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_mount.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_attr_sf.h"
32 #include "xfs_dir2_sf.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
36 #include "xfs_dir2_data.h"
37 #include "xfs_dir2_leaf.h"
38 #include "xfs_dir2_block.h"
39 #include "xfs_dir2_node.h"
40 #include "xfs_error.h"
41 #include "xfs_trace.h"
44 * Local function declarations.
47 static void xfs_dir2_leaf_check(xfs_inode_t
*dp
, xfs_dabuf_t
*bp
);
49 #define xfs_dir2_leaf_check(dp, bp)
51 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t
*args
, xfs_dabuf_t
**lbpp
,
52 int *indexp
, xfs_dabuf_t
**dbpp
);
53 static void xfs_dir2_leaf_log_bests(struct xfs_trans
*tp
, struct xfs_dabuf
*bp
,
55 static void xfs_dir2_leaf_log_tail(struct xfs_trans
*tp
, struct xfs_dabuf
*bp
);
59 * Convert a block form directory to a leaf form directory.
62 xfs_dir2_block_to_leaf(
63 xfs_da_args_t
*args
, /* operation arguments */
64 xfs_dabuf_t
*dbp
) /* input block's buffer */
66 __be16
*bestsp
; /* leaf's bestsp entries */
67 xfs_dablk_t blkno
; /* leaf block's bno */
68 xfs_dir2_block_t
*block
; /* block structure */
69 xfs_dir2_leaf_entry_t
*blp
; /* block's leaf entries */
70 xfs_dir2_block_tail_t
*btp
; /* block's tail */
71 xfs_inode_t
*dp
; /* incore directory inode */
72 int error
; /* error return code */
73 xfs_dabuf_t
*lbp
; /* leaf block's buffer */
74 xfs_dir2_db_t ldb
; /* leaf block's bno */
75 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
76 xfs_dir2_leaf_tail_t
*ltp
; /* leaf's tail */
77 xfs_mount_t
*mp
; /* filesystem mount point */
78 int needlog
; /* need to log block header */
79 int needscan
; /* need to rescan bestfree */
80 xfs_trans_t
*tp
; /* transaction pointer */
82 trace_xfs_dir2_block_to_leaf(args
);
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 trace_xfs_dir2_leaf_addname(args
);
197 * Read the leaf block.
199 error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirleafblk
, -1, &lbp
,
206 * Look up the entry by hash value and name.
207 * We know it's not there, our caller has already done a lookup.
208 * So the index is of the entry to insert in front of.
209 * But if there are dup hash values the index is of the first of those.
211 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
213 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
214 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
215 length
= xfs_dir2_data_entsize(args
->namelen
);
217 * See if there are any entries with the same hash value
218 * and space in their block for the new entry.
219 * This is good because it puts multiple same-hash value entries
220 * in a data block, improving the lookup of those entries.
222 for (use_block
= -1, lep
= &leaf
->ents
[index
];
223 index
< be16_to_cpu(leaf
->hdr
.count
) && be32_to_cpu(lep
->hashval
) == args
->hashval
;
225 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
227 i
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
228 ASSERT(i
< be32_to_cpu(ltp
->bestcount
));
229 ASSERT(be16_to_cpu(bestsp
[i
]) != NULLDATAOFF
);
230 if (be16_to_cpu(bestsp
[i
]) >= length
) {
236 * Didn't find a block yet, linear search all the data blocks.
238 if (use_block
== -1) {
239 for (i
= 0; i
< be32_to_cpu(ltp
->bestcount
); i
++) {
241 * Remember a block we see that's missing.
243 if (be16_to_cpu(bestsp
[i
]) == NULLDATAOFF
&& use_block
== -1)
245 else if (be16_to_cpu(bestsp
[i
]) >= length
) {
252 * How many bytes do we need in the leaf block?
255 (leaf
->hdr
.stale
? 0 : (uint
)sizeof(leaf
->ents
[0])) +
256 (use_block
!= -1 ? 0 : (uint
)sizeof(leaf
->bests
[0]));
258 * Now kill use_block if it refers to a missing block, so we
259 * can use it as an indication of allocation needed.
261 if (use_block
!= -1 && be16_to_cpu(bestsp
[use_block
]) == NULLDATAOFF
)
264 * If we don't have enough free bytes but we can make enough
265 * by compacting out stale entries, we'll do that.
267 if ((char *)bestsp
- (char *)&leaf
->ents
[be16_to_cpu(leaf
->hdr
.count
)] <
268 needbytes
&& be16_to_cpu(leaf
->hdr
.stale
) > 1) {
272 * Otherwise if we don't have enough free bytes we need to
273 * convert to node form.
275 else if ((char *)bestsp
- (char *)&leaf
->ents
[be16_to_cpu(
276 leaf
->hdr
.count
)] < needbytes
) {
278 * Just checking or no space reservation, give up.
280 if ((args
->op_flags
& XFS_DA_OP_JUSTCHECK
) ||
282 xfs_da_brelse(tp
, lbp
);
283 return XFS_ERROR(ENOSPC
);
286 * Convert to node form.
288 error
= xfs_dir2_leaf_to_node(args
, lbp
);
289 xfs_da_buf_done(lbp
);
293 * Then add the new entry.
295 return xfs_dir2_node_addname(args
);
298 * Otherwise it will fit without compaction.
303 * If just checking, then it will fit unless we needed to allocate
306 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
307 xfs_da_brelse(tp
, lbp
);
308 return use_block
== -1 ? XFS_ERROR(ENOSPC
) : 0;
311 * If no allocations are allowed, return now before we've
314 if (args
->total
== 0 && use_block
== -1) {
315 xfs_da_brelse(tp
, lbp
);
316 return XFS_ERROR(ENOSPC
);
319 * Need to compact the leaf entries, removing stale ones.
320 * Leave one stale entry behind - the one closest to our
321 * insertion index - and we'll shift that one to our insertion
325 xfs_dir2_leaf_compact_x1(lbp
, &index
, &lowstale
, &highstale
,
326 &lfloglow
, &lfloghigh
);
329 * There are stale entries, so we'll need log-low and log-high
330 * impossibly bad values later.
332 else if (be16_to_cpu(leaf
->hdr
.stale
)) {
333 lfloglow
= be16_to_cpu(leaf
->hdr
.count
);
337 * If there was no data block space found, we need to allocate
340 if (use_block
== -1) {
342 * Add the new data block.
344 if ((error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
,
346 xfs_da_brelse(tp
, lbp
);
350 * Initialize the block.
352 if ((error
= xfs_dir2_data_init(args
, use_block
, &dbp
))) {
353 xfs_da_brelse(tp
, lbp
);
357 * If we're adding a new data block on the end we need to
358 * extend the bests table. Copy it up one entry.
360 if (use_block
>= be32_to_cpu(ltp
->bestcount
)) {
362 memmove(&bestsp
[0], &bestsp
[1],
363 be32_to_cpu(ltp
->bestcount
) * sizeof(bestsp
[0]));
364 be32_add_cpu(<p
->bestcount
, 1);
365 xfs_dir2_leaf_log_tail(tp
, lbp
);
366 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
369 * If we're filling in a previously empty block just log it.
372 xfs_dir2_leaf_log_bests(tp
, lbp
, use_block
, use_block
);
374 bestsp
[use_block
] = data
->hdr
.bestfree
[0].length
;
378 * Already had space in some data block.
379 * Just read that one in.
383 xfs_da_read_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, use_block
),
384 -1, &dbp
, XFS_DATA_FORK
))) {
385 xfs_da_brelse(tp
, lbp
);
391 xfs_dir2_data_check(dp
, dbp
);
393 * Point to the biggest freespace in our data block.
395 dup
= (xfs_dir2_data_unused_t
*)
396 ((char *)data
+ be16_to_cpu(data
->hdr
.bestfree
[0].offset
));
397 ASSERT(be16_to_cpu(dup
->length
) >= length
);
398 needscan
= needlog
= 0;
400 * Mark the initial part of our freespace in use for the new entry.
402 xfs_dir2_data_use_free(tp
, dbp
, dup
,
403 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)data
), length
,
404 &needlog
, &needscan
);
406 * Initialize our new entry (at last).
408 dep
= (xfs_dir2_data_entry_t
*)dup
;
409 dep
->inumber
= cpu_to_be64(args
->inumber
);
410 dep
->namelen
= args
->namelen
;
411 memcpy(dep
->name
, args
->name
, dep
->namelen
);
412 tagp
= xfs_dir2_data_entry_tag_p(dep
);
413 *tagp
= cpu_to_be16((char *)dep
- (char *)data
);
415 * Need to scan fix up the bestfree table.
418 xfs_dir2_data_freescan(mp
, data
, &needlog
);
420 * Need to log the data block's header.
423 xfs_dir2_data_log_header(tp
, dbp
);
424 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
426 * If the bests table needs to be changed, do it.
427 * Log the change unless we've already done that.
429 if (be16_to_cpu(bestsp
[use_block
]) != be16_to_cpu(data
->hdr
.bestfree
[0].length
)) {
430 bestsp
[use_block
] = data
->hdr
.bestfree
[0].length
;
432 xfs_dir2_leaf_log_bests(tp
, lbp
, use_block
, use_block
);
435 * Now we need to make room to insert the leaf entry.
436 * If there are no stale entries, we just insert a hole at index.
438 if (!leaf
->hdr
.stale
) {
440 * lep is still good as the index leaf entry.
442 if (index
< be16_to_cpu(leaf
->hdr
.count
))
443 memmove(lep
+ 1, lep
,
444 (be16_to_cpu(leaf
->hdr
.count
) - index
) * sizeof(*lep
));
446 * Record low and high logging indices for the leaf.
449 lfloghigh
= be16_to_cpu(leaf
->hdr
.count
);
450 be16_add_cpu(&leaf
->hdr
.count
, 1);
453 * There are stale entries.
454 * We will use one of them for the new entry.
455 * It's probably not at the right location, so we'll have to
456 * shift some up or down first.
460 * If we didn't compact before, we need to find the nearest
461 * stale entries before and after our insertion point.
465 * Find the first stale entry before the insertion
468 for (lowstale
= index
- 1;
470 be32_to_cpu(leaf
->ents
[lowstale
].address
) !=
471 XFS_DIR2_NULL_DATAPTR
;
475 * Find the next stale entry at or after the insertion
476 * point, if any. Stop if we go so far that the
477 * lowstale entry would be better.
479 for (highstale
= index
;
480 highstale
< be16_to_cpu(leaf
->hdr
.count
) &&
481 be32_to_cpu(leaf
->ents
[highstale
].address
) !=
482 XFS_DIR2_NULL_DATAPTR
&&
484 index
- lowstale
- 1 >= highstale
- index
);
489 * If the low one is better, use it.
492 (highstale
== be16_to_cpu(leaf
->hdr
.count
) ||
493 index
- lowstale
- 1 < highstale
- index
)) {
494 ASSERT(index
- lowstale
- 1 >= 0);
495 ASSERT(be32_to_cpu(leaf
->ents
[lowstale
].address
) ==
496 XFS_DIR2_NULL_DATAPTR
);
498 * Copy entries up to cover the stale entry
499 * and make room for the new entry.
501 if (index
- lowstale
- 1 > 0)
502 memmove(&leaf
->ents
[lowstale
],
503 &leaf
->ents
[lowstale
+ 1],
504 (index
- lowstale
- 1) * sizeof(*lep
));
505 lep
= &leaf
->ents
[index
- 1];
506 lfloglow
= MIN(lowstale
, lfloglow
);
507 lfloghigh
= MAX(index
- 1, lfloghigh
);
510 * The high one is better, so use that one.
513 ASSERT(highstale
- index
>= 0);
514 ASSERT(be32_to_cpu(leaf
->ents
[highstale
].address
) ==
515 XFS_DIR2_NULL_DATAPTR
);
517 * Copy entries down to cover the stale entry
518 * and make room for the new entry.
520 if (highstale
- index
> 0)
521 memmove(&leaf
->ents
[index
+ 1],
523 (highstale
- index
) * sizeof(*lep
));
524 lep
= &leaf
->ents
[index
];
525 lfloglow
= MIN(index
, lfloglow
);
526 lfloghigh
= MAX(highstale
, lfloghigh
);
528 be16_add_cpu(&leaf
->hdr
.stale
, -1);
531 * Fill in the new leaf entry.
533 lep
->hashval
= cpu_to_be32(args
->hashval
);
534 lep
->address
= cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp
, use_block
,
535 be16_to_cpu(*tagp
)));
537 * Log the leaf fields and give up the buffers.
539 xfs_dir2_leaf_log_header(tp
, lbp
);
540 xfs_dir2_leaf_log_ents(tp
, lbp
, lfloglow
, lfloghigh
);
541 xfs_dir2_leaf_check(dp
, lbp
);
542 xfs_da_buf_done(lbp
);
543 xfs_dir2_data_check(dp
, dbp
);
544 xfs_da_buf_done(dbp
);
550 * Check the internal consistency of a leaf1 block.
551 * Pop an assert if something is wrong.
555 xfs_inode_t
*dp
, /* incore directory inode */
556 xfs_dabuf_t
*bp
) /* leaf's buffer */
558 int i
; /* leaf index */
559 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
560 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail pointer */
561 xfs_mount_t
*mp
; /* filesystem mount point */
562 int stale
; /* count of stale leaves */
566 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
);
568 * This value is not restrictive enough.
569 * Should factor in the size of the bests table as well.
570 * We can deduce a value for that from di_size.
572 ASSERT(be16_to_cpu(leaf
->hdr
.count
) <= xfs_dir2_max_leaf_ents(mp
));
573 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
575 * Leaves and bests don't overlap.
577 ASSERT((char *)&leaf
->ents
[be16_to_cpu(leaf
->hdr
.count
)] <=
578 (char *)xfs_dir2_leaf_bests_p(ltp
));
580 * Check hash value order, count stale entries.
582 for (i
= stale
= 0; i
< be16_to_cpu(leaf
->hdr
.count
); i
++) {
583 if (i
+ 1 < be16_to_cpu(leaf
->hdr
.count
))
584 ASSERT(be32_to_cpu(leaf
->ents
[i
].hashval
) <=
585 be32_to_cpu(leaf
->ents
[i
+ 1].hashval
));
586 if (be32_to_cpu(leaf
->ents
[i
].address
) == XFS_DIR2_NULL_DATAPTR
)
589 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) == stale
);
594 * Compact out any stale entries in the leaf.
595 * Log the header and changed leaf entries, if any.
598 xfs_dir2_leaf_compact(
599 xfs_da_args_t
*args
, /* operation arguments */
600 xfs_dabuf_t
*bp
) /* leaf buffer */
602 int from
; /* source leaf index */
603 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
604 int loglow
; /* first leaf entry to log */
605 int to
; /* target leaf index */
608 if (!leaf
->hdr
.stale
) {
612 * Compress out the stale entries in place.
614 for (from
= to
= 0, loglow
= -1; from
< be16_to_cpu(leaf
->hdr
.count
); from
++) {
615 if (be32_to_cpu(leaf
->ents
[from
].address
) == XFS_DIR2_NULL_DATAPTR
)
618 * Only actually copy the entries that are different.
623 leaf
->ents
[to
] = leaf
->ents
[from
];
628 * Update and log the header, log the leaf entries.
630 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) == from
- to
);
631 be16_add_cpu(&leaf
->hdr
.count
, -(be16_to_cpu(leaf
->hdr
.stale
)));
633 xfs_dir2_leaf_log_header(args
->trans
, bp
);
635 xfs_dir2_leaf_log_ents(args
->trans
, bp
, loglow
, to
- 1);
639 * Compact the leaf entries, removing stale ones.
640 * Leave one stale entry behind - the one closest to our
641 * insertion index - and the caller will shift that one to our insertion
643 * Return new insertion index, where the remaining stale entry is,
644 * and leaf logging indices.
647 xfs_dir2_leaf_compact_x1(
648 xfs_dabuf_t
*bp
, /* leaf buffer */
649 int *indexp
, /* insertion index */
650 int *lowstalep
, /* out: stale entry before us */
651 int *highstalep
, /* out: stale entry after us */
652 int *lowlogp
, /* out: low log index */
653 int *highlogp
) /* out: high log index */
655 int from
; /* source copy index */
656 int highstale
; /* stale entry at/after index */
657 int index
; /* insertion index */
658 int keepstale
; /* source index of kept stale */
659 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
660 int lowstale
; /* stale entry before index */
661 int newindex
=0; /* new insertion index */
662 int to
; /* destination copy index */
665 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) > 1);
668 * Find the first stale entry before our index, if any.
670 for (lowstale
= index
- 1;
672 be32_to_cpu(leaf
->ents
[lowstale
].address
) != XFS_DIR2_NULL_DATAPTR
;
676 * Find the first stale entry at or after our index, if any.
677 * Stop if the answer would be worse than lowstale.
679 for (highstale
= index
;
680 highstale
< be16_to_cpu(leaf
->hdr
.count
) &&
681 be32_to_cpu(leaf
->ents
[highstale
].address
) != XFS_DIR2_NULL_DATAPTR
&&
682 (lowstale
< 0 || index
- lowstale
> highstale
- index
);
686 * Pick the better of lowstale and highstale.
689 (highstale
== be16_to_cpu(leaf
->hdr
.count
) ||
690 index
- lowstale
<= highstale
- index
))
691 keepstale
= lowstale
;
693 keepstale
= highstale
;
695 * Copy the entries in place, removing all the stale entries
698 for (from
= to
= 0; from
< be16_to_cpu(leaf
->hdr
.count
); from
++) {
700 * Notice the new value of index.
704 if (from
!= keepstale
&&
705 be32_to_cpu(leaf
->ents
[from
].address
) == XFS_DIR2_NULL_DATAPTR
) {
711 * Record the new keepstale value for the insertion.
713 if (from
== keepstale
)
714 lowstale
= highstale
= to
;
716 * Copy only the entries that have moved.
719 leaf
->ents
[to
] = leaf
->ents
[from
];
724 * If the insertion point was past the last entry,
725 * set the new insertion point accordingly.
731 * Adjust the leaf header values.
733 be16_add_cpu(&leaf
->hdr
.count
, -(from
- to
));
734 leaf
->hdr
.stale
= cpu_to_be16(1);
736 * Remember the low/high stale value only in the "right"
739 if (lowstale
>= newindex
)
742 highstale
= be16_to_cpu(leaf
->hdr
.count
);
743 *highlogp
= be16_to_cpu(leaf
->hdr
.count
) - 1;
744 *lowstalep
= lowstale
;
745 *highstalep
= highstale
;
749 * Getdents (readdir) for leaf and node directories.
750 * This reads the data blocks only, so is the same for both forms.
753 xfs_dir2_leaf_getdents(
754 xfs_inode_t
*dp
, /* incore directory inode */
760 xfs_dabuf_t
*bp
; /* data block buffer */
761 int byteoff
; /* offset in current block */
762 xfs_dir2_db_t curdb
; /* db for current block */
763 xfs_dir2_off_t curoff
; /* current overall offset */
764 xfs_dir2_data_t
*data
; /* data block structure */
765 xfs_dir2_data_entry_t
*dep
; /* data entry */
766 xfs_dir2_data_unused_t
*dup
; /* unused entry */
767 int error
= 0; /* error return value */
768 int i
; /* temporary loop index */
769 int j
; /* temporary loop index */
770 int length
; /* temporary length value */
771 xfs_bmbt_irec_t
*map
; /* map vector for blocks */
772 xfs_extlen_t map_blocks
; /* number of fsbs in map */
773 xfs_dablk_t map_off
; /* last mapped file offset */
774 int map_size
; /* total entries in *map */
775 int map_valid
; /* valid entries in *map */
776 xfs_mount_t
*mp
; /* filesystem mount point */
777 xfs_dir2_off_t newoff
; /* new curoff after new blk */
778 int nmap
; /* mappings to ask xfs_bmapi */
779 char *ptr
= NULL
; /* pointer to current data */
780 int ra_current
; /* number of read-ahead blks */
781 int ra_index
; /* *map index for read-ahead */
782 int ra_offset
; /* map entry offset for ra */
783 int ra_want
; /* readahead count wanted */
786 * If the offset is at or past the largest allowed value,
787 * give up right away.
789 if (*offset
>= XFS_DIR2_MAX_DATAPTR
)
795 * Set up to bmap a number of blocks based on the caller's
796 * buffer size, the directory block size, and the filesystem
799 map_size
= howmany(bufsize
+ mp
->m_dirblksize
, mp
->m_sb
.sb_blocksize
);
800 map
= kmem_alloc(map_size
* sizeof(*map
), KM_SLEEP
);
801 map_valid
= ra_index
= ra_offset
= ra_current
= map_blocks
= 0;
805 * Inside the loop we keep the main offset value as a byte offset
806 * in the directory file.
808 curoff
= xfs_dir2_dataptr_to_byte(mp
, *offset
);
811 * Force this conversion through db so we truncate the offset
812 * down to get the start of the data block.
814 map_off
= xfs_dir2_db_to_da(mp
, xfs_dir2_byte_to_db(mp
, curoff
));
816 * Loop over directory entries until we reach the end offset.
817 * Get more blocks and readahead as necessary.
819 while (curoff
< XFS_DIR2_LEAF_OFFSET
) {
821 * If we have no buffer, or we're off the end of the
822 * current buffer, need to get another one.
824 if (!bp
|| ptr
>= (char *)bp
->data
+ mp
->m_dirblksize
) {
826 * If we have a buffer, we need to release it and
827 * take it out of the mapping.
830 xfs_da_brelse(NULL
, bp
);
832 map_blocks
-= mp
->m_dirblkfsbs
;
834 * Loop to get rid of the extents for the
837 for (i
= mp
->m_dirblkfsbs
; i
> 0; ) {
838 j
= MIN((int)map
->br_blockcount
, i
);
839 map
->br_blockcount
-= j
;
840 map
->br_startblock
+= j
;
841 map
->br_startoff
+= j
;
843 * If mapping is done, pitch it from
846 if (!map
->br_blockcount
&& --map_valid
)
847 memmove(&map
[0], &map
[1],
854 * Recalculate the readahead blocks wanted.
856 ra_want
= howmany(bufsize
+ mp
->m_dirblksize
,
857 mp
->m_sb
.sb_blocksize
) - 1;
858 ASSERT(ra_want
>= 0);
861 * If we don't have as many as we want, and we haven't
862 * run out of data blocks, get some more mappings.
864 if (1 + ra_want
> map_blocks
&&
866 xfs_dir2_byte_to_da(mp
, XFS_DIR2_LEAF_OFFSET
)) {
868 * Get more bmaps, fill in after the ones
869 * we already have in the table.
871 nmap
= map_size
- map_valid
;
872 error
= xfs_bmapi(NULL
, dp
,
874 xfs_dir2_byte_to_da(mp
,
875 XFS_DIR2_LEAF_OFFSET
) - map_off
,
876 XFS_BMAPI_METADATA
, NULL
, 0,
877 &map
[map_valid
], &nmap
, NULL
, NULL
);
879 * Don't know if we should ignore this or
880 * try to return an error.
881 * The trouble with returning errors
882 * is that readdir will just stop without
883 * actually passing the error through.
888 * If we got all the mappings we asked for,
889 * set the final map offset based on the
890 * last bmap value received.
891 * Otherwise, we've reached the end.
893 if (nmap
== map_size
- map_valid
)
895 map
[map_valid
+ nmap
- 1].br_startoff
+
896 map
[map_valid
+ nmap
- 1].br_blockcount
;
899 xfs_dir2_byte_to_da(mp
,
900 XFS_DIR2_LEAF_OFFSET
);
902 * Look for holes in the mapping, and
903 * eliminate them. Count up the valid blocks.
905 for (i
= map_valid
; i
< map_valid
+ nmap
; ) {
906 if (map
[i
].br_startblock
==
909 length
= map_valid
+ nmap
- i
;
917 map
[i
].br_blockcount
;
924 * No valid mappings, so no more data blocks.
927 curoff
= xfs_dir2_da_to_byte(mp
, map_off
);
931 * Read the directory block starting at the first
934 curdb
= xfs_dir2_da_to_db(mp
, map
->br_startoff
);
935 error
= xfs_da_read_buf(NULL
, dp
, map
->br_startoff
,
936 map
->br_blockcount
>= mp
->m_dirblkfsbs
?
937 XFS_FSB_TO_DADDR(mp
, map
->br_startblock
) :
941 * Should just skip over the data block instead
947 * Adjust the current amount of read-ahead: we just
948 * read a block that was previously ra.
951 ra_current
-= mp
->m_dirblkfsbs
;
953 * Do we need more readahead?
955 for (ra_index
= ra_offset
= i
= 0;
956 ra_want
> ra_current
&& i
< map_blocks
;
957 i
+= mp
->m_dirblkfsbs
) {
958 ASSERT(ra_index
< map_valid
);
960 * Read-ahead a contiguous directory block.
962 if (i
> ra_current
&&
963 map
[ra_index
].br_blockcount
>=
965 xfs_baread(mp
->m_ddev_targp
,
967 map
[ra_index
].br_startblock
+
969 (int)BTOBB(mp
->m_dirblksize
));
973 * Read-ahead a non-contiguous directory block.
974 * This doesn't use our mapping, but this
975 * is a very rare case.
977 else if (i
> ra_current
) {
978 (void)xfs_da_reada_buf(NULL
, dp
,
979 map
[ra_index
].br_startoff
+
980 ra_offset
, XFS_DATA_FORK
);
984 * Advance offset through the mapping table.
986 for (j
= 0; j
< mp
->m_dirblkfsbs
; j
++) {
988 * The rest of this extent but not
989 * more than a dir block.
991 length
= MIN(mp
->m_dirblkfsbs
,
992 (int)(map
[ra_index
].br_blockcount
-
997 * Advance to the next mapping if
998 * this one is used up.
1001 map
[ra_index
].br_blockcount
) {
1008 * Having done a read, we need to set a new offset.
1010 newoff
= xfs_dir2_db_off_to_byte(mp
, curdb
, 0);
1012 * Start of the current block.
1014 if (curoff
< newoff
)
1017 * Make sure we're in the right block.
1019 else if (curoff
> newoff
)
1020 ASSERT(xfs_dir2_byte_to_db(mp
, curoff
) ==
1023 xfs_dir2_data_check(dp
, bp
);
1025 * Find our position in the block.
1027 ptr
= (char *)&data
->u
;
1028 byteoff
= xfs_dir2_byte_to_off(mp
, curoff
);
1030 * Skip past the header.
1033 curoff
+= (uint
)sizeof(data
->hdr
);
1035 * Skip past entries until we reach our offset.
1038 while ((char *)ptr
- (char *)data
< byteoff
) {
1039 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1041 if (be16_to_cpu(dup
->freetag
)
1042 == XFS_DIR2_DATA_FREE_TAG
) {
1044 length
= be16_to_cpu(dup
->length
);
1048 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1050 xfs_dir2_data_entsize(dep
->namelen
);
1054 * Now set our real offset.
1057 xfs_dir2_db_off_to_byte(mp
,
1058 xfs_dir2_byte_to_db(mp
, curoff
),
1059 (char *)ptr
- (char *)data
);
1060 if (ptr
>= (char *)data
+ mp
->m_dirblksize
) {
1066 * We have a pointer to an entry.
1069 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1071 * No, it's unused, skip over it.
1073 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
1074 length
= be16_to_cpu(dup
->length
);
1080 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1081 length
= xfs_dir2_data_entsize(dep
->namelen
);
1083 if (filldir(dirent
, (char *)dep
->name
, dep
->namelen
,
1084 xfs_dir2_byte_to_dataptr(mp
, curoff
) & 0x7fffffff,
1085 be64_to_cpu(dep
->inumber
), DT_UNKNOWN
))
1089 * Advance to next entry in the block.
1093 /* bufsize may have just been a guess; don't go negative */
1094 bufsize
= bufsize
> length
? bufsize
- length
: 0;
1098 * All done. Set output offset value to current offset.
1100 if (curoff
> xfs_dir2_dataptr_to_byte(mp
, XFS_DIR2_MAX_DATAPTR
))
1101 *offset
= XFS_DIR2_MAX_DATAPTR
& 0x7fffffff;
1103 *offset
= xfs_dir2_byte_to_dataptr(mp
, curoff
) & 0x7fffffff;
1106 xfs_da_brelse(NULL
, bp
);
1111 * Initialize a new leaf block, leaf1 or leafn magic accepted.
1115 xfs_da_args_t
*args
, /* operation arguments */
1116 xfs_dir2_db_t bno
, /* directory block number */
1117 xfs_dabuf_t
**bpp
, /* out: leaf buffer */
1118 int magic
) /* magic number for block */
1120 xfs_dabuf_t
*bp
; /* leaf buffer */
1121 xfs_inode_t
*dp
; /* incore directory inode */
1122 int error
; /* error return code */
1123 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1124 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1125 xfs_mount_t
*mp
; /* filesystem mount point */
1126 xfs_trans_t
*tp
; /* transaction pointer */
1132 ASSERT(bno
>= XFS_DIR2_LEAF_FIRSTDB(mp
) &&
1133 bno
< XFS_DIR2_FREE_FIRSTDB(mp
));
1135 * Get the buffer for the block.
1137 error
= xfs_da_get_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, bno
), -1, &bp
,
1145 * Initialize the header.
1147 leaf
->hdr
.info
.magic
= cpu_to_be16(magic
);
1148 leaf
->hdr
.info
.forw
= 0;
1149 leaf
->hdr
.info
.back
= 0;
1150 leaf
->hdr
.count
= 0;
1151 leaf
->hdr
.stale
= 0;
1152 xfs_dir2_leaf_log_header(tp
, bp
);
1154 * If it's a leaf-format directory initialize the tail.
1155 * In this case our caller has the real bests table to copy into
1158 if (magic
== XFS_DIR2_LEAF1_MAGIC
) {
1159 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1161 xfs_dir2_leaf_log_tail(tp
, bp
);
1168 * Log the bests entries indicated from a leaf1 block.
1171 xfs_dir2_leaf_log_bests(
1172 xfs_trans_t
*tp
, /* transaction pointer */
1173 xfs_dabuf_t
*bp
, /* leaf buffer */
1174 int first
, /* first entry to log */
1175 int last
) /* last entry to log */
1177 __be16
*firstb
; /* pointer to first entry */
1178 __be16
*lastb
; /* pointer to last entry */
1179 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1180 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1183 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
);
1184 ltp
= xfs_dir2_leaf_tail_p(tp
->t_mountp
, leaf
);
1185 firstb
= xfs_dir2_leaf_bests_p(ltp
) + first
;
1186 lastb
= xfs_dir2_leaf_bests_p(ltp
) + last
;
1187 xfs_da_log_buf(tp
, bp
, (uint
)((char *)firstb
- (char *)leaf
),
1188 (uint
)((char *)lastb
- (char *)leaf
+ sizeof(*lastb
) - 1));
1192 * Log the leaf entries indicated from a leaf1 or leafn block.
1195 xfs_dir2_leaf_log_ents(
1196 xfs_trans_t
*tp
, /* transaction pointer */
1197 xfs_dabuf_t
*bp
, /* leaf buffer */
1198 int first
, /* first entry to log */
1199 int last
) /* last entry to log */
1201 xfs_dir2_leaf_entry_t
*firstlep
; /* pointer to first entry */
1202 xfs_dir2_leaf_entry_t
*lastlep
; /* pointer to last entry */
1203 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1206 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
||
1207 be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
1208 firstlep
= &leaf
->ents
[first
];
1209 lastlep
= &leaf
->ents
[last
];
1210 xfs_da_log_buf(tp
, bp
, (uint
)((char *)firstlep
- (char *)leaf
),
1211 (uint
)((char *)lastlep
- (char *)leaf
+ sizeof(*lastlep
) - 1));
1215 * Log the header of the leaf1 or leafn block.
1218 xfs_dir2_leaf_log_header(
1219 xfs_trans_t
*tp
, /* transaction pointer */
1220 xfs_dabuf_t
*bp
) /* leaf buffer */
1222 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1225 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
||
1226 be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
1227 xfs_da_log_buf(tp
, bp
, (uint
)((char *)&leaf
->hdr
- (char *)leaf
),
1228 (uint
)(sizeof(leaf
->hdr
) - 1));
1232 * Log the tail of the leaf1 block.
1235 xfs_dir2_leaf_log_tail(
1236 xfs_trans_t
*tp
, /* transaction pointer */
1237 xfs_dabuf_t
*bp
) /* leaf buffer */
1239 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1240 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1241 xfs_mount_t
*mp
; /* filesystem mount point */
1245 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
);
1246 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1247 xfs_da_log_buf(tp
, bp
, (uint
)((char *)ltp
- (char *)leaf
),
1248 (uint
)(mp
->m_dirblksize
- 1));
1252 * Look up the entry referred to by args in the leaf format directory.
1253 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1254 * is also used by the node-format code.
1257 xfs_dir2_leaf_lookup(
1258 xfs_da_args_t
*args
) /* operation arguments */
1260 xfs_dabuf_t
*dbp
; /* data block buffer */
1261 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1262 xfs_inode_t
*dp
; /* incore directory inode */
1263 int error
; /* error return code */
1264 int index
; /* found entry index */
1265 xfs_dabuf_t
*lbp
; /* leaf buffer */
1266 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1267 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1268 xfs_trans_t
*tp
; /* transaction pointer */
1270 trace_xfs_dir2_leaf_lookup(args
);
1273 * Look up name in the leaf block, returning both buffers and index.
1275 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1280 xfs_dir2_leaf_check(dp
, lbp
);
1283 * Get to the leaf entry and contained data entry address.
1285 lep
= &leaf
->ents
[index
];
1287 * Point to the data entry.
1289 dep
= (xfs_dir2_data_entry_t
*)
1290 ((char *)dbp
->data
+
1291 xfs_dir2_dataptr_to_off(dp
->i_mount
, be32_to_cpu(lep
->address
)));
1293 * Return the found inode number & CI name if appropriate
1295 args
->inumber
= be64_to_cpu(dep
->inumber
);
1296 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
1297 xfs_da_brelse(tp
, dbp
);
1298 xfs_da_brelse(tp
, lbp
);
1299 return XFS_ERROR(error
);
1303 * Look up name/hash in the leaf block.
1304 * Fill in indexp with the found index, and dbpp with the data buffer.
1305 * If not found dbpp will be NULL, and ENOENT comes back.
1306 * lbpp will always be filled in with the leaf buffer unless there's an error.
1308 static int /* error */
1309 xfs_dir2_leaf_lookup_int(
1310 xfs_da_args_t
*args
, /* operation arguments */
1311 xfs_dabuf_t
**lbpp
, /* out: leaf buffer */
1312 int *indexp
, /* out: index in leaf block */
1313 xfs_dabuf_t
**dbpp
) /* out: data buffer */
1315 xfs_dir2_db_t curdb
= -1; /* current data block number */
1316 xfs_dabuf_t
*dbp
= NULL
; /* data buffer */
1317 xfs_dir2_data_entry_t
*dep
; /* data entry */
1318 xfs_inode_t
*dp
; /* incore directory inode */
1319 int error
; /* error return code */
1320 int index
; /* index in leaf block */
1321 xfs_dabuf_t
*lbp
; /* leaf buffer */
1322 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1323 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1324 xfs_mount_t
*mp
; /* filesystem mount point */
1325 xfs_dir2_db_t newdb
; /* new data block number */
1326 xfs_trans_t
*tp
; /* transaction pointer */
1327 xfs_dir2_db_t cidb
= -1; /* case match data block no. */
1328 enum xfs_dacmp cmp
; /* name compare result */
1334 * Read the leaf block into the buffer.
1336 error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirleafblk
, -1, &lbp
,
1342 xfs_dir2_leaf_check(dp
, lbp
);
1344 * Look for the first leaf entry with our hash value.
1346 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
1348 * Loop over all the entries with the right hash value
1349 * looking to match the name.
1351 for (lep
= &leaf
->ents
[index
]; index
< be16_to_cpu(leaf
->hdr
.count
) &&
1352 be32_to_cpu(lep
->hashval
) == args
->hashval
;
1355 * Skip over stale leaf entries.
1357 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
1360 * Get the new data block number.
1362 newdb
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
1364 * If it's not the same as the old data block number,
1365 * need to pitch the old one and read the new one.
1367 if (newdb
!= curdb
) {
1369 xfs_da_brelse(tp
, dbp
);
1370 error
= xfs_da_read_buf(tp
, dp
,
1371 xfs_dir2_db_to_da(mp
, newdb
),
1372 -1, &dbp
, XFS_DATA_FORK
);
1374 xfs_da_brelse(tp
, lbp
);
1377 xfs_dir2_data_check(dp
, dbp
);
1381 * Point to the data entry.
1383 dep
= (xfs_dir2_data_entry_t
*)((char *)dbp
->data
+
1384 xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
1386 * Compare name and if it's an exact match, return the index
1387 * and buffer. If it's the first case-insensitive match, store
1388 * the index and buffer and continue looking for an exact match.
1390 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
1391 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
1392 args
->cmpresult
= cmp
;
1394 /* case exact match: return the current buffer. */
1395 if (cmp
== XFS_CMP_EXACT
) {
1402 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
1404 * Here, we can only be doing a lookup (not a rename or remove).
1405 * If a case-insensitive match was found earlier, re-read the
1406 * appropriate data block if required and return it.
1408 if (args
->cmpresult
== XFS_CMP_CASE
) {
1410 if (cidb
!= curdb
) {
1411 xfs_da_brelse(tp
, dbp
);
1412 error
= xfs_da_read_buf(tp
, dp
,
1413 xfs_dir2_db_to_da(mp
, cidb
),
1414 -1, &dbp
, XFS_DATA_FORK
);
1416 xfs_da_brelse(tp
, lbp
);
1424 * No match found, return ENOENT.
1428 xfs_da_brelse(tp
, dbp
);
1429 xfs_da_brelse(tp
, lbp
);
1430 return XFS_ERROR(ENOENT
);
1434 * Remove an entry from a leaf format directory.
1437 xfs_dir2_leaf_removename(
1438 xfs_da_args_t
*args
) /* operation arguments */
1440 __be16
*bestsp
; /* leaf block best freespace */
1441 xfs_dir2_data_t
*data
; /* data block structure */
1442 xfs_dir2_db_t db
; /* data block number */
1443 xfs_dabuf_t
*dbp
; /* data block buffer */
1444 xfs_dir2_data_entry_t
*dep
; /* data entry structure */
1445 xfs_inode_t
*dp
; /* incore directory inode */
1446 int error
; /* error return code */
1447 xfs_dir2_db_t i
; /* temporary data block # */
1448 int index
; /* index into leaf entries */
1449 xfs_dabuf_t
*lbp
; /* leaf buffer */
1450 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1451 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1452 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1453 xfs_mount_t
*mp
; /* filesystem mount point */
1454 int needlog
; /* need to log data header */
1455 int needscan
; /* need to rescan data frees */
1456 xfs_dir2_data_off_t oldbest
; /* old value of best free */
1457 xfs_trans_t
*tp
; /* transaction pointer */
1459 trace_xfs_dir2_leaf_removename(args
);
1462 * Lookup the leaf entry, get the leaf and data blocks read in.
1464 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1472 xfs_dir2_data_check(dp
, dbp
);
1474 * Point to the leaf entry, use that to point to the data entry.
1476 lep
= &leaf
->ents
[index
];
1477 db
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
1478 dep
= (xfs_dir2_data_entry_t
*)
1479 ((char *)data
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
1480 needscan
= needlog
= 0;
1481 oldbest
= be16_to_cpu(data
->hdr
.bestfree
[0].length
);
1482 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1483 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1484 ASSERT(be16_to_cpu(bestsp
[db
]) == oldbest
);
1486 * Mark the former data entry unused.
1488 xfs_dir2_data_make_free(tp
, dbp
,
1489 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)data
),
1490 xfs_dir2_data_entsize(dep
->namelen
), &needlog
, &needscan
);
1492 * We just mark the leaf entry stale by putting a null in it.
1494 be16_add_cpu(&leaf
->hdr
.stale
, 1);
1495 xfs_dir2_leaf_log_header(tp
, lbp
);
1496 lep
->address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
1497 xfs_dir2_leaf_log_ents(tp
, lbp
, index
, index
);
1499 * Scan the freespace in the data block again if necessary,
1500 * log the data block header if necessary.
1503 xfs_dir2_data_freescan(mp
, data
, &needlog
);
1505 xfs_dir2_data_log_header(tp
, dbp
);
1507 * If the longest freespace in the data block has changed,
1508 * put the new value in the bests table and log that.
1510 if (be16_to_cpu(data
->hdr
.bestfree
[0].length
) != oldbest
) {
1511 bestsp
[db
] = data
->hdr
.bestfree
[0].length
;
1512 xfs_dir2_leaf_log_bests(tp
, lbp
, db
, db
);
1514 xfs_dir2_data_check(dp
, dbp
);
1516 * If the data block is now empty then get rid of the data block.
1518 if (be16_to_cpu(data
->hdr
.bestfree
[0].length
) ==
1519 mp
->m_dirblksize
- (uint
)sizeof(data
->hdr
)) {
1520 ASSERT(db
!= mp
->m_dirdatablk
);
1521 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1523 * Nope, can't get rid of it because it caused
1524 * allocation of a bmap btree block to do so.
1525 * Just go on, returning success, leaving the
1526 * empty block in place.
1528 if (error
== ENOSPC
&& args
->total
== 0) {
1529 xfs_da_buf_done(dbp
);
1532 xfs_dir2_leaf_check(dp
, lbp
);
1533 xfs_da_buf_done(lbp
);
1538 * If this is the last data block then compact the
1539 * bests table by getting rid of entries.
1541 if (db
== be32_to_cpu(ltp
->bestcount
) - 1) {
1543 * Look for the last active entry (i).
1545 for (i
= db
- 1; i
> 0; i
--) {
1546 if (be16_to_cpu(bestsp
[i
]) != NULLDATAOFF
)
1550 * Copy the table down so inactive entries at the
1553 memmove(&bestsp
[db
- i
], bestsp
,
1554 (be32_to_cpu(ltp
->bestcount
) - (db
- i
)) * sizeof(*bestsp
));
1555 be32_add_cpu(<p
->bestcount
, -(db
- i
));
1556 xfs_dir2_leaf_log_tail(tp
, lbp
);
1557 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1559 bestsp
[db
] = cpu_to_be16(NULLDATAOFF
);
1562 * If the data block was not the first one, drop it.
1564 else if (db
!= mp
->m_dirdatablk
&& dbp
!= NULL
) {
1565 xfs_da_buf_done(dbp
);
1568 xfs_dir2_leaf_check(dp
, lbp
);
1570 * See if we can convert to block form.
1572 return xfs_dir2_leaf_to_block(args
, lbp
, dbp
);
1576 * Replace the inode number in a leaf format directory entry.
1579 xfs_dir2_leaf_replace(
1580 xfs_da_args_t
*args
) /* operation arguments */
1582 xfs_dabuf_t
*dbp
; /* data block buffer */
1583 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1584 xfs_inode_t
*dp
; /* incore directory inode */
1585 int error
; /* error return code */
1586 int index
; /* index of leaf entry */
1587 xfs_dabuf_t
*lbp
; /* leaf buffer */
1588 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1589 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1590 xfs_trans_t
*tp
; /* transaction pointer */
1592 trace_xfs_dir2_leaf_replace(args
);
1595 * Look up the entry.
1597 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1603 * Point to the leaf entry, get data address from it.
1605 lep
= &leaf
->ents
[index
];
1607 * Point to the data entry.
1609 dep
= (xfs_dir2_data_entry_t
*)
1610 ((char *)dbp
->data
+
1611 xfs_dir2_dataptr_to_off(dp
->i_mount
, be32_to_cpu(lep
->address
)));
1612 ASSERT(args
->inumber
!= be64_to_cpu(dep
->inumber
));
1614 * Put the new inode number in, log it.
1616 dep
->inumber
= cpu_to_be64(args
->inumber
);
1618 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
1619 xfs_da_buf_done(dbp
);
1620 xfs_dir2_leaf_check(dp
, lbp
);
1621 xfs_da_brelse(tp
, lbp
);
1626 * Return index in the leaf block (lbp) which is either the first
1627 * one with this hash value, or if there are none, the insert point
1628 * for that hash value.
1630 int /* index value */
1631 xfs_dir2_leaf_search_hash(
1632 xfs_da_args_t
*args
, /* operation arguments */
1633 xfs_dabuf_t
*lbp
) /* leaf buffer */
1635 xfs_dahash_t hash
=0; /* hash from this entry */
1636 xfs_dahash_t hashwant
; /* hash value looking for */
1637 int high
; /* high leaf index */
1638 int low
; /* low leaf index */
1639 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1640 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1641 int mid
=0; /* current leaf index */
1645 if (!leaf
->hdr
.count
)
1649 * Note, the table cannot be empty, so we have to go through the loop.
1650 * Binary search the leaf entries looking for our hash value.
1652 for (lep
= leaf
->ents
, low
= 0, high
= be16_to_cpu(leaf
->hdr
.count
) - 1,
1653 hashwant
= args
->hashval
;
1655 mid
= (low
+ high
) >> 1;
1656 if ((hash
= be32_to_cpu(lep
[mid
].hashval
)) == hashwant
)
1658 if (hash
< hashwant
)
1664 * Found one, back up through all the equal hash values.
1666 if (hash
== hashwant
) {
1667 while (mid
> 0 && be32_to_cpu(lep
[mid
- 1].hashval
) == hashwant
) {
1672 * Need to point to an entry higher than ours.
1674 else if (hash
< hashwant
)
1680 * Trim off a trailing data block. We know it's empty since the leaf
1681 * freespace table says so.
1684 xfs_dir2_leaf_trim_data(
1685 xfs_da_args_t
*args
, /* operation arguments */
1686 xfs_dabuf_t
*lbp
, /* leaf buffer */
1687 xfs_dir2_db_t db
) /* data block number */
1689 __be16
*bestsp
; /* leaf bests table */
1691 xfs_dir2_data_t
*data
; /* data block structure */
1693 xfs_dabuf_t
*dbp
; /* data block buffer */
1694 xfs_inode_t
*dp
; /* incore directory inode */
1695 int error
; /* error return value */
1696 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1697 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1698 xfs_mount_t
*mp
; /* filesystem mount point */
1699 xfs_trans_t
*tp
; /* transaction pointer */
1705 * Read the offending data block. We need its buffer.
1707 if ((error
= xfs_da_read_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, db
), -1, &dbp
,
1713 ASSERT(be32_to_cpu(data
->hdr
.magic
) == XFS_DIR2_DATA_MAGIC
);
1715 /* this seems to be an error
1716 * data is only valid if DEBUG is defined?
1721 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1722 ASSERT(be16_to_cpu(data
->hdr
.bestfree
[0].length
) ==
1723 mp
->m_dirblksize
- (uint
)sizeof(data
->hdr
));
1724 ASSERT(db
== be32_to_cpu(ltp
->bestcount
) - 1);
1726 * Get rid of the data block.
1728 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1729 ASSERT(error
!= ENOSPC
);
1730 xfs_da_brelse(tp
, dbp
);
1734 * Eliminate the last bests entry from the table.
1736 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1737 be32_add_cpu(<p
->bestcount
, -1);
1738 memmove(&bestsp
[1], &bestsp
[0], be32_to_cpu(ltp
->bestcount
) * sizeof(*bestsp
));
1739 xfs_dir2_leaf_log_tail(tp
, lbp
);
1740 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1745 * Convert node form directory to leaf form directory.
1746 * The root of the node form dir needs to already be a LEAFN block.
1747 * Just return if we can't do anything.
1750 xfs_dir2_node_to_leaf(
1751 xfs_da_state_t
*state
) /* directory operation state */
1753 xfs_da_args_t
*args
; /* operation arguments */
1754 xfs_inode_t
*dp
; /* incore directory inode */
1755 int error
; /* error return code */
1756 xfs_dabuf_t
*fbp
; /* buffer for freespace block */
1757 xfs_fileoff_t fo
; /* freespace file offset */
1758 xfs_dir2_free_t
*free
; /* freespace structure */
1759 xfs_dabuf_t
*lbp
; /* buffer for leaf block */
1760 xfs_dir2_leaf_tail_t
*ltp
; /* tail of leaf structure */
1761 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1762 xfs_mount_t
*mp
; /* filesystem mount point */
1763 int rval
; /* successful free trim? */
1764 xfs_trans_t
*tp
; /* transaction pointer */
1767 * There's more than a leaf level in the btree, so there must
1768 * be multiple leafn blocks. Give up.
1770 if (state
->path
.active
> 1)
1774 trace_xfs_dir2_node_to_leaf(args
);
1780 * Get the last offset in the file.
1782 if ((error
= xfs_bmap_last_offset(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1785 fo
-= mp
->m_dirblkfsbs
;
1787 * If there are freespace blocks other than the first one,
1788 * take this opportunity to remove trailing empty freespace blocks
1789 * that may have been left behind during no-space-reservation
1792 while (fo
> mp
->m_dirfreeblk
) {
1793 if ((error
= xfs_dir2_node_trim_free(args
, fo
, &rval
))) {
1797 fo
-= mp
->m_dirblkfsbs
;
1802 * Now find the block just before the freespace block.
1804 if ((error
= xfs_bmap_last_before(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1808 * If it's not the single leaf block, give up.
1810 if (XFS_FSB_TO_B(mp
, fo
) > XFS_DIR2_LEAF_OFFSET
+ mp
->m_dirblksize
)
1812 lbp
= state
->path
.blk
[0].bp
;
1814 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
1816 * Read the freespace block.
1818 if ((error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirfreeblk
, -1, &fbp
,
1823 ASSERT(be32_to_cpu(free
->hdr
.magic
) == XFS_DIR2_FREE_MAGIC
);
1824 ASSERT(!free
->hdr
.firstdb
);
1826 * Now see if the leafn and free data will fit in a leaf1.
1827 * If not, release the buffer and give up.
1829 if ((uint
)sizeof(leaf
->hdr
) +
1830 (be16_to_cpu(leaf
->hdr
.count
) - be16_to_cpu(leaf
->hdr
.stale
)) * (uint
)sizeof(leaf
->ents
[0]) +
1831 be32_to_cpu(free
->hdr
.nvalid
) * (uint
)sizeof(leaf
->bests
[0]) +
1832 (uint
)sizeof(leaf
->tail
) >
1834 xfs_da_brelse(tp
, fbp
);
1838 * If the leaf has any stale entries in it, compress them out.
1839 * The compact routine will log the header.
1841 if (be16_to_cpu(leaf
->hdr
.stale
))
1842 xfs_dir2_leaf_compact(args
, lbp
);
1844 xfs_dir2_leaf_log_header(tp
, lbp
);
1845 leaf
->hdr
.info
.magic
= cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
);
1847 * Set up the leaf tail from the freespace block.
1849 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1850 ltp
->bestcount
= free
->hdr
.nvalid
;
1852 * Set up the leaf bests table.
1854 memcpy(xfs_dir2_leaf_bests_p(ltp
), free
->bests
,
1855 be32_to_cpu(ltp
->bestcount
) * sizeof(leaf
->bests
[0]));
1856 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1857 xfs_dir2_leaf_log_tail(tp
, lbp
);
1858 xfs_dir2_leaf_check(dp
, lbp
);
1860 * Get rid of the freespace block.
1862 error
= xfs_dir2_shrink_inode(args
, XFS_DIR2_FREE_FIRSTDB(mp
), fbp
);
1865 * This can't fail here because it can only happen when
1866 * punching out the middle of an extent, and this is an
1869 ASSERT(error
!= ENOSPC
);
1874 * Now see if we can convert the single-leaf directory
1875 * down to a block form directory.
1876 * This routine always kills the dabuf for the leaf, so
1877 * eliminate it from the path.
1879 error
= xfs_dir2_leaf_to_block(args
, lbp
, NULL
);
1880 state
->path
.blk
[0].bp
= NULL
;