1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_trans.h"
16 #include "xfs_buf_item.h"
17 #include "xfs_btree.h"
18 #include "xfs_errortag.h"
19 #include "xfs_error.h"
20 #include "xfs_trace.h"
21 #include "xfs_alloc.h"
23 #include "xfs_btree_staging.h"
26 * Cursor allocation zone.
28 kmem_zone_t *xfs_btree_cur_zone;
31 * Btree magic numbers.
33 static const uint32_t xfs_magics[2][XFS_BTNUM_MAX] = {
34 { XFS_ABTB_MAGIC, XFS_ABTC_MAGIC, 0, XFS_BMAP_MAGIC, XFS_IBT_MAGIC,
36 { XFS_ABTB_CRC_MAGIC, XFS_ABTC_CRC_MAGIC, XFS_RMAP_CRC_MAGIC,
37 XFS_BMAP_CRC_MAGIC, XFS_IBT_CRC_MAGIC, XFS_FIBT_CRC_MAGIC,
46 uint32_t magic = xfs_magics[crc][btnum];
48 /* Ensure we asked for crc for crc-only magics. */
54 * Check a long btree block header. Return the address of the failing check,
55 * or NULL if everything is ok.
58 __xfs_btree_check_lblock(
59 struct xfs_btree_cur *cur,
60 struct xfs_btree_block *block,
64 struct xfs_mount *mp = cur->bc_mp;
65 xfs_btnum_t btnum = cur->bc_btnum;
66 int crc = xfs_sb_version_hascrc(&mp->m_sb);
69 if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
70 return __this_address;
71 if (block->bb_u.l.bb_blkno !=
72 cpu_to_be64(bp ? bp->b_bn : XFS_BUF_DADDR_NULL))
73 return __this_address;
74 if (block->bb_u.l.bb_pad != cpu_to_be32(0))
75 return __this_address;
78 if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum))
79 return __this_address;
80 if (be16_to_cpu(block->bb_level) != level)
81 return __this_address;
82 if (be16_to_cpu(block->bb_numrecs) >
83 cur->bc_ops->get_maxrecs(cur, level))
84 return __this_address;
85 if (block->bb_u.l.bb_leftsib != cpu_to_be64(NULLFSBLOCK) &&
86 !xfs_btree_check_lptr(cur, be64_to_cpu(block->bb_u.l.bb_leftsib),
88 return __this_address;
89 if (block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK) &&
90 !xfs_btree_check_lptr(cur, be64_to_cpu(block->bb_u.l.bb_rightsib),
92 return __this_address;
97 /* Check a long btree block header. */
99 xfs_btree_check_lblock(
100 struct xfs_btree_cur *cur,
101 struct xfs_btree_block *block,
105 struct xfs_mount *mp = cur->bc_mp;
108 fa = __xfs_btree_check_lblock(cur, block, level, bp);
109 if (XFS_IS_CORRUPT(mp, fa != NULL) ||
110 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BTREE_CHECK_LBLOCK)) {
112 trace_xfs_btree_corrupt(bp, _RET_IP_);
113 return -EFSCORRUPTED;
119 * Check a short btree block header. Return the address of the failing check,
120 * or NULL if everything is ok.
123 __xfs_btree_check_sblock(
124 struct xfs_btree_cur *cur,
125 struct xfs_btree_block *block,
129 struct xfs_mount *mp = cur->bc_mp;
130 xfs_btnum_t btnum = cur->bc_btnum;
131 int crc = xfs_sb_version_hascrc(&mp->m_sb);
134 if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
135 return __this_address;
136 if (block->bb_u.s.bb_blkno !=
137 cpu_to_be64(bp ? bp->b_bn : XFS_BUF_DADDR_NULL))
138 return __this_address;
141 if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum))
142 return __this_address;
143 if (be16_to_cpu(block->bb_level) != level)
144 return __this_address;
145 if (be16_to_cpu(block->bb_numrecs) >
146 cur->bc_ops->get_maxrecs(cur, level))
147 return __this_address;
148 if (block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK) &&
149 !xfs_btree_check_sptr(cur, be32_to_cpu(block->bb_u.s.bb_leftsib),
151 return __this_address;
152 if (block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK) &&
153 !xfs_btree_check_sptr(cur, be32_to_cpu(block->bb_u.s.bb_rightsib),
155 return __this_address;
160 /* Check a short btree block header. */
162 xfs_btree_check_sblock(
163 struct xfs_btree_cur *cur,
164 struct xfs_btree_block *block,
168 struct xfs_mount *mp = cur->bc_mp;
171 fa = __xfs_btree_check_sblock(cur, block, level, bp);
172 if (XFS_IS_CORRUPT(mp, fa != NULL) ||
173 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BTREE_CHECK_SBLOCK)) {
175 trace_xfs_btree_corrupt(bp, _RET_IP_);
176 return -EFSCORRUPTED;
182 * Debug routine: check that block header is ok.
185 xfs_btree_check_block(
186 struct xfs_btree_cur *cur, /* btree cursor */
187 struct xfs_btree_block *block, /* generic btree block pointer */
188 int level, /* level of the btree block */
189 struct xfs_buf *bp) /* buffer containing block, if any */
191 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
192 return xfs_btree_check_lblock(cur, block, level, bp);
194 return xfs_btree_check_sblock(cur, block, level, bp);
197 /* Check that this long pointer is valid and points within the fs. */
199 xfs_btree_check_lptr(
200 struct xfs_btree_cur *cur,
206 return xfs_verify_fsbno(cur->bc_mp, fsbno);
209 /* Check that this short pointer is valid and points within the AG. */
211 xfs_btree_check_sptr(
212 struct xfs_btree_cur *cur,
218 return xfs_verify_agbno(cur->bc_mp, cur->bc_ag.agno, agbno);
222 * Check that a given (indexed) btree pointer at a certain level of a
223 * btree is valid and doesn't point past where it should.
227 struct xfs_btree_cur *cur,
228 union xfs_btree_ptr *ptr,
232 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
233 if (xfs_btree_check_lptr(cur, be64_to_cpu((&ptr->l)[index]),
237 "Inode %llu fork %d: Corrupt btree %d pointer at level %d index %d.",
238 cur->bc_ino.ip->i_ino,
239 cur->bc_ino.whichfork, cur->bc_btnum,
242 if (xfs_btree_check_sptr(cur, be32_to_cpu((&ptr->s)[index]),
246 "AG %u: Corrupt btree %d pointer at level %d index %d.",
247 cur->bc_ag.agno, cur->bc_btnum,
251 return -EFSCORRUPTED;
255 # define xfs_btree_debug_check_ptr xfs_btree_check_ptr
257 # define xfs_btree_debug_check_ptr(...) (0)
261 * Calculate CRC on the whole btree block and stuff it into the
262 * long-form btree header.
264 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
265 * it into the buffer so recovery knows what the last modification was that made
269 xfs_btree_lblock_calc_crc(
272 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
273 struct xfs_buf_log_item *bip = bp->b_log_item;
275 if (!xfs_sb_version_hascrc(&bp->b_mount->m_sb))
278 block->bb_u.l.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
279 xfs_buf_update_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
283 xfs_btree_lblock_verify_crc(
286 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
287 struct xfs_mount *mp = bp->b_mount;
289 if (xfs_sb_version_hascrc(&mp->m_sb)) {
290 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.l.bb_lsn)))
292 return xfs_buf_verify_cksum(bp, XFS_BTREE_LBLOCK_CRC_OFF);
299 * Calculate CRC on the whole btree block and stuff it into the
300 * short-form btree header.
302 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
303 * it into the buffer so recovery knows what the last modification was that made
307 xfs_btree_sblock_calc_crc(
310 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
311 struct xfs_buf_log_item *bip = bp->b_log_item;
313 if (!xfs_sb_version_hascrc(&bp->b_mount->m_sb))
316 block->bb_u.s.bb_lsn = cpu_to_be64(bip->bli_item.li_lsn);
317 xfs_buf_update_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
321 xfs_btree_sblock_verify_crc(
324 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
325 struct xfs_mount *mp = bp->b_mount;
327 if (xfs_sb_version_hascrc(&mp->m_sb)) {
328 if (!xfs_log_check_lsn(mp, be64_to_cpu(block->bb_u.s.bb_lsn)))
330 return xfs_buf_verify_cksum(bp, XFS_BTREE_SBLOCK_CRC_OFF);
337 xfs_btree_free_block(
338 struct xfs_btree_cur *cur,
343 error = cur->bc_ops->free_block(cur, bp);
345 xfs_trans_binval(cur->bc_tp, bp);
346 XFS_BTREE_STATS_INC(cur, free);
352 * Delete the btree cursor.
355 xfs_btree_del_cursor(
356 struct xfs_btree_cur *cur, /* btree cursor */
357 int error) /* del because of error */
359 int i; /* btree level */
362 * Clear the buffer pointers and release the buffers. If we're doing
363 * this because of an error, inspect all of the entries in the bc_bufs
364 * array for buffers to be unlocked. This is because some of the btree
365 * code works from level n down to 0, and if we get an error along the
366 * way we won't have initialized all the entries down to 0.
368 for (i = 0; i < cur->bc_nlevels; i++) {
370 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
376 * If we are doing a BMBT update, the number of unaccounted blocks
377 * allocated during this cursor life time should be zero. If it's not
378 * zero, then we should be shut down or on our way to shutdown due to
379 * cancelling a dirty transaction on error.
381 ASSERT(cur->bc_btnum != XFS_BTNUM_BMAP || cur->bc_ino.allocated == 0 ||
382 XFS_FORCED_SHUTDOWN(cur->bc_mp) || error != 0);
383 if (unlikely(cur->bc_flags & XFS_BTREE_STAGING))
384 kmem_free(cur->bc_ops);
385 kmem_cache_free(xfs_btree_cur_zone, cur);
389 * Duplicate the btree cursor.
390 * Allocate a new one, copy the record, re-get the buffers.
393 xfs_btree_dup_cursor(
394 xfs_btree_cur_t *cur, /* input cursor */
395 xfs_btree_cur_t **ncur) /* output cursor */
397 xfs_buf_t *bp; /* btree block's buffer pointer */
398 int error; /* error return value */
399 int i; /* level number of btree block */
400 xfs_mount_t *mp; /* mount structure for filesystem */
401 xfs_btree_cur_t *new; /* new cursor value */
402 xfs_trans_t *tp; /* transaction pointer, can be NULL */
408 * Allocate a new cursor like the old one.
410 new = cur->bc_ops->dup_cursor(cur);
413 * Copy the record currently in the cursor.
415 new->bc_rec = cur->bc_rec;
418 * For each level current, re-get the buffer and copy the ptr value.
420 for (i = 0; i < new->bc_nlevels; i++) {
421 new->bc_ptrs[i] = cur->bc_ptrs[i];
422 new->bc_ra[i] = cur->bc_ra[i];
423 bp = cur->bc_bufs[i];
425 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
426 XFS_BUF_ADDR(bp), mp->m_bsize,
428 cur->bc_ops->buf_ops);
430 xfs_btree_del_cursor(new, error);
435 new->bc_bufs[i] = bp;
442 * XFS btree block layout and addressing:
444 * There are two types of blocks in the btree: leaf and non-leaf blocks.
446 * The leaf record start with a header then followed by records containing
447 * the values. A non-leaf block also starts with the same header, and
448 * then first contains lookup keys followed by an equal number of pointers
449 * to the btree blocks at the previous level.
451 * +--------+-------+-------+-------+-------+-------+-------+
452 * Leaf: | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
453 * +--------+-------+-------+-------+-------+-------+-------+
455 * +--------+-------+-------+-------+-------+-------+-------+
456 * Non-Leaf: | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
457 * +--------+-------+-------+-------+-------+-------+-------+
459 * The header is called struct xfs_btree_block for reasons better left unknown
460 * and comes in different versions for short (32bit) and long (64bit) block
461 * pointers. The record and key structures are defined by the btree instances
462 * and opaque to the btree core. The block pointers are simple disk endian
463 * integers, available in a short (32bit) and long (64bit) variant.
465 * The helpers below calculate the offset of a given record, key or pointer
466 * into a btree block (xfs_btree_*_offset) or return a pointer to the given
467 * record, key or pointer (xfs_btree_*_addr). Note that all addressing
468 * inside the btree block is done using indices starting at one, not zero!
470 * If XFS_BTREE_OVERLAPPING is set, then this btree supports keys containing
471 * overlapping intervals. In such a tree, records are still sorted lowest to
472 * highest and indexed by the smallest key value that refers to the record.
473 * However, nodes are different: each pointer has two associated keys -- one
474 * indexing the lowest key available in the block(s) below (the same behavior
475 * as the key in a regular btree) and another indexing the highest key
476 * available in the block(s) below. Because records are /not/ sorted by the
477 * highest key, all leaf block updates require us to compute the highest key
478 * that matches any record in the leaf and to recursively update the high keys
479 * in the nodes going further up in the tree, if necessary. Nodes look like
482 * +--------+-----+-----+-----+-----+-----+-------+-------+-----+
483 * Non-Leaf: | header | lo1 | hi1 | lo2 | hi2 | ... | ptr 1 | ptr 2 | ... |
484 * +--------+-----+-----+-----+-----+-----+-------+-------+-----+
486 * To perform an interval query on an overlapped tree, perform the usual
487 * depth-first search and use the low and high keys to decide if we can skip
488 * that particular node. If a leaf node is reached, return the records that
489 * intersect the interval. Note that an interval query may return numerous
490 * entries. For a non-overlapped tree, simply search for the record associated
491 * with the lowest key and iterate forward until a non-matching record is
492 * found. Section 14.3 ("Interval Trees") of _Introduction to Algorithms_ by
493 * Cormen, Leiserson, Rivest, and Stein (2nd or 3rd ed. only) discuss this in
496 * Why do we care about overlapping intervals? Let's say you have a bunch of
497 * reverse mapping records on a reflink filesystem:
499 * 1: +- file A startblock B offset C length D -----------+
500 * 2: +- file E startblock F offset G length H --------------+
501 * 3: +- file I startblock F offset J length K --+
502 * 4: +- file L... --+
504 * Now say we want to map block (B+D) into file A at offset (C+D). Ideally,
505 * we'd simply increment the length of record 1. But how do we find the record
506 * that ends at (B+D-1) (i.e. record 1)? A LE lookup of (B+D-1) would return
507 * record 3 because the keys are ordered first by startblock. An interval
508 * query would return records 1 and 2 because they both overlap (B+D-1), and
509 * from that we can pick out record 1 as the appropriate left neighbor.
511 * In the non-overlapped case you can do a LE lookup and decrement the cursor
512 * because a record's interval must end before the next record.
516 * Return size of the btree block header for this btree instance.
518 static inline size_t xfs_btree_block_len(struct xfs_btree_cur *cur)
520 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
521 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
522 return XFS_BTREE_LBLOCK_CRC_LEN;
523 return XFS_BTREE_LBLOCK_LEN;
525 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS)
526 return XFS_BTREE_SBLOCK_CRC_LEN;
527 return XFS_BTREE_SBLOCK_LEN;
531 * Return size of btree block pointers for this btree instance.
533 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur *cur)
535 return (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
536 sizeof(__be64) : sizeof(__be32);
540 * Calculate offset of the n-th record in a btree block.
543 xfs_btree_rec_offset(
544 struct xfs_btree_cur *cur,
547 return xfs_btree_block_len(cur) +
548 (n - 1) * cur->bc_ops->rec_len;
552 * Calculate offset of the n-th key in a btree block.
555 xfs_btree_key_offset(
556 struct xfs_btree_cur *cur,
559 return xfs_btree_block_len(cur) +
560 (n - 1) * cur->bc_ops->key_len;
564 * Calculate offset of the n-th high key in a btree block.
567 xfs_btree_high_key_offset(
568 struct xfs_btree_cur *cur,
571 return xfs_btree_block_len(cur) +
572 (n - 1) * cur->bc_ops->key_len + (cur->bc_ops->key_len / 2);
576 * Calculate offset of the n-th block pointer in a btree block.
579 xfs_btree_ptr_offset(
580 struct xfs_btree_cur *cur,
584 return xfs_btree_block_len(cur) +
585 cur->bc_ops->get_maxrecs(cur, level) * cur->bc_ops->key_len +
586 (n - 1) * xfs_btree_ptr_len(cur);
590 * Return a pointer to the n-th record in the btree block.
592 union xfs_btree_rec *
594 struct xfs_btree_cur *cur,
596 struct xfs_btree_block *block)
598 return (union xfs_btree_rec *)
599 ((char *)block + xfs_btree_rec_offset(cur, n));
603 * Return a pointer to the n-th key in the btree block.
605 union xfs_btree_key *
607 struct xfs_btree_cur *cur,
609 struct xfs_btree_block *block)
611 return (union xfs_btree_key *)
612 ((char *)block + xfs_btree_key_offset(cur, n));
616 * Return a pointer to the n-th high key in the btree block.
618 union xfs_btree_key *
619 xfs_btree_high_key_addr(
620 struct xfs_btree_cur *cur,
622 struct xfs_btree_block *block)
624 return (union xfs_btree_key *)
625 ((char *)block + xfs_btree_high_key_offset(cur, n));
629 * Return a pointer to the n-th block pointer in the btree block.
631 union xfs_btree_ptr *
633 struct xfs_btree_cur *cur,
635 struct xfs_btree_block *block)
637 int level = xfs_btree_get_level(block);
639 ASSERT(block->bb_level != 0);
641 return (union xfs_btree_ptr *)
642 ((char *)block + xfs_btree_ptr_offset(cur, n, level));
647 struct xfs_btree_cur *cur)
649 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
651 if (cur->bc_flags & XFS_BTREE_STAGING)
652 return cur->bc_ino.ifake->if_fork;
653 return XFS_IFORK_PTR(cur->bc_ino.ip, cur->bc_ino.whichfork);
657 * Get the root block which is stored in the inode.
659 * For now this btree implementation assumes the btree root is always
660 * stored in the if_broot field of an inode fork.
662 STATIC struct xfs_btree_block *
664 struct xfs_btree_cur *cur)
666 struct xfs_ifork *ifp = xfs_btree_ifork_ptr(cur);
668 return (struct xfs_btree_block *)ifp->if_broot;
672 * Retrieve the block pointer from the cursor at the given level.
673 * This may be an inode btree root or from a buffer.
675 struct xfs_btree_block * /* generic btree block pointer */
677 struct xfs_btree_cur *cur, /* btree cursor */
678 int level, /* level in btree */
679 struct xfs_buf **bpp) /* buffer containing the block */
681 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
682 (level == cur->bc_nlevels - 1)) {
684 return xfs_btree_get_iroot(cur);
687 *bpp = cur->bc_bufs[level];
688 return XFS_BUF_TO_BLOCK(*bpp);
692 * Change the cursor to point to the first record at the given level.
693 * Other levels are unaffected.
695 STATIC int /* success=1, failure=0 */
697 xfs_btree_cur_t *cur, /* btree cursor */
698 int level) /* level to change */
700 struct xfs_btree_block *block; /* generic btree block pointer */
701 xfs_buf_t *bp; /* buffer containing block */
704 * Get the block pointer for this level.
706 block = xfs_btree_get_block(cur, level, &bp);
707 if (xfs_btree_check_block(cur, block, level, bp))
710 * It's empty, there is no such record.
712 if (!block->bb_numrecs)
715 * Set the ptr value to 1, that's the first record/key.
717 cur->bc_ptrs[level] = 1;
722 * Change the cursor to point to the last record in the current block
723 * at the given level. Other levels are unaffected.
725 STATIC int /* success=1, failure=0 */
727 xfs_btree_cur_t *cur, /* btree cursor */
728 int level) /* level to change */
730 struct xfs_btree_block *block; /* generic btree block pointer */
731 xfs_buf_t *bp; /* buffer containing block */
734 * Get the block pointer for this level.
736 block = xfs_btree_get_block(cur, level, &bp);
737 if (xfs_btree_check_block(cur, block, level, bp))
740 * It's empty, there is no such record.
742 if (!block->bb_numrecs)
745 * Set the ptr value to numrecs, that's the last record/key.
747 cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
752 * Compute first and last byte offsets for the fields given.
753 * Interprets the offsets table, which contains struct field offsets.
757 int64_t fields, /* bitmask of fields */
758 const short *offsets, /* table of field offsets */
759 int nbits, /* number of bits to inspect */
760 int *first, /* output: first byte offset */
761 int *last) /* output: last byte offset */
763 int i; /* current bit number */
764 int64_t imask; /* mask for current bit number */
768 * Find the lowest bit, so the first byte offset.
770 for (i = 0, imask = 1LL; ; i++, imask <<= 1) {
771 if (imask & fields) {
777 * Find the highest bit, so the last byte offset.
779 for (i = nbits - 1, imask = 1LL << i; ; i--, imask >>= 1) {
780 if (imask & fields) {
781 *last = offsets[i + 1] - 1;
788 * Get a buffer for the block, return it read in.
789 * Long-form addressing.
793 struct xfs_mount *mp, /* file system mount point */
794 struct xfs_trans *tp, /* transaction pointer */
795 xfs_fsblock_t fsbno, /* file system block number */
796 struct xfs_buf **bpp, /* buffer for fsbno */
797 int refval, /* ref count value for buffer */
798 const struct xfs_buf_ops *ops)
800 struct xfs_buf *bp; /* return value */
801 xfs_daddr_t d; /* real disk block address */
804 if (!xfs_verify_fsbno(mp, fsbno))
805 return -EFSCORRUPTED;
806 d = XFS_FSB_TO_DADDR(mp, fsbno);
807 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, d,
808 mp->m_bsize, 0, &bp, ops);
812 xfs_buf_set_ref(bp, refval);
818 * Read-ahead the block, don't wait for it, don't return a buffer.
819 * Long-form addressing.
823 xfs_btree_reada_bufl(
824 struct xfs_mount *mp, /* file system mount point */
825 xfs_fsblock_t fsbno, /* file system block number */
826 xfs_extlen_t count, /* count of filesystem blocks */
827 const struct xfs_buf_ops *ops)
831 ASSERT(fsbno != NULLFSBLOCK);
832 d = XFS_FSB_TO_DADDR(mp, fsbno);
833 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
837 * Read-ahead the block, don't wait for it, don't return a buffer.
838 * Short-form addressing.
842 xfs_btree_reada_bufs(
843 struct xfs_mount *mp, /* file system mount point */
844 xfs_agnumber_t agno, /* allocation group number */
845 xfs_agblock_t agbno, /* allocation group block number */
846 xfs_extlen_t count, /* count of filesystem blocks */
847 const struct xfs_buf_ops *ops)
851 ASSERT(agno != NULLAGNUMBER);
852 ASSERT(agbno != NULLAGBLOCK);
853 d = XFS_AGB_TO_DADDR(mp, agno, agbno);
854 xfs_buf_readahead(mp->m_ddev_targp, d, mp->m_bsize * count, ops);
858 xfs_btree_readahead_lblock(
859 struct xfs_btree_cur *cur,
861 struct xfs_btree_block *block)
864 xfs_fsblock_t left = be64_to_cpu(block->bb_u.l.bb_leftsib);
865 xfs_fsblock_t right = be64_to_cpu(block->bb_u.l.bb_rightsib);
867 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLFSBLOCK) {
868 xfs_btree_reada_bufl(cur->bc_mp, left, 1,
869 cur->bc_ops->buf_ops);
873 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLFSBLOCK) {
874 xfs_btree_reada_bufl(cur->bc_mp, right, 1,
875 cur->bc_ops->buf_ops);
883 xfs_btree_readahead_sblock(
884 struct xfs_btree_cur *cur,
886 struct xfs_btree_block *block)
889 xfs_agblock_t left = be32_to_cpu(block->bb_u.s.bb_leftsib);
890 xfs_agblock_t right = be32_to_cpu(block->bb_u.s.bb_rightsib);
893 if ((lr & XFS_BTCUR_LEFTRA) && left != NULLAGBLOCK) {
894 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_ag.agno,
895 left, 1, cur->bc_ops->buf_ops);
899 if ((lr & XFS_BTCUR_RIGHTRA) && right != NULLAGBLOCK) {
900 xfs_btree_reada_bufs(cur->bc_mp, cur->bc_ag.agno,
901 right, 1, cur->bc_ops->buf_ops);
909 * Read-ahead btree blocks, at the given level.
910 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
914 struct xfs_btree_cur *cur, /* btree cursor */
915 int lev, /* level in btree */
916 int lr) /* left/right bits */
918 struct xfs_btree_block *block;
921 * No readahead needed if we are at the root level and the
922 * btree root is stored in the inode.
924 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
925 (lev == cur->bc_nlevels - 1))
928 if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
931 cur->bc_ra[lev] |= lr;
932 block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
934 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
935 return xfs_btree_readahead_lblock(cur, lr, block);
936 return xfs_btree_readahead_sblock(cur, lr, block);
940 xfs_btree_ptr_to_daddr(
941 struct xfs_btree_cur *cur,
942 union xfs_btree_ptr *ptr,
949 error = xfs_btree_check_ptr(cur, ptr, 0, 1);
953 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
954 fsbno = be64_to_cpu(ptr->l);
955 *daddr = XFS_FSB_TO_DADDR(cur->bc_mp, fsbno);
957 agbno = be32_to_cpu(ptr->s);
958 *daddr = XFS_AGB_TO_DADDR(cur->bc_mp, cur->bc_ag.agno,
966 * Readahead @count btree blocks at the given @ptr location.
968 * We don't need to care about long or short form btrees here as we have a
969 * method of converting the ptr directly to a daddr available to us.
972 xfs_btree_readahead_ptr(
973 struct xfs_btree_cur *cur,
974 union xfs_btree_ptr *ptr,
979 if (xfs_btree_ptr_to_daddr(cur, ptr, &daddr))
981 xfs_buf_readahead(cur->bc_mp->m_ddev_targp, daddr,
982 cur->bc_mp->m_bsize * count, cur->bc_ops->buf_ops);
986 * Set the buffer for level "lev" in the cursor to bp, releasing
987 * any previous buffer.
991 xfs_btree_cur_t *cur, /* btree cursor */
992 int lev, /* level in btree */
993 xfs_buf_t *bp) /* new buffer to set */
995 struct xfs_btree_block *b; /* btree block */
997 if (cur->bc_bufs[lev])
998 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
999 cur->bc_bufs[lev] = bp;
1000 cur->bc_ra[lev] = 0;
1002 b = XFS_BUF_TO_BLOCK(bp);
1003 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1004 if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK))
1005 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1006 if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK))
1007 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1009 if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
1010 cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
1011 if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
1012 cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
1017 xfs_btree_ptr_is_null(
1018 struct xfs_btree_cur *cur,
1019 union xfs_btree_ptr *ptr)
1021 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1022 return ptr->l == cpu_to_be64(NULLFSBLOCK);
1024 return ptr->s == cpu_to_be32(NULLAGBLOCK);
1028 xfs_btree_set_ptr_null(
1029 struct xfs_btree_cur *cur,
1030 union xfs_btree_ptr *ptr)
1032 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1033 ptr->l = cpu_to_be64(NULLFSBLOCK);
1035 ptr->s = cpu_to_be32(NULLAGBLOCK);
1039 * Get/set/init sibling pointers
1042 xfs_btree_get_sibling(
1043 struct xfs_btree_cur *cur,
1044 struct xfs_btree_block *block,
1045 union xfs_btree_ptr *ptr,
1048 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1050 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1051 if (lr == XFS_BB_RIGHTSIB)
1052 ptr->l = block->bb_u.l.bb_rightsib;
1054 ptr->l = block->bb_u.l.bb_leftsib;
1056 if (lr == XFS_BB_RIGHTSIB)
1057 ptr->s = block->bb_u.s.bb_rightsib;
1059 ptr->s = block->bb_u.s.bb_leftsib;
1064 xfs_btree_set_sibling(
1065 struct xfs_btree_cur *cur,
1066 struct xfs_btree_block *block,
1067 union xfs_btree_ptr *ptr,
1070 ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
1072 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
1073 if (lr == XFS_BB_RIGHTSIB)
1074 block->bb_u.l.bb_rightsib = ptr->l;
1076 block->bb_u.l.bb_leftsib = ptr->l;
1078 if (lr == XFS_BB_RIGHTSIB)
1079 block->bb_u.s.bb_rightsib = ptr->s;
1081 block->bb_u.s.bb_leftsib = ptr->s;
1086 xfs_btree_init_block_int(
1087 struct xfs_mount *mp,
1088 struct xfs_btree_block *buf,
1096 int crc = xfs_sb_version_hascrc(&mp->m_sb);
1097 __u32 magic = xfs_btree_magic(crc, btnum);
1099 buf->bb_magic = cpu_to_be32(magic);
1100 buf->bb_level = cpu_to_be16(level);
1101 buf->bb_numrecs = cpu_to_be16(numrecs);
1103 if (flags & XFS_BTREE_LONG_PTRS) {
1104 buf->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK);
1105 buf->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK);
1107 buf->bb_u.l.bb_blkno = cpu_to_be64(blkno);
1108 buf->bb_u.l.bb_owner = cpu_to_be64(owner);
1109 uuid_copy(&buf->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid);
1110 buf->bb_u.l.bb_pad = 0;
1111 buf->bb_u.l.bb_lsn = 0;
1114 /* owner is a 32 bit value on short blocks */
1115 __u32 __owner = (__u32)owner;
1117 buf->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
1118 buf->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
1120 buf->bb_u.s.bb_blkno = cpu_to_be64(blkno);
1121 buf->bb_u.s.bb_owner = cpu_to_be32(__owner);
1122 uuid_copy(&buf->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid);
1123 buf->bb_u.s.bb_lsn = 0;
1129 xfs_btree_init_block(
1130 struct xfs_mount *mp,
1137 xfs_btree_init_block_int(mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1138 btnum, level, numrecs, owner, 0);
1142 xfs_btree_init_block_cur(
1143 struct xfs_btree_cur *cur,
1151 * we can pull the owner from the cursor right now as the different
1152 * owners align directly with the pointer size of the btree. This may
1153 * change in future, but is safe for current users of the generic btree
1156 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1157 owner = cur->bc_ino.ip->i_ino;
1159 owner = cur->bc_ag.agno;
1161 xfs_btree_init_block_int(cur->bc_mp, XFS_BUF_TO_BLOCK(bp), bp->b_bn,
1162 cur->bc_btnum, level, numrecs,
1163 owner, cur->bc_flags);
1167 * Return true if ptr is the last record in the btree and
1168 * we need to track updates to this record. The decision
1169 * will be further refined in the update_lastrec method.
1172 xfs_btree_is_lastrec(
1173 struct xfs_btree_cur *cur,
1174 struct xfs_btree_block *block,
1177 union xfs_btree_ptr ptr;
1181 if (!(cur->bc_flags & XFS_BTREE_LASTREC_UPDATE))
1184 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1185 if (!xfs_btree_ptr_is_null(cur, &ptr))
1191 xfs_btree_buf_to_ptr(
1192 struct xfs_btree_cur *cur,
1194 union xfs_btree_ptr *ptr)
1196 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
1197 ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
1200 ptr->s = cpu_to_be32(xfs_daddr_to_agbno(cur->bc_mp,
1207 struct xfs_btree_cur *cur,
1210 switch (cur->bc_btnum) {
1213 xfs_buf_set_ref(bp, XFS_ALLOC_BTREE_REF);
1216 case XFS_BTNUM_FINO:
1217 xfs_buf_set_ref(bp, XFS_INO_BTREE_REF);
1219 case XFS_BTNUM_BMAP:
1220 xfs_buf_set_ref(bp, XFS_BMAP_BTREE_REF);
1222 case XFS_BTNUM_RMAP:
1223 xfs_buf_set_ref(bp, XFS_RMAP_BTREE_REF);
1225 case XFS_BTNUM_REFC:
1226 xfs_buf_set_ref(bp, XFS_REFC_BTREE_REF);
1234 xfs_btree_get_buf_block(
1235 struct xfs_btree_cur *cur,
1236 union xfs_btree_ptr *ptr,
1237 struct xfs_btree_block **block,
1238 struct xfs_buf **bpp)
1240 struct xfs_mount *mp = cur->bc_mp;
1244 error = xfs_btree_ptr_to_daddr(cur, ptr, &d);
1247 error = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d, mp->m_bsize,
1252 (*bpp)->b_ops = cur->bc_ops->buf_ops;
1253 *block = XFS_BUF_TO_BLOCK(*bpp);
1258 * Read in the buffer at the given ptr and return the buffer and
1259 * the block pointer within the buffer.
1262 xfs_btree_read_buf_block(
1263 struct xfs_btree_cur *cur,
1264 union xfs_btree_ptr *ptr,
1266 struct xfs_btree_block **block,
1267 struct xfs_buf **bpp)
1269 struct xfs_mount *mp = cur->bc_mp;
1273 /* need to sort out how callers deal with failures first */
1274 ASSERT(!(flags & XBF_TRYLOCK));
1276 error = xfs_btree_ptr_to_daddr(cur, ptr, &d);
1279 error = xfs_trans_read_buf(mp, cur->bc_tp, mp->m_ddev_targp, d,
1280 mp->m_bsize, flags, bpp,
1281 cur->bc_ops->buf_ops);
1285 xfs_btree_set_refs(cur, *bpp);
1286 *block = XFS_BUF_TO_BLOCK(*bpp);
1291 * Copy keys from one btree block to another.
1294 xfs_btree_copy_keys(
1295 struct xfs_btree_cur *cur,
1296 union xfs_btree_key *dst_key,
1297 union xfs_btree_key *src_key,
1300 ASSERT(numkeys >= 0);
1301 memcpy(dst_key, src_key, numkeys * cur->bc_ops->key_len);
1305 * Copy records from one btree block to another.
1308 xfs_btree_copy_recs(
1309 struct xfs_btree_cur *cur,
1310 union xfs_btree_rec *dst_rec,
1311 union xfs_btree_rec *src_rec,
1314 ASSERT(numrecs >= 0);
1315 memcpy(dst_rec, src_rec, numrecs * cur->bc_ops->rec_len);
1319 * Copy block pointers from one btree block to another.
1322 xfs_btree_copy_ptrs(
1323 struct xfs_btree_cur *cur,
1324 union xfs_btree_ptr *dst_ptr,
1325 const union xfs_btree_ptr *src_ptr,
1328 ASSERT(numptrs >= 0);
1329 memcpy(dst_ptr, src_ptr, numptrs * xfs_btree_ptr_len(cur));
1333 * Shift keys one index left/right inside a single btree block.
1336 xfs_btree_shift_keys(
1337 struct xfs_btree_cur *cur,
1338 union xfs_btree_key *key,
1344 ASSERT(numkeys >= 0);
1345 ASSERT(dir == 1 || dir == -1);
1347 dst_key = (char *)key + (dir * cur->bc_ops->key_len);
1348 memmove(dst_key, key, numkeys * cur->bc_ops->key_len);
1352 * Shift records one index left/right inside a single btree block.
1355 xfs_btree_shift_recs(
1356 struct xfs_btree_cur *cur,
1357 union xfs_btree_rec *rec,
1363 ASSERT(numrecs >= 0);
1364 ASSERT(dir == 1 || dir == -1);
1366 dst_rec = (char *)rec + (dir * cur->bc_ops->rec_len);
1367 memmove(dst_rec, rec, numrecs * cur->bc_ops->rec_len);
1371 * Shift block pointers one index left/right inside a single btree block.
1374 xfs_btree_shift_ptrs(
1375 struct xfs_btree_cur *cur,
1376 union xfs_btree_ptr *ptr,
1382 ASSERT(numptrs >= 0);
1383 ASSERT(dir == 1 || dir == -1);
1385 dst_ptr = (char *)ptr + (dir * xfs_btree_ptr_len(cur));
1386 memmove(dst_ptr, ptr, numptrs * xfs_btree_ptr_len(cur));
1390 * Log key values from the btree block.
1394 struct xfs_btree_cur *cur,
1401 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1402 xfs_trans_log_buf(cur->bc_tp, bp,
1403 xfs_btree_key_offset(cur, first),
1404 xfs_btree_key_offset(cur, last + 1) - 1);
1406 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1407 xfs_ilog_fbroot(cur->bc_ino.whichfork));
1412 * Log record values from the btree block.
1416 struct xfs_btree_cur *cur,
1422 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1423 xfs_trans_log_buf(cur->bc_tp, bp,
1424 xfs_btree_rec_offset(cur, first),
1425 xfs_btree_rec_offset(cur, last + 1) - 1);
1430 * Log block pointer fields from a btree block (nonleaf).
1434 struct xfs_btree_cur *cur, /* btree cursor */
1435 struct xfs_buf *bp, /* buffer containing btree block */
1436 int first, /* index of first pointer to log */
1437 int last) /* index of last pointer to log */
1441 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
1442 int level = xfs_btree_get_level(block);
1444 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1445 xfs_trans_log_buf(cur->bc_tp, bp,
1446 xfs_btree_ptr_offset(cur, first, level),
1447 xfs_btree_ptr_offset(cur, last + 1, level) - 1);
1449 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1450 xfs_ilog_fbroot(cur->bc_ino.whichfork));
1456 * Log fields from a btree block header.
1459 xfs_btree_log_block(
1460 struct xfs_btree_cur *cur, /* btree cursor */
1461 struct xfs_buf *bp, /* buffer containing btree block */
1462 int fields) /* mask of fields: XFS_BB_... */
1464 int first; /* first byte offset logged */
1465 int last; /* last byte offset logged */
1466 static const short soffsets[] = { /* table of offsets (short) */
1467 offsetof(struct xfs_btree_block, bb_magic),
1468 offsetof(struct xfs_btree_block, bb_level),
1469 offsetof(struct xfs_btree_block, bb_numrecs),
1470 offsetof(struct xfs_btree_block, bb_u.s.bb_leftsib),
1471 offsetof(struct xfs_btree_block, bb_u.s.bb_rightsib),
1472 offsetof(struct xfs_btree_block, bb_u.s.bb_blkno),
1473 offsetof(struct xfs_btree_block, bb_u.s.bb_lsn),
1474 offsetof(struct xfs_btree_block, bb_u.s.bb_uuid),
1475 offsetof(struct xfs_btree_block, bb_u.s.bb_owner),
1476 offsetof(struct xfs_btree_block, bb_u.s.bb_crc),
1477 XFS_BTREE_SBLOCK_CRC_LEN
1479 static const short loffsets[] = { /* table of offsets (long) */
1480 offsetof(struct xfs_btree_block, bb_magic),
1481 offsetof(struct xfs_btree_block, bb_level),
1482 offsetof(struct xfs_btree_block, bb_numrecs),
1483 offsetof(struct xfs_btree_block, bb_u.l.bb_leftsib),
1484 offsetof(struct xfs_btree_block, bb_u.l.bb_rightsib),
1485 offsetof(struct xfs_btree_block, bb_u.l.bb_blkno),
1486 offsetof(struct xfs_btree_block, bb_u.l.bb_lsn),
1487 offsetof(struct xfs_btree_block, bb_u.l.bb_uuid),
1488 offsetof(struct xfs_btree_block, bb_u.l.bb_owner),
1489 offsetof(struct xfs_btree_block, bb_u.l.bb_crc),
1490 offsetof(struct xfs_btree_block, bb_u.l.bb_pad),
1491 XFS_BTREE_LBLOCK_CRC_LEN
1497 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
1499 * We don't log the CRC when updating a btree
1500 * block but instead recreate it during log
1501 * recovery. As the log buffers have checksums
1502 * of their own this is safe and avoids logging a crc
1503 * update in a lot of places.
1505 if (fields == XFS_BB_ALL_BITS)
1506 fields = XFS_BB_ALL_BITS_CRC;
1507 nbits = XFS_BB_NUM_BITS_CRC;
1509 nbits = XFS_BB_NUM_BITS;
1511 xfs_btree_offsets(fields,
1512 (cur->bc_flags & XFS_BTREE_LONG_PTRS) ?
1513 loffsets : soffsets,
1514 nbits, &first, &last);
1515 xfs_trans_buf_set_type(cur->bc_tp, bp, XFS_BLFT_BTREE_BUF);
1516 xfs_trans_log_buf(cur->bc_tp, bp, first, last);
1518 xfs_trans_log_inode(cur->bc_tp, cur->bc_ino.ip,
1519 xfs_ilog_fbroot(cur->bc_ino.whichfork));
1524 * Increment cursor by one record at the level.
1525 * For nonzero levels the leaf-ward information is untouched.
1528 xfs_btree_increment(
1529 struct xfs_btree_cur *cur,
1531 int *stat) /* success/failure */
1533 struct xfs_btree_block *block;
1534 union xfs_btree_ptr ptr;
1536 int error; /* error return value */
1539 ASSERT(level < cur->bc_nlevels);
1541 /* Read-ahead to the right at this level. */
1542 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
1544 /* Get a pointer to the btree block. */
1545 block = xfs_btree_get_block(cur, level, &bp);
1548 error = xfs_btree_check_block(cur, block, level, bp);
1553 /* We're done if we remain in the block after the increment. */
1554 if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
1557 /* Fail if we just went off the right edge of the tree. */
1558 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1559 if (xfs_btree_ptr_is_null(cur, &ptr))
1562 XFS_BTREE_STATS_INC(cur, increment);
1565 * March up the tree incrementing pointers.
1566 * Stop when we don't go off the right edge of a block.
1568 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1569 block = xfs_btree_get_block(cur, lev, &bp);
1572 error = xfs_btree_check_block(cur, block, lev, bp);
1577 if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
1580 /* Read-ahead the right block for the next loop. */
1581 xfs_btree_readahead(cur, lev, XFS_BTCUR_RIGHTRA);
1585 * If we went off the root then we are either seriously
1586 * confused or have the tree root in an inode.
1588 if (lev == cur->bc_nlevels) {
1589 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1592 error = -EFSCORRUPTED;
1595 ASSERT(lev < cur->bc_nlevels);
1598 * Now walk back down the tree, fixing up the cursor's buffer
1599 * pointers and key numbers.
1601 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1602 union xfs_btree_ptr *ptrp;
1604 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1606 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1610 xfs_btree_setbuf(cur, lev, bp);
1611 cur->bc_ptrs[lev] = 1;
1626 * Decrement cursor by one record at the level.
1627 * For nonzero levels the leaf-ward information is untouched.
1630 xfs_btree_decrement(
1631 struct xfs_btree_cur *cur,
1633 int *stat) /* success/failure */
1635 struct xfs_btree_block *block;
1637 int error; /* error return value */
1639 union xfs_btree_ptr ptr;
1641 ASSERT(level < cur->bc_nlevels);
1643 /* Read-ahead to the left at this level. */
1644 xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
1646 /* We're done if we remain in the block after the decrement. */
1647 if (--cur->bc_ptrs[level] > 0)
1650 /* Get a pointer to the btree block. */
1651 block = xfs_btree_get_block(cur, level, &bp);
1654 error = xfs_btree_check_block(cur, block, level, bp);
1659 /* Fail if we just went off the left edge of the tree. */
1660 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
1661 if (xfs_btree_ptr_is_null(cur, &ptr))
1664 XFS_BTREE_STATS_INC(cur, decrement);
1667 * March up the tree decrementing pointers.
1668 * Stop when we don't go off the left edge of a block.
1670 for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
1671 if (--cur->bc_ptrs[lev] > 0)
1673 /* Read-ahead the left block for the next loop. */
1674 xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
1678 * If we went off the root then we are seriously confused.
1679 * or the root of the tree is in an inode.
1681 if (lev == cur->bc_nlevels) {
1682 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
1685 error = -EFSCORRUPTED;
1688 ASSERT(lev < cur->bc_nlevels);
1691 * Now walk back down the tree, fixing up the cursor's buffer
1692 * pointers and key numbers.
1694 for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
1695 union xfs_btree_ptr *ptrp;
1697 ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
1699 error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
1702 xfs_btree_setbuf(cur, lev, bp);
1703 cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
1718 xfs_btree_lookup_get_block(
1719 struct xfs_btree_cur *cur, /* btree cursor */
1720 int level, /* level in the btree */
1721 union xfs_btree_ptr *pp, /* ptr to btree block */
1722 struct xfs_btree_block **blkp) /* return btree block */
1724 struct xfs_buf *bp; /* buffer pointer for btree block */
1728 /* special case the root block if in an inode */
1729 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
1730 (level == cur->bc_nlevels - 1)) {
1731 *blkp = xfs_btree_get_iroot(cur);
1736 * If the old buffer at this level for the disk address we are
1737 * looking for re-use it.
1739 * Otherwise throw it away and get a new one.
1741 bp = cur->bc_bufs[level];
1742 error = xfs_btree_ptr_to_daddr(cur, pp, &daddr);
1745 if (bp && XFS_BUF_ADDR(bp) == daddr) {
1746 *blkp = XFS_BUF_TO_BLOCK(bp);
1750 error = xfs_btree_read_buf_block(cur, pp, 0, blkp, &bp);
1754 /* Check the inode owner since the verifiers don't. */
1755 if (xfs_sb_version_hascrc(&cur->bc_mp->m_sb) &&
1756 !(cur->bc_ino.flags & XFS_BTCUR_BMBT_INVALID_OWNER) &&
1757 (cur->bc_flags & XFS_BTREE_LONG_PTRS) &&
1758 be64_to_cpu((*blkp)->bb_u.l.bb_owner) !=
1759 cur->bc_ino.ip->i_ino)
1762 /* Did we get the level we were looking for? */
1763 if (be16_to_cpu((*blkp)->bb_level) != level)
1766 /* Check that internal nodes have at least one record. */
1767 if (level != 0 && be16_to_cpu((*blkp)->bb_numrecs) == 0)
1770 xfs_btree_setbuf(cur, level, bp);
1775 xfs_buf_mark_corrupt(bp);
1776 xfs_trans_brelse(cur->bc_tp, bp);
1777 return -EFSCORRUPTED;
1781 * Get current search key. For level 0 we don't actually have a key
1782 * structure so we make one up from the record. For all other levels
1783 * we just return the right key.
1785 STATIC union xfs_btree_key *
1786 xfs_lookup_get_search_key(
1787 struct xfs_btree_cur *cur,
1790 struct xfs_btree_block *block,
1791 union xfs_btree_key *kp)
1794 cur->bc_ops->init_key_from_rec(kp,
1795 xfs_btree_rec_addr(cur, keyno, block));
1799 return xfs_btree_key_addr(cur, keyno, block);
1803 * Lookup the record. The cursor is made to point to it, based on dir.
1804 * stat is set to 0 if can't find any such record, 1 for success.
1808 struct xfs_btree_cur *cur, /* btree cursor */
1809 xfs_lookup_t dir, /* <=, ==, or >= */
1810 int *stat) /* success/failure */
1812 struct xfs_btree_block *block; /* current btree block */
1813 int64_t diff; /* difference for the current key */
1814 int error; /* error return value */
1815 int keyno; /* current key number */
1816 int level; /* level in the btree */
1817 union xfs_btree_ptr *pp; /* ptr to btree block */
1818 union xfs_btree_ptr ptr; /* ptr to btree block */
1820 XFS_BTREE_STATS_INC(cur, lookup);
1822 /* No such thing as a zero-level tree. */
1823 if (XFS_IS_CORRUPT(cur->bc_mp, cur->bc_nlevels == 0))
1824 return -EFSCORRUPTED;
1829 /* initialise start pointer from cursor */
1830 cur->bc_ops->init_ptr_from_cur(cur, &ptr);
1834 * Iterate over each level in the btree, starting at the root.
1835 * For each level above the leaves, find the key we need, based
1836 * on the lookup record, then follow the corresponding block
1837 * pointer down to the next level.
1839 for (level = cur->bc_nlevels - 1, diff = 1; level >= 0; level--) {
1840 /* Get the block we need to do the lookup on. */
1841 error = xfs_btree_lookup_get_block(cur, level, pp, &block);
1847 * If we already had a key match at a higher level, we
1848 * know we need to use the first entry in this block.
1852 /* Otherwise search this block. Do a binary search. */
1854 int high; /* high entry number */
1855 int low; /* low entry number */
1857 /* Set low and high entry numbers, 1-based. */
1859 high = xfs_btree_get_numrecs(block);
1861 /* Block is empty, must be an empty leaf. */
1862 if (level != 0 || cur->bc_nlevels != 1) {
1863 XFS_CORRUPTION_ERROR(__func__,
1867 return -EFSCORRUPTED;
1870 cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
1875 /* Binary search the block. */
1876 while (low <= high) {
1877 union xfs_btree_key key;
1878 union xfs_btree_key *kp;
1880 XFS_BTREE_STATS_INC(cur, compare);
1882 /* keyno is average of low and high. */
1883 keyno = (low + high) >> 1;
1885 /* Get current search key */
1886 kp = xfs_lookup_get_search_key(cur, level,
1887 keyno, block, &key);
1890 * Compute difference to get next direction:
1891 * - less than, move right
1892 * - greater than, move left
1893 * - equal, we're done
1895 diff = cur->bc_ops->key_diff(cur, kp);
1906 * If there are more levels, set up for the next level
1907 * by getting the block number and filling in the cursor.
1911 * If we moved left, need the previous key number,
1912 * unless there isn't one.
1914 if (diff > 0 && --keyno < 1)
1916 pp = xfs_btree_ptr_addr(cur, keyno, block);
1918 error = xfs_btree_debug_check_ptr(cur, pp, 0, level);
1922 cur->bc_ptrs[level] = keyno;
1926 /* Done with the search. See if we need to adjust the results. */
1927 if (dir != XFS_LOOKUP_LE && diff < 0) {
1930 * If ge search and we went off the end of the block, but it's
1931 * not the last block, we're in the wrong block.
1933 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
1934 if (dir == XFS_LOOKUP_GE &&
1935 keyno > xfs_btree_get_numrecs(block) &&
1936 !xfs_btree_ptr_is_null(cur, &ptr)) {
1939 cur->bc_ptrs[0] = keyno;
1940 error = xfs_btree_increment(cur, 0, &i);
1943 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1944 return -EFSCORRUPTED;
1948 } else if (dir == XFS_LOOKUP_LE && diff > 0)
1950 cur->bc_ptrs[0] = keyno;
1952 /* Return if we succeeded or not. */
1953 if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
1955 else if (dir != XFS_LOOKUP_EQ || diff == 0)
1965 /* Find the high key storage area from a regular key. */
1966 union xfs_btree_key *
1967 xfs_btree_high_key_from_key(
1968 struct xfs_btree_cur *cur,
1969 union xfs_btree_key *key)
1971 ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
1972 return (union xfs_btree_key *)((char *)key +
1973 (cur->bc_ops->key_len / 2));
1976 /* Determine the low (and high if overlapped) keys of a leaf block */
1978 xfs_btree_get_leaf_keys(
1979 struct xfs_btree_cur *cur,
1980 struct xfs_btree_block *block,
1981 union xfs_btree_key *key)
1983 union xfs_btree_key max_hkey;
1984 union xfs_btree_key hkey;
1985 union xfs_btree_rec *rec;
1986 union xfs_btree_key *high;
1989 rec = xfs_btree_rec_addr(cur, 1, block);
1990 cur->bc_ops->init_key_from_rec(key, rec);
1992 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
1994 cur->bc_ops->init_high_key_from_rec(&max_hkey, rec);
1995 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
1996 rec = xfs_btree_rec_addr(cur, n, block);
1997 cur->bc_ops->init_high_key_from_rec(&hkey, rec);
1998 if (cur->bc_ops->diff_two_keys(cur, &hkey, &max_hkey)
2003 high = xfs_btree_high_key_from_key(cur, key);
2004 memcpy(high, &max_hkey, cur->bc_ops->key_len / 2);
2008 /* Determine the low (and high if overlapped) keys of a node block */
2010 xfs_btree_get_node_keys(
2011 struct xfs_btree_cur *cur,
2012 struct xfs_btree_block *block,
2013 union xfs_btree_key *key)
2015 union xfs_btree_key *hkey;
2016 union xfs_btree_key *max_hkey;
2017 union xfs_btree_key *high;
2020 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2021 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2022 cur->bc_ops->key_len / 2);
2024 max_hkey = xfs_btree_high_key_addr(cur, 1, block);
2025 for (n = 2; n <= xfs_btree_get_numrecs(block); n++) {
2026 hkey = xfs_btree_high_key_addr(cur, n, block);
2027 if (cur->bc_ops->diff_two_keys(cur, hkey, max_hkey) > 0)
2031 high = xfs_btree_high_key_from_key(cur, key);
2032 memcpy(high, max_hkey, cur->bc_ops->key_len / 2);
2034 memcpy(key, xfs_btree_key_addr(cur, 1, block),
2035 cur->bc_ops->key_len);
2039 /* Derive the keys for any btree block. */
2042 struct xfs_btree_cur *cur,
2043 struct xfs_btree_block *block,
2044 union xfs_btree_key *key)
2046 if (be16_to_cpu(block->bb_level) == 0)
2047 xfs_btree_get_leaf_keys(cur, block, key);
2049 xfs_btree_get_node_keys(cur, block, key);
2053 * Decide if we need to update the parent keys of a btree block. For
2054 * a standard btree this is only necessary if we're updating the first
2055 * record/key. For an overlapping btree, we must always update the
2056 * keys because the highest key can be in any of the records or keys
2060 xfs_btree_needs_key_update(
2061 struct xfs_btree_cur *cur,
2064 return (cur->bc_flags & XFS_BTREE_OVERLAPPING) || ptr == 1;
2068 * Update the low and high parent keys of the given level, progressing
2069 * towards the root. If force_all is false, stop if the keys for a given
2070 * level do not need updating.
2073 __xfs_btree_updkeys(
2074 struct xfs_btree_cur *cur,
2076 struct xfs_btree_block *block,
2077 struct xfs_buf *bp0,
2080 union xfs_btree_key key; /* keys from current level */
2081 union xfs_btree_key *lkey; /* keys from the next level up */
2082 union xfs_btree_key *hkey;
2083 union xfs_btree_key *nlkey; /* keys from the next level up */
2084 union xfs_btree_key *nhkey;
2088 ASSERT(cur->bc_flags & XFS_BTREE_OVERLAPPING);
2090 /* Exit if there aren't any parent levels to update. */
2091 if (level + 1 >= cur->bc_nlevels)
2094 trace_xfs_btree_updkeys(cur, level, bp0);
2097 hkey = xfs_btree_high_key_from_key(cur, lkey);
2098 xfs_btree_get_keys(cur, block, lkey);
2099 for (level++; level < cur->bc_nlevels; level++) {
2103 block = xfs_btree_get_block(cur, level, &bp);
2104 trace_xfs_btree_updkeys(cur, level, bp);
2106 error = xfs_btree_check_block(cur, block, level, bp);
2110 ptr = cur->bc_ptrs[level];
2111 nlkey = xfs_btree_key_addr(cur, ptr, block);
2112 nhkey = xfs_btree_high_key_addr(cur, ptr, block);
2114 !(cur->bc_ops->diff_two_keys(cur, nlkey, lkey) != 0 ||
2115 cur->bc_ops->diff_two_keys(cur, nhkey, hkey) != 0))
2117 xfs_btree_copy_keys(cur, nlkey, lkey, 1);
2118 xfs_btree_log_keys(cur, bp, ptr, ptr);
2119 if (level + 1 >= cur->bc_nlevels)
2121 xfs_btree_get_node_keys(cur, block, lkey);
2127 /* Update all the keys from some level in cursor back to the root. */
2129 xfs_btree_updkeys_force(
2130 struct xfs_btree_cur *cur,
2134 struct xfs_btree_block *block;
2136 block = xfs_btree_get_block(cur, level, &bp);
2137 return __xfs_btree_updkeys(cur, level, block, bp, true);
2141 * Update the parent keys of the given level, progressing towards the root.
2144 xfs_btree_update_keys(
2145 struct xfs_btree_cur *cur,
2148 struct xfs_btree_block *block;
2150 union xfs_btree_key *kp;
2151 union xfs_btree_key key;
2156 block = xfs_btree_get_block(cur, level, &bp);
2157 if (cur->bc_flags & XFS_BTREE_OVERLAPPING)
2158 return __xfs_btree_updkeys(cur, level, block, bp, false);
2161 * Go up the tree from this level toward the root.
2162 * At each level, update the key value to the value input.
2163 * Stop when we reach a level where the cursor isn't pointing
2164 * at the first entry in the block.
2166 xfs_btree_get_keys(cur, block, &key);
2167 for (level++, ptr = 1; ptr == 1 && level < cur->bc_nlevels; level++) {
2171 block = xfs_btree_get_block(cur, level, &bp);
2173 error = xfs_btree_check_block(cur, block, level, bp);
2177 ptr = cur->bc_ptrs[level];
2178 kp = xfs_btree_key_addr(cur, ptr, block);
2179 xfs_btree_copy_keys(cur, kp, &key, 1);
2180 xfs_btree_log_keys(cur, bp, ptr, ptr);
2187 * Update the record referred to by cur to the value in the
2188 * given record. This either works (return 0) or gets an
2189 * EFSCORRUPTED error.
2193 struct xfs_btree_cur *cur,
2194 union xfs_btree_rec *rec)
2196 struct xfs_btree_block *block;
2200 union xfs_btree_rec *rp;
2202 /* Pick up the current block. */
2203 block = xfs_btree_get_block(cur, 0, &bp);
2206 error = xfs_btree_check_block(cur, block, 0, bp);
2210 /* Get the address of the rec to be updated. */
2211 ptr = cur->bc_ptrs[0];
2212 rp = xfs_btree_rec_addr(cur, ptr, block);
2214 /* Fill in the new contents and log them. */
2215 xfs_btree_copy_recs(cur, rp, rec, 1);
2216 xfs_btree_log_recs(cur, bp, ptr, ptr);
2219 * If we are tracking the last record in the tree and
2220 * we are at the far right edge of the tree, update it.
2222 if (xfs_btree_is_lastrec(cur, block, 0)) {
2223 cur->bc_ops->update_lastrec(cur, block, rec,
2224 ptr, LASTREC_UPDATE);
2227 /* Pass new key value up to our parent. */
2228 if (xfs_btree_needs_key_update(cur, ptr)) {
2229 error = xfs_btree_update_keys(cur, 0);
2241 * Move 1 record left from cur/level if possible.
2242 * Update cur to reflect the new path.
2244 STATIC int /* error */
2246 struct xfs_btree_cur *cur,
2248 int *stat) /* success/failure */
2250 struct xfs_buf *lbp; /* left buffer pointer */
2251 struct xfs_btree_block *left; /* left btree block */
2252 int lrecs; /* left record count */
2253 struct xfs_buf *rbp; /* right buffer pointer */
2254 struct xfs_btree_block *right; /* right btree block */
2255 struct xfs_btree_cur *tcur; /* temporary btree cursor */
2256 int rrecs; /* right record count */
2257 union xfs_btree_ptr lptr; /* left btree pointer */
2258 union xfs_btree_key *rkp = NULL; /* right btree key */
2259 union xfs_btree_ptr *rpp = NULL; /* right address pointer */
2260 union xfs_btree_rec *rrp = NULL; /* right record pointer */
2261 int error; /* error return value */
2264 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2265 level == cur->bc_nlevels - 1)
2268 /* Set up variables for this block as "right". */
2269 right = xfs_btree_get_block(cur, level, &rbp);
2272 error = xfs_btree_check_block(cur, right, level, rbp);
2277 /* If we've got no left sibling then we can't shift an entry left. */
2278 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2279 if (xfs_btree_ptr_is_null(cur, &lptr))
2283 * If the cursor entry is the one that would be moved, don't
2284 * do it... it's too complicated.
2286 if (cur->bc_ptrs[level] <= 1)
2289 /* Set up the left neighbor as "left". */
2290 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
2294 /* If it's full, it can't take another entry. */
2295 lrecs = xfs_btree_get_numrecs(left);
2296 if (lrecs == cur->bc_ops->get_maxrecs(cur, level))
2299 rrecs = xfs_btree_get_numrecs(right);
2302 * We add one entry to the left side and remove one for the right side.
2303 * Account for it here, the changes will be updated on disk and logged
2309 XFS_BTREE_STATS_INC(cur, lshift);
2310 XFS_BTREE_STATS_ADD(cur, moves, 1);
2313 * If non-leaf, copy a key and a ptr to the left block.
2314 * Log the changes to the left block.
2317 /* It's a non-leaf. Move keys and pointers. */
2318 union xfs_btree_key *lkp; /* left btree key */
2319 union xfs_btree_ptr *lpp; /* left address pointer */
2321 lkp = xfs_btree_key_addr(cur, lrecs, left);
2322 rkp = xfs_btree_key_addr(cur, 1, right);
2324 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2325 rpp = xfs_btree_ptr_addr(cur, 1, right);
2327 error = xfs_btree_debug_check_ptr(cur, rpp, 0, level);
2331 xfs_btree_copy_keys(cur, lkp, rkp, 1);
2332 xfs_btree_copy_ptrs(cur, lpp, rpp, 1);
2334 xfs_btree_log_keys(cur, lbp, lrecs, lrecs);
2335 xfs_btree_log_ptrs(cur, lbp, lrecs, lrecs);
2337 ASSERT(cur->bc_ops->keys_inorder(cur,
2338 xfs_btree_key_addr(cur, lrecs - 1, left), lkp));
2340 /* It's a leaf. Move records. */
2341 union xfs_btree_rec *lrp; /* left record pointer */
2343 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2344 rrp = xfs_btree_rec_addr(cur, 1, right);
2346 xfs_btree_copy_recs(cur, lrp, rrp, 1);
2347 xfs_btree_log_recs(cur, lbp, lrecs, lrecs);
2349 ASSERT(cur->bc_ops->recs_inorder(cur,
2350 xfs_btree_rec_addr(cur, lrecs - 1, left), lrp));
2353 xfs_btree_set_numrecs(left, lrecs);
2354 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2356 xfs_btree_set_numrecs(right, rrecs);
2357 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2360 * Slide the contents of right down one entry.
2362 XFS_BTREE_STATS_ADD(cur, moves, rrecs - 1);
2364 /* It's a nonleaf. operate on keys and ptrs */
2365 for (i = 0; i < rrecs; i++) {
2366 error = xfs_btree_debug_check_ptr(cur, rpp, i + 1, level);
2371 xfs_btree_shift_keys(cur,
2372 xfs_btree_key_addr(cur, 2, right),
2374 xfs_btree_shift_ptrs(cur,
2375 xfs_btree_ptr_addr(cur, 2, right),
2378 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2379 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2381 /* It's a leaf. operate on records */
2382 xfs_btree_shift_recs(cur,
2383 xfs_btree_rec_addr(cur, 2, right),
2385 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2389 * Using a temporary cursor, update the parent key values of the
2390 * block on the left.
2392 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2393 error = xfs_btree_dup_cursor(cur, &tcur);
2396 i = xfs_btree_firstrec(tcur, level);
2397 if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2398 error = -EFSCORRUPTED;
2402 error = xfs_btree_decrement(tcur, level, &i);
2406 /* Update the parent high keys of the left block, if needed. */
2407 error = xfs_btree_update_keys(tcur, level);
2411 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2414 /* Update the parent keys of the right block. */
2415 error = xfs_btree_update_keys(cur, level);
2419 /* Slide the cursor value left one. */
2420 cur->bc_ptrs[level]--;
2433 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2438 * Move 1 record right from cur/level if possible.
2439 * Update cur to reflect the new path.
2441 STATIC int /* error */
2443 struct xfs_btree_cur *cur,
2445 int *stat) /* success/failure */
2447 struct xfs_buf *lbp; /* left buffer pointer */
2448 struct xfs_btree_block *left; /* left btree block */
2449 struct xfs_buf *rbp; /* right buffer pointer */
2450 struct xfs_btree_block *right; /* right btree block */
2451 struct xfs_btree_cur *tcur; /* temporary btree cursor */
2452 union xfs_btree_ptr rptr; /* right block pointer */
2453 union xfs_btree_key *rkp; /* right btree key */
2454 int rrecs; /* right record count */
2455 int lrecs; /* left record count */
2456 int error; /* error return value */
2457 int i; /* loop counter */
2459 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
2460 (level == cur->bc_nlevels - 1))
2463 /* Set up variables for this block as "left". */
2464 left = xfs_btree_get_block(cur, level, &lbp);
2467 error = xfs_btree_check_block(cur, left, level, lbp);
2472 /* If we've got no right sibling then we can't shift an entry right. */
2473 xfs_btree_get_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2474 if (xfs_btree_ptr_is_null(cur, &rptr))
2478 * If the cursor entry is the one that would be moved, don't
2479 * do it... it's too complicated.
2481 lrecs = xfs_btree_get_numrecs(left);
2482 if (cur->bc_ptrs[level] >= lrecs)
2485 /* Set up the right neighbor as "right". */
2486 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
2490 /* If it's full, it can't take another entry. */
2491 rrecs = xfs_btree_get_numrecs(right);
2492 if (rrecs == cur->bc_ops->get_maxrecs(cur, level))
2495 XFS_BTREE_STATS_INC(cur, rshift);
2496 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2499 * Make a hole at the start of the right neighbor block, then
2500 * copy the last left block entry to the hole.
2503 /* It's a nonleaf. make a hole in the keys and ptrs */
2504 union xfs_btree_key *lkp;
2505 union xfs_btree_ptr *lpp;
2506 union xfs_btree_ptr *rpp;
2508 lkp = xfs_btree_key_addr(cur, lrecs, left);
2509 lpp = xfs_btree_ptr_addr(cur, lrecs, left);
2510 rkp = xfs_btree_key_addr(cur, 1, right);
2511 rpp = xfs_btree_ptr_addr(cur, 1, right);
2513 for (i = rrecs - 1; i >= 0; i--) {
2514 error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
2519 xfs_btree_shift_keys(cur, rkp, 1, rrecs);
2520 xfs_btree_shift_ptrs(cur, rpp, 1, rrecs);
2522 error = xfs_btree_debug_check_ptr(cur, lpp, 0, level);
2526 /* Now put the new data in, and log it. */
2527 xfs_btree_copy_keys(cur, rkp, lkp, 1);
2528 xfs_btree_copy_ptrs(cur, rpp, lpp, 1);
2530 xfs_btree_log_keys(cur, rbp, 1, rrecs + 1);
2531 xfs_btree_log_ptrs(cur, rbp, 1, rrecs + 1);
2533 ASSERT(cur->bc_ops->keys_inorder(cur, rkp,
2534 xfs_btree_key_addr(cur, 2, right)));
2536 /* It's a leaf. make a hole in the records */
2537 union xfs_btree_rec *lrp;
2538 union xfs_btree_rec *rrp;
2540 lrp = xfs_btree_rec_addr(cur, lrecs, left);
2541 rrp = xfs_btree_rec_addr(cur, 1, right);
2543 xfs_btree_shift_recs(cur, rrp, 1, rrecs);
2545 /* Now put the new data in, and log it. */
2546 xfs_btree_copy_recs(cur, rrp, lrp, 1);
2547 xfs_btree_log_recs(cur, rbp, 1, rrecs + 1);
2551 * Decrement and log left's numrecs, bump and log right's numrecs.
2553 xfs_btree_set_numrecs(left, --lrecs);
2554 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS);
2556 xfs_btree_set_numrecs(right, ++rrecs);
2557 xfs_btree_log_block(cur, rbp, XFS_BB_NUMRECS);
2560 * Using a temporary cursor, update the parent key values of the
2561 * block on the right.
2563 error = xfs_btree_dup_cursor(cur, &tcur);
2566 i = xfs_btree_lastrec(tcur, level);
2567 if (XFS_IS_CORRUPT(tcur->bc_mp, i != 1)) {
2568 error = -EFSCORRUPTED;
2572 error = xfs_btree_increment(tcur, level, &i);
2576 /* Update the parent high keys of the left block, if needed. */
2577 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2578 error = xfs_btree_update_keys(cur, level);
2583 /* Update the parent keys of the right block. */
2584 error = xfs_btree_update_keys(tcur, level);
2588 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
2601 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
2606 * Split cur/level block in half.
2607 * Return new block number and the key to its first
2608 * record (to be inserted into parent).
2610 STATIC int /* error */
2612 struct xfs_btree_cur *cur,
2614 union xfs_btree_ptr *ptrp,
2615 union xfs_btree_key *key,
2616 struct xfs_btree_cur **curp,
2617 int *stat) /* success/failure */
2619 union xfs_btree_ptr lptr; /* left sibling block ptr */
2620 struct xfs_buf *lbp; /* left buffer pointer */
2621 struct xfs_btree_block *left; /* left btree block */
2622 union xfs_btree_ptr rptr; /* right sibling block ptr */
2623 struct xfs_buf *rbp; /* right buffer pointer */
2624 struct xfs_btree_block *right; /* right btree block */
2625 union xfs_btree_ptr rrptr; /* right-right sibling ptr */
2626 struct xfs_buf *rrbp; /* right-right buffer pointer */
2627 struct xfs_btree_block *rrblock; /* right-right btree block */
2631 int error; /* error return value */
2634 XFS_BTREE_STATS_INC(cur, split);
2636 /* Set up left block (current one). */
2637 left = xfs_btree_get_block(cur, level, &lbp);
2640 error = xfs_btree_check_block(cur, left, level, lbp);
2645 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
2647 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2648 error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, stat);
2653 XFS_BTREE_STATS_INC(cur, alloc);
2655 /* Set up the new block as "right". */
2656 error = xfs_btree_get_buf_block(cur, &rptr, &right, &rbp);
2660 /* Fill in the btree header for the new right block. */
2661 xfs_btree_init_block_cur(cur, rbp, xfs_btree_get_level(left), 0);
2664 * Split the entries between the old and the new block evenly.
2665 * Make sure that if there's an odd number of entries now, that
2666 * each new block will have the same number of entries.
2668 lrecs = xfs_btree_get_numrecs(left);
2670 if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
2672 src_index = (lrecs - rrecs + 1);
2674 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
2676 /* Adjust numrecs for the later get_*_keys() calls. */
2678 xfs_btree_set_numrecs(left, lrecs);
2679 xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
2682 * Copy btree block entries from the left block over to the
2683 * new block, the right. Update the right block and log the
2687 /* It's a non-leaf. Move keys and pointers. */
2688 union xfs_btree_key *lkp; /* left btree key */
2689 union xfs_btree_ptr *lpp; /* left address pointer */
2690 union xfs_btree_key *rkp; /* right btree key */
2691 union xfs_btree_ptr *rpp; /* right address pointer */
2693 lkp = xfs_btree_key_addr(cur, src_index, left);
2694 lpp = xfs_btree_ptr_addr(cur, src_index, left);
2695 rkp = xfs_btree_key_addr(cur, 1, right);
2696 rpp = xfs_btree_ptr_addr(cur, 1, right);
2698 for (i = src_index; i < rrecs; i++) {
2699 error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
2704 /* Copy the keys & pointers to the new block. */
2705 xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
2706 xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
2708 xfs_btree_log_keys(cur, rbp, 1, rrecs);
2709 xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
2711 /* Stash the keys of the new block for later insertion. */
2712 xfs_btree_get_node_keys(cur, right, key);
2714 /* It's a leaf. Move records. */
2715 union xfs_btree_rec *lrp; /* left record pointer */
2716 union xfs_btree_rec *rrp; /* right record pointer */
2718 lrp = xfs_btree_rec_addr(cur, src_index, left);
2719 rrp = xfs_btree_rec_addr(cur, 1, right);
2721 /* Copy records to the new block. */
2722 xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
2723 xfs_btree_log_recs(cur, rbp, 1, rrecs);
2725 /* Stash the keys of the new block for later insertion. */
2726 xfs_btree_get_leaf_keys(cur, right, key);
2730 * Find the left block number by looking in the buffer.
2731 * Adjust sibling pointers.
2733 xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
2734 xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
2735 xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
2736 xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
2738 xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
2739 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
2742 * If there's a block to the new block's right, make that block
2743 * point back to right instead of to left.
2745 if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
2746 error = xfs_btree_read_buf_block(cur, &rrptr,
2747 0, &rrblock, &rrbp);
2750 xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
2751 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
2754 /* Update the parent high keys of the left block, if needed. */
2755 if (cur->bc_flags & XFS_BTREE_OVERLAPPING) {
2756 error = xfs_btree_update_keys(cur, level);
2762 * If the cursor is really in the right block, move it there.
2763 * If it's just pointing past the last entry in left, then we'll
2764 * insert there, so don't change anything in that case.
2766 if (cur->bc_ptrs[level] > lrecs + 1) {
2767 xfs_btree_setbuf(cur, level, rbp);
2768 cur->bc_ptrs[level] -= lrecs;
2771 * If there are more levels, we'll need another cursor which refers
2772 * the right block, no matter where this cursor was.
2774 if (level + 1 < cur->bc_nlevels) {
2775 error = xfs_btree_dup_cursor(cur, curp);
2778 (*curp)->bc_ptrs[level + 1]++;
2791 struct xfs_btree_split_args {
2792 struct xfs_btree_cur *cur;
2794 union xfs_btree_ptr *ptrp;
2795 union xfs_btree_key *key;
2796 struct xfs_btree_cur **curp;
2797 int *stat; /* success/failure */
2799 bool kswapd; /* allocation in kswapd context */
2800 struct completion *done;
2801 struct work_struct work;
2805 * Stack switching interfaces for allocation
2808 xfs_btree_split_worker(
2809 struct work_struct *work)
2811 struct xfs_btree_split_args *args = container_of(work,
2812 struct xfs_btree_split_args, work);
2813 unsigned long pflags;
2814 unsigned long new_pflags = 0;
2817 * we are in a transaction context here, but may also be doing work
2818 * in kswapd context, and hence we may need to inherit that state
2819 * temporarily to ensure that we don't block waiting for memory reclaim
2823 new_pflags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
2825 current_set_flags_nested(&pflags, new_pflags);
2826 xfs_trans_set_context(args->cur->bc_tp);
2828 args->result = __xfs_btree_split(args->cur, args->level, args->ptrp,
2829 args->key, args->curp, args->stat);
2831 xfs_trans_clear_context(args->cur->bc_tp);
2832 current_restore_flags_nested(&pflags, new_pflags);
2835 * Do not access args after complete() has run here. We don't own args
2836 * and the owner may run and free args before we return here.
2838 complete(args->done);
2843 * BMBT split requests often come in with little stack to work on. Push
2844 * them off to a worker thread so there is lots of stack to use. For the other
2845 * btree types, just call directly to avoid the context switch overhead here.
2847 STATIC int /* error */
2849 struct xfs_btree_cur *cur,
2851 union xfs_btree_ptr *ptrp,
2852 union xfs_btree_key *key,
2853 struct xfs_btree_cur **curp,
2854 int *stat) /* success/failure */
2856 struct xfs_btree_split_args args;
2857 DECLARE_COMPLETION_ONSTACK(done);
2859 if (cur->bc_btnum != XFS_BTNUM_BMAP)
2860 return __xfs_btree_split(cur, level, ptrp, key, curp, stat);
2869 args.kswapd = current_is_kswapd();
2870 INIT_WORK_ONSTACK(&args.work, xfs_btree_split_worker);
2871 queue_work(xfs_alloc_wq, &args.work);
2872 wait_for_completion(&done);
2873 destroy_work_on_stack(&args.work);
2879 * Copy the old inode root contents into a real block and make the
2880 * broot point to it.
2883 xfs_btree_new_iroot(
2884 struct xfs_btree_cur *cur, /* btree cursor */
2885 int *logflags, /* logging flags for inode */
2886 int *stat) /* return status - 0 fail */
2888 struct xfs_buf *cbp; /* buffer for cblock */
2889 struct xfs_btree_block *block; /* btree block */
2890 struct xfs_btree_block *cblock; /* child btree block */
2891 union xfs_btree_key *ckp; /* child key pointer */
2892 union xfs_btree_ptr *cpp; /* child ptr pointer */
2893 union xfs_btree_key *kp; /* pointer to btree key */
2894 union xfs_btree_ptr *pp; /* pointer to block addr */
2895 union xfs_btree_ptr nptr; /* new block addr */
2896 int level; /* btree level */
2897 int error; /* error return code */
2898 int i; /* loop counter */
2900 XFS_BTREE_STATS_INC(cur, newroot);
2902 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
2904 level = cur->bc_nlevels - 1;
2906 block = xfs_btree_get_iroot(cur);
2907 pp = xfs_btree_ptr_addr(cur, 1, block);
2909 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2910 error = cur->bc_ops->alloc_block(cur, pp, &nptr, stat);
2916 XFS_BTREE_STATS_INC(cur, alloc);
2918 /* Copy the root into a real block. */
2919 error = xfs_btree_get_buf_block(cur, &nptr, &cblock, &cbp);
2924 * we can't just memcpy() the root in for CRC enabled btree blocks.
2925 * In that case have to also ensure the blkno remains correct
2927 memcpy(cblock, block, xfs_btree_block_len(cur));
2928 if (cur->bc_flags & XFS_BTREE_CRC_BLOCKS) {
2929 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
2930 cblock->bb_u.l.bb_blkno = cpu_to_be64(cbp->b_bn);
2932 cblock->bb_u.s.bb_blkno = cpu_to_be64(cbp->b_bn);
2935 be16_add_cpu(&block->bb_level, 1);
2936 xfs_btree_set_numrecs(block, 1);
2938 cur->bc_ptrs[level + 1] = 1;
2940 kp = xfs_btree_key_addr(cur, 1, block);
2941 ckp = xfs_btree_key_addr(cur, 1, cblock);
2942 xfs_btree_copy_keys(cur, ckp, kp, xfs_btree_get_numrecs(cblock));
2944 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
2945 for (i = 0; i < be16_to_cpu(cblock->bb_numrecs); i++) {
2946 error = xfs_btree_debug_check_ptr(cur, pp, i, level);
2951 xfs_btree_copy_ptrs(cur, cpp, pp, xfs_btree_get_numrecs(cblock));
2953 error = xfs_btree_debug_check_ptr(cur, &nptr, 0, level);
2957 xfs_btree_copy_ptrs(cur, pp, &nptr, 1);
2959 xfs_iroot_realloc(cur->bc_ino.ip,
2960 1 - xfs_btree_get_numrecs(cblock),
2961 cur->bc_ino.whichfork);
2963 xfs_btree_setbuf(cur, level, cbp);
2966 * Do all this logging at the end so that
2967 * the root is at the right level.
2969 xfs_btree_log_block(cur, cbp, XFS_BB_ALL_BITS);
2970 xfs_btree_log_keys(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2971 xfs_btree_log_ptrs(cur, cbp, 1, be16_to_cpu(cblock->bb_numrecs));
2974 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork);
2982 * Allocate a new root block, fill it in.
2984 STATIC int /* error */
2986 struct xfs_btree_cur *cur, /* btree cursor */
2987 int *stat) /* success/failure */
2989 struct xfs_btree_block *block; /* one half of the old root block */
2990 struct xfs_buf *bp; /* buffer containing block */
2991 int error; /* error return value */
2992 struct xfs_buf *lbp; /* left buffer pointer */
2993 struct xfs_btree_block *left; /* left btree block */
2994 struct xfs_buf *nbp; /* new (root) buffer */
2995 struct xfs_btree_block *new; /* new (root) btree block */
2996 int nptr; /* new value for key index, 1 or 2 */
2997 struct xfs_buf *rbp; /* right buffer pointer */
2998 struct xfs_btree_block *right; /* right btree block */
2999 union xfs_btree_ptr rptr;
3000 union xfs_btree_ptr lptr;
3002 XFS_BTREE_STATS_INC(cur, newroot);
3004 /* initialise our start point from the cursor */
3005 cur->bc_ops->init_ptr_from_cur(cur, &rptr);
3007 /* Allocate the new block. If we can't do it, we're toast. Give up. */
3008 error = cur->bc_ops->alloc_block(cur, &rptr, &lptr, stat);
3013 XFS_BTREE_STATS_INC(cur, alloc);
3015 /* Set up the new block. */
3016 error = xfs_btree_get_buf_block(cur, &lptr, &new, &nbp);
3020 /* Set the root in the holding structure increasing the level by 1. */
3021 cur->bc_ops->set_root(cur, &lptr, 1);
3024 * At the previous root level there are now two blocks: the old root,
3025 * and the new block generated when it was split. We don't know which
3026 * one the cursor is pointing at, so we set up variables "left" and
3027 * "right" for each case.
3029 block = xfs_btree_get_block(cur, cur->bc_nlevels - 1, &bp);
3032 error = xfs_btree_check_block(cur, block, cur->bc_nlevels - 1, bp);
3037 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3038 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3039 /* Our block is left, pick up the right block. */
3041 xfs_btree_buf_to_ptr(cur, lbp, &lptr);
3043 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
3049 /* Our block is right, pick up the left block. */
3051 xfs_btree_buf_to_ptr(cur, rbp, &rptr);
3053 xfs_btree_get_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
3054 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
3061 /* Fill in the new block's btree header and log it. */
3062 xfs_btree_init_block_cur(cur, nbp, cur->bc_nlevels, 2);
3063 xfs_btree_log_block(cur, nbp, XFS_BB_ALL_BITS);
3064 ASSERT(!xfs_btree_ptr_is_null(cur, &lptr) &&
3065 !xfs_btree_ptr_is_null(cur, &rptr));
3067 /* Fill in the key data in the new root. */
3068 if (xfs_btree_get_level(left) > 0) {
3070 * Get the keys for the left block's keys and put them directly
3071 * in the parent block. Do the same for the right block.
3073 xfs_btree_get_node_keys(cur, left,
3074 xfs_btree_key_addr(cur, 1, new));
3075 xfs_btree_get_node_keys(cur, right,
3076 xfs_btree_key_addr(cur, 2, new));
3079 * Get the keys for the left block's records and put them
3080 * directly in the parent block. Do the same for the right
3083 xfs_btree_get_leaf_keys(cur, left,
3084 xfs_btree_key_addr(cur, 1, new));
3085 xfs_btree_get_leaf_keys(cur, right,
3086 xfs_btree_key_addr(cur, 2, new));
3088 xfs_btree_log_keys(cur, nbp, 1, 2);
3090 /* Fill in the pointer data in the new root. */
3091 xfs_btree_copy_ptrs(cur,
3092 xfs_btree_ptr_addr(cur, 1, new), &lptr, 1);
3093 xfs_btree_copy_ptrs(cur,
3094 xfs_btree_ptr_addr(cur, 2, new), &rptr, 1);
3095 xfs_btree_log_ptrs(cur, nbp, 1, 2);
3097 /* Fix up the cursor. */
3098 xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
3099 cur->bc_ptrs[cur->bc_nlevels] = nptr;
3111 xfs_btree_make_block_unfull(
3112 struct xfs_btree_cur *cur, /* btree cursor */
3113 int level, /* btree level */
3114 int numrecs,/* # of recs in block */
3115 int *oindex,/* old tree index */
3116 int *index, /* new tree index */
3117 union xfs_btree_ptr *nptr, /* new btree ptr */
3118 struct xfs_btree_cur **ncur, /* new btree cursor */
3119 union xfs_btree_key *key, /* key of new block */
3124 if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3125 level == cur->bc_nlevels - 1) {
3126 struct xfs_inode *ip = cur->bc_ino.ip;
3128 if (numrecs < cur->bc_ops->get_dmaxrecs(cur, level)) {
3129 /* A root block that can be made bigger. */
3130 xfs_iroot_realloc(ip, 1, cur->bc_ino.whichfork);
3133 /* A root block that needs replacing */
3136 error = xfs_btree_new_iroot(cur, &logflags, stat);
3137 if (error || *stat == 0)
3140 xfs_trans_log_inode(cur->bc_tp, ip, logflags);
3146 /* First, try shifting an entry to the right neighbor. */
3147 error = xfs_btree_rshift(cur, level, stat);
3151 /* Next, try shifting an entry to the left neighbor. */
3152 error = xfs_btree_lshift(cur, level, stat);
3157 *oindex = *index = cur->bc_ptrs[level];
3162 * Next, try splitting the current block in half.
3164 * If this works we have to re-set our variables because we
3165 * could be in a different block now.
3167 error = xfs_btree_split(cur, level, nptr, key, ncur, stat);
3168 if (error || *stat == 0)
3172 *index = cur->bc_ptrs[level];
3177 * Insert one record/level. Return information to the caller
3178 * allowing the next level up to proceed if necessary.
3182 struct xfs_btree_cur *cur, /* btree cursor */
3183 int level, /* level to insert record at */
3184 union xfs_btree_ptr *ptrp, /* i/o: block number inserted */
3185 union xfs_btree_rec *rec, /* record to insert */
3186 union xfs_btree_key *key, /* i/o: block key for ptrp */
3187 struct xfs_btree_cur **curp, /* output: new cursor replacing cur */
3188 int *stat) /* success/failure */
3190 struct xfs_btree_block *block; /* btree block */
3191 struct xfs_buf *bp; /* buffer for block */
3192 union xfs_btree_ptr nptr; /* new block ptr */
3193 struct xfs_btree_cur *ncur; /* new btree cursor */
3194 union xfs_btree_key nkey; /* new block key */
3195 union xfs_btree_key *lkey;
3196 int optr; /* old key/record index */
3197 int ptr; /* key/record index */
3198 int numrecs;/* number of records */
3199 int error; /* error return value */
3207 * If we have an external root pointer, and we've made it to the
3208 * root level, allocate a new root block and we're done.
3210 if (!(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) &&
3211 (level >= cur->bc_nlevels)) {
3212 error = xfs_btree_new_root(cur, stat);
3213 xfs_btree_set_ptr_null(cur, ptrp);
3218 /* If we're off the left edge, return failure. */
3219 ptr = cur->bc_ptrs[level];
3227 XFS_BTREE_STATS_INC(cur, insrec);
3229 /* Get pointers to the btree buffer and block. */
3230 block = xfs_btree_get_block(cur, level, &bp);
3231 old_bn = bp ? bp->b_bn : XFS_BUF_DADDR_NULL;
3232 numrecs = xfs_btree_get_numrecs(block);
3235 error = xfs_btree_check_block(cur, block, level, bp);
3239 /* Check that the new entry is being inserted in the right place. */
3240 if (ptr <= numrecs) {
3242 ASSERT(cur->bc_ops->recs_inorder(cur, rec,
3243 xfs_btree_rec_addr(cur, ptr, block)));
3245 ASSERT(cur->bc_ops->keys_inorder(cur, key,
3246 xfs_btree_key_addr(cur, ptr, block)));
3252 * If the block is full, we can't insert the new entry until we
3253 * make the block un-full.
3255 xfs_btree_set_ptr_null(cur, &nptr);
3256 if (numrecs == cur->bc_ops->get_maxrecs(cur, level)) {
3257 error = xfs_btree_make_block_unfull(cur, level, numrecs,
3258 &optr, &ptr, &nptr, &ncur, lkey, stat);
3259 if (error || *stat == 0)
3264 * The current block may have changed if the block was
3265 * previously full and we have just made space in it.
3267 block = xfs_btree_get_block(cur, level, &bp);
3268 numrecs = xfs_btree_get_numrecs(block);
3271 error = xfs_btree_check_block(cur, block, level, bp);
3277 * At this point we know there's room for our new entry in the block
3278 * we're pointing at.
3280 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr + 1);
3283 /* It's a nonleaf. make a hole in the keys and ptrs */
3284 union xfs_btree_key *kp;
3285 union xfs_btree_ptr *pp;
3287 kp = xfs_btree_key_addr(cur, ptr, block);
3288 pp = xfs_btree_ptr_addr(cur, ptr, block);
3290 for (i = numrecs - ptr; i >= 0; i--) {
3291 error = xfs_btree_debug_check_ptr(cur, pp, i, level);
3296 xfs_btree_shift_keys(cur, kp, 1, numrecs - ptr + 1);
3297 xfs_btree_shift_ptrs(cur, pp, 1, numrecs - ptr + 1);
3299 error = xfs_btree_debug_check_ptr(cur, ptrp, 0, level);
3303 /* Now put the new data in, bump numrecs and log it. */
3304 xfs_btree_copy_keys(cur, kp, key, 1);
3305 xfs_btree_copy_ptrs(cur, pp, ptrp, 1);
3307 xfs_btree_set_numrecs(block, numrecs);
3308 xfs_btree_log_ptrs(cur, bp, ptr, numrecs);
3309 xfs_btree_log_keys(cur, bp, ptr, numrecs);
3311 if (ptr < numrecs) {
3312 ASSERT(cur->bc_ops->keys_inorder(cur, kp,
3313 xfs_btree_key_addr(cur, ptr + 1, block)));
3317 /* It's a leaf. make a hole in the records */
3318 union xfs_btree_rec *rp;
3320 rp = xfs_btree_rec_addr(cur, ptr, block);
3322 xfs_btree_shift_recs(cur, rp, 1, numrecs - ptr + 1);
3324 /* Now put the new data in, bump numrecs and log it. */
3325 xfs_btree_copy_recs(cur, rp, rec, 1);
3326 xfs_btree_set_numrecs(block, ++numrecs);
3327 xfs_btree_log_recs(cur, bp, ptr, numrecs);
3329 if (ptr < numrecs) {
3330 ASSERT(cur->bc_ops->recs_inorder(cur, rp,
3331 xfs_btree_rec_addr(cur, ptr + 1, block)));
3336 /* Log the new number of records in the btree header. */
3337 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3340 * If we just inserted into a new tree block, we have to
3341 * recalculate nkey here because nkey is out of date.
3343 * Otherwise we're just updating an existing block (having shoved
3344 * some records into the new tree block), so use the regular key
3347 if (bp && bp->b_bn != old_bn) {
3348 xfs_btree_get_keys(cur, block, lkey);
3349 } else if (xfs_btree_needs_key_update(cur, optr)) {
3350 error = xfs_btree_update_keys(cur, level);
3356 * If we are tracking the last record in the tree and
3357 * we are at the far right edge of the tree, update it.
3359 if (xfs_btree_is_lastrec(cur, block, level)) {
3360 cur->bc_ops->update_lastrec(cur, block, rec,
3361 ptr, LASTREC_INSREC);
3365 * Return the new block number, if any.
3366 * If there is one, give back a record value and a cursor too.
3369 if (!xfs_btree_ptr_is_null(cur, &nptr)) {
3370 xfs_btree_copy_keys(cur, key, lkey, 1);
3382 * Insert the record at the point referenced by cur.
3384 * A multi-level split of the tree on insert will invalidate the original
3385 * cursor. All callers of this function should assume that the cursor is
3386 * no longer valid and revalidate it.
3390 struct xfs_btree_cur *cur,
3393 int error; /* error return value */
3394 int i; /* result value, 0 for failure */
3395 int level; /* current level number in btree */
3396 union xfs_btree_ptr nptr; /* new block number (split result) */
3397 struct xfs_btree_cur *ncur; /* new cursor (split result) */
3398 struct xfs_btree_cur *pcur; /* previous level's cursor */
3399 union xfs_btree_key bkey; /* key of block to insert */
3400 union xfs_btree_key *key;
3401 union xfs_btree_rec rec; /* record to insert */
3408 xfs_btree_set_ptr_null(cur, &nptr);
3410 /* Make a key out of the record data to be inserted, and save it. */
3411 cur->bc_ops->init_rec_from_cur(cur, &rec);
3412 cur->bc_ops->init_key_from_rec(key, &rec);
3415 * Loop going up the tree, starting at the leaf level.
3416 * Stop when we don't get a split block, that must mean that
3417 * the insert is finished with this level.
3421 * Insert nrec/nptr into this level of the tree.
3422 * Note if we fail, nptr will be null.
3424 error = xfs_btree_insrec(pcur, level, &nptr, &rec, key,
3428 xfs_btree_del_cursor(pcur, XFS_BTREE_ERROR);
3432 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3433 error = -EFSCORRUPTED;
3439 * See if the cursor we just used is trash.
3440 * Can't trash the caller's cursor, but otherwise we should
3441 * if ncur is a new cursor or we're about to be done.
3444 (ncur || xfs_btree_ptr_is_null(cur, &nptr))) {
3445 /* Save the state from the cursor before we trash it */
3446 if (cur->bc_ops->update_cursor)
3447 cur->bc_ops->update_cursor(pcur, cur);
3448 cur->bc_nlevels = pcur->bc_nlevels;
3449 xfs_btree_del_cursor(pcur, XFS_BTREE_NOERROR);
3451 /* If we got a new cursor, switch to it. */
3456 } while (!xfs_btree_ptr_is_null(cur, &nptr));
3465 * Try to merge a non-leaf block back into the inode root.
3467 * Note: the killroot names comes from the fact that we're effectively
3468 * killing the old root block. But because we can't just delete the
3469 * inode we have to copy the single block it was pointing to into the
3473 xfs_btree_kill_iroot(
3474 struct xfs_btree_cur *cur)
3476 int whichfork = cur->bc_ino.whichfork;
3477 struct xfs_inode *ip = cur->bc_ino.ip;
3478 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
3479 struct xfs_btree_block *block;
3480 struct xfs_btree_block *cblock;
3481 union xfs_btree_key *kp;
3482 union xfs_btree_key *ckp;
3483 union xfs_btree_ptr *pp;
3484 union xfs_btree_ptr *cpp;
3485 struct xfs_buf *cbp;
3491 union xfs_btree_ptr ptr;
3495 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
3496 ASSERT(cur->bc_nlevels > 1);
3499 * Don't deal with the root block needs to be a leaf case.
3500 * We're just going to turn the thing back into extents anyway.
3502 level = cur->bc_nlevels - 1;
3507 * Give up if the root has multiple children.
3509 block = xfs_btree_get_iroot(cur);
3510 if (xfs_btree_get_numrecs(block) != 1)
3513 cblock = xfs_btree_get_block(cur, level - 1, &cbp);
3514 numrecs = xfs_btree_get_numrecs(cblock);
3517 * Only do this if the next level will fit.
3518 * Then the data must be copied up to the inode,
3519 * instead of freeing the root you free the next level.
3521 if (numrecs > cur->bc_ops->get_dmaxrecs(cur, level))
3524 XFS_BTREE_STATS_INC(cur, killroot);
3527 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_LEFTSIB);
3528 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3529 xfs_btree_get_sibling(cur, block, &ptr, XFS_BB_RIGHTSIB);
3530 ASSERT(xfs_btree_ptr_is_null(cur, &ptr));
3533 index = numrecs - cur->bc_ops->get_maxrecs(cur, level);
3535 xfs_iroot_realloc(cur->bc_ino.ip, index,
3536 cur->bc_ino.whichfork);
3537 block = ifp->if_broot;
3540 be16_add_cpu(&block->bb_numrecs, index);
3541 ASSERT(block->bb_numrecs == cblock->bb_numrecs);
3543 kp = xfs_btree_key_addr(cur, 1, block);
3544 ckp = xfs_btree_key_addr(cur, 1, cblock);
3545 xfs_btree_copy_keys(cur, kp, ckp, numrecs);
3547 pp = xfs_btree_ptr_addr(cur, 1, block);
3548 cpp = xfs_btree_ptr_addr(cur, 1, cblock);
3550 for (i = 0; i < numrecs; i++) {
3551 error = xfs_btree_debug_check_ptr(cur, cpp, i, level - 1);
3556 xfs_btree_copy_ptrs(cur, pp, cpp, numrecs);
3558 error = xfs_btree_free_block(cur, cbp);
3562 cur->bc_bufs[level - 1] = NULL;
3563 be16_add_cpu(&block->bb_level, -1);
3564 xfs_trans_log_inode(cur->bc_tp, ip,
3565 XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork));
3572 * Kill the current root node, and replace it with it's only child node.
3575 xfs_btree_kill_root(
3576 struct xfs_btree_cur *cur,
3579 union xfs_btree_ptr *newroot)
3583 XFS_BTREE_STATS_INC(cur, killroot);
3586 * Update the root pointer, decreasing the level by 1 and then
3587 * free the old root.
3589 cur->bc_ops->set_root(cur, newroot, -1);
3591 error = xfs_btree_free_block(cur, bp);
3595 cur->bc_bufs[level] = NULL;
3596 cur->bc_ra[level] = 0;
3603 xfs_btree_dec_cursor(
3604 struct xfs_btree_cur *cur,
3612 error = xfs_btree_decrement(cur, level, &i);
3622 * Single level of the btree record deletion routine.
3623 * Delete record pointed to by cur/level.
3624 * Remove the record from its block then rebalance the tree.
3625 * Return 0 for error, 1 for done, 2 to go on to the next level.
3627 STATIC int /* error */
3629 struct xfs_btree_cur *cur, /* btree cursor */
3630 int level, /* level removing record from */
3631 int *stat) /* fail/done/go-on */
3633 struct xfs_btree_block *block; /* btree block */
3634 union xfs_btree_ptr cptr; /* current block ptr */
3635 struct xfs_buf *bp; /* buffer for block */
3636 int error; /* error return value */
3637 int i; /* loop counter */
3638 union xfs_btree_ptr lptr; /* left sibling block ptr */
3639 struct xfs_buf *lbp; /* left buffer pointer */
3640 struct xfs_btree_block *left; /* left btree block */
3641 int lrecs = 0; /* left record count */
3642 int ptr; /* key/record index */
3643 union xfs_btree_ptr rptr; /* right sibling block ptr */
3644 struct xfs_buf *rbp; /* right buffer pointer */
3645 struct xfs_btree_block *right; /* right btree block */
3646 struct xfs_btree_block *rrblock; /* right-right btree block */
3647 struct xfs_buf *rrbp; /* right-right buffer pointer */
3648 int rrecs = 0; /* right record count */
3649 struct xfs_btree_cur *tcur; /* temporary btree cursor */
3650 int numrecs; /* temporary numrec count */
3654 /* Get the index of the entry being deleted, check for nothing there. */
3655 ptr = cur->bc_ptrs[level];
3661 /* Get the buffer & block containing the record or key/ptr. */
3662 block = xfs_btree_get_block(cur, level, &bp);
3663 numrecs = xfs_btree_get_numrecs(block);
3666 error = xfs_btree_check_block(cur, block, level, bp);
3671 /* Fail if we're off the end of the block. */
3672 if (ptr > numrecs) {
3677 XFS_BTREE_STATS_INC(cur, delrec);
3678 XFS_BTREE_STATS_ADD(cur, moves, numrecs - ptr);
3680 /* Excise the entries being deleted. */
3682 /* It's a nonleaf. operate on keys and ptrs */
3683 union xfs_btree_key *lkp;
3684 union xfs_btree_ptr *lpp;
3686 lkp = xfs_btree_key_addr(cur, ptr + 1, block);
3687 lpp = xfs_btree_ptr_addr(cur, ptr + 1, block);
3689 for (i = 0; i < numrecs - ptr; i++) {
3690 error = xfs_btree_debug_check_ptr(cur, lpp, i, level);
3695 if (ptr < numrecs) {
3696 xfs_btree_shift_keys(cur, lkp, -1, numrecs - ptr);
3697 xfs_btree_shift_ptrs(cur, lpp, -1, numrecs - ptr);
3698 xfs_btree_log_keys(cur, bp, ptr, numrecs - 1);
3699 xfs_btree_log_ptrs(cur, bp, ptr, numrecs - 1);
3702 /* It's a leaf. operate on records */
3703 if (ptr < numrecs) {
3704 xfs_btree_shift_recs(cur,
3705 xfs_btree_rec_addr(cur, ptr + 1, block),
3707 xfs_btree_log_recs(cur, bp, ptr, numrecs - 1);
3712 * Decrement and log the number of entries in the block.
3714 xfs_btree_set_numrecs(block, --numrecs);
3715 xfs_btree_log_block(cur, bp, XFS_BB_NUMRECS);
3718 * If we are tracking the last record in the tree and
3719 * we are at the far right edge of the tree, update it.
3721 if (xfs_btree_is_lastrec(cur, block, level)) {
3722 cur->bc_ops->update_lastrec(cur, block, NULL,
3723 ptr, LASTREC_DELREC);
3727 * We're at the root level. First, shrink the root block in-memory.
3728 * Try to get rid of the next level down. If we can't then there's
3729 * nothing left to do.
3731 if (level == cur->bc_nlevels - 1) {
3732 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3733 xfs_iroot_realloc(cur->bc_ino.ip, -1,
3734 cur->bc_ino.whichfork);
3736 error = xfs_btree_kill_iroot(cur);
3740 error = xfs_btree_dec_cursor(cur, level, stat);
3748 * If this is the root level, and there's only one entry left,
3749 * and it's NOT the leaf level, then we can get rid of this
3752 if (numrecs == 1 && level > 0) {
3753 union xfs_btree_ptr *pp;
3755 * pp is still set to the first pointer in the block.
3756 * Make it the new root of the btree.
3758 pp = xfs_btree_ptr_addr(cur, 1, block);
3759 error = xfs_btree_kill_root(cur, bp, level, pp);
3762 } else if (level > 0) {
3763 error = xfs_btree_dec_cursor(cur, level, stat);
3772 * If we deleted the leftmost entry in the block, update the
3773 * key values above us in the tree.
3775 if (xfs_btree_needs_key_update(cur, ptr)) {
3776 error = xfs_btree_update_keys(cur, level);
3782 * If the number of records remaining in the block is at least
3783 * the minimum, we're done.
3785 if (numrecs >= cur->bc_ops->get_minrecs(cur, level)) {
3786 error = xfs_btree_dec_cursor(cur, level, stat);
3793 * Otherwise, we have to move some records around to keep the
3794 * tree balanced. Look at the left and right sibling blocks to
3795 * see if we can re-balance by moving only one record.
3797 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
3798 xfs_btree_get_sibling(cur, block, &lptr, XFS_BB_LEFTSIB);
3800 if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) {
3802 * One child of root, need to get a chance to copy its contents
3803 * into the root and delete it. Can't go up to next level,
3804 * there's nothing to delete there.
3806 if (xfs_btree_ptr_is_null(cur, &rptr) &&
3807 xfs_btree_ptr_is_null(cur, &lptr) &&
3808 level == cur->bc_nlevels - 2) {
3809 error = xfs_btree_kill_iroot(cur);
3811 error = xfs_btree_dec_cursor(cur, level, stat);
3818 ASSERT(!xfs_btree_ptr_is_null(cur, &rptr) ||
3819 !xfs_btree_ptr_is_null(cur, &lptr));
3822 * Duplicate the cursor so our btree manipulations here won't
3823 * disrupt the next level up.
3825 error = xfs_btree_dup_cursor(cur, &tcur);
3830 * If there's a right sibling, see if it's ok to shift an entry
3833 if (!xfs_btree_ptr_is_null(cur, &rptr)) {
3835 * Move the temp cursor to the last entry in the next block.
3836 * Actually any entry but the first would suffice.
3838 i = xfs_btree_lastrec(tcur, level);
3839 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3840 error = -EFSCORRUPTED;
3844 error = xfs_btree_increment(tcur, level, &i);
3847 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3848 error = -EFSCORRUPTED;
3852 i = xfs_btree_lastrec(tcur, level);
3853 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3854 error = -EFSCORRUPTED;
3858 /* Grab a pointer to the block. */
3859 right = xfs_btree_get_block(tcur, level, &rbp);
3861 error = xfs_btree_check_block(tcur, right, level, rbp);
3865 /* Grab the current block number, for future use. */
3866 xfs_btree_get_sibling(tcur, right, &cptr, XFS_BB_LEFTSIB);
3869 * If right block is full enough so that removing one entry
3870 * won't make it too empty, and left-shifting an entry out
3871 * of right to us works, we're done.
3873 if (xfs_btree_get_numrecs(right) - 1 >=
3874 cur->bc_ops->get_minrecs(tcur, level)) {
3875 error = xfs_btree_lshift(tcur, level, &i);
3879 ASSERT(xfs_btree_get_numrecs(block) >=
3880 cur->bc_ops->get_minrecs(tcur, level));
3882 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3885 error = xfs_btree_dec_cursor(cur, level, stat);
3893 * Otherwise, grab the number of records in right for
3894 * future reference, and fix up the temp cursor to point
3895 * to our block again (last record).
3897 rrecs = xfs_btree_get_numrecs(right);
3898 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3899 i = xfs_btree_firstrec(tcur, level);
3900 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3901 error = -EFSCORRUPTED;
3905 error = xfs_btree_decrement(tcur, level, &i);
3908 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3909 error = -EFSCORRUPTED;
3916 * If there's a left sibling, see if it's ok to shift an entry
3919 if (!xfs_btree_ptr_is_null(cur, &lptr)) {
3921 * Move the temp cursor to the first entry in the
3924 i = xfs_btree_firstrec(tcur, level);
3925 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3926 error = -EFSCORRUPTED;
3930 error = xfs_btree_decrement(tcur, level, &i);
3933 i = xfs_btree_firstrec(tcur, level);
3934 if (XFS_IS_CORRUPT(cur->bc_mp, i != 1)) {
3935 error = -EFSCORRUPTED;
3939 /* Grab a pointer to the block. */
3940 left = xfs_btree_get_block(tcur, level, &lbp);
3942 error = xfs_btree_check_block(cur, left, level, lbp);
3946 /* Grab the current block number, for future use. */
3947 xfs_btree_get_sibling(tcur, left, &cptr, XFS_BB_RIGHTSIB);
3950 * If left block is full enough so that removing one entry
3951 * won't make it too empty, and right-shifting an entry out
3952 * of left to us works, we're done.
3954 if (xfs_btree_get_numrecs(left) - 1 >=
3955 cur->bc_ops->get_minrecs(tcur, level)) {
3956 error = xfs_btree_rshift(tcur, level, &i);
3960 ASSERT(xfs_btree_get_numrecs(block) >=
3961 cur->bc_ops->get_minrecs(tcur, level));
3962 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3973 * Otherwise, grab the number of records in right for
3976 lrecs = xfs_btree_get_numrecs(left);
3979 /* Delete the temp cursor, we're done with it. */
3980 xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
3983 /* If here, we need to do a join to keep the tree balanced. */
3984 ASSERT(!xfs_btree_ptr_is_null(cur, &cptr));
3986 if (!xfs_btree_ptr_is_null(cur, &lptr) &&
3987 lrecs + xfs_btree_get_numrecs(block) <=
3988 cur->bc_ops->get_maxrecs(cur, level)) {
3990 * Set "right" to be the starting block,
3991 * "left" to be the left neighbor.
3996 error = xfs_btree_read_buf_block(cur, &lptr, 0, &left, &lbp);
4001 * If that won't work, see if we can join with the right neighbor block.
4003 } else if (!xfs_btree_ptr_is_null(cur, &rptr) &&
4004 rrecs + xfs_btree_get_numrecs(block) <=
4005 cur->bc_ops->get_maxrecs(cur, level)) {
4007 * Set "left" to be the starting block,
4008 * "right" to be the right neighbor.
4013 error = xfs_btree_read_buf_block(cur, &rptr, 0, &right, &rbp);
4018 * Otherwise, we can't fix the imbalance.
4019 * Just return. This is probably a logic error, but it's not fatal.
4022 error = xfs_btree_dec_cursor(cur, level, stat);
4028 rrecs = xfs_btree_get_numrecs(right);
4029 lrecs = xfs_btree_get_numrecs(left);
4032 * We're now going to join "left" and "right" by moving all the stuff
4033 * in "right" to "left" and deleting "right".
4035 XFS_BTREE_STATS_ADD(cur, moves, rrecs);
4037 /* It's a non-leaf. Move keys and pointers. */
4038 union xfs_btree_key *lkp; /* left btree key */
4039 union xfs_btree_ptr *lpp; /* left address pointer */
4040 union xfs_btree_key *rkp; /* right btree key */
4041 union xfs_btree_ptr *rpp; /* right address pointer */
4043 lkp = xfs_btree_key_addr(cur, lrecs + 1, left);
4044 lpp = xfs_btree_ptr_addr(cur, lrecs + 1, left);
4045 rkp = xfs_btree_key_addr(cur, 1, right);
4046 rpp = xfs_btree_ptr_addr(cur, 1, right);
4048 for (i = 1; i < rrecs; i++) {
4049 error = xfs_btree_debug_check_ptr(cur, rpp, i, level);
4054 xfs_btree_copy_keys(cur, lkp, rkp, rrecs);
4055 xfs_btree_copy_ptrs(cur, lpp, rpp, rrecs);
4057 xfs_btree_log_keys(cur, lbp, lrecs + 1, lrecs + rrecs);
4058 xfs_btree_log_ptrs(cur, lbp, lrecs + 1, lrecs + rrecs);
4060 /* It's a leaf. Move records. */
4061 union xfs_btree_rec *lrp; /* left record pointer */
4062 union xfs_btree_rec *rrp; /* right record pointer */
4064 lrp = xfs_btree_rec_addr(cur, lrecs + 1, left);
4065 rrp = xfs_btree_rec_addr(cur, 1, right);
4067 xfs_btree_copy_recs(cur, lrp, rrp, rrecs);
4068 xfs_btree_log_recs(cur, lbp, lrecs + 1, lrecs + rrecs);
4071 XFS_BTREE_STATS_INC(cur, join);
4074 * Fix up the number of records and right block pointer in the
4075 * surviving block, and log it.
4077 xfs_btree_set_numrecs(left, lrecs + rrecs);
4078 xfs_btree_get_sibling(cur, right, &cptr, XFS_BB_RIGHTSIB),
4079 xfs_btree_set_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4080 xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
4082 /* If there is a right sibling, point it to the remaining block. */
4083 xfs_btree_get_sibling(cur, left, &cptr, XFS_BB_RIGHTSIB);
4084 if (!xfs_btree_ptr_is_null(cur, &cptr)) {
4085 error = xfs_btree_read_buf_block(cur, &cptr, 0, &rrblock, &rrbp);
4088 xfs_btree_set_sibling(cur, rrblock, &lptr, XFS_BB_LEFTSIB);
4089 xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
4092 /* Free the deleted block. */
4093 error = xfs_btree_free_block(cur, rbp);
4098 * If we joined with the left neighbor, set the buffer in the
4099 * cursor to the left block, and fix up the index.
4102 cur->bc_bufs[level] = lbp;
4103 cur->bc_ptrs[level] += lrecs;
4104 cur->bc_ra[level] = 0;
4107 * If we joined with the right neighbor and there's a level above
4108 * us, increment the cursor at that level.
4110 else if ((cur->bc_flags & XFS_BTREE_ROOT_IN_INODE) ||
4111 (level + 1 < cur->bc_nlevels)) {
4112 error = xfs_btree_increment(cur, level + 1, &i);
4118 * Readjust the ptr at this level if it's not a leaf, since it's
4119 * still pointing at the deletion point, which makes the cursor
4120 * inconsistent. If this makes the ptr 0, the caller fixes it up.
4121 * We can't use decrement because it would change the next level up.
4124 cur->bc_ptrs[level]--;
4127 * We combined blocks, so we have to update the parent keys if the
4128 * btree supports overlapped intervals. However, bc_ptrs[level + 1]
4129 * points to the old block so that the caller knows which record to
4130 * delete. Therefore, the caller must be savvy enough to call updkeys
4131 * for us if we return stat == 2. The other exit points from this
4132 * function don't require deletions further up the tree, so they can
4133 * call updkeys directly.
4136 /* Return value means the next level up has something to do. */
4142 xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
4147 * Delete the record pointed to by cur.
4148 * The cursor refers to the place where the record was (could be inserted)
4149 * when the operation returns.
4153 struct xfs_btree_cur *cur,
4154 int *stat) /* success/failure */
4156 int error; /* error return value */
4159 bool joined = false;
4162 * Go up the tree, starting at leaf level.
4164 * If 2 is returned then a join was done; go to the next level.
4165 * Otherwise we are done.
4167 for (level = 0, i = 2; i == 2; level++) {
4168 error = xfs_btree_delrec(cur, level, &i);
4176 * If we combined blocks as part of deleting the record, delrec won't
4177 * have updated the parent high keys so we have to do that here.
4179 if (joined && (cur->bc_flags & XFS_BTREE_OVERLAPPING)) {
4180 error = xfs_btree_updkeys_force(cur, 0);
4186 for (level = 1; level < cur->bc_nlevels; level++) {
4187 if (cur->bc_ptrs[level] == 0) {
4188 error = xfs_btree_decrement(cur, level, &i);
4203 * Get the data from the pointed-to record.
4207 struct xfs_btree_cur *cur, /* btree cursor */
4208 union xfs_btree_rec **recp, /* output: btree record */
4209 int *stat) /* output: success/failure */
4211 struct xfs_btree_block *block; /* btree block */
4212 struct xfs_buf *bp; /* buffer pointer */
4213 int ptr; /* record number */
4215 int error; /* error return value */
4218 ptr = cur->bc_ptrs[0];
4219 block = xfs_btree_get_block(cur, 0, &bp);
4222 error = xfs_btree_check_block(cur, block, 0, bp);
4228 * Off the right end or left end, return failure.
4230 if (ptr > xfs_btree_get_numrecs(block) || ptr <= 0) {
4236 * Point to the record and extract its data.
4238 *recp = xfs_btree_rec_addr(cur, ptr, block);
4243 /* Visit a block in a btree. */
4245 xfs_btree_visit_block(
4246 struct xfs_btree_cur *cur,
4248 xfs_btree_visit_blocks_fn fn,
4251 struct xfs_btree_block *block;
4253 union xfs_btree_ptr rptr;
4256 /* do right sibling readahead */
4257 xfs_btree_readahead(cur, level, XFS_BTCUR_RIGHTRA);
4258 block = xfs_btree_get_block(cur, level, &bp);
4260 /* process the block */
4261 error = fn(cur, level, data);
4265 /* now read rh sibling block for next iteration */
4266 xfs_btree_get_sibling(cur, block, &rptr, XFS_BB_RIGHTSIB);
4267 if (xfs_btree_ptr_is_null(cur, &rptr))
4270 return xfs_btree_lookup_get_block(cur, level, &rptr, &block);
4274 /* Visit every block in a btree. */
4276 xfs_btree_visit_blocks(
4277 struct xfs_btree_cur *cur,
4278 xfs_btree_visit_blocks_fn fn,
4282 union xfs_btree_ptr lptr;
4284 struct xfs_btree_block *block = NULL;
4287 cur->bc_ops->init_ptr_from_cur(cur, &lptr);
4289 /* for each level */
4290 for (level = cur->bc_nlevels - 1; level >= 0; level--) {
4291 /* grab the left hand block */
4292 error = xfs_btree_lookup_get_block(cur, level, &lptr, &block);
4296 /* readahead the left most block for the next level down */
4298 union xfs_btree_ptr *ptr;
4300 ptr = xfs_btree_ptr_addr(cur, 1, block);
4301 xfs_btree_readahead_ptr(cur, ptr, 1);
4303 /* save for the next iteration of the loop */
4304 xfs_btree_copy_ptrs(cur, &lptr, ptr, 1);
4306 if (!(flags & XFS_BTREE_VISIT_LEAVES))
4308 } else if (!(flags & XFS_BTREE_VISIT_RECORDS)) {
4312 /* for each buffer in the level */
4314 error = xfs_btree_visit_block(cur, level, fn, data);
4317 if (error != -ENOENT)
4325 * Change the owner of a btree.
4327 * The mechanism we use here is ordered buffer logging. Because we don't know
4328 * how many buffers were are going to need to modify, we don't really want to
4329 * have to make transaction reservations for the worst case of every buffer in a
4330 * full size btree as that may be more space that we can fit in the log....
4332 * We do the btree walk in the most optimal manner possible - we have sibling
4333 * pointers so we can just walk all the blocks on each level from left to right
4334 * in a single pass, and then move to the next level and do the same. We can
4335 * also do readahead on the sibling pointers to get IO moving more quickly,
4336 * though for slow disks this is unlikely to make much difference to performance
4337 * as the amount of CPU work we have to do before moving to the next block is
4340 * For each btree block that we load, modify the owner appropriately, set the
4341 * buffer as an ordered buffer and log it appropriately. We need to ensure that
4342 * we mark the region we change dirty so that if the buffer is relogged in
4343 * a subsequent transaction the changes we make here as an ordered buffer are
4344 * correctly relogged in that transaction. If we are in recovery context, then
4345 * just queue the modified buffer as delayed write buffer so the transaction
4346 * recovery completion writes the changes to disk.
4348 struct xfs_btree_block_change_owner_info {
4350 struct list_head *buffer_list;
4354 xfs_btree_block_change_owner(
4355 struct xfs_btree_cur *cur,
4359 struct xfs_btree_block_change_owner_info *bbcoi = data;
4360 struct xfs_btree_block *block;
4363 /* modify the owner */
4364 block = xfs_btree_get_block(cur, level, &bp);
4365 if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
4366 if (block->bb_u.l.bb_owner == cpu_to_be64(bbcoi->new_owner))
4368 block->bb_u.l.bb_owner = cpu_to_be64(bbcoi->new_owner);
4370 if (block->bb_u.s.bb_owner == cpu_to_be32(bbcoi->new_owner))
4372 block->bb_u.s.bb_owner = cpu_to_be32(bbcoi->new_owner);
4376 * If the block is a root block hosted in an inode, we might not have a
4377 * buffer pointer here and we shouldn't attempt to log the change as the
4378 * information is already held in the inode and discarded when the root
4379 * block is formatted into the on-disk inode fork. We still change it,
4380 * though, so everything is consistent in memory.
4383 ASSERT(cur->bc_flags & XFS_BTREE_ROOT_IN_INODE);
4384 ASSERT(level == cur->bc_nlevels - 1);
4389 if (!xfs_trans_ordered_buf(cur->bc_tp, bp)) {
4390 xfs_btree_log_block(cur, bp, XFS_BB_OWNER);
4394 xfs_buf_delwri_queue(bp, bbcoi->buffer_list);
4401 xfs_btree_change_owner(
4402 struct xfs_btree_cur *cur,
4404 struct list_head *buffer_list)
4406 struct xfs_btree_block_change_owner_info bbcoi;
4408 bbcoi.new_owner = new_owner;
4409 bbcoi.buffer_list = buffer_list;
4411 return xfs_btree_visit_blocks(cur, xfs_btree_block_change_owner,
4412 XFS_BTREE_VISIT_ALL, &bbcoi);
4415 /* Verify the v5 fields of a long-format btree block. */
4417 xfs_btree_lblock_v5hdr_verify(
4421 struct xfs_mount *mp = bp->b_mount;
4422 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4424 if (!xfs_sb_version_hascrc(&mp->m_sb))
4425 return __this_address;
4426 if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
4427 return __this_address;
4428 if (block->bb_u.l.bb_blkno != cpu_to_be64(bp->b_bn))
4429 return __this_address;
4430 if (owner != XFS_RMAP_OWN_UNKNOWN &&
4431 be64_to_cpu(block->bb_u.l.bb_owner) != owner)
4432 return __this_address;
4436 /* Verify a long-format btree block. */
4438 xfs_btree_lblock_verify(
4440 unsigned int max_recs)
4442 struct xfs_mount *mp = bp->b_mount;
4443 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4445 /* numrecs verification */
4446 if (be16_to_cpu(block->bb_numrecs) > max_recs)
4447 return __this_address;
4449 /* sibling pointer verification */
4450 if (block->bb_u.l.bb_leftsib != cpu_to_be64(NULLFSBLOCK) &&
4451 !xfs_verify_fsbno(mp, be64_to_cpu(block->bb_u.l.bb_leftsib)))
4452 return __this_address;
4453 if (block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK) &&
4454 !xfs_verify_fsbno(mp, be64_to_cpu(block->bb_u.l.bb_rightsib)))
4455 return __this_address;
4461 * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format
4464 * @bp: buffer containing the btree block
4467 xfs_btree_sblock_v5hdr_verify(
4470 struct xfs_mount *mp = bp->b_mount;
4471 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4472 struct xfs_perag *pag = bp->b_pag;
4474 if (!xfs_sb_version_hascrc(&mp->m_sb))
4475 return __this_address;
4476 if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
4477 return __this_address;
4478 if (block->bb_u.s.bb_blkno != cpu_to_be64(bp->b_bn))
4479 return __this_address;
4480 if (pag && be32_to_cpu(block->bb_u.s.bb_owner) != pag->pag_agno)
4481 return __this_address;
4486 * xfs_btree_sblock_verify() -- verify a short-format btree block
4488 * @bp: buffer containing the btree block
4489 * @max_recs: maximum records allowed in this btree node
4492 xfs_btree_sblock_verify(
4494 unsigned int max_recs)
4496 struct xfs_mount *mp = bp->b_mount;
4497 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
4500 /* numrecs verification */
4501 if (be16_to_cpu(block->bb_numrecs) > max_recs)
4502 return __this_address;
4504 /* sibling pointer verification */
4505 agno = xfs_daddr_to_agno(mp, XFS_BUF_ADDR(bp));
4506 if (block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK) &&
4507 !xfs_verify_agbno(mp, agno, be32_to_cpu(block->bb_u.s.bb_leftsib)))
4508 return __this_address;
4509 if (block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK) &&
4510 !xfs_verify_agbno(mp, agno, be32_to_cpu(block->bb_u.s.bb_rightsib)))
4511 return __this_address;
4517 * Calculate the number of btree levels needed to store a given number of
4518 * records in a short-format btree.
4521 xfs_btree_compute_maxlevels(
4526 unsigned long maxblocks;
4528 maxblocks = (len + limits[0] - 1) / limits[0];
4529 for (level = 1; maxblocks > 1; level++)
4530 maxblocks = (maxblocks + limits[1] - 1) / limits[1];
4535 * Query a regular btree for all records overlapping a given interval.
4536 * Start with a LE lookup of the key of low_rec and return all records
4537 * until we find a record with a key greater than the key of high_rec.
4540 xfs_btree_simple_query_range(
4541 struct xfs_btree_cur *cur,
4542 union xfs_btree_key *low_key,
4543 union xfs_btree_key *high_key,
4544 xfs_btree_query_range_fn fn,
4547 union xfs_btree_rec *recp;
4548 union xfs_btree_key rec_key;
4551 bool firstrec = true;
4554 ASSERT(cur->bc_ops->init_high_key_from_rec);
4555 ASSERT(cur->bc_ops->diff_two_keys);
4558 * Find the leftmost record. The btree cursor must be set
4559 * to the low record used to generate low_key.
4562 error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
4566 /* Nothing? See if there's anything to the right. */
4568 error = xfs_btree_increment(cur, 0, &stat);
4574 /* Find the record. */
4575 error = xfs_btree_get_rec(cur, &recp, &stat);
4579 /* Skip if high_key(rec) < low_key. */
4581 cur->bc_ops->init_high_key_from_rec(&rec_key, recp);
4583 diff = cur->bc_ops->diff_two_keys(cur, low_key,
4589 /* Stop if high_key < low_key(rec). */
4590 cur->bc_ops->init_key_from_rec(&rec_key, recp);
4591 diff = cur->bc_ops->diff_two_keys(cur, &rec_key, high_key);
4596 error = fn(cur, recp, priv);
4601 /* Move on to the next record. */
4602 error = xfs_btree_increment(cur, 0, &stat);
4612 * Query an overlapped interval btree for all records overlapping a given
4613 * interval. This function roughly follows the algorithm given in
4614 * "Interval Trees" of _Introduction to Algorithms_, which is section
4615 * 14.3 in the 2nd and 3rd editions.
4617 * First, generate keys for the low and high records passed in.
4619 * For any leaf node, generate the high and low keys for the record.
4620 * If the record keys overlap with the query low/high keys, pass the
4621 * record to the function iterator.
4623 * For any internal node, compare the low and high keys of each
4624 * pointer against the query low/high keys. If there's an overlap,
4625 * follow the pointer.
4627 * As an optimization, we stop scanning a block when we find a low key
4628 * that is greater than the query's high key.
4631 xfs_btree_overlapped_query_range(
4632 struct xfs_btree_cur *cur,
4633 union xfs_btree_key *low_key,
4634 union xfs_btree_key *high_key,
4635 xfs_btree_query_range_fn fn,
4638 union xfs_btree_ptr ptr;
4639 union xfs_btree_ptr *pp;
4640 union xfs_btree_key rec_key;
4641 union xfs_btree_key rec_hkey;
4642 union xfs_btree_key *lkp;
4643 union xfs_btree_key *hkp;
4644 union xfs_btree_rec *recp;
4645 struct xfs_btree_block *block;
4653 /* Load the root of the btree. */
4654 level = cur->bc_nlevels - 1;
4655 cur->bc_ops->init_ptr_from_cur(cur, &ptr);
4656 error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
4659 xfs_btree_get_block(cur, level, &bp);
4660 trace_xfs_btree_overlapped_query_range(cur, level, bp);
4662 error = xfs_btree_check_block(cur, block, level, bp);
4666 cur->bc_ptrs[level] = 1;
4668 while (level < cur->bc_nlevels) {
4669 block = xfs_btree_get_block(cur, level, &bp);
4671 /* End of node, pop back towards the root. */
4672 if (cur->bc_ptrs[level] > be16_to_cpu(block->bb_numrecs)) {
4674 if (level < cur->bc_nlevels - 1)
4675 cur->bc_ptrs[level + 1]++;
4681 /* Handle a leaf node. */
4682 recp = xfs_btree_rec_addr(cur, cur->bc_ptrs[0], block);
4684 cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
4685 ldiff = cur->bc_ops->diff_two_keys(cur, &rec_hkey,
4688 cur->bc_ops->init_key_from_rec(&rec_key, recp);
4689 hdiff = cur->bc_ops->diff_two_keys(cur, high_key,
4693 * If (record's high key >= query's low key) and
4694 * (query's high key >= record's low key), then
4695 * this record overlaps the query range; callback.
4697 if (ldiff >= 0 && hdiff >= 0) {
4698 error = fn(cur, recp, priv);
4701 } else if (hdiff < 0) {
4702 /* Record is larger than high key; pop. */
4705 cur->bc_ptrs[level]++;
4709 /* Handle an internal node. */
4710 lkp = xfs_btree_key_addr(cur, cur->bc_ptrs[level], block);
4711 hkp = xfs_btree_high_key_addr(cur, cur->bc_ptrs[level], block);
4712 pp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[level], block);
4714 ldiff = cur->bc_ops->diff_two_keys(cur, hkp, low_key);
4715 hdiff = cur->bc_ops->diff_two_keys(cur, high_key, lkp);
4718 * If (pointer's high key >= query's low key) and
4719 * (query's high key >= pointer's low key), then
4720 * this record overlaps the query range; follow pointer.
4722 if (ldiff >= 0 && hdiff >= 0) {
4724 error = xfs_btree_lookup_get_block(cur, level, pp,
4728 xfs_btree_get_block(cur, level, &bp);
4729 trace_xfs_btree_overlapped_query_range(cur, level, bp);
4731 error = xfs_btree_check_block(cur, block, level, bp);
4735 cur->bc_ptrs[level] = 1;
4737 } else if (hdiff < 0) {
4738 /* The low key is larger than the upper range; pop. */
4741 cur->bc_ptrs[level]++;
4746 * If we don't end this function with the cursor pointing at a record
4747 * block, a subsequent non-error cursor deletion will not release
4748 * node-level buffers, causing a buffer leak. This is quite possible
4749 * with a zero-results range query, so release the buffers if we
4750 * failed to return any results.
4752 if (cur->bc_bufs[0] == NULL) {
4753 for (i = 0; i < cur->bc_nlevels; i++) {
4754 if (cur->bc_bufs[i]) {
4755 xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
4756 cur->bc_bufs[i] = NULL;
4757 cur->bc_ptrs[i] = 0;
4767 * Query a btree for all records overlapping a given interval of keys. The
4768 * supplied function will be called with each record found; return one of the
4769 * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
4770 * code. This function returns -ECANCELED, zero, or a negative error code.
4773 xfs_btree_query_range(
4774 struct xfs_btree_cur *cur,
4775 union xfs_btree_irec *low_rec,
4776 union xfs_btree_irec *high_rec,
4777 xfs_btree_query_range_fn fn,
4780 union xfs_btree_rec rec;
4781 union xfs_btree_key low_key;
4782 union xfs_btree_key high_key;
4784 /* Find the keys of both ends of the interval. */
4785 cur->bc_rec = *high_rec;
4786 cur->bc_ops->init_rec_from_cur(cur, &rec);
4787 cur->bc_ops->init_key_from_rec(&high_key, &rec);
4789 cur->bc_rec = *low_rec;
4790 cur->bc_ops->init_rec_from_cur(cur, &rec);
4791 cur->bc_ops->init_key_from_rec(&low_key, &rec);
4793 /* Enforce low key < high key. */
4794 if (cur->bc_ops->diff_two_keys(cur, &low_key, &high_key) > 0)
4797 if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING))
4798 return xfs_btree_simple_query_range(cur, &low_key,
4799 &high_key, fn, priv);
4800 return xfs_btree_overlapped_query_range(cur, &low_key, &high_key,
4804 /* Query a btree for all records. */
4806 xfs_btree_query_all(
4807 struct xfs_btree_cur *cur,
4808 xfs_btree_query_range_fn fn,
4811 union xfs_btree_key low_key;
4812 union xfs_btree_key high_key;
4814 memset(&cur->bc_rec, 0, sizeof(cur->bc_rec));
4815 memset(&low_key, 0, sizeof(low_key));
4816 memset(&high_key, 0xFF, sizeof(high_key));
4818 return xfs_btree_simple_query_range(cur, &low_key, &high_key, fn, priv);
4822 * Calculate the number of blocks needed to store a given number of records
4823 * in a short-format (per-AG metadata) btree.
4826 xfs_btree_calc_size(
4828 unsigned long long len)
4832 unsigned long long rval;
4834 maxrecs = limits[0];
4835 for (level = 0, rval = 0; len > 1; level++) {
4837 do_div(len, maxrecs);
4838 maxrecs = limits[1];
4845 xfs_btree_count_blocks_helper(
4846 struct xfs_btree_cur *cur,
4850 xfs_extlen_t *blocks = data;
4856 /* Count the blocks in a btree and return the result in *blocks. */
4858 xfs_btree_count_blocks(
4859 struct xfs_btree_cur *cur,
4860 xfs_extlen_t *blocks)
4863 return xfs_btree_visit_blocks(cur, xfs_btree_count_blocks_helper,
4864 XFS_BTREE_VISIT_ALL, blocks);
4867 /* Compare two btree pointers. */
4869 xfs_btree_diff_two_ptrs(
4870 struct xfs_btree_cur *cur,
4871 const union xfs_btree_ptr *a,
4872 const union xfs_btree_ptr *b)
4874 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
4875 return (int64_t)be64_to_cpu(a->l) - be64_to_cpu(b->l);
4876 return (int64_t)be32_to_cpu(a->s) - be32_to_cpu(b->s);
4879 /* If there's an extent, we're done. */
4881 xfs_btree_has_record_helper(
4882 struct xfs_btree_cur *cur,
4883 union xfs_btree_rec *rec,
4889 /* Is there a record covering a given range of keys? */
4891 xfs_btree_has_record(
4892 struct xfs_btree_cur *cur,
4893 union xfs_btree_irec *low,
4894 union xfs_btree_irec *high,
4899 error = xfs_btree_query_range(cur, low, high,
4900 &xfs_btree_has_record_helper, NULL);
4901 if (error == -ECANCELED) {
4909 /* Are there more records in this btree? */
4911 xfs_btree_has_more_records(
4912 struct xfs_btree_cur *cur)
4914 struct xfs_btree_block *block;
4917 block = xfs_btree_get_block(cur, 0, &bp);
4919 /* There are still records in this block. */
4920 if (cur->bc_ptrs[0] < xfs_btree_get_numrecs(block))
4923 /* There are more record blocks. */
4924 if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
4925 return block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK);
4927 return block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK);