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_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
30 #include "xfs_dir2_priv.h"
31 #include "xfs_error.h"
32 #include "xfs_trace.h"
33 #include "xfs_trans.h"
34 #include "xfs_buf_item.h"
35 #include "xfs_cksum.h"
38 * Local function declarations.
40 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t
*args
, struct xfs_buf
**lbpp
,
41 int *indexp
, struct xfs_buf
**dbpp
);
42 static void xfs_dir3_leaf_log_bests(struct xfs_da_args
*args
,
43 struct xfs_buf
*bp
, int first
, int last
);
44 static void xfs_dir3_leaf_log_tail(struct xfs_da_args
*args
,
48 * Check the internal consistency of a leaf1 block.
49 * Pop an assert if something is wrong.
52 #define xfs_dir3_leaf_check(dp, bp) \
54 if (!xfs_dir3_leaf1_check((dp), (bp))) \
63 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
64 struct xfs_dir3_icleaf_hdr leafhdr
;
66 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
68 if (leafhdr
.magic
== XFS_DIR3_LEAF1_MAGIC
) {
69 struct xfs_dir3_leaf_hdr
*leaf3
= bp
->b_addr
;
70 if (be64_to_cpu(leaf3
->info
.blkno
) != bp
->b_bn
)
72 } else if (leafhdr
.magic
!= XFS_DIR2_LEAF1_MAGIC
)
75 return xfs_dir3_leaf_check_int(dp
->i_mount
, dp
, &leafhdr
, leaf
);
78 #define xfs_dir3_leaf_check(dp, bp)
82 xfs_dir3_leaf_check_int(
85 struct xfs_dir3_icleaf_hdr
*hdr
,
86 struct xfs_dir2_leaf
*leaf
)
88 struct xfs_dir2_leaf_entry
*ents
;
89 xfs_dir2_leaf_tail_t
*ltp
;
92 const struct xfs_dir_ops
*ops
;
93 struct xfs_dir3_icleaf_hdr leafhdr
;
94 struct xfs_da_geometry
*geo
= mp
->m_dir_geo
;
97 * we can be passed a null dp here from a verifier, so we need to go the
98 * hard way to get them.
100 ops
= xfs_dir_get_ops(mp
, dp
);
103 ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
107 ents
= ops
->leaf_ents_p(leaf
);
108 ltp
= xfs_dir2_leaf_tail_p(geo
, leaf
);
111 * XXX (dgc): This value is not restrictive enough.
112 * Should factor in the size of the bests table as well.
113 * We can deduce a value for that from di_size.
115 if (hdr
->count
> ops
->leaf_max_ents(geo
))
118 /* Leaves and bests don't overlap in leaf format. */
119 if ((hdr
->magic
== XFS_DIR2_LEAF1_MAGIC
||
120 hdr
->magic
== XFS_DIR3_LEAF1_MAGIC
) &&
121 (char *)&ents
[hdr
->count
] > (char *)xfs_dir2_leaf_bests_p(ltp
))
124 /* Check hash value order, count stale entries. */
125 for (i
= stale
= 0; i
< hdr
->count
; i
++) {
126 if (i
+ 1 < hdr
->count
) {
127 if (be32_to_cpu(ents
[i
].hashval
) >
128 be32_to_cpu(ents
[i
+ 1].hashval
))
131 if (ents
[i
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
134 if (hdr
->stale
!= stale
)
140 * We verify the magic numbers before decoding the leaf header so that on debug
141 * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
142 * to incorrect magic numbers.
145 xfs_dir3_leaf_verify(
149 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
150 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
152 ASSERT(magic
== XFS_DIR2_LEAF1_MAGIC
|| magic
== XFS_DIR2_LEAFN_MAGIC
);
154 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
155 struct xfs_dir3_leaf_hdr
*leaf3
= bp
->b_addr
;
158 magic3
= (magic
== XFS_DIR2_LEAF1_MAGIC
) ? XFS_DIR3_LEAF1_MAGIC
159 : XFS_DIR3_LEAFN_MAGIC
;
161 if (leaf3
->info
.hdr
.magic
!= cpu_to_be16(magic3
))
163 if (!uuid_equal(&leaf3
->info
.uuid
, &mp
->m_sb
.sb_uuid
))
165 if (be64_to_cpu(leaf3
->info
.blkno
) != bp
->b_bn
)
168 if (leaf
->hdr
.info
.magic
!= cpu_to_be16(magic
))
172 return xfs_dir3_leaf_check_int(mp
, NULL
, NULL
, leaf
);
180 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
182 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
183 !xfs_buf_verify_cksum(bp
, XFS_DIR3_LEAF_CRC_OFF
))
184 xfs_buf_ioerror(bp
, -EFSBADCRC
);
185 else if (!xfs_dir3_leaf_verify(bp
, magic
))
186 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
189 xfs_verifier_error(bp
);
197 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
198 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
199 struct xfs_dir3_leaf_hdr
*hdr3
= bp
->b_addr
;
201 if (!xfs_dir3_leaf_verify(bp
, magic
)) {
202 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
203 xfs_verifier_error(bp
);
207 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
211 hdr3
->info
.lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
213 xfs_buf_update_cksum(bp
, XFS_DIR3_LEAF_CRC_OFF
);
217 xfs_dir3_leaf1_read_verify(
220 __read_verify(bp
, XFS_DIR2_LEAF1_MAGIC
);
224 xfs_dir3_leaf1_write_verify(
227 __write_verify(bp
, XFS_DIR2_LEAF1_MAGIC
);
231 xfs_dir3_leafn_read_verify(
234 __read_verify(bp
, XFS_DIR2_LEAFN_MAGIC
);
238 xfs_dir3_leafn_write_verify(
241 __write_verify(bp
, XFS_DIR2_LEAFN_MAGIC
);
244 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops
= {
245 .verify_read
= xfs_dir3_leaf1_read_verify
,
246 .verify_write
= xfs_dir3_leaf1_write_verify
,
249 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops
= {
250 .verify_read
= xfs_dir3_leafn_read_verify
,
251 .verify_write
= xfs_dir3_leafn_write_verify
,
256 struct xfs_trans
*tp
,
257 struct xfs_inode
*dp
,
259 xfs_daddr_t mappedbno
,
260 struct xfs_buf
**bpp
)
264 err
= xfs_da_read_buf(tp
, dp
, fbno
, mappedbno
, bpp
,
265 XFS_DATA_FORK
, &xfs_dir3_leaf1_buf_ops
);
267 xfs_trans_buf_set_type(tp
, *bpp
, XFS_BLFT_DIR_LEAF1_BUF
);
273 struct xfs_trans
*tp
,
274 struct xfs_inode
*dp
,
276 xfs_daddr_t mappedbno
,
277 struct xfs_buf
**bpp
)
281 err
= xfs_da_read_buf(tp
, dp
, fbno
, mappedbno
, bpp
,
282 XFS_DATA_FORK
, &xfs_dir3_leafn_buf_ops
);
284 xfs_trans_buf_set_type(tp
, *bpp
, XFS_BLFT_DIR_LEAFN_BUF
);
289 * Initialize a new leaf block, leaf1 or leafn magic accepted.
293 struct xfs_mount
*mp
,
294 struct xfs_trans
*tp
,
299 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
301 ASSERT(type
== XFS_DIR2_LEAF1_MAGIC
|| type
== XFS_DIR2_LEAFN_MAGIC
);
303 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
304 struct xfs_dir3_leaf_hdr
*leaf3
= bp
->b_addr
;
306 memset(leaf3
, 0, sizeof(*leaf3
));
308 leaf3
->info
.hdr
.magic
= (type
== XFS_DIR2_LEAF1_MAGIC
)
309 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC
)
310 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
);
311 leaf3
->info
.blkno
= cpu_to_be64(bp
->b_bn
);
312 leaf3
->info
.owner
= cpu_to_be64(owner
);
313 uuid_copy(&leaf3
->info
.uuid
, &mp
->m_sb
.sb_uuid
);
315 memset(leaf
, 0, sizeof(*leaf
));
316 leaf
->hdr
.info
.magic
= cpu_to_be16(type
);
320 * If it's a leaf-format directory initialize the tail.
321 * Caller is responsible for initialising the bests table.
323 if (type
== XFS_DIR2_LEAF1_MAGIC
) {
324 struct xfs_dir2_leaf_tail
*ltp
;
326 ltp
= xfs_dir2_leaf_tail_p(mp
->m_dir_geo
, leaf
);
328 bp
->b_ops
= &xfs_dir3_leaf1_buf_ops
;
329 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DIR_LEAF1_BUF
);
331 bp
->b_ops
= &xfs_dir3_leafn_buf_ops
;
332 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DIR_LEAFN_BUF
);
337 xfs_dir3_leaf_get_buf(
340 struct xfs_buf
**bpp
,
343 struct xfs_inode
*dp
= args
->dp
;
344 struct xfs_trans
*tp
= args
->trans
;
345 struct xfs_mount
*mp
= dp
->i_mount
;
349 ASSERT(magic
== XFS_DIR2_LEAF1_MAGIC
|| magic
== XFS_DIR2_LEAFN_MAGIC
);
350 ASSERT(bno
>= xfs_dir2_byte_to_db(args
->geo
, XFS_DIR2_LEAF_OFFSET
) &&
351 bno
< xfs_dir2_byte_to_db(args
->geo
, XFS_DIR2_FREE_OFFSET
));
353 error
= xfs_da_get_buf(tp
, dp
, xfs_dir2_db_to_da(args
->geo
, bno
),
354 -1, &bp
, XFS_DATA_FORK
);
358 xfs_dir3_leaf_init(mp
, tp
, bp
, dp
->i_ino
, magic
);
359 xfs_dir3_leaf_log_header(args
, bp
);
360 if (magic
== XFS_DIR2_LEAF1_MAGIC
)
361 xfs_dir3_leaf_log_tail(args
, bp
);
367 * Convert a block form directory to a leaf form directory.
370 xfs_dir2_block_to_leaf(
371 xfs_da_args_t
*args
, /* operation arguments */
372 struct xfs_buf
*dbp
) /* input block's buffer */
374 __be16
*bestsp
; /* leaf's bestsp entries */
375 xfs_dablk_t blkno
; /* leaf block's bno */
376 xfs_dir2_data_hdr_t
*hdr
; /* block header */
377 xfs_dir2_leaf_entry_t
*blp
; /* block's leaf entries */
378 xfs_dir2_block_tail_t
*btp
; /* block's tail */
379 xfs_inode_t
*dp
; /* incore directory inode */
380 int error
; /* error return code */
381 struct xfs_buf
*lbp
; /* leaf block's buffer */
382 xfs_dir2_db_t ldb
; /* leaf block's bno */
383 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
384 xfs_dir2_leaf_tail_t
*ltp
; /* leaf's tail */
385 int needlog
; /* need to log block header */
386 int needscan
; /* need to rescan bestfree */
387 xfs_trans_t
*tp
; /* transaction pointer */
388 struct xfs_dir2_data_free
*bf
;
389 struct xfs_dir2_leaf_entry
*ents
;
390 struct xfs_dir3_icleaf_hdr leafhdr
;
392 trace_xfs_dir2_block_to_leaf(args
);
397 * Add the leaf block to the inode.
398 * This interface will only put blocks in the leaf/node range.
399 * Since that's empty now, we'll get the root (block 0 in range).
401 if ((error
= xfs_da_grow_inode(args
, &blkno
))) {
404 ldb
= xfs_dir2_da_to_db(args
->geo
, blkno
);
405 ASSERT(ldb
== xfs_dir2_byte_to_db(args
->geo
, XFS_DIR2_LEAF_OFFSET
));
407 * Initialize the leaf block, get a buffer for it.
409 error
= xfs_dir3_leaf_get_buf(args
, ldb
, &lbp
, XFS_DIR2_LEAF1_MAGIC
);
415 xfs_dir3_data_check(dp
, dbp
);
416 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
417 blp
= xfs_dir2_block_leaf_p(btp
);
418 bf
= dp
->d_ops
->data_bestfree_p(hdr
);
419 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
422 * Set the counts in the leaf header.
424 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
425 leafhdr
.count
= be32_to_cpu(btp
->count
);
426 leafhdr
.stale
= be32_to_cpu(btp
->stale
);
427 dp
->d_ops
->leaf_hdr_to_disk(leaf
, &leafhdr
);
428 xfs_dir3_leaf_log_header(args
, lbp
);
431 * Could compact these but I think we always do the conversion
432 * after squeezing out stale entries.
434 memcpy(ents
, blp
, be32_to_cpu(btp
->count
) * sizeof(xfs_dir2_leaf_entry_t
));
435 xfs_dir3_leaf_log_ents(args
, lbp
, 0, leafhdr
.count
- 1);
439 * Make the space formerly occupied by the leaf entries and block
442 xfs_dir2_data_make_free(args
, dbp
,
443 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)hdr
),
444 (xfs_dir2_data_aoff_t
)((char *)hdr
+ args
->geo
->blksize
-
446 &needlog
, &needscan
);
448 * Fix up the block header, make it a data block.
450 dbp
->b_ops
= &xfs_dir3_data_buf_ops
;
451 xfs_trans_buf_set_type(tp
, dbp
, XFS_BLFT_DIR_DATA_BUF
);
452 if (hdr
->magic
== cpu_to_be32(XFS_DIR2_BLOCK_MAGIC
))
453 hdr
->magic
= cpu_to_be32(XFS_DIR2_DATA_MAGIC
);
455 hdr
->magic
= cpu_to_be32(XFS_DIR3_DATA_MAGIC
);
458 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
460 * Set up leaf tail and bests table.
462 ltp
= xfs_dir2_leaf_tail_p(args
->geo
, leaf
);
463 ltp
->bestcount
= cpu_to_be32(1);
464 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
465 bestsp
[0] = bf
[0].length
;
467 * Log the data header and leaf bests table.
470 xfs_dir2_data_log_header(args
, dbp
);
471 xfs_dir3_leaf_check(dp
, lbp
);
472 xfs_dir3_data_check(dp
, dbp
);
473 xfs_dir3_leaf_log_bests(args
, lbp
, 0, 0);
478 xfs_dir3_leaf_find_stale(
479 struct xfs_dir3_icleaf_hdr
*leafhdr
,
480 struct xfs_dir2_leaf_entry
*ents
,
486 * Find the first stale entry before our index, if any.
488 for (*lowstale
= index
- 1; *lowstale
>= 0; --*lowstale
) {
489 if (ents
[*lowstale
].address
==
490 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
495 * Find the first stale entry at or after our index, if any.
496 * Stop if the result would require moving more entries than using
499 for (*highstale
= index
; *highstale
< leafhdr
->count
; ++*highstale
) {
500 if (ents
[*highstale
].address
==
501 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
503 if (*lowstale
>= 0 && index
- *lowstale
<= *highstale
- index
)
508 struct xfs_dir2_leaf_entry
*
509 xfs_dir3_leaf_find_entry(
510 struct xfs_dir3_icleaf_hdr
*leafhdr
,
511 struct xfs_dir2_leaf_entry
*ents
,
512 int index
, /* leaf table position */
513 int compact
, /* need to compact leaves */
514 int lowstale
, /* index of prev stale leaf */
515 int highstale
, /* index of next stale leaf */
516 int *lfloglow
, /* low leaf logging index */
517 int *lfloghigh
) /* high leaf logging index */
519 if (!leafhdr
->stale
) {
520 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry table pointer */
523 * Now we need to make room to insert the leaf entry.
525 * If there are no stale entries, just insert a hole at index.
528 if (index
< leafhdr
->count
)
529 memmove(lep
+ 1, lep
,
530 (leafhdr
->count
- index
) * sizeof(*lep
));
533 * Record low and high logging indices for the leaf.
536 *lfloghigh
= leafhdr
->count
++;
541 * There are stale entries.
543 * We will use one of them for the new entry. It's probably not at
544 * the right location, so we'll have to shift some up or down first.
546 * If we didn't compact before, we need to find the nearest stale
547 * entries before and after our insertion point.
550 xfs_dir3_leaf_find_stale(leafhdr
, ents
, index
,
551 &lowstale
, &highstale
);
554 * If the low one is better, use it.
557 (highstale
== leafhdr
->count
||
558 index
- lowstale
- 1 < highstale
- index
)) {
559 ASSERT(index
- lowstale
- 1 >= 0);
560 ASSERT(ents
[lowstale
].address
==
561 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
));
564 * Copy entries up to cover the stale entry and make room
567 if (index
- lowstale
- 1 > 0) {
568 memmove(&ents
[lowstale
], &ents
[lowstale
+ 1],
569 (index
- lowstale
- 1) *
570 sizeof(xfs_dir2_leaf_entry_t
));
572 *lfloglow
= MIN(lowstale
, *lfloglow
);
573 *lfloghigh
= MAX(index
- 1, *lfloghigh
);
575 return &ents
[index
- 1];
579 * The high one is better, so use that one.
581 ASSERT(highstale
- index
>= 0);
582 ASSERT(ents
[highstale
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
));
585 * Copy entries down to cover the stale entry and make room for the
588 if (highstale
- index
> 0) {
589 memmove(&ents
[index
+ 1], &ents
[index
],
590 (highstale
- index
) * sizeof(xfs_dir2_leaf_entry_t
));
592 *lfloglow
= MIN(index
, *lfloglow
);
593 *lfloghigh
= MAX(highstale
, *lfloghigh
);
599 * Add an entry to a leaf form directory.
602 xfs_dir2_leaf_addname(
603 xfs_da_args_t
*args
) /* operation arguments */
605 __be16
*bestsp
; /* freespace table in leaf */
606 int compact
; /* need to compact leaves */
607 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
608 struct xfs_buf
*dbp
; /* data block buffer */
609 xfs_dir2_data_entry_t
*dep
; /* data block entry */
610 xfs_inode_t
*dp
; /* incore directory inode */
611 xfs_dir2_data_unused_t
*dup
; /* data unused entry */
612 int error
; /* error return value */
613 int grown
; /* allocated new data block */
614 int highstale
; /* index of next stale leaf */
615 int i
; /* temporary, index */
616 int index
; /* leaf table position */
617 struct xfs_buf
*lbp
; /* leaf's buffer */
618 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
619 int length
; /* length of new entry */
620 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry table pointer */
621 int lfloglow
; /* low leaf logging index */
622 int lfloghigh
; /* high leaf logging index */
623 int lowstale
; /* index of prev stale leaf */
624 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail pointer */
625 int needbytes
; /* leaf block bytes needed */
626 int needlog
; /* need to log data header */
627 int needscan
; /* need to rescan data free */
628 __be16
*tagp
; /* end of data entry */
629 xfs_trans_t
*tp
; /* transaction pointer */
630 xfs_dir2_db_t use_block
; /* data block number */
631 struct xfs_dir2_data_free
*bf
; /* bestfree table */
632 struct xfs_dir2_leaf_entry
*ents
;
633 struct xfs_dir3_icleaf_hdr leafhdr
;
635 trace_xfs_dir2_leaf_addname(args
);
640 error
= xfs_dir3_leaf_read(tp
, dp
, args
->geo
->leafblk
, -1, &lbp
);
645 * Look up the entry by hash value and name.
646 * We know it's not there, our caller has already done a lookup.
647 * So the index is of the entry to insert in front of.
648 * But if there are dup hash values the index is of the first of those.
650 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
652 ltp
= xfs_dir2_leaf_tail_p(args
->geo
, leaf
);
653 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
654 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
655 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
656 length
= dp
->d_ops
->data_entsize(args
->namelen
);
659 * See if there are any entries with the same hash value
660 * and space in their block for the new entry.
661 * This is good because it puts multiple same-hash value entries
662 * in a data block, improving the lookup of those entries.
664 for (use_block
= -1, lep
= &ents
[index
];
665 index
< leafhdr
.count
&& be32_to_cpu(lep
->hashval
) == args
->hashval
;
667 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
669 i
= xfs_dir2_dataptr_to_db(args
->geo
, be32_to_cpu(lep
->address
));
670 ASSERT(i
< be32_to_cpu(ltp
->bestcount
));
671 ASSERT(bestsp
[i
] != cpu_to_be16(NULLDATAOFF
));
672 if (be16_to_cpu(bestsp
[i
]) >= length
) {
678 * Didn't find a block yet, linear search all the data blocks.
680 if (use_block
== -1) {
681 for (i
= 0; i
< be32_to_cpu(ltp
->bestcount
); i
++) {
683 * Remember a block we see that's missing.
685 if (bestsp
[i
] == cpu_to_be16(NULLDATAOFF
) &&
688 else if (be16_to_cpu(bestsp
[i
]) >= length
) {
695 * How many bytes do we need in the leaf block?
699 needbytes
+= sizeof(xfs_dir2_leaf_entry_t
);
701 needbytes
+= sizeof(xfs_dir2_data_off_t
);
704 * Now kill use_block if it refers to a missing block, so we
705 * can use it as an indication of allocation needed.
707 if (use_block
!= -1 && bestsp
[use_block
] == cpu_to_be16(NULLDATAOFF
))
710 * If we don't have enough free bytes but we can make enough
711 * by compacting out stale entries, we'll do that.
713 if ((char *)bestsp
- (char *)&ents
[leafhdr
.count
] < needbytes
&&
718 * Otherwise if we don't have enough free bytes we need to
719 * convert to node form.
721 else if ((char *)bestsp
- (char *)&ents
[leafhdr
.count
] < needbytes
) {
723 * Just checking or no space reservation, give up.
725 if ((args
->op_flags
& XFS_DA_OP_JUSTCHECK
) ||
727 xfs_trans_brelse(tp
, lbp
);
731 * Convert to node form.
733 error
= xfs_dir2_leaf_to_node(args
, lbp
);
737 * Then add the new entry.
739 return xfs_dir2_node_addname(args
);
742 * Otherwise it will fit without compaction.
747 * If just checking, then it will fit unless we needed to allocate
750 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
751 xfs_trans_brelse(tp
, lbp
);
752 return use_block
== -1 ? -ENOSPC
: 0;
755 * If no allocations are allowed, return now before we've
758 if (args
->total
== 0 && use_block
== -1) {
759 xfs_trans_brelse(tp
, lbp
);
763 * Need to compact the leaf entries, removing stale ones.
764 * Leave one stale entry behind - the one closest to our
765 * insertion index - and we'll shift that one to our insertion
769 xfs_dir3_leaf_compact_x1(&leafhdr
, ents
, &index
, &lowstale
,
770 &highstale
, &lfloglow
, &lfloghigh
);
773 * There are stale entries, so we'll need log-low and log-high
774 * impossibly bad values later.
776 else if (leafhdr
.stale
) {
777 lfloglow
= leafhdr
.count
;
781 * If there was no data block space found, we need to allocate
784 if (use_block
== -1) {
786 * Add the new data block.
788 if ((error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
,
790 xfs_trans_brelse(tp
, lbp
);
794 * Initialize the block.
796 if ((error
= xfs_dir3_data_init(args
, use_block
, &dbp
))) {
797 xfs_trans_brelse(tp
, lbp
);
801 * If we're adding a new data block on the end we need to
802 * extend the bests table. Copy it up one entry.
804 if (use_block
>= be32_to_cpu(ltp
->bestcount
)) {
806 memmove(&bestsp
[0], &bestsp
[1],
807 be32_to_cpu(ltp
->bestcount
) * sizeof(bestsp
[0]));
808 be32_add_cpu(<p
->bestcount
, 1);
809 xfs_dir3_leaf_log_tail(args
, lbp
);
810 xfs_dir3_leaf_log_bests(args
, lbp
, 0,
811 be32_to_cpu(ltp
->bestcount
) - 1);
814 * If we're filling in a previously empty block just log it.
817 xfs_dir3_leaf_log_bests(args
, lbp
, use_block
, use_block
);
819 bf
= dp
->d_ops
->data_bestfree_p(hdr
);
820 bestsp
[use_block
] = bf
[0].length
;
824 * Already had space in some data block.
825 * Just read that one in.
827 error
= xfs_dir3_data_read(tp
, dp
,
828 xfs_dir2_db_to_da(args
->geo
, use_block
),
831 xfs_trans_brelse(tp
, lbp
);
835 bf
= dp
->d_ops
->data_bestfree_p(hdr
);
839 * Point to the biggest freespace in our data block.
841 dup
= (xfs_dir2_data_unused_t
*)
842 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
843 ASSERT(be16_to_cpu(dup
->length
) >= length
);
844 needscan
= needlog
= 0;
846 * Mark the initial part of our freespace in use for the new entry.
848 xfs_dir2_data_use_free(args
, dbp
, dup
,
849 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
), length
,
850 &needlog
, &needscan
);
852 * Initialize our new entry (at last).
854 dep
= (xfs_dir2_data_entry_t
*)dup
;
855 dep
->inumber
= cpu_to_be64(args
->inumber
);
856 dep
->namelen
= args
->namelen
;
857 memcpy(dep
->name
, args
->name
, dep
->namelen
);
858 dp
->d_ops
->data_put_ftype(dep
, args
->filetype
);
859 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
860 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
862 * Need to scan fix up the bestfree table.
865 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
867 * Need to log the data block's header.
870 xfs_dir2_data_log_header(args
, dbp
);
871 xfs_dir2_data_log_entry(args
, dbp
, dep
);
873 * If the bests table needs to be changed, do it.
874 * Log the change unless we've already done that.
876 if (be16_to_cpu(bestsp
[use_block
]) != be16_to_cpu(bf
[0].length
)) {
877 bestsp
[use_block
] = bf
[0].length
;
879 xfs_dir3_leaf_log_bests(args
, lbp
, use_block
, use_block
);
882 lep
= xfs_dir3_leaf_find_entry(&leafhdr
, ents
, index
, compact
, lowstale
,
883 highstale
, &lfloglow
, &lfloghigh
);
886 * Fill in the new leaf entry.
888 lep
->hashval
= cpu_to_be32(args
->hashval
);
889 lep
->address
= cpu_to_be32(
890 xfs_dir2_db_off_to_dataptr(args
->geo
, use_block
,
891 be16_to_cpu(*tagp
)));
893 * Log the leaf fields and give up the buffers.
895 dp
->d_ops
->leaf_hdr_to_disk(leaf
, &leafhdr
);
896 xfs_dir3_leaf_log_header(args
, lbp
);
897 xfs_dir3_leaf_log_ents(args
, lbp
, lfloglow
, lfloghigh
);
898 xfs_dir3_leaf_check(dp
, lbp
);
899 xfs_dir3_data_check(dp
, dbp
);
904 * Compact out any stale entries in the leaf.
905 * Log the header and changed leaf entries, if any.
908 xfs_dir3_leaf_compact(
909 xfs_da_args_t
*args
, /* operation arguments */
910 struct xfs_dir3_icleaf_hdr
*leafhdr
,
911 struct xfs_buf
*bp
) /* leaf buffer */
913 int from
; /* source leaf index */
914 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
915 int loglow
; /* first leaf entry to log */
916 int to
; /* target leaf index */
917 struct xfs_dir2_leaf_entry
*ents
;
918 struct xfs_inode
*dp
= args
->dp
;
925 * Compress out the stale entries in place.
927 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
928 for (from
= to
= 0, loglow
= -1; from
< leafhdr
->count
; from
++) {
929 if (ents
[from
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
932 * Only actually copy the entries that are different.
937 ents
[to
] = ents
[from
];
942 * Update and log the header, log the leaf entries.
944 ASSERT(leafhdr
->stale
== from
- to
);
945 leafhdr
->count
-= leafhdr
->stale
;
948 dp
->d_ops
->leaf_hdr_to_disk(leaf
, leafhdr
);
949 xfs_dir3_leaf_log_header(args
, bp
);
951 xfs_dir3_leaf_log_ents(args
, bp
, loglow
, to
- 1);
955 * Compact the leaf entries, removing stale ones.
956 * Leave one stale entry behind - the one closest to our
957 * insertion index - and the caller will shift that one to our insertion
959 * Return new insertion index, where the remaining stale entry is,
960 * and leaf logging indices.
963 xfs_dir3_leaf_compact_x1(
964 struct xfs_dir3_icleaf_hdr
*leafhdr
,
965 struct xfs_dir2_leaf_entry
*ents
,
966 int *indexp
, /* insertion index */
967 int *lowstalep
, /* out: stale entry before us */
968 int *highstalep
, /* out: stale entry after us */
969 int *lowlogp
, /* out: low log index */
970 int *highlogp
) /* out: high log index */
972 int from
; /* source copy index */
973 int highstale
; /* stale entry at/after index */
974 int index
; /* insertion index */
975 int keepstale
; /* source index of kept stale */
976 int lowstale
; /* stale entry before index */
977 int newindex
=0; /* new insertion index */
978 int to
; /* destination copy index */
980 ASSERT(leafhdr
->stale
> 1);
983 xfs_dir3_leaf_find_stale(leafhdr
, ents
, index
, &lowstale
, &highstale
);
986 * Pick the better of lowstale and highstale.
989 (highstale
== leafhdr
->count
||
990 index
- lowstale
<= highstale
- index
))
991 keepstale
= lowstale
;
993 keepstale
= highstale
;
995 * Copy the entries in place, removing all the stale entries
998 for (from
= to
= 0; from
< leafhdr
->count
; from
++) {
1000 * Notice the new value of index.
1004 if (from
!= keepstale
&&
1005 ents
[from
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
)) {
1011 * Record the new keepstale value for the insertion.
1013 if (from
== keepstale
)
1014 lowstale
= highstale
= to
;
1016 * Copy only the entries that have moved.
1019 ents
[to
] = ents
[from
];
1024 * If the insertion point was past the last entry,
1025 * set the new insertion point accordingly.
1031 * Adjust the leaf header values.
1033 leafhdr
->count
-= from
- to
;
1036 * Remember the low/high stale value only in the "right"
1039 if (lowstale
>= newindex
)
1042 highstale
= leafhdr
->count
;
1043 *highlogp
= leafhdr
->count
- 1;
1044 *lowstalep
= lowstale
;
1045 *highstalep
= highstale
;
1049 * Log the bests entries indicated from a leaf1 block.
1052 xfs_dir3_leaf_log_bests(
1053 struct xfs_da_args
*args
,
1054 struct xfs_buf
*bp
, /* leaf buffer */
1055 int first
, /* first entry to log */
1056 int last
) /* last entry to log */
1058 __be16
*firstb
; /* pointer to first entry */
1059 __be16
*lastb
; /* pointer to last entry */
1060 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
1061 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1063 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
1064 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAF1_MAGIC
));
1066 ltp
= xfs_dir2_leaf_tail_p(args
->geo
, leaf
);
1067 firstb
= xfs_dir2_leaf_bests_p(ltp
) + first
;
1068 lastb
= xfs_dir2_leaf_bests_p(ltp
) + last
;
1069 xfs_trans_log_buf(args
->trans
, bp
,
1070 (uint
)((char *)firstb
- (char *)leaf
),
1071 (uint
)((char *)lastb
- (char *)leaf
+ sizeof(*lastb
) - 1));
1075 * Log the leaf entries indicated from a leaf1 or leafn block.
1078 xfs_dir3_leaf_log_ents(
1079 struct xfs_da_args
*args
,
1084 xfs_dir2_leaf_entry_t
*firstlep
; /* pointer to first entry */
1085 xfs_dir2_leaf_entry_t
*lastlep
; /* pointer to last entry */
1086 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
1087 struct xfs_dir2_leaf_entry
*ents
;
1089 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
1090 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAF1_MAGIC
) ||
1091 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1092 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
));
1094 ents
= args
->dp
->d_ops
->leaf_ents_p(leaf
);
1095 firstlep
= &ents
[first
];
1096 lastlep
= &ents
[last
];
1097 xfs_trans_log_buf(args
->trans
, bp
,
1098 (uint
)((char *)firstlep
- (char *)leaf
),
1099 (uint
)((char *)lastlep
- (char *)leaf
+ sizeof(*lastlep
) - 1));
1103 * Log the header of the leaf1 or leafn block.
1106 xfs_dir3_leaf_log_header(
1107 struct xfs_da_args
*args
,
1110 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
1112 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
1113 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAF1_MAGIC
) ||
1114 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1115 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
));
1117 xfs_trans_log_buf(args
->trans
, bp
,
1118 (uint
)((char *)&leaf
->hdr
- (char *)leaf
),
1119 args
->dp
->d_ops
->leaf_hdr_size
- 1);
1123 * Log the tail of the leaf1 block.
1126 xfs_dir3_leaf_log_tail(
1127 struct xfs_da_args
*args
,
1130 struct xfs_dir2_leaf
*leaf
= bp
->b_addr
;
1131 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1133 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
1134 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAF1_MAGIC
) ||
1135 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
) ||
1136 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR3_LEAFN_MAGIC
));
1138 ltp
= xfs_dir2_leaf_tail_p(args
->geo
, leaf
);
1139 xfs_trans_log_buf(args
->trans
, bp
, (uint
)((char *)ltp
- (char *)leaf
),
1140 (uint
)(args
->geo
->blksize
- 1));
1144 * Look up the entry referred to by args in the leaf format directory.
1145 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1146 * is also used by the node-format code.
1149 xfs_dir2_leaf_lookup(
1150 xfs_da_args_t
*args
) /* operation arguments */
1152 struct xfs_buf
*dbp
; /* data block buffer */
1153 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1154 xfs_inode_t
*dp
; /* incore directory inode */
1155 int error
; /* error return code */
1156 int index
; /* found entry index */
1157 struct xfs_buf
*lbp
; /* leaf buffer */
1158 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1159 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1160 xfs_trans_t
*tp
; /* transaction pointer */
1161 struct xfs_dir2_leaf_entry
*ents
;
1163 trace_xfs_dir2_leaf_lookup(args
);
1166 * Look up name in the leaf block, returning both buffers and index.
1168 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1173 xfs_dir3_leaf_check(dp
, lbp
);
1175 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
1177 * Get to the leaf entry and contained data entry address.
1182 * Point to the data entry.
1184 dep
= (xfs_dir2_data_entry_t
*)
1185 ((char *)dbp
->b_addr
+
1186 xfs_dir2_dataptr_to_off(args
->geo
, be32_to_cpu(lep
->address
)));
1188 * Return the found inode number & CI name if appropriate
1190 args
->inumber
= be64_to_cpu(dep
->inumber
);
1191 args
->filetype
= dp
->d_ops
->data_get_ftype(dep
);
1192 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
1193 xfs_trans_brelse(tp
, dbp
);
1194 xfs_trans_brelse(tp
, lbp
);
1199 * Look up name/hash in the leaf block.
1200 * Fill in indexp with the found index, and dbpp with the data buffer.
1201 * If not found dbpp will be NULL, and ENOENT comes back.
1202 * lbpp will always be filled in with the leaf buffer unless there's an error.
1204 static int /* error */
1205 xfs_dir2_leaf_lookup_int(
1206 xfs_da_args_t
*args
, /* operation arguments */
1207 struct xfs_buf
**lbpp
, /* out: leaf buffer */
1208 int *indexp
, /* out: index in leaf block */
1209 struct xfs_buf
**dbpp
) /* out: data buffer */
1211 xfs_dir2_db_t curdb
= -1; /* current data block number */
1212 struct xfs_buf
*dbp
= NULL
; /* data buffer */
1213 xfs_dir2_data_entry_t
*dep
; /* data entry */
1214 xfs_inode_t
*dp
; /* incore directory inode */
1215 int error
; /* error return code */
1216 int index
; /* index in leaf block */
1217 struct xfs_buf
*lbp
; /* leaf buffer */
1218 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1219 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1220 xfs_mount_t
*mp
; /* filesystem mount point */
1221 xfs_dir2_db_t newdb
; /* new data block number */
1222 xfs_trans_t
*tp
; /* transaction pointer */
1223 xfs_dir2_db_t cidb
= -1; /* case match data block no. */
1224 enum xfs_dacmp cmp
; /* name compare result */
1225 struct xfs_dir2_leaf_entry
*ents
;
1226 struct xfs_dir3_icleaf_hdr leafhdr
;
1232 error
= xfs_dir3_leaf_read(tp
, dp
, args
->geo
->leafblk
, -1, &lbp
);
1238 xfs_dir3_leaf_check(dp
, lbp
);
1239 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
1240 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
1243 * Look for the first leaf entry with our hash value.
1245 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
1247 * Loop over all the entries with the right hash value
1248 * looking to match the name.
1250 for (lep
= &ents
[index
];
1251 index
< leafhdr
.count
&& be32_to_cpu(lep
->hashval
) == args
->hashval
;
1254 * Skip over stale leaf entries.
1256 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
1259 * Get the new data block number.
1261 newdb
= xfs_dir2_dataptr_to_db(args
->geo
,
1262 be32_to_cpu(lep
->address
));
1264 * If it's not the same as the old data block number,
1265 * need to pitch the old one and read the new one.
1267 if (newdb
!= curdb
) {
1269 xfs_trans_brelse(tp
, dbp
);
1270 error
= xfs_dir3_data_read(tp
, dp
,
1271 xfs_dir2_db_to_da(args
->geo
, newdb
),
1274 xfs_trans_brelse(tp
, lbp
);
1280 * Point to the data entry.
1282 dep
= (xfs_dir2_data_entry_t
*)((char *)dbp
->b_addr
+
1283 xfs_dir2_dataptr_to_off(args
->geo
,
1284 be32_to_cpu(lep
->address
)));
1286 * Compare name and if it's an exact match, return the index
1287 * and buffer. If it's the first case-insensitive match, store
1288 * the index and buffer and continue looking for an exact match.
1290 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
1291 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
1292 args
->cmpresult
= cmp
;
1294 /* case exact match: return the current buffer. */
1295 if (cmp
== XFS_CMP_EXACT
) {
1302 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
1304 * Here, we can only be doing a lookup (not a rename or remove).
1305 * If a case-insensitive match was found earlier, re-read the
1306 * appropriate data block if required and return it.
1308 if (args
->cmpresult
== XFS_CMP_CASE
) {
1310 if (cidb
!= curdb
) {
1311 xfs_trans_brelse(tp
, dbp
);
1312 error
= xfs_dir3_data_read(tp
, dp
,
1313 xfs_dir2_db_to_da(args
->geo
, cidb
),
1316 xfs_trans_brelse(tp
, lbp
);
1324 * No match found, return -ENOENT.
1328 xfs_trans_brelse(tp
, dbp
);
1329 xfs_trans_brelse(tp
, lbp
);
1334 * Remove an entry from a leaf format directory.
1337 xfs_dir2_leaf_removename(
1338 xfs_da_args_t
*args
) /* operation arguments */
1340 __be16
*bestsp
; /* leaf block best freespace */
1341 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
1342 xfs_dir2_db_t db
; /* data block number */
1343 struct xfs_buf
*dbp
; /* data block buffer */
1344 xfs_dir2_data_entry_t
*dep
; /* data entry structure */
1345 xfs_inode_t
*dp
; /* incore directory inode */
1346 int error
; /* error return code */
1347 xfs_dir2_db_t i
; /* temporary data block # */
1348 int index
; /* index into leaf entries */
1349 struct xfs_buf
*lbp
; /* leaf buffer */
1350 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1351 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1352 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1353 int needlog
; /* need to log data header */
1354 int needscan
; /* need to rescan data frees */
1355 xfs_dir2_data_off_t oldbest
; /* old value of best free */
1356 struct xfs_dir2_data_free
*bf
; /* bestfree table */
1357 struct xfs_dir2_leaf_entry
*ents
;
1358 struct xfs_dir3_icleaf_hdr leafhdr
;
1360 trace_xfs_dir2_leaf_removename(args
);
1363 * Lookup the leaf entry, get the leaf and data blocks read in.
1365 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1371 xfs_dir3_data_check(dp
, dbp
);
1372 bf
= dp
->d_ops
->data_bestfree_p(hdr
);
1373 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
1374 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
1376 * Point to the leaf entry, use that to point to the data entry.
1379 db
= xfs_dir2_dataptr_to_db(args
->geo
, be32_to_cpu(lep
->address
));
1380 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
1381 xfs_dir2_dataptr_to_off(args
->geo
, be32_to_cpu(lep
->address
)));
1382 needscan
= needlog
= 0;
1383 oldbest
= be16_to_cpu(bf
[0].length
);
1384 ltp
= xfs_dir2_leaf_tail_p(args
->geo
, leaf
);
1385 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1386 ASSERT(be16_to_cpu(bestsp
[db
]) == oldbest
);
1388 * Mark the former data entry unused.
1390 xfs_dir2_data_make_free(args
, dbp
,
1391 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)hdr
),
1392 dp
->d_ops
->data_entsize(dep
->namelen
), &needlog
, &needscan
);
1394 * We just mark the leaf entry stale by putting a null in it.
1397 dp
->d_ops
->leaf_hdr_to_disk(leaf
, &leafhdr
);
1398 xfs_dir3_leaf_log_header(args
, lbp
);
1400 lep
->address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
1401 xfs_dir3_leaf_log_ents(args
, lbp
, index
, index
);
1404 * Scan the freespace in the data block again if necessary,
1405 * log the data block header if necessary.
1408 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
1410 xfs_dir2_data_log_header(args
, dbp
);
1412 * If the longest freespace in the data block has changed,
1413 * put the new value in the bests table and log that.
1415 if (be16_to_cpu(bf
[0].length
) != oldbest
) {
1416 bestsp
[db
] = bf
[0].length
;
1417 xfs_dir3_leaf_log_bests(args
, lbp
, db
, db
);
1419 xfs_dir3_data_check(dp
, dbp
);
1421 * If the data block is now empty then get rid of the data block.
1423 if (be16_to_cpu(bf
[0].length
) ==
1424 args
->geo
->blksize
- dp
->d_ops
->data_entry_offset
) {
1425 ASSERT(db
!= args
->geo
->datablk
);
1426 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1428 * Nope, can't get rid of it because it caused
1429 * allocation of a bmap btree block to do so.
1430 * Just go on, returning success, leaving the
1431 * empty block in place.
1433 if (error
== -ENOSPC
&& args
->total
== 0)
1435 xfs_dir3_leaf_check(dp
, lbp
);
1440 * If this is the last data block then compact the
1441 * bests table by getting rid of entries.
1443 if (db
== be32_to_cpu(ltp
->bestcount
) - 1) {
1445 * Look for the last active entry (i).
1447 for (i
= db
- 1; i
> 0; i
--) {
1448 if (bestsp
[i
] != cpu_to_be16(NULLDATAOFF
))
1452 * Copy the table down so inactive entries at the
1455 memmove(&bestsp
[db
- i
], bestsp
,
1456 (be32_to_cpu(ltp
->bestcount
) - (db
- i
)) * sizeof(*bestsp
));
1457 be32_add_cpu(<p
->bestcount
, -(db
- i
));
1458 xfs_dir3_leaf_log_tail(args
, lbp
);
1459 xfs_dir3_leaf_log_bests(args
, lbp
, 0,
1460 be32_to_cpu(ltp
->bestcount
) - 1);
1462 bestsp
[db
] = cpu_to_be16(NULLDATAOFF
);
1465 * If the data block was not the first one, drop it.
1467 else if (db
!= args
->geo
->datablk
)
1470 xfs_dir3_leaf_check(dp
, lbp
);
1472 * See if we can convert to block form.
1474 return xfs_dir2_leaf_to_block(args
, lbp
, dbp
);
1478 * Replace the inode number in a leaf format directory entry.
1481 xfs_dir2_leaf_replace(
1482 xfs_da_args_t
*args
) /* operation arguments */
1484 struct xfs_buf
*dbp
; /* data block buffer */
1485 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1486 xfs_inode_t
*dp
; /* incore directory inode */
1487 int error
; /* error return code */
1488 int index
; /* index of leaf entry */
1489 struct xfs_buf
*lbp
; /* leaf buffer */
1490 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1491 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1492 xfs_trans_t
*tp
; /* transaction pointer */
1493 struct xfs_dir2_leaf_entry
*ents
;
1495 trace_xfs_dir2_leaf_replace(args
);
1498 * Look up the entry.
1500 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1505 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
1507 * Point to the leaf entry, get data address from it.
1511 * Point to the data entry.
1513 dep
= (xfs_dir2_data_entry_t
*)
1514 ((char *)dbp
->b_addr
+
1515 xfs_dir2_dataptr_to_off(args
->geo
, be32_to_cpu(lep
->address
)));
1516 ASSERT(args
->inumber
!= be64_to_cpu(dep
->inumber
));
1518 * Put the new inode number in, log it.
1520 dep
->inumber
= cpu_to_be64(args
->inumber
);
1521 dp
->d_ops
->data_put_ftype(dep
, args
->filetype
);
1523 xfs_dir2_data_log_entry(args
, dbp
, dep
);
1524 xfs_dir3_leaf_check(dp
, lbp
);
1525 xfs_trans_brelse(tp
, lbp
);
1530 * Return index in the leaf block (lbp) which is either the first
1531 * one with this hash value, or if there are none, the insert point
1532 * for that hash value.
1534 int /* index value */
1535 xfs_dir2_leaf_search_hash(
1536 xfs_da_args_t
*args
, /* operation arguments */
1537 struct xfs_buf
*lbp
) /* leaf buffer */
1539 xfs_dahash_t hash
=0; /* hash from this entry */
1540 xfs_dahash_t hashwant
; /* hash value looking for */
1541 int high
; /* high leaf index */
1542 int low
; /* low leaf index */
1543 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1544 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1545 int mid
=0; /* current leaf index */
1546 struct xfs_dir2_leaf_entry
*ents
;
1547 struct xfs_dir3_icleaf_hdr leafhdr
;
1550 ents
= args
->dp
->d_ops
->leaf_ents_p(leaf
);
1551 args
->dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
1554 * Note, the table cannot be empty, so we have to go through the loop.
1555 * Binary search the leaf entries looking for our hash value.
1557 for (lep
= ents
, low
= 0, high
= leafhdr
.count
- 1,
1558 hashwant
= args
->hashval
;
1560 mid
= (low
+ high
) >> 1;
1561 if ((hash
= be32_to_cpu(lep
[mid
].hashval
)) == hashwant
)
1563 if (hash
< hashwant
)
1569 * Found one, back up through all the equal hash values.
1571 if (hash
== hashwant
) {
1572 while (mid
> 0 && be32_to_cpu(lep
[mid
- 1].hashval
) == hashwant
) {
1577 * Need to point to an entry higher than ours.
1579 else if (hash
< hashwant
)
1585 * Trim off a trailing data block. We know it's empty since the leaf
1586 * freespace table says so.
1589 xfs_dir2_leaf_trim_data(
1590 xfs_da_args_t
*args
, /* operation arguments */
1591 struct xfs_buf
*lbp
, /* leaf buffer */
1592 xfs_dir2_db_t db
) /* data block number */
1594 __be16
*bestsp
; /* leaf bests table */
1595 struct xfs_buf
*dbp
; /* data block buffer */
1596 xfs_inode_t
*dp
; /* incore directory inode */
1597 int error
; /* error return value */
1598 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1599 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1600 xfs_trans_t
*tp
; /* transaction pointer */
1605 * Read the offending data block. We need its buffer.
1607 error
= xfs_dir3_data_read(tp
, dp
, xfs_dir2_db_to_da(args
->geo
, db
),
1613 ltp
= xfs_dir2_leaf_tail_p(args
->geo
, leaf
);
1617 struct xfs_dir2_data_hdr
*hdr
= dbp
->b_addr
;
1618 struct xfs_dir2_data_free
*bf
= dp
->d_ops
->data_bestfree_p(hdr
);
1620 ASSERT(hdr
->magic
== cpu_to_be32(XFS_DIR2_DATA_MAGIC
) ||
1621 hdr
->magic
== cpu_to_be32(XFS_DIR3_DATA_MAGIC
));
1622 ASSERT(be16_to_cpu(bf
[0].length
) ==
1623 args
->geo
->blksize
- dp
->d_ops
->data_entry_offset
);
1624 ASSERT(db
== be32_to_cpu(ltp
->bestcount
) - 1);
1629 * Get rid of the data block.
1631 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1632 ASSERT(error
!= -ENOSPC
);
1633 xfs_trans_brelse(tp
, dbp
);
1637 * Eliminate the last bests entry from the table.
1639 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1640 be32_add_cpu(<p
->bestcount
, -1);
1641 memmove(&bestsp
[1], &bestsp
[0], be32_to_cpu(ltp
->bestcount
) * sizeof(*bestsp
));
1642 xfs_dir3_leaf_log_tail(args
, lbp
);
1643 xfs_dir3_leaf_log_bests(args
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1647 static inline size_t
1649 struct xfs_dir3_icleaf_hdr
*hdr
,
1655 entries
= hdr
->count
- hdr
->stale
;
1656 if (hdr
->magic
== XFS_DIR2_LEAF1_MAGIC
||
1657 hdr
->magic
== XFS_DIR2_LEAFN_MAGIC
)
1658 hdrsize
= sizeof(struct xfs_dir2_leaf_hdr
);
1660 hdrsize
= sizeof(struct xfs_dir3_leaf_hdr
);
1662 return hdrsize
+ entries
* sizeof(xfs_dir2_leaf_entry_t
)
1663 + counts
* sizeof(xfs_dir2_data_off_t
)
1664 + sizeof(xfs_dir2_leaf_tail_t
);
1668 * Convert node form directory to leaf form directory.
1669 * The root of the node form dir needs to already be a LEAFN block.
1670 * Just return if we can't do anything.
1673 xfs_dir2_node_to_leaf(
1674 xfs_da_state_t
*state
) /* directory operation state */
1676 xfs_da_args_t
*args
; /* operation arguments */
1677 xfs_inode_t
*dp
; /* incore directory inode */
1678 int error
; /* error return code */
1679 struct xfs_buf
*fbp
; /* buffer for freespace block */
1680 xfs_fileoff_t fo
; /* freespace file offset */
1681 xfs_dir2_free_t
*free
; /* freespace structure */
1682 struct xfs_buf
*lbp
; /* buffer for leaf block */
1683 xfs_dir2_leaf_tail_t
*ltp
; /* tail of leaf structure */
1684 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1685 xfs_mount_t
*mp
; /* filesystem mount point */
1686 int rval
; /* successful free trim? */
1687 xfs_trans_t
*tp
; /* transaction pointer */
1688 struct xfs_dir3_icleaf_hdr leafhdr
;
1689 struct xfs_dir3_icfree_hdr freehdr
;
1692 * There's more than a leaf level in the btree, so there must
1693 * be multiple leafn blocks. Give up.
1695 if (state
->path
.active
> 1)
1699 trace_xfs_dir2_node_to_leaf(args
);
1705 * Get the last offset in the file.
1707 if ((error
= xfs_bmap_last_offset(dp
, &fo
, XFS_DATA_FORK
))) {
1710 fo
-= args
->geo
->fsbcount
;
1712 * If there are freespace blocks other than the first one,
1713 * take this opportunity to remove trailing empty freespace blocks
1714 * that may have been left behind during no-space-reservation
1717 while (fo
> args
->geo
->freeblk
) {
1718 if ((error
= xfs_dir2_node_trim_free(args
, fo
, &rval
))) {
1722 fo
-= args
->geo
->fsbcount
;
1727 * Now find the block just before the freespace block.
1729 if ((error
= xfs_bmap_last_before(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1733 * If it's not the single leaf block, give up.
1735 if (XFS_FSB_TO_B(mp
, fo
) > XFS_DIR2_LEAF_OFFSET
+ args
->geo
->blksize
)
1737 lbp
= state
->path
.blk
[0].bp
;
1739 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
1741 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAFN_MAGIC
||
1742 leafhdr
.magic
== XFS_DIR3_LEAFN_MAGIC
);
1745 * Read the freespace block.
1747 error
= xfs_dir2_free_read(tp
, dp
, args
->geo
->freeblk
, &fbp
);
1751 dp
->d_ops
->free_hdr_from_disk(&freehdr
, free
);
1753 ASSERT(!freehdr
.firstdb
);
1756 * Now see if the leafn and free data will fit in a leaf1.
1757 * If not, release the buffer and give up.
1759 if (xfs_dir3_leaf_size(&leafhdr
, freehdr
.nvalid
) > args
->geo
->blksize
) {
1760 xfs_trans_brelse(tp
, fbp
);
1765 * If the leaf has any stale entries in it, compress them out.
1768 xfs_dir3_leaf_compact(args
, &leafhdr
, lbp
);
1770 lbp
->b_ops
= &xfs_dir3_leaf1_buf_ops
;
1771 xfs_trans_buf_set_type(tp
, lbp
, XFS_BLFT_DIR_LEAF1_BUF
);
1772 leafhdr
.magic
= (leafhdr
.magic
== XFS_DIR2_LEAFN_MAGIC
)
1773 ? XFS_DIR2_LEAF1_MAGIC
1774 : XFS_DIR3_LEAF1_MAGIC
;
1777 * Set up the leaf tail from the freespace block.
1779 ltp
= xfs_dir2_leaf_tail_p(args
->geo
, leaf
);
1780 ltp
->bestcount
= cpu_to_be32(freehdr
.nvalid
);
1783 * Set up the leaf bests table.
1785 memcpy(xfs_dir2_leaf_bests_p(ltp
), dp
->d_ops
->free_bests_p(free
),
1786 freehdr
.nvalid
* sizeof(xfs_dir2_data_off_t
));
1788 dp
->d_ops
->leaf_hdr_to_disk(leaf
, &leafhdr
);
1789 xfs_dir3_leaf_log_header(args
, lbp
);
1790 xfs_dir3_leaf_log_bests(args
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1791 xfs_dir3_leaf_log_tail(args
, lbp
);
1792 xfs_dir3_leaf_check(dp
, lbp
);
1795 * Get rid of the freespace block.
1797 error
= xfs_dir2_shrink_inode(args
,
1798 xfs_dir2_byte_to_db(args
->geo
, XFS_DIR2_FREE_OFFSET
),
1802 * This can't fail here because it can only happen when
1803 * punching out the middle of an extent, and this is an
1806 ASSERT(error
!= -ENOSPC
);
1811 * Now see if we can convert the single-leaf directory
1812 * down to a block form directory.
1813 * This routine always kills the dabuf for the leaf, so
1814 * eliminate it from the path.
1816 error
= xfs_dir2_leaf_to_block(args
, lbp
, NULL
);
1817 state
->path
.blk
[0].bp
= NULL
;