2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
23 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
32 #include "xfs_dir2_format.h"
33 #include "xfs_dir2_priv.h"
34 #include "xfs_error.h"
35 #include "xfs_trace.h"
38 * Local function declarations.
41 static void xfs_dir2_leaf_check(struct xfs_inode
*dp
, struct xfs_buf
*bp
);
43 #define xfs_dir2_leaf_check(dp, bp)
45 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t
*args
, struct xfs_buf
**lbpp
,
46 int *indexp
, struct xfs_buf
**dbpp
);
47 static void xfs_dir2_leaf_log_bests(struct xfs_trans
*tp
, struct xfs_buf
*bp
,
49 static void xfs_dir2_leaf_log_tail(struct xfs_trans
*tp
, struct xfs_buf
*bp
);
53 * Convert a block form directory to a leaf form directory.
56 xfs_dir2_block_to_leaf(
57 xfs_da_args_t
*args
, /* operation arguments */
58 struct xfs_buf
*dbp
) /* input block's buffer */
60 __be16
*bestsp
; /* leaf's bestsp entries */
61 xfs_dablk_t blkno
; /* leaf block's bno */
62 xfs_dir2_data_hdr_t
*hdr
; /* block header */
63 xfs_dir2_leaf_entry_t
*blp
; /* block's leaf entries */
64 xfs_dir2_block_tail_t
*btp
; /* block's tail */
65 xfs_inode_t
*dp
; /* incore directory inode */
66 int error
; /* error return code */
67 struct xfs_buf
*lbp
; /* leaf block's buffer */
68 xfs_dir2_db_t ldb
; /* leaf block's bno */
69 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
70 xfs_dir2_leaf_tail_t
*ltp
; /* leaf's tail */
71 xfs_mount_t
*mp
; /* filesystem mount point */
72 int needlog
; /* need to log block header */
73 int needscan
; /* need to rescan bestfree */
74 xfs_trans_t
*tp
; /* transaction pointer */
76 trace_xfs_dir2_block_to_leaf(args
);
82 * Add the leaf block to the inode.
83 * This interface will only put blocks in the leaf/node range.
84 * Since that's empty now, we'll get the root (block 0 in range).
86 if ((error
= xfs_da_grow_inode(args
, &blkno
))) {
89 ldb
= xfs_dir2_da_to_db(mp
, blkno
);
90 ASSERT(ldb
== XFS_DIR2_LEAF_FIRSTDB(mp
));
92 * Initialize the leaf block, get a buffer for it.
94 if ((error
= xfs_dir2_leaf_init(args
, ldb
, &lbp
, XFS_DIR2_LEAF1_MAGIC
))) {
100 xfs_dir2_data_check(dp
, dbp
);
101 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
102 blp
= xfs_dir2_block_leaf_p(btp
);
104 * Set the counts in the leaf header.
106 leaf
->hdr
.count
= cpu_to_be16(be32_to_cpu(btp
->count
));
107 leaf
->hdr
.stale
= cpu_to_be16(be32_to_cpu(btp
->stale
));
109 * Could compact these but I think we always do the conversion
110 * after squeezing out stale entries.
112 memcpy(leaf
->ents
, blp
, be32_to_cpu(btp
->count
) * sizeof(xfs_dir2_leaf_entry_t
));
113 xfs_dir2_leaf_log_ents(tp
, lbp
, 0, be16_to_cpu(leaf
->hdr
.count
) - 1);
117 * Make the space formerly occupied by the leaf entries and block
120 xfs_dir2_data_make_free(tp
, dbp
,
121 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)hdr
),
122 (xfs_dir2_data_aoff_t
)((char *)hdr
+ mp
->m_dirblksize
-
124 &needlog
, &needscan
);
126 * Fix up the block header, make it a data block.
128 hdr
->magic
= cpu_to_be32(XFS_DIR2_DATA_MAGIC
);
130 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
132 * Set up leaf tail and bests table.
134 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
135 ltp
->bestcount
= cpu_to_be32(1);
136 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
137 bestsp
[0] = hdr
->bestfree
[0].length
;
139 * Log the data header and leaf bests table.
142 xfs_dir2_data_log_header(tp
, dbp
);
143 xfs_dir2_leaf_check(dp
, lbp
);
144 xfs_dir2_data_check(dp
, dbp
);
145 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, 0);
150 xfs_dir2_leaf_find_stale(
151 struct xfs_dir2_leaf
*leaf
,
157 * Find the first stale entry before our index, if any.
159 for (*lowstale
= index
- 1; *lowstale
>= 0; --*lowstale
) {
160 if (leaf
->ents
[*lowstale
].address
==
161 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
166 * Find the first stale entry at or after our index, if any.
167 * Stop if the result would require moving more entries than using
170 for (*highstale
= index
;
171 *highstale
< be16_to_cpu(leaf
->hdr
.count
);
173 if (leaf
->ents
[*highstale
].address
==
174 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
176 if (*lowstale
>= 0 && index
- *lowstale
<= *highstale
- index
)
181 struct xfs_dir2_leaf_entry
*
182 xfs_dir2_leaf_find_entry(
183 xfs_dir2_leaf_t
*leaf
, /* leaf structure */
184 int index
, /* leaf table position */
185 int compact
, /* need to compact leaves */
186 int lowstale
, /* index of prev stale leaf */
187 int highstale
, /* index of next stale leaf */
188 int *lfloglow
, /* low leaf logging index */
189 int *lfloghigh
) /* high leaf logging index */
191 if (!leaf
->hdr
.stale
) {
192 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry table pointer */
195 * Now we need to make room to insert the leaf entry.
197 * If there are no stale entries, just insert a hole at index.
199 lep
= &leaf
->ents
[index
];
200 if (index
< be16_to_cpu(leaf
->hdr
.count
))
201 memmove(lep
+ 1, lep
,
202 (be16_to_cpu(leaf
->hdr
.count
) - index
) *
206 * Record low and high logging indices for the leaf.
209 *lfloghigh
= be16_to_cpu(leaf
->hdr
.count
);
210 be16_add_cpu(&leaf
->hdr
.count
, 1);
215 * There are stale entries.
217 * We will use one of them for the new entry. It's probably not at
218 * the right location, so we'll have to shift some up or down first.
220 * If we didn't compact before, we need to find the nearest stale
221 * entries before and after our insertion point.
224 xfs_dir2_leaf_find_stale(leaf
, index
, &lowstale
, &highstale
);
227 * If the low one is better, use it.
230 (highstale
== be16_to_cpu(leaf
->hdr
.count
) ||
231 index
- lowstale
- 1 < highstale
- index
)) {
232 ASSERT(index
- lowstale
- 1 >= 0);
233 ASSERT(leaf
->ents
[lowstale
].address
==
234 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
));
237 * Copy entries up to cover the stale entry and make room
240 if (index
- lowstale
- 1 > 0) {
241 memmove(&leaf
->ents
[lowstale
],
242 &leaf
->ents
[lowstale
+ 1],
243 (index
- lowstale
- 1) *
244 sizeof(xfs_dir2_leaf_entry_t
));
246 *lfloglow
= MIN(lowstale
, *lfloglow
);
247 *lfloghigh
= MAX(index
- 1, *lfloghigh
);
248 be16_add_cpu(&leaf
->hdr
.stale
, -1);
249 return &leaf
->ents
[index
- 1];
253 * The high one is better, so use that one.
255 ASSERT(highstale
- index
>= 0);
256 ASSERT(leaf
->ents
[highstale
].address
==
257 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
));
260 * Copy entries down to cover the stale entry and make room for the
263 if (highstale
- index
> 0) {
264 memmove(&leaf
->ents
[index
+ 1],
266 (highstale
- index
) * sizeof(xfs_dir2_leaf_entry_t
));
268 *lfloglow
= MIN(index
, *lfloglow
);
269 *lfloghigh
= MAX(highstale
, *lfloghigh
);
270 be16_add_cpu(&leaf
->hdr
.stale
, -1);
271 return &leaf
->ents
[index
];
275 * Add an entry to a leaf form directory.
278 xfs_dir2_leaf_addname(
279 xfs_da_args_t
*args
) /* operation arguments */
281 __be16
*bestsp
; /* freespace table in leaf */
282 int compact
; /* need to compact leaves */
283 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
284 struct xfs_buf
*dbp
; /* data block buffer */
285 xfs_dir2_data_entry_t
*dep
; /* data block entry */
286 xfs_inode_t
*dp
; /* incore directory inode */
287 xfs_dir2_data_unused_t
*dup
; /* data unused entry */
288 int error
; /* error return value */
289 int grown
; /* allocated new data block */
290 int highstale
; /* index of next stale leaf */
291 int i
; /* temporary, index */
292 int index
; /* leaf table position */
293 struct xfs_buf
*lbp
; /* leaf's buffer */
294 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
295 int length
; /* length of new entry */
296 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry table pointer */
297 int lfloglow
; /* low leaf logging index */
298 int lfloghigh
; /* high leaf logging index */
299 int lowstale
; /* index of prev stale leaf */
300 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail pointer */
301 xfs_mount_t
*mp
; /* filesystem mount point */
302 int needbytes
; /* leaf block bytes needed */
303 int needlog
; /* need to log data header */
304 int needscan
; /* need to rescan data free */
305 __be16
*tagp
; /* end of data entry */
306 xfs_trans_t
*tp
; /* transaction pointer */
307 xfs_dir2_db_t use_block
; /* data block number */
309 trace_xfs_dir2_leaf_addname(args
);
315 * Read the leaf block.
317 error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirleafblk
, -1, &lbp
,
324 * Look up the entry by hash value and name.
325 * We know it's not there, our caller has already done a lookup.
326 * So the index is of the entry to insert in front of.
327 * But if there are dup hash values the index is of the first of those.
329 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
331 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
332 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
333 length
= xfs_dir2_data_entsize(args
->namelen
);
335 * See if there are any entries with the same hash value
336 * and space in their block for the new entry.
337 * This is good because it puts multiple same-hash value entries
338 * in a data block, improving the lookup of those entries.
340 for (use_block
= -1, lep
= &leaf
->ents
[index
];
341 index
< be16_to_cpu(leaf
->hdr
.count
) && be32_to_cpu(lep
->hashval
) == args
->hashval
;
343 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
345 i
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
346 ASSERT(i
< be32_to_cpu(ltp
->bestcount
));
347 ASSERT(bestsp
[i
] != cpu_to_be16(NULLDATAOFF
));
348 if (be16_to_cpu(bestsp
[i
]) >= length
) {
354 * Didn't find a block yet, linear search all the data blocks.
356 if (use_block
== -1) {
357 for (i
= 0; i
< be32_to_cpu(ltp
->bestcount
); i
++) {
359 * Remember a block we see that's missing.
361 if (bestsp
[i
] == cpu_to_be16(NULLDATAOFF
) &&
364 else if (be16_to_cpu(bestsp
[i
]) >= length
) {
371 * How many bytes do we need in the leaf block?
374 if (!leaf
->hdr
.stale
)
375 needbytes
+= sizeof(xfs_dir2_leaf_entry_t
);
377 needbytes
+= sizeof(xfs_dir2_data_off_t
);
380 * Now kill use_block if it refers to a missing block, so we
381 * can use it as an indication of allocation needed.
383 if (use_block
!= -1 && bestsp
[use_block
] == cpu_to_be16(NULLDATAOFF
))
386 * If we don't have enough free bytes but we can make enough
387 * by compacting out stale entries, we'll do that.
389 if ((char *)bestsp
- (char *)&leaf
->ents
[be16_to_cpu(leaf
->hdr
.count
)] <
390 needbytes
&& be16_to_cpu(leaf
->hdr
.stale
) > 1) {
394 * Otherwise if we don't have enough free bytes we need to
395 * convert to node form.
397 else if ((char *)bestsp
- (char *)&leaf
->ents
[be16_to_cpu(
398 leaf
->hdr
.count
)] < needbytes
) {
400 * Just checking or no space reservation, give up.
402 if ((args
->op_flags
& XFS_DA_OP_JUSTCHECK
) ||
404 xfs_trans_brelse(tp
, lbp
);
405 return XFS_ERROR(ENOSPC
);
408 * Convert to node form.
410 error
= xfs_dir2_leaf_to_node(args
, lbp
);
414 * Then add the new entry.
416 return xfs_dir2_node_addname(args
);
419 * Otherwise it will fit without compaction.
424 * If just checking, then it will fit unless we needed to allocate
427 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
428 xfs_trans_brelse(tp
, lbp
);
429 return use_block
== -1 ? XFS_ERROR(ENOSPC
) : 0;
432 * If no allocations are allowed, return now before we've
435 if (args
->total
== 0 && use_block
== -1) {
436 xfs_trans_brelse(tp
, lbp
);
437 return XFS_ERROR(ENOSPC
);
440 * Need to compact the leaf entries, removing stale ones.
441 * Leave one stale entry behind - the one closest to our
442 * insertion index - and we'll shift that one to our insertion
446 xfs_dir2_leaf_compact_x1(lbp
, &index
, &lowstale
, &highstale
,
447 &lfloglow
, &lfloghigh
);
450 * There are stale entries, so we'll need log-low and log-high
451 * impossibly bad values later.
453 else if (be16_to_cpu(leaf
->hdr
.stale
)) {
454 lfloglow
= be16_to_cpu(leaf
->hdr
.count
);
458 * If there was no data block space found, we need to allocate
461 if (use_block
== -1) {
463 * Add the new data block.
465 if ((error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
,
467 xfs_trans_brelse(tp
, lbp
);
471 * Initialize the block.
473 if ((error
= xfs_dir2_data_init(args
, use_block
, &dbp
))) {
474 xfs_trans_brelse(tp
, lbp
);
478 * If we're adding a new data block on the end we need to
479 * extend the bests table. Copy it up one entry.
481 if (use_block
>= be32_to_cpu(ltp
->bestcount
)) {
483 memmove(&bestsp
[0], &bestsp
[1],
484 be32_to_cpu(ltp
->bestcount
) * sizeof(bestsp
[0]));
485 be32_add_cpu(<p
->bestcount
, 1);
486 xfs_dir2_leaf_log_tail(tp
, lbp
);
487 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
490 * If we're filling in a previously empty block just log it.
493 xfs_dir2_leaf_log_bests(tp
, lbp
, use_block
, use_block
);
495 bestsp
[use_block
] = hdr
->bestfree
[0].length
;
499 * Already had space in some data block.
500 * Just read that one in.
504 xfs_da_read_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, use_block
),
505 -1, &dbp
, XFS_DATA_FORK
))) {
506 xfs_trans_brelse(tp
, lbp
);
512 xfs_dir2_data_check(dp
, dbp
);
514 * Point to the biggest freespace in our data block.
516 dup
= (xfs_dir2_data_unused_t
*)
517 ((char *)hdr
+ be16_to_cpu(hdr
->bestfree
[0].offset
));
518 ASSERT(be16_to_cpu(dup
->length
) >= length
);
519 needscan
= needlog
= 0;
521 * Mark the initial part of our freespace in use for the new entry.
523 xfs_dir2_data_use_free(tp
, dbp
, dup
,
524 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
), length
,
525 &needlog
, &needscan
);
527 * Initialize our new entry (at last).
529 dep
= (xfs_dir2_data_entry_t
*)dup
;
530 dep
->inumber
= cpu_to_be64(args
->inumber
);
531 dep
->namelen
= args
->namelen
;
532 memcpy(dep
->name
, args
->name
, dep
->namelen
);
533 tagp
= xfs_dir2_data_entry_tag_p(dep
);
534 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
536 * Need to scan fix up the bestfree table.
539 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
541 * Need to log the data block's header.
544 xfs_dir2_data_log_header(tp
, dbp
);
545 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
547 * If the bests table needs to be changed, do it.
548 * Log the change unless we've already done that.
550 if (be16_to_cpu(bestsp
[use_block
]) != be16_to_cpu(hdr
->bestfree
[0].length
)) {
551 bestsp
[use_block
] = hdr
->bestfree
[0].length
;
553 xfs_dir2_leaf_log_bests(tp
, lbp
, use_block
, use_block
);
556 lep
= xfs_dir2_leaf_find_entry(leaf
, index
, compact
, lowstale
,
557 highstale
, &lfloglow
, &lfloghigh
);
560 * Fill in the new leaf entry.
562 lep
->hashval
= cpu_to_be32(args
->hashval
);
563 lep
->address
= cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp
, use_block
,
564 be16_to_cpu(*tagp
)));
566 * Log the leaf fields and give up the buffers.
568 xfs_dir2_leaf_log_header(tp
, lbp
);
569 xfs_dir2_leaf_log_ents(tp
, lbp
, lfloglow
, lfloghigh
);
570 xfs_dir2_leaf_check(dp
, lbp
);
571 xfs_dir2_data_check(dp
, dbp
);
577 * Check the internal consistency of a leaf1 block.
578 * Pop an assert if something is wrong.
582 struct xfs_inode
*dp
, /* incore directory inode */
583 struct xfs_buf
*bp
) /* leaf's buffer */
585 int i
; /* leaf index */
586 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
587 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail pointer */
588 xfs_mount_t
*mp
; /* filesystem mount point */
589 int stale
; /* count of stale leaves */
593 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
));
595 * This value is not restrictive enough.
596 * Should factor in the size of the bests table as well.
597 * We can deduce a value for that from di_size.
599 ASSERT(be16_to_cpu(leaf
->hdr
.count
) <= xfs_dir2_max_leaf_ents(mp
));
600 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
602 * Leaves and bests don't overlap.
604 ASSERT((char *)&leaf
->ents
[be16_to_cpu(leaf
->hdr
.count
)] <=
605 (char *)xfs_dir2_leaf_bests_p(ltp
));
607 * Check hash value order, count stale entries.
609 for (i
= stale
= 0; i
< be16_to_cpu(leaf
->hdr
.count
); i
++) {
610 if (i
+ 1 < be16_to_cpu(leaf
->hdr
.count
))
611 ASSERT(be32_to_cpu(leaf
->ents
[i
].hashval
) <=
612 be32_to_cpu(leaf
->ents
[i
+ 1].hashval
));
613 if (leaf
->ents
[i
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
616 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) == stale
);
621 * Compact out any stale entries in the leaf.
622 * Log the header and changed leaf entries, if any.
625 xfs_dir2_leaf_compact(
626 xfs_da_args_t
*args
, /* operation arguments */
627 struct xfs_buf
*bp
) /* leaf buffer */
629 int from
; /* source leaf index */
630 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
631 int loglow
; /* first leaf entry to log */
632 int to
; /* target leaf index */
635 if (!leaf
->hdr
.stale
) {
639 * Compress out the stale entries in place.
641 for (from
= to
= 0, loglow
= -1; from
< be16_to_cpu(leaf
->hdr
.count
); from
++) {
642 if (leaf
->ents
[from
].address
==
643 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
646 * Only actually copy the entries that are different.
651 leaf
->ents
[to
] = leaf
->ents
[from
];
656 * Update and log the header, log the leaf entries.
658 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) == from
- to
);
659 be16_add_cpu(&leaf
->hdr
.count
, -(be16_to_cpu(leaf
->hdr
.stale
)));
661 xfs_dir2_leaf_log_header(args
->trans
, bp
);
663 xfs_dir2_leaf_log_ents(args
->trans
, bp
, loglow
, to
- 1);
667 * Compact the leaf entries, removing stale ones.
668 * Leave one stale entry behind - the one closest to our
669 * insertion index - and the caller will shift that one to our insertion
671 * Return new insertion index, where the remaining stale entry is,
672 * and leaf logging indices.
675 xfs_dir2_leaf_compact_x1(
676 struct xfs_buf
*bp
, /* leaf buffer */
677 int *indexp
, /* insertion index */
678 int *lowstalep
, /* out: stale entry before us */
679 int *highstalep
, /* out: stale entry after us */
680 int *lowlogp
, /* out: low log index */
681 int *highlogp
) /* out: high log index */
683 int from
; /* source copy index */
684 int highstale
; /* stale entry at/after index */
685 int index
; /* insertion index */
686 int keepstale
; /* source index of kept stale */
687 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
688 int lowstale
; /* stale entry before index */
689 int newindex
=0; /* new insertion index */
690 int to
; /* destination copy index */
693 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) > 1);
696 xfs_dir2_leaf_find_stale(leaf
, index
, &lowstale
, &highstale
);
699 * Pick the better of lowstale and highstale.
702 (highstale
== be16_to_cpu(leaf
->hdr
.count
) ||
703 index
- lowstale
<= highstale
- index
))
704 keepstale
= lowstale
;
706 keepstale
= highstale
;
708 * Copy the entries in place, removing all the stale entries
711 for (from
= to
= 0; from
< be16_to_cpu(leaf
->hdr
.count
); from
++) {
713 * Notice the new value of index.
717 if (from
!= keepstale
&&
718 leaf
->ents
[from
].address
==
719 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
)) {
725 * Record the new keepstale value for the insertion.
727 if (from
== keepstale
)
728 lowstale
= highstale
= to
;
730 * Copy only the entries that have moved.
733 leaf
->ents
[to
] = leaf
->ents
[from
];
738 * If the insertion point was past the last entry,
739 * set the new insertion point accordingly.
745 * Adjust the leaf header values.
747 be16_add_cpu(&leaf
->hdr
.count
, -(from
- to
));
748 leaf
->hdr
.stale
= cpu_to_be16(1);
750 * Remember the low/high stale value only in the "right"
753 if (lowstale
>= newindex
)
756 highstale
= be16_to_cpu(leaf
->hdr
.count
);
757 *highlogp
= be16_to_cpu(leaf
->hdr
.count
) - 1;
758 *lowstalep
= lowstale
;
759 *highstalep
= highstale
;
762 struct xfs_dir2_leaf_map_info
{
763 xfs_extlen_t map_blocks
; /* number of fsbs in map */
764 xfs_dablk_t map_off
; /* last mapped file offset */
765 int map_size
; /* total entries in *map */
766 int map_valid
; /* valid entries in *map */
767 int nmap
; /* mappings to ask xfs_bmapi */
768 xfs_dir2_db_t curdb
; /* db for current block */
769 int ra_current
; /* number of read-ahead blks */
770 int ra_index
; /* *map index for read-ahead */
771 int ra_offset
; /* map entry offset for ra */
772 int ra_want
; /* readahead count wanted */
773 struct xfs_bmbt_irec map
[]; /* map vector for blocks */
777 xfs_dir2_leaf_readbuf(
778 struct xfs_inode
*dp
,
780 struct xfs_dir2_leaf_map_info
*mip
,
781 xfs_dir2_off_t
*curoff
,
782 struct xfs_buf
**bpp
)
784 struct xfs_mount
*mp
= dp
->i_mount
;
785 struct xfs_buf
*bp
= *bpp
;
786 struct xfs_bmbt_irec
*map
= mip
->map
;
793 * If we have a buffer, we need to release it and
794 * take it out of the mapping.
798 xfs_trans_brelse(NULL
, bp
);
800 mip
->map_blocks
-= mp
->m_dirblkfsbs
;
802 * Loop to get rid of the extents for the
805 for (i
= mp
->m_dirblkfsbs
; i
> 0; ) {
806 j
= min_t(int, map
->br_blockcount
, i
);
807 map
->br_blockcount
-= j
;
808 map
->br_startblock
+= j
;
809 map
->br_startoff
+= j
;
811 * If mapping is done, pitch it from
814 if (!map
->br_blockcount
&& --mip
->map_valid
)
815 memmove(&map
[0], &map
[1],
816 sizeof(map
[0]) * mip
->map_valid
);
822 * Recalculate the readahead blocks wanted.
824 mip
->ra_want
= howmany(bufsize
+ mp
->m_dirblksize
,
825 mp
->m_sb
.sb_blocksize
) - 1;
826 ASSERT(mip
->ra_want
>= 0);
829 * If we don't have as many as we want, and we haven't
830 * run out of data blocks, get some more mappings.
832 if (1 + mip
->ra_want
> mip
->map_blocks
&&
833 mip
->map_off
< xfs_dir2_byte_to_da(mp
, XFS_DIR2_LEAF_OFFSET
)) {
835 * Get more bmaps, fill in after the ones
836 * we already have in the table.
838 mip
->nmap
= mip
->map_size
- mip
->map_valid
;
839 error
= xfs_bmapi_read(dp
, mip
->map_off
,
840 xfs_dir2_byte_to_da(mp
, XFS_DIR2_LEAF_OFFSET
) -
842 &map
[mip
->map_valid
], &mip
->nmap
, 0);
845 * Don't know if we should ignore this or try to return an
846 * error. The trouble with returning errors is that readdir
847 * will just stop without actually passing the error through.
853 * If we got all the mappings we asked for, set the final map
854 * offset based on the last bmap value received. Otherwise,
855 * we've reached the end.
857 if (mip
->nmap
== mip
->map_size
- mip
->map_valid
) {
858 i
= mip
->map_valid
+ mip
->nmap
- 1;
859 mip
->map_off
= map
[i
].br_startoff
+ map
[i
].br_blockcount
;
861 mip
->map_off
= xfs_dir2_byte_to_da(mp
,
862 XFS_DIR2_LEAF_OFFSET
);
865 * Look for holes in the mapping, and eliminate them. Count up
868 for (i
= mip
->map_valid
; i
< mip
->map_valid
+ mip
->nmap
; ) {
869 if (map
[i
].br_startblock
== HOLESTARTBLOCK
) {
871 length
= mip
->map_valid
+ mip
->nmap
- i
;
873 memmove(&map
[i
], &map
[i
+ 1],
874 sizeof(map
[i
]) * length
);
876 mip
->map_blocks
+= map
[i
].br_blockcount
;
880 mip
->map_valid
+= mip
->nmap
;
884 * No valid mappings, so no more data blocks.
886 if (!mip
->map_valid
) {
887 *curoff
= xfs_dir2_da_to_byte(mp
, mip
->map_off
);
892 * Read the directory block starting at the first mapping.
894 mip
->curdb
= xfs_dir2_da_to_db(mp
, map
->br_startoff
);
895 error
= xfs_da_read_buf(NULL
, dp
, map
->br_startoff
,
896 map
->br_blockcount
>= mp
->m_dirblkfsbs
?
897 XFS_FSB_TO_DADDR(mp
, map
->br_startblock
) : -1,
901 * Should just skip over the data block instead of giving up.
907 * Adjust the current amount of read-ahead: we just read a block that
911 mip
->ra_current
-= mp
->m_dirblkfsbs
;
914 * Do we need more readahead?
916 for (mip
->ra_index
= mip
->ra_offset
= i
= 0;
917 mip
->ra_want
> mip
->ra_current
&& i
< mip
->map_blocks
;
918 i
+= mp
->m_dirblkfsbs
) {
919 ASSERT(mip
->ra_index
< mip
->map_valid
);
921 * Read-ahead a contiguous directory block.
923 if (i
> mip
->ra_current
&&
924 map
[mip
->ra_index
].br_blockcount
>= mp
->m_dirblkfsbs
) {
925 xfs_buf_readahead(mp
->m_ddev_targp
,
927 map
[mip
->ra_index
].br_startblock
+
929 (int)BTOBB(mp
->m_dirblksize
));
934 * Read-ahead a non-contiguous directory block. This doesn't
935 * use our mapping, but this is a very rare case.
937 else if (i
> mip
->ra_current
) {
938 xfs_da_reada_buf(NULL
, dp
,
939 map
[mip
->ra_index
].br_startoff
+
946 * Advance offset through the mapping table.
948 for (j
= 0; j
< mp
->m_dirblkfsbs
; j
++) {
950 * The rest of this extent but not more than a dir
953 length
= min_t(int, mp
->m_dirblkfsbs
,
954 map
[mip
->ra_index
].br_blockcount
-
957 mip
->ra_offset
+= length
;
960 * Advance to the next mapping if this one is used up.
962 if (mip
->ra_offset
== map
[mip
->ra_index
].br_blockcount
) {
975 * Getdents (readdir) for leaf and node directories.
976 * This reads the data blocks only, so is the same for both forms.
979 xfs_dir2_leaf_getdents(
980 xfs_inode_t
*dp
, /* incore directory inode */
986 struct xfs_buf
*bp
= NULL
; /* data block buffer */
987 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
988 xfs_dir2_data_entry_t
*dep
; /* data entry */
989 xfs_dir2_data_unused_t
*dup
; /* unused entry */
990 int error
= 0; /* error return value */
991 int length
; /* temporary length value */
992 xfs_mount_t
*mp
; /* filesystem mount point */
993 int byteoff
; /* offset in current block */
994 xfs_dir2_off_t curoff
; /* current overall offset */
995 xfs_dir2_off_t newoff
; /* new curoff after new blk */
996 char *ptr
= NULL
; /* pointer to current data */
997 struct xfs_dir2_leaf_map_info
*map_info
;
1000 * If the offset is at or past the largest allowed value,
1001 * give up right away.
1003 if (*offset
>= XFS_DIR2_MAX_DATAPTR
)
1009 * Set up to bmap a number of blocks based on the caller's
1010 * buffer size, the directory block size, and the filesystem
1013 length
= howmany(bufsize
+ mp
->m_dirblksize
,
1014 mp
->m_sb
.sb_blocksize
);
1015 map_info
= kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info
, map
) +
1016 (length
* sizeof(struct xfs_bmbt_irec
)),
1018 map_info
->map_size
= length
;
1021 * Inside the loop we keep the main offset value as a byte offset
1022 * in the directory file.
1024 curoff
= xfs_dir2_dataptr_to_byte(mp
, *offset
);
1027 * Force this conversion through db so we truncate the offset
1028 * down to get the start of the data block.
1030 map_info
->map_off
= xfs_dir2_db_to_da(mp
,
1031 xfs_dir2_byte_to_db(mp
, curoff
));
1034 * Loop over directory entries until we reach the end offset.
1035 * Get more blocks and readahead as necessary.
1037 while (curoff
< XFS_DIR2_LEAF_OFFSET
) {
1039 * If we have no buffer, or we're off the end of the
1040 * current buffer, need to get another one.
1042 if (!bp
|| ptr
>= (char *)bp
->b_addr
+ mp
->m_dirblksize
) {
1044 error
= xfs_dir2_leaf_readbuf(dp
, bufsize
, map_info
,
1046 if (error
|| !map_info
->map_valid
)
1050 * Having done a read, we need to set a new offset.
1052 newoff
= xfs_dir2_db_off_to_byte(mp
, map_info
->curdb
, 0);
1054 * Start of the current block.
1056 if (curoff
< newoff
)
1059 * Make sure we're in the right block.
1061 else if (curoff
> newoff
)
1062 ASSERT(xfs_dir2_byte_to_db(mp
, curoff
) ==
1065 xfs_dir2_data_check(dp
, bp
);
1067 * Find our position in the block.
1069 ptr
= (char *)(hdr
+ 1);
1070 byteoff
= xfs_dir2_byte_to_off(mp
, curoff
);
1072 * Skip past the header.
1075 curoff
+= (uint
)sizeof(*hdr
);
1077 * Skip past entries until we reach our offset.
1080 while ((char *)ptr
- (char *)hdr
< byteoff
) {
1081 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1083 if (be16_to_cpu(dup
->freetag
)
1084 == XFS_DIR2_DATA_FREE_TAG
) {
1086 length
= be16_to_cpu(dup
->length
);
1090 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1092 xfs_dir2_data_entsize(dep
->namelen
);
1096 * Now set our real offset.
1099 xfs_dir2_db_off_to_byte(mp
,
1100 xfs_dir2_byte_to_db(mp
, curoff
),
1101 (char *)ptr
- (char *)hdr
);
1102 if (ptr
>= (char *)hdr
+ mp
->m_dirblksize
) {
1108 * We have a pointer to an entry.
1111 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1113 * No, it's unused, skip over it.
1115 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
1116 length
= be16_to_cpu(dup
->length
);
1122 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1123 length
= xfs_dir2_data_entsize(dep
->namelen
);
1125 if (filldir(dirent
, (char *)dep
->name
, dep
->namelen
,
1126 xfs_dir2_byte_to_dataptr(mp
, curoff
) & 0x7fffffff,
1127 be64_to_cpu(dep
->inumber
), DT_UNKNOWN
))
1131 * Advance to next entry in the block.
1135 /* bufsize may have just been a guess; don't go negative */
1136 bufsize
= bufsize
> length
? bufsize
- length
: 0;
1140 * All done. Set output offset value to current offset.
1142 if (curoff
> xfs_dir2_dataptr_to_byte(mp
, XFS_DIR2_MAX_DATAPTR
))
1143 *offset
= XFS_DIR2_MAX_DATAPTR
& 0x7fffffff;
1145 *offset
= xfs_dir2_byte_to_dataptr(mp
, curoff
) & 0x7fffffff;
1146 kmem_free(map_info
);
1148 xfs_trans_brelse(NULL
, bp
);
1153 * Initialize a new leaf block, leaf1 or leafn magic accepted.
1157 xfs_da_args_t
*args
, /* operation arguments */
1158 xfs_dir2_db_t bno
, /* directory block number */
1159 struct xfs_buf
**bpp
, /* out: leaf buffer */
1160 int magic
) /* magic number for block */
1162 struct xfs_buf
*bp
; /* leaf buffer */
1163 xfs_inode_t
*dp
; /* incore directory inode */
1164 int error
; /* error return code */
1165 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1166 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1167 xfs_mount_t
*mp
; /* filesystem mount point */
1168 xfs_trans_t
*tp
; /* transaction pointer */
1174 ASSERT(bno
>= XFS_DIR2_LEAF_FIRSTDB(mp
) &&
1175 bno
< XFS_DIR2_FREE_FIRSTDB(mp
));
1177 * Get the buffer for the block.
1179 error
= xfs_da_get_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, bno
), -1, &bp
,
1187 * Initialize the header.
1189 leaf
->hdr
.info
.magic
= cpu_to_be16(magic
);
1190 leaf
->hdr
.info
.forw
= 0;
1191 leaf
->hdr
.info
.back
= 0;
1192 leaf
->hdr
.count
= 0;
1193 leaf
->hdr
.stale
= 0;
1194 xfs_dir2_leaf_log_header(tp
, bp
);
1196 * If it's a leaf-format directory initialize the tail.
1197 * In this case our caller has the real bests table to copy into
1200 if (magic
== XFS_DIR2_LEAF1_MAGIC
) {
1201 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1203 xfs_dir2_leaf_log_tail(tp
, bp
);
1210 * Log the bests entries indicated from a leaf1 block.
1213 xfs_dir2_leaf_log_bests(
1214 xfs_trans_t
*tp
, /* transaction pointer */
1215 struct xfs_buf
*bp
, /* leaf buffer */
1216 int first
, /* first entry to log */
1217 int last
) /* last entry to log */
1219 __be16
*firstb
; /* pointer to first entry */
1220 __be16
*lastb
; /* pointer to last entry */
1221 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1222 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1225 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
));
1226 ltp
= xfs_dir2_leaf_tail_p(tp
->t_mountp
, leaf
);
1227 firstb
= xfs_dir2_leaf_bests_p(ltp
) + first
;
1228 lastb
= xfs_dir2_leaf_bests_p(ltp
) + last
;
1229 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)firstb
- (char *)leaf
),
1230 (uint
)((char *)lastb
- (char *)leaf
+ sizeof(*lastb
) - 1));
1234 * Log the leaf entries indicated from a leaf1 or leafn block.
1237 xfs_dir2_leaf_log_ents(
1238 xfs_trans_t
*tp
, /* transaction pointer */
1239 struct xfs_buf
*bp
, /* leaf buffer */
1240 int first
, /* first entry to log */
1241 int last
) /* last entry to log */
1243 xfs_dir2_leaf_entry_t
*firstlep
; /* pointer to first entry */
1244 xfs_dir2_leaf_entry_t
*lastlep
; /* pointer to last entry */
1245 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1248 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
1249 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
));
1250 firstlep
= &leaf
->ents
[first
];
1251 lastlep
= &leaf
->ents
[last
];
1252 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)firstlep
- (char *)leaf
),
1253 (uint
)((char *)lastlep
- (char *)leaf
+ sizeof(*lastlep
) - 1));
1257 * Log the header of the leaf1 or leafn block.
1260 xfs_dir2_leaf_log_header(
1261 struct xfs_trans
*tp
,
1264 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1267 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
) ||
1268 leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
));
1269 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)&leaf
->hdr
- (char *)leaf
),
1270 (uint
)(sizeof(leaf
->hdr
) - 1));
1274 * Log the tail of the leaf1 block.
1277 xfs_dir2_leaf_log_tail(
1278 struct xfs_trans
*tp
,
1281 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1282 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1283 xfs_mount_t
*mp
; /* filesystem mount point */
1287 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
));
1288 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1289 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)ltp
- (char *)leaf
),
1290 (uint
)(mp
->m_dirblksize
- 1));
1294 * Look up the entry referred to by args in the leaf format directory.
1295 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1296 * is also used by the node-format code.
1299 xfs_dir2_leaf_lookup(
1300 xfs_da_args_t
*args
) /* operation arguments */
1302 struct xfs_buf
*dbp
; /* data block buffer */
1303 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1304 xfs_inode_t
*dp
; /* incore directory inode */
1305 int error
; /* error return code */
1306 int index
; /* found entry index */
1307 struct xfs_buf
*lbp
; /* leaf buffer */
1308 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1309 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1310 xfs_trans_t
*tp
; /* transaction pointer */
1312 trace_xfs_dir2_leaf_lookup(args
);
1315 * Look up name in the leaf block, returning both buffers and index.
1317 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1322 xfs_dir2_leaf_check(dp
, lbp
);
1325 * Get to the leaf entry and contained data entry address.
1327 lep
= &leaf
->ents
[index
];
1329 * Point to the data entry.
1331 dep
= (xfs_dir2_data_entry_t
*)
1332 ((char *)dbp
->b_addr
+
1333 xfs_dir2_dataptr_to_off(dp
->i_mount
, be32_to_cpu(lep
->address
)));
1335 * Return the found inode number & CI name if appropriate
1337 args
->inumber
= be64_to_cpu(dep
->inumber
);
1338 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
1339 xfs_trans_brelse(tp
, dbp
);
1340 xfs_trans_brelse(tp
, lbp
);
1341 return XFS_ERROR(error
);
1345 * Look up name/hash in the leaf block.
1346 * Fill in indexp with the found index, and dbpp with the data buffer.
1347 * If not found dbpp will be NULL, and ENOENT comes back.
1348 * lbpp will always be filled in with the leaf buffer unless there's an error.
1350 static int /* error */
1351 xfs_dir2_leaf_lookup_int(
1352 xfs_da_args_t
*args
, /* operation arguments */
1353 struct xfs_buf
**lbpp
, /* out: leaf buffer */
1354 int *indexp
, /* out: index in leaf block */
1355 struct xfs_buf
**dbpp
) /* out: data buffer */
1357 xfs_dir2_db_t curdb
= -1; /* current data block number */
1358 struct xfs_buf
*dbp
= NULL
; /* data buffer */
1359 xfs_dir2_data_entry_t
*dep
; /* data entry */
1360 xfs_inode_t
*dp
; /* incore directory inode */
1361 int error
; /* error return code */
1362 int index
; /* index in leaf block */
1363 struct xfs_buf
*lbp
; /* leaf buffer */
1364 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1365 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1366 xfs_mount_t
*mp
; /* filesystem mount point */
1367 xfs_dir2_db_t newdb
; /* new data block number */
1368 xfs_trans_t
*tp
; /* transaction pointer */
1369 xfs_dir2_db_t cidb
= -1; /* case match data block no. */
1370 enum xfs_dacmp cmp
; /* name compare result */
1376 * Read the leaf block into the buffer.
1378 error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirleafblk
, -1, &lbp
,
1384 xfs_dir2_leaf_check(dp
, lbp
);
1386 * Look for the first leaf entry with our hash value.
1388 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
1390 * Loop over all the entries with the right hash value
1391 * looking to match the name.
1393 for (lep
= &leaf
->ents
[index
]; index
< be16_to_cpu(leaf
->hdr
.count
) &&
1394 be32_to_cpu(lep
->hashval
) == args
->hashval
;
1397 * Skip over stale leaf entries.
1399 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
1402 * Get the new data block number.
1404 newdb
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
1406 * If it's not the same as the old data block number,
1407 * need to pitch the old one and read the new one.
1409 if (newdb
!= curdb
) {
1411 xfs_trans_brelse(tp
, dbp
);
1412 error
= xfs_da_read_buf(tp
, dp
,
1413 xfs_dir2_db_to_da(mp
, newdb
),
1414 -1, &dbp
, XFS_DATA_FORK
);
1416 xfs_trans_brelse(tp
, lbp
);
1419 xfs_dir2_data_check(dp
, dbp
);
1423 * Point to the data entry.
1425 dep
= (xfs_dir2_data_entry_t
*)((char *)dbp
->b_addr
+
1426 xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
1428 * Compare name and if it's an exact match, return the index
1429 * and buffer. If it's the first case-insensitive match, store
1430 * the index and buffer and continue looking for an exact match.
1432 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
1433 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
1434 args
->cmpresult
= cmp
;
1436 /* case exact match: return the current buffer. */
1437 if (cmp
== XFS_CMP_EXACT
) {
1444 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
1446 * Here, we can only be doing a lookup (not a rename or remove).
1447 * If a case-insensitive match was found earlier, re-read the
1448 * appropriate data block if required and return it.
1450 if (args
->cmpresult
== XFS_CMP_CASE
) {
1452 if (cidb
!= curdb
) {
1453 xfs_trans_brelse(tp
, dbp
);
1454 error
= xfs_da_read_buf(tp
, dp
,
1455 xfs_dir2_db_to_da(mp
, cidb
),
1456 -1, &dbp
, XFS_DATA_FORK
);
1458 xfs_trans_brelse(tp
, lbp
);
1466 * No match found, return ENOENT.
1470 xfs_trans_brelse(tp
, dbp
);
1471 xfs_trans_brelse(tp
, lbp
);
1472 return XFS_ERROR(ENOENT
);
1476 * Remove an entry from a leaf format directory.
1479 xfs_dir2_leaf_removename(
1480 xfs_da_args_t
*args
) /* operation arguments */
1482 __be16
*bestsp
; /* leaf block best freespace */
1483 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
1484 xfs_dir2_db_t db
; /* data block number */
1485 struct xfs_buf
*dbp
; /* data block buffer */
1486 xfs_dir2_data_entry_t
*dep
; /* data entry structure */
1487 xfs_inode_t
*dp
; /* incore directory inode */
1488 int error
; /* error return code */
1489 xfs_dir2_db_t i
; /* temporary data block # */
1490 int index
; /* index into leaf entries */
1491 struct xfs_buf
*lbp
; /* leaf buffer */
1492 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1493 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1494 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1495 xfs_mount_t
*mp
; /* filesystem mount point */
1496 int needlog
; /* need to log data header */
1497 int needscan
; /* need to rescan data frees */
1498 xfs_dir2_data_off_t oldbest
; /* old value of best free */
1499 xfs_trans_t
*tp
; /* transaction pointer */
1501 trace_xfs_dir2_leaf_removename(args
);
1504 * Lookup the leaf entry, get the leaf and data blocks read in.
1506 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1514 xfs_dir2_data_check(dp
, dbp
);
1516 * Point to the leaf entry, use that to point to the data entry.
1518 lep
= &leaf
->ents
[index
];
1519 db
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
1520 dep
= (xfs_dir2_data_entry_t
*)
1521 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
1522 needscan
= needlog
= 0;
1523 oldbest
= be16_to_cpu(hdr
->bestfree
[0].length
);
1524 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1525 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1526 ASSERT(be16_to_cpu(bestsp
[db
]) == oldbest
);
1528 * Mark the former data entry unused.
1530 xfs_dir2_data_make_free(tp
, dbp
,
1531 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)hdr
),
1532 xfs_dir2_data_entsize(dep
->namelen
), &needlog
, &needscan
);
1534 * We just mark the leaf entry stale by putting a null in it.
1536 be16_add_cpu(&leaf
->hdr
.stale
, 1);
1537 xfs_dir2_leaf_log_header(tp
, lbp
);
1538 lep
->address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
1539 xfs_dir2_leaf_log_ents(tp
, lbp
, index
, index
);
1541 * Scan the freespace in the data block again if necessary,
1542 * log the data block header if necessary.
1545 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
1547 xfs_dir2_data_log_header(tp
, dbp
);
1549 * If the longest freespace in the data block has changed,
1550 * put the new value in the bests table and log that.
1552 if (be16_to_cpu(hdr
->bestfree
[0].length
) != oldbest
) {
1553 bestsp
[db
] = hdr
->bestfree
[0].length
;
1554 xfs_dir2_leaf_log_bests(tp
, lbp
, db
, db
);
1556 xfs_dir2_data_check(dp
, dbp
);
1558 * If the data block is now empty then get rid of the data block.
1560 if (be16_to_cpu(hdr
->bestfree
[0].length
) ==
1561 mp
->m_dirblksize
- (uint
)sizeof(*hdr
)) {
1562 ASSERT(db
!= mp
->m_dirdatablk
);
1563 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1565 * Nope, can't get rid of it because it caused
1566 * allocation of a bmap btree block to do so.
1567 * Just go on, returning success, leaving the
1568 * empty block in place.
1570 if (error
== ENOSPC
&& args
->total
== 0)
1572 xfs_dir2_leaf_check(dp
, lbp
);
1577 * If this is the last data block then compact the
1578 * bests table by getting rid of entries.
1580 if (db
== be32_to_cpu(ltp
->bestcount
) - 1) {
1582 * Look for the last active entry (i).
1584 for (i
= db
- 1; i
> 0; i
--) {
1585 if (bestsp
[i
] != cpu_to_be16(NULLDATAOFF
))
1589 * Copy the table down so inactive entries at the
1592 memmove(&bestsp
[db
- i
], bestsp
,
1593 (be32_to_cpu(ltp
->bestcount
) - (db
- i
)) * sizeof(*bestsp
));
1594 be32_add_cpu(<p
->bestcount
, -(db
- i
));
1595 xfs_dir2_leaf_log_tail(tp
, lbp
);
1596 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1598 bestsp
[db
] = cpu_to_be16(NULLDATAOFF
);
1601 * If the data block was not the first one, drop it.
1603 else if (db
!= mp
->m_dirdatablk
)
1606 xfs_dir2_leaf_check(dp
, lbp
);
1608 * See if we can convert to block form.
1610 return xfs_dir2_leaf_to_block(args
, lbp
, dbp
);
1614 * Replace the inode number in a leaf format directory entry.
1617 xfs_dir2_leaf_replace(
1618 xfs_da_args_t
*args
) /* operation arguments */
1620 struct xfs_buf
*dbp
; /* data block buffer */
1621 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1622 xfs_inode_t
*dp
; /* incore directory inode */
1623 int error
; /* error return code */
1624 int index
; /* index of leaf entry */
1625 struct xfs_buf
*lbp
; /* leaf buffer */
1626 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1627 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1628 xfs_trans_t
*tp
; /* transaction pointer */
1630 trace_xfs_dir2_leaf_replace(args
);
1633 * Look up the entry.
1635 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1641 * Point to the leaf entry, get data address from it.
1643 lep
= &leaf
->ents
[index
];
1645 * Point to the data entry.
1647 dep
= (xfs_dir2_data_entry_t
*)
1648 ((char *)dbp
->b_addr
+
1649 xfs_dir2_dataptr_to_off(dp
->i_mount
, be32_to_cpu(lep
->address
)));
1650 ASSERT(args
->inumber
!= be64_to_cpu(dep
->inumber
));
1652 * Put the new inode number in, log it.
1654 dep
->inumber
= cpu_to_be64(args
->inumber
);
1656 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
1657 xfs_dir2_leaf_check(dp
, lbp
);
1658 xfs_trans_brelse(tp
, lbp
);
1663 * Return index in the leaf block (lbp) which is either the first
1664 * one with this hash value, or if there are none, the insert point
1665 * for that hash value.
1667 int /* index value */
1668 xfs_dir2_leaf_search_hash(
1669 xfs_da_args_t
*args
, /* operation arguments */
1670 struct xfs_buf
*lbp
) /* leaf buffer */
1672 xfs_dahash_t hash
=0; /* hash from this entry */
1673 xfs_dahash_t hashwant
; /* hash value looking for */
1674 int high
; /* high leaf index */
1675 int low
; /* low leaf index */
1676 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1677 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1678 int mid
=0; /* current leaf index */
1682 if (!leaf
->hdr
.count
)
1686 * Note, the table cannot be empty, so we have to go through the loop.
1687 * Binary search the leaf entries looking for our hash value.
1689 for (lep
= leaf
->ents
, low
= 0, high
= be16_to_cpu(leaf
->hdr
.count
) - 1,
1690 hashwant
= args
->hashval
;
1692 mid
= (low
+ high
) >> 1;
1693 if ((hash
= be32_to_cpu(lep
[mid
].hashval
)) == hashwant
)
1695 if (hash
< hashwant
)
1701 * Found one, back up through all the equal hash values.
1703 if (hash
== hashwant
) {
1704 while (mid
> 0 && be32_to_cpu(lep
[mid
- 1].hashval
) == hashwant
) {
1709 * Need to point to an entry higher than ours.
1711 else if (hash
< hashwant
)
1717 * Trim off a trailing data block. We know it's empty since the leaf
1718 * freespace table says so.
1721 xfs_dir2_leaf_trim_data(
1722 xfs_da_args_t
*args
, /* operation arguments */
1723 struct xfs_buf
*lbp
, /* leaf buffer */
1724 xfs_dir2_db_t db
) /* data block number */
1726 __be16
*bestsp
; /* leaf bests table */
1727 struct xfs_buf
*dbp
; /* data block buffer */
1728 xfs_inode_t
*dp
; /* incore directory inode */
1729 int error
; /* error return value */
1730 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1731 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1732 xfs_mount_t
*mp
; /* filesystem mount point */
1733 xfs_trans_t
*tp
; /* transaction pointer */
1739 * Read the offending data block. We need its buffer.
1741 if ((error
= xfs_da_read_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, db
), -1, &dbp
,
1747 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1751 struct xfs_dir2_data_hdr
*hdr
= dbp
->b_addr
;
1753 ASSERT(hdr
->magic
== cpu_to_be32(XFS_DIR2_DATA_MAGIC
));
1754 ASSERT(be16_to_cpu(hdr
->bestfree
[0].length
) ==
1755 mp
->m_dirblksize
- (uint
)sizeof(*hdr
));
1756 ASSERT(db
== be32_to_cpu(ltp
->bestcount
) - 1);
1761 * Get rid of the data block.
1763 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1764 ASSERT(error
!= ENOSPC
);
1765 xfs_trans_brelse(tp
, dbp
);
1769 * Eliminate the last bests entry from the table.
1771 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1772 be32_add_cpu(<p
->bestcount
, -1);
1773 memmove(&bestsp
[1], &bestsp
[0], be32_to_cpu(ltp
->bestcount
) * sizeof(*bestsp
));
1774 xfs_dir2_leaf_log_tail(tp
, lbp
);
1775 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1779 static inline size_t
1781 struct xfs_dir2_leaf_hdr
*hdr
,
1786 entries
= be16_to_cpu(hdr
->count
) - be16_to_cpu(hdr
->stale
);
1787 return sizeof(xfs_dir2_leaf_hdr_t
) +
1788 entries
* sizeof(xfs_dir2_leaf_entry_t
) +
1789 counts
* sizeof(xfs_dir2_data_off_t
) +
1790 sizeof(xfs_dir2_leaf_tail_t
);
1794 * Convert node form directory to leaf form directory.
1795 * The root of the node form dir needs to already be a LEAFN block.
1796 * Just return if we can't do anything.
1799 xfs_dir2_node_to_leaf(
1800 xfs_da_state_t
*state
) /* directory operation state */
1802 xfs_da_args_t
*args
; /* operation arguments */
1803 xfs_inode_t
*dp
; /* incore directory inode */
1804 int error
; /* error return code */
1805 struct xfs_buf
*fbp
; /* buffer for freespace block */
1806 xfs_fileoff_t fo
; /* freespace file offset */
1807 xfs_dir2_free_t
*free
; /* freespace structure */
1808 struct xfs_buf
*lbp
; /* buffer for leaf block */
1809 xfs_dir2_leaf_tail_t
*ltp
; /* tail of leaf structure */
1810 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1811 xfs_mount_t
*mp
; /* filesystem mount point */
1812 int rval
; /* successful free trim? */
1813 xfs_trans_t
*tp
; /* transaction pointer */
1816 * There's more than a leaf level in the btree, so there must
1817 * be multiple leafn blocks. Give up.
1819 if (state
->path
.active
> 1)
1823 trace_xfs_dir2_node_to_leaf(args
);
1829 * Get the last offset in the file.
1831 if ((error
= xfs_bmap_last_offset(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1834 fo
-= mp
->m_dirblkfsbs
;
1836 * If there are freespace blocks other than the first one,
1837 * take this opportunity to remove trailing empty freespace blocks
1838 * that may have been left behind during no-space-reservation
1841 while (fo
> mp
->m_dirfreeblk
) {
1842 if ((error
= xfs_dir2_node_trim_free(args
, fo
, &rval
))) {
1846 fo
-= mp
->m_dirblkfsbs
;
1851 * Now find the block just before the freespace block.
1853 if ((error
= xfs_bmap_last_before(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1857 * If it's not the single leaf block, give up.
1859 if (XFS_FSB_TO_B(mp
, fo
) > XFS_DIR2_LEAF_OFFSET
+ mp
->m_dirblksize
)
1861 lbp
= state
->path
.blk
[0].bp
;
1863 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
));
1865 * Read the freespace block.
1867 if ((error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirfreeblk
, -1, &fbp
,
1872 ASSERT(free
->hdr
.magic
== cpu_to_be32(XFS_DIR2_FREE_MAGIC
));
1873 ASSERT(!free
->hdr
.firstdb
);
1876 * Now see if the leafn and free data will fit in a leaf1.
1877 * If not, release the buffer and give up.
1879 if (xfs_dir2_leaf_size(&leaf
->hdr
, be32_to_cpu(free
->hdr
.nvalid
)) >
1881 xfs_trans_brelse(tp
, fbp
);
1886 * If the leaf has any stale entries in it, compress them out.
1887 * The compact routine will log the header.
1889 if (be16_to_cpu(leaf
->hdr
.stale
))
1890 xfs_dir2_leaf_compact(args
, lbp
);
1892 xfs_dir2_leaf_log_header(tp
, lbp
);
1893 leaf
->hdr
.info
.magic
= cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
);
1895 * Set up the leaf tail from the freespace block.
1897 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1898 ltp
->bestcount
= free
->hdr
.nvalid
;
1900 * Set up the leaf bests table.
1902 memcpy(xfs_dir2_leaf_bests_p(ltp
), free
->bests
,
1903 be32_to_cpu(ltp
->bestcount
) * sizeof(xfs_dir2_data_off_t
));
1904 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1905 xfs_dir2_leaf_log_tail(tp
, lbp
);
1906 xfs_dir2_leaf_check(dp
, lbp
);
1908 * Get rid of the freespace block.
1910 error
= xfs_dir2_shrink_inode(args
, XFS_DIR2_FREE_FIRSTDB(mp
), fbp
);
1913 * This can't fail here because it can only happen when
1914 * punching out the middle of an extent, and this is an
1917 ASSERT(error
!= ENOSPC
);
1922 * Now see if we can convert the single-leaf directory
1923 * down to a block form directory.
1924 * This routine always kills the dabuf for the leaf, so
1925 * eliminate it from the path.
1927 error
= xfs_dir2_leaf_to_block(args
, lbp
, NULL
);
1928 state
->path
.blk
[0].bp
= NULL
;