2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_types.h"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_dinode.h"
31 #include "xfs_inode.h"
33 #include "xfs_dir2_format.h"
35 #include "xfs_dir2_priv.h"
36 #include "xfs_error.h"
37 #include "xfs_trace.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_cksum.h"
42 * Local function declarations.
44 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t
*args
, struct xfs_buf
**lbpp
,
45 int *indexp
, struct xfs_buf
**dbpp
);
46 static void xfs_dir3_leaf_log_bests(struct xfs_trans
*tp
, struct xfs_buf
*bp
,
48 static void xfs_dir3_leaf_log_tail(struct xfs_trans
*tp
, struct xfs_buf
*bp
);
51 * Check the internal consistency of a leaf1 block.
52 * Pop an assert if something is wrong.
55 #define xfs_dir3_leaf_check(mp, bp) \
57 if (!xfs_dir3_leaf1_check((mp), (bp))) \
66 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
67 struct xfs_dir3_icleaf_hdr leafhdr
;
69 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
71 if (leafhdr
.magic
== XFS_DIR3_LEAF1_MAGIC
) {
72 struct xfs_dir3_leaf_hdr
*leaf3
= bp
->b_addr
;
73 if (be64_to_cpu(leaf3
->info
.blkno
) != bp
->b_bn
)
75 } else if (leafhdr
.magic
!= XFS_DIR2_LEAF1_MAGIC
)
78 return xfs_dir3_leaf_check_int(mp
, &leafhdr
, leaf
);
81 #define xfs_dir3_leaf_check(mp, bp)
85 xfs_dir3_leaf_hdr_from_disk(
86 struct xfs_dir3_icleaf_hdr
*to
,
87 struct xfs_dir2_leaf
*from
)
89 if (from
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
90 from
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
)) {
91 to
->forw
= be32_to_cpu(from
->hdr
.info
.forw
);
92 to
->back
= be32_to_cpu(from
->hdr
.info
.back
);
93 to
->magic
= be16_to_cpu(from
->hdr
.info
.magic
);
94 to
->count
= be16_to_cpu(from
->hdr
.count
);
95 to
->stale
= be16_to_cpu(from
->hdr
.stale
);
97 struct xfs_dir3_leaf_hdr
*hdr3
= (struct xfs_dir3_leaf_hdr
*)from
;
99 to
->forw
= be32_to_cpu(hdr3
->info
.hdr
.forw
);
100 to
->back
= be32_to_cpu(hdr3
->info
.hdr
.back
);
101 to
->magic
= be16_to_cpu(hdr3
->info
.hdr
.magic
);
102 to
->count
= be16_to_cpu(hdr3
->count
);
103 to
->stale
= be16_to_cpu(hdr3
->stale
);
106 ASSERT(to
->magic
== XFS_DIR2_LEAF1_MAGIC
||
107 to
->magic
== XFS_DIR3_LEAF1_MAGIC
||
108 to
->magic
== XFS_DIR2_LEAFN_MAGIC
||
109 to
->magic
== XFS_DIR3_LEAFN_MAGIC
);
113 xfs_dir3_leaf_hdr_to_disk(
114 struct xfs_dir2_leaf
*to
,
115 struct xfs_dir3_icleaf_hdr
*from
)
117 ASSERT(from
->magic
== XFS_DIR2_LEAF1_MAGIC
||
118 from
->magic
== XFS_DIR3_LEAF1_MAGIC
||
119 from
->magic
== XFS_DIR2_LEAFN_MAGIC
||
120 from
->magic
== XFS_DIR3_LEAFN_MAGIC
);
122 if (from
->magic
== XFS_DIR2_LEAF1_MAGIC
||
123 from
->magic
== XFS_DIR2_LEAFN_MAGIC
) {
124 to
->hdr
.info
.forw
= cpu_to_be32(from
->forw
);
125 to
->hdr
.info
.back
= cpu_to_be32(from
->back
);
126 to
->hdr
.info
.magic
= cpu_to_be16(from
->magic
);
127 to
->hdr
.count
= cpu_to_be16(from
->count
);
128 to
->hdr
.stale
= cpu_to_be16(from
->stale
);
130 struct xfs_dir3_leaf_hdr
*hdr3
= (struct xfs_dir3_leaf_hdr
*)to
;
132 hdr3
->info
.hdr
.forw
= cpu_to_be32(from
->forw
);
133 hdr3
->info
.hdr
.back
= cpu_to_be32(from
->back
);
134 hdr3
->info
.hdr
.magic
= cpu_to_be16(from
->magic
);
135 hdr3
->count
= cpu_to_be16(from
->count
);
136 hdr3
->stale
= cpu_to_be16(from
->stale
);
141 xfs_dir3_leaf_check_int(
142 struct xfs_mount
*mp
,
143 struct xfs_dir3_icleaf_hdr
*hdr
,
144 struct xfs_dir2_leaf
*leaf
)
146 struct xfs_dir2_leaf_entry
*ents
;
147 xfs_dir2_leaf_tail_t
*ltp
;
151 ents
= xfs_dir3_leaf_ents_p(leaf
);
152 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
155 * XXX (dgc): This value is not restrictive enough.
156 * Should factor in the size of the bests table as well.
157 * We can deduce a value for that from di_size.
159 if (hdr
->count
> xfs_dir3_max_leaf_ents(mp
, leaf
))
162 /* Leaves and bests don't overlap in leaf format. */
163 if ((hdr
->magic
== XFS_DIR2_LEAF1_MAGIC
||
164 hdr
->magic
== XFS_DIR3_LEAF1_MAGIC
) &&
165 (char *)&ents
[hdr
->count
] > (char *)xfs_dir2_leaf_bests_p(ltp
))
168 /* Check hash value order, count stale entries. */
169 for (i
= stale
= 0; i
< hdr
->count
; i
++) {
170 if (i
+ 1 < hdr
->count
) {
171 if (be32_to_cpu(ents
[i
].hashval
) >
172 be32_to_cpu(ents
[i
+ 1].hashval
))
175 if (ents
[i
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
178 if (hdr
->stale
!= stale
)
184 xfs_dir3_leaf_verify(
188 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
189 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
190 struct xfs_dir3_icleaf_hdr leafhdr
;
192 ASSERT(magic
== XFS_DIR2_LEAF1_MAGIC
|| magic
== XFS_DIR2_LEAFN_MAGIC
);
194 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
195 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
196 struct xfs_dir3_leaf_hdr
*leaf3
= bp
->b_addr
;
198 if ((magic
== XFS_DIR2_LEAF1_MAGIC
&&
199 leafhdr
.magic
!= XFS_DIR3_LEAF1_MAGIC
) ||
200 (magic
== XFS_DIR2_LEAFN_MAGIC
&&
201 leafhdr
.magic
!= XFS_DIR3_LEAFN_MAGIC
))
204 if (!uuid_equal(&leaf3
->info
.uuid
, &mp
->m_sb
.sb_uuid
))
206 if (be64_to_cpu(leaf3
->info
.blkno
) != bp
->b_bn
)
209 if (leafhdr
.magic
!= magic
)
212 return xfs_dir3_leaf_check_int(mp
, &leafhdr
, leaf
);
220 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
222 if ((xfs_sb_version_hascrc(&mp
->m_sb
) &&
223 !xfs_verify_cksum(bp
->b_addr
, BBTOB(bp
->b_length
),
224 XFS_DIR3_LEAF_CRC_OFF
)) ||
225 !xfs_dir3_leaf_verify(bp
, magic
)) {
226 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, bp
->b_addr
);
227 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
236 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
237 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
238 struct xfs_dir3_leaf_hdr
*hdr3
= bp
->b_addr
;
240 if (!xfs_dir3_leaf_verify(bp
, magic
)) {
241 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, bp
->b_addr
);
242 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
246 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
250 hdr3
->info
.lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
252 xfs_update_cksum(bp
->b_addr
, BBTOB(bp
->b_length
), XFS_DIR3_LEAF_CRC_OFF
);
256 xfs_dir3_leaf1_read_verify(
259 __read_verify(bp
, XFS_DIR2_LEAF1_MAGIC
);
263 xfs_dir3_leaf1_write_verify(
266 __write_verify(bp
, XFS_DIR2_LEAF1_MAGIC
);
270 xfs_dir3_leafn_read_verify(
273 __read_verify(bp
, XFS_DIR2_LEAFN_MAGIC
);
277 xfs_dir3_leafn_write_verify(
280 __write_verify(bp
, XFS_DIR2_LEAFN_MAGIC
);
283 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops
= {
284 .verify_read
= xfs_dir3_leaf1_read_verify
,
285 .verify_write
= xfs_dir3_leaf1_write_verify
,
288 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops
= {
289 .verify_read
= xfs_dir3_leafn_read_verify
,
290 .verify_write
= xfs_dir3_leafn_write_verify
,
295 struct xfs_trans
*tp
,
296 struct xfs_inode
*dp
,
298 xfs_daddr_t mappedbno
,
299 struct xfs_buf
**bpp
)
303 err
= xfs_da_read_buf(tp
, dp
, fbno
, mappedbno
, bpp
,
304 XFS_DATA_FORK
, &xfs_dir3_leaf1_buf_ops
);
306 xfs_trans_buf_set_type(tp
, *bpp
, XFS_BLFT_DIR_LEAF1_BUF
);
312 struct xfs_trans
*tp
,
313 struct xfs_inode
*dp
,
315 xfs_daddr_t mappedbno
,
316 struct xfs_buf
**bpp
)
320 err
= xfs_da_read_buf(tp
, dp
, fbno
, mappedbno
, bpp
,
321 XFS_DATA_FORK
, &xfs_dir3_leafn_buf_ops
);
323 xfs_trans_buf_set_type(tp
, *bpp
, XFS_BLFT_DIR_LEAFN_BUF
);
328 * Initialize a new leaf block, leaf1 or leafn magic accepted.
332 struct xfs_mount
*mp
,
333 struct xfs_trans
*tp
,
338 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
340 ASSERT(type
== XFS_DIR2_LEAF1_MAGIC
|| type
== XFS_DIR2_LEAFN_MAGIC
);
342 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
343 struct xfs_dir3_leaf_hdr
*leaf3
= bp
->b_addr
;
345 memset(leaf3
, 0, sizeof(*leaf3
));
347 leaf3
->info
.hdr
.magic
= (type
== XFS_DIR2_LEAF1_MAGIC
)
348 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC
)
349 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
);
350 leaf3
->info
.blkno
= cpu_to_be64(bp
->b_bn
);
351 leaf3
->info
.owner
= cpu_to_be64(owner
);
352 uuid_copy(&leaf3
->info
.uuid
, &mp
->m_sb
.sb_uuid
);
354 memset(leaf
, 0, sizeof(*leaf
));
355 leaf
->hdr
.info
.magic
= cpu_to_be16(type
);
359 * If it's a leaf-format directory initialize the tail.
360 * Caller is responsible for initialising the bests table.
362 if (type
== XFS_DIR2_LEAF1_MAGIC
) {
363 struct xfs_dir2_leaf_tail
*ltp
;
365 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
367 bp
->b_ops
= &xfs_dir3_leaf1_buf_ops
;
368 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DIR_LEAF1_BUF
);
370 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
371 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DIR_LEAFN_BUF
);
376 xfs_dir3_leaf_get_buf(
379 struct xfs_buf
**bpp
,
382 struct xfs_inode
*dp
= args
->dp
;
383 struct xfs_trans
*tp
= args
->trans
;
384 struct xfs_mount
*mp
= dp
->i_mount
;
388 ASSERT(magic
== XFS_DIR2_LEAF1_MAGIC
|| magic
== XFS_DIR2_LEAFN_MAGIC
);
389 ASSERT(bno
>= XFS_DIR2_LEAF_FIRSTDB(mp
) &&
390 bno
< XFS_DIR2_FREE_FIRSTDB(mp
));
392 error
= xfs_da_get_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, bno
), -1, &bp
,
397 xfs_dir3_leaf_init(mp
, tp
, bp
, dp
->i_ino
, magic
);
398 xfs_dir3_leaf_log_header(tp
, bp
);
399 if (magic
== XFS_DIR2_LEAF1_MAGIC
)
400 xfs_dir3_leaf_log_tail(tp
, bp
);
406 * Convert a block form directory to a leaf form directory.
409 xfs_dir2_block_to_leaf(
410 xfs_da_args_t
*args
, /* operation arguments */
411 struct xfs_buf
*dbp
) /* input block's buffer */
413 __be16
*bestsp
; /* leaf's bestsp entries */
414 xfs_dablk_t blkno
; /* leaf block's bno */
415 xfs_dir2_data_hdr_t
*hdr
; /* block header */
416 xfs_dir2_leaf_entry_t
*blp
; /* block's leaf entries */
417 xfs_dir2_block_tail_t
*btp
; /* block's tail */
418 xfs_inode_t
*dp
; /* incore directory inode */
419 int error
; /* error return code */
420 struct xfs_buf
*lbp
; /* leaf block's buffer */
421 xfs_dir2_db_t ldb
; /* leaf block's bno */
422 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
423 xfs_dir2_leaf_tail_t
*ltp
; /* leaf's tail */
424 xfs_mount_t
*mp
; /* filesystem mount point */
425 int needlog
; /* need to log block header */
426 int needscan
; /* need to rescan bestfree */
427 xfs_trans_t
*tp
; /* transaction pointer */
428 struct xfs_dir2_data_free
*bf
;
429 struct xfs_dir2_leaf_entry
*ents
;
430 struct xfs_dir3_icleaf_hdr leafhdr
;
432 trace_xfs_dir2_block_to_leaf(args
);
438 * Add the leaf block to the inode.
439 * This interface will only put blocks in the leaf/node range.
440 * Since that's empty now, we'll get the root (block 0 in range).
442 if ((error
= xfs_da_grow_inode(args
, &blkno
))) {
445 ldb
= xfs_dir2_da_to_db(mp
, blkno
);
446 ASSERT(ldb
== XFS_DIR2_LEAF_FIRSTDB(mp
));
448 * Initialize the leaf block, get a buffer for it.
450 error
= xfs_dir3_leaf_get_buf(args
, ldb
, &lbp
, XFS_DIR2_LEAF1_MAGIC
);
456 xfs_dir3_data_check(dp
, dbp
);
457 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
458 blp
= xfs_dir2_block_leaf_p(btp
);
459 bf
= xfs_dir3_data_bestfree_p(hdr
);
460 ents
= xfs_dir3_leaf_ents_p(leaf
);
463 * Set the counts in the leaf header.
465 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
466 leafhdr
.count
= be32_to_cpu(btp
->count
);
467 leafhdr
.stale
= be32_to_cpu(btp
->stale
);
468 xfs_dir3_leaf_hdr_to_disk(leaf
, &leafhdr
);
469 xfs_dir3_leaf_log_header(tp
, lbp
);
472 * Could compact these but I think we always do the conversion
473 * after squeezing out stale entries.
475 memcpy(ents
, blp
, be32_to_cpu(btp
->count
) * sizeof(xfs_dir2_leaf_entry_t
));
476 xfs_dir3_leaf_log_ents(tp
, lbp
, 0, leafhdr
.count
- 1);
480 * Make the space formerly occupied by the leaf entries and block
483 xfs_dir2_data_make_free(tp
, dbp
,
484 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)hdr
),
485 (xfs_dir2_data_aoff_t
)((char *)hdr
+ mp
->m_dirblksize
-
487 &needlog
, &needscan
);
489 * Fix up the block header, make it a data block.
491 dbp
->b_ops
= &xfs_dir3_data_buf_ops
;
492 xfs_trans_buf_set_type(tp
, dbp
, XFS_BLFT_DIR_DATA_BUF
);
493 if (hdr
->magic
== cpu_to_be32(XFS_DIR2_BLOCK_MAGIC
))
494 hdr
->magic
= cpu_to_be32(XFS_DIR2_DATA_MAGIC
);
496 hdr
->magic
= cpu_to_be32(XFS_DIR3_DATA_MAGIC
);
499 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
501 * Set up leaf tail and bests table.
503 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
504 ltp
->bestcount
= cpu_to_be32(1);
505 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
506 bestsp
[0] = bf
[0].length
;
508 * Log the data header and leaf bests table.
511 xfs_dir2_data_log_header(tp
, dbp
);
512 xfs_dir3_leaf_check(mp
, lbp
);
513 xfs_dir3_data_check(dp
, dbp
);
514 xfs_dir3_leaf_log_bests(tp
, lbp
, 0, 0);
519 xfs_dir3_leaf_find_stale(
520 struct xfs_dir3_icleaf_hdr
*leafhdr
,
521 struct xfs_dir2_leaf_entry
*ents
,
527 * Find the first stale entry before our index, if any.
529 for (*lowstale
= index
- 1; *lowstale
>= 0; --*lowstale
) {
530 if (ents
[*lowstale
].address
==
531 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
536 * Find the first stale entry at or after our index, if any.
537 * Stop if the result would require moving more entries than using
540 for (*highstale
= index
; *highstale
< leafhdr
->count
; ++*highstale
) {
541 if (ents
[*highstale
].address
==
542 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
544 if (*lowstale
>= 0 && index
- *lowstale
<= *highstale
- index
)
549 struct xfs_dir2_leaf_entry
*
550 xfs_dir3_leaf_find_entry(
551 struct xfs_dir3_icleaf_hdr
*leafhdr
,
552 struct xfs_dir2_leaf_entry
*ents
,
553 int index
, /* leaf table position */
554 int compact
, /* need to compact leaves */
555 int lowstale
, /* index of prev stale leaf */
556 int highstale
, /* index of next stale leaf */
557 int *lfloglow
, /* low leaf logging index */
558 int *lfloghigh
) /* high leaf logging index */
560 if (!leafhdr
->stale
) {
561 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry table pointer */
564 * Now we need to make room to insert the leaf entry.
566 * If there are no stale entries, just insert a hole at index.
569 if (index
< leafhdr
->count
)
570 memmove(lep
+ 1, lep
,
571 (leafhdr
->count
- index
) * sizeof(*lep
));
574 * Record low and high logging indices for the leaf.
577 *lfloghigh
= leafhdr
->count
++;
582 * There are stale entries.
584 * We will use one of them for the new entry. It's probably not at
585 * the right location, so we'll have to shift some up or down first.
587 * If we didn't compact before, we need to find the nearest stale
588 * entries before and after our insertion point.
591 xfs_dir3_leaf_find_stale(leafhdr
, ents
, index
,
592 &lowstale
, &highstale
);
595 * If the low one is better, use it.
598 (highstale
== leafhdr
->count
||
599 index
- lowstale
- 1 < highstale
- index
)) {
600 ASSERT(index
- lowstale
- 1 >= 0);
601 ASSERT(ents
[lowstale
].address
==
602 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
));
605 * Copy entries up to cover the stale entry and make room
608 if (index
- lowstale
- 1 > 0) {
609 memmove(&ents
[lowstale
], &ents
[lowstale
+ 1],
610 (index
- lowstale
- 1) *
611 sizeof(xfs_dir2_leaf_entry_t
));
613 *lfloglow
= MIN(lowstale
, *lfloglow
);
614 *lfloghigh
= MAX(index
- 1, *lfloghigh
);
616 return &ents
[index
- 1];
620 * The high one is better, so use that one.
622 ASSERT(highstale
- index
>= 0);
623 ASSERT(ents
[highstale
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
));
626 * Copy entries down to cover the stale entry and make room for the
629 if (highstale
- index
> 0) {
630 memmove(&ents
[index
+ 1], &ents
[index
],
631 (highstale
- index
) * sizeof(xfs_dir2_leaf_entry_t
));
633 *lfloglow
= MIN(index
, *lfloglow
);
634 *lfloghigh
= MAX(highstale
, *lfloghigh
);
640 * Add an entry to a leaf form directory.
643 xfs_dir2_leaf_addname(
644 xfs_da_args_t
*args
) /* operation arguments */
646 __be16
*bestsp
; /* freespace table in leaf */
647 int compact
; /* need to compact leaves */
648 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
649 struct xfs_buf
*dbp
; /* data block buffer */
650 xfs_dir2_data_entry_t
*dep
; /* data block entry */
651 xfs_inode_t
*dp
; /* incore directory inode */
652 xfs_dir2_data_unused_t
*dup
; /* data unused entry */
653 int error
; /* error return value */
654 int grown
; /* allocated new data block */
655 int highstale
; /* index of next stale leaf */
656 int i
; /* temporary, index */
657 int index
; /* leaf table position */
658 struct xfs_buf
*lbp
; /* leaf's buffer */
659 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
660 int length
; /* length of new entry */
661 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry table pointer */
662 int lfloglow
; /* low leaf logging index */
663 int lfloghigh
; /* high leaf logging index */
664 int lowstale
; /* index of prev stale leaf */
665 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail pointer */
666 xfs_mount_t
*mp
; /* filesystem mount point */
667 int needbytes
; /* leaf block bytes needed */
668 int needlog
; /* need to log data header */
669 int needscan
; /* need to rescan data free */
670 __be16
*tagp
; /* end of data entry */
671 xfs_trans_t
*tp
; /* transaction pointer */
672 xfs_dir2_db_t use_block
; /* data block number */
673 struct xfs_dir2_data_free
*bf
; /* bestfree table */
674 struct xfs_dir2_leaf_entry
*ents
;
675 struct xfs_dir3_icleaf_hdr leafhdr
;
677 trace_xfs_dir2_leaf_addname(args
);
683 error
= xfs_dir3_leaf_read(tp
, dp
, mp
->m_dirleafblk
, -1, &lbp
);
688 * Look up the entry by hash value and name.
689 * We know it's not there, our caller has already done a lookup.
690 * So the index is of the entry to insert in front of.
691 * But if there are dup hash values the index is of the first of those.
693 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
695 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
696 ents
= xfs_dir3_leaf_ents_p(leaf
);
697 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
698 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
699 length
= xfs_dir2_data_entsize(args
->namelen
);
702 * See if there are any entries with the same hash value
703 * and space in their block for the new entry.
704 * This is good because it puts multiple same-hash value entries
705 * in a data block, improving the lookup of those entries.
707 for (use_block
= -1, lep
= &ents
[index
];
708 index
< leafhdr
.count
&& be32_to_cpu(lep
->hashval
) == args
->hashval
;
710 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
712 i
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
713 ASSERT(i
< be32_to_cpu(ltp
->bestcount
));
714 ASSERT(bestsp
[i
] != cpu_to_be16(NULLDATAOFF
));
715 if (be16_to_cpu(bestsp
[i
]) >= length
) {
721 * Didn't find a block yet, linear search all the data blocks.
723 if (use_block
== -1) {
724 for (i
= 0; i
< be32_to_cpu(ltp
->bestcount
); i
++) {
726 * Remember a block we see that's missing.
728 if (bestsp
[i
] == cpu_to_be16(NULLDATAOFF
) &&
731 else if (be16_to_cpu(bestsp
[i
]) >= length
) {
738 * How many bytes do we need in the leaf block?
742 needbytes
+= sizeof(xfs_dir2_leaf_entry_t
);
744 needbytes
+= sizeof(xfs_dir2_data_off_t
);
747 * Now kill use_block if it refers to a missing block, so we
748 * can use it as an indication of allocation needed.
750 if (use_block
!= -1 && bestsp
[use_block
] == cpu_to_be16(NULLDATAOFF
))
753 * If we don't have enough free bytes but we can make enough
754 * by compacting out stale entries, we'll do that.
756 if ((char *)bestsp
- (char *)&ents
[leafhdr
.count
] < needbytes
&&
761 * Otherwise if we don't have enough free bytes we need to
762 * convert to node form.
764 else if ((char *)bestsp
- (char *)&ents
[leafhdr
.count
] < needbytes
) {
766 * Just checking or no space reservation, give up.
768 if ((args
->op_flags
& XFS_DA_OP_JUSTCHECK
) ||
770 xfs_trans_brelse(tp
, lbp
);
771 return XFS_ERROR(ENOSPC
);
774 * Convert to node form.
776 error
= xfs_dir2_leaf_to_node(args
, lbp
);
780 * Then add the new entry.
782 return xfs_dir2_node_addname(args
);
785 * Otherwise it will fit without compaction.
790 * If just checking, then it will fit unless we needed to allocate
793 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
794 xfs_trans_brelse(tp
, lbp
);
795 return use_block
== -1 ? XFS_ERROR(ENOSPC
) : 0;
798 * If no allocations are allowed, return now before we've
801 if (args
->total
== 0 && use_block
== -1) {
802 xfs_trans_brelse(tp
, lbp
);
803 return XFS_ERROR(ENOSPC
);
806 * Need to compact the leaf entries, removing stale ones.
807 * Leave one stale entry behind - the one closest to our
808 * insertion index - and we'll shift that one to our insertion
812 xfs_dir3_leaf_compact_x1(&leafhdr
, ents
, &index
, &lowstale
,
813 &highstale
, &lfloglow
, &lfloghigh
);
816 * There are stale entries, so we'll need log-low and log-high
817 * impossibly bad values later.
819 else if (leafhdr
.stale
) {
820 lfloglow
= leafhdr
.count
;
824 * If there was no data block space found, we need to allocate
827 if (use_block
== -1) {
829 * Add the new data block.
831 if ((error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
,
833 xfs_trans_brelse(tp
, lbp
);
837 * Initialize the block.
839 if ((error
= xfs_dir3_data_init(args
, use_block
, &dbp
))) {
840 xfs_trans_brelse(tp
, lbp
);
844 * If we're adding a new data block on the end we need to
845 * extend the bests table. Copy it up one entry.
847 if (use_block
>= be32_to_cpu(ltp
->bestcount
)) {
849 memmove(&bestsp
[0], &bestsp
[1],
850 be32_to_cpu(ltp
->bestcount
) * sizeof(bestsp
[0]));
851 be32_add_cpu(<p
->bestcount
, 1);
852 xfs_dir3_leaf_log_tail(tp
, lbp
);
853 xfs_dir3_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
856 * If we're filling in a previously empty block just log it.
859 xfs_dir3_leaf_log_bests(tp
, lbp
, use_block
, use_block
);
861 bf
= xfs_dir3_data_bestfree_p(hdr
);
862 bestsp
[use_block
] = bf
[0].length
;
866 * Already had space in some data block.
867 * Just read that one in.
869 error
= xfs_dir3_data_read(tp
, dp
,
870 xfs_dir2_db_to_da(mp
, use_block
),
873 xfs_trans_brelse(tp
, lbp
);
877 bf
= xfs_dir3_data_bestfree_p(hdr
);
881 * Point to the biggest freespace in our data block.
883 dup
= (xfs_dir2_data_unused_t
*)
884 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
885 ASSERT(be16_to_cpu(dup
->length
) >= length
);
886 needscan
= needlog
= 0;
888 * Mark the initial part of our freespace in use for the new entry.
890 xfs_dir2_data_use_free(tp
, dbp
, dup
,
891 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
), length
,
892 &needlog
, &needscan
);
894 * Initialize our new entry (at last).
896 dep
= (xfs_dir2_data_entry_t
*)dup
;
897 dep
->inumber
= cpu_to_be64(args
->inumber
);
898 dep
->namelen
= args
->namelen
;
899 memcpy(dep
->name
, args
->name
, dep
->namelen
);
900 tagp
= xfs_dir2_data_entry_tag_p(dep
);
901 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
903 * Need to scan fix up the bestfree table.
906 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
908 * Need to log the data block's header.
911 xfs_dir2_data_log_header(tp
, dbp
);
912 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
914 * If the bests table needs to be changed, do it.
915 * Log the change unless we've already done that.
917 if (be16_to_cpu(bestsp
[use_block
]) != be16_to_cpu(bf
[0].length
)) {
918 bestsp
[use_block
] = bf
[0].length
;
920 xfs_dir3_leaf_log_bests(tp
, lbp
, use_block
, use_block
);
923 lep
= xfs_dir3_leaf_find_entry(&leafhdr
, ents
, index
, compact
, lowstale
,
924 highstale
, &lfloglow
, &lfloghigh
);
927 * Fill in the new leaf entry.
929 lep
->hashval
= cpu_to_be32(args
->hashval
);
930 lep
->address
= cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp
, use_block
,
931 be16_to_cpu(*tagp
)));
933 * Log the leaf fields and give up the buffers.
935 xfs_dir3_leaf_hdr_to_disk(leaf
, &leafhdr
);
936 xfs_dir3_leaf_log_header(tp
, lbp
);
937 xfs_dir3_leaf_log_ents(tp
, lbp
, lfloglow
, lfloghigh
);
938 xfs_dir3_leaf_check(mp
, lbp
);
939 xfs_dir3_data_check(dp
, dbp
);
944 * Compact out any stale entries in the leaf.
945 * Log the header and changed leaf entries, if any.
948 xfs_dir3_leaf_compact(
949 xfs_da_args_t
*args
, /* operation arguments */
950 struct xfs_dir3_icleaf_hdr
*leafhdr
,
951 struct xfs_buf
*bp
) /* leaf buffer */
953 int from
; /* source leaf index */
954 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
955 int loglow
; /* first leaf entry to log */
956 int to
; /* target leaf index */
957 struct xfs_dir2_leaf_entry
*ents
;
964 * Compress out the stale entries in place.
966 ents
= xfs_dir3_leaf_ents_p(leaf
);
967 for (from
= to
= 0, loglow
= -1; from
< leafhdr
->count
; from
++) {
968 if (ents
[from
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
971 * Only actually copy the entries that are different.
976 ents
[to
] = ents
[from
];
981 * Update and log the header, log the leaf entries.
983 ASSERT(leafhdr
->stale
== from
- to
);
984 leafhdr
->count
-= leafhdr
->stale
;
987 xfs_dir3_leaf_hdr_to_disk(leaf
, leafhdr
);
988 xfs_dir3_leaf_log_header(args
->trans
, bp
);
990 xfs_dir3_leaf_log_ents(args
->trans
, bp
, loglow
, to
- 1);
994 * Compact the leaf entries, removing stale ones.
995 * Leave one stale entry behind - the one closest to our
996 * insertion index - and the caller will shift that one to our insertion
998 * Return new insertion index, where the remaining stale entry is,
999 * and leaf logging indices.
1002 xfs_dir3_leaf_compact_x1(
1003 struct xfs_dir3_icleaf_hdr
*leafhdr
,
1004 struct xfs_dir2_leaf_entry
*ents
,
1005 int *indexp
, /* insertion index */
1006 int *lowstalep
, /* out: stale entry before us */
1007 int *highstalep
, /* out: stale entry after us */
1008 int *lowlogp
, /* out: low log index */
1009 int *highlogp
) /* out: high log index */
1011 int from
; /* source copy index */
1012 int highstale
; /* stale entry at/after index */
1013 int index
; /* insertion index */
1014 int keepstale
; /* source index of kept stale */
1015 int lowstale
; /* stale entry before index */
1016 int newindex
=0; /* new insertion index */
1017 int to
; /* destination copy index */
1019 ASSERT(leafhdr
->stale
> 1);
1022 xfs_dir3_leaf_find_stale(leafhdr
, ents
, index
, &lowstale
, &highstale
);
1025 * Pick the better of lowstale and highstale.
1027 if (lowstale
>= 0 &&
1028 (highstale
== leafhdr
->count
||
1029 index
- lowstale
<= highstale
- index
))
1030 keepstale
= lowstale
;
1032 keepstale
= highstale
;
1034 * Copy the entries in place, removing all the stale entries
1037 for (from
= to
= 0; from
< leafhdr
->count
; from
++) {
1039 * Notice the new value of index.
1043 if (from
!= keepstale
&&
1044 ents
[from
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
)) {
1050 * Record the new keepstale value for the insertion.
1052 if (from
== keepstale
)
1053 lowstale
= highstale
= to
;
1055 * Copy only the entries that have moved.
1058 ents
[to
] = ents
[from
];
1063 * If the insertion point was past the last entry,
1064 * set the new insertion point accordingly.
1070 * Adjust the leaf header values.
1072 leafhdr
->count
-= from
- to
;
1075 * Remember the low/high stale value only in the "right"
1078 if (lowstale
>= newindex
)
1081 highstale
= leafhdr
->count
;
1082 *highlogp
= leafhdr
->count
- 1;
1083 *lowstalep
= lowstale
;
1084 *highstalep
= highstale
;
1088 * Log the bests entries indicated from a leaf1 block.
1091 xfs_dir3_leaf_log_bests(
1092 xfs_trans_t
*tp
, /* transaction pointer */
1093 struct xfs_buf
*bp
, /* leaf buffer */
1094 int first
, /* first entry to log */
1095 int last
) /* last entry to log */
1097 __be16
*firstb
; /* pointer to first entry */
1098 __be16
*lastb
; /* pointer to last entry */
1099 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
1100 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1102 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
1103 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAF1_MAGIC
));
1105 ltp
= xfs_dir2_leaf_tail_p(tp
->t_mountp
, leaf
);
1106 firstb
= xfs_dir2_leaf_bests_p(ltp
) + first
;
1107 lastb
= xfs_dir2_leaf_bests_p(ltp
) + last
;
1108 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)firstb
- (char *)leaf
),
1109 (uint
)((char *)lastb
- (char *)leaf
+ sizeof(*lastb
) - 1));
1113 * Log the leaf entries indicated from a leaf1 or leafn block.
1116 xfs_dir3_leaf_log_ents(
1117 xfs_trans_t
*tp
, /* transaction pointer */
1118 struct xfs_buf
*bp
, /* leaf buffer */
1119 int first
, /* first entry to log */
1120 int last
) /* last entry to log */
1122 xfs_dir2_leaf_entry_t
*firstlep
; /* pointer to first entry */
1123 xfs_dir2_leaf_entry_t
*lastlep
; /* pointer to last entry */
1124 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
1125 struct xfs_dir2_leaf_entry
*ents
;
1127 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
1128 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAF1_MAGIC
) ||
1129 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1130 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
));
1132 ents
= xfs_dir3_leaf_ents_p(leaf
);
1133 firstlep
= &ents
[first
];
1134 lastlep
= &ents
[last
];
1135 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)firstlep
- (char *)leaf
),
1136 (uint
)((char *)lastlep
- (char *)leaf
+ sizeof(*lastlep
) - 1));
1140 * Log the header of the leaf1 or leafn block.
1143 xfs_dir3_leaf_log_header(
1144 struct xfs_trans
*tp
,
1147 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
1149 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
1150 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAF1_MAGIC
) ||
1151 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1152 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
));
1154 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)&leaf
->hdr
- (char *)leaf
),
1155 xfs_dir3_leaf_hdr_size(leaf
) - 1);
1159 * Log the tail of the leaf1 block.
1162 xfs_dir3_leaf_log_tail(
1163 struct xfs_trans
*tp
,
1166 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
1167 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1168 struct xfs_mount
*mp
= tp
->t_mountp
;
1170 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
1171 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAF1_MAGIC
) ||
1172 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1173 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
));
1175 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1176 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)ltp
- (char *)leaf
),
1177 (uint
)(mp
->m_dirblksize
- 1));
1181 * Look up the entry referred to by args in the leaf format directory.
1182 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1183 * is also used by the node-format code.
1186 xfs_dir2_leaf_lookup(
1187 xfs_da_args_t
*args
) /* operation arguments */
1189 struct xfs_buf
*dbp
; /* data block buffer */
1190 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1191 xfs_inode_t
*dp
; /* incore directory inode */
1192 int error
; /* error return code */
1193 int index
; /* found entry index */
1194 struct xfs_buf
*lbp
; /* leaf buffer */
1195 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1196 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1197 xfs_trans_t
*tp
; /* transaction pointer */
1198 struct xfs_dir2_leaf_entry
*ents
;
1200 trace_xfs_dir2_leaf_lookup(args
);
1203 * Look up name in the leaf block, returning both buffers and index.
1205 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1210 xfs_dir3_leaf_check(dp
->i_mount
, lbp
);
1212 ents
= xfs_dir3_leaf_ents_p(leaf
);
1214 * Get to the leaf entry and contained data entry address.
1219 * Point to the data entry.
1221 dep
= (xfs_dir2_data_entry_t
*)
1222 ((char *)dbp
->b_addr
+
1223 xfs_dir2_dataptr_to_off(dp
->i_mount
, be32_to_cpu(lep
->address
)));
1225 * Return the found inode number & CI name if appropriate
1227 args
->inumber
= be64_to_cpu(dep
->inumber
);
1228 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
1229 xfs_trans_brelse(tp
, dbp
);
1230 xfs_trans_brelse(tp
, lbp
);
1231 return XFS_ERROR(error
);
1235 * Look up name/hash in the leaf block.
1236 * Fill in indexp with the found index, and dbpp with the data buffer.
1237 * If not found dbpp will be NULL, and ENOENT comes back.
1238 * lbpp will always be filled in with the leaf buffer unless there's an error.
1240 static int /* error */
1241 xfs_dir2_leaf_lookup_int(
1242 xfs_da_args_t
*args
, /* operation arguments */
1243 struct xfs_buf
**lbpp
, /* out: leaf buffer */
1244 int *indexp
, /* out: index in leaf block */
1245 struct xfs_buf
**dbpp
) /* out: data buffer */
1247 xfs_dir2_db_t curdb
= -1; /* current data block number */
1248 struct xfs_buf
*dbp
= NULL
; /* data buffer */
1249 xfs_dir2_data_entry_t
*dep
; /* data entry */
1250 xfs_inode_t
*dp
; /* incore directory inode */
1251 int error
; /* error return code */
1252 int index
; /* index in leaf block */
1253 struct xfs_buf
*lbp
; /* leaf buffer */
1254 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1255 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1256 xfs_mount_t
*mp
; /* filesystem mount point */
1257 xfs_dir2_db_t newdb
; /* new data block number */
1258 xfs_trans_t
*tp
; /* transaction pointer */
1259 xfs_dir2_db_t cidb
= -1; /* case match data block no. */
1260 enum xfs_dacmp cmp
; /* name compare result */
1261 struct xfs_dir2_leaf_entry
*ents
;
1262 struct xfs_dir3_icleaf_hdr leafhdr
;
1268 error
= xfs_dir3_leaf_read(tp
, dp
, mp
->m_dirleafblk
, -1, &lbp
);
1274 xfs_dir3_leaf_check(mp
, lbp
);
1275 ents
= xfs_dir3_leaf_ents_p(leaf
);
1276 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
1279 * Look for the first leaf entry with our hash value.
1281 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
1283 * Loop over all the entries with the right hash value
1284 * looking to match the name.
1286 for (lep
= &ents
[index
];
1287 index
< leafhdr
.count
&& be32_to_cpu(lep
->hashval
) == args
->hashval
;
1290 * Skip over stale leaf entries.
1292 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
1295 * Get the new data block number.
1297 newdb
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
1299 * If it's not the same as the old data block number,
1300 * need to pitch the old one and read the new one.
1302 if (newdb
!= curdb
) {
1304 xfs_trans_brelse(tp
, dbp
);
1305 error
= xfs_dir3_data_read(tp
, dp
,
1306 xfs_dir2_db_to_da(mp
, newdb
),
1309 xfs_trans_brelse(tp
, lbp
);
1315 * Point to the data entry.
1317 dep
= (xfs_dir2_data_entry_t
*)((char *)dbp
->b_addr
+
1318 xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
1320 * Compare name and if it's an exact match, return the index
1321 * and buffer. If it's the first case-insensitive match, store
1322 * the index and buffer and continue looking for an exact match.
1324 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
1325 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
1326 args
->cmpresult
= cmp
;
1328 /* case exact match: return the current buffer. */
1329 if (cmp
== XFS_CMP_EXACT
) {
1336 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
1338 * Here, we can only be doing a lookup (not a rename or remove).
1339 * If a case-insensitive match was found earlier, re-read the
1340 * appropriate data block if required and return it.
1342 if (args
->cmpresult
== XFS_CMP_CASE
) {
1344 if (cidb
!= curdb
) {
1345 xfs_trans_brelse(tp
, dbp
);
1346 error
= xfs_dir3_data_read(tp
, dp
,
1347 xfs_dir2_db_to_da(mp
, cidb
),
1350 xfs_trans_brelse(tp
, lbp
);
1358 * No match found, return ENOENT.
1362 xfs_trans_brelse(tp
, dbp
);
1363 xfs_trans_brelse(tp
, lbp
);
1364 return XFS_ERROR(ENOENT
);
1368 * Remove an entry from a leaf format directory.
1371 xfs_dir2_leaf_removename(
1372 xfs_da_args_t
*args
) /* operation arguments */
1374 __be16
*bestsp
; /* leaf block best freespace */
1375 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
1376 xfs_dir2_db_t db
; /* data block number */
1377 struct xfs_buf
*dbp
; /* data block buffer */
1378 xfs_dir2_data_entry_t
*dep
; /* data entry structure */
1379 xfs_inode_t
*dp
; /* incore directory inode */
1380 int error
; /* error return code */
1381 xfs_dir2_db_t i
; /* temporary data block # */
1382 int index
; /* index into leaf entries */
1383 struct xfs_buf
*lbp
; /* leaf buffer */
1384 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1385 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1386 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1387 xfs_mount_t
*mp
; /* filesystem mount point */
1388 int needlog
; /* need to log data header */
1389 int needscan
; /* need to rescan data frees */
1390 xfs_dir2_data_off_t oldbest
; /* old value of best free */
1391 xfs_trans_t
*tp
; /* transaction pointer */
1392 struct xfs_dir2_data_free
*bf
; /* bestfree table */
1393 struct xfs_dir2_leaf_entry
*ents
;
1394 struct xfs_dir3_icleaf_hdr leafhdr
;
1396 trace_xfs_dir2_leaf_removename(args
);
1399 * Lookup the leaf entry, get the leaf and data blocks read in.
1401 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1409 xfs_dir3_data_check(dp
, dbp
);
1410 bf
= xfs_dir3_data_bestfree_p(hdr
);
1411 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
1412 ents
= xfs_dir3_leaf_ents_p(leaf
);
1414 * Point to the leaf entry, use that to point to the data entry.
1417 db
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
1418 dep
= (xfs_dir2_data_entry_t
*)
1419 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
1420 needscan
= needlog
= 0;
1421 oldbest
= be16_to_cpu(bf
[0].length
);
1422 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1423 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1424 ASSERT(be16_to_cpu(bestsp
[db
]) == oldbest
);
1426 * Mark the former data entry unused.
1428 xfs_dir2_data_make_free(tp
, dbp
,
1429 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)hdr
),
1430 xfs_dir2_data_entsize(dep
->namelen
), &needlog
, &needscan
);
1432 * We just mark the leaf entry stale by putting a null in it.
1435 xfs_dir3_leaf_hdr_to_disk(leaf
, &leafhdr
);
1436 xfs_dir3_leaf_log_header(tp
, lbp
);
1438 lep
->address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
1439 xfs_dir3_leaf_log_ents(tp
, lbp
, index
, index
);
1442 * Scan the freespace in the data block again if necessary,
1443 * log the data block header if necessary.
1446 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
1448 xfs_dir2_data_log_header(tp
, dbp
);
1450 * If the longest freespace in the data block has changed,
1451 * put the new value in the bests table and log that.
1453 if (be16_to_cpu(bf
[0].length
) != oldbest
) {
1454 bestsp
[db
] = bf
[0].length
;
1455 xfs_dir3_leaf_log_bests(tp
, lbp
, db
, db
);
1457 xfs_dir3_data_check(dp
, dbp
);
1459 * If the data block is now empty then get rid of the data block.
1461 if (be16_to_cpu(bf
[0].length
) ==
1462 mp
->m_dirblksize
- xfs_dir3_data_entry_offset(hdr
)) {
1463 ASSERT(db
!= mp
->m_dirdatablk
);
1464 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1466 * Nope, can't get rid of it because it caused
1467 * allocation of a bmap btree block to do so.
1468 * Just go on, returning success, leaving the
1469 * empty block in place.
1471 if (error
== ENOSPC
&& args
->total
== 0)
1473 xfs_dir3_leaf_check(mp
, lbp
);
1478 * If this is the last data block then compact the
1479 * bests table by getting rid of entries.
1481 if (db
== be32_to_cpu(ltp
->bestcount
) - 1) {
1483 * Look for the last active entry (i).
1485 for (i
= db
- 1; i
> 0; i
--) {
1486 if (bestsp
[i
] != cpu_to_be16(NULLDATAOFF
))
1490 * Copy the table down so inactive entries at the
1493 memmove(&bestsp
[db
- i
], bestsp
,
1494 (be32_to_cpu(ltp
->bestcount
) - (db
- i
)) * sizeof(*bestsp
));
1495 be32_add_cpu(<p
->bestcount
, -(db
- i
));
1496 xfs_dir3_leaf_log_tail(tp
, lbp
);
1497 xfs_dir3_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1499 bestsp
[db
] = cpu_to_be16(NULLDATAOFF
);
1502 * If the data block was not the first one, drop it.
1504 else if (db
!= mp
->m_dirdatablk
)
1507 xfs_dir3_leaf_check(mp
, lbp
);
1509 * See if we can convert to block form.
1511 return xfs_dir2_leaf_to_block(args
, lbp
, dbp
);
1515 * Replace the inode number in a leaf format directory entry.
1518 xfs_dir2_leaf_replace(
1519 xfs_da_args_t
*args
) /* operation arguments */
1521 struct xfs_buf
*dbp
; /* data block buffer */
1522 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1523 xfs_inode_t
*dp
; /* incore directory inode */
1524 int error
; /* error return code */
1525 int index
; /* index of leaf entry */
1526 struct xfs_buf
*lbp
; /* leaf buffer */
1527 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1528 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1529 xfs_trans_t
*tp
; /* transaction pointer */
1530 struct xfs_dir2_leaf_entry
*ents
;
1532 trace_xfs_dir2_leaf_replace(args
);
1535 * Look up the entry.
1537 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1542 ents
= xfs_dir3_leaf_ents_p(leaf
);
1544 * Point to the leaf entry, get data address from it.
1548 * Point to the data entry.
1550 dep
= (xfs_dir2_data_entry_t
*)
1551 ((char *)dbp
->b_addr
+
1552 xfs_dir2_dataptr_to_off(dp
->i_mount
, be32_to_cpu(lep
->address
)));
1553 ASSERT(args
->inumber
!= be64_to_cpu(dep
->inumber
));
1555 * Put the new inode number in, log it.
1557 dep
->inumber
= cpu_to_be64(args
->inumber
);
1559 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
1560 xfs_dir3_leaf_check(dp
->i_mount
, lbp
);
1561 xfs_trans_brelse(tp
, lbp
);
1566 * Return index in the leaf block (lbp) which is either the first
1567 * one with this hash value, or if there are none, the insert point
1568 * for that hash value.
1570 int /* index value */
1571 xfs_dir2_leaf_search_hash(
1572 xfs_da_args_t
*args
, /* operation arguments */
1573 struct xfs_buf
*lbp
) /* leaf buffer */
1575 xfs_dahash_t hash
=0; /* hash from this entry */
1576 xfs_dahash_t hashwant
; /* hash value looking for */
1577 int high
; /* high leaf index */
1578 int low
; /* low leaf index */
1579 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1580 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1581 int mid
=0; /* current leaf index */
1582 struct xfs_dir2_leaf_entry
*ents
;
1583 struct xfs_dir3_icleaf_hdr leafhdr
;
1586 ents
= xfs_dir3_leaf_ents_p(leaf
);
1587 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
1594 * Note, the table cannot be empty, so we have to go through the loop.
1595 * Binary search the leaf entries looking for our hash value.
1597 for (lep
= ents
, low
= 0, high
= leafhdr
.count
- 1,
1598 hashwant
= args
->hashval
;
1600 mid
= (low
+ high
) >> 1;
1601 if ((hash
= be32_to_cpu(lep
[mid
].hashval
)) == hashwant
)
1603 if (hash
< hashwant
)
1609 * Found one, back up through all the equal hash values.
1611 if (hash
== hashwant
) {
1612 while (mid
> 0 && be32_to_cpu(lep
[mid
- 1].hashval
) == hashwant
) {
1617 * Need to point to an entry higher than ours.
1619 else if (hash
< hashwant
)
1625 * Trim off a trailing data block. We know it's empty since the leaf
1626 * freespace table says so.
1629 xfs_dir2_leaf_trim_data(
1630 xfs_da_args_t
*args
, /* operation arguments */
1631 struct xfs_buf
*lbp
, /* leaf buffer */
1632 xfs_dir2_db_t db
) /* data block number */
1634 __be16
*bestsp
; /* leaf bests table */
1635 struct xfs_buf
*dbp
; /* data block buffer */
1636 xfs_inode_t
*dp
; /* incore directory inode */
1637 int error
; /* error return value */
1638 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1639 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1640 xfs_mount_t
*mp
; /* filesystem mount point */
1641 xfs_trans_t
*tp
; /* transaction pointer */
1647 * Read the offending data block. We need its buffer.
1649 error
= xfs_dir3_data_read(tp
, dp
, xfs_dir2_db_to_da(mp
, db
), -1, &dbp
);
1654 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1658 struct xfs_dir2_data_hdr
*hdr
= dbp
->b_addr
;
1659 struct xfs_dir2_data_free
*bf
= xfs_dir3_data_bestfree_p(hdr
);
1661 ASSERT(hdr
->magic
== cpu_to_be32(XFS_DIR2_DATA_MAGIC
) ||
1662 hdr
->magic
== cpu_to_be32(XFS_DIR3_DATA_MAGIC
));
1663 ASSERT(be16_to_cpu(bf
[0].length
) ==
1664 mp
->m_dirblksize
- xfs_dir3_data_entry_offset(hdr
));
1665 ASSERT(db
== be32_to_cpu(ltp
->bestcount
) - 1);
1670 * Get rid of the data block.
1672 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1673 ASSERT(error
!= ENOSPC
);
1674 xfs_trans_brelse(tp
, dbp
);
1678 * Eliminate the last bests entry from the table.
1680 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1681 be32_add_cpu(<p
->bestcount
, -1);
1682 memmove(&bestsp
[1], &bestsp
[0], be32_to_cpu(ltp
->bestcount
) * sizeof(*bestsp
));
1683 xfs_dir3_leaf_log_tail(tp
, lbp
);
1684 xfs_dir3_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1688 static inline size_t
1690 struct xfs_dir3_icleaf_hdr
*hdr
,
1696 entries
= hdr
->count
- hdr
->stale
;
1697 if (hdr
->magic
== XFS_DIR2_LEAF1_MAGIC
||
1698 hdr
->magic
== XFS_DIR2_LEAFN_MAGIC
)
1699 hdrsize
= sizeof(struct xfs_dir2_leaf_hdr
);
1701 hdrsize
= sizeof(struct xfs_dir3_leaf_hdr
);
1703 return hdrsize
+ entries
* sizeof(xfs_dir2_leaf_entry_t
)
1704 + counts
* sizeof(xfs_dir2_data_off_t
)
1705 + sizeof(xfs_dir2_leaf_tail_t
);
1709 * Convert node form directory to leaf form directory.
1710 * The root of the node form dir needs to already be a LEAFN block.
1711 * Just return if we can't do anything.
1714 xfs_dir2_node_to_leaf(
1715 xfs_da_state_t
*state
) /* directory operation state */
1717 xfs_da_args_t
*args
; /* operation arguments */
1718 xfs_inode_t
*dp
; /* incore directory inode */
1719 int error
; /* error return code */
1720 struct xfs_buf
*fbp
; /* buffer for freespace block */
1721 xfs_fileoff_t fo
; /* freespace file offset */
1722 xfs_dir2_free_t
*free
; /* freespace structure */
1723 struct xfs_buf
*lbp
; /* buffer for leaf block */
1724 xfs_dir2_leaf_tail_t
*ltp
; /* tail of leaf structure */
1725 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1726 xfs_mount_t
*mp
; /* filesystem mount point */
1727 int rval
; /* successful free trim? */
1728 xfs_trans_t
*tp
; /* transaction pointer */
1729 struct xfs_dir3_icleaf_hdr leafhdr
;
1730 struct xfs_dir3_icfree_hdr freehdr
;
1733 * There's more than a leaf level in the btree, so there must
1734 * be multiple leafn blocks. Give up.
1736 if (state
->path
.active
> 1)
1740 trace_xfs_dir2_node_to_leaf(args
);
1746 * Get the last offset in the file.
1748 if ((error
= xfs_bmap_last_offset(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1751 fo
-= mp
->m_dirblkfsbs
;
1753 * If there are freespace blocks other than the first one,
1754 * take this opportunity to remove trailing empty freespace blocks
1755 * that may have been left behind during no-space-reservation
1758 while (fo
> mp
->m_dirfreeblk
) {
1759 if ((error
= xfs_dir2_node_trim_free(args
, fo
, &rval
))) {
1763 fo
-= mp
->m_dirblkfsbs
;
1768 * Now find the block just before the freespace block.
1770 if ((error
= xfs_bmap_last_before(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1774 * If it's not the single leaf block, give up.
1776 if (XFS_FSB_TO_B(mp
, fo
) > XFS_DIR2_LEAF_OFFSET
+ mp
->m_dirblksize
)
1778 lbp
= state
->path
.blk
[0].bp
;
1780 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
1782 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAFN_MAGIC
||
1783 leafhdr
.magic
== XFS_DIR3_LEAFN_MAGIC
);
1786 * Read the freespace block.
1788 error
= xfs_dir2_free_read(tp
, dp
, mp
->m_dirfreeblk
, &fbp
);
1792 xfs_dir3_free_hdr_from_disk(&freehdr
, free
);
1794 ASSERT(!freehdr
.firstdb
);
1797 * Now see if the leafn and free data will fit in a leaf1.
1798 * If not, release the buffer and give up.
1800 if (xfs_dir3_leaf_size(&leafhdr
, freehdr
.nvalid
) > mp
->m_dirblksize
) {
1801 xfs_trans_brelse(tp
, fbp
);
1806 * If the leaf has any stale entries in it, compress them out.
1809 xfs_dir3_leaf_compact(args
, &leafhdr
, lbp
);
1811 lbp
->b_ops
= &xfs_dir3_leaf1_buf_ops
;
1812 xfs_trans_buf_set_type(tp
, lbp
, XFS_BLFT_DIR_LEAF1_BUF
);
1813 leafhdr
.magic
= (leafhdr
.magic
== XFS_DIR2_LEAFN_MAGIC
)
1814 ? XFS_DIR2_LEAF1_MAGIC
1815 : XFS_DIR3_LEAF1_MAGIC
;
1818 * Set up the leaf tail from the freespace block.
1820 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1821 ltp
->bestcount
= cpu_to_be32(freehdr
.nvalid
);
1824 * Set up the leaf bests table.
1826 memcpy(xfs_dir2_leaf_bests_p(ltp
), xfs_dir3_free_bests_p(mp
, free
),
1827 freehdr
.nvalid
* sizeof(xfs_dir2_data_off_t
));
1829 xfs_dir3_leaf_hdr_to_disk(leaf
, &leafhdr
);
1830 xfs_dir3_leaf_log_header(tp
, lbp
);
1831 xfs_dir3_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1832 xfs_dir3_leaf_log_tail(tp
, lbp
);
1833 xfs_dir3_leaf_check(mp
, lbp
);
1836 * Get rid of the freespace block.
1838 error
= xfs_dir2_shrink_inode(args
, XFS_DIR2_FREE_FIRSTDB(mp
), fbp
);
1841 * This can't fail here because it can only happen when
1842 * punching out the middle of an extent, and this is an
1845 ASSERT(error
!= ENOSPC
);
1850 * Now see if we can convert the single-leaf directory
1851 * down to a block form directory.
1852 * This routine always kills the dabuf for the leaf, so
1853 * eliminate it from the path.
1855 error
= xfs_dir2_leaf_to_block(args
, lbp
, NULL
);
1856 state
->path
.blk
[0].bp
= NULL
;