2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
23 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
32 #include "xfs_dir2_format.h"
33 #include "xfs_dir2_priv.h"
34 #include "xfs_error.h"
35 #include "xfs_trace.h"
38 * Local function declarations.
41 static void xfs_dir2_leaf_check(struct xfs_inode *dp, struct xfs_buf *bp);
43 #define xfs_dir2_leaf_check(dp, bp)
45 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
46 int *indexp, struct xfs_buf **dbpp);
47 static void xfs_dir2_leaf_log_bests(struct xfs_trans *tp, struct xfs_buf *bp,
49 static void xfs_dir2_leaf_log_tail(struct xfs_trans *tp, struct xfs_buf *bp);
56 struct xfs_mount *mp = bp->b_target->bt_mount;
57 struct xfs_dir2_leaf_hdr *hdr = bp->b_addr;
60 block_ok = hdr->info.magic == magic;
62 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, hdr);
63 xfs_buf_ioerror(bp, EFSCORRUPTED);
68 xfs_dir2_leaf1_read_verify(
71 xfs_dir2_leaf_verify(bp, cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
75 xfs_dir2_leaf1_write_verify(
78 xfs_dir2_leaf_verify(bp, cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
82 xfs_dir2_leafn_read_verify(
85 xfs_dir2_leaf_verify(bp, cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
89 xfs_dir2_leafn_write_verify(
92 xfs_dir2_leaf_verify(bp, cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
95 static const struct xfs_buf_ops xfs_dir2_leaf1_buf_ops = {
96 .verify_read = xfs_dir2_leaf1_read_verify,
97 .verify_write = xfs_dir2_leaf1_write_verify,
100 const struct xfs_buf_ops xfs_dir2_leafn_buf_ops = {
101 .verify_read = xfs_dir2_leafn_read_verify,
102 .verify_write = xfs_dir2_leafn_write_verify,
107 struct xfs_trans *tp,
108 struct xfs_inode *dp,
110 xfs_daddr_t mappedbno,
111 struct xfs_buf **bpp)
113 return xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
114 XFS_DATA_FORK, &xfs_dir2_leaf1_buf_ops);
119 struct xfs_trans *tp,
120 struct xfs_inode *dp,
122 xfs_daddr_t mappedbno,
123 struct xfs_buf **bpp)
125 return xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
126 XFS_DATA_FORK, &xfs_dir2_leafn_buf_ops);
130 * Convert a block form directory to a leaf form directory.
133 xfs_dir2_block_to_leaf(
134 xfs_da_args_t *args, /* operation arguments */
135 struct xfs_buf *dbp) /* input block's buffer */
137 __be16 *bestsp; /* leaf's bestsp entries */
138 xfs_dablk_t blkno; /* leaf block's bno */
139 xfs_dir2_data_hdr_t *hdr; /* block header */
140 xfs_dir2_leaf_entry_t *blp; /* block's leaf entries */
141 xfs_dir2_block_tail_t *btp; /* block's tail */
142 xfs_inode_t *dp; /* incore directory inode */
143 int error; /* error return code */
144 struct xfs_buf *lbp; /* leaf block's buffer */
145 xfs_dir2_db_t ldb; /* leaf block's bno */
146 xfs_dir2_leaf_t *leaf; /* leaf structure */
147 xfs_dir2_leaf_tail_t *ltp; /* leaf's tail */
148 xfs_mount_t *mp; /* filesystem mount point */
149 int needlog; /* need to log block header */
150 int needscan; /* need to rescan bestfree */
151 xfs_trans_t *tp; /* transaction pointer */
152 struct xfs_dir2_data_free *bf;
154 trace_xfs_dir2_block_to_leaf(args);
160 * Add the leaf block to the inode.
161 * This interface will only put blocks in the leaf/node range.
162 * Since that's empty now, we'll get the root (block 0 in range).
164 if ((error = xfs_da_grow_inode(args, &blkno))) {
167 ldb = xfs_dir2_da_to_db(mp, blkno);
168 ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
170 * Initialize the leaf block, get a buffer for it.
172 if ((error = xfs_dir2_leaf_init(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC))) {
178 xfs_dir2_data_check(dp, dbp);
179 btp = xfs_dir2_block_tail_p(mp, hdr);
180 blp = xfs_dir2_block_leaf_p(btp);
181 bf = xfs_dir3_data_bestfree_p(hdr);
183 * Set the counts in the leaf header.
185 leaf->hdr.count = cpu_to_be16(be32_to_cpu(btp->count));
186 leaf->hdr.stale = cpu_to_be16(be32_to_cpu(btp->stale));
188 * Could compact these but I think we always do the conversion
189 * after squeezing out stale entries.
191 memcpy(leaf->ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
192 xfs_dir2_leaf_log_ents(tp, lbp, 0, be16_to_cpu(leaf->hdr.count) - 1);
196 * Make the space formerly occupied by the leaf entries and block
199 xfs_dir2_data_make_free(tp, dbp,
200 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
201 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
203 &needlog, &needscan);
205 * Fix up the block header, make it a data block.
207 dbp->b_ops = &xfs_dir2_data_buf_ops;
208 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
210 xfs_dir2_data_freescan(mp, hdr, &needlog);
212 * Set up leaf tail and bests table.
214 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
215 ltp->bestcount = cpu_to_be32(1);
216 bestsp = xfs_dir2_leaf_bests_p(ltp);
217 bestsp[0] = bf[0].length;
219 * Log the data header and leaf bests table.
222 xfs_dir2_data_log_header(tp, dbp);
223 xfs_dir2_leaf_check(dp, lbp);
224 xfs_dir2_data_check(dp, dbp);
225 xfs_dir2_leaf_log_bests(tp, lbp, 0, 0);
230 xfs_dir2_leaf_find_stale(
231 struct xfs_dir2_leaf *leaf,
237 * Find the first stale entry before our index, if any.
239 for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
240 if (leaf->ents[*lowstale].address ==
241 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
246 * Find the first stale entry at or after our index, if any.
247 * Stop if the result would require moving more entries than using
250 for (*highstale = index;
251 *highstale < be16_to_cpu(leaf->hdr.count);
253 if (leaf->ents[*highstale].address ==
254 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
256 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
261 struct xfs_dir2_leaf_entry *
262 xfs_dir2_leaf_find_entry(
263 xfs_dir2_leaf_t *leaf, /* leaf structure */
264 int index, /* leaf table position */
265 int compact, /* need to compact leaves */
266 int lowstale, /* index of prev stale leaf */
267 int highstale, /* index of next stale leaf */
268 int *lfloglow, /* low leaf logging index */
269 int *lfloghigh) /* high leaf logging index */
271 if (!leaf->hdr.stale) {
272 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
275 * Now we need to make room to insert the leaf entry.
277 * If there are no stale entries, just insert a hole at index.
279 lep = &leaf->ents[index];
280 if (index < be16_to_cpu(leaf->hdr.count))
281 memmove(lep + 1, lep,
282 (be16_to_cpu(leaf->hdr.count) - index) *
286 * Record low and high logging indices for the leaf.
289 *lfloghigh = be16_to_cpu(leaf->hdr.count);
290 be16_add_cpu(&leaf->hdr.count, 1);
295 * There are stale entries.
297 * We will use one of them for the new entry. It's probably not at
298 * the right location, so we'll have to shift some up or down first.
300 * If we didn't compact before, we need to find the nearest stale
301 * entries before and after our insertion point.
304 xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
307 * If the low one is better, use it.
310 (highstale == be16_to_cpu(leaf->hdr.count) ||
311 index - lowstale - 1 < highstale - index)) {
312 ASSERT(index - lowstale - 1 >= 0);
313 ASSERT(leaf->ents[lowstale].address ==
314 cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
317 * Copy entries up to cover the stale entry and make room
320 if (index - lowstale - 1 > 0) {
321 memmove(&leaf->ents[lowstale],
322 &leaf->ents[lowstale + 1],
323 (index - lowstale - 1) *
324 sizeof(xfs_dir2_leaf_entry_t));
326 *lfloglow = MIN(lowstale, *lfloglow);
327 *lfloghigh = MAX(index - 1, *lfloghigh);
328 be16_add_cpu(&leaf->hdr.stale, -1);
329 return &leaf->ents[index - 1];
333 * The high one is better, so use that one.
335 ASSERT(highstale - index >= 0);
336 ASSERT(leaf->ents[highstale].address ==
337 cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
340 * Copy entries down to cover the stale entry and make room for the
343 if (highstale - index > 0) {
344 memmove(&leaf->ents[index + 1],
346 (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
348 *lfloglow = MIN(index, *lfloglow);
349 *lfloghigh = MAX(highstale, *lfloghigh);
350 be16_add_cpu(&leaf->hdr.stale, -1);
351 return &leaf->ents[index];
355 * Add an entry to a leaf form directory.
358 xfs_dir2_leaf_addname(
359 xfs_da_args_t *args) /* operation arguments */
361 __be16 *bestsp; /* freespace table in leaf */
362 int compact; /* need to compact leaves */
363 xfs_dir2_data_hdr_t *hdr; /* data block header */
364 struct xfs_buf *dbp; /* data block buffer */
365 xfs_dir2_data_entry_t *dep; /* data block entry */
366 xfs_inode_t *dp; /* incore directory inode */
367 xfs_dir2_data_unused_t *dup; /* data unused entry */
368 int error; /* error return value */
369 int grown; /* allocated new data block */
370 int highstale; /* index of next stale leaf */
371 int i; /* temporary, index */
372 int index; /* leaf table position */
373 struct xfs_buf *lbp; /* leaf's buffer */
374 xfs_dir2_leaf_t *leaf; /* leaf structure */
375 int length; /* length of new entry */
376 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
377 int lfloglow; /* low leaf logging index */
378 int lfloghigh; /* high leaf logging index */
379 int lowstale; /* index of prev stale leaf */
380 xfs_dir2_leaf_tail_t *ltp; /* leaf tail pointer */
381 xfs_mount_t *mp; /* filesystem mount point */
382 int needbytes; /* leaf block bytes needed */
383 int needlog; /* need to log data header */
384 int needscan; /* need to rescan data free */
385 __be16 *tagp; /* end of data entry */
386 xfs_trans_t *tp; /* transaction pointer */
387 xfs_dir2_db_t use_block; /* data block number */
389 trace_xfs_dir2_leaf_addname(args);
395 error = xfs_dir2_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
400 * Look up the entry by hash value and name.
401 * We know it's not there, our caller has already done a lookup.
402 * So the index is of the entry to insert in front of.
403 * But if there are dup hash values the index is of the first of those.
405 index = xfs_dir2_leaf_search_hash(args, lbp);
407 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
408 bestsp = xfs_dir2_leaf_bests_p(ltp);
409 length = xfs_dir2_data_entsize(args->namelen);
411 * See if there are any entries with the same hash value
412 * and space in their block for the new entry.
413 * This is good because it puts multiple same-hash value entries
414 * in a data block, improving the lookup of those entries.
416 for (use_block = -1, lep = &leaf->ents[index];
417 index < be16_to_cpu(leaf->hdr.count) && be32_to_cpu(lep->hashval) == args->hashval;
419 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
421 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
422 ASSERT(i < be32_to_cpu(ltp->bestcount));
423 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
424 if (be16_to_cpu(bestsp[i]) >= length) {
430 * Didn't find a block yet, linear search all the data blocks.
432 if (use_block == -1) {
433 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
435 * Remember a block we see that's missing.
437 if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
440 else if (be16_to_cpu(bestsp[i]) >= length) {
447 * How many bytes do we need in the leaf block?
450 if (!leaf->hdr.stale)
451 needbytes += sizeof(xfs_dir2_leaf_entry_t);
453 needbytes += sizeof(xfs_dir2_data_off_t);
456 * Now kill use_block if it refers to a missing block, so we
457 * can use it as an indication of allocation needed.
459 if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
462 * If we don't have enough free bytes but we can make enough
463 * by compacting out stale entries, we'll do that.
465 if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <
466 needbytes && be16_to_cpu(leaf->hdr.stale) > 1) {
470 * Otherwise if we don't have enough free bytes we need to
471 * convert to node form.
473 else if ((char *)bestsp - (char *)&leaf->ents[be16_to_cpu(
474 leaf->hdr.count)] < needbytes) {
476 * Just checking or no space reservation, give up.
478 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
480 xfs_trans_brelse(tp, lbp);
481 return XFS_ERROR(ENOSPC);
484 * Convert to node form.
486 error = xfs_dir2_leaf_to_node(args, lbp);
490 * Then add the new entry.
492 return xfs_dir2_node_addname(args);
495 * Otherwise it will fit without compaction.
500 * If just checking, then it will fit unless we needed to allocate
503 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
504 xfs_trans_brelse(tp, lbp);
505 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
508 * If no allocations are allowed, return now before we've
511 if (args->total == 0 && use_block == -1) {
512 xfs_trans_brelse(tp, lbp);
513 return XFS_ERROR(ENOSPC);
516 * Need to compact the leaf entries, removing stale ones.
517 * Leave one stale entry behind - the one closest to our
518 * insertion index - and we'll shift that one to our insertion
522 xfs_dir2_leaf_compact_x1(lbp, &index, &lowstale, &highstale,
523 &lfloglow, &lfloghigh);
526 * There are stale entries, so we'll need log-low and log-high
527 * impossibly bad values later.
529 else if (be16_to_cpu(leaf->hdr.stale)) {
530 lfloglow = be16_to_cpu(leaf->hdr.count);
534 * If there was no data block space found, we need to allocate
537 if (use_block == -1) {
539 * Add the new data block.
541 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
543 xfs_trans_brelse(tp, lbp);
547 * Initialize the block.
549 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
550 xfs_trans_brelse(tp, lbp);
554 * If we're adding a new data block on the end we need to
555 * extend the bests table. Copy it up one entry.
557 if (use_block >= be32_to_cpu(ltp->bestcount)) {
559 memmove(&bestsp[0], &bestsp[1],
560 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
561 be32_add_cpu(<p->bestcount, 1);
562 xfs_dir2_leaf_log_tail(tp, lbp);
563 xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
566 * If we're filling in a previously empty block just log it.
569 xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
571 bestsp[use_block] = hdr->bestfree[0].length;
575 * Already had space in some data block.
576 * Just read that one in.
578 error = xfs_dir2_data_read(tp, dp,
579 xfs_dir2_db_to_da(mp, use_block),
582 xfs_trans_brelse(tp, lbp);
589 * Point to the biggest freespace in our data block.
591 dup = (xfs_dir2_data_unused_t *)
592 ((char *)hdr + be16_to_cpu(hdr->bestfree[0].offset));
593 ASSERT(be16_to_cpu(dup->length) >= length);
594 needscan = needlog = 0;
596 * Mark the initial part of our freespace in use for the new entry.
598 xfs_dir2_data_use_free(tp, dbp, dup,
599 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
600 &needlog, &needscan);
602 * Initialize our new entry (at last).
604 dep = (xfs_dir2_data_entry_t *)dup;
605 dep->inumber = cpu_to_be64(args->inumber);
606 dep->namelen = args->namelen;
607 memcpy(dep->name, args->name, dep->namelen);
608 tagp = xfs_dir2_data_entry_tag_p(dep);
609 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
611 * Need to scan fix up the bestfree table.
614 xfs_dir2_data_freescan(mp, hdr, &needlog);
616 * Need to log the data block's header.
619 xfs_dir2_data_log_header(tp, dbp);
620 xfs_dir2_data_log_entry(tp, dbp, dep);
622 * If the bests table needs to be changed, do it.
623 * Log the change unless we've already done that.
625 if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(hdr->bestfree[0].length)) {
626 bestsp[use_block] = hdr->bestfree[0].length;
628 xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
631 lep = xfs_dir2_leaf_find_entry(leaf, index, compact, lowstale,
632 highstale, &lfloglow, &lfloghigh);
635 * Fill in the new leaf entry.
637 lep->hashval = cpu_to_be32(args->hashval);
638 lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
639 be16_to_cpu(*tagp)));
641 * Log the leaf fields and give up the buffers.
643 xfs_dir2_leaf_log_header(tp, lbp);
644 xfs_dir2_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
645 xfs_dir2_leaf_check(dp, lbp);
646 xfs_dir2_data_check(dp, dbp);
652 * Check the internal consistency of a leaf1 block.
653 * Pop an assert if something is wrong.
657 struct xfs_inode *dp, /* incore directory inode */
658 struct xfs_buf *bp) /* leaf's buffer */
660 int i; /* leaf index */
661 xfs_dir2_leaf_t *leaf; /* leaf structure */
662 xfs_dir2_leaf_tail_t *ltp; /* leaf tail pointer */
663 xfs_mount_t *mp; /* filesystem mount point */
664 int stale; /* count of stale leaves */
668 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
670 * This value is not restrictive enough.
671 * Should factor in the size of the bests table as well.
672 * We can deduce a value for that from di_size.
674 ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
675 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
677 * Leaves and bests don't overlap.
679 ASSERT((char *)&leaf->ents[be16_to_cpu(leaf->hdr.count)] <=
680 (char *)xfs_dir2_leaf_bests_p(ltp));
682 * Check hash value order, count stale entries.
684 for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
685 if (i + 1 < be16_to_cpu(leaf->hdr.count))
686 ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
687 be32_to_cpu(leaf->ents[i + 1].hashval));
688 if (leaf->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
691 ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
696 * Compact out any stale entries in the leaf.
697 * Log the header and changed leaf entries, if any.
700 xfs_dir2_leaf_compact(
701 xfs_da_args_t *args, /* operation arguments */
702 struct xfs_buf *bp) /* leaf buffer */
704 int from; /* source leaf index */
705 xfs_dir2_leaf_t *leaf; /* leaf structure */
706 int loglow; /* first leaf entry to log */
707 int to; /* target leaf index */
710 if (!leaf->hdr.stale) {
714 * Compress out the stale entries in place.
716 for (from = to = 0, loglow = -1; from < be16_to_cpu(leaf->hdr.count); from++) {
717 if (leaf->ents[from].address ==
718 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
721 * Only actually copy the entries that are different.
726 leaf->ents[to] = leaf->ents[from];
731 * Update and log the header, log the leaf entries.
733 ASSERT(be16_to_cpu(leaf->hdr.stale) == from - to);
734 be16_add_cpu(&leaf->hdr.count, -(be16_to_cpu(leaf->hdr.stale)));
736 xfs_dir2_leaf_log_header(args->trans, bp);
738 xfs_dir2_leaf_log_ents(args->trans, bp, loglow, to - 1);
742 * Compact the leaf entries, removing stale ones.
743 * Leave one stale entry behind - the one closest to our
744 * insertion index - and the caller will shift that one to our insertion
746 * Return new insertion index, where the remaining stale entry is,
747 * and leaf logging indices.
750 xfs_dir2_leaf_compact_x1(
751 struct xfs_buf *bp, /* leaf buffer */
752 int *indexp, /* insertion index */
753 int *lowstalep, /* out: stale entry before us */
754 int *highstalep, /* out: stale entry after us */
755 int *lowlogp, /* out: low log index */
756 int *highlogp) /* out: high log index */
758 int from; /* source copy index */
759 int highstale; /* stale entry at/after index */
760 int index; /* insertion index */
761 int keepstale; /* source index of kept stale */
762 xfs_dir2_leaf_t *leaf; /* leaf structure */
763 int lowstale; /* stale entry before index */
764 int newindex=0; /* new insertion index */
765 int to; /* destination copy index */
768 ASSERT(be16_to_cpu(leaf->hdr.stale) > 1);
771 xfs_dir2_leaf_find_stale(leaf, index, &lowstale, &highstale);
774 * Pick the better of lowstale and highstale.
777 (highstale == be16_to_cpu(leaf->hdr.count) ||
778 index - lowstale <= highstale - index))
779 keepstale = lowstale;
781 keepstale = highstale;
783 * Copy the entries in place, removing all the stale entries
786 for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
788 * Notice the new value of index.
792 if (from != keepstale &&
793 leaf->ents[from].address ==
794 cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
800 * Record the new keepstale value for the insertion.
802 if (from == keepstale)
803 lowstale = highstale = to;
805 * Copy only the entries that have moved.
808 leaf->ents[to] = leaf->ents[from];
813 * If the insertion point was past the last entry,
814 * set the new insertion point accordingly.
820 * Adjust the leaf header values.
822 be16_add_cpu(&leaf->hdr.count, -(from - to));
823 leaf->hdr.stale = cpu_to_be16(1);
825 * Remember the low/high stale value only in the "right"
828 if (lowstale >= newindex)
831 highstale = be16_to_cpu(leaf->hdr.count);
832 *highlogp = be16_to_cpu(leaf->hdr.count) - 1;
833 *lowstalep = lowstale;
834 *highstalep = highstale;
837 struct xfs_dir2_leaf_map_info {
838 xfs_extlen_t map_blocks; /* number of fsbs in map */
839 xfs_dablk_t map_off; /* last mapped file offset */
840 int map_size; /* total entries in *map */
841 int map_valid; /* valid entries in *map */
842 int nmap; /* mappings to ask xfs_bmapi */
843 xfs_dir2_db_t curdb; /* db for current block */
844 int ra_current; /* number of read-ahead blks */
845 int ra_index; /* *map index for read-ahead */
846 int ra_offset; /* map entry offset for ra */
847 int ra_want; /* readahead count wanted */
848 struct xfs_bmbt_irec map[]; /* map vector for blocks */
852 xfs_dir2_leaf_readbuf(
853 struct xfs_inode *dp,
855 struct xfs_dir2_leaf_map_info *mip,
856 xfs_dir2_off_t *curoff,
857 struct xfs_buf **bpp)
859 struct xfs_mount *mp = dp->i_mount;
860 struct xfs_buf *bp = *bpp;
861 struct xfs_bmbt_irec *map = mip->map;
868 * If we have a buffer, we need to release it and
869 * take it out of the mapping.
873 xfs_trans_brelse(NULL, bp);
875 mip->map_blocks -= mp->m_dirblkfsbs;
877 * Loop to get rid of the extents for the
880 for (i = mp->m_dirblkfsbs; i > 0; ) {
881 j = min_t(int, map->br_blockcount, i);
882 map->br_blockcount -= j;
883 map->br_startblock += j;
884 map->br_startoff += j;
886 * If mapping is done, pitch it from
889 if (!map->br_blockcount && --mip->map_valid)
890 memmove(&map[0], &map[1],
891 sizeof(map[0]) * mip->map_valid);
897 * Recalculate the readahead blocks wanted.
899 mip->ra_want = howmany(bufsize + mp->m_dirblksize,
900 mp->m_sb.sb_blocksize) - 1;
901 ASSERT(mip->ra_want >= 0);
904 * If we don't have as many as we want, and we haven't
905 * run out of data blocks, get some more mappings.
907 if (1 + mip->ra_want > mip->map_blocks &&
908 mip->map_off < xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET)) {
910 * Get more bmaps, fill in after the ones
911 * we already have in the table.
913 mip->nmap = mip->map_size - mip->map_valid;
914 error = xfs_bmapi_read(dp, mip->map_off,
915 xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET) -
917 &map[mip->map_valid], &mip->nmap, 0);
920 * Don't know if we should ignore this or try to return an
921 * error. The trouble with returning errors is that readdir
922 * will just stop without actually passing the error through.
928 * If we got all the mappings we asked for, set the final map
929 * offset based on the last bmap value received. Otherwise,
930 * we've reached the end.
932 if (mip->nmap == mip->map_size - mip->map_valid) {
933 i = mip->map_valid + mip->nmap - 1;
934 mip->map_off = map[i].br_startoff + map[i].br_blockcount;
936 mip->map_off = xfs_dir2_byte_to_da(mp,
937 XFS_DIR2_LEAF_OFFSET);
940 * Look for holes in the mapping, and eliminate them. Count up
943 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
944 if (map[i].br_startblock == HOLESTARTBLOCK) {
946 length = mip->map_valid + mip->nmap - i;
948 memmove(&map[i], &map[i + 1],
949 sizeof(map[i]) * length);
951 mip->map_blocks += map[i].br_blockcount;
955 mip->map_valid += mip->nmap;
959 * No valid mappings, so no more data blocks.
961 if (!mip->map_valid) {
962 *curoff = xfs_dir2_da_to_byte(mp, mip->map_off);
967 * Read the directory block starting at the first mapping.
969 mip->curdb = xfs_dir2_da_to_db(mp, map->br_startoff);
970 error = xfs_dir2_data_read(NULL, dp, map->br_startoff,
971 map->br_blockcount >= mp->m_dirblkfsbs ?
972 XFS_FSB_TO_DADDR(mp, map->br_startblock) : -1, &bp);
975 * Should just skip over the data block instead of giving up.
981 * Adjust the current amount of read-ahead: we just read a block that
985 mip->ra_current -= mp->m_dirblkfsbs;
988 * Do we need more readahead?
990 for (mip->ra_index = mip->ra_offset = i = 0;
991 mip->ra_want > mip->ra_current && i < mip->map_blocks;
992 i += mp->m_dirblkfsbs) {
993 ASSERT(mip->ra_index < mip->map_valid);
995 * Read-ahead a contiguous directory block.
997 if (i > mip->ra_current &&
998 map[mip->ra_index].br_blockcount >= mp->m_dirblkfsbs) {
999 xfs_dir2_data_readahead(NULL, dp,
1000 map[mip->ra_index].br_startoff + mip->ra_offset,
1001 XFS_FSB_TO_DADDR(mp,
1002 map[mip->ra_index].br_startblock +
1004 mip->ra_current = i;
1008 * Read-ahead a non-contiguous directory block. This doesn't
1009 * use our mapping, but this is a very rare case.
1011 else if (i > mip->ra_current) {
1012 xfs_dir2_data_readahead(NULL, dp,
1013 map[mip->ra_index].br_startoff +
1014 mip->ra_offset, -1);
1015 mip->ra_current = i;
1019 * Advance offset through the mapping table.
1021 for (j = 0; j < mp->m_dirblkfsbs; j++) {
1023 * The rest of this extent but not more than a dir
1026 length = min_t(int, mp->m_dirblkfsbs,
1027 map[mip->ra_index].br_blockcount -
1030 mip->ra_offset += length;
1033 * Advance to the next mapping if this one is used up.
1035 if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
1048 * Getdents (readdir) for leaf and node directories.
1049 * This reads the data blocks only, so is the same for both forms.
1052 xfs_dir2_leaf_getdents(
1053 xfs_inode_t *dp, /* incore directory inode */
1059 struct xfs_buf *bp = NULL; /* data block buffer */
1060 xfs_dir2_data_hdr_t *hdr; /* data block header */
1061 xfs_dir2_data_entry_t *dep; /* data entry */
1062 xfs_dir2_data_unused_t *dup; /* unused entry */
1063 int error = 0; /* error return value */
1064 int length; /* temporary length value */
1065 xfs_mount_t *mp; /* filesystem mount point */
1066 int byteoff; /* offset in current block */
1067 xfs_dir2_off_t curoff; /* current overall offset */
1068 xfs_dir2_off_t newoff; /* new curoff after new blk */
1069 char *ptr = NULL; /* pointer to current data */
1070 struct xfs_dir2_leaf_map_info *map_info;
1073 * If the offset is at or past the largest allowed value,
1074 * give up right away.
1076 if (*offset >= XFS_DIR2_MAX_DATAPTR)
1082 * Set up to bmap a number of blocks based on the caller's
1083 * buffer size, the directory block size, and the filesystem
1086 length = howmany(bufsize + mp->m_dirblksize,
1087 mp->m_sb.sb_blocksize);
1088 map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
1089 (length * sizeof(struct xfs_bmbt_irec)),
1091 map_info->map_size = length;
1094 * Inside the loop we keep the main offset value as a byte offset
1095 * in the directory file.
1097 curoff = xfs_dir2_dataptr_to_byte(mp, *offset);
1100 * Force this conversion through db so we truncate the offset
1101 * down to get the start of the data block.
1103 map_info->map_off = xfs_dir2_db_to_da(mp,
1104 xfs_dir2_byte_to_db(mp, curoff));
1107 * Loop over directory entries until we reach the end offset.
1108 * Get more blocks and readahead as necessary.
1110 while (curoff < XFS_DIR2_LEAF_OFFSET) {
1112 * If we have no buffer, or we're off the end of the
1113 * current buffer, need to get another one.
1115 if (!bp || ptr >= (char *)bp->b_addr + mp->m_dirblksize) {
1117 error = xfs_dir2_leaf_readbuf(dp, bufsize, map_info,
1119 if (error || !map_info->map_valid)
1123 * Having done a read, we need to set a new offset.
1125 newoff = xfs_dir2_db_off_to_byte(mp, map_info->curdb, 0);
1127 * Start of the current block.
1129 if (curoff < newoff)
1132 * Make sure we're in the right block.
1134 else if (curoff > newoff)
1135 ASSERT(xfs_dir2_byte_to_db(mp, curoff) ==
1138 xfs_dir2_data_check(dp, bp);
1140 * Find our position in the block.
1142 ptr = (char *)(hdr + 1);
1143 byteoff = xfs_dir2_byte_to_off(mp, curoff);
1145 * Skip past the header.
1148 curoff += (uint)sizeof(*hdr);
1150 * Skip past entries until we reach our offset.
1153 while ((char *)ptr - (char *)hdr < byteoff) {
1154 dup = (xfs_dir2_data_unused_t *)ptr;
1156 if (be16_to_cpu(dup->freetag)
1157 == XFS_DIR2_DATA_FREE_TAG) {
1159 length = be16_to_cpu(dup->length);
1163 dep = (xfs_dir2_data_entry_t *)ptr;
1165 xfs_dir2_data_entsize(dep->namelen);
1169 * Now set our real offset.
1172 xfs_dir2_db_off_to_byte(mp,
1173 xfs_dir2_byte_to_db(mp, curoff),
1174 (char *)ptr - (char *)hdr);
1175 if (ptr >= (char *)hdr + mp->m_dirblksize) {
1181 * We have a pointer to an entry.
1184 dup = (xfs_dir2_data_unused_t *)ptr;
1186 * No, it's unused, skip over it.
1188 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
1189 length = be16_to_cpu(dup->length);
1195 dep = (xfs_dir2_data_entry_t *)ptr;
1196 length = xfs_dir2_data_entsize(dep->namelen);
1198 if (filldir(dirent, (char *)dep->name, dep->namelen,
1199 xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff,
1200 be64_to_cpu(dep->inumber), DT_UNKNOWN))
1204 * Advance to next entry in the block.
1208 /* bufsize may have just been a guess; don't go negative */
1209 bufsize = bufsize > length ? bufsize - length : 0;
1213 * All done. Set output offset value to current offset.
1215 if (curoff > xfs_dir2_dataptr_to_byte(mp, XFS_DIR2_MAX_DATAPTR))
1216 *offset = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
1218 *offset = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
1219 kmem_free(map_info);
1221 xfs_trans_brelse(NULL, bp);
1226 * Initialize a new leaf block, leaf1 or leafn magic accepted.
1230 xfs_da_args_t *args, /* operation arguments */
1231 xfs_dir2_db_t bno, /* directory block number */
1232 struct xfs_buf **bpp, /* out: leaf buffer */
1233 int magic) /* magic number for block */
1235 struct xfs_buf *bp; /* leaf buffer */
1236 xfs_inode_t *dp; /* incore directory inode */
1237 int error; /* error return code */
1238 xfs_dir2_leaf_t *leaf; /* leaf structure */
1239 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1240 xfs_mount_t *mp; /* filesystem mount point */
1241 xfs_trans_t *tp; /* transaction pointer */
1247 ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
1248 bno < XFS_DIR2_FREE_FIRSTDB(mp));
1250 * Get the buffer for the block.
1252 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
1258 * Initialize the header.
1261 leaf->hdr.info.magic = cpu_to_be16(magic);
1262 leaf->hdr.info.forw = 0;
1263 leaf->hdr.info.back = 0;
1264 leaf->hdr.count = 0;
1265 leaf->hdr.stale = 0;
1266 xfs_dir2_leaf_log_header(tp, bp);
1268 * If it's a leaf-format directory initialize the tail.
1269 * In this case our caller has the real bests table to copy into
1272 if (magic == XFS_DIR2_LEAF1_MAGIC) {
1273 bp->b_ops = &xfs_dir2_leaf1_buf_ops;
1274 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1276 xfs_dir2_leaf_log_tail(tp, bp);
1278 bp->b_ops = &xfs_dir2_leafn_buf_ops;
1284 * Log the bests entries indicated from a leaf1 block.
1287 xfs_dir2_leaf_log_bests(
1288 xfs_trans_t *tp, /* transaction pointer */
1289 struct xfs_buf *bp, /* leaf buffer */
1290 int first, /* first entry to log */
1291 int last) /* last entry to log */
1293 __be16 *firstb; /* pointer to first entry */
1294 __be16 *lastb; /* pointer to last entry */
1295 xfs_dir2_leaf_t *leaf; /* leaf structure */
1296 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1299 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1300 ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1301 firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1302 lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1303 xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1304 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1308 * Log the leaf entries indicated from a leaf1 or leafn block.
1311 xfs_dir2_leaf_log_ents(
1312 xfs_trans_t *tp, /* transaction pointer */
1313 struct xfs_buf *bp, /* leaf buffer */
1314 int first, /* first entry to log */
1315 int last) /* last entry to log */
1317 xfs_dir2_leaf_entry_t *firstlep; /* pointer to first entry */
1318 xfs_dir2_leaf_entry_t *lastlep; /* pointer to last entry */
1319 xfs_dir2_leaf_t *leaf; /* leaf structure */
1322 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1323 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1324 firstlep = &leaf->ents[first];
1325 lastlep = &leaf->ents[last];
1326 xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1327 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1331 * Log the header of the leaf1 or leafn block.
1334 xfs_dir2_leaf_log_header(
1335 struct xfs_trans *tp,
1338 xfs_dir2_leaf_t *leaf; /* leaf structure */
1341 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1342 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1343 xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1344 (uint)(sizeof(leaf->hdr) - 1));
1348 * Log the tail of the leaf1 block.
1351 xfs_dir2_leaf_log_tail(
1352 struct xfs_trans *tp,
1355 xfs_dir2_leaf_t *leaf; /* leaf structure */
1356 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1357 xfs_mount_t *mp; /* filesystem mount point */
1361 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
1362 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1363 xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1364 (uint)(mp->m_dirblksize - 1));
1368 * Look up the entry referred to by args in the leaf format directory.
1369 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1370 * is also used by the node-format code.
1373 xfs_dir2_leaf_lookup(
1374 xfs_da_args_t *args) /* operation arguments */
1376 struct xfs_buf *dbp; /* data block buffer */
1377 xfs_dir2_data_entry_t *dep; /* data block entry */
1378 xfs_inode_t *dp; /* incore directory inode */
1379 int error; /* error return code */
1380 int index; /* found entry index */
1381 struct xfs_buf *lbp; /* leaf buffer */
1382 xfs_dir2_leaf_t *leaf; /* leaf structure */
1383 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1384 xfs_trans_t *tp; /* transaction pointer */
1386 trace_xfs_dir2_leaf_lookup(args);
1389 * Look up name in the leaf block, returning both buffers and index.
1391 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1396 xfs_dir2_leaf_check(dp, lbp);
1399 * Get to the leaf entry and contained data entry address.
1401 lep = &leaf->ents[index];
1403 * Point to the data entry.
1405 dep = (xfs_dir2_data_entry_t *)
1406 ((char *)dbp->b_addr +
1407 xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1409 * Return the found inode number & CI name if appropriate
1411 args->inumber = be64_to_cpu(dep->inumber);
1412 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1413 xfs_trans_brelse(tp, dbp);
1414 xfs_trans_brelse(tp, lbp);
1415 return XFS_ERROR(error);
1419 * Look up name/hash in the leaf block.
1420 * Fill in indexp with the found index, and dbpp with the data buffer.
1421 * If not found dbpp will be NULL, and ENOENT comes back.
1422 * lbpp will always be filled in with the leaf buffer unless there's an error.
1424 static int /* error */
1425 xfs_dir2_leaf_lookup_int(
1426 xfs_da_args_t *args, /* operation arguments */
1427 struct xfs_buf **lbpp, /* out: leaf buffer */
1428 int *indexp, /* out: index in leaf block */
1429 struct xfs_buf **dbpp) /* out: data buffer */
1431 xfs_dir2_db_t curdb = -1; /* current data block number */
1432 struct xfs_buf *dbp = NULL; /* data buffer */
1433 xfs_dir2_data_entry_t *dep; /* data entry */
1434 xfs_inode_t *dp; /* incore directory inode */
1435 int error; /* error return code */
1436 int index; /* index in leaf block */
1437 struct xfs_buf *lbp; /* leaf buffer */
1438 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1439 xfs_dir2_leaf_t *leaf; /* leaf structure */
1440 xfs_mount_t *mp; /* filesystem mount point */
1441 xfs_dir2_db_t newdb; /* new data block number */
1442 xfs_trans_t *tp; /* transaction pointer */
1443 xfs_dir2_db_t cidb = -1; /* case match data block no. */
1444 enum xfs_dacmp cmp; /* name compare result */
1450 error = xfs_dir2_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
1456 xfs_dir2_leaf_check(dp, lbp);
1458 * Look for the first leaf entry with our hash value.
1460 index = xfs_dir2_leaf_search_hash(args, lbp);
1462 * Loop over all the entries with the right hash value
1463 * looking to match the name.
1465 for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
1466 be32_to_cpu(lep->hashval) == args->hashval;
1469 * Skip over stale leaf entries.
1471 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1474 * Get the new data block number.
1476 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1478 * If it's not the same as the old data block number,
1479 * need to pitch the old one and read the new one.
1481 if (newdb != curdb) {
1483 xfs_trans_brelse(tp, dbp);
1484 error = xfs_dir2_data_read(tp, dp,
1485 xfs_dir2_db_to_da(mp, newdb),
1488 xfs_trans_brelse(tp, lbp);
1494 * Point to the data entry.
1496 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1497 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1499 * Compare name and if it's an exact match, return the index
1500 * and buffer. If it's the first case-insensitive match, store
1501 * the index and buffer and continue looking for an exact match.
1503 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1504 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1505 args->cmpresult = cmp;
1507 /* case exact match: return the current buffer. */
1508 if (cmp == XFS_CMP_EXACT) {
1515 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1517 * Here, we can only be doing a lookup (not a rename or remove).
1518 * If a case-insensitive match was found earlier, re-read the
1519 * appropriate data block if required and return it.
1521 if (args->cmpresult == XFS_CMP_CASE) {
1523 if (cidb != curdb) {
1524 xfs_trans_brelse(tp, dbp);
1525 error = xfs_dir2_data_read(tp, dp,
1526 xfs_dir2_db_to_da(mp, cidb),
1529 xfs_trans_brelse(tp, lbp);
1537 * No match found, return ENOENT.
1541 xfs_trans_brelse(tp, dbp);
1542 xfs_trans_brelse(tp, lbp);
1543 return XFS_ERROR(ENOENT);
1547 * Remove an entry from a leaf format directory.
1550 xfs_dir2_leaf_removename(
1551 xfs_da_args_t *args) /* operation arguments */
1553 __be16 *bestsp; /* leaf block best freespace */
1554 xfs_dir2_data_hdr_t *hdr; /* data block header */
1555 xfs_dir2_db_t db; /* data block number */
1556 struct xfs_buf *dbp; /* data block buffer */
1557 xfs_dir2_data_entry_t *dep; /* data entry structure */
1558 xfs_inode_t *dp; /* incore directory inode */
1559 int error; /* error return code */
1560 xfs_dir2_db_t i; /* temporary data block # */
1561 int index; /* index into leaf entries */
1562 struct xfs_buf *lbp; /* leaf buffer */
1563 xfs_dir2_leaf_t *leaf; /* leaf structure */
1564 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1565 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1566 xfs_mount_t *mp; /* filesystem mount point */
1567 int needlog; /* need to log data header */
1568 int needscan; /* need to rescan data frees */
1569 xfs_dir2_data_off_t oldbest; /* old value of best free */
1570 xfs_trans_t *tp; /* transaction pointer */
1572 trace_xfs_dir2_leaf_removename(args);
1575 * Lookup the leaf entry, get the leaf and data blocks read in.
1577 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1585 xfs_dir2_data_check(dp, dbp);
1587 * Point to the leaf entry, use that to point to the data entry.
1589 lep = &leaf->ents[index];
1590 db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1591 dep = (xfs_dir2_data_entry_t *)
1592 ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1593 needscan = needlog = 0;
1594 oldbest = be16_to_cpu(hdr->bestfree[0].length);
1595 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1596 bestsp = xfs_dir2_leaf_bests_p(ltp);
1597 ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1599 * Mark the former data entry unused.
1601 xfs_dir2_data_make_free(tp, dbp,
1602 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1603 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1605 * We just mark the leaf entry stale by putting a null in it.
1607 be16_add_cpu(&leaf->hdr.stale, 1);
1608 xfs_dir2_leaf_log_header(tp, lbp);
1609 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1610 xfs_dir2_leaf_log_ents(tp, lbp, index, index);
1612 * Scan the freespace in the data block again if necessary,
1613 * log the data block header if necessary.
1616 xfs_dir2_data_freescan(mp, hdr, &needlog);
1618 xfs_dir2_data_log_header(tp, dbp);
1620 * If the longest freespace in the data block has changed,
1621 * put the new value in the bests table and log that.
1623 if (be16_to_cpu(hdr->bestfree[0].length) != oldbest) {
1624 bestsp[db] = hdr->bestfree[0].length;
1625 xfs_dir2_leaf_log_bests(tp, lbp, db, db);
1627 xfs_dir2_data_check(dp, dbp);
1629 * If the data block is now empty then get rid of the data block.
1631 if (be16_to_cpu(hdr->bestfree[0].length) ==
1632 mp->m_dirblksize - (uint)sizeof(*hdr)) {
1633 ASSERT(db != mp->m_dirdatablk);
1634 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1636 * Nope, can't get rid of it because it caused
1637 * allocation of a bmap btree block to do so.
1638 * Just go on, returning success, leaving the
1639 * empty block in place.
1641 if (error == ENOSPC && args->total == 0)
1643 xfs_dir2_leaf_check(dp, lbp);
1648 * If this is the last data block then compact the
1649 * bests table by getting rid of entries.
1651 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1653 * Look for the last active entry (i).
1655 for (i = db - 1; i > 0; i--) {
1656 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1660 * Copy the table down so inactive entries at the
1663 memmove(&bestsp[db - i], bestsp,
1664 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1665 be32_add_cpu(<p->bestcount, -(db - i));
1666 xfs_dir2_leaf_log_tail(tp, lbp);
1667 xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1669 bestsp[db] = cpu_to_be16(NULLDATAOFF);
1672 * If the data block was not the first one, drop it.
1674 else if (db != mp->m_dirdatablk)
1677 xfs_dir2_leaf_check(dp, lbp);
1679 * See if we can convert to block form.
1681 return xfs_dir2_leaf_to_block(args, lbp, dbp);
1685 * Replace the inode number in a leaf format directory entry.
1688 xfs_dir2_leaf_replace(
1689 xfs_da_args_t *args) /* operation arguments */
1691 struct xfs_buf *dbp; /* data block buffer */
1692 xfs_dir2_data_entry_t *dep; /* data block entry */
1693 xfs_inode_t *dp; /* incore directory inode */
1694 int error; /* error return code */
1695 int index; /* index of leaf entry */
1696 struct xfs_buf *lbp; /* leaf buffer */
1697 xfs_dir2_leaf_t *leaf; /* leaf structure */
1698 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1699 xfs_trans_t *tp; /* transaction pointer */
1701 trace_xfs_dir2_leaf_replace(args);
1704 * Look up the entry.
1706 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1712 * Point to the leaf entry, get data address from it.
1714 lep = &leaf->ents[index];
1716 * Point to the data entry.
1718 dep = (xfs_dir2_data_entry_t *)
1719 ((char *)dbp->b_addr +
1720 xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1721 ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1723 * Put the new inode number in, log it.
1725 dep->inumber = cpu_to_be64(args->inumber);
1727 xfs_dir2_data_log_entry(tp, dbp, dep);
1728 xfs_dir2_leaf_check(dp, lbp);
1729 xfs_trans_brelse(tp, lbp);
1734 * Return index in the leaf block (lbp) which is either the first
1735 * one with this hash value, or if there are none, the insert point
1736 * for that hash value.
1738 int /* index value */
1739 xfs_dir2_leaf_search_hash(
1740 xfs_da_args_t *args, /* operation arguments */
1741 struct xfs_buf *lbp) /* leaf buffer */
1743 xfs_dahash_t hash=0; /* hash from this entry */
1744 xfs_dahash_t hashwant; /* hash value looking for */
1745 int high; /* high leaf index */
1746 int low; /* low leaf index */
1747 xfs_dir2_leaf_t *leaf; /* leaf structure */
1748 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1749 int mid=0; /* current leaf index */
1753 if (!leaf->hdr.count)
1757 * Note, the table cannot be empty, so we have to go through the loop.
1758 * Binary search the leaf entries looking for our hash value.
1760 for (lep = leaf->ents, low = 0, high = be16_to_cpu(leaf->hdr.count) - 1,
1761 hashwant = args->hashval;
1763 mid = (low + high) >> 1;
1764 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1766 if (hash < hashwant)
1772 * Found one, back up through all the equal hash values.
1774 if (hash == hashwant) {
1775 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1780 * Need to point to an entry higher than ours.
1782 else if (hash < hashwant)
1788 * Trim off a trailing data block. We know it's empty since the leaf
1789 * freespace table says so.
1792 xfs_dir2_leaf_trim_data(
1793 xfs_da_args_t *args, /* operation arguments */
1794 struct xfs_buf *lbp, /* leaf buffer */
1795 xfs_dir2_db_t db) /* data block number */
1797 __be16 *bestsp; /* leaf bests table */
1798 struct xfs_buf *dbp; /* data block buffer */
1799 xfs_inode_t *dp; /* incore directory inode */
1800 int error; /* error return value */
1801 xfs_dir2_leaf_t *leaf; /* leaf structure */
1802 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1803 xfs_mount_t *mp; /* filesystem mount point */
1804 xfs_trans_t *tp; /* transaction pointer */
1810 * Read the offending data block. We need its buffer.
1812 error = xfs_dir2_data_read(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp);
1817 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1821 struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1823 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
1824 ASSERT(be16_to_cpu(hdr->bestfree[0].length) ==
1825 mp->m_dirblksize - (uint)sizeof(*hdr));
1826 ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1831 * Get rid of the data block.
1833 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1834 ASSERT(error != ENOSPC);
1835 xfs_trans_brelse(tp, dbp);
1839 * Eliminate the last bests entry from the table.
1841 bestsp = xfs_dir2_leaf_bests_p(ltp);
1842 be32_add_cpu(<p->bestcount, -1);
1843 memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1844 xfs_dir2_leaf_log_tail(tp, lbp);
1845 xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1849 static inline size_t
1851 struct xfs_dir2_leaf_hdr *hdr,
1856 entries = be16_to_cpu(hdr->count) - be16_to_cpu(hdr->stale);
1857 return sizeof(xfs_dir2_leaf_hdr_t) +
1858 entries * sizeof(xfs_dir2_leaf_entry_t) +
1859 counts * sizeof(xfs_dir2_data_off_t) +
1860 sizeof(xfs_dir2_leaf_tail_t);
1864 * Convert node form directory to leaf form directory.
1865 * The root of the node form dir needs to already be a LEAFN block.
1866 * Just return if we can't do anything.
1869 xfs_dir2_node_to_leaf(
1870 xfs_da_state_t *state) /* directory operation state */
1872 xfs_da_args_t *args; /* operation arguments */
1873 xfs_inode_t *dp; /* incore directory inode */
1874 int error; /* error return code */
1875 struct xfs_buf *fbp; /* buffer for freespace block */
1876 xfs_fileoff_t fo; /* freespace file offset */
1877 xfs_dir2_free_t *free; /* freespace structure */
1878 struct xfs_buf *lbp; /* buffer for leaf block */
1879 xfs_dir2_leaf_tail_t *ltp; /* tail of leaf structure */
1880 xfs_dir2_leaf_t *leaf; /* leaf structure */
1881 xfs_mount_t *mp; /* filesystem mount point */
1882 int rval; /* successful free trim? */
1883 xfs_trans_t *tp; /* transaction pointer */
1886 * There's more than a leaf level in the btree, so there must
1887 * be multiple leafn blocks. Give up.
1889 if (state->path.active > 1)
1893 trace_xfs_dir2_node_to_leaf(args);
1899 * Get the last offset in the file.
1901 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1904 fo -= mp->m_dirblkfsbs;
1906 * If there are freespace blocks other than the first one,
1907 * take this opportunity to remove trailing empty freespace blocks
1908 * that may have been left behind during no-space-reservation
1911 while (fo > mp->m_dirfreeblk) {
1912 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1916 fo -= mp->m_dirblkfsbs;
1921 * Now find the block just before the freespace block.
1923 if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1927 * If it's not the single leaf block, give up.
1929 if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1931 lbp = state->path.blk[0].bp;
1933 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1935 * Read the freespace block.
1937 error = xfs_dir2_free_read(tp, dp, mp->m_dirfreeblk, &fbp);
1941 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1942 ASSERT(!free->hdr.firstdb);
1945 * Now see if the leafn and free data will fit in a leaf1.
1946 * If not, release the buffer and give up.
1948 if (xfs_dir2_leaf_size(&leaf->hdr, be32_to_cpu(free->hdr.nvalid)) >
1950 xfs_trans_brelse(tp, fbp);
1955 * If the leaf has any stale entries in it, compress them out.
1956 * The compact routine will log the header.
1958 if (be16_to_cpu(leaf->hdr.stale))
1959 xfs_dir2_leaf_compact(args, lbp);
1961 xfs_dir2_leaf_log_header(tp, lbp);
1963 lbp->b_ops = &xfs_dir2_leaf1_buf_ops;
1964 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAF1_MAGIC);
1967 * Set up the leaf tail from the freespace block.
1969 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1970 ltp->bestcount = free->hdr.nvalid;
1972 * Set up the leaf bests table.
1974 memcpy(xfs_dir2_leaf_bests_p(ltp), free->bests,
1975 be32_to_cpu(ltp->bestcount) * sizeof(xfs_dir2_data_off_t));
1976 xfs_dir2_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1977 xfs_dir2_leaf_log_tail(tp, lbp);
1978 xfs_dir2_leaf_check(dp, lbp);
1980 * Get rid of the freespace block.
1982 error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1985 * This can't fail here because it can only happen when
1986 * punching out the middle of an extent, and this is an
1989 ASSERT(error != ENOSPC);
1994 * Now see if we can convert the single-leaf directory
1995 * down to a block form directory.
1996 * This routine always kills the dabuf for the leaf, so
1997 * eliminate it from the path.
1999 error = xfs_dir2_leaf_to_block(args, lbp, NULL);
2000 state->path.blk[0].bp = NULL;