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,
59 __uint32_t magic = xfs_magics[crc][btnum];
61 /* Ensure we asked for crc for crc-only magics. */
66 STATIC int /* error (0 or EFSCORRUPTED) */
67 xfs_btree_check_lblock(
68 struct xfs_btree_cur *cur, /* btree cursor */
69 struct xfs_btree_block *block, /* btree long form block pointer */
70 int level, /* level of the btree block */
71 struct xfs_buf *bp) /* buffer for block, if any */
73 int lblock_ok = 1; /* block passes checks */
74 struct xfs_mount *mp; /* file system mount point */
75 xfs_btnum_t btnum = cur->bc_btnum;
79 crc = xfs_sb_version_hascrc(&mp->m_sb);
82 lblock_ok = lblock_ok &&
83 uuid_equal(&block->bb_u.l.bb_uuid,
84 &mp->m_sb.sb_meta_uuid) &&
85 block->bb_u.l.bb_blkno == cpu_to_be64(
86 bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
89 lblock_ok = lblock_ok &&
90 be32_to_cpu(block->bb_magic) == xfs_btree_magic(crc, btnum) &&
91 be16_to_cpu(block->bb_level) == level &&
92 be16_to_cpu(block->bb_numrecs) <=
93 cur->bc_ops->get_maxrecs(cur, level) &&
94 block->bb_u.l.bb_leftsib &&
95 (block->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK) ||
96 XFS_FSB_SANITY_CHECK(mp,
97 be64_to_cpu(block->bb_u.l.bb_leftsib))) &&
98 block->bb_u.l.bb_rightsib &&
99 (block->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK) ||
100 XFS_FSB_SANITY_CHECK(mp,
101 be64_to_cpu(block->bb_u.l.bb_rightsib)));
103 if (unlikely(XFS_TEST_ERROR(!lblock_ok, mp,
104 XFS_ERRTAG_BTREE_CHECK_LBLOCK,
105 XFS_RANDOM_BTREE_CHECK_LBLOCK))) {
107 trace_xfs_btree_corrupt(bp, _RET_IP_);
108 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
109 return -EFSCORRUPTED;
114 STATIC int /* error (0 or EFSCORRUPTED) */
115 xfs_btree_check_sblock(
116 struct xfs_btree_cur *cur, /* btree cursor */
117 struct xfs_btree_block *block, /* btree short form block pointer */
118 int level, /* level of the btree block */
119 struct xfs_buf *bp) /* buffer containing block */
121 struct xfs_mount *mp; /* file system mount point */
122 struct xfs_buf *agbp; /* buffer for ag. freespace struct */
123 struct xfs_agf *agf; /* ag. freespace structure */
124 xfs_agblock_t agflen; /* native ag. freespace length */
125 int sblock_ok = 1; /* block passes checks */
126 xfs_btnum_t btnum = cur->bc_btnum;
130 crc = xfs_sb_version_hascrc(&mp->m_sb);
131 agbp = cur->bc_private.a.agbp;
132 agf = XFS_BUF_TO_AGF(agbp);
133 agflen = be32_to_cpu(agf->agf_length);
136 sblock_ok = sblock_ok &&
137 uuid_equal(&block->bb_u.s.bb_uuid,
138 &mp->m_sb.sb_meta_uuid) &&
139 block->bb_u.s.bb_blkno == cpu_to_be64(
140 bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
143 sblock_ok = sblock_ok &&
144 be32_to_cpu(block->bb_magic) == xfs_btree_magic(crc, btnum) &&
145 be16_to_cpu(block->bb_level) == level &&
146 be16_to_cpu(block->bb_numrecs) <=
147 cur->bc_ops->get_maxrecs(cur, level) &&
148 (block->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) ||
149 be32_to_cpu(block->bb_u.s.bb_leftsib) < agflen) &&
150 block->bb_u.s.bb_leftsib &&
151 (block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK) ||
152 be32_to_cpu(block->bb_u.s.bb_rightsib) < agflen) &&
153 block->bb_u.s.bb_rightsib;
155 if (unlikely(XFS_TEST_ERROR(!sblock_ok, mp,
156 XFS_ERRTAG_BTREE_CHECK_SBLOCK,
157 XFS_RANDOM_BTREE_CHECK_SBLOCK))) {
159 trace_xfs_btree_corrupt(bp, _RET_IP_);
160 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
161 return -EFSCORRUPTED;
167 * Debug routine: check that block header is ok.
170 xfs_btree_check_block(
171 struct xfs_btree_cur *cur, /* btree cursor */
172 struct xfs_btree_block *block, /* generic btree block pointer */
173 int level, /* level of the btree block */
174 struct xfs_buf *bp) /* buffer containing block, if any */
176 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
177 return xfs_btree_check_lblock(cur, block, level, bp);
179 return xfs_btree_check_sblock(cur, block, level, bp);
183 * Check that (long) pointer is ok.
185 int /* error (0 or EFSCORRUPTED) */
186 xfs_btree_check_lptr(
187 struct xfs_btree_cur *cur, /* btree cursor */
188 xfs_fsblock_t bno, /* btree block disk address */
189 int level) /* btree block level */
191 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
193 bno != NULLFSBLOCK &&
194 XFS_FSB_SANITY_CHECK(cur->bc_mp, bno));
200 * Check that (short) pointer is ok.
202 STATIC int /* error (0 or EFSCORRUPTED) */
203 xfs_btree_check_sptr(
204 struct xfs_btree_cur *cur, /* btree cursor */
205 xfs_agblock_t bno, /* btree block disk address */
206 int level) /* btree block level */
208 xfs_agblock_t agblocks = cur->bc_mp->m_sb.sb_agblocks;
210 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
212 bno != NULLAGBLOCK &&
219 * Check that block ptr is ok.
221 STATIC int /* error (0 or EFSCORRUPTED) */
223 struct xfs_btree_cur *cur, /* btree cursor */
224 union xfs_btree_ptr *ptr, /* btree block disk address */
225 int index, /* offset from ptr to check */
226 int level) /* btree block level */
228 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
229 return xfs_btree_check_lptr(cur,
230 be64_to_cpu((&ptr->l)[index]), level);
232 return xfs_btree_check_sptr(cur,
233 be32_to_cpu((&ptr->s)[index]), level);
239 * Calculate CRC on the whole btree block and stuff it into the
240 * long-form btree header.
242 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
243 * it into the buffer so recovery knows what the last modification was that made
247 xfs_btree_lblock_calc_crc(
250 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
251 struct xfs_buf_log_item *bip = bp->b_fspriv;
253 if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
256 block->bb_u.l.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
257 xfs_buf_update_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
261 xfs_btree_lblock_verify_crc(
264 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
265 struct xfs_mount *mp = bp->b_target->bt_mount;
267 if (xfs_sb_version_hascrc(&mp->m_sb)) {
268 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.l.bb_lsn)))
270 return xfs_buf_verify_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
277 * Calculate CRC on the whole btree block and stuff it into the
278 * short-form btree header.
280 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
281 * it into the buffer so recovery knows what the last modification was that made
285 xfs_btree_sblock_calc_crc(
288 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
289 struct xfs_buf_log_item *bip = bp->b_fspriv;
291 if (!xfs_sb_version_hascrc(&bp->b_target->bt_mount->m_sb))
294 block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
295 xfs_buf_update_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
299 xfs_btree_sblock_verify_crc(
302 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
303 struct xfs_mount *mp = bp->b_target->bt_mount;
305 if (xfs_sb_version_hascrc(&mp->m_sb)) {
306 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.s.bb_lsn)))
308 return xfs_buf_verify_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
315 xfs_btree_free_block(
316 struct xfs_btree_cur *cur,
321 error = cur->bc_ops->free_block(cur, bp);
323 xfs_trans_binval(cur->bc_tp, bp);
324 XFS_BTREE_STATS_INC(cur, free);
330 * Delete the btree cursor.
333 xfs_btree_del_cursor(
334 xfs_btree_cur_t *cur, /* btree cursor */
335 int error) /* del because of error */
337 int i; /* btree level */
340 * Clear the buffer pointers, and release the buffers.
341 * If we're doing this in the face of an error, we
342 * need to make sure to inspect all of the entries
343 * in the bc_bufs array for buffers to be unlocked.
344 * This is because some of the btree code works from
345 * level n down to 0, and if we get an error along
346 * the way we won't have initialized all the entries
349 for (i = 0; i < cur->bc_nlevels; i++) {
351 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
356 * Can't free a bmap cursor without having dealt with the
357 * allocated indirect blocks' accounting.
359 ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP ||
360 cur->bc_private.b.allocated == 0);
364 kmem_zone_free(xfs_btree_cur_zone, cur);
368 * Duplicate the btree cursor.
369 * Allocate a new one, copy the record, re-get the buffers.
372 xfs_btree_dup_cursor(
373 xfs_btree_cur_t *cur, /* input cursor */
374 xfs_btree_cur_t **ncur) /* output cursor */
376 xfs_buf_t *bp; /* btree block's buffer pointer */
377 int error; /* error return value */
378 int i; /* level number of btree block */
379 xfs_mount_t *mp; /* mount structure for filesystem */
380 xfs_btree_cur_t *new; /* new cursor value */
381 xfs_trans_t *tp; /* transaction pointer, can be NULL */
387 * Allocate a new cursor like the old one.
389 new = cur->bc_ops->dup_cursor(cur);
392 * Copy the record currently in the cursor.
394 new->bc_rec = cur->bc_rec;
397 * For each level current, re-get the buffer and copy the ptr value.
399 for (i = 0; i < new->bc_nlevels; i++) {
400 new->bc_ptrs[i] = cur->bc_ptrs[i];
401 new->bc_ra[i] = cur->bc_ra[i];
402 bp = cur->bc_bufs[i];
404 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
405 XFS_BUF_ADDR(bp), mp->m_bsize,
407 cur->bc_ops->buf_ops);
409 xfs_btree_del_cursor(new, error);
414 new->bc_bufs[i] = bp;
421 * XFS btree block layout and addressing:
423 * There are two types of blocks in the btree: leaf and non-leaf blocks.
425 * The leaf record start with a header then followed by records containing
426 * the values. A non-leaf block also starts with the same header, and
427 * then first contains lookup keys followed by an equal number of pointers
428 * to the btree blocks at the previous level.
430 * +--------+-------+-------+-------+-------+-------+-------+
431 * Leaf: | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
432 * +--------+-------+-------+-------+-------+-------+-------+
434 * +--------+-------+-------+-------+-------+-------+-------+
435 * Non-Leaf: | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
436 * +--------+-------+-------+-------+-------+-------+-------+
438 * The header is called struct xfs_btree_block for reasons better left unknown
439 * and comes in different versions for short (32bit) and long (64bit) block
440 * pointers. The record and key structures are defined by the btree instances
441 * and opaque to the btree core. The block pointers are simple disk endian
442 * integers, available in a short (32bit) and long (64bit) variant.
444 * The helpers below calculate the offset of a given record, key or pointer
445 * into a btree block (xfs_btree_*_offset) or return a pointer to the given
446 * record, key or pointer (xfs_btree_*_addr). Note that all addressing
447 * inside the btree block is done using indices starting at one, not zero!
449 * If XFS_BTREE_OVERLAPPING is set, then this btree supports keys containing
450 * overlapping intervals. In such a tree, records are still sorted lowest to
451 * highest and indexed by the smallest key value that refers to the record.
452 * However, nodes are different: each pointer has two associated keys -- one
453 * indexing the lowest key available in the block(s) below (the same behavior
454 * as the key in a regular btree) and another indexing the highest key
455 * available in the block(s) below. Because records are /not/ sorted by the
456 * highest key, all leaf block updates require us to compute the highest key
457 * that matches any record in the leaf and to recursively update the high keys
458 * in the nodes going further up in the tree, if necessary. Nodes look like
461 * +--------+-----+-----+-----+-----+-----+-------+-------+-----+
462 * Non-Leaf: | header | lo1 | hi1 | lo2 | hi2 | ... | ptr 1 | ptr 2 | ... |
463 * +--------+-----+-----+-----+-----+-----+-------+-------+-----+
465 * To perform an interval query on an overlapped tree, perform the usual
466 * depth-first search and use the low and high keys to decide if we can skip
467 * that particular node. If a leaf node is reached, return the records that
468 * intersect the interval. Note that an interval query may return numerous
469 * entries. For a non-overlapped tree, simply search for the record associated
470 * with the lowest key and iterate forward until a non-matching record is
471 * found. Section 14.3 ("Interval Trees") of _Introduction to Algorithms_ by
472 * Cormen, Leiserson, Rivest, and Stein (2nd or 3rd ed. only) discuss this in
475 * Why do we care about overlapping intervals? Let's say you have a bunch of
476 * reverse mapping records on a reflink filesystem:
478 * 1: +- file A startblock B offset C length D -----------+
479 * 2: +- file E startblock F offset G length H --------------+
480 * 3: +- file I startblock F offset J length K --+
481 * 4: +- file L... --+
483 * Now say we want to map block (B+D) into file A at offset (C+D). Ideally,
484 * we'd simply increment the length of record 1. But how do we find the record
485 * that ends at (B+D-1) (i.e. record 1)? A LE lookup of (B+D-1) would return
486 * record 3 because the keys are ordered first by startblock. An interval
487 * query would return records 1 and 2 because they both overlap (B+D-1), and
488 * from that we can pick out record 1 as the appropriate left neighbor.
490 * In the non-overlapped case you can do a LE lookup and decrement the cursor
491 * because a record's interval must end before the next record.
495 * Return size of the btree block header for this btree instance.
497 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
499 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
500 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
501 return XFS_BTREE_LBLOCK_CRC_LEN;
502 return XFS_BTREE_LBLOCK_LEN;
504 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
505 return XFS_BTREE_SBLOCK_CRC_LEN;
506 return XFS_BTREE_SBLOCK_LEN;
510 * Return size of btree block pointers for this btree instance.
512 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
514 return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
515 sizeof(__be64) : sizeof(__be32);
519 * Calculate offset of the n-th record in a btree block.
522 xfs_btree_rec_offset(
523 struct xfs_btree_cur *cur,
526 return xfs_btree_block_len(cur) +
527 (n - 1) * cur->bc_ops->rec_len;
531 * Calculate offset of the n-th key in a btree block.
534 xfs_btree_key_offset(
535 struct xfs_btree_cur *cur,
538 return xfs_btree_block_len(cur) +
539 (n - 1) * cur->bc_ops->key_len;
543 * Calculate offset of the n-th high key in a btree block.
546 xfs_btree_high_key_offset(
547 struct xfs_btree_cur *cur,
550 return xfs_btree_block_len(cur) +
551 (n - 1) * cur->bc_ops->key_len + (cur->bc_ops->key_len / 2);
555 * Calculate offset of the n-th block pointer in a btree block.
558 xfs_btree_ptr_offset(
559 struct xfs_btree_cur *cur,
563 return xfs_btree_block_len(cur) +
564 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
565 (n - 1) * xfs_btree_ptr_len(cur);
569 * Return a pointer to the n-th record in the btree block.
571 STATIC union xfs_btree_rec *
573 struct xfs_btree_cur *cur,
575 struct xfs_btree_block *block)
577 return (union xfs_btree_rec *)
578 ((char *)block + xfs_btree_rec_offset(cur, n));
582 * Return a pointer to the n-th key in the btree block.
584 STATIC union xfs_btree_key *
586 struct xfs_btree_cur *cur,
588 struct xfs_btree_block *block)
590 return (union xfs_btree_key *)
591 ((char *)block + xfs_btree_key_offset(cur, n));
595 * Return a pointer to the n-th high key in the btree block.
597 STATIC union xfs_btree_key *
598 xfs_btree_high_key_addr(
599 struct xfs_btree_cur *cur,
601 struct xfs_btree_block *block)
603 return (union xfs_btree_key *)
604 ((char *)block + xfs_btree_high_key_offset(cur, n));
608 * Return a pointer to the n-th block pointer in the btree block.
610 STATIC union xfs_btree_ptr *
612 struct xfs_btree_cur *cur,
614 struct xfs_btree_block *block)
616 int level = xfs_btree_get_level(block);
618 ASSERT(block->bb_level != 0);
620 return (union xfs_btree_ptr *)
621 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
625 * Get the root block which is stored in the inode.
627 * For now this btree implementation assumes the btree root is always
628 * stored in the if_broot field of an inode fork.
630 STATIC struct xfs_btree_block *
632 struct xfs_btree_cur *cur)
634 struct xfs_ifork *ifp;
636 ifp = XFS_IFORK_PTR(cur->bc_private.b.ip, cur->bc_private.b.whichfork);
637 return (struct xfs_btree_block *)ifp->if_broot;
641 * Retrieve the block pointer from the cursor at the given level.
642 * This may be an inode btree root or from a buffer.
644 STATIC struct xfs_btree_block * /* generic btree block pointer */
646 struct xfs_btree_cur *cur, /* btree cursor */
647 int level, /* level in btree */
648 struct xfs_buf **bpp) /* buffer containing the block */
650 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
651 (level == cur->bc_nlevels - 1)) {
653 return xfs_btree_get_iroot(cur);
656 *bpp = cur->bc_bufs[level];
657 return XFS_BUF_TO_BLOCK(*bpp);
661 * Get a buffer for the block, return it with no data read.
662 * Long-form addressing.
664 xfs_buf_t * /* buffer for fsbno */
666 xfs_mount_t *mp, /* file system mount point */
667 xfs_trans_t *tp, /* transaction pointer */
668 xfs_fsblock_t fsbno, /* file system block number */
669 uint lock) /* lock flags for get_buf */
671 xfs_daddr_t d; /* real disk block address */
673 ASSERT(fsbno != NULLFSBLOCK);
674 d = XFS_FSB_TO_DADDR(mp, fsbno);
675 return xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
679 * Get a buffer for the block, return it with no data read.
680 * Short-form addressing.
682 xfs_buf_t * /* buffer for agno/agbno */
684 xfs_mount_t *mp, /* file system mount point */
685 xfs_trans_t *tp, /* transaction pointer */
686 xfs_agnumber_t agno, /* allocation group number */
687 xfs_agblock_t agbno, /* allocation group block number */
688 uint lock) /* lock flags for get_buf */
690 xfs_daddr_t d; /* real disk block address */
692 ASSERT(agno != NULLAGNUMBER);
693 ASSERT(agbno != NULLAGBLOCK);
694 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
695 return xfs_trans_get_buf(tp, mp->m_ddev_targp, d, mp->m_bsize, lock);
699 * Check for the cursor referring to the last block at the given level.
701 int /* 1=is last block, 0=not last block */
702 xfs_btree_islastblock(
703 xfs_btree_cur_t *cur, /* btree cursor */
704 int level) /* level to check */
706 struct xfs_btree_block *block; /* generic btree block pointer */
707 xfs_buf_t *bp; /* buffer containing block */
709 block = xfs_btree_get_block(cur, level, &bp);
710 xfs_btree_check_block(cur, block, level, bp);
711 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
712 return block->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK);
714 return block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK);
718 * Change the cursor to point to the first record at the given level.
719 * Other levels are unaffected.
721 STATIC int /* success=1, failure=0 */
723 xfs_btree_cur_t *cur, /* btree cursor */
724 int level) /* level to change */
726 struct xfs_btree_block *block; /* generic btree block pointer */
727 xfs_buf_t *bp; /* buffer containing block */
730 * Get the block pointer for this level.
732 block = xfs_btree_get_block(cur, level, &bp);
733 xfs_btree_check_block(cur, block, level, bp);
735 * It's empty, there is no such record.
737 if (!block->bb_numrecs)
740 * Set the ptr value to 1, that's the first record/key.
742 cur->bc_ptrs[level] = 1;
747 * Change the cursor to point to the last record in the current block
748 * at the given level. Other levels are unaffected.
750 STATIC int /* success=1, failure=0 */
752 xfs_btree_cur_t *cur, /* btree cursor */
753 int level) /* level to change */
755 struct xfs_btree_block *block; /* generic btree block pointer */
756 xfs_buf_t *bp; /* buffer containing block */
759 * Get the block pointer for this level.
761 block = xfs_btree_get_block(cur, level, &bp);
762 xfs_btree_check_block(cur, block, level, bp);
764 * It's empty, there is no such record.
766 if (!block->bb_numrecs)
769 * Set the ptr value to numrecs, that's the last record/key.
771 cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
776 * Compute first and last byte offsets for the fields given.
777 * Interprets the offsets table, which contains struct field offsets.
781 __int64_t fields, /* bitmask of fields */
782 const short *offsets, /* table of field offsets */
783 int nbits, /* number of bits to inspect */
784 int *first, /* output: first byte offset */
785 int *last) /* output: last byte offset */
787 int i; /* current bit number */
788 __int64_t imask; /* mask for current bit number */
792 * Find the lowest bit, so the first byte offset.
794 for (i = 0, imask = 1LL; ; i++, imask <<= 1) {
795 if (imask & fields) {
801 * Find the highest bit, so the last byte offset.
803 for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) {
804 if (imask & fields) {
805 *last = offsets[i + 1] - 1;
812 * Get a buffer for the block, return it read in.
813 * Long-form addressing.
817 struct xfs_mount *mp, /* file system mount point */
818 struct xfs_trans *tp, /* transaction pointer */
819 xfs_fsblock_t fsbno, /* file system block number */
820 uint lock, /* lock flags for read_buf */
821 struct xfs_buf **bpp, /* buffer for fsbno */
822 int refval, /* ref count value for buffer */
823 const struct xfs_buf_ops *ops)
825 struct xfs_buf *bp; /* return value */
826 xfs_daddr_t d; /* real disk block address */
829 ASSERT(fsbno != NULLFSBLOCK);
830 d = XFS_FSB_TO_DADDR(mp, fsbno);
831 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
832 mp->m_bsize, lock, &bp, ops);
836 xfs_buf_set_ref(bp, refval);
842 * Read-ahead the block, don't wait for it, don't return a buffer.
843 * Long-form addressing.
847 xfs_btree_reada_bufl(
848 struct xfs_mount *mp, /* file system mount point */
849 xfs_fsblock_t fsbno, /* file system block number */
850 xfs_extlen_t count, /* count of filesystem blocks */
851 const struct xfs_buf_ops *ops)
855 ASSERT(fsbno != NULLFSBLOCK);
856 d = XFS_FSB_TO_DADDR(mp, fsbno);
857 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
861 * Read-ahead the block, don't wait for it, don't return a buffer.
862 * Short-form addressing.
866 xfs_btree_reada_bufs(
867 struct xfs_mount *mp, /* file system mount point */
868 xfs_agnumber_t agno, /* allocation group number */
869 xfs_agblock_t agbno, /* allocation group block number */
870 xfs_extlen_t count, /* count of filesystem blocks */
871 const struct xfs_buf_ops *ops)
875 ASSERT(agno != NULLAGNUMBER);
876 ASSERT(agbno != NULLAGBLOCK);
877 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
878 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
882 xfs_btree_readahead_lblock(
883 struct xfs_btree_cur *cur,
885 struct xfs_btree_block *block)
888 xfs_fsblock_t left = be64_to_cpu(block->bb_u.l.bb_leftsib);
889 xfs_fsblock_t right = be64_to_cpu(block->bb_u.l.bb_rightsib);
891 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLFSBLOCK) {
892 xfs_btree_reada_bufl(cur->bc_mp, left, 1,
893 cur->bc_ops->buf_ops);
897 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLFSBLOCK) {
898 xfs_btree_reada_bufl(cur->bc_mp, right, 1,
899 cur->bc_ops->buf_ops);
907 xfs_btree_readahead_sblock(
908 struct xfs_btree_cur *cur,
910 struct xfs_btree_block *block)
913 xfs_agblock_t left = be32_to_cpu(block->bb_u.s.bb_leftsib);
914 xfs_agblock_t right = be32_to_cpu(block->bb_u.s.bb_rightsib);
917 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
918 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
919 left, 1, cur->bc_ops->buf_ops);
923 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
924 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_private.a.agno,
925 right, 1, cur->bc_ops->buf_ops);
933 * Read-ahead btree blocks, at the given level.
934 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
938 struct xfs_btree_cur *cur, /* btree cursor */
939 int lev, /* level in btree */
940 int lr) /* left/right bits */
942 struct xfs_btree_block *block;
945 * No readahead needed if we are at the root level and the
946 * btree root is stored in the inode.
948 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
949 (lev == cur->bc_nlevels - 1))
952 if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
955 cur->bc_ra[lev] |= lr;
956 block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
958 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
959 return xfs_btree_readahead_lblock(cur, lr, block);
960 return xfs_btree_readahead_sblock(cur, lr, block);
964 xfs_btree_ptr_to_daddr(
965 struct xfs_btree_cur *cur,
966 union xfs_btree_ptr *ptr)
968 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
969 ASSERT(ptr->l != cpu_to_be64(NULLFSBLOCK));
971 return XFS_FSB_TO_DADDR(cur->bc_mp, be64_to_cpu(ptr->l));
973 ASSERT(cur->bc_private.a.agno != NULLAGNUMBER);
974 ASSERT(ptr->s != cpu_to_be32(NULLAGBLOCK));
976 return XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_private.a.agno,
977 be32_to_cpu(ptr->s));
982 * Readahead @count btree blocks at the given @ptr location.
984 * We don't need to care about long or short form btrees here as we have a
985 * method of converting the ptr directly to a daddr available to us.
988 xfs_btree_readahead_ptr(
989 struct xfs_btree_cur *cur,
990 union xfs_btree_ptr *ptr,
993 xfs_buf_readahead(cur->bc_mp->m_ddev_targp,
994 xfs_btree_ptr_to_daddr(cur, ptr),
995 cur->bc_mp->m_bsize * count, cur->bc_ops->buf_ops);
999 * Set the buffer for level "lev" in the cursor to bp, releasing
1000 * any previous buffer.
1004 xfs_btree_cur_t *cur, /* btree cursor */
1005 int lev, /* level in btree */
1006 xfs_buf_t *bp) /* new buffer to set */
1008 struct xfs_btree_block *b; /* btree block */
1010 if (cur->bc_bufs[lev])
1011 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
1012 cur->bc_bufs[lev] = bp;
1013 cur->bc_ra[lev] = 0;
1015 b = XFS_BUF_TO_BLOCK(bp);
1016 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1017 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK))
1018 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1019 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK))
1020 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1022 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
1023 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1024 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
1025 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1030 xfs_btree_ptr_is_null(
1031 struct xfs_btree_cur *cur,
1032 union xfs_btree_ptr *ptr)
1034 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1035 return ptr->l == cpu_to_be64(NULLFSBLOCK);
1037 return ptr->s == cpu_to_be32(NULLAGBLOCK);
1041 xfs_btree_set_ptr_null(
1042 struct xfs_btree_cur *cur,
1043 union xfs_btree_ptr *ptr)
1045 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1046 ptr->l = cpu_to_be64(NULLFSBLOCK);
1048 ptr->s = cpu_to_be32(NULLAGBLOCK);
1052 * Get/set/init sibling pointers
1055 xfs_btree_get_sibling(
1056 struct xfs_btree_cur *cur,
1057 struct xfs_btree_block *block,
1058 union xfs_btree_ptr *ptr,
1061 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1063 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1064 if (lr == XFS_BB_RIGHTSIB)
1065 ptr->l = block->bb_u.l.bb_rightsib;
1067 ptr->l = block->bb_u.l.bb_leftsib;
1069 if (lr == XFS_BB_RIGHTSIB)
1070 ptr->s = block->bb_u.s.bb_rightsib;
1072 ptr->s = block->bb_u.s.bb_leftsib;
1077 xfs_btree_set_sibling(
1078 struct xfs_btree_cur *cur,
1079 struct xfs_btree_block *block,
1080 union xfs_btree_ptr *ptr,
1083 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1085 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1086 if (lr == XFS_BB_RIGHTSIB)
1087 block->bb_u.l.bb_rightsib = ptr->l;
1089 block->bb_u.l.bb_leftsib = ptr->l;
1091 if (lr == XFS_BB_RIGHTSIB)
1092 block->bb_u.s.bb_rightsib = ptr->s;
1094 block->bb_u.s.bb_leftsib = ptr->s;
1099 xfs_btree_init_block_int(
1100 struct xfs_mount *mp,
1101 struct xfs_btree_block *buf,
1109 int crc = xfs_sb_version_hascrc(&mp->m_sb);
1111 buf->bb_magic = cpu_to_be32(magic);
1112 buf->bb_level = cpu_to_be16(level);
1113 buf->bb_numrecs = cpu_to_be16(numrecs);
1115 if (flags & XFS_BTREE_LONG_PTRS) {
1116 buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK);
1117 buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK);
1119 buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
1120 buf->bb_u.l.bb_owner = cpu_to_be64(owner);
1121 uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid);
1122 buf->bb_u.l.bb_pad = 0;
1123 buf->bb_u.l.bb_lsn = 0;
1126 /* owner is a 32 bit value on short blocks */
1127 __u32 __owner = (__u32)owner;
1129 buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1130 buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1132 buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
1133 buf->bb_u.s.bb_owner = cpu_to_be32(__owner);
1134 uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid);
1135 buf->bb_u.s.bb_lsn = 0;
1141 xfs_btree_init_block(
1142 struct xfs_mount *mp,
1150 xfs_btree_init_block_int(mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1151 magic, level, numrecs, owner, flags);
1155 xfs_btree_init_block_cur(
1156 struct xfs_btree_cur *cur,
1162 int crc = xfs_sb_version_hascrc(&cur->bc_mp->m_sb);
1163 xfs_btnum_t btnum = cur->bc_btnum;
1166 * we can pull the owner from the cursor right now as the different
1167 * owners align directly with the pointer size of the btree. This may
1168 * change in future, but is safe for current users of the generic btree
1171 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1172 owner = cur->bc_private.b.ip->i_ino;
1174 owner = cur->bc_private.a.agno;
1176 xfs_btree_init_block_int(cur->bc_mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1177 xfs_btree_magic(crc, btnum), level, numrecs,
1178 owner, cur->bc_flags);
1182 * Return true if ptr is the last record in the btree and
1183 * we need to track updates to this record. The decision
1184 * will be further refined in the update_lastrec method.
1187 xfs_btree_is_lastrec(
1188 struct xfs_btree_cur *cur,
1189 struct xfs_btree_block *block,
1192 union xfs_btree_ptr ptr;
1196 if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
1199 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1200 if (!xfs_btree_ptr_is_null(cur, &ptr))
1206 xfs_btree_buf_to_ptr(
1207 struct xfs_btree_cur *cur,
1209 union xfs_btree_ptr *ptr)
1211 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1212 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1215 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
1222 struct xfs_btree_cur *cur,
1225 switch (cur->bc_btnum) {
1228 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
1231 case XFS_BTNUM_FINO:
1232 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
1234 case XFS_BTNUM_BMAP:
1235 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
1237 case XFS_BTNUM_RMAP:
1238 xfs_buf_set_ref(bp, XFS_RMAP_BTREE_REF);
1240 case XFS_BTNUM_REFC:
1241 xfs_buf_set_ref(bp, XFS_REFC_BTREE_REF);
1249 xfs_btree_get_buf_block(
1250 struct xfs_btree_cur *cur,
1251 union xfs_btree_ptr *ptr,
1253 struct xfs_btree_block **block,
1254 struct xfs_buf **bpp)
1256 struct xfs_mount *mp = cur->bc_mp;
1259 /* need to sort out how callers deal with failures first */
1260 ASSERT(!(flags & XBF_TRYLOCK));
1262 d = xfs_btree_ptr_to_daddr(cur, ptr);
1263 *bpp = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d,
1264 mp->m_bsize, flags);
1269 (*bpp)->b_ops = cur->bc_ops->buf_ops;
1270 *block = XFS_BUF_TO_BLOCK(*bpp);
1275 * Read in the buffer at the given ptr and return the buffer and
1276 * the block pointer within the buffer.
1279 xfs_btree_read_buf_block(
1280 struct xfs_btree_cur *cur,
1281 union xfs_btree_ptr *ptr,
1283 struct xfs_btree_block **block,
1284 struct xfs_buf **bpp)
1286 struct xfs_mount *mp = cur->bc_mp;
1290 /* need to sort out how callers deal with failures first */
1291 ASSERT(!(flags & XBF_TRYLOCK));
1293 d = xfs_btree_ptr_to_daddr(cur, ptr);
1294 error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
1295 mp->m_bsize, flags, bpp,
1296 cur->bc_ops->buf_ops);
1300 xfs_btree_set_refs(cur, *bpp);
1301 *block = XFS_BUF_TO_BLOCK(*bpp);
1306 * Copy keys from one btree block to another.
1309 xfs_btree_copy_keys(
1310 struct xfs_btree_cur *cur,
1311 union xfs_btree_key *dst_key,
1312 union xfs_btree_key *src_key,
1315 ASSERT(numkeys >= 0);
1316 memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1320 * Copy records from one btree block to another.
1323 xfs_btree_copy_recs(
1324 struct xfs_btree_cur *cur,
1325 union xfs_btree_rec *dst_rec,
1326 union xfs_btree_rec *src_rec,
1329 ASSERT(numrecs >= 0);
1330 memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1334 * Copy block pointers from one btree block to another.
1337 xfs_btree_copy_ptrs(
1338 struct xfs_btree_cur *cur,
1339 union xfs_btree_ptr *dst_ptr,
1340 union xfs_btree_ptr *src_ptr,
1343 ASSERT(numptrs >= 0);
1344 memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1348 * Shift keys one index left/right inside a single btree block.
1351 xfs_btree_shift_keys(
1352 struct xfs_btree_cur *cur,
1353 union xfs_btree_key *key,
1359 ASSERT(numkeys >= 0);
1360 ASSERT(dir == 1 || dir == -1);
1362 dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1363 memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1367 * Shift records one index left/right inside a single btree block.
1370 xfs_btree_shift_recs(
1371 struct xfs_btree_cur *cur,
1372 union xfs_btree_rec *rec,
1378 ASSERT(numrecs >= 0);
1379 ASSERT(dir == 1 || dir == -1);
1381 dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1382 memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1386 * Shift block pointers one index left/right inside a single btree block.
1389 xfs_btree_shift_ptrs(
1390 struct xfs_btree_cur *cur,
1391 union xfs_btree_ptr *ptr,
1397 ASSERT(numptrs >= 0);
1398 ASSERT(dir == 1 || dir == -1);
1400 dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1401 memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1405 * Log key values from the btree block.
1409 struct xfs_btree_cur *cur,
1414 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1415 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1418 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1419 xfs_trans_log_buf(cur->bc_tp, bp,
1420 xfs_btree_key_offset(cur, first),
1421 xfs_btree_key_offset(cur, last + 1) - 1);
1423 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1424 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1427 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1431 * Log record values from the btree block.
1435 struct xfs_btree_cur *cur,
1440 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1441 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1443 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1444 xfs_trans_log_buf(cur->bc_tp, bp,
1445 xfs_btree_rec_offset(cur, first),
1446 xfs_btree_rec_offset(cur, last + 1) - 1);
1448 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1452 * Log block pointer fields from a btree block (nonleaf).
1456 struct xfs_btree_cur *cur, /* btree cursor */
1457 struct xfs_buf *bp, /* buffer containing btree block */
1458 int first, /* index of first pointer to log */
1459 int last) /* index of last pointer to log */
1461 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1462 XFS_BTREE_TRACE_ARGBII(cur, bp, first, last);
1465 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
1466 int level = xfs_btree_get_level(block);
1468 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1469 xfs_trans_log_buf(cur->bc_tp, bp,
1470 xfs_btree_ptr_offset(cur, first, level),
1471 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1473 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1474 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1477 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1481 * Log fields from a btree block header.
1484 xfs_btree_log_block(
1485 struct xfs_btree_cur *cur, /* btree cursor */
1486 struct xfs_buf *bp, /* buffer containing btree block */
1487 int fields) /* mask of fields: XFS_BB_... */
1489 int first; /* first byte offset logged */
1490 int last; /* last byte offset logged */
1491 static const short soffsets[] = { /* table of offsets (short) */
1492 offsetof(struct xfs_btree_block, bb_magic),
1493 offsetof(struct xfs_btree_block, bb_level),
1494 offsetof(struct xfs_btree_block, bb_numrecs),
1495 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1496 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1497 offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1498 offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1499 offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1500 offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1501 offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1502 XFS_BTREE_SBLOCK_CRC_LEN
1504 static const short loffsets[] = { /* table of offsets (long) */
1505 offsetof(struct xfs_btree_block, bb_magic),
1506 offsetof(struct xfs_btree_block, bb_level),
1507 offsetof(struct xfs_btree_block, bb_numrecs),
1508 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1509 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1510 offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1511 offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1512 offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1513 offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1514 offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1515 offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1516 XFS_BTREE_LBLOCK_CRC_LEN
1519 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1520 XFS_BTREE_TRACE_ARGBI(cur, bp, fields);
1525 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
1527 * We don't log the CRC when updating a btree
1528 * block but instead recreate it during log
1529 * recovery. As the log buffers have checksums
1530 * of their own this is safe and avoids logging a crc
1531 * update in a lot of places.
1533 if (fields == XFS_BB_ALL_BITS)
1534 fields = XFS_BB_ALL_BITS_CRC;
1535 nbits = XFS_BB_NUM_BITS_CRC;
1537 nbits = XFS_BB_NUM_BITS;
1539 xfs_btree_offsets(fields,
1540 (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1541 loffsets : soffsets,
1542 nbits, &first, &last);
1543 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1544 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1546 xfs_trans_log_inode(cur->bc_tp, cur->bc_private.b.ip,
1547 xfs_ilog_fbroot(cur->bc_private.b.whichfork));
1550 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1554 * Increment cursor by one record at the level.
1555 * For nonzero levels the leaf-ward information is untouched.
1558 xfs_btree_increment(
1559 struct xfs_btree_cur *cur,
1561 int *stat) /* success/failure */
1563 struct xfs_btree_block *block;
1564 union xfs_btree_ptr ptr;
1566 int error; /* error return value */
1569 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1570 XFS_BTREE_TRACE_ARGI(cur, level);
1572 ASSERT(level < cur->bc_nlevels);
1574 /* Read-ahead to the right at this level. */
1575 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1577 /* Get a pointer to the btree block. */
1578 block = xfs_btree_get_block(cur, level, &bp);
1581 error = xfs_btree_check_block(cur, block, level, bp);
1586 /* We're done if we remain in the block after the increment. */
1587 if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
1590 /* Fail if we just went off the right edge of the tree. */
1591 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1592 if (xfs_btree_ptr_is_null(cur, &ptr))
1595 XFS_BTREE_STATS_INC(cur, increment);
1598 * March up the tree incrementing pointers.
1599 * Stop when we don't go off the right edge of a block.
1601 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1602 block = xfs_btree_get_block(cur, lev, &bp);
1605 error = xfs_btree_check_block(cur, block, lev, bp);
1610 if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
1613 /* Read-ahead the right block for the next loop. */
1614 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1618 * If we went off the root then we are either seriously
1619 * confused or have the tree root in an inode.
1621 if (lev == cur->bc_nlevels) {
1622 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1625 error = -EFSCORRUPTED;
1628 ASSERT(lev < cur->bc_nlevels);
1631 * Now walk back down the tree, fixing up the cursor's buffer
1632 * pointers and key numbers.
1634 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1635 union xfs_btree_ptr *ptrp;
1637 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1639 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1643 xfs_btree_setbuf(cur, lev, bp);
1644 cur->bc_ptrs[lev] = 1;
1647 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1652 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1657 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1662 * Decrement cursor by one record at the level.
1663 * For nonzero levels the leaf-ward information is untouched.
1666 xfs_btree_decrement(
1667 struct xfs_btree_cur *cur,
1669 int *stat) /* success/failure */
1671 struct xfs_btree_block *block;
1673 int error; /* error return value */
1675 union xfs_btree_ptr ptr;
1677 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1678 XFS_BTREE_TRACE_ARGI(cur, level);
1680 ASSERT(level < cur->bc_nlevels);
1682 /* Read-ahead to the left at this level. */
1683 xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1685 /* We're done if we remain in the block after the decrement. */
1686 if (--cur->bc_ptrs[level] > 0)
1689 /* Get a pointer to the btree block. */
1690 block = xfs_btree_get_block(cur, level, &bp);
1693 error = xfs_btree_check_block(cur, block, level, bp);
1698 /* Fail if we just went off the left edge of the tree. */
1699 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1700 if (xfs_btree_ptr_is_null(cur, &ptr))
1703 XFS_BTREE_STATS_INC(cur, decrement);
1706 * March up the tree decrementing pointers.
1707 * Stop when we don't go off the left edge of a block.
1709 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1710 if (--cur->bc_ptrs[lev] > 0)
1712 /* Read-ahead the left block for the next loop. */
1713 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1717 * If we went off the root then we are seriously confused.
1718 * or the root of the tree is in an inode.
1720 if (lev == cur->bc_nlevels) {
1721 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1724 error = -EFSCORRUPTED;
1727 ASSERT(lev < cur->bc_nlevels);
1730 * Now walk back down the tree, fixing up the cursor's buffer
1731 * pointers and key numbers.
1733 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1734 union xfs_btree_ptr *ptrp;
1736 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1738 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1741 xfs_btree_setbuf(cur, lev, bp);
1742 cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
1745 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1750 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1755 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
1760 xfs_btree_lookup_get_block(
1761 struct xfs_btree_cur *cur, /* btree cursor */
1762 int level, /* level in the btree */
1763 union xfs_btree_ptr *pp, /* ptr to btree block */
1764 struct xfs_btree_block **blkp) /* return btree block */
1766 struct xfs_buf *bp; /* buffer pointer for btree block */
1769 /* special case the root block if in an inode */
1770 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1771 (level == cur->bc_nlevels - 1)) {
1772 *blkp = xfs_btree_get_iroot(cur);
1777 * If the old buffer at this level for the disk address we are
1778 * looking for re-use it.
1780 * Otherwise throw it away and get a new one.
1782 bp = cur->bc_bufs[level];
1783 if (bp && XFS_BUF_ADDR(bp) == xfs_btree_ptr_to_daddr(cur, pp)) {
1784 *blkp = XFS_BUF_TO_BLOCK(bp);
1788 error = xfs_btree_read_buf_block(cur, pp, 0, blkp, &bp);
1792 /* Check the inode owner since the verifiers don't. */
1793 if (xfs_sb_version_hascrc(&cur->bc_mp->m_sb) &&
1794 (cur->bc_flags & XFS_BTREE_LONG_PTRS) &&
1795 be64_to_cpu((*blkp)->bb_u.l.bb_owner) !=
1796 cur->bc_private.b.ip->i_ino)
1799 /* Did we get the level we were looking for? */
1800 if (be16_to_cpu((*blkp)->bb_level) != level)
1803 /* Check that internal nodes have at least one record. */
1804 if (level != 0 && be16_to_cpu((*blkp)->bb_numrecs) == 0)
1807 xfs_btree_setbuf(cur, level, bp);
1812 xfs_trans_brelse(cur->bc_tp, bp);
1813 return -EFSCORRUPTED;
1817 * Get current search key. For level 0 we don't actually have a key
1818 * structure so we make one up from the record. For all other levels
1819 * we just return the right key.
1821 STATIC union xfs_btree_key *
1822 xfs_lookup_get_search_key(
1823 struct xfs_btree_cur *cur,
1826 struct xfs_btree_block *block,
1827 union xfs_btree_key *kp)
1830 cur->bc_ops->init_key_from_rec(kp,
1831 xfs_btree_rec_addr(cur, keyno, block));
1835 return xfs_btree_key_addr(cur, keyno, block);
1839 * Lookup the record. The cursor is made to point to it, based on dir.
1840 * stat is set to 0 if can't find any such record, 1 for success.
1844 struct xfs_btree_cur *cur, /* btree cursor */
1845 xfs_lookup_t dir, /* <=, ==, or >= */
1846 int *stat) /* success/failure */
1848 struct xfs_btree_block *block; /* current btree block */
1849 __int64_t diff; /* difference for the current key */
1850 int error; /* error return value */
1851 int keyno; /* current key number */
1852 int level; /* level in the btree */
1853 union xfs_btree_ptr *pp; /* ptr to btree block */
1854 union xfs_btree_ptr ptr; /* ptr to btree block */
1856 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
1857 XFS_BTREE_TRACE_ARGI(cur, dir);
1859 XFS_BTREE_STATS_INC(cur, lookup);
1861 /* No such thing as a zero-level tree. */
1862 if (cur->bc_nlevels == 0)
1863 return -EFSCORRUPTED;
1868 /* initialise start pointer from cursor */
1869 cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1873 * Iterate over each level in the btree, starting at the root.
1874 * For each level above the leaves, find the key we need, based
1875 * on the lookup record, then follow the corresponding block
1876 * pointer down to the next level.
1878 for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1879 /* Get the block we need to do the lookup on. */
1880 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1886 * If we already had a key match at a higher level, we
1887 * know we need to use the first entry in this block.
1891 /* Otherwise search this block. Do a binary search. */
1893 int high; /* high entry number */
1894 int low; /* low entry number */
1896 /* Set low and high entry numbers, 1-based. */
1898 high = xfs_btree_get_numrecs(block);
1900 /* Block is empty, must be an empty leaf. */
1901 ASSERT(level == 0 && cur->bc_nlevels == 1);
1903 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1904 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1909 /* Binary search the block. */
1910 while (low <= high) {
1911 union xfs_btree_key key;
1912 union xfs_btree_key *kp;
1914 XFS_BTREE_STATS_INC(cur, compare);
1916 /* keyno is average of low and high. */
1917 keyno = (low + high) >> 1;
1919 /* Get current search key */
1920 kp = xfs_lookup_get_search_key(cur, level,
1921 keyno, block, &key);
1924 * Compute difference to get next direction:
1925 * - less than, move right
1926 * - greater than, move left
1927 * - equal, we're done
1929 diff = cur->bc_ops->key_diff(cur, kp);
1940 * If there are more levels, set up for the next level
1941 * by getting the block number and filling in the cursor.
1945 * If we moved left, need the previous key number,
1946 * unless there isn't one.
1948 if (diff > 0 && --keyno < 1)
1950 pp = xfs_btree_ptr_addr(cur, keyno, block);
1953 error = xfs_btree_check_ptr(cur, pp, 0, level);
1957 cur->bc_ptrs[level] = keyno;
1961 /* Done with the search. See if we need to adjust the results. */
1962 if (dir != XFS_LOOKUP_LE && diff < 0) {
1965 * If ge search and we went off the end of the block, but it's
1966 * not the last block, we're in the wrong block.
1968 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1969 if (dir == XFS_LOOKUP_GE &&
1970 keyno > xfs_btree_get_numrecs(block) &&
1971 !xfs_btree_ptr_is_null(cur, &ptr)) {
1974 cur->bc_ptrs[0] = keyno;
1975 error = xfs_btree_increment(cur, 0, &i);
1978 XFS_WANT_CORRUPTED_RETURN(cur->bc_mp, i == 1);
1979 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1983 } else if (dir == XFS_LOOKUP_LE && diff > 0)
1985 cur->bc_ptrs[0] = keyno;
1987 /* Return if we succeeded or not. */
1988 if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
1990 else if (dir != XFS_LOOKUP_EQ || diff == 0)
1994 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
1998 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2002 /* Find the high key storage area from a regular key. */
2003 STATIC union xfs_btree_key *
2004 xfs_btree_high_key_from_key(
2005 struct xfs_btree_cur *cur,
2006 union xfs_btree_key *key)
2008 ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2009 return (union xfs_btree_key *)((char *)key +
2010 (cur->bc_ops->key_len / 2));
2013 /* Determine the low (and high if overlapped) keys of a leaf block */
2015 xfs_btree_get_leaf_keys(
2016 struct xfs_btree_cur *cur,
2017 struct xfs_btree_block *block,
2018 union xfs_btree_key *key)
2020 union xfs_btree_key max_hkey;
2021 union xfs_btree_key hkey;
2022 union xfs_btree_rec *rec;
2023 union xfs_btree_key *high;
2026 rec = xfs_btree_rec_addr(cur, 1, block);
2027 cur->bc_ops->init_key_from_rec(key, rec);
2029 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2031 cur->bc_ops->init_high_key_from_rec(&max_hkey, rec);
2032 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2033 rec = xfs_btree_rec_addr(cur, n, block);
2034 cur->bc_ops->init_high_key_from_rec(&hkey, rec);
2035 if (cur->bc_ops->diff_two_keys(cur, &hkey, &max_hkey)
2040 high = xfs_btree_high_key_from_key(cur, key);
2041 memcpy(high, &max_hkey, cur->bc_ops->key_len / 2);
2045 /* Determine the low (and high if overlapped) keys of a node block */
2047 xfs_btree_get_node_keys(
2048 struct xfs_btree_cur *cur,
2049 struct xfs_btree_block *block,
2050 union xfs_btree_key *key)
2052 union xfs_btree_key *hkey;
2053 union xfs_btree_key *max_hkey;
2054 union xfs_btree_key *high;
2057 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2058 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2059 cur->bc_ops->key_len / 2);
2061 max_hkey = xfs_btree_high_key_addr(cur, 1, block);
2062 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2063 hkey = xfs_btree_high_key_addr(cur, n, block);
2064 if (cur->bc_ops->diff_two_keys(cur, hkey, max_hkey) > 0)
2068 high = xfs_btree_high_key_from_key(cur, key);
2069 memcpy(high, max_hkey, cur->bc_ops->key_len / 2);
2071 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2072 cur->bc_ops->key_len);
2076 /* Derive the keys for any btree block. */
2079 struct xfs_btree_cur *cur,
2080 struct xfs_btree_block *block,
2081 union xfs_btree_key *key)
2083 if (be16_to_cpu(block->bb_level) == 0)
2084 xfs_btree_get_leaf_keys(cur, block, key);
2086 xfs_btree_get_node_keys(cur, block, key);
2090 * Decide if we need to update the parent keys of a btree block. For
2091 * a standard btree this is only necessary if we're updating the first
2092 * record/key. For an overlapping btree, we must always update the
2093 * keys because the highest key can be in any of the records or keys
2097 xfs_btree_needs_key_update(
2098 struct xfs_btree_cur *cur,
2101 return (cur->bc_flags & XFS_BTREE_OVERLAPPING) || ptr == 1;
2105 * Update the low and high parent keys of the given level, progressing
2106 * towards the root. If force_all is false, stop if the keys for a given
2107 * level do not need updating.
2110 __xfs_btree_updkeys(
2111 struct xfs_btree_cur *cur,
2113 struct xfs_btree_block *block,
2114 struct xfs_buf *bp0,
2117 union xfs_btree_key key; /* keys from current level */
2118 union xfs_btree_key *lkey; /* keys from the next level up */
2119 union xfs_btree_key *hkey;
2120 union xfs_btree_key *nlkey; /* keys from the next level up */
2121 union xfs_btree_key *nhkey;
2125 ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2127 /* Exit if there aren't any parent levels to update. */
2128 if (level + 1 >= cur->bc_nlevels)
2131 trace_xfs_btree_updkeys(cur, level, bp0);
2134 hkey = xfs_btree_high_key_from_key(cur, lkey);
2135 xfs_btree_get_keys(cur, block, lkey);
2136 for (level++; level < cur->bc_nlevels; level++) {
2140 block = xfs_btree_get_block(cur, level, &bp);
2141 trace_xfs_btree_updkeys(cur, level, bp);
2143 error = xfs_btree_check_block(cur, block, level, bp);
2145 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2149 ptr = cur->bc_ptrs[level];
2150 nlkey = xfs_btree_key_addr(cur, ptr, block);
2151 nhkey = xfs_btree_high_key_addr(cur, ptr, block);
2153 !(cur->bc_ops->diff_two_keys(cur, nlkey, lkey) != 0 ||
2154 cur->bc_ops->diff_two_keys(cur, nhkey, hkey) != 0))
2156 xfs_btree_copy_keys(cur, nlkey, lkey, 1);
2157 xfs_btree_log_keys(cur, bp, ptr, ptr);
2158 if (level + 1 >= cur->bc_nlevels)
2160 xfs_btree_get_node_keys(cur, block, lkey);
2166 /* Update all the keys from some level in cursor back to the root. */
2168 xfs_btree_updkeys_force(
2169 struct xfs_btree_cur *cur,
2173 struct xfs_btree_block *block;
2175 block = xfs_btree_get_block(cur, level, &bp);
2176 return __xfs_btree_updkeys(cur, level, block, bp, true);
2180 * Update the parent keys of the given level, progressing towards the root.
2183 xfs_btree_update_keys(
2184 struct xfs_btree_cur *cur,
2187 struct xfs_btree_block *block;
2189 union xfs_btree_key *kp;
2190 union xfs_btree_key key;
2195 block = xfs_btree_get_block(cur, level, &bp);
2196 if (cur->bc_flags & XFS_BTREE_OVERLAPPING)
2197 return __xfs_btree_updkeys(cur, level, block, bp, false);
2199 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2200 XFS_BTREE_TRACE_ARGIK(cur, level, keyp);
2203 * Go up the tree from this level toward the root.
2204 * At each level, update the key value to the value input.
2205 * Stop when we reach a level where the cursor isn't pointing
2206 * at the first entry in the block.
2208 xfs_btree_get_keys(cur, block, &key);
2209 for (level++, ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
2213 block = xfs_btree_get_block(cur, level, &bp);
2215 error = xfs_btree_check_block(cur, block, level, bp);
2217 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2221 ptr = cur->bc_ptrs[level];
2222 kp = xfs_btree_key_addr(cur, ptr, block);
2223 xfs_btree_copy_keys(cur, kp, &key, 1);
2224 xfs_btree_log_keys(cur, bp, ptr, ptr);
2227 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2232 * Update the record referred to by cur to the value in the
2233 * given record. This either works (return 0) or gets an
2234 * EFSCORRUPTED error.
2238 struct xfs_btree_cur *cur,
2239 union xfs_btree_rec *rec)
2241 struct xfs_btree_block *block;
2245 union xfs_btree_rec *rp;
2247 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2248 XFS_BTREE_TRACE_ARGR(cur, rec);
2250 /* Pick up the current block. */
2251 block = xfs_btree_get_block(cur, 0, &bp);
2254 error = xfs_btree_check_block(cur, block, 0, bp);
2258 /* Get the address of the rec to be updated. */
2259 ptr = cur->bc_ptrs[0];
2260 rp = xfs_btree_rec_addr(cur, ptr, block);
2262 /* Fill in the new contents and log them. */
2263 xfs_btree_copy_recs(cur, rp, rec, 1);
2264 xfs_btree_log_recs(cur, bp, ptr, ptr);
2267 * If we are tracking the last record in the tree and
2268 * we are at the far right edge of the tree, update it.
2270 if (xfs_btree_is_lastrec(cur, block, 0)) {
2271 cur->bc_ops->update_lastrec(cur, block, rec,
2272 ptr, LASTREC_UPDATE);
2275 /* Pass new key value up to our parent. */
2276 if (xfs_btree_needs_key_update(cur, ptr)) {
2277 error = xfs_btree_update_keys(cur, 0);
2282 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2286 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2291 * Move 1 record left from cur/level if possible.
2292 * Update cur to reflect the new path.
2294 STATIC int /* error */
2296 struct xfs_btree_cur *cur,
2298 int *stat) /* success/failure */
2300 struct xfs_buf *lbp; /* left buffer pointer */
2301 struct xfs_btree_block *left; /* left btree block */
2302 int lrecs; /* left record count */
2303 struct xfs_buf *rbp; /* right buffer pointer */
2304 struct xfs_btree_block *right; /* right btree block */
2305 struct xfs_btree_cur *tcur; /* temporary btree cursor */
2306 int rrecs; /* right record count */
2307 union xfs_btree_ptr lptr; /* left btree pointer */
2308 union xfs_btree_key *rkp = NULL; /* right btree key */
2309 union xfs_btree_ptr *rpp = NULL; /* right address pointer */
2310 union xfs_btree_rec *rrp = NULL; /* right record pointer */
2311 int error; /* error return value */
2314 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2315 XFS_BTREE_TRACE_ARGI(cur, level);
2317 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2318 level == cur->bc_nlevels - 1)
2321 /* Set up variables for this block as "right". */
2322 right = xfs_btree_get_block(cur, level, &rbp);
2325 error = xfs_btree_check_block(cur, right, level, rbp);
2330 /* If we've got no left sibling then we can't shift an entry left. */
2331 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2332 if (xfs_btree_ptr_is_null(cur, &lptr))
2336 * If the cursor entry is the one that would be moved, don't
2337 * do it... it's too complicated.
2339 if (cur->bc_ptrs[level] <= 1)
2342 /* Set up the left neighbor as "left". */
2343 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2347 /* If it's full, it can't take another entry. */
2348 lrecs = xfs_btree_get_numrecs(left);
2349 if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2352 rrecs = xfs_btree_get_numrecs(right);
2355 * We add one entry to the left side and remove one for the right side.
2356 * Account for it here, the changes will be updated on disk and logged
2362 XFS_BTREE_STATS_INC(cur, lshift);
2363 XFS_BTREE_STATS_ADD(cur, moves, 1);
2366 * If non-leaf, copy a key and a ptr to the left block.
2367 * Log the changes to the left block.
2370 /* It's a non-leaf. Move keys and pointers. */
2371 union xfs_btree_key *lkp; /* left btree key */
2372 union xfs_btree_ptr *lpp; /* left address pointer */
2374 lkp = xfs_btree_key_addr(cur, lrecs, left);
2375 rkp = xfs_btree_key_addr(cur, 1, right);
2377 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2378 rpp = xfs_btree_ptr_addr(cur, 1, right);
2380 error = xfs_btree_check_ptr(cur, rpp, 0, level);
2384 xfs_btree_copy_keys(cur, lkp, rkp, 1);
2385 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2387 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2388 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2390 ASSERT(cur->bc_ops->keys_inorder(cur,
2391 xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2393 /* It's a leaf. Move records. */
2394 union xfs_btree_rec *lrp; /* left record pointer */
2396 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2397 rrp = xfs_btree_rec_addr(cur, 1, right);
2399 xfs_btree_copy_recs(cur, lrp, rrp, 1);
2400 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2402 ASSERT(cur->bc_ops->recs_inorder(cur,
2403 xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2406 xfs_btree_set_numrecs(left, lrecs);
2407 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2409 xfs_btree_set_numrecs(right, rrecs);
2410 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2413 * Slide the contents of right down one entry.
2415 XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2417 /* It's a nonleaf. operate on keys and ptrs */
2419 int i; /* loop index */
2421 for (i = 0; i < rrecs; i++) {
2422 error = xfs_btree_check_ptr(cur, rpp, i + 1, level);
2427 xfs_btree_shift_keys(cur,
2428 xfs_btree_key_addr(cur, 2, right),
2430 xfs_btree_shift_ptrs(cur,
2431 xfs_btree_ptr_addr(cur, 2, right),
2434 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2435 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2437 /* It's a leaf. operate on records */
2438 xfs_btree_shift_recs(cur,
2439 xfs_btree_rec_addr(cur, 2, right),
2441 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2445 * Using a temporary cursor, update the parent key values of the
2446 * block on the left.
2448 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2449 error = xfs_btree_dup_cursor(cur, &tcur);
2452 i = xfs_btree_firstrec(tcur, level);
2453 XFS_WANT_CORRUPTED_GOTO(tcur->bc_mp, i == 1, error0);
2455 error = xfs_btree_decrement(tcur, level, &i);
2459 /* Update the parent high keys of the left block, if needed. */
2460 error = xfs_btree_update_keys(tcur, level);
2464 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2467 /* Update the parent keys of the right block. */
2468 error = xfs_btree_update_keys(cur, level);
2472 /* Slide the cursor value left one. */
2473 cur->bc_ptrs[level]--;
2475 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2480 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2485 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2489 XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2490 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2495 * Move 1 record right from cur/level if possible.
2496 * Update cur to reflect the new path.
2498 STATIC int /* error */
2500 struct xfs_btree_cur *cur,
2502 int *stat) /* success/failure */
2504 struct xfs_buf *lbp; /* left buffer pointer */
2505 struct xfs_btree_block *left; /* left btree block */
2506 struct xfs_buf *rbp; /* right buffer pointer */
2507 struct xfs_btree_block *right; /* right btree block */
2508 struct xfs_btree_cur *tcur; /* temporary btree cursor */
2509 union xfs_btree_ptr rptr; /* right block pointer */
2510 union xfs_btree_key *rkp; /* right btree key */
2511 int rrecs; /* right record count */
2512 int lrecs; /* left record count */
2513 int error; /* error return value */
2514 int i; /* loop counter */
2516 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2517 XFS_BTREE_TRACE_ARGI(cur, level);
2519 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2520 (level == cur->bc_nlevels - 1))
2523 /* Set up variables for this block as "left". */
2524 left = xfs_btree_get_block(cur, level, &lbp);
2527 error = xfs_btree_check_block(cur, left, level, lbp);
2532 /* If we've got no right sibling then we can't shift an entry right. */
2533 xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2534 if (xfs_btree_ptr_is_null(cur, &rptr))
2538 * If the cursor entry is the one that would be moved, don't
2539 * do it... it's too complicated.
2541 lrecs = xfs_btree_get_numrecs(left);
2542 if (cur->bc_ptrs[level] >= lrecs)
2545 /* Set up the right neighbor as "right". */
2546 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2550 /* If it's full, it can't take another entry. */
2551 rrecs = xfs_btree_get_numrecs(right);
2552 if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2555 XFS_BTREE_STATS_INC(cur, rshift);
2556 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2559 * Make a hole at the start of the right neighbor block, then
2560 * copy the last left block entry to the hole.
2563 /* It's a nonleaf. make a hole in the keys and ptrs */
2564 union xfs_btree_key *lkp;
2565 union xfs_btree_ptr *lpp;
2566 union xfs_btree_ptr *rpp;
2568 lkp = xfs_btree_key_addr(cur, lrecs, left);
2569 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2570 rkp = xfs_btree_key_addr(cur, 1, right);
2571 rpp = xfs_btree_ptr_addr(cur, 1, right);
2574 for (i = rrecs - 1; i >= 0; i--) {
2575 error = xfs_btree_check_ptr(cur, rpp, i, level);
2581 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2582 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2585 error = xfs_btree_check_ptr(cur, lpp, 0, level);
2590 /* Now put the new data in, and log it. */
2591 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2592 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2594 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2595 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2597 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2598 xfs_btree_key_addr(cur, 2, right)));
2600 /* It's a leaf. make a hole in the records */
2601 union xfs_btree_rec *lrp;
2602 union xfs_btree_rec *rrp;
2604 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2605 rrp = xfs_btree_rec_addr(cur, 1, right);
2607 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2609 /* Now put the new data in, and log it. */
2610 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2611 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2615 * Decrement and log left's numrecs, bump and log right's numrecs.
2617 xfs_btree_set_numrecs(left, --lrecs);
2618 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2620 xfs_btree_set_numrecs(right, ++rrecs);
2621 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2624 * Using a temporary cursor, update the parent key values of the
2625 * block on the right.
2627 error = xfs_btree_dup_cursor(cur, &tcur);
2630 i = xfs_btree_lastrec(tcur, level);
2631 XFS_WANT_CORRUPTED_GOTO(tcur->bc_mp, i == 1, error0);
2633 error = xfs_btree_increment(tcur, level, &i);
2637 /* Update the parent high keys of the left block, if needed. */
2638 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2639 error = xfs_btree_update_keys(cur, level);
2644 /* Update the parent keys of the right block. */
2645 error = xfs_btree_update_keys(tcur, level);
2649 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2651 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2656 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2661 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2665 XFS_BTREE_TRACE_CURSOR(tcur, XBT_ERROR);
2666 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2671 * Split cur/level block in half.
2672 * Return new block number and the key to its first
2673 * record (to be inserted into parent).
2675 STATIC int /* error */
2677 struct xfs_btree_cur *cur,
2679 union xfs_btree_ptr *ptrp,
2680 union xfs_btree_key *key,
2681 struct xfs_btree_cur **curp,
2682 int *stat) /* success/failure */
2684 union xfs_btree_ptr lptr; /* left sibling block ptr */
2685 struct xfs_buf *lbp; /* left buffer pointer */
2686 struct xfs_btree_block *left; /* left btree block */
2687 union xfs_btree_ptr rptr; /* right sibling block ptr */
2688 struct xfs_buf *rbp; /* right buffer pointer */
2689 struct xfs_btree_block *right; /* right btree block */
2690 union xfs_btree_ptr rrptr; /* right-right sibling ptr */
2691 struct xfs_buf *rrbp; /* right-right buffer pointer */
2692 struct xfs_btree_block *rrblock; /* right-right btree block */
2696 int error; /* error return value */
2701 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2702 XFS_BTREE_TRACE_ARGIPK(cur, level, *ptrp, key);
2704 XFS_BTREE_STATS_INC(cur, split);
2706 /* Set up left block (current one). */
2707 left = xfs_btree_get_block(cur, level, &lbp);
2710 error = xfs_btree_check_block(cur, left, level, lbp);
2715 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2717 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2718 error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, stat);
2723 XFS_BTREE_STATS_INC(cur, alloc);
2725 /* Set up the new block as "right". */
2726 error = xfs_btree_get_buf_block(cur, &rptr, 0, &right, &rbp);
2730 /* Fill in the btree header for the new right block. */
2731 xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2734 * Split the entries between the old and the new block evenly.
2735 * Make sure that if there's an odd number of entries now, that
2736 * each new block will have the same number of entries.
2738 lrecs = xfs_btree_get_numrecs(left);
2740 if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
2742 src_index = (lrecs - rrecs + 1);
2744 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2746 /* Adjust numrecs for the later get_*_keys() calls. */
2748 xfs_btree_set_numrecs(left, lrecs);
2749 xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2752 * Copy btree block entries from the left block over to the
2753 * new block, the right. Update the right block and log the
2757 /* It's a non-leaf. Move keys and pointers. */
2758 union xfs_btree_key *lkp; /* left btree key */
2759 union xfs_btree_ptr *lpp; /* left address pointer */
2760 union xfs_btree_key *rkp; /* right btree key */
2761 union xfs_btree_ptr *rpp; /* right address pointer */
2763 lkp = xfs_btree_key_addr(cur, src_index, left);
2764 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2765 rkp = xfs_btree_key_addr(cur, 1, right);
2766 rpp = xfs_btree_ptr_addr(cur, 1, right);
2769 for (i = src_index; i < rrecs; i++) {
2770 error = xfs_btree_check_ptr(cur, lpp, i, level);
2776 /* Copy the keys & pointers to the new block. */
2777 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2778 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2780 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2781 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2783 /* Stash the keys of the new block for later insertion. */
2784 xfs_btree_get_node_keys(cur, right, key);
2786 /* It's a leaf. Move records. */
2787 union xfs_btree_rec *lrp; /* left record pointer */
2788 union xfs_btree_rec *rrp; /* right record pointer */
2790 lrp = xfs_btree_rec_addr(cur, src_index, left);
2791 rrp = xfs_btree_rec_addr(cur, 1, right);
2793 /* Copy records to the new block. */
2794 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2795 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2797 /* Stash the keys of the new block for later insertion. */
2798 xfs_btree_get_leaf_keys(cur, right, key);
2802 * Find the left block number by looking in the buffer.
2803 * Adjust sibling pointers.
2805 xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2806 xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2807 xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2808 xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2810 xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2811 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2814 * If there's a block to the new block's right, make that block
2815 * point back to right instead of to left.
2817 if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2818 error = xfs_btree_read_buf_block(cur, &rrptr,
2819 0, &rrblock, &rrbp);
2822 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2823 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2826 /* Update the parent high keys of the left block, if needed. */
2827 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2828 error = xfs_btree_update_keys(cur, level);
2834 * If the cursor is really in the right block, move it there.
2835 * If it's just pointing past the last entry in left, then we'll
2836 * insert there, so don't change anything in that case.
2838 if (cur->bc_ptrs[level] > lrecs + 1) {
2839 xfs_btree_setbuf(cur, level, rbp);
2840 cur->bc_ptrs[level] -= lrecs;
2843 * If there are more levels, we'll need another cursor which refers
2844 * the right block, no matter where this cursor was.
2846 if (level + 1 < cur->bc_nlevels) {
2847 error = xfs_btree_dup_cursor(cur, curp);
2850 (*curp)->bc_ptrs[level + 1]++;
2853 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2857 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2862 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
2866 struct xfs_btree_split_args {
2867 struct xfs_btree_cur *cur;
2869 union xfs_btree_ptr *ptrp;
2870 union xfs_btree_key *key;
2871 struct xfs_btree_cur **curp;
2872 int *stat; /* success/failure */
2874 bool kswapd; /* allocation in kswapd context */
2875 struct completion *done;
2876 struct work_struct work;
2880 * Stack switching interfaces for allocation
2883 xfs_btree_split_worker(
2884 struct work_struct *work)
2886 struct xfs_btree_split_args *args = container_of(work,
2887 struct xfs_btree_split_args, work);
2888 unsigned long pflags;
2889 unsigned long new_pflags = PF_FSTRANS;
2892 * we are in a transaction context here, but may also be doing work
2893 * in kswapd context, and hence we may need to inherit that state
2894 * temporarily to ensure that we don't block waiting for memory reclaim
2898 new_pflags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
2900 current_set_flags_nested(&pflags, new_pflags);
2902 args->result = __xfs_btree_split(args->cur, args->level, args->ptrp,
2903 args->key, args->curp, args->stat);
2904 complete(args->done);
2906 current_restore_flags_nested(&pflags, new_pflags);
2910 * BMBT split requests often come in with little stack to work on. Push
2911 * them off to a worker thread so there is lots of stack to use. For the other
2912 * btree types, just call directly to avoid the context switch overhead here.
2914 STATIC int /* error */
2916 struct xfs_btree_cur *cur,
2918 union xfs_btree_ptr *ptrp,
2919 union xfs_btree_key *key,
2920 struct xfs_btree_cur **curp,
2921 int *stat) /* success/failure */
2923 struct xfs_btree_split_args args;
2924 DECLARE_COMPLETION_ONSTACK(done);
2926 if (cur->bc_btnum != XFS_BTNUM_BMAP)
2927 return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
2936 args.kswapd = current_is_kswapd();
2937 INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker);
2938 queue_work(xfs_alloc_wq, &args.work);
2939 wait_for_completion(&done);
2940 destroy_work_on_stack(&args.work);
2946 * Copy the old inode root contents into a real block and make the
2947 * broot point to it.
2950 xfs_btree_new_iroot(
2951 struct xfs_btree_cur *cur, /* btree cursor */
2952 int *logflags, /* logging flags for inode */
2953 int *stat) /* return status - 0 fail */
2955 struct xfs_buf *cbp; /* buffer for cblock */
2956 struct xfs_btree_block *block; /* btree block */
2957 struct xfs_btree_block *cblock; /* child btree block */
2958 union xfs_btree_key *ckp; /* child key pointer */
2959 union xfs_btree_ptr *cpp; /* child ptr pointer */
2960 union xfs_btree_key *kp; /* pointer to btree key */
2961 union xfs_btree_ptr *pp; /* pointer to block addr */
2962 union xfs_btree_ptr nptr; /* new block addr */
2963 int level; /* btree level */
2964 int error; /* error return code */
2966 int i; /* loop counter */
2969 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
2970 XFS_BTREE_STATS_INC(cur, newroot);
2972 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2974 level = cur->bc_nlevels - 1;
2976 block = xfs_btree_get_iroot(cur);
2977 pp = xfs_btree_ptr_addr(cur, 1, block);
2979 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2980 error = cur->bc_ops->alloc_block(cur, pp, &nptr, stat);
2984 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
2987 XFS_BTREE_STATS_INC(cur, alloc);
2989 /* Copy the root into a real block. */
2990 error = xfs_btree_get_buf_block(cur, &nptr, 0, &cblock, &cbp);
2995 * we can't just memcpy() the root in for CRC enabled btree blocks.
2996 * In that case have to also ensure the blkno remains correct
2998 memcpy(cblock, block, xfs_btree_block_len(cur));
2999 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
3000 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
3001 cblock->bb_u.l.bb_blkno = cpu_to_be64(cbp->b_bn);
3003 cblock->bb_u.s.bb_blkno = cpu_to_be64(cbp->b_bn);
3006 be16_add_cpu(&block->bb_level, 1);
3007 xfs_btree_set_numrecs(block, 1);
3009 cur->bc_ptrs[level + 1] = 1;
3011 kp = xfs_btree_key_addr(cur, 1, block);
3012 ckp = xfs_btree_key_addr(cur, 1, cblock);
3013 xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
3015 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3017 for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
3018 error = xfs_btree_check_ptr(cur, pp, i, level);
3023 xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
3026 error = xfs_btree_check_ptr(cur, &nptr, 0, level);
3030 xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
3032 xfs_iroot_realloc(cur->bc_private.b.ip,
3033 1 - xfs_btree_get_numrecs(cblock),
3034 cur->bc_private.b.whichfork);
3036 xfs_btree_setbuf(cur, level, cbp);
3039 * Do all this logging at the end so that
3040 * the root is at the right level.
3042 xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
3043 xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3044 xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
3047 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork);
3049 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3052 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3057 * Allocate a new root block, fill it in.
3059 STATIC int /* error */
3061 struct xfs_btree_cur *cur, /* btree cursor */
3062 int *stat) /* success/failure */
3064 struct xfs_btree_block *block; /* one half of the old root block */
3065 struct xfs_buf *bp; /* buffer containing block */
3066 int error; /* error return value */
3067 struct xfs_buf *lbp; /* left buffer pointer */
3068 struct xfs_btree_block *left; /* left btree block */
3069 struct xfs_buf *nbp; /* new (root) buffer */
3070 struct xfs_btree_block *new; /* new (root) btree block */
3071 int nptr; /* new value for key index, 1 or 2 */
3072 struct xfs_buf *rbp; /* right buffer pointer */
3073 struct xfs_btree_block *right; /* right btree block */
3074 union xfs_btree_ptr rptr;
3075 union xfs_btree_ptr lptr;
3077 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3078 XFS_BTREE_STATS_INC(cur, newroot);
3080 /* initialise our start point from the cursor */
3081 cur->bc_ops->init_ptr_from_cur(cur, &rptr);
3083 /* Allocate the new block. If we can't do it, we're toast. Give up. */
3084 error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, stat);
3089 XFS_BTREE_STATS_INC(cur, alloc);
3091 /* Set up the new block. */
3092 error = xfs_btree_get_buf_block(cur, &lptr, 0, &new, &nbp);
3096 /* Set the root in the holding structure increasing the level by 1. */
3097 cur->bc_ops->set_root(cur, &lptr, 1);
3100 * At the previous root level there are now two blocks: the old root,
3101 * and the new block generated when it was split. We don't know which
3102 * one the cursor is pointing at, so we set up variables "left" and
3103 * "right" for each case.
3105 block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
3108 error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
3113 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3114 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3115 /* Our block is left, pick up the right block. */
3117 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
3119 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
3125 /* Our block is right, pick up the left block. */
3127 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
3129 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
3130 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3137 /* Fill in the new block's btree header and log it. */
3138 xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
3139 xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
3140 ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
3141 !xfs_btree_ptr_is_null(cur, &rptr));
3143 /* Fill in the key data in the new root. */
3144 if (xfs_btree_get_level(left) > 0) {
3146 * Get the keys for the left block's keys and put them directly
3147 * in the parent block. Do the same for the right block.
3149 xfs_btree_get_node_keys(cur, left,
3150 xfs_btree_key_addr(cur, 1, new));
3151 xfs_btree_get_node_keys(cur, right,
3152 xfs_btree_key_addr(cur, 2, new));
3155 * Get the keys for the left block's records and put them
3156 * directly in the parent block. Do the same for the right
3159 xfs_btree_get_leaf_keys(cur, left,
3160 xfs_btree_key_addr(cur, 1, new));
3161 xfs_btree_get_leaf_keys(cur, right,
3162 xfs_btree_key_addr(cur, 2, new));
3164 xfs_btree_log_keys(cur, nbp, 1, 2);
3166 /* Fill in the pointer data in the new root. */
3167 xfs_btree_copy_ptrs(cur,
3168 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
3169 xfs_btree_copy_ptrs(cur,
3170 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
3171 xfs_btree_log_ptrs(cur, nbp, 1, 2);
3173 /* Fix up the cursor. */
3174 xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
3175 cur->bc_ptrs[cur->bc_nlevels] = nptr;
3177 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3181 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3184 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3190 xfs_btree_make_block_unfull(
3191 struct xfs_btree_cur *cur, /* btree cursor */
3192 int level, /* btree level */
3193 int numrecs,/* # of recs in block */
3194 int *oindex,/* old tree index */
3195 int *index, /* new tree index */
3196 union xfs_btree_ptr *nptr, /* new btree ptr */
3197 struct xfs_btree_cur **ncur, /* new btree cursor */
3198 union xfs_btree_key *key, /* key of new block */
3203 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3204 level == cur->bc_nlevels - 1) {
3205 struct xfs_inode *ip = cur->bc_private.b.ip;
3207 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
3208 /* A root block that can be made bigger. */
3209 xfs_iroot_realloc(ip, 1, cur->bc_private.b.whichfork);
3212 /* A root block that needs replacing */
3215 error = xfs_btree_new_iroot(cur, &logflags, stat);
3216 if (error || *stat == 0)
3219 xfs_trans_log_inode(cur->bc_tp, ip, logflags);
3225 /* First, try shifting an entry to the right neighbor. */
3226 error = xfs_btree_rshift(cur, level, stat);
3230 /* Next, try shifting an entry to the left neighbor. */
3231 error = xfs_btree_lshift(cur, level, stat);
3236 *oindex = *index = cur->bc_ptrs[level];
3241 * Next, try splitting the current block in half.
3243 * If this works we have to re-set our variables because we
3244 * could be in a different block now.
3246 error = xfs_btree_split(cur, level, nptr, key, ncur, stat);
3247 if (error || *stat == 0)
3251 *index = cur->bc_ptrs[level];
3256 * Insert one record/level. Return information to the caller
3257 * allowing the next level up to proceed if necessary.
3261 struct xfs_btree_cur *cur, /* btree cursor */
3262 int level, /* level to insert record at */
3263 union xfs_btree_ptr *ptrp, /* i/o: block number inserted */
3264 union xfs_btree_rec *rec, /* record to insert */
3265 union xfs_btree_key *key, /* i/o: block key for ptrp */
3266 struct xfs_btree_cur **curp, /* output: new cursor replacing cur */
3267 int *stat) /* success/failure */
3269 struct xfs_btree_block *block; /* btree block */
3270 struct xfs_buf *bp; /* buffer for block */
3271 union xfs_btree_ptr nptr; /* new block ptr */
3272 struct xfs_btree_cur *ncur; /* new btree cursor */
3273 union xfs_btree_key nkey; /* new block key */
3274 union xfs_btree_key *lkey;
3275 int optr; /* old key/record index */
3276 int ptr; /* key/record index */
3277 int numrecs;/* number of records */
3278 int error; /* error return value */
3284 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3285 XFS_BTREE_TRACE_ARGIPR(cur, level, *ptrp, &rec);
3291 * If we have an external root pointer, and we've made it to the
3292 * root level, allocate a new root block and we're done.
3294 if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3295 (level >= cur->bc_nlevels)) {
3296 error = xfs_btree_new_root(cur, stat);
3297 xfs_btree_set_ptr_null(cur, ptrp);
3299 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3303 /* If we're off the left edge, return failure. */
3304 ptr = cur->bc_ptrs[level];
3306 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3313 XFS_BTREE_STATS_INC(cur, insrec);
3315 /* Get pointers to the btree buffer and block. */
3316 block = xfs_btree_get_block(cur, level, &bp);
3317 old_bn = bp ? bp->b_bn : XFS_BUF_DADDR_NULL;
3318 numrecs = xfs_btree_get_numrecs(block);
3321 error = xfs_btree_check_block(cur, block, level, bp);
3325 /* Check that the new entry is being inserted in the right place. */
3326 if (ptr <= numrecs) {
3328 ASSERT(cur->bc_ops->recs_inorder(cur, rec,
3329 xfs_btree_rec_addr(cur, ptr, block)));
3331 ASSERT(cur->bc_ops->keys_inorder(cur, key,
3332 xfs_btree_key_addr(cur, ptr, block)));
3338 * If the block is full, we can't insert the new entry until we
3339 * make the block un-full.
3341 xfs_btree_set_ptr_null(cur, &nptr);
3342 if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
3343 error = xfs_btree_make_block_unfull(cur, level, numrecs,
3344 &optr, &ptr, &nptr, &ncur, lkey, stat);
3345 if (error || *stat == 0)
3350 * The current block may have changed if the block was
3351 * previously full and we have just made space in it.
3353 block = xfs_btree_get_block(cur, level, &bp);
3354 numrecs = xfs_btree_get_numrecs(block);
3357 error = xfs_btree_check_block(cur, block, level, bp);
3363 * At this point we know there's room for our new entry in the block
3364 * we're pointing at.
3366 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
3369 /* It's a nonleaf. make a hole in the keys and ptrs */
3370 union xfs_btree_key *kp;
3371 union xfs_btree_ptr *pp;
3373 kp = xfs_btree_key_addr(cur, ptr, block);
3374 pp = xfs_btree_ptr_addr(cur, ptr, block);
3377 for (i = numrecs - ptr; i >= 0; i--) {
3378 error = xfs_btree_check_ptr(cur, pp, i, level);
3384 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
3385 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
3388 error = xfs_btree_check_ptr(cur, ptrp, 0, level);
3393 /* Now put the new data in, bump numrecs and log it. */
3394 xfs_btree_copy_keys(cur, kp, key, 1);
3395 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
3397 xfs_btree_set_numrecs(block, numrecs);
3398 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
3399 xfs_btree_log_keys(cur, bp, ptr, numrecs);
3401 if (ptr < numrecs) {
3402 ASSERT(cur->bc_ops->keys_inorder(cur, kp,
3403 xfs_btree_key_addr(cur, ptr + 1, block)));
3407 /* It's a leaf. make a hole in the records */
3408 union xfs_btree_rec *rp;
3410 rp = xfs_btree_rec_addr(cur, ptr, block);
3412 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
3414 /* Now put the new data in, bump numrecs and log it. */
3415 xfs_btree_copy_recs(cur, rp, rec, 1);
3416 xfs_btree_set_numrecs(block, ++numrecs);
3417 xfs_btree_log_recs(cur, bp, ptr, numrecs);
3419 if (ptr < numrecs) {
3420 ASSERT(cur->bc_ops->recs_inorder(cur, rp,
3421 xfs_btree_rec_addr(cur, ptr + 1, block)));
3426 /* Log the new number of records in the btree header. */
3427 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3430 * If we just inserted into a new tree block, we have to
3431 * recalculate nkey here because nkey is out of date.
3433 * Otherwise we're just updating an existing block (having shoved
3434 * some records into the new tree block), so use the regular key
3437 if (bp && bp->b_bn != old_bn) {
3438 xfs_btree_get_keys(cur, block, lkey);
3439 } else if (xfs_btree_needs_key_update(cur, optr)) {
3440 error = xfs_btree_update_keys(cur, level);
3446 * If we are tracking the last record in the tree and
3447 * we are at the far right edge of the tree, update it.
3449 if (xfs_btree_is_lastrec(cur, block, level)) {
3450 cur->bc_ops->update_lastrec(cur, block, rec,
3451 ptr, LASTREC_INSREC);
3455 * Return the new block number, if any.
3456 * If there is one, give back a record value and a cursor too.
3459 if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3460 xfs_btree_copy_keys(cur, key, lkey, 1);
3464 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3469 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3474 * Insert the record at the point referenced by cur.
3476 * A multi-level split of the tree on insert will invalidate the original
3477 * cursor. All callers of this function should assume that the cursor is
3478 * no longer valid and revalidate it.
3482 struct xfs_btree_cur *cur,
3485 int error; /* error return value */
3486 int i; /* result value, 0 for failure */
3487 int level; /* current level number in btree */
3488 union xfs_btree_ptr nptr; /* new block number (split result) */
3489 struct xfs_btree_cur *ncur; /* new cursor (split result) */
3490 struct xfs_btree_cur *pcur; /* previous level's cursor */
3491 union xfs_btree_key bkey; /* key of block to insert */
3492 union xfs_btree_key *key;
3493 union xfs_btree_rec rec; /* record to insert */
3500 xfs_btree_set_ptr_null(cur, &nptr);
3502 /* Make a key out of the record data to be inserted, and save it. */
3503 cur->bc_ops->init_rec_from_cur(cur, &rec);
3504 cur->bc_ops->init_key_from_rec(key, &rec);
3507 * Loop going up the tree, starting at the leaf level.
3508 * Stop when we don't get a split block, that must mean that
3509 * the insert is finished with this level.
3513 * Insert nrec/nptr into this level of the tree.
3514 * Note if we fail, nptr will be null.
3516 error = xfs_btree_insrec(pcur, level, &nptr, &rec, key,
3520 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3524 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3528 * See if the cursor we just used is trash.
3529 * Can't trash the caller's cursor, but otherwise we should
3530 * if ncur is a new cursor or we're about to be done.
3533 (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3534 /* Save the state from the cursor before we trash it */
3535 if (cur->bc_ops->update_cursor)
3536 cur->bc_ops->update_cursor(pcur, cur);
3537 cur->bc_nlevels = pcur->bc_nlevels;
3538 xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3540 /* If we got a new cursor, switch to it. */
3545 } while (!xfs_btree_ptr_is_null(cur, &nptr));
3547 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3551 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3556 * Try to merge a non-leaf block back into the inode root.
3558 * Note: the killroot names comes from the fact that we're effectively
3559 * killing the old root block. But because we can't just delete the
3560 * inode we have to copy the single block it was pointing to into the
3564 xfs_btree_kill_iroot(
3565 struct xfs_btree_cur *cur)
3567 int whichfork = cur->bc_private.b.whichfork;
3568 struct xfs_inode *ip = cur->bc_private.b.ip;
3569 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
3570 struct xfs_btree_block *block;
3571 struct xfs_btree_block *cblock;
3572 union xfs_btree_key *kp;
3573 union xfs_btree_key *ckp;
3574 union xfs_btree_ptr *pp;
3575 union xfs_btree_ptr *cpp;
3576 struct xfs_buf *cbp;
3582 union xfs_btree_ptr ptr;
3586 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3588 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3589 ASSERT(cur->bc_nlevels > 1);
3592 * Don't deal with the root block needs to be a leaf case.
3593 * We're just going to turn the thing back into extents anyway.
3595 level = cur->bc_nlevels - 1;
3600 * Give up if the root has multiple children.
3602 block = xfs_btree_get_iroot(cur);
3603 if (xfs_btree_get_numrecs(block) != 1)
3606 cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3607 numrecs = xfs_btree_get_numrecs(cblock);
3610 * Only do this if the next level will fit.
3611 * Then the data must be copied up to the inode,
3612 * instead of freeing the root you free the next level.
3614 if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3617 XFS_BTREE_STATS_INC(cur, killroot);
3620 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3621 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3622 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3623 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3626 index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3628 xfs_iroot_realloc(cur->bc_private.b.ip, index,
3629 cur->bc_private.b.whichfork);
3630 block = ifp->if_broot;
3633 be16_add_cpu(&block->bb_numrecs, index);
3634 ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3636 kp = xfs_btree_key_addr(cur, 1, block);
3637 ckp = xfs_btree_key_addr(cur, 1, cblock);
3638 xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3640 pp = xfs_btree_ptr_addr(cur, 1, block);
3641 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3643 for (i = 0; i < numrecs; i++) {
3644 error = xfs_btree_check_ptr(cur, cpp, i, level - 1);
3646 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3651 xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3653 error = xfs_btree_free_block(cur, cbp);
3655 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3659 cur->bc_bufs[level - 1] = NULL;
3660 be16_add_cpu(&block->bb_level, -1);
3661 xfs_trans_log_inode(cur->bc_tp, ip,
3662 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_private.b.whichfork));
3665 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3670 * Kill the current root node, and replace it with it's only child node.
3673 xfs_btree_kill_root(
3674 struct xfs_btree_cur *cur,
3677 union xfs_btree_ptr *newroot)
3681 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3682 XFS_BTREE_STATS_INC(cur, killroot);
3685 * Update the root pointer, decreasing the level by 1 and then
3686 * free the old root.
3688 cur->bc_ops->set_root(cur, newroot, -1);
3690 error = xfs_btree_free_block(cur, bp);
3692 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
3696 cur->bc_bufs[level] = NULL;
3697 cur->bc_ra[level] = 0;
3700 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3705 xfs_btree_dec_cursor(
3706 struct xfs_btree_cur *cur,
3714 error = xfs_btree_decrement(cur, level, &i);
3719 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3725 * Single level of the btree record deletion routine.
3726 * Delete record pointed to by cur/level.
3727 * Remove the record from its block then rebalance the tree.
3728 * Return 0 for error, 1 for done, 2 to go on to the next level.
3730 STATIC int /* error */
3732 struct xfs_btree_cur *cur, /* btree cursor */
3733 int level, /* level removing record from */
3734 int *stat) /* fail/done/go-on */
3736 struct xfs_btree_block *block; /* btree block */
3737 union xfs_btree_ptr cptr; /* current block ptr */
3738 struct xfs_buf *bp; /* buffer for block */
3739 int error; /* error return value */
3740 int i; /* loop counter */
3741 union xfs_btree_ptr lptr; /* left sibling block ptr */
3742 struct xfs_buf *lbp; /* left buffer pointer */
3743 struct xfs_btree_block *left; /* left btree block */
3744 int lrecs = 0; /* left record count */
3745 int ptr; /* key/record index */
3746 union xfs_btree_ptr rptr; /* right sibling block ptr */
3747 struct xfs_buf *rbp; /* right buffer pointer */
3748 struct xfs_btree_block *right; /* right btree block */
3749 struct xfs_btree_block *rrblock; /* right-right btree block */
3750 struct xfs_buf *rrbp; /* right-right buffer pointer */
3751 int rrecs = 0; /* right record count */
3752 struct xfs_btree_cur *tcur; /* temporary btree cursor */
3753 int numrecs; /* temporary numrec count */
3755 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
3756 XFS_BTREE_TRACE_ARGI(cur, level);
3760 /* Get the index of the entry being deleted, check for nothing there. */
3761 ptr = cur->bc_ptrs[level];
3763 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3768 /* Get the buffer & block containing the record or key/ptr. */
3769 block = xfs_btree_get_block(cur, level, &bp);
3770 numrecs = xfs_btree_get_numrecs(block);
3773 error = xfs_btree_check_block(cur, block, level, bp);
3778 /* Fail if we're off the end of the block. */
3779 if (ptr > numrecs) {
3780 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
3785 XFS_BTREE_STATS_INC(cur, delrec);
3786 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3788 /* Excise the entries being deleted. */
3790 /* It's a nonleaf. operate on keys and ptrs */
3791 union xfs_btree_key *lkp;
3792 union xfs_btree_ptr *lpp;
3794 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3795 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3798 for (i = 0; i < numrecs - ptr; i++) {
3799 error = xfs_btree_check_ptr(cur, lpp, i, level);
3805 if (ptr < numrecs) {
3806 xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3807 xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3808 xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3809 xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3812 /* It's a leaf. operate on records */
3813 if (ptr < numrecs) {
3814 xfs_btree_shift_recs(cur,
3815 xfs_btree_rec_addr(cur, ptr + 1, block),
3817 xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3822 * Decrement and log the number of entries in the block.
3824 xfs_btree_set_numrecs(block, --numrecs);
3825 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3828 * If we are tracking the last record in the tree and
3829 * we are at the far right edge of the tree, update it.
3831 if (xfs_btree_is_lastrec(cur, block, level)) {
3832 cur->bc_ops->update_lastrec(cur, block, NULL,
3833 ptr, LASTREC_DELREC);
3837 * We're at the root level. First, shrink the root block in-memory.
3838 * Try to get rid of the next level down. If we can't then there's
3839 * nothing left to do.
3841 if (level == cur->bc_nlevels - 1) {
3842 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3843 xfs_iroot_realloc(cur->bc_private.b.ip, -1,
3844 cur->bc_private.b.whichfork);
3846 error = xfs_btree_kill_iroot(cur);
3850 error = xfs_btree_dec_cursor(cur, level, stat);
3858 * If this is the root level, and there's only one entry left,
3859 * and it's NOT the leaf level, then we can get rid of this
3862 if (numrecs == 1 && level > 0) {
3863 union xfs_btree_ptr *pp;
3865 * pp is still set to the first pointer in the block.
3866 * Make it the new root of the btree.
3868 pp = xfs_btree_ptr_addr(cur, 1, block);
3869 error = xfs_btree_kill_root(cur, bp, level, pp);
3872 } else if (level > 0) {
3873 error = xfs_btree_dec_cursor(cur, level, stat);
3882 * If we deleted the leftmost entry in the block, update the
3883 * key values above us in the tree.
3885 if (xfs_btree_needs_key_update(cur, ptr)) {
3886 error = xfs_btree_update_keys(cur, level);
3892 * If the number of records remaining in the block is at least
3893 * the minimum, we're done.
3895 if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3896 error = xfs_btree_dec_cursor(cur, level, stat);
3903 * Otherwise, we have to move some records around to keep the
3904 * tree balanced. Look at the left and right sibling blocks to
3905 * see if we can re-balance by moving only one record.
3907 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3908 xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3910 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3912 * One child of root, need to get a chance to copy its contents
3913 * into the root and delete it. Can't go up to next level,
3914 * there's nothing to delete there.
3916 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3917 xfs_btree_ptr_is_null(cur, &lptr) &&
3918 level == cur->bc_nlevels - 2) {
3919 error = xfs_btree_kill_iroot(cur);
3921 error = xfs_btree_dec_cursor(cur, level, stat);
3928 ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3929 !xfs_btree_ptr_is_null(cur, &lptr));
3932 * Duplicate the cursor so our btree manipulations here won't
3933 * disrupt the next level up.
3935 error = xfs_btree_dup_cursor(cur, &tcur);
3940 * If there's a right sibling, see if it's ok to shift an entry
3943 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3945 * Move the temp cursor to the last entry in the next block.
3946 * Actually any entry but the first would suffice.
3948 i = xfs_btree_lastrec(tcur, level);
3949 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3951 error = xfs_btree_increment(tcur, level, &i);
3954 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3956 i = xfs_btree_lastrec(tcur, level);
3957 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
3959 /* Grab a pointer to the block. */
3960 right = xfs_btree_get_block(tcur, level, &rbp);
3962 error = xfs_btree_check_block(tcur, right, level, rbp);
3966 /* Grab the current block number, for future use. */
3967 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3970 * If right block is full enough so that removing one entry
3971 * won't make it too empty, and left-shifting an entry out
3972 * of right to us works, we're done.
3974 if (xfs_btree_get_numrecs(right) - 1 >=
3975 cur->bc_ops->get_minrecs(tcur, level)) {
3976 error = xfs_btree_lshift(tcur, level, &i);
3980 ASSERT(xfs_btree_get_numrecs(block) >=
3981 cur->bc_ops->get_minrecs(tcur, level));
3983 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3986 error = xfs_btree_dec_cursor(cur, level, stat);
3994 * Otherwise, grab the number of records in right for
3995 * future reference, and fix up the temp cursor to point
3996 * to our block again (last record).
3998 rrecs = xfs_btree_get_numrecs(right);
3999 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
4000 i = xfs_btree_firstrec(tcur, level);
4001 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
4003 error = xfs_btree_decrement(tcur, level, &i);
4006 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
4011 * If there's a left sibling, see if it's ok to shift an entry
4014 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
4016 * Move the temp cursor to the first entry in the
4019 i = xfs_btree_firstrec(tcur, level);
4020 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
4022 error = xfs_btree_decrement(tcur, level, &i);
4025 i = xfs_btree_firstrec(tcur, level);
4026 XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, i == 1, error0);
4028 /* Grab a pointer to the block. */
4029 left = xfs_btree_get_block(tcur, level, &lbp);
4031 error = xfs_btree_check_block(cur, left, level, lbp);
4035 /* Grab the current block number, for future use. */
4036 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
4039 * If left block is full enough so that removing one entry
4040 * won't make it too empty, and right-shifting an entry out
4041 * of left to us works, we're done.
4043 if (xfs_btree_get_numrecs(left) - 1 >=
4044 cur->bc_ops->get_minrecs(tcur, level)) {
4045 error = xfs_btree_rshift(tcur, level, &i);
4049 ASSERT(xfs_btree_get_numrecs(block) >=
4050 cur->bc_ops->get_minrecs(tcur, level));
4051 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4055 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
4062 * Otherwise, grab the number of records in right for
4065 lrecs = xfs_btree_get_numrecs(left);
4068 /* Delete the temp cursor, we're done with it. */
4069 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
4072 /* If here, we need to do a join to keep the tree balanced. */
4073 ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
4075 if (!xfs_btree_ptr_is_null(cur, &lptr) &&
4076 lrecs + xfs_btree_get_numrecs(block) <=
4077 cur->bc_ops->get_maxrecs(cur, level)) {
4079 * Set "right" to be the starting block,
4080 * "left" to be the left neighbor.
4085 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
4090 * If that won't work, see if we can join with the right neighbor block.
4092 } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
4093 rrecs + xfs_btree_get_numrecs(block) <=
4094 cur->bc_ops->get_maxrecs(cur, level)) {
4096 * Set "left" to be the starting block,
4097 * "right" to be the right neighbor.
4102 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
4107 * Otherwise, we can't fix the imbalance.
4108 * Just return. This is probably a logic error, but it's not fatal.
4111 error = xfs_btree_dec_cursor(cur, level, stat);
4117 rrecs = xfs_btree_get_numrecs(right);
4118 lrecs = xfs_btree_get_numrecs(left);
4121 * We're now going to join "left" and "right" by moving all the stuff
4122 * in "right" to "left" and deleting "right".
4124 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
4126 /* It's a non-leaf. Move keys and pointers. */
4127 union xfs_btree_key *lkp; /* left btree key */
4128 union xfs_btree_ptr *lpp; /* left address pointer */
4129 union xfs_btree_key *rkp; /* right btree key */
4130 union xfs_btree_ptr *rpp; /* right address pointer */
4132 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
4133 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
4134 rkp = xfs_btree_key_addr(cur, 1, right);
4135 rpp = xfs_btree_ptr_addr(cur, 1, right);
4137 for (i = 1; i < rrecs; i++) {
4138 error = xfs_btree_check_ptr(cur, rpp, i, level);
4143 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
4144 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
4146 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
4147 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
4149 /* It's a leaf. Move records. */
4150 union xfs_btree_rec *lrp; /* left record pointer */
4151 union xfs_btree_rec *rrp; /* right record pointer */
4153 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
4154 rrp = xfs_btree_rec_addr(cur, 1, right);
4156 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
4157 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
4160 XFS_BTREE_STATS_INC(cur, join);
4163 * Fix up the number of records and right block pointer in the
4164 * surviving block, and log it.
4166 xfs_btree_set_numrecs(left, lrecs + rrecs);
4167 xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB),
4168 xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4169 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
4171 /* If there is a right sibling, point it to the remaining block. */
4172 xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4173 if (!xfs_btree_ptr_is_null(cur, &cptr)) {
4174 error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp);
4177 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
4178 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
4181 /* Free the deleted block. */
4182 error = xfs_btree_free_block(cur, rbp);
4187 * If we joined with the left neighbor, set the buffer in the
4188 * cursor to the left block, and fix up the index.
4191 cur->bc_bufs[level] = lbp;
4192 cur->bc_ptrs[level] += lrecs;
4193 cur->bc_ra[level] = 0;
4196 * If we joined with the right neighbor and there's a level above
4197 * us, increment the cursor at that level.
4199 else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
4200 (level + 1 < cur->bc_nlevels)) {
4201 error = xfs_btree_increment(cur, level + 1, &i);
4207 * Readjust the ptr at this level if it's not a leaf, since it's
4208 * still pointing at the deletion point, which makes the cursor
4209 * inconsistent. If this makes the ptr 0, the caller fixes it up.
4210 * We can't use decrement because it would change the next level up.
4213 cur->bc_ptrs[level]--;
4216 * We combined blocks, so we have to update the parent keys if the
4217 * btree supports overlapped intervals. However, bc_ptrs[level + 1]
4218 * points to the old block so that the caller knows which record to
4219 * delete. Therefore, the caller must be savvy enough to call updkeys
4220 * for us if we return stat == 2. The other exit points from this
4221 * function don't require deletions further up the tree, so they can
4222 * call updkeys directly.
4225 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
4226 /* Return value means the next level up has something to do. */
4231 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
4233 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
4238 * Delete the record pointed to by cur.
4239 * The cursor refers to the place where the record was (could be inserted)
4240 * when the operation returns.
4244 struct xfs_btree_cur *cur,
4245 int *stat) /* success/failure */
4247 int error; /* error return value */
4250 bool joined = false;
4252 XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
4255 * Go up the tree, starting at leaf level.
4257 * If 2 is returned then a join was done; go to the next level.
4258 * Otherwise we are done.
4260 for (level = 0, i = 2; i == 2; level++) {
4261 error = xfs_btree_delrec(cur, level, &i);
4269 * If we combined blocks as part of deleting the record, delrec won't
4270 * have updated the parent high keys so we have to do that here.
4272 if (joined && (cur->bc_flags & XFS_BTREE_OVERLAPPING)) {
4273 error = xfs_btree_updkeys_force(cur, 0);
4279 for (level = 1; level < cur->bc_nlevels; level++) {
4280 if (cur->bc_ptrs[level] == 0) {
4281 error = xfs_btree_decrement(cur, level, &i);
4289 XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
4293 XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
4298 * Get the data from the pointed-to record.
4302 struct xfs_btree_cur *cur, /* btree cursor */
4303 union xfs_btree_rec **recp, /* output: btree record */
4304 int *stat) /* output: success/failure */
4306 struct xfs_btree_block *block; /* btree block */
4307 struct xfs_buf *bp; /* buffer pointer */
4308 int ptr; /* record number */
4310 int error; /* error return value */
4313 ptr = cur->bc_ptrs[0];
4314 block = xfs_btree_get_block(cur, 0, &bp);
4317 error = xfs_btree_check_block(cur, block, 0, bp);
4323 * Off the right end or left end, return failure.
4325 if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
4331 * Point to the record and extract its data.
4333 *recp = xfs_btree_rec_addr(cur, ptr, block);
4338 /* Visit a block in a btree. */
4340 xfs_btree_visit_block(
4341 struct xfs_btree_cur *cur,
4343 xfs_btree_visit_blocks_fn fn,
4346 struct xfs_btree_block *block;
4348 union xfs_btree_ptr rptr;
4351 /* do right sibling readahead */
4352 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
4353 block = xfs_btree_get_block(cur, level, &bp);
4355 /* process the block */
4356 error = fn(cur, level, data);
4360 /* now read rh sibling block for next iteration */
4361 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4362 if (xfs_btree_ptr_is_null(cur, &rptr))
4365 return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
4369 /* Visit every block in a btree. */
4371 xfs_btree_visit_blocks(
4372 struct xfs_btree_cur *cur,
4373 xfs_btree_visit_blocks_fn fn,
4376 union xfs_btree_ptr lptr;
4378 struct xfs_btree_block *block = NULL;
4381 cur->bc_ops->init_ptr_from_cur(cur, &lptr);
4383 /* for each level */
4384 for (level = cur->bc_nlevels - 1; level >= 0; level--) {
4385 /* grab the left hand block */
4386 error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
4390 /* readahead the left most block for the next level down */
4392 union xfs_btree_ptr *ptr;
4394 ptr = xfs_btree_ptr_addr(cur, 1, block);
4395 xfs_btree_readahead_ptr(cur, ptr, 1);
4397 /* save for the next iteration of the loop */
4401 /* for each buffer in the level */
4403 error = xfs_btree_visit_block(cur, level, fn, data);
4406 if (error != -ENOENT)
4414 * Change the owner of a btree.
4416 * The mechanism we use here is ordered buffer logging. Because we don't know
4417 * how many buffers were are going to need to modify, we don't really want to
4418 * have to make transaction reservations for the worst case of every buffer in a
4419 * full size btree as that may be more space that we can fit in the log....
4421 * We do the btree walk in the most optimal manner possible - we have sibling
4422 * pointers so we can just walk all the blocks on each level from left to right
4423 * in a single pass, and then move to the next level and do the same. We can
4424 * also do readahead on the sibling pointers to get IO moving more quickly,
4425 * though for slow disks this is unlikely to make much difference to performance
4426 * as the amount of CPU work we have to do before moving to the next block is
4429 * For each btree block that we load, modify the owner appropriately, set the
4430 * buffer as an ordered buffer and log it appropriately. We need to ensure that
4431 * we mark the region we change dirty so that if the buffer is relogged in
4432 * a subsequent transaction the changes we make here as an ordered buffer are
4433 * correctly relogged in that transaction. If we are in recovery context, then
4434 * just queue the modified buffer as delayed write buffer so the transaction
4435 * recovery completion writes the changes to disk.
4437 struct xfs_btree_block_change_owner_info {
4438 __uint64_t new_owner;
4439 struct list_head *buffer_list;
4443 xfs_btree_block_change_owner(
4444 struct xfs_btree_cur *cur,
4448 struct xfs_btree_block_change_owner_info *bbcoi = data;
4449 struct xfs_btree_block *block;
4452 /* modify the owner */
4453 block = xfs_btree_get_block(cur, level, &bp);
4454 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
4455 block->bb_u.l.bb_owner = cpu_to_be64(bbcoi->new_owner);
4457 block->bb_u.s.bb_owner = cpu_to_be32(bbcoi->new_owner);
4460 * If the block is a root block hosted in an inode, we might not have a
4461 * buffer pointer here and we shouldn't attempt to log the change as the
4462 * information is already held in the inode and discarded when the root
4463 * block is formatted into the on-disk inode fork. We still change it,
4464 * though, so everything is consistent in memory.
4468 xfs_trans_ordered_buf(cur->bc_tp, bp);
4469 xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
4471 xfs_buf_delwri_queue(bp, bbcoi->buffer_list);
4474 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
4475 ASSERT(level == cur->bc_nlevels - 1);
4482 xfs_btree_change_owner(
4483 struct xfs_btree_cur *cur,
4484 __uint64_t new_owner,
4485 struct list_head *buffer_list)
4487 struct xfs_btree_block_change_owner_info bbcoi;
4489 bbcoi.new_owner = new_owner;
4490 bbcoi.buffer_list = buffer_list;
4492 return xfs_btree_visit_blocks(cur, xfs_btree_block_change_owner,
4497 * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format
4500 * @bp: buffer containing the btree block
4501 * @max_recs: pointer to the m_*_mxr max records field in the xfs mount
4502 * @pag_max_level: pointer to the per-ag max level field
4505 xfs_btree_sblock_v5hdr_verify(
4508 struct xfs_mount *mp = bp->b_target->bt_mount;
4509 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4510 struct xfs_perag *pag = bp->b_pag;
4512 if (!xfs_sb_version_hascrc(&mp->m_sb))
4514 if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
4516 if (block->bb_u.s.bb_blkno != cpu_to_be64(bp->b_bn))
4518 if (pag && be32_to_cpu(block->bb_u.s.bb_owner) != pag->pag_agno)
4524 * xfs_btree_sblock_verify() -- verify a short-format btree block
4526 * @bp: buffer containing the btree block
4527 * @max_recs: maximum records allowed in this btree node
4530 xfs_btree_sblock_verify(
4532 unsigned int max_recs)
4534 struct xfs_mount *mp = bp->b_target->bt_mount;
4535 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4537 /* numrecs verification */
4538 if (be16_to_cpu(block->bb_numrecs) > max_recs)
4541 /* sibling pointer verification */
4542 if (!block->bb_u.s.bb_leftsib ||
4543 (be32_to_cpu(block->bb_u.s.bb_leftsib) >= mp->m_sb.sb_agblocks &&
4544 block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK)))
4546 if (!block->bb_u.s.bb_rightsib ||
4547 (be32_to_cpu(block->bb_u.s.bb_rightsib) >= mp->m_sb.sb_agblocks &&
4548 block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK)))
4555 * Calculate the number of btree levels needed to store a given number of
4556 * records in a short-format btree.
4559 xfs_btree_compute_maxlevels(
4560 struct xfs_mount *mp,
4565 unsigned long maxblocks;
4567 maxblocks = (len + limits[0] - 1) / limits[0];
4568 for (level = 1; maxblocks > 1; level++)
4569 maxblocks = (maxblocks + limits[1] - 1) / limits[1];
4574 * Query a regular btree for all records overlapping a given interval.
4575 * Start with a LE lookup of the key of low_rec and return all records
4576 * until we find a record with a key greater than the key of high_rec.
4579 xfs_btree_simple_query_range(
4580 struct xfs_btree_cur *cur,
4581 union xfs_btree_key *low_key,
4582 union xfs_btree_key *high_key,
4583 xfs_btree_query_range_fn fn,
4586 union xfs_btree_rec *recp;
4587 union xfs_btree_key rec_key;
4590 bool firstrec = true;
4593 ASSERT(cur->bc_ops->init_high_key_from_rec);
4594 ASSERT(cur->bc_ops->diff_two_keys);
4597 * Find the leftmost record. The btree cursor must be set
4598 * to the low record used to generate low_key.
4601 error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
4605 /* Nothing? See if there's anything to the right. */
4607 error = xfs_btree_increment(cur, 0, &stat);
4613 /* Find the record. */
4614 error = xfs_btree_get_rec(cur, &recp, &stat);
4618 /* Skip if high_key(rec) < low_key. */
4620 cur->bc_ops->init_high_key_from_rec(&rec_key, recp);
4622 diff = cur->bc_ops->diff_two_keys(cur, low_key,
4628 /* Stop if high_key < low_key(rec). */
4629 cur->bc_ops->init_key_from_rec(&rec_key, recp);
4630 diff = cur->bc_ops->diff_two_keys(cur, &rec_key, high_key);
4635 error = fn(cur, recp, priv);
4636 if (error < 0 || error == XFS_BTREE_QUERY_RANGE_ABORT)
4640 /* Move on to the next record. */
4641 error = xfs_btree_increment(cur, 0, &stat);
4651 * Query an overlapped interval btree for all records overlapping a given
4652 * interval. This function roughly follows the algorithm given in
4653 * "Interval Trees" of _Introduction to Algorithms_, which is section
4654 * 14.3 in the 2nd and 3rd editions.
4656 * First, generate keys for the low and high records passed in.
4658 * For any leaf node, generate the high and low keys for the record.
4659 * If the record keys overlap with the query low/high keys, pass the
4660 * record to the function iterator.
4662 * For any internal node, compare the low and high keys of each
4663 * pointer against the query low/high keys. If there's an overlap,
4664 * follow the pointer.
4666 * As an optimization, we stop scanning a block when we find a low key
4667 * that is greater than the query's high key.
4670 xfs_btree_overlapped_query_range(
4671 struct xfs_btree_cur *cur,
4672 union xfs_btree_key *low_key,
4673 union xfs_btree_key *high_key,
4674 xfs_btree_query_range_fn fn,
4677 union xfs_btree_ptr ptr;
4678 union xfs_btree_ptr *pp;
4679 union xfs_btree_key rec_key;
4680 union xfs_btree_key rec_hkey;
4681 union xfs_btree_key *lkp;
4682 union xfs_btree_key *hkp;
4683 union xfs_btree_rec *recp;
4684 struct xfs_btree_block *block;
4692 /* Load the root of the btree. */
4693 level = cur->bc_nlevels - 1;
4694 cur->bc_ops->init_ptr_from_cur(cur, &ptr);
4695 error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
4698 xfs_btree_get_block(cur, level, &bp);
4699 trace_xfs_btree_overlapped_query_range(cur, level, bp);
4701 error = xfs_btree_check_block(cur, block, level, bp);
4705 cur->bc_ptrs[level] = 1;
4707 while (level < cur->bc_nlevels) {
4708 block = xfs_btree_get_block(cur, level, &bp);
4710 /* End of node, pop back towards the root. */
4711 if (cur->bc_ptrs[level] > be16_to_cpu(block->bb_numrecs)) {
4713 if (level < cur->bc_nlevels - 1)
4714 cur->bc_ptrs[level + 1]++;
4720 /* Handle a leaf node. */
4721 recp = xfs_btree_rec_addr(cur, cur->bc_ptrs[0], block);
4723 cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
4724 ldiff = cur->bc_ops->diff_two_keys(cur, &rec_hkey,
4727 cur->bc_ops->init_key_from_rec(&rec_key, recp);
4728 hdiff = cur->bc_ops->diff_two_keys(cur, high_key,
4732 * If (record's high key >= query's low key) and
4733 * (query's high key >= record's low key), then
4734 * this record overlaps the query range; callback.
4736 if (ldiff >= 0 && hdiff >= 0) {
4737 error = fn(cur, recp, priv);
4739 error == XFS_BTREE_QUERY_RANGE_ABORT)
4741 } else if (hdiff < 0) {
4742 /* Record is larger than high key; pop. */
4745 cur->bc_ptrs[level]++;
4749 /* Handle an internal node. */
4750 lkp = xfs_btree_key_addr(cur, cur->bc_ptrs[level], block);
4751 hkp = xfs_btree_high_key_addr(cur, cur->bc_ptrs[level], block);
4752 pp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[level], block);
4754 ldiff = cur->bc_ops->diff_two_keys(cur, hkp, low_key);
4755 hdiff = cur->bc_ops->diff_two_keys(cur, high_key, lkp);
4758 * If (pointer's high key >= query's low key) and
4759 * (query's high key >= pointer's low key), then
4760 * this record overlaps the query range; follow pointer.
4762 if (ldiff >= 0 && hdiff >= 0) {
4764 error = xfs_btree_lookup_get_block(cur, level, pp,
4768 xfs_btree_get_block(cur, level, &bp);
4769 trace_xfs_btree_overlapped_query_range(cur, level, bp);
4771 error = xfs_btree_check_block(cur, block, level, bp);
4775 cur->bc_ptrs[level] = 1;
4777 } else if (hdiff < 0) {
4778 /* The low key is larger than the upper range; pop. */
4781 cur->bc_ptrs[level]++;
4786 * If we don't end this function with the cursor pointing at a record
4787 * block, a subsequent non-error cursor deletion will not release
4788 * node-level buffers, causing a buffer leak. This is quite possible
4789 * with a zero-results range query, so release the buffers if we
4790 * failed to return any results.
4792 if (cur->bc_bufs[0] == NULL) {
4793 for (i = 0; i < cur->bc_nlevels; i++) {
4794 if (cur->bc_bufs[i]) {
4795 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
4796 cur->bc_bufs[i] = NULL;
4797 cur->bc_ptrs[i] = 0;
4807 * Query a btree for all records overlapping a given interval of keys. The
4808 * supplied function will be called with each record found; return one of the
4809 * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
4810 * code. This function returns XFS_BTREE_QUERY_RANGE_ABORT, zero, or a
4811 * negative error code.
4814 xfs_btree_query_range(
4815 struct xfs_btree_cur *cur,
4816 union xfs_btree_irec *low_rec,
4817 union xfs_btree_irec *high_rec,
4818 xfs_btree_query_range_fn fn,
4821 union xfs_btree_rec rec;
4822 union xfs_btree_key low_key;
4823 union xfs_btree_key high_key;
4825 /* Find the keys of both ends of the interval. */
4826 cur->bc_rec = *high_rec;
4827 cur->bc_ops->init_rec_from_cur(cur, &rec);
4828 cur->bc_ops->init_key_from_rec(&high_key, &rec);
4830 cur->bc_rec = *low_rec;
4831 cur->bc_ops->init_rec_from_cur(cur, &rec);
4832 cur->bc_ops->init_key_from_rec(&low_key, &rec);
4834 /* Enforce low key < high key. */
4835 if (cur->bc_ops->diff_two_keys(cur, &low_key, &high_key) > 0)
4838 if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING))
4839 return xfs_btree_simple_query_range(cur, &low_key,
4840 &high_key, fn, priv);
4841 return xfs_btree_overlapped_query_range(cur, &low_key, &high_key,
4846 * Calculate the number of blocks needed to store a given number of records
4847 * in a short-format (per-AG metadata) btree.
4850 xfs_btree_calc_size(
4851 struct xfs_mount *mp,
4853 unsigned long long len)
4859 maxrecs = limits[0];
4860 for (level = 0, rval = 0; len > 1; level++) {
4862 do_div(len, maxrecs);
4863 maxrecs = limits[1];
4870 xfs_btree_count_blocks_helper(
4871 struct xfs_btree_cur *cur,
4875 xfs_extlen_t *blocks = data;
4881 /* Count the blocks in a btree and return the result in *blocks. */
4883 xfs_btree_count_blocks(
4884 struct xfs_btree_cur *cur,
4885 xfs_extlen_t *blocks)
4888 return xfs_btree_visit_blocks(cur, xfs_btree_count_blocks_helper,