2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include "libxfs_priv.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_defer.h"
27 #include "xfs_inode.h"
28 #include "xfs_trans.h"
29 #include "xfs_btree.h"
30 #include "xfs_errortag.h"
31 #include "xfs_trace.h"
32 #include "xfs_cksum.h"
33 #include "xfs_alloc.h"
36 * Cursor allocation zone.
38 kmem_zone_t
*xfs_btree_cur_zone
;
41 * Btree magic numbers.
43 static const uint32_t xfs_magics
[2][XFS_BTNUM_MAX
] = {
44 { XFS_ABTB_MAGIC
, XFS_ABTC_MAGIC
, 0, XFS_BMAP_MAGIC
, XFS_IBT_MAGIC
,
46 { XFS_ABTB_CRC_MAGIC
, XFS_ABTC_CRC_MAGIC
, XFS_RMAP_CRC_MAGIC
,
47 XFS_BMAP_CRC_MAGIC
, XFS_IBT_CRC_MAGIC
, XFS_FIBT_CRC_MAGIC
,
56 uint32_t magic
= xfs_magics
[crc
][btnum
];
58 /* Ensure we asked for crc for crc-only magics. */
64 * Check a long btree block header. Return the address of the failing check,
65 * or NULL if everything is ok.
68 __xfs_btree_check_lblock(
69 struct xfs_btree_cur
*cur
,
70 struct xfs_btree_block
*block
,
74 struct xfs_mount
*mp
= cur
->bc_mp
;
75 xfs_btnum_t btnum
= cur
->bc_btnum
;
76 int crc
= xfs_sb_version_hascrc(&mp
->m_sb
);
79 if (!uuid_equal(&block
->bb_u
.l
.bb_uuid
, &mp
->m_sb
.sb_meta_uuid
))
80 return __this_address
;
81 if (block
->bb_u
.l
.bb_blkno
!=
82 cpu_to_be64(bp
? bp
->b_bn
: XFS_BUF_DADDR_NULL
))
83 return __this_address
;
84 if (block
->bb_u
.l
.bb_pad
!= cpu_to_be32(0))
85 return __this_address
;
88 if (be32_to_cpu(block
->bb_magic
) != xfs_btree_magic(crc
, btnum
))
89 return __this_address
;
90 if (be16_to_cpu(block
->bb_level
) != level
)
91 return __this_address
;
92 if (be16_to_cpu(block
->bb_numrecs
) >
93 cur
->bc_ops
->get_maxrecs(cur
, level
))
94 return __this_address
;
95 if (block
->bb_u
.l
.bb_leftsib
!= cpu_to_be64(NULLFSBLOCK
) &&
96 !xfs_btree_check_lptr(cur
, be64_to_cpu(block
->bb_u
.l
.bb_leftsib
),
98 return __this_address
;
99 if (block
->bb_u
.l
.bb_rightsib
!= cpu_to_be64(NULLFSBLOCK
) &&
100 !xfs_btree_check_lptr(cur
, be64_to_cpu(block
->bb_u
.l
.bb_rightsib
),
102 return __this_address
;
107 /* Check a long btree block header. */
109 xfs_btree_check_lblock(
110 struct xfs_btree_cur
*cur
,
111 struct xfs_btree_block
*block
,
115 struct xfs_mount
*mp
= cur
->bc_mp
;
118 fa
= __xfs_btree_check_lblock(cur
, block
, level
, bp
);
119 if (unlikely(XFS_TEST_ERROR(fa
!= NULL
, mp
,
120 XFS_ERRTAG_BTREE_CHECK_LBLOCK
))) {
122 trace_xfs_btree_corrupt(bp
, _RET_IP_
);
123 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_LOW
, mp
);
124 return -EFSCORRUPTED
;
130 * Check a short btree block header. Return the address of the failing check,
131 * or NULL if everything is ok.
134 __xfs_btree_check_sblock(
135 struct xfs_btree_cur
*cur
,
136 struct xfs_btree_block
*block
,
140 struct xfs_mount
*mp
= cur
->bc_mp
;
141 xfs_btnum_t btnum
= cur
->bc_btnum
;
142 int crc
= xfs_sb_version_hascrc(&mp
->m_sb
);
145 if (!uuid_equal(&block
->bb_u
.s
.bb_uuid
, &mp
->m_sb
.sb_meta_uuid
))
146 return __this_address
;
147 if (block
->bb_u
.s
.bb_blkno
!=
148 cpu_to_be64(bp
? bp
->b_bn
: XFS_BUF_DADDR_NULL
))
149 return __this_address
;
152 if (be32_to_cpu(block
->bb_magic
) != xfs_btree_magic(crc
, btnum
))
153 return __this_address
;
154 if (be16_to_cpu(block
->bb_level
) != level
)
155 return __this_address
;
156 if (be16_to_cpu(block
->bb_numrecs
) >
157 cur
->bc_ops
->get_maxrecs(cur
, level
))
158 return __this_address
;
159 if (block
->bb_u
.s
.bb_leftsib
!= cpu_to_be32(NULLAGBLOCK
) &&
160 !xfs_btree_check_sptr(cur
, be32_to_cpu(block
->bb_u
.s
.bb_leftsib
),
162 return __this_address
;
163 if (block
->bb_u
.s
.bb_rightsib
!= cpu_to_be32(NULLAGBLOCK
) &&
164 !xfs_btree_check_sptr(cur
, be32_to_cpu(block
->bb_u
.s
.bb_rightsib
),
166 return __this_address
;
171 /* Check a short btree block header. */
173 xfs_btree_check_sblock(
174 struct xfs_btree_cur
*cur
,
175 struct xfs_btree_block
*block
,
179 struct xfs_mount
*mp
= cur
->bc_mp
;
182 fa
= __xfs_btree_check_sblock(cur
, block
, level
, bp
);
183 if (unlikely(XFS_TEST_ERROR(fa
!= NULL
, mp
,
184 XFS_ERRTAG_BTREE_CHECK_SBLOCK
))) {
186 trace_xfs_btree_corrupt(bp
, _RET_IP_
);
187 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_LOW
, mp
);
188 return -EFSCORRUPTED
;
194 * Debug routine: check that block header is ok.
197 xfs_btree_check_block(
198 struct xfs_btree_cur
*cur
, /* btree cursor */
199 struct xfs_btree_block
*block
, /* generic btree block pointer */
200 int level
, /* level of the btree block */
201 struct xfs_buf
*bp
) /* buffer containing block, if any */
203 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
204 return xfs_btree_check_lblock(cur
, block
, level
, bp
);
206 return xfs_btree_check_sblock(cur
, block
, level
, bp
);
209 /* Check that this long pointer is valid and points within the fs. */
211 xfs_btree_check_lptr(
212 struct xfs_btree_cur
*cur
,
218 return xfs_verify_fsbno(cur
->bc_mp
, fsbno
);
221 /* Check that this short pointer is valid and points within the AG. */
223 xfs_btree_check_sptr(
224 struct xfs_btree_cur
*cur
,
230 return xfs_verify_agbno(cur
->bc_mp
, cur
->bc_private
.a
.agno
, agbno
);
235 * Check that a given (indexed) btree pointer at a certain level of a
236 * btree is valid and doesn't point past where it should.
240 struct xfs_btree_cur
*cur
,
241 union xfs_btree_ptr
*ptr
,
245 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
246 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
,
247 xfs_btree_check_lptr(cur
,
248 be64_to_cpu((&ptr
->l
)[index
]), level
));
250 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
,
251 xfs_btree_check_sptr(cur
,
252 be32_to_cpu((&ptr
->s
)[index
]), level
));
260 * Calculate CRC on the whole btree block and stuff it into the
261 * long-form btree header.
263 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
264 * it into the buffer so recovery knows what the last modification was that made
268 xfs_btree_lblock_calc_crc(
271 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
272 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
274 if (!xfs_sb_version_hascrc(&bp
->b_target
->bt_mount
->m_sb
))
277 block
->bb_u
.l
.bb_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
278 xfs_buf_update_cksum(bp
, XFS_BTREE_LBLOCK_CRC_OFF
);
282 xfs_btree_lblock_verify_crc(
285 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
286 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
288 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
289 if (!xfs_log_check_lsn(mp
, be64_to_cpu(block
->bb_u
.l
.bb_lsn
)))
291 return xfs_buf_verify_cksum(bp
, XFS_BTREE_LBLOCK_CRC_OFF
);
298 * Calculate CRC on the whole btree block and stuff it into the
299 * short-form btree header.
301 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
302 * it into the buffer so recovery knows what the last modification was that made
306 xfs_btree_sblock_calc_crc(
309 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
310 struct xfs_buf_log_item
*bip
= bp
->b_log_item
;
312 if (!xfs_sb_version_hascrc(&bp
->b_target
->bt_mount
->m_sb
))
315 block
->bb_u
.s
.bb_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
316 xfs_buf_update_cksum(bp
, XFS_BTREE_SBLOCK_CRC_OFF
);
320 xfs_btree_sblock_verify_crc(
323 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
324 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
326 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
327 if (!xfs_log_check_lsn(mp
, be64_to_cpu(block
->bb_u
.s
.bb_lsn
)))
328 return __this_address
;
329 return xfs_buf_verify_cksum(bp
, XFS_BTREE_SBLOCK_CRC_OFF
);
336 xfs_btree_free_block(
337 struct xfs_btree_cur
*cur
,
342 error
= cur
->bc_ops
->free_block(cur
, bp
);
344 xfs_trans_binval(cur
->bc_tp
, bp
);
345 XFS_BTREE_STATS_INC(cur
, free
);
351 * Delete the btree cursor.
354 xfs_btree_del_cursor(
355 xfs_btree_cur_t
*cur
, /* btree cursor */
356 int error
) /* del because of error */
358 int i
; /* btree level */
361 * Clear the buffer pointers, and release the buffers.
362 * If we're doing this in the face of an error, we
363 * need to make sure to inspect all of the entries
364 * in the bc_bufs array for buffers to be unlocked.
365 * This is because some of the btree code works from
366 * level n down to 0, and if we get an error along
367 * the way we won't have initialized all the entries
370 for (i
= 0; i
< cur
->bc_nlevels
; i
++) {
372 xfs_trans_brelse(cur
->bc_tp
, cur
->bc_bufs
[i
]);
377 * Can't free a bmap cursor without having dealt with the
378 * allocated indirect blocks' accounting.
380 ASSERT(cur
->bc_btnum
!= XFS_BTNUM_BMAP
||
381 cur
->bc_private
.b
.allocated
== 0);
385 kmem_zone_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
));
646 * Get the root block which is stored in the inode.
648 * For now this btree implementation assumes the btree root is always
649 * stored in the if_broot field of an inode fork.
651 STATIC
struct xfs_btree_block
*
653 struct xfs_btree_cur
*cur
)
655 struct xfs_ifork
*ifp
;
657 ifp
= XFS_IFORK_PTR(cur
->bc_private
.b
.ip
, cur
->bc_private
.b
.whichfork
);
658 return (struct xfs_btree_block
*)ifp
->if_broot
;
662 * Retrieve the block pointer from the cursor at the given level.
663 * This may be an inode btree root or from a buffer.
665 struct xfs_btree_block
* /* generic btree block pointer */
667 struct xfs_btree_cur
*cur
, /* btree cursor */
668 int level
, /* level in btree */
669 struct xfs_buf
**bpp
) /* buffer containing the block */
671 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
672 (level
== cur
->bc_nlevels
- 1)) {
674 return xfs_btree_get_iroot(cur
);
677 *bpp
= cur
->bc_bufs
[level
];
678 return XFS_BUF_TO_BLOCK(*bpp
);
682 * Get a buffer for the block, return it with no data read.
683 * Long-form addressing.
685 xfs_buf_t
* /* buffer for fsbno */
687 xfs_mount_t
*mp
, /* file system mount point */
688 xfs_trans_t
*tp
, /* transaction pointer */
689 xfs_fsblock_t fsbno
, /* file system block number */
690 uint lock
) /* lock flags for get_buf */
692 xfs_daddr_t d
; /* real disk block address */
694 ASSERT(fsbno
!= NULLFSBLOCK
);
695 d
= XFS_FSB_TO_DADDR(mp
, fsbno
);
696 return xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
, mp
->m_bsize
, lock
);
700 * Get a buffer for the block, return it with no data read.
701 * Short-form addressing.
703 xfs_buf_t
* /* buffer for agno/agbno */
705 xfs_mount_t
*mp
, /* file system mount point */
706 xfs_trans_t
*tp
, /* transaction pointer */
707 xfs_agnumber_t agno
, /* allocation group number */
708 xfs_agblock_t agbno
, /* allocation group block number */
709 uint lock
) /* lock flags for get_buf */
711 xfs_daddr_t d
; /* real disk block address */
713 ASSERT(agno
!= NULLAGNUMBER
);
714 ASSERT(agbno
!= NULLAGBLOCK
);
715 d
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
);
716 return xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
, mp
->m_bsize
, lock
);
720 * Check for the cursor referring to the last block at the given level.
722 int /* 1=is last block, 0=not last block */
723 xfs_btree_islastblock(
724 xfs_btree_cur_t
*cur
, /* btree cursor */
725 int level
) /* level to check */
727 struct xfs_btree_block
*block
; /* generic btree block pointer */
728 xfs_buf_t
*bp
; /* buffer containing block */
730 block
= xfs_btree_get_block(cur
, level
, &bp
);
731 xfs_btree_check_block(cur
, block
, level
, bp
);
732 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
733 return block
->bb_u
.l
.bb_rightsib
== cpu_to_be64(NULLFSBLOCK
);
735 return block
->bb_u
.s
.bb_rightsib
== cpu_to_be32(NULLAGBLOCK
);
739 * Change the cursor to point to the first record at the given level.
740 * Other levels are unaffected.
742 STATIC
int /* success=1, failure=0 */
744 xfs_btree_cur_t
*cur
, /* btree cursor */
745 int level
) /* level to change */
747 struct xfs_btree_block
*block
; /* generic btree block pointer */
748 xfs_buf_t
*bp
; /* buffer containing block */
751 * Get the block pointer for this level.
753 block
= xfs_btree_get_block(cur
, level
, &bp
);
754 if (xfs_btree_check_block(cur
, block
, level
, bp
))
757 * It's empty, there is no such record.
759 if (!block
->bb_numrecs
)
762 * Set the ptr value to 1, that's the first record/key.
764 cur
->bc_ptrs
[level
] = 1;
769 * Change the cursor to point to the last record in the current block
770 * at the given level. Other levels are unaffected.
772 STATIC
int /* success=1, failure=0 */
774 xfs_btree_cur_t
*cur
, /* btree cursor */
775 int level
) /* level to change */
777 struct xfs_btree_block
*block
; /* generic btree block pointer */
778 xfs_buf_t
*bp
; /* buffer containing block */
781 * Get the block pointer for this level.
783 block
= xfs_btree_get_block(cur
, level
, &bp
);
784 if (xfs_btree_check_block(cur
, block
, level
, bp
))
787 * It's empty, there is no such record.
789 if (!block
->bb_numrecs
)
792 * Set the ptr value to numrecs, that's the last record/key.
794 cur
->bc_ptrs
[level
] = be16_to_cpu(block
->bb_numrecs
);
799 * Compute first and last byte offsets for the fields given.
800 * Interprets the offsets table, which contains struct field offsets.
804 int64_t fields
, /* bitmask of fields */
805 const short *offsets
, /* table of field offsets */
806 int nbits
, /* number of bits to inspect */
807 int *first
, /* output: first byte offset */
808 int *last
) /* output: last byte offset */
810 int i
; /* current bit number */
811 int64_t imask
; /* mask for current bit number */
815 * Find the lowest bit, so the first byte offset.
817 for (i
= 0, imask
= 1LL; ; i
++, imask
<<= 1) {
818 if (imask
& fields
) {
824 * Find the highest bit, so the last byte offset.
826 for (i
= nbits
- 1, imask
= 1LL << i
; ; i
--, imask
>>= 1) {
827 if (imask
& fields
) {
828 *last
= offsets
[i
+ 1] - 1;
835 * Get a buffer for the block, return it read in.
836 * Long-form addressing.
840 struct xfs_mount
*mp
, /* file system mount point */
841 struct xfs_trans
*tp
, /* transaction pointer */
842 xfs_fsblock_t fsbno
, /* file system block number */
843 uint lock
, /* lock flags for read_buf */
844 struct xfs_buf
**bpp
, /* buffer for fsbno */
845 int refval
, /* ref count value for buffer */
846 const struct xfs_buf_ops
*ops
)
848 struct xfs_buf
*bp
; /* return value */
849 xfs_daddr_t d
; /* real disk block address */
852 if (!xfs_verify_fsbno(mp
, fsbno
))
853 return -EFSCORRUPTED
;
854 d
= XFS_FSB_TO_DADDR(mp
, fsbno
);
855 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, d
,
856 mp
->m_bsize
, lock
, &bp
, ops
);
860 xfs_buf_set_ref(bp
, refval
);
866 * Read-ahead the block, don't wait for it, don't return a buffer.
867 * Long-form addressing.
871 xfs_btree_reada_bufl(
872 struct xfs_mount
*mp
, /* file system mount point */
873 xfs_fsblock_t fsbno
, /* file system block number */
874 xfs_extlen_t count
, /* count of filesystem blocks */
875 const struct xfs_buf_ops
*ops
)
879 ASSERT(fsbno
!= NULLFSBLOCK
);
880 d
= XFS_FSB_TO_DADDR(mp
, fsbno
);
881 xfs_buf_readahead(mp
->m_ddev_targp
, d
, mp
->m_bsize
* count
, ops
);
885 * Read-ahead the block, don't wait for it, don't return a buffer.
886 * Short-form addressing.
890 xfs_btree_reada_bufs(
891 struct xfs_mount
*mp
, /* file system mount point */
892 xfs_agnumber_t agno
, /* allocation group number */
893 xfs_agblock_t agbno
, /* allocation group block number */
894 xfs_extlen_t count
, /* count of filesystem blocks */
895 const struct xfs_buf_ops
*ops
)
899 ASSERT(agno
!= NULLAGNUMBER
);
900 ASSERT(agbno
!= NULLAGBLOCK
);
901 d
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
);
902 xfs_buf_readahead(mp
->m_ddev_targp
, d
, mp
->m_bsize
* count
, ops
);
906 xfs_btree_readahead_lblock(
907 struct xfs_btree_cur
*cur
,
909 struct xfs_btree_block
*block
)
912 xfs_fsblock_t left
= be64_to_cpu(block
->bb_u
.l
.bb_leftsib
);
913 xfs_fsblock_t right
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
915 if ((lr
& XFS_BTCUR_LEFTRA
) && left
!= NULLFSBLOCK
) {
916 xfs_btree_reada_bufl(cur
->bc_mp
, left
, 1,
917 cur
->bc_ops
->buf_ops
);
921 if ((lr
& XFS_BTCUR_RIGHTRA
) && right
!= NULLFSBLOCK
) {
922 xfs_btree_reada_bufl(cur
->bc_mp
, right
, 1,
923 cur
->bc_ops
->buf_ops
);
931 xfs_btree_readahead_sblock(
932 struct xfs_btree_cur
*cur
,
934 struct xfs_btree_block
*block
)
937 xfs_agblock_t left
= be32_to_cpu(block
->bb_u
.s
.bb_leftsib
);
938 xfs_agblock_t right
= be32_to_cpu(block
->bb_u
.s
.bb_rightsib
);
941 if ((lr
& XFS_BTCUR_LEFTRA
) && left
!= NULLAGBLOCK
) {
942 xfs_btree_reada_bufs(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
943 left
, 1, cur
->bc_ops
->buf_ops
);
947 if ((lr
& XFS_BTCUR_RIGHTRA
) && right
!= NULLAGBLOCK
) {
948 xfs_btree_reada_bufs(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
949 right
, 1, cur
->bc_ops
->buf_ops
);
957 * Read-ahead btree blocks, at the given level.
958 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
962 struct xfs_btree_cur
*cur
, /* btree cursor */
963 int lev
, /* level in btree */
964 int lr
) /* left/right bits */
966 struct xfs_btree_block
*block
;
969 * No readahead needed if we are at the root level and the
970 * btree root is stored in the inode.
972 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
973 (lev
== cur
->bc_nlevels
- 1))
976 if ((cur
->bc_ra
[lev
] | lr
) == cur
->bc_ra
[lev
])
979 cur
->bc_ra
[lev
] |= lr
;
980 block
= XFS_BUF_TO_BLOCK(cur
->bc_bufs
[lev
]);
982 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
983 return xfs_btree_readahead_lblock(cur
, lr
, block
);
984 return xfs_btree_readahead_sblock(cur
, lr
, block
);
988 xfs_btree_ptr_to_daddr(
989 struct xfs_btree_cur
*cur
,
990 union xfs_btree_ptr
*ptr
)
992 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
993 ASSERT(ptr
->l
!= cpu_to_be64(NULLFSBLOCK
));
995 return XFS_FSB_TO_DADDR(cur
->bc_mp
, be64_to_cpu(ptr
->l
));
997 ASSERT(cur
->bc_private
.a
.agno
!= NULLAGNUMBER
);
998 ASSERT(ptr
->s
!= cpu_to_be32(NULLAGBLOCK
));
1000 return XFS_AGB_TO_DADDR(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
1001 be32_to_cpu(ptr
->s
));
1006 * Readahead @count btree blocks at the given @ptr location.
1008 * We don't need to care about long or short form btrees here as we have a
1009 * method of converting the ptr directly to a daddr available to us.
1012 xfs_btree_readahead_ptr(
1013 struct xfs_btree_cur
*cur
,
1014 union xfs_btree_ptr
*ptr
,
1017 xfs_buf_readahead(cur
->bc_mp
->m_ddev_targp
,
1018 xfs_btree_ptr_to_daddr(cur
, ptr
),
1019 cur
->bc_mp
->m_bsize
* count
, cur
->bc_ops
->buf_ops
);
1023 * Set the buffer for level "lev" in the cursor to bp, releasing
1024 * any previous buffer.
1028 xfs_btree_cur_t
*cur
, /* btree cursor */
1029 int lev
, /* level in btree */
1030 xfs_buf_t
*bp
) /* new buffer to set */
1032 struct xfs_btree_block
*b
; /* btree block */
1034 if (cur
->bc_bufs
[lev
])
1035 xfs_trans_brelse(cur
->bc_tp
, cur
->bc_bufs
[lev
]);
1036 cur
->bc_bufs
[lev
] = bp
;
1037 cur
->bc_ra
[lev
] = 0;
1039 b
= XFS_BUF_TO_BLOCK(bp
);
1040 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
1041 if (b
->bb_u
.l
.bb_leftsib
== cpu_to_be64(NULLFSBLOCK
))
1042 cur
->bc_ra
[lev
] |= XFS_BTCUR_LEFTRA
;
1043 if (b
->bb_u
.l
.bb_rightsib
== cpu_to_be64(NULLFSBLOCK
))
1044 cur
->bc_ra
[lev
] |= XFS_BTCUR_RIGHTRA
;
1046 if (b
->bb_u
.s
.bb_leftsib
== cpu_to_be32(NULLAGBLOCK
))
1047 cur
->bc_ra
[lev
] |= XFS_BTCUR_LEFTRA
;
1048 if (b
->bb_u
.s
.bb_rightsib
== cpu_to_be32(NULLAGBLOCK
))
1049 cur
->bc_ra
[lev
] |= XFS_BTCUR_RIGHTRA
;
1054 xfs_btree_ptr_is_null(
1055 struct xfs_btree_cur
*cur
,
1056 union xfs_btree_ptr
*ptr
)
1058 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
1059 return ptr
->l
== cpu_to_be64(NULLFSBLOCK
);
1061 return ptr
->s
== cpu_to_be32(NULLAGBLOCK
);
1065 xfs_btree_set_ptr_null(
1066 struct xfs_btree_cur
*cur
,
1067 union xfs_btree_ptr
*ptr
)
1069 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
1070 ptr
->l
= cpu_to_be64(NULLFSBLOCK
);
1072 ptr
->s
= cpu_to_be32(NULLAGBLOCK
);
1076 * Get/set/init sibling pointers
1079 xfs_btree_get_sibling(
1080 struct xfs_btree_cur
*cur
,
1081 struct xfs_btree_block
*block
,
1082 union xfs_btree_ptr
*ptr
,
1085 ASSERT(lr
== XFS_BB_LEFTSIB
|| lr
== XFS_BB_RIGHTSIB
);
1087 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
1088 if (lr
== XFS_BB_RIGHTSIB
)
1089 ptr
->l
= block
->bb_u
.l
.bb_rightsib
;
1091 ptr
->l
= block
->bb_u
.l
.bb_leftsib
;
1093 if (lr
== XFS_BB_RIGHTSIB
)
1094 ptr
->s
= block
->bb_u
.s
.bb_rightsib
;
1096 ptr
->s
= block
->bb_u
.s
.bb_leftsib
;
1101 xfs_btree_set_sibling(
1102 struct xfs_btree_cur
*cur
,
1103 struct xfs_btree_block
*block
,
1104 union xfs_btree_ptr
*ptr
,
1107 ASSERT(lr
== XFS_BB_LEFTSIB
|| lr
== XFS_BB_RIGHTSIB
);
1109 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
1110 if (lr
== XFS_BB_RIGHTSIB
)
1111 block
->bb_u
.l
.bb_rightsib
= ptr
->l
;
1113 block
->bb_u
.l
.bb_leftsib
= ptr
->l
;
1115 if (lr
== XFS_BB_RIGHTSIB
)
1116 block
->bb_u
.s
.bb_rightsib
= ptr
->s
;
1118 block
->bb_u
.s
.bb_leftsib
= ptr
->s
;
1123 xfs_btree_init_block_int(
1124 struct xfs_mount
*mp
,
1125 struct xfs_btree_block
*buf
,
1133 int crc
= xfs_sb_version_hascrc(&mp
->m_sb
);
1134 __u32 magic
= xfs_btree_magic(crc
, btnum
);
1136 buf
->bb_magic
= cpu_to_be32(magic
);
1137 buf
->bb_level
= cpu_to_be16(level
);
1138 buf
->bb_numrecs
= cpu_to_be16(numrecs
);
1140 if (flags
& XFS_BTREE_LONG_PTRS
) {
1141 buf
->bb_u
.l
.bb_leftsib
= cpu_to_be64(NULLFSBLOCK
);
1142 buf
->bb_u
.l
.bb_rightsib
= cpu_to_be64(NULLFSBLOCK
);
1144 buf
->bb_u
.l
.bb_blkno
= cpu_to_be64(blkno
);
1145 buf
->bb_u
.l
.bb_owner
= cpu_to_be64(owner
);
1146 uuid_copy(&buf
->bb_u
.l
.bb_uuid
, &mp
->m_sb
.sb_meta_uuid
);
1147 buf
->bb_u
.l
.bb_pad
= 0;
1148 buf
->bb_u
.l
.bb_lsn
= 0;
1151 /* owner is a 32 bit value on short blocks */
1152 __u32 __owner
= (__u32
)owner
;
1154 buf
->bb_u
.s
.bb_leftsib
= cpu_to_be32(NULLAGBLOCK
);
1155 buf
->bb_u
.s
.bb_rightsib
= cpu_to_be32(NULLAGBLOCK
);
1157 buf
->bb_u
.s
.bb_blkno
= cpu_to_be64(blkno
);
1158 buf
->bb_u
.s
.bb_owner
= cpu_to_be32(__owner
);
1159 uuid_copy(&buf
->bb_u
.s
.bb_uuid
, &mp
->m_sb
.sb_meta_uuid
);
1160 buf
->bb_u
.s
.bb_lsn
= 0;
1166 xfs_btree_init_block(
1167 struct xfs_mount
*mp
,
1175 xfs_btree_init_block_int(mp
, XFS_BUF_TO_BLOCK(bp
), bp
->b_bn
,
1176 btnum
, level
, numrecs
, owner
, flags
);
1180 xfs_btree_init_block_cur(
1181 struct xfs_btree_cur
*cur
,
1189 * we can pull the owner from the cursor right now as the different
1190 * owners align directly with the pointer size of the btree. This may
1191 * change in future, but is safe for current users of the generic btree
1194 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
1195 owner
= cur
->bc_private
.b
.ip
->i_ino
;
1197 owner
= cur
->bc_private
.a
.agno
;
1199 xfs_btree_init_block_int(cur
->bc_mp
, XFS_BUF_TO_BLOCK(bp
), bp
->b_bn
,
1200 cur
->bc_btnum
, level
, numrecs
,
1201 owner
, cur
->bc_flags
);
1205 * Return true if ptr is the last record in the btree and
1206 * we need to track updates to this record. The decision
1207 * will be further refined in the update_lastrec method.
1210 xfs_btree_is_lastrec(
1211 struct xfs_btree_cur
*cur
,
1212 struct xfs_btree_block
*block
,
1215 union xfs_btree_ptr ptr
;
1219 if (!(cur
->bc_flags
& XFS_BTREE_LASTREC_UPDATE
))
1222 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_RIGHTSIB
);
1223 if (!xfs_btree_ptr_is_null(cur
, &ptr
))
1229 xfs_btree_buf_to_ptr(
1230 struct xfs_btree_cur
*cur
,
1232 union xfs_btree_ptr
*ptr
)
1234 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
1235 ptr
->l
= cpu_to_be64(XFS_DADDR_TO_FSB(cur
->bc_mp
,
1238 ptr
->s
= cpu_to_be32(xfs_daddr_to_agbno(cur
->bc_mp
,
1245 struct xfs_btree_cur
*cur
,
1248 switch (cur
->bc_btnum
) {
1251 xfs_buf_set_ref(bp
, XFS_ALLOC_BTREE_REF
);
1254 case XFS_BTNUM_FINO
:
1255 xfs_buf_set_ref(bp
, XFS_INO_BTREE_REF
);
1257 case XFS_BTNUM_BMAP
:
1258 xfs_buf_set_ref(bp
, XFS_BMAP_BTREE_REF
);
1260 case XFS_BTNUM_RMAP
:
1261 xfs_buf_set_ref(bp
, XFS_RMAP_BTREE_REF
);
1263 case XFS_BTNUM_REFC
:
1264 xfs_buf_set_ref(bp
, XFS_REFC_BTREE_REF
);
1272 xfs_btree_get_buf_block(
1273 struct xfs_btree_cur
*cur
,
1274 union xfs_btree_ptr
*ptr
,
1276 struct xfs_btree_block
**block
,
1277 struct xfs_buf
**bpp
)
1279 struct xfs_mount
*mp
= cur
->bc_mp
;
1282 /* need to sort out how callers deal with failures first */
1283 ASSERT(!(flags
& XBF_TRYLOCK
));
1285 d
= xfs_btree_ptr_to_daddr(cur
, ptr
);
1286 *bpp
= xfs_trans_get_buf(cur
->bc_tp
, mp
->m_ddev_targp
, d
,
1287 mp
->m_bsize
, flags
);
1292 (*bpp
)->b_ops
= cur
->bc_ops
->buf_ops
;
1293 *block
= XFS_BUF_TO_BLOCK(*bpp
);
1298 * Read in the buffer at the given ptr and return the buffer and
1299 * the block pointer within the buffer.
1302 xfs_btree_read_buf_block(
1303 struct xfs_btree_cur
*cur
,
1304 union xfs_btree_ptr
*ptr
,
1306 struct xfs_btree_block
**block
,
1307 struct xfs_buf
**bpp
)
1309 struct xfs_mount
*mp
= cur
->bc_mp
;
1313 /* need to sort out how callers deal with failures first */
1314 ASSERT(!(flags
& XBF_TRYLOCK
));
1316 d
= xfs_btree_ptr_to_daddr(cur
, ptr
);
1317 error
= xfs_trans_read_buf(mp
, cur
->bc_tp
, mp
->m_ddev_targp
, d
,
1318 mp
->m_bsize
, flags
, bpp
,
1319 cur
->bc_ops
->buf_ops
);
1323 xfs_btree_set_refs(cur
, *bpp
);
1324 *block
= XFS_BUF_TO_BLOCK(*bpp
);
1329 * Copy keys from one btree block to another.
1332 xfs_btree_copy_keys(
1333 struct xfs_btree_cur
*cur
,
1334 union xfs_btree_key
*dst_key
,
1335 union xfs_btree_key
*src_key
,
1338 ASSERT(numkeys
>= 0);
1339 memcpy(dst_key
, src_key
, numkeys
* cur
->bc_ops
->key_len
);
1343 * Copy records from one btree block to another.
1346 xfs_btree_copy_recs(
1347 struct xfs_btree_cur
*cur
,
1348 union xfs_btree_rec
*dst_rec
,
1349 union xfs_btree_rec
*src_rec
,
1352 ASSERT(numrecs
>= 0);
1353 memcpy(dst_rec
, src_rec
, numrecs
* cur
->bc_ops
->rec_len
);
1357 * Copy block pointers from one btree block to another.
1360 xfs_btree_copy_ptrs(
1361 struct xfs_btree_cur
*cur
,
1362 union xfs_btree_ptr
*dst_ptr
,
1363 union xfs_btree_ptr
*src_ptr
,
1366 ASSERT(numptrs
>= 0);
1367 memcpy(dst_ptr
, src_ptr
, numptrs
* xfs_btree_ptr_len(cur
));
1371 * Shift keys one index left/right inside a single btree block.
1374 xfs_btree_shift_keys(
1375 struct xfs_btree_cur
*cur
,
1376 union xfs_btree_key
*key
,
1382 ASSERT(numkeys
>= 0);
1383 ASSERT(dir
== 1 || dir
== -1);
1385 dst_key
= (char *)key
+ (dir
* cur
->bc_ops
->key_len
);
1386 memmove(dst_key
, key
, numkeys
* cur
->bc_ops
->key_len
);
1390 * Shift records one index left/right inside a single btree block.
1393 xfs_btree_shift_recs(
1394 struct xfs_btree_cur
*cur
,
1395 union xfs_btree_rec
*rec
,
1401 ASSERT(numrecs
>= 0);
1402 ASSERT(dir
== 1 || dir
== -1);
1404 dst_rec
= (char *)rec
+ (dir
* cur
->bc_ops
->rec_len
);
1405 memmove(dst_rec
, rec
, numrecs
* cur
->bc_ops
->rec_len
);
1409 * Shift block pointers one index left/right inside a single btree block.
1412 xfs_btree_shift_ptrs(
1413 struct xfs_btree_cur
*cur
,
1414 union xfs_btree_ptr
*ptr
,
1420 ASSERT(numptrs
>= 0);
1421 ASSERT(dir
== 1 || dir
== -1);
1423 dst_ptr
= (char *)ptr
+ (dir
* xfs_btree_ptr_len(cur
));
1424 memmove(dst_ptr
, ptr
, numptrs
* xfs_btree_ptr_len(cur
));
1428 * Log key values from the btree block.
1432 struct xfs_btree_cur
*cur
,
1439 xfs_trans_buf_set_type(cur
->bc_tp
, bp
, XFS_BLFT_BTREE_BUF
);
1440 xfs_trans_log_buf(cur
->bc_tp
, bp
,
1441 xfs_btree_key_offset(cur
, first
),
1442 xfs_btree_key_offset(cur
, last
+ 1) - 1);
1444 xfs_trans_log_inode(cur
->bc_tp
, cur
->bc_private
.b
.ip
,
1445 xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
));
1450 * Log record values from the btree block.
1454 struct xfs_btree_cur
*cur
,
1460 xfs_trans_buf_set_type(cur
->bc_tp
, bp
, XFS_BLFT_BTREE_BUF
);
1461 xfs_trans_log_buf(cur
->bc_tp
, bp
,
1462 xfs_btree_rec_offset(cur
, first
),
1463 xfs_btree_rec_offset(cur
, last
+ 1) - 1);
1468 * Log block pointer fields from a btree block (nonleaf).
1472 struct xfs_btree_cur
*cur
, /* btree cursor */
1473 struct xfs_buf
*bp
, /* buffer containing btree block */
1474 int first
, /* index of first pointer to log */
1475 int last
) /* index of last pointer to log */
1479 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
1480 int level
= xfs_btree_get_level(block
);
1482 xfs_trans_buf_set_type(cur
->bc_tp
, bp
, XFS_BLFT_BTREE_BUF
);
1483 xfs_trans_log_buf(cur
->bc_tp
, bp
,
1484 xfs_btree_ptr_offset(cur
, first
, level
),
1485 xfs_btree_ptr_offset(cur
, last
+ 1, level
) - 1);
1487 xfs_trans_log_inode(cur
->bc_tp
, cur
->bc_private
.b
.ip
,
1488 xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
));
1494 * Log fields from a btree block header.
1497 xfs_btree_log_block(
1498 struct xfs_btree_cur
*cur
, /* btree cursor */
1499 struct xfs_buf
*bp
, /* buffer containing btree block */
1500 int fields
) /* mask of fields: XFS_BB_... */
1502 int first
; /* first byte offset logged */
1503 int last
; /* last byte offset logged */
1504 static const short soffsets
[] = { /* table of offsets (short) */
1505 offsetof(struct xfs_btree_block
, bb_magic
),
1506 offsetof(struct xfs_btree_block
, bb_level
),
1507 offsetof(struct xfs_btree_block
, bb_numrecs
),
1508 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_leftsib
),
1509 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_rightsib
),
1510 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_blkno
),
1511 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_lsn
),
1512 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_uuid
),
1513 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_owner
),
1514 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_crc
),
1515 XFS_BTREE_SBLOCK_CRC_LEN
1517 static const short loffsets
[] = { /* table of offsets (long) */
1518 offsetof(struct xfs_btree_block
, bb_magic
),
1519 offsetof(struct xfs_btree_block
, bb_level
),
1520 offsetof(struct xfs_btree_block
, bb_numrecs
),
1521 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_leftsib
),
1522 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_rightsib
),
1523 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_blkno
),
1524 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_lsn
),
1525 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_uuid
),
1526 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_owner
),
1527 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_crc
),
1528 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_pad
),
1529 XFS_BTREE_LBLOCK_CRC_LEN
1535 if (cur
->bc_flags
& XFS_BTREE_CRC_BLOCKS
) {
1537 * We don't log the CRC when updating a btree
1538 * block but instead recreate it during log
1539 * recovery. As the log buffers have checksums
1540 * of their own this is safe and avoids logging a crc
1541 * update in a lot of places.
1543 if (fields
== XFS_BB_ALL_BITS
)
1544 fields
= XFS_BB_ALL_BITS_CRC
;
1545 nbits
= XFS_BB_NUM_BITS_CRC
;
1547 nbits
= XFS_BB_NUM_BITS
;
1549 xfs_btree_offsets(fields
,
1550 (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) ?
1551 loffsets
: soffsets
,
1552 nbits
, &first
, &last
);
1553 xfs_trans_buf_set_type(cur
->bc_tp
, bp
, XFS_BLFT_BTREE_BUF
);
1554 xfs_trans_log_buf(cur
->bc_tp
, bp
, first
, last
);
1556 xfs_trans_log_inode(cur
->bc_tp
, cur
->bc_private
.b
.ip
,
1557 xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
));
1562 * Increment cursor by one record at the level.
1563 * For nonzero levels the leaf-ward information is untouched.
1566 xfs_btree_increment(
1567 struct xfs_btree_cur
*cur
,
1569 int *stat
) /* success/failure */
1571 struct xfs_btree_block
*block
;
1572 union xfs_btree_ptr ptr
;
1574 int error
; /* error return value */
1577 ASSERT(level
< cur
->bc_nlevels
);
1579 /* Read-ahead to the right at this level. */
1580 xfs_btree_readahead(cur
, level
, XFS_BTCUR_RIGHTRA
);
1582 /* Get a pointer to the btree block. */
1583 block
= xfs_btree_get_block(cur
, level
, &bp
);
1586 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
1591 /* We're done if we remain in the block after the increment. */
1592 if (++cur
->bc_ptrs
[level
] <= xfs_btree_get_numrecs(block
))
1595 /* Fail if we just went off the right edge of the tree. */
1596 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_RIGHTSIB
);
1597 if (xfs_btree_ptr_is_null(cur
, &ptr
))
1600 XFS_BTREE_STATS_INC(cur
, increment
);
1603 * March up the tree incrementing pointers.
1604 * Stop when we don't go off the right edge of a block.
1606 for (lev
= level
+ 1; lev
< cur
->bc_nlevels
; lev
++) {
1607 block
= xfs_btree_get_block(cur
, lev
, &bp
);
1610 error
= xfs_btree_check_block(cur
, block
, lev
, bp
);
1615 if (++cur
->bc_ptrs
[lev
] <= xfs_btree_get_numrecs(block
))
1618 /* Read-ahead the right block for the next loop. */
1619 xfs_btree_readahead(cur
, lev
, XFS_BTCUR_RIGHTRA
);
1623 * If we went off the root then we are either seriously
1624 * confused or have the tree root in an inode.
1626 if (lev
== cur
->bc_nlevels
) {
1627 if (cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
)
1630 error
= -EFSCORRUPTED
;
1633 ASSERT(lev
< cur
->bc_nlevels
);
1636 * Now walk back down the tree, fixing up the cursor's buffer
1637 * pointers and key numbers.
1639 for (block
= xfs_btree_get_block(cur
, lev
, &bp
); lev
> level
; ) {
1640 union xfs_btree_ptr
*ptrp
;
1642 ptrp
= xfs_btree_ptr_addr(cur
, cur
->bc_ptrs
[lev
], block
);
1644 error
= xfs_btree_read_buf_block(cur
, ptrp
, 0, &block
, &bp
);
1648 xfs_btree_setbuf(cur
, lev
, bp
);
1649 cur
->bc_ptrs
[lev
] = 1;
1664 * Decrement cursor by one record at the level.
1665 * For nonzero levels the leaf-ward information is untouched.
1668 xfs_btree_decrement(
1669 struct xfs_btree_cur
*cur
,
1671 int *stat
) /* success/failure */
1673 struct xfs_btree_block
*block
;
1675 int error
; /* error return value */
1677 union xfs_btree_ptr ptr
;
1679 ASSERT(level
< cur
->bc_nlevels
);
1681 /* Read-ahead to the left at this level. */
1682 xfs_btree_readahead(cur
, level
, XFS_BTCUR_LEFTRA
);
1684 /* We're done if we remain in the block after the decrement. */
1685 if (--cur
->bc_ptrs
[level
] > 0)
1688 /* Get a pointer to the btree block. */
1689 block
= xfs_btree_get_block(cur
, level
, &bp
);
1692 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
1697 /* Fail if we just went off the left edge of the tree. */
1698 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_LEFTSIB
);
1699 if (xfs_btree_ptr_is_null(cur
, &ptr
))
1702 XFS_BTREE_STATS_INC(cur
, decrement
);
1705 * March up the tree decrementing pointers.
1706 * Stop when we don't go off the left edge of a block.
1708 for (lev
= level
+ 1; lev
< cur
->bc_nlevels
; lev
++) {
1709 if (--cur
->bc_ptrs
[lev
] > 0)
1711 /* Read-ahead the left block for the next loop. */
1712 xfs_btree_readahead(cur
, lev
, XFS_BTCUR_LEFTRA
);
1716 * If we went off the root then we are seriously confused.
1717 * or the root of the tree is in an inode.
1719 if (lev
== cur
->bc_nlevels
) {
1720 if (cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
)
1723 error
= -EFSCORRUPTED
;
1726 ASSERT(lev
< cur
->bc_nlevels
);
1729 * Now walk back down the tree, fixing up the cursor's buffer
1730 * pointers and key numbers.
1732 for (block
= xfs_btree_get_block(cur
, lev
, &bp
); lev
> level
; ) {
1733 union xfs_btree_ptr
*ptrp
;
1735 ptrp
= xfs_btree_ptr_addr(cur
, cur
->bc_ptrs
[lev
], block
);
1737 error
= xfs_btree_read_buf_block(cur
, ptrp
, 0, &block
, &bp
);
1740 xfs_btree_setbuf(cur
, lev
, bp
);
1741 cur
->bc_ptrs
[lev
] = xfs_btree_get_numrecs(block
);
1756 xfs_btree_lookup_get_block(
1757 struct xfs_btree_cur
*cur
, /* btree cursor */
1758 int level
, /* level in the btree */
1759 union xfs_btree_ptr
*pp
, /* ptr to btree block */
1760 struct xfs_btree_block
**blkp
) /* return btree block */
1762 struct xfs_buf
*bp
; /* buffer pointer for btree block */
1765 /* special case the root block if in an inode */
1766 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
1767 (level
== cur
->bc_nlevels
- 1)) {
1768 *blkp
= xfs_btree_get_iroot(cur
);
1773 * If the old buffer at this level for the disk address we are
1774 * looking for re-use it.
1776 * Otherwise throw it away and get a new one.
1778 bp
= cur
->bc_bufs
[level
];
1779 if (bp
&& XFS_BUF_ADDR(bp
) == xfs_btree_ptr_to_daddr(cur
, pp
)) {
1780 *blkp
= XFS_BUF_TO_BLOCK(bp
);
1784 error
= xfs_btree_read_buf_block(cur
, pp
, 0, blkp
, &bp
);
1788 /* Check the inode owner since the verifiers don't. */
1789 if (xfs_sb_version_hascrc(&cur
->bc_mp
->m_sb
) &&
1790 !(cur
->bc_private
.b
.flags
& XFS_BTCUR_BPRV_INVALID_OWNER
) &&
1791 (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) &&
1792 be64_to_cpu((*blkp
)->bb_u
.l
.bb_owner
) !=
1793 cur
->bc_private
.b
.ip
->i_ino
)
1796 /* Did we get the level we were looking for? */
1797 if (be16_to_cpu((*blkp
)->bb_level
) != level
)
1800 /* Check that internal nodes have at least one record. */
1801 if (level
!= 0 && be16_to_cpu((*blkp
)->bb_numrecs
) == 0)
1804 xfs_btree_setbuf(cur
, level
, bp
);
1809 xfs_trans_brelse(cur
->bc_tp
, bp
);
1810 return -EFSCORRUPTED
;
1814 * Get current search key. For level 0 we don't actually have a key
1815 * structure so we make one up from the record. For all other levels
1816 * we just return the right key.
1818 STATIC
union xfs_btree_key
*
1819 xfs_lookup_get_search_key(
1820 struct xfs_btree_cur
*cur
,
1823 struct xfs_btree_block
*block
,
1824 union xfs_btree_key
*kp
)
1827 cur
->bc_ops
->init_key_from_rec(kp
,
1828 xfs_btree_rec_addr(cur
, keyno
, block
));
1832 return xfs_btree_key_addr(cur
, keyno
, block
);
1836 * Lookup the record. The cursor is made to point to it, based on dir.
1837 * stat is set to 0 if can't find any such record, 1 for success.
1841 struct xfs_btree_cur
*cur
, /* btree cursor */
1842 xfs_lookup_t dir
, /* <=, ==, or >= */
1843 int *stat
) /* success/failure */
1845 struct xfs_btree_block
*block
; /* current btree block */
1846 int64_t diff
; /* difference for the current key */
1847 int error
; /* error return value */
1848 int keyno
; /* current key number */
1849 int level
; /* level in the btree */
1850 union xfs_btree_ptr
*pp
; /* ptr to btree block */
1851 union xfs_btree_ptr ptr
; /* ptr to btree block */
1853 XFS_BTREE_STATS_INC(cur
, lookup
);
1855 /* No such thing as a zero-level tree. */
1856 if (cur
->bc_nlevels
== 0)
1857 return -EFSCORRUPTED
;
1862 /* initialise start pointer from cursor */
1863 cur
->bc_ops
->init_ptr_from_cur(cur
, &ptr
);
1867 * Iterate over each level in the btree, starting at the root.
1868 * For each level above the leaves, find the key we need, based
1869 * on the lookup record, then follow the corresponding block
1870 * pointer down to the next level.
1872 for (level
= cur
->bc_nlevels
- 1, diff
= 1; level
>= 0; level
--) {
1873 /* Get the block we need to do the lookup on. */
1874 error
= xfs_btree_lookup_get_block(cur
, level
, pp
, &block
);
1880 * If we already had a key match at a higher level, we
1881 * know we need to use the first entry in this block.
1885 /* Otherwise search this block. Do a binary search. */
1887 int high
; /* high entry number */
1888 int low
; /* low entry number */
1890 /* Set low and high entry numbers, 1-based. */
1892 high
= xfs_btree_get_numrecs(block
);
1894 /* Block is empty, must be an empty leaf. */
1895 ASSERT(level
== 0 && cur
->bc_nlevels
== 1);
1897 cur
->bc_ptrs
[0] = dir
!= XFS_LOOKUP_LE
;
1902 /* Binary search the block. */
1903 while (low
<= high
) {
1904 union xfs_btree_key key
;
1905 union xfs_btree_key
*kp
;
1907 XFS_BTREE_STATS_INC(cur
, compare
);
1909 /* keyno is average of low and high. */
1910 keyno
= (low
+ high
) >> 1;
1912 /* Get current search key */
1913 kp
= xfs_lookup_get_search_key(cur
, level
,
1914 keyno
, block
, &key
);
1917 * Compute difference to get next direction:
1918 * - less than, move right
1919 * - greater than, move left
1920 * - equal, we're done
1922 diff
= cur
->bc_ops
->key_diff(cur
, kp
);
1933 * If there are more levels, set up for the next level
1934 * by getting the block number and filling in the cursor.
1938 * If we moved left, need the previous key number,
1939 * unless there isn't one.
1941 if (diff
> 0 && --keyno
< 1)
1943 pp
= xfs_btree_ptr_addr(cur
, keyno
, block
);
1946 error
= xfs_btree_check_ptr(cur
, pp
, 0, level
);
1950 cur
->bc_ptrs
[level
] = keyno
;
1954 /* Done with the search. See if we need to adjust the results. */
1955 if (dir
!= XFS_LOOKUP_LE
&& diff
< 0) {
1958 * If ge search and we went off the end of the block, but it's
1959 * not the last block, we're in the wrong block.
1961 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_RIGHTSIB
);
1962 if (dir
== XFS_LOOKUP_GE
&&
1963 keyno
> xfs_btree_get_numrecs(block
) &&
1964 !xfs_btree_ptr_is_null(cur
, &ptr
)) {
1967 cur
->bc_ptrs
[0] = keyno
;
1968 error
= xfs_btree_increment(cur
, 0, &i
);
1971 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1975 } else if (dir
== XFS_LOOKUP_LE
&& diff
> 0)
1977 cur
->bc_ptrs
[0] = keyno
;
1979 /* Return if we succeeded or not. */
1980 if (keyno
== 0 || keyno
> xfs_btree_get_numrecs(block
))
1982 else if (dir
!= XFS_LOOKUP_EQ
|| diff
== 0)
1992 /* Find the high key storage area from a regular key. */
1993 union xfs_btree_key
*
1994 xfs_btree_high_key_from_key(
1995 struct xfs_btree_cur
*cur
,
1996 union xfs_btree_key
*key
)
1998 ASSERT(cur
->bc_flags
& XFS_BTREE_OVERLAPPING
);
1999 return (union xfs_btree_key
*)((char *)key
+
2000 (cur
->bc_ops
->key_len
/ 2));
2003 /* Determine the low (and high if overlapped) keys of a leaf block */
2005 xfs_btree_get_leaf_keys(
2006 struct xfs_btree_cur
*cur
,
2007 struct xfs_btree_block
*block
,
2008 union xfs_btree_key
*key
)
2010 union xfs_btree_key max_hkey
;
2011 union xfs_btree_key hkey
;
2012 union xfs_btree_rec
*rec
;
2013 union xfs_btree_key
*high
;
2016 rec
= xfs_btree_rec_addr(cur
, 1, block
);
2017 cur
->bc_ops
->init_key_from_rec(key
, rec
);
2019 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2021 cur
->bc_ops
->init_high_key_from_rec(&max_hkey
, rec
);
2022 for (n
= 2; n
<= xfs_btree_get_numrecs(block
); n
++) {
2023 rec
= xfs_btree_rec_addr(cur
, n
, block
);
2024 cur
->bc_ops
->init_high_key_from_rec(&hkey
, rec
);
2025 if (cur
->bc_ops
->diff_two_keys(cur
, &hkey
, &max_hkey
)
2030 high
= xfs_btree_high_key_from_key(cur
, key
);
2031 memcpy(high
, &max_hkey
, cur
->bc_ops
->key_len
/ 2);
2035 /* Determine the low (and high if overlapped) keys of a node block */
2037 xfs_btree_get_node_keys(
2038 struct xfs_btree_cur
*cur
,
2039 struct xfs_btree_block
*block
,
2040 union xfs_btree_key
*key
)
2042 union xfs_btree_key
*hkey
;
2043 union xfs_btree_key
*max_hkey
;
2044 union xfs_btree_key
*high
;
2047 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2048 memcpy(key
, xfs_btree_key_addr(cur
, 1, block
),
2049 cur
->bc_ops
->key_len
/ 2);
2051 max_hkey
= xfs_btree_high_key_addr(cur
, 1, block
);
2052 for (n
= 2; n
<= xfs_btree_get_numrecs(block
); n
++) {
2053 hkey
= xfs_btree_high_key_addr(cur
, n
, block
);
2054 if (cur
->bc_ops
->diff_two_keys(cur
, hkey
, max_hkey
) > 0)
2058 high
= xfs_btree_high_key_from_key(cur
, key
);
2059 memcpy(high
, max_hkey
, cur
->bc_ops
->key_len
/ 2);
2061 memcpy(key
, xfs_btree_key_addr(cur
, 1, block
),
2062 cur
->bc_ops
->key_len
);
2066 /* Derive the keys for any btree block. */
2069 struct xfs_btree_cur
*cur
,
2070 struct xfs_btree_block
*block
,
2071 union xfs_btree_key
*key
)
2073 if (be16_to_cpu(block
->bb_level
) == 0)
2074 xfs_btree_get_leaf_keys(cur
, block
, key
);
2076 xfs_btree_get_node_keys(cur
, block
, key
);
2080 * Decide if we need to update the parent keys of a btree block. For
2081 * a standard btree this is only necessary if we're updating the first
2082 * record/key. For an overlapping btree, we must always update the
2083 * keys because the highest key can be in any of the records or keys
2087 xfs_btree_needs_key_update(
2088 struct xfs_btree_cur
*cur
,
2091 return (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) || ptr
== 1;
2095 * Update the low and high parent keys of the given level, progressing
2096 * towards the root. If force_all is false, stop if the keys for a given
2097 * level do not need updating.
2100 __xfs_btree_updkeys(
2101 struct xfs_btree_cur
*cur
,
2103 struct xfs_btree_block
*block
,
2104 struct xfs_buf
*bp0
,
2107 union xfs_btree_key key
; /* keys from current level */
2108 union xfs_btree_key
*lkey
; /* keys from the next level up */
2109 union xfs_btree_key
*hkey
;
2110 union xfs_btree_key
*nlkey
; /* keys from the next level up */
2111 union xfs_btree_key
*nhkey
;
2115 ASSERT(cur
->bc_flags
& XFS_BTREE_OVERLAPPING
);
2117 /* Exit if there aren't any parent levels to update. */
2118 if (level
+ 1 >= cur
->bc_nlevels
)
2121 trace_xfs_btree_updkeys(cur
, level
, bp0
);
2124 hkey
= xfs_btree_high_key_from_key(cur
, lkey
);
2125 xfs_btree_get_keys(cur
, block
, lkey
);
2126 for (level
++; level
< cur
->bc_nlevels
; level
++) {
2130 block
= xfs_btree_get_block(cur
, level
, &bp
);
2131 trace_xfs_btree_updkeys(cur
, level
, bp
);
2133 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
2137 ptr
= cur
->bc_ptrs
[level
];
2138 nlkey
= xfs_btree_key_addr(cur
, ptr
, block
);
2139 nhkey
= xfs_btree_high_key_addr(cur
, ptr
, block
);
2141 !(cur
->bc_ops
->diff_two_keys(cur
, nlkey
, lkey
) != 0 ||
2142 cur
->bc_ops
->diff_two_keys(cur
, nhkey
, hkey
) != 0))
2144 xfs_btree_copy_keys(cur
, nlkey
, lkey
, 1);
2145 xfs_btree_log_keys(cur
, bp
, ptr
, ptr
);
2146 if (level
+ 1 >= cur
->bc_nlevels
)
2148 xfs_btree_get_node_keys(cur
, block
, lkey
);
2154 /* Update all the keys from some level in cursor back to the root. */
2156 xfs_btree_updkeys_force(
2157 struct xfs_btree_cur
*cur
,
2161 struct xfs_btree_block
*block
;
2163 block
= xfs_btree_get_block(cur
, level
, &bp
);
2164 return __xfs_btree_updkeys(cur
, level
, block
, bp
, true);
2168 * Update the parent keys of the given level, progressing towards the root.
2171 xfs_btree_update_keys(
2172 struct xfs_btree_cur
*cur
,
2175 struct xfs_btree_block
*block
;
2177 union xfs_btree_key
*kp
;
2178 union xfs_btree_key key
;
2183 block
= xfs_btree_get_block(cur
, level
, &bp
);
2184 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
)
2185 return __xfs_btree_updkeys(cur
, level
, block
, bp
, false);
2188 * Go up the tree from this level toward the root.
2189 * At each level, update the key value to the value input.
2190 * Stop when we reach a level where the cursor isn't pointing
2191 * at the first entry in the block.
2193 xfs_btree_get_keys(cur
, block
, &key
);
2194 for (level
++, ptr
= 1; ptr
== 1 && level
< cur
->bc_nlevels
; level
++) {
2198 block
= xfs_btree_get_block(cur
, level
, &bp
);
2200 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
2204 ptr
= cur
->bc_ptrs
[level
];
2205 kp
= xfs_btree_key_addr(cur
, ptr
, block
);
2206 xfs_btree_copy_keys(cur
, kp
, &key
, 1);
2207 xfs_btree_log_keys(cur
, bp
, ptr
, ptr
);
2214 * Update the record referred to by cur to the value in the
2215 * given record. This either works (return 0) or gets an
2216 * EFSCORRUPTED error.
2220 struct xfs_btree_cur
*cur
,
2221 union xfs_btree_rec
*rec
)
2223 struct xfs_btree_block
*block
;
2227 union xfs_btree_rec
*rp
;
2229 /* Pick up the current block. */
2230 block
= xfs_btree_get_block(cur
, 0, &bp
);
2233 error
= xfs_btree_check_block(cur
, block
, 0, bp
);
2237 /* Get the address of the rec to be updated. */
2238 ptr
= cur
->bc_ptrs
[0];
2239 rp
= xfs_btree_rec_addr(cur
, ptr
, block
);
2241 /* Fill in the new contents and log them. */
2242 xfs_btree_copy_recs(cur
, rp
, rec
, 1);
2243 xfs_btree_log_recs(cur
, bp
, ptr
, ptr
);
2246 * If we are tracking the last record in the tree and
2247 * we are at the far right edge of the tree, update it.
2249 if (xfs_btree_is_lastrec(cur
, block
, 0)) {
2250 cur
->bc_ops
->update_lastrec(cur
, block
, rec
,
2251 ptr
, LASTREC_UPDATE
);
2254 /* Pass new key value up to our parent. */
2255 if (xfs_btree_needs_key_update(cur
, ptr
)) {
2256 error
= xfs_btree_update_keys(cur
, 0);
2268 * Move 1 record left from cur/level if possible.
2269 * Update cur to reflect the new path.
2271 STATIC
int /* error */
2273 struct xfs_btree_cur
*cur
,
2275 int *stat
) /* success/failure */
2277 struct xfs_buf
*lbp
; /* left buffer pointer */
2278 struct xfs_btree_block
*left
; /* left btree block */
2279 int lrecs
; /* left record count */
2280 struct xfs_buf
*rbp
; /* right buffer pointer */
2281 struct xfs_btree_block
*right
; /* right btree block */
2282 struct xfs_btree_cur
*tcur
; /* temporary btree cursor */
2283 int rrecs
; /* right record count */
2284 union xfs_btree_ptr lptr
; /* left btree pointer */
2285 union xfs_btree_key
*rkp
= NULL
; /* right btree key */
2286 union xfs_btree_ptr
*rpp
= NULL
; /* right address pointer */
2287 union xfs_btree_rec
*rrp
= NULL
; /* right record pointer */
2288 int error
; /* error return value */
2291 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
2292 level
== cur
->bc_nlevels
- 1)
2295 /* Set up variables for this block as "right". */
2296 right
= xfs_btree_get_block(cur
, level
, &rbp
);
2299 error
= xfs_btree_check_block(cur
, right
, level
, rbp
);
2304 /* If we've got no left sibling then we can't shift an entry left. */
2305 xfs_btree_get_sibling(cur
, right
, &lptr
, XFS_BB_LEFTSIB
);
2306 if (xfs_btree_ptr_is_null(cur
, &lptr
))
2310 * If the cursor entry is the one that would be moved, don't
2311 * do it... it's too complicated.
2313 if (cur
->bc_ptrs
[level
] <= 1)
2316 /* Set up the left neighbor as "left". */
2317 error
= xfs_btree_read_buf_block(cur
, &lptr
, 0, &left
, &lbp
);
2321 /* If it's full, it can't take another entry. */
2322 lrecs
= xfs_btree_get_numrecs(left
);
2323 if (lrecs
== cur
->bc_ops
->get_maxrecs(cur
, level
))
2326 rrecs
= xfs_btree_get_numrecs(right
);
2329 * We add one entry to the left side and remove one for the right side.
2330 * Account for it here, the changes will be updated on disk and logged
2336 XFS_BTREE_STATS_INC(cur
, lshift
);
2337 XFS_BTREE_STATS_ADD(cur
, moves
, 1);
2340 * If non-leaf, copy a key and a ptr to the left block.
2341 * Log the changes to the left block.
2344 /* It's a non-leaf. Move keys and pointers. */
2345 union xfs_btree_key
*lkp
; /* left btree key */
2346 union xfs_btree_ptr
*lpp
; /* left address pointer */
2348 lkp
= xfs_btree_key_addr(cur
, lrecs
, left
);
2349 rkp
= xfs_btree_key_addr(cur
, 1, right
);
2351 lpp
= xfs_btree_ptr_addr(cur
, lrecs
, left
);
2352 rpp
= xfs_btree_ptr_addr(cur
, 1, right
);
2354 error
= xfs_btree_check_ptr(cur
, rpp
, 0, level
);
2358 xfs_btree_copy_keys(cur
, lkp
, rkp
, 1);
2359 xfs_btree_copy_ptrs(cur
, lpp
, rpp
, 1);
2361 xfs_btree_log_keys(cur
, lbp
, lrecs
, lrecs
);
2362 xfs_btree_log_ptrs(cur
, lbp
, lrecs
, lrecs
);
2364 ASSERT(cur
->bc_ops
->keys_inorder(cur
,
2365 xfs_btree_key_addr(cur
, lrecs
- 1, left
), lkp
));
2367 /* It's a leaf. Move records. */
2368 union xfs_btree_rec
*lrp
; /* left record pointer */
2370 lrp
= xfs_btree_rec_addr(cur
, lrecs
, left
);
2371 rrp
= xfs_btree_rec_addr(cur
, 1, right
);
2373 xfs_btree_copy_recs(cur
, lrp
, rrp
, 1);
2374 xfs_btree_log_recs(cur
, lbp
, lrecs
, lrecs
);
2376 ASSERT(cur
->bc_ops
->recs_inorder(cur
,
2377 xfs_btree_rec_addr(cur
, lrecs
- 1, left
), lrp
));
2380 xfs_btree_set_numrecs(left
, lrecs
);
2381 xfs_btree_log_block(cur
, lbp
, XFS_BB_NUMRECS
);
2383 xfs_btree_set_numrecs(right
, rrecs
);
2384 xfs_btree_log_block(cur
, rbp
, XFS_BB_NUMRECS
);
2387 * Slide the contents of right down one entry.
2389 XFS_BTREE_STATS_ADD(cur
, moves
, rrecs
- 1);
2391 /* It's a nonleaf. operate on keys and ptrs */
2393 int i
; /* loop index */
2395 for (i
= 0; i
< rrecs
; i
++) {
2396 error
= xfs_btree_check_ptr(cur
, rpp
, i
+ 1, level
);
2401 xfs_btree_shift_keys(cur
,
2402 xfs_btree_key_addr(cur
, 2, right
),
2404 xfs_btree_shift_ptrs(cur
,
2405 xfs_btree_ptr_addr(cur
, 2, right
),
2408 xfs_btree_log_keys(cur
, rbp
, 1, rrecs
);
2409 xfs_btree_log_ptrs(cur
, rbp
, 1, rrecs
);
2411 /* It's a leaf. operate on records */
2412 xfs_btree_shift_recs(cur
,
2413 xfs_btree_rec_addr(cur
, 2, right
),
2415 xfs_btree_log_recs(cur
, rbp
, 1, rrecs
);
2419 * Using a temporary cursor, update the parent key values of the
2420 * block on the left.
2422 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2423 error
= xfs_btree_dup_cursor(cur
, &tcur
);
2426 i
= xfs_btree_firstrec(tcur
, level
);
2427 XFS_WANT_CORRUPTED_GOTO(tcur
->bc_mp
, i
== 1, error0
);
2429 error
= xfs_btree_decrement(tcur
, level
, &i
);
2433 /* Update the parent high keys of the left block, if needed. */
2434 error
= xfs_btree_update_keys(tcur
, level
);
2438 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
2441 /* Update the parent keys of the right block. */
2442 error
= xfs_btree_update_keys(cur
, level
);
2446 /* Slide the cursor value left one. */
2447 cur
->bc_ptrs
[level
]--;
2460 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
2465 * Move 1 record right from cur/level if possible.
2466 * Update cur to reflect the new path.
2468 STATIC
int /* error */
2470 struct xfs_btree_cur
*cur
,
2472 int *stat
) /* success/failure */
2474 struct xfs_buf
*lbp
; /* left buffer pointer */
2475 struct xfs_btree_block
*left
; /* left btree block */
2476 struct xfs_buf
*rbp
; /* right buffer pointer */
2477 struct xfs_btree_block
*right
; /* right btree block */
2478 struct xfs_btree_cur
*tcur
; /* temporary btree cursor */
2479 union xfs_btree_ptr rptr
; /* right block pointer */
2480 union xfs_btree_key
*rkp
; /* right btree key */
2481 int rrecs
; /* right record count */
2482 int lrecs
; /* left record count */
2483 int error
; /* error return value */
2484 int i
; /* loop counter */
2486 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
2487 (level
== cur
->bc_nlevels
- 1))
2490 /* Set up variables for this block as "left". */
2491 left
= xfs_btree_get_block(cur
, level
, &lbp
);
2494 error
= xfs_btree_check_block(cur
, left
, level
, lbp
);
2499 /* If we've got no right sibling then we can't shift an entry right. */
2500 xfs_btree_get_sibling(cur
, left
, &rptr
, XFS_BB_RIGHTSIB
);
2501 if (xfs_btree_ptr_is_null(cur
, &rptr
))
2505 * If the cursor entry is the one that would be moved, don't
2506 * do it... it's too complicated.
2508 lrecs
= xfs_btree_get_numrecs(left
);
2509 if (cur
->bc_ptrs
[level
] >= lrecs
)
2512 /* Set up the right neighbor as "right". */
2513 error
= xfs_btree_read_buf_block(cur
, &rptr
, 0, &right
, &rbp
);
2517 /* If it's full, it can't take another entry. */
2518 rrecs
= xfs_btree_get_numrecs(right
);
2519 if (rrecs
== cur
->bc_ops
->get_maxrecs(cur
, level
))
2522 XFS_BTREE_STATS_INC(cur
, rshift
);
2523 XFS_BTREE_STATS_ADD(cur
, moves
, rrecs
);
2526 * Make a hole at the start of the right neighbor block, then
2527 * copy the last left block entry to the hole.
2530 /* It's a nonleaf. make a hole in the keys and ptrs */
2531 union xfs_btree_key
*lkp
;
2532 union xfs_btree_ptr
*lpp
;
2533 union xfs_btree_ptr
*rpp
;
2535 lkp
= xfs_btree_key_addr(cur
, lrecs
, left
);
2536 lpp
= xfs_btree_ptr_addr(cur
, lrecs
, left
);
2537 rkp
= xfs_btree_key_addr(cur
, 1, right
);
2538 rpp
= xfs_btree_ptr_addr(cur
, 1, right
);
2541 for (i
= rrecs
- 1; i
>= 0; i
--) {
2542 error
= xfs_btree_check_ptr(cur
, rpp
, i
, level
);
2548 xfs_btree_shift_keys(cur
, rkp
, 1, rrecs
);
2549 xfs_btree_shift_ptrs(cur
, rpp
, 1, rrecs
);
2552 error
= xfs_btree_check_ptr(cur
, lpp
, 0, level
);
2557 /* Now put the new data in, and log it. */
2558 xfs_btree_copy_keys(cur
, rkp
, lkp
, 1);
2559 xfs_btree_copy_ptrs(cur
, rpp
, lpp
, 1);
2561 xfs_btree_log_keys(cur
, rbp
, 1, rrecs
+ 1);
2562 xfs_btree_log_ptrs(cur
, rbp
, 1, rrecs
+ 1);
2564 ASSERT(cur
->bc_ops
->keys_inorder(cur
, rkp
,
2565 xfs_btree_key_addr(cur
, 2, right
)));
2567 /* It's a leaf. make a hole in the records */
2568 union xfs_btree_rec
*lrp
;
2569 union xfs_btree_rec
*rrp
;
2571 lrp
= xfs_btree_rec_addr(cur
, lrecs
, left
);
2572 rrp
= xfs_btree_rec_addr(cur
, 1, right
);
2574 xfs_btree_shift_recs(cur
, rrp
, 1, rrecs
);
2576 /* Now put the new data in, and log it. */
2577 xfs_btree_copy_recs(cur
, rrp
, lrp
, 1);
2578 xfs_btree_log_recs(cur
, rbp
, 1, rrecs
+ 1);
2582 * Decrement and log left's numrecs, bump and log right's numrecs.
2584 xfs_btree_set_numrecs(left
, --lrecs
);
2585 xfs_btree_log_block(cur
, lbp
, XFS_BB_NUMRECS
);
2587 xfs_btree_set_numrecs(right
, ++rrecs
);
2588 xfs_btree_log_block(cur
, rbp
, XFS_BB_NUMRECS
);
2591 * Using a temporary cursor, update the parent key values of the
2592 * block on the right.
2594 error
= xfs_btree_dup_cursor(cur
, &tcur
);
2597 i
= xfs_btree_lastrec(tcur
, level
);
2598 XFS_WANT_CORRUPTED_GOTO(tcur
->bc_mp
, i
== 1, error0
);
2600 error
= xfs_btree_increment(tcur
, level
, &i
);
2604 /* Update the parent high keys of the left block, if needed. */
2605 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2606 error
= xfs_btree_update_keys(cur
, level
);
2611 /* Update the parent keys of the right block. */
2612 error
= xfs_btree_update_keys(tcur
, level
);
2616 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
2629 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
2634 * Split cur/level block in half.
2635 * Return new block number and the key to its first
2636 * record (to be inserted into parent).
2638 STATIC
int /* error */
2640 struct xfs_btree_cur
*cur
,
2642 union xfs_btree_ptr
*ptrp
,
2643 union xfs_btree_key
*key
,
2644 struct xfs_btree_cur
**curp
,
2645 int *stat
) /* success/failure */
2647 union xfs_btree_ptr lptr
; /* left sibling block ptr */
2648 struct xfs_buf
*lbp
; /* left buffer pointer */
2649 struct xfs_btree_block
*left
; /* left btree block */
2650 union xfs_btree_ptr rptr
; /* right sibling block ptr */
2651 struct xfs_buf
*rbp
; /* right buffer pointer */
2652 struct xfs_btree_block
*right
; /* right btree block */
2653 union xfs_btree_ptr rrptr
; /* right-right sibling ptr */
2654 struct xfs_buf
*rrbp
; /* right-right buffer pointer */
2655 struct xfs_btree_block
*rrblock
; /* right-right btree block */
2659 int error
; /* error return value */
2664 XFS_BTREE_STATS_INC(cur
, split
);
2666 /* Set up left block (current one). */
2667 left
= xfs_btree_get_block(cur
, level
, &lbp
);
2670 error
= xfs_btree_check_block(cur
, left
, level
, lbp
);
2675 xfs_btree_buf_to_ptr(cur
, lbp
, &lptr
);
2677 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2678 error
= cur
->bc_ops
->alloc_block(cur
, &lptr
, &rptr
, stat
);
2683 XFS_BTREE_STATS_INC(cur
, alloc
);
2685 /* Set up the new block as "right". */
2686 error
= xfs_btree_get_buf_block(cur
, &rptr
, 0, &right
, &rbp
);
2690 /* Fill in the btree header for the new right block. */
2691 xfs_btree_init_block_cur(cur
, rbp
, xfs_btree_get_level(left
), 0);
2694 * Split the entries between the old and the new block evenly.
2695 * Make sure that if there's an odd number of entries now, that
2696 * each new block will have the same number of entries.
2698 lrecs
= xfs_btree_get_numrecs(left
);
2700 if ((lrecs
& 1) && cur
->bc_ptrs
[level
] <= rrecs
+ 1)
2702 src_index
= (lrecs
- rrecs
+ 1);
2704 XFS_BTREE_STATS_ADD(cur
, moves
, rrecs
);
2706 /* Adjust numrecs for the later get_*_keys() calls. */
2708 xfs_btree_set_numrecs(left
, lrecs
);
2709 xfs_btree_set_numrecs(right
, xfs_btree_get_numrecs(right
) + rrecs
);
2712 * Copy btree block entries from the left block over to the
2713 * new block, the right. Update the right block and log the
2717 /* It's a non-leaf. Move keys and pointers. */
2718 union xfs_btree_key
*lkp
; /* left btree key */
2719 union xfs_btree_ptr
*lpp
; /* left address pointer */
2720 union xfs_btree_key
*rkp
; /* right btree key */
2721 union xfs_btree_ptr
*rpp
; /* right address pointer */
2723 lkp
= xfs_btree_key_addr(cur
, src_index
, left
);
2724 lpp
= xfs_btree_ptr_addr(cur
, src_index
, left
);
2725 rkp
= xfs_btree_key_addr(cur
, 1, right
);
2726 rpp
= xfs_btree_ptr_addr(cur
, 1, right
);
2729 for (i
= src_index
; i
< rrecs
; i
++) {
2730 error
= xfs_btree_check_ptr(cur
, lpp
, i
, level
);
2736 /* Copy the keys & pointers to the new block. */
2737 xfs_btree_copy_keys(cur
, rkp
, lkp
, rrecs
);
2738 xfs_btree_copy_ptrs(cur
, rpp
, lpp
, rrecs
);
2740 xfs_btree_log_keys(cur
, rbp
, 1, rrecs
);
2741 xfs_btree_log_ptrs(cur
, rbp
, 1, rrecs
);
2743 /* Stash the keys of the new block for later insertion. */
2744 xfs_btree_get_node_keys(cur
, right
, key
);
2746 /* It's a leaf. Move records. */
2747 union xfs_btree_rec
*lrp
; /* left record pointer */
2748 union xfs_btree_rec
*rrp
; /* right record pointer */
2750 lrp
= xfs_btree_rec_addr(cur
, src_index
, left
);
2751 rrp
= xfs_btree_rec_addr(cur
, 1, right
);
2753 /* Copy records to the new block. */
2754 xfs_btree_copy_recs(cur
, rrp
, lrp
, rrecs
);
2755 xfs_btree_log_recs(cur
, rbp
, 1, rrecs
);
2757 /* Stash the keys of the new block for later insertion. */
2758 xfs_btree_get_leaf_keys(cur
, right
, key
);
2762 * Find the left block number by looking in the buffer.
2763 * Adjust sibling pointers.
2765 xfs_btree_get_sibling(cur
, left
, &rrptr
, XFS_BB_RIGHTSIB
);
2766 xfs_btree_set_sibling(cur
, right
, &rrptr
, XFS_BB_RIGHTSIB
);
2767 xfs_btree_set_sibling(cur
, right
, &lptr
, XFS_BB_LEFTSIB
);
2768 xfs_btree_set_sibling(cur
, left
, &rptr
, XFS_BB_RIGHTSIB
);
2770 xfs_btree_log_block(cur
, rbp
, XFS_BB_ALL_BITS
);
2771 xfs_btree_log_block(cur
, lbp
, XFS_BB_NUMRECS
| XFS_BB_RIGHTSIB
);
2774 * If there's a block to the new block's right, make that block
2775 * point back to right instead of to left.
2777 if (!xfs_btree_ptr_is_null(cur
, &rrptr
)) {
2778 error
= xfs_btree_read_buf_block(cur
, &rrptr
,
2779 0, &rrblock
, &rrbp
);
2782 xfs_btree_set_sibling(cur
, rrblock
, &rptr
, XFS_BB_LEFTSIB
);
2783 xfs_btree_log_block(cur
, rrbp
, XFS_BB_LEFTSIB
);
2786 /* Update the parent high keys of the left block, if needed. */
2787 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2788 error
= xfs_btree_update_keys(cur
, level
);
2794 * If the cursor is really in the right block, move it there.
2795 * If it's just pointing past the last entry in left, then we'll
2796 * insert there, so don't change anything in that case.
2798 if (cur
->bc_ptrs
[level
] > lrecs
+ 1) {
2799 xfs_btree_setbuf(cur
, level
, rbp
);
2800 cur
->bc_ptrs
[level
] -= lrecs
;
2803 * If there are more levels, we'll need another cursor which refers
2804 * the right block, no matter where this cursor was.
2806 if (level
+ 1 < cur
->bc_nlevels
) {
2807 error
= xfs_btree_dup_cursor(cur
, curp
);
2810 (*curp
)->bc_ptrs
[level
+ 1]++;
2824 struct xfs_btree_split_args
{
2825 struct xfs_btree_cur
*cur
;
2827 union xfs_btree_ptr
*ptrp
;
2828 union xfs_btree_key
*key
;
2829 struct xfs_btree_cur
**curp
;
2830 int *stat
; /* success/failure */
2832 bool kswapd
; /* allocation in kswapd context */
2833 struct completion
*done
;
2834 struct work_struct work
;
2838 * Stack switching interfaces for allocation
2841 xfs_btree_split_worker(
2842 struct work_struct
*work
)
2844 struct xfs_btree_split_args
*args
= container_of(work
,
2845 struct xfs_btree_split_args
, work
);
2846 unsigned long pflags
;
2847 unsigned long new_pflags
= PF_MEMALLOC_NOFS
;
2850 * we are in a transaction context here, but may also be doing work
2851 * in kswapd context, and hence we may need to inherit that state
2852 * temporarily to ensure that we don't block waiting for memory reclaim
2856 new_pflags
|= PF_MEMALLOC
| PF_SWAPWRITE
| PF_KSWAPD
;
2858 current_set_flags_nested(&pflags
, new_pflags
);
2860 args
->result
= __xfs_btree_split(args
->cur
, args
->level
, args
->ptrp
,
2861 args
->key
, args
->curp
, args
->stat
);
2862 complete(args
->done
);
2864 current_restore_flags_nested(&pflags
, new_pflags
);
2868 * BMBT split requests often come in with little stack to work on. Push
2869 * them off to a worker thread so there is lots of stack to use. For the other
2870 * btree types, just call directly to avoid the context switch overhead here.
2872 STATIC
int /* error */
2874 struct xfs_btree_cur
*cur
,
2876 union xfs_btree_ptr
*ptrp
,
2877 union xfs_btree_key
*key
,
2878 struct xfs_btree_cur
**curp
,
2879 int *stat
) /* success/failure */
2881 struct xfs_btree_split_args args
;
2882 DECLARE_COMPLETION_ONSTACK(done
);
2884 if (cur
->bc_btnum
!= XFS_BTNUM_BMAP
)
2885 return __xfs_btree_split(cur
, level
, ptrp
, key
, curp
, stat
);
2894 args
.kswapd
= current_is_kswapd();
2895 INIT_WORK_ONSTACK(&args
.work
, xfs_btree_split_worker
);
2896 queue_work(xfs_alloc_wq
, &args
.work
);
2897 wait_for_completion(&done
);
2898 destroy_work_on_stack(&args
.work
);
2902 #define xfs_btree_split __xfs_btree_split
2907 * Copy the old inode root contents into a real block and make the
2908 * broot point to it.
2911 xfs_btree_new_iroot(
2912 struct xfs_btree_cur
*cur
, /* btree cursor */
2913 int *logflags
, /* logging flags for inode */
2914 int *stat
) /* return status - 0 fail */
2916 struct xfs_buf
*cbp
; /* buffer for cblock */
2917 struct xfs_btree_block
*block
; /* btree block */
2918 struct xfs_btree_block
*cblock
; /* child btree block */
2919 union xfs_btree_key
*ckp
; /* child key pointer */
2920 union xfs_btree_ptr
*cpp
; /* child ptr pointer */
2921 union xfs_btree_key
*kp
; /* pointer to btree key */
2922 union xfs_btree_ptr
*pp
; /* pointer to block addr */
2923 union xfs_btree_ptr nptr
; /* new block addr */
2924 int level
; /* btree level */
2925 int error
; /* error return code */
2927 int i
; /* loop counter */
2930 XFS_BTREE_STATS_INC(cur
, newroot
);
2932 ASSERT(cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
);
2934 level
= cur
->bc_nlevels
- 1;
2936 block
= xfs_btree_get_iroot(cur
);
2937 pp
= xfs_btree_ptr_addr(cur
, 1, block
);
2939 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2940 error
= cur
->bc_ops
->alloc_block(cur
, pp
, &nptr
, stat
);
2946 XFS_BTREE_STATS_INC(cur
, alloc
);
2948 /* Copy the root into a real block. */
2949 error
= xfs_btree_get_buf_block(cur
, &nptr
, 0, &cblock
, &cbp
);
2954 * we can't just memcpy() the root in for CRC enabled btree blocks.
2955 * In that case have to also ensure the blkno remains correct
2957 memcpy(cblock
, block
, xfs_btree_block_len(cur
));
2958 if (cur
->bc_flags
& XFS_BTREE_CRC_BLOCKS
) {
2959 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
2960 cblock
->bb_u
.l
.bb_blkno
= cpu_to_be64(cbp
->b_bn
);
2962 cblock
->bb_u
.s
.bb_blkno
= cpu_to_be64(cbp
->b_bn
);
2965 be16_add_cpu(&block
->bb_level
, 1);
2966 xfs_btree_set_numrecs(block
, 1);
2968 cur
->bc_ptrs
[level
+ 1] = 1;
2970 kp
= xfs_btree_key_addr(cur
, 1, block
);
2971 ckp
= xfs_btree_key_addr(cur
, 1, cblock
);
2972 xfs_btree_copy_keys(cur
, ckp
, kp
, xfs_btree_get_numrecs(cblock
));
2974 cpp
= xfs_btree_ptr_addr(cur
, 1, cblock
);
2976 for (i
= 0; i
< be16_to_cpu(cblock
->bb_numrecs
); i
++) {
2977 error
= xfs_btree_check_ptr(cur
, pp
, i
, level
);
2982 xfs_btree_copy_ptrs(cur
, cpp
, pp
, xfs_btree_get_numrecs(cblock
));
2985 error
= xfs_btree_check_ptr(cur
, &nptr
, 0, level
);
2989 xfs_btree_copy_ptrs(cur
, pp
, &nptr
, 1);
2991 xfs_iroot_realloc(cur
->bc_private
.b
.ip
,
2992 1 - xfs_btree_get_numrecs(cblock
),
2993 cur
->bc_private
.b
.whichfork
);
2995 xfs_btree_setbuf(cur
, level
, cbp
);
2998 * Do all this logging at the end so that
2999 * the root is at the right level.
3001 xfs_btree_log_block(cur
, cbp
, XFS_BB_ALL_BITS
);
3002 xfs_btree_log_keys(cur
, cbp
, 1, be16_to_cpu(cblock
->bb_numrecs
));
3003 xfs_btree_log_ptrs(cur
, cbp
, 1, be16_to_cpu(cblock
->bb_numrecs
));
3006 XFS_ILOG_CORE
| xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
);
3014 * Allocate a new root block, fill it in.
3016 STATIC
int /* error */
3018 struct xfs_btree_cur
*cur
, /* btree cursor */
3019 int *stat
) /* success/failure */
3021 struct xfs_btree_block
*block
; /* one half of the old root block */
3022 struct xfs_buf
*bp
; /* buffer containing block */
3023 int error
; /* error return value */
3024 struct xfs_buf
*lbp
; /* left buffer pointer */
3025 struct xfs_btree_block
*left
; /* left btree block */
3026 struct xfs_buf
*nbp
; /* new (root) buffer */
3027 struct xfs_btree_block
*new; /* new (root) btree block */
3028 int nptr
; /* new value for key index, 1 or 2 */
3029 struct xfs_buf
*rbp
; /* right buffer pointer */
3030 struct xfs_btree_block
*right
; /* right btree block */
3031 union xfs_btree_ptr rptr
;
3032 union xfs_btree_ptr lptr
;
3034 XFS_BTREE_STATS_INC(cur
, newroot
);
3036 /* initialise our start point from the cursor */
3037 cur
->bc_ops
->init_ptr_from_cur(cur
, &rptr
);
3039 /* Allocate the new block. If we can't do it, we're toast. Give up. */
3040 error
= cur
->bc_ops
->alloc_block(cur
, &rptr
, &lptr
, stat
);
3045 XFS_BTREE_STATS_INC(cur
, alloc
);
3047 /* Set up the new block. */
3048 error
= xfs_btree_get_buf_block(cur
, &lptr
, 0, &new, &nbp
);
3052 /* Set the root in the holding structure increasing the level by 1. */
3053 cur
->bc_ops
->set_root(cur
, &lptr
, 1);
3056 * At the previous root level there are now two blocks: the old root,
3057 * and the new block generated when it was split. We don't know which
3058 * one the cursor is pointing at, so we set up variables "left" and
3059 * "right" for each case.
3061 block
= xfs_btree_get_block(cur
, cur
->bc_nlevels
- 1, &bp
);
3064 error
= xfs_btree_check_block(cur
, block
, cur
->bc_nlevels
- 1, bp
);
3069 xfs_btree_get_sibling(cur
, block
, &rptr
, XFS_BB_RIGHTSIB
);
3070 if (!xfs_btree_ptr_is_null(cur
, &rptr
)) {
3071 /* Our block is left, pick up the right block. */
3073 xfs_btree_buf_to_ptr(cur
, lbp
, &lptr
);
3075 error
= xfs_btree_read_buf_block(cur
, &rptr
, 0, &right
, &rbp
);
3081 /* Our block is right, pick up the left block. */
3083 xfs_btree_buf_to_ptr(cur
, rbp
, &rptr
);
3085 xfs_btree_get_sibling(cur
, right
, &lptr
, XFS_BB_LEFTSIB
);
3086 error
= xfs_btree_read_buf_block(cur
, &lptr
, 0, &left
, &lbp
);
3093 /* Fill in the new block's btree header and log it. */
3094 xfs_btree_init_block_cur(cur
, nbp
, cur
->bc_nlevels
, 2);
3095 xfs_btree_log_block(cur
, nbp
, XFS_BB_ALL_BITS
);
3096 ASSERT(!xfs_btree_ptr_is_null(cur
, &lptr
) &&
3097 !xfs_btree_ptr_is_null(cur
, &rptr
));
3099 /* Fill in the key data in the new root. */
3100 if (xfs_btree_get_level(left
) > 0) {
3102 * Get the keys for the left block's keys and put them directly
3103 * in the parent block. Do the same for the right block.
3105 xfs_btree_get_node_keys(cur
, left
,
3106 xfs_btree_key_addr(cur
, 1, new));
3107 xfs_btree_get_node_keys(cur
, right
,
3108 xfs_btree_key_addr(cur
, 2, new));
3111 * Get the keys for the left block's records and put them
3112 * directly in the parent block. Do the same for the right
3115 xfs_btree_get_leaf_keys(cur
, left
,
3116 xfs_btree_key_addr(cur
, 1, new));
3117 xfs_btree_get_leaf_keys(cur
, right
,
3118 xfs_btree_key_addr(cur
, 2, new));
3120 xfs_btree_log_keys(cur
, nbp
, 1, 2);
3122 /* Fill in the pointer data in the new root. */
3123 xfs_btree_copy_ptrs(cur
,
3124 xfs_btree_ptr_addr(cur
, 1, new), &lptr
, 1);
3125 xfs_btree_copy_ptrs(cur
,
3126 xfs_btree_ptr_addr(cur
, 2, new), &rptr
, 1);
3127 xfs_btree_log_ptrs(cur
, nbp
, 1, 2);
3129 /* Fix up the cursor. */
3130 xfs_btree_setbuf(cur
, cur
->bc_nlevels
, nbp
);
3131 cur
->bc_ptrs
[cur
->bc_nlevels
] = nptr
;
3143 xfs_btree_make_block_unfull(
3144 struct xfs_btree_cur
*cur
, /* btree cursor */
3145 int level
, /* btree level */
3146 int numrecs
,/* # of recs in block */
3147 int *oindex
,/* old tree index */
3148 int *index
, /* new tree index */
3149 union xfs_btree_ptr
*nptr
, /* new btree ptr */
3150 struct xfs_btree_cur
**ncur
, /* new btree cursor */
3151 union xfs_btree_key
*key
, /* key of new block */
3156 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
3157 level
== cur
->bc_nlevels
- 1) {
3158 struct xfs_inode
*ip
= cur
->bc_private
.b
.ip
;
3160 if (numrecs
< cur
->bc_ops
->get_dmaxrecs(cur
, level
)) {
3161 /* A root block that can be made bigger. */
3162 xfs_iroot_realloc(ip
, 1, cur
->bc_private
.b
.whichfork
);
3165 /* A root block that needs replacing */
3168 error
= xfs_btree_new_iroot(cur
, &logflags
, stat
);
3169 if (error
|| *stat
== 0)
3172 xfs_trans_log_inode(cur
->bc_tp
, ip
, logflags
);
3178 /* First, try shifting an entry to the right neighbor. */
3179 error
= xfs_btree_rshift(cur
, level
, stat
);
3183 /* Next, try shifting an entry to the left neighbor. */
3184 error
= xfs_btree_lshift(cur
, level
, stat
);
3189 *oindex
= *index
= cur
->bc_ptrs
[level
];
3194 * Next, try splitting the current block in half.
3196 * If this works we have to re-set our variables because we
3197 * could be in a different block now.
3199 error
= xfs_btree_split(cur
, level
, nptr
, key
, ncur
, stat
);
3200 if (error
|| *stat
== 0)
3204 *index
= cur
->bc_ptrs
[level
];
3209 * Insert one record/level. Return information to the caller
3210 * allowing the next level up to proceed if necessary.
3214 struct xfs_btree_cur
*cur
, /* btree cursor */
3215 int level
, /* level to insert record at */
3216 union xfs_btree_ptr
*ptrp
, /* i/o: block number inserted */
3217 union xfs_btree_rec
*rec
, /* record to insert */
3218 union xfs_btree_key
*key
, /* i/o: block key for ptrp */
3219 struct xfs_btree_cur
**curp
, /* output: new cursor replacing cur */
3220 int *stat
) /* success/failure */
3222 struct xfs_btree_block
*block
; /* btree block */
3223 struct xfs_buf
*bp
; /* buffer for block */
3224 union xfs_btree_ptr nptr
; /* new block ptr */
3225 struct xfs_btree_cur
*ncur
; /* new btree cursor */
3226 union xfs_btree_key nkey
; /* new block key */
3227 union xfs_btree_key
*lkey
;
3228 int optr
; /* old key/record index */
3229 int ptr
; /* key/record index */
3230 int numrecs
;/* number of records */
3231 int error
; /* error return value */
3241 * If we have an external root pointer, and we've made it to the
3242 * root level, allocate a new root block and we're done.
3244 if (!(cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
3245 (level
>= cur
->bc_nlevels
)) {
3246 error
= xfs_btree_new_root(cur
, stat
);
3247 xfs_btree_set_ptr_null(cur
, ptrp
);
3252 /* If we're off the left edge, return failure. */
3253 ptr
= cur
->bc_ptrs
[level
];
3261 XFS_BTREE_STATS_INC(cur
, insrec
);
3263 /* Get pointers to the btree buffer and block. */
3264 block
= xfs_btree_get_block(cur
, level
, &bp
);
3265 old_bn
= bp
? bp
->b_bn
: XFS_BUF_DADDR_NULL
;
3266 numrecs
= xfs_btree_get_numrecs(block
);
3269 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
3273 /* Check that the new entry is being inserted in the right place. */
3274 if (ptr
<= numrecs
) {
3276 ASSERT(cur
->bc_ops
->recs_inorder(cur
, rec
,
3277 xfs_btree_rec_addr(cur
, ptr
, block
)));
3279 ASSERT(cur
->bc_ops
->keys_inorder(cur
, key
,
3280 xfs_btree_key_addr(cur
, ptr
, block
)));
3286 * If the block is full, we can't insert the new entry until we
3287 * make the block un-full.
3289 xfs_btree_set_ptr_null(cur
, &nptr
);
3290 if (numrecs
== cur
->bc_ops
->get_maxrecs(cur
, level
)) {
3291 error
= xfs_btree_make_block_unfull(cur
, level
, numrecs
,
3292 &optr
, &ptr
, &nptr
, &ncur
, lkey
, stat
);
3293 if (error
|| *stat
== 0)
3298 * The current block may have changed if the block was
3299 * previously full and we have just made space in it.
3301 block
= xfs_btree_get_block(cur
, level
, &bp
);
3302 numrecs
= xfs_btree_get_numrecs(block
);
3305 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
3311 * At this point we know there's room for our new entry in the block
3312 * we're pointing at.
3314 XFS_BTREE_STATS_ADD(cur
, moves
, numrecs
- ptr
+ 1);
3317 /* It's a nonleaf. make a hole in the keys and ptrs */
3318 union xfs_btree_key
*kp
;
3319 union xfs_btree_ptr
*pp
;
3321 kp
= xfs_btree_key_addr(cur
, ptr
, block
);
3322 pp
= xfs_btree_ptr_addr(cur
, ptr
, block
);
3325 for (i
= numrecs
- ptr
; i
>= 0; i
--) {
3326 error
= xfs_btree_check_ptr(cur
, pp
, i
, level
);
3332 xfs_btree_shift_keys(cur
, kp
, 1, numrecs
- ptr
+ 1);
3333 xfs_btree_shift_ptrs(cur
, pp
, 1, numrecs
- ptr
+ 1);
3336 error
= xfs_btree_check_ptr(cur
, ptrp
, 0, level
);
3341 /* Now put the new data in, bump numrecs and log it. */
3342 xfs_btree_copy_keys(cur
, kp
, key
, 1);
3343 xfs_btree_copy_ptrs(cur
, pp
, ptrp
, 1);
3345 xfs_btree_set_numrecs(block
, numrecs
);
3346 xfs_btree_log_ptrs(cur
, bp
, ptr
, numrecs
);
3347 xfs_btree_log_keys(cur
, bp
, ptr
, numrecs
);
3349 if (ptr
< numrecs
) {
3350 ASSERT(cur
->bc_ops
->keys_inorder(cur
, kp
,
3351 xfs_btree_key_addr(cur
, ptr
+ 1, block
)));
3355 /* It's a leaf. make a hole in the records */
3356 union xfs_btree_rec
*rp
;
3358 rp
= xfs_btree_rec_addr(cur
, ptr
, block
);
3360 xfs_btree_shift_recs(cur
, rp
, 1, numrecs
- ptr
+ 1);
3362 /* Now put the new data in, bump numrecs and log it. */
3363 xfs_btree_copy_recs(cur
, rp
, rec
, 1);
3364 xfs_btree_set_numrecs(block
, ++numrecs
);
3365 xfs_btree_log_recs(cur
, bp
, ptr
, numrecs
);
3367 if (ptr
< numrecs
) {
3368 ASSERT(cur
->bc_ops
->recs_inorder(cur
, rp
,
3369 xfs_btree_rec_addr(cur
, ptr
+ 1, block
)));
3374 /* Log the new number of records in the btree header. */
3375 xfs_btree_log_block(cur
, bp
, XFS_BB_NUMRECS
);
3378 * If we just inserted into a new tree block, we have to
3379 * recalculate nkey here because nkey is out of date.
3381 * Otherwise we're just updating an existing block (having shoved
3382 * some records into the new tree block), so use the regular key
3385 if (bp
&& bp
->b_bn
!= old_bn
) {
3386 xfs_btree_get_keys(cur
, block
, lkey
);
3387 } else if (xfs_btree_needs_key_update(cur
, optr
)) {
3388 error
= xfs_btree_update_keys(cur
, level
);
3394 * If we are tracking the last record in the tree and
3395 * we are at the far right edge of the tree, update it.
3397 if (xfs_btree_is_lastrec(cur
, block
, level
)) {
3398 cur
->bc_ops
->update_lastrec(cur
, block
, rec
,
3399 ptr
, LASTREC_INSREC
);
3403 * Return the new block number, if any.
3404 * If there is one, give back a record value and a cursor too.
3407 if (!xfs_btree_ptr_is_null(cur
, &nptr
)) {
3408 xfs_btree_copy_keys(cur
, key
, lkey
, 1);
3420 * Insert the record at the point referenced by cur.
3422 * A multi-level split of the tree on insert will invalidate the original
3423 * cursor. All callers of this function should assume that the cursor is
3424 * no longer valid and revalidate it.
3428 struct xfs_btree_cur
*cur
,
3431 int error
; /* error return value */
3432 int i
; /* result value, 0 for failure */
3433 int level
; /* current level number in btree */
3434 union xfs_btree_ptr nptr
; /* new block number (split result) */
3435 struct xfs_btree_cur
*ncur
; /* new cursor (split result) */
3436 struct xfs_btree_cur
*pcur
; /* previous level's cursor */
3437 union xfs_btree_key bkey
; /* key of block to insert */
3438 union xfs_btree_key
*key
;
3439 union xfs_btree_rec rec
; /* record to insert */
3446 xfs_btree_set_ptr_null(cur
, &nptr
);
3448 /* Make a key out of the record data to be inserted, and save it. */
3449 cur
->bc_ops
->init_rec_from_cur(cur
, &rec
);
3450 cur
->bc_ops
->init_key_from_rec(key
, &rec
);
3453 * Loop going up the tree, starting at the leaf level.
3454 * Stop when we don't get a split block, that must mean that
3455 * the insert is finished with this level.
3459 * Insert nrec/nptr into this level of the tree.
3460 * Note if we fail, nptr will be null.
3462 error
= xfs_btree_insrec(pcur
, level
, &nptr
, &rec
, key
,
3466 xfs_btree_del_cursor(pcur
, XFS_BTREE_ERROR
);
3470 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3474 * See if the cursor we just used is trash.
3475 * Can't trash the caller's cursor, but otherwise we should
3476 * if ncur is a new cursor or we're about to be done.
3479 (ncur
|| xfs_btree_ptr_is_null(cur
, &nptr
))) {
3480 /* Save the state from the cursor before we trash it */
3481 if (cur
->bc_ops
->update_cursor
)
3482 cur
->bc_ops
->update_cursor(pcur
, cur
);
3483 cur
->bc_nlevels
= pcur
->bc_nlevels
;
3484 xfs_btree_del_cursor(pcur
, XFS_BTREE_NOERROR
);
3486 /* If we got a new cursor, switch to it. */
3491 } while (!xfs_btree_ptr_is_null(cur
, &nptr
));
3500 * Try to merge a non-leaf block back into the inode root.
3502 * Note: the killroot names comes from the fact that we're effectively
3503 * killing the old root block. But because we can't just delete the
3504 * inode we have to copy the single block it was pointing to into the
3508 xfs_btree_kill_iroot(
3509 struct xfs_btree_cur
*cur
)
3511 int whichfork
= cur
->bc_private
.b
.whichfork
;
3512 struct xfs_inode
*ip
= cur
->bc_private
.b
.ip
;
3513 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3514 struct xfs_btree_block
*block
;
3515 struct xfs_btree_block
*cblock
;
3516 union xfs_btree_key
*kp
;
3517 union xfs_btree_key
*ckp
;
3518 union xfs_btree_ptr
*pp
;
3519 union xfs_btree_ptr
*cpp
;
3520 struct xfs_buf
*cbp
;
3526 union xfs_btree_ptr ptr
;
3530 ASSERT(cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
);
3531 ASSERT(cur
->bc_nlevels
> 1);
3534 * Don't deal with the root block needs to be a leaf case.
3535 * We're just going to turn the thing back into extents anyway.
3537 level
= cur
->bc_nlevels
- 1;
3542 * Give up if the root has multiple children.
3544 block
= xfs_btree_get_iroot(cur
);
3545 if (xfs_btree_get_numrecs(block
) != 1)
3548 cblock
= xfs_btree_get_block(cur
, level
- 1, &cbp
);
3549 numrecs
= xfs_btree_get_numrecs(cblock
);
3552 * Only do this if the next level will fit.
3553 * Then the data must be copied up to the inode,
3554 * instead of freeing the root you free the next level.
3556 if (numrecs
> cur
->bc_ops
->get_dmaxrecs(cur
, level
))
3559 XFS_BTREE_STATS_INC(cur
, killroot
);
3562 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_LEFTSIB
);
3563 ASSERT(xfs_btree_ptr_is_null(cur
, &ptr
));
3564 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_RIGHTSIB
);
3565 ASSERT(xfs_btree_ptr_is_null(cur
, &ptr
));
3568 index
= numrecs
- cur
->bc_ops
->get_maxrecs(cur
, level
);
3570 xfs_iroot_realloc(cur
->bc_private
.b
.ip
, index
,
3571 cur
->bc_private
.b
.whichfork
);
3572 block
= ifp
->if_broot
;
3575 be16_add_cpu(&block
->bb_numrecs
, index
);
3576 ASSERT(block
->bb_numrecs
== cblock
->bb_numrecs
);
3578 kp
= xfs_btree_key_addr(cur
, 1, block
);
3579 ckp
= xfs_btree_key_addr(cur
, 1, cblock
);
3580 xfs_btree_copy_keys(cur
, kp
, ckp
, numrecs
);
3582 pp
= xfs_btree_ptr_addr(cur
, 1, block
);
3583 cpp
= xfs_btree_ptr_addr(cur
, 1, cblock
);
3585 for (i
= 0; i
< numrecs
; i
++) {
3586 error
= xfs_btree_check_ptr(cur
, cpp
, i
, level
- 1);
3591 xfs_btree_copy_ptrs(cur
, pp
, cpp
, numrecs
);
3593 error
= xfs_btree_free_block(cur
, cbp
);
3597 cur
->bc_bufs
[level
- 1] = NULL
;
3598 be16_add_cpu(&block
->bb_level
, -1);
3599 xfs_trans_log_inode(cur
->bc_tp
, ip
,
3600 XFS_ILOG_CORE
| xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
));
3607 * Kill the current root node, and replace it with it's only child node.
3610 xfs_btree_kill_root(
3611 struct xfs_btree_cur
*cur
,
3614 union xfs_btree_ptr
*newroot
)
3618 XFS_BTREE_STATS_INC(cur
, killroot
);
3621 * Update the root pointer, decreasing the level by 1 and then
3622 * free the old root.
3624 cur
->bc_ops
->set_root(cur
, newroot
, -1);
3626 error
= xfs_btree_free_block(cur
, bp
);
3630 cur
->bc_bufs
[level
] = NULL
;
3631 cur
->bc_ra
[level
] = 0;
3638 xfs_btree_dec_cursor(
3639 struct xfs_btree_cur
*cur
,
3647 error
= xfs_btree_decrement(cur
, level
, &i
);
3657 * Single level of the btree record deletion routine.
3658 * Delete record pointed to by cur/level.
3659 * Remove the record from its block then rebalance the tree.
3660 * Return 0 for error, 1 for done, 2 to go on to the next level.
3662 STATIC
int /* error */
3664 struct xfs_btree_cur
*cur
, /* btree cursor */
3665 int level
, /* level removing record from */
3666 int *stat
) /* fail/done/go-on */
3668 struct xfs_btree_block
*block
; /* btree block */
3669 union xfs_btree_ptr cptr
; /* current block ptr */
3670 struct xfs_buf
*bp
; /* buffer for block */
3671 int error
; /* error return value */
3672 int i
; /* loop counter */
3673 union xfs_btree_ptr lptr
; /* left sibling block ptr */
3674 struct xfs_buf
*lbp
; /* left buffer pointer */
3675 struct xfs_btree_block
*left
; /* left btree block */
3676 int lrecs
= 0; /* left record count */
3677 int ptr
; /* key/record index */
3678 union xfs_btree_ptr rptr
; /* right sibling block ptr */
3679 struct xfs_buf
*rbp
; /* right buffer pointer */
3680 struct xfs_btree_block
*right
; /* right btree block */
3681 struct xfs_btree_block
*rrblock
; /* right-right btree block */
3682 struct xfs_buf
*rrbp
; /* right-right buffer pointer */
3683 int rrecs
= 0; /* right record count */
3684 struct xfs_btree_cur
*tcur
; /* temporary btree cursor */
3685 int numrecs
; /* temporary numrec count */
3689 /* Get the index of the entry being deleted, check for nothing there. */
3690 ptr
= cur
->bc_ptrs
[level
];
3696 /* Get the buffer & block containing the record or key/ptr. */
3697 block
= xfs_btree_get_block(cur
, level
, &bp
);
3698 numrecs
= xfs_btree_get_numrecs(block
);
3701 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
3706 /* Fail if we're off the end of the block. */
3707 if (ptr
> numrecs
) {
3712 XFS_BTREE_STATS_INC(cur
, delrec
);
3713 XFS_BTREE_STATS_ADD(cur
, moves
, numrecs
- ptr
);
3715 /* Excise the entries being deleted. */
3717 /* It's a nonleaf. operate on keys and ptrs */
3718 union xfs_btree_key
*lkp
;
3719 union xfs_btree_ptr
*lpp
;
3721 lkp
= xfs_btree_key_addr(cur
, ptr
+ 1, block
);
3722 lpp
= xfs_btree_ptr_addr(cur
, ptr
+ 1, block
);
3725 for (i
= 0; i
< numrecs
- ptr
; i
++) {
3726 error
= xfs_btree_check_ptr(cur
, lpp
, i
, level
);
3732 if (ptr
< numrecs
) {
3733 xfs_btree_shift_keys(cur
, lkp
, -1, numrecs
- ptr
);
3734 xfs_btree_shift_ptrs(cur
, lpp
, -1, numrecs
- ptr
);
3735 xfs_btree_log_keys(cur
, bp
, ptr
, numrecs
- 1);
3736 xfs_btree_log_ptrs(cur
, bp
, ptr
, numrecs
- 1);
3739 /* It's a leaf. operate on records */
3740 if (ptr
< numrecs
) {
3741 xfs_btree_shift_recs(cur
,
3742 xfs_btree_rec_addr(cur
, ptr
+ 1, block
),
3744 xfs_btree_log_recs(cur
, bp
, ptr
, numrecs
- 1);
3749 * Decrement and log the number of entries in the block.
3751 xfs_btree_set_numrecs(block
, --numrecs
);
3752 xfs_btree_log_block(cur
, bp
, XFS_BB_NUMRECS
);
3755 * If we are tracking the last record in the tree and
3756 * we are at the far right edge of the tree, update it.
3758 if (xfs_btree_is_lastrec(cur
, block
, level
)) {
3759 cur
->bc_ops
->update_lastrec(cur
, block
, NULL
,
3760 ptr
, LASTREC_DELREC
);
3764 * We're at the root level. First, shrink the root block in-memory.
3765 * Try to get rid of the next level down. If we can't then there's
3766 * nothing left to do.
3768 if (level
== cur
->bc_nlevels
- 1) {
3769 if (cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) {
3770 xfs_iroot_realloc(cur
->bc_private
.b
.ip
, -1,
3771 cur
->bc_private
.b
.whichfork
);
3773 error
= xfs_btree_kill_iroot(cur
);
3777 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3785 * If this is the root level, and there's only one entry left,
3786 * and it's NOT the leaf level, then we can get rid of this
3789 if (numrecs
== 1 && level
> 0) {
3790 union xfs_btree_ptr
*pp
;
3792 * pp is still set to the first pointer in the block.
3793 * Make it the new root of the btree.
3795 pp
= xfs_btree_ptr_addr(cur
, 1, block
);
3796 error
= xfs_btree_kill_root(cur
, bp
, level
, pp
);
3799 } else if (level
> 0) {
3800 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3809 * If we deleted the leftmost entry in the block, update the
3810 * key values above us in the tree.
3812 if (xfs_btree_needs_key_update(cur
, ptr
)) {
3813 error
= xfs_btree_update_keys(cur
, level
);
3819 * If the number of records remaining in the block is at least
3820 * the minimum, we're done.
3822 if (numrecs
>= cur
->bc_ops
->get_minrecs(cur
, level
)) {
3823 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3830 * Otherwise, we have to move some records around to keep the
3831 * tree balanced. Look at the left and right sibling blocks to
3832 * see if we can re-balance by moving only one record.
3834 xfs_btree_get_sibling(cur
, block
, &rptr
, XFS_BB_RIGHTSIB
);
3835 xfs_btree_get_sibling(cur
, block
, &lptr
, XFS_BB_LEFTSIB
);
3837 if (cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) {
3839 * One child of root, need to get a chance to copy its contents
3840 * into the root and delete it. Can't go up to next level,
3841 * there's nothing to delete there.
3843 if (xfs_btree_ptr_is_null(cur
, &rptr
) &&
3844 xfs_btree_ptr_is_null(cur
, &lptr
) &&
3845 level
== cur
->bc_nlevels
- 2) {
3846 error
= xfs_btree_kill_iroot(cur
);
3848 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3855 ASSERT(!xfs_btree_ptr_is_null(cur
, &rptr
) ||
3856 !xfs_btree_ptr_is_null(cur
, &lptr
));
3859 * Duplicate the cursor so our btree manipulations here won't
3860 * disrupt the next level up.
3862 error
= xfs_btree_dup_cursor(cur
, &tcur
);
3867 * If there's a right sibling, see if it's ok to shift an entry
3870 if (!xfs_btree_ptr_is_null(cur
, &rptr
)) {
3872 * Move the temp cursor to the last entry in the next block.
3873 * Actually any entry but the first would suffice.
3875 i
= xfs_btree_lastrec(tcur
, level
);
3876 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3878 error
= xfs_btree_increment(tcur
, level
, &i
);
3881 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3883 i
= xfs_btree_lastrec(tcur
, level
);
3884 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3886 /* Grab a pointer to the block. */
3887 right
= xfs_btree_get_block(tcur
, level
, &rbp
);
3889 error
= xfs_btree_check_block(tcur
, right
, level
, rbp
);
3893 /* Grab the current block number, for future use. */
3894 xfs_btree_get_sibling(tcur
, right
, &cptr
, XFS_BB_LEFTSIB
);
3897 * If right block is full enough so that removing one entry
3898 * won't make it too empty, and left-shifting an entry out
3899 * of right to us works, we're done.
3901 if (xfs_btree_get_numrecs(right
) - 1 >=
3902 cur
->bc_ops
->get_minrecs(tcur
, level
)) {
3903 error
= xfs_btree_lshift(tcur
, level
, &i
);
3907 ASSERT(xfs_btree_get_numrecs(block
) >=
3908 cur
->bc_ops
->get_minrecs(tcur
, level
));
3910 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
3913 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3921 * Otherwise, grab the number of records in right for
3922 * future reference, and fix up the temp cursor to point
3923 * to our block again (last record).
3925 rrecs
= xfs_btree_get_numrecs(right
);
3926 if (!xfs_btree_ptr_is_null(cur
, &lptr
)) {
3927 i
= xfs_btree_firstrec(tcur
, level
);
3928 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3930 error
= xfs_btree_decrement(tcur
, level
, &i
);
3933 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3938 * If there's a left sibling, see if it's ok to shift an entry
3941 if (!xfs_btree_ptr_is_null(cur
, &lptr
)) {
3943 * Move the temp cursor to the first entry in the
3946 i
= xfs_btree_firstrec(tcur
, level
);
3947 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3949 error
= xfs_btree_decrement(tcur
, level
, &i
);
3952 i
= xfs_btree_firstrec(tcur
, level
);
3953 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3955 /* Grab a pointer to the block. */
3956 left
= xfs_btree_get_block(tcur
, level
, &lbp
);
3958 error
= xfs_btree_check_block(cur
, left
, level
, lbp
);
3962 /* Grab the current block number, for future use. */
3963 xfs_btree_get_sibling(tcur
, left
, &cptr
, XFS_BB_RIGHTSIB
);
3966 * If left block is full enough so that removing one entry
3967 * won't make it too empty, and right-shifting an entry out
3968 * of left to us works, we're done.
3970 if (xfs_btree_get_numrecs(left
) - 1 >=
3971 cur
->bc_ops
->get_minrecs(tcur
, level
)) {
3972 error
= xfs_btree_rshift(tcur
, level
, &i
);
3976 ASSERT(xfs_btree_get_numrecs(block
) >=
3977 cur
->bc_ops
->get_minrecs(tcur
, level
));
3978 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
3989 * Otherwise, grab the number of records in right for
3992 lrecs
= xfs_btree_get_numrecs(left
);
3995 /* Delete the temp cursor, we're done with it. */
3996 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
3999 /* If here, we need to do a join to keep the tree balanced. */
4000 ASSERT(!xfs_btree_ptr_is_null(cur
, &cptr
));
4002 if (!xfs_btree_ptr_is_null(cur
, &lptr
) &&
4003 lrecs
+ xfs_btree_get_numrecs(block
) <=
4004 cur
->bc_ops
->get_maxrecs(cur
, level
)) {
4006 * Set "right" to be the starting block,
4007 * "left" to be the left neighbor.
4012 error
= xfs_btree_read_buf_block(cur
, &lptr
, 0, &left
, &lbp
);
4017 * If that won't work, see if we can join with the right neighbor block.
4019 } else if (!xfs_btree_ptr_is_null(cur
, &rptr
) &&
4020 rrecs
+ xfs_btree_get_numrecs(block
) <=
4021 cur
->bc_ops
->get_maxrecs(cur
, level
)) {
4023 * Set "left" to be the starting block,
4024 * "right" to be the right neighbor.
4029 error
= xfs_btree_read_buf_block(cur
, &rptr
, 0, &right
, &rbp
);
4034 * Otherwise, we can't fix the imbalance.
4035 * Just return. This is probably a logic error, but it's not fatal.
4038 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
4044 rrecs
= xfs_btree_get_numrecs(right
);
4045 lrecs
= xfs_btree_get_numrecs(left
);
4048 * We're now going to join "left" and "right" by moving all the stuff
4049 * in "right" to "left" and deleting "right".
4051 XFS_BTREE_STATS_ADD(cur
, moves
, rrecs
);
4053 /* It's a non-leaf. Move keys and pointers. */
4054 union xfs_btree_key
*lkp
; /* left btree key */
4055 union xfs_btree_ptr
*lpp
; /* left address pointer */
4056 union xfs_btree_key
*rkp
; /* right btree key */
4057 union xfs_btree_ptr
*rpp
; /* right address pointer */
4059 lkp
= xfs_btree_key_addr(cur
, lrecs
+ 1, left
);
4060 lpp
= xfs_btree_ptr_addr(cur
, lrecs
+ 1, left
);
4061 rkp
= xfs_btree_key_addr(cur
, 1, right
);
4062 rpp
= xfs_btree_ptr_addr(cur
, 1, right
);
4064 for (i
= 1; i
< rrecs
; i
++) {
4065 error
= xfs_btree_check_ptr(cur
, rpp
, i
, level
);
4070 xfs_btree_copy_keys(cur
, lkp
, rkp
, rrecs
);
4071 xfs_btree_copy_ptrs(cur
, lpp
, rpp
, rrecs
);
4073 xfs_btree_log_keys(cur
, lbp
, lrecs
+ 1, lrecs
+ rrecs
);
4074 xfs_btree_log_ptrs(cur
, lbp
, lrecs
+ 1, lrecs
+ rrecs
);
4076 /* It's a leaf. Move records. */
4077 union xfs_btree_rec
*lrp
; /* left record pointer */
4078 union xfs_btree_rec
*rrp
; /* right record pointer */
4080 lrp
= xfs_btree_rec_addr(cur
, lrecs
+ 1, left
);
4081 rrp
= xfs_btree_rec_addr(cur
, 1, right
);
4083 xfs_btree_copy_recs(cur
, lrp
, rrp
, rrecs
);
4084 xfs_btree_log_recs(cur
, lbp
, lrecs
+ 1, lrecs
+ rrecs
);
4087 XFS_BTREE_STATS_INC(cur
, join
);
4090 * Fix up the number of records and right block pointer in the
4091 * surviving block, and log it.
4093 xfs_btree_set_numrecs(left
, lrecs
+ rrecs
);
4094 xfs_btree_get_sibling(cur
, right
, &cptr
, XFS_BB_RIGHTSIB
),
4095 xfs_btree_set_sibling(cur
, left
, &cptr
, XFS_BB_RIGHTSIB
);
4096 xfs_btree_log_block(cur
, lbp
, XFS_BB_NUMRECS
| XFS_BB_RIGHTSIB
);
4098 /* If there is a right sibling, point it to the remaining block. */
4099 xfs_btree_get_sibling(cur
, left
, &cptr
, XFS_BB_RIGHTSIB
);
4100 if (!xfs_btree_ptr_is_null(cur
, &cptr
)) {
4101 error
= xfs_btree_read_buf_block(cur
, &cptr
, 0, &rrblock
, &rrbp
);
4104 xfs_btree_set_sibling(cur
, rrblock
, &lptr
, XFS_BB_LEFTSIB
);
4105 xfs_btree_log_block(cur
, rrbp
, XFS_BB_LEFTSIB
);
4108 /* Free the deleted block. */
4109 error
= xfs_btree_free_block(cur
, rbp
);
4114 * If we joined with the left neighbor, set the buffer in the
4115 * cursor to the left block, and fix up the index.
4118 cur
->bc_bufs
[level
] = lbp
;
4119 cur
->bc_ptrs
[level
] += lrecs
;
4120 cur
->bc_ra
[level
] = 0;
4123 * If we joined with the right neighbor and there's a level above
4124 * us, increment the cursor at that level.
4126 else if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) ||
4127 (level
+ 1 < cur
->bc_nlevels
)) {
4128 error
= xfs_btree_increment(cur
, level
+ 1, &i
);
4134 * Readjust the ptr at this level if it's not a leaf, since it's
4135 * still pointing at the deletion point, which makes the cursor
4136 * inconsistent. If this makes the ptr 0, the caller fixes it up.
4137 * We can't use decrement because it would change the next level up.
4140 cur
->bc_ptrs
[level
]--;
4143 * We combined blocks, so we have to update the parent keys if the
4144 * btree supports overlapped intervals. However, bc_ptrs[level + 1]
4145 * points to the old block so that the caller knows which record to
4146 * delete. Therefore, the caller must be savvy enough to call updkeys
4147 * for us if we return stat == 2. The other exit points from this
4148 * function don't require deletions further up the tree, so they can
4149 * call updkeys directly.
4152 /* Return value means the next level up has something to do. */
4158 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
4163 * Delete the record pointed to by cur.
4164 * The cursor refers to the place where the record was (could be inserted)
4165 * when the operation returns.
4169 struct xfs_btree_cur
*cur
,
4170 int *stat
) /* success/failure */
4172 int error
; /* error return value */
4175 bool joined
= false;
4178 * Go up the tree, starting at leaf level.
4180 * If 2 is returned then a join was done; go to the next level.
4181 * Otherwise we are done.
4183 for (level
= 0, i
= 2; i
== 2; level
++) {
4184 error
= xfs_btree_delrec(cur
, level
, &i
);
4192 * If we combined blocks as part of deleting the record, delrec won't
4193 * have updated the parent high keys so we have to do that here.
4195 if (joined
&& (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
)) {
4196 error
= xfs_btree_updkeys_force(cur
, 0);
4202 for (level
= 1; level
< cur
->bc_nlevels
; level
++) {
4203 if (cur
->bc_ptrs
[level
] == 0) {
4204 error
= xfs_btree_decrement(cur
, level
, &i
);
4219 * Get the data from the pointed-to record.
4223 struct xfs_btree_cur
*cur
, /* btree cursor */
4224 union xfs_btree_rec
**recp
, /* output: btree record */
4225 int *stat
) /* output: success/failure */
4227 struct xfs_btree_block
*block
; /* btree block */
4228 struct xfs_buf
*bp
; /* buffer pointer */
4229 int ptr
; /* record number */
4231 int error
; /* error return value */
4234 ptr
= cur
->bc_ptrs
[0];
4235 block
= xfs_btree_get_block(cur
, 0, &bp
);
4238 error
= xfs_btree_check_block(cur
, block
, 0, bp
);
4244 * Off the right end or left end, return failure.
4246 if (ptr
> xfs_btree_get_numrecs(block
) || ptr
<= 0) {
4252 * Point to the record and extract its data.
4254 *recp
= xfs_btree_rec_addr(cur
, ptr
, block
);
4259 /* Visit a block in a btree. */
4261 xfs_btree_visit_block(
4262 struct xfs_btree_cur
*cur
,
4264 xfs_btree_visit_blocks_fn fn
,
4267 struct xfs_btree_block
*block
;
4269 union xfs_btree_ptr rptr
;
4272 /* do right sibling readahead */
4273 xfs_btree_readahead(cur
, level
, XFS_BTCUR_RIGHTRA
);
4274 block
= xfs_btree_get_block(cur
, level
, &bp
);
4276 /* process the block */
4277 error
= fn(cur
, level
, data
);
4281 /* now read rh sibling block for next iteration */
4282 xfs_btree_get_sibling(cur
, block
, &rptr
, XFS_BB_RIGHTSIB
);
4283 if (xfs_btree_ptr_is_null(cur
, &rptr
))
4286 return xfs_btree_lookup_get_block(cur
, level
, &rptr
, &block
);
4290 /* Visit every block in a btree. */
4292 xfs_btree_visit_blocks(
4293 struct xfs_btree_cur
*cur
,
4294 xfs_btree_visit_blocks_fn fn
,
4297 union xfs_btree_ptr lptr
;
4299 struct xfs_btree_block
*block
= NULL
;
4302 cur
->bc_ops
->init_ptr_from_cur(cur
, &lptr
);
4304 /* for each level */
4305 for (level
= cur
->bc_nlevels
- 1; level
>= 0; level
--) {
4306 /* grab the left hand block */
4307 error
= xfs_btree_lookup_get_block(cur
, level
, &lptr
, &block
);
4311 /* readahead the left most block for the next level down */
4313 union xfs_btree_ptr
*ptr
;
4315 ptr
= xfs_btree_ptr_addr(cur
, 1, block
);
4316 xfs_btree_readahead_ptr(cur
, ptr
, 1);
4318 /* save for the next iteration of the loop */
4319 xfs_btree_copy_ptrs(cur
, &lptr
, ptr
, 1);
4322 /* for each buffer in the level */
4324 error
= xfs_btree_visit_block(cur
, level
, fn
, data
);
4327 if (error
!= -ENOENT
)
4335 * Change the owner of a btree.
4337 * The mechanism we use here is ordered buffer logging. Because we don't know
4338 * how many buffers were are going to need to modify, we don't really want to
4339 * have to make transaction reservations for the worst case of every buffer in a
4340 * full size btree as that may be more space that we can fit in the log....
4342 * We do the btree walk in the most optimal manner possible - we have sibling
4343 * pointers so we can just walk all the blocks on each level from left to right
4344 * in a single pass, and then move to the next level and do the same. We can
4345 * also do readahead on the sibling pointers to get IO moving more quickly,
4346 * though for slow disks this is unlikely to make much difference to performance
4347 * as the amount of CPU work we have to do before moving to the next block is
4350 * For each btree block that we load, modify the owner appropriately, set the
4351 * buffer as an ordered buffer and log it appropriately. We need to ensure that
4352 * we mark the region we change dirty so that if the buffer is relogged in
4353 * a subsequent transaction the changes we make here as an ordered buffer are
4354 * correctly relogged in that transaction. If we are in recovery context, then
4355 * just queue the modified buffer as delayed write buffer so the transaction
4356 * recovery completion writes the changes to disk.
4358 struct xfs_btree_block_change_owner_info
{
4360 struct list_head
*buffer_list
;
4364 xfs_btree_block_change_owner(
4365 struct xfs_btree_cur
*cur
,
4369 struct xfs_btree_block_change_owner_info
*bbcoi
= data
;
4370 struct xfs_btree_block
*block
;
4373 /* modify the owner */
4374 block
= xfs_btree_get_block(cur
, level
, &bp
);
4375 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
4376 if (block
->bb_u
.l
.bb_owner
== cpu_to_be64(bbcoi
->new_owner
))
4378 block
->bb_u
.l
.bb_owner
= cpu_to_be64(bbcoi
->new_owner
);
4380 if (block
->bb_u
.s
.bb_owner
== cpu_to_be32(bbcoi
->new_owner
))
4382 block
->bb_u
.s
.bb_owner
= cpu_to_be32(bbcoi
->new_owner
);
4386 * If the block is a root block hosted in an inode, we might not have a
4387 * buffer pointer here and we shouldn't attempt to log the change as the
4388 * information is already held in the inode and discarded when the root
4389 * block is formatted into the on-disk inode fork. We still change it,
4390 * though, so everything is consistent in memory.
4393 ASSERT(cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
);
4394 ASSERT(level
== cur
->bc_nlevels
- 1);
4399 if (!xfs_trans_ordered_buf(cur
->bc_tp
, bp
)) {
4400 xfs_btree_log_block(cur
, bp
, XFS_BB_OWNER
);
4404 xfs_buf_delwri_queue(bp
, bbcoi
->buffer_list
);
4411 xfs_btree_change_owner(
4412 struct xfs_btree_cur
*cur
,
4414 struct list_head
*buffer_list
)
4416 struct xfs_btree_block_change_owner_info bbcoi
;
4418 bbcoi
.new_owner
= new_owner
;
4419 bbcoi
.buffer_list
= buffer_list
;
4421 return xfs_btree_visit_blocks(cur
, xfs_btree_block_change_owner
,
4425 /* Verify the v5 fields of a long-format btree block. */
4427 xfs_btree_lblock_v5hdr_verify(
4431 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
4432 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
4434 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
4435 return __this_address
;
4436 if (!uuid_equal(&block
->bb_u
.l
.bb_uuid
, &mp
->m_sb
.sb_meta_uuid
))
4437 return __this_address
;
4438 if (block
->bb_u
.l
.bb_blkno
!= cpu_to_be64(bp
->b_bn
))
4439 return __this_address
;
4440 if (owner
!= XFS_RMAP_OWN_UNKNOWN
&&
4441 be64_to_cpu(block
->bb_u
.l
.bb_owner
) != owner
)
4442 return __this_address
;
4446 /* Verify a long-format btree block. */
4448 xfs_btree_lblock_verify(
4450 unsigned int max_recs
)
4452 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
4453 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
4455 /* numrecs verification */
4456 if (be16_to_cpu(block
->bb_numrecs
) > max_recs
)
4457 return __this_address
;
4459 /* sibling pointer verification */
4460 if (block
->bb_u
.l
.bb_leftsib
!= cpu_to_be64(NULLFSBLOCK
) &&
4461 !xfs_verify_fsbno(mp
, be64_to_cpu(block
->bb_u
.l
.bb_leftsib
)))
4462 return __this_address
;
4463 if (block
->bb_u
.l
.bb_rightsib
!= cpu_to_be64(NULLFSBLOCK
) &&
4464 !xfs_verify_fsbno(mp
, be64_to_cpu(block
->bb_u
.l
.bb_rightsib
)))
4465 return __this_address
;
4471 * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format
4474 * @bp: buffer containing the btree block
4475 * @max_recs: pointer to the m_*_mxr max records field in the xfs mount
4476 * @pag_max_level: pointer to the per-ag max level field
4479 xfs_btree_sblock_v5hdr_verify(
4482 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
4483 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
4484 struct xfs_perag
*pag
= bp
->b_pag
;
4486 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
4487 return __this_address
;
4488 if (!uuid_equal(&block
->bb_u
.s
.bb_uuid
, &mp
->m_sb
.sb_meta_uuid
))
4489 return __this_address
;
4490 if (block
->bb_u
.s
.bb_blkno
!= cpu_to_be64(bp
->b_bn
))
4491 return __this_address
;
4492 if (pag
&& be32_to_cpu(block
->bb_u
.s
.bb_owner
) != pag
->pag_agno
)
4493 return __this_address
;
4498 * xfs_btree_sblock_verify() -- verify a short-format btree block
4500 * @bp: buffer containing the btree block
4501 * @max_recs: maximum records allowed in this btree node
4504 xfs_btree_sblock_verify(
4506 unsigned int max_recs
)
4508 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
4509 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
4512 /* numrecs verification */
4513 if (be16_to_cpu(block
->bb_numrecs
) > max_recs
)
4514 return __this_address
;
4516 /* sibling pointer verification */
4517 agno
= xfs_daddr_to_agno(mp
, XFS_BUF_ADDR(bp
));
4518 if (block
->bb_u
.s
.bb_leftsib
!= cpu_to_be32(NULLAGBLOCK
) &&
4519 !xfs_verify_agbno(mp
, agno
, be32_to_cpu(block
->bb_u
.s
.bb_leftsib
)))
4520 return __this_address
;
4521 if (block
->bb_u
.s
.bb_rightsib
!= cpu_to_be32(NULLAGBLOCK
) &&
4522 !xfs_verify_agbno(mp
, agno
, be32_to_cpu(block
->bb_u
.s
.bb_rightsib
)))
4523 return __this_address
;
4529 * Calculate the number of btree levels needed to store a given number of
4530 * records in a short-format btree.
4533 xfs_btree_compute_maxlevels(
4538 unsigned long maxblocks
;
4540 maxblocks
= (len
+ limits
[0] - 1) / limits
[0];
4541 for (level
= 1; maxblocks
> 1; level
++)
4542 maxblocks
= (maxblocks
+ limits
[1] - 1) / limits
[1];
4547 * Query a regular btree for all records overlapping a given interval.
4548 * Start with a LE lookup of the key of low_rec and return all records
4549 * until we find a record with a key greater than the key of high_rec.
4552 xfs_btree_simple_query_range(
4553 struct xfs_btree_cur
*cur
,
4554 union xfs_btree_key
*low_key
,
4555 union xfs_btree_key
*high_key
,
4556 xfs_btree_query_range_fn fn
,
4559 union xfs_btree_rec
*recp
;
4560 union xfs_btree_key rec_key
;
4563 bool firstrec
= true;
4566 ASSERT(cur
->bc_ops
->init_high_key_from_rec
);
4567 ASSERT(cur
->bc_ops
->diff_two_keys
);
4570 * Find the leftmost record. The btree cursor must be set
4571 * to the low record used to generate low_key.
4574 error
= xfs_btree_lookup(cur
, XFS_LOOKUP_LE
, &stat
);
4578 /* Nothing? See if there's anything to the right. */
4580 error
= xfs_btree_increment(cur
, 0, &stat
);
4586 /* Find the record. */
4587 error
= xfs_btree_get_rec(cur
, &recp
, &stat
);
4591 /* Skip if high_key(rec) < low_key. */
4593 cur
->bc_ops
->init_high_key_from_rec(&rec_key
, recp
);
4595 diff
= cur
->bc_ops
->diff_two_keys(cur
, low_key
,
4601 /* Stop if high_key < low_key(rec). */
4602 cur
->bc_ops
->init_key_from_rec(&rec_key
, recp
);
4603 diff
= cur
->bc_ops
->diff_two_keys(cur
, &rec_key
, high_key
);
4608 error
= fn(cur
, recp
, priv
);
4609 if (error
< 0 || error
== XFS_BTREE_QUERY_RANGE_ABORT
)
4613 /* Move on to the next record. */
4614 error
= xfs_btree_increment(cur
, 0, &stat
);
4624 * Query an overlapped interval btree for all records overlapping a given
4625 * interval. This function roughly follows the algorithm given in
4626 * "Interval Trees" of _Introduction to Algorithms_, which is section
4627 * 14.3 in the 2nd and 3rd editions.
4629 * First, generate keys for the low and high records passed in.
4631 * For any leaf node, generate the high and low keys for the record.
4632 * If the record keys overlap with the query low/high keys, pass the
4633 * record to the function iterator.
4635 * For any internal node, compare the low and high keys of each
4636 * pointer against the query low/high keys. If there's an overlap,
4637 * follow the pointer.
4639 * As an optimization, we stop scanning a block when we find a low key
4640 * that is greater than the query's high key.
4643 xfs_btree_overlapped_query_range(
4644 struct xfs_btree_cur
*cur
,
4645 union xfs_btree_key
*low_key
,
4646 union xfs_btree_key
*high_key
,
4647 xfs_btree_query_range_fn fn
,
4650 union xfs_btree_ptr ptr
;
4651 union xfs_btree_ptr
*pp
;
4652 union xfs_btree_key rec_key
;
4653 union xfs_btree_key rec_hkey
;
4654 union xfs_btree_key
*lkp
;
4655 union xfs_btree_key
*hkp
;
4656 union xfs_btree_rec
*recp
;
4657 struct xfs_btree_block
*block
;
4665 /* Load the root of the btree. */
4666 level
= cur
->bc_nlevels
- 1;
4667 cur
->bc_ops
->init_ptr_from_cur(cur
, &ptr
);
4668 error
= xfs_btree_lookup_get_block(cur
, level
, &ptr
, &block
);
4671 xfs_btree_get_block(cur
, level
, &bp
);
4672 trace_xfs_btree_overlapped_query_range(cur
, level
, bp
);
4674 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
4678 cur
->bc_ptrs
[level
] = 1;
4680 while (level
< cur
->bc_nlevels
) {
4681 block
= xfs_btree_get_block(cur
, level
, &bp
);
4683 /* End of node, pop back towards the root. */
4684 if (cur
->bc_ptrs
[level
] > be16_to_cpu(block
->bb_numrecs
)) {
4686 if (level
< cur
->bc_nlevels
- 1)
4687 cur
->bc_ptrs
[level
+ 1]++;
4693 /* Handle a leaf node. */
4694 recp
= xfs_btree_rec_addr(cur
, cur
->bc_ptrs
[0], block
);
4696 cur
->bc_ops
->init_high_key_from_rec(&rec_hkey
, recp
);
4697 ldiff
= cur
->bc_ops
->diff_two_keys(cur
, &rec_hkey
,
4700 cur
->bc_ops
->init_key_from_rec(&rec_key
, recp
);
4701 hdiff
= cur
->bc_ops
->diff_two_keys(cur
, high_key
,
4705 * If (record's high key >= query's low key) and
4706 * (query's high key >= record's low key), then
4707 * this record overlaps the query range; callback.
4709 if (ldiff
>= 0 && hdiff
>= 0) {
4710 error
= fn(cur
, recp
, priv
);
4712 error
== XFS_BTREE_QUERY_RANGE_ABORT
)
4714 } else if (hdiff
< 0) {
4715 /* Record is larger than high key; pop. */
4718 cur
->bc_ptrs
[level
]++;
4722 /* Handle an internal node. */
4723 lkp
= xfs_btree_key_addr(cur
, cur
->bc_ptrs
[level
], block
);
4724 hkp
= xfs_btree_high_key_addr(cur
, cur
->bc_ptrs
[level
], block
);
4725 pp
= xfs_btree_ptr_addr(cur
, cur
->bc_ptrs
[level
], block
);
4727 ldiff
= cur
->bc_ops
->diff_two_keys(cur
, hkp
, low_key
);
4728 hdiff
= cur
->bc_ops
->diff_two_keys(cur
, high_key
, lkp
);
4731 * If (pointer's high key >= query's low key) and
4732 * (query's high key >= pointer's low key), then
4733 * this record overlaps the query range; follow pointer.
4735 if (ldiff
>= 0 && hdiff
>= 0) {
4737 error
= xfs_btree_lookup_get_block(cur
, level
, pp
,
4741 xfs_btree_get_block(cur
, level
, &bp
);
4742 trace_xfs_btree_overlapped_query_range(cur
, level
, bp
);
4744 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
4748 cur
->bc_ptrs
[level
] = 1;
4750 } else if (hdiff
< 0) {
4751 /* The low key is larger than the upper range; pop. */
4754 cur
->bc_ptrs
[level
]++;
4759 * If we don't end this function with the cursor pointing at a record
4760 * block, a subsequent non-error cursor deletion will not release
4761 * node-level buffers, causing a buffer leak. This is quite possible
4762 * with a zero-results range query, so release the buffers if we
4763 * failed to return any results.
4765 if (cur
->bc_bufs
[0] == NULL
) {
4766 for (i
= 0; i
< cur
->bc_nlevels
; i
++) {
4767 if (cur
->bc_bufs
[i
]) {
4768 xfs_trans_brelse(cur
->bc_tp
, cur
->bc_bufs
[i
]);
4769 cur
->bc_bufs
[i
] = NULL
;
4770 cur
->bc_ptrs
[i
] = 0;
4780 * Query a btree for all records overlapping a given interval of keys. The
4781 * supplied function will be called with each record found; return one of the
4782 * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
4783 * code. This function returns XFS_BTREE_QUERY_RANGE_ABORT, zero, or a
4784 * negative error code.
4787 xfs_btree_query_range(
4788 struct xfs_btree_cur
*cur
,
4789 union xfs_btree_irec
*low_rec
,
4790 union xfs_btree_irec
*high_rec
,
4791 xfs_btree_query_range_fn fn
,
4794 union xfs_btree_rec rec
;
4795 union xfs_btree_key low_key
;
4796 union xfs_btree_key high_key
;
4798 /* Find the keys of both ends of the interval. */
4799 cur
->bc_rec
= *high_rec
;
4800 cur
->bc_ops
->init_rec_from_cur(cur
, &rec
);
4801 cur
->bc_ops
->init_key_from_rec(&high_key
, &rec
);
4803 cur
->bc_rec
= *low_rec
;
4804 cur
->bc_ops
->init_rec_from_cur(cur
, &rec
);
4805 cur
->bc_ops
->init_key_from_rec(&low_key
, &rec
);
4807 /* Enforce low key < high key. */
4808 if (cur
->bc_ops
->diff_two_keys(cur
, &low_key
, &high_key
) > 0)
4811 if (!(cur
->bc_flags
& XFS_BTREE_OVERLAPPING
))
4812 return xfs_btree_simple_query_range(cur
, &low_key
,
4813 &high_key
, fn
, priv
);
4814 return xfs_btree_overlapped_query_range(cur
, &low_key
, &high_key
,
4818 /* Query a btree for all records. */
4820 xfs_btree_query_all(
4821 struct xfs_btree_cur
*cur
,
4822 xfs_btree_query_range_fn fn
,
4825 union xfs_btree_key low_key
;
4826 union xfs_btree_key high_key
;
4828 memset(&cur
->bc_rec
, 0, sizeof(cur
->bc_rec
));
4829 memset(&low_key
, 0, sizeof(low_key
));
4830 memset(&high_key
, 0xFF, sizeof(high_key
));
4832 return xfs_btree_simple_query_range(cur
, &low_key
, &high_key
, fn
, priv
);
4836 * Calculate the number of blocks needed to store a given number of records
4837 * in a short-format (per-AG metadata) btree.
4840 xfs_btree_calc_size(
4842 unsigned long long len
)
4848 maxrecs
= limits
[0];
4849 for (level
= 0, rval
= 0; len
> 1; level
++) {
4851 do_div(len
, maxrecs
);
4852 maxrecs
= limits
[1];
4859 xfs_btree_count_blocks_helper(
4860 struct xfs_btree_cur
*cur
,
4864 xfs_extlen_t
*blocks
= data
;
4870 /* Count the blocks in a btree and return the result in *blocks. */
4872 xfs_btree_count_blocks(
4873 struct xfs_btree_cur
*cur
,
4874 xfs_extlen_t
*blocks
)
4877 return xfs_btree_visit_blocks(cur
, xfs_btree_count_blocks_helper
,
4881 /* Compare two btree pointers. */
4883 xfs_btree_diff_two_ptrs(
4884 struct xfs_btree_cur
*cur
,
4885 const union xfs_btree_ptr
*a
,
4886 const union xfs_btree_ptr
*b
)
4888 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
4889 return (int64_t)be64_to_cpu(a
->l
) - be64_to_cpu(b
->l
);
4890 return (int64_t)be32_to_cpu(a
->s
) - be32_to_cpu(b
->s
);
4893 /* If there's an extent, we're done. */
4895 xfs_btree_has_record_helper(
4896 struct xfs_btree_cur
*cur
,
4897 union xfs_btree_rec
*rec
,
4900 return XFS_BTREE_QUERY_RANGE_ABORT
;
4903 /* Is there a record covering a given range of keys? */
4905 xfs_btree_has_record(
4906 struct xfs_btree_cur
*cur
,
4907 union xfs_btree_irec
*low
,
4908 union xfs_btree_irec
*high
,
4913 error
= xfs_btree_query_range(cur
, low
, high
,
4914 &xfs_btree_has_record_helper
, NULL
);
4915 if (error
== XFS_BTREE_QUERY_RANGE_ABORT
) {