2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
32 #include "xfs_dir2_priv.h"
33 #include "xfs_error.h"
34 #include "xfs_trace.h"
35 #include "xfs_trans.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_cksum.h"
40 * Local function declarations.
42 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
43 int *indexp, struct xfs_buf **dbpp);
44 static void xfs_dir3_leaf_log_bests(struct xfs_trans *tp, struct xfs_buf *bp,
46 static void xfs_dir3_leaf_log_tail(struct xfs_trans *tp, struct xfs_buf *bp);
49 * Check the internal consistency of a leaf1 block.
50 * Pop an assert if something is wrong.
53 #define xfs_dir3_leaf_check(dp, bp) \
55 if (!xfs_dir3_leaf1_check((dp), (bp))) \
64 struct xfs_dir2_leaf *leaf = bp->b_addr;
65 struct xfs_dir3_icleaf_hdr leafhdr;
67 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
69 if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
70 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
71 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
73 } else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
76 return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
79 #define xfs_dir3_leaf_check(dp, bp)
83 xfs_dir3_leaf_check_int(
86 struct xfs_dir3_icleaf_hdr *hdr,
87 struct xfs_dir2_leaf *leaf)
89 struct xfs_dir2_leaf_entry *ents;
90 xfs_dir2_leaf_tail_t *ltp;
93 const struct xfs_dir_ops *ops;
94 struct xfs_dir3_icleaf_hdr leafhdr;
97 * we can be passed a null dp here from a verifier, so we need to go the
98 * hard way to get them.
100 ops = xfs_dir_get_ops(mp, dp);
103 ops->leaf_hdr_from_disk(&leafhdr, leaf);
107 ents = ops->leaf_ents_p(leaf);
108 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
111 * XXX (dgc): This value is not restrictive enough.
112 * Should factor in the size of the bests table as well.
113 * We can deduce a value for that from di_size.
115 if (hdr->count > ops->leaf_max_ents(mp))
118 /* Leaves and bests don't overlap in leaf format. */
119 if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
120 hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
121 (char *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
124 /* Check hash value order, count stale entries. */
125 for (i = stale = 0; i < hdr->count; i++) {
126 if (i + 1 < hdr->count) {
127 if (be32_to_cpu(ents[i].hashval) >
128 be32_to_cpu(ents[i + 1].hashval))
131 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
134 if (hdr->stale != stale)
140 * We verify the magic numbers before decoding the leaf header so that on debug
141 * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
142 * to incorrect magic numbers.
145 xfs_dir3_leaf_verify(
149 struct xfs_mount *mp = bp->b_target->bt_mount;
150 struct xfs_dir2_leaf *leaf = bp->b_addr;
152 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
154 if (xfs_sb_version_hascrc(&mp->m_sb)) {
155 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
158 magic3 = (magic == XFS_DIR2_LEAF1_MAGIC) ? XFS_DIR3_LEAF1_MAGIC
159 : XFS_DIR3_LEAFN_MAGIC;
161 if (leaf3->info.hdr.magic != cpu_to_be16(magic3))
163 if (!uuid_equal(&leaf3->info.uuid, &mp->m_sb.sb_uuid))
165 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
168 if (leaf->hdr.info.magic != cpu_to_be16(magic))
172 return xfs_dir3_leaf_check_int(mp, NULL, NULL, leaf);
180 struct xfs_mount *mp = bp->b_target->bt_mount;
182 if (xfs_sb_version_hascrc(&mp->m_sb) &&
183 !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
184 xfs_buf_ioerror(bp, EFSBADCRC);
185 else if (!xfs_dir3_leaf_verify(bp, magic))
186 xfs_buf_ioerror(bp, EFSCORRUPTED);
189 xfs_verifier_error(bp);
197 struct xfs_mount *mp = bp->b_target->bt_mount;
198 struct xfs_buf_log_item *bip = bp->b_fspriv;
199 struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
201 if (!xfs_dir3_leaf_verify(bp, magic)) {
202 xfs_buf_ioerror(bp, EFSCORRUPTED);
203 xfs_verifier_error(bp);
207 if (!xfs_sb_version_hascrc(&mp->m_sb))
211 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
213 xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
217 xfs_dir3_leaf1_read_verify(
220 __read_verify(bp, XFS_DIR2_LEAF1_MAGIC);
224 xfs_dir3_leaf1_write_verify(
227 __write_verify(bp, XFS_DIR2_LEAF1_MAGIC);
231 xfs_dir3_leafn_read_verify(
234 __read_verify(bp, XFS_DIR2_LEAFN_MAGIC);
238 xfs_dir3_leafn_write_verify(
241 __write_verify(bp, XFS_DIR2_LEAFN_MAGIC);
244 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
245 .verify_read = xfs_dir3_leaf1_read_verify,
246 .verify_write = xfs_dir3_leaf1_write_verify,
249 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
250 .verify_read = xfs_dir3_leafn_read_verify,
251 .verify_write = xfs_dir3_leafn_write_verify,
256 struct xfs_trans *tp,
257 struct xfs_inode *dp,
259 xfs_daddr_t mappedbno,
260 struct xfs_buf **bpp)
264 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
265 XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
267 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
273 struct xfs_trans *tp,
274 struct xfs_inode *dp,
276 xfs_daddr_t mappedbno,
277 struct xfs_buf **bpp)
281 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
282 XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
284 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
289 * Initialize a new leaf block, leaf1 or leafn magic accepted.
293 struct xfs_mount *mp,
294 struct xfs_trans *tp,
299 struct xfs_dir2_leaf *leaf = bp->b_addr;
301 ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
303 if (xfs_sb_version_hascrc(&mp->m_sb)) {
304 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
306 memset(leaf3, 0, sizeof(*leaf3));
308 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
309 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
310 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
311 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
312 leaf3->info.owner = cpu_to_be64(owner);
313 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_uuid);
315 memset(leaf, 0, sizeof(*leaf));
316 leaf->hdr.info.magic = cpu_to_be16(type);
320 * If it's a leaf-format directory initialize the tail.
321 * Caller is responsible for initialising the bests table.
323 if (type == XFS_DIR2_LEAF1_MAGIC) {
324 struct xfs_dir2_leaf_tail *ltp;
326 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
328 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
329 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
331 bp->b_ops = &xfs_dir3_leafn_buf_ops;
332 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
337 xfs_dir3_leaf_get_buf(
340 struct xfs_buf **bpp,
343 struct xfs_inode *dp = args->dp;
344 struct xfs_trans *tp = args->trans;
345 struct xfs_mount *mp = dp->i_mount;
349 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
350 ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
351 bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
353 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
354 -1, &bp, XFS_DATA_FORK);
358 xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
359 xfs_dir3_leaf_log_header(tp, dp, bp);
360 if (magic == XFS_DIR2_LEAF1_MAGIC)
361 xfs_dir3_leaf_log_tail(tp, bp);
367 * Convert a block form directory to a leaf form directory.
370 xfs_dir2_block_to_leaf(
371 xfs_da_args_t *args, /* operation arguments */
372 struct xfs_buf *dbp) /* input block's buffer */
374 __be16 *bestsp; /* leaf's bestsp entries */
375 xfs_dablk_t blkno; /* leaf block's bno */
376 xfs_dir2_data_hdr_t *hdr; /* block header */
377 xfs_dir2_leaf_entry_t *blp; /* block's leaf entries */
378 xfs_dir2_block_tail_t *btp; /* block's tail */
379 xfs_inode_t *dp; /* incore directory inode */
380 int error; /* error return code */
381 struct xfs_buf *lbp; /* leaf block's buffer */
382 xfs_dir2_db_t ldb; /* leaf block's bno */
383 xfs_dir2_leaf_t *leaf; /* leaf structure */
384 xfs_dir2_leaf_tail_t *ltp; /* leaf's tail */
385 xfs_mount_t *mp; /* filesystem mount point */
386 int needlog; /* need to log block header */
387 int needscan; /* need to rescan bestfree */
388 xfs_trans_t *tp; /* transaction pointer */
389 struct xfs_dir2_data_free *bf;
390 struct xfs_dir2_leaf_entry *ents;
391 struct xfs_dir3_icleaf_hdr leafhdr;
393 trace_xfs_dir2_block_to_leaf(args);
399 * Add the leaf block to the inode.
400 * This interface will only put blocks in the leaf/node range.
401 * Since that's empty now, we'll get the root (block 0 in range).
403 if ((error = xfs_da_grow_inode(args, &blkno))) {
406 ldb = xfs_dir2_da_to_db(args->geo, blkno);
407 ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
409 * Initialize the leaf block, get a buffer for it.
411 error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
417 xfs_dir3_data_check(dp, dbp);
418 btp = xfs_dir2_block_tail_p(mp, hdr);
419 blp = xfs_dir2_block_leaf_p(btp);
420 bf = dp->d_ops->data_bestfree_p(hdr);
421 ents = dp->d_ops->leaf_ents_p(leaf);
424 * Set the counts in the leaf header.
426 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
427 leafhdr.count = be32_to_cpu(btp->count);
428 leafhdr.stale = be32_to_cpu(btp->stale);
429 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
430 xfs_dir3_leaf_log_header(tp, dp, lbp);
433 * Could compact these but I think we always do the conversion
434 * after squeezing out stale entries.
436 memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
437 xfs_dir3_leaf_log_ents(tp, dp, lbp, 0, leafhdr.count - 1);
441 * Make the space formerly occupied by the leaf entries and block
444 xfs_dir2_data_make_free(tp, dp, dbp,
445 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
446 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
448 &needlog, &needscan);
450 * Fix up the block header, make it a data block.
452 dbp->b_ops = &xfs_dir3_data_buf_ops;
453 xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
454 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
455 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
457 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
460 xfs_dir2_data_freescan(dp, hdr, &needlog);
462 * Set up leaf tail and bests table.
464 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
465 ltp->bestcount = cpu_to_be32(1);
466 bestsp = xfs_dir2_leaf_bests_p(ltp);
467 bestsp[0] = bf[0].length;
469 * Log the data header and leaf bests table.
472 xfs_dir2_data_log_header(tp, dp, dbp);
473 xfs_dir3_leaf_check(dp, lbp);
474 xfs_dir3_data_check(dp, dbp);
475 xfs_dir3_leaf_log_bests(tp, lbp, 0, 0);
480 xfs_dir3_leaf_find_stale(
481 struct xfs_dir3_icleaf_hdr *leafhdr,
482 struct xfs_dir2_leaf_entry *ents,
488 * Find the first stale entry before our index, if any.
490 for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
491 if (ents[*lowstale].address ==
492 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
497 * Find the first stale entry at or after our index, if any.
498 * Stop if the result would require moving more entries than using
501 for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
502 if (ents[*highstale].address ==
503 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
505 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
510 struct xfs_dir2_leaf_entry *
511 xfs_dir3_leaf_find_entry(
512 struct xfs_dir3_icleaf_hdr *leafhdr,
513 struct xfs_dir2_leaf_entry *ents,
514 int index, /* leaf table position */
515 int compact, /* need to compact leaves */
516 int lowstale, /* index of prev stale leaf */
517 int highstale, /* index of next stale leaf */
518 int *lfloglow, /* low leaf logging index */
519 int *lfloghigh) /* high leaf logging index */
521 if (!leafhdr->stale) {
522 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
525 * Now we need to make room to insert the leaf entry.
527 * If there are no stale entries, just insert a hole at index.
530 if (index < leafhdr->count)
531 memmove(lep + 1, lep,
532 (leafhdr->count - index) * sizeof(*lep));
535 * Record low and high logging indices for the leaf.
538 *lfloghigh = leafhdr->count++;
543 * There are stale entries.
545 * We will use one of them for the new entry. It's probably not at
546 * the right location, so we'll have to shift some up or down first.
548 * If we didn't compact before, we need to find the nearest stale
549 * entries before and after our insertion point.
552 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
553 &lowstale, &highstale);
556 * If the low one is better, use it.
559 (highstale == leafhdr->count ||
560 index - lowstale - 1 < highstale - index)) {
561 ASSERT(index - lowstale - 1 >= 0);
562 ASSERT(ents[lowstale].address ==
563 cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
566 * Copy entries up to cover the stale entry and make room
569 if (index - lowstale - 1 > 0) {
570 memmove(&ents[lowstale], &ents[lowstale + 1],
571 (index - lowstale - 1) *
572 sizeof(xfs_dir2_leaf_entry_t));
574 *lfloglow = MIN(lowstale, *lfloglow);
575 *lfloghigh = MAX(index - 1, *lfloghigh);
577 return &ents[index - 1];
581 * The high one is better, so use that one.
583 ASSERT(highstale - index >= 0);
584 ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
587 * Copy entries down to cover the stale entry and make room for the
590 if (highstale - index > 0) {
591 memmove(&ents[index + 1], &ents[index],
592 (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
594 *lfloglow = MIN(index, *lfloglow);
595 *lfloghigh = MAX(highstale, *lfloghigh);
601 * Add an entry to a leaf form directory.
604 xfs_dir2_leaf_addname(
605 xfs_da_args_t *args) /* operation arguments */
607 __be16 *bestsp; /* freespace table in leaf */
608 int compact; /* need to compact leaves */
609 xfs_dir2_data_hdr_t *hdr; /* data block header */
610 struct xfs_buf *dbp; /* data block buffer */
611 xfs_dir2_data_entry_t *dep; /* data block entry */
612 xfs_inode_t *dp; /* incore directory inode */
613 xfs_dir2_data_unused_t *dup; /* data unused entry */
614 int error; /* error return value */
615 int grown; /* allocated new data block */
616 int highstale; /* index of next stale leaf */
617 int i; /* temporary, index */
618 int index; /* leaf table position */
619 struct xfs_buf *lbp; /* leaf's buffer */
620 xfs_dir2_leaf_t *leaf; /* leaf structure */
621 int length; /* length of new entry */
622 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
623 int lfloglow; /* low leaf logging index */
624 int lfloghigh; /* high leaf logging index */
625 int lowstale; /* index of prev stale leaf */
626 xfs_dir2_leaf_tail_t *ltp; /* leaf tail pointer */
627 xfs_mount_t *mp; /* filesystem mount point */
628 int needbytes; /* leaf block bytes needed */
629 int needlog; /* need to log data header */
630 int needscan; /* need to rescan data free */
631 __be16 *tagp; /* end of data entry */
632 xfs_trans_t *tp; /* transaction pointer */
633 xfs_dir2_db_t use_block; /* data block number */
634 struct xfs_dir2_data_free *bf; /* bestfree table */
635 struct xfs_dir2_leaf_entry *ents;
636 struct xfs_dir3_icleaf_hdr leafhdr;
638 trace_xfs_dir2_leaf_addname(args);
644 error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
649 * Look up the entry by hash value and name.
650 * We know it's not there, our caller has already done a lookup.
651 * So the index is of the entry to insert in front of.
652 * But if there are dup hash values the index is of the first of those.
654 index = xfs_dir2_leaf_search_hash(args, lbp);
656 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
657 ents = dp->d_ops->leaf_ents_p(leaf);
658 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
659 bestsp = xfs_dir2_leaf_bests_p(ltp);
660 length = dp->d_ops->data_entsize(args->namelen);
663 * See if there are any entries with the same hash value
664 * and space in their block for the new entry.
665 * This is good because it puts multiple same-hash value entries
666 * in a data block, improving the lookup of those entries.
668 for (use_block = -1, lep = &ents[index];
669 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
671 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
673 i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
674 ASSERT(i < be32_to_cpu(ltp->bestcount));
675 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
676 if (be16_to_cpu(bestsp[i]) >= length) {
682 * Didn't find a block yet, linear search all the data blocks.
684 if (use_block == -1) {
685 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
687 * Remember a block we see that's missing.
689 if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
692 else if (be16_to_cpu(bestsp[i]) >= length) {
699 * How many bytes do we need in the leaf block?
703 needbytes += sizeof(xfs_dir2_leaf_entry_t);
705 needbytes += sizeof(xfs_dir2_data_off_t);
708 * Now kill use_block if it refers to a missing block, so we
709 * can use it as an indication of allocation needed.
711 if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
714 * If we don't have enough free bytes but we can make enough
715 * by compacting out stale entries, we'll do that.
717 if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
722 * Otherwise if we don't have enough free bytes we need to
723 * convert to node form.
725 else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
727 * Just checking or no space reservation, give up.
729 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
731 xfs_trans_brelse(tp, lbp);
732 return XFS_ERROR(ENOSPC);
735 * Convert to node form.
737 error = xfs_dir2_leaf_to_node(args, lbp);
741 * Then add the new entry.
743 return xfs_dir2_node_addname(args);
746 * Otherwise it will fit without compaction.
751 * If just checking, then it will fit unless we needed to allocate
754 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
755 xfs_trans_brelse(tp, lbp);
756 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
759 * If no allocations are allowed, return now before we've
762 if (args->total == 0 && use_block == -1) {
763 xfs_trans_brelse(tp, lbp);
764 return XFS_ERROR(ENOSPC);
767 * Need to compact the leaf entries, removing stale ones.
768 * Leave one stale entry behind - the one closest to our
769 * insertion index - and we'll shift that one to our insertion
773 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
774 &highstale, &lfloglow, &lfloghigh);
777 * There are stale entries, so we'll need log-low and log-high
778 * impossibly bad values later.
780 else if (leafhdr.stale) {
781 lfloglow = leafhdr.count;
785 * If there was no data block space found, we need to allocate
788 if (use_block == -1) {
790 * Add the new data block.
792 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
794 xfs_trans_brelse(tp, lbp);
798 * Initialize the block.
800 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
801 xfs_trans_brelse(tp, lbp);
805 * If we're adding a new data block on the end we need to
806 * extend the bests table. Copy it up one entry.
808 if (use_block >= be32_to_cpu(ltp->bestcount)) {
810 memmove(&bestsp[0], &bestsp[1],
811 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
812 be32_add_cpu(<p->bestcount, 1);
813 xfs_dir3_leaf_log_tail(tp, lbp);
814 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
817 * If we're filling in a previously empty block just log it.
820 xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
822 bf = dp->d_ops->data_bestfree_p(hdr);
823 bestsp[use_block] = bf[0].length;
827 * Already had space in some data block.
828 * Just read that one in.
830 error = xfs_dir3_data_read(tp, dp,
831 xfs_dir2_db_to_da(args->geo, use_block),
834 xfs_trans_brelse(tp, lbp);
838 bf = dp->d_ops->data_bestfree_p(hdr);
842 * Point to the biggest freespace in our data block.
844 dup = (xfs_dir2_data_unused_t *)
845 ((char *)hdr + be16_to_cpu(bf[0].offset));
846 ASSERT(be16_to_cpu(dup->length) >= length);
847 needscan = needlog = 0;
849 * Mark the initial part of our freespace in use for the new entry.
851 xfs_dir2_data_use_free(tp, dp, dbp, dup,
852 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
853 &needlog, &needscan);
855 * Initialize our new entry (at last).
857 dep = (xfs_dir2_data_entry_t *)dup;
858 dep->inumber = cpu_to_be64(args->inumber);
859 dep->namelen = args->namelen;
860 memcpy(dep->name, args->name, dep->namelen);
861 dp->d_ops->data_put_ftype(dep, args->filetype);
862 tagp = dp->d_ops->data_entry_tag_p(dep);
863 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
865 * Need to scan fix up the bestfree table.
868 xfs_dir2_data_freescan(dp, hdr, &needlog);
870 * Need to log the data block's header.
873 xfs_dir2_data_log_header(tp, dp, dbp);
874 xfs_dir2_data_log_entry(tp, dp, dbp, dep);
876 * If the bests table needs to be changed, do it.
877 * Log the change unless we've already done that.
879 if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
880 bestsp[use_block] = bf[0].length;
882 xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
885 lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
886 highstale, &lfloglow, &lfloghigh);
889 * Fill in the new leaf entry.
891 lep->hashval = cpu_to_be32(args->hashval);
892 lep->address = cpu_to_be32(
893 xfs_dir2_db_off_to_dataptr(args->geo, use_block,
894 be16_to_cpu(*tagp)));
896 * Log the leaf fields and give up the buffers.
898 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
899 xfs_dir3_leaf_log_header(tp, dp, lbp);
900 xfs_dir3_leaf_log_ents(tp, dp, lbp, lfloglow, lfloghigh);
901 xfs_dir3_leaf_check(dp, lbp);
902 xfs_dir3_data_check(dp, dbp);
907 * Compact out any stale entries in the leaf.
908 * Log the header and changed leaf entries, if any.
911 xfs_dir3_leaf_compact(
912 xfs_da_args_t *args, /* operation arguments */
913 struct xfs_dir3_icleaf_hdr *leafhdr,
914 struct xfs_buf *bp) /* leaf buffer */
916 int from; /* source leaf index */
917 xfs_dir2_leaf_t *leaf; /* leaf structure */
918 int loglow; /* first leaf entry to log */
919 int to; /* target leaf index */
920 struct xfs_dir2_leaf_entry *ents;
921 struct xfs_inode *dp = args->dp;
928 * Compress out the stale entries in place.
930 ents = dp->d_ops->leaf_ents_p(leaf);
931 for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
932 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
935 * Only actually copy the entries that are different.
940 ents[to] = ents[from];
945 * Update and log the header, log the leaf entries.
947 ASSERT(leafhdr->stale == from - to);
948 leafhdr->count -= leafhdr->stale;
951 dp->d_ops->leaf_hdr_to_disk(leaf, leafhdr);
952 xfs_dir3_leaf_log_header(args->trans, dp, bp);
954 xfs_dir3_leaf_log_ents(args->trans, dp, bp, loglow, to - 1);
958 * Compact the leaf entries, removing stale ones.
959 * Leave one stale entry behind - the one closest to our
960 * insertion index - and the caller will shift that one to our insertion
962 * Return new insertion index, where the remaining stale entry is,
963 * and leaf logging indices.
966 xfs_dir3_leaf_compact_x1(
967 struct xfs_dir3_icleaf_hdr *leafhdr,
968 struct xfs_dir2_leaf_entry *ents,
969 int *indexp, /* insertion index */
970 int *lowstalep, /* out: stale entry before us */
971 int *highstalep, /* out: stale entry after us */
972 int *lowlogp, /* out: low log index */
973 int *highlogp) /* out: high log index */
975 int from; /* source copy index */
976 int highstale; /* stale entry at/after index */
977 int index; /* insertion index */
978 int keepstale; /* source index of kept stale */
979 int lowstale; /* stale entry before index */
980 int newindex=0; /* new insertion index */
981 int to; /* destination copy index */
983 ASSERT(leafhdr->stale > 1);
986 xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
989 * Pick the better of lowstale and highstale.
992 (highstale == leafhdr->count ||
993 index - lowstale <= highstale - index))
994 keepstale = lowstale;
996 keepstale = highstale;
998 * Copy the entries in place, removing all the stale entries
1001 for (from = to = 0; from < leafhdr->count; from++) {
1003 * Notice the new value of index.
1007 if (from != keepstale &&
1008 ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1014 * Record the new keepstale value for the insertion.
1016 if (from == keepstale)
1017 lowstale = highstale = to;
1019 * Copy only the entries that have moved.
1022 ents[to] = ents[from];
1027 * If the insertion point was past the last entry,
1028 * set the new insertion point accordingly.
1034 * Adjust the leaf header values.
1036 leafhdr->count -= from - to;
1039 * Remember the low/high stale value only in the "right"
1042 if (lowstale >= newindex)
1045 highstale = leafhdr->count;
1046 *highlogp = leafhdr->count - 1;
1047 *lowstalep = lowstale;
1048 *highstalep = highstale;
1052 * Log the bests entries indicated from a leaf1 block.
1055 xfs_dir3_leaf_log_bests(
1056 xfs_trans_t *tp, /* transaction pointer */
1057 struct xfs_buf *bp, /* leaf buffer */
1058 int first, /* first entry to log */
1059 int last) /* last entry to log */
1061 __be16 *firstb; /* pointer to first entry */
1062 __be16 *lastb; /* pointer to last entry */
1063 struct xfs_dir2_leaf *leaf = bp->b_addr;
1064 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1066 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1067 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1069 ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1070 firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1071 lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1072 xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1073 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1077 * Log the leaf entries indicated from a leaf1 or leafn block.
1080 xfs_dir3_leaf_log_ents(
1081 struct xfs_trans *tp,
1082 struct xfs_inode *dp,
1087 xfs_dir2_leaf_entry_t *firstlep; /* pointer to first entry */
1088 xfs_dir2_leaf_entry_t *lastlep; /* pointer to last entry */
1089 struct xfs_dir2_leaf *leaf = bp->b_addr;
1090 struct xfs_dir2_leaf_entry *ents;
1092 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1093 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1094 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1095 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1097 ents = dp->d_ops->leaf_ents_p(leaf);
1098 firstlep = &ents[first];
1099 lastlep = &ents[last];
1100 xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1101 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1105 * Log the header of the leaf1 or leafn block.
1108 xfs_dir3_leaf_log_header(
1109 struct xfs_trans *tp,
1110 struct xfs_inode *dp,
1113 struct xfs_dir2_leaf *leaf = bp->b_addr;
1115 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1116 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1117 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1118 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1120 xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1121 dp->d_ops->leaf_hdr_size - 1);
1125 * Log the tail of the leaf1 block.
1128 xfs_dir3_leaf_log_tail(
1129 struct xfs_trans *tp,
1132 struct xfs_dir2_leaf *leaf = bp->b_addr;
1133 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1134 struct xfs_mount *mp = tp->t_mountp;
1136 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1137 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1138 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1139 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1141 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1142 xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1143 (uint)(mp->m_dirblksize - 1));
1147 * Look up the entry referred to by args in the leaf format directory.
1148 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1149 * is also used by the node-format code.
1152 xfs_dir2_leaf_lookup(
1153 xfs_da_args_t *args) /* operation arguments */
1155 struct xfs_buf *dbp; /* data block buffer */
1156 xfs_dir2_data_entry_t *dep; /* data block entry */
1157 xfs_inode_t *dp; /* incore directory inode */
1158 int error; /* error return code */
1159 int index; /* found entry index */
1160 struct xfs_buf *lbp; /* leaf buffer */
1161 xfs_dir2_leaf_t *leaf; /* leaf structure */
1162 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1163 xfs_trans_t *tp; /* transaction pointer */
1164 struct xfs_dir2_leaf_entry *ents;
1166 trace_xfs_dir2_leaf_lookup(args);
1169 * Look up name in the leaf block, returning both buffers and index.
1171 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1176 xfs_dir3_leaf_check(dp, lbp);
1178 ents = dp->d_ops->leaf_ents_p(leaf);
1180 * Get to the leaf entry and contained data entry address.
1185 * Point to the data entry.
1187 dep = (xfs_dir2_data_entry_t *)
1188 ((char *)dbp->b_addr +
1189 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1191 * Return the found inode number & CI name if appropriate
1193 args->inumber = be64_to_cpu(dep->inumber);
1194 args->filetype = dp->d_ops->data_get_ftype(dep);
1195 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1196 xfs_trans_brelse(tp, dbp);
1197 xfs_trans_brelse(tp, lbp);
1198 return XFS_ERROR(error);
1202 * Look up name/hash in the leaf block.
1203 * Fill in indexp with the found index, and dbpp with the data buffer.
1204 * If not found dbpp will be NULL, and ENOENT comes back.
1205 * lbpp will always be filled in with the leaf buffer unless there's an error.
1207 static int /* error */
1208 xfs_dir2_leaf_lookup_int(
1209 xfs_da_args_t *args, /* operation arguments */
1210 struct xfs_buf **lbpp, /* out: leaf buffer */
1211 int *indexp, /* out: index in leaf block */
1212 struct xfs_buf **dbpp) /* out: data buffer */
1214 xfs_dir2_db_t curdb = -1; /* current data block number */
1215 struct xfs_buf *dbp = NULL; /* data buffer */
1216 xfs_dir2_data_entry_t *dep; /* data entry */
1217 xfs_inode_t *dp; /* incore directory inode */
1218 int error; /* error return code */
1219 int index; /* index in leaf block */
1220 struct xfs_buf *lbp; /* leaf buffer */
1221 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1222 xfs_dir2_leaf_t *leaf; /* leaf structure */
1223 xfs_mount_t *mp; /* filesystem mount point */
1224 xfs_dir2_db_t newdb; /* new data block number */
1225 xfs_trans_t *tp; /* transaction pointer */
1226 xfs_dir2_db_t cidb = -1; /* case match data block no. */
1227 enum xfs_dacmp cmp; /* name compare result */
1228 struct xfs_dir2_leaf_entry *ents;
1229 struct xfs_dir3_icleaf_hdr leafhdr;
1235 error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
1241 xfs_dir3_leaf_check(dp, lbp);
1242 ents = dp->d_ops->leaf_ents_p(leaf);
1243 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1246 * Look for the first leaf entry with our hash value.
1248 index = xfs_dir2_leaf_search_hash(args, lbp);
1250 * Loop over all the entries with the right hash value
1251 * looking to match the name.
1253 for (lep = &ents[index];
1254 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1257 * Skip over stale leaf entries.
1259 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1262 * Get the new data block number.
1264 newdb = xfs_dir2_dataptr_to_db(args->geo,
1265 be32_to_cpu(lep->address));
1267 * If it's not the same as the old data block number,
1268 * need to pitch the old one and read the new one.
1270 if (newdb != curdb) {
1272 xfs_trans_brelse(tp, dbp);
1273 error = xfs_dir3_data_read(tp, dp,
1274 xfs_dir2_db_to_da(args->geo, newdb),
1277 xfs_trans_brelse(tp, lbp);
1283 * Point to the data entry.
1285 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1286 xfs_dir2_dataptr_to_off(args->geo,
1287 be32_to_cpu(lep->address)));
1289 * Compare name and if it's an exact match, return the index
1290 * and buffer. If it's the first case-insensitive match, store
1291 * the index and buffer and continue looking for an exact match.
1293 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1294 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1295 args->cmpresult = cmp;
1297 /* case exact match: return the current buffer. */
1298 if (cmp == XFS_CMP_EXACT) {
1305 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1307 * Here, we can only be doing a lookup (not a rename or remove).
1308 * If a case-insensitive match was found earlier, re-read the
1309 * appropriate data block if required and return it.
1311 if (args->cmpresult == XFS_CMP_CASE) {
1313 if (cidb != curdb) {
1314 xfs_trans_brelse(tp, dbp);
1315 error = xfs_dir3_data_read(tp, dp,
1316 xfs_dir2_db_to_da(args->geo, cidb),
1319 xfs_trans_brelse(tp, lbp);
1327 * No match found, return ENOENT.
1331 xfs_trans_brelse(tp, dbp);
1332 xfs_trans_brelse(tp, lbp);
1333 return XFS_ERROR(ENOENT);
1337 * Remove an entry from a leaf format directory.
1340 xfs_dir2_leaf_removename(
1341 xfs_da_args_t *args) /* operation arguments */
1343 __be16 *bestsp; /* leaf block best freespace */
1344 xfs_dir2_data_hdr_t *hdr; /* data block header */
1345 xfs_dir2_db_t db; /* data block number */
1346 struct xfs_buf *dbp; /* data block buffer */
1347 xfs_dir2_data_entry_t *dep; /* data entry structure */
1348 xfs_inode_t *dp; /* incore directory inode */
1349 int error; /* error return code */
1350 xfs_dir2_db_t i; /* temporary data block # */
1351 int index; /* index into leaf entries */
1352 struct xfs_buf *lbp; /* leaf buffer */
1353 xfs_dir2_leaf_t *leaf; /* leaf structure */
1354 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1355 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1356 xfs_mount_t *mp; /* filesystem mount point */
1357 int needlog; /* need to log data header */
1358 int needscan; /* need to rescan data frees */
1359 xfs_dir2_data_off_t oldbest; /* old value of best free */
1360 xfs_trans_t *tp; /* transaction pointer */
1361 struct xfs_dir2_data_free *bf; /* bestfree table */
1362 struct xfs_dir2_leaf_entry *ents;
1363 struct xfs_dir3_icleaf_hdr leafhdr;
1365 trace_xfs_dir2_leaf_removename(args);
1368 * Lookup the leaf entry, get the leaf and data blocks read in.
1370 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1378 xfs_dir3_data_check(dp, dbp);
1379 bf = dp->d_ops->data_bestfree_p(hdr);
1380 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1381 ents = dp->d_ops->leaf_ents_p(leaf);
1383 * Point to the leaf entry, use that to point to the data entry.
1386 db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1387 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1388 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1389 needscan = needlog = 0;
1390 oldbest = be16_to_cpu(bf[0].length);
1391 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1392 bestsp = xfs_dir2_leaf_bests_p(ltp);
1393 ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1395 * Mark the former data entry unused.
1397 xfs_dir2_data_make_free(tp, dp, dbp,
1398 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1399 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1401 * We just mark the leaf entry stale by putting a null in it.
1404 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1405 xfs_dir3_leaf_log_header(tp, dp, lbp);
1407 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1408 xfs_dir3_leaf_log_ents(tp, dp, lbp, index, index);
1411 * Scan the freespace in the data block again if necessary,
1412 * log the data block header if necessary.
1415 xfs_dir2_data_freescan(dp, hdr, &needlog);
1417 xfs_dir2_data_log_header(tp, dp, dbp);
1419 * If the longest freespace in the data block has changed,
1420 * put the new value in the bests table and log that.
1422 if (be16_to_cpu(bf[0].length) != oldbest) {
1423 bestsp[db] = bf[0].length;
1424 xfs_dir3_leaf_log_bests(tp, lbp, db, db);
1426 xfs_dir3_data_check(dp, dbp);
1428 * If the data block is now empty then get rid of the data block.
1430 if (be16_to_cpu(bf[0].length) ==
1431 mp->m_dirblksize - dp->d_ops->data_entry_offset) {
1432 ASSERT(db != args->geo->datablk);
1433 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1435 * Nope, can't get rid of it because it caused
1436 * allocation of a bmap btree block to do so.
1437 * Just go on, returning success, leaving the
1438 * empty block in place.
1440 if (error == ENOSPC && args->total == 0)
1442 xfs_dir3_leaf_check(dp, lbp);
1447 * If this is the last data block then compact the
1448 * bests table by getting rid of entries.
1450 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1452 * Look for the last active entry (i).
1454 for (i = db - 1; i > 0; i--) {
1455 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1459 * Copy the table down so inactive entries at the
1462 memmove(&bestsp[db - i], bestsp,
1463 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1464 be32_add_cpu(<p->bestcount, -(db - i));
1465 xfs_dir3_leaf_log_tail(tp, lbp);
1466 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1468 bestsp[db] = cpu_to_be16(NULLDATAOFF);
1471 * If the data block was not the first one, drop it.
1473 else if (db != args->geo->datablk)
1476 xfs_dir3_leaf_check(dp, lbp);
1478 * See if we can convert to block form.
1480 return xfs_dir2_leaf_to_block(args, lbp, dbp);
1484 * Replace the inode number in a leaf format directory entry.
1487 xfs_dir2_leaf_replace(
1488 xfs_da_args_t *args) /* operation arguments */
1490 struct xfs_buf *dbp; /* data block buffer */
1491 xfs_dir2_data_entry_t *dep; /* data block entry */
1492 xfs_inode_t *dp; /* incore directory inode */
1493 int error; /* error return code */
1494 int index; /* index of leaf entry */
1495 struct xfs_buf *lbp; /* leaf buffer */
1496 xfs_dir2_leaf_t *leaf; /* leaf structure */
1497 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1498 xfs_trans_t *tp; /* transaction pointer */
1499 struct xfs_dir2_leaf_entry *ents;
1501 trace_xfs_dir2_leaf_replace(args);
1504 * Look up the entry.
1506 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1511 ents = dp->d_ops->leaf_ents_p(leaf);
1513 * Point to the leaf entry, get data address from it.
1517 * Point to the data entry.
1519 dep = (xfs_dir2_data_entry_t *)
1520 ((char *)dbp->b_addr +
1521 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1522 ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1524 * Put the new inode number in, log it.
1526 dep->inumber = cpu_to_be64(args->inumber);
1527 dp->d_ops->data_put_ftype(dep, args->filetype);
1529 xfs_dir2_data_log_entry(tp, dp, dbp, dep);
1530 xfs_dir3_leaf_check(dp, lbp);
1531 xfs_trans_brelse(tp, lbp);
1536 * Return index in the leaf block (lbp) which is either the first
1537 * one with this hash value, or if there are none, the insert point
1538 * for that hash value.
1540 int /* index value */
1541 xfs_dir2_leaf_search_hash(
1542 xfs_da_args_t *args, /* operation arguments */
1543 struct xfs_buf *lbp) /* leaf buffer */
1545 xfs_dahash_t hash=0; /* hash from this entry */
1546 xfs_dahash_t hashwant; /* hash value looking for */
1547 int high; /* high leaf index */
1548 int low; /* low leaf index */
1549 xfs_dir2_leaf_t *leaf; /* leaf structure */
1550 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1551 int mid=0; /* current leaf index */
1552 struct xfs_dir2_leaf_entry *ents;
1553 struct xfs_dir3_icleaf_hdr leafhdr;
1556 ents = args->dp->d_ops->leaf_ents_p(leaf);
1557 args->dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1560 * Note, the table cannot be empty, so we have to go through the loop.
1561 * Binary search the leaf entries looking for our hash value.
1563 for (lep = ents, low = 0, high = leafhdr.count - 1,
1564 hashwant = args->hashval;
1566 mid = (low + high) >> 1;
1567 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1569 if (hash < hashwant)
1575 * Found one, back up through all the equal hash values.
1577 if (hash == hashwant) {
1578 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1583 * Need to point to an entry higher than ours.
1585 else if (hash < hashwant)
1591 * Trim off a trailing data block. We know it's empty since the leaf
1592 * freespace table says so.
1595 xfs_dir2_leaf_trim_data(
1596 xfs_da_args_t *args, /* operation arguments */
1597 struct xfs_buf *lbp, /* leaf buffer */
1598 xfs_dir2_db_t db) /* data block number */
1600 __be16 *bestsp; /* leaf bests table */
1601 struct xfs_buf *dbp; /* data block buffer */
1602 xfs_inode_t *dp; /* incore directory inode */
1603 int error; /* error return value */
1604 xfs_dir2_leaf_t *leaf; /* leaf structure */
1605 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1606 xfs_mount_t *mp; /* filesystem mount point */
1607 xfs_trans_t *tp; /* transaction pointer */
1613 * Read the offending data block. We need its buffer.
1615 error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(args->geo, db),
1621 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1625 struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1626 struct xfs_dir2_data_free *bf = dp->d_ops->data_bestfree_p(hdr);
1628 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1629 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1630 ASSERT(be16_to_cpu(bf[0].length) ==
1631 mp->m_dirblksize - dp->d_ops->data_entry_offset);
1632 ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1637 * Get rid of the data block.
1639 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1640 ASSERT(error != ENOSPC);
1641 xfs_trans_brelse(tp, dbp);
1645 * Eliminate the last bests entry from the table.
1647 bestsp = xfs_dir2_leaf_bests_p(ltp);
1648 be32_add_cpu(<p->bestcount, -1);
1649 memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1650 xfs_dir3_leaf_log_tail(tp, lbp);
1651 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1655 static inline size_t
1657 struct xfs_dir3_icleaf_hdr *hdr,
1663 entries = hdr->count - hdr->stale;
1664 if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1665 hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1666 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1668 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1670 return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1671 + counts * sizeof(xfs_dir2_data_off_t)
1672 + sizeof(xfs_dir2_leaf_tail_t);
1676 * Convert node form directory to leaf form directory.
1677 * The root of the node form dir needs to already be a LEAFN block.
1678 * Just return if we can't do anything.
1681 xfs_dir2_node_to_leaf(
1682 xfs_da_state_t *state) /* directory operation state */
1684 xfs_da_args_t *args; /* operation arguments */
1685 xfs_inode_t *dp; /* incore directory inode */
1686 int error; /* error return code */
1687 struct xfs_buf *fbp; /* buffer for freespace block */
1688 xfs_fileoff_t fo; /* freespace file offset */
1689 xfs_dir2_free_t *free; /* freespace structure */
1690 struct xfs_buf *lbp; /* buffer for leaf block */
1691 xfs_dir2_leaf_tail_t *ltp; /* tail of leaf structure */
1692 xfs_dir2_leaf_t *leaf; /* leaf structure */
1693 xfs_mount_t *mp; /* filesystem mount point */
1694 int rval; /* successful free trim? */
1695 xfs_trans_t *tp; /* transaction pointer */
1696 struct xfs_dir3_icleaf_hdr leafhdr;
1697 struct xfs_dir3_icfree_hdr freehdr;
1700 * There's more than a leaf level in the btree, so there must
1701 * be multiple leafn blocks. Give up.
1703 if (state->path.active > 1)
1707 trace_xfs_dir2_node_to_leaf(args);
1713 * Get the last offset in the file.
1715 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1718 fo -= args->geo->fsbcount;
1720 * If there are freespace blocks other than the first one,
1721 * take this opportunity to remove trailing empty freespace blocks
1722 * that may have been left behind during no-space-reservation
1725 while (fo > args->geo->freeblk) {
1726 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1730 fo -= args->geo->fsbcount;
1735 * Now find the block just before the freespace block.
1737 if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1741 * If it's not the single leaf block, give up.
1743 if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1745 lbp = state->path.blk[0].bp;
1747 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1749 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1750 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1753 * Read the freespace block.
1755 error = xfs_dir2_free_read(tp, dp, args->geo->freeblk, &fbp);
1759 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1761 ASSERT(!freehdr.firstdb);
1764 * Now see if the leafn and free data will fit in a leaf1.
1765 * If not, release the buffer and give up.
1767 if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > mp->m_dirblksize) {
1768 xfs_trans_brelse(tp, fbp);
1773 * If the leaf has any stale entries in it, compress them out.
1776 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1778 lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1779 xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1780 leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1781 ? XFS_DIR2_LEAF1_MAGIC
1782 : XFS_DIR3_LEAF1_MAGIC;
1785 * Set up the leaf tail from the freespace block.
1787 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1788 ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1791 * Set up the leaf bests table.
1793 memcpy(xfs_dir2_leaf_bests_p(ltp), dp->d_ops->free_bests_p(free),
1794 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1796 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1797 xfs_dir3_leaf_log_header(tp, dp, lbp);
1798 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1799 xfs_dir3_leaf_log_tail(tp, lbp);
1800 xfs_dir3_leaf_check(dp, lbp);
1803 * Get rid of the freespace block.
1805 error = xfs_dir2_shrink_inode(args,
1806 xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET),
1810 * This can't fail here because it can only happen when
1811 * punching out the middle of an extent, and this is an
1814 ASSERT(error != ENOSPC);
1819 * Now see if we can convert the single-leaf directory
1820 * down to a block form directory.
1821 * This routine always kills the dabuf for the leaf, so
1822 * eliminate it from the path.
1824 error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1825 state->path.blk[0].bp = NULL;