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_dir2_sf.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
35 #include "xfs_dir2_data.h"
36 #include "xfs_dir2_leaf.h"
37 #include "xfs_dir2_block.h"
38 #include "xfs_dir2_node.h"
39 #include "xfs_error.h"
40 #include "xfs_trace.h"
43 * Local function declarations.
46 static void xfs_dir2_leaf_check(xfs_inode_t
*dp
, xfs_dabuf_t
*bp
);
48 #define xfs_dir2_leaf_check(dp, bp)
50 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t
*args
, xfs_dabuf_t
**lbpp
,
51 int *indexp
, xfs_dabuf_t
**dbpp
);
52 static void xfs_dir2_leaf_log_bests(struct xfs_trans
*tp
, struct xfs_dabuf
*bp
,
54 static void xfs_dir2_leaf_log_tail(struct xfs_trans
*tp
, struct xfs_dabuf
*bp
);
58 * Convert a block form directory to a leaf form directory.
61 xfs_dir2_block_to_leaf(
62 xfs_da_args_t
*args
, /* operation arguments */
63 xfs_dabuf_t
*dbp
) /* input block's buffer */
65 __be16
*bestsp
; /* leaf's bestsp entries */
66 xfs_dablk_t blkno
; /* leaf block's bno */
67 xfs_dir2_data_hdr_t
*hdr
; /* block header */
68 xfs_dir2_leaf_entry_t
*blp
; /* block's leaf entries */
69 xfs_dir2_block_tail_t
*btp
; /* block's tail */
70 xfs_inode_t
*dp
; /* incore directory inode */
71 int error
; /* error return code */
72 xfs_dabuf_t
*lbp
; /* leaf block's buffer */
73 xfs_dir2_db_t ldb
; /* leaf block's bno */
74 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
75 xfs_dir2_leaf_tail_t
*ltp
; /* leaf's tail */
76 xfs_mount_t
*mp
; /* filesystem mount point */
77 int needlog
; /* need to log block header */
78 int needscan
; /* need to rescan bestfree */
79 xfs_trans_t
*tp
; /* transaction pointer */
81 trace_xfs_dir2_block_to_leaf(args
);
87 * Add the leaf block to the inode.
88 * This interface will only put blocks in the leaf/node range.
89 * Since that's empty now, we'll get the root (block 0 in range).
91 if ((error
= xfs_da_grow_inode(args
, &blkno
))) {
94 ldb
= xfs_dir2_da_to_db(mp
, blkno
);
95 ASSERT(ldb
== XFS_DIR2_LEAF_FIRSTDB(mp
));
97 * Initialize the leaf block, get a buffer for it.
99 if ((error
= xfs_dir2_leaf_init(args
, ldb
, &lbp
, XFS_DIR2_LEAF1_MAGIC
))) {
105 xfs_dir2_data_check(dp
, dbp
);
106 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
107 blp
= xfs_dir2_block_leaf_p(btp
);
109 * Set the counts in the leaf header.
111 leaf
->hdr
.count
= cpu_to_be16(be32_to_cpu(btp
->count
));
112 leaf
->hdr
.stale
= cpu_to_be16(be32_to_cpu(btp
->stale
));
114 * Could compact these but I think we always do the conversion
115 * after squeezing out stale entries.
117 memcpy(leaf
->ents
, blp
, be32_to_cpu(btp
->count
) * sizeof(xfs_dir2_leaf_entry_t
));
118 xfs_dir2_leaf_log_ents(tp
, lbp
, 0, be16_to_cpu(leaf
->hdr
.count
) - 1);
122 * Make the space formerly occupied by the leaf entries and block
125 xfs_dir2_data_make_free(tp
, dbp
,
126 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)hdr
),
127 (xfs_dir2_data_aoff_t
)((char *)hdr
+ mp
->m_dirblksize
-
129 &needlog
, &needscan
);
131 * Fix up the block header, make it a data block.
133 hdr
->magic
= cpu_to_be32(XFS_DIR2_DATA_MAGIC
);
135 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
137 * Set up leaf tail and bests table.
139 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
140 ltp
->bestcount
= cpu_to_be32(1);
141 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
142 bestsp
[0] = hdr
->bestfree
[0].length
;
144 * Log the data header and leaf bests table.
147 xfs_dir2_data_log_header(tp
, dbp
);
148 xfs_dir2_leaf_check(dp
, lbp
);
149 xfs_dir2_data_check(dp
, dbp
);
150 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, 0);
151 xfs_da_buf_done(lbp
);
155 struct xfs_dir2_leaf_entry
*
156 xfs_dir2_leaf_find_entry(
157 xfs_dir2_leaf_t
*leaf
, /* leaf structure */
158 int index
, /* leaf table position */
159 int compact
, /* need to compact leaves */
160 int lowstale
, /* index of prev stale leaf */
161 int highstale
, /* index of next stale leaf */
162 int *lfloglow
, /* low leaf logging index */
163 int *lfloghigh
) /* high leaf logging index */
165 if (!leaf
->hdr
.stale
) {
166 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry table pointer */
169 * Now we need to make room to insert the leaf entry.
171 * If there are no stale entries, just insert a hole at index.
173 lep
= &leaf
->ents
[index
];
174 if (index
< be16_to_cpu(leaf
->hdr
.count
))
175 memmove(lep
+ 1, lep
,
176 (be16_to_cpu(leaf
->hdr
.count
) - index
) *
180 * Record low and high logging indices for the leaf.
183 *lfloghigh
= be16_to_cpu(leaf
->hdr
.count
);
184 be16_add_cpu(&leaf
->hdr
.count
, 1);
189 * There are stale entries.
191 * We will use one of them for the new entry. It's probably not at
192 * the right location, so we'll have to shift some up or down first.
194 * If we didn't compact before, we need to find the nearest stale
195 * entries before and after our insertion point.
199 * Find the first stale entry before the insertion point,
202 for (lowstale
= index
- 1;
204 be32_to_cpu(leaf
->ents
[lowstale
].address
) !=
205 XFS_DIR2_NULL_DATAPTR
;
210 * Find the next stale entry at or after the insertion point,
211 * if any. Stop if we go so far that the lowstale entry
214 for (highstale
= index
;
215 highstale
< be16_to_cpu(leaf
->hdr
.count
) &&
216 be32_to_cpu(leaf
->ents
[highstale
].address
) !=
217 XFS_DIR2_NULL_DATAPTR
&&
219 index
- lowstale
- 1 >= highstale
- index
);
225 * If the low one is better, use it.
228 (highstale
== be16_to_cpu(leaf
->hdr
.count
) ||
229 index
- lowstale
- 1 < highstale
- index
)) {
230 ASSERT(index
- lowstale
- 1 >= 0);
231 ASSERT(be32_to_cpu(leaf
->ents
[lowstale
].address
) ==
232 XFS_DIR2_NULL_DATAPTR
);
235 * Copy entries up to cover the stale entry and make room
238 if (index
- lowstale
- 1 > 0) {
239 memmove(&leaf
->ents
[lowstale
],
240 &leaf
->ents
[lowstale
+ 1],
241 (index
- lowstale
- 1) *
242 sizeof(xfs_dir2_leaf_entry_t
));
244 *lfloglow
= MIN(lowstale
, *lfloglow
);
245 *lfloghigh
= MAX(index
- 1, *lfloghigh
);
246 be16_add_cpu(&leaf
->hdr
.stale
, -1);
247 return &leaf
->ents
[index
- 1];
251 * The high one is better, so use that one.
253 ASSERT(highstale
- index
>= 0);
254 ASSERT(be32_to_cpu(leaf
->ents
[highstale
].address
) ==
255 XFS_DIR2_NULL_DATAPTR
);
258 * Copy entries down to cover the stale entry and make room for the
261 if (highstale
- index
> 0) {
262 memmove(&leaf
->ents
[index
+ 1],
264 (highstale
- index
) * sizeof(xfs_dir2_leaf_entry_t
));
266 *lfloglow
= MIN(index
, *lfloglow
);
267 *lfloghigh
= MAX(highstale
, *lfloghigh
);
268 be16_add_cpu(&leaf
->hdr
.stale
, -1);
269 return &leaf
->ents
[index
];
273 * Add an entry to a leaf form directory.
276 xfs_dir2_leaf_addname(
277 xfs_da_args_t
*args
) /* operation arguments */
279 __be16
*bestsp
; /* freespace table in leaf */
280 int compact
; /* need to compact leaves */
281 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
282 xfs_dabuf_t
*dbp
; /* data block buffer */
283 xfs_dir2_data_entry_t
*dep
; /* data block entry */
284 xfs_inode_t
*dp
; /* incore directory inode */
285 xfs_dir2_data_unused_t
*dup
; /* data unused entry */
286 int error
; /* error return value */
287 int grown
; /* allocated new data block */
288 int highstale
; /* index of next stale leaf */
289 int i
; /* temporary, index */
290 int index
; /* leaf table position */
291 xfs_dabuf_t
*lbp
; /* leaf's buffer */
292 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
293 int length
; /* length of new entry */
294 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry table pointer */
295 int lfloglow
; /* low leaf logging index */
296 int lfloghigh
; /* high leaf logging index */
297 int lowstale
; /* index of prev stale leaf */
298 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail pointer */
299 xfs_mount_t
*mp
; /* filesystem mount point */
300 int needbytes
; /* leaf block bytes needed */
301 int needlog
; /* need to log data header */
302 int needscan
; /* need to rescan data free */
303 __be16
*tagp
; /* end of data entry */
304 xfs_trans_t
*tp
; /* transaction pointer */
305 xfs_dir2_db_t use_block
; /* data block number */
307 trace_xfs_dir2_leaf_addname(args
);
313 * Read the leaf block.
315 error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirleafblk
, -1, &lbp
,
322 * Look up the entry by hash value and name.
323 * We know it's not there, our caller has already done a lookup.
324 * So the index is of the entry to insert in front of.
325 * But if there are dup hash values the index is of the first of those.
327 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
329 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
330 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
331 length
= xfs_dir2_data_entsize(args
->namelen
);
333 * See if there are any entries with the same hash value
334 * and space in their block for the new entry.
335 * This is good because it puts multiple same-hash value entries
336 * in a data block, improving the lookup of those entries.
338 for (use_block
= -1, lep
= &leaf
->ents
[index
];
339 index
< be16_to_cpu(leaf
->hdr
.count
) && be32_to_cpu(lep
->hashval
) == args
->hashval
;
341 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
343 i
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
344 ASSERT(i
< be32_to_cpu(ltp
->bestcount
));
345 ASSERT(be16_to_cpu(bestsp
[i
]) != NULLDATAOFF
);
346 if (be16_to_cpu(bestsp
[i
]) >= length
) {
352 * Didn't find a block yet, linear search all the data blocks.
354 if (use_block
== -1) {
355 for (i
= 0; i
< be32_to_cpu(ltp
->bestcount
); i
++) {
357 * Remember a block we see that's missing.
359 if (be16_to_cpu(bestsp
[i
]) == NULLDATAOFF
&& use_block
== -1)
361 else if (be16_to_cpu(bestsp
[i
]) >= length
) {
368 * How many bytes do we need in the leaf block?
371 if (!leaf
->hdr
.stale
)
372 needbytes
+= sizeof(xfs_dir2_leaf_entry_t
);
374 needbytes
+= sizeof(xfs_dir2_data_off_t
);
377 * Now kill use_block if it refers to a missing block, so we
378 * can use it as an indication of allocation needed.
380 if (use_block
!= -1 && be16_to_cpu(bestsp
[use_block
]) == NULLDATAOFF
)
383 * If we don't have enough free bytes but we can make enough
384 * by compacting out stale entries, we'll do that.
386 if ((char *)bestsp
- (char *)&leaf
->ents
[be16_to_cpu(leaf
->hdr
.count
)] <
387 needbytes
&& be16_to_cpu(leaf
->hdr
.stale
) > 1) {
391 * Otherwise if we don't have enough free bytes we need to
392 * convert to node form.
394 else if ((char *)bestsp
- (char *)&leaf
->ents
[be16_to_cpu(
395 leaf
->hdr
.count
)] < needbytes
) {
397 * Just checking or no space reservation, give up.
399 if ((args
->op_flags
& XFS_DA_OP_JUSTCHECK
) ||
401 xfs_da_brelse(tp
, lbp
);
402 return XFS_ERROR(ENOSPC
);
405 * Convert to node form.
407 error
= xfs_dir2_leaf_to_node(args
, lbp
);
408 xfs_da_buf_done(lbp
);
412 * Then add the new entry.
414 return xfs_dir2_node_addname(args
);
417 * Otherwise it will fit without compaction.
422 * If just checking, then it will fit unless we needed to allocate
425 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
426 xfs_da_brelse(tp
, lbp
);
427 return use_block
== -1 ? XFS_ERROR(ENOSPC
) : 0;
430 * If no allocations are allowed, return now before we've
433 if (args
->total
== 0 && use_block
== -1) {
434 xfs_da_brelse(tp
, lbp
);
435 return XFS_ERROR(ENOSPC
);
438 * Need to compact the leaf entries, removing stale ones.
439 * Leave one stale entry behind - the one closest to our
440 * insertion index - and we'll shift that one to our insertion
444 xfs_dir2_leaf_compact_x1(lbp
, &index
, &lowstale
, &highstale
,
445 &lfloglow
, &lfloghigh
);
448 * There are stale entries, so we'll need log-low and log-high
449 * impossibly bad values later.
451 else if (be16_to_cpu(leaf
->hdr
.stale
)) {
452 lfloglow
= be16_to_cpu(leaf
->hdr
.count
);
456 * If there was no data block space found, we need to allocate
459 if (use_block
== -1) {
461 * Add the new data block.
463 if ((error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
,
465 xfs_da_brelse(tp
, lbp
);
469 * Initialize the block.
471 if ((error
= xfs_dir2_data_init(args
, use_block
, &dbp
))) {
472 xfs_da_brelse(tp
, lbp
);
476 * If we're adding a new data block on the end we need to
477 * extend the bests table. Copy it up one entry.
479 if (use_block
>= be32_to_cpu(ltp
->bestcount
)) {
481 memmove(&bestsp
[0], &bestsp
[1],
482 be32_to_cpu(ltp
->bestcount
) * sizeof(bestsp
[0]));
483 be32_add_cpu(<p
->bestcount
, 1);
484 xfs_dir2_leaf_log_tail(tp
, lbp
);
485 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
488 * If we're filling in a previously empty block just log it.
491 xfs_dir2_leaf_log_bests(tp
, lbp
, use_block
, use_block
);
493 bestsp
[use_block
] = hdr
->bestfree
[0].length
;
497 * Already had space in some data block.
498 * Just read that one in.
502 xfs_da_read_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, use_block
),
503 -1, &dbp
, XFS_DATA_FORK
))) {
504 xfs_da_brelse(tp
, lbp
);
510 xfs_dir2_data_check(dp
, dbp
);
512 * Point to the biggest freespace in our data block.
514 dup
= (xfs_dir2_data_unused_t
*)
515 ((char *)hdr
+ be16_to_cpu(hdr
->bestfree
[0].offset
));
516 ASSERT(be16_to_cpu(dup
->length
) >= length
);
517 needscan
= needlog
= 0;
519 * Mark the initial part of our freespace in use for the new entry.
521 xfs_dir2_data_use_free(tp
, dbp
, dup
,
522 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
), length
,
523 &needlog
, &needscan
);
525 * Initialize our new entry (at last).
527 dep
= (xfs_dir2_data_entry_t
*)dup
;
528 dep
->inumber
= cpu_to_be64(args
->inumber
);
529 dep
->namelen
= args
->namelen
;
530 memcpy(dep
->name
, args
->name
, dep
->namelen
);
531 tagp
= xfs_dir2_data_entry_tag_p(dep
);
532 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
534 * Need to scan fix up the bestfree table.
537 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
539 * Need to log the data block's header.
542 xfs_dir2_data_log_header(tp
, dbp
);
543 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
545 * If the bests table needs to be changed, do it.
546 * Log the change unless we've already done that.
548 if (be16_to_cpu(bestsp
[use_block
]) != be16_to_cpu(hdr
->bestfree
[0].length
)) {
549 bestsp
[use_block
] = hdr
->bestfree
[0].length
;
551 xfs_dir2_leaf_log_bests(tp
, lbp
, use_block
, use_block
);
554 lep
= xfs_dir2_leaf_find_entry(leaf
, index
, compact
, lowstale
,
555 highstale
, &lfloglow
, &lfloghigh
);
558 * Fill in the new leaf entry.
560 lep
->hashval
= cpu_to_be32(args
->hashval
);
561 lep
->address
= cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp
, use_block
,
562 be16_to_cpu(*tagp
)));
564 * Log the leaf fields and give up the buffers.
566 xfs_dir2_leaf_log_header(tp
, lbp
);
567 xfs_dir2_leaf_log_ents(tp
, lbp
, lfloglow
, lfloghigh
);
568 xfs_dir2_leaf_check(dp
, lbp
);
569 xfs_da_buf_done(lbp
);
570 xfs_dir2_data_check(dp
, dbp
);
571 xfs_da_buf_done(dbp
);
577 * Check the internal consistency of a leaf1 block.
578 * Pop an assert if something is wrong.
582 xfs_inode_t
*dp
, /* incore directory inode */
583 xfs_dabuf_t
*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(be16_to_cpu(leaf
->hdr
.info
.magic
) == 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 (be32_to_cpu(leaf
->ents
[i
].address
) == 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 xfs_dabuf_t
*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 (be32_to_cpu(leaf
->ents
[from
].address
) == XFS_DIR2_NULL_DATAPTR
)
645 * Only actually copy the entries that are different.
650 leaf
->ents
[to
] = leaf
->ents
[from
];
655 * Update and log the header, log the leaf entries.
657 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) == from
- to
);
658 be16_add_cpu(&leaf
->hdr
.count
, -(be16_to_cpu(leaf
->hdr
.stale
)));
660 xfs_dir2_leaf_log_header(args
->trans
, bp
);
662 xfs_dir2_leaf_log_ents(args
->trans
, bp
, loglow
, to
- 1);
666 * Compact the leaf entries, removing stale ones.
667 * Leave one stale entry behind - the one closest to our
668 * insertion index - and the caller will shift that one to our insertion
670 * Return new insertion index, where the remaining stale entry is,
671 * and leaf logging indices.
674 xfs_dir2_leaf_compact_x1(
675 xfs_dabuf_t
*bp
, /* leaf buffer */
676 int *indexp
, /* insertion index */
677 int *lowstalep
, /* out: stale entry before us */
678 int *highstalep
, /* out: stale entry after us */
679 int *lowlogp
, /* out: low log index */
680 int *highlogp
) /* out: high log index */
682 int from
; /* source copy index */
683 int highstale
; /* stale entry at/after index */
684 int index
; /* insertion index */
685 int keepstale
; /* source index of kept stale */
686 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
687 int lowstale
; /* stale entry before index */
688 int newindex
=0; /* new insertion index */
689 int to
; /* destination copy index */
692 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) > 1);
695 * Find the first stale entry before our index, if any.
697 for (lowstale
= index
- 1;
699 be32_to_cpu(leaf
->ents
[lowstale
].address
) != XFS_DIR2_NULL_DATAPTR
;
703 * Find the first stale entry at or after our index, if any.
704 * Stop if the answer would be worse than lowstale.
706 for (highstale
= index
;
707 highstale
< be16_to_cpu(leaf
->hdr
.count
) &&
708 be32_to_cpu(leaf
->ents
[highstale
].address
) != XFS_DIR2_NULL_DATAPTR
&&
709 (lowstale
< 0 || index
- lowstale
> highstale
- index
);
713 * Pick the better of lowstale and highstale.
716 (highstale
== be16_to_cpu(leaf
->hdr
.count
) ||
717 index
- lowstale
<= highstale
- index
))
718 keepstale
= lowstale
;
720 keepstale
= highstale
;
722 * Copy the entries in place, removing all the stale entries
725 for (from
= to
= 0; from
< be16_to_cpu(leaf
->hdr
.count
); from
++) {
727 * Notice the new value of index.
731 if (from
!= keepstale
&&
732 be32_to_cpu(leaf
->ents
[from
].address
) == XFS_DIR2_NULL_DATAPTR
) {
738 * Record the new keepstale value for the insertion.
740 if (from
== keepstale
)
741 lowstale
= highstale
= to
;
743 * Copy only the entries that have moved.
746 leaf
->ents
[to
] = leaf
->ents
[from
];
751 * If the insertion point was past the last entry,
752 * set the new insertion point accordingly.
758 * Adjust the leaf header values.
760 be16_add_cpu(&leaf
->hdr
.count
, -(from
- to
));
761 leaf
->hdr
.stale
= cpu_to_be16(1);
763 * Remember the low/high stale value only in the "right"
766 if (lowstale
>= newindex
)
769 highstale
= be16_to_cpu(leaf
->hdr
.count
);
770 *highlogp
= be16_to_cpu(leaf
->hdr
.count
) - 1;
771 *lowstalep
= lowstale
;
772 *highstalep
= highstale
;
776 * Getdents (readdir) for leaf and node directories.
777 * This reads the data blocks only, so is the same for both forms.
780 xfs_dir2_leaf_getdents(
781 xfs_inode_t
*dp
, /* incore directory inode */
787 xfs_dabuf_t
*bp
; /* data block buffer */
788 int byteoff
; /* offset in current block */
789 xfs_dir2_db_t curdb
; /* db for current block */
790 xfs_dir2_off_t curoff
; /* current overall offset */
791 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
792 xfs_dir2_data_entry_t
*dep
; /* data entry */
793 xfs_dir2_data_unused_t
*dup
; /* unused entry */
794 int error
= 0; /* error return value */
795 int i
; /* temporary loop index */
796 int j
; /* temporary loop index */
797 int length
; /* temporary length value */
798 xfs_bmbt_irec_t
*map
; /* map vector for blocks */
799 xfs_extlen_t map_blocks
; /* number of fsbs in map */
800 xfs_dablk_t map_off
; /* last mapped file offset */
801 int map_size
; /* total entries in *map */
802 int map_valid
; /* valid entries in *map */
803 xfs_mount_t
*mp
; /* filesystem mount point */
804 xfs_dir2_off_t newoff
; /* new curoff after new blk */
805 int nmap
; /* mappings to ask xfs_bmapi */
806 char *ptr
= NULL
; /* pointer to current data */
807 int ra_current
; /* number of read-ahead blks */
808 int ra_index
; /* *map index for read-ahead */
809 int ra_offset
; /* map entry offset for ra */
810 int ra_want
; /* readahead count wanted */
813 * If the offset is at or past the largest allowed value,
814 * give up right away.
816 if (*offset
>= XFS_DIR2_MAX_DATAPTR
)
822 * Set up to bmap a number of blocks based on the caller's
823 * buffer size, the directory block size, and the filesystem
826 map_size
= howmany(bufsize
+ mp
->m_dirblksize
, mp
->m_sb
.sb_blocksize
);
827 map
= kmem_alloc(map_size
* sizeof(*map
), KM_SLEEP
);
828 map_valid
= ra_index
= ra_offset
= ra_current
= map_blocks
= 0;
832 * Inside the loop we keep the main offset value as a byte offset
833 * in the directory file.
835 curoff
= xfs_dir2_dataptr_to_byte(mp
, *offset
);
838 * Force this conversion through db so we truncate the offset
839 * down to get the start of the data block.
841 map_off
= xfs_dir2_db_to_da(mp
, xfs_dir2_byte_to_db(mp
, curoff
));
843 * Loop over directory entries until we reach the end offset.
844 * Get more blocks and readahead as necessary.
846 while (curoff
< XFS_DIR2_LEAF_OFFSET
) {
848 * If we have no buffer, or we're off the end of the
849 * current buffer, need to get another one.
851 if (!bp
|| ptr
>= (char *)bp
->data
+ mp
->m_dirblksize
) {
853 * If we have a buffer, we need to release it and
854 * take it out of the mapping.
857 xfs_da_brelse(NULL
, bp
);
859 map_blocks
-= mp
->m_dirblkfsbs
;
861 * Loop to get rid of the extents for the
864 for (i
= mp
->m_dirblkfsbs
; i
> 0; ) {
865 j
= MIN((int)map
->br_blockcount
, i
);
866 map
->br_blockcount
-= j
;
867 map
->br_startblock
+= j
;
868 map
->br_startoff
+= j
;
870 * If mapping is done, pitch it from
873 if (!map
->br_blockcount
&& --map_valid
)
874 memmove(&map
[0], &map
[1],
881 * Recalculate the readahead blocks wanted.
883 ra_want
= howmany(bufsize
+ mp
->m_dirblksize
,
884 mp
->m_sb
.sb_blocksize
) - 1;
885 ASSERT(ra_want
>= 0);
888 * If we don't have as many as we want, and we haven't
889 * run out of data blocks, get some more mappings.
891 if (1 + ra_want
> map_blocks
&&
893 xfs_dir2_byte_to_da(mp
, XFS_DIR2_LEAF_OFFSET
)) {
895 * Get more bmaps, fill in after the ones
896 * we already have in the table.
898 nmap
= map_size
- map_valid
;
899 error
= xfs_bmapi(NULL
, dp
,
901 xfs_dir2_byte_to_da(mp
,
902 XFS_DIR2_LEAF_OFFSET
) - map_off
,
903 XFS_BMAPI_METADATA
, NULL
, 0,
904 &map
[map_valid
], &nmap
, NULL
);
906 * Don't know if we should ignore this or
907 * try to return an error.
908 * The trouble with returning errors
909 * is that readdir will just stop without
910 * actually passing the error through.
915 * If we got all the mappings we asked for,
916 * set the final map offset based on the
917 * last bmap value received.
918 * Otherwise, we've reached the end.
920 if (nmap
== map_size
- map_valid
)
922 map
[map_valid
+ nmap
- 1].br_startoff
+
923 map
[map_valid
+ nmap
- 1].br_blockcount
;
926 xfs_dir2_byte_to_da(mp
,
927 XFS_DIR2_LEAF_OFFSET
);
929 * Look for holes in the mapping, and
930 * eliminate them. Count up the valid blocks.
932 for (i
= map_valid
; i
< map_valid
+ nmap
; ) {
933 if (map
[i
].br_startblock
==
936 length
= map_valid
+ nmap
- i
;
944 map
[i
].br_blockcount
;
951 * No valid mappings, so no more data blocks.
954 curoff
= xfs_dir2_da_to_byte(mp
, map_off
);
958 * Read the directory block starting at the first
961 curdb
= xfs_dir2_da_to_db(mp
, map
->br_startoff
);
962 error
= xfs_da_read_buf(NULL
, dp
, map
->br_startoff
,
963 map
->br_blockcount
>= mp
->m_dirblkfsbs
?
964 XFS_FSB_TO_DADDR(mp
, map
->br_startblock
) :
968 * Should just skip over the data block instead
974 * Adjust the current amount of read-ahead: we just
975 * read a block that was previously ra.
978 ra_current
-= mp
->m_dirblkfsbs
;
980 * Do we need more readahead?
982 for (ra_index
= ra_offset
= i
= 0;
983 ra_want
> ra_current
&& i
< map_blocks
;
984 i
+= mp
->m_dirblkfsbs
) {
985 ASSERT(ra_index
< map_valid
);
987 * Read-ahead a contiguous directory block.
989 if (i
> ra_current
&&
990 map
[ra_index
].br_blockcount
>=
992 xfs_buf_readahead(mp
->m_ddev_targp
,
994 map
[ra_index
].br_startblock
+
996 (int)BTOBB(mp
->m_dirblksize
));
1000 * Read-ahead a non-contiguous directory block.
1001 * This doesn't use our mapping, but this
1002 * is a very rare case.
1004 else if (i
> ra_current
) {
1005 (void)xfs_da_reada_buf(NULL
, dp
,
1006 map
[ra_index
].br_startoff
+
1007 ra_offset
, XFS_DATA_FORK
);
1011 * Advance offset through the mapping table.
1013 for (j
= 0; j
< mp
->m_dirblkfsbs
; j
++) {
1015 * The rest of this extent but not
1016 * more than a dir block.
1018 length
= MIN(mp
->m_dirblkfsbs
,
1019 (int)(map
[ra_index
].br_blockcount
-
1022 ra_offset
+= length
;
1024 * Advance to the next mapping if
1025 * this one is used up.
1028 map
[ra_index
].br_blockcount
) {
1035 * Having done a read, we need to set a new offset.
1037 newoff
= xfs_dir2_db_off_to_byte(mp
, curdb
, 0);
1039 * Start of the current block.
1041 if (curoff
< newoff
)
1044 * Make sure we're in the right block.
1046 else if (curoff
> newoff
)
1047 ASSERT(xfs_dir2_byte_to_db(mp
, curoff
) ==
1050 xfs_dir2_data_check(dp
, bp
);
1052 * Find our position in the block.
1054 ptr
= (char *)(hdr
+ 1);
1055 byteoff
= xfs_dir2_byte_to_off(mp
, curoff
);
1057 * Skip past the header.
1060 curoff
+= (uint
)sizeof(*hdr
);
1062 * Skip past entries until we reach our offset.
1065 while ((char *)ptr
- (char *)hdr
< byteoff
) {
1066 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1068 if (be16_to_cpu(dup
->freetag
)
1069 == XFS_DIR2_DATA_FREE_TAG
) {
1071 length
= be16_to_cpu(dup
->length
);
1075 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1077 xfs_dir2_data_entsize(dep
->namelen
);
1081 * Now set our real offset.
1084 xfs_dir2_db_off_to_byte(mp
,
1085 xfs_dir2_byte_to_db(mp
, curoff
),
1086 (char *)ptr
- (char *)hdr
);
1087 if (ptr
>= (char *)hdr
+ mp
->m_dirblksize
) {
1093 * We have a pointer to an entry.
1096 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1098 * No, it's unused, skip over it.
1100 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
1101 length
= be16_to_cpu(dup
->length
);
1107 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1108 length
= xfs_dir2_data_entsize(dep
->namelen
);
1110 if (filldir(dirent
, (char *)dep
->name
, dep
->namelen
,
1111 xfs_dir2_byte_to_dataptr(mp
, curoff
) & 0x7fffffff,
1112 be64_to_cpu(dep
->inumber
), DT_UNKNOWN
))
1116 * Advance to next entry in the block.
1120 /* bufsize may have just been a guess; don't go negative */
1121 bufsize
= bufsize
> length
? bufsize
- length
: 0;
1125 * All done. Set output offset value to current offset.
1127 if (curoff
> xfs_dir2_dataptr_to_byte(mp
, XFS_DIR2_MAX_DATAPTR
))
1128 *offset
= XFS_DIR2_MAX_DATAPTR
& 0x7fffffff;
1130 *offset
= xfs_dir2_byte_to_dataptr(mp
, curoff
) & 0x7fffffff;
1133 xfs_da_brelse(NULL
, bp
);
1138 * Initialize a new leaf block, leaf1 or leafn magic accepted.
1142 xfs_da_args_t
*args
, /* operation arguments */
1143 xfs_dir2_db_t bno
, /* directory block number */
1144 xfs_dabuf_t
**bpp
, /* out: leaf buffer */
1145 int magic
) /* magic number for block */
1147 xfs_dabuf_t
*bp
; /* leaf buffer */
1148 xfs_inode_t
*dp
; /* incore directory inode */
1149 int error
; /* error return code */
1150 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1151 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1152 xfs_mount_t
*mp
; /* filesystem mount point */
1153 xfs_trans_t
*tp
; /* transaction pointer */
1159 ASSERT(bno
>= XFS_DIR2_LEAF_FIRSTDB(mp
) &&
1160 bno
< XFS_DIR2_FREE_FIRSTDB(mp
));
1162 * Get the buffer for the block.
1164 error
= xfs_da_get_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, bno
), -1, &bp
,
1172 * Initialize the header.
1174 leaf
->hdr
.info
.magic
= cpu_to_be16(magic
);
1175 leaf
->hdr
.info
.forw
= 0;
1176 leaf
->hdr
.info
.back
= 0;
1177 leaf
->hdr
.count
= 0;
1178 leaf
->hdr
.stale
= 0;
1179 xfs_dir2_leaf_log_header(tp
, bp
);
1181 * If it's a leaf-format directory initialize the tail.
1182 * In this case our caller has the real bests table to copy into
1185 if (magic
== XFS_DIR2_LEAF1_MAGIC
) {
1186 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1188 xfs_dir2_leaf_log_tail(tp
, bp
);
1195 * Log the bests entries indicated from a leaf1 block.
1198 xfs_dir2_leaf_log_bests(
1199 xfs_trans_t
*tp
, /* transaction pointer */
1200 xfs_dabuf_t
*bp
, /* leaf buffer */
1201 int first
, /* first entry to log */
1202 int last
) /* last entry to log */
1204 __be16
*firstb
; /* pointer to first entry */
1205 __be16
*lastb
; /* pointer to last entry */
1206 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1207 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1210 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
);
1211 ltp
= xfs_dir2_leaf_tail_p(tp
->t_mountp
, leaf
);
1212 firstb
= xfs_dir2_leaf_bests_p(ltp
) + first
;
1213 lastb
= xfs_dir2_leaf_bests_p(ltp
) + last
;
1214 xfs_da_log_buf(tp
, bp
, (uint
)((char *)firstb
- (char *)leaf
),
1215 (uint
)((char *)lastb
- (char *)leaf
+ sizeof(*lastb
) - 1));
1219 * Log the leaf entries indicated from a leaf1 or leafn block.
1222 xfs_dir2_leaf_log_ents(
1223 xfs_trans_t
*tp
, /* transaction pointer */
1224 xfs_dabuf_t
*bp
, /* leaf buffer */
1225 int first
, /* first entry to log */
1226 int last
) /* last entry to log */
1228 xfs_dir2_leaf_entry_t
*firstlep
; /* pointer to first entry */
1229 xfs_dir2_leaf_entry_t
*lastlep
; /* pointer to last entry */
1230 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1233 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
||
1234 be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
1235 firstlep
= &leaf
->ents
[first
];
1236 lastlep
= &leaf
->ents
[last
];
1237 xfs_da_log_buf(tp
, bp
, (uint
)((char *)firstlep
- (char *)leaf
),
1238 (uint
)((char *)lastlep
- (char *)leaf
+ sizeof(*lastlep
) - 1));
1242 * Log the header of the leaf1 or leafn block.
1245 xfs_dir2_leaf_log_header(
1246 xfs_trans_t
*tp
, /* transaction pointer */
1247 xfs_dabuf_t
*bp
) /* leaf buffer */
1249 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1252 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
||
1253 be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
1254 xfs_da_log_buf(tp
, bp
, (uint
)((char *)&leaf
->hdr
- (char *)leaf
),
1255 (uint
)(sizeof(leaf
->hdr
) - 1));
1259 * Log the tail of the leaf1 block.
1262 xfs_dir2_leaf_log_tail(
1263 xfs_trans_t
*tp
, /* transaction pointer */
1264 xfs_dabuf_t
*bp
) /* leaf buffer */
1266 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1267 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1268 xfs_mount_t
*mp
; /* filesystem mount point */
1272 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAF1_MAGIC
);
1273 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1274 xfs_da_log_buf(tp
, bp
, (uint
)((char *)ltp
- (char *)leaf
),
1275 (uint
)(mp
->m_dirblksize
- 1));
1279 * Look up the entry referred to by args in the leaf format directory.
1280 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1281 * is also used by the node-format code.
1284 xfs_dir2_leaf_lookup(
1285 xfs_da_args_t
*args
) /* operation arguments */
1287 xfs_dabuf_t
*dbp
; /* data block buffer */
1288 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1289 xfs_inode_t
*dp
; /* incore directory inode */
1290 int error
; /* error return code */
1291 int index
; /* found entry index */
1292 xfs_dabuf_t
*lbp
; /* leaf buffer */
1293 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1294 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1295 xfs_trans_t
*tp
; /* transaction pointer */
1297 trace_xfs_dir2_leaf_lookup(args
);
1300 * Look up name in the leaf block, returning both buffers and index.
1302 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1307 xfs_dir2_leaf_check(dp
, lbp
);
1310 * Get to the leaf entry and contained data entry address.
1312 lep
= &leaf
->ents
[index
];
1314 * Point to the data entry.
1316 dep
= (xfs_dir2_data_entry_t
*)
1317 ((char *)dbp
->data
+
1318 xfs_dir2_dataptr_to_off(dp
->i_mount
, be32_to_cpu(lep
->address
)));
1320 * Return the found inode number & CI name if appropriate
1322 args
->inumber
= be64_to_cpu(dep
->inumber
);
1323 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
1324 xfs_da_brelse(tp
, dbp
);
1325 xfs_da_brelse(tp
, lbp
);
1326 return XFS_ERROR(error
);
1330 * Look up name/hash in the leaf block.
1331 * Fill in indexp with the found index, and dbpp with the data buffer.
1332 * If not found dbpp will be NULL, and ENOENT comes back.
1333 * lbpp will always be filled in with the leaf buffer unless there's an error.
1335 static int /* error */
1336 xfs_dir2_leaf_lookup_int(
1337 xfs_da_args_t
*args
, /* operation arguments */
1338 xfs_dabuf_t
**lbpp
, /* out: leaf buffer */
1339 int *indexp
, /* out: index in leaf block */
1340 xfs_dabuf_t
**dbpp
) /* out: data buffer */
1342 xfs_dir2_db_t curdb
= -1; /* current data block number */
1343 xfs_dabuf_t
*dbp
= NULL
; /* data buffer */
1344 xfs_dir2_data_entry_t
*dep
; /* data entry */
1345 xfs_inode_t
*dp
; /* incore directory inode */
1346 int error
; /* error return code */
1347 int index
; /* index in leaf block */
1348 xfs_dabuf_t
*lbp
; /* leaf buffer */
1349 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1350 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1351 xfs_mount_t
*mp
; /* filesystem mount point */
1352 xfs_dir2_db_t newdb
; /* new data block number */
1353 xfs_trans_t
*tp
; /* transaction pointer */
1354 xfs_dir2_db_t cidb
= -1; /* case match data block no. */
1355 enum xfs_dacmp cmp
; /* name compare result */
1361 * Read the leaf block into the buffer.
1363 error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirleafblk
, -1, &lbp
,
1369 xfs_dir2_leaf_check(dp
, lbp
);
1371 * Look for the first leaf entry with our hash value.
1373 index
= xfs_dir2_leaf_search_hash(args
, lbp
);
1375 * Loop over all the entries with the right hash value
1376 * looking to match the name.
1378 for (lep
= &leaf
->ents
[index
]; index
< be16_to_cpu(leaf
->hdr
.count
) &&
1379 be32_to_cpu(lep
->hashval
) == args
->hashval
;
1382 * Skip over stale leaf entries.
1384 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
1387 * Get the new data block number.
1389 newdb
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
1391 * If it's not the same as the old data block number,
1392 * need to pitch the old one and read the new one.
1394 if (newdb
!= curdb
) {
1396 xfs_da_brelse(tp
, dbp
);
1397 error
= xfs_da_read_buf(tp
, dp
,
1398 xfs_dir2_db_to_da(mp
, newdb
),
1399 -1, &dbp
, XFS_DATA_FORK
);
1401 xfs_da_brelse(tp
, lbp
);
1404 xfs_dir2_data_check(dp
, dbp
);
1408 * Point to the data entry.
1410 dep
= (xfs_dir2_data_entry_t
*)((char *)dbp
->data
+
1411 xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
1413 * Compare name and if it's an exact match, return the index
1414 * and buffer. If it's the first case-insensitive match, store
1415 * the index and buffer and continue looking for an exact match.
1417 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
1418 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
1419 args
->cmpresult
= cmp
;
1421 /* case exact match: return the current buffer. */
1422 if (cmp
== XFS_CMP_EXACT
) {
1429 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
1431 * Here, we can only be doing a lookup (not a rename or remove).
1432 * If a case-insensitive match was found earlier, re-read the
1433 * appropriate data block if required and return it.
1435 if (args
->cmpresult
== XFS_CMP_CASE
) {
1437 if (cidb
!= curdb
) {
1438 xfs_da_brelse(tp
, dbp
);
1439 error
= xfs_da_read_buf(tp
, dp
,
1440 xfs_dir2_db_to_da(mp
, cidb
),
1441 -1, &dbp
, XFS_DATA_FORK
);
1443 xfs_da_brelse(tp
, lbp
);
1451 * No match found, return ENOENT.
1455 xfs_da_brelse(tp
, dbp
);
1456 xfs_da_brelse(tp
, lbp
);
1457 return XFS_ERROR(ENOENT
);
1461 * Remove an entry from a leaf format directory.
1464 xfs_dir2_leaf_removename(
1465 xfs_da_args_t
*args
) /* operation arguments */
1467 __be16
*bestsp
; /* leaf block best freespace */
1468 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
1469 xfs_dir2_db_t db
; /* data block number */
1470 xfs_dabuf_t
*dbp
; /* data block buffer */
1471 xfs_dir2_data_entry_t
*dep
; /* data entry structure */
1472 xfs_inode_t
*dp
; /* incore directory inode */
1473 int error
; /* error return code */
1474 xfs_dir2_db_t i
; /* temporary data block # */
1475 int index
; /* index into leaf entries */
1476 xfs_dabuf_t
*lbp
; /* leaf buffer */
1477 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1478 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1479 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1480 xfs_mount_t
*mp
; /* filesystem mount point */
1481 int needlog
; /* need to log data header */
1482 int needscan
; /* need to rescan data frees */
1483 xfs_dir2_data_off_t oldbest
; /* old value of best free */
1484 xfs_trans_t
*tp
; /* transaction pointer */
1486 trace_xfs_dir2_leaf_removename(args
);
1489 * Lookup the leaf entry, get the leaf and data blocks read in.
1491 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1499 xfs_dir2_data_check(dp
, dbp
);
1501 * Point to the leaf entry, use that to point to the data entry.
1503 lep
= &leaf
->ents
[index
];
1504 db
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
1505 dep
= (xfs_dir2_data_entry_t
*)
1506 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
1507 needscan
= needlog
= 0;
1508 oldbest
= be16_to_cpu(hdr
->bestfree
[0].length
);
1509 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1510 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1511 ASSERT(be16_to_cpu(bestsp
[db
]) == oldbest
);
1513 * Mark the former data entry unused.
1515 xfs_dir2_data_make_free(tp
, dbp
,
1516 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)hdr
),
1517 xfs_dir2_data_entsize(dep
->namelen
), &needlog
, &needscan
);
1519 * We just mark the leaf entry stale by putting a null in it.
1521 be16_add_cpu(&leaf
->hdr
.stale
, 1);
1522 xfs_dir2_leaf_log_header(tp
, lbp
);
1523 lep
->address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
1524 xfs_dir2_leaf_log_ents(tp
, lbp
, index
, index
);
1526 * Scan the freespace in the data block again if necessary,
1527 * log the data block header if necessary.
1530 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
1532 xfs_dir2_data_log_header(tp
, dbp
);
1534 * If the longest freespace in the data block has changed,
1535 * put the new value in the bests table and log that.
1537 if (be16_to_cpu(hdr
->bestfree
[0].length
) != oldbest
) {
1538 bestsp
[db
] = hdr
->bestfree
[0].length
;
1539 xfs_dir2_leaf_log_bests(tp
, lbp
, db
, db
);
1541 xfs_dir2_data_check(dp
, dbp
);
1543 * If the data block is now empty then get rid of the data block.
1545 if (be16_to_cpu(hdr
->bestfree
[0].length
) ==
1546 mp
->m_dirblksize
- (uint
)sizeof(*hdr
)) {
1547 ASSERT(db
!= mp
->m_dirdatablk
);
1548 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1550 * Nope, can't get rid of it because it caused
1551 * allocation of a bmap btree block to do so.
1552 * Just go on, returning success, leaving the
1553 * empty block in place.
1555 if (error
== ENOSPC
&& args
->total
== 0) {
1556 xfs_da_buf_done(dbp
);
1559 xfs_dir2_leaf_check(dp
, lbp
);
1560 xfs_da_buf_done(lbp
);
1565 * If this is the last data block then compact the
1566 * bests table by getting rid of entries.
1568 if (db
== be32_to_cpu(ltp
->bestcount
) - 1) {
1570 * Look for the last active entry (i).
1572 for (i
= db
- 1; i
> 0; i
--) {
1573 if (be16_to_cpu(bestsp
[i
]) != NULLDATAOFF
)
1577 * Copy the table down so inactive entries at the
1580 memmove(&bestsp
[db
- i
], bestsp
,
1581 (be32_to_cpu(ltp
->bestcount
) - (db
- i
)) * sizeof(*bestsp
));
1582 be32_add_cpu(<p
->bestcount
, -(db
- i
));
1583 xfs_dir2_leaf_log_tail(tp
, lbp
);
1584 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1586 bestsp
[db
] = cpu_to_be16(NULLDATAOFF
);
1589 * If the data block was not the first one, drop it.
1591 else if (db
!= mp
->m_dirdatablk
&& dbp
!= NULL
) {
1592 xfs_da_buf_done(dbp
);
1595 xfs_dir2_leaf_check(dp
, lbp
);
1597 * See if we can convert to block form.
1599 return xfs_dir2_leaf_to_block(args
, lbp
, dbp
);
1603 * Replace the inode number in a leaf format directory entry.
1606 xfs_dir2_leaf_replace(
1607 xfs_da_args_t
*args
) /* operation arguments */
1609 xfs_dabuf_t
*dbp
; /* data block buffer */
1610 xfs_dir2_data_entry_t
*dep
; /* data block entry */
1611 xfs_inode_t
*dp
; /* incore directory inode */
1612 int error
; /* error return code */
1613 int index
; /* index of leaf entry */
1614 xfs_dabuf_t
*lbp
; /* leaf buffer */
1615 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1616 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1617 xfs_trans_t
*tp
; /* transaction pointer */
1619 trace_xfs_dir2_leaf_replace(args
);
1622 * Look up the entry.
1624 if ((error
= xfs_dir2_leaf_lookup_int(args
, &lbp
, &index
, &dbp
))) {
1630 * Point to the leaf entry, get data address from it.
1632 lep
= &leaf
->ents
[index
];
1634 * Point to the data entry.
1636 dep
= (xfs_dir2_data_entry_t
*)
1637 ((char *)dbp
->data
+
1638 xfs_dir2_dataptr_to_off(dp
->i_mount
, be32_to_cpu(lep
->address
)));
1639 ASSERT(args
->inumber
!= be64_to_cpu(dep
->inumber
));
1641 * Put the new inode number in, log it.
1643 dep
->inumber
= cpu_to_be64(args
->inumber
);
1645 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
1646 xfs_da_buf_done(dbp
);
1647 xfs_dir2_leaf_check(dp
, lbp
);
1648 xfs_da_brelse(tp
, lbp
);
1653 * Return index in the leaf block (lbp) which is either the first
1654 * one with this hash value, or if there are none, the insert point
1655 * for that hash value.
1657 int /* index value */
1658 xfs_dir2_leaf_search_hash(
1659 xfs_da_args_t
*args
, /* operation arguments */
1660 xfs_dabuf_t
*lbp
) /* leaf buffer */
1662 xfs_dahash_t hash
=0; /* hash from this entry */
1663 xfs_dahash_t hashwant
; /* hash value looking for */
1664 int high
; /* high leaf index */
1665 int low
; /* low leaf index */
1666 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1667 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1668 int mid
=0; /* current leaf index */
1672 if (!leaf
->hdr
.count
)
1676 * Note, the table cannot be empty, so we have to go through the loop.
1677 * Binary search the leaf entries looking for our hash value.
1679 for (lep
= leaf
->ents
, low
= 0, high
= be16_to_cpu(leaf
->hdr
.count
) - 1,
1680 hashwant
= args
->hashval
;
1682 mid
= (low
+ high
) >> 1;
1683 if ((hash
= be32_to_cpu(lep
[mid
].hashval
)) == hashwant
)
1685 if (hash
< hashwant
)
1691 * Found one, back up through all the equal hash values.
1693 if (hash
== hashwant
) {
1694 while (mid
> 0 && be32_to_cpu(lep
[mid
- 1].hashval
) == hashwant
) {
1699 * Need to point to an entry higher than ours.
1701 else if (hash
< hashwant
)
1707 * Trim off a trailing data block. We know it's empty since the leaf
1708 * freespace table says so.
1711 xfs_dir2_leaf_trim_data(
1712 xfs_da_args_t
*args
, /* operation arguments */
1713 xfs_dabuf_t
*lbp
, /* leaf buffer */
1714 xfs_dir2_db_t db
) /* data block number */
1716 __be16
*bestsp
; /* leaf bests table */
1717 xfs_dabuf_t
*dbp
; /* data block buffer */
1718 xfs_inode_t
*dp
; /* incore directory inode */
1719 int error
; /* error return value */
1720 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1721 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1722 xfs_mount_t
*mp
; /* filesystem mount point */
1723 xfs_trans_t
*tp
; /* transaction pointer */
1729 * Read the offending data block. We need its buffer.
1731 if ((error
= xfs_da_read_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, db
), -1, &dbp
,
1737 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1741 struct xfs_dir2_data_hdr
*hdr
= dbp
->data
;
1743 ASSERT(be32_to_cpu(hdr
->magic
) == XFS_DIR2_DATA_MAGIC
);
1744 ASSERT(be16_to_cpu(hdr
->bestfree
[0].length
) ==
1745 mp
->m_dirblksize
- (uint
)sizeof(*hdr
));
1746 ASSERT(db
== be32_to_cpu(ltp
->bestcount
) - 1);
1751 * Get rid of the data block.
1753 if ((error
= xfs_dir2_shrink_inode(args
, db
, dbp
))) {
1754 ASSERT(error
!= ENOSPC
);
1755 xfs_da_brelse(tp
, dbp
);
1759 * Eliminate the last bests entry from the table.
1761 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1762 be32_add_cpu(<p
->bestcount
, -1);
1763 memmove(&bestsp
[1], &bestsp
[0], be32_to_cpu(ltp
->bestcount
) * sizeof(*bestsp
));
1764 xfs_dir2_leaf_log_tail(tp
, lbp
);
1765 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1769 static inline size_t
1771 struct xfs_dir2_leaf_hdr
*hdr
,
1776 entries
= be16_to_cpu(hdr
->count
) - be16_to_cpu(hdr
->stale
);
1777 return sizeof(xfs_dir2_leaf_hdr_t
) +
1778 entries
* sizeof(xfs_dir2_leaf_entry_t
) +
1779 counts
* sizeof(xfs_dir2_data_off_t
) +
1780 sizeof(xfs_dir2_leaf_tail_t
);
1784 * Convert node form directory to leaf form directory.
1785 * The root of the node form dir needs to already be a LEAFN block.
1786 * Just return if we can't do anything.
1789 xfs_dir2_node_to_leaf(
1790 xfs_da_state_t
*state
) /* directory operation state */
1792 xfs_da_args_t
*args
; /* operation arguments */
1793 xfs_inode_t
*dp
; /* incore directory inode */
1794 int error
; /* error return code */
1795 xfs_dabuf_t
*fbp
; /* buffer for freespace block */
1796 xfs_fileoff_t fo
; /* freespace file offset */
1797 xfs_dir2_free_t
*free
; /* freespace structure */
1798 xfs_dabuf_t
*lbp
; /* buffer for leaf block */
1799 xfs_dir2_leaf_tail_t
*ltp
; /* tail of leaf structure */
1800 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1801 xfs_mount_t
*mp
; /* filesystem mount point */
1802 int rval
; /* successful free trim? */
1803 xfs_trans_t
*tp
; /* transaction pointer */
1806 * There's more than a leaf level in the btree, so there must
1807 * be multiple leafn blocks. Give up.
1809 if (state
->path
.active
> 1)
1813 trace_xfs_dir2_node_to_leaf(args
);
1819 * Get the last offset in the file.
1821 if ((error
= xfs_bmap_last_offset(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1824 fo
-= mp
->m_dirblkfsbs
;
1826 * If there are freespace blocks other than the first one,
1827 * take this opportunity to remove trailing empty freespace blocks
1828 * that may have been left behind during no-space-reservation
1831 while (fo
> mp
->m_dirfreeblk
) {
1832 if ((error
= xfs_dir2_node_trim_free(args
, fo
, &rval
))) {
1836 fo
-= mp
->m_dirblkfsbs
;
1841 * Now find the block just before the freespace block.
1843 if ((error
= xfs_bmap_last_before(tp
, dp
, &fo
, XFS_DATA_FORK
))) {
1847 * If it's not the single leaf block, give up.
1849 if (XFS_FSB_TO_B(mp
, fo
) > XFS_DIR2_LEAF_OFFSET
+ mp
->m_dirblksize
)
1851 lbp
= state
->path
.blk
[0].bp
;
1853 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
1855 * Read the freespace block.
1857 if ((error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirfreeblk
, -1, &fbp
,
1862 ASSERT(be32_to_cpu(free
->hdr
.magic
) == XFS_DIR2_FREE_MAGIC
);
1863 ASSERT(!free
->hdr
.firstdb
);
1866 * Now see if the leafn and free data will fit in a leaf1.
1867 * If not, release the buffer and give up.
1869 if (xfs_dir2_leaf_size(&leaf
->hdr
, be32_to_cpu(free
->hdr
.nvalid
)) >
1871 xfs_da_brelse(tp
, fbp
);
1876 * If the leaf has any stale entries in it, compress them out.
1877 * The compact routine will log the header.
1879 if (be16_to_cpu(leaf
->hdr
.stale
))
1880 xfs_dir2_leaf_compact(args
, lbp
);
1882 xfs_dir2_leaf_log_header(tp
, lbp
);
1883 leaf
->hdr
.info
.magic
= cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
);
1885 * Set up the leaf tail from the freespace block.
1887 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1888 ltp
->bestcount
= free
->hdr
.nvalid
;
1890 * Set up the leaf bests table.
1892 memcpy(xfs_dir2_leaf_bests_p(ltp
), free
->bests
,
1893 be32_to_cpu(ltp
->bestcount
) * sizeof(xfs_dir2_data_off_t
));
1894 xfs_dir2_leaf_log_bests(tp
, lbp
, 0, be32_to_cpu(ltp
->bestcount
) - 1);
1895 xfs_dir2_leaf_log_tail(tp
, lbp
);
1896 xfs_dir2_leaf_check(dp
, lbp
);
1898 * Get rid of the freespace block.
1900 error
= xfs_dir2_shrink_inode(args
, XFS_DIR2_FREE_FIRSTDB(mp
), fbp
);
1903 * This can't fail here because it can only happen when
1904 * punching out the middle of an extent, and this is an
1907 ASSERT(error
!= ENOSPC
);
1912 * Now see if we can convert the single-leaf directory
1913 * down to a block form directory.
1914 * This routine always kills the dabuf for the leaf, so
1915 * eliminate it from the path.
1917 error
= xfs_dir2_leaf_to_block(args
, lbp
, NULL
);
1918 state
->path
.blk
[0].bp
= NULL
;