2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_defer.h"
27 #include "xfs_inode.h"
28 #include "xfs_trans.h"
29 #include "xfs_inode_item.h"
30 #include "xfs_buf_item.h"
31 #include "xfs_btree.h"
32 #include "xfs_error.h"
33 #include "xfs_trace.h"
34 #include "xfs_cksum.h"
35 #include "xfs_alloc.h"
39 * Cursor allocation zone.
41 kmem_zone_t
*xfs_btree_cur_zone
;
44 * Btree magic numbers.
46 static const __uint32_t xfs_magics
[2][XFS_BTNUM_MAX
] = {
47 { XFS_ABTB_MAGIC
, XFS_ABTC_MAGIC
, 0, XFS_BMAP_MAGIC
, XFS_IBT_MAGIC
,
49 { XFS_ABTB_CRC_MAGIC
, XFS_ABTC_CRC_MAGIC
, XFS_RMAP_CRC_MAGIC
,
50 XFS_BMAP_CRC_MAGIC
, XFS_IBT_CRC_MAGIC
, XFS_FIBT_CRC_MAGIC
,
53 #define xfs_btree_magic(cur) \
54 xfs_magics[!!((cur)->bc_flags & XFS_BTREE_CRC_BLOCKS)][cur->bc_btnum]
56 STATIC
int /* error (0 or EFSCORRUPTED) */
57 xfs_btree_check_lblock(
58 struct xfs_btree_cur
*cur
, /* btree cursor */
59 struct xfs_btree_block
*block
, /* btree long form block pointer */
60 int level
, /* level of the btree block */
61 struct xfs_buf
*bp
) /* buffer for block, if any */
63 int lblock_ok
= 1; /* block passes checks */
64 struct xfs_mount
*mp
; /* file system mount point */
68 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
69 lblock_ok
= lblock_ok
&&
70 uuid_equal(&block
->bb_u
.l
.bb_uuid
,
71 &mp
->m_sb
.sb_meta_uuid
) &&
72 block
->bb_u
.l
.bb_blkno
== cpu_to_be64(
73 bp
? bp
->b_bn
: XFS_BUF_DADDR_NULL
);
76 lblock_ok
= lblock_ok
&&
77 be32_to_cpu(block
->bb_magic
) == xfs_btree_magic(cur
) &&
78 be16_to_cpu(block
->bb_level
) == level
&&
79 be16_to_cpu(block
->bb_numrecs
) <=
80 cur
->bc_ops
->get_maxrecs(cur
, level
) &&
81 block
->bb_u
.l
.bb_leftsib
&&
82 (block
->bb_u
.l
.bb_leftsib
== cpu_to_be64(NULLFSBLOCK
) ||
83 XFS_FSB_SANITY_CHECK(mp
,
84 be64_to_cpu(block
->bb_u
.l
.bb_leftsib
))) &&
85 block
->bb_u
.l
.bb_rightsib
&&
86 (block
->bb_u
.l
.bb_rightsib
== cpu_to_be64(NULLFSBLOCK
) ||
87 XFS_FSB_SANITY_CHECK(mp
,
88 be64_to_cpu(block
->bb_u
.l
.bb_rightsib
)));
90 if (unlikely(XFS_TEST_ERROR(!lblock_ok
, mp
,
91 XFS_ERRTAG_BTREE_CHECK_LBLOCK
,
92 XFS_RANDOM_BTREE_CHECK_LBLOCK
))) {
94 trace_xfs_btree_corrupt(bp
, _RET_IP_
);
95 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_LOW
, mp
);
101 STATIC
int /* error (0 or EFSCORRUPTED) */
102 xfs_btree_check_sblock(
103 struct xfs_btree_cur
*cur
, /* btree cursor */
104 struct xfs_btree_block
*block
, /* btree short form block pointer */
105 int level
, /* level of the btree block */
106 struct xfs_buf
*bp
) /* buffer containing block */
108 struct xfs_mount
*mp
; /* file system mount point */
109 struct xfs_buf
*agbp
; /* buffer for ag. freespace struct */
110 struct xfs_agf
*agf
; /* ag. freespace structure */
111 xfs_agblock_t agflen
; /* native ag. freespace length */
112 int sblock_ok
= 1; /* block passes checks */
115 agbp
= cur
->bc_private
.a
.agbp
;
116 agf
= XFS_BUF_TO_AGF(agbp
);
117 agflen
= be32_to_cpu(agf
->agf_length
);
119 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
120 sblock_ok
= sblock_ok
&&
121 uuid_equal(&block
->bb_u
.s
.bb_uuid
,
122 &mp
->m_sb
.sb_meta_uuid
) &&
123 block
->bb_u
.s
.bb_blkno
== cpu_to_be64(
124 bp
? bp
->b_bn
: XFS_BUF_DADDR_NULL
);
127 sblock_ok
= sblock_ok
&&
128 be32_to_cpu(block
->bb_magic
) == xfs_btree_magic(cur
) &&
129 be16_to_cpu(block
->bb_level
) == level
&&
130 be16_to_cpu(block
->bb_numrecs
) <=
131 cur
->bc_ops
->get_maxrecs(cur
, level
) &&
132 (block
->bb_u
.s
.bb_leftsib
== cpu_to_be32(NULLAGBLOCK
) ||
133 be32_to_cpu(block
->bb_u
.s
.bb_leftsib
) < agflen
) &&
134 block
->bb_u
.s
.bb_leftsib
&&
135 (block
->bb_u
.s
.bb_rightsib
== cpu_to_be32(NULLAGBLOCK
) ||
136 be32_to_cpu(block
->bb_u
.s
.bb_rightsib
) < agflen
) &&
137 block
->bb_u
.s
.bb_rightsib
;
139 if (unlikely(XFS_TEST_ERROR(!sblock_ok
, mp
,
140 XFS_ERRTAG_BTREE_CHECK_SBLOCK
,
141 XFS_RANDOM_BTREE_CHECK_SBLOCK
))) {
143 trace_xfs_btree_corrupt(bp
, _RET_IP_
);
144 XFS_ERROR_REPORT(__func__
, XFS_ERRLEVEL_LOW
, mp
);
145 return -EFSCORRUPTED
;
151 * Debug routine: check that block header is ok.
154 xfs_btree_check_block(
155 struct xfs_btree_cur
*cur
, /* btree cursor */
156 struct xfs_btree_block
*block
, /* generic btree block pointer */
157 int level
, /* level of the btree block */
158 struct xfs_buf
*bp
) /* buffer containing block, if any */
160 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
161 return xfs_btree_check_lblock(cur
, block
, level
, bp
);
163 return xfs_btree_check_sblock(cur
, block
, level
, bp
);
167 * Check that (long) pointer is ok.
169 int /* error (0 or EFSCORRUPTED) */
170 xfs_btree_check_lptr(
171 struct xfs_btree_cur
*cur
, /* btree cursor */
172 xfs_fsblock_t bno
, /* btree block disk address */
173 int level
) /* btree block level */
175 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
,
177 bno
!= NULLFSBLOCK
&&
178 XFS_FSB_SANITY_CHECK(cur
->bc_mp
, bno
));
184 * Check that (short) pointer is ok.
186 STATIC
int /* error (0 or EFSCORRUPTED) */
187 xfs_btree_check_sptr(
188 struct xfs_btree_cur
*cur
, /* btree cursor */
189 xfs_agblock_t bno
, /* btree block disk address */
190 int level
) /* btree block level */
192 xfs_agblock_t agblocks
= cur
->bc_mp
->m_sb
.sb_agblocks
;
194 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
,
196 bno
!= NULLAGBLOCK
&&
203 * Check that block ptr is ok.
205 STATIC
int /* error (0 or EFSCORRUPTED) */
207 struct xfs_btree_cur
*cur
, /* btree cursor */
208 union xfs_btree_ptr
*ptr
, /* btree block disk address */
209 int index
, /* offset from ptr to check */
210 int level
) /* btree block level */
212 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
213 return xfs_btree_check_lptr(cur
,
214 be64_to_cpu((&ptr
->l
)[index
]), level
);
216 return xfs_btree_check_sptr(cur
,
217 be32_to_cpu((&ptr
->s
)[index
]), level
);
223 * Calculate CRC on the whole btree block and stuff it into the
224 * long-form btree header.
226 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
227 * it into the buffer so recovery knows what the last modification was that made
231 xfs_btree_lblock_calc_crc(
234 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
235 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
237 if (!xfs_sb_version_hascrc(&bp
->b_target
->bt_mount
->m_sb
))
240 block
->bb_u
.l
.bb_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
241 xfs_buf_update_cksum(bp
, XFS_BTREE_LBLOCK_CRC_OFF
);
245 xfs_btree_lblock_verify_crc(
248 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
249 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
251 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
252 if (!xfs_log_check_lsn(mp
, be64_to_cpu(block
->bb_u
.l
.bb_lsn
)))
254 return xfs_buf_verify_cksum(bp
, XFS_BTREE_LBLOCK_CRC_OFF
);
261 * Calculate CRC on the whole btree block and stuff it into the
262 * short-form btree header.
264 * Prior to calculting the CRC, pull the LSN out of the buffer log item and put
265 * it into the buffer so recovery knows what the last modification was that made
269 xfs_btree_sblock_calc_crc(
272 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
273 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
275 if (!xfs_sb_version_hascrc(&bp
->b_target
->bt_mount
->m_sb
))
278 block
->bb_u
.s
.bb_lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
279 xfs_buf_update_cksum(bp
, XFS_BTREE_SBLOCK_CRC_OFF
);
283 xfs_btree_sblock_verify_crc(
286 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
287 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
289 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
290 if (!xfs_log_check_lsn(mp
, be64_to_cpu(block
->bb_u
.s
.bb_lsn
)))
292 return xfs_buf_verify_cksum(bp
, XFS_BTREE_SBLOCK_CRC_OFF
);
299 xfs_btree_free_block(
300 struct xfs_btree_cur
*cur
,
305 error
= cur
->bc_ops
->free_block(cur
, bp
);
307 xfs_trans_binval(cur
->bc_tp
, bp
);
308 XFS_BTREE_STATS_INC(cur
, free
);
314 * Delete the btree cursor.
317 xfs_btree_del_cursor(
318 xfs_btree_cur_t
*cur
, /* btree cursor */
319 int error
) /* del because of error */
321 int i
; /* btree level */
324 * Clear the buffer pointers, and release the buffers.
325 * If we're doing this in the face of an error, we
326 * need to make sure to inspect all of the entries
327 * in the bc_bufs array for buffers to be unlocked.
328 * This is because some of the btree code works from
329 * level n down to 0, and if we get an error along
330 * the way we won't have initialized all the entries
333 for (i
= 0; i
< cur
->bc_nlevels
; i
++) {
335 xfs_trans_brelse(cur
->bc_tp
, cur
->bc_bufs
[i
]);
340 * Can't free a bmap cursor without having dealt with the
341 * allocated indirect blocks' accounting.
343 ASSERT(cur
->bc_btnum
!= XFS_BTNUM_BMAP
||
344 cur
->bc_private
.b
.allocated
== 0);
348 kmem_zone_free(xfs_btree_cur_zone
, cur
);
352 * Duplicate the btree cursor.
353 * Allocate a new one, copy the record, re-get the buffers.
356 xfs_btree_dup_cursor(
357 xfs_btree_cur_t
*cur
, /* input cursor */
358 xfs_btree_cur_t
**ncur
) /* output cursor */
360 xfs_buf_t
*bp
; /* btree block's buffer pointer */
361 int error
; /* error return value */
362 int i
; /* level number of btree block */
363 xfs_mount_t
*mp
; /* mount structure for filesystem */
364 xfs_btree_cur_t
*new; /* new cursor value */
365 xfs_trans_t
*tp
; /* transaction pointer, can be NULL */
371 * Allocate a new cursor like the old one.
373 new = cur
->bc_ops
->dup_cursor(cur
);
376 * Copy the record currently in the cursor.
378 new->bc_rec
= cur
->bc_rec
;
381 * For each level current, re-get the buffer and copy the ptr value.
383 for (i
= 0; i
< new->bc_nlevels
; i
++) {
384 new->bc_ptrs
[i
] = cur
->bc_ptrs
[i
];
385 new->bc_ra
[i
] = cur
->bc_ra
[i
];
386 bp
= cur
->bc_bufs
[i
];
388 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
,
389 XFS_BUF_ADDR(bp
), mp
->m_bsize
,
391 cur
->bc_ops
->buf_ops
);
393 xfs_btree_del_cursor(new, error
);
398 new->bc_bufs
[i
] = bp
;
405 * XFS btree block layout and addressing:
407 * There are two types of blocks in the btree: leaf and non-leaf blocks.
409 * The leaf record start with a header then followed by records containing
410 * the values. A non-leaf block also starts with the same header, and
411 * then first contains lookup keys followed by an equal number of pointers
412 * to the btree blocks at the previous level.
414 * +--------+-------+-------+-------+-------+-------+-------+
415 * Leaf: | header | rec 1 | rec 2 | rec 3 | rec 4 | rec 5 | rec N |
416 * +--------+-------+-------+-------+-------+-------+-------+
418 * +--------+-------+-------+-------+-------+-------+-------+
419 * Non-Leaf: | header | key 1 | key 2 | key N | ptr 1 | ptr 2 | ptr N |
420 * +--------+-------+-------+-------+-------+-------+-------+
422 * The header is called struct xfs_btree_block for reasons better left unknown
423 * and comes in different versions for short (32bit) and long (64bit) block
424 * pointers. The record and key structures are defined by the btree instances
425 * and opaque to the btree core. The block pointers are simple disk endian
426 * integers, available in a short (32bit) and long (64bit) variant.
428 * The helpers below calculate the offset of a given record, key or pointer
429 * into a btree block (xfs_btree_*_offset) or return a pointer to the given
430 * record, key or pointer (xfs_btree_*_addr). Note that all addressing
431 * inside the btree block is done using indices starting at one, not zero!
433 * If XFS_BTREE_OVERLAPPING is set, then this btree supports keys containing
434 * overlapping intervals. In such a tree, records are still sorted lowest to
435 * highest and indexed by the smallest key value that refers to the record.
436 * However, nodes are different: each pointer has two associated keys -- one
437 * indexing the lowest key available in the block(s) below (the same behavior
438 * as the key in a regular btree) and another indexing the highest key
439 * available in the block(s) below. Because records are /not/ sorted by the
440 * highest key, all leaf block updates require us to compute the highest key
441 * that matches any record in the leaf and to recursively update the high keys
442 * in the nodes going further up in the tree, if necessary. Nodes look like
445 * +--------+-----+-----+-----+-----+-----+-------+-------+-----+
446 * Non-Leaf: | header | lo1 | hi1 | lo2 | hi2 | ... | ptr 1 | ptr 2 | ... |
447 * +--------+-----+-----+-----+-----+-----+-------+-------+-----+
449 * To perform an interval query on an overlapped tree, perform the usual
450 * depth-first search and use the low and high keys to decide if we can skip
451 * that particular node. If a leaf node is reached, return the records that
452 * intersect the interval. Note that an interval query may return numerous
453 * entries. For a non-overlapped tree, simply search for the record associated
454 * with the lowest key and iterate forward until a non-matching record is
455 * found. Section 14.3 ("Interval Trees") of _Introduction to Algorithms_ by
456 * Cormen, Leiserson, Rivest, and Stein (2nd or 3rd ed. only) discuss this in
459 * Why do we care about overlapping intervals? Let's say you have a bunch of
460 * reverse mapping records on a reflink filesystem:
462 * 1: +- file A startblock B offset C length D -----------+
463 * 2: +- file E startblock F offset G length H --------------+
464 * 3: +- file I startblock F offset J length K --+
465 * 4: +- file L... --+
467 * Now say we want to map block (B+D) into file A at offset (C+D). Ideally,
468 * we'd simply increment the length of record 1. But how do we find the record
469 * that ends at (B+D-1) (i.e. record 1)? A LE lookup of (B+D-1) would return
470 * record 3 because the keys are ordered first by startblock. An interval
471 * query would return records 1 and 2 because they both overlap (B+D-1), and
472 * from that we can pick out record 1 as the appropriate left neighbor.
474 * In the non-overlapped case you can do a LE lookup and decrement the cursor
475 * because a record's interval must end before the next record.
479 * Return size of the btree block header for this btree instance.
481 static inline size_t xfs_btree_block_len(struct xfs_btree_cur
*cur
)
483 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
484 if (cur
->bc_flags
& XFS_BTREE_CRC_BLOCKS
)
485 return XFS_BTREE_LBLOCK_CRC_LEN
;
486 return XFS_BTREE_LBLOCK_LEN
;
488 if (cur
->bc_flags
& XFS_BTREE_CRC_BLOCKS
)
489 return XFS_BTREE_SBLOCK_CRC_LEN
;
490 return XFS_BTREE_SBLOCK_LEN
;
494 * Return size of btree block pointers for this btree instance.
496 static inline size_t xfs_btree_ptr_len(struct xfs_btree_cur
*cur
)
498 return (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) ?
499 sizeof(__be64
) : sizeof(__be32
);
503 * Calculate offset of the n-th record in a btree block.
506 xfs_btree_rec_offset(
507 struct xfs_btree_cur
*cur
,
510 return xfs_btree_block_len(cur
) +
511 (n
- 1) * cur
->bc_ops
->rec_len
;
515 * Calculate offset of the n-th key in a btree block.
518 xfs_btree_key_offset(
519 struct xfs_btree_cur
*cur
,
522 return xfs_btree_block_len(cur
) +
523 (n
- 1) * cur
->bc_ops
->key_len
;
527 * Calculate offset of the n-th high key in a btree block.
530 xfs_btree_high_key_offset(
531 struct xfs_btree_cur
*cur
,
534 return xfs_btree_block_len(cur
) +
535 (n
- 1) * cur
->bc_ops
->key_len
+ (cur
->bc_ops
->key_len
/ 2);
539 * Calculate offset of the n-th block pointer in a btree block.
542 xfs_btree_ptr_offset(
543 struct xfs_btree_cur
*cur
,
547 return xfs_btree_block_len(cur
) +
548 cur
->bc_ops
->get_maxrecs(cur
, level
) * cur
->bc_ops
->key_len
+
549 (n
- 1) * xfs_btree_ptr_len(cur
);
553 * Return a pointer to the n-th record in the btree block.
555 STATIC
union xfs_btree_rec
*
557 struct xfs_btree_cur
*cur
,
559 struct xfs_btree_block
*block
)
561 return (union xfs_btree_rec
*)
562 ((char *)block
+ xfs_btree_rec_offset(cur
, n
));
566 * Return a pointer to the n-th key in the btree block.
568 STATIC
union xfs_btree_key
*
570 struct xfs_btree_cur
*cur
,
572 struct xfs_btree_block
*block
)
574 return (union xfs_btree_key
*)
575 ((char *)block
+ xfs_btree_key_offset(cur
, n
));
579 * Return a pointer to the n-th high key in the btree block.
581 STATIC
union xfs_btree_key
*
582 xfs_btree_high_key_addr(
583 struct xfs_btree_cur
*cur
,
585 struct xfs_btree_block
*block
)
587 return (union xfs_btree_key
*)
588 ((char *)block
+ xfs_btree_high_key_offset(cur
, n
));
592 * Return a pointer to the n-th block pointer in the btree block.
594 STATIC
union xfs_btree_ptr
*
596 struct xfs_btree_cur
*cur
,
598 struct xfs_btree_block
*block
)
600 int level
= xfs_btree_get_level(block
);
602 ASSERT(block
->bb_level
!= 0);
604 return (union xfs_btree_ptr
*)
605 ((char *)block
+ xfs_btree_ptr_offset(cur
, n
, level
));
609 * Get the root block which is stored in the inode.
611 * For now this btree implementation assumes the btree root is always
612 * stored in the if_broot field of an inode fork.
614 STATIC
struct xfs_btree_block
*
616 struct xfs_btree_cur
*cur
)
618 struct xfs_ifork
*ifp
;
620 ifp
= XFS_IFORK_PTR(cur
->bc_private
.b
.ip
, cur
->bc_private
.b
.whichfork
);
621 return (struct xfs_btree_block
*)ifp
->if_broot
;
625 * Retrieve the block pointer from the cursor at the given level.
626 * This may be an inode btree root or from a buffer.
628 STATIC
struct xfs_btree_block
* /* generic btree block pointer */
630 struct xfs_btree_cur
*cur
, /* btree cursor */
631 int level
, /* level in btree */
632 struct xfs_buf
**bpp
) /* buffer containing the block */
634 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
635 (level
== cur
->bc_nlevels
- 1)) {
637 return xfs_btree_get_iroot(cur
);
640 *bpp
= cur
->bc_bufs
[level
];
641 return XFS_BUF_TO_BLOCK(*bpp
);
645 * Get a buffer for the block, return it with no data read.
646 * Long-form addressing.
648 xfs_buf_t
* /* buffer for fsbno */
650 xfs_mount_t
*mp
, /* file system mount point */
651 xfs_trans_t
*tp
, /* transaction pointer */
652 xfs_fsblock_t fsbno
, /* file system block number */
653 uint lock
) /* lock flags for get_buf */
655 xfs_daddr_t d
; /* real disk block address */
657 ASSERT(fsbno
!= NULLFSBLOCK
);
658 d
= XFS_FSB_TO_DADDR(mp
, fsbno
);
659 return xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
, mp
->m_bsize
, lock
);
663 * Get a buffer for the block, return it with no data read.
664 * Short-form addressing.
666 xfs_buf_t
* /* buffer for agno/agbno */
668 xfs_mount_t
*mp
, /* file system mount point */
669 xfs_trans_t
*tp
, /* transaction pointer */
670 xfs_agnumber_t agno
, /* allocation group number */
671 xfs_agblock_t agbno
, /* allocation group block number */
672 uint lock
) /* lock flags for get_buf */
674 xfs_daddr_t d
; /* real disk block address */
676 ASSERT(agno
!= NULLAGNUMBER
);
677 ASSERT(agbno
!= NULLAGBLOCK
);
678 d
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
);
679 return xfs_trans_get_buf(tp
, mp
->m_ddev_targp
, d
, mp
->m_bsize
, lock
);
683 * Check for the cursor referring to the last block at the given level.
685 int /* 1=is last block, 0=not last block */
686 xfs_btree_islastblock(
687 xfs_btree_cur_t
*cur
, /* btree cursor */
688 int level
) /* level to check */
690 struct xfs_btree_block
*block
; /* generic btree block pointer */
691 xfs_buf_t
*bp
; /* buffer containing block */
693 block
= xfs_btree_get_block(cur
, level
, &bp
);
694 xfs_btree_check_block(cur
, block
, level
, bp
);
695 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
696 return block
->bb_u
.l
.bb_rightsib
== cpu_to_be64(NULLFSBLOCK
);
698 return block
->bb_u
.s
.bb_rightsib
== cpu_to_be32(NULLAGBLOCK
);
702 * Change the cursor to point to the first record at the given level.
703 * Other levels are unaffected.
705 STATIC
int /* success=1, failure=0 */
707 xfs_btree_cur_t
*cur
, /* btree cursor */
708 int level
) /* level to change */
710 struct xfs_btree_block
*block
; /* generic btree block pointer */
711 xfs_buf_t
*bp
; /* buffer containing block */
714 * Get the block pointer for this level.
716 block
= xfs_btree_get_block(cur
, level
, &bp
);
717 xfs_btree_check_block(cur
, block
, level
, bp
);
719 * It's empty, there is no such record.
721 if (!block
->bb_numrecs
)
724 * Set the ptr value to 1, that's the first record/key.
726 cur
->bc_ptrs
[level
] = 1;
731 * Change the cursor to point to the last record in the current block
732 * at the given level. Other levels are unaffected.
734 STATIC
int /* success=1, failure=0 */
736 xfs_btree_cur_t
*cur
, /* btree cursor */
737 int level
) /* level to change */
739 struct xfs_btree_block
*block
; /* generic btree block pointer */
740 xfs_buf_t
*bp
; /* buffer containing block */
743 * Get the block pointer for this level.
745 block
= xfs_btree_get_block(cur
, level
, &bp
);
746 xfs_btree_check_block(cur
, block
, level
, bp
);
748 * It's empty, there is no such record.
750 if (!block
->bb_numrecs
)
753 * Set the ptr value to numrecs, that's the last record/key.
755 cur
->bc_ptrs
[level
] = be16_to_cpu(block
->bb_numrecs
);
760 * Compute first and last byte offsets for the fields given.
761 * Interprets the offsets table, which contains struct field offsets.
765 __int64_t fields
, /* bitmask of fields */
766 const short *offsets
, /* table of field offsets */
767 int nbits
, /* number of bits to inspect */
768 int *first
, /* output: first byte offset */
769 int *last
) /* output: last byte offset */
771 int i
; /* current bit number */
772 __int64_t imask
; /* mask for current bit number */
776 * Find the lowest bit, so the first byte offset.
778 for (i
= 0, imask
= 1LL; ; i
++, imask
<<= 1) {
779 if (imask
& fields
) {
785 * Find the highest bit, so the last byte offset.
787 for (i
= nbits
- 1, imask
= 1LL << i
; ; i
--, imask
>>= 1) {
788 if (imask
& fields
) {
789 *last
= offsets
[i
+ 1] - 1;
796 * Get a buffer for the block, return it read in.
797 * Long-form addressing.
801 struct xfs_mount
*mp
, /* file system mount point */
802 struct xfs_trans
*tp
, /* transaction pointer */
803 xfs_fsblock_t fsbno
, /* file system block number */
804 uint lock
, /* lock flags for read_buf */
805 struct xfs_buf
**bpp
, /* buffer for fsbno */
806 int refval
, /* ref count value for buffer */
807 const struct xfs_buf_ops
*ops
)
809 struct xfs_buf
*bp
; /* return value */
810 xfs_daddr_t d
; /* real disk block address */
813 ASSERT(fsbno
!= NULLFSBLOCK
);
814 d
= XFS_FSB_TO_DADDR(mp
, fsbno
);
815 error
= xfs_trans_read_buf(mp
, tp
, mp
->m_ddev_targp
, d
,
816 mp
->m_bsize
, lock
, &bp
, ops
);
820 xfs_buf_set_ref(bp
, refval
);
826 * Read-ahead the block, don't wait for it, don't return a buffer.
827 * Long-form addressing.
831 xfs_btree_reada_bufl(
832 struct xfs_mount
*mp
, /* file system mount point */
833 xfs_fsblock_t fsbno
, /* file system block number */
834 xfs_extlen_t count
, /* count of filesystem blocks */
835 const struct xfs_buf_ops
*ops
)
839 ASSERT(fsbno
!= NULLFSBLOCK
);
840 d
= XFS_FSB_TO_DADDR(mp
, fsbno
);
841 xfs_buf_readahead(mp
->m_ddev_targp
, d
, mp
->m_bsize
* count
, ops
);
845 * Read-ahead the block, don't wait for it, don't return a buffer.
846 * Short-form addressing.
850 xfs_btree_reada_bufs(
851 struct xfs_mount
*mp
, /* file system mount point */
852 xfs_agnumber_t agno
, /* allocation group number */
853 xfs_agblock_t agbno
, /* allocation group block number */
854 xfs_extlen_t count
, /* count of filesystem blocks */
855 const struct xfs_buf_ops
*ops
)
859 ASSERT(agno
!= NULLAGNUMBER
);
860 ASSERT(agbno
!= NULLAGBLOCK
);
861 d
= XFS_AGB_TO_DADDR(mp
, agno
, agbno
);
862 xfs_buf_readahead(mp
->m_ddev_targp
, d
, mp
->m_bsize
* count
, ops
);
866 xfs_btree_readahead_lblock(
867 struct xfs_btree_cur
*cur
,
869 struct xfs_btree_block
*block
)
872 xfs_fsblock_t left
= be64_to_cpu(block
->bb_u
.l
.bb_leftsib
);
873 xfs_fsblock_t right
= be64_to_cpu(block
->bb_u
.l
.bb_rightsib
);
875 if ((lr
& XFS_BTCUR_LEFTRA
) && left
!= NULLFSBLOCK
) {
876 xfs_btree_reada_bufl(cur
->bc_mp
, left
, 1,
877 cur
->bc_ops
->buf_ops
);
881 if ((lr
& XFS_BTCUR_RIGHTRA
) && right
!= NULLFSBLOCK
) {
882 xfs_btree_reada_bufl(cur
->bc_mp
, right
, 1,
883 cur
->bc_ops
->buf_ops
);
891 xfs_btree_readahead_sblock(
892 struct xfs_btree_cur
*cur
,
894 struct xfs_btree_block
*block
)
897 xfs_agblock_t left
= be32_to_cpu(block
->bb_u
.s
.bb_leftsib
);
898 xfs_agblock_t right
= be32_to_cpu(block
->bb_u
.s
.bb_rightsib
);
901 if ((lr
& XFS_BTCUR_LEFTRA
) && left
!= NULLAGBLOCK
) {
902 xfs_btree_reada_bufs(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
903 left
, 1, cur
->bc_ops
->buf_ops
);
907 if ((lr
& XFS_BTCUR_RIGHTRA
) && right
!= NULLAGBLOCK
) {
908 xfs_btree_reada_bufs(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
909 right
, 1, cur
->bc_ops
->buf_ops
);
917 * Read-ahead btree blocks, at the given level.
918 * Bits in lr are set from XFS_BTCUR_{LEFT,RIGHT}RA.
922 struct xfs_btree_cur
*cur
, /* btree cursor */
923 int lev
, /* level in btree */
924 int lr
) /* left/right bits */
926 struct xfs_btree_block
*block
;
929 * No readahead needed if we are at the root level and the
930 * btree root is stored in the inode.
932 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
933 (lev
== cur
->bc_nlevels
- 1))
936 if ((cur
->bc_ra
[lev
] | lr
) == cur
->bc_ra
[lev
])
939 cur
->bc_ra
[lev
] |= lr
;
940 block
= XFS_BUF_TO_BLOCK(cur
->bc_bufs
[lev
]);
942 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
943 return xfs_btree_readahead_lblock(cur
, lr
, block
);
944 return xfs_btree_readahead_sblock(cur
, lr
, block
);
948 xfs_btree_ptr_to_daddr(
949 struct xfs_btree_cur
*cur
,
950 union xfs_btree_ptr
*ptr
)
952 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
953 ASSERT(ptr
->l
!= cpu_to_be64(NULLFSBLOCK
));
955 return XFS_FSB_TO_DADDR(cur
->bc_mp
, be64_to_cpu(ptr
->l
));
957 ASSERT(cur
->bc_private
.a
.agno
!= NULLAGNUMBER
);
958 ASSERT(ptr
->s
!= cpu_to_be32(NULLAGBLOCK
));
960 return XFS_AGB_TO_DADDR(cur
->bc_mp
, cur
->bc_private
.a
.agno
,
961 be32_to_cpu(ptr
->s
));
966 * Readahead @count btree blocks at the given @ptr location.
968 * We don't need to care about long or short form btrees here as we have a
969 * method of converting the ptr directly to a daddr available to us.
972 xfs_btree_readahead_ptr(
973 struct xfs_btree_cur
*cur
,
974 union xfs_btree_ptr
*ptr
,
977 xfs_buf_readahead(cur
->bc_mp
->m_ddev_targp
,
978 xfs_btree_ptr_to_daddr(cur
, ptr
),
979 cur
->bc_mp
->m_bsize
* count
, cur
->bc_ops
->buf_ops
);
983 * Set the buffer for level "lev" in the cursor to bp, releasing
984 * any previous buffer.
988 xfs_btree_cur_t
*cur
, /* btree cursor */
989 int lev
, /* level in btree */
990 xfs_buf_t
*bp
) /* new buffer to set */
992 struct xfs_btree_block
*b
; /* btree block */
994 if (cur
->bc_bufs
[lev
])
995 xfs_trans_brelse(cur
->bc_tp
, cur
->bc_bufs
[lev
]);
996 cur
->bc_bufs
[lev
] = bp
;
999 b
= XFS_BUF_TO_BLOCK(bp
);
1000 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
1001 if (b
->bb_u
.l
.bb_leftsib
== cpu_to_be64(NULLFSBLOCK
))
1002 cur
->bc_ra
[lev
] |= XFS_BTCUR_LEFTRA
;
1003 if (b
->bb_u
.l
.bb_rightsib
== cpu_to_be64(NULLFSBLOCK
))
1004 cur
->bc_ra
[lev
] |= XFS_BTCUR_RIGHTRA
;
1006 if (b
->bb_u
.s
.bb_leftsib
== cpu_to_be32(NULLAGBLOCK
))
1007 cur
->bc_ra
[lev
] |= XFS_BTCUR_LEFTRA
;
1008 if (b
->bb_u
.s
.bb_rightsib
== cpu_to_be32(NULLAGBLOCK
))
1009 cur
->bc_ra
[lev
] |= XFS_BTCUR_RIGHTRA
;
1014 xfs_btree_ptr_is_null(
1015 struct xfs_btree_cur
*cur
,
1016 union xfs_btree_ptr
*ptr
)
1018 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
1019 return ptr
->l
== cpu_to_be64(NULLFSBLOCK
);
1021 return ptr
->s
== cpu_to_be32(NULLAGBLOCK
);
1025 xfs_btree_set_ptr_null(
1026 struct xfs_btree_cur
*cur
,
1027 union xfs_btree_ptr
*ptr
)
1029 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
1030 ptr
->l
= cpu_to_be64(NULLFSBLOCK
);
1032 ptr
->s
= cpu_to_be32(NULLAGBLOCK
);
1036 * Get/set/init sibling pointers
1039 xfs_btree_get_sibling(
1040 struct xfs_btree_cur
*cur
,
1041 struct xfs_btree_block
*block
,
1042 union xfs_btree_ptr
*ptr
,
1045 ASSERT(lr
== XFS_BB_LEFTSIB
|| lr
== XFS_BB_RIGHTSIB
);
1047 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
1048 if (lr
== XFS_BB_RIGHTSIB
)
1049 ptr
->l
= block
->bb_u
.l
.bb_rightsib
;
1051 ptr
->l
= block
->bb_u
.l
.bb_leftsib
;
1053 if (lr
== XFS_BB_RIGHTSIB
)
1054 ptr
->s
= block
->bb_u
.s
.bb_rightsib
;
1056 ptr
->s
= block
->bb_u
.s
.bb_leftsib
;
1061 xfs_btree_set_sibling(
1062 struct xfs_btree_cur
*cur
,
1063 struct xfs_btree_block
*block
,
1064 union xfs_btree_ptr
*ptr
,
1067 ASSERT(lr
== XFS_BB_LEFTSIB
|| lr
== XFS_BB_RIGHTSIB
);
1069 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) {
1070 if (lr
== XFS_BB_RIGHTSIB
)
1071 block
->bb_u
.l
.bb_rightsib
= ptr
->l
;
1073 block
->bb_u
.l
.bb_leftsib
= ptr
->l
;
1075 if (lr
== XFS_BB_RIGHTSIB
)
1076 block
->bb_u
.s
.bb_rightsib
= ptr
->s
;
1078 block
->bb_u
.s
.bb_leftsib
= ptr
->s
;
1083 xfs_btree_init_block_int(
1084 struct xfs_mount
*mp
,
1085 struct xfs_btree_block
*buf
,
1093 buf
->bb_magic
= cpu_to_be32(magic
);
1094 buf
->bb_level
= cpu_to_be16(level
);
1095 buf
->bb_numrecs
= cpu_to_be16(numrecs
);
1097 if (flags
& XFS_BTREE_LONG_PTRS
) {
1098 buf
->bb_u
.l
.bb_leftsib
= cpu_to_be64(NULLFSBLOCK
);
1099 buf
->bb_u
.l
.bb_rightsib
= cpu_to_be64(NULLFSBLOCK
);
1100 if (flags
& XFS_BTREE_CRC_BLOCKS
) {
1101 buf
->bb_u
.l
.bb_blkno
= cpu_to_be64(blkno
);
1102 buf
->bb_u
.l
.bb_owner
= cpu_to_be64(owner
);
1103 uuid_copy(&buf
->bb_u
.l
.bb_uuid
, &mp
->m_sb
.sb_meta_uuid
);
1104 buf
->bb_u
.l
.bb_pad
= 0;
1105 buf
->bb_u
.l
.bb_lsn
= 0;
1108 /* owner is a 32 bit value on short blocks */
1109 __u32 __owner
= (__u32
)owner
;
1111 buf
->bb_u
.s
.bb_leftsib
= cpu_to_be32(NULLAGBLOCK
);
1112 buf
->bb_u
.s
.bb_rightsib
= cpu_to_be32(NULLAGBLOCK
);
1113 if (flags
& XFS_BTREE_CRC_BLOCKS
) {
1114 buf
->bb_u
.s
.bb_blkno
= cpu_to_be64(blkno
);
1115 buf
->bb_u
.s
.bb_owner
= cpu_to_be32(__owner
);
1116 uuid_copy(&buf
->bb_u
.s
.bb_uuid
, &mp
->m_sb
.sb_meta_uuid
);
1117 buf
->bb_u
.s
.bb_lsn
= 0;
1123 xfs_btree_init_block(
1124 struct xfs_mount
*mp
,
1132 xfs_btree_init_block_int(mp
, XFS_BUF_TO_BLOCK(bp
), bp
->b_bn
,
1133 magic
, level
, numrecs
, owner
, flags
);
1137 xfs_btree_init_block_cur(
1138 struct xfs_btree_cur
*cur
,
1146 * we can pull the owner from the cursor right now as the different
1147 * owners align directly with the pointer size of the btree. This may
1148 * change in future, but is safe for current users of the generic btree
1151 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
1152 owner
= cur
->bc_private
.b
.ip
->i_ino
;
1154 owner
= cur
->bc_private
.a
.agno
;
1156 xfs_btree_init_block_int(cur
->bc_mp
, XFS_BUF_TO_BLOCK(bp
), bp
->b_bn
,
1157 xfs_btree_magic(cur
), level
, numrecs
,
1158 owner
, cur
->bc_flags
);
1162 * Return true if ptr is the last record in the btree and
1163 * we need to track updates to this record. The decision
1164 * will be further refined in the update_lastrec method.
1167 xfs_btree_is_lastrec(
1168 struct xfs_btree_cur
*cur
,
1169 struct xfs_btree_block
*block
,
1172 union xfs_btree_ptr ptr
;
1176 if (!(cur
->bc_flags
& XFS_BTREE_LASTREC_UPDATE
))
1179 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_RIGHTSIB
);
1180 if (!xfs_btree_ptr_is_null(cur
, &ptr
))
1186 xfs_btree_buf_to_ptr(
1187 struct xfs_btree_cur
*cur
,
1189 union xfs_btree_ptr
*ptr
)
1191 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
1192 ptr
->l
= cpu_to_be64(XFS_DADDR_TO_FSB(cur
->bc_mp
,
1195 ptr
->s
= cpu_to_be32(xfs_daddr_to_agbno(cur
->bc_mp
,
1202 struct xfs_btree_cur
*cur
,
1205 switch (cur
->bc_btnum
) {
1208 xfs_buf_set_ref(bp
, XFS_ALLOC_BTREE_REF
);
1211 case XFS_BTNUM_FINO
:
1212 xfs_buf_set_ref(bp
, XFS_INO_BTREE_REF
);
1214 case XFS_BTNUM_BMAP
:
1215 xfs_buf_set_ref(bp
, XFS_BMAP_BTREE_REF
);
1217 case XFS_BTNUM_RMAP
:
1218 xfs_buf_set_ref(bp
, XFS_RMAP_BTREE_REF
);
1220 case XFS_BTNUM_REFC
:
1221 xfs_buf_set_ref(bp
, XFS_REFC_BTREE_REF
);
1229 xfs_btree_get_buf_block(
1230 struct xfs_btree_cur
*cur
,
1231 union xfs_btree_ptr
*ptr
,
1233 struct xfs_btree_block
**block
,
1234 struct xfs_buf
**bpp
)
1236 struct xfs_mount
*mp
= cur
->bc_mp
;
1239 /* need to sort out how callers deal with failures first */
1240 ASSERT(!(flags
& XBF_TRYLOCK
));
1242 d
= xfs_btree_ptr_to_daddr(cur
, ptr
);
1243 *bpp
= xfs_trans_get_buf(cur
->bc_tp
, mp
->m_ddev_targp
, d
,
1244 mp
->m_bsize
, flags
);
1249 (*bpp
)->b_ops
= cur
->bc_ops
->buf_ops
;
1250 *block
= XFS_BUF_TO_BLOCK(*bpp
);
1255 * Read in the buffer at the given ptr and return the buffer and
1256 * the block pointer within the buffer.
1259 xfs_btree_read_buf_block(
1260 struct xfs_btree_cur
*cur
,
1261 union xfs_btree_ptr
*ptr
,
1263 struct xfs_btree_block
**block
,
1264 struct xfs_buf
**bpp
)
1266 struct xfs_mount
*mp
= cur
->bc_mp
;
1270 /* need to sort out how callers deal with failures first */
1271 ASSERT(!(flags
& XBF_TRYLOCK
));
1273 d
= xfs_btree_ptr_to_daddr(cur
, ptr
);
1274 error
= xfs_trans_read_buf(mp
, cur
->bc_tp
, mp
->m_ddev_targp
, d
,
1275 mp
->m_bsize
, flags
, bpp
,
1276 cur
->bc_ops
->buf_ops
);
1280 xfs_btree_set_refs(cur
, *bpp
);
1281 *block
= XFS_BUF_TO_BLOCK(*bpp
);
1286 * Copy keys from one btree block to another.
1289 xfs_btree_copy_keys(
1290 struct xfs_btree_cur
*cur
,
1291 union xfs_btree_key
*dst_key
,
1292 union xfs_btree_key
*src_key
,
1295 ASSERT(numkeys
>= 0);
1296 memcpy(dst_key
, src_key
, numkeys
* cur
->bc_ops
->key_len
);
1300 * Copy records from one btree block to another.
1303 xfs_btree_copy_recs(
1304 struct xfs_btree_cur
*cur
,
1305 union xfs_btree_rec
*dst_rec
,
1306 union xfs_btree_rec
*src_rec
,
1309 ASSERT(numrecs
>= 0);
1310 memcpy(dst_rec
, src_rec
, numrecs
* cur
->bc_ops
->rec_len
);
1314 * Copy block pointers from one btree block to another.
1317 xfs_btree_copy_ptrs(
1318 struct xfs_btree_cur
*cur
,
1319 union xfs_btree_ptr
*dst_ptr
,
1320 union xfs_btree_ptr
*src_ptr
,
1323 ASSERT(numptrs
>= 0);
1324 memcpy(dst_ptr
, src_ptr
, numptrs
* xfs_btree_ptr_len(cur
));
1328 * Shift keys one index left/right inside a single btree block.
1331 xfs_btree_shift_keys(
1332 struct xfs_btree_cur
*cur
,
1333 union xfs_btree_key
*key
,
1339 ASSERT(numkeys
>= 0);
1340 ASSERT(dir
== 1 || dir
== -1);
1342 dst_key
= (char *)key
+ (dir
* cur
->bc_ops
->key_len
);
1343 memmove(dst_key
, key
, numkeys
* cur
->bc_ops
->key_len
);
1347 * Shift records one index left/right inside a single btree block.
1350 xfs_btree_shift_recs(
1351 struct xfs_btree_cur
*cur
,
1352 union xfs_btree_rec
*rec
,
1358 ASSERT(numrecs
>= 0);
1359 ASSERT(dir
== 1 || dir
== -1);
1361 dst_rec
= (char *)rec
+ (dir
* cur
->bc_ops
->rec_len
);
1362 memmove(dst_rec
, rec
, numrecs
* cur
->bc_ops
->rec_len
);
1366 * Shift block pointers one index left/right inside a single btree block.
1369 xfs_btree_shift_ptrs(
1370 struct xfs_btree_cur
*cur
,
1371 union xfs_btree_ptr
*ptr
,
1377 ASSERT(numptrs
>= 0);
1378 ASSERT(dir
== 1 || dir
== -1);
1380 dst_ptr
= (char *)ptr
+ (dir
* xfs_btree_ptr_len(cur
));
1381 memmove(dst_ptr
, ptr
, numptrs
* xfs_btree_ptr_len(cur
));
1385 * Log key values from the btree block.
1389 struct xfs_btree_cur
*cur
,
1394 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1395 XFS_BTREE_TRACE_ARGBII(cur
, bp
, first
, last
);
1398 xfs_trans_buf_set_type(cur
->bc_tp
, bp
, XFS_BLFT_BTREE_BUF
);
1399 xfs_trans_log_buf(cur
->bc_tp
, bp
,
1400 xfs_btree_key_offset(cur
, first
),
1401 xfs_btree_key_offset(cur
, last
+ 1) - 1);
1403 xfs_trans_log_inode(cur
->bc_tp
, cur
->bc_private
.b
.ip
,
1404 xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
));
1407 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1411 * Log record values from the btree block.
1415 struct xfs_btree_cur
*cur
,
1420 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1421 XFS_BTREE_TRACE_ARGBII(cur
, bp
, first
, last
);
1423 xfs_trans_buf_set_type(cur
->bc_tp
, bp
, XFS_BLFT_BTREE_BUF
);
1424 xfs_trans_log_buf(cur
->bc_tp
, bp
,
1425 xfs_btree_rec_offset(cur
, first
),
1426 xfs_btree_rec_offset(cur
, last
+ 1) - 1);
1428 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1432 * Log block pointer fields from a btree block (nonleaf).
1436 struct xfs_btree_cur
*cur
, /* btree cursor */
1437 struct xfs_buf
*bp
, /* buffer containing btree block */
1438 int first
, /* index of first pointer to log */
1439 int last
) /* index of last pointer to log */
1441 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1442 XFS_BTREE_TRACE_ARGBII(cur
, bp
, first
, last
);
1445 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
1446 int level
= xfs_btree_get_level(block
);
1448 xfs_trans_buf_set_type(cur
->bc_tp
, bp
, XFS_BLFT_BTREE_BUF
);
1449 xfs_trans_log_buf(cur
->bc_tp
, bp
,
1450 xfs_btree_ptr_offset(cur
, first
, level
),
1451 xfs_btree_ptr_offset(cur
, last
+ 1, level
) - 1);
1453 xfs_trans_log_inode(cur
->bc_tp
, cur
->bc_private
.b
.ip
,
1454 xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
));
1457 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1461 * Log fields from a btree block header.
1464 xfs_btree_log_block(
1465 struct xfs_btree_cur
*cur
, /* btree cursor */
1466 struct xfs_buf
*bp
, /* buffer containing btree block */
1467 int fields
) /* mask of fields: XFS_BB_... */
1469 int first
; /* first byte offset logged */
1470 int last
; /* last byte offset logged */
1471 static const short soffsets
[] = { /* table of offsets (short) */
1472 offsetof(struct xfs_btree_block
, bb_magic
),
1473 offsetof(struct xfs_btree_block
, bb_level
),
1474 offsetof(struct xfs_btree_block
, bb_numrecs
),
1475 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_leftsib
),
1476 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_rightsib
),
1477 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_blkno
),
1478 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_lsn
),
1479 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_uuid
),
1480 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_owner
),
1481 offsetof(struct xfs_btree_block
, bb_u
.s
.bb_crc
),
1482 XFS_BTREE_SBLOCK_CRC_LEN
1484 static const short loffsets
[] = { /* table of offsets (long) */
1485 offsetof(struct xfs_btree_block
, bb_magic
),
1486 offsetof(struct xfs_btree_block
, bb_level
),
1487 offsetof(struct xfs_btree_block
, bb_numrecs
),
1488 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_leftsib
),
1489 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_rightsib
),
1490 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_blkno
),
1491 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_lsn
),
1492 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_uuid
),
1493 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_owner
),
1494 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_crc
),
1495 offsetof(struct xfs_btree_block
, bb_u
.l
.bb_pad
),
1496 XFS_BTREE_LBLOCK_CRC_LEN
1499 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1500 XFS_BTREE_TRACE_ARGBI(cur
, bp
, fields
);
1505 if (cur
->bc_flags
& XFS_BTREE_CRC_BLOCKS
) {
1507 * We don't log the CRC when updating a btree
1508 * block but instead recreate it during log
1509 * recovery. As the log buffers have checksums
1510 * of their own this is safe and avoids logging a crc
1511 * update in a lot of places.
1513 if (fields
== XFS_BB_ALL_BITS
)
1514 fields
= XFS_BB_ALL_BITS_CRC
;
1515 nbits
= XFS_BB_NUM_BITS_CRC
;
1517 nbits
= XFS_BB_NUM_BITS
;
1519 xfs_btree_offsets(fields
,
1520 (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
) ?
1521 loffsets
: soffsets
,
1522 nbits
, &first
, &last
);
1523 xfs_trans_buf_set_type(cur
->bc_tp
, bp
, XFS_BLFT_BTREE_BUF
);
1524 xfs_trans_log_buf(cur
->bc_tp
, bp
, first
, last
);
1526 xfs_trans_log_inode(cur
->bc_tp
, cur
->bc_private
.b
.ip
,
1527 xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
));
1530 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1534 * Increment cursor by one record at the level.
1535 * For nonzero levels the leaf-ward information is untouched.
1538 xfs_btree_increment(
1539 struct xfs_btree_cur
*cur
,
1541 int *stat
) /* success/failure */
1543 struct xfs_btree_block
*block
;
1544 union xfs_btree_ptr ptr
;
1546 int error
; /* error return value */
1549 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1550 XFS_BTREE_TRACE_ARGI(cur
, level
);
1552 ASSERT(level
< cur
->bc_nlevels
);
1554 /* Read-ahead to the right at this level. */
1555 xfs_btree_readahead(cur
, level
, XFS_BTCUR_RIGHTRA
);
1557 /* Get a pointer to the btree block. */
1558 block
= xfs_btree_get_block(cur
, level
, &bp
);
1561 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
1566 /* We're done if we remain in the block after the increment. */
1567 if (++cur
->bc_ptrs
[level
] <= xfs_btree_get_numrecs(block
))
1570 /* Fail if we just went off the right edge of the tree. */
1571 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_RIGHTSIB
);
1572 if (xfs_btree_ptr_is_null(cur
, &ptr
))
1575 XFS_BTREE_STATS_INC(cur
, increment
);
1578 * March up the tree incrementing pointers.
1579 * Stop when we don't go off the right edge of a block.
1581 for (lev
= level
+ 1; lev
< cur
->bc_nlevels
; lev
++) {
1582 block
= xfs_btree_get_block(cur
, lev
, &bp
);
1585 error
= xfs_btree_check_block(cur
, block
, lev
, bp
);
1590 if (++cur
->bc_ptrs
[lev
] <= xfs_btree_get_numrecs(block
))
1593 /* Read-ahead the right block for the next loop. */
1594 xfs_btree_readahead(cur
, lev
, XFS_BTCUR_RIGHTRA
);
1598 * If we went off the root then we are either seriously
1599 * confused or have the tree root in an inode.
1601 if (lev
== cur
->bc_nlevels
) {
1602 if (cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
)
1605 error
= -EFSCORRUPTED
;
1608 ASSERT(lev
< cur
->bc_nlevels
);
1611 * Now walk back down the tree, fixing up the cursor's buffer
1612 * pointers and key numbers.
1614 for (block
= xfs_btree_get_block(cur
, lev
, &bp
); lev
> level
; ) {
1615 union xfs_btree_ptr
*ptrp
;
1617 ptrp
= xfs_btree_ptr_addr(cur
, cur
->bc_ptrs
[lev
], block
);
1619 error
= xfs_btree_read_buf_block(cur
, ptrp
, 0, &block
, &bp
);
1623 xfs_btree_setbuf(cur
, lev
, bp
);
1624 cur
->bc_ptrs
[lev
] = 1;
1627 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1632 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1637 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
1642 * Decrement cursor by one record at the level.
1643 * For nonzero levels the leaf-ward information is untouched.
1646 xfs_btree_decrement(
1647 struct xfs_btree_cur
*cur
,
1649 int *stat
) /* success/failure */
1651 struct xfs_btree_block
*block
;
1653 int error
; /* error return value */
1655 union xfs_btree_ptr ptr
;
1657 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1658 XFS_BTREE_TRACE_ARGI(cur
, level
);
1660 ASSERT(level
< cur
->bc_nlevels
);
1662 /* Read-ahead to the left at this level. */
1663 xfs_btree_readahead(cur
, level
, XFS_BTCUR_LEFTRA
);
1665 /* We're done if we remain in the block after the decrement. */
1666 if (--cur
->bc_ptrs
[level
] > 0)
1669 /* Get a pointer to the btree block. */
1670 block
= xfs_btree_get_block(cur
, level
, &bp
);
1673 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
1678 /* Fail if we just went off the left edge of the tree. */
1679 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_LEFTSIB
);
1680 if (xfs_btree_ptr_is_null(cur
, &ptr
))
1683 XFS_BTREE_STATS_INC(cur
, decrement
);
1686 * March up the tree decrementing pointers.
1687 * Stop when we don't go off the left edge of a block.
1689 for (lev
= level
+ 1; lev
< cur
->bc_nlevels
; lev
++) {
1690 if (--cur
->bc_ptrs
[lev
] > 0)
1692 /* Read-ahead the left block for the next loop. */
1693 xfs_btree_readahead(cur
, lev
, XFS_BTCUR_LEFTRA
);
1697 * If we went off the root then we are seriously confused.
1698 * or the root of the tree is in an inode.
1700 if (lev
== cur
->bc_nlevels
) {
1701 if (cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
)
1704 error
= -EFSCORRUPTED
;
1707 ASSERT(lev
< cur
->bc_nlevels
);
1710 * Now walk back down the tree, fixing up the cursor's buffer
1711 * pointers and key numbers.
1713 for (block
= xfs_btree_get_block(cur
, lev
, &bp
); lev
> level
; ) {
1714 union xfs_btree_ptr
*ptrp
;
1716 ptrp
= xfs_btree_ptr_addr(cur
, cur
->bc_ptrs
[lev
], block
);
1718 error
= xfs_btree_read_buf_block(cur
, ptrp
, 0, &block
, &bp
);
1721 xfs_btree_setbuf(cur
, lev
, bp
);
1722 cur
->bc_ptrs
[lev
] = xfs_btree_get_numrecs(block
);
1725 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1730 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1735 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
1740 xfs_btree_lookup_get_block(
1741 struct xfs_btree_cur
*cur
, /* btree cursor */
1742 int level
, /* level in the btree */
1743 union xfs_btree_ptr
*pp
, /* ptr to btree block */
1744 struct xfs_btree_block
**blkp
) /* return btree block */
1746 struct xfs_buf
*bp
; /* buffer pointer for btree block */
1749 /* special case the root block if in an inode */
1750 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
1751 (level
== cur
->bc_nlevels
- 1)) {
1752 *blkp
= xfs_btree_get_iroot(cur
);
1757 * If the old buffer at this level for the disk address we are
1758 * looking for re-use it.
1760 * Otherwise throw it away and get a new one.
1762 bp
= cur
->bc_bufs
[level
];
1763 if (bp
&& XFS_BUF_ADDR(bp
) == xfs_btree_ptr_to_daddr(cur
, pp
)) {
1764 *blkp
= XFS_BUF_TO_BLOCK(bp
);
1768 error
= xfs_btree_read_buf_block(cur
, pp
, 0, blkp
, &bp
);
1772 xfs_btree_setbuf(cur
, level
, bp
);
1777 * Get current search key. For level 0 we don't actually have a key
1778 * structure so we make one up from the record. For all other levels
1779 * we just return the right key.
1781 STATIC
union xfs_btree_key
*
1782 xfs_lookup_get_search_key(
1783 struct xfs_btree_cur
*cur
,
1786 struct xfs_btree_block
*block
,
1787 union xfs_btree_key
*kp
)
1790 cur
->bc_ops
->init_key_from_rec(kp
,
1791 xfs_btree_rec_addr(cur
, keyno
, block
));
1795 return xfs_btree_key_addr(cur
, keyno
, block
);
1799 * Lookup the record. The cursor is made to point to it, based on dir.
1800 * stat is set to 0 if can't find any such record, 1 for success.
1804 struct xfs_btree_cur
*cur
, /* btree cursor */
1805 xfs_lookup_t dir
, /* <=, ==, or >= */
1806 int *stat
) /* success/failure */
1808 struct xfs_btree_block
*block
; /* current btree block */
1809 __int64_t diff
; /* difference for the current key */
1810 int error
; /* error return value */
1811 int keyno
; /* current key number */
1812 int level
; /* level in the btree */
1813 union xfs_btree_ptr
*pp
; /* ptr to btree block */
1814 union xfs_btree_ptr ptr
; /* ptr to btree block */
1816 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
1817 XFS_BTREE_TRACE_ARGI(cur
, dir
);
1819 XFS_BTREE_STATS_INC(cur
, lookup
);
1821 /* No such thing as a zero-level tree. */
1822 if (cur
->bc_nlevels
== 0)
1823 return -EFSCORRUPTED
;
1828 /* initialise start pointer from cursor */
1829 cur
->bc_ops
->init_ptr_from_cur(cur
, &ptr
);
1833 * Iterate over each level in the btree, starting at the root.
1834 * For each level above the leaves, find the key we need, based
1835 * on the lookup record, then follow the corresponding block
1836 * pointer down to the next level.
1838 for (level
= cur
->bc_nlevels
- 1, diff
= 1; level
>= 0; level
--) {
1839 /* Get the block we need to do the lookup on. */
1840 error
= xfs_btree_lookup_get_block(cur
, level
, pp
, &block
);
1846 * If we already had a key match at a higher level, we
1847 * know we need to use the first entry in this block.
1851 /* Otherwise search this block. Do a binary search. */
1853 int high
; /* high entry number */
1854 int low
; /* low entry number */
1856 /* Set low and high entry numbers, 1-based. */
1858 high
= xfs_btree_get_numrecs(block
);
1860 /* Block is empty, must be an empty leaf. */
1861 ASSERT(level
== 0 && cur
->bc_nlevels
== 1);
1863 cur
->bc_ptrs
[0] = dir
!= XFS_LOOKUP_LE
;
1864 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1869 /* Binary search the block. */
1870 while (low
<= high
) {
1871 union xfs_btree_key key
;
1872 union xfs_btree_key
*kp
;
1874 XFS_BTREE_STATS_INC(cur
, compare
);
1876 /* keyno is average of low and high. */
1877 keyno
= (low
+ high
) >> 1;
1879 /* Get current search key */
1880 kp
= xfs_lookup_get_search_key(cur
, level
,
1881 keyno
, block
, &key
);
1884 * Compute difference to get next direction:
1885 * - less than, move right
1886 * - greater than, move left
1887 * - equal, we're done
1889 diff
= cur
->bc_ops
->key_diff(cur
, kp
);
1900 * If there are more levels, set up for the next level
1901 * by getting the block number and filling in the cursor.
1905 * If we moved left, need the previous key number,
1906 * unless there isn't one.
1908 if (diff
> 0 && --keyno
< 1)
1910 pp
= xfs_btree_ptr_addr(cur
, keyno
, block
);
1913 error
= xfs_btree_check_ptr(cur
, pp
, 0, level
);
1917 cur
->bc_ptrs
[level
] = keyno
;
1921 /* Done with the search. See if we need to adjust the results. */
1922 if (dir
!= XFS_LOOKUP_LE
&& diff
< 0) {
1925 * If ge search and we went off the end of the block, but it's
1926 * not the last block, we're in the wrong block.
1928 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_RIGHTSIB
);
1929 if (dir
== XFS_LOOKUP_GE
&&
1930 keyno
> xfs_btree_get_numrecs(block
) &&
1931 !xfs_btree_ptr_is_null(cur
, &ptr
)) {
1934 cur
->bc_ptrs
[0] = keyno
;
1935 error
= xfs_btree_increment(cur
, 0, &i
);
1938 XFS_WANT_CORRUPTED_RETURN(cur
->bc_mp
, i
== 1);
1939 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1943 } else if (dir
== XFS_LOOKUP_LE
&& diff
> 0)
1945 cur
->bc_ptrs
[0] = keyno
;
1947 /* Return if we succeeded or not. */
1948 if (keyno
== 0 || keyno
> xfs_btree_get_numrecs(block
))
1950 else if (dir
!= XFS_LOOKUP_EQ
|| diff
== 0)
1954 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
1958 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
1962 /* Find the high key storage area from a regular key. */
1963 STATIC
union xfs_btree_key
*
1964 xfs_btree_high_key_from_key(
1965 struct xfs_btree_cur
*cur
,
1966 union xfs_btree_key
*key
)
1968 ASSERT(cur
->bc_flags
& XFS_BTREE_OVERLAPPING
);
1969 return (union xfs_btree_key
*)((char *)key
+
1970 (cur
->bc_ops
->key_len
/ 2));
1973 /* Determine the low (and high if overlapped) keys of a leaf block */
1975 xfs_btree_get_leaf_keys(
1976 struct xfs_btree_cur
*cur
,
1977 struct xfs_btree_block
*block
,
1978 union xfs_btree_key
*key
)
1980 union xfs_btree_key max_hkey
;
1981 union xfs_btree_key hkey
;
1982 union xfs_btree_rec
*rec
;
1983 union xfs_btree_key
*high
;
1986 rec
= xfs_btree_rec_addr(cur
, 1, block
);
1987 cur
->bc_ops
->init_key_from_rec(key
, rec
);
1989 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
1991 cur
->bc_ops
->init_high_key_from_rec(&max_hkey
, rec
);
1992 for (n
= 2; n
<= xfs_btree_get_numrecs(block
); n
++) {
1993 rec
= xfs_btree_rec_addr(cur
, n
, block
);
1994 cur
->bc_ops
->init_high_key_from_rec(&hkey
, rec
);
1995 if (cur
->bc_ops
->diff_two_keys(cur
, &hkey
, &max_hkey
)
2000 high
= xfs_btree_high_key_from_key(cur
, key
);
2001 memcpy(high
, &max_hkey
, cur
->bc_ops
->key_len
/ 2);
2005 /* Determine the low (and high if overlapped) keys of a node block */
2007 xfs_btree_get_node_keys(
2008 struct xfs_btree_cur
*cur
,
2009 struct xfs_btree_block
*block
,
2010 union xfs_btree_key
*key
)
2012 union xfs_btree_key
*hkey
;
2013 union xfs_btree_key
*max_hkey
;
2014 union xfs_btree_key
*high
;
2017 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2018 memcpy(key
, xfs_btree_key_addr(cur
, 1, block
),
2019 cur
->bc_ops
->key_len
/ 2);
2021 max_hkey
= xfs_btree_high_key_addr(cur
, 1, block
);
2022 for (n
= 2; n
<= xfs_btree_get_numrecs(block
); n
++) {
2023 hkey
= xfs_btree_high_key_addr(cur
, n
, block
);
2024 if (cur
->bc_ops
->diff_two_keys(cur
, hkey
, max_hkey
) > 0)
2028 high
= xfs_btree_high_key_from_key(cur
, key
);
2029 memcpy(high
, max_hkey
, cur
->bc_ops
->key_len
/ 2);
2031 memcpy(key
, xfs_btree_key_addr(cur
, 1, block
),
2032 cur
->bc_ops
->key_len
);
2036 /* Derive the keys for any btree block. */
2039 struct xfs_btree_cur
*cur
,
2040 struct xfs_btree_block
*block
,
2041 union xfs_btree_key
*key
)
2043 if (be16_to_cpu(block
->bb_level
) == 0)
2044 xfs_btree_get_leaf_keys(cur
, block
, key
);
2046 xfs_btree_get_node_keys(cur
, block
, key
);
2050 * Decide if we need to update the parent keys of a btree block. For
2051 * a standard btree this is only necessary if we're updating the first
2052 * record/key. For an overlapping btree, we must always update the
2053 * keys because the highest key can be in any of the records or keys
2057 xfs_btree_needs_key_update(
2058 struct xfs_btree_cur
*cur
,
2061 return (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) || ptr
== 1;
2065 * Update the low and high parent keys of the given level, progressing
2066 * towards the root. If force_all is false, stop if the keys for a given
2067 * level do not need updating.
2070 __xfs_btree_updkeys(
2071 struct xfs_btree_cur
*cur
,
2073 struct xfs_btree_block
*block
,
2074 struct xfs_buf
*bp0
,
2077 union xfs_btree_key key
; /* keys from current level */
2078 union xfs_btree_key
*lkey
; /* keys from the next level up */
2079 union xfs_btree_key
*hkey
;
2080 union xfs_btree_key
*nlkey
; /* keys from the next level up */
2081 union xfs_btree_key
*nhkey
;
2085 ASSERT(cur
->bc_flags
& XFS_BTREE_OVERLAPPING
);
2087 /* Exit if there aren't any parent levels to update. */
2088 if (level
+ 1 >= cur
->bc_nlevels
)
2091 trace_xfs_btree_updkeys(cur
, level
, bp0
);
2094 hkey
= xfs_btree_high_key_from_key(cur
, lkey
);
2095 xfs_btree_get_keys(cur
, block
, lkey
);
2096 for (level
++; level
< cur
->bc_nlevels
; level
++) {
2100 block
= xfs_btree_get_block(cur
, level
, &bp
);
2101 trace_xfs_btree_updkeys(cur
, level
, bp
);
2103 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
2105 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2109 ptr
= cur
->bc_ptrs
[level
];
2110 nlkey
= xfs_btree_key_addr(cur
, ptr
, block
);
2111 nhkey
= xfs_btree_high_key_addr(cur
, ptr
, block
);
2113 !(cur
->bc_ops
->diff_two_keys(cur
, nlkey
, lkey
) != 0 ||
2114 cur
->bc_ops
->diff_two_keys(cur
, nhkey
, hkey
) != 0))
2116 xfs_btree_copy_keys(cur
, nlkey
, lkey
, 1);
2117 xfs_btree_log_keys(cur
, bp
, ptr
, ptr
);
2118 if (level
+ 1 >= cur
->bc_nlevels
)
2120 xfs_btree_get_node_keys(cur
, block
, lkey
);
2126 /* Update all the keys from some level in cursor back to the root. */
2128 xfs_btree_updkeys_force(
2129 struct xfs_btree_cur
*cur
,
2133 struct xfs_btree_block
*block
;
2135 block
= xfs_btree_get_block(cur
, level
, &bp
);
2136 return __xfs_btree_updkeys(cur
, level
, block
, bp
, true);
2140 * Update the parent keys of the given level, progressing towards the root.
2143 xfs_btree_update_keys(
2144 struct xfs_btree_cur
*cur
,
2147 struct xfs_btree_block
*block
;
2149 union xfs_btree_key
*kp
;
2150 union xfs_btree_key key
;
2155 block
= xfs_btree_get_block(cur
, level
, &bp
);
2156 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
)
2157 return __xfs_btree_updkeys(cur
, level
, block
, bp
, false);
2159 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
2160 XFS_BTREE_TRACE_ARGIK(cur
, level
, keyp
);
2163 * Go up the tree from this level toward the root.
2164 * At each level, update the key value to the value input.
2165 * Stop when we reach a level where the cursor isn't pointing
2166 * at the first entry in the block.
2168 xfs_btree_get_keys(cur
, block
, &key
);
2169 for (level
++, ptr
= 1; ptr
== 1 && level
< cur
->bc_nlevels
; level
++) {
2173 block
= xfs_btree_get_block(cur
, level
, &bp
);
2175 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
2177 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2181 ptr
= cur
->bc_ptrs
[level
];
2182 kp
= xfs_btree_key_addr(cur
, ptr
, block
);
2183 xfs_btree_copy_keys(cur
, kp
, &key
, 1);
2184 xfs_btree_log_keys(cur
, bp
, ptr
, ptr
);
2187 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2192 * Update the record referred to by cur to the value in the
2193 * given record. This either works (return 0) or gets an
2194 * EFSCORRUPTED error.
2198 struct xfs_btree_cur
*cur
,
2199 union xfs_btree_rec
*rec
)
2201 struct xfs_btree_block
*block
;
2205 union xfs_btree_rec
*rp
;
2207 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
2208 XFS_BTREE_TRACE_ARGR(cur
, rec
);
2210 /* Pick up the current block. */
2211 block
= xfs_btree_get_block(cur
, 0, &bp
);
2214 error
= xfs_btree_check_block(cur
, block
, 0, bp
);
2218 /* Get the address of the rec to be updated. */
2219 ptr
= cur
->bc_ptrs
[0];
2220 rp
= xfs_btree_rec_addr(cur
, ptr
, block
);
2222 /* Fill in the new contents and log them. */
2223 xfs_btree_copy_recs(cur
, rp
, rec
, 1);
2224 xfs_btree_log_recs(cur
, bp
, ptr
, ptr
);
2227 * If we are tracking the last record in the tree and
2228 * we are at the far right edge of the tree, update it.
2230 if (xfs_btree_is_lastrec(cur
, block
, 0)) {
2231 cur
->bc_ops
->update_lastrec(cur
, block
, rec
,
2232 ptr
, LASTREC_UPDATE
);
2235 /* Pass new key value up to our parent. */
2236 if (xfs_btree_needs_key_update(cur
, ptr
)) {
2237 error
= xfs_btree_update_keys(cur
, 0);
2242 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2246 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2251 * Move 1 record left from cur/level if possible.
2252 * Update cur to reflect the new path.
2254 STATIC
int /* error */
2256 struct xfs_btree_cur
*cur
,
2258 int *stat
) /* success/failure */
2260 struct xfs_buf
*lbp
; /* left buffer pointer */
2261 struct xfs_btree_block
*left
; /* left btree block */
2262 int lrecs
; /* left record count */
2263 struct xfs_buf
*rbp
; /* right buffer pointer */
2264 struct xfs_btree_block
*right
; /* right btree block */
2265 struct xfs_btree_cur
*tcur
; /* temporary btree cursor */
2266 int rrecs
; /* right record count */
2267 union xfs_btree_ptr lptr
; /* left btree pointer */
2268 union xfs_btree_key
*rkp
= NULL
; /* right btree key */
2269 union xfs_btree_ptr
*rpp
= NULL
; /* right address pointer */
2270 union xfs_btree_rec
*rrp
= NULL
; /* right record pointer */
2271 int error
; /* error return value */
2274 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
2275 XFS_BTREE_TRACE_ARGI(cur
, level
);
2277 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
2278 level
== cur
->bc_nlevels
- 1)
2281 /* Set up variables for this block as "right". */
2282 right
= xfs_btree_get_block(cur
, level
, &rbp
);
2285 error
= xfs_btree_check_block(cur
, right
, level
, rbp
);
2290 /* If we've got no left sibling then we can't shift an entry left. */
2291 xfs_btree_get_sibling(cur
, right
, &lptr
, XFS_BB_LEFTSIB
);
2292 if (xfs_btree_ptr_is_null(cur
, &lptr
))
2296 * If the cursor entry is the one that would be moved, don't
2297 * do it... it's too complicated.
2299 if (cur
->bc_ptrs
[level
] <= 1)
2302 /* Set up the left neighbor as "left". */
2303 error
= xfs_btree_read_buf_block(cur
, &lptr
, 0, &left
, &lbp
);
2307 /* If it's full, it can't take another entry. */
2308 lrecs
= xfs_btree_get_numrecs(left
);
2309 if (lrecs
== cur
->bc_ops
->get_maxrecs(cur
, level
))
2312 rrecs
= xfs_btree_get_numrecs(right
);
2315 * We add one entry to the left side and remove one for the right side.
2316 * Account for it here, the changes will be updated on disk and logged
2322 XFS_BTREE_STATS_INC(cur
, lshift
);
2323 XFS_BTREE_STATS_ADD(cur
, moves
, 1);
2326 * If non-leaf, copy a key and a ptr to the left block.
2327 * Log the changes to the left block.
2330 /* It's a non-leaf. Move keys and pointers. */
2331 union xfs_btree_key
*lkp
; /* left btree key */
2332 union xfs_btree_ptr
*lpp
; /* left address pointer */
2334 lkp
= xfs_btree_key_addr(cur
, lrecs
, left
);
2335 rkp
= xfs_btree_key_addr(cur
, 1, right
);
2337 lpp
= xfs_btree_ptr_addr(cur
, lrecs
, left
);
2338 rpp
= xfs_btree_ptr_addr(cur
, 1, right
);
2340 error
= xfs_btree_check_ptr(cur
, rpp
, 0, level
);
2344 xfs_btree_copy_keys(cur
, lkp
, rkp
, 1);
2345 xfs_btree_copy_ptrs(cur
, lpp
, rpp
, 1);
2347 xfs_btree_log_keys(cur
, lbp
, lrecs
, lrecs
);
2348 xfs_btree_log_ptrs(cur
, lbp
, lrecs
, lrecs
);
2350 ASSERT(cur
->bc_ops
->keys_inorder(cur
,
2351 xfs_btree_key_addr(cur
, lrecs
- 1, left
), lkp
));
2353 /* It's a leaf. Move records. */
2354 union xfs_btree_rec
*lrp
; /* left record pointer */
2356 lrp
= xfs_btree_rec_addr(cur
, lrecs
, left
);
2357 rrp
= xfs_btree_rec_addr(cur
, 1, right
);
2359 xfs_btree_copy_recs(cur
, lrp
, rrp
, 1);
2360 xfs_btree_log_recs(cur
, lbp
, lrecs
, lrecs
);
2362 ASSERT(cur
->bc_ops
->recs_inorder(cur
,
2363 xfs_btree_rec_addr(cur
, lrecs
- 1, left
), lrp
));
2366 xfs_btree_set_numrecs(left
, lrecs
);
2367 xfs_btree_log_block(cur
, lbp
, XFS_BB_NUMRECS
);
2369 xfs_btree_set_numrecs(right
, rrecs
);
2370 xfs_btree_log_block(cur
, rbp
, XFS_BB_NUMRECS
);
2373 * Slide the contents of right down one entry.
2375 XFS_BTREE_STATS_ADD(cur
, moves
, rrecs
- 1);
2377 /* It's a nonleaf. operate on keys and ptrs */
2379 int i
; /* loop index */
2381 for (i
= 0; i
< rrecs
; i
++) {
2382 error
= xfs_btree_check_ptr(cur
, rpp
, i
+ 1, level
);
2387 xfs_btree_shift_keys(cur
,
2388 xfs_btree_key_addr(cur
, 2, right
),
2390 xfs_btree_shift_ptrs(cur
,
2391 xfs_btree_ptr_addr(cur
, 2, right
),
2394 xfs_btree_log_keys(cur
, rbp
, 1, rrecs
);
2395 xfs_btree_log_ptrs(cur
, rbp
, 1, rrecs
);
2397 /* It's a leaf. operate on records */
2398 xfs_btree_shift_recs(cur
,
2399 xfs_btree_rec_addr(cur
, 2, right
),
2401 xfs_btree_log_recs(cur
, rbp
, 1, rrecs
);
2405 * Using a temporary cursor, update the parent key values of the
2406 * block on the left.
2408 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2409 error
= xfs_btree_dup_cursor(cur
, &tcur
);
2412 i
= xfs_btree_firstrec(tcur
, level
);
2413 XFS_WANT_CORRUPTED_GOTO(tcur
->bc_mp
, i
== 1, error0
);
2415 error
= xfs_btree_decrement(tcur
, level
, &i
);
2419 /* Update the parent high keys of the left block, if needed. */
2420 error
= xfs_btree_update_keys(tcur
, level
);
2424 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
2427 /* Update the parent keys of the right block. */
2428 error
= xfs_btree_update_keys(cur
, level
);
2432 /* Slide the cursor value left one. */
2433 cur
->bc_ptrs
[level
]--;
2435 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2440 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2445 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2449 XFS_BTREE_TRACE_CURSOR(tcur
, XBT_ERROR
);
2450 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
2455 * Move 1 record right from cur/level if possible.
2456 * Update cur to reflect the new path.
2458 STATIC
int /* error */
2460 struct xfs_btree_cur
*cur
,
2462 int *stat
) /* success/failure */
2464 struct xfs_buf
*lbp
; /* left buffer pointer */
2465 struct xfs_btree_block
*left
; /* left btree block */
2466 struct xfs_buf
*rbp
; /* right buffer pointer */
2467 struct xfs_btree_block
*right
; /* right btree block */
2468 struct xfs_btree_cur
*tcur
; /* temporary btree cursor */
2469 union xfs_btree_ptr rptr
; /* right block pointer */
2470 union xfs_btree_key
*rkp
; /* right btree key */
2471 int rrecs
; /* right record count */
2472 int lrecs
; /* left record count */
2473 int error
; /* error return value */
2474 int i
; /* loop counter */
2476 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
2477 XFS_BTREE_TRACE_ARGI(cur
, level
);
2479 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
2480 (level
== cur
->bc_nlevels
- 1))
2483 /* Set up variables for this block as "left". */
2484 left
= xfs_btree_get_block(cur
, level
, &lbp
);
2487 error
= xfs_btree_check_block(cur
, left
, level
, lbp
);
2492 /* If we've got no right sibling then we can't shift an entry right. */
2493 xfs_btree_get_sibling(cur
, left
, &rptr
, XFS_BB_RIGHTSIB
);
2494 if (xfs_btree_ptr_is_null(cur
, &rptr
))
2498 * If the cursor entry is the one that would be moved, don't
2499 * do it... it's too complicated.
2501 lrecs
= xfs_btree_get_numrecs(left
);
2502 if (cur
->bc_ptrs
[level
] >= lrecs
)
2505 /* Set up the right neighbor as "right". */
2506 error
= xfs_btree_read_buf_block(cur
, &rptr
, 0, &right
, &rbp
);
2510 /* If it's full, it can't take another entry. */
2511 rrecs
= xfs_btree_get_numrecs(right
);
2512 if (rrecs
== cur
->bc_ops
->get_maxrecs(cur
, level
))
2515 XFS_BTREE_STATS_INC(cur
, rshift
);
2516 XFS_BTREE_STATS_ADD(cur
, moves
, rrecs
);
2519 * Make a hole at the start of the right neighbor block, then
2520 * copy the last left block entry to the hole.
2523 /* It's a nonleaf. make a hole in the keys and ptrs */
2524 union xfs_btree_key
*lkp
;
2525 union xfs_btree_ptr
*lpp
;
2526 union xfs_btree_ptr
*rpp
;
2528 lkp
= xfs_btree_key_addr(cur
, lrecs
, left
);
2529 lpp
= xfs_btree_ptr_addr(cur
, lrecs
, left
);
2530 rkp
= xfs_btree_key_addr(cur
, 1, right
);
2531 rpp
= xfs_btree_ptr_addr(cur
, 1, right
);
2534 for (i
= rrecs
- 1; i
>= 0; i
--) {
2535 error
= xfs_btree_check_ptr(cur
, rpp
, i
, level
);
2541 xfs_btree_shift_keys(cur
, rkp
, 1, rrecs
);
2542 xfs_btree_shift_ptrs(cur
, rpp
, 1, rrecs
);
2545 error
= xfs_btree_check_ptr(cur
, lpp
, 0, level
);
2550 /* Now put the new data in, and log it. */
2551 xfs_btree_copy_keys(cur
, rkp
, lkp
, 1);
2552 xfs_btree_copy_ptrs(cur
, rpp
, lpp
, 1);
2554 xfs_btree_log_keys(cur
, rbp
, 1, rrecs
+ 1);
2555 xfs_btree_log_ptrs(cur
, rbp
, 1, rrecs
+ 1);
2557 ASSERT(cur
->bc_ops
->keys_inorder(cur
, rkp
,
2558 xfs_btree_key_addr(cur
, 2, right
)));
2560 /* It's a leaf. make a hole in the records */
2561 union xfs_btree_rec
*lrp
;
2562 union xfs_btree_rec
*rrp
;
2564 lrp
= xfs_btree_rec_addr(cur
, lrecs
, left
);
2565 rrp
= xfs_btree_rec_addr(cur
, 1, right
);
2567 xfs_btree_shift_recs(cur
, rrp
, 1, rrecs
);
2569 /* Now put the new data in, and log it. */
2570 xfs_btree_copy_recs(cur
, rrp
, lrp
, 1);
2571 xfs_btree_log_recs(cur
, rbp
, 1, rrecs
+ 1);
2575 * Decrement and log left's numrecs, bump and log right's numrecs.
2577 xfs_btree_set_numrecs(left
, --lrecs
);
2578 xfs_btree_log_block(cur
, lbp
, XFS_BB_NUMRECS
);
2580 xfs_btree_set_numrecs(right
, ++rrecs
);
2581 xfs_btree_log_block(cur
, rbp
, XFS_BB_NUMRECS
);
2584 * Using a temporary cursor, update the parent key values of the
2585 * block on the right.
2587 error
= xfs_btree_dup_cursor(cur
, &tcur
);
2590 i
= xfs_btree_lastrec(tcur
, level
);
2591 XFS_WANT_CORRUPTED_GOTO(tcur
->bc_mp
, i
== 1, error0
);
2593 error
= xfs_btree_increment(tcur
, level
, &i
);
2597 /* Update the parent high keys of the left block, if needed. */
2598 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2599 error
= xfs_btree_update_keys(cur
, level
);
2604 /* Update the parent keys of the right block. */
2605 error
= xfs_btree_update_keys(tcur
, level
);
2609 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
2611 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2616 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2621 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2625 XFS_BTREE_TRACE_CURSOR(tcur
, XBT_ERROR
);
2626 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
2631 * Split cur/level block in half.
2632 * Return new block number and the key to its first
2633 * record (to be inserted into parent).
2635 STATIC
int /* error */
2637 struct xfs_btree_cur
*cur
,
2639 union xfs_btree_ptr
*ptrp
,
2640 union xfs_btree_key
*key
,
2641 struct xfs_btree_cur
**curp
,
2642 int *stat
) /* success/failure */
2644 union xfs_btree_ptr lptr
; /* left sibling block ptr */
2645 struct xfs_buf
*lbp
; /* left buffer pointer */
2646 struct xfs_btree_block
*left
; /* left btree block */
2647 union xfs_btree_ptr rptr
; /* right sibling block ptr */
2648 struct xfs_buf
*rbp
; /* right buffer pointer */
2649 struct xfs_btree_block
*right
; /* right btree block */
2650 union xfs_btree_ptr rrptr
; /* right-right sibling ptr */
2651 struct xfs_buf
*rrbp
; /* right-right buffer pointer */
2652 struct xfs_btree_block
*rrblock
; /* right-right btree block */
2656 int error
; /* error return value */
2661 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
2662 XFS_BTREE_TRACE_ARGIPK(cur
, level
, *ptrp
, key
);
2664 XFS_BTREE_STATS_INC(cur
, split
);
2666 /* Set up left block (current one). */
2667 left
= xfs_btree_get_block(cur
, level
, &lbp
);
2670 error
= xfs_btree_check_block(cur
, left
, level
, lbp
);
2675 xfs_btree_buf_to_ptr(cur
, lbp
, &lptr
);
2677 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2678 error
= cur
->bc_ops
->alloc_block(cur
, &lptr
, &rptr
, stat
);
2683 XFS_BTREE_STATS_INC(cur
, alloc
);
2685 /* Set up the new block as "right". */
2686 error
= xfs_btree_get_buf_block(cur
, &rptr
, 0, &right
, &rbp
);
2690 /* Fill in the btree header for the new right block. */
2691 xfs_btree_init_block_cur(cur
, rbp
, xfs_btree_get_level(left
), 0);
2694 * Split the entries between the old and the new block evenly.
2695 * Make sure that if there's an odd number of entries now, that
2696 * each new block will have the same number of entries.
2698 lrecs
= xfs_btree_get_numrecs(left
);
2700 if ((lrecs
& 1) && cur
->bc_ptrs
[level
] <= rrecs
+ 1)
2702 src_index
= (lrecs
- rrecs
+ 1);
2704 XFS_BTREE_STATS_ADD(cur
, moves
, rrecs
);
2706 /* Adjust numrecs for the later get_*_keys() calls. */
2708 xfs_btree_set_numrecs(left
, lrecs
);
2709 xfs_btree_set_numrecs(right
, xfs_btree_get_numrecs(right
) + rrecs
);
2712 * Copy btree block entries from the left block over to the
2713 * new block, the right. Update the right block and log the
2717 /* It's a non-leaf. Move keys and pointers. */
2718 union xfs_btree_key
*lkp
; /* left btree key */
2719 union xfs_btree_ptr
*lpp
; /* left address pointer */
2720 union xfs_btree_key
*rkp
; /* right btree key */
2721 union xfs_btree_ptr
*rpp
; /* right address pointer */
2723 lkp
= xfs_btree_key_addr(cur
, src_index
, left
);
2724 lpp
= xfs_btree_ptr_addr(cur
, src_index
, left
);
2725 rkp
= xfs_btree_key_addr(cur
, 1, right
);
2726 rpp
= xfs_btree_ptr_addr(cur
, 1, right
);
2729 for (i
= src_index
; i
< rrecs
; i
++) {
2730 error
= xfs_btree_check_ptr(cur
, lpp
, i
, level
);
2736 /* Copy the keys & pointers to the new block. */
2737 xfs_btree_copy_keys(cur
, rkp
, lkp
, rrecs
);
2738 xfs_btree_copy_ptrs(cur
, rpp
, lpp
, rrecs
);
2740 xfs_btree_log_keys(cur
, rbp
, 1, rrecs
);
2741 xfs_btree_log_ptrs(cur
, rbp
, 1, rrecs
);
2743 /* Stash the keys of the new block for later insertion. */
2744 xfs_btree_get_node_keys(cur
, right
, key
);
2746 /* It's a leaf. Move records. */
2747 union xfs_btree_rec
*lrp
; /* left record pointer */
2748 union xfs_btree_rec
*rrp
; /* right record pointer */
2750 lrp
= xfs_btree_rec_addr(cur
, src_index
, left
);
2751 rrp
= xfs_btree_rec_addr(cur
, 1, right
);
2753 /* Copy records to the new block. */
2754 xfs_btree_copy_recs(cur
, rrp
, lrp
, rrecs
);
2755 xfs_btree_log_recs(cur
, rbp
, 1, rrecs
);
2757 /* Stash the keys of the new block for later insertion. */
2758 xfs_btree_get_leaf_keys(cur
, right
, key
);
2762 * Find the left block number by looking in the buffer.
2763 * Adjust sibling pointers.
2765 xfs_btree_get_sibling(cur
, left
, &rrptr
, XFS_BB_RIGHTSIB
);
2766 xfs_btree_set_sibling(cur
, right
, &rrptr
, XFS_BB_RIGHTSIB
);
2767 xfs_btree_set_sibling(cur
, right
, &lptr
, XFS_BB_LEFTSIB
);
2768 xfs_btree_set_sibling(cur
, left
, &rptr
, XFS_BB_RIGHTSIB
);
2770 xfs_btree_log_block(cur
, rbp
, XFS_BB_ALL_BITS
);
2771 xfs_btree_log_block(cur
, lbp
, XFS_BB_NUMRECS
| XFS_BB_RIGHTSIB
);
2774 * If there's a block to the new block's right, make that block
2775 * point back to right instead of to left.
2777 if (!xfs_btree_ptr_is_null(cur
, &rrptr
)) {
2778 error
= xfs_btree_read_buf_block(cur
, &rrptr
,
2779 0, &rrblock
, &rrbp
);
2782 xfs_btree_set_sibling(cur
, rrblock
, &rptr
, XFS_BB_LEFTSIB
);
2783 xfs_btree_log_block(cur
, rrbp
, XFS_BB_LEFTSIB
);
2786 /* Update the parent high keys of the left block, if needed. */
2787 if (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
) {
2788 error
= xfs_btree_update_keys(cur
, level
);
2794 * If the cursor is really in the right block, move it there.
2795 * If it's just pointing past the last entry in left, then we'll
2796 * insert there, so don't change anything in that case.
2798 if (cur
->bc_ptrs
[level
] > lrecs
+ 1) {
2799 xfs_btree_setbuf(cur
, level
, rbp
);
2800 cur
->bc_ptrs
[level
] -= lrecs
;
2803 * If there are more levels, we'll need another cursor which refers
2804 * the right block, no matter where this cursor was.
2806 if (level
+ 1 < cur
->bc_nlevels
) {
2807 error
= xfs_btree_dup_cursor(cur
, curp
);
2810 (*curp
)->bc_ptrs
[level
+ 1]++;
2813 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2817 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2822 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
2826 struct xfs_btree_split_args
{
2827 struct xfs_btree_cur
*cur
;
2829 union xfs_btree_ptr
*ptrp
;
2830 union xfs_btree_key
*key
;
2831 struct xfs_btree_cur
**curp
;
2832 int *stat
; /* success/failure */
2834 bool kswapd
; /* allocation in kswapd context */
2835 struct completion
*done
;
2836 struct work_struct work
;
2840 * Stack switching interfaces for allocation
2843 xfs_btree_split_worker(
2844 struct work_struct
*work
)
2846 struct xfs_btree_split_args
*args
= container_of(work
,
2847 struct xfs_btree_split_args
, work
);
2848 unsigned long pflags
;
2849 unsigned long new_pflags
= PF_FSTRANS
;
2852 * we are in a transaction context here, but may also be doing work
2853 * in kswapd context, and hence we may need to inherit that state
2854 * temporarily to ensure that we don't block waiting for memory reclaim
2858 new_pflags
|= PF_MEMALLOC
| PF_SWAPWRITE
| PF_KSWAPD
;
2860 current_set_flags_nested(&pflags
, new_pflags
);
2862 args
->result
= __xfs_btree_split(args
->cur
, args
->level
, args
->ptrp
,
2863 args
->key
, args
->curp
, args
->stat
);
2864 complete(args
->done
);
2866 current_restore_flags_nested(&pflags
, new_pflags
);
2870 * BMBT split requests often come in with little stack to work on. Push
2871 * them off to a worker thread so there is lots of stack to use. For the other
2872 * btree types, just call directly to avoid the context switch overhead here.
2874 STATIC
int /* error */
2876 struct xfs_btree_cur
*cur
,
2878 union xfs_btree_ptr
*ptrp
,
2879 union xfs_btree_key
*key
,
2880 struct xfs_btree_cur
**curp
,
2881 int *stat
) /* success/failure */
2883 struct xfs_btree_split_args args
;
2884 DECLARE_COMPLETION_ONSTACK(done
);
2886 if (cur
->bc_btnum
!= XFS_BTNUM_BMAP
)
2887 return __xfs_btree_split(cur
, level
, ptrp
, key
, curp
, stat
);
2896 args
.kswapd
= current_is_kswapd();
2897 INIT_WORK_ONSTACK(&args
.work
, xfs_btree_split_worker
);
2898 queue_work(xfs_alloc_wq
, &args
.work
);
2899 wait_for_completion(&done
);
2900 destroy_work_on_stack(&args
.work
);
2906 * Copy the old inode root contents into a real block and make the
2907 * broot point to it.
2910 xfs_btree_new_iroot(
2911 struct xfs_btree_cur
*cur
, /* btree cursor */
2912 int *logflags
, /* logging flags for inode */
2913 int *stat
) /* return status - 0 fail */
2915 struct xfs_buf
*cbp
; /* buffer for cblock */
2916 struct xfs_btree_block
*block
; /* btree block */
2917 struct xfs_btree_block
*cblock
; /* child btree block */
2918 union xfs_btree_key
*ckp
; /* child key pointer */
2919 union xfs_btree_ptr
*cpp
; /* child ptr pointer */
2920 union xfs_btree_key
*kp
; /* pointer to btree key */
2921 union xfs_btree_ptr
*pp
; /* pointer to block addr */
2922 union xfs_btree_ptr nptr
; /* new block addr */
2923 int level
; /* btree level */
2924 int error
; /* error return code */
2926 int i
; /* loop counter */
2929 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
2930 XFS_BTREE_STATS_INC(cur
, newroot
);
2932 ASSERT(cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
);
2934 level
= cur
->bc_nlevels
- 1;
2936 block
= xfs_btree_get_iroot(cur
);
2937 pp
= xfs_btree_ptr_addr(cur
, 1, block
);
2939 /* Allocate the new block. If we can't do it, we're toast. Give up. */
2940 error
= cur
->bc_ops
->alloc_block(cur
, pp
, &nptr
, stat
);
2944 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
2947 XFS_BTREE_STATS_INC(cur
, alloc
);
2949 /* Copy the root into a real block. */
2950 error
= xfs_btree_get_buf_block(cur
, &nptr
, 0, &cblock
, &cbp
);
2955 * we can't just memcpy() the root in for CRC enabled btree blocks.
2956 * In that case have to also ensure the blkno remains correct
2958 memcpy(cblock
, block
, xfs_btree_block_len(cur
));
2959 if (cur
->bc_flags
& XFS_BTREE_CRC_BLOCKS
) {
2960 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
2961 cblock
->bb_u
.l
.bb_blkno
= cpu_to_be64(cbp
->b_bn
);
2963 cblock
->bb_u
.s
.bb_blkno
= cpu_to_be64(cbp
->b_bn
);
2966 be16_add_cpu(&block
->bb_level
, 1);
2967 xfs_btree_set_numrecs(block
, 1);
2969 cur
->bc_ptrs
[level
+ 1] = 1;
2971 kp
= xfs_btree_key_addr(cur
, 1, block
);
2972 ckp
= xfs_btree_key_addr(cur
, 1, cblock
);
2973 xfs_btree_copy_keys(cur
, ckp
, kp
, xfs_btree_get_numrecs(cblock
));
2975 cpp
= xfs_btree_ptr_addr(cur
, 1, cblock
);
2977 for (i
= 0; i
< be16_to_cpu(cblock
->bb_numrecs
); i
++) {
2978 error
= xfs_btree_check_ptr(cur
, pp
, i
, level
);
2983 xfs_btree_copy_ptrs(cur
, cpp
, pp
, xfs_btree_get_numrecs(cblock
));
2986 error
= xfs_btree_check_ptr(cur
, &nptr
, 0, level
);
2990 xfs_btree_copy_ptrs(cur
, pp
, &nptr
, 1);
2992 xfs_iroot_realloc(cur
->bc_private
.b
.ip
,
2993 1 - xfs_btree_get_numrecs(cblock
),
2994 cur
->bc_private
.b
.whichfork
);
2996 xfs_btree_setbuf(cur
, level
, cbp
);
2999 * Do all this logging at the end so that
3000 * the root is at the right level.
3002 xfs_btree_log_block(cur
, cbp
, XFS_BB_ALL_BITS
);
3003 xfs_btree_log_keys(cur
, cbp
, 1, be16_to_cpu(cblock
->bb_numrecs
));
3004 xfs_btree_log_ptrs(cur
, cbp
, 1, be16_to_cpu(cblock
->bb_numrecs
));
3007 XFS_ILOG_CORE
| xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
);
3009 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3012 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3017 * Allocate a new root block, fill it in.
3019 STATIC
int /* error */
3021 struct xfs_btree_cur
*cur
, /* btree cursor */
3022 int *stat
) /* success/failure */
3024 struct xfs_btree_block
*block
; /* one half of the old root block */
3025 struct xfs_buf
*bp
; /* buffer containing block */
3026 int error
; /* error return value */
3027 struct xfs_buf
*lbp
; /* left buffer pointer */
3028 struct xfs_btree_block
*left
; /* left btree block */
3029 struct xfs_buf
*nbp
; /* new (root) buffer */
3030 struct xfs_btree_block
*new; /* new (root) btree block */
3031 int nptr
; /* new value for key index, 1 or 2 */
3032 struct xfs_buf
*rbp
; /* right buffer pointer */
3033 struct xfs_btree_block
*right
; /* right btree block */
3034 union xfs_btree_ptr rptr
;
3035 union xfs_btree_ptr lptr
;
3037 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
3038 XFS_BTREE_STATS_INC(cur
, newroot
);
3040 /* initialise our start point from the cursor */
3041 cur
->bc_ops
->init_ptr_from_cur(cur
, &rptr
);
3043 /* Allocate the new block. If we can't do it, we're toast. Give up. */
3044 error
= cur
->bc_ops
->alloc_block(cur
, &rptr
, &lptr
, stat
);
3049 XFS_BTREE_STATS_INC(cur
, alloc
);
3051 /* Set up the new block. */
3052 error
= xfs_btree_get_buf_block(cur
, &lptr
, 0, &new, &nbp
);
3056 /* Set the root in the holding structure increasing the level by 1. */
3057 cur
->bc_ops
->set_root(cur
, &lptr
, 1);
3060 * At the previous root level there are now two blocks: the old root,
3061 * and the new block generated when it was split. We don't know which
3062 * one the cursor is pointing at, so we set up variables "left" and
3063 * "right" for each case.
3065 block
= xfs_btree_get_block(cur
, cur
->bc_nlevels
- 1, &bp
);
3068 error
= xfs_btree_check_block(cur
, block
, cur
->bc_nlevels
- 1, bp
);
3073 xfs_btree_get_sibling(cur
, block
, &rptr
, XFS_BB_RIGHTSIB
);
3074 if (!xfs_btree_ptr_is_null(cur
, &rptr
)) {
3075 /* Our block is left, pick up the right block. */
3077 xfs_btree_buf_to_ptr(cur
, lbp
, &lptr
);
3079 error
= xfs_btree_read_buf_block(cur
, &rptr
, 0, &right
, &rbp
);
3085 /* Our block is right, pick up the left block. */
3087 xfs_btree_buf_to_ptr(cur
, rbp
, &rptr
);
3089 xfs_btree_get_sibling(cur
, right
, &lptr
, XFS_BB_LEFTSIB
);
3090 error
= xfs_btree_read_buf_block(cur
, &lptr
, 0, &left
, &lbp
);
3097 /* Fill in the new block's btree header and log it. */
3098 xfs_btree_init_block_cur(cur
, nbp
, cur
->bc_nlevels
, 2);
3099 xfs_btree_log_block(cur
, nbp
, XFS_BB_ALL_BITS
);
3100 ASSERT(!xfs_btree_ptr_is_null(cur
, &lptr
) &&
3101 !xfs_btree_ptr_is_null(cur
, &rptr
));
3103 /* Fill in the key data in the new root. */
3104 if (xfs_btree_get_level(left
) > 0) {
3106 * Get the keys for the left block's keys and put them directly
3107 * in the parent block. Do the same for the right block.
3109 xfs_btree_get_node_keys(cur
, left
,
3110 xfs_btree_key_addr(cur
, 1, new));
3111 xfs_btree_get_node_keys(cur
, right
,
3112 xfs_btree_key_addr(cur
, 2, new));
3115 * Get the keys for the left block's records and put them
3116 * directly in the parent block. Do the same for the right
3119 xfs_btree_get_leaf_keys(cur
, left
,
3120 xfs_btree_key_addr(cur
, 1, new));
3121 xfs_btree_get_leaf_keys(cur
, right
,
3122 xfs_btree_key_addr(cur
, 2, new));
3124 xfs_btree_log_keys(cur
, nbp
, 1, 2);
3126 /* Fill in the pointer data in the new root. */
3127 xfs_btree_copy_ptrs(cur
,
3128 xfs_btree_ptr_addr(cur
, 1, new), &lptr
, 1);
3129 xfs_btree_copy_ptrs(cur
,
3130 xfs_btree_ptr_addr(cur
, 2, new), &rptr
, 1);
3131 xfs_btree_log_ptrs(cur
, nbp
, 1, 2);
3133 /* Fix up the cursor. */
3134 xfs_btree_setbuf(cur
, cur
->bc_nlevels
, nbp
);
3135 cur
->bc_ptrs
[cur
->bc_nlevels
] = nptr
;
3137 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3141 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3144 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3150 xfs_btree_make_block_unfull(
3151 struct xfs_btree_cur
*cur
, /* btree cursor */
3152 int level
, /* btree level */
3153 int numrecs
,/* # of recs in block */
3154 int *oindex
,/* old tree index */
3155 int *index
, /* new tree index */
3156 union xfs_btree_ptr
*nptr
, /* new btree ptr */
3157 struct xfs_btree_cur
**ncur
, /* new btree cursor */
3158 union xfs_btree_key
*key
, /* key of new block */
3163 if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
3164 level
== cur
->bc_nlevels
- 1) {
3165 struct xfs_inode
*ip
= cur
->bc_private
.b
.ip
;
3167 if (numrecs
< cur
->bc_ops
->get_dmaxrecs(cur
, level
)) {
3168 /* A root block that can be made bigger. */
3169 xfs_iroot_realloc(ip
, 1, cur
->bc_private
.b
.whichfork
);
3172 /* A root block that needs replacing */
3175 error
= xfs_btree_new_iroot(cur
, &logflags
, stat
);
3176 if (error
|| *stat
== 0)
3179 xfs_trans_log_inode(cur
->bc_tp
, ip
, logflags
);
3185 /* First, try shifting an entry to the right neighbor. */
3186 error
= xfs_btree_rshift(cur
, level
, stat
);
3190 /* Next, try shifting an entry to the left neighbor. */
3191 error
= xfs_btree_lshift(cur
, level
, stat
);
3196 *oindex
= *index
= cur
->bc_ptrs
[level
];
3201 * Next, try splitting the current block in half.
3203 * If this works we have to re-set our variables because we
3204 * could be in a different block now.
3206 error
= xfs_btree_split(cur
, level
, nptr
, key
, ncur
, stat
);
3207 if (error
|| *stat
== 0)
3211 *index
= cur
->bc_ptrs
[level
];
3216 * Insert one record/level. Return information to the caller
3217 * allowing the next level up to proceed if necessary.
3221 struct xfs_btree_cur
*cur
, /* btree cursor */
3222 int level
, /* level to insert record at */
3223 union xfs_btree_ptr
*ptrp
, /* i/o: block number inserted */
3224 union xfs_btree_rec
*rec
, /* record to insert */
3225 union xfs_btree_key
*key
, /* i/o: block key for ptrp */
3226 struct xfs_btree_cur
**curp
, /* output: new cursor replacing cur */
3227 int *stat
) /* success/failure */
3229 struct xfs_btree_block
*block
; /* btree block */
3230 struct xfs_buf
*bp
; /* buffer for block */
3231 union xfs_btree_ptr nptr
; /* new block ptr */
3232 struct xfs_btree_cur
*ncur
; /* new btree cursor */
3233 union xfs_btree_key nkey
; /* new block key */
3234 union xfs_btree_key
*lkey
;
3235 int optr
; /* old key/record index */
3236 int ptr
; /* key/record index */
3237 int numrecs
;/* number of records */
3238 int error
; /* error return value */
3244 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
3245 XFS_BTREE_TRACE_ARGIPR(cur
, level
, *ptrp
, &rec
);
3251 * If we have an external root pointer, and we've made it to the
3252 * root level, allocate a new root block and we're done.
3254 if (!(cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) &&
3255 (level
>= cur
->bc_nlevels
)) {
3256 error
= xfs_btree_new_root(cur
, stat
);
3257 xfs_btree_set_ptr_null(cur
, ptrp
);
3259 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3263 /* If we're off the left edge, return failure. */
3264 ptr
= cur
->bc_ptrs
[level
];
3266 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3273 XFS_BTREE_STATS_INC(cur
, insrec
);
3275 /* Get pointers to the btree buffer and block. */
3276 block
= xfs_btree_get_block(cur
, level
, &bp
);
3277 old_bn
= bp
? bp
->b_bn
: XFS_BUF_DADDR_NULL
;
3278 numrecs
= xfs_btree_get_numrecs(block
);
3281 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
3285 /* Check that the new entry is being inserted in the right place. */
3286 if (ptr
<= numrecs
) {
3288 ASSERT(cur
->bc_ops
->recs_inorder(cur
, rec
,
3289 xfs_btree_rec_addr(cur
, ptr
, block
)));
3291 ASSERT(cur
->bc_ops
->keys_inorder(cur
, key
,
3292 xfs_btree_key_addr(cur
, ptr
, block
)));
3298 * If the block is full, we can't insert the new entry until we
3299 * make the block un-full.
3301 xfs_btree_set_ptr_null(cur
, &nptr
);
3302 if (numrecs
== cur
->bc_ops
->get_maxrecs(cur
, level
)) {
3303 error
= xfs_btree_make_block_unfull(cur
, level
, numrecs
,
3304 &optr
, &ptr
, &nptr
, &ncur
, lkey
, stat
);
3305 if (error
|| *stat
== 0)
3310 * The current block may have changed if the block was
3311 * previously full and we have just made space in it.
3313 block
= xfs_btree_get_block(cur
, level
, &bp
);
3314 numrecs
= xfs_btree_get_numrecs(block
);
3317 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
3323 * At this point we know there's room for our new entry in the block
3324 * we're pointing at.
3326 XFS_BTREE_STATS_ADD(cur
, moves
, numrecs
- ptr
+ 1);
3329 /* It's a nonleaf. make a hole in the keys and ptrs */
3330 union xfs_btree_key
*kp
;
3331 union xfs_btree_ptr
*pp
;
3333 kp
= xfs_btree_key_addr(cur
, ptr
, block
);
3334 pp
= xfs_btree_ptr_addr(cur
, ptr
, block
);
3337 for (i
= numrecs
- ptr
; i
>= 0; i
--) {
3338 error
= xfs_btree_check_ptr(cur
, pp
, i
, level
);
3344 xfs_btree_shift_keys(cur
, kp
, 1, numrecs
- ptr
+ 1);
3345 xfs_btree_shift_ptrs(cur
, pp
, 1, numrecs
- ptr
+ 1);
3348 error
= xfs_btree_check_ptr(cur
, ptrp
, 0, level
);
3353 /* Now put the new data in, bump numrecs and log it. */
3354 xfs_btree_copy_keys(cur
, kp
, key
, 1);
3355 xfs_btree_copy_ptrs(cur
, pp
, ptrp
, 1);
3357 xfs_btree_set_numrecs(block
, numrecs
);
3358 xfs_btree_log_ptrs(cur
, bp
, ptr
, numrecs
);
3359 xfs_btree_log_keys(cur
, bp
, ptr
, numrecs
);
3361 if (ptr
< numrecs
) {
3362 ASSERT(cur
->bc_ops
->keys_inorder(cur
, kp
,
3363 xfs_btree_key_addr(cur
, ptr
+ 1, block
)));
3367 /* It's a leaf. make a hole in the records */
3368 union xfs_btree_rec
*rp
;
3370 rp
= xfs_btree_rec_addr(cur
, ptr
, block
);
3372 xfs_btree_shift_recs(cur
, rp
, 1, numrecs
- ptr
+ 1);
3374 /* Now put the new data in, bump numrecs and log it. */
3375 xfs_btree_copy_recs(cur
, rp
, rec
, 1);
3376 xfs_btree_set_numrecs(block
, ++numrecs
);
3377 xfs_btree_log_recs(cur
, bp
, ptr
, numrecs
);
3379 if (ptr
< numrecs
) {
3380 ASSERT(cur
->bc_ops
->recs_inorder(cur
, rp
,
3381 xfs_btree_rec_addr(cur
, ptr
+ 1, block
)));
3386 /* Log the new number of records in the btree header. */
3387 xfs_btree_log_block(cur
, bp
, XFS_BB_NUMRECS
);
3390 * If we just inserted into a new tree block, we have to
3391 * recalculate nkey here because nkey is out of date.
3393 * Otherwise we're just updating an existing block (having shoved
3394 * some records into the new tree block), so use the regular key
3397 if (bp
&& bp
->b_bn
!= old_bn
) {
3398 xfs_btree_get_keys(cur
, block
, lkey
);
3399 } else if (xfs_btree_needs_key_update(cur
, optr
)) {
3400 error
= xfs_btree_update_keys(cur
, level
);
3406 * If we are tracking the last record in the tree and
3407 * we are at the far right edge of the tree, update it.
3409 if (xfs_btree_is_lastrec(cur
, block
, level
)) {
3410 cur
->bc_ops
->update_lastrec(cur
, block
, rec
,
3411 ptr
, LASTREC_INSREC
);
3415 * Return the new block number, if any.
3416 * If there is one, give back a record value and a cursor too.
3419 if (!xfs_btree_ptr_is_null(cur
, &nptr
)) {
3420 xfs_btree_copy_keys(cur
, key
, lkey
, 1);
3424 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3429 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3434 * Insert the record at the point referenced by cur.
3436 * A multi-level split of the tree on insert will invalidate the original
3437 * cursor. All callers of this function should assume that the cursor is
3438 * no longer valid and revalidate it.
3442 struct xfs_btree_cur
*cur
,
3445 int error
; /* error return value */
3446 int i
; /* result value, 0 for failure */
3447 int level
; /* current level number in btree */
3448 union xfs_btree_ptr nptr
; /* new block number (split result) */
3449 struct xfs_btree_cur
*ncur
; /* new cursor (split result) */
3450 struct xfs_btree_cur
*pcur
; /* previous level's cursor */
3451 union xfs_btree_key bkey
; /* key of block to insert */
3452 union xfs_btree_key
*key
;
3453 union xfs_btree_rec rec
; /* record to insert */
3460 xfs_btree_set_ptr_null(cur
, &nptr
);
3462 /* Make a key out of the record data to be inserted, and save it. */
3463 cur
->bc_ops
->init_rec_from_cur(cur
, &rec
);
3464 cur
->bc_ops
->init_key_from_rec(key
, &rec
);
3467 * Loop going up the tree, starting at the leaf level.
3468 * Stop when we don't get a split block, that must mean that
3469 * the insert is finished with this level.
3473 * Insert nrec/nptr into this level of the tree.
3474 * Note if we fail, nptr will be null.
3476 error
= xfs_btree_insrec(pcur
, level
, &nptr
, &rec
, key
,
3480 xfs_btree_del_cursor(pcur
, XFS_BTREE_ERROR
);
3484 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3488 * See if the cursor we just used is trash.
3489 * Can't trash the caller's cursor, but otherwise we should
3490 * if ncur is a new cursor or we're about to be done.
3493 (ncur
|| xfs_btree_ptr_is_null(cur
, &nptr
))) {
3494 /* Save the state from the cursor before we trash it */
3495 if (cur
->bc_ops
->update_cursor
)
3496 cur
->bc_ops
->update_cursor(pcur
, cur
);
3497 cur
->bc_nlevels
= pcur
->bc_nlevels
;
3498 xfs_btree_del_cursor(pcur
, XFS_BTREE_NOERROR
);
3500 /* If we got a new cursor, switch to it. */
3505 } while (!xfs_btree_ptr_is_null(cur
, &nptr
));
3507 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3511 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3516 * Try to merge a non-leaf block back into the inode root.
3518 * Note: the killroot names comes from the fact that we're effectively
3519 * killing the old root block. But because we can't just delete the
3520 * inode we have to copy the single block it was pointing to into the
3524 xfs_btree_kill_iroot(
3525 struct xfs_btree_cur
*cur
)
3527 int whichfork
= cur
->bc_private
.b
.whichfork
;
3528 struct xfs_inode
*ip
= cur
->bc_private
.b
.ip
;
3529 struct xfs_ifork
*ifp
= XFS_IFORK_PTR(ip
, whichfork
);
3530 struct xfs_btree_block
*block
;
3531 struct xfs_btree_block
*cblock
;
3532 union xfs_btree_key
*kp
;
3533 union xfs_btree_key
*ckp
;
3534 union xfs_btree_ptr
*pp
;
3535 union xfs_btree_ptr
*cpp
;
3536 struct xfs_buf
*cbp
;
3542 union xfs_btree_ptr ptr
;
3546 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
3548 ASSERT(cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
);
3549 ASSERT(cur
->bc_nlevels
> 1);
3552 * Don't deal with the root block needs to be a leaf case.
3553 * We're just going to turn the thing back into extents anyway.
3555 level
= cur
->bc_nlevels
- 1;
3560 * Give up if the root has multiple children.
3562 block
= xfs_btree_get_iroot(cur
);
3563 if (xfs_btree_get_numrecs(block
) != 1)
3566 cblock
= xfs_btree_get_block(cur
, level
- 1, &cbp
);
3567 numrecs
= xfs_btree_get_numrecs(cblock
);
3570 * Only do this if the next level will fit.
3571 * Then the data must be copied up to the inode,
3572 * instead of freeing the root you free the next level.
3574 if (numrecs
> cur
->bc_ops
->get_dmaxrecs(cur
, level
))
3577 XFS_BTREE_STATS_INC(cur
, killroot
);
3580 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_LEFTSIB
);
3581 ASSERT(xfs_btree_ptr_is_null(cur
, &ptr
));
3582 xfs_btree_get_sibling(cur
, block
, &ptr
, XFS_BB_RIGHTSIB
);
3583 ASSERT(xfs_btree_ptr_is_null(cur
, &ptr
));
3586 index
= numrecs
- cur
->bc_ops
->get_maxrecs(cur
, level
);
3588 xfs_iroot_realloc(cur
->bc_private
.b
.ip
, index
,
3589 cur
->bc_private
.b
.whichfork
);
3590 block
= ifp
->if_broot
;
3593 be16_add_cpu(&block
->bb_numrecs
, index
);
3594 ASSERT(block
->bb_numrecs
== cblock
->bb_numrecs
);
3596 kp
= xfs_btree_key_addr(cur
, 1, block
);
3597 ckp
= xfs_btree_key_addr(cur
, 1, cblock
);
3598 xfs_btree_copy_keys(cur
, kp
, ckp
, numrecs
);
3600 pp
= xfs_btree_ptr_addr(cur
, 1, block
);
3601 cpp
= xfs_btree_ptr_addr(cur
, 1, cblock
);
3603 for (i
= 0; i
< numrecs
; i
++) {
3604 error
= xfs_btree_check_ptr(cur
, cpp
, i
, level
- 1);
3606 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3611 xfs_btree_copy_ptrs(cur
, pp
, cpp
, numrecs
);
3613 error
= xfs_btree_free_block(cur
, cbp
);
3615 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3619 cur
->bc_bufs
[level
- 1] = NULL
;
3620 be16_add_cpu(&block
->bb_level
, -1);
3621 xfs_trans_log_inode(cur
->bc_tp
, ip
,
3622 XFS_ILOG_CORE
| xfs_ilog_fbroot(cur
->bc_private
.b
.whichfork
));
3625 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3630 * Kill the current root node, and replace it with it's only child node.
3633 xfs_btree_kill_root(
3634 struct xfs_btree_cur
*cur
,
3637 union xfs_btree_ptr
*newroot
)
3641 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
3642 XFS_BTREE_STATS_INC(cur
, killroot
);
3645 * Update the root pointer, decreasing the level by 1 and then
3646 * free the old root.
3648 cur
->bc_ops
->set_root(cur
, newroot
, -1);
3650 error
= xfs_btree_free_block(cur
, bp
);
3652 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
3656 cur
->bc_bufs
[level
] = NULL
;
3657 cur
->bc_ra
[level
] = 0;
3660 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3665 xfs_btree_dec_cursor(
3666 struct xfs_btree_cur
*cur
,
3674 error
= xfs_btree_decrement(cur
, level
, &i
);
3679 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3685 * Single level of the btree record deletion routine.
3686 * Delete record pointed to by cur/level.
3687 * Remove the record from its block then rebalance the tree.
3688 * Return 0 for error, 1 for done, 2 to go on to the next level.
3690 STATIC
int /* error */
3692 struct xfs_btree_cur
*cur
, /* btree cursor */
3693 int level
, /* level removing record from */
3694 int *stat
) /* fail/done/go-on */
3696 struct xfs_btree_block
*block
; /* btree block */
3697 union xfs_btree_ptr cptr
; /* current block ptr */
3698 struct xfs_buf
*bp
; /* buffer for block */
3699 int error
; /* error return value */
3700 int i
; /* loop counter */
3701 union xfs_btree_ptr lptr
; /* left sibling block ptr */
3702 struct xfs_buf
*lbp
; /* left buffer pointer */
3703 struct xfs_btree_block
*left
; /* left btree block */
3704 int lrecs
= 0; /* left record count */
3705 int ptr
; /* key/record index */
3706 union xfs_btree_ptr rptr
; /* right sibling block ptr */
3707 struct xfs_buf
*rbp
; /* right buffer pointer */
3708 struct xfs_btree_block
*right
; /* right btree block */
3709 struct xfs_btree_block
*rrblock
; /* right-right btree block */
3710 struct xfs_buf
*rrbp
; /* right-right buffer pointer */
3711 int rrecs
= 0; /* right record count */
3712 struct xfs_btree_cur
*tcur
; /* temporary btree cursor */
3713 int numrecs
; /* temporary numrec count */
3715 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
3716 XFS_BTREE_TRACE_ARGI(cur
, level
);
3720 /* Get the index of the entry being deleted, check for nothing there. */
3721 ptr
= cur
->bc_ptrs
[level
];
3723 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3728 /* Get the buffer & block containing the record or key/ptr. */
3729 block
= xfs_btree_get_block(cur
, level
, &bp
);
3730 numrecs
= xfs_btree_get_numrecs(block
);
3733 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
3738 /* Fail if we're off the end of the block. */
3739 if (ptr
> numrecs
) {
3740 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
3745 XFS_BTREE_STATS_INC(cur
, delrec
);
3746 XFS_BTREE_STATS_ADD(cur
, moves
, numrecs
- ptr
);
3748 /* Excise the entries being deleted. */
3750 /* It's a nonleaf. operate on keys and ptrs */
3751 union xfs_btree_key
*lkp
;
3752 union xfs_btree_ptr
*lpp
;
3754 lkp
= xfs_btree_key_addr(cur
, ptr
+ 1, block
);
3755 lpp
= xfs_btree_ptr_addr(cur
, ptr
+ 1, block
);
3758 for (i
= 0; i
< numrecs
- ptr
; i
++) {
3759 error
= xfs_btree_check_ptr(cur
, lpp
, i
, level
);
3765 if (ptr
< numrecs
) {
3766 xfs_btree_shift_keys(cur
, lkp
, -1, numrecs
- ptr
);
3767 xfs_btree_shift_ptrs(cur
, lpp
, -1, numrecs
- ptr
);
3768 xfs_btree_log_keys(cur
, bp
, ptr
, numrecs
- 1);
3769 xfs_btree_log_ptrs(cur
, bp
, ptr
, numrecs
- 1);
3772 /* It's a leaf. operate on records */
3773 if (ptr
< numrecs
) {
3774 xfs_btree_shift_recs(cur
,
3775 xfs_btree_rec_addr(cur
, ptr
+ 1, block
),
3777 xfs_btree_log_recs(cur
, bp
, ptr
, numrecs
- 1);
3782 * Decrement and log the number of entries in the block.
3784 xfs_btree_set_numrecs(block
, --numrecs
);
3785 xfs_btree_log_block(cur
, bp
, XFS_BB_NUMRECS
);
3788 * If we are tracking the last record in the tree and
3789 * we are at the far right edge of the tree, update it.
3791 if (xfs_btree_is_lastrec(cur
, block
, level
)) {
3792 cur
->bc_ops
->update_lastrec(cur
, block
, NULL
,
3793 ptr
, LASTREC_DELREC
);
3797 * We're at the root level. First, shrink the root block in-memory.
3798 * Try to get rid of the next level down. If we can't then there's
3799 * nothing left to do.
3801 if (level
== cur
->bc_nlevels
- 1) {
3802 if (cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) {
3803 xfs_iroot_realloc(cur
->bc_private
.b
.ip
, -1,
3804 cur
->bc_private
.b
.whichfork
);
3806 error
= xfs_btree_kill_iroot(cur
);
3810 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3818 * If this is the root level, and there's only one entry left,
3819 * and it's NOT the leaf level, then we can get rid of this
3822 if (numrecs
== 1 && level
> 0) {
3823 union xfs_btree_ptr
*pp
;
3825 * pp is still set to the first pointer in the block.
3826 * Make it the new root of the btree.
3828 pp
= xfs_btree_ptr_addr(cur
, 1, block
);
3829 error
= xfs_btree_kill_root(cur
, bp
, level
, pp
);
3832 } else if (level
> 0) {
3833 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3842 * If we deleted the leftmost entry in the block, update the
3843 * key values above us in the tree.
3845 if (xfs_btree_needs_key_update(cur
, ptr
)) {
3846 error
= xfs_btree_update_keys(cur
, level
);
3852 * If the number of records remaining in the block is at least
3853 * the minimum, we're done.
3855 if (numrecs
>= cur
->bc_ops
->get_minrecs(cur
, level
)) {
3856 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3863 * Otherwise, we have to move some records around to keep the
3864 * tree balanced. Look at the left and right sibling blocks to
3865 * see if we can re-balance by moving only one record.
3867 xfs_btree_get_sibling(cur
, block
, &rptr
, XFS_BB_RIGHTSIB
);
3868 xfs_btree_get_sibling(cur
, block
, &lptr
, XFS_BB_LEFTSIB
);
3870 if (cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) {
3872 * One child of root, need to get a chance to copy its contents
3873 * into the root and delete it. Can't go up to next level,
3874 * there's nothing to delete there.
3876 if (xfs_btree_ptr_is_null(cur
, &rptr
) &&
3877 xfs_btree_ptr_is_null(cur
, &lptr
) &&
3878 level
== cur
->bc_nlevels
- 2) {
3879 error
= xfs_btree_kill_iroot(cur
);
3881 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3888 ASSERT(!xfs_btree_ptr_is_null(cur
, &rptr
) ||
3889 !xfs_btree_ptr_is_null(cur
, &lptr
));
3892 * Duplicate the cursor so our btree manipulations here won't
3893 * disrupt the next level up.
3895 error
= xfs_btree_dup_cursor(cur
, &tcur
);
3900 * If there's a right sibling, see if it's ok to shift an entry
3903 if (!xfs_btree_ptr_is_null(cur
, &rptr
)) {
3905 * Move the temp cursor to the last entry in the next block.
3906 * Actually any entry but the first would suffice.
3908 i
= xfs_btree_lastrec(tcur
, level
);
3909 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3911 error
= xfs_btree_increment(tcur
, level
, &i
);
3914 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3916 i
= xfs_btree_lastrec(tcur
, level
);
3917 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3919 /* Grab a pointer to the block. */
3920 right
= xfs_btree_get_block(tcur
, level
, &rbp
);
3922 error
= xfs_btree_check_block(tcur
, right
, level
, rbp
);
3926 /* Grab the current block number, for future use. */
3927 xfs_btree_get_sibling(tcur
, right
, &cptr
, XFS_BB_LEFTSIB
);
3930 * If right block is full enough so that removing one entry
3931 * won't make it too empty, and left-shifting an entry out
3932 * of right to us works, we're done.
3934 if (xfs_btree_get_numrecs(right
) - 1 >=
3935 cur
->bc_ops
->get_minrecs(tcur
, level
)) {
3936 error
= xfs_btree_lshift(tcur
, level
, &i
);
3940 ASSERT(xfs_btree_get_numrecs(block
) >=
3941 cur
->bc_ops
->get_minrecs(tcur
, level
));
3943 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
3946 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
3954 * Otherwise, grab the number of records in right for
3955 * future reference, and fix up the temp cursor to point
3956 * to our block again (last record).
3958 rrecs
= xfs_btree_get_numrecs(right
);
3959 if (!xfs_btree_ptr_is_null(cur
, &lptr
)) {
3960 i
= xfs_btree_firstrec(tcur
, level
);
3961 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3963 error
= xfs_btree_decrement(tcur
, level
, &i
);
3966 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3971 * If there's a left sibling, see if it's ok to shift an entry
3974 if (!xfs_btree_ptr_is_null(cur
, &lptr
)) {
3976 * Move the temp cursor to the first entry in the
3979 i
= xfs_btree_firstrec(tcur
, level
);
3980 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3982 error
= xfs_btree_decrement(tcur
, level
, &i
);
3985 i
= xfs_btree_firstrec(tcur
, level
);
3986 XFS_WANT_CORRUPTED_GOTO(cur
->bc_mp
, i
== 1, error0
);
3988 /* Grab a pointer to the block. */
3989 left
= xfs_btree_get_block(tcur
, level
, &lbp
);
3991 error
= xfs_btree_check_block(cur
, left
, level
, lbp
);
3995 /* Grab the current block number, for future use. */
3996 xfs_btree_get_sibling(tcur
, left
, &cptr
, XFS_BB_RIGHTSIB
);
3999 * If left block is full enough so that removing one entry
4000 * won't make it too empty, and right-shifting an entry out
4001 * of left to us works, we're done.
4003 if (xfs_btree_get_numrecs(left
) - 1 >=
4004 cur
->bc_ops
->get_minrecs(tcur
, level
)) {
4005 error
= xfs_btree_rshift(tcur
, level
, &i
);
4009 ASSERT(xfs_btree_get_numrecs(block
) >=
4010 cur
->bc_ops
->get_minrecs(tcur
, level
));
4011 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
4015 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
4022 * Otherwise, grab the number of records in right for
4025 lrecs
= xfs_btree_get_numrecs(left
);
4028 /* Delete the temp cursor, we're done with it. */
4029 xfs_btree_del_cursor(tcur
, XFS_BTREE_NOERROR
);
4032 /* If here, we need to do a join to keep the tree balanced. */
4033 ASSERT(!xfs_btree_ptr_is_null(cur
, &cptr
));
4035 if (!xfs_btree_ptr_is_null(cur
, &lptr
) &&
4036 lrecs
+ xfs_btree_get_numrecs(block
) <=
4037 cur
->bc_ops
->get_maxrecs(cur
, level
)) {
4039 * Set "right" to be the starting block,
4040 * "left" to be the left neighbor.
4045 error
= xfs_btree_read_buf_block(cur
, &lptr
, 0, &left
, &lbp
);
4050 * If that won't work, see if we can join with the right neighbor block.
4052 } else if (!xfs_btree_ptr_is_null(cur
, &rptr
) &&
4053 rrecs
+ xfs_btree_get_numrecs(block
) <=
4054 cur
->bc_ops
->get_maxrecs(cur
, level
)) {
4056 * Set "left" to be the starting block,
4057 * "right" to be the right neighbor.
4062 error
= xfs_btree_read_buf_block(cur
, &rptr
, 0, &right
, &rbp
);
4067 * Otherwise, we can't fix the imbalance.
4068 * Just return. This is probably a logic error, but it's not fatal.
4071 error
= xfs_btree_dec_cursor(cur
, level
, stat
);
4077 rrecs
= xfs_btree_get_numrecs(right
);
4078 lrecs
= xfs_btree_get_numrecs(left
);
4081 * We're now going to join "left" and "right" by moving all the stuff
4082 * in "right" to "left" and deleting "right".
4084 XFS_BTREE_STATS_ADD(cur
, moves
, rrecs
);
4086 /* It's a non-leaf. Move keys and pointers. */
4087 union xfs_btree_key
*lkp
; /* left btree key */
4088 union xfs_btree_ptr
*lpp
; /* left address pointer */
4089 union xfs_btree_key
*rkp
; /* right btree key */
4090 union xfs_btree_ptr
*rpp
; /* right address pointer */
4092 lkp
= xfs_btree_key_addr(cur
, lrecs
+ 1, left
);
4093 lpp
= xfs_btree_ptr_addr(cur
, lrecs
+ 1, left
);
4094 rkp
= xfs_btree_key_addr(cur
, 1, right
);
4095 rpp
= xfs_btree_ptr_addr(cur
, 1, right
);
4097 for (i
= 1; i
< rrecs
; i
++) {
4098 error
= xfs_btree_check_ptr(cur
, rpp
, i
, level
);
4103 xfs_btree_copy_keys(cur
, lkp
, rkp
, rrecs
);
4104 xfs_btree_copy_ptrs(cur
, lpp
, rpp
, rrecs
);
4106 xfs_btree_log_keys(cur
, lbp
, lrecs
+ 1, lrecs
+ rrecs
);
4107 xfs_btree_log_ptrs(cur
, lbp
, lrecs
+ 1, lrecs
+ rrecs
);
4109 /* It's a leaf. Move records. */
4110 union xfs_btree_rec
*lrp
; /* left record pointer */
4111 union xfs_btree_rec
*rrp
; /* right record pointer */
4113 lrp
= xfs_btree_rec_addr(cur
, lrecs
+ 1, left
);
4114 rrp
= xfs_btree_rec_addr(cur
, 1, right
);
4116 xfs_btree_copy_recs(cur
, lrp
, rrp
, rrecs
);
4117 xfs_btree_log_recs(cur
, lbp
, lrecs
+ 1, lrecs
+ rrecs
);
4120 XFS_BTREE_STATS_INC(cur
, join
);
4123 * Fix up the number of records and right block pointer in the
4124 * surviving block, and log it.
4126 xfs_btree_set_numrecs(left
, lrecs
+ rrecs
);
4127 xfs_btree_get_sibling(cur
, right
, &cptr
, XFS_BB_RIGHTSIB
),
4128 xfs_btree_set_sibling(cur
, left
, &cptr
, XFS_BB_RIGHTSIB
);
4129 xfs_btree_log_block(cur
, lbp
, XFS_BB_NUMRECS
| XFS_BB_RIGHTSIB
);
4131 /* If there is a right sibling, point it to the remaining block. */
4132 xfs_btree_get_sibling(cur
, left
, &cptr
, XFS_BB_RIGHTSIB
);
4133 if (!xfs_btree_ptr_is_null(cur
, &cptr
)) {
4134 error
= xfs_btree_read_buf_block(cur
, &cptr
, 0, &rrblock
, &rrbp
);
4137 xfs_btree_set_sibling(cur
, rrblock
, &lptr
, XFS_BB_LEFTSIB
);
4138 xfs_btree_log_block(cur
, rrbp
, XFS_BB_LEFTSIB
);
4141 /* Free the deleted block. */
4142 error
= xfs_btree_free_block(cur
, rbp
);
4147 * If we joined with the left neighbor, set the buffer in the
4148 * cursor to the left block, and fix up the index.
4151 cur
->bc_bufs
[level
] = lbp
;
4152 cur
->bc_ptrs
[level
] += lrecs
;
4153 cur
->bc_ra
[level
] = 0;
4156 * If we joined with the right neighbor and there's a level above
4157 * us, increment the cursor at that level.
4159 else if ((cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
) ||
4160 (level
+ 1 < cur
->bc_nlevels
)) {
4161 error
= xfs_btree_increment(cur
, level
+ 1, &i
);
4167 * Readjust the ptr at this level if it's not a leaf, since it's
4168 * still pointing at the deletion point, which makes the cursor
4169 * inconsistent. If this makes the ptr 0, the caller fixes it up.
4170 * We can't use decrement because it would change the next level up.
4173 cur
->bc_ptrs
[level
]--;
4176 * We combined blocks, so we have to update the parent keys if the
4177 * btree supports overlapped intervals. However, bc_ptrs[level + 1]
4178 * points to the old block so that the caller knows which record to
4179 * delete. Therefore, the caller must be savvy enough to call updkeys
4180 * for us if we return stat == 2. The other exit points from this
4181 * function don't require deletions further up the tree, so they can
4182 * call updkeys directly.
4185 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
4186 /* Return value means the next level up has something to do. */
4191 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
4193 xfs_btree_del_cursor(tcur
, XFS_BTREE_ERROR
);
4198 * Delete the record pointed to by cur.
4199 * The cursor refers to the place where the record was (could be inserted)
4200 * when the operation returns.
4204 struct xfs_btree_cur
*cur
,
4205 int *stat
) /* success/failure */
4207 int error
; /* error return value */
4210 bool joined
= false;
4212 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ENTRY
);
4215 * Go up the tree, starting at leaf level.
4217 * If 2 is returned then a join was done; go to the next level.
4218 * Otherwise we are done.
4220 for (level
= 0, i
= 2; i
== 2; level
++) {
4221 error
= xfs_btree_delrec(cur
, level
, &i
);
4229 * If we combined blocks as part of deleting the record, delrec won't
4230 * have updated the parent high keys so we have to do that here.
4232 if (joined
&& (cur
->bc_flags
& XFS_BTREE_OVERLAPPING
)) {
4233 error
= xfs_btree_updkeys_force(cur
, 0);
4239 for (level
= 1; level
< cur
->bc_nlevels
; level
++) {
4240 if (cur
->bc_ptrs
[level
] == 0) {
4241 error
= xfs_btree_decrement(cur
, level
, &i
);
4249 XFS_BTREE_TRACE_CURSOR(cur
, XBT_EXIT
);
4253 XFS_BTREE_TRACE_CURSOR(cur
, XBT_ERROR
);
4258 * Get the data from the pointed-to record.
4262 struct xfs_btree_cur
*cur
, /* btree cursor */
4263 union xfs_btree_rec
**recp
, /* output: btree record */
4264 int *stat
) /* output: success/failure */
4266 struct xfs_btree_block
*block
; /* btree block */
4267 struct xfs_buf
*bp
; /* buffer pointer */
4268 int ptr
; /* record number */
4270 int error
; /* error return value */
4273 ptr
= cur
->bc_ptrs
[0];
4274 block
= xfs_btree_get_block(cur
, 0, &bp
);
4277 error
= xfs_btree_check_block(cur
, block
, 0, bp
);
4283 * Off the right end or left end, return failure.
4285 if (ptr
> xfs_btree_get_numrecs(block
) || ptr
<= 0) {
4291 * Point to the record and extract its data.
4293 *recp
= xfs_btree_rec_addr(cur
, ptr
, block
);
4298 /* Visit a block in a btree. */
4300 xfs_btree_visit_block(
4301 struct xfs_btree_cur
*cur
,
4303 xfs_btree_visit_blocks_fn fn
,
4306 struct xfs_btree_block
*block
;
4308 union xfs_btree_ptr rptr
;
4311 /* do right sibling readahead */
4312 xfs_btree_readahead(cur
, level
, XFS_BTCUR_RIGHTRA
);
4313 block
= xfs_btree_get_block(cur
, level
, &bp
);
4315 /* process the block */
4316 error
= fn(cur
, level
, data
);
4320 /* now read rh sibling block for next iteration */
4321 xfs_btree_get_sibling(cur
, block
, &rptr
, XFS_BB_RIGHTSIB
);
4322 if (xfs_btree_ptr_is_null(cur
, &rptr
))
4325 return xfs_btree_lookup_get_block(cur
, level
, &rptr
, &block
);
4329 /* Visit every block in a btree. */
4331 xfs_btree_visit_blocks(
4332 struct xfs_btree_cur
*cur
,
4333 xfs_btree_visit_blocks_fn fn
,
4336 union xfs_btree_ptr lptr
;
4338 struct xfs_btree_block
*block
= NULL
;
4341 cur
->bc_ops
->init_ptr_from_cur(cur
, &lptr
);
4343 /* for each level */
4344 for (level
= cur
->bc_nlevels
- 1; level
>= 0; level
--) {
4345 /* grab the left hand block */
4346 error
= xfs_btree_lookup_get_block(cur
, level
, &lptr
, &block
);
4350 /* readahead the left most block for the next level down */
4352 union xfs_btree_ptr
*ptr
;
4354 ptr
= xfs_btree_ptr_addr(cur
, 1, block
);
4355 xfs_btree_readahead_ptr(cur
, ptr
, 1);
4357 /* save for the next iteration of the loop */
4361 /* for each buffer in the level */
4363 error
= xfs_btree_visit_block(cur
, level
, fn
, data
);
4366 if (error
!= -ENOENT
)
4374 * Change the owner of a btree.
4376 * The mechanism we use here is ordered buffer logging. Because we don't know
4377 * how many buffers were are going to need to modify, we don't really want to
4378 * have to make transaction reservations for the worst case of every buffer in a
4379 * full size btree as that may be more space that we can fit in the log....
4381 * We do the btree walk in the most optimal manner possible - we have sibling
4382 * pointers so we can just walk all the blocks on each level from left to right
4383 * in a single pass, and then move to the next level and do the same. We can
4384 * also do readahead on the sibling pointers to get IO moving more quickly,
4385 * though for slow disks this is unlikely to make much difference to performance
4386 * as the amount of CPU work we have to do before moving to the next block is
4389 * For each btree block that we load, modify the owner appropriately, set the
4390 * buffer as an ordered buffer and log it appropriately. We need to ensure that
4391 * we mark the region we change dirty so that if the buffer is relogged in
4392 * a subsequent transaction the changes we make here as an ordered buffer are
4393 * correctly relogged in that transaction. If we are in recovery context, then
4394 * just queue the modified buffer as delayed write buffer so the transaction
4395 * recovery completion writes the changes to disk.
4397 struct xfs_btree_block_change_owner_info
{
4398 __uint64_t new_owner
;
4399 struct list_head
*buffer_list
;
4403 xfs_btree_block_change_owner(
4404 struct xfs_btree_cur
*cur
,
4408 struct xfs_btree_block_change_owner_info
*bbcoi
= data
;
4409 struct xfs_btree_block
*block
;
4412 /* modify the owner */
4413 block
= xfs_btree_get_block(cur
, level
, &bp
);
4414 if (cur
->bc_flags
& XFS_BTREE_LONG_PTRS
)
4415 block
->bb_u
.l
.bb_owner
= cpu_to_be64(bbcoi
->new_owner
);
4417 block
->bb_u
.s
.bb_owner
= cpu_to_be32(bbcoi
->new_owner
);
4420 * If the block is a root block hosted in an inode, we might not have a
4421 * buffer pointer here and we shouldn't attempt to log the change as the
4422 * information is already held in the inode and discarded when the root
4423 * block is formatted into the on-disk inode fork. We still change it,
4424 * though, so everything is consistent in memory.
4428 xfs_trans_ordered_buf(cur
->bc_tp
, bp
);
4429 xfs_btree_log_block(cur
, bp
, XFS_BB_OWNER
);
4431 xfs_buf_delwri_queue(bp
, bbcoi
->buffer_list
);
4434 ASSERT(cur
->bc_flags
& XFS_BTREE_ROOT_IN_INODE
);
4435 ASSERT(level
== cur
->bc_nlevels
- 1);
4442 xfs_btree_change_owner(
4443 struct xfs_btree_cur
*cur
,
4444 __uint64_t new_owner
,
4445 struct list_head
*buffer_list
)
4447 struct xfs_btree_block_change_owner_info bbcoi
;
4449 bbcoi
.new_owner
= new_owner
;
4450 bbcoi
.buffer_list
= buffer_list
;
4452 return xfs_btree_visit_blocks(cur
, xfs_btree_block_change_owner
,
4457 * xfs_btree_sblock_v5hdr_verify() -- verify the v5 fields of a short-format
4460 * @bp: buffer containing the btree block
4461 * @max_recs: pointer to the m_*_mxr max records field in the xfs mount
4462 * @pag_max_level: pointer to the per-ag max level field
4465 xfs_btree_sblock_v5hdr_verify(
4468 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
4469 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
4470 struct xfs_perag
*pag
= bp
->b_pag
;
4472 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
4474 if (!uuid_equal(&block
->bb_u
.s
.bb_uuid
, &mp
->m_sb
.sb_meta_uuid
))
4476 if (block
->bb_u
.s
.bb_blkno
!= cpu_to_be64(bp
->b_bn
))
4478 if (pag
&& be32_to_cpu(block
->bb_u
.s
.bb_owner
) != pag
->pag_agno
)
4484 * xfs_btree_sblock_verify() -- verify a short-format btree block
4486 * @bp: buffer containing the btree block
4487 * @max_recs: maximum records allowed in this btree node
4490 xfs_btree_sblock_verify(
4492 unsigned int max_recs
)
4494 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
4495 struct xfs_btree_block
*block
= XFS_BUF_TO_BLOCK(bp
);
4497 /* numrecs verification */
4498 if (be16_to_cpu(block
->bb_numrecs
) > max_recs
)
4501 /* sibling pointer verification */
4502 if (!block
->bb_u
.s
.bb_leftsib
||
4503 (be32_to_cpu(block
->bb_u
.s
.bb_leftsib
) >= mp
->m_sb
.sb_agblocks
&&
4504 block
->bb_u
.s
.bb_leftsib
!= cpu_to_be32(NULLAGBLOCK
)))
4506 if (!block
->bb_u
.s
.bb_rightsib
||
4507 (be32_to_cpu(block
->bb_u
.s
.bb_rightsib
) >= mp
->m_sb
.sb_agblocks
&&
4508 block
->bb_u
.s
.bb_rightsib
!= cpu_to_be32(NULLAGBLOCK
)))
4515 * Calculate the number of btree levels needed to store a given number of
4516 * records in a short-format btree.
4519 xfs_btree_compute_maxlevels(
4520 struct xfs_mount
*mp
,
4525 unsigned long maxblocks
;
4527 maxblocks
= (len
+ limits
[0] - 1) / limits
[0];
4528 for (level
= 1; maxblocks
> 1; level
++)
4529 maxblocks
= (maxblocks
+ limits
[1] - 1) / limits
[1];
4534 * Query a regular btree for all records overlapping a given interval.
4535 * Start with a LE lookup of the key of low_rec and return all records
4536 * until we find a record with a key greater than the key of high_rec.
4539 xfs_btree_simple_query_range(
4540 struct xfs_btree_cur
*cur
,
4541 union xfs_btree_key
*low_key
,
4542 union xfs_btree_key
*high_key
,
4543 xfs_btree_query_range_fn fn
,
4546 union xfs_btree_rec
*recp
;
4547 union xfs_btree_key rec_key
;
4550 bool firstrec
= true;
4553 ASSERT(cur
->bc_ops
->init_high_key_from_rec
);
4554 ASSERT(cur
->bc_ops
->diff_two_keys
);
4557 * Find the leftmost record. The btree cursor must be set
4558 * to the low record used to generate low_key.
4561 error
= xfs_btree_lookup(cur
, XFS_LOOKUP_LE
, &stat
);
4565 /* Nothing? See if there's anything to the right. */
4567 error
= xfs_btree_increment(cur
, 0, &stat
);
4573 /* Find the record. */
4574 error
= xfs_btree_get_rec(cur
, &recp
, &stat
);
4578 /* Skip if high_key(rec) < low_key. */
4580 cur
->bc_ops
->init_high_key_from_rec(&rec_key
, recp
);
4582 diff
= cur
->bc_ops
->diff_two_keys(cur
, low_key
,
4588 /* Stop if high_key < low_key(rec). */
4589 cur
->bc_ops
->init_key_from_rec(&rec_key
, recp
);
4590 diff
= cur
->bc_ops
->diff_two_keys(cur
, &rec_key
, high_key
);
4595 error
= fn(cur
, recp
, priv
);
4596 if (error
< 0 || error
== XFS_BTREE_QUERY_RANGE_ABORT
)
4600 /* Move on to the next record. */
4601 error
= xfs_btree_increment(cur
, 0, &stat
);
4611 * Query an overlapped interval btree for all records overlapping a given
4612 * interval. This function roughly follows the algorithm given in
4613 * "Interval Trees" of _Introduction to Algorithms_, which is section
4614 * 14.3 in the 2nd and 3rd editions.
4616 * First, generate keys for the low and high records passed in.
4618 * For any leaf node, generate the high and low keys for the record.
4619 * If the record keys overlap with the query low/high keys, pass the
4620 * record to the function iterator.
4622 * For any internal node, compare the low and high keys of each
4623 * pointer against the query low/high keys. If there's an overlap,
4624 * follow the pointer.
4626 * As an optimization, we stop scanning a block when we find a low key
4627 * that is greater than the query's high key.
4630 xfs_btree_overlapped_query_range(
4631 struct xfs_btree_cur
*cur
,
4632 union xfs_btree_key
*low_key
,
4633 union xfs_btree_key
*high_key
,
4634 xfs_btree_query_range_fn fn
,
4637 union xfs_btree_ptr ptr
;
4638 union xfs_btree_ptr
*pp
;
4639 union xfs_btree_key rec_key
;
4640 union xfs_btree_key rec_hkey
;
4641 union xfs_btree_key
*lkp
;
4642 union xfs_btree_key
*hkp
;
4643 union xfs_btree_rec
*recp
;
4644 struct xfs_btree_block
*block
;
4652 /* Load the root of the btree. */
4653 level
= cur
->bc_nlevels
- 1;
4654 cur
->bc_ops
->init_ptr_from_cur(cur
, &ptr
);
4655 error
= xfs_btree_lookup_get_block(cur
, level
, &ptr
, &block
);
4658 xfs_btree_get_block(cur
, level
, &bp
);
4659 trace_xfs_btree_overlapped_query_range(cur
, level
, bp
);
4661 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
4665 cur
->bc_ptrs
[level
] = 1;
4667 while (level
< cur
->bc_nlevels
) {
4668 block
= xfs_btree_get_block(cur
, level
, &bp
);
4670 /* End of node, pop back towards the root. */
4671 if (cur
->bc_ptrs
[level
] > be16_to_cpu(block
->bb_numrecs
)) {
4673 if (level
< cur
->bc_nlevels
- 1)
4674 cur
->bc_ptrs
[level
+ 1]++;
4680 /* Handle a leaf node. */
4681 recp
= xfs_btree_rec_addr(cur
, cur
->bc_ptrs
[0], block
);
4683 cur
->bc_ops
->init_high_key_from_rec(&rec_hkey
, recp
);
4684 ldiff
= cur
->bc_ops
->diff_two_keys(cur
, &rec_hkey
,
4687 cur
->bc_ops
->init_key_from_rec(&rec_key
, recp
);
4688 hdiff
= cur
->bc_ops
->diff_two_keys(cur
, high_key
,
4692 * If (record's high key >= query's low key) and
4693 * (query's high key >= record's low key), then
4694 * this record overlaps the query range; callback.
4696 if (ldiff
>= 0 && hdiff
>= 0) {
4697 error
= fn(cur
, recp
, priv
);
4699 error
== XFS_BTREE_QUERY_RANGE_ABORT
)
4701 } else if (hdiff
< 0) {
4702 /* Record is larger than high key; pop. */
4705 cur
->bc_ptrs
[level
]++;
4709 /* Handle an internal node. */
4710 lkp
= xfs_btree_key_addr(cur
, cur
->bc_ptrs
[level
], block
);
4711 hkp
= xfs_btree_high_key_addr(cur
, cur
->bc_ptrs
[level
], block
);
4712 pp
= xfs_btree_ptr_addr(cur
, cur
->bc_ptrs
[level
], block
);
4714 ldiff
= cur
->bc_ops
->diff_two_keys(cur
, hkp
, low_key
);
4715 hdiff
= cur
->bc_ops
->diff_two_keys(cur
, high_key
, lkp
);
4718 * If (pointer's high key >= query's low key) and
4719 * (query's high key >= pointer's low key), then
4720 * this record overlaps the query range; follow pointer.
4722 if (ldiff
>= 0 && hdiff
>= 0) {
4724 error
= xfs_btree_lookup_get_block(cur
, level
, pp
,
4728 xfs_btree_get_block(cur
, level
, &bp
);
4729 trace_xfs_btree_overlapped_query_range(cur
, level
, bp
);
4731 error
= xfs_btree_check_block(cur
, block
, level
, bp
);
4735 cur
->bc_ptrs
[level
] = 1;
4737 } else if (hdiff
< 0) {
4738 /* The low key is larger than the upper range; pop. */
4741 cur
->bc_ptrs
[level
]++;
4746 * If we don't end this function with the cursor pointing at a record
4747 * block, a subsequent non-error cursor deletion will not release
4748 * node-level buffers, causing a buffer leak. This is quite possible
4749 * with a zero-results range query, so release the buffers if we
4750 * failed to return any results.
4752 if (cur
->bc_bufs
[0] == NULL
) {
4753 for (i
= 0; i
< cur
->bc_nlevels
; i
++) {
4754 if (cur
->bc_bufs
[i
]) {
4755 xfs_trans_brelse(cur
->bc_tp
, cur
->bc_bufs
[i
]);
4756 cur
->bc_bufs
[i
] = NULL
;
4757 cur
->bc_ptrs
[i
] = 0;
4767 * Query a btree for all records overlapping a given interval of keys. The
4768 * supplied function will be called with each record found; return one of the
4769 * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
4770 * code. This function returns XFS_BTREE_QUERY_RANGE_ABORT, zero, or a
4771 * negative error code.
4774 xfs_btree_query_range(
4775 struct xfs_btree_cur
*cur
,
4776 union xfs_btree_irec
*low_rec
,
4777 union xfs_btree_irec
*high_rec
,
4778 xfs_btree_query_range_fn fn
,
4781 union xfs_btree_rec rec
;
4782 union xfs_btree_key low_key
;
4783 union xfs_btree_key high_key
;
4785 /* Find the keys of both ends of the interval. */
4786 cur
->bc_rec
= *high_rec
;
4787 cur
->bc_ops
->init_rec_from_cur(cur
, &rec
);
4788 cur
->bc_ops
->init_key_from_rec(&high_key
, &rec
);
4790 cur
->bc_rec
= *low_rec
;
4791 cur
->bc_ops
->init_rec_from_cur(cur
, &rec
);
4792 cur
->bc_ops
->init_key_from_rec(&low_key
, &rec
);
4794 /* Enforce low key < high key. */
4795 if (cur
->bc_ops
->diff_two_keys(cur
, &low_key
, &high_key
) > 0)
4798 if (!(cur
->bc_flags
& XFS_BTREE_OVERLAPPING
))
4799 return xfs_btree_simple_query_range(cur
, &low_key
,
4800 &high_key
, fn
, priv
);
4801 return xfs_btree_overlapped_query_range(cur
, &low_key
, &high_key
,
4806 * Calculate the number of blocks needed to store a given number of records
4807 * in a short-format (per-AG metadata) btree.
4810 xfs_btree_calc_size(
4811 struct xfs_mount
*mp
,
4813 unsigned long long len
)
4819 maxrecs
= limits
[0];
4820 for (level
= 0, rval
= 0; len
> 1; level
++) {
4822 do_div(len
, maxrecs
);
4823 maxrecs
= limits
[1];
4830 xfs_btree_count_blocks_helper(
4831 struct xfs_btree_cur
*cur
,
4835 xfs_extlen_t
*blocks
= data
;
4841 /* Count the blocks in a btree and return the result in *blocks. */
4843 xfs_btree_count_blocks(
4844 struct xfs_btree_cur
*cur
,
4845 xfs_extlen_t
*blocks
)
4848 return xfs_btree_visit_blocks(cur
, xfs_btree_count_blocks_helper
,