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_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_defer.h"
27 #include "xfs_inode.h"
28 #include "xfs_trans.h"
29 #include "xfs_inode_item.h"
30 #include "xfs_buf_item.h"
31 #include "xfs_btree.h"
32 #include "xfs_error.h"
33 #include "xfs_trace.h"
34 #include "xfs_cksum.h"
35 #include "xfs_alloc.h"
39 * Cursor allocation zone.
41 kmem_zone_t *xfs_btree_cur_zone;
44 * Btree magic numbers.
46 static const __uint32_t xfs_magics[2][XFS_BTNUM_MAX] = {
47 { XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, 0, XFS_BMAP_MAGIC, XFS_IBT_MAGIC,
49 { XFS_ABTB_CRC_MAGIC, XFS_ABTC_CRC_MAGIC, XFS_RMAP_CRC_MAGIC,
50 XFS_BMAP_CRC_MAGIC, XFS_IBT_CRC_MAGIC, XFS_FIBT_CRC_MAGIC }
52 #define xfs_btree_magic(cur) \
53 xfs_magics[!!((cur)->bc_flags & XFS_BTREE_CRC_BLOCKS)][cur->bc_btnum]
55 STATIC int /* error (0 or EFSCORRUPTED) */
56 xfs_btree_check_lblock(
57 struct xfs_btree_cur *cur, /* btree cursor */
58 struct xfs_btree_block *block, /* btree long form block pointer */
59 int level, /* level of the btree block */
60 struct xfs_buf *bp) /* buffer for block, if any */
62 int lblock_ok = 1; /* block passes checks */
63 struct xfs_mount *mp; /* file system mount point */
67 if (xfs_sb_version_hascrc(&mp->m_sb)) {
68 lblock_ok = lblock_ok &&
69 uuid_equal(&block->bb_u.l.bb_uuid,
70 &mp->m_sb.sb_meta_uuid) &&
71 block->bb_u.l.bb_blkno == cpu_to_be64(
72 bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
75 lblock_ok = lblock_ok &&
76 be32_to_cpu(block->bb_magic) == xfs_btree_magic(cur) &&
77 be16_to_cpu(block->bb_level) == level &&
78 be16_to_cpu(block->bb_numrecs) <=
79 cur->bc_ops->get_maxrecs(cur, level) &&
80 block->bb_u.l.bb_leftsib &&
81 (block->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK) ||
82 XFS_FSB_SANITY_CHECK(mp,
83 be64_to_cpu(block->bb_u.l.bb_leftsib))) &&
84 block->bb_u.l.bb_rightsib &&
85 (block->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK) ||
86 XFS_FSB_SANITY_CHECK(mp,
87 be64_to_cpu(block->bb_u.l.bb_rightsib)));
89 if (unlikely(XFS_TEST_ERROR(!lblock_ok, mp,
90 XFS_ERRTAG_BTREE_CHECK_LBLOCK,
91 XFS_RANDOM_BTREE_CHECK_LBLOCK))) {
93 trace_xfs_btree_corrupt(bp, _RET_IP_);
94 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
100 STATIC int /* error (0 or EFSCORRUPTED) */
101 xfs_btree_check_sblock(
102 struct xfs_btree_cur *cur, /* btree cursor */
103 struct xfs_btree_block *block, /* btree short form block pointer */
104 int level, /* level of the btree block */
105 struct xfs_buf *bp) /* buffer containing block */
107 struct xfs_mount *mp; /* file system mount point */
108 struct xfs_buf *agbp; /* buffer for ag. freespace struct */
109 struct xfs_agf *agf; /* ag. freespace structure */
110 xfs_agblock_t agflen; /* native ag. freespace length */
111 int sblock_ok = 1; /* block passes checks */
114 agbp = cur->bc_private.a.agbp;
115 agf = XFS_BUF_TO_AGF(agbp);
116 agflen = be32_to_cpu(agf->agf_length);
118 if (xfs_sb_version_hascrc(&mp->m_sb)) {
119 sblock_ok = sblock_ok &&
120 uuid_equal(&block->bb_u.s.bb_uuid,
121 &mp->m_sb.sb_meta_uuid) &&
122 block->bb_u.s.bb_blkno == cpu_to_be64(
123 bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
126 sblock_ok = sblock_ok &&
127 be32_to_cpu(block->bb_magic) == xfs_btree_magic(cur) &&
128 be16_to_cpu(block->bb_level) == level &&
129 be16_to_cpu(block->bb_numrecs) <=
130 cur->bc_ops->get_maxrecs(cur, level) &&
131 (block->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) ||
132 be32_to_cpu(block->bb_u.s.bb_leftsib) < agflen) &&
133 block->bb_u.s.bb_leftsib &&
134 (block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK) ||
135 be32_to_cpu(block->bb_u.s.bb_rightsib) < agflen) &&
136 block->bb_u.s.bb_rightsib;
138 if (unlikely(XFS_TEST_ERROR(!sblock_ok, mp,
139 XFS_ERRTAG_BTREE_CHECK_SBLOCK,
140 XFS_RANDOM_BTREE_CHECK_SBLOCK))) {
142 trace_xfs_btree_corrupt(bp, _RET_IP_);
143 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
144 return -EFSCORRUPTED;
150 * Debug routine: check that block header is ok.
153 xfs_btree_check_block(
154 struct xfs_btree_cur *cur, /* btree cursor */
155 struct xfs_btree_block *block, /* generic btree block pointer */
156 int level, /* level of the btree block */
157 struct xfs_buf *bp) /* buffer containing block, if any */
159 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
160 return xfs_btree_check_lblock(cur, block, level, bp);
162 return xfs_btree_check_sblock(cur, block, level, bp);
166 * Check that (long) pointer is ok.
168 int /* error (0 or EFSCORRUPTED) */
169 xfs_btree_check_lptr(
170 struct xfs_btree_cur *cur, /* btree cursor */
171 xfs_fsblock_t bno, /* btree block disk address */
172 int level) /* btree block level */
174 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
176 bno != NULLFSBLOCK &&
177 XFS_FSB_SANITY_CHECK(cur->bc_mp, bno));
183 * Check that (short) pointer is ok.
185 STATIC int /* error (0 or EFSCORRUPTED) */
186 xfs_btree_check_sptr(
187 struct xfs_btree_cur *cur, /* btree cursor */
188 xfs_agblock_t bno, /* btree block disk address */
189 int level) /* btree block level */
191 xfs_agblock_t agblocks = cur->bc_mp->m_sb.sb_agblocks;
193 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
195 bno != NULLAGBLOCK &&
202 * Check that block ptr is ok.
204 STATIC int /* error (0 or EFSCORRUPTED) */
206 struct xfs_btree_cur *cur, /* btree cursor */
207 union xfs_btree_ptr *ptr, /* btree block disk address */
208 int index, /* offset from ptr to check */
209 int level) /* btree block level */
211 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
212 return xfs_btree_check_lptr(cur,
213 be64_to_cpu((&ptr->l)[index]), level);
215 return xfs_btree_check_sptr(cur,
216 be32_to_cpu((&ptr->s)[index]), level);
222 * Calculate CRC on the whole btree block and stuff it into the
223 * long-form btree header.
225 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
226 * it into the buffer so recovery knows what the last modification was that made
230 xfs_btree_lblock_calc_crc(
233 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
234 struct xfs_buf_log_item *bip = bp->b_fspriv;
236 if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
239 block->bb_u.l.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
240 xfs_buf_update_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
244 xfs_btree_lblock_verify_crc(
247 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
248 struct xfs_mount *mp = bp->b_target->bt_mount;
250 if (xfs_sb_version_hascrc(&mp->m_sb)) {
251 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.l.bb_lsn)))
253 return xfs_buf_verify_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
260 * Calculate CRC on the whole btree block and stuff it into the
261 * short-form btree header.
263 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
264 * it into the buffer so recovery knows what the last modification was that made
268 xfs_btree_sblock_calc_crc(
271 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
272 struct xfs_buf_log_item *bip = bp->b_fspriv;
274 if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
277 block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
278 xfs_buf_update_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
282 xfs_btree_sblock_verify_crc(
285 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
286 struct xfs_mount *mp = bp->b_target->bt_mount;
288 if (xfs_sb_version_hascrc(&mp->m_sb)) {
289 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.s.bb_lsn)))
291 return xfs_buf_verify_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
298 xfs_btree_free_block(
299 struct xfs_btree_cur *cur,
304 error = cur->bc_ops->free_block(cur, bp);
306 xfs_trans_binval(cur->bc_tp, bp);
307 XFS_BTREE_STATS_INC(cur, free);
313 * Delete the btree cursor.
316 xfs_btree_del_cursor(
317 xfs_btree_cur_t *cur, /* btree cursor */
318 int error) /* del because of error */
320 int i; /* btree level */
323 * Clear the buffer pointers, and release the buffers.
324 * If we're doing this in the face of an error, we
325 * need to make sure to inspect all of the entries
326 * in the bc_bufs array for buffers to be unlocked.
327 * This is because some of the btree code works from
328 * level n down to 0, and if we get an error along
329 * the way we won't have initialized all the entries
332 for (i = 0; i < cur->bc_nlevels; i++) {
334 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
339 * Can't free a bmap cursor without having dealt with the
340 * allocated indirect blocks' accounting.
342 ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP ||
343 cur->bc_private.b.allocated == 0);
347 kmem_zone_free(xfs_btree_cur_zone, cur);
351 * Duplicate the btree cursor.
352 * Allocate a new one, copy the record, re-get the buffers.
355 xfs_btree_dup_cursor(
356 xfs_btree_cur_t *cur, /* input cursor */
357 xfs_btree_cur_t **ncur) /* output cursor */
359 xfs_buf_t *bp; /* btree block's buffer pointer */
360 int error; /* error return value */
361 int i; /* level number of btree block */
362 xfs_mount_t *mp; /* mount structure for filesystem */
363 xfs_btree_cur_t *new; /* new cursor value */
364 xfs_trans_t *tp; /* transaction pointer, can be NULL */
370 * Allocate a new cursor like the old one.
372 new = cur->bc_ops->dup_cursor(cur);
375 * Copy the record currently in the cursor.
377 new->bc_rec = cur->bc_rec;
380 * For each level current, re-get the buffer and copy the ptr value.
382 for (i = 0; i < new->bc_nlevels; i++) {
383 new->bc_ptrs[i] = cur->bc_ptrs[i];
384 new->bc_ra[i] = cur->bc_ra[i];
385 bp = cur->bc_bufs[i];
387 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
388 XFS_BUF_ADDR(bp), mp->m_bsize,
390 cur->bc_ops->buf_ops);
392 xfs_btree_del_cursor(new, error);
397 new->bc_bufs[i] = bp;
404 * XFS btree block layout and addressing:
406 * There are two types of blocks in the btree: leaf and non-leaf blocks.
408 * The leaf record start with a header then followed by records containing
409 * the values. A non-leaf block also starts with the same header, and
410 * then first contains lookup keys followed by an equal number of pointers
411 * to the btree blocks at the previous level.
413 * +--------+-------+-------+-------+-------+-------+-------+
414 * Leaf: | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
415 * +--------+-------+-------+-------+-------+-------+-------+
417 * +--------+-------+-------+-------+-------+-------+-------+
418 * Non-Leaf: | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
419 * +--------+-------+-------+-------+-------+-------+-------+
421 * The header is called struct xfs_btree_block for reasons better left unknown
422 * and comes in different versions for short (32bit) and long (64bit) block
423 * pointers. The record and key structures are defined by the btree instances
424 * and opaque to the btree core. The block pointers are simple disk endian
425 * integers, available in a short (32bit) and long (64bit) variant.
427 * The helpers below calculate the offset of a given record, key or pointer
428 * into a btree block (xfs_btree_*_offset) or return a pointer to the given
429 * record, key or pointer (xfs_btree_*_addr). Note that all addressing
430 * inside the btree block is done using indices starting at one, not zero!
432 * If XFS_BTREE_OVERLAPPING is set, then this btree supports keys containing
433 * overlapping intervals. In such a tree, records are still sorted lowest to
434 * highest and indexed by the smallest key value that refers to the record.
435 * However, nodes are different: each pointer has two associated keys -- one
436 * indexing the lowest key available in the block(s) below (the same behavior
437 * as the key in a regular btree) and another indexing the highest key
438 * available in the block(s) below. Because records are /not/ sorted by the
439 * highest key, all leaf block updates require us to compute the highest key
440 * that matches any record in the leaf and to recursively update the high keys
441 * in the nodes going further up in the tree, if necessary. Nodes look like
444 * +--------+-----+-----+-----+-----+-----+-------+-------+-----+
445 * Non-Leaf: | header | lo1 | hi1 | lo2 | hi2 | ... | ptr 1 | ptr 2 | ... |
446 * +--------+-----+-----+-----+-----+-----+-------+-------+-----+
448 * To perform an interval query on an overlapped tree, perform the usual
449 * depth-first search and use the low and high keys to decide if we can skip
450 * that particular node. If a leaf node is reached, return the records that
451 * intersect the interval. Note that an interval query may return numerous
452 * entries. For a non-overlapped tree, simply search for the record associated
453 * with the lowest key and iterate forward until a non-matching record is
454 * found. Section 14.3 ("Interval Trees") of _Introduction to Algorithms_ by
455 * Cormen, Leiserson, Rivest, and Stein (2nd or 3rd ed. only) discuss this in
458 * Why do we care about overlapping intervals? Let's say you have a bunch of
459 * reverse mapping records on a reflink filesystem:
461 * 1: +- file A startblock B offset C length D -----------+
462 * 2: +- file E startblock F offset G length H --------------+
463 * 3: +- file I startblock F offset J length K --+
464 * 4: +- file L... --+
466 * Now say we want to map block (B+D) into file A at offset (C+D). Ideally,
467 * we'd simply increment the length of record 1. But how do we find the record
468 * that ends at (B+D-1) (i.e. record 1)? A LE lookup of (B+D-1) would return
469 * record 3 because the keys are ordered first by startblock. An interval
470 * query would return records 1 and 2 because they both overlap (B+D-1), and
471 * from that we can pick out record 1 as the appropriate left neighbor.
473 * In the non-overlapped case you can do a LE lookup and decrement the cursor
474 * because a record's interval must end before the next record.
478 * Return size of the btree block header for this btree instance.
480 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
482 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
483 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
484 return XFS_BTREE_LBLOCK_CRC_LEN;
485 return XFS_BTREE_LBLOCK_LEN;
487 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
488 return XFS_BTREE_SBLOCK_CRC_LEN;
489 return XFS_BTREE_SBLOCK_LEN;
493 * Return size of btree block pointers for this btree instance.
495 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
497 return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
498 sizeof(__be64) : sizeof(__be32);
502 * Calculate offset of the n-th record in a btree block.
505 xfs_btree_rec_offset(
506 struct xfs_btree_cur *cur,
509 return xfs_btree_block_len(cur) +
510 (n - 1) * cur->bc_ops->rec_len;
514 * Calculate offset of the n-th key in a btree block.
517 xfs_btree_key_offset(
518 struct xfs_btree_cur *cur,
521 return xfs_btree_block_len(cur) +
522 (n - 1) * cur->bc_ops->key_len;
526 * Calculate offset of the n-th high key in a btree block.
529 xfs_btree_high_key_offset(
530 struct xfs_btree_cur *cur,
533 return xfs_btree_block_len(cur) +
534 (n - 1) * cur->bc_ops->key_len + (cur->bc_ops->key_len / 2);
538 * Calculate offset of the n-th block pointer in a btree block.
541 xfs_btree_ptr_offset(
542 struct xfs_btree_cur *cur,
546 return xfs_btree_block_len(cur) +
547 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
548 (n - 1) * xfs_btree_ptr_len(cur);
552 * Return a pointer to the n-th record in the btree block.
554 STATIC union xfs_btree_rec *
556 struct xfs_btree_cur *cur,
558 struct xfs_btree_block *block)
560 return (union xfs_btree_rec *)
561 ((char *)block + xfs_btree_rec_offset(cur, n));
565 * Return a pointer to the n-th key in the btree block.
567 STATIC union xfs_btree_key *
569 struct xfs_btree_cur *cur,
571 struct xfs_btree_block *block)
573 return (union xfs_btree_key *)
574 ((char *)block + xfs_btree_key_offset(cur, n));
578 * Return a pointer to the n-th high key in the btree block.
580 STATIC union xfs_btree_key *
581 xfs_btree_high_key_addr(
582 struct xfs_btree_cur *cur,
584 struct xfs_btree_block *block)
586 return (union xfs_btree_key *)
587 ((char *)block + xfs_btree_high_key_offset(cur, n));
591 * Return a pointer to the n-th block pointer in the btree block.
593 STATIC union xfs_btree_ptr *
595 struct xfs_btree_cur *cur,
597 struct xfs_btree_block *block)
599 int level = xfs_btree_get_level(block);
601 ASSERT(block->bb_level != 0);
603 return (union xfs_btree_ptr *)
604 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
608 * Get the root block which is stored in the inode.
610 * For now this btree implementation assumes the btree root is always
611 * stored in the if_broot field of an inode fork.
613 STATIC struct xfs_btree_block *
615 struct xfs_btree_cur *cur)
617 struct xfs_ifork *ifp;
619 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip, cur->bc_private.b.whichfork);
620 return (struct xfs_btree_block *)ifp->if_broot;
624 * Retrieve the block pointer from the cursor at the given level.
625 * This may be an inode btree root or from a buffer.
627 STATIC struct xfs_btree_block * /* generic btree block pointer */
629 struct xfs_btree_cur *cur, /* btree cursor */
630 int level, /* level in btree */
631 struct xfs_buf **bpp) /* buffer containing the block */
633 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
634 (level == cur->bc_nlevels - 1)) {
636 return xfs_btree_get_iroot(cur);
639 *bpp = cur->bc_bufs[level];
640 return XFS_BUF_TO_BLOCK(*bpp);
644 * Get a buffer for the block, return it with no data read.
645 * Long-form addressing.
647 xfs_buf_t * /* buffer for fsbno */
649 xfs_mount_t *mp, /* file system mount point */
650 xfs_trans_t *tp, /* transaction pointer */
651 xfs_fsblock_t fsbno, /* file system block number */
652 uint lock) /* lock flags for get_buf */
654 xfs_daddr_t d; /* real disk block address */
656 ASSERT(fsbno != NULLFSBLOCK);
657 d = XFS_FSB_TO_DADDR(mp, fsbno);
658 return xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
662 * Get a buffer for the block, return it with no data read.
663 * Short-form addressing.
665 xfs_buf_t * /* buffer for agno/agbno */
667 xfs_mount_t *mp, /* file system mount point */
668 xfs_trans_t *tp, /* transaction pointer */
669 xfs_agnumber_t agno, /* allocation group number */
670 xfs_agblock_t agbno, /* allocation group block number */
671 uint lock) /* lock flags for get_buf */
673 xfs_daddr_t d; /* real disk block address */
675 ASSERT(agno != NULLAGNUMBER);
676 ASSERT(agbno != NULLAGBLOCK);
677 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
678 return xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
682 * Check for the cursor referring to the last block at the given level.
684 int /* 1=is last block, 0=not last block */
685 xfs_btree_islastblock(
686 xfs_btree_cur_t *cur, /* btree cursor */
687 int level) /* level to check */
689 struct xfs_btree_block *block; /* generic btree block pointer */
690 xfs_buf_t *bp; /* buffer containing block */
692 block = xfs_btree_get_block(cur, level, &bp);
693 xfs_btree_check_block(cur, block, level, bp);
694 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
695 return block->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK);
697 return block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK);
701 * Change the cursor to point to the first record at the given level.
702 * Other levels are unaffected.
704 STATIC int /* success=1, failure=0 */
706 xfs_btree_cur_t *cur, /* btree cursor */
707 int level) /* level to change */
709 struct xfs_btree_block *block; /* generic btree block pointer */
710 xfs_buf_t *bp; /* buffer containing block */
713 * Get the block pointer for this level.
715 block = xfs_btree_get_block(cur, level, &bp);
716 xfs_btree_check_block(cur, block, level, bp);
718 * It's empty, there is no such record.
720 if (!block->bb_numrecs)
723 * Set the ptr value to 1, that's the first record/key.
725 cur->bc_ptrs[level] = 1;
730 * Change the cursor to point to the last record in the current block
731 * at the given level. Other levels are unaffected.
733 STATIC int /* success=1, failure=0 */
735 xfs_btree_cur_t *cur, /* btree cursor */
736 int level) /* level to change */
738 struct xfs_btree_block *block; /* generic btree block pointer */
739 xfs_buf_t *bp; /* buffer containing block */
742 * Get the block pointer for this level.
744 block = xfs_btree_get_block(cur, level, &bp);
745 xfs_btree_check_block(cur, block, level, bp);
747 * It's empty, there is no such record.
749 if (!block->bb_numrecs)
752 * Set the ptr value to numrecs, that's the last record/key.
754 cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
759 * Compute first and last byte offsets for the fields given.
760 * Interprets the offsets table, which contains struct field offsets.
764 __int64_t fields, /* bitmask of fields */
765 const short *offsets, /* table of field offsets */
766 int nbits, /* number of bits to inspect */
767 int *first, /* output: first byte offset */
768 int *last) /* output: last byte offset */
770 int i; /* current bit number */
771 __int64_t imask; /* mask for current bit number */
775 * Find the lowest bit, so the first byte offset.
777 for (i = 0, imask = 1LL; ; i++, imask <<= 1) {
778 if (imask & fields) {
784 * Find the highest bit, so the last byte offset.
786 for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) {
787 if (imask & fields) {
788 *last = offsets[i + 1] - 1;
795 * Get a buffer for the block, return it read in.
796 * Long-form addressing.
800 struct xfs_mount *mp, /* file system mount point */
801 struct xfs_trans *tp, /* transaction pointer */
802 xfs_fsblock_t fsbno, /* file system block number */
803 uint lock, /* lock flags for read_buf */
804 struct xfs_buf **bpp, /* buffer for fsbno */
805 int refval, /* ref count value for buffer */
806 const struct xfs_buf_ops *ops)
808 struct xfs_buf *bp; /* return value */
809 xfs_daddr_t d; /* real disk block address */
812 ASSERT(fsbno != NULLFSBLOCK);
813 d = XFS_FSB_TO_DADDR(mp, fsbno);
814 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
815 mp->m_bsize, lock, &bp, ops);
819 xfs_buf_set_ref(bp, refval);
825 * Read-ahead the block, don't wait for it, don't return a buffer.
826 * Long-form addressing.
830 xfs_btree_reada_bufl(
831 struct xfs_mount *mp, /* file system mount point */
832 xfs_fsblock_t fsbno, /* file system block number */
833 xfs_extlen_t count, /* count of filesystem blocks */
834 const struct xfs_buf_ops *ops)
838 ASSERT(fsbno != NULLFSBLOCK);
839 d = XFS_FSB_TO_DADDR(mp, fsbno);
840 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
844 * Read-ahead the block, don't wait for it, don't return a buffer.
845 * Short-form addressing.
849 xfs_btree_reada_bufs(
850 struct xfs_mount *mp, /* file system mount point */
851 xfs_agnumber_t agno, /* allocation group number */
852 xfs_agblock_t agbno, /* allocation group block number */
853 xfs_extlen_t count, /* count of filesystem blocks */
854 const struct xfs_buf_ops *ops)
858 ASSERT(agno != NULLAGNUMBER);
859 ASSERT(agbno != NULLAGBLOCK);
860 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
861 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
865 xfs_btree_readahead_lblock(
866 struct xfs_btree_cur *cur,
868 struct xfs_btree_block *block)
871 xfs_fsblock_t left = be64_to_cpu(block->bb_u.l.bb_leftsib);
872 xfs_fsblock_t right = be64_to_cpu(block->bb_u.l.bb_rightsib);
874 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLFSBLOCK) {
875 xfs_btree_reada_bufl(cur->bc_mp, left, 1,
876 cur->bc_ops->buf_ops);
880 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLFSBLOCK) {
881 xfs_btree_reada_bufl(cur->bc_mp, right, 1,
882 cur->bc_ops->buf_ops);
890 xfs_btree_readahead_sblock(
891 struct xfs_btree_cur *cur,
893 struct xfs_btree_block *block)
896 xfs_agblock_t left = be32_to_cpu(block->bb_u.s.bb_leftsib);
897 xfs_agblock_t right = be32_to_cpu(block->bb_u.s.bb_rightsib);
900 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
901 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
902 left, 1, cur->bc_ops->buf_ops);
906 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
907 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
908 right, 1, cur->bc_ops->buf_ops);
916 * Read-ahead btree blocks, at the given level.
917 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
921 struct xfs_btree_cur *cur, /* btree cursor */
922 int lev, /* level in btree */
923 int lr) /* left/right bits */
925 struct xfs_btree_block *block;
928 * No readahead needed if we are at the root level and the
929 * btree root is stored in the inode.
931 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
932 (lev == cur->bc_nlevels - 1))
935 if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
938 cur->bc_ra[lev] |= lr;
939 block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
941 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
942 return xfs_btree_readahead_lblock(cur, lr, block);
943 return xfs_btree_readahead_sblock(cur, lr, block);
947 xfs_btree_ptr_to_daddr(
948 struct xfs_btree_cur *cur,
949 union xfs_btree_ptr *ptr)
951 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
952 ASSERT(ptr->l != cpu_to_be64(NULLFSBLOCK));
954 return XFS_FSB_TO_DADDR(cur->bc_mp, be64_to_cpu(ptr->l));
956 ASSERT(cur->bc_private.a.agno != NULLAGNUMBER);
957 ASSERT(ptr->s != cpu_to_be32(NULLAGBLOCK));
959 return XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_private.a.agno,
960 be32_to_cpu(ptr->s));
965 * Readahead @count btree blocks at the given @ptr location.
967 * We don't need to care about long or short form btrees here as we have a
968 * method of converting the ptr directly to a daddr available to us.
971 xfs_btree_readahead_ptr(
972 struct xfs_btree_cur *cur,
973 union xfs_btree_ptr *ptr,
976 xfs_buf_readahead(cur->bc_mp->m_ddev_targp,
977 xfs_btree_ptr_to_daddr(cur, ptr),
978 cur->bc_mp->m_bsize * count, cur->bc_ops->buf_ops);
982 * Set the buffer for level "lev" in the cursor to bp, releasing
983 * any previous buffer.
987 xfs_btree_cur_t *cur, /* btree cursor */
988 int lev, /* level in btree */
989 xfs_buf_t *bp) /* new buffer to set */
991 struct xfs_btree_block *b; /* btree block */
993 if (cur->bc_bufs[lev])
994 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
995 cur->bc_bufs[lev] = bp;
998 b = XFS_BUF_TO_BLOCK(bp);
999 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1000 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK))
1001 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1002 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK))
1003 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1005 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
1006 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1007 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
1008 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1013 xfs_btree_ptr_is_null(
1014 struct xfs_btree_cur *cur,
1015 union xfs_btree_ptr *ptr)
1017 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1018 return ptr->l == cpu_to_be64(NULLFSBLOCK);
1020 return ptr->s == cpu_to_be32(NULLAGBLOCK);
1024 xfs_btree_set_ptr_null(
1025 struct xfs_btree_cur *cur,
1026 union xfs_btree_ptr *ptr)
1028 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1029 ptr->l = cpu_to_be64(NULLFSBLOCK);
1031 ptr->s = cpu_to_be32(NULLAGBLOCK);
1035 * Get/set/init sibling pointers
1038 xfs_btree_get_sibling(
1039 struct xfs_btree_cur *cur,
1040 struct xfs_btree_block *block,
1041 union xfs_btree_ptr *ptr,
1044 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1046 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1047 if (lr == XFS_BB_RIGHTSIB)
1048 ptr->l = block->bb_u.l.bb_rightsib;
1050 ptr->l = block->bb_u.l.bb_leftsib;
1052 if (lr == XFS_BB_RIGHTSIB)
1053 ptr->s = block->bb_u.s.bb_rightsib;
1055 ptr->s = block->bb_u.s.bb_leftsib;
1060 xfs_btree_set_sibling(
1061 struct xfs_btree_cur *cur,
1062 struct xfs_btree_block *block,
1063 union xfs_btree_ptr *ptr,
1066 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1068 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1069 if (lr == XFS_BB_RIGHTSIB)
1070 block->bb_u.l.bb_rightsib = ptr->l;
1072 block->bb_u.l.bb_leftsib = ptr->l;
1074 if (lr == XFS_BB_RIGHTSIB)
1075 block->bb_u.s.bb_rightsib = ptr->s;
1077 block->bb_u.s.bb_leftsib = ptr->s;
1082 xfs_btree_init_block_int(
1083 struct xfs_mount *mp,
1084 struct xfs_btree_block *buf,
1092 buf->bb_magic = cpu_to_be32(magic);
1093 buf->bb_level = cpu_to_be16(level);
1094 buf->bb_numrecs = cpu_to_be16(numrecs);
1096 if (flags & XFS_BTREE_LONG_PTRS) {
1097 buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK);
1098 buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK);
1099 if (flags & XFS_BTREE_CRC_BLOCKS) {
1100 buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
1101 buf->bb_u.l.bb_owner = cpu_to_be64(owner);
1102 uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid);
1103 buf->bb_u.l.bb_pad = 0;
1104 buf->bb_u.l.bb_lsn = 0;
1107 /* owner is a 32 bit value on short blocks */
1108 __u32 __owner = (__u32)owner;
1110 buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1111 buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1112 if (flags & XFS_BTREE_CRC_BLOCKS) {
1113 buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
1114 buf->bb_u.s.bb_owner = cpu_to_be32(__owner);
1115 uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid);
1116 buf->bb_u.s.bb_lsn = 0;
1122 xfs_btree_init_block(
1123 struct xfs_mount *mp,
1131 xfs_btree_init_block_int(mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1132 magic, level, numrecs, owner, flags);
1136 xfs_btree_init_block_cur(
1137 struct xfs_btree_cur *cur,
1145 * we can pull the owner from the cursor right now as the different
1146 * owners align directly with the pointer size of the btree. This may
1147 * change in future, but is safe for current users of the generic btree
1150 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1151 owner = cur->bc_private.b.ip->i_ino;
1153 owner = cur->bc_private.a.agno;
1155 xfs_btree_init_block_int(cur->bc_mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1156 xfs_btree_magic(cur), level, numrecs,
1157 owner, cur->bc_flags);
1161 * Return true if ptr is the last record in the btree and
1162 * we need to track updates to this record. The decision
1163 * will be further refined in the update_lastrec method.
1166 xfs_btree_is_lastrec(
1167 struct xfs_btree_cur *cur,
1168 struct xfs_btree_block *block,
1171 union xfs_btree_ptr ptr;
1175 if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
1178 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1179 if (!xfs_btree_ptr_is_null(cur, &ptr))
1185 xfs_btree_buf_to_ptr(
1186 struct xfs_btree_cur *cur,
1188 union xfs_btree_ptr *ptr)
1190 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1191 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1194 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
1201 struct xfs_btree_cur *cur,
1204 switch (cur->bc_btnum) {
1207 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
1210 case XFS_BTNUM_FINO:
1211 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
1213 case XFS_BTNUM_BMAP:
1214 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
1216 case XFS_BTNUM_RMAP:
1217 xfs_buf_set_ref(bp, XFS_RMAP_BTREE_REF);
1225 xfs_btree_get_buf_block(
1226 struct xfs_btree_cur *cur,
1227 union xfs_btree_ptr *ptr,
1229 struct xfs_btree_block **block,
1230 struct xfs_buf **bpp)
1232 struct xfs_mount *mp = cur->bc_mp;
1235 /* need to sort out how callers deal with failures first */
1236 ASSERT(!(flags & XBF_TRYLOCK));
1238 d = xfs_btree_ptr_to_daddr(cur, ptr);
1239 *bpp = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d,
1240 mp->m_bsize, flags);
1245 (*bpp)->b_ops = cur->bc_ops->buf_ops;
1246 *block = XFS_BUF_TO_BLOCK(*bpp);
1251 * Read in the buffer at the given ptr and return the buffer and
1252 * the block pointer within the buffer.
1255 xfs_btree_read_buf_block(
1256 struct xfs_btree_cur *cur,
1257 union xfs_btree_ptr *ptr,
1259 struct xfs_btree_block **block,
1260 struct xfs_buf **bpp)
1262 struct xfs_mount *mp = cur->bc_mp;
1266 /* need to sort out how callers deal with failures first */
1267 ASSERT(!(flags & XBF_TRYLOCK));
1269 d = xfs_btree_ptr_to_daddr(cur, ptr);
1270 error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
1271 mp->m_bsize, flags, bpp,
1272 cur->bc_ops->buf_ops);
1276 xfs_btree_set_refs(cur, *bpp);
1277 *block = XFS_BUF_TO_BLOCK(*bpp);
1282 * Copy keys from one btree block to another.
1285 xfs_btree_copy_keys(
1286 struct xfs_btree_cur *cur,
1287 union xfs_btree_key *dst_key,
1288 union xfs_btree_key *src_key,
1291 ASSERT(numkeys >= 0);
1292 memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1296 * Copy records from one btree block to another.
1299 xfs_btree_copy_recs(
1300 struct xfs_btree_cur *cur,
1301 union xfs_btree_rec *dst_rec,
1302 union xfs_btree_rec *src_rec,
1305 ASSERT(numrecs >= 0);
1306 memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1310 * Copy block pointers from one btree block to another.
1313 xfs_btree_copy_ptrs(
1314 struct xfs_btree_cur *cur,
1315 union xfs_btree_ptr *dst_ptr,
1316 union xfs_btree_ptr *src_ptr,
1319 ASSERT(numptrs >= 0);
1320 memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1324 * Shift keys one index left/right inside a single btree block.
1327 xfs_btree_shift_keys(
1328 struct xfs_btree_cur *cur,
1329 union xfs_btree_key *key,
1335 ASSERT(numkeys >= 0);
1336 ASSERT(dir == 1 || dir == -1);
1338 dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1339 memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1343 * Shift records one index left/right inside a single btree block.
1346 xfs_btree_shift_recs(
1347 struct xfs_btree_cur *cur,
1348 union xfs_btree_rec *rec,
1354 ASSERT(numrecs >= 0);
1355 ASSERT(dir == 1 || dir == -1);
1357 dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1358 memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1362 * Shift block pointers one index left/right inside a single btree block.
1365 xfs_btree_shift_ptrs(
1366 struct xfs_btree_cur *cur,
1367 union xfs_btree_ptr *ptr,
1373 ASSERT(numptrs >= 0);
1374 ASSERT(dir == 1 || dir == -1);
1376 dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1377 memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1381 * Log key values from the btree block.
1385 struct xfs_btree_cur *cur,
1390 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1391 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1394 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1395 xfs_trans_log_buf(cur->bc_tp, bp,
1396 xfs_btree_key_offset(cur, first),
1397 xfs_btree_key_offset(cur, last + 1) - 1);
1399 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1400 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1403 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1407 * Log record values from the btree block.
1411 struct xfs_btree_cur *cur,
1416 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1417 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1419 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1420 xfs_trans_log_buf(cur->bc_tp, bp,
1421 xfs_btree_rec_offset(cur, first),
1422 xfs_btree_rec_offset(cur, last + 1) - 1);
1424 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1428 * Log block pointer fields from a btree block (nonleaf).
1432 struct xfs_btree_cur *cur, /* btree cursor */
1433 struct xfs_buf *bp, /* buffer containing btree block */
1434 int first, /* index of first pointer to log */
1435 int last) /* index of last pointer to log */
1437 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1438 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1441 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
1442 int level = xfs_btree_get_level(block);
1444 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1445 xfs_trans_log_buf(cur->bc_tp, bp,
1446 xfs_btree_ptr_offset(cur, first, level),
1447 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1449 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1450 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1453 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1457 * Log fields from a btree block header.
1460 xfs_btree_log_block(
1461 struct xfs_btree_cur *cur, /* btree cursor */
1462 struct xfs_buf *bp, /* buffer containing btree block */
1463 int fields) /* mask of fields: XFS_BB_... */
1465 int first; /* first byte offset logged */
1466 int last; /* last byte offset logged */
1467 static const short soffsets[] = { /* table of offsets (short) */
1468 offsetof(struct xfs_btree_block, bb_magic),
1469 offsetof(struct xfs_btree_block, bb_level),
1470 offsetof(struct xfs_btree_block, bb_numrecs),
1471 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1472 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1473 offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1474 offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1475 offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1476 offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1477 offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1478 XFS_BTREE_SBLOCK_CRC_LEN
1480 static const short loffsets[] = { /* table of offsets (long) */
1481 offsetof(struct xfs_btree_block, bb_magic),
1482 offsetof(struct xfs_btree_block, bb_level),
1483 offsetof(struct xfs_btree_block, bb_numrecs),
1484 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1485 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1486 offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1487 offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1488 offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1489 offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1490 offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1491 offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1492 XFS_BTREE_LBLOCK_CRC_LEN
1495 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1496 XFS_BTREE_TRACE_ARGBI(cur, bp, fields);
1501 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
1503 * We don't log the CRC when updating a btree
1504 * block but instead recreate it during log
1505 * recovery. As the log buffers have checksums
1506 * of their own this is safe and avoids logging a crc
1507 * update in a lot of places.
1509 if (fields == XFS_BB_ALL_BITS)
1510 fields = XFS_BB_ALL_BITS_CRC;
1511 nbits = XFS_BB_NUM_BITS_CRC;
1513 nbits = XFS_BB_NUM_BITS;
1515 xfs_btree_offsets(fields,
1516 (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1517 loffsets : soffsets,
1518 nbits, &first, &last);
1519 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1520 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1522 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1523 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1526 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1530 * Increment cursor by one record at the level.
1531 * For nonzero levels the leaf-ward information is untouched.
1534 xfs_btree_increment(
1535 struct xfs_btree_cur *cur,
1537 int *stat) /* success/failure */
1539 struct xfs_btree_block *block;
1540 union xfs_btree_ptr ptr;
1542 int error; /* error return value */
1545 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1546 XFS_BTREE_TRACE_ARGI(cur, level);
1548 ASSERT(level < cur->bc_nlevels);
1550 /* Read-ahead to the right at this level. */
1551 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1553 /* Get a pointer to the btree block. */
1554 block = xfs_btree_get_block(cur, level, &bp);
1557 error = xfs_btree_check_block(cur, block, level, bp);
1562 /* We're done if we remain in the block after the increment. */
1563 if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
1566 /* Fail if we just went off the right edge of the tree. */
1567 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1568 if (xfs_btree_ptr_is_null(cur, &ptr))
1571 XFS_BTREE_STATS_INC(cur, increment);
1574 * March up the tree incrementing pointers.
1575 * Stop when we don't go off the right edge of a block.
1577 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1578 block = xfs_btree_get_block(cur, lev, &bp);
1581 error = xfs_btree_check_block(cur, block, lev, bp);
1586 if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
1589 /* Read-ahead the right block for the next loop. */
1590 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1594 * If we went off the root then we are either seriously
1595 * confused or have the tree root in an inode.
1597 if (lev == cur->bc_nlevels) {
1598 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1601 error = -EFSCORRUPTED;
1604 ASSERT(lev < cur->bc_nlevels);
1607 * Now walk back down the tree, fixing up the cursor's buffer
1608 * pointers and key numbers.
1610 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1611 union xfs_btree_ptr *ptrp;
1613 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1615 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1619 xfs_btree_setbuf(cur, lev, bp);
1620 cur->bc_ptrs[lev] = 1;
1623 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1628 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1633 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1638 * Decrement cursor by one record at the level.
1639 * For nonzero levels the leaf-ward information is untouched.
1642 xfs_btree_decrement(
1643 struct xfs_btree_cur *cur,
1645 int *stat) /* success/failure */
1647 struct xfs_btree_block *block;
1649 int error; /* error return value */
1651 union xfs_btree_ptr ptr;
1653 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1654 XFS_BTREE_TRACE_ARGI(cur, level);
1656 ASSERT(level < cur->bc_nlevels);
1658 /* Read-ahead to the left at this level. */
1659 xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1661 /* We're done if we remain in the block after the decrement. */
1662 if (--cur->bc_ptrs[level] > 0)
1665 /* Get a pointer to the btree block. */
1666 block = xfs_btree_get_block(cur, level, &bp);
1669 error = xfs_btree_check_block(cur, block, level, bp);
1674 /* Fail if we just went off the left edge of the tree. */
1675 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1676 if (xfs_btree_ptr_is_null(cur, &ptr))
1679 XFS_BTREE_STATS_INC(cur, decrement);
1682 * March up the tree decrementing pointers.
1683 * Stop when we don't go off the left edge of a block.
1685 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1686 if (--cur->bc_ptrs[lev] > 0)
1688 /* Read-ahead the left block for the next loop. */
1689 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1693 * If we went off the root then we are seriously confused.
1694 * or the root of the tree is in an inode.
1696 if (lev == cur->bc_nlevels) {
1697 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1700 error = -EFSCORRUPTED;
1703 ASSERT(lev < cur->bc_nlevels);
1706 * Now walk back down the tree, fixing up the cursor's buffer
1707 * pointers and key numbers.
1709 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1710 union xfs_btree_ptr *ptrp;
1712 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1714 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1717 xfs_btree_setbuf(cur, lev, bp);
1718 cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
1721 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1726 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1731 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1736 xfs_btree_lookup_get_block(
1737 struct xfs_btree_cur *cur, /* btree cursor */
1738 int level, /* level in the btree */
1739 union xfs_btree_ptr *pp, /* ptr to btree block */
1740 struct xfs_btree_block **blkp) /* return btree block */
1742 struct xfs_buf *bp; /* buffer pointer for btree block */
1745 /* special case the root block if in an inode */
1746 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1747 (level == cur->bc_nlevels - 1)) {
1748 *blkp = xfs_btree_get_iroot(cur);
1753 * If the old buffer at this level for the disk address we are
1754 * looking for re-use it.
1756 * Otherwise throw it away and get a new one.
1758 bp = cur->bc_bufs[level];
1759 if (bp && XFS_BUF_ADDR(bp) == xfs_btree_ptr_to_daddr(cur, pp)) {
1760 *blkp = XFS_BUF_TO_BLOCK(bp);
1764 error = xfs_btree_read_buf_block(cur, pp, 0, blkp, &bp);
1768 xfs_btree_setbuf(cur, level, bp);
1773 * Get current search key. For level 0 we don't actually have a key
1774 * structure so we make one up from the record. For all other levels
1775 * we just return the right key.
1777 STATIC union xfs_btree_key *
1778 xfs_lookup_get_search_key(
1779 struct xfs_btree_cur *cur,
1782 struct xfs_btree_block *block,
1783 union xfs_btree_key *kp)
1786 cur->bc_ops->init_key_from_rec(kp,
1787 xfs_btree_rec_addr(cur, keyno, block));
1791 return xfs_btree_key_addr(cur, keyno, block);
1795 * Lookup the record. The cursor is made to point to it, based on dir.
1796 * stat is set to 0 if can't find any such record, 1 for success.
1800 struct xfs_btree_cur *cur, /* btree cursor */
1801 xfs_lookup_t dir, /* <=, ==, or >= */
1802 int *stat) /* success/failure */
1804 struct xfs_btree_block *block; /* current btree block */
1805 __int64_t diff; /* difference for the current key */
1806 int error; /* error return value */
1807 int keyno; /* current key number */
1808 int level; /* level in the btree */
1809 union xfs_btree_ptr *pp; /* ptr to btree block */
1810 union xfs_btree_ptr ptr; /* ptr to btree block */
1812 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1813 XFS_BTREE_TRACE_ARGI(cur, dir);
1815 XFS_BTREE_STATS_INC(cur, lookup);
1820 /* initialise start pointer from cursor */
1821 cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1825 * Iterate over each level in the btree, starting at the root.
1826 * For each level above the leaves, find the key we need, based
1827 * on the lookup record, then follow the corresponding block
1828 * pointer down to the next level.
1830 for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1831 /* Get the block we need to do the lookup on. */
1832 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1838 * If we already had a key match at a higher level, we
1839 * know we need to use the first entry in this block.
1843 /* Otherwise search this block. Do a binary search. */
1845 int high; /* high entry number */
1846 int low; /* low entry number */
1848 /* Set low and high entry numbers, 1-based. */
1850 high = xfs_btree_get_numrecs(block);
1852 /* Block is empty, must be an empty leaf. */
1853 ASSERT(level == 0 && cur->bc_nlevels == 1);
1855 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1856 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1861 /* Binary search the block. */
1862 while (low <= high) {
1863 union xfs_btree_key key;
1864 union xfs_btree_key *kp;
1866 XFS_BTREE_STATS_INC(cur, compare);
1868 /* keyno is average of low and high. */
1869 keyno = (low + high) >> 1;
1871 /* Get current search key */
1872 kp = xfs_lookup_get_search_key(cur, level,
1873 keyno, block, &key);
1876 * Compute difference to get next direction:
1877 * - less than, move right
1878 * - greater than, move left
1879 * - equal, we're done
1881 diff = cur->bc_ops->key_diff(cur, kp);
1892 * If there are more levels, set up for the next level
1893 * by getting the block number and filling in the cursor.
1897 * If we moved left, need the previous key number,
1898 * unless there isn't one.
1900 if (diff > 0 && --keyno < 1)
1902 pp = xfs_btree_ptr_addr(cur, keyno, block);
1905 error = xfs_btree_check_ptr(cur, pp, 0, level);
1909 cur->bc_ptrs[level] = keyno;
1913 /* Done with the search. See if we need to adjust the results. */
1914 if (dir != XFS_LOOKUP_LE && diff < 0) {
1917 * If ge search and we went off the end of the block, but it's
1918 * not the last block, we're in the wrong block.
1920 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1921 if (dir == XFS_LOOKUP_GE &&
1922 keyno > xfs_btree_get_numrecs(block) &&
1923 !xfs_btree_ptr_is_null(cur, &ptr)) {
1926 cur->bc_ptrs[0] = keyno;
1927 error = xfs_btree_increment(cur, 0, &i);
1930 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1931 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1935 } else if (dir == XFS_LOOKUP_LE && diff > 0)
1937 cur->bc_ptrs[0] = keyno;
1939 /* Return if we succeeded or not. */
1940 if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
1942 else if (dir != XFS_LOOKUP_EQ || diff == 0)
1946 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1950 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1954 /* Find the high key storage area from a regular key. */
1955 STATIC union xfs_btree_key *
1956 xfs_btree_high_key_from_key(
1957 struct xfs_btree_cur *cur,
1958 union xfs_btree_key *key)
1960 ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
1961 return (union xfs_btree_key *)((char *)key +
1962 (cur->bc_ops->key_len / 2));
1965 /* Determine the low (and high if overlapped) keys of a leaf block */
1967 xfs_btree_get_leaf_keys(
1968 struct xfs_btree_cur *cur,
1969 struct xfs_btree_block *block,
1970 union xfs_btree_key *key)
1972 union xfs_btree_key max_hkey;
1973 union xfs_btree_key hkey;
1974 union xfs_btree_rec *rec;
1975 union xfs_btree_key *high;
1978 rec = xfs_btree_rec_addr(cur, 1, block);
1979 cur->bc_ops->init_key_from_rec(key, rec);
1981 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
1983 cur->bc_ops->init_high_key_from_rec(&max_hkey, rec);
1984 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
1985 rec = xfs_btree_rec_addr(cur, n, block);
1986 cur->bc_ops->init_high_key_from_rec(&hkey, rec);
1987 if (cur->bc_ops->diff_two_keys(cur, &hkey, &max_hkey)
1992 high = xfs_btree_high_key_from_key(cur, key);
1993 memcpy(high, &max_hkey, cur->bc_ops->key_len / 2);
1997 /* Determine the low (and high if overlapped) keys of a node block */
1999 xfs_btree_get_node_keys(
2000 struct xfs_btree_cur *cur,
2001 struct xfs_btree_block *block,
2002 union xfs_btree_key *key)
2004 union xfs_btree_key *hkey;
2005 union xfs_btree_key *max_hkey;
2006 union xfs_btree_key *high;
2009 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2010 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2011 cur->bc_ops->key_len / 2);
2013 max_hkey = xfs_btree_high_key_addr(cur, 1, block);
2014 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2015 hkey = xfs_btree_high_key_addr(cur, n, block);
2016 if (cur->bc_ops->diff_two_keys(cur, hkey, max_hkey) > 0)
2020 high = xfs_btree_high_key_from_key(cur, key);
2021 memcpy(high, max_hkey, cur->bc_ops->key_len / 2);
2023 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2024 cur->bc_ops->key_len);
2028 /* Derive the keys for any btree block. */
2031 struct xfs_btree_cur *cur,
2032 struct xfs_btree_block *block,
2033 union xfs_btree_key *key)
2035 if (be16_to_cpu(block->bb_level) == 0)
2036 xfs_btree_get_leaf_keys(cur, block, key);
2038 xfs_btree_get_node_keys(cur, block, key);
2042 * Decide if we need to update the parent keys of a btree block. For
2043 * a standard btree this is only necessary if we're updating the first
2044 * record/key. For an overlapping btree, we must always update the
2045 * keys because the highest key can be in any of the records or keys
2049 xfs_btree_needs_key_update(
2050 struct xfs_btree_cur *cur,
2053 return (cur->bc_flags & XFS_BTREE_OVERLAPPING) || ptr == 1;
2057 * Update the low and high parent keys of the given level, progressing
2058 * towards the root. If force_all is false, stop if the keys for a given
2059 * level do not need updating.
2062 __xfs_btree_updkeys(
2063 struct xfs_btree_cur *cur,
2065 struct xfs_btree_block *block,
2066 struct xfs_buf *bp0,
2069 union xfs_btree_bigkey key; /* keys from current level */
2070 union xfs_btree_key *lkey; /* keys from the next level up */
2071 union xfs_btree_key *hkey;
2072 union xfs_btree_key *nlkey; /* keys from the next level up */
2073 union xfs_btree_key *nhkey;
2077 ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2079 /* Exit if there aren't any parent levels to update. */
2080 if (level + 1 >= cur->bc_nlevels)
2083 trace_xfs_btree_updkeys(cur, level, bp0);
2085 lkey = (union xfs_btree_key *)&key;
2086 hkey = xfs_btree_high_key_from_key(cur, lkey);
2087 xfs_btree_get_keys(cur, block, lkey);
2088 for (level++; level < cur->bc_nlevels; level++) {
2092 block = xfs_btree_get_block(cur, level, &bp);
2093 trace_xfs_btree_updkeys(cur, level, bp);
2095 error = xfs_btree_check_block(cur, block, level, bp);
2097 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2101 ptr = cur->bc_ptrs[level];
2102 nlkey = xfs_btree_key_addr(cur, ptr, block);
2103 nhkey = xfs_btree_high_key_addr(cur, ptr, block);
2105 !(cur->bc_ops->diff_two_keys(cur, nlkey, lkey) != 0 ||
2106 cur->bc_ops->diff_two_keys(cur, nhkey, hkey) != 0))
2108 xfs_btree_copy_keys(cur, nlkey, lkey, 1);
2109 xfs_btree_log_keys(cur, bp, ptr, ptr);
2110 if (level + 1 >= cur->bc_nlevels)
2112 xfs_btree_get_node_keys(cur, block, lkey);
2118 /* Update all the keys from some level in cursor back to the root. */
2120 xfs_btree_updkeys_force(
2121 struct xfs_btree_cur *cur,
2125 struct xfs_btree_block *block;
2127 block = xfs_btree_get_block(cur, level, &bp);
2128 return __xfs_btree_updkeys(cur, level, block, bp, true);
2132 * Update the parent keys of the given level, progressing towards the root.
2135 xfs_btree_update_keys(
2136 struct xfs_btree_cur *cur,
2139 struct xfs_btree_block *block;
2141 union xfs_btree_key *kp;
2142 union xfs_btree_key key;
2147 block = xfs_btree_get_block(cur, level, &bp);
2148 if (cur->bc_flags & XFS_BTREE_OVERLAPPING)
2149 return __xfs_btree_updkeys(cur, level, block, bp, false);
2151 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2152 XFS_BTREE_TRACE_ARGIK(cur, level, keyp);
2155 * Go up the tree from this level toward the root.
2156 * At each level, update the key value to the value input.
2157 * Stop when we reach a level where the cursor isn't pointing
2158 * at the first entry in the block.
2160 xfs_btree_get_keys(cur, block, &key);
2161 for (level++, ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
2165 block = xfs_btree_get_block(cur, level, &bp);
2167 error = xfs_btree_check_block(cur, block, level, bp);
2169 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2173 ptr = cur->bc_ptrs[level];
2174 kp = xfs_btree_key_addr(cur, ptr, block);
2175 xfs_btree_copy_keys(cur, kp, &key, 1);
2176 xfs_btree_log_keys(cur, bp, ptr, ptr);
2179 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2184 * Update the record referred to by cur to the value in the
2185 * given record. This either works (return 0) or gets an
2186 * EFSCORRUPTED error.
2190 struct xfs_btree_cur *cur,
2191 union xfs_btree_rec *rec)
2193 struct xfs_btree_block *block;
2197 union xfs_btree_rec *rp;
2199 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2200 XFS_BTREE_TRACE_ARGR(cur, rec);
2202 /* Pick up the current block. */
2203 block = xfs_btree_get_block(cur, 0, &bp);
2206 error = xfs_btree_check_block(cur, block, 0, bp);
2210 /* Get the address of the rec to be updated. */
2211 ptr = cur->bc_ptrs[0];
2212 rp = xfs_btree_rec_addr(cur, ptr, block);
2214 /* Fill in the new contents and log them. */
2215 xfs_btree_copy_recs(cur, rp, rec, 1);
2216 xfs_btree_log_recs(cur, bp, ptr, ptr);
2219 * If we are tracking the last record in the tree and
2220 * we are at the far right edge of the tree, update it.
2222 if (xfs_btree_is_lastrec(cur, block, 0)) {
2223 cur->bc_ops->update_lastrec(cur, block, rec,
2224 ptr, LASTREC_UPDATE);
2227 /* Pass new key value up to our parent. */
2228 if (xfs_btree_needs_key_update(cur, ptr)) {
2229 error = xfs_btree_update_keys(cur, 0);
2234 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2238 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2243 * Move 1 record left from cur/level if possible.
2244 * Update cur to reflect the new path.
2246 STATIC int /* error */
2248 struct xfs_btree_cur *cur,
2250 int *stat) /* success/failure */
2252 struct xfs_buf *lbp; /* left buffer pointer */
2253 struct xfs_btree_block *left; /* left btree block */
2254 int lrecs; /* left record count */
2255 struct xfs_buf *rbp; /* right buffer pointer */
2256 struct xfs_btree_block *right; /* right btree block */
2257 struct xfs_btree_cur *tcur; /* temporary btree cursor */
2258 int rrecs; /* right record count */
2259 union xfs_btree_ptr lptr; /* left btree pointer */
2260 union xfs_btree_key *rkp = NULL; /* right btree key */
2261 union xfs_btree_ptr *rpp = NULL; /* right address pointer */
2262 union xfs_btree_rec *rrp = NULL; /* right record pointer */
2263 int error; /* error return value */
2266 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2267 XFS_BTREE_TRACE_ARGI(cur, level);
2269 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2270 level == cur->bc_nlevels - 1)
2273 /* Set up variables for this block as "right". */
2274 right = xfs_btree_get_block(cur, level, &rbp);
2277 error = xfs_btree_check_block(cur, right, level, rbp);
2282 /* If we've got no left sibling then we can't shift an entry left. */
2283 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2284 if (xfs_btree_ptr_is_null(cur, &lptr))
2288 * If the cursor entry is the one that would be moved, don't
2289 * do it... it's too complicated.
2291 if (cur->bc_ptrs[level] <= 1)
2294 /* Set up the left neighbor as "left". */
2295 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2299 /* If it's full, it can't take another entry. */
2300 lrecs = xfs_btree_get_numrecs(left);
2301 if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2304 rrecs = xfs_btree_get_numrecs(right);
2307 * We add one entry to the left side and remove one for the right side.
2308 * Account for it here, the changes will be updated on disk and logged
2314 XFS_BTREE_STATS_INC(cur, lshift);
2315 XFS_BTREE_STATS_ADD(cur, moves, 1);
2318 * If non-leaf, copy a key and a ptr to the left block.
2319 * Log the changes to the left block.
2322 /* It's a non-leaf. Move keys and pointers. */
2323 union xfs_btree_key *lkp; /* left btree key */
2324 union xfs_btree_ptr *lpp; /* left address pointer */
2326 lkp = xfs_btree_key_addr(cur, lrecs, left);
2327 rkp = xfs_btree_key_addr(cur, 1, right);
2329 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2330 rpp = xfs_btree_ptr_addr(cur, 1, right);
2332 error = xfs_btree_check_ptr(cur, rpp, 0, level);
2336 xfs_btree_copy_keys(cur, lkp, rkp, 1);
2337 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2339 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2340 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2342 ASSERT(cur->bc_ops->keys_inorder(cur,
2343 xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2345 /* It's a leaf. Move records. */
2346 union xfs_btree_rec *lrp; /* left record pointer */
2348 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2349 rrp = xfs_btree_rec_addr(cur, 1, right);
2351 xfs_btree_copy_recs(cur, lrp, rrp, 1);
2352 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2354 ASSERT(cur->bc_ops->recs_inorder(cur,
2355 xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2358 xfs_btree_set_numrecs(left, lrecs);
2359 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2361 xfs_btree_set_numrecs(right, rrecs);
2362 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2365 * Slide the contents of right down one entry.
2367 XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2369 /* It's a nonleaf. operate on keys and ptrs */
2371 int i; /* loop index */
2373 for (i = 0; i < rrecs; i++) {
2374 error = xfs_btree_check_ptr(cur, rpp, i + 1, level);
2379 xfs_btree_shift_keys(cur,
2380 xfs_btree_key_addr(cur, 2, right),
2382 xfs_btree_shift_ptrs(cur,
2383 xfs_btree_ptr_addr(cur, 2, right),
2386 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2387 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2389 /* It's a leaf. operate on records */
2390 xfs_btree_shift_recs(cur,
2391 xfs_btree_rec_addr(cur, 2, right),
2393 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2397 * Using a temporary cursor, update the parent key values of the
2398 * block on the left.
2400 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2401 error = xfs_btree_dup_cursor(cur, &tcur);
2404 i = xfs_btree_firstrec(tcur, level);
2405 XFS_WANT_CORRUPTED_GOTO(tcur->bc_mp, i == 1, error0);
2407 error = xfs_btree_decrement(tcur, level, &i);
2411 /* Update the parent high keys of the left block, if needed. */
2412 error = xfs_btree_update_keys(tcur, level);
2416 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2419 /* Update the parent keys of the right block. */
2420 error = xfs_btree_update_keys(cur, level);
2424 /* Slide the cursor value left one. */
2425 cur->bc_ptrs[level]--;
2427 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2432 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2437 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2441 XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2442 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2447 * Move 1 record right from cur/level if possible.
2448 * Update cur to reflect the new path.
2450 STATIC int /* error */
2452 struct xfs_btree_cur *cur,
2454 int *stat) /* success/failure */
2456 struct xfs_buf *lbp; /* left buffer pointer */
2457 struct xfs_btree_block *left; /* left btree block */
2458 struct xfs_buf *rbp; /* right buffer pointer */
2459 struct xfs_btree_block *right; /* right btree block */
2460 struct xfs_btree_cur *tcur; /* temporary btree cursor */
2461 union xfs_btree_ptr rptr; /* right block pointer */
2462 union xfs_btree_key *rkp; /* right btree key */
2463 int rrecs; /* right record count */
2464 int lrecs; /* left record count */
2465 int error; /* error return value */
2466 int i; /* loop counter */
2468 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2469 XFS_BTREE_TRACE_ARGI(cur, level);
2471 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2472 (level == cur->bc_nlevels - 1))
2475 /* Set up variables for this block as "left". */
2476 left = xfs_btree_get_block(cur, level, &lbp);
2479 error = xfs_btree_check_block(cur, left, level, lbp);
2484 /* If we've got no right sibling then we can't shift an entry right. */
2485 xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2486 if (xfs_btree_ptr_is_null(cur, &rptr))
2490 * If the cursor entry is the one that would be moved, don't
2491 * do it... it's too complicated.
2493 lrecs = xfs_btree_get_numrecs(left);
2494 if (cur->bc_ptrs[level] >= lrecs)
2497 /* Set up the right neighbor as "right". */
2498 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2502 /* If it's full, it can't take another entry. */
2503 rrecs = xfs_btree_get_numrecs(right);
2504 if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2507 XFS_BTREE_STATS_INC(cur, rshift);
2508 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2511 * Make a hole at the start of the right neighbor block, then
2512 * copy the last left block entry to the hole.
2515 /* It's a nonleaf. make a hole in the keys and ptrs */
2516 union xfs_btree_key *lkp;
2517 union xfs_btree_ptr *lpp;
2518 union xfs_btree_ptr *rpp;
2520 lkp = xfs_btree_key_addr(cur, lrecs, left);
2521 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2522 rkp = xfs_btree_key_addr(cur, 1, right);
2523 rpp = xfs_btree_ptr_addr(cur, 1, right);
2526 for (i = rrecs - 1; i >= 0; i--) {
2527 error = xfs_btree_check_ptr(cur, rpp, i, level);
2533 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2534 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2537 error = xfs_btree_check_ptr(cur, lpp, 0, level);
2542 /* Now put the new data in, and log it. */
2543 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2544 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2546 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2547 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2549 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2550 xfs_btree_key_addr(cur, 2, right)));
2552 /* It's a leaf. make a hole in the records */
2553 union xfs_btree_rec *lrp;
2554 union xfs_btree_rec *rrp;
2556 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2557 rrp = xfs_btree_rec_addr(cur, 1, right);
2559 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2561 /* Now put the new data in, and log it. */
2562 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2563 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2567 * Decrement and log left's numrecs, bump and log right's numrecs.
2569 xfs_btree_set_numrecs(left, --lrecs);
2570 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2572 xfs_btree_set_numrecs(right, ++rrecs);
2573 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2576 * Using a temporary cursor, update the parent key values of the
2577 * block on the right.
2579 error = xfs_btree_dup_cursor(cur, &tcur);
2582 i = xfs_btree_lastrec(tcur, level);
2583 XFS_WANT_CORRUPTED_GOTO(tcur->bc_mp, i == 1, error0);
2585 error = xfs_btree_increment(tcur, level, &i);
2589 /* Update the parent high keys of the left block, if needed. */
2590 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2591 error = xfs_btree_update_keys(cur, level);
2596 /* Update the parent keys of the right block. */
2597 error = xfs_btree_update_keys(tcur, level);
2601 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2603 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2608 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2613 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2617 XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2618 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2623 * Split cur/level block in half.
2624 * Return new block number and the key to its first
2625 * record (to be inserted into parent).
2627 STATIC int /* error */
2629 struct xfs_btree_cur *cur,
2631 union xfs_btree_ptr *ptrp,
2632 union xfs_btree_key *key,
2633 struct xfs_btree_cur **curp,
2634 int *stat) /* success/failure */
2636 union xfs_btree_ptr lptr; /* left sibling block ptr */
2637 struct xfs_buf *lbp; /* left buffer pointer */
2638 struct xfs_btree_block *left; /* left btree block */
2639 union xfs_btree_ptr rptr; /* right sibling block ptr */
2640 struct xfs_buf *rbp; /* right buffer pointer */
2641 struct xfs_btree_block *right; /* right btree block */
2642 union xfs_btree_ptr rrptr; /* right-right sibling ptr */
2643 struct xfs_buf *rrbp; /* right-right buffer pointer */
2644 struct xfs_btree_block *rrblock; /* right-right btree block */
2648 int error; /* error return value */
2653 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2654 XFS_BTREE_TRACE_ARGIPK(cur, level, *ptrp, key);
2656 XFS_BTREE_STATS_INC(cur, split);
2658 /* Set up left block (current one). */
2659 left = xfs_btree_get_block(cur, level, &lbp);
2662 error = xfs_btree_check_block(cur, left, level, lbp);
2667 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2669 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2670 error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, stat);
2675 XFS_BTREE_STATS_INC(cur, alloc);
2677 /* Set up the new block as "right". */
2678 error = xfs_btree_get_buf_block(cur, &rptr, 0, &right, &rbp);
2682 /* Fill in the btree header for the new right block. */
2683 xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2686 * Split the entries between the old and the new block evenly.
2687 * Make sure that if there's an odd number of entries now, that
2688 * each new block will have the same number of entries.
2690 lrecs = xfs_btree_get_numrecs(left);
2692 if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
2694 src_index = (lrecs - rrecs + 1);
2696 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2698 /* Adjust numrecs for the later get_*_keys() calls. */
2700 xfs_btree_set_numrecs(left, lrecs);
2701 xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2704 * Copy btree block entries from the left block over to the
2705 * new block, the right. Update the right block and log the
2709 /* It's a non-leaf. Move keys and pointers. */
2710 union xfs_btree_key *lkp; /* left btree key */
2711 union xfs_btree_ptr *lpp; /* left address pointer */
2712 union xfs_btree_key *rkp; /* right btree key */
2713 union xfs_btree_ptr *rpp; /* right address pointer */
2715 lkp = xfs_btree_key_addr(cur, src_index, left);
2716 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2717 rkp = xfs_btree_key_addr(cur, 1, right);
2718 rpp = xfs_btree_ptr_addr(cur, 1, right);
2721 for (i = src_index; i < rrecs; i++) {
2722 error = xfs_btree_check_ptr(cur, lpp, i, level);
2728 /* Copy the keys & pointers to the new block. */
2729 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2730 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2732 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2733 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2735 /* Stash the keys of the new block for later insertion. */
2736 xfs_btree_get_node_keys(cur, right, key);
2738 /* It's a leaf. Move records. */
2739 union xfs_btree_rec *lrp; /* left record pointer */
2740 union xfs_btree_rec *rrp; /* right record pointer */
2742 lrp = xfs_btree_rec_addr(cur, src_index, left);
2743 rrp = xfs_btree_rec_addr(cur, 1, right);
2745 /* Copy records to the new block. */
2746 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2747 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2749 /* Stash the keys of the new block for later insertion. */
2750 xfs_btree_get_leaf_keys(cur, right, key);
2754 * Find the left block number by looking in the buffer.
2755 * Adjust sibling pointers.
2757 xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2758 xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2759 xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2760 xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2762 xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2763 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2766 * If there's a block to the new block's right, make that block
2767 * point back to right instead of to left.
2769 if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2770 error = xfs_btree_read_buf_block(cur, &rrptr,
2771 0, &rrblock, &rrbp);
2774 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2775 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2778 /* Update the parent high keys of the left block, if needed. */
2779 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2780 error = xfs_btree_update_keys(cur, level);
2786 * If the cursor is really in the right block, move it there.
2787 * If it's just pointing past the last entry in left, then we'll
2788 * insert there, so don't change anything in that case.
2790 if (cur->bc_ptrs[level] > lrecs + 1) {
2791 xfs_btree_setbuf(cur, level, rbp);
2792 cur->bc_ptrs[level] -= lrecs;
2795 * If there are more levels, we'll need another cursor which refers
2796 * the right block, no matter where this cursor was.
2798 if (level + 1 < cur->bc_nlevels) {
2799 error = xfs_btree_dup_cursor(cur, curp);
2802 (*curp)->bc_ptrs[level + 1]++;
2805 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2809 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2814 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2818 struct xfs_btree_split_args {
2819 struct xfs_btree_cur *cur;
2821 union xfs_btree_ptr *ptrp;
2822 union xfs_btree_key *key;
2823 struct xfs_btree_cur **curp;
2824 int *stat; /* success/failure */
2826 bool kswapd; /* allocation in kswapd context */
2827 struct completion *done;
2828 struct work_struct work;
2832 * Stack switching interfaces for allocation
2835 xfs_btree_split_worker(
2836 struct work_struct *work)
2838 struct xfs_btree_split_args *args = container_of(work,
2839 struct xfs_btree_split_args, work);
2840 unsigned long pflags;
2841 unsigned long new_pflags = PF_FSTRANS;
2844 * we are in a transaction context here, but may also be doing work
2845 * in kswapd context, and hence we may need to inherit that state
2846 * temporarily to ensure that we don't block waiting for memory reclaim
2850 new_pflags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
2852 current_set_flags_nested(&pflags, new_pflags);
2854 args->result = __xfs_btree_split(args->cur, args->level, args->ptrp,
2855 args->key, args->curp, args->stat);
2856 complete(args->done);
2858 current_restore_flags_nested(&pflags, new_pflags);
2862 * BMBT split requests often come in with little stack to work on. Push
2863 * them off to a worker thread so there is lots of stack to use. For the other
2864 * btree types, just call directly to avoid the context switch overhead here.
2866 STATIC int /* error */
2868 struct xfs_btree_cur *cur,
2870 union xfs_btree_ptr *ptrp,
2871 union xfs_btree_key *key,
2872 struct xfs_btree_cur **curp,
2873 int *stat) /* success/failure */
2875 struct xfs_btree_split_args args;
2876 DECLARE_COMPLETION_ONSTACK(done);
2878 if (cur->bc_btnum != XFS_BTNUM_BMAP)
2879 return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
2888 args.kswapd = current_is_kswapd();
2889 INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker);
2890 queue_work(xfs_alloc_wq, &args.work);
2891 wait_for_completion(&done);
2892 destroy_work_on_stack(&args.work);
2898 * Copy the old inode root contents into a real block and make the
2899 * broot point to it.
2902 xfs_btree_new_iroot(
2903 struct xfs_btree_cur *cur, /* btree cursor */
2904 int *logflags, /* logging flags for inode */
2905 int *stat) /* return status - 0 fail */
2907 struct xfs_buf *cbp; /* buffer for cblock */
2908 struct xfs_btree_block *block; /* btree block */
2909 struct xfs_btree_block *cblock; /* child btree block */
2910 union xfs_btree_key *ckp; /* child key pointer */
2911 union xfs_btree_ptr *cpp; /* child ptr pointer */
2912 union xfs_btree_key *kp; /* pointer to btree key */
2913 union xfs_btree_ptr *pp; /* pointer to block addr */
2914 union xfs_btree_ptr nptr; /* new block addr */
2915 int level; /* btree level */
2916 int error; /* error return code */
2918 int i; /* loop counter */
2921 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2922 XFS_BTREE_STATS_INC(cur, newroot);
2924 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2926 level = cur->bc_nlevels - 1;
2928 block = xfs_btree_get_iroot(cur);
2929 pp = xfs_btree_ptr_addr(cur, 1, block);
2931 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2932 error = cur->bc_ops->alloc_block(cur, pp, &nptr, stat);
2936 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2939 XFS_BTREE_STATS_INC(cur, alloc);
2941 /* Copy the root into a real block. */
2942 error = xfs_btree_get_buf_block(cur, &nptr, 0, &cblock, &cbp);
2947 * we can't just memcpy() the root in for CRC enabled btree blocks.
2948 * In that case have to also ensure the blkno remains correct
2950 memcpy(cblock, block, xfs_btree_block_len(cur));
2951 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
2952 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
2953 cblock->bb_u.l.bb_blkno = cpu_to_be64(cbp->b_bn);
2955 cblock->bb_u.s.bb_blkno = cpu_to_be64(cbp->b_bn);
2958 be16_add_cpu(&block->bb_level, 1);
2959 xfs_btree_set_numrecs(block, 1);
2961 cur->bc_ptrs[level + 1] = 1;
2963 kp = xfs_btree_key_addr(cur, 1, block);
2964 ckp = xfs_btree_key_addr(cur, 1, cblock);
2965 xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
2967 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
2969 for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
2970 error = xfs_btree_check_ptr(cur, pp, i, level);
2975 xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
2978 error = xfs_btree_check_ptr(cur, &nptr, 0, level);
2982 xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
2984 xfs_iroot_realloc(cur->bc_private.b.ip,
2985 1 - xfs_btree_get_numrecs(cblock),
2986 cur->bc_private.b.whichfork);
2988 xfs_btree_setbuf(cur, level, cbp);
2991 * Do all this logging at the end so that
2992 * the root is at the right level.
2994 xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
2995 xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2996 xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2999 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork);
3001 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3004 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3009 * Allocate a new root block, fill it in.
3011 STATIC int /* error */
3013 struct xfs_btree_cur *cur, /* btree cursor */
3014 int *stat) /* success/failure */
3016 struct xfs_btree_block *block; /* one half of the old root block */
3017 struct xfs_buf *bp; /* buffer containing block */
3018 int error; /* error return value */
3019 struct xfs_buf *lbp; /* left buffer pointer */
3020 struct xfs_btree_block *left; /* left btree block */
3021 struct xfs_buf *nbp; /* new (root) buffer */
3022 struct xfs_btree_block *new; /* new (root) btree block */
3023 int nptr; /* new value for key index, 1 or 2 */
3024 struct xfs_buf *rbp; /* right buffer pointer */
3025 struct xfs_btree_block *right; /* right btree block */
3026 union xfs_btree_ptr rptr;
3027 union xfs_btree_ptr lptr;
3029 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3030 XFS_BTREE_STATS_INC(cur, newroot);
3032 /* initialise our start point from the cursor */
3033 cur->bc_ops->init_ptr_from_cur(cur, &rptr);
3035 /* Allocate the new block. If we can't do it, we're toast. Give up. */
3036 error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, stat);
3041 XFS_BTREE_STATS_INC(cur, alloc);
3043 /* Set up the new block. */
3044 error = xfs_btree_get_buf_block(cur, &lptr, 0, &new, &nbp);
3048 /* Set the root in the holding structure increasing the level by 1. */
3049 cur->bc_ops->set_root(cur, &lptr, 1);
3052 * At the previous root level there are now two blocks: the old root,
3053 * and the new block generated when it was split. We don't know which
3054 * one the cursor is pointing at, so we set up variables "left" and
3055 * "right" for each case.
3057 block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
3060 error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
3065 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3066 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3067 /* Our block is left, pick up the right block. */
3069 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
3071 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
3077 /* Our block is right, pick up the left block. */
3079 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
3081 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
3082 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3089 /* Fill in the new block's btree header and log it. */
3090 xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
3091 xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
3092 ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
3093 !xfs_btree_ptr_is_null(cur, &rptr));
3095 /* Fill in the key data in the new root. */
3096 if (xfs_btree_get_level(left) > 0) {
3098 * Get the keys for the left block's keys and put them directly
3099 * in the parent block. Do the same for the right block.
3101 xfs_btree_get_node_keys(cur, left,
3102 xfs_btree_key_addr(cur, 1, new));
3103 xfs_btree_get_node_keys(cur, right,
3104 xfs_btree_key_addr(cur, 2, new));
3107 * Get the keys for the left block's records and put them
3108 * directly in the parent block. Do the same for the right
3111 xfs_btree_get_leaf_keys(cur, left,
3112 xfs_btree_key_addr(cur, 1, new));
3113 xfs_btree_get_leaf_keys(cur, right,
3114 xfs_btree_key_addr(cur, 2, new));
3116 xfs_btree_log_keys(cur, nbp, 1, 2);
3118 /* Fill in the pointer data in the new root. */
3119 xfs_btree_copy_ptrs(cur,
3120 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
3121 xfs_btree_copy_ptrs(cur,
3122 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
3123 xfs_btree_log_ptrs(cur, nbp, 1, 2);
3125 /* Fix up the cursor. */
3126 xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
3127 cur->bc_ptrs[cur->bc_nlevels] = nptr;
3129 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3133 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3136 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3142 xfs_btree_make_block_unfull(
3143 struct xfs_btree_cur *cur, /* btree cursor */
3144 int level, /* btree level */
3145 int numrecs,/* # of recs in block */
3146 int *oindex,/* old tree index */
3147 int *index, /* new tree index */
3148 union xfs_btree_ptr *nptr, /* new btree ptr */
3149 struct xfs_btree_cur **ncur, /* new btree cursor */
3150 union xfs_btree_key *key, /* key of new block */
3155 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3156 level == cur->bc_nlevels - 1) {
3157 struct xfs_inode *ip = cur->bc_private.b.ip;
3159 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
3160 /* A root block that can be made bigger. */
3161 xfs_iroot_realloc(ip, 1, cur->bc_private.b.whichfork);
3164 /* A root block that needs replacing */
3167 error = xfs_btree_new_iroot(cur, &logflags, stat);
3168 if (error || *stat == 0)
3171 xfs_trans_log_inode(cur->bc_tp, ip, logflags);
3177 /* First, try shifting an entry to the right neighbor. */
3178 error = xfs_btree_rshift(cur, level, stat);
3182 /* Next, try shifting an entry to the left neighbor. */
3183 error = xfs_btree_lshift(cur, level, stat);
3188 *oindex = *index = cur->bc_ptrs[level];
3193 * Next, try splitting the current block in half.
3195 * If this works we have to re-set our variables because we
3196 * could be in a different block now.
3198 error = xfs_btree_split(cur, level, nptr, key, ncur, stat);
3199 if (error || *stat == 0)
3203 *index = cur->bc_ptrs[level];
3208 * Insert one record/level. Return information to the caller
3209 * allowing the next level up to proceed if necessary.
3213 struct xfs_btree_cur *cur, /* btree cursor */
3214 int level, /* level to insert record at */
3215 union xfs_btree_ptr *ptrp, /* i/o: block number inserted */
3216 union xfs_btree_rec *rec, /* record to insert */
3217 union xfs_btree_key *key, /* i/o: block key for ptrp */
3218 struct xfs_btree_cur **curp, /* output: new cursor replacing cur */
3219 int *stat) /* success/failure */
3221 struct xfs_btree_block *block; /* btree block */
3222 struct xfs_buf *bp; /* buffer for block */
3223 union xfs_btree_ptr nptr; /* new block ptr */
3224 struct xfs_btree_cur *ncur; /* new btree cursor */
3225 union xfs_btree_bigkey nkey; /* new block key */
3226 union xfs_btree_key *lkey;
3227 int optr; /* old key/record index */
3228 int ptr; /* key/record index */
3229 int numrecs;/* number of records */
3230 int error; /* error return value */
3236 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3237 XFS_BTREE_TRACE_ARGIPR(cur, level, *ptrp, &rec);
3240 lkey = (union xfs_btree_key *)&nkey;
3243 * If we have an external root pointer, and we've made it to the
3244 * root level, allocate a new root block and we're done.
3246 if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3247 (level >= cur->bc_nlevels)) {
3248 error = xfs_btree_new_root(cur, stat);
3249 xfs_btree_set_ptr_null(cur, ptrp);
3251 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3255 /* If we're off the left edge, return failure. */
3256 ptr = cur->bc_ptrs[level];
3258 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3265 XFS_BTREE_STATS_INC(cur, insrec);
3267 /* Get pointers to the btree buffer and block. */
3268 block = xfs_btree_get_block(cur, level, &bp);
3269 old_bn = bp ? bp->b_bn : XFS_BUF_DADDR_NULL;
3270 numrecs = xfs_btree_get_numrecs(block);
3273 error = xfs_btree_check_block(cur, block, level, bp);
3277 /* Check that the new entry is being inserted in the right place. */
3278 if (ptr <= numrecs) {
3280 ASSERT(cur->bc_ops->recs_inorder(cur, rec,
3281 xfs_btree_rec_addr(cur, ptr, block)));
3283 ASSERT(cur->bc_ops->keys_inorder(cur, key,
3284 xfs_btree_key_addr(cur, ptr, block)));
3290 * If the block is full, we can't insert the new entry until we
3291 * make the block un-full.
3293 xfs_btree_set_ptr_null(cur, &nptr);
3294 if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
3295 error = xfs_btree_make_block_unfull(cur, level, numrecs,
3296 &optr, &ptr, &nptr, &ncur, lkey, stat);
3297 if (error || *stat == 0)
3302 * The current block may have changed if the block was
3303 * previously full and we have just made space in it.
3305 block = xfs_btree_get_block(cur, level, &bp);
3306 numrecs = xfs_btree_get_numrecs(block);
3309 error = xfs_btree_check_block(cur, block, level, bp);
3315 * At this point we know there's room for our new entry in the block
3316 * we're pointing at.
3318 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
3321 /* It's a nonleaf. make a hole in the keys and ptrs */
3322 union xfs_btree_key *kp;
3323 union xfs_btree_ptr *pp;
3325 kp = xfs_btree_key_addr(cur, ptr, block);
3326 pp = xfs_btree_ptr_addr(cur, ptr, block);
3329 for (i = numrecs - ptr; i >= 0; i--) {
3330 error = xfs_btree_check_ptr(cur, pp, i, level);
3336 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
3337 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
3340 error = xfs_btree_check_ptr(cur, ptrp, 0, level);
3345 /* Now put the new data in, bump numrecs and log it. */
3346 xfs_btree_copy_keys(cur, kp, key, 1);
3347 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
3349 xfs_btree_set_numrecs(block, numrecs);
3350 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
3351 xfs_btree_log_keys(cur, bp, ptr, numrecs);
3353 if (ptr < numrecs) {
3354 ASSERT(cur->bc_ops->keys_inorder(cur, kp,
3355 xfs_btree_key_addr(cur, ptr + 1, block)));
3359 /* It's a leaf. make a hole in the records */
3360 union xfs_btree_rec *rp;
3362 rp = xfs_btree_rec_addr(cur, ptr, block);
3364 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
3366 /* Now put the new data in, bump numrecs and log it. */
3367 xfs_btree_copy_recs(cur, rp, rec, 1);
3368 xfs_btree_set_numrecs(block, ++numrecs);
3369 xfs_btree_log_recs(cur, bp, ptr, numrecs);
3371 if (ptr < numrecs) {
3372 ASSERT(cur->bc_ops->recs_inorder(cur, rp,
3373 xfs_btree_rec_addr(cur, ptr + 1, block)));
3378 /* Log the new number of records in the btree header. */
3379 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3382 * If we just inserted into a new tree block, we have to
3383 * recalculate nkey here because nkey is out of date.
3385 * Otherwise we're just updating an existing block (having shoved
3386 * some records into the new tree block), so use the regular key
3389 if (bp && bp->b_bn != old_bn) {
3390 xfs_btree_get_keys(cur, block, lkey);
3391 } else if (xfs_btree_needs_key_update(cur, optr)) {
3392 error = xfs_btree_update_keys(cur, level);
3398 * If we are tracking the last record in the tree and
3399 * we are at the far right edge of the tree, update it.
3401 if (xfs_btree_is_lastrec(cur, block, level)) {
3402 cur->bc_ops->update_lastrec(cur, block, rec,
3403 ptr, LASTREC_INSREC);
3407 * Return the new block number, if any.
3408 * If there is one, give back a record value and a cursor too.
3411 if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3412 xfs_btree_copy_keys(cur, key, lkey, 1);
3416 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3421 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3426 * Insert the record at the point referenced by cur.
3428 * A multi-level split of the tree on insert will invalidate the original
3429 * cursor. All callers of this function should assume that the cursor is
3430 * no longer valid and revalidate it.
3434 struct xfs_btree_cur *cur,
3437 int error; /* error return value */
3438 int i; /* result value, 0 for failure */
3439 int level; /* current level number in btree */
3440 union xfs_btree_ptr nptr; /* new block number (split result) */
3441 struct xfs_btree_cur *ncur; /* new cursor (split result) */
3442 struct xfs_btree_cur *pcur; /* previous level's cursor */
3443 union xfs_btree_bigkey bkey; /* key of block to insert */
3444 union xfs_btree_key *key;
3445 union xfs_btree_rec rec; /* record to insert */
3450 key = (union xfs_btree_key *)&bkey;
3452 xfs_btree_set_ptr_null(cur, &nptr);
3454 /* Make a key out of the record data to be inserted, and save it. */
3455 cur->bc_ops->init_rec_from_cur(cur, &rec);
3456 cur->bc_ops->init_key_from_rec(key, &rec);
3459 * Loop going up the tree, starting at the leaf level.
3460 * Stop when we don't get a split block, that must mean that
3461 * the insert is finished with this level.
3465 * Insert nrec/nptr into this level of the tree.
3466 * Note if we fail, nptr will be null.
3468 error = xfs_btree_insrec(pcur, level, &nptr, &rec, key,
3472 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3476 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3480 * See if the cursor we just used is trash.
3481 * Can't trash the caller's cursor, but otherwise we should
3482 * if ncur is a new cursor or we're about to be done.
3485 (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3486 /* Save the state from the cursor before we trash it */
3487 if (cur->bc_ops->update_cursor)
3488 cur->bc_ops->update_cursor(pcur, cur);
3489 cur->bc_nlevels = pcur->bc_nlevels;
3490 xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3492 /* If we got a new cursor, switch to it. */
3497 } while (!xfs_btree_ptr_is_null(cur, &nptr));
3499 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3503 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3508 * Try to merge a non-leaf block back into the inode root.
3510 * Note: the killroot names comes from the fact that we're effectively
3511 * killing the old root block. But because we can't just delete the
3512 * inode we have to copy the single block it was pointing to into the
3516 xfs_btree_kill_iroot(
3517 struct xfs_btree_cur *cur)
3519 int whichfork = cur->bc_private.b.whichfork;
3520 struct xfs_inode *ip = cur->bc_private.b.ip;
3521 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
3522 struct xfs_btree_block *block;
3523 struct xfs_btree_block *cblock;
3524 union xfs_btree_key *kp;
3525 union xfs_btree_key *ckp;
3526 union xfs_btree_ptr *pp;
3527 union xfs_btree_ptr *cpp;
3528 struct xfs_buf *cbp;
3534 union xfs_btree_ptr ptr;
3538 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3540 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3541 ASSERT(cur->bc_nlevels > 1);
3544 * Don't deal with the root block needs to be a leaf case.
3545 * We're just going to turn the thing back into extents anyway.
3547 level = cur->bc_nlevels - 1;
3552 * Give up if the root has multiple children.
3554 block = xfs_btree_get_iroot(cur);
3555 if (xfs_btree_get_numrecs(block) != 1)
3558 cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3559 numrecs = xfs_btree_get_numrecs(cblock);
3562 * Only do this if the next level will fit.
3563 * Then the data must be copied up to the inode,
3564 * instead of freeing the root you free the next level.
3566 if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3569 XFS_BTREE_STATS_INC(cur, killroot);
3572 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3573 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3574 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3575 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3578 index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3580 xfs_iroot_realloc(cur->bc_private.b.ip, index,
3581 cur->bc_private.b.whichfork);
3582 block = ifp->if_broot;
3585 be16_add_cpu(&block->bb_numrecs, index);
3586 ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3588 kp = xfs_btree_key_addr(cur, 1, block);
3589 ckp = xfs_btree_key_addr(cur, 1, cblock);
3590 xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3592 pp = xfs_btree_ptr_addr(cur, 1, block);
3593 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3595 for (i = 0; i < numrecs; i++) {
3596 error = xfs_btree_check_ptr(cur, cpp, i, level - 1);
3598 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3603 xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3605 error = xfs_btree_free_block(cur, cbp);
3607 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3611 cur->bc_bufs[level - 1] = NULL;
3612 be16_add_cpu(&block->bb_level, -1);
3613 xfs_trans_log_inode(cur->bc_tp, ip,
3614 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork));
3617 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3622 * Kill the current root node, and replace it with it's only child node.
3625 xfs_btree_kill_root(
3626 struct xfs_btree_cur *cur,
3629 union xfs_btree_ptr *newroot)
3633 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3634 XFS_BTREE_STATS_INC(cur, killroot);
3637 * Update the root pointer, decreasing the level by 1 and then
3638 * free the old root.
3640 cur->bc_ops->set_root(cur, newroot, -1);
3642 error = xfs_btree_free_block(cur, bp);
3644 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3648 cur->bc_bufs[level] = NULL;
3649 cur->bc_ra[level] = 0;
3652 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3657 xfs_btree_dec_cursor(
3658 struct xfs_btree_cur *cur,
3666 error = xfs_btree_decrement(cur, level, &i);
3671 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3677 * Single level of the btree record deletion routine.
3678 * Delete record pointed to by cur/level.
3679 * Remove the record from its block then rebalance the tree.
3680 * Return 0 for error, 1 for done, 2 to go on to the next level.
3682 STATIC int /* error */
3684 struct xfs_btree_cur *cur, /* btree cursor */
3685 int level, /* level removing record from */
3686 int *stat) /* fail/done/go-on */
3688 struct xfs_btree_block *block; /* btree block */
3689 union xfs_btree_ptr cptr; /* current block ptr */
3690 struct xfs_buf *bp; /* buffer for block */
3691 int error; /* error return value */
3692 int i; /* loop counter */
3693 union xfs_btree_ptr lptr; /* left sibling block ptr */
3694 struct xfs_buf *lbp; /* left buffer pointer */
3695 struct xfs_btree_block *left; /* left btree block */
3696 int lrecs = 0; /* left record count */
3697 int ptr; /* key/record index */
3698 union xfs_btree_ptr rptr; /* right sibling block ptr */
3699 struct xfs_buf *rbp; /* right buffer pointer */
3700 struct xfs_btree_block *right; /* right btree block */
3701 struct xfs_btree_block *rrblock; /* right-right btree block */
3702 struct xfs_buf *rrbp; /* right-right buffer pointer */
3703 int rrecs = 0; /* right record count */
3704 struct xfs_btree_cur *tcur; /* temporary btree cursor */
3705 int numrecs; /* temporary numrec count */
3707 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3708 XFS_BTREE_TRACE_ARGI(cur, level);
3712 /* Get the index of the entry being deleted, check for nothing there. */
3713 ptr = cur->bc_ptrs[level];
3715 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3720 /* Get the buffer & block containing the record or key/ptr. */
3721 block = xfs_btree_get_block(cur, level, &bp);
3722 numrecs = xfs_btree_get_numrecs(block);
3725 error = xfs_btree_check_block(cur, block, level, bp);
3730 /* Fail if we're off the end of the block. */
3731 if (ptr > numrecs) {
3732 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3737 XFS_BTREE_STATS_INC(cur, delrec);
3738 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3740 /* Excise the entries being deleted. */
3742 /* It's a nonleaf. operate on keys and ptrs */
3743 union xfs_btree_key *lkp;
3744 union xfs_btree_ptr *lpp;
3746 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3747 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3750 for (i = 0; i < numrecs - ptr; i++) {
3751 error = xfs_btree_check_ptr(cur, lpp, i, level);
3757 if (ptr < numrecs) {
3758 xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3759 xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3760 xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3761 xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3764 /* It's a leaf. operate on records */
3765 if (ptr < numrecs) {
3766 xfs_btree_shift_recs(cur,
3767 xfs_btree_rec_addr(cur, ptr + 1, block),
3769 xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3774 * Decrement and log the number of entries in the block.
3776 xfs_btree_set_numrecs(block, --numrecs);
3777 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3780 * If we are tracking the last record in the tree and
3781 * we are at the far right edge of the tree, update it.
3783 if (xfs_btree_is_lastrec(cur, block, level)) {
3784 cur->bc_ops->update_lastrec(cur, block, NULL,
3785 ptr, LASTREC_DELREC);
3789 * We're at the root level. First, shrink the root block in-memory.
3790 * Try to get rid of the next level down. If we can't then there's
3791 * nothing left to do.
3793 if (level == cur->bc_nlevels - 1) {
3794 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3795 xfs_iroot_realloc(cur->bc_private.b.ip, -1,
3796 cur->bc_private.b.whichfork);
3798 error = xfs_btree_kill_iroot(cur);
3802 error = xfs_btree_dec_cursor(cur, level, stat);
3810 * If this is the root level, and there's only one entry left,
3811 * and it's NOT the leaf level, then we can get rid of this
3814 if (numrecs == 1 && level > 0) {
3815 union xfs_btree_ptr *pp;
3817 * pp is still set to the first pointer in the block.
3818 * Make it the new root of the btree.
3820 pp = xfs_btree_ptr_addr(cur, 1, block);
3821 error = xfs_btree_kill_root(cur, bp, level, pp);
3824 } else if (level > 0) {
3825 error = xfs_btree_dec_cursor(cur, level, stat);
3834 * If we deleted the leftmost entry in the block, update the
3835 * key values above us in the tree.
3837 if (xfs_btree_needs_key_update(cur, ptr)) {
3838 error = xfs_btree_update_keys(cur, level);
3844 * If the number of records remaining in the block is at least
3845 * the minimum, we're done.
3847 if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3848 error = xfs_btree_dec_cursor(cur, level, stat);
3855 * Otherwise, we have to move some records around to keep the
3856 * tree balanced. Look at the left and right sibling blocks to
3857 * see if we can re-balance by moving only one record.
3859 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3860 xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3862 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3864 * One child of root, need to get a chance to copy its contents
3865 * into the root and delete it. Can't go up to next level,
3866 * there's nothing to delete there.
3868 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3869 xfs_btree_ptr_is_null(cur, &lptr) &&
3870 level == cur->bc_nlevels - 2) {
3871 error = xfs_btree_kill_iroot(cur);
3873 error = xfs_btree_dec_cursor(cur, level, stat);
3880 ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3881 !xfs_btree_ptr_is_null(cur, &lptr));
3884 * Duplicate the cursor so our btree manipulations here won't
3885 * disrupt the next level up.
3887 error = xfs_btree_dup_cursor(cur, &tcur);
3892 * If there's a right sibling, see if it's ok to shift an entry
3895 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3897 * Move the temp cursor to the last entry in the next block.
3898 * Actually any entry but the first would suffice.
3900 i = xfs_btree_lastrec(tcur, level);
3901 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3903 error = xfs_btree_increment(tcur, level, &i);
3906 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3908 i = xfs_btree_lastrec(tcur, level);
3909 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3911 /* Grab a pointer to the block. */
3912 right = xfs_btree_get_block(tcur, level, &rbp);
3914 error = xfs_btree_check_block(tcur, right, level, rbp);
3918 /* Grab the current block number, for future use. */
3919 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3922 * If right block is full enough so that removing one entry
3923 * won't make it too empty, and left-shifting an entry out
3924 * of right to us works, we're done.
3926 if (xfs_btree_get_numrecs(right) - 1 >=
3927 cur->bc_ops->get_minrecs(tcur, level)) {
3928 error = xfs_btree_lshift(tcur, level, &i);
3932 ASSERT(xfs_btree_get_numrecs(block) >=
3933 cur->bc_ops->get_minrecs(tcur, level));
3935 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3938 error = xfs_btree_dec_cursor(cur, level, stat);
3946 * Otherwise, grab the number of records in right for
3947 * future reference, and fix up the temp cursor to point
3948 * to our block again (last record).
3950 rrecs = xfs_btree_get_numrecs(right);
3951 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3952 i = xfs_btree_firstrec(tcur, level);
3953 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3955 error = xfs_btree_decrement(tcur, level, &i);
3958 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3963 * If there's a left sibling, see if it's ok to shift an entry
3966 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3968 * Move the temp cursor to the first entry in the
3971 i = xfs_btree_firstrec(tcur, level);
3972 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3974 error = xfs_btree_decrement(tcur, level, &i);
3977 i = xfs_btree_firstrec(tcur, level);
3978 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3980 /* Grab a pointer to the block. */
3981 left = xfs_btree_get_block(tcur, level, &lbp);
3983 error = xfs_btree_check_block(cur, left, level, lbp);
3987 /* Grab the current block number, for future use. */
3988 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
3991 * If left block is full enough so that removing one entry
3992 * won't make it too empty, and right-shifting an entry out
3993 * of left to us works, we're done.
3995 if (xfs_btree_get_numrecs(left) - 1 >=
3996 cur->bc_ops->get_minrecs(tcur, level)) {
3997 error = xfs_btree_rshift(tcur, level, &i);
4001 ASSERT(xfs_btree_get_numrecs(block) >=
4002 cur->bc_ops->get_minrecs(tcur, level));
4003 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4007 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
4014 * Otherwise, grab the number of records in right for
4017 lrecs = xfs_btree_get_numrecs(left);
4020 /* Delete the temp cursor, we're done with it. */
4021 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4024 /* If here, we need to do a join to keep the tree balanced. */
4025 ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
4027 if (!xfs_btree_ptr_is_null(cur, &lptr) &&
4028 lrecs + xfs_btree_get_numrecs(block) <=
4029 cur->bc_ops->get_maxrecs(cur, level)) {
4031 * Set "right" to be the starting block,
4032 * "left" to be the left neighbor.
4037 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
4042 * If that won't work, see if we can join with the right neighbor block.
4044 } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
4045 rrecs + xfs_btree_get_numrecs(block) <=
4046 cur->bc_ops->get_maxrecs(cur, level)) {
4048 * Set "left" to be the starting block,
4049 * "right" to be the right neighbor.
4054 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
4059 * Otherwise, we can't fix the imbalance.
4060 * Just return. This is probably a logic error, but it's not fatal.
4063 error = xfs_btree_dec_cursor(cur, level, stat);
4069 rrecs = xfs_btree_get_numrecs(right);
4070 lrecs = xfs_btree_get_numrecs(left);
4073 * We're now going to join "left" and "right" by moving all the stuff
4074 * in "right" to "left" and deleting "right".
4076 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
4078 /* It's a non-leaf. Move keys and pointers. */
4079 union xfs_btree_key *lkp; /* left btree key */
4080 union xfs_btree_ptr *lpp; /* left address pointer */
4081 union xfs_btree_key *rkp; /* right btree key */
4082 union xfs_btree_ptr *rpp; /* right address pointer */
4084 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
4085 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
4086 rkp = xfs_btree_key_addr(cur, 1, right);
4087 rpp = xfs_btree_ptr_addr(cur, 1, right);
4089 for (i = 1; i < rrecs; i++) {
4090 error = xfs_btree_check_ptr(cur, rpp, i, level);
4095 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
4096 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
4098 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
4099 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
4101 /* It's a leaf. Move records. */
4102 union xfs_btree_rec *lrp; /* left record pointer */
4103 union xfs_btree_rec *rrp; /* right record pointer */
4105 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
4106 rrp = xfs_btree_rec_addr(cur, 1, right);
4108 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
4109 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
4112 XFS_BTREE_STATS_INC(cur, join);
4115 * Fix up the number of records and right block pointer in the
4116 * surviving block, and log it.
4118 xfs_btree_set_numrecs(left, lrecs + rrecs);
4119 xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB),
4120 xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4121 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
4123 /* If there is a right sibling, point it to the remaining block. */
4124 xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4125 if (!xfs_btree_ptr_is_null(cur, &cptr)) {
4126 error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp);
4129 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
4130 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
4133 /* Free the deleted block. */
4134 error = xfs_btree_free_block(cur, rbp);
4139 * If we joined with the left neighbor, set the buffer in the
4140 * cursor to the left block, and fix up the index.
4143 cur->bc_bufs[level] = lbp;
4144 cur->bc_ptrs[level] += lrecs;
4145 cur->bc_ra[level] = 0;
4148 * If we joined with the right neighbor and there's a level above
4149 * us, increment the cursor at that level.
4151 else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
4152 (level + 1 < cur->bc_nlevels)) {
4153 error = xfs_btree_increment(cur, level + 1, &i);
4159 * Readjust the ptr at this level if it's not a leaf, since it's
4160 * still pointing at the deletion point, which makes the cursor
4161 * inconsistent. If this makes the ptr 0, the caller fixes it up.
4162 * We can't use decrement because it would change the next level up.
4165 cur->bc_ptrs[level]--;
4168 * We combined blocks, so we have to update the parent keys if the
4169 * btree supports overlapped intervals. However, bc_ptrs[level + 1]
4170 * points to the old block so that the caller knows which record to
4171 * delete. Therefore, the caller must be savvy enough to call updkeys
4172 * for us if we return stat == 2. The other exit points from this
4173 * function don't require deletions further up the tree, so they can
4174 * call updkeys directly.
4177 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
4178 /* Return value means the next level up has something to do. */
4183 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
4185 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
4190 * Delete the record pointed to by cur.
4191 * The cursor refers to the place where the record was (could be inserted)
4192 * when the operation returns.
4196 struct xfs_btree_cur *cur,
4197 int *stat) /* success/failure */
4199 int error; /* error return value */
4202 bool joined = false;
4204 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
4207 * Go up the tree, starting at leaf level.
4209 * If 2 is returned then a join was done; go to the next level.
4210 * Otherwise we are done.
4212 for (level = 0, i = 2; i == 2; level++) {
4213 error = xfs_btree_delrec(cur, level, &i);
4221 * If we combined blocks as part of deleting the record, delrec won't
4222 * have updated the parent high keys so we have to do that here.
4224 if (joined && (cur->bc_flags & XFS_BTREE_OVERLAPPING)) {
4225 error = xfs_btree_updkeys_force(cur, 0);
4231 for (level = 1; level < cur->bc_nlevels; level++) {
4232 if (cur->bc_ptrs[level] == 0) {
4233 error = xfs_btree_decrement(cur, level, &i);
4241 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
4245 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
4250 * Get the data from the pointed-to record.
4254 struct xfs_btree_cur *cur, /* btree cursor */
4255 union xfs_btree_rec **recp, /* output: btree record */
4256 int *stat) /* output: success/failure */
4258 struct xfs_btree_block *block; /* btree block */
4259 struct xfs_buf *bp; /* buffer pointer */
4260 int ptr; /* record number */
4262 int error; /* error return value */
4265 ptr = cur->bc_ptrs[0];
4266 block = xfs_btree_get_block(cur, 0, &bp);
4269 error = xfs_btree_check_block(cur, block, 0, bp);
4275 * Off the right end or left end, return failure.
4277 if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
4283 * Point to the record and extract its data.
4285 *recp = xfs_btree_rec_addr(cur, ptr, block);
4290 /* Visit a block in a btree. */
4292 xfs_btree_visit_block(
4293 struct xfs_btree_cur *cur,
4295 xfs_btree_visit_blocks_fn fn,
4298 struct xfs_btree_block *block;
4300 union xfs_btree_ptr rptr;
4303 /* do right sibling readahead */
4304 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
4305 block = xfs_btree_get_block(cur, level, &bp);
4307 /* process the block */
4308 error = fn(cur, level, data);
4312 /* now read rh sibling block for next iteration */
4313 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4314 if (xfs_btree_ptr_is_null(cur, &rptr))
4317 return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
4321 /* Visit every block in a btree. */
4323 xfs_btree_visit_blocks(
4324 struct xfs_btree_cur *cur,
4325 xfs_btree_visit_blocks_fn fn,
4328 union xfs_btree_ptr lptr;
4330 struct xfs_btree_block *block = NULL;
4333 cur->bc_ops->init_ptr_from_cur(cur, &lptr);
4335 /* for each level */
4336 for (level = cur->bc_nlevels - 1; level >= 0; level--) {
4337 /* grab the left hand block */
4338 error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
4342 /* readahead the left most block for the next level down */
4344 union xfs_btree_ptr *ptr;
4346 ptr = xfs_btree_ptr_addr(cur, 1, block);
4347 xfs_btree_readahead_ptr(cur, ptr, 1);
4349 /* save for the next iteration of the loop */
4353 /* for each buffer in the level */
4355 error = xfs_btree_visit_block(cur, level, fn, data);
4358 if (error != -ENOENT)
4366 * Change the owner of a btree.
4368 * The mechanism we use here is ordered buffer logging. Because we don't know
4369 * how many buffers were are going to need to modify, we don't really want to
4370 * have to make transaction reservations for the worst case of every buffer in a
4371 * full size btree as that may be more space that we can fit in the log....
4373 * We do the btree walk in the most optimal manner possible - we have sibling
4374 * pointers so we can just walk all the blocks on each level from left to right
4375 * in a single pass, and then move to the next level and do the same. We can
4376 * also do readahead on the sibling pointers to get IO moving more quickly,
4377 * though for slow disks this is unlikely to make much difference to performance
4378 * as the amount of CPU work we have to do before moving to the next block is
4381 * For each btree block that we load, modify the owner appropriately, set the
4382 * buffer as an ordered buffer and log it appropriately. We need to ensure that
4383 * we mark the region we change dirty so that if the buffer is relogged in
4384 * a subsequent transaction the changes we make here as an ordered buffer are
4385 * correctly relogged in that transaction. If we are in recovery context, then
4386 * just queue the modified buffer as delayed write buffer so the transaction
4387 * recovery completion writes the changes to disk.
4389 struct xfs_btree_block_change_owner_info {
4390 __uint64_t new_owner;
4391 struct list_head *buffer_list;
4395 xfs_btree_block_change_owner(
4396 struct xfs_btree_cur *cur,
4400 struct xfs_btree_block_change_owner_info *bbcoi = data;
4401 struct xfs_btree_block *block;
4404 /* modify the owner */
4405 block = xfs_btree_get_block(cur, level, &bp);
4406 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
4407 block->bb_u.l.bb_owner = cpu_to_be64(bbcoi->new_owner);
4409 block->bb_u.s.bb_owner = cpu_to_be32(bbcoi->new_owner);
4412 * If the block is a root block hosted in an inode, we might not have a
4413 * buffer pointer here and we shouldn't attempt to log the change as the
4414 * information is already held in the inode and discarded when the root
4415 * block is formatted into the on-disk inode fork. We still change it,
4416 * though, so everything is consistent in memory.
4420 xfs_trans_ordered_buf(cur->bc_tp, bp);
4421 xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
4423 xfs_buf_delwri_queue(bp, bbcoi->buffer_list);
4426 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
4427 ASSERT(level == cur->bc_nlevels - 1);
4434 xfs_btree_change_owner(
4435 struct xfs_btree_cur *cur,
4436 __uint64_t new_owner,
4437 struct list_head *buffer_list)
4439 struct xfs_btree_block_change_owner_info bbcoi;
4441 bbcoi.new_owner = new_owner;
4442 bbcoi.buffer_list = buffer_list;
4444 return xfs_btree_visit_blocks(cur, xfs_btree_block_change_owner,
4449 * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format
4452 * @bp: buffer containing the btree block
4453 * @max_recs: pointer to the m_*_mxr max records field in the xfs mount
4454 * @pag_max_level: pointer to the per-ag max level field
4457 xfs_btree_sblock_v5hdr_verify(
4460 struct xfs_mount *mp = bp->b_target->bt_mount;
4461 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4462 struct xfs_perag *pag = bp->b_pag;
4464 if (!xfs_sb_version_hascrc(&mp->m_sb))
4466 if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
4468 if (block->bb_u.s.bb_blkno != cpu_to_be64(bp->b_bn))
4470 if (pag && be32_to_cpu(block->bb_u.s.bb_owner) != pag->pag_agno)
4476 * xfs_btree_sblock_verify() -- verify a short-format btree block
4478 * @bp: buffer containing the btree block
4479 * @max_recs: maximum records allowed in this btree node
4482 xfs_btree_sblock_verify(
4484 unsigned int max_recs)
4486 struct xfs_mount *mp = bp->b_target->bt_mount;
4487 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4489 /* numrecs verification */
4490 if (be16_to_cpu(block->bb_numrecs) > max_recs)
4493 /* sibling pointer verification */
4494 if (!block->bb_u.s.bb_leftsib ||
4495 (be32_to_cpu(block->bb_u.s.bb_leftsib) >= mp->m_sb.sb_agblocks &&
4496 block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK)))
4498 if (!block->bb_u.s.bb_rightsib ||
4499 (be32_to_cpu(block->bb_u.s.bb_rightsib) >= mp->m_sb.sb_agblocks &&
4500 block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK)))
4507 * Calculate the number of btree levels needed to store a given number of
4508 * records in a short-format btree.
4511 xfs_btree_compute_maxlevels(
4512 struct xfs_mount *mp,
4517 unsigned long maxblocks;
4519 maxblocks = (len + limits[0] - 1) / limits[0];
4520 for (level = 1; maxblocks > 1; level++)
4521 maxblocks = (maxblocks + limits[1] - 1) / limits[1];
4526 * Query a regular btree for all records overlapping a given interval.
4527 * Start with a LE lookup of the key of low_rec and return all records
4528 * until we find a record with a key greater than the key of high_rec.
4531 xfs_btree_simple_query_range(
4532 struct xfs_btree_cur *cur,
4533 union xfs_btree_key *low_key,
4534 union xfs_btree_key *high_key,
4535 xfs_btree_query_range_fn fn,
4538 union xfs_btree_rec *recp;
4539 union xfs_btree_key rec_key;
4542 bool firstrec = true;
4545 ASSERT(cur->bc_ops->init_high_key_from_rec);
4546 ASSERT(cur->bc_ops->diff_two_keys);
4549 * Find the leftmost record. The btree cursor must be set
4550 * to the low record used to generate low_key.
4553 error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
4558 /* Find the record. */
4559 error = xfs_btree_get_rec(cur, &recp, &stat);
4562 cur->bc_ops->init_high_key_from_rec(&rec_key, recp);
4564 /* Skip if high_key(rec) < low_key. */
4567 diff = cur->bc_ops->diff_two_keys(cur, low_key,
4573 /* Stop if high_key < low_key(rec). */
4574 diff = cur->bc_ops->diff_two_keys(cur, &rec_key, high_key);
4579 error = fn(cur, recp, priv);
4580 if (error < 0 || error == XFS_BTREE_QUERY_RANGE_ABORT)
4584 /* Move on to the next record. */
4585 error = xfs_btree_increment(cur, 0, &stat);
4595 * Query an overlapped interval btree for all records overlapping a given
4596 * interval. This function roughly follows the algorithm given in
4597 * "Interval Trees" of _Introduction to Algorithms_, which is section
4598 * 14.3 in the 2nd and 3rd editions.
4600 * First, generate keys for the low and high records passed in.
4602 * For any leaf node, generate the high and low keys for the record.
4603 * If the record keys overlap with the query low/high keys, pass the
4604 * record to the function iterator.
4606 * For any internal node, compare the low and high keys of each
4607 * pointer against the query low/high keys. If there's an overlap,
4608 * follow the pointer.
4610 * As an optimization, we stop scanning a block when we find a low key
4611 * that is greater than the query's high key.
4614 xfs_btree_overlapped_query_range(
4615 struct xfs_btree_cur *cur,
4616 union xfs_btree_key *low_key,
4617 union xfs_btree_key *high_key,
4618 xfs_btree_query_range_fn fn,
4621 union xfs_btree_ptr ptr;
4622 union xfs_btree_ptr *pp;
4623 union xfs_btree_key rec_key;
4624 union xfs_btree_key rec_hkey;
4625 union xfs_btree_key *lkp;
4626 union xfs_btree_key *hkp;
4627 union xfs_btree_rec *recp;
4628 struct xfs_btree_block *block;
4636 /* Load the root of the btree. */
4637 level = cur->bc_nlevels - 1;
4638 cur->bc_ops->init_ptr_from_cur(cur, &ptr);
4639 error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
4642 xfs_btree_get_block(cur, level, &bp);
4643 trace_xfs_btree_overlapped_query_range(cur, level, bp);
4645 error = xfs_btree_check_block(cur, block, level, bp);
4649 cur->bc_ptrs[level] = 1;
4651 while (level < cur->bc_nlevels) {
4652 block = xfs_btree_get_block(cur, level, &bp);
4654 /* End of node, pop back towards the root. */
4655 if (cur->bc_ptrs[level] > be16_to_cpu(block->bb_numrecs)) {
4657 if (level < cur->bc_nlevels - 1)
4658 cur->bc_ptrs[level + 1]++;
4664 /* Handle a leaf node. */
4665 recp = xfs_btree_rec_addr(cur, cur->bc_ptrs[0], block);
4667 cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
4668 ldiff = cur->bc_ops->diff_two_keys(cur, &rec_hkey,
4671 cur->bc_ops->init_key_from_rec(&rec_key, recp);
4672 hdiff = cur->bc_ops->diff_two_keys(cur, high_key,
4676 * If (record's high key >= query's low key) and
4677 * (query's high key >= record's low key), then
4678 * this record overlaps the query range; callback.
4680 if (ldiff >= 0 && hdiff >= 0) {
4681 error = fn(cur, recp, priv);
4683 error == XFS_BTREE_QUERY_RANGE_ABORT)
4685 } else if (hdiff < 0) {
4686 /* Record is larger than high key; pop. */
4689 cur->bc_ptrs[level]++;
4693 /* Handle an internal node. */
4694 lkp = xfs_btree_key_addr(cur, cur->bc_ptrs[level], block);
4695 hkp = xfs_btree_high_key_addr(cur, cur->bc_ptrs[level], block);
4696 pp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[level], block);
4698 ldiff = cur->bc_ops->diff_two_keys(cur, hkp, low_key);
4699 hdiff = cur->bc_ops->diff_two_keys(cur, high_key, lkp);
4702 * If (pointer's high key >= query's low key) and
4703 * (query's high key >= pointer's low key), then
4704 * this record overlaps the query range; follow pointer.
4706 if (ldiff >= 0 && hdiff >= 0) {
4708 error = xfs_btree_lookup_get_block(cur, level, pp,
4712 xfs_btree_get_block(cur, level, &bp);
4713 trace_xfs_btree_overlapped_query_range(cur, level, bp);
4715 error = xfs_btree_check_block(cur, block, level, bp);
4719 cur->bc_ptrs[level] = 1;
4721 } else if (hdiff < 0) {
4722 /* The low key is larger than the upper range; pop. */
4725 cur->bc_ptrs[level]++;
4730 * If we don't end this function with the cursor pointing at a record
4731 * block, a subsequent non-error cursor deletion will not release
4732 * node-level buffers, causing a buffer leak. This is quite possible
4733 * with a zero-results range query, so release the buffers if we
4734 * failed to return any results.
4736 if (cur->bc_bufs[0] == NULL) {
4737 for (i = 0; i < cur->bc_nlevels; i++) {
4738 if (cur->bc_bufs[i]) {
4739 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
4740 cur->bc_bufs[i] = NULL;
4741 cur->bc_ptrs[i] = 0;
4751 * Query a btree for all records overlapping a given interval of keys. The
4752 * supplied function will be called with each record found; return one of the
4753 * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
4754 * code. This function returns XFS_BTREE_QUERY_RANGE_ABORT, zero, or a
4755 * negative error code.
4758 xfs_btree_query_range(
4759 struct xfs_btree_cur *cur,
4760 union xfs_btree_irec *low_rec,
4761 union xfs_btree_irec *high_rec,
4762 xfs_btree_query_range_fn fn,
4765 union xfs_btree_rec rec;
4766 union xfs_btree_key low_key;
4767 union xfs_btree_key high_key;
4769 /* Find the keys of both ends of the interval. */
4770 cur->bc_rec = *high_rec;
4771 cur->bc_ops->init_rec_from_cur(cur, &rec);
4772 cur->bc_ops->init_key_from_rec(&high_key, &rec);
4774 cur->bc_rec = *low_rec;
4775 cur->bc_ops->init_rec_from_cur(cur, &rec);
4776 cur->bc_ops->init_key_from_rec(&low_key, &rec);
4778 /* Enforce low key < high key. */
4779 if (cur->bc_ops->diff_two_keys(cur, &low_key, &high_key) > 0)
4782 if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING))
4783 return xfs_btree_simple_query_range(cur, &low_key,
4784 &high_key, fn, priv);
4785 return xfs_btree_overlapped_query_range(cur, &low_key, &high_key,