and copy the ptr value.
266 for (i = 0; i < new->bc_nlevels; i++) {
267 new->bc_ptrs[i] = cur->bc_ptrs[i];
268 new->bc_ra[i] = cur->bc_ra[i];
269 if ((bp = cur->bc_bufs[i])) {
270 if ((error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
271 XFS_BUF_ADDR(bp), mp->m_bsize, 0, &bp))) {
272 xfs_btree_del_cursor(new, error);
276 new->bc_bufs[i] = bp;
277 ASSERT(!xfs_buf_geterror(bp));
279 new->bc_bufs[i] = NULL;
286 * XFS btree block layout and addressing:
288 * There are two types of blocks in the btree: leaf and non-leaf blocks.
290 * The leaf record start with a header then followed by records containing
291 * the values. A non-leaf block also starts with the same header, and
292 * then first contains lookup keys followed by an equal number of pointers
293 * to the btree blocks at the previous level.
295 * +--------+-------+-------+-------+-------+-------+-------+
296 * Leaf: | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
297 * +--------+-------+-------+-------+-------+-------+-------+
299 * +--------+-------+-------+-------+-------+-------+-------+
300 * Non-Leaf: | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
301 * +--------+-------+-------+-------+-------+-------+-------+
303 * The header is called struct xfs_btree_block for reasons better left unknown
304 * and comes in different versions for short (32bit) and long (64bit) block
305 * pointers. The record and key structures are defined by the btree instances
306 * and opaque to the btree core. The block pointers are simple disk endian
307 * integers, available in a short (32bit) and long (64bit) variant.
309 * The helpers below calculate the offset of a given record, key or pointer
310 * into a btree block (xfs_btree_*_offset) or return a pointer to the given
311 * record, key or pointer (xfs_btree_*_addr). Note that all addressing
312 * inside the btree block is done using indices starting at one, not zero!
316 * Return size of the btree block header for this btree instance.
318 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
320 return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
321 XFS_BTREE_LBLOCK_LEN :
322 XFS_BTREE_SBLOCK_LEN;
326 * Return size of btree block pointers for this btree instance.
328 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
330 return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
331 sizeof(__be64) : sizeof(__be32);
335 * Calculate offset of the n-th record in a btree block.
338 xfs_btree_rec_offset(
339 struct xfs_btree_cur *cur,
342 return xfs_btree_block_len(cur) +
343 (n - 1) * cur->bc_ops->rec_len;
347 * Calculate offset of the n-th key in a btree block.
350 xfs_btree_key_offset(
351 struct xfs_btree_cur *cur,
354 return xfs_btree_block_len(cur) +
355 (n - 1) * cur->bc_ops->key_len;
359 * Calculate offset of the n-th block pointer in a btree block.
362 xfs_btree_ptr_offset(
363 struct xfs_btree_cur *cur,
367 return xfs_btree_block_len(cur) +
368 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
369 (n - 1) * xfs_btree_ptr_len(cur);
373 * Return a pointer to the n-th record in the btree block.
375 STATIC union xfs_btree_rec *
377 struct xfs_btree_cur *cur,
379 struct xfs_btree_block *block)
381 return (union xfs_btree_rec *)
382 ((char *)block + xfs_btree_rec_offset(cur, n));
386 * Return a pointer to the n-th key in the btree block.
388 STATIC union xfs_btree_key *
390 struct xfs_btree_cur *cur,
392 struct xfs_btree_block *block)
394 return (union xfs_btree_key *)
395 ((char *)block + xfs_btree_key_offset(cur, n));
399 * Return a pointer to the n-th block pointer in the btree block.
401 STATIC union xfs_btree_ptr *
403 struct xfs_btree_cur *cur,
405 struct xfs_btree_block *block)
407 int level = xfs_btree_get_level(block);
409 ASSERT(block->bb_level != 0);
411 return (union xfs_btree_ptr *)
412 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
416 * Get a the root block which is stored in the inode.
418 * For now this btree implementation assumes the btree root is always
419 * stored in the if_broot field of an inode fork.
421 STATIC struct xfs_btree_block *
423 struct xfs_btree_cur *cur)
425 struct xfs_ifork *ifp;
427 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip, cur->bc_private.b.whichfork);
428 return (struct xfs_btree_block *)ifp->if_broot;
432 * Retrieve the block pointer from the cursor at the given level.
433 * This may be an inode btree root or from a buffer.
435 STATIC struct xfs_btree_block * /* generic btree block pointer */
437 struct xfs_btree_cur *cur, /* btree cursor */
438 int level, /* level in btree */
439 struct xfs_buf **bpp) /* buffer containing the block */
441 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
442 (level == cur->bc_nlevels - 1)) {
444 return xfs_btree_get_iroot(cur);
447 *bpp = cur->bc_bufs[level];
448 return XFS_BUF_TO_BLOCK(*bpp);
452 * Get a buffer for the block, return it with no data read.
453 * Long-form addressing.
455 xfs_buf_t * /* buffer for fsbno */
457 xfs_mount_t *mp, /* file system mount point */
458 xfs_trans_t *tp, /* transaction pointer */
459 xfs_fsblock_t fsbno, /* file system block number */
460 uint lock) /* lock flags for get_buf */
462 xfs_buf_t *bp; /* buffer pointer (return value) */
463 xfs_daddr_t d; /* real disk block address */
465 ASSERT(fsbno != NULLFSBLOCK);
466 d = XFS_FSB_TO_DADDR(mp, fsbno);
467 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
468 ASSERT(!xfs_buf_geterror(bp));
473 * Get a buffer for the block, return it with no data read.
474 * Short-form addressing.
476 xfs_buf_t * /* buffer for agno/agbno */
478 xfs_mount_t *mp, /* file system mount point */
479 xfs_trans_t *tp, /* transaction pointer */
480 xfs_agnumber_t agno, /* allocation group number */
481 xfs_agblock_t agbno, /* allocation group block number */
482 uint lock) /* lock flags for get_buf */
484 xfs_buf_t *bp; /* buffer pointer (return value) */
485 xfs_daddr_t d; /* real disk block address */
487 ASSERT(agno != NULLAGNUMBER);
488 ASSERT(agbno != NULLAGBLOCK);
489 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
490 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
491 ASSERT(!xfs_buf_geterror(bp));
496 * Check for the cursor referring to the last block at the given level.
498 int /* 1=is last block, 0=not last block */
499 xfs_btree_islastblock(
500 xfs_btree_cur_t *cur, /* btree cursor */
501 int level) /* level to check */
503 struct xfs_btree_block *block; /* generic btree block pointer */
504 xfs_buf_t *bp; /* buffer containing block */
506 block = xfs_btree_get_block(cur, level, &bp);
507 xfs_btree_check_block(cur, block, level, bp);
508 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
509 return block->bb_u.l.bb_rightsib == cpu_to_be64(NULLDFSBNO);
511 return block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK);
515 * Change the cursor to point to the first record at the given level.
516 * Other levels are unaffected.
518 STATIC int /* success=1, failure=0 */
520 xfs_btree_cur_t *cur, /* btree cursor */
521 int level) /* level to change */
523 struct xfs_btree_block *block; /* generic btree block pointer */
524 xfs_buf_t *bp; /* buffer containing block */
527 * Get the block pointer for this level.
529 block = xfs_btree_get_block(cur, level, &bp);
530 xfs_btree_check_block(cur, block, level, bp);
532 * It's empty, there is no such record.
534 if (!block->bb_numrecs)
537 * Set the ptr value to 1, that's the first record/key.
539 cur->bc_ptrs[level] = 1;
544 * Change the cursor to point to the last record in the current block
545 * at the given level. Other levels are unaffected.
547 STATIC int /* success=1, failure=0 */
549 xfs_btree_cur_t *cur, /* btree cursor */
550 int level) /* level to change */
552 struct xfs_btree_block *block; /* generic btree block pointer */
553 xfs_buf_t *bp; /* buffer containing block */
556 * Get the block pointer for this level.
558 block = xfs_btree_get_block(cur, level, &bp);
559 xfs_btree_check_block(cur, block, level, bp);
561 * It's empty, there is no such record.
563 if (!block->bb_numrecs)
566 * Set the ptr value to numrecs, that's the last record/key.
568 cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
573 * Compute first and last byte offsets for the fields given.
574 * Interprets the offsets table, which contains struct field offsets.
578 __int64_t fields, /* bitmask of fields */
579 const short *offsets, /* table of field offsets */
580 int nbits, /* number of bits to inspect */
581 int *first, /* output: first byte offset */
582 int *last) /* output: last byte offset */
584 int i; /* current bit number */
585 __int64_t imask; /* mask for current bit number */
589 * Find the lowest bit, so the first byte offset.
591 for (i = 0, imask = 1LL; ; i++, imask <<= 1) {
592 if (imask & fields) {
598 * Find the highest bit, so the last byte offset.
600 for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) {
601 if (imask & fields) {
602 *last = offsets[i + 1] - 1;
609 * Get a buffer for the block, return it read in.
610 * Long-form addressing.
614 xfs_mount_t *mp, /* file system mount point */
615 xfs_trans_t *tp, /* transaction pointer */
616 xfs_fsblock_t fsbno, /* file system block number */
617 uint lock, /* lock flags for read_buf */
618 xfs_buf_t **bpp, /* buffer for fsbno */
619 int refval) /* ref count value for buffer */
621 xfs_buf_t *bp; /* return value */
622 xfs_daddr_t d; /* real disk block address */
625 ASSERT(fsbno != NULLFSBLOCK);
626 d = XFS_FSB_TO_DADDR(mp, fsbno);
627 if ((error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
628 mp->m_bsize, lock, &bp))) {
631 ASSERT(!xfs_buf_geterror(bp));
633 xfs_buf_set_ref(bp, refval);
639 * Read-ahead the block, don't wait for it, don't return a buffer.
640 * Long-form addressing.
644 xfs_btree_reada_bufl(
645 xfs_mount_t *mp, /* file system mount point */
646 xfs_fsblock_t fsbno, /* file system block number */
647 xfs_extlen_t count) /* count of filesystem blocks */
651 ASSERT(fsbno != NULLFSBLOCK);
652 d = XFS_FSB_TO_DADDR(mp, fsbno);
653 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count);
657 * Read-ahead the block, don't wait for it, don't return a buffer.
658 * Short-form addressing.
662 xfs_btree_reada_bufs(
663 xfs_mount_t *mp, /* file system mount point */
664 xfs_agnumber_t agno, /* allocation group number */
665 xfs_agblock_t agbno, /* allocation group block number */
666 xfs_extlen_t count) /* count of filesystem blocks */
670 ASSERT(agno != NULLAGNUMBER);
671 ASSERT(agbno != NULLAGBLOCK);
672 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
673 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count);
677 xfs_btree_readahead_lblock(
678 struct xfs_btree_cur *cur,
680 struct xfs_btree_block *block)
683 xfs_dfsbno_t left = be64_to_cpu(block->bb_u.l.bb_leftsib);
684 xfs_dfsbno_t right = be64_to_cpu(block->bb_u.l.bb_rightsib);
686 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLDFSBNO) {
687 xfs_btree_reada_bufl(cur->bc_mp, left, 1);
691 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLDFSBNO) {
692 xfs_btree_reada_bufl(cur->bc_mp, right, 1);
700 xfs_btree_readahead_sblock(
701 struct xfs_btree_cur *cur,
703 struct xfs_btree_block *block)
706 xfs_agblock_t left = be32_to_cpu(block->bb_u.s.bb_leftsib);
707 xfs_agblock_t right = be32_to_cpu(block->bb_u.s.bb_rightsib);
710 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
711 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
716 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
717 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
726 * Read-ahead btree blocks, at the given level.
727 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
731 struct xfs_btree_cur *cur, /* btree cursor */
732 int lev, /* level in btree */
733 int lr) /* left/right bits */
735 struct xfs_btree_block *block;
738 * No readahead needed if we are at the root level and the
739 * btree root is stored in the inode.
741 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
742 (lev == cur->bc_nlevels - 1))
745 if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
748 cur->bc_ra[lev] |= lr;
749 block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
751 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
752 return xfs_btree_readahead_lblock(cur, lr, block);
753 return xfs_btree_readahead_sblock(cur, lr, block);
757 * Set the buffer for level "lev" in the cursor to bp, releasing
758 * any previous buffer.
762 xfs_btree_cur_t *cur, /* btree cursor */
763 int lev, /* level in btree */
764 xfs_buf_t *bp) /* new buffer to set */
766 struct xfs_btree_block *b; /* btree block */
768 if (cur->bc_bufs[lev])
769 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
770 cur->bc_bufs[lev] = bp;
773 b = XFS_BUF_TO_BLOCK(bp);
774 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
775 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLDFSBNO))
776 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
777 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLDFSBNO))
778 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
780 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
781 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
782 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
783 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
788 xfs_btree_ptr_is_null(
789 struct xfs_btree_cur *cur,
790 union xfs_btree_ptr *ptr)
792 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
793 return ptr->l == cpu_to_be64(NULLDFSBNO);
795 return ptr->s == cpu_to_be32(NULLAGBLOCK);
799 xfs_btree_set_ptr_null(
800 struct xfs_btree_cur *cur,
801 union xfs_btree_ptr *ptr)
803 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
804 ptr->l = cpu_to_be64(NULLDFSBNO);
806 ptr->s = cpu_to_be32(NULLAGBLOCK);
810 * Get/set/init sibling pointers
813 xfs_btree_get_sibling(
814 struct xfs_btree_cur *cur,
815 struct xfs_btree_block *block,
816 union xfs_btree_ptr *ptr,
819 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
821 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
822 if (lr == XFS_BB_RIGHTSIB)
823 ptr->l = block->bb_u.l.bb_rightsib;
825 ptr->l = block->bb_u.l.bb_leftsib;
827 if (lr == XFS_BB_RIGHTSIB)
828 ptr->s = block->bb_u.s.bb_rightsib;
830 ptr->s = block->bb_u.s.bb_leftsib;
835 xfs_btree_set_sibling(
836 struct xfs_btree_cur *cur,
837 struct xfs_btree_block *block,
838 union xfs_btree_ptr *ptr,
841 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
843 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
844 if (lr == XFS_BB_RIGHTSIB)
845 block->bb_u.l.bb_rightsib = ptr->l;
847 block->bb_u.l.bb_leftsib = ptr->l;
849 if (lr == XFS_BB_RIGHTSIB)
850 block->bb_u.s.bb_rightsib = ptr->s;
852 block->bb_u.s.bb_leftsib = ptr->s;
857 xfs_btree_init_block(
858 struct xfs_btree_cur *cur,
861 struct xfs_btree_block *new) /* new block */
863 new->bb_magic = cpu_to_be32(xfs_magics[cur->bc_btnum]);
864 new->bb_level = cpu_to_be16(level);
865 new->bb_numrecs = cpu_to_be16(numrecs);
867 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
868 new->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
869 new->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
871 new->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
872 new->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
877 * Return true if ptr is the last record in the btree and
878 * we need to track updateѕ to this record. The decision
879 * will be further refined in the update_lastrec method.
882 xfs_btree_is_lastrec(
883 struct xfs_btree_cur *cur,
884 struct xfs_btree_block *block,
887 union xfs_btree_ptr ptr;
891 if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
894 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
895 if (!xfs_btree_ptr_is_null(cur, &ptr))
901 xfs_btree_buf_to_ptr(
902 struct xfs_btree_cur *cur,
904 union xfs_btree_ptr *ptr)
906 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
907 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
910 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
916 xfs_btree_ptr_to_daddr(
917 struct xfs_btree_cur *cur,
918 union xfs_btree_ptr *ptr)
920 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
921 ASSERT(ptr->l != cpu_to_be64(NULLDFSBNO));
923 return XFS_FSB_TO_DADDR(cur->bc_mp, be64_to_cpu(ptr->l));
925 ASSERT(cur->bc_private.a.agno != NULLAGNUMBER);
926 ASSERT(ptr->s != cpu_to_be32(NULLAGBLOCK));
928 return XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_private.a.agno,
929 be32_to_cpu(ptr->s));
935 struct xfs_btree_cur *cur,
938 switch (cur->bc_btnum) {
941 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
944 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
947 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
955 xfs_btree_get_buf_block(
956 struct xfs_btree_cur *cur,
957 union xfs_btree_ptr *ptr,
959 struct xfs_btree_block **block,
960 struct xfs_buf **bpp)
962 struct xfs_mount *mp = cur->bc_mp;
965 /* need to sort out how callers deal with failures first */
966 ASSERT(!(flags & XBF_TRYLOCK));
968 d = xfs_btree_ptr_to_daddr(cur, ptr);
969 *bpp = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d,
975 *block = XFS_BUF_TO_BLOCK(*bpp);
980 * Read in the buffer at the given ptr and return the buffer and
981 * the block pointer within the buffer.
984 xfs_btree_read_buf_block(
985 struct xfs_btree_cur *cur,
986 union xfs_btree_ptr *ptr,
989 struct xfs_btree_block **block,
990 struct xfs_buf **bpp)
992 struct xfs_mount *mp = cur->bc_mp;
996 /* need to sort out how callers deal with failures first */
997 ASSERT(!(flags & XBF_TRYLOCK));
999 d = xfs_btree_ptr_to_daddr(cur, ptr);
1000 error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
1001 mp->m_bsize, flags, bpp);
1005 ASSERT(!xfs_buf_geterror(*bpp));
1007 xfs_btree_set_refs(cur, *bpp);
1008 *block = XFS_BUF_TO_BLOCK(*bpp);
1010 error = xfs_btree_check_block(cur, *block, level, *bpp);
1012 xfs_trans_brelse(cur->bc_tp, *bpp);
1017 * Copy keys from one btree block to another.
1020 xfs_btree_copy_keys(
1021 struct xfs_btree_cur *cur,
1022 union xfs_btree_key *dst_key,
1023 union xfs_btree_key *src_key,
1026 ASSERT(numkeys >= 0);
1027 memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1031 * Copy records from one btree block to another.
1034 xfs_btree_copy_recs(
1035 struct xfs_btree_cur *cur,
1036 union xfs_btree_rec *dst_rec,
1037 union xfs_btree_rec *src_rec,
1040 ASSERT(numrecs >= 0);
1041 memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1045 * Copy block pointers from one btree block to another.
1048 xfs_btree_copy_ptrs(
1049 struct xfs_btree_cur *cur,
1050 union xfs_btree_ptr *dst_ptr,
1051 union xfs_btree_ptr *src_ptr,
1054 ASSERT(numptrs >= 0);
1055 memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1059 * Shift keys one index left/right inside a single btree block.
1062 xfs_btree_shift_keys(
1063 struct xfs_btree_cur *cur,
1064 union xfs_btree_key *key,
1070 ASSERT(numkeys >= 0);
1071 ASSERT(dir == 1 || dir == -1);
1073 dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1074 memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1078 * Shift records one index left/right inside a single btree block.
1081 xfs_btree_shift_recs(
1082 struct xfs_btree_cur *cur,
1083 union xfs_btree_rec *rec,
1089 ASSERT(numrecs >= 0);
1090 ASSERT(dir == 1 || dir == -1);
1092 dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1093 memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1097 * Shift block pointers one index left/right inside a single btree block.
1100 xfs_btree_shift_ptrs(
1101 struct xfs_btree_cur *cur,
1102 union xfs_btree_ptr *ptr,
1108 ASSERT(numptrs >= 0);
1109 ASSERT(dir == 1 || dir == -1);
1111 dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1112 memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1116 * Log key values from the btree block.
1120 struct xfs_btree_cur *cur,
1125 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1126 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1129 xfs_trans_log_buf(cur->bc_tp, bp,
1130 xfs_btree_key_offset(cur, first),
1131 xfs_btree_key_offset(cur, last + 1) - 1);
1133 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1134 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1137 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1141 * Log record values from the btree block.
1145 struct xfs_btree_cur *cur,
1150 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1151 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1153 xfs_trans_log_buf(cur->bc_tp, bp,
1154 xfs_btree_rec_offset(cur, first),
1155 xfs_btree_rec_offset(cur, last + 1) - 1);
1157 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1161 * Log block pointer fields from a btree block (nonleaf).
1165 struct xfs_btree_cur *cur, /* btree cursor */
1166 struct xfs_buf *bp, /* buffer containing btree block */
1167 int first, /* index of first pointer to log */
1168 int last) /* index of last pointer to log */
1170 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1171 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1174 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
1175 int level = xfs_btree_get_level(block);
1177 xfs_trans_log_buf(cur->bc_tp, bp,
1178 xfs_btree_ptr_offset(cur, first, level),
1179 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1181 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1182 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1185 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1189 * Log fields from a btree block header.
1192 xfs_btree_log_block(
1193 struct xfs_btree_cur *cur, /* btree cursor */
1194 struct xfs_buf *bp, /* buffer containing btree block */
1195 int fields) /* mask of fields: XFS_BB_... */
1197 int first; /* first byte offset logged */
1198 int last; /* last byte offset logged */
1199 static const short soffsets[] = { /* table of offsets (short) */
1200 offsetof(struct xfs_btree_block, bb_magic),
1201 offsetof(struct xfs_btree_block, bb_level),
1202 offsetof(struct xfs_btree_block, bb_numrecs),
1203 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1204 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1205 XFS_BTREE_SBLOCK_LEN
1207 static const short loffsets[] = { /* table of offsets (long) */
1208 offsetof(struct xfs_btree_block, bb_magic),
1209 offsetof(struct xfs_btree_block, bb_level),
1210 offsetof(struct xfs_btree_block, bb_numrecs),
1211 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1212 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1213 XFS_BTREE_LBLOCK_LEN
1216 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1217 XFS_BTREE_TRACE_ARGBI(cur, bp, fields);
1220 xfs_btree_offsets(fields,
1221 (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1222 loffsets : soffsets,
1223 XFS_BB_NUM_BITS, &first, &last);
1224 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1226 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1227 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1230 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1234 * Increment cursor by one record at the level.
1235 * For nonzero levels the leaf-ward information is untouched.
1238 xfs_btree_increment(
1239 struct xfs_btree_cur *cur,
1241 int *stat) /* success/failure */
1243 struct xfs_btree_block *block;
1244 union xfs_btree_ptr ptr;
1246 int error; /* error return value */
1249 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1250 XFS_BTREE_TRACE_ARGI(cur, level);
1252 ASSERT(level < cur->bc_nlevels);
1254 /* Read-ahead to the right at this level. */
1255 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1257 /* Get a pointer to the btree block. */
1258 block = xfs_btree_get_block(cur, level, &bp);
1261 error = xfs_btree_check_block(cur, block, level, bp);
1266 /* We're done if we remain in the block after the increment. */
1267 if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
1270 /* Fail if we just went off the right edge of the tree. */
1271 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1272 if (xfs_btree_ptr_is_null(cur, &ptr))
1275 XFS_BTREE_STATS_INC(cur, increment);
1278 * March up the tree incrementing pointers.
1279 * Stop when we don't go off the right edge of a block.
1281 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1282 block = xfs_btree_get_block(cur, lev, &bp);
1285 error = xfs_btree_check_block(cur, block, lev, bp);
1290 if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
1293 /* Read-ahead the right block for the next loop. */
1294 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1298 * If we went off the root then we are either seriously
1299 * confused or have the tree root in an inode.
1301 if (lev == cur->bc_nlevels) {
1302 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1305 error = EFSCORRUPTED;
1308 ASSERT(lev < cur->bc_nlevels);
1311 * Now walk back down the tree, fixing up the cursor's buffer
1312 * pointers and key numbers.
1314 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1315 union xfs_btree_ptr *ptrp;
1317 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1318 error = xfs_btree_read_buf_block(cur, ptrp, --lev,
1323 xfs_btree_setbuf(cur, lev, bp);
1324 cur->bc_ptrs[lev] = 1;
1327 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1332 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1337 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1342 * Decrement cursor by one record at the level.
1343 * For nonzero levels the leaf-ward information is untouched.
1346 xfs_btree_decrement(
1347 struct xfs_btree_cur *cur,
1349 int *stat) /* success/failure */
1351 struct xfs_btree_block *block;
1353 int error; /* error return value */
1355 union xfs_btree_ptr ptr;
1357 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1358 XFS_BTREE_TRACE_ARGI(cur, level);
1360 ASSERT(level < cur->bc_nlevels);
1362 /* Read-ahead to the left at this level. */
1363 xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1365 /* We're done if we remain in the block after the decrement. */
1366 if (--cur->bc_ptrs[level] > 0)
1369 /* Get a pointer to the btree block. */
1370 block = xfs_btree_get_block(cur, level, &bp);
1373 error = xfs_btree_check_block(cur, block, level, bp);
1378 /* Fail if we just went off the left edge of the tree. */
1379 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1380 if (xfs_btree_ptr_is_null(cur, &ptr))
1383 XFS_BTREE_STATS_INC(cur, decrement);
1386 * March up the tree decrementing pointers.
1387 * Stop when we don't go off the left edge of a block.
1389 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1390 if (--cur->bc_ptrs[lev] > 0)
1392 /* Read-ahead the left block for the next loop. */
1393 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1397 * If we went off the root then we are seriously confused.
1398 * or the root of the tree is in an inode.
1400 if (lev == cur->bc_nlevels) {
1401 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1404 error = EFSCORRUPTED;
1407 ASSERT(lev < cur->bc_nlevels);
1410 * Now walk back down the tree, fixing up the cursor's buffer
1411 * pointers and key numbers.
1413 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1414 union xfs_btree_ptr *ptrp;
1416 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1417 error = xfs_btree_read_buf_block(cur, ptrp, --lev,
1421 xfs_btree_setbuf(cur, lev, bp);
1422 cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
1425 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1430 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1435 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1440 xfs_btree_lookup_get_block(
1441 struct xfs_btree_cur *cur, /* btree cursor */
1442 int level, /* level in the btree */
1443 union xfs_btree_ptr *pp, /* ptr to btree block */
1444 struct xfs_btree_block **blkp) /* return btree block */
1446 struct xfs_buf *bp; /* buffer pointer for btree block */
1449 /* special case the root block if in an inode */
1450 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1451 (level == cur->bc_nlevels - 1)) {
1452 *blkp = xfs_btree_get_iroot(cur);
1457 * If the old buffer at this level for the disk address we are
1458 * looking for re-use it.
1460 * Otherwise throw it away and get a new one.
1462 bp = cur->bc_bufs[level];
1463 if (bp && XFS_BUF_ADDR(bp) == xfs_btree_ptr_to_daddr(cur, pp)) {
1464 *blkp = XFS_BUF_TO_BLOCK(bp);
1468 error = xfs_btree_read_buf_block(cur, pp, level, 0, blkp, &bp);
1472 xfs_btree_setbuf(cur, level, bp);
1477 * Get current search key. For level 0 we don't actually have a key
1478 * structure so we make one up from the record. For all other levels
1479 * we just return the right key.
1481 STATIC union xfs_btree_key *
1482 xfs_lookup_get_search_key(
1483 struct xfs_btree_cur *cur,
1486 struct xfs_btree_block *block,
1487 union xfs_btree_key *kp)
1490 cur->bc_ops->init_key_from_rec(kp,
1491 xfs_btree_rec_addr(cur, keyno, block));
1495 return xfs_btree_key_addr(cur, keyno, block);
1499 * Lookup the record. The cursor is made to point to it, based on dir.
1500 * Return 0 if can't find any such record, 1 for success.
1504 struct xfs_btree_cur *cur, /* btree cursor */
1505 xfs_lookup_t dir, /* <=, ==, or >= */
1506 int *stat) /* success/failure */
1508 struct xfs_btree_block *block; /* current btree block */
1509 __int64_t diff; /* difference for the current key */
1510 int error; /* error return value */
1511 int keyno; /* current key number */
1512 int level; /* level in the btree */
1513 union xfs_btree_ptr *pp; /* ptr to btree block */
1514 union xfs_btree_ptr ptr; /* ptr to btree block */
1516 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1517 XFS_BTREE_TRACE_ARGI(cur, dir);
1519 XFS_BTREE_STATS_INC(cur, lookup);
1524 /* initialise start pointer from cursor */
1525 cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1529 * Iterate over each level in the btree, starting at the root.
1530 * For each level above the leaves, find the key we need, based
1531 * on the lookup record, then follow the corresponding block
1532 * pointer down to the next level.
1534 for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1535 /* Get the block we need to do the lookup on. */
1536 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1542 * If we already had a key match at a higher level, we
1543 * know we need to use the first entry in this block.
1547 /* Otherwise search this block. Do a binary search. */
1549 int high; /* high entry number */
1550 int low; /* low entry number */
1552 /* Set low and high entry numbers, 1-based. */
1554 high = xfs_btree_get_numrecs(block);
1556 /* Block is empty, must be an empty leaf. */
1557 ASSERT(level == 0 && cur->bc_nlevels == 1);
1559 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1560 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1565 /* Binary search the block. */
1566 while (low <= high) {
1567 union xfs_btree_key key;
1568 union xfs_btree_key *kp;
1570 XFS_BTREE_STATS_INC(cur, compare);
1572 /* keyno is average of low and high. */
1573 keyno = (low + high) >> 1;
1575 /* Get current search key */
1576 kp = xfs_lookup_get_search_key(cur, level,
1577 keyno, block, &key);
1580 * Compute difference to get next direction:
1581 * - less than, move right
1582 * - greater than, move left
1583 * - equal, we're done
1585 diff = cur->bc_ops->key_diff(cur, kp);
1596 * If there are more levels, set up for the next level
1597 * by getting the block number and filling in the cursor.
1601 * If we moved left, need the previous key number,
1602 * unless there isn't one.
1604 if (diff > 0 && --keyno < 1)
1606 pp = xfs_btree_ptr_addr(cur, keyno, block);
1609 error = xfs_btree_check_ptr(cur, pp, 0, level);
1613 cur->bc_ptrs[level] = keyno;
1617 /* Done with the search. See if we need to adjust the results. */
1618 if (dir != XFS_LOOKUP_LE && diff < 0) {
1621 * If ge search and we went off the end of the block, but it's
1622 * not the last block, we're in the wrong block.
1624 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1625 if (dir == XFS_LOOKUP_GE &&
1626 keyno > xfs_btree_get_numrecs(block) &&
1627 !xfs_btree_ptr_is_null(cur, &ptr)) {
1630 cur->bc_ptrs[0] = keyno;
1631 error = xfs_btree_increment(cur, 0, &i);
1634 XFS_WANT_CORRUPTED_RETURN(i == 1);
1635 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1639 } else if (dir == XFS_LOOKUP_LE && diff > 0)
1641 cur->bc_ptrs[0] = keyno;
1643 /* Return if we succeeded or not. */
1644 if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
1646 else if (dir != XFS_LOOKUP_EQ || diff == 0)
1650 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1654 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1659 * Update keys at all levels from here to the root along the cursor's path.
1663 struct xfs_btree_cur *cur,
1664 union xfs_btree_key *keyp,
1667 struct xfs_btree_block *block;
1669 union xfs_btree_key *kp;
1672 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1673 XFS_BTREE_TRACE_ARGIK(cur, level, keyp);
1675 ASSERT(!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) || level >= 1);
1678 * Go up the tree from this level toward the root.
1679 * At each level, update the key value to the value input.
1680 * Stop when we reach a level where the cursor isn't pointing
1681 * at the first entry in the block.
1683 for (ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
1687 block = xfs_btree_get_block(cur, level, &bp);
1689 error = xfs_btree_check_block(cur, block, level, bp);
1691 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1695 ptr = cur->bc_ptrs[level];
1696 kp = xfs_btree_key_addr(cur, ptr, block);
1697 xfs_btree_copy_keys(cur, kp, keyp, 1);
1698 xfs_btree_log_keys(cur, bp, ptr, ptr);
1701 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1706 * Update the record referred to by cur to the value in the
1707 * given record. This either works (return 0) or gets an
1708 * EFSCORRUPTED error.
1712 struct xfs_btree_cur *cur,
1713 union xfs_btree_rec *rec)
1715 struct xfs_btree_block *block;
1719 union xfs_btree_rec *rp;
1721 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1722 XFS_BTREE_TRACE_ARGR(cur, rec);
1724 /* Pick up the current block. */
1725 block = xfs_btree_get_block(cur, 0, &bp);
1728 error = xfs_btree_check_block(cur, block, 0, bp);
1732 /* Get the address of the rec to be updated. */
1733 ptr = cur->bc_ptrs[0];
1734 rp = xfs_btree_rec_addr(cur, ptr, block);
1736 /* Fill in the new contents and log them. */
1737 xfs_btree_copy_recs(cur, rp, rec, 1);
1738 xfs_btree_log_recs(cur, bp, ptr, ptr);
1741 * If we are tracking the last record in the tree and
1742 * we are at the far right edge of the tree, update it.
1744 if (xfs_btree_is_lastrec(cur, block, 0)) {
1745 cur->bc_ops->update_lastrec(cur, block, rec,
1746 ptr, LASTREC_UPDATE);
1749 /* Updating first rec in leaf. Pass new key value up to our parent. */
1751 union xfs_btree_key key;
1753 cur->bc_ops->init_key_from_rec(&key, rec);
1754 error = xfs_btree_updkey(cur, &key, 1);
1759 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1763 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1768 * Move 1 record left from cur/level if possible.
1769 * Update cur to reflect the new path.
1771 STATIC int /* error */
1773 struct xfs_btree_cur *cur,
1775 int *stat) /* success/failure */
1777 union xfs_btree_key key; /* btree key */
1778 struct xfs_buf *lbp; /* left buffer pointer */
1779 struct xfs_btree_block *left; /* left btree block */
1780 int lrecs; /* left record count */
1781 struct xfs_buf *rbp; /* right buffer pointer */
1782 struct xfs_btree_block *right; /* right btree block */
1783 int rrecs; /* right record count */
1784 union xfs_btree_ptr lptr; /* left btree pointer */
1785 union xfs_btree_key *rkp = NULL; /* right btree key */
1786 union xfs_btree_ptr *rpp = NULL; /* right address pointer */
1787 union xfs_btree_rec *rrp = NULL; /* right record pointer */
1788 int error; /* error return value */
1790 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1791 XFS_BTREE_TRACE_ARGI(cur, level);
1793 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1794 level == cur->bc_nlevels - 1)
1797 /* Set up variables for this block as "right". */
1798 right = xfs_btree_get_block(cur, level, &rbp);
1801 error = xfs_btree_check_block(cur, right, level, rbp);
1806 /* If we've got no left sibling then we can't shift an entry left. */
1807 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
1808 if (xfs_btree_ptr_is_null(cur, &lptr))
1812 * If the cursor entry is the one that would be moved, don't
1813 * do it... it's too complicated.
1815 if (cur->bc_ptrs[level] <= 1)
1818 /* Set up the left neighbor as "left". */
1819 error = xfs_btree_read_buf_block(cur, &lptr, level, 0, &left, &lbp);
1823 /* If it's full, it can't take another entry. */
1824 lrecs = xfs_btree_get_numrecs(left);
1825 if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
1828 rrecs = xfs_btree_get_numrecs(right);
1831 * We add one entry to the left side and remove one for the right side.
1832 * Account for it here, the changes will be updated on disk and logged
1838 XFS_BTREE_STATS_INC(cur, lshift);
1839 XFS_BTREE_STATS_ADD(cur, moves, 1);
1842 * If non-leaf, copy a key and a ptr to the left block.
1843 * Log the changes to the left block.
1846 /* It's a non-leaf. Move keys and pointers. */
1847 union xfs_btree_key *lkp; /* left btree key */
1848 union xfs_btree_ptr *lpp; /* left address pointer */
1850 lkp = xfs_btree_key_addr(cur, lrecs, left);
1851 rkp = xfs_btree_key_addr(cur, 1, right);
1853 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
1854 rpp = xfs_btree_ptr_addr(cur, 1, right);
1856 error = xfs_btree_check_ptr(cur, rpp, 0, level);
1860 xfs_btree_copy_keys(cur, lkp, rkp, 1);
1861 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
1863 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
1864 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
1866 ASSERT(cur->bc_ops->keys_inorder(cur,
1867 xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
1869 /* It's a leaf. Move records. */
1870 union xfs_btree_rec *lrp; /* left record pointer */
1872 lrp = xfs_btree_rec_addr(cur, lrecs, left);
1873 rrp = xfs_btree_rec_addr(cur, 1, right);
1875 xfs_btree_copy_recs(cur, lrp, rrp, 1);
1876 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
1878 ASSERT(cur->bc_ops->recs_inorder(cur,
1879 xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
1882 xfs_btree_set_numrecs(left, lrecs);
1883 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
1885 xfs_btree_set_numrecs(right, rrecs);
1886 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
1889 * Slide the contents of right down one entry.
1891 XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
1893 /* It's a nonleaf. operate on keys and ptrs */
1895 int i; /* loop index */
1897 for (i = 0; i < rrecs; i++) {
1898 error = xfs_btree_check_ptr(cur, rpp, i + 1, level);
1903 xfs_btree_shift_keys(cur,
1904 xfs_btree_key_addr(cur, 2, right),
1906 xfs_btree_shift_ptrs(cur,
1907 xfs_btree_ptr_addr(cur, 2, right),
1910 xfs_btree_log_keys(cur, rbp, 1, rrecs);
1911 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
1913 /* It's a leaf. operate on records */
1914 xfs_btree_shift_recs(cur,
1915 xfs_btree_rec_addr(cur, 2, right),
1917 xfs_btree_log_recs(cur, rbp, 1, rrecs);
1920 * If it's the first record in the block, we'll need a key
1921 * structure to pass up to the next level (updkey).
1923 cur->bc_ops->init_key_from_rec(&key,
1924 xfs_btree_rec_addr(cur, 1, right));
1928 /* Update the parent key values of right. */
1929 error = xfs_btree_updkey(cur, rkp, level + 1);
1933 /* Slide the cursor value left one. */
1934 cur->bc_ptrs[level]--;
1936 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1941 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1946 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1951 * Move 1 record right from cur/level if possible.
1952 * Update cur to reflect the new path.
1954 STATIC int /* error */
1956 struct xfs_btree_cur *cur,
1958 int *stat) /* success/failure */
1960 union xfs_btree_key key; /* btree key */
1961 struct xfs_buf *lbp; /* left buffer pointer */
1962 struct xfs_btree_block *left; /* left btree block */
1963 struct xfs_buf *rbp; /* right buffer pointer */
1964 struct xfs_btree_block *right; /* right btree block */
1965 struct xfs_btree_cur *tcur; /* temporary btree cursor */
1966 union xfs_btree_ptr rptr; /* right block pointer */
1967 union xfs_btree_key *rkp; /* right btree key */
1968 int rrecs; /* right record count */
1969 int lrecs; /* left record count */
1970 int error; /* error return value */
1971 int i; /* loop counter */
1973 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1974 XFS_BTREE_TRACE_ARGI(cur, level);
1976 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1977 (level == cur->bc_nlevels - 1))
1980 /* Set up variables for this block as "left". */
1981 left = xfs_btree_get_block(cur, level, &lbp);
1984 error = xfs_btree_check_block(cur, left, level, lbp);
1989 /* If we've got no right sibling then we can't shift an entry right. */
1990 xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
1991 if (xfs_btree_ptr_is_null(cur, &rptr))
1995 * If the cursor entry is the one that would be moved, don't
1996 * do it... it's too complicated.
1998 lrecs = xfs_btree_get_numrecs(left);
1999 if (cur->bc_ptrs[level] >= lrecs)
2002 /* Set up the right neighbor as "right". */
2003 error = xfs_btree_read_buf_block(cur, &rptr, level, 0, &right, &rbp);
2007 /* If it's full, it can't take another entry. */
2008 rrecs = xfs_btree_get_numrecs(right);
2009 if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2012 XFS_BTREE_STATS_INC(cur, rshift);
2013 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2016 * Make a hole at the start of the right neighbor block, then
2017 * copy the last left block entry to the hole.
2020 /* It's a nonleaf. make a hole in the keys and ptrs */
2021 union xfs_btree_key *lkp;
2022 union xfs_btree_ptr *lpp;
2023 union xfs_btree_ptr *rpp;
2025 lkp = xfs_btree_key_addr(cur, lrecs, left);
2026 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2027 rkp = xfs_btree_key_addr(cur, 1, right);
2028 rpp = xfs_btree_ptr_addr(cur, 1, right);
2031 for (i = rrecs - 1; i >= 0; i--) {
2032 error = xfs_btree_check_ptr(cur, rpp, i, level);
2038 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2039 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2042 error = xfs_btree_check_ptr(cur, lpp, 0, level);
2047 /* Now put the new data in, and log it. */
2048 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2049 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2051 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2052 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2054 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2055 xfs_btree_key_addr(cur, 2, right)));
2057 /* It's a leaf. make a hole in the records */
2058 union xfs_btree_rec *lrp;
2059 union xfs_btree_rec *rrp;
2061 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2062 rrp = xfs_btree_rec_addr(cur, 1, right);
2064 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2066 /* Now put the new data in, and log it. */
2067 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2068 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2070 cur->bc_ops->init_key_from_rec(&key, rrp);
2073 ASSERT(cur->bc_ops->recs_inorder(cur, rrp,
2074 xfs_btree_rec_addr(cur, 2, right)));
2078 * Decrement and log left's numrecs, bump and log right's numrecs.
2080 xfs_btree_set_numrecs(left, --lrecs);
2081 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2083 xfs_btree_set_numrecs(right, ++rrecs);
2084 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2087 * Using a temporary cursor, update the parent key values of the
2088 * block on the right.
2090 error = xfs_btree_dup_cursor(cur, &tcur);
2093 i = xfs_btree_lastrec(tcur, level);
2094 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
2096 error = xfs_btree_increment(tcur, level, &i);
2100 error = xfs_btree_updkey(tcur, rkp, level + 1);
2104 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2106 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2111 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2116 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2120 XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2121 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2126 * Split cur/level block in half.
2127 * Return new block number and the key to its first
2128 * record (to be inserted into parent).
2130 STATIC int /* error */
2132 struct xfs_btree_cur *cur,
2134 union xfs_btree_ptr *ptrp,
2135 union xfs_btree_key *key,
2136 struct xfs_btree_cur **curp,
2137 int *stat) /* success/failure */
2139 union xfs_btree_ptr lptr; /* left sibling block ptr */
2140 struct xfs_buf *lbp; /* left buffer pointer */
2141 struct xfs_btree_block *left; /* left btree block */
2142 union xfs_btree_ptr rptr; /* right sibling block ptr */
2143 struct xfs_buf *rbp; /* right buffer pointer */
2144 struct xfs_btree_block *right; /* right btree block */
2145 union xfs_btree_ptr rrptr; /* right-right sibling ptr */
2146 struct xfs_buf *rrbp; /* right-right buffer pointer */
2147 struct xfs_btree_block *rrblock; /* right-right btree block */
2151 int error; /* error return value */
2156 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2157 XFS_BTREE_TRACE_ARGIPK(cur, level, *ptrp, key);
2159 XFS_BTREE_STATS_INC(cur, split);
2161 /* Set up left block (current one). */
2162 left = xfs_btree_get_block(cur, level, &lbp);
2165 error = xfs_btree_check_block(cur, left, level, lbp);
2170 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2172 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2173 error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, 1, stat);
2178 XFS_BTREE_STATS_INC(cur, alloc);
2180 /* Set up the new block as "right". */
2181 error = xfs_btree_get_buf_block(cur, &rptr, 0, &right, &rbp);
2185 /* Fill in the btree header for the new right block. */
2186 xfs_btree_init_block(cur, xfs_btree_get_level(left), 0, right);
2189 * Split the entries between the old and the new block evenly.
2190 * Make sure that if there's an odd number of entries now, that
2191 * each new block will have the same number of entries.
2193 lrecs = xfs_btree_get_numrecs(left);
2195 if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
2197 src_index = (lrecs - rrecs + 1);
2199 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2202 * Copy btree block entries from the left block over to the
2203 * new block, the right. Update the right block and log the
2207 /* It's a non-leaf. Move keys and pointers. */
2208 union xfs_btree_key *lkp; /* left btree key */
2209 union xfs_btree_ptr *lpp; /* left address pointer */
2210 union xfs_btree_key *rkp; /* right btree key */
2211 union xfs_btree_ptr *rpp; /* right address pointer */
2213 lkp = xfs_btree_key_addr(cur, src_index, left);
2214 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2215 rkp = xfs_btree_key_addr(cur, 1, right);
2216 rpp = xfs_btree_ptr_addr(cur, 1, right);
2219 for (i = src_index; i < rrecs; i++) {
2220 error = xfs_btree_check_ptr(cur, lpp, i, level);
2226 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2227 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2229 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2230 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2232 /* Grab the keys to the entries moved to the right block */
2233 xfs_btree_copy_keys(cur, key, rkp, 1);
2235 /* It's a leaf. Move records. */
2236 union xfs_btree_rec *lrp; /* left record pointer */
2237 union xfs_btree_rec *rrp; /* right record pointer */
2239 lrp = xfs_btree_rec_addr(cur, src_index, left);
2240 rrp = xfs_btree_rec_addr(cur, 1, right);
2242 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2243 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2245 cur->bc_ops->init_key_from_rec(key,
2246 xfs_btree_rec_addr(cur, 1, right));
2251 * Find the left block number by looking in the buffer.
2252 * Adjust numrecs, sibling pointers.
2254 xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2255 xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2256 xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2257 xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2260 xfs_btree_set_numrecs(left, lrecs);
2261 xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2263 xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2264 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2267 * If there's a block to the new block's right, make that block
2268 * point back to right instead of to left.
2270 if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2271 error = xfs_btree_read_buf_block(cur, &rrptr, level,
2272 0, &rrblock, &rrbp);
2275 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2276 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2279 * If the cursor is really in the right block, move it there.
2280 * If it's just pointing past the last entry in left, then we'll
2281 * insert there, so don't change anything in that case.
2283 if (cur->bc_ptrs[level] > lrecs + 1) {
2284 xfs_btree_setbuf(cur, level, rbp);
2285 cur->bc_ptrs[level] -= lrecs;
2288 * If there are more levels, we'll need another cursor which refers
2289 * the right block, no matter where this cursor was.
2291 if (level + 1 < cur->bc_nlevels) {
2292 error = xfs_btree_dup_cursor(cur, curp);
2295 (*curp)->bc_ptrs[level + 1]++;
2298 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2302 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2307 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2312 * Copy the old inode root contents into a real block and make the
2313 * broot point to it.
2316 xfs_btree_new_iroot(
2317 struct xfs_btree_cur *cur, /* btree cursor */
2318 int *logflags, /* logging flags for inode */
2319 int *stat) /* return status - 0 fail */
2321 struct xfs_buf *cbp; /* buffer for cblock */
2322 struct xfs_btree_block *block; /* btree block */
2323 struct xfs_btree_block *cblock; /* child btree block */
2324 union xfs_btree_key *ckp; /* child key pointer */
2325 union xfs_btree_ptr *cpp; /* child ptr pointer */
2326 union xfs_btree_key *kp; /* pointer to btree key */
2327 union xfs_btree_ptr *pp; /* pointer to block addr */
2328 union xfs_btree_ptr nptr; /* new block addr */
2329 int level; /* btree level */
2330 int error; /* error return code */
2332 int i; /* loop counter */
2335 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2336 XFS_BTREE_STATS_INC(cur, newroot);
2338 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2340 level = cur->bc_nlevels - 1;
2342 block = xfs_btree_get_iroot(cur);
2343 pp = xfs_btree_ptr_addr(cur, 1, block);
2345 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2346 error = cur->bc_ops->alloc_block(cur, pp, &nptr, 1, stat);
2350 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2353 XFS_BTREE_STATS_INC(cur, alloc);
2355 /* Copy the root into a real block. */
2356 error = xfs_btree_get_buf_block(cur, &nptr, 0, &cblock, &cbp);
2360 memcpy(cblock, block, xfs_btree_block_len(cur));
2362 be16_add_cpu(&block->bb_level, 1);
2363 xfs_btree_set_numrecs(block, 1);
2365 cur->bc_ptrs[level + 1] = 1;
2367 kp = xfs_btree_key_addr(cur, 1, block);
2368 ckp = xfs_btree_key_addr(cur, 1, cblock);
2369 xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
2371 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
2373 for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
2374 error = xfs_btree_check_ptr(cur, pp, i, level);
2379 xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
2382 error = xfs_btree_check_ptr(cur, &nptr, 0, level);
2386 xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
2388 xfs_iroot_realloc(cur->bc_private.b.ip,
2389 1 - xfs_btree_get_numrecs(cblock),
2390 cur->bc_private.b.whichfork);
2392 xfs_btree_setbuf(cur, level, cbp);
2395 * Do all this logging at the end so that
2396 * the root is at the right level.
2398 xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
2399 xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2400 xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2403 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork);
2405 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2408 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2413 * Allocate a new root block, fill it in.
2415 STATIC int /* error */
2417 struct xfs_btree_cur *cur, /* btree cursor */
2418 int *stat) /* success/failure */
2420 struct xfs_btree_block *block; /* one half of the old root block */
2421 struct xfs_buf *bp; /* buffer containing block */
2422 int error; /* error return value */
2423 struct xfs_buf *lbp; /* left buffer pointer */
2424 struct xfs_btree_block *left; /* left btree block */
2425 struct xfs_buf *nbp; /* new (root) buffer */
2426 struct xfs_btree_block *new; /* new (root) btree block */
2427 int nptr; /* new value for key index, 1 or 2 */
2428 struct xfs_buf *rbp; /* right buffer pointer */
2429 struct xfs_btree_block *right; /* right btree block */
2430 union xfs_btree_ptr rptr;
2431 union xfs_btree_ptr lptr;
2433 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2434 XFS_BTREE_STATS_INC(cur, newroot);
2436 /* initialise our start point from the cursor */
2437 cur->bc_ops->init_ptr_from_cur(cur, &rptr);
2439 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2440 error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, 1, stat);
2445 XFS_BTREE_STATS_INC(cur, alloc);
2447 /* Set up the new block. */
2448 error = xfs_btree_get_buf_block(cur, &lptr, 0, &new, &nbp);
2452 /* Set the root in the holding structure increasing the level by 1. */
2453 cur->bc_ops->set_root(cur, &lptr, 1);
2456 * At the previous root level there are now two blocks: the old root,
2457 * and the new block generated when it was split. We don't know which
2458 * one the cursor is pointing at, so we set up variables "left" and
2459 * "right" for each case.
2461 block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
2464 error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
2469 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
2470 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
2471 /* Our block is left, pick up the right block. */
2473 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2475 error = xfs_btree_read_buf_block(cur, &rptr,
2476 cur->bc_nlevels - 1, 0, &right, &rbp);
2482 /* Our block is right, pick up the left block. */
2484 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
2486 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2487 error = xfs_btree_read_buf_block(cur, &lptr,
2488 cur->bc_nlevels - 1, 0, &left, &lbp);
2494 /* Fill in the new block's btree header and log it. */
2495 xfs_btree_init_block(cur, cur->bc_nlevels, 2, new);
2496 xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
2497 ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
2498 !xfs_btree_ptr_is_null(cur, &rptr));
2500 /* Fill in the key data in the new root. */
2501 if (xfs_btree_get_level(left) > 0) {
2502 xfs_btree_copy_keys(cur,
2503 xfs_btree_key_addr(cur, 1, new),
2504 xfs_btree_key_addr(cur, 1, left), 1);
2505 xfs_btree_copy_keys(cur,
2506 xfs_btree_key_addr(cur, 2, new),
2507 xfs_btree_key_addr(cur, 1, right), 1);
2509 cur->bc_ops->init_key_from_rec(
2510 xfs_btree_key_addr(cur, 1, new),
2511 xfs_btree_rec_addr(cur, 1, left));
2512 cur->bc_ops->init_key_from_rec(
2513 xfs_btree_key_addr(cur, 2, new),
2514 xfs_btree_rec_addr(cur, 1, right));
2516 xfs_btree_log_keys(cur, nbp, 1, 2);
2518 /* Fill in the pointer data in the new root. */
2519 xfs_btree_copy_ptrs(cur,
2520 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
2521 xfs_btree_copy_ptrs(cur,
2522 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
2523 xfs_btree_log_ptrs(cur, nbp, 1, 2);
2525 /* Fix up the cursor. */
2526 xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
2527 cur->bc_ptrs[cur->bc_nlevels] = nptr;
2529 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2533 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2536 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2542 xfs_btree_make_block_unfull(
2543 struct xfs_btree_cur *cur, /* btree cursor */
2544 int level, /* btree level */
2545 int numrecs,/* # of recs in block */
2546 int *oindex,/* old tree index */
2547 int *index, /* new tree index */
2548 union xfs_btree_ptr *nptr, /* new btree ptr */
2549 struct xfs_btree_cur **ncur, /* new btree cursor */
2550 union xfs_btree_rec *nrec, /* new record */
2553 union xfs_btree_key key; /* new btree key value */
2556 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2557 level == cur->bc_nlevels - 1) {
2558 struct xfs_inode *ip = cur->bc_private.b.ip;
2560 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
2561 /* A root block that can be made bigger. */
2563 xfs_iroot_realloc(ip, 1, cur->bc_private.b.whichfork);
2565 /* A root block that needs replacing */
2568 error = xfs_btree_new_iroot(cur, &logflags, stat);
2569 if (error || *stat == 0)
2572 xfs_trans_log_inode(cur->bc_tp, ip, logflags);
2578 /* First, try shifting an entry to the right neighbor. */
2579 error = xfs_btree_rshift(cur, level, stat);
2583 /* Next, try shifting an entry to the left neighbor. */
2584 error = xfs_btree_lshift(cur, level, stat);
2589 *oindex = *index = cur->bc_ptrs[level];
2594 * Next, try splitting the current block in half.
2596 * If this works we have to re-set our variables because we
2597 * could be in a different block now.
2599 error = xfs_btree_split(cur, level, nptr, &key, ncur, stat);
2600 if (error || *stat == 0)
2604 *index = cur->bc_ptrs[level];
2605 cur->bc_ops->init_rec_from_key(&key, nrec);
2610 * Insert one record/level. Return information to the caller
2611 * allowing the next level up to proceed if necessary.
2615 struct xfs_btree_cur *cur, /* btree cursor */
2616 int level, /* level to insert record at */
2617 union xfs_btree_ptr *ptrp, /* i/o: block number inserted */
2618 union xfs_btree_rec *recp, /* i/o: record data inserted */
2619 struct xfs_btree_cur **curp, /* output: new cursor replacing cur */
2620 int *stat) /* success/failure */
2622 struct xfs_btree_block *block; /* btree block */
2623 struct xfs_buf *bp; /* buffer for block */
2624 union xfs_btree_key key; /* btree key */
2625 union xfs_btree_ptr nptr; /* new block ptr */
2626 struct xfs_btree_cur *ncur; /* new btree cursor */
2627 union xfs_btree_rec nrec; /* new record count */
2628 int optr; /* old key/record index */
2629 int ptr; /* key/record index */
2630 int numrecs;/* number of records */
2631 int error; /* error return value */
2636 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2637 XFS_BTREE_TRACE_ARGIPR(cur, level, *ptrp, recp);
2642 * If we have an external root pointer, and we've made it to the
2643 * root level, allocate a new root block and we're done.
2645 if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2646 (level >= cur->bc_nlevels)) {
2647 error = xfs_btree_new_root(cur, stat);
2648 xfs_btree_set_ptr_null(cur, ptrp);
2650 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2654 /* If we're off the left edge, return failure. */
2655 ptr = cur->bc_ptrs[level];
2657 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2662 /* Make a key out of the record data to be inserted, and save it. */
2663 cur->bc_ops->init_key_from_rec(&key, recp);
2667 XFS_BTREE_STATS_INC(cur, insrec);
2669 /* Get pointers to the btree buffer and block. */
2670 block = xfs_btree_get_block(cur, level, &bp);
2671 numrecs = xfs_btree_get_numrecs(block);
2674 error = xfs_btree_check_block(cur, block, level, bp);
2678 /* Check that the new entry is being inserted in the right place. */
2679 if (ptr <= numrecs) {
2681 ASSERT(cur->bc_ops->recs_inorder(cur, recp,
2682 xfs_btree_rec_addr(cur, ptr, block)));
2684 ASSERT(cur->bc_ops->keys_inorder(cur, &key,
2685 xfs_btree_key_addr(cur, ptr, block)));
2691 * If the block is full, we can't insert the new entry until we
2692 * make the block un-full.
2694 xfs_btree_set_ptr_null(cur, &nptr);
2695 if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
2696 error = xfs_btree_make_block_unfull(cur, level, numrecs,
2697 &optr, &ptr, &nptr, &ncur, &nrec, stat);
2698 if (error || *stat == 0)
2703 * The current block may have changed if the block was
2704 * previously full and we have just made space in it.
2706 block = xfs_btree_get_block(cur, level, &bp);
2707 numrecs = xfs_btree_get_numrecs(block);
2710 error = xfs_btree_check_block(cur, block, level, bp);
2716 * At this point we know there's room for our new entry in the block
2717 * we're pointing at.
2719 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
2722 /* It's a nonleaf. make a hole in the keys and ptrs */
2723 union xfs_btree_key *kp;
2724 union xfs_btree_ptr *pp;
2726 kp = xfs_btree_key_addr(cur, ptr, block);
2727 pp = xfs_btree_ptr_addr(cur, ptr, block);
2730 for (i = numrecs - ptr; i >= 0; i--) {
2731 error = xfs_btree_check_ptr(cur, pp, i, level);
2737 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
2738 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
2741 error = xfs_btree_check_ptr(cur, ptrp, 0, level);
2746 /* Now put the new data in, bump numrecs and log it. */
2747 xfs_btree_copy_keys(cur, kp, &key, 1);
2748 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
2750 xfs_btree_set_numrecs(block, numrecs);
2751 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
2752 xfs_btree_log_keys(cur, bp, ptr, numrecs);
2754 if (ptr < numrecs) {
2755 ASSERT(cur->bc_ops->keys_inorder(cur, kp,
2756 xfs_btree_key_addr(cur, ptr + 1, block)));
2760 /* It's a leaf. make a hole in the records */
2761 union xfs_btree_rec *rp;
2763 rp = xfs_btree_rec_addr(cur, ptr, block);
2765 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
2767 /* Now put the new data in, bump numrecs and log it. */
2768 xfs_btree_copy_recs(cur, rp, recp, 1);
2769 xfs_btree_set_numrecs(block, ++numrecs);
2770 xfs_btree_log_recs(cur, bp, ptr, numrecs);
2772 if (ptr < numrecs) {
2773 ASSERT(cur->bc_ops->recs_inorder(cur, rp,
2774 xfs_btree_rec_addr(cur, ptr + 1, block)));
2779 /* Log the new number of records in the btree header. */
2780 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
2782 /* If we inserted at the start of a block, update the parents' keys. */
2784 error = xfs_btree_updkey(cur, &key, level + 1);
2790 * If we are tracking the last record in the tree and
2791 * we are at the far right edge of the tree, update it.
2793 if (xfs_btree_is_lastrec(cur, block, level)) {
2794 cur->bc_ops->update_lastrec(cur, block, recp,
2795 ptr, LASTREC_INSREC);
2799 * Return the new block number, if any.
2800 * If there is one, give back a record value and a cursor too.
2803 if (!xfs_btree_ptr_is_null(cur, &nptr)) {
2808 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2813 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2818 * Insert the record at the point referenced by cur.
2820 * A multi-level split of the tree on insert will invalidate the original
2821 * cursor. All callers of this function should assume that the cursor is
2822 * no longer valid and revalidate it.
2826 struct xfs_btree_cur *cur,
2829 int error; /* error return value */
2830 int i; /* result value, 0 for failure */
2831 int level; /* current level number in btree */
2832 union xfs_btree_ptr nptr; /* new block number (split result) */
2833 struct xfs_btree_cur *ncur; /* new cursor (split result) */
2834 struct xfs_btree_cur *pcur; /* previous level's cursor */
2835 union xfs_btree_rec rec; /* record to insert */
2841 xfs_btree_set_ptr_null(cur, &nptr);
2842 cur->bc_ops->init_rec_from_cur(cur, &rec);
2845 * Loop going up the tree, starting at the leaf level.
2846 * Stop when we don't get a split block, that must mean that
2847 * the insert is finished with this level.
2851 * Insert nrec/nptr into this level of the tree.
2852 * Note if we fail, nptr will be null.
2854 error = xfs_btree_insrec(pcur, level, &nptr, &rec, &ncur, &i);
2857 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
2861 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
2865 * See if the cursor we just used is trash.
2866 * Can't trash the caller's cursor, but otherwise we should
2867 * if ncur is a new cursor or we're about to be done.
2870 (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
2871 /* Save the state from the cursor before we trash it */
2872 if (cur->bc_ops->update_cursor)
2873 cur->bc_ops->update_cursor(pcur, cur);
2874 cur->bc_nlevels = pcur->bc_nlevels;
2875 xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
2877 /* If we got a new cursor, switch to it. */
2882 } while (!xfs_btree_ptr_is_null(cur, &nptr));
2884 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2888 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2893 * Try to merge a non-leaf block back into the inode root.
2895 * Note: the killroot names comes from the fact that we're effectively
2896 * killing the old root block. But because we can't just delete the
2897 * inode we have to copy the single block it was pointing to into the
2901 xfs_btree_kill_iroot(
2902 struct xfs_btree_cur *cur)
2904 int whichfork = cur->bc_private.b.whichfork;
2905 struct xfs_inode *ip = cur->bc_private.b.ip;
2906 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
2907 struct xfs_btree_block *block;
2908 struct xfs_btree_block *cblock;
2909 union xfs_btree_key *kp;
2910 union xfs_btree_key *ckp;
2911 union xfs_btree_ptr *pp;
2912 union xfs_btree_ptr *cpp;
2913 struct xfs_buf *cbp;
2918 union xfs_btree_ptr ptr;
2922 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2924 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2925 ASSERT(cur->bc_nlevels > 1);
2928 * Don't deal with the root block needs to be a leaf case.
2929 * We're just going to turn the thing back into extents anyway.
2931 level = cur->bc_nlevels - 1;
2936 * Give up if the root has multiple children.
2938 block = xfs_btree_get_iroot(cur);
2939 if (xfs_btree_get_numrecs(block) != 1)
2942 cblock = xfs_btree_get_block(cur, level - 1, &cbp);
2943 numrecs = xfs_btree_get_numrecs(cblock);
2946 * Only do this if the next level will fit.
2947 * Then the data must be copied up to the inode,
2948 * instead of freeing the root you free the next level.
2950 if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
2953 XFS_BTREE_STATS_INC(cur, killroot);
2956 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
2957 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
2958 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
2959 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
2962 index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
2964 xfs_iroot_realloc(cur->bc_private.b.ip, index,
2965 cur->bc_private.b.whichfork);
2966 block = ifp->if_broot;
2969 be16_add_cpu(&block->bb_numrecs, index);
2970 ASSERT(block->bb_numrecs == cblock->bb_numrecs);
2972 kp = xfs_btree_key_addr(cur, 1, block);
2973 ckp = xfs_btree_key_addr(cur, 1, cblock);
2974 xfs_btree_copy_keys(cur, kp, ckp, numrecs);
2976 pp = xfs_btree_ptr_addr(cur, 1, block);
2977 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
2979 for (i = 0; i < numrecs; i++) {
2982 error = xfs_btree_check_ptr(cur, cpp, i, level - 1);
2984 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2989 xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
2991 cur->bc_ops->free_block(cur, cbp);
2992 XFS_BTREE_STATS_INC(cur, free);
2994 cur->bc_bufs[level - 1] = NULL;
2995 be16_add_cpu(&block->bb_level, -1);
2996 xfs_trans_log_inode(cur->bc_tp, ip,
2997 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork));
3000 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3005 * Kill the current root node, and replace it with it's only child node.
3008 xfs_btree_kill_root(
3009 struct xfs_btree_cur *cur,
3012 union xfs_btree_ptr *newroot)
3016 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3017 XFS_BTREE_STATS_INC(cur, killroot);
3020 * Update the root pointer, decreasing the level by 1 and then
3021 * free the old root.
3023 cur->bc_ops->set_root(cur, newroot, -1);
3025 error = cur->bc_ops->free_block(cur, bp);
3027 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3031 XFS_BTREE_STATS_INC(cur, free);
3033 cur->bc_bufs[level] = NULL;
3034 cur->bc_ra[level] = 0;
3037 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3042 xfs_btree_dec_cursor(
3043 struct xfs_btree_cur *cur,
3051 error = xfs_btree_decrement(cur, level, &i);
3056 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3062 * Single level of the btree record deletion routine.
3063 * Delete record pointed to by cur/level.
3064 * Remove the record from its block then rebalance the tree.
3065 * Return 0 for error, 1 for done, 2 to go on to the next level.
3067 STATIC int /* error */
3069 struct xfs_btree_cur *cur, /* btree cursor */
3070 int level, /* level removing record from */
3071 int *stat) /* fail/done/go-on */
3073 struct xfs_btree_block *block; /* btree block */
3074 union xfs_btree_ptr cptr; /* current block ptr */
3075 struct xfs_buf *bp; /* buffer for block */
3076 int error; /* error return value */
3077 int i; /* loop counter */
3078 union xfs_btree_key key; /* storage for keyp */
3079 union xfs_btree_key *keyp = &key; /* passed to the next level */
3080 union xfs_btree_ptr lptr; /* left sibling block ptr */
3081 struct xfs_buf *lbp; /* left buffer pointer */
3082 struct xfs_btree_block *left; /* left btree block */
3083 int lrecs = 0; /* left record count */
3084 int ptr; /* key/record index */
3085 union xfs_btree_ptr rptr; /* right sibling block ptr */
3086 struct xfs_buf *rbp; /* right buffer pointer */
3087 struct xfs_btree_block *right; /* right btree block */
3088 struct xfs_btree_block *rrblock; /* right-right btree block */
3089 struct xfs_buf *rrbp; /* right-right buffer pointer */
3090 int rrecs = 0; /* right record count */
3091 struct xfs_btree_cur *tcur; /* temporary btree cursor */
3092 int numrecs; /* temporary numrec count */
3094 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3095 XFS_BTREE_TRACE_ARGI(cur, level);
3099 /* Get the index of the entry being deleted, check for nothing there. */
3100 ptr = cur->bc_ptrs[level];
3102 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3107 /* Get the buffer & block containing the record or key/ptr. */
3108 block = xfs_btree_get_block(cur, level, &bp);
3109 numrecs = xfs_btree_get_numrecs(block);
3112 error = xfs_btree_check_block(cur, block, level, bp);
3117 /* Fail if we're off the end of the block. */
3118 if (ptr > numrecs) {
3119 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3124 XFS_BTREE_STATS_INC(cur, delrec);
3125 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3127 /* Excise the entries being deleted. */
3129 /* It's a nonleaf. operate on keys and ptrs */
3130 union xfs_btree_key *lkp;
3131 union xfs_btree_ptr *lpp;
3133 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3134 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3137 for (i = 0; i < numrecs - ptr; i++) {
3138 error = xfs_btree_check_ptr(cur, lpp, i, level);
3144 if (ptr < numrecs) {
3145 xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3146 xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3147 xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3148 xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3152 * If it's the first record in the block, we'll need to pass a
3153 * key up to the next level (updkey).
3156 keyp = xfs_btree_key_addr(cur, 1, block);
3158 /* It's a leaf. operate on records */
3159 if (ptr < numrecs) {
3160 xfs_btree_shift_recs(cur,
3161 xfs_btree_rec_addr(cur, ptr + 1, block),
3163 xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3167 * If it's the first record in the block, we'll need a key
3168 * structure to pass up to the next level (updkey).
3171 cur->bc_ops->init_key_from_rec(&key,
3172 xfs_btree_rec_addr(cur, 1, block));
3178 * Decrement and log the number of entries in the block.
3180 xfs_btree_set_numrecs(block, --numrecs);
3181 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3184 * If we are tracking the last record in the tree and
3185 * we are at the far right edge of the tree, update it.
3187 if (xfs_btree_is_lastrec(cur, block, level)) {
3188 cur->bc_ops->update_lastrec(cur, block, NULL,
3189 ptr, LASTREC_DELREC);
3193 * We're at the root level. First, shrink the root block in-memory.
3194 * Try to get rid of the next level down. If we can't then there's
3195 * nothing left to do.
3197 if (level == cur->bc_nlevels - 1) {
3198 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3199 xfs_iroot_realloc(cur->bc_private.b.ip, -1,
3200 cur->bc_private.b.whichfork);
3202 error = xfs_btree_kill_iroot(cur);
3206 error = xfs_btree_dec_cursor(cur, level, stat);
3214 * If this is the root level, and there's only one entry left,
3215 * and it's NOT the leaf level, then we can get rid of this
3218 if (numrecs == 1 && level > 0) {
3219 union xfs_btree_ptr *pp;
3221 * pp is still set to the first pointer in the block.
3222 * Make it the new root of the btree.
3224 pp = xfs_btree_ptr_addr(cur, 1, block);
3225 error = xfs_btree_kill_root(cur, bp, level, pp);
3228 } else if (level > 0) {
3229 error = xfs_btree_dec_cursor(cur, level, stat);
3238 * If we deleted the leftmost entry in the block, update the
3239 * key values above us in the tree.
3242 error = xfs_btree_updkey(cur, keyp, level + 1);
3248 * If the number of records remaining in the block is at least
3249 * the minimum, we're done.
3251 if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3252 error = xfs_btree_dec_cursor(cur, level, stat);
3259 * Otherwise, we have to move some records around to keep the
3260 * tree balanced. Look at the left and right sibling blocks to
3261 * see if we can re-balance by moving only one record.
3263 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3264 xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3266 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3268 * One child of root, need to get a chance to copy its contents
3269 * into the root and delete it. Can't go up to next level,
3270 * there's nothing to delete there.
3272 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3273 xfs_btree_ptr_is_null(cur, &lptr) &&
3274 level == cur->bc_nlevels - 2) {
3275 error = xfs_btree_kill_iroot(cur);
3277 error = xfs_btree_dec_cursor(cur, level, stat);
3284 ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3285 !xfs_btree_ptr_is_null(cur, &lptr));
3288 * Duplicate the cursor so our btree manipulations here won't
3289 * disrupt the next level up.
3291 error = xfs_btree_dup_cursor(cur, &tcur);
3296 * If there's a right sibling, see if it's ok to shift an entry
3299 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3301 * Move the temp cursor to the last entry in the next block.
3302 * Actually any entry but the first would suffice.
3304 i = xfs_btree_lastrec(tcur, level);
3305 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3307 error = xfs_btree_increment(tcur, level, &i);
3310 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3312 i = xfs_btree_lastrec(tcur, level);
3313 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3315 /* Grab a pointer to the block. */
3316 right = xfs_btree_get_block(tcur, level, &rbp);
3318 error = xfs_btree_check_block(tcur, right, level, rbp);
3322 /* Grab the current block number, for future use. */
3323 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3326 * If right block is full enough so that removing one entry
3327 * won't make it too empty, and left-shifting an entry out
3328 * of right to us works, we're done.
3330 if (xfs_btree_get_numrecs(right) - 1 >=
3331 cur->bc_ops->get_minrecs(tcur, level)) {
3332 error = xfs_btree_lshift(tcur, level, &i);
3336 ASSERT(xfs_btree_get_numrecs(block) >=
3337 cur->bc_ops->get_minrecs(tcur, level));
3339 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3342 error = xfs_btree_dec_cursor(cur, level, stat);
3350 * Otherwise, grab the number of records in right for
3351 * future reference, and fix up the temp cursor to point
3352 * to our block again (last record).
3354 rrecs = xfs_btree_get_numrecs(right);
3355 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3356 i = xfs_btree_firstrec(tcur, level);
3357 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3359 error = xfs_btree_decrement(tcur, level, &i);
3362 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3367 * If there's a left sibling, see if it's ok to shift an entry
3370 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3372 * Move the temp cursor to the first entry in the
3375 i = xfs_btree_firstrec(tcur, level);
3376 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3378 error = xfs_btree_decrement(tcur, level, &i);
3381 i = xfs_btree_firstrec(tcur, level);
3382 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3384 /* Grab a pointer to the block. */
3385 left = xfs_btree_get_block(tcur, level, &lbp);
3387 error = xfs_btree_check_block(cur, left, level, lbp);
3391 /* Grab the current block number, for future use. */
3392 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
3395 * If left block is full enough so that removing one entry
3396 * won't make it too empty, and right-shifting an entry out
3397 * of left to us works, we're done.
3399 if (xfs_btree_get_numrecs(left) - 1 >=
3400 cur->bc_ops->get_minrecs(tcur, level)) {
3401 error = xfs_btree_rshift(tcur, level, &i);
3405 ASSERT(xfs_btree_get_numrecs(block) >=
3406 cur->bc_ops->get_minrecs(tcur, level));
3407 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3411 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3418 * Otherwise, grab the number of records in right for
3421 lrecs = xfs_btree_get_numrecs(left);
3424 /* Delete the temp cursor, we're done with it. */
3425 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3428 /* If here, we need to do a join to keep the tree balanced. */
3429 ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
3431 if (!xfs_btree_ptr_is_null(cur, &lptr) &&
3432 lrecs + xfs_btree_get_numrecs(block) <=
3433 cur->bc_ops->get_maxrecs(cur, level)) {
3435 * Set "right" to be the starting block,
3436 * "left" to be the left neighbor.
3441 error = xfs_btree_read_buf_block(cur, &lptr, level,
3447 * If that won't work, see if we can join with the right neighbor block.
3449 } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
3450 rrecs + xfs_btree_get_numrecs(block) <=
3451 cur->bc_ops->get_maxrecs(cur, level)) {
3453 * Set "left" to be the starting block,
3454 * "right" to be the right neighbor.
3459 error = xfs_btree_read_buf_block(cur, &rptr, level,
3465 * Otherwise, we can't fix the imbalance.
3466 * Just return. This is probably a logic error, but it's not fatal.
3469 error = xfs_btree_dec_cursor(cur, level, stat);
3475 rrecs = xfs_btree_get_numrecs(right);
3476 lrecs = xfs_btree_get_numrecs(left);
3479 * We're now going to join "left" and "right" by moving all the stuff
3480 * in "right" to "left" and deleting "right".
3482 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
3484 /* It's a non-leaf. Move keys and pointers. */
3485 union xfs_btree_key *lkp; /* left btree key */
3486 union xfs_btree_ptr *lpp; /* left address pointer */
3487 union xfs_btree_key *rkp; /* right btree key */
3488 union xfs_btree_ptr *rpp; /* right address pointer */
3490 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
3491 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
3492 rkp = xfs_btree_key_addr(cur, 1, right);
3493 rpp = xfs_btree_ptr_addr(cur, 1, right);
3495 for (i = 1; i < rrecs; i++) {
3496 error = xfs_btree_check_ptr(cur, rpp, i, level);
3501 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
3502 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
3504 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
3505 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
3507 /* It's a leaf. Move records. */
3508 union xfs_btree_rec *lrp; /* left record pointer */
3509 union xfs_btree_rec *rrp; /* right record pointer */
3511 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
3512 rrp = xfs_btree_rec_addr(cur, 1, right);
3514 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
3515 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
3518 XFS_BTREE_STATS_INC(cur, join);
3521 * Fix up the number of records and right block pointer in the
3522 * surviving block, and log it.
3524 xfs_btree_set_numrecs(left, lrecs + rrecs);
3525 xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB),
3526 xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
3527 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
3529 /* If there is a right sibling, point it to the remaining block. */
3530 xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
3531 if (!xfs_btree_ptr_is_null(cur, &cptr)) {
3532 error = xfs_btree_read_buf_block(cur, &cptr, level,
3533 0, &rrblock, &rrbp);
3536 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
3537 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
3540 /* Free the deleted block. */
3541 error = cur->bc_ops->free_block(cur, rbp);
3544 XFS_BTREE_STATS_INC(cur, free);
3547 * If we joined with the left neighbor, set the buffer in the
3548 * cursor to the left block, and fix up the index.
3551 cur->bc_bufs[level] = lbp;
3552 cur->bc_ptrs[level] += lrecs;
3553 cur->bc_ra[level] = 0;
3556 * If we joined with the right neighbor and there's a level above
3557 * us, increment the cursor at that level.
3559 else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
3560 (level + 1 < cur->bc_nlevels)) {
3561 error = xfs_btree_increment(cur, level + 1, &i);
3567 * Readjust the ptr at this level if it's not a leaf, since it's
3568 * still pointing at the deletion point, which makes the cursor
3569 * inconsistent. If this makes the ptr 0, the caller fixes it up.
3570 * We can't use decrement because it would change the next level up.
3573 cur->bc_ptrs[level]--;
3575 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3576 /* Return value means the next level up has something to do. */
3581 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3583 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
3588 * Delete the record pointed to by cur.
3589 * The cursor refers to the place where the record was (could be inserted)
3590 * when the operation returns.
3594 struct xfs_btree_cur *cur,
3595 int *stat) /* success/failure */
3597 int error; /* error return value */
3601 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3604 * Go up the tree, starting at leaf level.
3606 * If 2 is returned then a join was done; go to the next level.
3607 * Otherwise we are done.
3609 for (level = 0, i = 2; i == 2; level++) {
3610 error = xfs_btree_delrec(cur, level, &i);
3616 for (level = 1; level < cur->bc_nlevels; level++) {
3617 if (cur->bc_ptrs[level] == 0) {
3618 error = xfs_btree_decrement(cur, level, &i);
3626 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3630 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3635 * Get the data from the pointed-to record.
3639 struct xfs_btree_cur *cur, /* btree cursor */
3640 union xfs_btree_rec **recp, /* output: btree record */
3641 int *stat) /* output: success/failure */
3643 struct xfs_btree_block *block; /* btree block */
3644 struct xfs_buf *bp; /* buffer pointer */
3645 int ptr; /* record number */
3647 int error; /* error return value */
3650 ptr = cur->bc_ptrs[0];
3651 block = xfs_btree_get_block(cur, 0, &bp);
3654 error = xfs_btree_check_block(cur, block, 0, bp);
3660 * Off the right end or left end, return failure.
3662 if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
3668 * Point to the record and extract its data.
3670 *recp = xfs_btree_rec_addr(cur, ptr, block);