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_fspriv
;
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_fspriv
;
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
,
1437 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1438 XFS_BTREE_TRACE_ARGBII(cur
, bp
, first
, last
);
1441 xfs_trans_buf_set_type(cur
->bc_tp
, bp
, XFS_BLFT_BTREE_BUF
);
1442 xfs_trans_log_buf(cur
->bc_tp
, bp
,
1443 xfs_btree_key_offset(cur
, first
),
1444 xfs_btree_key_offset(cur
, last
+ 1) - 1);
1446 xfs_trans_log_inode(cur
->bc_tp
, cur
->bc_private
.b
.ip
,
1447 xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
));
1450 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1454 * Log record values from the btree block.
1458 struct xfs_btree_cur
*cur
,
1463 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1464 XFS_BTREE_TRACE_ARGBII(cur
, bp
, first
, last
);
1466 xfs_trans_buf_set_type(cur
->bc_tp
, bp
, XFS_BLFT_BTREE_BUF
);
1467 xfs_trans_log_buf(cur
->bc_tp
, bp
,
1468 xfs_btree_rec_offset(cur
, first
),
1469 xfs_btree_rec_offset(cur
, last
+ 1) - 1);
1471 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1475 * Log block pointer fields from a btree block (nonleaf).
1479 struct xfs_btree_cur
*cur
, /* btree cursor */
1480 struct xfs_buf
*bp
, /* buffer containing btree block */
1481 int first
, /* index of first pointer to log */
1482 int last
) /* index of last pointer to log */
1484 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1485 XFS_BTREE_TRACE_ARGBII(cur
, bp
, first
, last
);
1488 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
1489 int level
= xfs_btree_get_level(block
);
1491 xfs_trans_buf_set_type(cur
->bc_tp
, bp
, XFS_BLFT_BTREE_BUF
);
1492 xfs_trans_log_buf(cur
->bc_tp
, bp
,
1493 xfs_btree_ptr_offset(cur
, first
, level
),
1494 xfs_btree_ptr_offset(cur
, last
+ 1, level
) - 1);
1496 xfs_trans_log_inode(cur
->bc_tp
, cur
->bc_private
.b
.ip
,
1497 xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
));
1500 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1504 * Log fields from a btree block header.
1507 xfs_btree_log_block(
1508 struct xfs_btree_cur
*cur
, /* btree cursor */
1509 struct xfs_buf
*bp
, /* buffer containing btree block */
1510 int fields
) /* mask of fields: XFS_BB_... */
1512 int first
; /* first byte offset logged */
1513 int last
; /* last byte offset logged */
1514 static const short soffsets
[] = { /* table of offsets (short) */
1515 offsetof(struct xfs_btree_block
, bb_magic
),
1516 offsetof(struct xfs_btree_block
, bb_level
),
1517 offsetof(struct xfs_btree_block
, bb_numrecs
),
1518 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_leftsib
),
1519 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_rightsib
),
1520 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_blkno
),
1521 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_lsn
),
1522 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_uuid
),
1523 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_owner
),
1524 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_crc
),
1525 XFS_BTREE_SBLOCK_CRC_LEN
1527 static const short loffsets
[] = { /* table of offsets (long) */
1528 offsetof(struct xfs_btree_block
, bb_magic
),
1529 offsetof(struct xfs_btree_block
, bb_level
),
1530 offsetof(struct xfs_btree_block
, bb_numrecs
),
1531 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_leftsib
),
1532 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_rightsib
),
1533 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_blkno
),
1534 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_lsn
),
1535 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_uuid
),
1536 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_owner
),
1537 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_crc
),
1538 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_pad
),
1539 XFS_BTREE_LBLOCK_CRC_LEN
1542 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1543 XFS_BTREE_TRACE_ARGBI(cur
, bp
, fields
);
1548 if (cur
->bc_flags
& XFS_BTREE_CRC_BLOCKS
) {
1550 * We don't log the CRC when updating a btree
1551 * block but instead recreate it during log
1552 * recovery. As the log buffers have checksums
1553 * of their own this is safe and avoids logging a crc
1554 * update in a lot of places.
1556 if (fields
== XFS_BB_ALL_BITS
)
1557 fields
= XFS_BB_ALL_BITS_CRC
;
1558 nbits
= XFS_BB_NUM_BITS_CRC
;
1560 nbits
= XFS_BB_NUM_BITS
;
1562 xfs_btree_offsets(fields
,
1563 (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) ?
1564 loffsets
: soffsets
,
1565 nbits
, &first
, &last
);
1566 xfs_trans_buf_set_type(cur
->bc_tp
, bp
, XFS_BLFT_BTREE_BUF
);
1567 xfs_trans_log_buf(cur
->bc_tp
, bp
, first
, last
);
1569 xfs_trans_log_inode(cur
->bc_tp
, cur
->bc_private
.b
.ip
,
1570 xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
));
1573 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1577 * Increment cursor by one record at the level.
1578 * For nonzero levels the leaf-ward information is untouched.
1581 xfs_btree_increment(
1582 struct xfs_btree_cur
*cur
,
1584 int *stat
) /* success/failure */
1586 struct xfs_btree_block
*block
;
1587 union xfs_btree_ptr ptr
;
1589 int error
; /* error return value */
1592 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1593 XFS_BTREE_TRACE_ARGI(cur
, level
);
1595 ASSERT(level
< cur
->bc_nlevels
);
1597 /* Read-ahead to the right at this level. */
1598 xfs_btree_readahead(cur
, level
, XFS_BTCUR_RIGHTRA
);
1600 /* Get a pointer to the btree block. */
1601 block
= xfs_btree_get_block(cur
, level
, &bp
);
1604 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
1609 /* We're done if we remain in the block after the increment. */
1610 if (++cur
->bc_ptrs
[level
] <= xfs_btree_get_numrecs(block
))
1613 /* Fail if we just went off the right edge of the tree. */
1614 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_RIGHTSIB
);
1615 if (xfs_btree_ptr_is_null(cur
, &ptr
))
1618 XFS_BTREE_STATS_INC(cur
, increment
);
1621 * March up the tree incrementing pointers.
1622 * Stop when we don't go off the right edge of a block.
1624 for (lev
= level
+ 1; lev
< cur
->bc_nlevels
; lev
++) {
1625 block
= xfs_btree_get_block(cur
, lev
, &bp
);
1628 error
= xfs_btree_check_block(cur
, block
, lev
, bp
);
1633 if (++cur
->bc_ptrs
[lev
] <= xfs_btree_get_numrecs(block
))
1636 /* Read-ahead the right block for the next loop. */
1637 xfs_btree_readahead(cur
, lev
, XFS_BTCUR_RIGHTRA
);
1641 * If we went off the root then we are either seriously
1642 * confused or have the tree root in an inode.
1644 if (lev
== cur
->bc_nlevels
) {
1645 if (cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
)
1648 error
= -EFSCORRUPTED
;
1651 ASSERT(lev
< cur
->bc_nlevels
);
1654 * Now walk back down the tree, fixing up the cursor's buffer
1655 * pointers and key numbers.
1657 for (block
= xfs_btree_get_block(cur
, lev
, &bp
); lev
> level
; ) {
1658 union xfs_btree_ptr
*ptrp
;
1660 ptrp
= xfs_btree_ptr_addr(cur
, cur
->bc_ptrs
[lev
], block
);
1662 error
= xfs_btree_read_buf_block(cur
, ptrp
, 0, &block
, &bp
);
1666 xfs_btree_setbuf(cur
, lev
, bp
);
1667 cur
->bc_ptrs
[lev
] = 1;
1670 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1675 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1680 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
1685 * Decrement cursor by one record at the level.
1686 * For nonzero levels the leaf-ward information is untouched.
1689 xfs_btree_decrement(
1690 struct xfs_btree_cur
*cur
,
1692 int *stat
) /* success/failure */
1694 struct xfs_btree_block
*block
;
1696 int error
; /* error return value */
1698 union xfs_btree_ptr ptr
;
1700 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1701 XFS_BTREE_TRACE_ARGI(cur
, level
);
1703 ASSERT(level
< cur
->bc_nlevels
);
1705 /* Read-ahead to the left at this level. */
1706 xfs_btree_readahead(cur
, level
, XFS_BTCUR_LEFTRA
);
1708 /* We're done if we remain in the block after the decrement. */
1709 if (--cur
->bc_ptrs
[level
] > 0)
1712 /* Get a pointer to the btree block. */
1713 block
= xfs_btree_get_block(cur
, level
, &bp
);
1716 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
1721 /* Fail if we just went off the left edge of the tree. */
1722 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_LEFTSIB
);
1723 if (xfs_btree_ptr_is_null(cur
, &ptr
))
1726 XFS_BTREE_STATS_INC(cur
, decrement
);
1729 * March up the tree decrementing pointers.
1730 * Stop when we don't go off the left edge of a block.
1732 for (lev
= level
+ 1; lev
< cur
->bc_nlevels
; lev
++) {
1733 if (--cur
->bc_ptrs
[lev
] > 0)
1735 /* Read-ahead the left block for the next loop. */
1736 xfs_btree_readahead(cur
, lev
, XFS_BTCUR_LEFTRA
);
1740 * If we went off the root then we are seriously confused.
1741 * or the root of the tree is in an inode.
1743 if (lev
== cur
->bc_nlevels
) {
1744 if (cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
)
1747 error
= -EFSCORRUPTED
;
1750 ASSERT(lev
< cur
->bc_nlevels
);
1753 * Now walk back down the tree, fixing up the cursor's buffer
1754 * pointers and key numbers.
1756 for (block
= xfs_btree_get_block(cur
, lev
, &bp
); lev
> level
; ) {
1757 union xfs_btree_ptr
*ptrp
;
1759 ptrp
= xfs_btree_ptr_addr(cur
, cur
->bc_ptrs
[lev
], block
);
1761 error
= xfs_btree_read_buf_block(cur
, ptrp
, 0, &block
, &bp
);
1764 xfs_btree_setbuf(cur
, lev
, bp
);
1765 cur
->bc_ptrs
[lev
] = xfs_btree_get_numrecs(block
);
1768 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1773 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1778 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
1783 xfs_btree_lookup_get_block(
1784 struct xfs_btree_cur
*cur
, /* btree cursor */
1785 int level
, /* level in the btree */
1786 union xfs_btree_ptr
*pp
, /* ptr to btree block */
1787 struct xfs_btree_block
**blkp
) /* return btree block */
1789 struct xfs_buf
*bp
; /* buffer pointer for btree block */
1792 /* special case the root block if in an inode */
1793 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
1794 (level
== cur
->bc_nlevels
- 1)) {
1795 *blkp
= xfs_btree_get_iroot(cur
);
1800 * If the old buffer at this level for the disk address we are
1801 * looking for re-use it.
1803 * Otherwise throw it away and get a new one.
1805 bp
= cur
->bc_bufs
[level
];
1806 if (bp
&& XFS_BUF_ADDR(bp
) == xfs_btree_ptr_to_daddr(cur
, pp
)) {
1807 *blkp
= XFS_BUF_TO_BLOCK(bp
);
1811 error
= xfs_btree_read_buf_block(cur
, pp
, 0, blkp
, &bp
);
1815 /* Check the inode owner since the verifiers don't. */
1816 if (xfs_sb_version_hascrc(&cur
->bc_mp
->m_sb
) &&
1817 !(cur
->bc_private
.b
.flags
& XFS_BTCUR_BPRV_INVALID_OWNER
) &&
1818 (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) &&
1819 be64_to_cpu((*blkp
)->bb_u
.l
.bb_owner
) !=
1820 cur
->bc_private
.b
.ip
->i_ino
)
1823 /* Did we get the level we were looking for? */
1824 if (be16_to_cpu((*blkp
)->bb_level
) != level
)
1827 /* Check that internal nodes have at least one record. */
1828 if (level
!= 0 && be16_to_cpu((*blkp
)->bb_numrecs
) == 0)
1831 xfs_btree_setbuf(cur
, level
, bp
);
1836 xfs_trans_brelse(cur
->bc_tp
, bp
);
1837 return -EFSCORRUPTED
;
1841 * Get current search key. For level 0 we don't actually have a key
1842 * structure so we make one up from the record. For all other levels
1843 * we just return the right key.
1845 STATIC
union xfs_btree_key
*
1846 xfs_lookup_get_search_key(
1847 struct xfs_btree_cur
*cur
,
1850 struct xfs_btree_block
*block
,
1851 union xfs_btree_key
*kp
)
1854 cur
->bc_ops
->init_key_from_rec(kp
,
1855 xfs_btree_rec_addr(cur
, keyno
, block
));
1859 return xfs_btree_key_addr(cur
, keyno
, block
);
1863 * Lookup the record. The cursor is made to point to it, based on dir.
1864 * stat is set to 0 if can't find any such record, 1 for success.
1868 struct xfs_btree_cur
*cur
, /* btree cursor */
1869 xfs_lookup_t dir
, /* <=, ==, or >= */
1870 int *stat
) /* success/failure */
1872 struct xfs_btree_block
*block
; /* current btree block */
1873 int64_t diff
; /* difference for the current key */
1874 int error
; /* error return value */
1875 int keyno
; /* current key number */
1876 int level
; /* level in the btree */
1877 union xfs_btree_ptr
*pp
; /* ptr to btree block */
1878 union xfs_btree_ptr ptr
; /* ptr to btree block */
1880 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1881 XFS_BTREE_TRACE_ARGI(cur
, dir
);
1883 XFS_BTREE_STATS_INC(cur
, lookup
);
1885 /* No such thing as a zero-level tree. */
1886 if (cur
->bc_nlevels
== 0)
1887 return -EFSCORRUPTED
;
1892 /* initialise start pointer from cursor */
1893 cur
->bc_ops
->init_ptr_from_cur(cur
, &ptr
);
1897 * Iterate over each level in the btree, starting at the root.
1898 * For each level above the leaves, find the key we need, based
1899 * on the lookup record, then follow the corresponding block
1900 * pointer down to the next level.
1902 for (level
= cur
->bc_nlevels
- 1, diff
= 1; level
>= 0; level
--) {
1903 /* Get the block we need to do the lookup on. */
1904 error
= xfs_btree_lookup_get_block(cur
, level
, pp
, &block
);
1910 * If we already had a key match at a higher level, we
1911 * know we need to use the first entry in this block.
1915 /* Otherwise search this block. Do a binary search. */
1917 int high
; /* high entry number */
1918 int low
; /* low entry number */
1920 /* Set low and high entry numbers, 1-based. */
1922 high
= xfs_btree_get_numrecs(block
);
1924 /* Block is empty, must be an empty leaf. */
1925 ASSERT(level
== 0 && cur
->bc_nlevels
== 1);
1927 cur
->bc_ptrs
[0] = dir
!= XFS_LOOKUP_LE
;
1928 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1933 /* Binary search the block. */
1934 while (low
<= high
) {
1935 union xfs_btree_key key
;
1936 union xfs_btree_key
*kp
;
1938 XFS_BTREE_STATS_INC(cur
, compare
);
1940 /* keyno is average of low and high. */
1941 keyno
= (low
+ high
) >> 1;
1943 /* Get current search key */
1944 kp
= xfs_lookup_get_search_key(cur
, level
,
1945 keyno
, block
, &key
);
1948 * Compute difference to get next direction:
1949 * - less than, move right
1950 * - greater than, move left
1951 * - equal, we're done
1953 diff
= cur
->bc_ops
->key_diff(cur
, kp
);
1964 * If there are more levels, set up for the next level
1965 * by getting the block number and filling in the cursor.
1969 * If we moved left, need the previous key number,
1970 * unless there isn't one.
1972 if (diff
> 0 && --keyno
< 1)
1974 pp
= xfs_btree_ptr_addr(cur
, keyno
, block
);
1977 error
= xfs_btree_check_ptr(cur
, pp
, 0, level
);
1981 cur
->bc_ptrs
[level
] = keyno
;
1985 /* Done with the search. See if we need to adjust the results. */
1986 if (dir
!= XFS_LOOKUP_LE
&& diff
< 0) {
1989 * If ge search and we went off the end of the block, but it's
1990 * not the last block, we're in the wrong block.
1992 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_RIGHTSIB
);
1993 if (dir
== XFS_LOOKUP_GE
&&
1994 keyno
> xfs_btree_get_numrecs(block
) &&
1995 !xfs_btree_ptr_is_null(cur
, &ptr
)) {
1998 cur
->bc_ptrs
[0] = keyno
;
1999 error
= xfs_btree_increment(cur
, 0, &i
);
2002 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
2003 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2007 } else if (dir
== XFS_LOOKUP_LE
&& diff
> 0)
2009 cur
->bc_ptrs
[0] = keyno
;
2011 /* Return if we succeeded or not. */
2012 if (keyno
== 0 || keyno
> xfs_btree_get_numrecs(block
))
2014 else if (dir
!= XFS_LOOKUP_EQ
|| diff
== 0)
2018 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2022 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2026 /* Find the high key storage area from a regular key. */
2027 union xfs_btree_key
*
2028 xfs_btree_high_key_from_key(
2029 struct xfs_btree_cur
*cur
,
2030 union xfs_btree_key
*key
)
2032 ASSERT(cur
->bc_flags
& XFS_BTREE_OVERLAPPING
);
2033 return (union xfs_btree_key
*)((char *)key
+
2034 (cur
->bc_ops
->key_len
/ 2));
2037 /* Determine the low (and high if overlapped) keys of a leaf block */
2039 xfs_btree_get_leaf_keys(
2040 struct xfs_btree_cur
*cur
,
2041 struct xfs_btree_block
*block
,
2042 union xfs_btree_key
*key
)
2044 union xfs_btree_key max_hkey
;
2045 union xfs_btree_key hkey
;
2046 union xfs_btree_rec
*rec
;
2047 union xfs_btree_key
*high
;
2050 rec
= xfs_btree_rec_addr(cur
, 1, block
);
2051 cur
->bc_ops
->init_key_from_rec(key
, rec
);
2053 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2055 cur
->bc_ops
->init_high_key_from_rec(&max_hkey
, rec
);
2056 for (n
= 2; n
<= xfs_btree_get_numrecs(block
); n
++) {
2057 rec
= xfs_btree_rec_addr(cur
, n
, block
);
2058 cur
->bc_ops
->init_high_key_from_rec(&hkey
, rec
);
2059 if (cur
->bc_ops
->diff_two_keys(cur
, &hkey
, &max_hkey
)
2064 high
= xfs_btree_high_key_from_key(cur
, key
);
2065 memcpy(high
, &max_hkey
, cur
->bc_ops
->key_len
/ 2);
2069 /* Determine the low (and high if overlapped) keys of a node block */
2071 xfs_btree_get_node_keys(
2072 struct xfs_btree_cur
*cur
,
2073 struct xfs_btree_block
*block
,
2074 union xfs_btree_key
*key
)
2076 union xfs_btree_key
*hkey
;
2077 union xfs_btree_key
*max_hkey
;
2078 union xfs_btree_key
*high
;
2081 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2082 memcpy(key
, xfs_btree_key_addr(cur
, 1, block
),
2083 cur
->bc_ops
->key_len
/ 2);
2085 max_hkey
= xfs_btree_high_key_addr(cur
, 1, block
);
2086 for (n
= 2; n
<= xfs_btree_get_numrecs(block
); n
++) {
2087 hkey
= xfs_btree_high_key_addr(cur
, n
, block
);
2088 if (cur
->bc_ops
->diff_two_keys(cur
, hkey
, max_hkey
) > 0)
2092 high
= xfs_btree_high_key_from_key(cur
, key
);
2093 memcpy(high
, max_hkey
, cur
->bc_ops
->key_len
/ 2);
2095 memcpy(key
, xfs_btree_key_addr(cur
, 1, block
),
2096 cur
->bc_ops
->key_len
);
2100 /* Derive the keys for any btree block. */
2103 struct xfs_btree_cur
*cur
,
2104 struct xfs_btree_block
*block
,
2105 union xfs_btree_key
*key
)
2107 if (be16_to_cpu(block
->bb_level
) == 0)
2108 xfs_btree_get_leaf_keys(cur
, block
, key
);
2110 xfs_btree_get_node_keys(cur
, block
, key
);
2114 * Decide if we need to update the parent keys of a btree block. For
2115 * a standard btree this is only necessary if we're updating the first
2116 * record/key. For an overlapping btree, we must always update the
2117 * keys because the highest key can be in any of the records or keys
2121 xfs_btree_needs_key_update(
2122 struct xfs_btree_cur
*cur
,
2125 return (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) || ptr
== 1;
2129 * Update the low and high parent keys of the given level, progressing
2130 * towards the root. If force_all is false, stop if the keys for a given
2131 * level do not need updating.
2134 __xfs_btree_updkeys(
2135 struct xfs_btree_cur
*cur
,
2137 struct xfs_btree_block
*block
,
2138 struct xfs_buf
*bp0
,
2141 union xfs_btree_key key
; /* keys from current level */
2142 union xfs_btree_key
*lkey
; /* keys from the next level up */
2143 union xfs_btree_key
*hkey
;
2144 union xfs_btree_key
*nlkey
; /* keys from the next level up */
2145 union xfs_btree_key
*nhkey
;
2149 ASSERT(cur
->bc_flags
& XFS_BTREE_OVERLAPPING
);
2151 /* Exit if there aren't any parent levels to update. */
2152 if (level
+ 1 >= cur
->bc_nlevels
)
2155 trace_xfs_btree_updkeys(cur
, level
, bp0
);
2158 hkey
= xfs_btree_high_key_from_key(cur
, lkey
);
2159 xfs_btree_get_keys(cur
, block
, lkey
);
2160 for (level
++; level
< cur
->bc_nlevels
; level
++) {
2164 block
= xfs_btree_get_block(cur
, level
, &bp
);
2165 trace_xfs_btree_updkeys(cur
, level
, bp
);
2167 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
2169 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2173 ptr
= cur
->bc_ptrs
[level
];
2174 nlkey
= xfs_btree_key_addr(cur
, ptr
, block
);
2175 nhkey
= xfs_btree_high_key_addr(cur
, ptr
, block
);
2177 !(cur
->bc_ops
->diff_two_keys(cur
, nlkey
, lkey
) != 0 ||
2178 cur
->bc_ops
->diff_two_keys(cur
, nhkey
, hkey
) != 0))
2180 xfs_btree_copy_keys(cur
, nlkey
, lkey
, 1);
2181 xfs_btree_log_keys(cur
, bp
, ptr
, ptr
);
2182 if (level
+ 1 >= cur
->bc_nlevels
)
2184 xfs_btree_get_node_keys(cur
, block
, lkey
);
2190 /* Update all the keys from some level in cursor back to the root. */
2192 xfs_btree_updkeys_force(
2193 struct xfs_btree_cur
*cur
,
2197 struct xfs_btree_block
*block
;
2199 block
= xfs_btree_get_block(cur
, level
, &bp
);
2200 return __xfs_btree_updkeys(cur
, level
, block
, bp
, true);
2204 * Update the parent keys of the given level, progressing towards the root.
2207 xfs_btree_update_keys(
2208 struct xfs_btree_cur
*cur
,
2211 struct xfs_btree_block
*block
;
2213 union xfs_btree_key
*kp
;
2214 union xfs_btree_key key
;
2219 block
= xfs_btree_get_block(cur
, level
, &bp
);
2220 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
)
2221 return __xfs_btree_updkeys(cur
, level
, block
, bp
, false);
2223 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
2224 XFS_BTREE_TRACE_ARGIK(cur
, level
, keyp
);
2227 * Go up the tree from this level toward the root.
2228 * At each level, update the key value to the value input.
2229 * Stop when we reach a level where the cursor isn't pointing
2230 * at the first entry in the block.
2232 xfs_btree_get_keys(cur
, block
, &key
);
2233 for (level
++, ptr
= 1; ptr
== 1 && level
< cur
->bc_nlevels
; level
++) {
2237 block
= xfs_btree_get_block(cur
, level
, &bp
);
2239 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
2241 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2245 ptr
= cur
->bc_ptrs
[level
];
2246 kp
= xfs_btree_key_addr(cur
, ptr
, block
);
2247 xfs_btree_copy_keys(cur
, kp
, &key
, 1);
2248 xfs_btree_log_keys(cur
, bp
, ptr
, ptr
);
2251 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2256 * Update the record referred to by cur to the value in the
2257 * given record. This either works (return 0) or gets an
2258 * EFSCORRUPTED error.
2262 struct xfs_btree_cur
*cur
,
2263 union xfs_btree_rec
*rec
)
2265 struct xfs_btree_block
*block
;
2269 union xfs_btree_rec
*rp
;
2271 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
2272 XFS_BTREE_TRACE_ARGR(cur
, rec
);
2274 /* Pick up the current block. */
2275 block
= xfs_btree_get_block(cur
, 0, &bp
);
2278 error
= xfs_btree_check_block(cur
, block
, 0, bp
);
2282 /* Get the address of the rec to be updated. */
2283 ptr
= cur
->bc_ptrs
[0];
2284 rp
= xfs_btree_rec_addr(cur
, ptr
, block
);
2286 /* Fill in the new contents and log them. */
2287 xfs_btree_copy_recs(cur
, rp
, rec
, 1);
2288 xfs_btree_log_recs(cur
, bp
, ptr
, ptr
);
2291 * If we are tracking the last record in the tree and
2292 * we are at the far right edge of the tree, update it.
2294 if (xfs_btree_is_lastrec(cur
, block
, 0)) {
2295 cur
->bc_ops
->update_lastrec(cur
, block
, rec
,
2296 ptr
, LASTREC_UPDATE
);
2299 /* Pass new key value up to our parent. */
2300 if (xfs_btree_needs_key_update(cur
, ptr
)) {
2301 error
= xfs_btree_update_keys(cur
, 0);
2306 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2310 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2315 * Move 1 record left from cur/level if possible.
2316 * Update cur to reflect the new path.
2318 STATIC
int /* error */
2320 struct xfs_btree_cur
*cur
,
2322 int *stat
) /* success/failure */
2324 struct xfs_buf
*lbp
; /* left buffer pointer */
2325 struct xfs_btree_block
*left
; /* left btree block */
2326 int lrecs
; /* left record count */
2327 struct xfs_buf
*rbp
; /* right buffer pointer */
2328 struct xfs_btree_block
*right
; /* right btree block */
2329 struct xfs_btree_cur
*tcur
; /* temporary btree cursor */
2330 int rrecs
; /* right record count */
2331 union xfs_btree_ptr lptr
; /* left btree pointer */
2332 union xfs_btree_key
*rkp
= NULL
; /* right btree key */
2333 union xfs_btree_ptr
*rpp
= NULL
; /* right address pointer */
2334 union xfs_btree_rec
*rrp
= NULL
; /* right record pointer */
2335 int error
; /* error return value */
2338 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
2339 XFS_BTREE_TRACE_ARGI(cur
, level
);
2341 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
2342 level
== cur
->bc_nlevels
- 1)
2345 /* Set up variables for this block as "right". */
2346 right
= xfs_btree_get_block(cur
, level
, &rbp
);
2349 error
= xfs_btree_check_block(cur
, right
, level
, rbp
);
2354 /* If we've got no left sibling then we can't shift an entry left. */
2355 xfs_btree_get_sibling(cur
, right
, &lptr
, XFS_BB_LEFTSIB
);
2356 if (xfs_btree_ptr_is_null(cur
, &lptr
))
2360 * If the cursor entry is the one that would be moved, don't
2361 * do it... it's too complicated.
2363 if (cur
->bc_ptrs
[level
] <= 1)
2366 /* Set up the left neighbor as "left". */
2367 error
= xfs_btree_read_buf_block(cur
, &lptr
, 0, &left
, &lbp
);
2371 /* If it's full, it can't take another entry. */
2372 lrecs
= xfs_btree_get_numrecs(left
);
2373 if (lrecs
== cur
->bc_ops
->get_maxrecs(cur
, level
))
2376 rrecs
= xfs_btree_get_numrecs(right
);
2379 * We add one entry to the left side and remove one for the right side.
2380 * Account for it here, the changes will be updated on disk and logged
2386 XFS_BTREE_STATS_INC(cur
, lshift
);
2387 XFS_BTREE_STATS_ADD(cur
, moves
, 1);
2390 * If non-leaf, copy a key and a ptr to the left block.
2391 * Log the changes to the left block.
2394 /* It's a non-leaf. Move keys and pointers. */
2395 union xfs_btree_key
*lkp
; /* left btree key */
2396 union xfs_btree_ptr
*lpp
; /* left address pointer */
2398 lkp
= xfs_btree_key_addr(cur
, lrecs
, left
);
2399 rkp
= xfs_btree_key_addr(cur
, 1, right
);
2401 lpp
= xfs_btree_ptr_addr(cur
, lrecs
, left
);
2402 rpp
= xfs_btree_ptr_addr(cur
, 1, right
);
2404 error
= xfs_btree_check_ptr(cur
, rpp
, 0, level
);
2408 xfs_btree_copy_keys(cur
, lkp
, rkp
, 1);
2409 xfs_btree_copy_ptrs(cur
, lpp
, rpp
, 1);
2411 xfs_btree_log_keys(cur
, lbp
, lrecs
, lrecs
);
2412 xfs_btree_log_ptrs(cur
, lbp
, lrecs
, lrecs
);
2414 ASSERT(cur
->bc_ops
->keys_inorder(cur
,
2415 xfs_btree_key_addr(cur
, lrecs
- 1, left
), lkp
));
2417 /* It's a leaf. Move records. */
2418 union xfs_btree_rec
*lrp
; /* left record pointer */
2420 lrp
= xfs_btree_rec_addr(cur
, lrecs
, left
);
2421 rrp
= xfs_btree_rec_addr(cur
, 1, right
);
2423 xfs_btree_copy_recs(cur
, lrp
, rrp
, 1);
2424 xfs_btree_log_recs(cur
, lbp
, lrecs
, lrecs
);
2426 ASSERT(cur
->bc_ops
->recs_inorder(cur
,
2427 xfs_btree_rec_addr(cur
, lrecs
- 1, left
), lrp
));
2430 xfs_btree_set_numrecs(left
, lrecs
);
2431 xfs_btree_log_block(cur
, lbp
, XFS_BB_NUMRECS
);
2433 xfs_btree_set_numrecs(right
, rrecs
);
2434 xfs_btree_log_block(cur
, rbp
, XFS_BB_NUMRECS
);
2437 * Slide the contents of right down one entry.
2439 XFS_BTREE_STATS_ADD(cur
, moves
, rrecs
- 1);
2441 /* It's a nonleaf. operate on keys and ptrs */
2443 int i
; /* loop index */
2445 for (i
= 0; i
< rrecs
; i
++) {
2446 error
= xfs_btree_check_ptr(cur
, rpp
, i
+ 1, level
);
2451 xfs_btree_shift_keys(cur
,
2452 xfs_btree_key_addr(cur
, 2, right
),
2454 xfs_btree_shift_ptrs(cur
,
2455 xfs_btree_ptr_addr(cur
, 2, right
),
2458 xfs_btree_log_keys(cur
, rbp
, 1, rrecs
);
2459 xfs_btree_log_ptrs(cur
, rbp
, 1, rrecs
);
2461 /* It's a leaf. operate on records */
2462 xfs_btree_shift_recs(cur
,
2463 xfs_btree_rec_addr(cur
, 2, right
),
2465 xfs_btree_log_recs(cur
, rbp
, 1, rrecs
);
2469 * Using a temporary cursor, update the parent key values of the
2470 * block on the left.
2472 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2473 error
= xfs_btree_dup_cursor(cur
, &tcur
);
2476 i
= xfs_btree_firstrec(tcur
, level
);
2477 XFS_WANT_CORRUPTED_GOTO(tcur
->bc_mp
, i
== 1, error0
);
2479 error
= xfs_btree_decrement(tcur
, level
, &i
);
2483 /* Update the parent high keys of the left block, if needed. */
2484 error
= xfs_btree_update_keys(tcur
, level
);
2488 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
2491 /* Update the parent keys of the right block. */
2492 error
= xfs_btree_update_keys(cur
, level
);
2496 /* Slide the cursor value left one. */
2497 cur
->bc_ptrs
[level
]--;
2499 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2504 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2509 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2513 XFS_BTREE_TRACE_CURSOR(tcur
, XBT_ERROR
);
2514 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
2519 * Move 1 record right from cur/level if possible.
2520 * Update cur to reflect the new path.
2522 STATIC
int /* error */
2524 struct xfs_btree_cur
*cur
,
2526 int *stat
) /* success/failure */
2528 struct xfs_buf
*lbp
; /* left buffer pointer */
2529 struct xfs_btree_block
*left
; /* left btree block */
2530 struct xfs_buf
*rbp
; /* right buffer pointer */
2531 struct xfs_btree_block
*right
; /* right btree block */
2532 struct xfs_btree_cur
*tcur
; /* temporary btree cursor */
2533 union xfs_btree_ptr rptr
; /* right block pointer */
2534 union xfs_btree_key
*rkp
; /* right btree key */
2535 int rrecs
; /* right record count */
2536 int lrecs
; /* left record count */
2537 int error
; /* error return value */
2538 int i
; /* loop counter */
2540 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
2541 XFS_BTREE_TRACE_ARGI(cur
, level
);
2543 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
2544 (level
== cur
->bc_nlevels
- 1))
2547 /* Set up variables for this block as "left". */
2548 left
= xfs_btree_get_block(cur
, level
, &lbp
);
2551 error
= xfs_btree_check_block(cur
, left
, level
, lbp
);
2556 /* If we've got no right sibling then we can't shift an entry right. */
2557 xfs_btree_get_sibling(cur
, left
, &rptr
, XFS_BB_RIGHTSIB
);
2558 if (xfs_btree_ptr_is_null(cur
, &rptr
))
2562 * If the cursor entry is the one that would be moved, don't
2563 * do it... it's too complicated.
2565 lrecs
= xfs_btree_get_numrecs(left
);
2566 if (cur
->bc_ptrs
[level
] >= lrecs
)
2569 /* Set up the right neighbor as "right". */
2570 error
= xfs_btree_read_buf_block(cur
, &rptr
, 0, &right
, &rbp
);
2574 /* If it's full, it can't take another entry. */
2575 rrecs
= xfs_btree_get_numrecs(right
);
2576 if (rrecs
== cur
->bc_ops
->get_maxrecs(cur
, level
))
2579 XFS_BTREE_STATS_INC(cur
, rshift
);
2580 XFS_BTREE_STATS_ADD(cur
, moves
, rrecs
);
2583 * Make a hole at the start of the right neighbor block, then
2584 * copy the last left block entry to the hole.
2587 /* It's a nonleaf. make a hole in the keys and ptrs */
2588 union xfs_btree_key
*lkp
;
2589 union xfs_btree_ptr
*lpp
;
2590 union xfs_btree_ptr
*rpp
;
2592 lkp
= xfs_btree_key_addr(cur
, lrecs
, left
);
2593 lpp
= xfs_btree_ptr_addr(cur
, lrecs
, left
);
2594 rkp
= xfs_btree_key_addr(cur
, 1, right
);
2595 rpp
= xfs_btree_ptr_addr(cur
, 1, right
);
2598 for (i
= rrecs
- 1; i
>= 0; i
--) {
2599 error
= xfs_btree_check_ptr(cur
, rpp
, i
, level
);
2605 xfs_btree_shift_keys(cur
, rkp
, 1, rrecs
);
2606 xfs_btree_shift_ptrs(cur
, rpp
, 1, rrecs
);
2609 error
= xfs_btree_check_ptr(cur
, lpp
, 0, level
);
2614 /* Now put the new data in, and log it. */
2615 xfs_btree_copy_keys(cur
, rkp
, lkp
, 1);
2616 xfs_btree_copy_ptrs(cur
, rpp
, lpp
, 1);
2618 xfs_btree_log_keys(cur
, rbp
, 1, rrecs
+ 1);
2619 xfs_btree_log_ptrs(cur
, rbp
, 1, rrecs
+ 1);
2621 ASSERT(cur
->bc_ops
->keys_inorder(cur
, rkp
,
2622 xfs_btree_key_addr(cur
, 2, right
)));
2624 /* It's a leaf. make a hole in the records */
2625 union xfs_btree_rec
*lrp
;
2626 union xfs_btree_rec
*rrp
;
2628 lrp
= xfs_btree_rec_addr(cur
, lrecs
, left
);
2629 rrp
= xfs_btree_rec_addr(cur
, 1, right
);
2631 xfs_btree_shift_recs(cur
, rrp
, 1, rrecs
);
2633 /* Now put the new data in, and log it. */
2634 xfs_btree_copy_recs(cur
, rrp
, lrp
, 1);
2635 xfs_btree_log_recs(cur
, rbp
, 1, rrecs
+ 1);
2639 * Decrement and log left's numrecs, bump and log right's numrecs.
2641 xfs_btree_set_numrecs(left
, --lrecs
);
2642 xfs_btree_log_block(cur
, lbp
, XFS_BB_NUMRECS
);
2644 xfs_btree_set_numrecs(right
, ++rrecs
);
2645 xfs_btree_log_block(cur
, rbp
, XFS_BB_NUMRECS
);
2648 * Using a temporary cursor, update the parent key values of the
2649 * block on the right.
2651 error
= xfs_btree_dup_cursor(cur
, &tcur
);
2654 i
= xfs_btree_lastrec(tcur
, level
);
2655 XFS_WANT_CORRUPTED_GOTO(tcur
->bc_mp
, i
== 1, error0
);
2657 error
= xfs_btree_increment(tcur
, level
, &i
);
2661 /* Update the parent high keys of the left block, if needed. */
2662 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2663 error
= xfs_btree_update_keys(cur
, level
);
2668 /* Update the parent keys of the right block. */
2669 error
= xfs_btree_update_keys(tcur
, level
);
2673 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
2675 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2680 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2685 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2689 XFS_BTREE_TRACE_CURSOR(tcur
, XBT_ERROR
);
2690 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
2695 * Split cur/level block in half.
2696 * Return new block number and the key to its first
2697 * record (to be inserted into parent).
2699 STATIC
int /* error */
2701 struct xfs_btree_cur
*cur
,
2703 union xfs_btree_ptr
*ptrp
,
2704 union xfs_btree_key
*key
,
2705 struct xfs_btree_cur
**curp
,
2706 int *stat
) /* success/failure */
2708 union xfs_btree_ptr lptr
; /* left sibling block ptr */
2709 struct xfs_buf
*lbp
; /* left buffer pointer */
2710 struct xfs_btree_block
*left
; /* left btree block */
2711 union xfs_btree_ptr rptr
; /* right sibling block ptr */
2712 struct xfs_buf
*rbp
; /* right buffer pointer */
2713 struct xfs_btree_block
*right
; /* right btree block */
2714 union xfs_btree_ptr rrptr
; /* right-right sibling ptr */
2715 struct xfs_buf
*rrbp
; /* right-right buffer pointer */
2716 struct xfs_btree_block
*rrblock
; /* right-right btree block */
2720 int error
; /* error return value */
2725 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
2726 XFS_BTREE_TRACE_ARGIPK(cur
, level
, *ptrp
, key
);
2728 XFS_BTREE_STATS_INC(cur
, split
);
2730 /* Set up left block (current one). */
2731 left
= xfs_btree_get_block(cur
, level
, &lbp
);
2734 error
= xfs_btree_check_block(cur
, left
, level
, lbp
);
2739 xfs_btree_buf_to_ptr(cur
, lbp
, &lptr
);
2741 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2742 error
= cur
->bc_ops
->alloc_block(cur
, &lptr
, &rptr
, stat
);
2747 XFS_BTREE_STATS_INC(cur
, alloc
);
2749 /* Set up the new block as "right". */
2750 error
= xfs_btree_get_buf_block(cur
, &rptr
, 0, &right
, &rbp
);
2754 /* Fill in the btree header for the new right block. */
2755 xfs_btree_init_block_cur(cur
, rbp
, xfs_btree_get_level(left
), 0);
2758 * Split the entries between the old and the new block evenly.
2759 * Make sure that if there's an odd number of entries now, that
2760 * each new block will have the same number of entries.
2762 lrecs
= xfs_btree_get_numrecs(left
);
2764 if ((lrecs
& 1) && cur
->bc_ptrs
[level
] <= rrecs
+ 1)
2766 src_index
= (lrecs
- rrecs
+ 1);
2768 XFS_BTREE_STATS_ADD(cur
, moves
, rrecs
);
2770 /* Adjust numrecs for the later get_*_keys() calls. */
2772 xfs_btree_set_numrecs(left
, lrecs
);
2773 xfs_btree_set_numrecs(right
, xfs_btree_get_numrecs(right
) + rrecs
);
2776 * Copy btree block entries from the left block over to the
2777 * new block, the right. Update the right block and log the
2781 /* It's a non-leaf. Move keys and pointers. */
2782 union xfs_btree_key
*lkp
; /* left btree key */
2783 union xfs_btree_ptr
*lpp
; /* left address pointer */
2784 union xfs_btree_key
*rkp
; /* right btree key */
2785 union xfs_btree_ptr
*rpp
; /* right address pointer */
2787 lkp
= xfs_btree_key_addr(cur
, src_index
, left
);
2788 lpp
= xfs_btree_ptr_addr(cur
, src_index
, left
);
2789 rkp
= xfs_btree_key_addr(cur
, 1, right
);
2790 rpp
= xfs_btree_ptr_addr(cur
, 1, right
);
2793 for (i
= src_index
; i
< rrecs
; i
++) {
2794 error
= xfs_btree_check_ptr(cur
, lpp
, i
, level
);
2800 /* Copy the keys & pointers to the new block. */
2801 xfs_btree_copy_keys(cur
, rkp
, lkp
, rrecs
);
2802 xfs_btree_copy_ptrs(cur
, rpp
, lpp
, rrecs
);
2804 xfs_btree_log_keys(cur
, rbp
, 1, rrecs
);
2805 xfs_btree_log_ptrs(cur
, rbp
, 1, rrecs
);
2807 /* Stash the keys of the new block for later insertion. */
2808 xfs_btree_get_node_keys(cur
, right
, key
);
2810 /* It's a leaf. Move records. */
2811 union xfs_btree_rec
*lrp
; /* left record pointer */
2812 union xfs_btree_rec
*rrp
; /* right record pointer */
2814 lrp
= xfs_btree_rec_addr(cur
, src_index
, left
);
2815 rrp
= xfs_btree_rec_addr(cur
, 1, right
);
2817 /* Copy records to the new block. */
2818 xfs_btree_copy_recs(cur
, rrp
, lrp
, rrecs
);
2819 xfs_btree_log_recs(cur
, rbp
, 1, rrecs
);
2821 /* Stash the keys of the new block for later insertion. */
2822 xfs_btree_get_leaf_keys(cur
, right
, key
);
2826 * Find the left block number by looking in the buffer.
2827 * Adjust sibling pointers.
2829 xfs_btree_get_sibling(cur
, left
, &rrptr
, XFS_BB_RIGHTSIB
);
2830 xfs_btree_set_sibling(cur
, right
, &rrptr
, XFS_BB_RIGHTSIB
);
2831 xfs_btree_set_sibling(cur
, right
, &lptr
, XFS_BB_LEFTSIB
);
2832 xfs_btree_set_sibling(cur
, left
, &rptr
, XFS_BB_RIGHTSIB
);
2834 xfs_btree_log_block(cur
, rbp
, XFS_BB_ALL_BITS
);
2835 xfs_btree_log_block(cur
, lbp
, XFS_BB_NUMRECS
| XFS_BB_RIGHTSIB
);
2838 * If there's a block to the new block's right, make that block
2839 * point back to right instead of to left.
2841 if (!xfs_btree_ptr_is_null(cur
, &rrptr
)) {
2842 error
= xfs_btree_read_buf_block(cur
, &rrptr
,
2843 0, &rrblock
, &rrbp
);
2846 xfs_btree_set_sibling(cur
, rrblock
, &rptr
, XFS_BB_LEFTSIB
);
2847 xfs_btree_log_block(cur
, rrbp
, XFS_BB_LEFTSIB
);
2850 /* Update the parent high keys of the left block, if needed. */
2851 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2852 error
= xfs_btree_update_keys(cur
, level
);
2858 * If the cursor is really in the right block, move it there.
2859 * If it's just pointing past the last entry in left, then we'll
2860 * insert there, so don't change anything in that case.
2862 if (cur
->bc_ptrs
[level
] > lrecs
+ 1) {
2863 xfs_btree_setbuf(cur
, level
, rbp
);
2864 cur
->bc_ptrs
[level
] -= lrecs
;
2867 * If there are more levels, we'll need another cursor which refers
2868 * the right block, no matter where this cursor was.
2870 if (level
+ 1 < cur
->bc_nlevels
) {
2871 error
= xfs_btree_dup_cursor(cur
, curp
);
2874 (*curp
)->bc_ptrs
[level
+ 1]++;
2877 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2881 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2886 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2891 struct xfs_btree_split_args
{
2892 struct xfs_btree_cur
*cur
;
2894 union xfs_btree_ptr
*ptrp
;
2895 union xfs_btree_key
*key
;
2896 struct xfs_btree_cur
**curp
;
2897 int *stat
; /* success/failure */
2899 bool kswapd
; /* allocation in kswapd context */
2900 struct completion
*done
;
2901 struct work_struct work
;
2905 * Stack switching interfaces for allocation
2908 xfs_btree_split_worker(
2909 struct work_struct
*work
)
2911 struct xfs_btree_split_args
*args
= container_of(work
,
2912 struct xfs_btree_split_args
, work
);
2913 unsigned long pflags
;
2914 unsigned long new_pflags
= PF_MEMALLOC_NOFS
;
2917 * we are in a transaction context here, but may also be doing work
2918 * in kswapd context, and hence we may need to inherit that state
2919 * temporarily to ensure that we don't block waiting for memory reclaim
2923 new_pflags
|= PF_MEMALLOC
| PF_SWAPWRITE
| PF_KSWAPD
;
2925 current_set_flags_nested(&pflags
, new_pflags
);
2927 args
->result
= __xfs_btree_split(args
->cur
, args
->level
, args
->ptrp
,
2928 args
->key
, args
->curp
, args
->stat
);
2929 complete(args
->done
);
2931 current_restore_flags_nested(&pflags
, new_pflags
);
2935 * BMBT split requests often come in with little stack to work on. Push
2936 * them off to a worker thread so there is lots of stack to use. For the other
2937 * btree types, just call directly to avoid the context switch overhead here.
2939 STATIC
int /* error */
2941 struct xfs_btree_cur
*cur
,
2943 union xfs_btree_ptr
*ptrp
,
2944 union xfs_btree_key
*key
,
2945 struct xfs_btree_cur
**curp
,
2946 int *stat
) /* success/failure */
2948 struct xfs_btree_split_args args
;
2949 DECLARE_COMPLETION_ONSTACK(done
);
2951 if (cur
->bc_btnum
!= XFS_BTNUM_BMAP
)
2952 return __xfs_btree_split(cur
, level
, ptrp
, key
, curp
, stat
);
2961 args
.kswapd
= current_is_kswapd();
2962 INIT_WORK_ONSTACK(&args
.work
, xfs_btree_split_worker
);
2963 queue_work(xfs_alloc_wq
, &args
.work
);
2964 wait_for_completion(&done
);
2965 destroy_work_on_stack(&args
.work
);
2969 #define xfs_btree_split __xfs_btree_split
2974 * Copy the old inode root contents into a real block and make the
2975 * broot point to it.
2978 xfs_btree_new_iroot(
2979 struct xfs_btree_cur
*cur
, /* btree cursor */
2980 int *logflags
, /* logging flags for inode */
2981 int *stat
) /* return status - 0 fail */
2983 struct xfs_buf
*cbp
; /* buffer for cblock */
2984 struct xfs_btree_block
*block
; /* btree block */
2985 struct xfs_btree_block
*cblock
; /* child btree block */
2986 union xfs_btree_key
*ckp
; /* child key pointer */
2987 union xfs_btree_ptr
*cpp
; /* child ptr pointer */
2988 union xfs_btree_key
*kp
; /* pointer to btree key */
2989 union xfs_btree_ptr
*pp
; /* pointer to block addr */
2990 union xfs_btree_ptr nptr
; /* new block addr */
2991 int level
; /* btree level */
2992 int error
; /* error return code */
2994 int i
; /* loop counter */
2997 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
2998 XFS_BTREE_STATS_INC(cur
, newroot
);
3000 ASSERT(cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
);
3002 level
= cur
->bc_nlevels
- 1;
3004 block
= xfs_btree_get_iroot(cur
);
3005 pp
= xfs_btree_ptr_addr(cur
, 1, block
);
3007 /* Allocate the new block. If we can't do it, we're toast. Give up. */
3008 error
= cur
->bc_ops
->alloc_block(cur
, pp
, &nptr
, stat
);
3012 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3015 XFS_BTREE_STATS_INC(cur
, alloc
);
3017 /* Copy the root into a real block. */
3018 error
= xfs_btree_get_buf_block(cur
, &nptr
, 0, &cblock
, &cbp
);
3023 * we can't just memcpy() the root in for CRC enabled btree blocks.
3024 * In that case have to also ensure the blkno remains correct
3026 memcpy(cblock
, block
, xfs_btree_block_len(cur
));
3027 if (cur
->bc_flags
& XFS_BTREE_CRC_BLOCKS
) {
3028 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
3029 cblock
->bb_u
.l
.bb_blkno
= cpu_to_be64(cbp
->b_bn
);
3031 cblock
->bb_u
.s
.bb_blkno
= cpu_to_be64(cbp
->b_bn
);
3034 be16_add_cpu(&block
->bb_level
, 1);
3035 xfs_btree_set_numrecs(block
, 1);
3037 cur
->bc_ptrs
[level
+ 1] = 1;
3039 kp
= xfs_btree_key_addr(cur
, 1, block
);
3040 ckp
= xfs_btree_key_addr(cur
, 1, cblock
);
3041 xfs_btree_copy_keys(cur
, ckp
, kp
, xfs_btree_get_numrecs(cblock
));
3043 cpp
= xfs_btree_ptr_addr(cur
, 1, cblock
);
3045 for (i
= 0; i
< be16_to_cpu(cblock
->bb_numrecs
); i
++) {
3046 error
= xfs_btree_check_ptr(cur
, pp
, i
, level
);
3051 xfs_btree_copy_ptrs(cur
, cpp
, pp
, xfs_btree_get_numrecs(cblock
));
3054 error
= xfs_btree_check_ptr(cur
, &nptr
, 0, level
);
3058 xfs_btree_copy_ptrs(cur
, pp
, &nptr
, 1);
3060 xfs_iroot_realloc(cur
->bc_private
.b
.ip
,
3061 1 - xfs_btree_get_numrecs(cblock
),
3062 cur
->bc_private
.b
.whichfork
);
3064 xfs_btree_setbuf(cur
, level
, cbp
);
3067 * Do all this logging at the end so that
3068 * the root is at the right level.
3070 xfs_btree_log_block(cur
, cbp
, XFS_BB_ALL_BITS
);
3071 xfs_btree_log_keys(cur
, cbp
, 1, be16_to_cpu(cblock
->bb_numrecs
));
3072 xfs_btree_log_ptrs(cur
, cbp
, 1, be16_to_cpu(cblock
->bb_numrecs
));
3075 XFS_ILOG_CORE
| xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
);
3077 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3080 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3085 * Allocate a new root block, fill it in.
3087 STATIC
int /* error */
3089 struct xfs_btree_cur
*cur
, /* btree cursor */
3090 int *stat
) /* success/failure */
3092 struct xfs_btree_block
*block
; /* one half of the old root block */
3093 struct xfs_buf
*bp
; /* buffer containing block */
3094 int error
; /* error return value */
3095 struct xfs_buf
*lbp
; /* left buffer pointer */
3096 struct xfs_btree_block
*left
; /* left btree block */
3097 struct xfs_buf
*nbp
; /* new (root) buffer */
3098 struct xfs_btree_block
*new; /* new (root) btree block */
3099 int nptr
; /* new value for key index, 1 or 2 */
3100 struct xfs_buf
*rbp
; /* right buffer pointer */
3101 struct xfs_btree_block
*right
; /* right btree block */
3102 union xfs_btree_ptr rptr
;
3103 union xfs_btree_ptr lptr
;
3105 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
3106 XFS_BTREE_STATS_INC(cur
, newroot
);
3108 /* initialise our start point from the cursor */
3109 cur
->bc_ops
->init_ptr_from_cur(cur
, &rptr
);
3111 /* Allocate the new block. If we can't do it, we're toast. Give up. */
3112 error
= cur
->bc_ops
->alloc_block(cur
, &rptr
, &lptr
, stat
);
3117 XFS_BTREE_STATS_INC(cur
, alloc
);
3119 /* Set up the new block. */
3120 error
= xfs_btree_get_buf_block(cur
, &lptr
, 0, &new, &nbp
);
3124 /* Set the root in the holding structure increasing the level by 1. */
3125 cur
->bc_ops
->set_root(cur
, &lptr
, 1);
3128 * At the previous root level there are now two blocks: the old root,
3129 * and the new block generated when it was split. We don't know which
3130 * one the cursor is pointing at, so we set up variables "left" and
3131 * "right" for each case.
3133 block
= xfs_btree_get_block(cur
, cur
->bc_nlevels
- 1, &bp
);
3136 error
= xfs_btree_check_block(cur
, block
, cur
->bc_nlevels
- 1, bp
);
3141 xfs_btree_get_sibling(cur
, block
, &rptr
, XFS_BB_RIGHTSIB
);
3142 if (!xfs_btree_ptr_is_null(cur
, &rptr
)) {
3143 /* Our block is left, pick up the right block. */
3145 xfs_btree_buf_to_ptr(cur
, lbp
, &lptr
);
3147 error
= xfs_btree_read_buf_block(cur
, &rptr
, 0, &right
, &rbp
);
3153 /* Our block is right, pick up the left block. */
3155 xfs_btree_buf_to_ptr(cur
, rbp
, &rptr
);
3157 xfs_btree_get_sibling(cur
, right
, &lptr
, XFS_BB_LEFTSIB
);
3158 error
= xfs_btree_read_buf_block(cur
, &lptr
, 0, &left
, &lbp
);
3165 /* Fill in the new block's btree header and log it. */
3166 xfs_btree_init_block_cur(cur
, nbp
, cur
->bc_nlevels
, 2);
3167 xfs_btree_log_block(cur
, nbp
, XFS_BB_ALL_BITS
);
3168 ASSERT(!xfs_btree_ptr_is_null(cur
, &lptr
) &&
3169 !xfs_btree_ptr_is_null(cur
, &rptr
));
3171 /* Fill in the key data in the new root. */
3172 if (xfs_btree_get_level(left
) > 0) {
3174 * Get the keys for the left block's keys and put them directly
3175 * in the parent block. Do the same for the right block.
3177 xfs_btree_get_node_keys(cur
, left
,
3178 xfs_btree_key_addr(cur
, 1, new));
3179 xfs_btree_get_node_keys(cur
, right
,
3180 xfs_btree_key_addr(cur
, 2, new));
3183 * Get the keys for the left block's records and put them
3184 * directly in the parent block. Do the same for the right
3187 xfs_btree_get_leaf_keys(cur
, left
,
3188 xfs_btree_key_addr(cur
, 1, new));
3189 xfs_btree_get_leaf_keys(cur
, right
,
3190 xfs_btree_key_addr(cur
, 2, new));
3192 xfs_btree_log_keys(cur
, nbp
, 1, 2);
3194 /* Fill in the pointer data in the new root. */
3195 xfs_btree_copy_ptrs(cur
,
3196 xfs_btree_ptr_addr(cur
, 1, new), &lptr
, 1);
3197 xfs_btree_copy_ptrs(cur
,
3198 xfs_btree_ptr_addr(cur
, 2, new), &rptr
, 1);
3199 xfs_btree_log_ptrs(cur
, nbp
, 1, 2);
3201 /* Fix up the cursor. */
3202 xfs_btree_setbuf(cur
, cur
->bc_nlevels
, nbp
);
3203 cur
->bc_ptrs
[cur
->bc_nlevels
] = nptr
;
3205 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3209 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3212 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3218 xfs_btree_make_block_unfull(
3219 struct xfs_btree_cur
*cur
, /* btree cursor */
3220 int level
, /* btree level */
3221 int numrecs
,/* # of recs in block */
3222 int *oindex
,/* old tree index */
3223 int *index
, /* new tree index */
3224 union xfs_btree_ptr
*nptr
, /* new btree ptr */
3225 struct xfs_btree_cur
**ncur
, /* new btree cursor */
3226 union xfs_btree_key
*key
, /* key of new block */
3231 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
3232 level
== cur
->bc_nlevels
- 1) {
3233 struct xfs_inode
*ip
= cur
->bc_private
.b
.ip
;
3235 if (numrecs
< cur
->bc_ops
->get_dmaxrecs(cur
, level
)) {
3236 /* A root block that can be made bigger. */
3237 xfs_iroot_realloc(ip
, 1, cur
->bc_private
.b
.whichfork
);
3240 /* A root block that needs replacing */
3243 error
= xfs_btree_new_iroot(cur
, &logflags
, stat
);
3244 if (error
|| *stat
== 0)
3247 xfs_trans_log_inode(cur
->bc_tp
, ip
, logflags
);
3253 /* First, try shifting an entry to the right neighbor. */
3254 error
= xfs_btree_rshift(cur
, level
, stat
);
3258 /* Next, try shifting an entry to the left neighbor. */
3259 error
= xfs_btree_lshift(cur
, level
, stat
);
3264 *oindex
= *index
= cur
->bc_ptrs
[level
];
3269 * Next, try splitting the current block in half.
3271 * If this works we have to re-set our variables because we
3272 * could be in a different block now.
3274 error
= xfs_btree_split(cur
, level
, nptr
, key
, ncur
, stat
);
3275 if (error
|| *stat
== 0)
3279 *index
= cur
->bc_ptrs
[level
];
3284 * Insert one record/level. Return information to the caller
3285 * allowing the next level up to proceed if necessary.
3289 struct xfs_btree_cur
*cur
, /* btree cursor */
3290 int level
, /* level to insert record at */
3291 union xfs_btree_ptr
*ptrp
, /* i/o: block number inserted */
3292 union xfs_btree_rec
*rec
, /* record to insert */
3293 union xfs_btree_key
*key
, /* i/o: block key for ptrp */
3294 struct xfs_btree_cur
**curp
, /* output: new cursor replacing cur */
3295 int *stat
) /* success/failure */
3297 struct xfs_btree_block
*block
; /* btree block */
3298 struct xfs_buf
*bp
; /* buffer for block */
3299 union xfs_btree_ptr nptr
; /* new block ptr */
3300 struct xfs_btree_cur
*ncur
; /* new btree cursor */
3301 union xfs_btree_key nkey
; /* new block key */
3302 union xfs_btree_key
*lkey
;
3303 int optr
; /* old key/record index */
3304 int ptr
; /* key/record index */
3305 int numrecs
;/* number of records */
3306 int error
; /* error return value */
3312 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
3313 XFS_BTREE_TRACE_ARGIPR(cur
, level
, *ptrp
, &rec
);
3319 * If we have an external root pointer, and we've made it to the
3320 * root level, allocate a new root block and we're done.
3322 if (!(cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
3323 (level
>= cur
->bc_nlevels
)) {
3324 error
= xfs_btree_new_root(cur
, stat
);
3325 xfs_btree_set_ptr_null(cur
, ptrp
);
3327 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3331 /* If we're off the left edge, return failure. */
3332 ptr
= cur
->bc_ptrs
[level
];
3334 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3341 XFS_BTREE_STATS_INC(cur
, insrec
);
3343 /* Get pointers to the btree buffer and block. */
3344 block
= xfs_btree_get_block(cur
, level
, &bp
);
3345 old_bn
= bp
? bp
->b_bn
: XFS_BUF_DADDR_NULL
;
3346 numrecs
= xfs_btree_get_numrecs(block
);
3349 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
3353 /* Check that the new entry is being inserted in the right place. */
3354 if (ptr
<= numrecs
) {
3356 ASSERT(cur
->bc_ops
->recs_inorder(cur
, rec
,
3357 xfs_btree_rec_addr(cur
, ptr
, block
)));
3359 ASSERT(cur
->bc_ops
->keys_inorder(cur
, key
,
3360 xfs_btree_key_addr(cur
, ptr
, block
)));
3366 * If the block is full, we can't insert the new entry until we
3367 * make the block un-full.
3369 xfs_btree_set_ptr_null(cur
, &nptr
);
3370 if (numrecs
== cur
->bc_ops
->get_maxrecs(cur
, level
)) {
3371 error
= xfs_btree_make_block_unfull(cur
, level
, numrecs
,
3372 &optr
, &ptr
, &nptr
, &ncur
, lkey
, stat
);
3373 if (error
|| *stat
== 0)
3378 * The current block may have changed if the block was
3379 * previously full and we have just made space in it.
3381 block
= xfs_btree_get_block(cur
, level
, &bp
);
3382 numrecs
= xfs_btree_get_numrecs(block
);
3385 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
3391 * At this point we know there's room for our new entry in the block
3392 * we're pointing at.
3394 XFS_BTREE_STATS_ADD(cur
, moves
, numrecs
- ptr
+ 1);
3397 /* It's a nonleaf. make a hole in the keys and ptrs */
3398 union xfs_btree_key
*kp
;
3399 union xfs_btree_ptr
*pp
;
3401 kp
= xfs_btree_key_addr(cur
, ptr
, block
);
3402 pp
= xfs_btree_ptr_addr(cur
, ptr
, block
);
3405 for (i
= numrecs
- ptr
; i
>= 0; i
--) {
3406 error
= xfs_btree_check_ptr(cur
, pp
, i
, level
);
3412 xfs_btree_shift_keys(cur
, kp
, 1, numrecs
- ptr
+ 1);
3413 xfs_btree_shift_ptrs(cur
, pp
, 1, numrecs
- ptr
+ 1);
3416 error
= xfs_btree_check_ptr(cur
, ptrp
, 0, level
);
3421 /* Now put the new data in, bump numrecs and log it. */
3422 xfs_btree_copy_keys(cur
, kp
, key
, 1);
3423 xfs_btree_copy_ptrs(cur
, pp
, ptrp
, 1);
3425 xfs_btree_set_numrecs(block
, numrecs
);
3426 xfs_btree_log_ptrs(cur
, bp
, ptr
, numrecs
);
3427 xfs_btree_log_keys(cur
, bp
, ptr
, numrecs
);
3429 if (ptr
< numrecs
) {
3430 ASSERT(cur
->bc_ops
->keys_inorder(cur
, kp
,
3431 xfs_btree_key_addr(cur
, ptr
+ 1, block
)));
3435 /* It's a leaf. make a hole in the records */
3436 union xfs_btree_rec
*rp
;
3438 rp
= xfs_btree_rec_addr(cur
, ptr
, block
);
3440 xfs_btree_shift_recs(cur
, rp
, 1, numrecs
- ptr
+ 1);
3442 /* Now put the new data in, bump numrecs and log it. */
3443 xfs_btree_copy_recs(cur
, rp
, rec
, 1);
3444 xfs_btree_set_numrecs(block
, ++numrecs
);
3445 xfs_btree_log_recs(cur
, bp
, ptr
, numrecs
);
3447 if (ptr
< numrecs
) {
3448 ASSERT(cur
->bc_ops
->recs_inorder(cur
, rp
,
3449 xfs_btree_rec_addr(cur
, ptr
+ 1, block
)));
3454 /* Log the new number of records in the btree header. */
3455 xfs_btree_log_block(cur
, bp
, XFS_BB_NUMRECS
);
3458 * If we just inserted into a new tree block, we have to
3459 * recalculate nkey here because nkey is out of date.
3461 * Otherwise we're just updating an existing block (having shoved
3462 * some records into the new tree block), so use the regular key
3465 if (bp
&& bp
->b_bn
!= old_bn
) {
3466 xfs_btree_get_keys(cur
, block
, lkey
);
3467 } else if (xfs_btree_needs_key_update(cur
, optr
)) {
3468 error
= xfs_btree_update_keys(cur
, level
);
3474 * If we are tracking the last record in the tree and
3475 * we are at the far right edge of the tree, update it.
3477 if (xfs_btree_is_lastrec(cur
, block
, level
)) {
3478 cur
->bc_ops
->update_lastrec(cur
, block
, rec
,
3479 ptr
, LASTREC_INSREC
);
3483 * Return the new block number, if any.
3484 * If there is one, give back a record value and a cursor too.
3487 if (!xfs_btree_ptr_is_null(cur
, &nptr
)) {
3488 xfs_btree_copy_keys(cur
, key
, lkey
, 1);
3492 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3497 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3502 * Insert the record at the point referenced by cur.
3504 * A multi-level split of the tree on insert will invalidate the original
3505 * cursor. All callers of this function should assume that the cursor is
3506 * no longer valid and revalidate it.
3510 struct xfs_btree_cur
*cur
,
3513 int error
; /* error return value */
3514 int i
; /* result value, 0 for failure */
3515 int level
; /* current level number in btree */
3516 union xfs_btree_ptr nptr
; /* new block number (split result) */
3517 struct xfs_btree_cur
*ncur
; /* new cursor (split result) */
3518 struct xfs_btree_cur
*pcur
; /* previous level's cursor */
3519 union xfs_btree_key bkey
; /* key of block to insert */
3520 union xfs_btree_key
*key
;
3521 union xfs_btree_rec rec
; /* record to insert */
3528 xfs_btree_set_ptr_null(cur
, &nptr
);
3530 /* Make a key out of the record data to be inserted, and save it. */
3531 cur
->bc_ops
->init_rec_from_cur(cur
, &rec
);
3532 cur
->bc_ops
->init_key_from_rec(key
, &rec
);
3535 * Loop going up the tree, starting at the leaf level.
3536 * Stop when we don't get a split block, that must mean that
3537 * the insert is finished with this level.
3541 * Insert nrec/nptr into this level of the tree.
3542 * Note if we fail, nptr will be null.
3544 error
= xfs_btree_insrec(pcur
, level
, &nptr
, &rec
, key
,
3548 xfs_btree_del_cursor(pcur
, XFS_BTREE_ERROR
);
3552 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3556 * See if the cursor we just used is trash.
3557 * Can't trash the caller's cursor, but otherwise we should
3558 * if ncur is a new cursor or we're about to be done.
3561 (ncur
|| xfs_btree_ptr_is_null(cur
, &nptr
))) {
3562 /* Save the state from the cursor before we trash it */
3563 if (cur
->bc_ops
->update_cursor
)
3564 cur
->bc_ops
->update_cursor(pcur
, cur
);
3565 cur
->bc_nlevels
= pcur
->bc_nlevels
;
3566 xfs_btree_del_cursor(pcur
, XFS_BTREE_NOERROR
);
3568 /* If we got a new cursor, switch to it. */
3573 } while (!xfs_btree_ptr_is_null(cur
, &nptr
));
3575 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3579 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3584 * Try to merge a non-leaf block back into the inode root.
3586 * Note: the killroot names comes from the fact that we're effectively
3587 * killing the old root block. But because we can't just delete the
3588 * inode we have to copy the single block it was pointing to into the
3592 xfs_btree_kill_iroot(
3593 struct xfs_btree_cur
*cur
)
3595 int whichfork
= cur
->bc_private
.b
.whichfork
;
3596 struct xfs_inode
*ip
= cur
->bc_private
.b
.ip
;
3597 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3598 struct xfs_btree_block
*block
;
3599 struct xfs_btree_block
*cblock
;
3600 union xfs_btree_key
*kp
;
3601 union xfs_btree_key
*ckp
;
3602 union xfs_btree_ptr
*pp
;
3603 union xfs_btree_ptr
*cpp
;
3604 struct xfs_buf
*cbp
;
3610 union xfs_btree_ptr ptr
;
3614 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
3616 ASSERT(cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
);
3617 ASSERT(cur
->bc_nlevels
> 1);
3620 * Don't deal with the root block needs to be a leaf case.
3621 * We're just going to turn the thing back into extents anyway.
3623 level
= cur
->bc_nlevels
- 1;
3628 * Give up if the root has multiple children.
3630 block
= xfs_btree_get_iroot(cur
);
3631 if (xfs_btree_get_numrecs(block
) != 1)
3634 cblock
= xfs_btree_get_block(cur
, level
- 1, &cbp
);
3635 numrecs
= xfs_btree_get_numrecs(cblock
);
3638 * Only do this if the next level will fit.
3639 * Then the data must be copied up to the inode,
3640 * instead of freeing the root you free the next level.
3642 if (numrecs
> cur
->bc_ops
->get_dmaxrecs(cur
, level
))
3645 XFS_BTREE_STATS_INC(cur
, killroot
);
3648 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_LEFTSIB
);
3649 ASSERT(xfs_btree_ptr_is_null(cur
, &ptr
));
3650 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_RIGHTSIB
);
3651 ASSERT(xfs_btree_ptr_is_null(cur
, &ptr
));
3654 index
= numrecs
- cur
->bc_ops
->get_maxrecs(cur
, level
);
3656 xfs_iroot_realloc(cur
->bc_private
.b
.ip
, index
,
3657 cur
->bc_private
.b
.whichfork
);
3658 block
= ifp
->if_broot
;
3661 be16_add_cpu(&block
->bb_numrecs
, index
);
3662 ASSERT(block
->bb_numrecs
== cblock
->bb_numrecs
);
3664 kp
= xfs_btree_key_addr(cur
, 1, block
);
3665 ckp
= xfs_btree_key_addr(cur
, 1, cblock
);
3666 xfs_btree_copy_keys(cur
, kp
, ckp
, numrecs
);
3668 pp
= xfs_btree_ptr_addr(cur
, 1, block
);
3669 cpp
= xfs_btree_ptr_addr(cur
, 1, cblock
);
3671 for (i
= 0; i
< numrecs
; i
++) {
3672 error
= xfs_btree_check_ptr(cur
, cpp
, i
, level
- 1);
3674 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3679 xfs_btree_copy_ptrs(cur
, pp
, cpp
, numrecs
);
3681 error
= xfs_btree_free_block(cur
, cbp
);
3683 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3687 cur
->bc_bufs
[level
- 1] = NULL
;
3688 be16_add_cpu(&block
->bb_level
, -1);
3689 xfs_trans_log_inode(cur
->bc_tp
, ip
,
3690 XFS_ILOG_CORE
| xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
));
3693 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3698 * Kill the current root node, and replace it with it's only child node.
3701 xfs_btree_kill_root(
3702 struct xfs_btree_cur
*cur
,
3705 union xfs_btree_ptr
*newroot
)
3709 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
3710 XFS_BTREE_STATS_INC(cur
, killroot
);
3713 * Update the root pointer, decreasing the level by 1 and then
3714 * free the old root.
3716 cur
->bc_ops
->set_root(cur
, newroot
, -1);
3718 error
= xfs_btree_free_block(cur
, bp
);
3720 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3724 cur
->bc_bufs
[level
] = NULL
;
3725 cur
->bc_ra
[level
] = 0;
3728 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3733 xfs_btree_dec_cursor(
3734 struct xfs_btree_cur
*cur
,
3742 error
= xfs_btree_decrement(cur
, level
, &i
);
3747 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3753 * Single level of the btree record deletion routine.
3754 * Delete record pointed to by cur/level.
3755 * Remove the record from its block then rebalance the tree.
3756 * Return 0 for error, 1 for done, 2 to go on to the next level.
3758 STATIC
int /* error */
3760 struct xfs_btree_cur
*cur
, /* btree cursor */
3761 int level
, /* level removing record from */
3762 int *stat
) /* fail/done/go-on */
3764 struct xfs_btree_block
*block
; /* btree block */
3765 union xfs_btree_ptr cptr
; /* current block ptr */
3766 struct xfs_buf
*bp
; /* buffer for block */
3767 int error
; /* error return value */
3768 int i
; /* loop counter */
3769 union xfs_btree_ptr lptr
; /* left sibling block ptr */
3770 struct xfs_buf
*lbp
; /* left buffer pointer */
3771 struct xfs_btree_block
*left
; /* left btree block */
3772 int lrecs
= 0; /* left record count */
3773 int ptr
; /* key/record index */
3774 union xfs_btree_ptr rptr
; /* right sibling block ptr */
3775 struct xfs_buf
*rbp
; /* right buffer pointer */
3776 struct xfs_btree_block
*right
; /* right btree block */
3777 struct xfs_btree_block
*rrblock
; /* right-right btree block */
3778 struct xfs_buf
*rrbp
; /* right-right buffer pointer */
3779 int rrecs
= 0; /* right record count */
3780 struct xfs_btree_cur
*tcur
; /* temporary btree cursor */
3781 int numrecs
; /* temporary numrec count */
3783 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
3784 XFS_BTREE_TRACE_ARGI(cur
, level
);
3788 /* Get the index of the entry being deleted, check for nothing there. */
3789 ptr
= cur
->bc_ptrs
[level
];
3791 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3796 /* Get the buffer & block containing the record or key/ptr. */
3797 block
= xfs_btree_get_block(cur
, level
, &bp
);
3798 numrecs
= xfs_btree_get_numrecs(block
);
3801 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
3806 /* Fail if we're off the end of the block. */
3807 if (ptr
> numrecs
) {
3808 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3813 XFS_BTREE_STATS_INC(cur
, delrec
);
3814 XFS_BTREE_STATS_ADD(cur
, moves
, numrecs
- ptr
);
3816 /* Excise the entries being deleted. */
3818 /* It's a nonleaf. operate on keys and ptrs */
3819 union xfs_btree_key
*lkp
;
3820 union xfs_btree_ptr
*lpp
;
3822 lkp
= xfs_btree_key_addr(cur
, ptr
+ 1, block
);
3823 lpp
= xfs_btree_ptr_addr(cur
, ptr
+ 1, block
);
3826 for (i
= 0; i
< numrecs
- ptr
; i
++) {
3827 error
= xfs_btree_check_ptr(cur
, lpp
, i
, level
);
3833 if (ptr
< numrecs
) {
3834 xfs_btree_shift_keys(cur
, lkp
, -1, numrecs
- ptr
);
3835 xfs_btree_shift_ptrs(cur
, lpp
, -1, numrecs
- ptr
);
3836 xfs_btree_log_keys(cur
, bp
, ptr
, numrecs
- 1);
3837 xfs_btree_log_ptrs(cur
, bp
, ptr
, numrecs
- 1);
3840 /* It's a leaf. operate on records */
3841 if (ptr
< numrecs
) {
3842 xfs_btree_shift_recs(cur
,
3843 xfs_btree_rec_addr(cur
, ptr
+ 1, block
),
3845 xfs_btree_log_recs(cur
, bp
, ptr
, numrecs
- 1);
3850 * Decrement and log the number of entries in the block.
3852 xfs_btree_set_numrecs(block
, --numrecs
);
3853 xfs_btree_log_block(cur
, bp
, XFS_BB_NUMRECS
);
3856 * If we are tracking the last record in the tree and
3857 * we are at the far right edge of the tree, update it.
3859 if (xfs_btree_is_lastrec(cur
, block
, level
)) {
3860 cur
->bc_ops
->update_lastrec(cur
, block
, NULL
,
3861 ptr
, LASTREC_DELREC
);
3865 * We're at the root level. First, shrink the root block in-memory.
3866 * Try to get rid of the next level down. If we can't then there's
3867 * nothing left to do.
3869 if (level
== cur
->bc_nlevels
- 1) {
3870 if (cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) {
3871 xfs_iroot_realloc(cur
->bc_private
.b
.ip
, -1,
3872 cur
->bc_private
.b
.whichfork
);
3874 error
= xfs_btree_kill_iroot(cur
);
3878 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3886 * If this is the root level, and there's only one entry left,
3887 * and it's NOT the leaf level, then we can get rid of this
3890 if (numrecs
== 1 && level
> 0) {
3891 union xfs_btree_ptr
*pp
;
3893 * pp is still set to the first pointer in the block.
3894 * Make it the new root of the btree.
3896 pp
= xfs_btree_ptr_addr(cur
, 1, block
);
3897 error
= xfs_btree_kill_root(cur
, bp
, level
, pp
);
3900 } else if (level
> 0) {
3901 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3910 * If we deleted the leftmost entry in the block, update the
3911 * key values above us in the tree.
3913 if (xfs_btree_needs_key_update(cur
, ptr
)) {
3914 error
= xfs_btree_update_keys(cur
, level
);
3920 * If the number of records remaining in the block is at least
3921 * the minimum, we're done.
3923 if (numrecs
>= cur
->bc_ops
->get_minrecs(cur
, level
)) {
3924 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3931 * Otherwise, we have to move some records around to keep the
3932 * tree balanced. Look at the left and right sibling blocks to
3933 * see if we can re-balance by moving only one record.
3935 xfs_btree_get_sibling(cur
, block
, &rptr
, XFS_BB_RIGHTSIB
);
3936 xfs_btree_get_sibling(cur
, block
, &lptr
, XFS_BB_LEFTSIB
);
3938 if (cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) {
3940 * One child of root, need to get a chance to copy its contents
3941 * into the root and delete it. Can't go up to next level,
3942 * there's nothing to delete there.
3944 if (xfs_btree_ptr_is_null(cur
, &rptr
) &&
3945 xfs_btree_ptr_is_null(cur
, &lptr
) &&
3946 level
== cur
->bc_nlevels
- 2) {
3947 error
= xfs_btree_kill_iroot(cur
);
3949 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3956 ASSERT(!xfs_btree_ptr_is_null(cur
, &rptr
) ||
3957 !xfs_btree_ptr_is_null(cur
, &lptr
));
3960 * Duplicate the cursor so our btree manipulations here won't
3961 * disrupt the next level up.
3963 error
= xfs_btree_dup_cursor(cur
, &tcur
);
3968 * If there's a right sibling, see if it's ok to shift an entry
3971 if (!xfs_btree_ptr_is_null(cur
, &rptr
)) {
3973 * Move the temp cursor to the last entry in the next block.
3974 * Actually any entry but the first would suffice.
3976 i
= xfs_btree_lastrec(tcur
, level
);
3977 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3979 error
= xfs_btree_increment(tcur
, level
, &i
);
3982 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3984 i
= xfs_btree_lastrec(tcur
, level
);
3985 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3987 /* Grab a pointer to the block. */
3988 right
= xfs_btree_get_block(tcur
, level
, &rbp
);
3990 error
= xfs_btree_check_block(tcur
, right
, level
, rbp
);
3994 /* Grab the current block number, for future use. */
3995 xfs_btree_get_sibling(tcur
, right
, &cptr
, XFS_BB_LEFTSIB
);
3998 * If right block is full enough so that removing one entry
3999 * won't make it too empty, and left-shifting an entry out
4000 * of right to us works, we're done.
4002 if (xfs_btree_get_numrecs(right
) - 1 >=
4003 cur
->bc_ops
->get_minrecs(tcur
, level
)) {
4004 error
= xfs_btree_lshift(tcur
, level
, &i
);
4008 ASSERT(xfs_btree_get_numrecs(block
) >=
4009 cur
->bc_ops
->get_minrecs(tcur
, level
));
4011 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
4014 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
4022 * Otherwise, grab the number of records in right for
4023 * future reference, and fix up the temp cursor to point
4024 * to our block again (last record).
4026 rrecs
= xfs_btree_get_numrecs(right
);
4027 if (!xfs_btree_ptr_is_null(cur
, &lptr
)) {
4028 i
= xfs_btree_firstrec(tcur
, level
);
4029 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
4031 error
= xfs_btree_decrement(tcur
, level
, &i
);
4034 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
4039 * If there's a left sibling, see if it's ok to shift an entry
4042 if (!xfs_btree_ptr_is_null(cur
, &lptr
)) {
4044 * Move the temp cursor to the first entry in the
4047 i
= xfs_btree_firstrec(tcur
, level
);
4048 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
4050 error
= xfs_btree_decrement(tcur
, level
, &i
);
4053 i
= xfs_btree_firstrec(tcur
, level
);
4054 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
4056 /* Grab a pointer to the block. */
4057 left
= xfs_btree_get_block(tcur
, level
, &lbp
);
4059 error
= xfs_btree_check_block(cur
, left
, level
, lbp
);
4063 /* Grab the current block number, for future use. */
4064 xfs_btree_get_sibling(tcur
, left
, &cptr
, XFS_BB_RIGHTSIB
);
4067 * If left block is full enough so that removing one entry
4068 * won't make it too empty, and right-shifting an entry out
4069 * of left to us works, we're done.
4071 if (xfs_btree_get_numrecs(left
) - 1 >=
4072 cur
->bc_ops
->get_minrecs(tcur
, level
)) {
4073 error
= xfs_btree_rshift(tcur
, level
, &i
);
4077 ASSERT(xfs_btree_get_numrecs(block
) >=
4078 cur
->bc_ops
->get_minrecs(tcur
, level
));
4079 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
4083 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
4090 * Otherwise, grab the number of records in right for
4093 lrecs
= xfs_btree_get_numrecs(left
);
4096 /* Delete the temp cursor, we're done with it. */
4097 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
4100 /* If here, we need to do a join to keep the tree balanced. */
4101 ASSERT(!xfs_btree_ptr_is_null(cur
, &cptr
));
4103 if (!xfs_btree_ptr_is_null(cur
, &lptr
) &&
4104 lrecs
+ xfs_btree_get_numrecs(block
) <=
4105 cur
->bc_ops
->get_maxrecs(cur
, level
)) {
4107 * Set "right" to be the starting block,
4108 * "left" to be the left neighbor.
4113 error
= xfs_btree_read_buf_block(cur
, &lptr
, 0, &left
, &lbp
);
4118 * If that won't work, see if we can join with the right neighbor block.
4120 } else if (!xfs_btree_ptr_is_null(cur
, &rptr
) &&
4121 rrecs
+ xfs_btree_get_numrecs(block
) <=
4122 cur
->bc_ops
->get_maxrecs(cur
, level
)) {
4124 * Set "left" to be the starting block,
4125 * "right" to be the right neighbor.
4130 error
= xfs_btree_read_buf_block(cur
, &rptr
, 0, &right
, &rbp
);
4135 * Otherwise, we can't fix the imbalance.
4136 * Just return. This is probably a logic error, but it's not fatal.
4139 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
4145 rrecs
= xfs_btree_get_numrecs(right
);
4146 lrecs
= xfs_btree_get_numrecs(left
);
4149 * We're now going to join "left" and "right" by moving all the stuff
4150 * in "right" to "left" and deleting "right".
4152 XFS_BTREE_STATS_ADD(cur
, moves
, rrecs
);
4154 /* It's a non-leaf. Move keys and pointers. */
4155 union xfs_btree_key
*lkp
; /* left btree key */
4156 union xfs_btree_ptr
*lpp
; /* left address pointer */
4157 union xfs_btree_key
*rkp
; /* right btree key */
4158 union xfs_btree_ptr
*rpp
; /* right address pointer */
4160 lkp
= xfs_btree_key_addr(cur
, lrecs
+ 1, left
);
4161 lpp
= xfs_btree_ptr_addr(cur
, lrecs
+ 1, left
);
4162 rkp
= xfs_btree_key_addr(cur
, 1, right
);
4163 rpp
= xfs_btree_ptr_addr(cur
, 1, right
);
4165 for (i
= 1; i
< rrecs
; i
++) {
4166 error
= xfs_btree_check_ptr(cur
, rpp
, i
, level
);
4171 xfs_btree_copy_keys(cur
, lkp
, rkp
, rrecs
);
4172 xfs_btree_copy_ptrs(cur
, lpp
, rpp
, rrecs
);
4174 xfs_btree_log_keys(cur
, lbp
, lrecs
+ 1, lrecs
+ rrecs
);
4175 xfs_btree_log_ptrs(cur
, lbp
, lrecs
+ 1, lrecs
+ rrecs
);
4177 /* It's a leaf. Move records. */
4178 union xfs_btree_rec
*lrp
; /* left record pointer */
4179 union xfs_btree_rec
*rrp
; /* right record pointer */
4181 lrp
= xfs_btree_rec_addr(cur
, lrecs
+ 1, left
);
4182 rrp
= xfs_btree_rec_addr(cur
, 1, right
);
4184 xfs_btree_copy_recs(cur
, lrp
, rrp
, rrecs
);
4185 xfs_btree_log_recs(cur
, lbp
, lrecs
+ 1, lrecs
+ rrecs
);
4188 XFS_BTREE_STATS_INC(cur
, join
);
4191 * Fix up the number of records and right block pointer in the
4192 * surviving block, and log it.
4194 xfs_btree_set_numrecs(left
, lrecs
+ rrecs
);
4195 xfs_btree_get_sibling(cur
, right
, &cptr
, XFS_BB_RIGHTSIB
),
4196 xfs_btree_set_sibling(cur
, left
, &cptr
, XFS_BB_RIGHTSIB
);
4197 xfs_btree_log_block(cur
, lbp
, XFS_BB_NUMRECS
| XFS_BB_RIGHTSIB
);
4199 /* If there is a right sibling, point it to the remaining block. */
4200 xfs_btree_get_sibling(cur
, left
, &cptr
, XFS_BB_RIGHTSIB
);
4201 if (!xfs_btree_ptr_is_null(cur
, &cptr
)) {
4202 error
= xfs_btree_read_buf_block(cur
, &cptr
, 0, &rrblock
, &rrbp
);
4205 xfs_btree_set_sibling(cur
, rrblock
, &lptr
, XFS_BB_LEFTSIB
);
4206 xfs_btree_log_block(cur
, rrbp
, XFS_BB_LEFTSIB
);
4209 /* Free the deleted block. */
4210 error
= xfs_btree_free_block(cur
, rbp
);
4215 * If we joined with the left neighbor, set the buffer in the
4216 * cursor to the left block, and fix up the index.
4219 cur
->bc_bufs
[level
] = lbp
;
4220 cur
->bc_ptrs
[level
] += lrecs
;
4221 cur
->bc_ra
[level
] = 0;
4224 * If we joined with the right neighbor and there's a level above
4225 * us, increment the cursor at that level.
4227 else if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) ||
4228 (level
+ 1 < cur
->bc_nlevels
)) {
4229 error
= xfs_btree_increment(cur
, level
+ 1, &i
);
4235 * Readjust the ptr at this level if it's not a leaf, since it's
4236 * still pointing at the deletion point, which makes the cursor
4237 * inconsistent. If this makes the ptr 0, the caller fixes it up.
4238 * We can't use decrement because it would change the next level up.
4241 cur
->bc_ptrs
[level
]--;
4244 * We combined blocks, so we have to update the parent keys if the
4245 * btree supports overlapped intervals. However, bc_ptrs[level + 1]
4246 * points to the old block so that the caller knows which record to
4247 * delete. Therefore, the caller must be savvy enough to call updkeys
4248 * for us if we return stat == 2. The other exit points from this
4249 * function don't require deletions further up the tree, so they can
4250 * call updkeys directly.
4253 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
4254 /* Return value means the next level up has something to do. */
4259 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
4261 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
4266 * Delete the record pointed to by cur.
4267 * The cursor refers to the place where the record was (could be inserted)
4268 * when the operation returns.
4272 struct xfs_btree_cur
*cur
,
4273 int *stat
) /* success/failure */
4275 int error
; /* error return value */
4278 bool joined
= false;
4280 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
4283 * Go up the tree, starting at leaf level.
4285 * If 2 is returned then a join was done; go to the next level.
4286 * Otherwise we are done.
4288 for (level
= 0, i
= 2; i
== 2; level
++) {
4289 error
= xfs_btree_delrec(cur
, level
, &i
);
4297 * If we combined blocks as part of deleting the record, delrec won't
4298 * have updated the parent high keys so we have to do that here.
4300 if (joined
&& (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
)) {
4301 error
= xfs_btree_updkeys_force(cur
, 0);
4307 for (level
= 1; level
< cur
->bc_nlevels
; level
++) {
4308 if (cur
->bc_ptrs
[level
] == 0) {
4309 error
= xfs_btree_decrement(cur
, level
, &i
);
4317 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
4321 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
4326 * Get the data from the pointed-to record.
4330 struct xfs_btree_cur
*cur
, /* btree cursor */
4331 union xfs_btree_rec
**recp
, /* output: btree record */
4332 int *stat
) /* output: success/failure */
4334 struct xfs_btree_block
*block
; /* btree block */
4335 struct xfs_buf
*bp
; /* buffer pointer */
4336 int ptr
; /* record number */
4338 int error
; /* error return value */
4341 ptr
= cur
->bc_ptrs
[0];
4342 block
= xfs_btree_get_block(cur
, 0, &bp
);
4345 error
= xfs_btree_check_block(cur
, block
, 0, bp
);
4351 * Off the right end or left end, return failure.
4353 if (ptr
> xfs_btree_get_numrecs(block
) || ptr
<= 0) {
4359 * Point to the record and extract its data.
4361 *recp
= xfs_btree_rec_addr(cur
, ptr
, block
);
4366 /* Visit a block in a btree. */
4368 xfs_btree_visit_block(
4369 struct xfs_btree_cur
*cur
,
4371 xfs_btree_visit_blocks_fn fn
,
4374 struct xfs_btree_block
*block
;
4376 union xfs_btree_ptr rptr
;
4379 /* do right sibling readahead */
4380 xfs_btree_readahead(cur
, level
, XFS_BTCUR_RIGHTRA
);
4381 block
= xfs_btree_get_block(cur
, level
, &bp
);
4383 /* process the block */
4384 error
= fn(cur
, level
, data
);
4388 /* now read rh sibling block for next iteration */
4389 xfs_btree_get_sibling(cur
, block
, &rptr
, XFS_BB_RIGHTSIB
);
4390 if (xfs_btree_ptr_is_null(cur
, &rptr
))
4393 return xfs_btree_lookup_get_block(cur
, level
, &rptr
, &block
);
4397 /* Visit every block in a btree. */
4399 xfs_btree_visit_blocks(
4400 struct xfs_btree_cur
*cur
,
4401 xfs_btree_visit_blocks_fn fn
,
4404 union xfs_btree_ptr lptr
;
4406 struct xfs_btree_block
*block
= NULL
;
4409 cur
->bc_ops
->init_ptr_from_cur(cur
, &lptr
);
4411 /* for each level */
4412 for (level
= cur
->bc_nlevels
- 1; level
>= 0; level
--) {
4413 /* grab the left hand block */
4414 error
= xfs_btree_lookup_get_block(cur
, level
, &lptr
, &block
);
4418 /* readahead the left most block for the next level down */
4420 union xfs_btree_ptr
*ptr
;
4422 ptr
= xfs_btree_ptr_addr(cur
, 1, block
);
4423 xfs_btree_readahead_ptr(cur
, ptr
, 1);
4425 /* save for the next iteration of the loop */
4426 xfs_btree_copy_ptrs(cur
, &lptr
, ptr
, 1);
4429 /* for each buffer in the level */
4431 error
= xfs_btree_visit_block(cur
, level
, fn
, data
);
4434 if (error
!= -ENOENT
)
4442 * Change the owner of a btree.
4444 * The mechanism we use here is ordered buffer logging. Because we don't know
4445 * how many buffers were are going to need to modify, we don't really want to
4446 * have to make transaction reservations for the worst case of every buffer in a
4447 * full size btree as that may be more space that we can fit in the log....
4449 * We do the btree walk in the most optimal manner possible - we have sibling
4450 * pointers so we can just walk all the blocks on each level from left to right
4451 * in a single pass, and then move to the next level and do the same. We can
4452 * also do readahead on the sibling pointers to get IO moving more quickly,
4453 * though for slow disks this is unlikely to make much difference to performance
4454 * as the amount of CPU work we have to do before moving to the next block is
4457 * For each btree block that we load, modify the owner appropriately, set the
4458 * buffer as an ordered buffer and log it appropriately. We need to ensure that
4459 * we mark the region we change dirty so that if the buffer is relogged in
4460 * a subsequent transaction the changes we make here as an ordered buffer are
4461 * correctly relogged in that transaction. If we are in recovery context, then
4462 * just queue the modified buffer as delayed write buffer so the transaction
4463 * recovery completion writes the changes to disk.
4465 struct xfs_btree_block_change_owner_info
{
4467 struct list_head
*buffer_list
;
4471 xfs_btree_block_change_owner(
4472 struct xfs_btree_cur
*cur
,
4476 struct xfs_btree_block_change_owner_info
*bbcoi
= data
;
4477 struct xfs_btree_block
*block
;
4480 /* modify the owner */
4481 block
= xfs_btree_get_block(cur
, level
, &bp
);
4482 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
4483 if (block
->bb_u
.l
.bb_owner
== cpu_to_be64(bbcoi
->new_owner
))
4485 block
->bb_u
.l
.bb_owner
= cpu_to_be64(bbcoi
->new_owner
);
4487 if (block
->bb_u
.s
.bb_owner
== cpu_to_be32(bbcoi
->new_owner
))
4489 block
->bb_u
.s
.bb_owner
= cpu_to_be32(bbcoi
->new_owner
);
4493 * If the block is a root block hosted in an inode, we might not have a
4494 * buffer pointer here and we shouldn't attempt to log the change as the
4495 * information is already held in the inode and discarded when the root
4496 * block is formatted into the on-disk inode fork. We still change it,
4497 * though, so everything is consistent in memory.
4500 ASSERT(cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
);
4501 ASSERT(level
== cur
->bc_nlevels
- 1);
4506 if (!xfs_trans_ordered_buf(cur
->bc_tp
, bp
)) {
4507 xfs_btree_log_block(cur
, bp
, XFS_BB_OWNER
);
4511 xfs_buf_delwri_queue(bp
, bbcoi
->buffer_list
);
4518 xfs_btree_change_owner(
4519 struct xfs_btree_cur
*cur
,
4521 struct list_head
*buffer_list
)
4523 struct xfs_btree_block_change_owner_info bbcoi
;
4525 bbcoi
.new_owner
= new_owner
;
4526 bbcoi
.buffer_list
= buffer_list
;
4528 return xfs_btree_visit_blocks(cur
, xfs_btree_block_change_owner
,
4532 /* Verify the v5 fields of a long-format btree block. */
4534 xfs_btree_lblock_v5hdr_verify(
4538 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
4539 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
4541 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
4542 return __this_address
;
4543 if (!uuid_equal(&block
->bb_u
.l
.bb_uuid
, &mp
->m_sb
.sb_meta_uuid
))
4544 return __this_address
;
4545 if (block
->bb_u
.l
.bb_blkno
!= cpu_to_be64(bp
->b_bn
))
4546 return __this_address
;
4547 if (owner
!= XFS_RMAP_OWN_UNKNOWN
&&
4548 be64_to_cpu(block
->bb_u
.l
.bb_owner
) != owner
)
4549 return __this_address
;
4553 /* Verify a long-format btree block. */
4555 xfs_btree_lblock_verify(
4557 unsigned int max_recs
)
4559 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
4560 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
4562 /* numrecs verification */
4563 if (be16_to_cpu(block
->bb_numrecs
) > max_recs
)
4564 return __this_address
;
4566 /* sibling pointer verification */
4567 if (block
->bb_u
.l
.bb_leftsib
!= cpu_to_be64(NULLFSBLOCK
) &&
4568 !xfs_verify_fsbno(mp
, be64_to_cpu(block
->bb_u
.l
.bb_leftsib
)))
4569 return __this_address
;
4570 if (block
->bb_u
.l
.bb_rightsib
!= cpu_to_be64(NULLFSBLOCK
) &&
4571 !xfs_verify_fsbno(mp
, be64_to_cpu(block
->bb_u
.l
.bb_rightsib
)))
4572 return __this_address
;
4578 * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format
4581 * @bp: buffer containing the btree block
4582 * @max_recs: pointer to the m_*_mxr max records field in the xfs mount
4583 * @pag_max_level: pointer to the per-ag max level field
4586 xfs_btree_sblock_v5hdr_verify(
4589 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
4590 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
4591 struct xfs_perag
*pag
= bp
->b_pag
;
4593 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
4594 return __this_address
;
4595 if (!uuid_equal(&block
->bb_u
.s
.bb_uuid
, &mp
->m_sb
.sb_meta_uuid
))
4596 return __this_address
;
4597 if (block
->bb_u
.s
.bb_blkno
!= cpu_to_be64(bp
->b_bn
))
4598 return __this_address
;
4599 if (pag
&& be32_to_cpu(block
->bb_u
.s
.bb_owner
) != pag
->pag_agno
)
4600 return __this_address
;
4605 * xfs_btree_sblock_verify() -- verify a short-format btree block
4607 * @bp: buffer containing the btree block
4608 * @max_recs: maximum records allowed in this btree node
4611 xfs_btree_sblock_verify(
4613 unsigned int max_recs
)
4615 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
4616 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
4619 /* numrecs verification */
4620 if (be16_to_cpu(block
->bb_numrecs
) > max_recs
)
4621 return __this_address
;
4623 /* sibling pointer verification */
4624 agno
= xfs_daddr_to_agno(mp
, XFS_BUF_ADDR(bp
));
4625 if (block
->bb_u
.s
.bb_leftsib
!= cpu_to_be32(NULLAGBLOCK
) &&
4626 !xfs_verify_agbno(mp
, agno
, be32_to_cpu(block
->bb_u
.s
.bb_leftsib
)))
4627 return __this_address
;
4628 if (block
->bb_u
.s
.bb_rightsib
!= cpu_to_be32(NULLAGBLOCK
) &&
4629 !xfs_verify_agbno(mp
, agno
, be32_to_cpu(block
->bb_u
.s
.bb_rightsib
)))
4630 return __this_address
;
4636 * Calculate the number of btree levels needed to store a given number of
4637 * records in a short-format btree.
4640 xfs_btree_compute_maxlevels(
4641 struct xfs_mount
*mp
,
4646 unsigned long maxblocks
;
4648 maxblocks
= (len
+ limits
[0] - 1) / limits
[0];
4649 for (level
= 1; maxblocks
> 1; level
++)
4650 maxblocks
= (maxblocks
+ limits
[1] - 1) / limits
[1];
4655 * Query a regular btree for all records overlapping a given interval.
4656 * Start with a LE lookup of the key of low_rec and return all records
4657 * until we find a record with a key greater than the key of high_rec.
4660 xfs_btree_simple_query_range(
4661 struct xfs_btree_cur
*cur
,
4662 union xfs_btree_key
*low_key
,
4663 union xfs_btree_key
*high_key
,
4664 xfs_btree_query_range_fn fn
,
4667 union xfs_btree_rec
*recp
;
4668 union xfs_btree_key rec_key
;
4671 bool firstrec
= true;
4674 ASSERT(cur
->bc_ops
->init_high_key_from_rec
);
4675 ASSERT(cur
->bc_ops
->diff_two_keys
);
4678 * Find the leftmost record. The btree cursor must be set
4679 * to the low record used to generate low_key.
4682 error
= xfs_btree_lookup(cur
, XFS_LOOKUP_LE
, &stat
);
4686 /* Nothing? See if there's anything to the right. */
4688 error
= xfs_btree_increment(cur
, 0, &stat
);
4694 /* Find the record. */
4695 error
= xfs_btree_get_rec(cur
, &recp
, &stat
);
4699 /* Skip if high_key(rec) < low_key. */
4701 cur
->bc_ops
->init_high_key_from_rec(&rec_key
, recp
);
4703 diff
= cur
->bc_ops
->diff_two_keys(cur
, low_key
,
4709 /* Stop if high_key < low_key(rec). */
4710 cur
->bc_ops
->init_key_from_rec(&rec_key
, recp
);
4711 diff
= cur
->bc_ops
->diff_two_keys(cur
, &rec_key
, high_key
);
4716 error
= fn(cur
, recp
, priv
);
4717 if (error
< 0 || error
== XFS_BTREE_QUERY_RANGE_ABORT
)
4721 /* Move on to the next record. */
4722 error
= xfs_btree_increment(cur
, 0, &stat
);
4732 * Query an overlapped interval btree for all records overlapping a given
4733 * interval. This function roughly follows the algorithm given in
4734 * "Interval Trees" of _Introduction to Algorithms_, which is section
4735 * 14.3 in the 2nd and 3rd editions.
4737 * First, generate keys for the low and high records passed in.
4739 * For any leaf node, generate the high and low keys for the record.
4740 * If the record keys overlap with the query low/high keys, pass the
4741 * record to the function iterator.
4743 * For any internal node, compare the low and high keys of each
4744 * pointer against the query low/high keys. If there's an overlap,
4745 * follow the pointer.
4747 * As an optimization, we stop scanning a block when we find a low key
4748 * that is greater than the query's high key.
4751 xfs_btree_overlapped_query_range(
4752 struct xfs_btree_cur
*cur
,
4753 union xfs_btree_key
*low_key
,
4754 union xfs_btree_key
*high_key
,
4755 xfs_btree_query_range_fn fn
,
4758 union xfs_btree_ptr ptr
;
4759 union xfs_btree_ptr
*pp
;
4760 union xfs_btree_key rec_key
;
4761 union xfs_btree_key rec_hkey
;
4762 union xfs_btree_key
*lkp
;
4763 union xfs_btree_key
*hkp
;
4764 union xfs_btree_rec
*recp
;
4765 struct xfs_btree_block
*block
;
4773 /* Load the root of the btree. */
4774 level
= cur
->bc_nlevels
- 1;
4775 cur
->bc_ops
->init_ptr_from_cur(cur
, &ptr
);
4776 error
= xfs_btree_lookup_get_block(cur
, level
, &ptr
, &block
);
4779 xfs_btree_get_block(cur
, level
, &bp
);
4780 trace_xfs_btree_overlapped_query_range(cur
, level
, bp
);
4782 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
4786 cur
->bc_ptrs
[level
] = 1;
4788 while (level
< cur
->bc_nlevels
) {
4789 block
= xfs_btree_get_block(cur
, level
, &bp
);
4791 /* End of node, pop back towards the root. */
4792 if (cur
->bc_ptrs
[level
] > be16_to_cpu(block
->bb_numrecs
)) {
4794 if (level
< cur
->bc_nlevels
- 1)
4795 cur
->bc_ptrs
[level
+ 1]++;
4801 /* Handle a leaf node. */
4802 recp
= xfs_btree_rec_addr(cur
, cur
->bc_ptrs
[0], block
);
4804 cur
->bc_ops
->init_high_key_from_rec(&rec_hkey
, recp
);
4805 ldiff
= cur
->bc_ops
->diff_two_keys(cur
, &rec_hkey
,
4808 cur
->bc_ops
->init_key_from_rec(&rec_key
, recp
);
4809 hdiff
= cur
->bc_ops
->diff_two_keys(cur
, high_key
,
4813 * If (record's high key >= query's low key) and
4814 * (query's high key >= record's low key), then
4815 * this record overlaps the query range; callback.
4817 if (ldiff
>= 0 && hdiff
>= 0) {
4818 error
= fn(cur
, recp
, priv
);
4820 error
== XFS_BTREE_QUERY_RANGE_ABORT
)
4822 } else if (hdiff
< 0) {
4823 /* Record is larger than high key; pop. */
4826 cur
->bc_ptrs
[level
]++;
4830 /* Handle an internal node. */
4831 lkp
= xfs_btree_key_addr(cur
, cur
->bc_ptrs
[level
], block
);
4832 hkp
= xfs_btree_high_key_addr(cur
, cur
->bc_ptrs
[level
], block
);
4833 pp
= xfs_btree_ptr_addr(cur
, cur
->bc_ptrs
[level
], block
);
4835 ldiff
= cur
->bc_ops
->diff_two_keys(cur
, hkp
, low_key
);
4836 hdiff
= cur
->bc_ops
->diff_two_keys(cur
, high_key
, lkp
);
4839 * If (pointer's high key >= query's low key) and
4840 * (query's high key >= pointer's low key), then
4841 * this record overlaps the query range; follow pointer.
4843 if (ldiff
>= 0 && hdiff
>= 0) {
4845 error
= xfs_btree_lookup_get_block(cur
, level
, pp
,
4849 xfs_btree_get_block(cur
, level
, &bp
);
4850 trace_xfs_btree_overlapped_query_range(cur
, level
, bp
);
4852 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
4856 cur
->bc_ptrs
[level
] = 1;
4858 } else if (hdiff
< 0) {
4859 /* The low key is larger than the upper range; pop. */
4862 cur
->bc_ptrs
[level
]++;
4867 * If we don't end this function with the cursor pointing at a record
4868 * block, a subsequent non-error cursor deletion will not release
4869 * node-level buffers, causing a buffer leak. This is quite possible
4870 * with a zero-results range query, so release the buffers if we
4871 * failed to return any results.
4873 if (cur
->bc_bufs
[0] == NULL
) {
4874 for (i
= 0; i
< cur
->bc_nlevels
; i
++) {
4875 if (cur
->bc_bufs
[i
]) {
4876 xfs_trans_brelse(cur
->bc_tp
, cur
->bc_bufs
[i
]);
4877 cur
->bc_bufs
[i
] = NULL
;
4878 cur
->bc_ptrs
[i
] = 0;
4888 * Query a btree for all records overlapping a given interval of keys. The
4889 * supplied function will be called with each record found; return one of the
4890 * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
4891 * code. This function returns XFS_BTREE_QUERY_RANGE_ABORT, zero, or a
4892 * negative error code.
4895 xfs_btree_query_range(
4896 struct xfs_btree_cur
*cur
,
4897 union xfs_btree_irec
*low_rec
,
4898 union xfs_btree_irec
*high_rec
,
4899 xfs_btree_query_range_fn fn
,
4902 union xfs_btree_rec rec
;
4903 union xfs_btree_key low_key
;
4904 union xfs_btree_key high_key
;
4906 /* Find the keys of both ends of the interval. */
4907 cur
->bc_rec
= *high_rec
;
4908 cur
->bc_ops
->init_rec_from_cur(cur
, &rec
);
4909 cur
->bc_ops
->init_key_from_rec(&high_key
, &rec
);
4911 cur
->bc_rec
= *low_rec
;
4912 cur
->bc_ops
->init_rec_from_cur(cur
, &rec
);
4913 cur
->bc_ops
->init_key_from_rec(&low_key
, &rec
);
4915 /* Enforce low key < high key. */
4916 if (cur
->bc_ops
->diff_two_keys(cur
, &low_key
, &high_key
) > 0)
4919 if (!(cur
->bc_flags
& XFS_BTREE_OVERLAPPING
))
4920 return xfs_btree_simple_query_range(cur
, &low_key
,
4921 &high_key
, fn
, priv
);
4922 return xfs_btree_overlapped_query_range(cur
, &low_key
, &high_key
,
4926 /* Query a btree for all records. */
4928 xfs_btree_query_all(
4929 struct xfs_btree_cur
*cur
,
4930 xfs_btree_query_range_fn fn
,
4933 union xfs_btree_key low_key
;
4934 union xfs_btree_key high_key
;
4936 memset(&cur
->bc_rec
, 0, sizeof(cur
->bc_rec
));
4937 memset(&low_key
, 0, sizeof(low_key
));
4938 memset(&high_key
, 0xFF, sizeof(high_key
));
4940 return xfs_btree_simple_query_range(cur
, &low_key
, &high_key
, fn
, priv
);
4944 * Calculate the number of blocks needed to store a given number of records
4945 * in a short-format (per-AG metadata) btree.
4948 xfs_btree_calc_size(
4949 struct xfs_mount
*mp
,
4951 unsigned long long len
)
4957 maxrecs
= limits
[0];
4958 for (level
= 0, rval
= 0; len
> 1; level
++) {
4960 do_div(len
, maxrecs
);
4961 maxrecs
= limits
[1];
4968 xfs_btree_count_blocks_helper(
4969 struct xfs_btree_cur
*cur
,
4973 xfs_extlen_t
*blocks
= data
;
4979 /* Count the blocks in a btree and return the result in *blocks. */
4981 xfs_btree_count_blocks(
4982 struct xfs_btree_cur
*cur
,
4983 xfs_extlen_t
*blocks
)
4986 return xfs_btree_visit_blocks(cur
, xfs_btree_count_blocks_helper
,
4990 /* Compare two btree pointers. */
4992 xfs_btree_diff_two_ptrs(
4993 struct xfs_btree_cur
*cur
,
4994 const union xfs_btree_ptr
*a
,
4995 const union xfs_btree_ptr
*b
)
4997 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
4998 return (int64_t)be64_to_cpu(a
->l
) - be64_to_cpu(b
->l
);
4999 return (int64_t)be32_to_cpu(a
->s
) - be32_to_cpu(b
->s
);