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"
24 #include "xfs_trans.h"
28 #include "xfs_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_da_btree.h"
31 #include "xfs_bmap_btree.h"
32 #include "xfs_attr_sf.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
37 #include "xfs_dir2_data.h"
38 #include "xfs_dir2_leaf.h"
39 #include "xfs_dir2_block.h"
40 #include "xfs_dir2_node.h"
41 #include "xfs_dir2_trace.h"
42 #include "xfs_error.h"
45 * Local function declarations.
48 static void xfs_dir2_leaf_check(xfs_inode_t
*dp
, xfs_dabuf_t
*bp
);
50 #define xfs_dir2_leaf_check(dp, bp)
52 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t
*args
, xfs_dabuf_t
**lbpp
,
53 int *indexp
, xfs_dabuf_t
**dbpp
);
54 static void xfs_dir2_leaf_log_bests(struct xfs_trans
*tp
, struct xfs_dabuf
*bp
,
56 static void xfs_dir2_leaf_log_tail(struct xfs_trans
*tp
, struct xfs_dabuf
*bp
);
60 * Convert a block form directory to a leaf form directory.
63 xfs_dir2_block_to_leaf(
64 xfs_da_args_t
*args
, /* operation arguments */
65 xfs_dabuf_t
*dbp
) /* input block's buffer */
67 __be16
*bestsp
; /* leaf's bestsp entries */
68 xfs_dablk_t blkno
; /* leaf block's bno */
69 xfs_dir2_block_t
*block
; /* block structure */
70 xfs_dir2_leaf_entry_t
*blp
; /* block's leaf entries */
71 xfs_dir2_block_tail_t
*btp
; /* block's tail */
72 xfs_inode_t
*dp
; /* incore directory inode */
73 int error
; /* error return code */
74 xfs_dabuf_t
*lbp
; /* leaf block's buffer */
75 xfs_dir2_db_t ldb
; /* leaf block's bno */
76 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
77 xfs_dir2_leaf_tail_t
*ltp
; /* leaf's tail */
78 xfs_mount_t
*mp
; /* filesystem mount point */
79 int needlog
; /* need to log block header */
80 int needscan
; /* need to rescan bestfree */
81 xfs_trans_t
*tp
; /* transaction pointer */
83 xfs_dir2_trace_args_b("block_to_leaf", args
, dbp
);
88 * Add the leaf block to the inode.
89 * This interface will only put blocks in the leaf/node range.
90 * Since that's empty now, we'll get the root (block 0 in range).
92 if ((error
= xfs_da_grow_inode(args
, &blkno
))) {
95 ldb
= xfs_dir2_da_to_db(mp
, blkno
);
96 ASSERT(ldb
== XFS_DIR2_LEAF_FIRSTDB(mp
));
98 * Initialize the leaf block, get a buffer for it.
100 if ((error
= xfs_dir2_leaf_init(args
, ldb
, &lbp
, XFS_DIR2_LEAF1_MAGIC
))) {
106 xfs_dir2_data_check(dp
, dbp
);
107 btp
= xfs_dir2_block_tail_p(mp
, block
);
108 blp
= xfs_dir2_block_leaf_p(btp
);
110 * Set the counts in the leaf header.
112 leaf
->hdr
.count
= cpu_to_be16(be32_to_cpu(btp
->count
));
113 leaf
->hdr
.stale
= cpu_to_be16(be32_to_cpu(btp
->stale
));
115 * Could compact these but I think we always do the conversion
116 * after squeezing out stale entries.
118 memcpy(leaf
->ents
, blp
, be32_to_cpu(btp
->count
) * sizeof(xfs_dir2_leaf_entry_t
));
119 xfs_dir2_leaf_log_ents(tp
, lbp
, 0, be16_to_cpu(leaf
->hdr
.count
) - 1);
123 * Make the space formerly occupied by the leaf entries and block
126 xfs_dir2_data_make_free(tp
, dbp
,
127 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)block
),
128 (xfs_dir2_data_aoff_t
)((char *)block
+ mp
->m_dirblksize
-
130 &needlog
, &needscan
);
132 * Fix up the block header, make it a data block.
134 block
->hdr
.magic
= cpu_to_be32(XFS_DIR2_DATA_MAGIC
);
136 xfs_dir2_data_freescan(mp
, (xfs_dir2_data_t
*)block
, &needlog
);
138 * Set up leaf tail and bests table.
140 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
141 ltp
->bestcount
= cpu_to_be32(1);
142 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
143 bestsp
[0] = block
->hdr
.bestfree
[0].length
;
145 * Log the data header and leaf bests table.
148 xfs_dir2_data_log_header(tp
, dbp
);
149 xfs_dir2_leaf_check(dp
, lbp
);
150 xfs_dir2_data_check(dp
, dbp
);
151 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, 0);
152 xfs_da_buf_done(lbp
);
157 * Add an entry to a leaf form directory.
160 xfs_dir2_leaf_addname(
161 xfs_da_args_t
*args
) /* operation arguments */
163 __be16
*bestsp
; /* freespace table in leaf */
164 int compact
; /* need to compact leaves */
165 xfs_dir2_data_t
*data
; /* data block structure */
166 xfs_dabuf_t
*dbp
; /* data block buffer */
167 xfs_dir2_data_entry_t
*dep
; /* data block entry */
168 xfs_inode_t
*dp
; /* incore directory inode */
169 xfs_dir2_data_unused_t
*dup
; /* data unused entry */
170 int error
; /* error return value */
171 int grown
; /* allocated new data block */
172 int highstale
; /* index of next stale leaf */
173 int i
; /* temporary, index */
174 int index
; /* leaf table position */
175 xfs_dabuf_t
*lbp
; /* leaf's buffer */
176 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
177 int length
; /* length of new entry */
178 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry table pointer */
179 int lfloglow
; /* low leaf logging index */
180 int lfloghigh
; /* high leaf logging index */
181 int lowstale
; /* index of prev stale leaf */
182 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail pointer */
183 xfs_mount_t
*mp
; /* filesystem mount point */
184 int needbytes
; /* leaf block bytes needed */
185 int needlog
; /* need to log data header */
186 int needscan
; /* need to rescan data free */
187 __be16
*tagp
; /* end of data entry */
188 xfs_trans_t
*tp
; /* transaction pointer */
189 xfs_dir2_db_t use_block
; /* data block number */
191 xfs_dir2_trace_args("leaf_addname", args
);
196 * Read the leaf block.
198 error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirleafblk
, -1, &lbp
,
205 * Look up the entry by hash value and name.
206 * We know it's not there, our caller has already done a lookup.
207 * So the index is of the entry to insert in front of.
208 * But if there are dup hash values the index is of the first of those.
210 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
212 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
213 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
214 length
= xfs_dir2_data_entsize(args
->namelen
);
216 * See if there are any entries with the same hash value
217 * and space in their block for the new entry.
218 * This is good because it puts multiple same-hash value entries
219 * in a data block, improving the lookup of those entries.
221 for (use_block
= -1, lep
= &leaf
->ents
[index
];
222 index
< be16_to_cpu(leaf
->hdr
.count
) && be32_to_cpu(lep
->hashval
) == args
->hashval
;
224 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
226 i
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
227 ASSERT(i
< be32_to_cpu(ltp
->bestcount
));
228 ASSERT(be16_to_cpu(bestsp
[i
]) != NULLDATAOFF
);
229 if (be16_to_cpu(bestsp
[i
]) >= length
) {
235 * Didn't find a block yet, linear search all the data blocks.
237 if (use_block
== -1) {
238 for (i
= 0; i
< be32_to_cpu(ltp
->bestcount
); i
++) {
240 * Remember a block we see that's missing.
242 if (be16_to_cpu(bestsp
[i
]) == NULLDATAOFF
&& use_block
== -1)
244 else if (be16_to_cpu(bestsp
[i
]) >= length
) {
251 * How many bytes do we need in the leaf block?
254 (leaf
->hdr
.stale
? 0 : (uint
)sizeof(leaf
->ents
[0])) +
255 (use_block
!= -1 ? 0 : (uint
)sizeof(leaf
->bests
[0]));
257 * Now kill use_block if it refers to a missing block, so we
258 * can use it as an indication of allocation needed.
260 if (use_block
!= -1 && be16_to_cpu(bestsp
[use_block
]) == NULLDATAOFF
)
263 * If we don't have enough free bytes but we can make enough
264 * by compacting out stale entries, we'll do that.
266 if ((char *)bestsp
- (char *)&leaf
->ents
[be16_to_cpu(leaf
->hdr
.count
)] <
267 needbytes
&& be16_to_cpu(leaf
->hdr
.stale
) > 1) {
271 * Otherwise if we don't have enough free bytes we need to
272 * convert to node form.
274 else if ((char *)bestsp
- (char *)&leaf
->ents
[be16_to_cpu(
275 leaf
->hdr
.count
)] < needbytes
) {
277 * Just checking or no space reservation, give up.
279 if ((args
->op_flags
& XFS_DA_OP_JUSTCHECK
) ||
281 xfs_da_brelse(tp
, lbp
);
282 return XFS_ERROR(ENOSPC
);
285 * Convert to node form.
287 error
= xfs_dir2_leaf_to_node(args
, lbp
);
288 xfs_da_buf_done(lbp
);
292 * Then add the new entry.
294 return xfs_dir2_node_addname(args
);
297 * Otherwise it will fit without compaction.
302 * If just checking, then it will fit unless we needed to allocate
305 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
306 xfs_da_brelse(tp
, lbp
);
307 return use_block
== -1 ? XFS_ERROR(ENOSPC
) : 0;
310 * If no allocations are allowed, return now before we've
313 if (args
->total
== 0 && use_block
== -1) {
314 xfs_da_brelse(tp
, lbp
);
315 return XFS_ERROR(ENOSPC
);
318 * Need to compact the leaf entries, removing stale ones.
319 * Leave one stale entry behind - the one closest to our
320 * insertion index - and we'll shift that one to our insertion
324 xfs_dir2_leaf_compact_x1(lbp
, &index
, &lowstale
, &highstale
,
325 &lfloglow
, &lfloghigh
);
328 * There are stale entries, so we'll need log-low and log-high
329 * impossibly bad values later.
331 else if (be16_to_cpu(leaf
->hdr
.stale
)) {
332 lfloglow
= be16_to_cpu(leaf
->hdr
.count
);
336 * If there was no data block space found, we need to allocate
339 if (use_block
== -1) {
341 * Add the new data block.
343 if ((error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
,
345 xfs_da_brelse(tp
, lbp
);
349 * Initialize the block.
351 if ((error
= xfs_dir2_data_init(args
, use_block
, &dbp
))) {
352 xfs_da_brelse(tp
, lbp
);
356 * If we're adding a new data block on the end we need to
357 * extend the bests table. Copy it up one entry.
359 if (use_block
>= be32_to_cpu(ltp
->bestcount
)) {
361 memmove(&bestsp
[0], &bestsp
[1],
362 be32_to_cpu(ltp
->bestcount
) * sizeof(bestsp
[0]));
363 be32_add_cpu(<p
->bestcount
, 1);
364 xfs_dir2_leaf_log_tail(tp
, lbp
);
365 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
368 * If we're filling in a previously empty block just log it.
371 xfs_dir2_leaf_log_bests(tp
, lbp
, use_block
, use_block
);
373 bestsp
[use_block
] = data
->hdr
.bestfree
[0].length
;
377 * Already had space in some data block.
378 * Just read that one in.
382 xfs_da_read_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, use_block
),
383 -1, &dbp
, XFS_DATA_FORK
))) {
384 xfs_da_brelse(tp
, lbp
);
390 xfs_dir2_data_check(dp
, dbp
);
392 * Point to the biggest freespace in our data block.
394 dup
= (xfs_dir2_data_unused_t
*)
395 ((char *)data
+ be16_to_cpu(data
->hdr
.bestfree
[0].offset
));
396 ASSERT(be16_to_cpu(dup
->length
) >= length
);
397 needscan
= needlog
= 0;
399 * Mark the initial part of our freespace in use for the new entry.
401 xfs_dir2_data_use_free(tp
, dbp
, dup
,
402 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)data
), length
,
403 &needlog
, &needscan
);
405 * Initialize our new entry (at last).
407 dep
= (xfs_dir2_data_entry_t
*)dup
;
408 dep
->inumber
= cpu_to_be64(args
->inumber
);
409 dep
->namelen
= args
->namelen
;
410 memcpy(dep
->name
, args
->name
, dep
->namelen
);
411 tagp
= xfs_dir2_data_entry_tag_p(dep
);
412 *tagp
= cpu_to_be16((char *)dep
- (char *)data
);
414 * Need to scan fix up the bestfree table.
417 xfs_dir2_data_freescan(mp
, data
, &needlog
);
419 * Need to log the data block's header.
422 xfs_dir2_data_log_header(tp
, dbp
);
423 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
425 * If the bests table needs to be changed, do it.
426 * Log the change unless we've already done that.
428 if (be16_to_cpu(bestsp
[use_block
]) != be16_to_cpu(data
->hdr
.bestfree
[0].length
)) {
429 bestsp
[use_block
] = data
->hdr
.bestfree
[0].length
;
431 xfs_dir2_leaf_log_bests(tp
, lbp
, use_block
, use_block
);
434 * Now we need to make room to insert the leaf entry.
435 * If there are no stale entries, we just insert a hole at index.
437 if (!leaf
->hdr
.stale
) {
439 * lep is still good as the index leaf entry.
441 if (index
< be16_to_cpu(leaf
->hdr
.count
))
442 memmove(lep
+ 1, lep
,
443 (be16_to_cpu(leaf
->hdr
.count
) - index
) * sizeof(*lep
));
445 * Record low and high logging indices for the leaf.
448 lfloghigh
= be16_to_cpu(leaf
->hdr
.count
);
449 be16_add_cpu(&leaf
->hdr
.count
, 1);
452 * There are stale entries.
453 * We will use one of them for the new entry.
454 * It's probably not at the right location, so we'll have to
455 * shift some up or down first.
459 * If we didn't compact before, we need to find the nearest
460 * stale entries before and after our insertion point.
464 * Find the first stale entry before the insertion
467 for (lowstale
= index
- 1;
469 be32_to_cpu(leaf
->ents
[lowstale
].address
) !=
470 XFS_DIR2_NULL_DATAPTR
;
474 * Find the next stale entry at or after the insertion
475 * point, if any. Stop if we go so far that the
476 * lowstale entry would be better.
478 for (highstale
= index
;
479 highstale
< be16_to_cpu(leaf
->hdr
.count
) &&
480 be32_to_cpu(leaf
->ents
[highstale
].address
) !=
481 XFS_DIR2_NULL_DATAPTR
&&
483 index
- lowstale
- 1 >= highstale
- index
);
488 * If the low one is better, use it.
491 (highstale
== be16_to_cpu(leaf
->hdr
.count
) ||
492 index
- lowstale
- 1 < highstale
- index
)) {
493 ASSERT(index
- lowstale
- 1 >= 0);
494 ASSERT(be32_to_cpu(leaf
->ents
[lowstale
].address
) ==
495 XFS_DIR2_NULL_DATAPTR
);
497 * Copy entries up to cover the stale entry
498 * and make room for the new entry.
500 if (index
- lowstale
- 1 > 0)
501 memmove(&leaf
->ents
[lowstale
],
502 &leaf
->ents
[lowstale
+ 1],
503 (index
- lowstale
- 1) * sizeof(*lep
));
504 lep
= &leaf
->ents
[index
- 1];
505 lfloglow
= MIN(lowstale
, lfloglow
);
506 lfloghigh
= MAX(index
- 1, lfloghigh
);
509 * The high one is better, so use that one.
512 ASSERT(highstale
- index
>= 0);
513 ASSERT(be32_to_cpu(leaf
->ents
[highstale
].address
) ==
514 XFS_DIR2_NULL_DATAPTR
);
516 * Copy entries down to cover the stale entry
517 * and make room for the new entry.
519 if (highstale
- index
> 0)
520 memmove(&leaf
->ents
[index
+ 1],
522 (highstale
- index
) * sizeof(*lep
));
523 lep
= &leaf
->ents
[index
];
524 lfloglow
= MIN(index
, lfloglow
);
525 lfloghigh
= MAX(highstale
, lfloghigh
);
527 be16_add_cpu(&leaf
->hdr
.stale
, -1);
530 * Fill in the new leaf entry.
532 lep
->hashval
= cpu_to_be32(args
->hashval
);
533 lep
->address
= cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp
, use_block
,
534 be16_to_cpu(*tagp
)));
536 * Log the leaf fields and give up the buffers.
538 xfs_dir2_leaf_log_header(tp
, lbp
);
539 xfs_dir2_leaf_log_ents(tp
, lbp
, lfloglow
, lfloghigh
);
540 xfs_dir2_leaf_check(dp
, lbp
);
541 xfs_da_buf_done(lbp
);
542 xfs_dir2_data_check(dp
, dbp
);
543 xfs_da_buf_done(dbp
);
549 * Check the internal consistency of a leaf1 block.
550 * Pop an assert if something is wrong.
554 xfs_inode_t
*dp
, /* incore directory inode */
555 xfs_dabuf_t
*bp
) /* leaf's buffer */
557 int i
; /* leaf index */
558 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
559 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail pointer */
560 xfs_mount_t
*mp
; /* filesystem mount point */
561 int stale
; /* count of stale leaves */
565 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
);
567 * This value is not restrictive enough.
568 * Should factor in the size of the bests table as well.
569 * We can deduce a value for that from di_size.
571 ASSERT(be16_to_cpu(leaf
->hdr
.count
) <= xfs_dir2_max_leaf_ents(mp
));
572 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
574 * Leaves and bests don't overlap.
576 ASSERT((char *)&leaf
->ents
[be16_to_cpu(leaf
->hdr
.count
)] <=
577 (char *)xfs_dir2_leaf_bests_p(ltp
));
579 * Check hash value order, count stale entries.
581 for (i
= stale
= 0; i
< be16_to_cpu(leaf
->hdr
.count
); i
++) {
582 if (i
+ 1 < be16_to_cpu(leaf
->hdr
.count
))
583 ASSERT(be32_to_cpu(leaf
->ents
[i
].hashval
) <=
584 be32_to_cpu(leaf
->ents
[i
+ 1].hashval
));
585 if (be32_to_cpu(leaf
->ents
[i
].address
) == XFS_DIR2_NULL_DATAPTR
)
588 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) == stale
);
593 * Compact out any stale entries in the leaf.
594 * Log the header and changed leaf entries, if any.
597 xfs_dir2_leaf_compact(
598 xfs_da_args_t
*args
, /* operation arguments */
599 xfs_dabuf_t
*bp
) /* leaf buffer */
601 int from
; /* source leaf index */
602 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
603 int loglow
; /* first leaf entry to log */
604 int to
; /* target leaf index */
607 if (!leaf
->hdr
.stale
) {
611 * Compress out the stale entries in place.
613 for (from
= to
= 0, loglow
= -1; from
< be16_to_cpu(leaf
->hdr
.count
); from
++) {
614 if (be32_to_cpu(leaf
->ents
[from
].address
) == XFS_DIR2_NULL_DATAPTR
)
617 * Only actually copy the entries that are different.
622 leaf
->ents
[to
] = leaf
->ents
[from
];
627 * Update and log the header, log the leaf entries.
629 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) == from
- to
);
630 be16_add_cpu(&leaf
->hdr
.count
, -(be16_to_cpu(leaf
->hdr
.stale
)));
632 xfs_dir2_leaf_log_header(args
->trans
, bp
);
634 xfs_dir2_leaf_log_ents(args
->trans
, bp
, loglow
, to
- 1);
638 * Compact the leaf entries, removing stale ones.
639 * Leave one stale entry behind - the one closest to our
640 * insertion index - and the caller will shift that one to our insertion
642 * Return new insertion index, where the remaining stale entry is,
643 * and leaf logging indices.
646 xfs_dir2_leaf_compact_x1(
647 xfs_dabuf_t
*bp
, /* leaf buffer */
648 int *indexp
, /* insertion index */
649 int *lowstalep
, /* out: stale entry before us */
650 int *highstalep
, /* out: stale entry after us */
651 int *lowlogp
, /* out: low log index */
652 int *highlogp
) /* out: high log index */
654 int from
; /* source copy index */
655 int highstale
; /* stale entry at/after index */
656 int index
; /* insertion index */
657 int keepstale
; /* source index of kept stale */
658 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
659 int lowstale
; /* stale entry before index */
660 int newindex
=0; /* new insertion index */
661 int to
; /* destination copy index */
664 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) > 1);
667 * Find the first stale entry before our index, if any.
669 for (lowstale
= index
- 1;
671 be32_to_cpu(leaf
->ents
[lowstale
].address
) != XFS_DIR2_NULL_DATAPTR
;
675 * Find the first stale entry at or after our index, if any.
676 * Stop if the answer would be worse than lowstale.
678 for (highstale
= index
;
679 highstale
< be16_to_cpu(leaf
->hdr
.count
) &&
680 be32_to_cpu(leaf
->ents
[highstale
].address
) != XFS_DIR2_NULL_DATAPTR
&&
681 (lowstale
< 0 || index
- lowstale
> highstale
- index
);
685 * Pick the better of lowstale and highstale.
688 (highstale
== be16_to_cpu(leaf
->hdr
.count
) ||
689 index
- lowstale
<= highstale
- index
))
690 keepstale
= lowstale
;
692 keepstale
= highstale
;
694 * Copy the entries in place, removing all the stale entries
697 for (from
= to
= 0; from
< be16_to_cpu(leaf
->hdr
.count
); from
++) {
699 * Notice the new value of index.
703 if (from
!= keepstale
&&
704 be32_to_cpu(leaf
->ents
[from
].address
) == XFS_DIR2_NULL_DATAPTR
) {
710 * Record the new keepstale value for the insertion.
712 if (from
== keepstale
)
713 lowstale
= highstale
= to
;
715 * Copy only the entries that have moved.
718 leaf
->ents
[to
] = leaf
->ents
[from
];
723 * If the insertion point was past the last entry,
724 * set the new insertion point accordingly.
730 * Adjust the leaf header values.
732 be16_add_cpu(&leaf
->hdr
.count
, -(from
- to
));
733 leaf
->hdr
.stale
= cpu_to_be16(1);
735 * Remember the low/high stale value only in the "right"
738 if (lowstale
>= newindex
)
741 highstale
= be16_to_cpu(leaf
->hdr
.count
);
742 *highlogp
= be16_to_cpu(leaf
->hdr
.count
) - 1;
743 *lowstalep
= lowstale
;
744 *highstalep
= highstale
;
748 * Getdents (readdir) for leaf and node directories.
749 * This reads the data blocks only, so is the same for both forms.
752 xfs_dir2_leaf_getdents(
753 xfs_inode_t
*dp
, /* incore directory inode */
759 xfs_dabuf_t
*bp
; /* data block buffer */
760 int byteoff
; /* offset in current block */
761 xfs_dir2_db_t curdb
; /* db for current block */
762 xfs_dir2_off_t curoff
; /* current overall offset */
763 xfs_dir2_data_t
*data
; /* data block structure */
764 xfs_dir2_data_entry_t
*dep
; /* data entry */
765 xfs_dir2_data_unused_t
*dup
; /* unused entry */
766 int error
= 0; /* error return value */
767 int i
; /* temporary loop index */
768 int j
; /* temporary loop index */
769 int length
; /* temporary length value */
770 xfs_bmbt_irec_t
*map
; /* map vector for blocks */
771 xfs_extlen_t map_blocks
; /* number of fsbs in map */
772 xfs_dablk_t map_off
; /* last mapped file offset */
773 int map_size
; /* total entries in *map */
774 int map_valid
; /* valid entries in *map */
775 xfs_mount_t
*mp
; /* filesystem mount point */
776 xfs_dir2_off_t newoff
; /* new curoff after new blk */
777 int nmap
; /* mappings to ask xfs_bmapi */
778 char *ptr
= NULL
; /* pointer to current data */
779 int ra_current
; /* number of read-ahead blks */
780 int ra_index
; /* *map index for read-ahead */
781 int ra_offset
; /* map entry offset for ra */
782 int ra_want
; /* readahead count wanted */
785 * If the offset is at or past the largest allowed value,
786 * give up right away.
788 if (*offset
>= XFS_DIR2_MAX_DATAPTR
)
794 * Set up to bmap a number of blocks based on the caller's
795 * buffer size, the directory block size, and the filesystem
798 map_size
= howmany(bufsize
+ mp
->m_dirblksize
, mp
->m_sb
.sb_blocksize
);
799 map
= kmem_alloc(map_size
* sizeof(*map
), KM_SLEEP
);
800 map_valid
= ra_index
= ra_offset
= ra_current
= map_blocks
= 0;
804 * Inside the loop we keep the main offset value as a byte offset
805 * in the directory file.
807 curoff
= xfs_dir2_dataptr_to_byte(mp
, *offset
);
810 * Force this conversion through db so we truncate the offset
811 * down to get the start of the data block.
813 map_off
= xfs_dir2_db_to_da(mp
, xfs_dir2_byte_to_db(mp
, curoff
));
815 * Loop over directory entries until we reach the end offset.
816 * Get more blocks and readahead as necessary.
818 while (curoff
< XFS_DIR2_LEAF_OFFSET
) {
820 * If we have no buffer, or we're off the end of the
821 * current buffer, need to get another one.
823 if (!bp
|| ptr
>= (char *)bp
->data
+ mp
->m_dirblksize
) {
825 * If we have a buffer, we need to release it and
826 * take it out of the mapping.
829 xfs_da_brelse(NULL
, bp
);
831 map_blocks
-= mp
->m_dirblkfsbs
;
833 * Loop to get rid of the extents for the
836 for (i
= mp
->m_dirblkfsbs
; i
> 0; ) {
837 j
= MIN((int)map
->br_blockcount
, i
);
838 map
->br_blockcount
-= j
;
839 map
->br_startblock
+= j
;
840 map
->br_startoff
+= j
;
842 * If mapping is done, pitch it from
845 if (!map
->br_blockcount
&& --map_valid
)
846 memmove(&map
[0], &map
[1],
853 * Recalculate the readahead blocks wanted.
855 ra_want
= howmany(bufsize
+ mp
->m_dirblksize
,
856 mp
->m_sb
.sb_blocksize
) - 1;
859 * If we don't have as many as we want, and we haven't
860 * run out of data blocks, get some more mappings.
862 if (1 + ra_want
> map_blocks
&&
864 xfs_dir2_byte_to_da(mp
, XFS_DIR2_LEAF_OFFSET
)) {
866 * Get more bmaps, fill in after the ones
867 * we already have in the table.
869 nmap
= map_size
- map_valid
;
870 error
= xfs_bmapi(NULL
, dp
,
872 xfs_dir2_byte_to_da(mp
,
873 XFS_DIR2_LEAF_OFFSET
) - map_off
,
874 XFS_BMAPI_METADATA
, NULL
, 0,
875 &map
[map_valid
], &nmap
, NULL
, NULL
);
877 * Don't know if we should ignore this or
878 * try to return an error.
879 * The trouble with returning errors
880 * is that readdir will just stop without
881 * actually passing the error through.
886 * If we got all the mappings we asked for,
887 * set the final map offset based on the
888 * last bmap value received.
889 * Otherwise, we've reached the end.
891 if (nmap
== map_size
- map_valid
)
893 map
[map_valid
+ nmap
- 1].br_startoff
+
894 map
[map_valid
+ nmap
- 1].br_blockcount
;
897 xfs_dir2_byte_to_da(mp
,
898 XFS_DIR2_LEAF_OFFSET
);
900 * Look for holes in the mapping, and
901 * eliminate them. Count up the valid blocks.
903 for (i
= map_valid
; i
< map_valid
+ nmap
; ) {
904 if (map
[i
].br_startblock
==
907 length
= map_valid
+ nmap
- i
;
915 map
[i
].br_blockcount
;
922 * No valid mappings, so no more data blocks.
925 curoff
= xfs_dir2_da_to_byte(mp
, map_off
);
929 * Read the directory block starting at the first
932 curdb
= xfs_dir2_da_to_db(mp
, map
->br_startoff
);
933 error
= xfs_da_read_buf(NULL
, dp
, map
->br_startoff
,
934 map
->br_blockcount
>= mp
->m_dirblkfsbs
?
935 XFS_FSB_TO_DADDR(mp
, map
->br_startblock
) :
939 * Should just skip over the data block instead
945 * Adjust the current amount of read-ahead: we just
946 * read a block that was previously ra.
949 ra_current
-= mp
->m_dirblkfsbs
;
951 * Do we need more readahead?
953 for (ra_index
= ra_offset
= i
= 0;
954 ra_want
> ra_current
&& i
< map_blocks
;
955 i
+= mp
->m_dirblkfsbs
) {
956 ASSERT(ra_index
< map_valid
);
958 * Read-ahead a contiguous directory block.
960 if (i
> ra_current
&&
961 map
[ra_index
].br_blockcount
>=
963 xfs_baread(mp
->m_ddev_targp
,
965 map
[ra_index
].br_startblock
+
967 (int)BTOBB(mp
->m_dirblksize
));
971 * Read-ahead a non-contiguous directory block.
972 * This doesn't use our mapping, but this
973 * is a very rare case.
975 else if (i
> ra_current
) {
976 (void)xfs_da_reada_buf(NULL
, dp
,
977 map
[ra_index
].br_startoff
+
978 ra_offset
, XFS_DATA_FORK
);
982 * Advance offset through the mapping table.
984 for (j
= 0; j
< mp
->m_dirblkfsbs
; j
++) {
986 * The rest of this extent but not
987 * more than a dir block.
989 length
= MIN(mp
->m_dirblkfsbs
,
990 (int)(map
[ra_index
].br_blockcount
-
995 * Advance to the next mapping if
996 * this one is used up.
999 map
[ra_index
].br_blockcount
) {
1006 * Having done a read, we need to set a new offset.
1008 newoff
= xfs_dir2_db_off_to_byte(mp
, curdb
, 0);
1010 * Start of the current block.
1012 if (curoff
< newoff
)
1015 * Make sure we're in the right block.
1017 else if (curoff
> newoff
)
1018 ASSERT(xfs_dir2_byte_to_db(mp
, curoff
) ==
1021 xfs_dir2_data_check(dp
, bp
);
1023 * Find our position in the block.
1025 ptr
= (char *)&data
->u
;
1026 byteoff
= xfs_dir2_byte_to_off(mp
, curoff
);
1028 * Skip past the header.
1031 curoff
+= (uint
)sizeof(data
->hdr
);
1033 * Skip past entries until we reach our offset.
1036 while ((char *)ptr
- (char *)data
< byteoff
) {
1037 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1039 if (be16_to_cpu(dup
->freetag
)
1040 == XFS_DIR2_DATA_FREE_TAG
) {
1042 length
= be16_to_cpu(dup
->length
);
1046 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1048 xfs_dir2_data_entsize(dep
->namelen
);
1052 * Now set our real offset.
1055 xfs_dir2_db_off_to_byte(mp
,
1056 xfs_dir2_byte_to_db(mp
, curoff
),
1057 (char *)ptr
- (char *)data
);
1058 if (ptr
>= (char *)data
+ mp
->m_dirblksize
) {
1064 * We have a pointer to an entry.
1067 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1069 * No, it's unused, skip over it.
1071 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
1072 length
= be16_to_cpu(dup
->length
);
1078 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1079 length
= xfs_dir2_data_entsize(dep
->namelen
);
1081 if (filldir(dirent
, dep
->name
, dep
->namelen
,
1082 xfs_dir2_byte_to_dataptr(mp
, curoff
) & 0x7fffffff,
1083 be64_to_cpu(dep
->inumber
), DT_UNKNOWN
))
1087 * Advance to next entry in the block.
1095 * All done. Set output offset value to current offset.
1097 if (curoff
> xfs_dir2_dataptr_to_byte(mp
, XFS_DIR2_MAX_DATAPTR
))
1098 *offset
= XFS_DIR2_MAX_DATAPTR
& 0x7fffffff;
1100 *offset
= xfs_dir2_byte_to_dataptr(mp
, curoff
) & 0x7fffffff;
1103 xfs_da_brelse(NULL
, bp
);
1108 * Initialize a new leaf block, leaf1 or leafn magic accepted.
1112 xfs_da_args_t
*args
, /* operation arguments */
1113 xfs_dir2_db_t bno
, /* directory block number */
1114 xfs_dabuf_t
**bpp
, /* out: leaf buffer */
1115 int magic
) /* magic number for block */
1117 xfs_dabuf_t
*bp
; /* leaf buffer */
1118 xfs_inode_t
*dp
; /* incore directory inode */
1119 int error
; /* error return code */
1120 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1121 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1122 xfs_mount_t
*mp
; /* filesystem mount point */
1123 xfs_trans_t
*tp
; /* transaction pointer */
1129 ASSERT(bno
>= XFS_DIR2_LEAF_FIRSTDB(mp
) &&
1130 bno
< XFS_DIR2_FREE_FIRSTDB(mp
));
1132 * Get the buffer for the block.
1134 error
= xfs_da_get_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, bno
), -1, &bp
,
1142 * Initialize the header.
1144 leaf
->hdr
.info
.magic
= cpu_to_be16(magic
);
1145 leaf
->hdr
.info
.forw
= 0;
1146 leaf
->hdr
.info
.back
= 0;
1147 leaf
->hdr
.count
= 0;
1148 leaf
->hdr
.stale
= 0;
1149 xfs_dir2_leaf_log_header(tp
, bp
);
1151 * If it's a leaf-format directory initialize the tail.
1152 * In this case our caller has the real bests table to copy into
1155 if (magic
== XFS_DIR2_LEAF1_MAGIC
) {
1156 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1158 xfs_dir2_leaf_log_tail(tp
, bp
);
1165 * Log the bests entries indicated from a leaf1 block.
1168 xfs_dir2_leaf_log_bests(
1169 xfs_trans_t
*tp
, /* transaction pointer */
1170 xfs_dabuf_t
*bp
, /* leaf buffer */
1171 int first
, /* first entry to log */
1172 int last
) /* last entry to log */
1174 __be16
*firstb
; /* pointer to first entry */
1175 __be16
*lastb
; /* pointer to last entry */
1176 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1177 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1180 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
);
1181 ltp
= xfs_dir2_leaf_tail_p(tp
->t_mountp
, leaf
);
1182 firstb
= xfs_dir2_leaf_bests_p(ltp
) + first
;
1183 lastb
= xfs_dir2_leaf_bests_p(ltp
) + last
;
1184 xfs_da_log_buf(tp
, bp
, (uint
)((char *)firstb
- (char *)leaf
),
1185 (uint
)((char *)lastb
- (char *)leaf
+ sizeof(*lastb
) - 1));
1189 * Log the leaf entries indicated from a leaf1 or leafn block.
1192 xfs_dir2_leaf_log_ents(
1193 xfs_trans_t
*tp
, /* transaction pointer */
1194 xfs_dabuf_t
*bp
, /* leaf buffer */
1195 int first
, /* first entry to log */
1196 int last
) /* last entry to log */
1198 xfs_dir2_leaf_entry_t
*firstlep
; /* pointer to first entry */
1199 xfs_dir2_leaf_entry_t
*lastlep
; /* pointer to last entry */
1200 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1203 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
||
1204 be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
1205 firstlep
= &leaf
->ents
[first
];
1206 lastlep
= &leaf
->ents
[last
];
1207 xfs_da_log_buf(tp
, bp
, (uint
)((char *)firstlep
- (char *)leaf
),
1208 (uint
)((char *)lastlep
- (char *)leaf
+ sizeof(*lastlep
) - 1));
1212 * Log the header of the leaf1 or leafn block.
1215 xfs_dir2_leaf_log_header(
1216 xfs_trans_t
*tp
, /* transaction pointer */
1217 xfs_dabuf_t
*bp
) /* leaf buffer */
1219 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1222 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
||
1223 be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
1224 xfs_da_log_buf(tp
, bp
, (uint
)((char *)&leaf
->hdr
- (char *)leaf
),
1225 (uint
)(sizeof(leaf
->hdr
) - 1));
1229 * Log the tail of the leaf1 block.
1232 xfs_dir2_leaf_log_tail(
1233 xfs_trans_t
*tp
, /* transaction pointer */
1234 xfs_dabuf_t
*bp
) /* leaf buffer */
1236 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1237 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1238 xfs_mount_t
*mp
; /* filesystem mount point */
1242 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
);
1243 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1244 xfs_da_log_buf(tp
, bp
, (uint
)((char *)ltp
- (char *)leaf
),
1245 (uint
)(mp
->m_dirblksize
- 1));
1249 * Look up the entry referred to by args in the leaf format directory.
1250 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1251 * is also used by the node-format code.
1254 xfs_dir2_leaf_lookup(
1255 xfs_da_args_t
*args
) /* operation arguments */
1257 xfs_dabuf_t
*dbp
; /* data block buffer */
1258 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1259 xfs_inode_t
*dp
; /* incore directory inode */
1260 int error
; /* error return code */
1261 int index
; /* found entry index */
1262 xfs_dabuf_t
*lbp
; /* leaf buffer */
1263 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1264 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1265 xfs_trans_t
*tp
; /* transaction pointer */
1267 xfs_dir2_trace_args("leaf_lookup", args
);
1269 * Look up name in the leaf block, returning both buffers and index.
1271 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1276 xfs_dir2_leaf_check(dp
, lbp
);
1279 * Get to the leaf entry and contained data entry address.
1281 lep
= &leaf
->ents
[index
];
1283 * Point to the data entry.
1285 dep
= (xfs_dir2_data_entry_t
*)
1286 ((char *)dbp
->data
+
1287 xfs_dir2_dataptr_to_off(dp
->i_mount
, be32_to_cpu(lep
->address
)));
1289 * Return the found inode number & CI name if appropriate
1291 args
->inumber
= be64_to_cpu(dep
->inumber
);
1292 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
1293 xfs_da_brelse(tp
, dbp
);
1294 xfs_da_brelse(tp
, lbp
);
1295 return XFS_ERROR(error
);
1299 * Look up name/hash in the leaf block.
1300 * Fill in indexp with the found index, and dbpp with the data buffer.
1301 * If not found dbpp will be NULL, and ENOENT comes back.
1302 * lbpp will always be filled in with the leaf buffer unless there's an error.
1304 static int /* error */
1305 xfs_dir2_leaf_lookup_int(
1306 xfs_da_args_t
*args
, /* operation arguments */
1307 xfs_dabuf_t
**lbpp
, /* out: leaf buffer */
1308 int *indexp
, /* out: index in leaf block */
1309 xfs_dabuf_t
**dbpp
) /* out: data buffer */
1311 xfs_dir2_db_t curdb
= -1; /* current data block number */
1312 xfs_dabuf_t
*dbp
= NULL
; /* data buffer */
1313 xfs_dir2_data_entry_t
*dep
; /* data entry */
1314 xfs_inode_t
*dp
; /* incore directory inode */
1315 int error
; /* error return code */
1316 int index
; /* index in leaf block */
1317 xfs_dabuf_t
*lbp
; /* leaf buffer */
1318 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1319 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1320 xfs_mount_t
*mp
; /* filesystem mount point */
1321 xfs_dir2_db_t newdb
; /* new data block number */
1322 xfs_trans_t
*tp
; /* transaction pointer */
1323 xfs_dir2_db_t cidb
= -1; /* case match data block no. */
1324 enum xfs_dacmp cmp
; /* name compare result */
1330 * Read the leaf block into the buffer.
1332 error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirleafblk
, -1, &lbp
,
1338 xfs_dir2_leaf_check(dp
, lbp
);
1340 * Look for the first leaf entry with our hash value.
1342 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
1344 * Loop over all the entries with the right hash value
1345 * looking to match the name.
1347 for (lep
= &leaf
->ents
[index
]; index
< be16_to_cpu(leaf
->hdr
.count
) &&
1348 be32_to_cpu(lep
->hashval
) == args
->hashval
;
1351 * Skip over stale leaf entries.
1353 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
1356 * Get the new data block number.
1358 newdb
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
1360 * If it's not the same as the old data block number,
1361 * need to pitch the old one and read the new one.
1363 if (newdb
!= curdb
) {
1365 xfs_da_brelse(tp
, dbp
);
1366 error
= xfs_da_read_buf(tp
, dp
,
1367 xfs_dir2_db_to_da(mp
, newdb
),
1368 -1, &dbp
, XFS_DATA_FORK
);
1370 xfs_da_brelse(tp
, lbp
);
1373 xfs_dir2_data_check(dp
, dbp
);
1377 * Point to the data entry.
1379 dep
= (xfs_dir2_data_entry_t
*)((char *)dbp
->data
+
1380 xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
1382 * Compare name and if it's an exact match, return the index
1383 * and buffer. If it's the first case-insensitive match, store
1384 * the index and buffer and continue looking for an exact match.
1386 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
1387 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
1388 args
->cmpresult
= cmp
;
1390 /* case exact match: return the current buffer. */
1391 if (cmp
== XFS_CMP_EXACT
) {
1398 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
1400 * Here, we can only be doing a lookup (not a rename or remove).
1401 * If a case-insensitive match was found earlier, re-read the
1402 * appropriate data block if required and return it.
1404 if (args
->cmpresult
== XFS_CMP_CASE
) {
1406 if (cidb
!= curdb
) {
1407 xfs_da_brelse(tp
, dbp
);
1408 error
= xfs_da_read_buf(tp
, dp
,
1409 xfs_dir2_db_to_da(mp
, cidb
),
1410 -1, &dbp
, XFS_DATA_FORK
);
1412 xfs_da_brelse(tp
, lbp
);
1420 * No match found, return ENOENT.
1424 xfs_da_brelse(tp
, dbp
);
1425 xfs_da_brelse(tp
, lbp
);
1426 return XFS_ERROR(ENOENT
);
1430 * Remove an entry from a leaf format directory.
1433 xfs_dir2_leaf_removename(
1434 xfs_da_args_t
*args
) /* operation arguments */
1436 __be16
*bestsp
; /* leaf block best freespace */
1437 xfs_dir2_data_t
*data
; /* data block structure */
1438 xfs_dir2_db_t db
; /* data block number */
1439 xfs_dabuf_t
*dbp
; /* data block buffer */
1440 xfs_dir2_data_entry_t
*dep
; /* data entry structure */
1441 xfs_inode_t
*dp
; /* incore directory inode */
1442 int error
; /* error return code */
1443 xfs_dir2_db_t i
; /* temporary data block # */
1444 int index
; /* index into leaf entries */
1445 xfs_dabuf_t
*lbp
; /* leaf buffer */
1446 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1447 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1448 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1449 xfs_mount_t
*mp
; /* filesystem mount point */
1450 int needlog
; /* need to log data header */
1451 int needscan
; /* need to rescan data frees */
1452 xfs_dir2_data_off_t oldbest
; /* old value of best free */
1453 xfs_trans_t
*tp
; /* transaction pointer */
1455 xfs_dir2_trace_args("leaf_removename", args
);
1457 * Lookup the leaf entry, get the leaf and data blocks read in.
1459 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1467 xfs_dir2_data_check(dp
, dbp
);
1469 * Point to the leaf entry, use that to point to the data entry.
1471 lep
= &leaf
->ents
[index
];
1472 db
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
1473 dep
= (xfs_dir2_data_entry_t
*)
1474 ((char *)data
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
1475 needscan
= needlog
= 0;
1476 oldbest
= be16_to_cpu(data
->hdr
.bestfree
[0].length
);
1477 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1478 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1479 ASSERT(be16_to_cpu(bestsp
[db
]) == oldbest
);
1481 * Mark the former data entry unused.
1483 xfs_dir2_data_make_free(tp
, dbp
,
1484 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)data
),
1485 xfs_dir2_data_entsize(dep
->namelen
), &needlog
, &needscan
);
1487 * We just mark the leaf entry stale by putting a null in it.
1489 be16_add_cpu(&leaf
->hdr
.stale
, 1);
1490 xfs_dir2_leaf_log_header(tp
, lbp
);
1491 lep
->address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
1492 xfs_dir2_leaf_log_ents(tp
, lbp
, index
, index
);
1494 * Scan the freespace in the data block again if necessary,
1495 * log the data block header if necessary.
1498 xfs_dir2_data_freescan(mp
, data
, &needlog
);
1500 xfs_dir2_data_log_header(tp
, dbp
);
1502 * If the longest freespace in the data block has changed,
1503 * put the new value in the bests table and log that.
1505 if (be16_to_cpu(data
->hdr
.bestfree
[0].length
) != oldbest
) {
1506 bestsp
[db
] = data
->hdr
.bestfree
[0].length
;
1507 xfs_dir2_leaf_log_bests(tp
, lbp
, db
, db
);
1509 xfs_dir2_data_check(dp
, dbp
);
1511 * If the data block is now empty then get rid of the data block.
1513 if (be16_to_cpu(data
->hdr
.bestfree
[0].length
) ==
1514 mp
->m_dirblksize
- (uint
)sizeof(data
->hdr
)) {
1515 ASSERT(db
!= mp
->m_dirdatablk
);
1516 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1518 * Nope, can't get rid of it because it caused
1519 * allocation of a bmap btree block to do so.
1520 * Just go on, returning success, leaving the
1521 * empty block in place.
1523 if (error
== ENOSPC
&& args
->total
== 0) {
1524 xfs_da_buf_done(dbp
);
1527 xfs_dir2_leaf_check(dp
, lbp
);
1528 xfs_da_buf_done(lbp
);
1533 * If this is the last data block then compact the
1534 * bests table by getting rid of entries.
1536 if (db
== be32_to_cpu(ltp
->bestcount
) - 1) {
1538 * Look for the last active entry (i).
1540 for (i
= db
- 1; i
> 0; i
--) {
1541 if (be16_to_cpu(bestsp
[i
]) != NULLDATAOFF
)
1545 * Copy the table down so inactive entries at the
1548 memmove(&bestsp
[db
- i
], bestsp
,
1549 (be32_to_cpu(ltp
->bestcount
) - (db
- i
)) * sizeof(*bestsp
));
1550 be32_add_cpu(<p
->bestcount
, -(db
- i
));
1551 xfs_dir2_leaf_log_tail(tp
, lbp
);
1552 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1554 bestsp
[db
] = cpu_to_be16(NULLDATAOFF
);
1557 * If the data block was not the first one, drop it.
1559 else if (db
!= mp
->m_dirdatablk
&& dbp
!= NULL
) {
1560 xfs_da_buf_done(dbp
);
1563 xfs_dir2_leaf_check(dp
, lbp
);
1565 * See if we can convert to block form.
1567 return xfs_dir2_leaf_to_block(args
, lbp
, dbp
);
1571 * Replace the inode number in a leaf format directory entry.
1574 xfs_dir2_leaf_replace(
1575 xfs_da_args_t
*args
) /* operation arguments */
1577 xfs_dabuf_t
*dbp
; /* data block buffer */
1578 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1579 xfs_inode_t
*dp
; /* incore directory inode */
1580 int error
; /* error return code */
1581 int index
; /* index of leaf entry */
1582 xfs_dabuf_t
*lbp
; /* leaf buffer */
1583 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1584 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1585 xfs_trans_t
*tp
; /* transaction pointer */
1587 xfs_dir2_trace_args("leaf_replace", args
);
1589 * Look up the entry.
1591 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1597 * Point to the leaf entry, get data address from it.
1599 lep
= &leaf
->ents
[index
];
1601 * Point to the data entry.
1603 dep
= (xfs_dir2_data_entry_t
*)
1604 ((char *)dbp
->data
+
1605 xfs_dir2_dataptr_to_off(dp
->i_mount
, be32_to_cpu(lep
->address
)));
1606 ASSERT(args
->inumber
!= be64_to_cpu(dep
->inumber
));
1608 * Put the new inode number in, log it.
1610 dep
->inumber
= cpu_to_be64(args
->inumber
);
1612 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
1613 xfs_da_buf_done(dbp
);
1614 xfs_dir2_leaf_check(dp
, lbp
);
1615 xfs_da_brelse(tp
, lbp
);
1620 * Return index in the leaf block (lbp) which is either the first
1621 * one with this hash value, or if there are none, the insert point
1622 * for that hash value.
1624 int /* index value */
1625 xfs_dir2_leaf_search_hash(
1626 xfs_da_args_t
*args
, /* operation arguments */
1627 xfs_dabuf_t
*lbp
) /* leaf buffer */
1629 xfs_dahash_t hash
=0; /* hash from this entry */
1630 xfs_dahash_t hashwant
; /* hash value looking for */
1631 int high
; /* high leaf index */
1632 int low
; /* low leaf index */
1633 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1634 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1635 int mid
=0; /* current leaf index */
1639 if (!leaf
->hdr
.count
)
1643 * Note, the table cannot be empty, so we have to go through the loop.
1644 * Binary search the leaf entries looking for our hash value.
1646 for (lep
= leaf
->ents
, low
= 0, high
= be16_to_cpu(leaf
->hdr
.count
) - 1,
1647 hashwant
= args
->hashval
;
1649 mid
= (low
+ high
) >> 1;
1650 if ((hash
= be32_to_cpu(lep
[mid
].hashval
)) == hashwant
)
1652 if (hash
< hashwant
)
1658 * Found one, back up through all the equal hash values.
1660 if (hash
== hashwant
) {
1661 while (mid
> 0 && be32_to_cpu(lep
[mid
- 1].hashval
) == hashwant
) {
1666 * Need to point to an entry higher than ours.
1668 else if (hash
< hashwant
)
1674 * Trim off a trailing data block. We know it's empty since the leaf
1675 * freespace table says so.
1678 xfs_dir2_leaf_trim_data(
1679 xfs_da_args_t
*args
, /* operation arguments */
1680 xfs_dabuf_t
*lbp
, /* leaf buffer */
1681 xfs_dir2_db_t db
) /* data block number */
1683 __be16
*bestsp
; /* leaf bests table */
1685 xfs_dir2_data_t
*data
; /* data block structure */
1687 xfs_dabuf_t
*dbp
; /* data block buffer */
1688 xfs_inode_t
*dp
; /* incore directory inode */
1689 int error
; /* error return value */
1690 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1691 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1692 xfs_mount_t
*mp
; /* filesystem mount point */
1693 xfs_trans_t
*tp
; /* transaction pointer */
1699 * Read the offending data block. We need its buffer.
1701 if ((error
= xfs_da_read_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, db
), -1, &dbp
,
1707 ASSERT(be32_to_cpu(data
->hdr
.magic
) == XFS_DIR2_DATA_MAGIC
);
1709 /* this seems to be an error
1710 * data is only valid if DEBUG is defined?
1715 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1716 ASSERT(be16_to_cpu(data
->hdr
.bestfree
[0].length
) ==
1717 mp
->m_dirblksize
- (uint
)sizeof(data
->hdr
));
1718 ASSERT(db
== be32_to_cpu(ltp
->bestcount
) - 1);
1720 * Get rid of the data block.
1722 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1723 ASSERT(error
!= ENOSPC
);
1724 xfs_da_brelse(tp
, dbp
);
1728 * Eliminate the last bests entry from the table.
1730 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1731 be32_add_cpu(<p
->bestcount
, -1);
1732 memmove(&bestsp
[1], &bestsp
[0], be32_to_cpu(ltp
->bestcount
) * sizeof(*bestsp
));
1733 xfs_dir2_leaf_log_tail(tp
, lbp
);
1734 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1739 * Convert node form directory to leaf form directory.
1740 * The root of the node form dir needs to already be a LEAFN block.
1741 * Just return if we can't do anything.
1744 xfs_dir2_node_to_leaf(
1745 xfs_da_state_t
*state
) /* directory operation state */
1747 xfs_da_args_t
*args
; /* operation arguments */
1748 xfs_inode_t
*dp
; /* incore directory inode */
1749 int error
; /* error return code */
1750 xfs_dabuf_t
*fbp
; /* buffer for freespace block */
1751 xfs_fileoff_t fo
; /* freespace file offset */
1752 xfs_dir2_free_t
*free
; /* freespace structure */
1753 xfs_dabuf_t
*lbp
; /* buffer for leaf block */
1754 xfs_dir2_leaf_tail_t
*ltp
; /* tail of leaf structure */
1755 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1756 xfs_mount_t
*mp
; /* filesystem mount point */
1757 int rval
; /* successful free trim? */
1758 xfs_trans_t
*tp
; /* transaction pointer */
1761 * There's more than a leaf level in the btree, so there must
1762 * be multiple leafn blocks. Give up.
1764 if (state
->path
.active
> 1)
1767 xfs_dir2_trace_args("node_to_leaf", args
);
1772 * Get the last offset in the file.
1774 if ((error
= xfs_bmap_last_offset(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1777 fo
-= mp
->m_dirblkfsbs
;
1779 * If there are freespace blocks other than the first one,
1780 * take this opportunity to remove trailing empty freespace blocks
1781 * that may have been left behind during no-space-reservation
1784 while (fo
> mp
->m_dirfreeblk
) {
1785 if ((error
= xfs_dir2_node_trim_free(args
, fo
, &rval
))) {
1789 fo
-= mp
->m_dirblkfsbs
;
1794 * Now find the block just before the freespace block.
1796 if ((error
= xfs_bmap_last_before(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1800 * If it's not the single leaf block, give up.
1802 if (XFS_FSB_TO_B(mp
, fo
) > XFS_DIR2_LEAF_OFFSET
+ mp
->m_dirblksize
)
1804 lbp
= state
->path
.blk
[0].bp
;
1806 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
1808 * Read the freespace block.
1810 if ((error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirfreeblk
, -1, &fbp
,
1815 ASSERT(be32_to_cpu(free
->hdr
.magic
) == XFS_DIR2_FREE_MAGIC
);
1816 ASSERT(!free
->hdr
.firstdb
);
1818 * Now see if the leafn and free data will fit in a leaf1.
1819 * If not, release the buffer and give up.
1821 if ((uint
)sizeof(leaf
->hdr
) +
1822 (be16_to_cpu(leaf
->hdr
.count
) - be16_to_cpu(leaf
->hdr
.stale
)) * (uint
)sizeof(leaf
->ents
[0]) +
1823 be32_to_cpu(free
->hdr
.nvalid
) * (uint
)sizeof(leaf
->bests
[0]) +
1824 (uint
)sizeof(leaf
->tail
) >
1826 xfs_da_brelse(tp
, fbp
);
1830 * If the leaf has any stale entries in it, compress them out.
1831 * The compact routine will log the header.
1833 if (be16_to_cpu(leaf
->hdr
.stale
))
1834 xfs_dir2_leaf_compact(args
, lbp
);
1836 xfs_dir2_leaf_log_header(tp
, lbp
);
1837 leaf
->hdr
.info
.magic
= cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
);
1839 * Set up the leaf tail from the freespace block.
1841 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1842 ltp
->bestcount
= free
->hdr
.nvalid
;
1844 * Set up the leaf bests table.
1846 memcpy(xfs_dir2_leaf_bests_p(ltp
), free
->bests
,
1847 be32_to_cpu(ltp
->bestcount
) * sizeof(leaf
->bests
[0]));
1848 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1849 xfs_dir2_leaf_log_tail(tp
, lbp
);
1850 xfs_dir2_leaf_check(dp
, lbp
);
1852 * Get rid of the freespace block.
1854 error
= xfs_dir2_shrink_inode(args
, XFS_DIR2_FREE_FIRSTDB(mp
), fbp
);
1857 * This can't fail here because it can only happen when
1858 * punching out the middle of an extent, and this is an
1861 ASSERT(error
!= ENOSPC
);
1866 * Now see if we can convert the single-leaf directory
1867 * down to a block form directory.
1868 * This routine always kills the dabuf for the leaf, so
1869 * eliminate it from the path.
1871 error
= xfs_dir2_leaf_to_block(args
, lbp
, NULL
);
1872 state
->path
.blk
[0].bp
= NULL
;