2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
23 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
32 #include "xfs_dir2_format.h"
33 #include "xfs_dir2_priv.h"
34 #include "xfs_error.h"
35 #include "xfs_trace.h"
38 * Local function declarations.
41 static void xfs_dir2_leaf_check(struct xfs_inode
*dp
, struct xfs_buf
*bp
);
43 #define xfs_dir2_leaf_check(dp, bp)
45 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t
*args
, struct xfs_buf
**lbpp
,
46 int *indexp
, struct xfs_buf
**dbpp
);
47 static void xfs_dir2_leaf_log_bests(struct xfs_trans
*tp
, struct xfs_buf
*bp
,
49 static void xfs_dir2_leaf_log_tail(struct xfs_trans
*tp
, struct xfs_buf
*bp
);
56 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
57 struct xfs_dir2_leaf_hdr
*hdr
= bp
->b_addr
;
60 block_ok
= hdr
->info
.magic
== magic
;
62 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, hdr
);
63 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
68 xfs_dir2_leaf1_write_verify(
71 xfs_dir2_leaf_verify(bp
, cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
));
75 xfs_dir2_leaf1_read_verify(
78 xfs_dir2_leaf_verify(bp
, cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
));
79 bp
->b_pre_io
= xfs_dir2_leaf1_write_verify
;
85 xfs_dir2_leafn_write_verify(
88 xfs_dir2_leaf_verify(bp
, cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
));
92 xfs_dir2_leafn_read_verify(
95 xfs_dir2_leaf_verify(bp
, cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
));
96 bp
->b_pre_io
= xfs_dir2_leafn_write_verify
;
103 struct xfs_trans
*tp
,
104 struct xfs_inode
*dp
,
106 xfs_daddr_t mappedbno
,
107 struct xfs_buf
**bpp
)
109 return xfs_da_read_buf(tp
, dp
, fbno
, mappedbno
, bpp
,
110 XFS_DATA_FORK
, xfs_dir2_leaf1_read_verify
);
115 struct xfs_trans
*tp
,
116 struct xfs_inode
*dp
,
118 xfs_daddr_t mappedbno
,
119 struct xfs_buf
**bpp
)
121 return xfs_da_read_buf(tp
, dp
, fbno
, mappedbno
, bpp
,
122 XFS_DATA_FORK
, xfs_dir2_leafn_read_verify
);
126 * Convert a block form directory to a leaf form directory.
129 xfs_dir2_block_to_leaf(
130 xfs_da_args_t
*args
, /* operation arguments */
131 struct xfs_buf
*dbp
) /* input block's buffer */
133 __be16
*bestsp
; /* leaf's bestsp entries */
134 xfs_dablk_t blkno
; /* leaf block's bno */
135 xfs_dir2_data_hdr_t
*hdr
; /* block header */
136 xfs_dir2_leaf_entry_t
*blp
; /* block's leaf entries */
137 xfs_dir2_block_tail_t
*btp
; /* block's tail */
138 xfs_inode_t
*dp
; /* incore directory inode */
139 int error
; /* error return code */
140 struct xfs_buf
*lbp
; /* leaf block's buffer */
141 xfs_dir2_db_t ldb
; /* leaf block's bno */
142 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
143 xfs_dir2_leaf_tail_t
*ltp
; /* leaf's tail */
144 xfs_mount_t
*mp
; /* filesystem mount point */
145 int needlog
; /* need to log block header */
146 int needscan
; /* need to rescan bestfree */
147 xfs_trans_t
*tp
; /* transaction pointer */
149 trace_xfs_dir2_block_to_leaf(args
);
155 * Add the leaf block to the inode.
156 * This interface will only put blocks in the leaf/node range.
157 * Since that's empty now, we'll get the root (block 0 in range).
159 if ((error
= xfs_da_grow_inode(args
, &blkno
))) {
162 ldb
= xfs_dir2_da_to_db(mp
, blkno
);
163 ASSERT(ldb
== XFS_DIR2_LEAF_FIRSTDB(mp
));
165 * Initialize the leaf block, get a buffer for it.
167 if ((error
= xfs_dir2_leaf_init(args
, ldb
, &lbp
, XFS_DIR2_LEAF1_MAGIC
))) {
173 xfs_dir2_data_check(dp
, dbp
);
174 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
175 blp
= xfs_dir2_block_leaf_p(btp
);
177 * Set the counts in the leaf header.
179 leaf
->hdr
.count
= cpu_to_be16(be32_to_cpu(btp
->count
));
180 leaf
->hdr
.stale
= cpu_to_be16(be32_to_cpu(btp
->stale
));
182 * Could compact these but I think we always do the conversion
183 * after squeezing out stale entries.
185 memcpy(leaf
->ents
, blp
, be32_to_cpu(btp
->count
) * sizeof(xfs_dir2_leaf_entry_t
));
186 xfs_dir2_leaf_log_ents(tp
, lbp
, 0, be16_to_cpu(leaf
->hdr
.count
) - 1);
190 * Make the space formerly occupied by the leaf entries and block
193 xfs_dir2_data_make_free(tp
, dbp
,
194 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)hdr
),
195 (xfs_dir2_data_aoff_t
)((char *)hdr
+ mp
->m_dirblksize
-
197 &needlog
, &needscan
);
199 * Fix up the block header, make it a data block.
201 hdr
->magic
= cpu_to_be32(XFS_DIR2_DATA_MAGIC
);
203 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
205 * Set up leaf tail and bests table.
207 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
208 ltp
->bestcount
= cpu_to_be32(1);
209 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
210 bestsp
[0] = hdr
->bestfree
[0].length
;
212 * Log the data header and leaf bests table.
215 xfs_dir2_data_log_header(tp
, dbp
);
216 xfs_dir2_leaf_check(dp
, lbp
);
217 xfs_dir2_data_check(dp
, dbp
);
218 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, 0);
223 xfs_dir2_leaf_find_stale(
224 struct xfs_dir2_leaf
*leaf
,
230 * Find the first stale entry before our index, if any.
232 for (*lowstale
= index
- 1; *lowstale
>= 0; --*lowstale
) {
233 if (leaf
->ents
[*lowstale
].address
==
234 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
239 * Find the first stale entry at or after our index, if any.
240 * Stop if the result would require moving more entries than using
243 for (*highstale
= index
;
244 *highstale
< be16_to_cpu(leaf
->hdr
.count
);
246 if (leaf
->ents
[*highstale
].address
==
247 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
249 if (*lowstale
>= 0 && index
- *lowstale
<= *highstale
- index
)
254 struct xfs_dir2_leaf_entry
*
255 xfs_dir2_leaf_find_entry(
256 xfs_dir2_leaf_t
*leaf
, /* leaf structure */
257 int index
, /* leaf table position */
258 int compact
, /* need to compact leaves */
259 int lowstale
, /* index of prev stale leaf */
260 int highstale
, /* index of next stale leaf */
261 int *lfloglow
, /* low leaf logging index */
262 int *lfloghigh
) /* high leaf logging index */
264 if (!leaf
->hdr
.stale
) {
265 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry table pointer */
268 * Now we need to make room to insert the leaf entry.
270 * If there are no stale entries, just insert a hole at index.
272 lep
= &leaf
->ents
[index
];
273 if (index
< be16_to_cpu(leaf
->hdr
.count
))
274 memmove(lep
+ 1, lep
,
275 (be16_to_cpu(leaf
->hdr
.count
) - index
) *
279 * Record low and high logging indices for the leaf.
282 *lfloghigh
= be16_to_cpu(leaf
->hdr
.count
);
283 be16_add_cpu(&leaf
->hdr
.count
, 1);
288 * There are stale entries.
290 * We will use one of them for the new entry. It's probably not at
291 * the right location, so we'll have to shift some up or down first.
293 * If we didn't compact before, we need to find the nearest stale
294 * entries before and after our insertion point.
297 xfs_dir2_leaf_find_stale(leaf
, index
, &lowstale
, &highstale
);
300 * If the low one is better, use it.
303 (highstale
== be16_to_cpu(leaf
->hdr
.count
) ||
304 index
- lowstale
- 1 < highstale
- index
)) {
305 ASSERT(index
- lowstale
- 1 >= 0);
306 ASSERT(leaf
->ents
[lowstale
].address
==
307 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
));
310 * Copy entries up to cover the stale entry and make room
313 if (index
- lowstale
- 1 > 0) {
314 memmove(&leaf
->ents
[lowstale
],
315 &leaf
->ents
[lowstale
+ 1],
316 (index
- lowstale
- 1) *
317 sizeof(xfs_dir2_leaf_entry_t
));
319 *lfloglow
= MIN(lowstale
, *lfloglow
);
320 *lfloghigh
= MAX(index
- 1, *lfloghigh
);
321 be16_add_cpu(&leaf
->hdr
.stale
, -1);
322 return &leaf
->ents
[index
- 1];
326 * The high one is better, so use that one.
328 ASSERT(highstale
- index
>= 0);
329 ASSERT(leaf
->ents
[highstale
].address
==
330 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
));
333 * Copy entries down to cover the stale entry and make room for the
336 if (highstale
- index
> 0) {
337 memmove(&leaf
->ents
[index
+ 1],
339 (highstale
- index
) * sizeof(xfs_dir2_leaf_entry_t
));
341 *lfloglow
= MIN(index
, *lfloglow
);
342 *lfloghigh
= MAX(highstale
, *lfloghigh
);
343 be16_add_cpu(&leaf
->hdr
.stale
, -1);
344 return &leaf
->ents
[index
];
348 * Add an entry to a leaf form directory.
351 xfs_dir2_leaf_addname(
352 xfs_da_args_t
*args
) /* operation arguments */
354 __be16
*bestsp
; /* freespace table in leaf */
355 int compact
; /* need to compact leaves */
356 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
357 struct xfs_buf
*dbp
; /* data block buffer */
358 xfs_dir2_data_entry_t
*dep
; /* data block entry */
359 xfs_inode_t
*dp
; /* incore directory inode */
360 xfs_dir2_data_unused_t
*dup
; /* data unused entry */
361 int error
; /* error return value */
362 int grown
; /* allocated new data block */
363 int highstale
; /* index of next stale leaf */
364 int i
; /* temporary, index */
365 int index
; /* leaf table position */
366 struct xfs_buf
*lbp
; /* leaf's buffer */
367 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
368 int length
; /* length of new entry */
369 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry table pointer */
370 int lfloglow
; /* low leaf logging index */
371 int lfloghigh
; /* high leaf logging index */
372 int lowstale
; /* index of prev stale leaf */
373 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail pointer */
374 xfs_mount_t
*mp
; /* filesystem mount point */
375 int needbytes
; /* leaf block bytes needed */
376 int needlog
; /* need to log data header */
377 int needscan
; /* need to rescan data free */
378 __be16
*tagp
; /* end of data entry */
379 xfs_trans_t
*tp
; /* transaction pointer */
380 xfs_dir2_db_t use_block
; /* data block number */
382 trace_xfs_dir2_leaf_addname(args
);
388 error
= xfs_dir2_leaf_read(tp
, dp
, mp
->m_dirleafblk
, -1, &lbp
);
393 * Look up the entry by hash value and name.
394 * We know it's not there, our caller has already done a lookup.
395 * So the index is of the entry to insert in front of.
396 * But if there are dup hash values the index is of the first of those.
398 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
400 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
401 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
402 length
= xfs_dir2_data_entsize(args
->namelen
);
404 * See if there are any entries with the same hash value
405 * and space in their block for the new entry.
406 * This is good because it puts multiple same-hash value entries
407 * in a data block, improving the lookup of those entries.
409 for (use_block
= -1, lep
= &leaf
->ents
[index
];
410 index
< be16_to_cpu(leaf
->hdr
.count
) && be32_to_cpu(lep
->hashval
) == args
->hashval
;
412 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
414 i
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
415 ASSERT(i
< be32_to_cpu(ltp
->bestcount
));
416 ASSERT(bestsp
[i
] != cpu_to_be16(NULLDATAOFF
));
417 if (be16_to_cpu(bestsp
[i
]) >= length
) {
423 * Didn't find a block yet, linear search all the data blocks.
425 if (use_block
== -1) {
426 for (i
= 0; i
< be32_to_cpu(ltp
->bestcount
); i
++) {
428 * Remember a block we see that's missing.
430 if (bestsp
[i
] == cpu_to_be16(NULLDATAOFF
) &&
433 else if (be16_to_cpu(bestsp
[i
]) >= length
) {
440 * How many bytes do we need in the leaf block?
443 if (!leaf
->hdr
.stale
)
444 needbytes
+= sizeof(xfs_dir2_leaf_entry_t
);
446 needbytes
+= sizeof(xfs_dir2_data_off_t
);
449 * Now kill use_block if it refers to a missing block, so we
450 * can use it as an indication of allocation needed.
452 if (use_block
!= -1 && bestsp
[use_block
] == cpu_to_be16(NULLDATAOFF
))
455 * If we don't have enough free bytes but we can make enough
456 * by compacting out stale entries, we'll do that.
458 if ((char *)bestsp
- (char *)&leaf
->ents
[be16_to_cpu(leaf
->hdr
.count
)] <
459 needbytes
&& be16_to_cpu(leaf
->hdr
.stale
) > 1) {
463 * Otherwise if we don't have enough free bytes we need to
464 * convert to node form.
466 else if ((char *)bestsp
- (char *)&leaf
->ents
[be16_to_cpu(
467 leaf
->hdr
.count
)] < needbytes
) {
469 * Just checking or no space reservation, give up.
471 if ((args
->op_flags
& XFS_DA_OP_JUSTCHECK
) ||
473 xfs_trans_brelse(tp
, lbp
);
474 return XFS_ERROR(ENOSPC
);
477 * Convert to node form.
479 error
= xfs_dir2_leaf_to_node(args
, lbp
);
483 * Then add the new entry.
485 return xfs_dir2_node_addname(args
);
488 * Otherwise it will fit without compaction.
493 * If just checking, then it will fit unless we needed to allocate
496 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
497 xfs_trans_brelse(tp
, lbp
);
498 return use_block
== -1 ? XFS_ERROR(ENOSPC
) : 0;
501 * If no allocations are allowed, return now before we've
504 if (args
->total
== 0 && use_block
== -1) {
505 xfs_trans_brelse(tp
, lbp
);
506 return XFS_ERROR(ENOSPC
);
509 * Need to compact the leaf entries, removing stale ones.
510 * Leave one stale entry behind - the one closest to our
511 * insertion index - and we'll shift that one to our insertion
515 xfs_dir2_leaf_compact_x1(lbp
, &index
, &lowstale
, &highstale
,
516 &lfloglow
, &lfloghigh
);
519 * There are stale entries, so we'll need log-low and log-high
520 * impossibly bad values later.
522 else if (be16_to_cpu(leaf
->hdr
.stale
)) {
523 lfloglow
= be16_to_cpu(leaf
->hdr
.count
);
527 * If there was no data block space found, we need to allocate
530 if (use_block
== -1) {
532 * Add the new data block.
534 if ((error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
,
536 xfs_trans_brelse(tp
, lbp
);
540 * Initialize the block.
542 if ((error
= xfs_dir2_data_init(args
, use_block
, &dbp
))) {
543 xfs_trans_brelse(tp
, lbp
);
547 * If we're adding a new data block on the end we need to
548 * extend the bests table. Copy it up one entry.
550 if (use_block
>= be32_to_cpu(ltp
->bestcount
)) {
552 memmove(&bestsp
[0], &bestsp
[1],
553 be32_to_cpu(ltp
->bestcount
) * sizeof(bestsp
[0]));
554 be32_add_cpu(<p
->bestcount
, 1);
555 xfs_dir2_leaf_log_tail(tp
, lbp
);
556 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
559 * If we're filling in a previously empty block just log it.
562 xfs_dir2_leaf_log_bests(tp
, lbp
, use_block
, use_block
);
564 bestsp
[use_block
] = hdr
->bestfree
[0].length
;
568 * Already had space in some data block.
569 * Just read that one in.
571 error
= xfs_dir2_data_read(tp
, dp
,
572 xfs_dir2_db_to_da(mp
, use_block
),
575 xfs_trans_brelse(tp
, lbp
);
582 * Point to the biggest freespace in our data block.
584 dup
= (xfs_dir2_data_unused_t
*)
585 ((char *)hdr
+ be16_to_cpu(hdr
->bestfree
[0].offset
));
586 ASSERT(be16_to_cpu(dup
->length
) >= length
);
587 needscan
= needlog
= 0;
589 * Mark the initial part of our freespace in use for the new entry.
591 xfs_dir2_data_use_free(tp
, dbp
, dup
,
592 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
), length
,
593 &needlog
, &needscan
);
595 * Initialize our new entry (at last).
597 dep
= (xfs_dir2_data_entry_t
*)dup
;
598 dep
->inumber
= cpu_to_be64(args
->inumber
);
599 dep
->namelen
= args
->namelen
;
600 memcpy(dep
->name
, args
->name
, dep
->namelen
);
601 tagp
= xfs_dir2_data_entry_tag_p(dep
);
602 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
604 * Need to scan fix up the bestfree table.
607 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
609 * Need to log the data block's header.
612 xfs_dir2_data_log_header(tp
, dbp
);
613 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
615 * If the bests table needs to be changed, do it.
616 * Log the change unless we've already done that.
618 if (be16_to_cpu(bestsp
[use_block
]) != be16_to_cpu(hdr
->bestfree
[0].length
)) {
619 bestsp
[use_block
] = hdr
->bestfree
[0].length
;
621 xfs_dir2_leaf_log_bests(tp
, lbp
, use_block
, use_block
);
624 lep
= xfs_dir2_leaf_find_entry(leaf
, index
, compact
, lowstale
,
625 highstale
, &lfloglow
, &lfloghigh
);
628 * Fill in the new leaf entry.
630 lep
->hashval
= cpu_to_be32(args
->hashval
);
631 lep
->address
= cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp
, use_block
,
632 be16_to_cpu(*tagp
)));
634 * Log the leaf fields and give up the buffers.
636 xfs_dir2_leaf_log_header(tp
, lbp
);
637 xfs_dir2_leaf_log_ents(tp
, lbp
, lfloglow
, lfloghigh
);
638 xfs_dir2_leaf_check(dp
, lbp
);
639 xfs_dir2_data_check(dp
, dbp
);
645 * Check the internal consistency of a leaf1 block.
646 * Pop an assert if something is wrong.
650 struct xfs_inode
*dp
, /* incore directory inode */
651 struct xfs_buf
*bp
) /* leaf's buffer */
653 int i
; /* leaf index */
654 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
655 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail pointer */
656 xfs_mount_t
*mp
; /* filesystem mount point */
657 int stale
; /* count of stale leaves */
661 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
));
663 * This value is not restrictive enough.
664 * Should factor in the size of the bests table as well.
665 * We can deduce a value for that from di_size.
667 ASSERT(be16_to_cpu(leaf
->hdr
.count
) <= xfs_dir2_max_leaf_ents(mp
));
668 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
670 * Leaves and bests don't overlap.
672 ASSERT((char *)&leaf
->ents
[be16_to_cpu(leaf
->hdr
.count
)] <=
673 (char *)xfs_dir2_leaf_bests_p(ltp
));
675 * Check hash value order, count stale entries.
677 for (i
= stale
= 0; i
< be16_to_cpu(leaf
->hdr
.count
); i
++) {
678 if (i
+ 1 < be16_to_cpu(leaf
->hdr
.count
))
679 ASSERT(be32_to_cpu(leaf
->ents
[i
].hashval
) <=
680 be32_to_cpu(leaf
->ents
[i
+ 1].hashval
));
681 if (leaf
->ents
[i
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
684 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) == stale
);
689 * Compact out any stale entries in the leaf.
690 * Log the header and changed leaf entries, if any.
693 xfs_dir2_leaf_compact(
694 xfs_da_args_t
*args
, /* operation arguments */
695 struct xfs_buf
*bp
) /* leaf buffer */
697 int from
; /* source leaf index */
698 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
699 int loglow
; /* first leaf entry to log */
700 int to
; /* target leaf index */
703 if (!leaf
->hdr
.stale
) {
707 * Compress out the stale entries in place.
709 for (from
= to
= 0, loglow
= -1; from
< be16_to_cpu(leaf
->hdr
.count
); from
++) {
710 if (leaf
->ents
[from
].address
==
711 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
714 * Only actually copy the entries that are different.
719 leaf
->ents
[to
] = leaf
->ents
[from
];
724 * Update and log the header, log the leaf entries.
726 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) == from
- to
);
727 be16_add_cpu(&leaf
->hdr
.count
, -(be16_to_cpu(leaf
->hdr
.stale
)));
729 xfs_dir2_leaf_log_header(args
->trans
, bp
);
731 xfs_dir2_leaf_log_ents(args
->trans
, bp
, loglow
, to
- 1);
735 * Compact the leaf entries, removing stale ones.
736 * Leave one stale entry behind - the one closest to our
737 * insertion index - and the caller will shift that one to our insertion
739 * Return new insertion index, where the remaining stale entry is,
740 * and leaf logging indices.
743 xfs_dir2_leaf_compact_x1(
744 struct xfs_buf
*bp
, /* leaf buffer */
745 int *indexp
, /* insertion index */
746 int *lowstalep
, /* out: stale entry before us */
747 int *highstalep
, /* out: stale entry after us */
748 int *lowlogp
, /* out: low log index */
749 int *highlogp
) /* out: high log index */
751 int from
; /* source copy index */
752 int highstale
; /* stale entry at/after index */
753 int index
; /* insertion index */
754 int keepstale
; /* source index of kept stale */
755 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
756 int lowstale
; /* stale entry before index */
757 int newindex
=0; /* new insertion index */
758 int to
; /* destination copy index */
761 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) > 1);
764 xfs_dir2_leaf_find_stale(leaf
, index
, &lowstale
, &highstale
);
767 * Pick the better of lowstale and highstale.
770 (highstale
== be16_to_cpu(leaf
->hdr
.count
) ||
771 index
- lowstale
<= highstale
- index
))
772 keepstale
= lowstale
;
774 keepstale
= highstale
;
776 * Copy the entries in place, removing all the stale entries
779 for (from
= to
= 0; from
< be16_to_cpu(leaf
->hdr
.count
); from
++) {
781 * Notice the new value of index.
785 if (from
!= keepstale
&&
786 leaf
->ents
[from
].address
==
787 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
)) {
793 * Record the new keepstale value for the insertion.
795 if (from
== keepstale
)
796 lowstale
= highstale
= to
;
798 * Copy only the entries that have moved.
801 leaf
->ents
[to
] = leaf
->ents
[from
];
806 * If the insertion point was past the last entry,
807 * set the new insertion point accordingly.
813 * Adjust the leaf header values.
815 be16_add_cpu(&leaf
->hdr
.count
, -(from
- to
));
816 leaf
->hdr
.stale
= cpu_to_be16(1);
818 * Remember the low/high stale value only in the "right"
821 if (lowstale
>= newindex
)
824 highstale
= be16_to_cpu(leaf
->hdr
.count
);
825 *highlogp
= be16_to_cpu(leaf
->hdr
.count
) - 1;
826 *lowstalep
= lowstale
;
827 *highstalep
= highstale
;
830 struct xfs_dir2_leaf_map_info
{
831 xfs_extlen_t map_blocks
; /* number of fsbs in map */
832 xfs_dablk_t map_off
; /* last mapped file offset */
833 int map_size
; /* total entries in *map */
834 int map_valid
; /* valid entries in *map */
835 int nmap
; /* mappings to ask xfs_bmapi */
836 xfs_dir2_db_t curdb
; /* db for current block */
837 int ra_current
; /* number of read-ahead blks */
838 int ra_index
; /* *map index for read-ahead */
839 int ra_offset
; /* map entry offset for ra */
840 int ra_want
; /* readahead count wanted */
841 struct xfs_bmbt_irec map
[]; /* map vector for blocks */
845 xfs_dir2_leaf_readbuf(
846 struct xfs_inode
*dp
,
848 struct xfs_dir2_leaf_map_info
*mip
,
849 xfs_dir2_off_t
*curoff
,
850 struct xfs_buf
**bpp
)
852 struct xfs_mount
*mp
= dp
->i_mount
;
853 struct xfs_buf
*bp
= *bpp
;
854 struct xfs_bmbt_irec
*map
= mip
->map
;
861 * If we have a buffer, we need to release it and
862 * take it out of the mapping.
866 xfs_trans_brelse(NULL
, bp
);
868 mip
->map_blocks
-= mp
->m_dirblkfsbs
;
870 * Loop to get rid of the extents for the
873 for (i
= mp
->m_dirblkfsbs
; i
> 0; ) {
874 j
= min_t(int, map
->br_blockcount
, i
);
875 map
->br_blockcount
-= j
;
876 map
->br_startblock
+= j
;
877 map
->br_startoff
+= j
;
879 * If mapping is done, pitch it from
882 if (!map
->br_blockcount
&& --mip
->map_valid
)
883 memmove(&map
[0], &map
[1],
884 sizeof(map
[0]) * mip
->map_valid
);
890 * Recalculate the readahead blocks wanted.
892 mip
->ra_want
= howmany(bufsize
+ mp
->m_dirblksize
,
893 mp
->m_sb
.sb_blocksize
) - 1;
894 ASSERT(mip
->ra_want
>= 0);
897 * If we don't have as many as we want, and we haven't
898 * run out of data blocks, get some more mappings.
900 if (1 + mip
->ra_want
> mip
->map_blocks
&&
901 mip
->map_off
< xfs_dir2_byte_to_da(mp
, XFS_DIR2_LEAF_OFFSET
)) {
903 * Get more bmaps, fill in after the ones
904 * we already have in the table.
906 mip
->nmap
= mip
->map_size
- mip
->map_valid
;
907 error
= xfs_bmapi_read(dp
, mip
->map_off
,
908 xfs_dir2_byte_to_da(mp
, XFS_DIR2_LEAF_OFFSET
) -
910 &map
[mip
->map_valid
], &mip
->nmap
, 0);
913 * Don't know if we should ignore this or try to return an
914 * error. The trouble with returning errors is that readdir
915 * will just stop without actually passing the error through.
921 * If we got all the mappings we asked for, set the final map
922 * offset based on the last bmap value received. Otherwise,
923 * we've reached the end.
925 if (mip
->nmap
== mip
->map_size
- mip
->map_valid
) {
926 i
= mip
->map_valid
+ mip
->nmap
- 1;
927 mip
->map_off
= map
[i
].br_startoff
+ map
[i
].br_blockcount
;
929 mip
->map_off
= xfs_dir2_byte_to_da(mp
,
930 XFS_DIR2_LEAF_OFFSET
);
933 * Look for holes in the mapping, and eliminate them. Count up
936 for (i
= mip
->map_valid
; i
< mip
->map_valid
+ mip
->nmap
; ) {
937 if (map
[i
].br_startblock
== HOLESTARTBLOCK
) {
939 length
= mip
->map_valid
+ mip
->nmap
- i
;
941 memmove(&map
[i
], &map
[i
+ 1],
942 sizeof(map
[i
]) * length
);
944 mip
->map_blocks
+= map
[i
].br_blockcount
;
948 mip
->map_valid
+= mip
->nmap
;
952 * No valid mappings, so no more data blocks.
954 if (!mip
->map_valid
) {
955 *curoff
= xfs_dir2_da_to_byte(mp
, mip
->map_off
);
960 * Read the directory block starting at the first mapping.
962 mip
->curdb
= xfs_dir2_da_to_db(mp
, map
->br_startoff
);
963 error
= xfs_dir2_data_read(NULL
, dp
, map
->br_startoff
,
964 map
->br_blockcount
>= mp
->m_dirblkfsbs
?
965 XFS_FSB_TO_DADDR(mp
, map
->br_startblock
) : -1, &bp
);
968 * Should just skip over the data block instead of giving up.
974 * Adjust the current amount of read-ahead: we just read a block that
978 mip
->ra_current
-= mp
->m_dirblkfsbs
;
981 * Do we need more readahead?
983 for (mip
->ra_index
= mip
->ra_offset
= i
= 0;
984 mip
->ra_want
> mip
->ra_current
&& i
< mip
->map_blocks
;
985 i
+= mp
->m_dirblkfsbs
) {
986 ASSERT(mip
->ra_index
< mip
->map_valid
);
988 * Read-ahead a contiguous directory block.
990 if (i
> mip
->ra_current
&&
991 map
[mip
->ra_index
].br_blockcount
>= mp
->m_dirblkfsbs
) {
992 xfs_dir2_data_readahead(NULL
, dp
,
993 map
[mip
->ra_index
].br_startoff
+ mip
->ra_offset
,
995 map
[mip
->ra_index
].br_startblock
+
1001 * Read-ahead a non-contiguous directory block. This doesn't
1002 * use our mapping, but this is a very rare case.
1004 else if (i
> mip
->ra_current
) {
1005 xfs_dir2_data_readahead(NULL
, dp
,
1006 map
[mip
->ra_index
].br_startoff
+
1007 mip
->ra_offset
, -1);
1008 mip
->ra_current
= i
;
1012 * Advance offset through the mapping table.
1014 for (j
= 0; j
< mp
->m_dirblkfsbs
; j
++) {
1016 * The rest of this extent but not more than a dir
1019 length
= min_t(int, mp
->m_dirblkfsbs
,
1020 map
[mip
->ra_index
].br_blockcount
-
1023 mip
->ra_offset
+= length
;
1026 * Advance to the next mapping if this one is used up.
1028 if (mip
->ra_offset
== map
[mip
->ra_index
].br_blockcount
) {
1041 * Getdents (readdir) for leaf and node directories.
1042 * This reads the data blocks only, so is the same for both forms.
1045 xfs_dir2_leaf_getdents(
1046 xfs_inode_t
*dp
, /* incore directory inode */
1052 struct xfs_buf
*bp
= NULL
; /* data block buffer */
1053 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
1054 xfs_dir2_data_entry_t
*dep
; /* data entry */
1055 xfs_dir2_data_unused_t
*dup
; /* unused entry */
1056 int error
= 0; /* error return value */
1057 int length
; /* temporary length value */
1058 xfs_mount_t
*mp
; /* filesystem mount point */
1059 int byteoff
; /* offset in current block */
1060 xfs_dir2_off_t curoff
; /* current overall offset */
1061 xfs_dir2_off_t newoff
; /* new curoff after new blk */
1062 char *ptr
= NULL
; /* pointer to current data */
1063 struct xfs_dir2_leaf_map_info
*map_info
;
1066 * If the offset is at or past the largest allowed value,
1067 * give up right away.
1069 if (*offset
>= XFS_DIR2_MAX_DATAPTR
)
1075 * Set up to bmap a number of blocks based on the caller's
1076 * buffer size, the directory block size, and the filesystem
1079 length
= howmany(bufsize
+ mp
->m_dirblksize
,
1080 mp
->m_sb
.sb_blocksize
);
1081 map_info
= kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info
, map
) +
1082 (length
* sizeof(struct xfs_bmbt_irec
)),
1084 map_info
->map_size
= length
;
1087 * Inside the loop we keep the main offset value as a byte offset
1088 * in the directory file.
1090 curoff
= xfs_dir2_dataptr_to_byte(mp
, *offset
);
1093 * Force this conversion through db so we truncate the offset
1094 * down to get the start of the data block.
1096 map_info
->map_off
= xfs_dir2_db_to_da(mp
,
1097 xfs_dir2_byte_to_db(mp
, curoff
));
1100 * Loop over directory entries until we reach the end offset.
1101 * Get more blocks and readahead as necessary.
1103 while (curoff
< XFS_DIR2_LEAF_OFFSET
) {
1105 * If we have no buffer, or we're off the end of the
1106 * current buffer, need to get another one.
1108 if (!bp
|| ptr
>= (char *)bp
->b_addr
+ mp
->m_dirblksize
) {
1110 error
= xfs_dir2_leaf_readbuf(dp
, bufsize
, map_info
,
1112 if (error
|| !map_info
->map_valid
)
1116 * Having done a read, we need to set a new offset.
1118 newoff
= xfs_dir2_db_off_to_byte(mp
, map_info
->curdb
, 0);
1120 * Start of the current block.
1122 if (curoff
< newoff
)
1125 * Make sure we're in the right block.
1127 else if (curoff
> newoff
)
1128 ASSERT(xfs_dir2_byte_to_db(mp
, curoff
) ==
1131 xfs_dir2_data_check(dp
, bp
);
1133 * Find our position in the block.
1135 ptr
= (char *)(hdr
+ 1);
1136 byteoff
= xfs_dir2_byte_to_off(mp
, curoff
);
1138 * Skip past the header.
1141 curoff
+= (uint
)sizeof(*hdr
);
1143 * Skip past entries until we reach our offset.
1146 while ((char *)ptr
- (char *)hdr
< byteoff
) {
1147 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1149 if (be16_to_cpu(dup
->freetag
)
1150 == XFS_DIR2_DATA_FREE_TAG
) {
1152 length
= be16_to_cpu(dup
->length
);
1156 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1158 xfs_dir2_data_entsize(dep
->namelen
);
1162 * Now set our real offset.
1165 xfs_dir2_db_off_to_byte(mp
,
1166 xfs_dir2_byte_to_db(mp
, curoff
),
1167 (char *)ptr
- (char *)hdr
);
1168 if (ptr
>= (char *)hdr
+ mp
->m_dirblksize
) {
1174 * We have a pointer to an entry.
1177 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1179 * No, it's unused, skip over it.
1181 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
1182 length
= be16_to_cpu(dup
->length
);
1188 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1189 length
= xfs_dir2_data_entsize(dep
->namelen
);
1191 if (filldir(dirent
, (char *)dep
->name
, dep
->namelen
,
1192 xfs_dir2_byte_to_dataptr(mp
, curoff
) & 0x7fffffff,
1193 be64_to_cpu(dep
->inumber
), DT_UNKNOWN
))
1197 * Advance to next entry in the block.
1201 /* bufsize may have just been a guess; don't go negative */
1202 bufsize
= bufsize
> length
? bufsize
- length
: 0;
1206 * All done. Set output offset value to current offset.
1208 if (curoff
> xfs_dir2_dataptr_to_byte(mp
, XFS_DIR2_MAX_DATAPTR
))
1209 *offset
= XFS_DIR2_MAX_DATAPTR
& 0x7fffffff;
1211 *offset
= xfs_dir2_byte_to_dataptr(mp
, curoff
) & 0x7fffffff;
1212 kmem_free(map_info
);
1214 xfs_trans_brelse(NULL
, bp
);
1219 * Initialize a new leaf block, leaf1 or leafn magic accepted.
1223 xfs_da_args_t
*args
, /* operation arguments */
1224 xfs_dir2_db_t bno
, /* directory block number */
1225 struct xfs_buf
**bpp
, /* out: leaf buffer */
1226 int magic
) /* magic number for block */
1228 struct xfs_buf
*bp
; /* leaf buffer */
1229 xfs_inode_t
*dp
; /* incore directory inode */
1230 int error
; /* error return code */
1231 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1232 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1233 xfs_mount_t
*mp
; /* filesystem mount point */
1234 xfs_trans_t
*tp
; /* transaction pointer */
1240 ASSERT(bno
>= XFS_DIR2_LEAF_FIRSTDB(mp
) &&
1241 bno
< XFS_DIR2_FREE_FIRSTDB(mp
));
1243 * Get the buffer for the block.
1245 error
= xfs_da_get_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, bno
), -1, &bp
,
1253 * Initialize the header.
1255 leaf
->hdr
.info
.magic
= cpu_to_be16(magic
);
1256 leaf
->hdr
.info
.forw
= 0;
1257 leaf
->hdr
.info
.back
= 0;
1258 leaf
->hdr
.count
= 0;
1259 leaf
->hdr
.stale
= 0;
1260 xfs_dir2_leaf_log_header(tp
, bp
);
1262 * If it's a leaf-format directory initialize the tail.
1263 * In this case our caller has the real bests table to copy into
1266 if (magic
== XFS_DIR2_LEAF1_MAGIC
) {
1267 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1269 xfs_dir2_leaf_log_tail(tp
, bp
);
1276 * Log the bests entries indicated from a leaf1 block.
1279 xfs_dir2_leaf_log_bests(
1280 xfs_trans_t
*tp
, /* transaction pointer */
1281 struct xfs_buf
*bp
, /* leaf buffer */
1282 int first
, /* first entry to log */
1283 int last
) /* last entry to log */
1285 __be16
*firstb
; /* pointer to first entry */
1286 __be16
*lastb
; /* pointer to last entry */
1287 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1288 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1291 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
));
1292 ltp
= xfs_dir2_leaf_tail_p(tp
->t_mountp
, leaf
);
1293 firstb
= xfs_dir2_leaf_bests_p(ltp
) + first
;
1294 lastb
= xfs_dir2_leaf_bests_p(ltp
) + last
;
1295 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)firstb
- (char *)leaf
),
1296 (uint
)((char *)lastb
- (char *)leaf
+ sizeof(*lastb
) - 1));
1300 * Log the leaf entries indicated from a leaf1 or leafn block.
1303 xfs_dir2_leaf_log_ents(
1304 xfs_trans_t
*tp
, /* transaction pointer */
1305 struct xfs_buf
*bp
, /* leaf buffer */
1306 int first
, /* first entry to log */
1307 int last
) /* last entry to log */
1309 xfs_dir2_leaf_entry_t
*firstlep
; /* pointer to first entry */
1310 xfs_dir2_leaf_entry_t
*lastlep
; /* pointer to last entry */
1311 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1314 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
1315 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
));
1316 firstlep
= &leaf
->ents
[first
];
1317 lastlep
= &leaf
->ents
[last
];
1318 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)firstlep
- (char *)leaf
),
1319 (uint
)((char *)lastlep
- (char *)leaf
+ sizeof(*lastlep
) - 1));
1323 * Log the header of the leaf1 or leafn block.
1326 xfs_dir2_leaf_log_header(
1327 struct xfs_trans
*tp
,
1330 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1333 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
1334 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
));
1335 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)&leaf
->hdr
- (char *)leaf
),
1336 (uint
)(sizeof(leaf
->hdr
) - 1));
1340 * Log the tail of the leaf1 block.
1343 xfs_dir2_leaf_log_tail(
1344 struct xfs_trans
*tp
,
1347 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1348 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1349 xfs_mount_t
*mp
; /* filesystem mount point */
1353 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
));
1354 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1355 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)ltp
- (char *)leaf
),
1356 (uint
)(mp
->m_dirblksize
- 1));
1360 * Look up the entry referred to by args in the leaf format directory.
1361 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1362 * is also used by the node-format code.
1365 xfs_dir2_leaf_lookup(
1366 xfs_da_args_t
*args
) /* operation arguments */
1368 struct xfs_buf
*dbp
; /* data block buffer */
1369 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1370 xfs_inode_t
*dp
; /* incore directory inode */
1371 int error
; /* error return code */
1372 int index
; /* found entry index */
1373 struct xfs_buf
*lbp
; /* leaf buffer */
1374 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1375 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1376 xfs_trans_t
*tp
; /* transaction pointer */
1378 trace_xfs_dir2_leaf_lookup(args
);
1381 * Look up name in the leaf block, returning both buffers and index.
1383 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1388 xfs_dir2_leaf_check(dp
, lbp
);
1391 * Get to the leaf entry and contained data entry address.
1393 lep
= &leaf
->ents
[index
];
1395 * Point to the data entry.
1397 dep
= (xfs_dir2_data_entry_t
*)
1398 ((char *)dbp
->b_addr
+
1399 xfs_dir2_dataptr_to_off(dp
->i_mount
, be32_to_cpu(lep
->address
)));
1401 * Return the found inode number & CI name if appropriate
1403 args
->inumber
= be64_to_cpu(dep
->inumber
);
1404 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
1405 xfs_trans_brelse(tp
, dbp
);
1406 xfs_trans_brelse(tp
, lbp
);
1407 return XFS_ERROR(error
);
1411 * Look up name/hash in the leaf block.
1412 * Fill in indexp with the found index, and dbpp with the data buffer.
1413 * If not found dbpp will be NULL, and ENOENT comes back.
1414 * lbpp will always be filled in with the leaf buffer unless there's an error.
1416 static int /* error */
1417 xfs_dir2_leaf_lookup_int(
1418 xfs_da_args_t
*args
, /* operation arguments */
1419 struct xfs_buf
**lbpp
, /* out: leaf buffer */
1420 int *indexp
, /* out: index in leaf block */
1421 struct xfs_buf
**dbpp
) /* out: data buffer */
1423 xfs_dir2_db_t curdb
= -1; /* current data block number */
1424 struct xfs_buf
*dbp
= NULL
; /* data buffer */
1425 xfs_dir2_data_entry_t
*dep
; /* data entry */
1426 xfs_inode_t
*dp
; /* incore directory inode */
1427 int error
; /* error return code */
1428 int index
; /* index in leaf block */
1429 struct xfs_buf
*lbp
; /* leaf buffer */
1430 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1431 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1432 xfs_mount_t
*mp
; /* filesystem mount point */
1433 xfs_dir2_db_t newdb
; /* new data block number */
1434 xfs_trans_t
*tp
; /* transaction pointer */
1435 xfs_dir2_db_t cidb
= -1; /* case match data block no. */
1436 enum xfs_dacmp cmp
; /* name compare result */
1442 error
= xfs_dir2_leaf_read(tp
, dp
, mp
->m_dirleafblk
, -1, &lbp
);
1448 xfs_dir2_leaf_check(dp
, lbp
);
1450 * Look for the first leaf entry with our hash value.
1452 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
1454 * Loop over all the entries with the right hash value
1455 * looking to match the name.
1457 for (lep
= &leaf
->ents
[index
]; index
< be16_to_cpu(leaf
->hdr
.count
) &&
1458 be32_to_cpu(lep
->hashval
) == args
->hashval
;
1461 * Skip over stale leaf entries.
1463 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
1466 * Get the new data block number.
1468 newdb
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
1470 * If it's not the same as the old data block number,
1471 * need to pitch the old one and read the new one.
1473 if (newdb
!= curdb
) {
1475 xfs_trans_brelse(tp
, dbp
);
1476 error
= xfs_dir2_data_read(tp
, dp
,
1477 xfs_dir2_db_to_da(mp
, newdb
),
1480 xfs_trans_brelse(tp
, lbp
);
1486 * Point to the data entry.
1488 dep
= (xfs_dir2_data_entry_t
*)((char *)dbp
->b_addr
+
1489 xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
1491 * Compare name and if it's an exact match, return the index
1492 * and buffer. If it's the first case-insensitive match, store
1493 * the index and buffer and continue looking for an exact match.
1495 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
1496 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
1497 args
->cmpresult
= cmp
;
1499 /* case exact match: return the current buffer. */
1500 if (cmp
== XFS_CMP_EXACT
) {
1507 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
1509 * Here, we can only be doing a lookup (not a rename or remove).
1510 * If a case-insensitive match was found earlier, re-read the
1511 * appropriate data block if required and return it.
1513 if (args
->cmpresult
== XFS_CMP_CASE
) {
1515 if (cidb
!= curdb
) {
1516 xfs_trans_brelse(tp
, dbp
);
1517 error
= xfs_dir2_data_read(tp
, dp
,
1518 xfs_dir2_db_to_da(mp
, cidb
),
1521 xfs_trans_brelse(tp
, lbp
);
1529 * No match found, return ENOENT.
1533 xfs_trans_brelse(tp
, dbp
);
1534 xfs_trans_brelse(tp
, lbp
);
1535 return XFS_ERROR(ENOENT
);
1539 * Remove an entry from a leaf format directory.
1542 xfs_dir2_leaf_removename(
1543 xfs_da_args_t
*args
) /* operation arguments */
1545 __be16
*bestsp
; /* leaf block best freespace */
1546 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
1547 xfs_dir2_db_t db
; /* data block number */
1548 struct xfs_buf
*dbp
; /* data block buffer */
1549 xfs_dir2_data_entry_t
*dep
; /* data entry structure */
1550 xfs_inode_t
*dp
; /* incore directory inode */
1551 int error
; /* error return code */
1552 xfs_dir2_db_t i
; /* temporary data block # */
1553 int index
; /* index into leaf entries */
1554 struct xfs_buf
*lbp
; /* leaf buffer */
1555 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1556 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1557 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1558 xfs_mount_t
*mp
; /* filesystem mount point */
1559 int needlog
; /* need to log data header */
1560 int needscan
; /* need to rescan data frees */
1561 xfs_dir2_data_off_t oldbest
; /* old value of best free */
1562 xfs_trans_t
*tp
; /* transaction pointer */
1564 trace_xfs_dir2_leaf_removename(args
);
1567 * Lookup the leaf entry, get the leaf and data blocks read in.
1569 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1577 xfs_dir2_data_check(dp
, dbp
);
1579 * Point to the leaf entry, use that to point to the data entry.
1581 lep
= &leaf
->ents
[index
];
1582 db
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
1583 dep
= (xfs_dir2_data_entry_t
*)
1584 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
1585 needscan
= needlog
= 0;
1586 oldbest
= be16_to_cpu(hdr
->bestfree
[0].length
);
1587 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1588 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1589 ASSERT(be16_to_cpu(bestsp
[db
]) == oldbest
);
1591 * Mark the former data entry unused.
1593 xfs_dir2_data_make_free(tp
, dbp
,
1594 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)hdr
),
1595 xfs_dir2_data_entsize(dep
->namelen
), &needlog
, &needscan
);
1597 * We just mark the leaf entry stale by putting a null in it.
1599 be16_add_cpu(&leaf
->hdr
.stale
, 1);
1600 xfs_dir2_leaf_log_header(tp
, lbp
);
1601 lep
->address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
1602 xfs_dir2_leaf_log_ents(tp
, lbp
, index
, index
);
1604 * Scan the freespace in the data block again if necessary,
1605 * log the data block header if necessary.
1608 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
1610 xfs_dir2_data_log_header(tp
, dbp
);
1612 * If the longest freespace in the data block has changed,
1613 * put the new value in the bests table and log that.
1615 if (be16_to_cpu(hdr
->bestfree
[0].length
) != oldbest
) {
1616 bestsp
[db
] = hdr
->bestfree
[0].length
;
1617 xfs_dir2_leaf_log_bests(tp
, lbp
, db
, db
);
1619 xfs_dir2_data_check(dp
, dbp
);
1621 * If the data block is now empty then get rid of the data block.
1623 if (be16_to_cpu(hdr
->bestfree
[0].length
) ==
1624 mp
->m_dirblksize
- (uint
)sizeof(*hdr
)) {
1625 ASSERT(db
!= mp
->m_dirdatablk
);
1626 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1628 * Nope, can't get rid of it because it caused
1629 * allocation of a bmap btree block to do so.
1630 * Just go on, returning success, leaving the
1631 * empty block in place.
1633 if (error
== ENOSPC
&& args
->total
== 0)
1635 xfs_dir2_leaf_check(dp
, lbp
);
1640 * If this is the last data block then compact the
1641 * bests table by getting rid of entries.
1643 if (db
== be32_to_cpu(ltp
->bestcount
) - 1) {
1645 * Look for the last active entry (i).
1647 for (i
= db
- 1; i
> 0; i
--) {
1648 if (bestsp
[i
] != cpu_to_be16(NULLDATAOFF
))
1652 * Copy the table down so inactive entries at the
1655 memmove(&bestsp
[db
- i
], bestsp
,
1656 (be32_to_cpu(ltp
->bestcount
) - (db
- i
)) * sizeof(*bestsp
));
1657 be32_add_cpu(<p
->bestcount
, -(db
- i
));
1658 xfs_dir2_leaf_log_tail(tp
, lbp
);
1659 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1661 bestsp
[db
] = cpu_to_be16(NULLDATAOFF
);
1664 * If the data block was not the first one, drop it.
1666 else if (db
!= mp
->m_dirdatablk
)
1669 xfs_dir2_leaf_check(dp
, lbp
);
1671 * See if we can convert to block form.
1673 return xfs_dir2_leaf_to_block(args
, lbp
, dbp
);
1677 * Replace the inode number in a leaf format directory entry.
1680 xfs_dir2_leaf_replace(
1681 xfs_da_args_t
*args
) /* operation arguments */
1683 struct xfs_buf
*dbp
; /* data block buffer */
1684 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1685 xfs_inode_t
*dp
; /* incore directory inode */
1686 int error
; /* error return code */
1687 int index
; /* index of leaf entry */
1688 struct xfs_buf
*lbp
; /* leaf buffer */
1689 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1690 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1691 xfs_trans_t
*tp
; /* transaction pointer */
1693 trace_xfs_dir2_leaf_replace(args
);
1696 * Look up the entry.
1698 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1704 * Point to the leaf entry, get data address from it.
1706 lep
= &leaf
->ents
[index
];
1708 * Point to the data entry.
1710 dep
= (xfs_dir2_data_entry_t
*)
1711 ((char *)dbp
->b_addr
+
1712 xfs_dir2_dataptr_to_off(dp
->i_mount
, be32_to_cpu(lep
->address
)));
1713 ASSERT(args
->inumber
!= be64_to_cpu(dep
->inumber
));
1715 * Put the new inode number in, log it.
1717 dep
->inumber
= cpu_to_be64(args
->inumber
);
1719 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
1720 xfs_dir2_leaf_check(dp
, lbp
);
1721 xfs_trans_brelse(tp
, lbp
);
1726 * Return index in the leaf block (lbp) which is either the first
1727 * one with this hash value, or if there are none, the insert point
1728 * for that hash value.
1730 int /* index value */
1731 xfs_dir2_leaf_search_hash(
1732 xfs_da_args_t
*args
, /* operation arguments */
1733 struct xfs_buf
*lbp
) /* leaf buffer */
1735 xfs_dahash_t hash
=0; /* hash from this entry */
1736 xfs_dahash_t hashwant
; /* hash value looking for */
1737 int high
; /* high leaf index */
1738 int low
; /* low leaf index */
1739 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1740 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1741 int mid
=0; /* current leaf index */
1745 if (!leaf
->hdr
.count
)
1749 * Note, the table cannot be empty, so we have to go through the loop.
1750 * Binary search the leaf entries looking for our hash value.
1752 for (lep
= leaf
->ents
, low
= 0, high
= be16_to_cpu(leaf
->hdr
.count
) - 1,
1753 hashwant
= args
->hashval
;
1755 mid
= (low
+ high
) >> 1;
1756 if ((hash
= be32_to_cpu(lep
[mid
].hashval
)) == hashwant
)
1758 if (hash
< hashwant
)
1764 * Found one, back up through all the equal hash values.
1766 if (hash
== hashwant
) {
1767 while (mid
> 0 && be32_to_cpu(lep
[mid
- 1].hashval
) == hashwant
) {
1772 * Need to point to an entry higher than ours.
1774 else if (hash
< hashwant
)
1780 * Trim off a trailing data block. We know it's empty since the leaf
1781 * freespace table says so.
1784 xfs_dir2_leaf_trim_data(
1785 xfs_da_args_t
*args
, /* operation arguments */
1786 struct xfs_buf
*lbp
, /* leaf buffer */
1787 xfs_dir2_db_t db
) /* data block number */
1789 __be16
*bestsp
; /* leaf bests table */
1790 struct xfs_buf
*dbp
; /* data block buffer */
1791 xfs_inode_t
*dp
; /* incore directory inode */
1792 int error
; /* error return value */
1793 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1794 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1795 xfs_mount_t
*mp
; /* filesystem mount point */
1796 xfs_trans_t
*tp
; /* transaction pointer */
1802 * Read the offending data block. We need its buffer.
1804 error
= xfs_dir2_data_read(tp
, dp
, xfs_dir2_db_to_da(mp
, db
), -1, &dbp
);
1809 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1813 struct xfs_dir2_data_hdr
*hdr
= dbp
->b_addr
;
1815 ASSERT(hdr
->magic
== cpu_to_be32(XFS_DIR2_DATA_MAGIC
));
1816 ASSERT(be16_to_cpu(hdr
->bestfree
[0].length
) ==
1817 mp
->m_dirblksize
- (uint
)sizeof(*hdr
));
1818 ASSERT(db
== be32_to_cpu(ltp
->bestcount
) - 1);
1823 * Get rid of the data block.
1825 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1826 ASSERT(error
!= ENOSPC
);
1827 xfs_trans_brelse(tp
, dbp
);
1831 * Eliminate the last bests entry from the table.
1833 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1834 be32_add_cpu(<p
->bestcount
, -1);
1835 memmove(&bestsp
[1], &bestsp
[0], be32_to_cpu(ltp
->bestcount
) * sizeof(*bestsp
));
1836 xfs_dir2_leaf_log_tail(tp
, lbp
);
1837 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1841 static inline size_t
1843 struct xfs_dir2_leaf_hdr
*hdr
,
1848 entries
= be16_to_cpu(hdr
->count
) - be16_to_cpu(hdr
->stale
);
1849 return sizeof(xfs_dir2_leaf_hdr_t
) +
1850 entries
* sizeof(xfs_dir2_leaf_entry_t
) +
1851 counts
* sizeof(xfs_dir2_data_off_t
) +
1852 sizeof(xfs_dir2_leaf_tail_t
);
1856 * Convert node form directory to leaf form directory.
1857 * The root of the node form dir needs to already be a LEAFN block.
1858 * Just return if we can't do anything.
1861 xfs_dir2_node_to_leaf(
1862 xfs_da_state_t
*state
) /* directory operation state */
1864 xfs_da_args_t
*args
; /* operation arguments */
1865 xfs_inode_t
*dp
; /* incore directory inode */
1866 int error
; /* error return code */
1867 struct xfs_buf
*fbp
; /* buffer for freespace block */
1868 xfs_fileoff_t fo
; /* freespace file offset */
1869 xfs_dir2_free_t
*free
; /* freespace structure */
1870 struct xfs_buf
*lbp
; /* buffer for leaf block */
1871 xfs_dir2_leaf_tail_t
*ltp
; /* tail of leaf structure */
1872 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1873 xfs_mount_t
*mp
; /* filesystem mount point */
1874 int rval
; /* successful free trim? */
1875 xfs_trans_t
*tp
; /* transaction pointer */
1878 * There's more than a leaf level in the btree, so there must
1879 * be multiple leafn blocks. Give up.
1881 if (state
->path
.active
> 1)
1885 trace_xfs_dir2_node_to_leaf(args
);
1891 * Get the last offset in the file.
1893 if ((error
= xfs_bmap_last_offset(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1896 fo
-= mp
->m_dirblkfsbs
;
1898 * If there are freespace blocks other than the first one,
1899 * take this opportunity to remove trailing empty freespace blocks
1900 * that may have been left behind during no-space-reservation
1903 while (fo
> mp
->m_dirfreeblk
) {
1904 if ((error
= xfs_dir2_node_trim_free(args
, fo
, &rval
))) {
1908 fo
-= mp
->m_dirblkfsbs
;
1913 * Now find the block just before the freespace block.
1915 if ((error
= xfs_bmap_last_before(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1919 * If it's not the single leaf block, give up.
1921 if (XFS_FSB_TO_B(mp
, fo
) > XFS_DIR2_LEAF_OFFSET
+ mp
->m_dirblksize
)
1923 lbp
= state
->path
.blk
[0].bp
;
1925 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
));
1927 * Read the freespace block.
1929 error
= xfs_dir2_free_read(tp
, dp
, mp
->m_dirfreeblk
, &fbp
);
1933 ASSERT(free
->hdr
.magic
== cpu_to_be32(XFS_DIR2_FREE_MAGIC
));
1934 ASSERT(!free
->hdr
.firstdb
);
1937 * Now see if the leafn and free data will fit in a leaf1.
1938 * If not, release the buffer and give up.
1940 if (xfs_dir2_leaf_size(&leaf
->hdr
, be32_to_cpu(free
->hdr
.nvalid
)) >
1942 xfs_trans_brelse(tp
, fbp
);
1947 * If the leaf has any stale entries in it, compress them out.
1948 * The compact routine will log the header.
1950 if (be16_to_cpu(leaf
->hdr
.stale
))
1951 xfs_dir2_leaf_compact(args
, lbp
);
1953 xfs_dir2_leaf_log_header(tp
, lbp
);
1954 leaf
->hdr
.info
.magic
= cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
);
1956 * Set up the leaf tail from the freespace block.
1958 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1959 ltp
->bestcount
= free
->hdr
.nvalid
;
1961 * Set up the leaf bests table.
1963 memcpy(xfs_dir2_leaf_bests_p(ltp
), free
->bests
,
1964 be32_to_cpu(ltp
->bestcount
) * sizeof(xfs_dir2_data_off_t
));
1965 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1966 xfs_dir2_leaf_log_tail(tp
, lbp
);
1967 xfs_dir2_leaf_check(dp
, lbp
);
1969 * Get rid of the freespace block.
1971 error
= xfs_dir2_shrink_inode(args
, XFS_DIR2_FREE_FIRSTDB(mp
), fbp
);
1974 * This can't fail here because it can only happen when
1975 * punching out the middle of an extent, and this is an
1978 ASSERT(error
!= ENOSPC
);
1983 * Now see if we can convert the single-leaf directory
1984 * down to a block form directory.
1985 * This routine always kills the dabuf for the leaf, so
1986 * eliminate it from the path.
1988 error
= xfs_dir2_leaf_to_block(args
, lbp
, NULL
);
1989 state
->path
.blk
[0].bp
= NULL
;