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"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_inode_item.h"
34 #include "xfs_btree.h"
35 #include "xfs_error.h"
36 #include "xfs_trace.h"
39 * Cursor allocation zone.
41 kmem_zone_t *xfs_btree_cur_zone;
44 * Btree magic numbers.
46 const __uint32_t xfs_magics[XFS_BTNUM_MAX] = {
47 XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, XFS_BMAP_MAGIC, XFS_IBT_MAGIC
51 STATIC int /* error (0 or EFSCORRUPTED) */
52 xfs_btree_check_lblock(
53 struct xfs_btree_cur *cur, /* btree cursor */
54 struct xfs_btree_block *block, /* btree long form block pointer */
55 int level, /* level of the btree block */
56 struct xfs_buf *bp) /* buffer for block, if any */
58 int lblock_ok; /* block passes checks */
59 struct xfs_mount *mp; /* file system mount point */
63 be32_to_cpu(block->bb_magic) == xfs_magics[cur->bc_btnum] &&
64 be16_to_cpu(block->bb_level) == level &&
65 be16_to_cpu(block->bb_numrecs) <=
66 cur->bc_ops->get_maxrecs(cur, level) &&
67 block->bb_u.l.bb_leftsib &&
68 (block->bb_u.l.bb_leftsib == cpu_to_be64(NULLDFSBNO) ||
69 XFS_FSB_SANITY_CHECK(mp,
70 be64_to_cpu(block->bb_u.l.bb_leftsib))) &&
71 block->bb_u.l.bb_rightsib &&
72 (block->bb_u.l.bb_rightsib == cpu_to_be64(NULLDFSBNO) ||
73 XFS_FSB_SANITY_CHECK(mp,
74 be64_to_cpu(block->bb_u.l.bb_rightsib)));
75 if (unlikely(XFS_TEST_ERROR(!lblock_ok, mp,
76 XFS_ERRTAG_BTREE_CHECK_LBLOCK,
77 XFS_RANDOM_BTREE_CHECK_LBLOCK))) {
79 trace_xfs_btree_corrupt(bp, _RET_IP_);
80 XFS_ERROR_REPORT("xfs_btree_check_lblock", XFS_ERRLEVEL_LOW,
82 return XFS_ERROR(EFSCORRUPTED);
87 STATIC int /* error (0 or EFSCORRUPTED) */
88 xfs_btree_check_sblock(
89 struct xfs_btree_cur *cur, /* btree cursor */
90 struct xfs_btree_block *block, /* btree short form block pointer */
91 int level, /* level of the btree block */
92 struct xfs_buf *bp) /* buffer containing block */
94 struct xfs_buf *agbp; /* buffer for ag. freespace struct */
95 struct xfs_agf *agf; /* ag. freespace structure */
96 xfs_agblock_t agflen; /* native ag. freespace length */
97 int sblock_ok; /* block passes checks */
99 agbp = cur->bc_private.a.agbp;
100 agf = XFS_BUF_TO_AGF(agbp);
101 agflen = be32_to_cpu(agf->agf_length);
103 be32_to_cpu(block->bb_magic) == xfs_magics[cur->bc_btnum] &&
104 be16_to_cpu(block->bb_level) == level &&
105 be16_to_cpu(block->bb_numrecs) <=
106 cur->bc_ops->get_maxrecs(cur, level) &&
107 (block->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) ||
108 be32_to_cpu(block->bb_u.s.bb_leftsib) < agflen) &&
109 block->bb_u.s.bb_leftsib &&
110 (block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK) ||
111 be32_to_cpu(block->bb_u.s.bb_rightsib) < agflen) &&
112 block->bb_u.s.bb_rightsib;
113 if (unlikely(XFS_TEST_ERROR(!sblock_ok, cur->bc_mp,
114 XFS_ERRTAG_BTREE_CHECK_SBLOCK,
115 XFS_RANDOM_BTREE_CHECK_SBLOCK))) {
117 trace_xfs_btree_corrupt(bp, _RET_IP_);
118 XFS_CORRUPTION_ERROR("xfs_btree_check_sblock",
119 XFS_ERRLEVEL_LOW, cur->bc_mp, block);
120 return XFS_ERROR(EFSCORRUPTED);
126 * Debug routine: check that block header is ok.
129 xfs_btree_check_block(
130 struct xfs_btree_cur *cur, /* btree cursor */
131 struct xfs_btree_block *block, /* generic btree block pointer */
132 int level, /* level of the btree block */
133 struct xfs_buf *bp) /* buffer containing block, if any */
135 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
136 return xfs_btree_check_lblock(cur, block, level, bp);
138 return xfs_btree_check_sblock(cur, block, level, bp);
142 * Check that (long) pointer is ok.
144 int /* error (0 or EFSCORRUPTED) */
145 xfs_btree_check_lptr(
146 struct xfs_btree_cur *cur, /* btree cursor */
147 xfs_dfsbno_t bno, /* btree block disk address */
148 int level) /* btree block level */
150 XFS_WANT_CORRUPTED_RETURN(
153 XFS_FSB_SANITY_CHECK(cur->bc_mp, bno));
159 * Check that (short) pointer is ok.
161 STATIC int /* error (0 or EFSCORRUPTED) */
162 xfs_btree_check_sptr(
163 struct xfs_btree_cur *cur, /* btree cursor */
164 xfs_agblock_t bno, /* btree block disk address */
165 int level) /* btree block level */
167 xfs_agblock_t agblocks = cur->bc_mp->m_sb.sb_agblocks;
169 XFS_WANT_CORRUPTED_RETURN(
171 bno != NULLAGBLOCK &&
178 * Check that block ptr is ok.
180 STATIC int /* error (0 or EFSCORRUPTED) */
182 struct xfs_btree_cur *cur, /* btree cursor */
183 union xfs_btree_ptr *ptr, /* btree block disk address */
184 int index, /* offset from ptr to check */
185 int level) /* btree block level */
187 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
188 return xfs_btree_check_lptr(cur,
189 be64_to_cpu((&ptr->l)[index]), level);
191 return xfs_btree_check_sptr(cur,
192 be32_to_cpu((&ptr->s)[index]), level);
198 * Delete the btree cursor.
201 xfs_btree_del_cursor(
202 xfs_btree_cur_t *cur, /* btree cursor */
203 int error) /* del because of error */
205 int i; /* btree level */
208 * Clear the buffer pointers, and release the buffers.
209 * If we're doing this in the face of an error, we
210 * need to make sure to inspect all of the entries
211 * in the bc_bufs array for buffers to be unlocked.
212 * This is because some of the btree code works from
213 * level n down to 0, and if we get an error along
214 * the way we won't have initialized all the entries
217 for (i = 0; i < cur->bc_nlevels; i++) {
219 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
224 * Can't free a bmap cursor without having dealt with the
225 * allocated indirect blocks' accounting.
227 ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP ||
228 cur->bc_private.b.allocated == 0);
232 kmem_zone_free(xfs_btree_cur_zone, cur);
236 * Duplicate the btree cursor.
237 * Allocate a new one, copy the record, re-get the buffers.
240 xfs_btree_dup_cursor(
241 xfs_btree_cur_t *cur, /* input cursor */
242 xfs_btree_cur_t **ncur) /* output cursor */
244 xfs_buf_t *bp; /* btree block's buffer pointer */
245 int error; /* error return value */
246 int i; /* level number of btree block */
247 xfs_mount_t *mp; /* mount structure for filesystem */
248 xfs_btree_cur_t *new; /* new cursor value */
249 xfs_trans_t *tp; /* transaction pointer, can be NULL */
255 * Allocate a new cursor like the old one.
257 new = cur->bc_ops->dup_cursor(cur);
260 * Copy the record currently in the cursor.
262 new->bc_rec = cur->bc_rec;
265 * For each level current, re-get the buffer and copy the ptr value.
267 for (i = 0; i < new->bc_nlevels; i++) {
268 new->bc_ptrs[i] = cur->bc_ptrs[i];
269 new->bc_ra[i] = cur->bc_ra[i];
270 if ((bp = cur->bc_bufs[i])) {
271 if ((error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
272 XFS_BUF_ADDR(bp), mp->m_bsize, 0, &bp))) {
273 xfs_btree_del_cursor(new, error);
277 new->bc_bufs[i] = bp;
278 ASSERT(!xfs_buf_geterror(bp));
280 new->bc_bufs[i] = NULL;
287 * XFS btree block layout and addressing:
289 * There are two types of blocks in the btree: leaf and non-leaf blocks.
291 * The leaf record start with a header then followed by records containing
292 * the values. A non-leaf block also starts with the same header, and
293 * then first contains lookup keys followed by an equal number of pointers
294 * to the btree blocks at the previous level.
296 * +--------+-------+-------+-------+-------+-------+-------+
297 * Leaf: | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
298 * +--------+-------+-------+-------+-------+-------+-------+
300 * +--------+-------+-------+-------+-------+-------+-------+
301 * Non-Leaf: | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
302 * +--------+-------+-------+-------+-------+-------+-------+
304 * The header is called struct xfs_btree_block for reasons better left unknown
305 * and comes in different versions for short (32bit) and long (64bit) block
306 * pointers. The record and key structures are defined by the btree instances
307 * and opaque to the btree core. The block pointers are simple disk endian
308 * integers, available in a short (32bit) and long (64bit) variant.
310 * The helpers below calculate the offset of a given record, key or pointer
311 * into a btree block (xfs_btree_*_offset) or return a pointer to the given
312 * record, key or pointer (xfs_btree_*_addr). Note that all addressing
313 * inside the btree block is done using indices starting at one, not zero!
317 * Return size of the btree block header for this btree instance.
319 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
321 return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
322 XFS_BTREE_LBLOCK_LEN :
323 XFS_BTREE_SBLOCK_LEN;
327 * Return size of btree block pointers for this btree instance.
329 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
331 return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
332 sizeof(__be64) : sizeof(__be32);
336 * Calculate offset of the n-th record in a btree block.
339 xfs_btree_rec_offset(
340 struct xfs_btree_cur *cur,
343 return xfs_btree_block_len(cur) +
344 (n - 1) * cur->bc_ops->rec_len;
348 * Calculate offset of the n-th key in a btree block.
351 xfs_btree_key_offset(
352 struct xfs_btree_cur *cur,
355 return xfs_btree_block_len(cur) +
356 (n - 1) * cur->bc_ops->key_len;
360 * Calculate offset of the n-th block pointer in a btree block.
363 xfs_btree_ptr_offset(
364 struct xfs_btree_cur *cur,
368 return xfs_btree_block_len(cur) +
369 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
370 (n - 1) * xfs_btree_ptr_len(cur);
374 * Return a pointer to the n-th record in the btree block.
376 STATIC union xfs_btree_rec *
378 struct xfs_btree_cur *cur,
380 struct xfs_btree_block *block)
382 return (union xfs_btree_rec *)
383 ((char *)block + xfs_btree_rec_offset(cur, n));
387 * Return a pointer to the n-th key in the btree block.
389 STATIC union xfs_btree_key *
391 struct xfs_btree_cur *cur,
393 struct xfs_btree_block *block)
395 return (union xfs_btree_key *)
396 ((char *)block + xfs_btree_key_offset(cur, n));
400 * Return a pointer to the n-th block pointer in the btree block.
402 STATIC union xfs_btree_ptr *
404 struct xfs_btree_cur *cur,
406 struct xfs_btree_block *block)
408 int level = xfs_btree_get_level(block);
410 ASSERT(block->bb_level != 0);
412 return (union xfs_btree_ptr *)
413 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
417 * Get a the root block which is stored in the inode.
419 * For now this btree implementation assumes the btree root is always
420 * stored in the if_broot field of an inode fork.
422 STATIC struct xfs_btree_block *
424 struct xfs_btree_cur *cur)
426 struct xfs_ifork *ifp;
428 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip, cur->bc_private.b.whichfork);
429 return (struct xfs_btree_block *)ifp->if_broot;
433 * Retrieve the block pointer from the cursor at the given level.
434 * This may be an inode btree root or from a buffer.
436 STATIC struct xfs_btree_block * /* generic btree block pointer */
438 struct xfs_btree_cur *cur, /* btree cursor */
439 int level, /* level in btree */
440 struct xfs_buf **bpp) /* buffer containing the block */
442 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
443 (level == cur->bc_nlevels - 1)) {
445 return xfs_btree_get_iroot(cur);
448 *bpp = cur->bc_bufs[level];
449 return XFS_BUF_TO_BLOCK(*bpp);
453 * Get a buffer for the block, return it with no data read.
454 * Long-form addressing.
456 xfs_buf_t * /* buffer for fsbno */
458 xfs_mount_t *mp, /* file system mount point */
459 xfs_trans_t *tp, /* transaction pointer */
460 xfs_fsblock_t fsbno, /* file system block number */
461 uint lock) /* lock flags for get_buf */
463 xfs_buf_t *bp; /* buffer pointer (return value) */
464 xfs_daddr_t d; /* real disk block address */
466 ASSERT(fsbno != NULLFSBLOCK);
467 d = XFS_FSB_TO_DADDR(mp, fsbno);
468 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
469 ASSERT(!xfs_buf_geterror(bp));
474 * Get a buffer for the block, return it with no data read.
475 * Short-form addressing.
477 xfs_buf_t * /* buffer for agno/agbno */
479 xfs_mount_t *mp, /* file system mount point */
480 xfs_trans_t *tp, /* transaction pointer */
481 xfs_agnumber_t agno, /* allocation group number */
482 xfs_agblock_t agbno, /* allocation group block number */
483 uint lock) /* lock flags for get_buf */
485 xfs_buf_t *bp; /* buffer pointer (return value) */
486 xfs_daddr_t d; /* real disk block address */
488 ASSERT(agno != NULLAGNUMBER);
489 ASSERT(agbno != NULLAGBLOCK);
490 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
491 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
492 ASSERT(!xfs_buf_geterror(bp));
497 * Check for the cursor referring to the last block at the given level.
499 int /* 1=is last block, 0=not last block */
500 xfs_btree_islastblock(
501 xfs_btree_cur_t *cur, /* btree cursor */
502 int level) /* level to check */
504 struct xfs_btree_block *block; /* generic btree block pointer */
505 xfs_buf_t *bp; /* buffer containing block */
507 block = xfs_btree_get_block(cur, level, &bp);
508 xfs_btree_check_block(cur, block, level, bp);
509 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
510 return block->bb_u.l.bb_rightsib == cpu_to_be64(NULLDFSBNO);
512 return block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK);
516 * Change the cursor to point to the first record at the given level.
517 * Other levels are unaffected.
519 STATIC int /* success=1, failure=0 */
521 xfs_btree_cur_t *cur, /* btree cursor */
522 int level) /* level to change */
524 struct xfs_btree_block *block; /* generic btree block pointer */
525 xfs_buf_t *bp; /* buffer containing block */
528 * Get the block pointer for this level.
530 block = xfs_btree_get_block(cur, level, &bp);
531 xfs_btree_check_block(cur, block, level, bp);
533 * It's empty, there is no such record.
535 if (!block->bb_numrecs)
538 * Set the ptr value to 1, that's the first record/key.
540 cur->bc_ptrs[level] = 1;
545 * Change the cursor to point to the last record in the current block
546 * at the given level. Other levels are unaffected.
548 STATIC int /* success=1, failure=0 */
550 xfs_btree_cur_t *cur, /* btree cursor */
551 int level) /* level to change */
553 struct xfs_btree_block *block; /* generic btree block pointer */
554 xfs_buf_t *bp; /* buffer containing block */
557 * Get the block pointer for this level.
559 block = xfs_btree_get_block(cur, level, &bp);
560 xfs_btree_check_block(cur, block, level, bp);
562 * It's empty, there is no such record.
564 if (!block->bb_numrecs)
567 * Set the ptr value to numrecs, that's the last record/key.
569 cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
574 * Compute first and last byte offsets for the fields given.
575 * Interprets the offsets table, which contains struct field offsets.
579 __int64_t fields, /* bitmask of fields */
580 const short *offsets, /* table of field offsets */
581 int nbits, /* number of bits to inspect */
582 int *first, /* output: first byte offset */
583 int *last) /* output: last byte offset */
585 int i; /* current bit number */
586 __int64_t imask; /* mask for current bit number */
590 * Find the lowest bit, so the first byte offset.
592 for (i = 0, imask = 1LL; ; i++, imask <<= 1) {
593 if (imask & fields) {
599 * Find the highest bit, so the last byte offset.
601 for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) {
602 if (imask & fields) {
603 *last = offsets[i + 1] - 1;
610 * Get a buffer for the block, return it read in.
611 * Long-form addressing.
615 xfs_mount_t *mp, /* file system mount point */
616 xfs_trans_t *tp, /* transaction pointer */
617 xfs_fsblock_t fsbno, /* file system block number */
618 uint lock, /* lock flags for read_buf */
619 xfs_buf_t **bpp, /* buffer for fsbno */
620 int refval) /* ref count value for buffer */
622 xfs_buf_t *bp; /* return value */
623 xfs_daddr_t d; /* real disk block address */
626 ASSERT(fsbno != NULLFSBLOCK);
627 d = XFS_FSB_TO_DADDR(mp, fsbno);
628 if ((error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
629 mp->m_bsize, lock, &bp))) {
632 ASSERT(!xfs_buf_geterror(bp));
634 xfs_buf_set_ref(bp, refval);
640 * Read-ahead the block, don't wait for it, don't return a buffer.
641 * Long-form addressing.
645 xfs_btree_reada_bufl(
646 xfs_mount_t *mp, /* file system mount point */
647 xfs_fsblock_t fsbno, /* file system block number */
648 xfs_extlen_t count) /* count of filesystem blocks */
652 ASSERT(fsbno != NULLFSBLOCK);
653 d = XFS_FSB_TO_DADDR(mp, fsbno);
654 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count);
658 * Read-ahead the block, don't wait for it, don't return a buffer.
659 * Short-form addressing.
663 xfs_btree_reada_bufs(
664 xfs_mount_t *mp, /* file system mount point */
665 xfs_agnumber_t agno, /* allocation group number */
666 xfs_agblock_t agbno, /* allocation group block number */
667 xfs_extlen_t count) /* count of filesystem blocks */
671 ASSERT(agno != NULLAGNUMBER);
672 ASSERT(agbno != NULLAGBLOCK);
673 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
674 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count);
678 xfs_btree_readahead_lblock(
679 struct xfs_btree_cur *cur,
681 struct xfs_btree_block *block)
684 xfs_dfsbno_t left = be64_to_cpu(block->bb_u.l.bb_leftsib);
685 xfs_dfsbno_t right = be64_to_cpu(block->bb_u.l.bb_rightsib);
687 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLDFSBNO) {
688 xfs_btree_reada_bufl(cur->bc_mp, left, 1);
692 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLDFSBNO) {
693 xfs_btree_reada_bufl(cur->bc_mp, right, 1);
701 xfs_btree_readahead_sblock(
702 struct xfs_btree_cur *cur,
704 struct xfs_btree_block *block)
707 xfs_agblock_t left = be32_to_cpu(block->bb_u.s.bb_leftsib);
708 xfs_agblock_t right = be32_to_cpu(block->bb_u.s.bb_rightsib);
711 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
712 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
717 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
718 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
727 * Read-ahead btree blocks, at the given level.
728 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
732 struct xfs_btree_cur *cur, /* btree cursor */
733 int lev, /* level in btree */
734 int lr) /* left/right bits */
736 struct xfs_btree_block *block;
739 * No readahead needed if we are at the root level and the
740 * btree root is stored in the inode.
742 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
743 (lev == cur->bc_nlevels - 1))
746 if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
749 cur->bc_ra[lev] |= lr;
750 block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
752 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
753 return xfs_btree_readahead_lblock(cur, lr, block);
754 return xfs_btree_readahead_sblock(cur, lr, block);
758 * Set the buffer for level "lev" in the cursor to bp, releasing
759 * any previous buffer.
763 xfs_btree_cur_t *cur, /* btree cursor */
764 int lev, /* level in btree */
765 xfs_buf_t *bp) /* new buffer to set */
767 struct xfs_btree_block *b; /* btree block */
769 if (cur->bc_bufs[lev])
770 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
771 cur->bc_bufs[lev] = bp;
774 b = XFS_BUF_TO_BLOCK(bp);
775 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
776 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLDFSBNO))
777 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
778 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLDFSBNO))
779 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
781 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
782 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
783 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
784 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
789 xfs_btree_ptr_is_null(
790 struct xfs_btree_cur *cur,
791 union xfs_btree_ptr *ptr)
793 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
794 return ptr->l == cpu_to_be64(NULLDFSBNO);
796 return ptr->s == cpu_to_be32(NULLAGBLOCK);
800 xfs_btree_set_ptr_null(
801 struct xfs_btree_cur *cur,
802 union xfs_btree_ptr *ptr)
804 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
805 ptr->l = cpu_to_be64(NULLDFSBNO);
807 ptr->s = cpu_to_be32(NULLAGBLOCK);
811 * Get/set/init sibling pointers
814 xfs_btree_get_sibling(
815 struct xfs_btree_cur *cur,
816 struct xfs_btree_block *block,
817 union xfs_btree_ptr *ptr,
820 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
822 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
823 if (lr == XFS_BB_RIGHTSIB)
824 ptr->l = block->bb_u.l.bb_rightsib;
826 ptr->l = block->bb_u.l.bb_leftsib;
828 if (lr == XFS_BB_RIGHTSIB)
829 ptr->s = block->bb_u.s.bb_rightsib;
831 ptr->s = block->bb_u.s.bb_leftsib;
836 xfs_btree_set_sibling(
837 struct xfs_btree_cur *cur,
838 struct xfs_btree_block *block,
839 union xfs_btree_ptr *ptr,
842 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
844 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
845 if (lr == XFS_BB_RIGHTSIB)
846 block->bb_u.l.bb_rightsib = ptr->l;
848 block->bb_u.l.bb_leftsib = ptr->l;
850 if (lr == XFS_BB_RIGHTSIB)
851 block->bb_u.s.bb_rightsib = ptr->s;
853 block->bb_u.s.bb_leftsib = ptr->s;
858 xfs_btree_init_block(
859 struct xfs_btree_cur *cur,
862 struct xfs_btree_block *new) /* new block */
864 new->bb_magic = cpu_to_be32(xfs_magics[cur->bc_btnum]);
865 new->bb_level = cpu_to_be16(level);
866 new->bb_numrecs = cpu_to_be16(numrecs);
868 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
869 new->bb_u.l.bb_leftsib = cpu_to_be64(NULLDFSBNO);
870 new->bb_u.l.bb_rightsib = cpu_to_be64(NULLDFSBNO);
872 new->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
873 new->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
878 * Return true if ptr is the last record in the btree and
879 * we need to track updateѕ to this record. The decision
880 * will be further refined in the update_lastrec method.
883 xfs_btree_is_lastrec(
884 struct xfs_btree_cur *cur,
885 struct xfs_btree_block *block,
888 union xfs_btree_ptr ptr;
892 if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
895 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
896 if (!xfs_btree_ptr_is_null(cur, &ptr))
902 xfs_btree_buf_to_ptr(
903 struct xfs_btree_cur *cur,
905 union xfs_btree_ptr *ptr)
907 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
908 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
911 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
917 xfs_btree_ptr_to_daddr(
918 struct xfs_btree_cur *cur,
919 union xfs_btree_ptr *ptr)
921 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
922 ASSERT(ptr->l != cpu_to_be64(NULLDFSBNO));
924 return XFS_FSB_TO_DADDR(cur->bc_mp, be64_to_cpu(ptr->l));
926 ASSERT(cur->bc_private.a.agno != NULLAGNUMBER);
927 ASSERT(ptr->s != cpu_to_be32(NULLAGBLOCK));
929 return XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_private.a.agno,
930 be32_to_cpu(ptr->s));
936 struct xfs_btree_cur *cur,
939 switch (cur->bc_btnum) {
942 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
945 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
948 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
956 xfs_btree_get_buf_block(
957 struct xfs_btree_cur *cur,
958 union xfs_btree_ptr *ptr,
960 struct xfs_btree_block **block,
961 struct xfs_buf **bpp)
963 struct xfs_mount *mp = cur->bc_mp;
966 /* need to sort out how callers deal with failures first */
967 ASSERT(!(flags & XBF_TRYLOCK));
969 d = xfs_btree_ptr_to_daddr(cur, ptr);
970 *bpp = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d,
976 *block = XFS_BUF_TO_BLOCK(*bpp);
981 * Read in the buffer at the given ptr and return the buffer and
982 * the block pointer within the buffer.
985 xfs_btree_read_buf_block(
986 struct xfs_btree_cur *cur,
987 union xfs_btree_ptr *ptr,
990 struct xfs_btree_block **block,
991 struct xfs_buf **bpp)
993 struct xfs_mount *mp = cur->bc_mp;
997 /* need to sort out how callers deal with failures first */
998 ASSERT(!(flags & XBF_TRYLOCK));
1000 d = xfs_btree_ptr_to_daddr(cur, ptr);
1001 error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
1002 mp->m_bsize, flags, bpp);
1006 ASSERT(!xfs_buf_geterror(*bpp));
1008 xfs_btree_set_refs(cur, *bpp);
1009 *block = XFS_BUF_TO_BLOCK(*bpp);
1011 error = xfs_btree_check_block(cur, *block, level, *bpp);
1013 xfs_trans_brelse(cur->bc_tp, *bpp);
1018 * Copy keys from one btree block to another.
1021 xfs_btree_copy_keys(
1022 struct xfs_btree_cur *cur,
1023 union xfs_btree_key *dst_key,
1024 union xfs_btree_key *src_key,
1027 ASSERT(numkeys >= 0);
1028 memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1032 * Copy records from one btree block to another.
1035 xfs_btree_copy_recs(
1036 struct xfs_btree_cur *cur,
1037 union xfs_btree_rec *dst_rec,
1038 union xfs_btree_rec *src_rec,
1041 ASSERT(numrecs >= 0);
1042 memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1046 * Copy block pointers from one btree block to another.
1049 xfs_btree_copy_ptrs(
1050 struct xfs_btree_cur *cur,
1051 union xfs_btree_ptr *dst_ptr,
1052 union xfs_btree_ptr *src_ptr,
1055 ASSERT(numptrs >= 0);
1056 memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1060 * Shift keys one index left/right inside a single btree block.
1063 xfs_btree_shift_keys(
1064 struct xfs_btree_cur *cur,
1065 union xfs_btree_key *key,
1071 ASSERT(numkeys >= 0);
1072 ASSERT(dir == 1 || dir == -1);
1074 dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1075 memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1079 * Shift records one index left/right inside a single btree block.
1082 xfs_btree_shift_recs(
1083 struct xfs_btree_cur *cur,
1084 union xfs_btree_rec *rec,
1090 ASSERT(numrecs >= 0);
1091 ASSERT(dir == 1 || dir == -1);
1093 dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1094 memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1098 * Shift block pointers one index left/right inside a single btree block.
1101 xfs_btree_shift_ptrs(
1102 struct xfs_btree_cur *cur,
1103 union xfs_btree_ptr *ptr,
1109 ASSERT(numptrs >= 0);
1110 ASSERT(dir == 1 || dir == -1);
1112 dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1113 memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1117 * Log key values from the btree block.
1121 struct xfs_btree_cur *cur,
1126 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1127 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1130 xfs_trans_log_buf(cur->bc_tp, bp,
1131 xfs_btree_key_offset(cur, first),
1132 xfs_btree_key_offset(cur, last + 1) - 1);
1134 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1135 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1138 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1142 * Log record values from the btree block.
1146 struct xfs_btree_cur *cur,
1151 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1152 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1154 xfs_trans_log_buf(cur->bc_tp, bp,
1155 xfs_btree_rec_offset(cur, first),
1156 xfs_btree_rec_offset(cur, last + 1) - 1);
1158 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1162 * Log block pointer fields from a btree block (nonleaf).
1166 struct xfs_btree_cur *cur, /* btree cursor */
1167 struct xfs_buf *bp, /* buffer containing btree block */
1168 int first, /* index of first pointer to log */
1169 int last) /* index of last pointer to log */
1171 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1172 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1175 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
1176 int level = xfs_btree_get_level(block);
1178 xfs_trans_log_buf(cur->bc_tp, bp,
1179 xfs_btree_ptr_offset(cur, first, level),
1180 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1182 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1183 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1186 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1190 * Log fields from a btree block header.
1193 xfs_btree_log_block(
1194 struct xfs_btree_cur *cur, /* btree cursor */
1195 struct xfs_buf *bp, /* buffer containing btree block */
1196 int fields) /* mask of fields: XFS_BB_... */
1198 int first; /* first byte offset logged */
1199 int last; /* last byte offset logged */
1200 static const short soffsets[] = { /* table of offsets (short) */
1201 offsetof(struct xfs_btree_block, bb_magic),
1202 offsetof(struct xfs_btree_block, bb_level),
1203 offsetof(struct xfs_btree_block, bb_numrecs),
1204 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1205 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1206 XFS_BTREE_SBLOCK_LEN
1208 static const short loffsets[] = { /* table of offsets (long) */
1209 offsetof(struct xfs_btree_block, bb_magic),
1210 offsetof(struct xfs_btree_block, bb_level),
1211 offsetof(struct xfs_btree_block, bb_numrecs),
1212 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1213 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1214 XFS_BTREE_LBLOCK_LEN
1217 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1218 XFS_BTREE_TRACE_ARGBI(cur, bp, fields);
1221 xfs_btree_offsets(fields,
1222 (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1223 loffsets : soffsets,
1224 XFS_BB_NUM_BITS, &first, &last);
1225 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1227 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1228 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1231 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1235 * Increment cursor by one record at the level.
1236 * For nonzero levels the leaf-ward information is untouched.
1239 xfs_btree_increment(
1240 struct xfs_btree_cur *cur,
1242 int *stat) /* success/failure */
1244 struct xfs_btree_block *block;
1245 union xfs_btree_ptr ptr;
1247 int error; /* error return value */
1250 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1251 XFS_BTREE_TRACE_ARGI(cur, level);
1253 ASSERT(level < cur->bc_nlevels);
1255 /* Read-ahead to the right at this level. */
1256 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1258 /* Get a pointer to the btree block. */
1259 block = xfs_btree_get_block(cur, level, &bp);
1262 error = xfs_btree_check_block(cur, block, level, bp);
1267 /* We're done if we remain in the block after the increment. */
1268 if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
1271 /* Fail if we just went off the right edge of the tree. */
1272 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1273 if (xfs_btree_ptr_is_null(cur, &ptr))
1276 XFS_BTREE_STATS_INC(cur, increment);
1279 * March up the tree incrementing pointers.
1280 * Stop when we don't go off the right edge of a block.
1282 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1283 block = xfs_btree_get_block(cur, lev, &bp);
1286 error = xfs_btree_check_block(cur, block, lev, bp);
1291 if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
1294 /* Read-ahead the right block for the next loop. */
1295 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1299 * If we went off the root then we are either seriously
1300 * confused or have the tree root in an inode.
1302 if (lev == cur->bc_nlevels) {
1303 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1306 error = EFSCORRUPTED;
1309 ASSERT(lev < cur->bc_nlevels);
1312 * Now walk back down the tree, fixing up the cursor's buffer
1313 * pointers and key numbers.
1315 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1316 union xfs_btree_ptr *ptrp;
1318 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1319 error = xfs_btree_read_buf_block(cur, ptrp, --lev,
1324 xfs_btree_setbuf(cur, lev, bp);
1325 cur->bc_ptrs[lev] = 1;
1328 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1333 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1338 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1343 * Decrement cursor by one record at the level.
1344 * For nonzero levels the leaf-ward information is untouched.
1347 xfs_btree_decrement(
1348 struct xfs_btree_cur *cur,
1350 int *stat) /* success/failure */
1352 struct xfs_btree_block *block;
1354 int error; /* error return value */
1356 union xfs_btree_ptr ptr;
1358 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1359 XFS_BTREE_TRACE_ARGI(cur, level);
1361 ASSERT(level < cur->bc_nlevels);
1363 /* Read-ahead to the left at this level. */
1364 xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1366 /* We're done if we remain in the block after the decrement. */
1367 if (--cur->bc_ptrs[level] > 0)
1370 /* Get a pointer to the btree block. */
1371 block = xfs_btree_get_block(cur, level, &bp);
1374 error = xfs_btree_check_block(cur, block, level, bp);
1379 /* Fail if we just went off the left edge of the tree. */
1380 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1381 if (xfs_btree_ptr_is_null(cur, &ptr))
1384 XFS_BTREE_STATS_INC(cur, decrement);
1387 * March up the tree decrementing pointers.
1388 * Stop when we don't go off the left edge of a block.
1390 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1391 if (--cur->bc_ptrs[lev] > 0)
1393 /* Read-ahead the left block for the next loop. */
1394 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1398 * If we went off the root then we are seriously confused.
1399 * or the root of the tree is in an inode.
1401 if (lev == cur->bc_nlevels) {
1402 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1405 error = EFSCORRUPTED;
1408 ASSERT(lev < cur->bc_nlevels);
1411 * Now walk back down the tree, fixing up the cursor's buffer
1412 * pointers and key numbers.
1414 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1415 union xfs_btree_ptr *ptrp;
1417 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1418 error = xfs_btree_read_buf_block(cur, ptrp, --lev,
1422 xfs_btree_setbuf(cur, lev, bp);
1423 cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
1426 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1431 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1436 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1441 xfs_btree_lookup_get_block(
1442 struct xfs_btree_cur *cur, /* btree cursor */
1443 int level, /* level in the btree */
1444 union xfs_btree_ptr *pp, /* ptr to btree block */
1445 struct xfs_btree_block **blkp) /* return btree block */
1447 struct xfs_buf *bp; /* buffer pointer for btree block */
1450 /* special case the root block if in an inode */
1451 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1452 (level == cur->bc_nlevels - 1)) {
1453 *blkp = xfs_btree_get_iroot(cur);
1458 * If the old buffer at this level for the disk address we are
1459 * looking for re-use it.
1461 * Otherwise throw it away and get a new one.
1463 bp = cur->bc_bufs[level];
1464 if (bp && XFS_BUF_ADDR(bp) == xfs_btree_ptr_to_daddr(cur, pp)) {
1465 *blkp = XFS_BUF_TO_BLOCK(bp);
1469 error = xfs_btree_read_buf_block(cur, pp, level, 0, blkp, &bp);
1473 xfs_btree_setbuf(cur, level, bp);
1478 * Get current search key. For level 0 we don't actually have a key
1479 * structure so we make one up from the record. For all other levels
1480 * we just return the right key.
1482 STATIC union xfs_btree_key *
1483 xfs_lookup_get_search_key(
1484 struct xfs_btree_cur *cur,
1487 struct xfs_btree_block *block,
1488 union xfs_btree_key *kp)
1491 cur->bc_ops->init_key_from_rec(kp,
1492 xfs_btree_rec_addr(cur, keyno, block));
1496 return xfs_btree_key_addr(cur, keyno, block);
1500 * Lookup the record. The cursor is made to point to it, based on dir.
1501 * Return 0 if can't find any such record, 1 for success.
1505 struct xfs_btree_cur *cur, /* btree cursor */
1506 xfs_lookup_t dir, /* <=, ==, or >= */
1507 int *stat) /* success/failure */
1509 struct xfs_btree_block *block; /* current btree block */
1510 __int64_t diff; /* difference for the current key */
1511 int error; /* error return value */
1512 int keyno; /* current key number */
1513 int level; /* level in the btree */
1514 union xfs_btree_ptr *pp; /* ptr to btree block */
1515 union xfs_btree_ptr ptr; /* ptr to btree block */
1517 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1518 XFS_BTREE_TRACE_ARGI(cur, dir);
1520 XFS_BTREE_STATS_INC(cur, lookup);
1525 /* initialise start pointer from cursor */
1526 cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1530 * Iterate over each level in the btree, starting at the root.
1531 * For each level above the leaves, find the key we need, based
1532 * on the lookup record, then follow the corresponding block
1533 * pointer down to the next level.
1535 for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1536 /* Get the block we need to do the lookup on. */
1537 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1543 * If we already had a key match at a higher level, we
1544 * know we need to use the first entry in this block.
1548 /* Otherwise search this block. Do a binary search. */
1550 int high; /* high entry number */
1551 int low; /* low entry number */
1553 /* Set low and high entry numbers, 1-based. */
1555 high = xfs_btree_get_numrecs(block);
1557 /* Block is empty, must be an empty leaf. */
1558 ASSERT(level == 0 && cur->bc_nlevels == 1);
1560 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1561 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1566 /* Binary search the block. */
1567 while (low <= high) {
1568 union xfs_btree_key key;
1569 union xfs_btree_key *kp;
1571 XFS_BTREE_STATS_INC(cur, compare);
1573 /* keyno is average of low and high. */
1574 keyno = (low + high) >> 1;
1576 /* Get current search key */
1577 kp = xfs_lookup_get_search_key(cur, level,
1578 keyno, block, &key);
1581 * Compute difference to get next direction:
1582 * - less than, move right
1583 * - greater than, move left
1584 * - equal, we're done
1586 diff = cur->bc_ops->key_diff(cur, kp);
1597 * If there are more levels, set up for the next level
1598 * by getting the block number and filling in the cursor.
1602 * If we moved left, need the previous key number,
1603 * unless there isn't one.
1605 if (diff > 0 && --keyno < 1)
1607 pp = xfs_btree_ptr_addr(cur, keyno, block);
1610 error = xfs_btree_check_ptr(cur, pp, 0, level);
1614 cur->bc_ptrs[level] = keyno;
1618 /* Done with the search. See if we need to adjust the results. */
1619 if (dir != XFS_LOOKUP_LE && diff < 0) {
1622 * If ge search and we went off the end of the block, but it's
1623 * not the last block, we're in the wrong block.
1625 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1626 if (dir == XFS_LOOKUP_GE &&
1627 keyno > xfs_btree_get_numrecs(block) &&
1628 !xfs_btree_ptr_is_null(cur, &ptr)) {
1631 cur->bc_ptrs[0] = keyno;
1632 error = xfs_btree_increment(cur, 0, &i);
1635 XFS_WANT_CORRUPTED_RETURN(i == 1);
1636 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1640 } else if (dir == XFS_LOOKUP_LE && diff > 0)
1642 cur->bc_ptrs[0] = keyno;
1644 /* Return if we succeeded or not. */
1645 if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
1647 else if (dir != XFS_LOOKUP_EQ || diff == 0)
1651 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1655 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1660 * Update keys at all levels from here to the root along the cursor's path.
1664 struct xfs_btree_cur *cur,
1665 union xfs_btree_key *keyp,
1668 struct xfs_btree_block *block;
1670 union xfs_btree_key *kp;
1673 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1674 XFS_BTREE_TRACE_ARGIK(cur, level, keyp);
1676 ASSERT(!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) || level >= 1);
1679 * Go up the tree from this level toward the root.
1680 * At each level, update the key value to the value input.
1681 * Stop when we reach a level where the cursor isn't pointing
1682 * at the first entry in the block.
1684 for (ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
1688 block = xfs_btree_get_block(cur, level, &bp);
1690 error = xfs_btree_check_block(cur, block, level, bp);
1692 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1696 ptr = cur->bc_ptrs[level];
1697 kp = xfs_btree_key_addr(cur, ptr, block);
1698 xfs_btree_copy_keys(cur, kp, keyp, 1);
1699 xfs_btree_log_keys(cur, bp, ptr, ptr);
1702 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1707 * Update the record referred to by cur to the value in the
1708 * given record. This either works (return 0) or gets an
1709 * EFSCORRUPTED error.
1713 struct xfs_btree_cur *cur,
1714 union xfs_btree_rec *rec)
1716 struct xfs_btree_block *block;
1720 union xfs_btree_rec *rp;
1722 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1723 XFS_BTREE_TRACE_ARGR(cur, rec);
1725 /* Pick up the current block. */
1726 block = xfs_btree_get_block(cur, 0, &bp);
1729 error = xfs_btree_check_block(cur, block, 0, bp);
1733 /* Get the address of the rec to be updated. */
1734 ptr = cur->bc_ptrs[0];
1735 rp = xfs_btree_rec_addr(cur, ptr, block);
1737 /* Fill in the new contents and log them. */
1738 xfs_btree_copy_recs(cur, rp, rec, 1);
1739 xfs_btree_log_recs(cur, bp, ptr, ptr);
1742 * If we are tracking the last record in the tree and
1743 * we are at the far right edge of the tree, update it.
1745 if (xfs_btree_is_lastrec(cur, block, 0)) {
1746 cur->bc_ops->update_lastrec(cur, block, rec,
1747 ptr, LASTREC_UPDATE);
1750 /* Updating first rec in leaf. Pass new key value up to our parent. */
1752 union xfs_btree_key key;
1754 cur->bc_ops->init_key_from_rec(&key, rec);
1755 error = xfs_btree_updkey(cur, &key, 1);
1760 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1764 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1769 * Move 1 record left from cur/level if possible.
1770 * Update cur to reflect the new path.
1772 STATIC int /* error */
1774 struct xfs_btree_cur *cur,
1776 int *stat) /* success/failure */
1778 union xfs_btree_key key; /* btree key */
1779 struct xfs_buf *lbp; /* left buffer pointer */
1780 struct xfs_btree_block *left; /* left btree block */
1781 int lrecs; /* left record count */
1782 struct xfs_buf *rbp; /* right buffer pointer */
1783 struct xfs_btree_block *right; /* right btree block */
1784 int rrecs; /* right record count */
1785 union xfs_btree_ptr lptr; /* left btree pointer */
1786 union xfs_btree_key *rkp = NULL; /* right btree key */
1787 union xfs_btree_ptr *rpp = NULL; /* right address pointer */
1788 union xfs_btree_rec *rrp = NULL; /* right record pointer */
1789 int error; /* error return value */
1791 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1792 XFS_BTREE_TRACE_ARGI(cur, level);
1794 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1795 level == cur->bc_nlevels - 1)
1798 /* Set up variables for this block as "right". */
1799 right = xfs_btree_get_block(cur, level, &rbp);
1802 error = xfs_btree_check_block(cur, right, level, rbp);
1807 /* If we've got no left sibling then we can't shift an entry left. */
1808 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
1809 if (xfs_btree_ptr_is_null(cur, &lptr))
1813 * If the cursor entry is the one that would be moved, don't
1814 * do it... it's too complicated.
1816 if (cur->bc_ptrs[level] <= 1)
1819 /* Set up the left neighbor as "left". */
1820 error = xfs_btree_read_buf_block(cur, &lptr, level, 0, &left, &lbp);
1824 /* If it's full, it can't take another entry. */
1825 lrecs = xfs_btree_get_numrecs(left);
1826 if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
1829 rrecs = xfs_btree_get_numrecs(right);
1832 * We add one entry to the left side and remove one for the right side.
1833 * Account for it here, the changes will be updated on disk and logged
1839 XFS_BTREE_STATS_INC(cur, lshift);
1840 XFS_BTREE_STATS_ADD(cur, moves, 1);
1843 * If non-leaf, copy a key and a ptr to the left block.
1844 * Log the changes to the left block.
1847 /* It's a non-leaf. Move keys and pointers. */
1848 union xfs_btree_key *lkp; /* left btree key */
1849 union xfs_btree_ptr *lpp; /* left address pointer */
1851 lkp = xfs_btree_key_addr(cur, lrecs, left);
1852 rkp = xfs_btree_key_addr(cur, 1, right);
1854 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
1855 rpp = xfs_btree_ptr_addr(cur, 1, right);
1857 error = xfs_btree_check_ptr(cur, rpp, 0, level);
1861 xfs_btree_copy_keys(cur, lkp, rkp, 1);
1862 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
1864 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
1865 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
1867 ASSERT(cur->bc_ops->keys_inorder(cur,
1868 xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
1870 /* It's a leaf. Move records. */
1871 union xfs_btree_rec *lrp; /* left record pointer */
1873 lrp = xfs_btree_rec_addr(cur, lrecs, left);
1874 rrp = xfs_btree_rec_addr(cur, 1, right);
1876 xfs_btree_copy_recs(cur, lrp, rrp, 1);
1877 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
1879 ASSERT(cur->bc_ops->recs_inorder(cur,
1880 xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
1883 xfs_btree_set_numrecs(left, lrecs);
1884 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
1886 xfs_btree_set_numrecs(right, rrecs);
1887 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
1890 * Slide the contents of right down one entry.
1892 XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
1894 /* It's a nonleaf. operate on keys and ptrs */
1896 int i; /* loop index */
1898 for (i = 0; i < rrecs; i++) {
1899 error = xfs_btree_check_ptr(cur, rpp, i + 1, level);
1904 xfs_btree_shift_keys(cur,
1905 xfs_btree_key_addr(cur, 2, right),
1907 xfs_btree_shift_ptrs(cur,
1908 xfs_btree_ptr_addr(cur, 2, right),
1911 xfs_btree_log_keys(cur, rbp, 1, rrecs);
1912 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
1914 /* It's a leaf. operate on records */
1915 xfs_btree_shift_recs(cur,
1916 xfs_btree_rec_addr(cur, 2, right),
1918 xfs_btree_log_recs(cur, rbp, 1, rrecs);
1921 * If it's the first record in the block, we'll need a key
1922 * structure to pass up to the next level (updkey).
1924 cur->bc_ops->init_key_from_rec(&key,
1925 xfs_btree_rec_addr(cur, 1, right));
1929 /* Update the parent key values of right. */
1930 error = xfs_btree_updkey(cur, rkp, level + 1);
1934 /* Slide the cursor value left one. */
1935 cur->bc_ptrs[level]--;
1937 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1942 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1947 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1952 * Move 1 record right from cur/level if possible.
1953 * Update cur to reflect the new path.
1955 STATIC int /* error */
1957 struct xfs_btree_cur *cur,
1959 int *stat) /* success/failure */
1961 union xfs_btree_key key; /* btree key */
1962 struct xfs_buf *lbp; /* left buffer pointer */
1963 struct xfs_btree_block *left; /* left btree block */
1964 struct xfs_buf *rbp; /* right buffer pointer */
1965 struct xfs_btree_block *right; /* right btree block */
1966 struct xfs_btree_cur *tcur; /* temporary btree cursor */
1967 union xfs_btree_ptr rptr; /* right block pointer */
1968 union xfs_btree_key *rkp; /* right btree key */
1969 int rrecs; /* right record count */
1970 int lrecs; /* left record count */
1971 int error; /* error return value */
1972 int i; /* loop counter */
1974 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1975 XFS_BTREE_TRACE_ARGI(cur, level);
1977 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1978 (level == cur->bc_nlevels - 1))
1981 /* Set up variables for this block as "left". */
1982 left = xfs_btree_get_block(cur, level, &lbp);
1985 error = xfs_btree_check_block(cur, left, level, lbp);
1990 /* If we've got no right sibling then we can't shift an entry right. */
1991 xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
1992 if (xfs_btree_ptr_is_null(cur, &rptr))
1996 * If the cursor entry is the one that would be moved, don't
1997 * do it... it's too complicated.
1999 lrecs = xfs_btree_get_numrecs(left);
2000 if (cur->bc_ptrs[level] >= lrecs)
2003 /* Set up the right neighbor as "right". */
2004 error = xfs_btree_read_buf_block(cur, &rptr, level, 0, &right, &rbp);
2008 /* If it's full, it can't take another entry. */
2009 rrecs = xfs_btree_get_numrecs(right);
2010 if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2013 XFS_BTREE_STATS_INC(cur, rshift);
2014 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2017 * Make a hole at the start of the right neighbor block, then
2018 * copy the last left block entry to the hole.
2021 /* It's a nonleaf. make a hole in the keys and ptrs */
2022 union xfs_btree_key *lkp;
2023 union xfs_btree_ptr *lpp;
2024 union xfs_btree_ptr *rpp;
2026 lkp = xfs_btree_key_addr(cur, lrecs, left);
2027 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2028 rkp = xfs_btree_key_addr(cur, 1, right);
2029 rpp = xfs_btree_ptr_addr(cur, 1, right);
2032 for (i = rrecs - 1; i >= 0; i--) {
2033 error = xfs_btree_check_ptr(cur, rpp, i, level);
2039 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2040 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2043 error = xfs_btree_check_ptr(cur, lpp, 0, level);
2048 /* Now put the new data in, and log it. */
2049 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2050 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2052 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2053 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2055 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2056 xfs_btree_key_addr(cur, 2, right)));
2058 /* It's a leaf. make a hole in the records */
2059 union xfs_btree_rec *lrp;
2060 union xfs_btree_rec *rrp;
2062 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2063 rrp = xfs_btree_rec_addr(cur, 1, right);
2065 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2067 /* Now put the new data in, and log it. */
2068 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2069 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2071 cur->bc_ops->init_key_from_rec(&key, rrp);
2074 ASSERT(cur->bc_ops->recs_inorder(cur, rrp,
2075 xfs_btree_rec_addr(cur, 2, right)));
2079 * Decrement and log left's numrecs, bump and log right's numrecs.
2081 xfs_btree_set_numrecs(left, --lrecs);
2082 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2084 xfs_btree_set_numrecs(right, ++rrecs);
2085 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2088 * Using a temporary cursor, update the parent key values of the
2089 * block on the right.
2091 error = xfs_btree_dup_cursor(cur, &tcur);
2094 i = xfs_btree_lastrec(tcur, level);
2095 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
2097 error = xfs_btree_increment(tcur, level, &i);
2101 error = xfs_btree_updkey(tcur, rkp, level + 1);
2105 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2107 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2112 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2117 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2121 XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2122 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2127 * Split cur/level block in half.
2128 * Return new block number and the key to its first
2129 * record (to be inserted into parent).
2131 STATIC int /* error */
2133 struct xfs_btree_cur *cur,
2135 union xfs_btree_ptr *ptrp,
2136 union xfs_btree_key *key,
2137 struct xfs_btree_cur **curp,
2138 int *stat) /* success/failure */
2140 union xfs_btree_ptr lptr; /* left sibling block ptr */
2141 struct xfs_buf *lbp; /* left buffer pointer */
2142 struct xfs_btree_block *left; /* left btree block */
2143 union xfs_btree_ptr rptr; /* right sibling block ptr */
2144 struct xfs_buf *rbp; /* right buffer pointer */
2145 struct xfs_btree_block *right; /* right btree block */
2146 union xfs_btree_ptr rrptr; /* right-right sibling ptr */
2147 struct xfs_buf *rrbp; /* right-right buffer pointer */
2148 struct xfs_btree_block *rrblock; /* right-right btree block */
2152 int error; /* error return value */
2157 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2158 XFS_BTREE_TRACE_ARGIPK(cur, level, *ptrp, key);
2160 XFS_BTREE_STATS_INC(cur, split);
2162 /* Set up left block (current one). */
2163 left = xfs_btree_get_block(cur, level, &lbp);
2166 error = xfs_btree_check_block(cur, left, level, lbp);
2171 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2173 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2174 error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, 1, stat);
2179 XFS_BTREE_STATS_INC(cur, alloc);
2181 /* Set up the new block as "right". */
2182 error = xfs_btree_get_buf_block(cur, &rptr, 0, &right, &rbp);
2186 /* Fill in the btree header for the new right block. */
2187 xfs_btree_init_block(cur, xfs_btree_get_level(left), 0, right);
2190 * Split the entries between the old and the new block evenly.
2191 * Make sure that if there's an odd number of entries now, that
2192 * each new block will have the same number of entries.
2194 lrecs = xfs_btree_get_numrecs(left);
2196 if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
2198 src_index = (lrecs - rrecs + 1);
2200 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2203 * Copy btree block entries from the left block over to the
2204 * new block, the right. Update the right block and log the
2208 /* It's a non-leaf. Move keys and pointers. */
2209 union xfs_btree_key *lkp; /* left btree key */
2210 union xfs_btree_ptr *lpp; /* left address pointer */
2211 union xfs_btree_key *rkp; /* right btree key */
2212 union xfs_btree_ptr *rpp; /* right address pointer */
2214 lkp = xfs_btree_key_addr(cur, src_index, left);
2215 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2216 rkp = xfs_btree_key_addr(cur, 1, right);
2217 rpp = xfs_btree_ptr_addr(cur, 1, right);
2220 for (i = src_index; i < rrecs; i++) {
2221 error = xfs_btree_check_ptr(cur, lpp, i, level);
2227 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2228 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2230 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2231 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2233 /* Grab the keys to the entries moved to the right block */
2234 xfs_btree_copy_keys(cur, key, rkp, 1);
2236 /* It's a leaf. Move records. */
2237 union xfs_btree_rec *lrp; /* left record pointer */
2238 union xfs_btree_rec *rrp; /* right record pointer */
2240 lrp = xfs_btree_rec_addr(cur, src_index, left);
2241 rrp = xfs_btree_rec_addr(cur, 1, right);
2243 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2244 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2246 cur->bc_ops->init_key_from_rec(key,
2247 xfs_btree_rec_addr(cur, 1, right));
2252 * Find the left block number by looking in the buffer.
2253 * Adjust numrecs, sibling pointers.
2255 xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2256 xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2257 xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2258 xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2261 xfs_btree_set_numrecs(left, lrecs);
2262 xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2264 xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2265 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2268 * If there's a block to the new block's right, make that block
2269 * point back to right instead of to left.
2271 if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2272 error = xfs_btree_read_buf_block(cur, &rrptr, level,
2273 0, &rrblock, &rrbp);
2276 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2277 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2280 * If the cursor is really in the right block, move it there.
2281 * If it's just pointing past the last entry in left, then we'll
2282 * insert there, so don't change anything in that case.
2284 if (cur->bc_ptrs[level] > lrecs + 1) {
2285 xfs_btree_setbuf(cur, level, rbp);
2286 cur->bc_ptrs[level] -= lrecs;
2289 * If there are more levels, we'll need another cursor which refers
2290 * the right block, no matter where this cursor was.
2292 if (level + 1 < cur->bc_nlevels) {
2293 error = xfs_btree_dup_cursor(cur, curp);
2296 (*curp)->bc_ptrs[level + 1]++;
2299 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2303 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2308 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2313 * Copy the old inode root contents into a real block and make the
2314 * broot point to it.
2317 xfs_btree_new_iroot(
2318 struct xfs_btree_cur *cur, /* btree cursor */
2319 int *logflags, /* logging flags for inode */
2320 int *stat) /* return status - 0 fail */
2322 struct xfs_buf *cbp; /* buffer for cblock */
2323 struct xfs_btree_block *block; /* btree block */
2324 struct xfs_btree_block *cblock; /* child btree block */
2325 union xfs_btree_key *ckp; /* child key pointer */
2326 union xfs_btree_ptr *cpp; /* child ptr pointer */
2327 union xfs_btree_key *kp; /* pointer to btree key */
2328 union xfs_btree_ptr *pp; /* pointer to block addr */
2329 union xfs_btree_ptr nptr; /* new block addr */
2330 int level; /* btree level */
2331 int error; /* error return code */
2333 int i; /* loop counter */
2336 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2337 XFS_BTREE_STATS_INC(cur, newroot);
2339 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2341 level = cur->bc_nlevels - 1;
2343 block = xfs_btree_get_iroot(cur);
2344 pp = xfs_btree_ptr_addr(cur, 1, block);
2346 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2347 error = cur->bc_ops->alloc_block(cur, pp, &nptr, 1, stat);
2351 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2354 XFS_BTREE_STATS_INC(cur, alloc);
2356 /* Copy the root into a real block. */
2357 error = xfs_btree_get_buf_block(cur, &nptr, 0, &cblock, &cbp);
2361 memcpy(cblock, block, xfs_btree_block_len(cur));
2363 be16_add_cpu(&block->bb_level, 1);
2364 xfs_btree_set_numrecs(block, 1);
2366 cur->bc_ptrs[level + 1] = 1;
2368 kp = xfs_btree_key_addr(cur, 1, block);
2369 ckp = xfs_btree_key_addr(cur, 1, cblock);
2370 xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
2372 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
2374 for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
2375 error = xfs_btree_check_ptr(cur, pp, i, level);
2380 xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
2383 error = xfs_btree_check_ptr(cur, &nptr, 0, level);
2387 xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
2389 xfs_iroot_realloc(cur->bc_private.b.ip,
2390 1 - xfs_btree_get_numrecs(cblock),
2391 cur->bc_private.b.whichfork);
2393 xfs_btree_setbuf(cur, level, cbp);
2396 * Do all this logging at the end so that
2397 * the root is at the right level.
2399 xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
2400 xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2401 xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2404 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork);
2406 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2409 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2414 * Allocate a new root block, fill it in.
2416 STATIC int /* error */
2418 struct xfs_btree_cur *cur, /* btree cursor */
2419 int *stat) /* success/failure */
2421 struct xfs_btree_block *block; /* one half of the old root block */
2422 struct xfs_buf *bp; /* buffer containing block */
2423 int error; /* error return value */
2424 struct xfs_buf *lbp; /* left buffer pointer */
2425 struct xfs_btree_block *left; /* left btree block */
2426 struct xfs_buf *nbp; /* new (root) buffer */
2427 struct xfs_btree_block *new; /* new (root) btree block */
2428 int nptr; /* new value for key index, 1 or 2 */
2429 struct xfs_buf *rbp; /* right buffer pointer */
2430 struct xfs_btree_block *right; /* right btree block */
2431 union xfs_btree_ptr rptr;
2432 union xfs_btree_ptr lptr;
2434 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2435 XFS_BTREE_STATS_INC(cur, newroot);
2437 /* initialise our start point from the cursor */
2438 cur->bc_ops->init_ptr_from_cur(cur, &rptr);
2440 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2441 error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, 1, stat);
2446 XFS_BTREE_STATS_INC(cur, alloc);
2448 /* Set up the new block. */
2449 error = xfs_btree_get_buf_block(cur, &lptr, 0, &new, &nbp);
2453 /* Set the root in the holding structure increasing the level by 1. */
2454 cur->bc_ops->set_root(cur, &lptr, 1);
2457 * At the previous root level there are now two blocks: the old root,
2458 * and the new block generated when it was split. We don't know which
2459 * one the cursor is pointing at, so we set up variables "left" and
2460 * "right" for each case.
2462 block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
2465 error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
2470 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
2471 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
2472 /* Our block is left, pick up the right block. */
2474 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2476 error = xfs_btree_read_buf_block(cur, &rptr,
2477 cur->bc_nlevels - 1, 0, &right, &rbp);
2483 /* Our block is right, pick up the left block. */
2485 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
2487 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2488 error = xfs_btree_read_buf_block(cur, &lptr,
2489 cur->bc_nlevels - 1, 0, &left, &lbp);
2495 /* Fill in the new block's btree header and log it. */
2496 xfs_btree_init_block(cur, cur->bc_nlevels, 2, new);
2497 xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
2498 ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
2499 !xfs_btree_ptr_is_null(cur, &rptr));
2501 /* Fill in the key data in the new root. */
2502 if (xfs_btree_get_level(left) > 0) {
2503 xfs_btree_copy_keys(cur,
2504 xfs_btree_key_addr(cur, 1, new),
2505 xfs_btree_key_addr(cur, 1, left), 1);
2506 xfs_btree_copy_keys(cur,
2507 xfs_btree_key_addr(cur, 2, new),
2508 xfs_btree_key_addr(cur, 1, right), 1);
2510 cur->bc_ops->init_key_from_rec(
2511 xfs_btree_key_addr(cur, 1, new),
2512 xfs_btree_rec_addr(cur, 1, left));
2513 cur->bc_ops->init_key_from_rec(
2514 xfs_btree_key_addr(cur, 2, new),
2515 xfs_btree_rec_addr(cur, 1, right));
2517 xfs_btree_log_keys(cur, nbp, 1, 2);
2519 /* Fill in the pointer data in the new root. */
2520 xfs_btree_copy_ptrs(cur,
2521 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
2522 xfs_btree_copy_ptrs(cur,
2523 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
2524 xfs_btree_log_ptrs(cur, nbp, 1, 2);
2526 /* Fix up the cursor. */
2527 xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
2528 cur->bc_ptrs[cur->bc_nlevels] = nptr;
2530 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2534 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2537 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2543 xfs_btree_make_block_unfull(
2544 struct xfs_btree_cur *cur, /* btree cursor */
2545 int level, /* btree level */
2546 int numrecs,/* # of recs in block */
2547 int *oindex,/* old tree index */
2548 int *index, /* new tree index */
2549 union xfs_btree_ptr *nptr, /* new btree ptr */
2550 struct xfs_btree_cur **ncur, /* new btree cursor */
2551 union xfs_btree_rec *nrec, /* new record */
2554 union xfs_btree_key key; /* new btree key value */
2557 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2558 level == cur->bc_nlevels - 1) {
2559 struct xfs_inode *ip = cur->bc_private.b.ip;
2561 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
2562 /* A root block that can be made bigger. */
2564 xfs_iroot_realloc(ip, 1, cur->bc_private.b.whichfork);
2566 /* A root block that needs replacing */
2569 error = xfs_btree_new_iroot(cur, &logflags, stat);
2570 if (error || *stat == 0)
2573 xfs_trans_log_inode(cur->bc_tp, ip, logflags);
2579 /* First, try shifting an entry to the right neighbor. */
2580 error = xfs_btree_rshift(cur, level, stat);
2584 /* Next, try shifting an entry to the left neighbor. */
2585 error = xfs_btree_lshift(cur, level, stat);
2590 *oindex = *index = cur->bc_ptrs[level];
2595 * Next, try splitting the current block in half.
2597 * If this works we have to re-set our variables because we
2598 * could be in a different block now.
2600 error = xfs_btree_split(cur, level, nptr, &key, ncur, stat);
2601 if (error || *stat == 0)
2605 *index = cur->bc_ptrs[level];
2606 cur->bc_ops->init_rec_from_key(&key, nrec);
2611 * Insert one record/level. Return information to the caller
2612 * allowing the next level up to proceed if necessary.
2616 struct xfs_btree_cur *cur, /* btree cursor */
2617 int level, /* level to insert record at */
2618 union xfs_btree_ptr *ptrp, /* i/o: block number inserted */
2619 union xfs_btree_rec *recp, /* i/o: record data inserted */
2620 struct xfs_btree_cur **curp, /* output: new cursor replacing cur */
2621 int *stat) /* success/failure */
2623 struct xfs_btree_block *block; /* btree block */
2624 struct xfs_buf *bp; /* buffer for block */
2625 union xfs_btree_key key; /* btree key */
2626 union xfs_btree_ptr nptr; /* new block ptr */
2627 struct xfs_btree_cur *ncur; /* new btree cursor */
2628 union xfs_btree_rec nrec; /* new record count */
2629 int optr; /* old key/record index */
2630 int ptr; /* key/record index */
2631 int numrecs;/* number of records */
2632 int error; /* error return value */
2637 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2638 XFS_BTREE_TRACE_ARGIPR(cur, level, *ptrp, recp);
2643 * If we have an external root pointer, and we've made it to the
2644 * root level, allocate a new root block and we're done.
2646 if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2647 (level >= cur->bc_nlevels)) {
2648 error = xfs_btree_new_root(cur, stat);
2649 xfs_btree_set_ptr_null(cur, ptrp);
2651 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2655 /* If we're off the left edge, return failure. */
2656 ptr = cur->bc_ptrs[level];
2658 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2663 /* Make a key out of the record data to be inserted, and save it. */
2664 cur->bc_ops->init_key_from_rec(&key, recp);
2668 XFS_BTREE_STATS_INC(cur, insrec);
2670 /* Get pointers to the btree buffer and block. */
2671 block = xfs_btree_get_block(cur, level, &bp);
2672 numrecs = xfs_btree_get_numrecs(block);
2675 error = xfs_btree_check_block(cur, block, level, bp);
2679 /* Check that the new entry is being inserted in the right place. */
2680 if (ptr <= numrecs) {
2682 ASSERT(cur->bc_ops->recs_inorder(cur, recp,
2683 xfs_btree_rec_addr(cur, ptr, block)));
2685 ASSERT(cur->bc_ops->keys_inorder(cur, &key,
2686 xfs_btree_key_addr(cur, ptr, block)));
2692 * If the block is full, we can't insert the new entry until we
2693 * make the block un-full.
2695 xfs_btree_set_ptr_null(cur, &nptr);
2696 if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
2697 error = xfs_btree_make_block_unfull(cur, level, numrecs,
2698 &optr, &ptr, &nptr, &ncur, &nrec, stat);
2699 if (error || *stat == 0)
2704 * The current block may have changed if the block was
2705 * previously full and we have just made space in it.
2707 block = xfs_btree_get_block(cur, level, &bp);
2708 numrecs = xfs_btree_get_numrecs(block);
2711 error = xfs_btree_check_block(cur, block, level, bp);
2717 * At this point we know there's room for our new entry in the block
2718 * we're pointing at.
2720 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
2723 /* It's a nonleaf. make a hole in the keys and ptrs */
2724 union xfs_btree_key *kp;
2725 union xfs_btree_ptr *pp;
2727 kp = xfs_btree_key_addr(cur, ptr, block);
2728 pp = xfs_btree_ptr_addr(cur, ptr, block);
2731 for (i = numrecs - ptr; i >= 0; i--) {
2732 error = xfs_btree_check_ptr(cur, pp, i, level);
2738 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
2739 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
2742 error = xfs_btree_check_ptr(cur, ptrp, 0, level);
2747 /* Now put the new data in, bump numrecs and log it. */
2748 xfs_btree_copy_keys(cur, kp, &key, 1);
2749 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
2751 xfs_btree_set_numrecs(block, numrecs);
2752 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
2753 xfs_btree_log_keys(cur, bp, ptr, numrecs);
2755 if (ptr < numrecs) {
2756 ASSERT(cur->bc_ops->keys_inorder(cur, kp,
2757 xfs_btree_key_addr(cur, ptr + 1, block)));
2761 /* It's a leaf. make a hole in the records */
2762 union xfs_btree_rec *rp;
2764 rp = xfs_btree_rec_addr(cur, ptr, block);
2766 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
2768 /* Now put the new data in, bump numrecs and log it. */
2769 xfs_btree_copy_recs(cur, rp, recp, 1);
2770 xfs_btree_set_numrecs(block, ++numrecs);
2771 xfs_btree_log_recs(cur, bp, ptr, numrecs);
2773 if (ptr < numrecs) {
2774 ASSERT(cur->bc_ops->recs_inorder(cur, rp,
2775 xfs_btree_rec_addr(cur, ptr + 1, block)));
2780 /* Log the new number of records in the btree header. */
2781 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
2783 /* If we inserted at the start of a block, update the parents' keys. */
2785 error = xfs_btree_updkey(cur, &key, level + 1);
2791 * If we are tracking the last record in the tree and
2792 * we are at the far right edge of the tree, update it.
2794 if (xfs_btree_is_lastrec(cur, block, level)) {
2795 cur->bc_ops->update_lastrec(cur, block, recp,
2796 ptr, LASTREC_INSREC);
2800 * Return the new block number, if any.
2801 * If there is one, give back a record value and a cursor too.
2804 if (!xfs_btree_ptr_is_null(cur, &nptr)) {
2809 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2814 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2819 * Insert the record at the point referenced by cur.
2821 * A multi-level split of the tree on insert will invalidate the original
2822 * cursor. All callers of this function should assume that the cursor is
2823 * no longer valid and revalidate it.
2827 struct xfs_btree_cur *cur,
2830 int error; /* error return value */
2831 int i; /* result value, 0 for failure */
2832 int level; /* current level number in btree */
2833 union xfs_btree_ptr nptr; /* new block number (split result) */
2834 struct xfs_btree_cur *ncur; /* new cursor (split result) */
2835 struct xfs_btree_cur *pcur; /* previous level's cursor */
2836 union xfs_btree_rec rec; /* record to insert */
2842 xfs_btree_set_ptr_null(cur, &nptr);
2843 cur->bc_ops->init_rec_from_cur(cur, &rec);
2846 * Loop going up the tree, starting at the leaf level.
2847 * Stop when we don't get a split block, that must mean that
2848 * the insert is finished with this level.
2852 * Insert nrec/nptr into this level of the tree.
2853 * Note if we fail, nptr will be null.
2855 error = xfs_btree_insrec(pcur, level, &nptr, &rec, &ncur, &i);
2858 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
2862 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
2866 * See if the cursor we just used is trash.
2867 * Can't trash the caller's cursor, but otherwise we should
2868 * if ncur is a new cursor or we're about to be done.
2871 (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
2872 /* Save the state from the cursor before we trash it */
2873 if (cur->bc_ops->update_cursor)
2874 cur->bc_ops->update_cursor(pcur, cur);
2875 cur->bc_nlevels = pcur->bc_nlevels;
2876 xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
2878 /* If we got a new cursor, switch to it. */
2883 } while (!xfs_btree_ptr_is_null(cur, &nptr));
2885 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2889 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2894 * Try to merge a non-leaf block back into the inode root.
2896 * Note: the killroot names comes from the fact that we're effectively
2897 * killing the old root block. But because we can't just delete the
2898 * inode we have to copy the single block it was pointing to into the
2902 xfs_btree_kill_iroot(
2903 struct xfs_btree_cur *cur)
2905 int whichfork = cur->bc_private.b.whichfork;
2906 struct xfs_inode *ip = cur->bc_private.b.ip;
2907 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
2908 struct xfs_btree_block *block;
2909 struct xfs_btree_block *cblock;
2910 union xfs_btree_key *kp;
2911 union xfs_btree_key *ckp;
2912 union xfs_btree_ptr *pp;
2913 union xfs_btree_ptr *cpp;
2914 struct xfs_buf *cbp;
2919 union xfs_btree_ptr ptr;
2923 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2925 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2926 ASSERT(cur->bc_nlevels > 1);
2929 * Don't deal with the root block needs to be a leaf case.
2930 * We're just going to turn the thing back into extents anyway.
2932 level = cur->bc_nlevels - 1;
2937 * Give up if the root has multiple children.
2939 block = xfs_btree_get_iroot(cur);
2940 if (xfs_btree_get_numrecs(block) != 1)
2943 cblock = xfs_btree_get_block(cur, level - 1, &cbp);
2944 numrecs = xfs_btree_get_numrecs(cblock);
2947 * Only do this if the next level will fit.
2948 * Then the data must be copied up to the inode,
2949 * instead of freeing the root you free the next level.
2951 if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
2954 XFS_BTREE_STATS_INC(cur, killroot);
2957 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
2958 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
2959 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
2960 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
2963 index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
2965 xfs_iroot_realloc(cur->bc_private.b.ip, index,
2966 cur->bc_private.b.whichfork);
2967 block = ifp->if_broot;
2970 be16_add_cpu(&block->bb_numrecs, index);
2971 ASSERT(block->bb_numrecs == cblock->bb_numrecs);
2973 kp = xfs_btree_key_addr(cur, 1, block);
2974 ckp = xfs_btree_key_addr(cur, 1, cblock);
2975 xfs_btree_copy_keys(cur, kp, ckp, numrecs);
2977 pp = xfs_btree_ptr_addr(cur, 1, block);
2978 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
2980 for (i = 0; i < numrecs; i++) {
2983 error = xfs_btree_check_ptr(cur, cpp, i, level - 1);
2985 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2990 xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
2992 cur->bc_ops->free_block(cur, cbp);
2993 XFS_BTREE_STATS_INC(cur, free);
2995 cur->bc_bufs[level - 1] = NULL;
2996 be16_add_cpu(&block->bb_level, -1);
2997 xfs_trans_log_inode(cur->bc_tp, ip,
2998 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork));
3001 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3006 * Kill the current root node, and replace it with it's only child node.
3009 xfs_btree_kill_root(
3010 struct xfs_btree_cur *cur,
3013 union xfs_btree_ptr *newroot)
3017 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3018 XFS_BTREE_STATS_INC(cur, killroot);
3021 * Update the root pointer, decreasing the level by 1 and then
3022 * free the old root.
3024 cur->bc_ops->set_root(cur, newroot, -1);
3026 error = cur->bc_ops->free_block(cur, bp);
3028 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3032 XFS_BTREE_STATS_INC(cur, free);
3034 cur->bc_bufs[level] = NULL;
3035 cur->bc_ra[level] = 0;
3038 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3043 xfs_btree_dec_cursor(
3044 struct xfs_btree_cur *cur,
3052 error = xfs_btree_decrement(cur, level, &i);
3057 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3063 * Single level of the btree record deletion routine.
3064 * Delete record pointed to by cur/level.
3065 * Remove the record from its block then rebalance the tree.
3066 * Return 0 for error, 1 for done, 2 to go on to the next level.
3068 STATIC int /* error */
3070 struct xfs_btree_cur *cur, /* btree cursor */
3071 int level, /* level removing record from */
3072 int *stat) /* fail/done/go-on */
3074 struct xfs_btree_block *block; /* btree block */
3075 union xfs_btree_ptr cptr; /* current block ptr */
3076 struct xfs_buf *bp; /* buffer for block */
3077 int error; /* error return value */
3078 int i; /* loop counter */
3079 union xfs_btree_key key; /* storage for keyp */
3080 union xfs_btree_key *keyp = &key; /* passed to the next level */
3081 union xfs_btree_ptr lptr; /* left sibling block ptr */
3082 struct xfs_buf *lbp; /* left buffer pointer */
3083 struct xfs_btree_block *left; /* left btree block */
3084 int lrecs = 0; /* left record count */
3085 int ptr; /* key/record index */
3086 union xfs_btree_ptr rptr; /* right sibling block ptr */
3087 struct xfs_buf *rbp; /* right buffer pointer */
3088 struct xfs_btree_block *right; /* right btree block */
3089 struct xfs_btree_block *rrblock; /* right-right btree block */
3090 struct xfs_buf *rrbp; /* right-right buffer pointer */
3091 int rrecs = 0; /* right record count */
3092 struct xfs_btree_cur *tcur; /* temporary btree cursor */
3093 int numrecs; /* temporary numrec count */
3095 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3096 XFS_BTREE_TRACE_ARGI(cur, level);
3100 /* Get the index of the entry being deleted, check for nothing there. */
3101 ptr = cur->bc_ptrs[level];
3103 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3108 /* Get the buffer & block containing the record or key/ptr. */
3109 block = xfs_btree_get_block(cur, level, &bp);
3110 numrecs = xfs_btree_get_numrecs(block);
3113 error = xfs_btree_check_block(cur, block, level, bp);
3118 /* Fail if we're off the end of the block. */
3119 if (ptr > numrecs) {
3120 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3125 XFS_BTREE_STATS_INC(cur, delrec);
3126 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3128 /* Excise the entries being deleted. */
3130 /* It's a nonleaf. operate on keys and ptrs */
3131 union xfs_btree_key *lkp;
3132 union xfs_btree_ptr *lpp;
3134 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3135 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3138 for (i = 0; i < numrecs - ptr; i++) {
3139 error = xfs_btree_check_ptr(cur, lpp, i, level);
3145 if (ptr < numrecs) {
3146 xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3147 xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3148 xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3149 xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3153 * If it's the first record in the block, we'll need to pass a
3154 * key up to the next level (updkey).
3157 keyp = xfs_btree_key_addr(cur, 1, block);
3159 /* It's a leaf. operate on records */
3160 if (ptr < numrecs) {
3161 xfs_btree_shift_recs(cur,
3162 xfs_btree_rec_addr(cur, ptr + 1, block),
3164 xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3168 * If it's the first record in the block, we'll need a key
3169 * structure to pass up to the next level (updkey).
3172 cur->bc_ops->init_key_from_rec(&key,
3173 xfs_btree_rec_addr(cur, 1, block));
3179 * Decrement and log the number of entries in the block.
3181 xfs_btree_set_numrecs(block, --numrecs);
3182 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3185 * If we are tracking the last record in the tree and
3186 * we are at the far right edge of the tree, update it.
3188 if (xfs_btree_is_lastrec(cur, block, level)) {
3189 cur->bc_ops->update_lastrec(cur, block, NULL,
3190 ptr, LASTREC_DELREC);
3194 * We're at the root level. First, shrink the root block in-memory.
3195 * Try to get rid of the next level down. If we can't then there's
3196 * nothing left to do.
3198 if (level == cur->bc_nlevels - 1) {
3199 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3200 xfs_iroot_realloc(cur->bc_private.b.ip, -1,
3201 cur->bc_private.b.whichfork);
3203 error = xfs_btree_kill_iroot(cur);
3207 error = xfs_btree_dec_cursor(cur, level, stat);
3215 * If this is the root level, and there's only one entry left,
3216 * and it's NOT the leaf level, then we can get rid of this
3219 if (numrecs == 1 && level > 0) {
3220 union xfs_btree_ptr *pp;
3222 * pp is still set to the first pointer in the block.
3223 * Make it the new root of the btree.
3225 pp = xfs_btree_ptr_addr(cur, 1, block);
3226 error = xfs_btree_kill_root(cur, bp, level, pp);
3229 } else if (level > 0) {
3230 error = xfs_btree_dec_cursor(cur, level, stat);
3239 * If we deleted the leftmost entry in the block, update the
3240 * key values above us in the tree.
3243 error = xfs_btree_updkey(cur, keyp, level + 1);
3249 * If the number of records remaining in the block is at least
3250 * the minimum, we're done.
3252 if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3253 error = xfs_btree_dec_cursor(cur, level, stat);
3260 * Otherwise, we have to move some records around to keep the
3261 * tree balanced. Look at the left and right sibling blocks to
3262 * see if we can re-balance by moving only one record.
3264 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3265 xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3267 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3269 * One child of root, need to get a chance to copy its contents
3270 * into the root and delete it. Can't go up to next level,
3271 * there's nothing to delete there.
3273 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3274 xfs_btree_ptr_is_null(cur, &lptr) &&
3275 level == cur->bc_nlevels - 2) {
3276 error = xfs_btree_kill_iroot(cur);
3278 error = xfs_btree_dec_cursor(cur, level, stat);
3285 ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3286 !xfs_btree_ptr_is_null(cur, &lptr));
3289 * Duplicate the cursor so our btree manipulations here won't
3290 * disrupt the next level up.
3292 error = xfs_btree_dup_cursor(cur, &tcur);
3297 * If there's a right sibling, see if it's ok to shift an entry
3300 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3302 * Move the temp cursor to the last entry in the next block.
3303 * Actually any entry but the first would suffice.
3305 i = xfs_btree_lastrec(tcur, level);
3306 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3308 error = xfs_btree_increment(tcur, level, &i);
3311 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3313 i = xfs_btree_lastrec(tcur, level);
3314 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3316 /* Grab a pointer to the block. */
3317 right = xfs_btree_get_block(tcur, level, &rbp);
3319 error = xfs_btree_check_block(tcur, right, level, rbp);
3323 /* Grab the current block number, for future use. */
3324 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3327 * If right block is full enough so that removing one entry
3328 * won't make it too empty, and left-shifting an entry out
3329 * of right to us works, we're done.
3331 if (xfs_btree_get_numrecs(right) - 1 >=
3332 cur->bc_ops->get_minrecs(tcur, level)) {
3333 error = xfs_btree_lshift(tcur, level, &i);
3337 ASSERT(xfs_btree_get_numrecs(block) >=
3338 cur->bc_ops->get_minrecs(tcur, level));
3340 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3343 error = xfs_btree_dec_cursor(cur, level, stat);
3351 * Otherwise, grab the number of records in right for
3352 * future reference, and fix up the temp cursor to point
3353 * to our block again (last record).
3355 rrecs = xfs_btree_get_numrecs(right);
3356 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3357 i = xfs_btree_firstrec(tcur, level);
3358 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3360 error = xfs_btree_decrement(tcur, level, &i);
3363 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3368 * If there's a left sibling, see if it's ok to shift an entry
3371 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3373 * Move the temp cursor to the first entry in the
3376 i = xfs_btree_firstrec(tcur, level);
3377 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3379 error = xfs_btree_decrement(tcur, level, &i);
3382 i = xfs_btree_firstrec(tcur, level);
3383 XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
3385 /* Grab a pointer to the block. */
3386 left = xfs_btree_get_block(tcur, level, &lbp);
3388 error = xfs_btree_check_block(cur, left, level, lbp);
3392 /* Grab the current block number, for future use. */
3393 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
3396 * If left block is full enough so that removing one entry
3397 * won't make it too empty, and right-shifting an entry out
3398 * of left to us works, we're done.
3400 if (xfs_btree_get_numrecs(left) - 1 >=
3401 cur->bc_ops->get_minrecs(tcur, level)) {
3402 error = xfs_btree_rshift(tcur, level, &i);
3406 ASSERT(xfs_btree_get_numrecs(block) >=
3407 cur->bc_ops->get_minrecs(tcur, level));
3408 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3412 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3419 * Otherwise, grab the number of records in right for
3422 lrecs = xfs_btree_get_numrecs(left);
3425 /* Delete the temp cursor, we're done with it. */
3426 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3429 /* If here, we need to do a join to keep the tree balanced. */
3430 ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
3432 if (!xfs_btree_ptr_is_null(cur, &lptr) &&
3433 lrecs + xfs_btree_get_numrecs(block) <=
3434 cur->bc_ops->get_maxrecs(cur, level)) {
3436 * Set "right" to be the starting block,
3437 * "left" to be the left neighbor.
3442 error = xfs_btree_read_buf_block(cur, &lptr, level,
3448 * If that won't work, see if we can join with the right neighbor block.
3450 } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
3451 rrecs + xfs_btree_get_numrecs(block) <=
3452 cur->bc_ops->get_maxrecs(cur, level)) {
3454 * Set "left" to be the starting block,
3455 * "right" to be the right neighbor.
3460 error = xfs_btree_read_buf_block(cur, &rptr, level,
3466 * Otherwise, we can't fix the imbalance.
3467 * Just return. This is probably a logic error, but it's not fatal.
3470 error = xfs_btree_dec_cursor(cur, level, stat);
3476 rrecs = xfs_btree_get_numrecs(right);
3477 lrecs = xfs_btree_get_numrecs(left);
3480 * We're now going to join "left" and "right" by moving all the stuff
3481 * in "right" to "left" and deleting "right".
3483 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
3485 /* It's a non-leaf. Move keys and pointers. */
3486 union xfs_btree_key *lkp; /* left btree key */
3487 union xfs_btree_ptr *lpp; /* left address pointer */
3488 union xfs_btree_key *rkp; /* right btree key */
3489 union xfs_btree_ptr *rpp; /* right address pointer */
3491 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
3492 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
3493 rkp = xfs_btree_key_addr(cur, 1, right);
3494 rpp = xfs_btree_ptr_addr(cur, 1, right);
3496 for (i = 1; i < rrecs; i++) {
3497 error = xfs_btree_check_ptr(cur, rpp, i, level);
3502 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
3503 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
3505 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
3506 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
3508 /* It's a leaf. Move records. */
3509 union xfs_btree_rec *lrp; /* left record pointer */
3510 union xfs_btree_rec *rrp; /* right record pointer */
3512 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
3513 rrp = xfs_btree_rec_addr(cur, 1, right);
3515 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
3516 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
3519 XFS_BTREE_STATS_INC(cur, join);
3522 * Fix up the number of records and right block pointer in the
3523 * surviving block, and log it.
3525 xfs_btree_set_numrecs(left, lrecs + rrecs);
3526 xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB),
3527 xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
3528 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
3530 /* If there is a right sibling, point it to the remaining block. */
3531 xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
3532 if (!xfs_btree_ptr_is_null(cur, &cptr)) {
3533 error = xfs_btree_read_buf_block(cur, &cptr, level,
3534 0, &rrblock, &rrbp);
3537 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
3538 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
3541 /* Free the deleted block. */
3542 error = cur->bc_ops->free_block(cur, rbp);
3545 XFS_BTREE_STATS_INC(cur, free);
3548 * If we joined with the left neighbor, set the buffer in the
3549 * cursor to the left block, and fix up the index.
3552 cur->bc_bufs[level] = lbp;
3553 cur->bc_ptrs[level] += lrecs;
3554 cur->bc_ra[level] = 0;
3557 * If we joined with the right neighbor and there's a level above
3558 * us, increment the cursor at that level.
3560 else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
3561 (level + 1 < cur->bc_nlevels)) {
3562 error = xfs_btree_increment(cur, level + 1, &i);
3568 * Readjust the ptr at this level if it's not a leaf, since it's
3569 * still pointing at the deletion point, which makes the cursor
3570 * inconsistent. If this makes the ptr 0, the caller fixes it up.
3571 * We can't use decrement because it would change the next level up.
3574 cur->bc_ptrs[level]--;
3576 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3577 /* Return value means the next level up has something to do. */
3582 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3584 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
3589 * Delete the record pointed to by cur.
3590 * The cursor refers to the place where the record was (could be inserted)
3591 * when the operation returns.
3595 struct xfs_btree_cur *cur,
3596 int *stat) /* success/failure */
3598 int error; /* error return value */
3602 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3605 * Go up the tree, starting at leaf level.
3607 * If 2 is returned then a join was done; go to the next level.
3608 * Otherwise we are done.
3610 for (level = 0, i = 2; i == 2; level++) {
3611 error = xfs_btree_delrec(cur, level, &i);
3617 for (level = 1; level < cur->bc_nlevels; level++) {
3618 if (cur->bc_ptrs[level] == 0) {
3619 error = xfs_btree_decrement(cur, level, &i);
3627 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3631 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3636 * Get the data from the pointed-to record.
3640 struct xfs_btree_cur *cur, /* btree cursor */
3641 union xfs_btree_rec **recp, /* output: btree record */
3642 int *stat) /* output: success/failure */
3644 struct xfs_btree_block *block; /* btree block */
3645 struct xfs_buf *bp; /* buffer pointer */
3646 int ptr; /* record number */
3648 int error; /* error return value */
3651 ptr = cur->bc_ptrs[0];
3652 block = xfs_btree_get_block(cur, 0, &bp);
3655 error = xfs_btree_check_block(cur, block, 0, bp);
3661 * Off the right end or left end, return failure.
3663 if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
3669 * Point to the record and extract its data.
3671 *recp = xfs_btree_rec_addr(cur, ptr, block);