1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
17 #include "xfs_dir2_priv.h"
18 #include "xfs_error.h"
19 #include "xfs_trace.h"
20 #include "xfs_trans.h"
21 #include "xfs_buf_item.h"
22 #include "xfs_health.h"
25 * Local function declarations.
27 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
28 int *indexp, struct xfs_buf **dbpp,
29 struct xfs_dir3_icleaf_hdr *leafhdr);
30 static void xfs_dir3_leaf_log_bests(struct xfs_da_args *args,
31 struct xfs_buf *bp, int first, int last);
32 static void xfs_dir3_leaf_log_tail(struct xfs_da_args *args,
36 xfs_dir2_leaf_hdr_from_disk(
38 struct xfs_dir3_icleaf_hdr *to,
39 struct xfs_dir2_leaf *from)
41 if (xfs_has_crc(mp)) {
42 struct xfs_dir3_leaf *from3 = (struct xfs_dir3_leaf *)from;
44 to->forw = be32_to_cpu(from3->hdr.info.hdr.forw);
45 to->back = be32_to_cpu(from3->hdr.info.hdr.back);
46 to->magic = be16_to_cpu(from3->hdr.info.hdr.magic);
47 to->count = be16_to_cpu(from3->hdr.count);
48 to->stale = be16_to_cpu(from3->hdr.stale);
49 to->ents = from3->__ents;
51 ASSERT(to->magic == XFS_DIR3_LEAF1_MAGIC ||
52 to->magic == XFS_DIR3_LEAFN_MAGIC);
54 to->forw = be32_to_cpu(from->hdr.info.forw);
55 to->back = be32_to_cpu(from->hdr.info.back);
56 to->magic = be16_to_cpu(from->hdr.info.magic);
57 to->count = be16_to_cpu(from->hdr.count);
58 to->stale = be16_to_cpu(from->hdr.stale);
59 to->ents = from->__ents;
61 ASSERT(to->magic == XFS_DIR2_LEAF1_MAGIC ||
62 to->magic == XFS_DIR2_LEAFN_MAGIC);
67 xfs_dir2_leaf_hdr_to_disk(
69 struct xfs_dir2_leaf *to,
70 struct xfs_dir3_icleaf_hdr *from)
72 if (xfs_has_crc(mp)) {
73 struct xfs_dir3_leaf *to3 = (struct xfs_dir3_leaf *)to;
75 ASSERT(from->magic == XFS_DIR3_LEAF1_MAGIC ||
76 from->magic == XFS_DIR3_LEAFN_MAGIC);
78 to3->hdr.info.hdr.forw = cpu_to_be32(from->forw);
79 to3->hdr.info.hdr.back = cpu_to_be32(from->back);
80 to3->hdr.info.hdr.magic = cpu_to_be16(from->magic);
81 to3->hdr.count = cpu_to_be16(from->count);
82 to3->hdr.stale = cpu_to_be16(from->stale);
84 ASSERT(from->magic == XFS_DIR2_LEAF1_MAGIC ||
85 from->magic == XFS_DIR2_LEAFN_MAGIC);
87 to->hdr.info.forw = cpu_to_be32(from->forw);
88 to->hdr.info.back = cpu_to_be32(from->back);
89 to->hdr.info.magic = cpu_to_be16(from->magic);
90 to->hdr.count = cpu_to_be16(from->count);
91 to->hdr.stale = cpu_to_be16(from->stale);
96 * Check the internal consistency of a leaf1 block.
97 * Pop an assert if something is wrong.
100 static xfs_failaddr_t
101 xfs_dir3_leaf1_check(
102 struct xfs_inode *dp,
105 struct xfs_dir2_leaf *leaf = bp->b_addr;
106 struct xfs_dir3_icleaf_hdr leafhdr;
108 xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
110 if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
111 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
112 if (be64_to_cpu(leaf3->info.blkno) != xfs_buf_daddr(bp))
113 return __this_address;
114 } else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
115 return __this_address;
117 return xfs_dir3_leaf_check_int(dp->i_mount, &leafhdr, leaf, false);
122 struct xfs_inode *dp,
127 fa = xfs_dir3_leaf1_check(dp, bp);
130 xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
131 bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
136 #define xfs_dir3_leaf_check(dp, bp)
140 xfs_dir3_leaf_check_int(
141 struct xfs_mount *mp,
142 struct xfs_dir3_icleaf_hdr *hdr,
143 struct xfs_dir2_leaf *leaf,
144 bool expensive_checking)
146 struct xfs_da_geometry *geo = mp->m_dir_geo;
147 xfs_dir2_leaf_tail_t *ltp;
150 bool isleaf1 = (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
151 hdr->magic == XFS_DIR3_LEAF1_MAGIC);
153 ltp = xfs_dir2_leaf_tail_p(geo, leaf);
156 * XXX (dgc): This value is not restrictive enough.
157 * Should factor in the size of the bests table as well.
158 * We can deduce a value for that from i_disk_size.
160 if (hdr->count > geo->leaf_max_ents)
161 return __this_address;
163 /* Leaves and bests don't overlap in leaf format. */
165 (char *)&hdr->ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
166 return __this_address;
168 if (!expensive_checking)
171 /* Check hash value order, count stale entries. */
172 for (i = stale = 0; i < hdr->count; i++) {
173 if (i + 1 < hdr->count) {
174 if (be32_to_cpu(hdr->ents[i].hashval) >
175 be32_to_cpu(hdr->ents[i + 1].hashval))
176 return __this_address;
178 if (hdr->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
180 if (isleaf1 && xfs_dir2_dataptr_to_db(geo,
181 be32_to_cpu(hdr->ents[i].address)) >=
182 be32_to_cpu(ltp->bestcount))
183 return __this_address;
185 if (hdr->stale != stale)
186 return __this_address;
191 * We verify the magic numbers before decoding the leaf header so that on debug
192 * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
193 * to incorrect magic numbers.
195 static xfs_failaddr_t
196 xfs_dir3_leaf_verify(
199 struct xfs_mount *mp = bp->b_mount;
200 struct xfs_dir3_icleaf_hdr leafhdr;
203 fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
207 xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, bp->b_addr);
208 return xfs_dir3_leaf_check_int(mp, &leafhdr, bp->b_addr, true);
212 xfs_dir3_leaf_read_verify(
215 struct xfs_mount *mp = bp->b_mount;
218 if (xfs_has_crc(mp) &&
219 !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
220 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
222 fa = xfs_dir3_leaf_verify(bp);
224 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
229 xfs_dir3_leaf_write_verify(
232 struct xfs_mount *mp = bp->b_mount;
233 struct xfs_buf_log_item *bip = bp->b_log_item;
234 struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
237 fa = xfs_dir3_leaf_verify(bp);
239 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
243 if (!xfs_has_crc(mp))
247 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
249 xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
252 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
253 .name = "xfs_dir3_leaf1",
254 .magic16 = { cpu_to_be16(XFS_DIR2_LEAF1_MAGIC),
255 cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) },
256 .verify_read = xfs_dir3_leaf_read_verify,
257 .verify_write = xfs_dir3_leaf_write_verify,
258 .verify_struct = xfs_dir3_leaf_verify,
261 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
262 .name = "xfs_dir3_leafn",
263 .magic16 = { cpu_to_be16(XFS_DIR2_LEAFN_MAGIC),
264 cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) },
265 .verify_read = xfs_dir3_leaf_read_verify,
266 .verify_write = xfs_dir3_leaf_write_verify,
267 .verify_struct = xfs_dir3_leaf_verify,
272 struct xfs_trans *tp,
273 struct xfs_inode *dp,
275 struct xfs_buf **bpp)
279 err = xfs_da_read_buf(tp, dp, fbno, 0, bpp, XFS_DATA_FORK,
280 &xfs_dir3_leaf1_buf_ops);
281 if (!err && tp && *bpp)
282 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
288 struct xfs_trans *tp,
289 struct xfs_inode *dp,
291 struct xfs_buf **bpp)
295 err = xfs_da_read_buf(tp, dp, fbno, 0, bpp, XFS_DATA_FORK,
296 &xfs_dir3_leafn_buf_ops);
297 if (!err && tp && *bpp)
298 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
303 * Initialize a new leaf block, leaf1 or leafn magic accepted.
307 struct xfs_mount *mp,
308 struct xfs_trans *tp,
313 struct xfs_dir2_leaf *leaf = bp->b_addr;
315 ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
317 if (xfs_has_crc(mp)) {
318 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
320 memset(leaf3, 0, sizeof(*leaf3));
322 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
323 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
324 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
325 leaf3->info.blkno = cpu_to_be64(xfs_buf_daddr(bp));
326 leaf3->info.owner = cpu_to_be64(owner);
327 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid);
329 memset(leaf, 0, sizeof(*leaf));
330 leaf->hdr.info.magic = cpu_to_be16(type);
334 * If it's a leaf-format directory initialize the tail.
335 * Caller is responsible for initialising the bests table.
337 if (type == XFS_DIR2_LEAF1_MAGIC) {
338 struct xfs_dir2_leaf_tail *ltp;
340 ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
342 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
343 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
345 bp->b_ops = &xfs_dir3_leafn_buf_ops;
346 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
351 xfs_dir3_leaf_get_buf(
354 struct xfs_buf **bpp,
357 struct xfs_inode *dp = args->dp;
358 struct xfs_trans *tp = args->trans;
359 struct xfs_mount *mp = dp->i_mount;
363 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
364 ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
365 bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
367 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
372 xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
373 xfs_dir3_leaf_log_header(args, bp);
374 if (magic == XFS_DIR2_LEAF1_MAGIC)
375 xfs_dir3_leaf_log_tail(args, bp);
381 * Convert a block form directory to a leaf form directory.
384 xfs_dir2_block_to_leaf(
385 xfs_da_args_t *args, /* operation arguments */
386 struct xfs_buf *dbp) /* input block's buffer */
388 __be16 *bestsp; /* leaf's bestsp entries */
389 xfs_dablk_t blkno; /* leaf block's bno */
390 xfs_dir2_data_hdr_t *hdr; /* block header */
391 xfs_dir2_leaf_entry_t *blp; /* block's leaf entries */
392 xfs_dir2_block_tail_t *btp; /* block's tail */
393 xfs_inode_t *dp; /* incore directory inode */
394 int error; /* error return code */
395 struct xfs_buf *lbp; /* leaf block's buffer */
396 xfs_dir2_db_t ldb; /* leaf block's bno */
397 xfs_dir2_leaf_t *leaf; /* leaf structure */
398 xfs_dir2_leaf_tail_t *ltp; /* leaf's tail */
399 int needlog; /* need to log block header */
400 int needscan; /* need to rescan bestfree */
401 xfs_trans_t *tp; /* transaction pointer */
402 struct xfs_dir2_data_free *bf;
403 struct xfs_dir3_icleaf_hdr leafhdr;
405 trace_xfs_dir2_block_to_leaf(args);
410 * Add the leaf block to the inode.
411 * This interface will only put blocks in the leaf/node range.
412 * Since that's empty now, we'll get the root (block 0 in range).
414 if ((error = xfs_da_grow_inode(args, &blkno))) {
417 ldb = xfs_dir2_da_to_db(args->geo, blkno);
418 ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
420 * Initialize the leaf block, get a buffer for it.
422 error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
428 xfs_dir3_data_check(dp, dbp);
429 btp = xfs_dir2_block_tail_p(args->geo, hdr);
430 blp = xfs_dir2_block_leaf_p(btp);
431 bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
434 * Set the counts in the leaf header.
436 xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
437 leafhdr.count = be32_to_cpu(btp->count);
438 leafhdr.stale = be32_to_cpu(btp->stale);
439 xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
440 xfs_dir3_leaf_log_header(args, lbp);
443 * Could compact these but I think we always do the conversion
444 * after squeezing out stale entries.
446 memcpy(leafhdr.ents, blp,
447 be32_to_cpu(btp->count) * sizeof(struct xfs_dir2_leaf_entry));
448 xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, 0, leafhdr.count - 1);
452 * Make the space formerly occupied by the leaf entries and block
455 xfs_dir2_data_make_free(args, dbp,
456 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
457 (xfs_dir2_data_aoff_t)((char *)hdr + args->geo->blksize -
459 &needlog, &needscan);
461 * Fix up the block header, make it a data block.
463 dbp->b_ops = &xfs_dir3_data_buf_ops;
464 xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
465 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
466 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
468 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
471 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
473 * Set up leaf tail and bests table.
475 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
476 ltp->bestcount = cpu_to_be32(1);
477 bestsp = xfs_dir2_leaf_bests_p(ltp);
478 bestsp[0] = bf[0].length;
480 * Log the data header and leaf bests table.
483 xfs_dir2_data_log_header(args, dbp);
484 xfs_dir3_leaf_check(dp, lbp);
485 xfs_dir3_data_check(dp, dbp);
486 xfs_dir3_leaf_log_bests(args, lbp, 0, 0);
491 xfs_dir3_leaf_find_stale(
492 struct xfs_dir3_icleaf_hdr *leafhdr,
493 struct xfs_dir2_leaf_entry *ents,
499 * Find the first stale entry before our index, if any.
501 for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
502 if (ents[*lowstale].address ==
503 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
508 * Find the first stale entry at or after our index, if any.
509 * Stop if the result would require moving more entries than using
512 for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
513 if (ents[*highstale].address ==
514 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
516 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
521 struct xfs_dir2_leaf_entry *
522 xfs_dir3_leaf_find_entry(
523 struct xfs_dir3_icleaf_hdr *leafhdr,
524 struct xfs_dir2_leaf_entry *ents,
525 int index, /* leaf table position */
526 int compact, /* need to compact leaves */
527 int lowstale, /* index of prev stale leaf */
528 int highstale, /* index of next stale leaf */
529 int *lfloglow, /* low leaf logging index */
530 int *lfloghigh) /* high leaf logging index */
532 if (!leafhdr->stale) {
533 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
536 * Now we need to make room to insert the leaf entry.
538 * If there are no stale entries, just insert a hole at index.
541 if (index < leafhdr->count)
542 memmove(lep + 1, lep,
543 (leafhdr->count - index) * sizeof(*lep));
546 * Record low and high logging indices for the leaf.
549 *lfloghigh = leafhdr->count++;
554 * There are stale entries.
556 * We will use one of them for the new entry. It's probably not at
557 * the right location, so we'll have to shift some up or down first.
559 * If we didn't compact before, we need to find the nearest stale
560 * entries before and after our insertion point.
563 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
564 &lowstale, &highstale);
567 * If the low one is better, use it.
570 (highstale == leafhdr->count ||
571 index - lowstale - 1 < highstale - index)) {
572 ASSERT(index - lowstale - 1 >= 0);
573 ASSERT(ents[lowstale].address ==
574 cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
577 * Copy entries up to cover the stale entry and make room
580 if (index - lowstale - 1 > 0) {
581 memmove(&ents[lowstale], &ents[lowstale + 1],
582 (index - lowstale - 1) *
583 sizeof(xfs_dir2_leaf_entry_t));
585 *lfloglow = min(lowstale, *lfloglow);
586 *lfloghigh = max(index - 1, *lfloghigh);
588 return &ents[index - 1];
592 * The high one is better, so use that one.
594 ASSERT(highstale - index >= 0);
595 ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
598 * Copy entries down to cover the stale entry and make room for the
601 if (highstale - index > 0) {
602 memmove(&ents[index + 1], &ents[index],
603 (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
605 *lfloglow = min(index, *lfloglow);
606 *lfloghigh = max(highstale, *lfloghigh);
612 * Add an entry to a leaf form directory.
615 xfs_dir2_leaf_addname(
616 struct xfs_da_args *args) /* operation arguments */
618 struct xfs_dir3_icleaf_hdr leafhdr;
619 struct xfs_trans *tp = args->trans;
620 __be16 *bestsp; /* freespace table in leaf */
621 __be16 *tagp; /* end of data entry */
622 struct xfs_buf *dbp; /* data block buffer */
623 struct xfs_buf *lbp; /* leaf's buffer */
624 struct xfs_dir2_leaf *leaf; /* leaf structure */
625 struct xfs_inode *dp = args->dp; /* incore directory inode */
626 struct xfs_dir2_data_hdr *hdr; /* data block header */
627 struct xfs_dir2_data_entry *dep; /* data block entry */
628 struct xfs_dir2_leaf_entry *lep; /* leaf entry table pointer */
629 struct xfs_dir2_leaf_entry *ents;
630 struct xfs_dir2_data_unused *dup; /* data unused entry */
631 struct xfs_dir2_leaf_tail *ltp; /* leaf tail pointer */
632 struct xfs_dir2_data_free *bf; /* bestfree table */
633 int compact; /* need to compact leaves */
634 int error; /* error return value */
635 int grown; /* allocated new data block */
636 int highstale = 0; /* index of next stale leaf */
637 int i; /* temporary, index */
638 int index; /* leaf table position */
639 int length; /* length of new entry */
640 int lfloglow; /* low leaf logging index */
641 int lfloghigh; /* high leaf logging index */
642 int lowstale = 0; /* index of prev stale leaf */
643 int needbytes; /* leaf block bytes needed */
644 int needlog; /* need to log data header */
645 int needscan; /* need to rescan data free */
646 xfs_dir2_db_t use_block; /* data block number */
648 trace_xfs_dir2_leaf_addname(args);
650 error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, &lbp);
655 * Look up the entry by hash value and name.
656 * We know it's not there, our caller has already done a lookup.
657 * So the index is of the entry to insert in front of.
658 * But if there are dup hash values the index is of the first of those.
660 index = xfs_dir2_leaf_search_hash(args, lbp);
662 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
663 xfs_dir2_leaf_hdr_from_disk(dp->i_mount, &leafhdr, leaf);
665 bestsp = xfs_dir2_leaf_bests_p(ltp);
666 length = xfs_dir2_data_entsize(dp->i_mount, args->namelen);
669 * See if there are any entries with the same hash value
670 * and space in their block for the new entry.
671 * This is good because it puts multiple same-hash value entries
672 * in a data block, improving the lookup of those entries.
674 for (use_block = -1, lep = &ents[index];
675 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
677 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
679 i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
680 ASSERT(i < be32_to_cpu(ltp->bestcount));
681 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
682 if (be16_to_cpu(bestsp[i]) >= length) {
688 * Didn't find a block yet, linear search all the data blocks.
690 if (use_block == -1) {
691 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
693 * Remember a block we see that's missing.
695 if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
698 else if (be16_to_cpu(bestsp[i]) >= length) {
705 * How many bytes do we need in the leaf block?
709 needbytes += sizeof(xfs_dir2_leaf_entry_t);
711 needbytes += sizeof(xfs_dir2_data_off_t);
714 * Now kill use_block if it refers to a missing block, so we
715 * can use it as an indication of allocation needed.
717 if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
720 * If we don't have enough free bytes but we can make enough
721 * by compacting out stale entries, we'll do that.
723 if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
728 * Otherwise if we don't have enough free bytes we need to
729 * convert to node form.
731 else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
733 * Just checking or no space reservation, give up.
735 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
737 xfs_trans_brelse(tp, lbp);
741 * Convert to node form.
743 error = xfs_dir2_leaf_to_node(args, lbp);
747 * Then add the new entry.
749 return xfs_dir2_node_addname(args);
752 * Otherwise it will fit without compaction.
757 * If just checking, then it will fit unless we needed to allocate
760 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
761 xfs_trans_brelse(tp, lbp);
762 return use_block == -1 ? -ENOSPC : 0;
765 * If no allocations are allowed, return now before we've
768 if (args->total == 0 && use_block == -1) {
769 xfs_trans_brelse(tp, lbp);
773 * Need to compact the leaf entries, removing stale ones.
774 * Leave one stale entry behind - the one closest to our
775 * insertion index - and we'll shift that one to our insertion
779 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
780 &highstale, &lfloglow, &lfloghigh);
783 * There are stale entries, so we'll need log-low and log-high
784 * impossibly bad values later.
786 else if (leafhdr.stale) {
787 lfloglow = leafhdr.count;
791 * If there was no data block space found, we need to allocate
794 if (use_block == -1) {
796 * Add the new data block.
798 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
800 xfs_trans_brelse(tp, lbp);
804 * Initialize the block.
806 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
807 xfs_trans_brelse(tp, lbp);
811 * If we're adding a new data block on the end we need to
812 * extend the bests table. Copy it up one entry.
814 if (use_block >= be32_to_cpu(ltp->bestcount)) {
816 memmove(&bestsp[0], &bestsp[1],
817 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
818 be32_add_cpu(<p->bestcount, 1);
819 xfs_dir3_leaf_log_tail(args, lbp);
820 xfs_dir3_leaf_log_bests(args, lbp, 0,
821 be32_to_cpu(ltp->bestcount) - 1);
824 * If we're filling in a previously empty block just log it.
827 xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
829 bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
830 bestsp[use_block] = bf[0].length;
834 * Already had space in some data block.
835 * Just read that one in.
837 error = xfs_dir3_data_read(tp, dp,
838 xfs_dir2_db_to_da(args->geo, use_block),
841 xfs_trans_brelse(tp, lbp);
845 bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
849 * Point to the biggest freespace in our data block.
851 dup = (xfs_dir2_data_unused_t *)
852 ((char *)hdr + be16_to_cpu(bf[0].offset));
853 needscan = needlog = 0;
855 * Mark the initial part of our freespace in use for the new entry.
857 error = xfs_dir2_data_use_free(args, dbp, dup,
858 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
859 length, &needlog, &needscan);
861 xfs_trans_brelse(tp, lbp);
865 * Initialize our new entry (at last).
867 dep = (xfs_dir2_data_entry_t *)dup;
868 dep->inumber = cpu_to_be64(args->inumber);
869 dep->namelen = args->namelen;
870 memcpy(dep->name, args->name, dep->namelen);
871 xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
872 tagp = xfs_dir2_data_entry_tag_p(dp->i_mount, dep);
873 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
875 * Need to scan fix up the bestfree table.
878 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
880 * Need to log the data block's header.
883 xfs_dir2_data_log_header(args, dbp);
884 xfs_dir2_data_log_entry(args, dbp, dep);
886 * If the bests table needs to be changed, do it.
887 * Log the change unless we've already done that.
889 if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
890 bestsp[use_block] = bf[0].length;
892 xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
895 lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
896 highstale, &lfloglow, &lfloghigh);
899 * Fill in the new leaf entry.
901 lep->hashval = cpu_to_be32(args->hashval);
902 lep->address = cpu_to_be32(
903 xfs_dir2_db_off_to_dataptr(args->geo, use_block,
904 be16_to_cpu(*tagp)));
906 * Log the leaf fields and give up the buffers.
908 xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
909 xfs_dir3_leaf_log_header(args, lbp);
910 xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, lfloglow, lfloghigh);
911 xfs_dir3_leaf_check(dp, lbp);
912 xfs_dir3_data_check(dp, dbp);
917 * Compact out any stale entries in the leaf.
918 * Log the header and changed leaf entries, if any.
921 xfs_dir3_leaf_compact(
922 xfs_da_args_t *args, /* operation arguments */
923 struct xfs_dir3_icleaf_hdr *leafhdr,
924 struct xfs_buf *bp) /* leaf buffer */
926 int from; /* source leaf index */
927 xfs_dir2_leaf_t *leaf; /* leaf structure */
928 int loglow; /* first leaf entry to log */
929 int to; /* target leaf index */
930 struct xfs_inode *dp = args->dp;
937 * Compress out the stale entries in place.
939 for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
940 if (leafhdr->ents[from].address ==
941 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
944 * Only actually copy the entries that are different.
949 leafhdr->ents[to] = leafhdr->ents[from];
954 * Update and log the header, log the leaf entries.
956 ASSERT(leafhdr->stale == from - to);
957 leafhdr->count -= leafhdr->stale;
960 xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, leafhdr);
961 xfs_dir3_leaf_log_header(args, bp);
963 xfs_dir3_leaf_log_ents(args, leafhdr, bp, loglow, to - 1);
967 * Compact the leaf entries, removing stale ones.
968 * Leave one stale entry behind - the one closest to our
969 * insertion index - and the caller will shift that one to our insertion
971 * Return new insertion index, where the remaining stale entry is,
972 * and leaf logging indices.
975 xfs_dir3_leaf_compact_x1(
976 struct xfs_dir3_icleaf_hdr *leafhdr,
977 struct xfs_dir2_leaf_entry *ents,
978 int *indexp, /* insertion index */
979 int *lowstalep, /* out: stale entry before us */
980 int *highstalep, /* out: stale entry after us */
981 int *lowlogp, /* out: low log index */
982 int *highlogp) /* out: high log index */
984 int from; /* source copy index */
985 int highstale; /* stale entry at/after index */
986 int index; /* insertion index */
987 int keepstale; /* source index of kept stale */
988 int lowstale; /* stale entry before index */
989 int newindex=0; /* new insertion index */
990 int to; /* destination copy index */
992 ASSERT(leafhdr->stale > 1);
995 xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
998 * Pick the better of lowstale and highstale.
1000 if (lowstale >= 0 &&
1001 (highstale == leafhdr->count ||
1002 index - lowstale <= highstale - index))
1003 keepstale = lowstale;
1005 keepstale = highstale;
1007 * Copy the entries in place, removing all the stale entries
1010 for (from = to = 0; from < leafhdr->count; from++) {
1012 * Notice the new value of index.
1016 if (from != keepstale &&
1017 ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1023 * Record the new keepstale value for the insertion.
1025 if (from == keepstale)
1026 lowstale = highstale = to;
1028 * Copy only the entries that have moved.
1031 ents[to] = ents[from];
1036 * If the insertion point was past the last entry,
1037 * set the new insertion point accordingly.
1043 * Adjust the leaf header values.
1045 leafhdr->count -= from - to;
1048 * Remember the low/high stale value only in the "right"
1051 if (lowstale >= newindex)
1054 highstale = leafhdr->count;
1055 *highlogp = leafhdr->count - 1;
1056 *lowstalep = lowstale;
1057 *highstalep = highstale;
1061 * Log the bests entries indicated from a leaf1 block.
1064 xfs_dir3_leaf_log_bests(
1065 struct xfs_da_args *args,
1066 struct xfs_buf *bp, /* leaf buffer */
1067 int first, /* first entry to log */
1068 int last) /* last entry to log */
1070 __be16 *firstb; /* pointer to first entry */
1071 __be16 *lastb; /* pointer to last entry */
1072 struct xfs_dir2_leaf *leaf = bp->b_addr;
1073 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1075 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1076 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1078 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1079 firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1080 lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1081 xfs_trans_log_buf(args->trans, bp,
1082 (uint)((char *)firstb - (char *)leaf),
1083 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1087 * Log the leaf entries indicated from a leaf1 or leafn block.
1090 xfs_dir3_leaf_log_ents(
1091 struct xfs_da_args *args,
1092 struct xfs_dir3_icleaf_hdr *hdr,
1097 xfs_dir2_leaf_entry_t *firstlep; /* pointer to first entry */
1098 xfs_dir2_leaf_entry_t *lastlep; /* pointer to last entry */
1099 struct xfs_dir2_leaf *leaf = bp->b_addr;
1101 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1102 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1103 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1104 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1106 firstlep = &hdr->ents[first];
1107 lastlep = &hdr->ents[last];
1108 xfs_trans_log_buf(args->trans, bp,
1109 (uint)((char *)firstlep - (char *)leaf),
1110 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1114 * Log the header of the leaf1 or leafn block.
1117 xfs_dir3_leaf_log_header(
1118 struct xfs_da_args *args,
1121 struct xfs_dir2_leaf *leaf = bp->b_addr;
1123 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1124 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1125 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1126 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1128 xfs_trans_log_buf(args->trans, bp,
1129 (uint)((char *)&leaf->hdr - (char *)leaf),
1130 args->geo->leaf_hdr_size - 1);
1134 * Log the tail of the leaf1 block.
1137 xfs_dir3_leaf_log_tail(
1138 struct xfs_da_args *args,
1141 struct xfs_dir2_leaf *leaf = bp->b_addr;
1142 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1144 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1145 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1146 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1147 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1149 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1150 xfs_trans_log_buf(args->trans, bp, (uint)((char *)ltp - (char *)leaf),
1151 (uint)(args->geo->blksize - 1));
1155 * Look up the entry referred to by args in the leaf format directory.
1156 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1157 * is also used by the node-format code.
1160 xfs_dir2_leaf_lookup(
1161 xfs_da_args_t *args) /* operation arguments */
1163 struct xfs_buf *dbp; /* data block buffer */
1164 xfs_dir2_data_entry_t *dep; /* data block entry */
1165 xfs_inode_t *dp; /* incore directory inode */
1166 int error; /* error return code */
1167 int index; /* found entry index */
1168 struct xfs_buf *lbp; /* leaf buffer */
1169 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1170 xfs_trans_t *tp; /* transaction pointer */
1171 struct xfs_dir3_icleaf_hdr leafhdr;
1173 trace_xfs_dir2_leaf_lookup(args);
1176 * Look up name in the leaf block, returning both buffers and index.
1178 error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1184 xfs_dir3_leaf_check(dp, lbp);
1187 * Get to the leaf entry and contained data entry address.
1189 lep = &leafhdr.ents[index];
1192 * Point to the data entry.
1194 dep = (xfs_dir2_data_entry_t *)
1195 ((char *)dbp->b_addr +
1196 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1198 * Return the found inode number & CI name if appropriate
1200 args->inumber = be64_to_cpu(dep->inumber);
1201 args->filetype = xfs_dir2_data_get_ftype(dp->i_mount, dep);
1202 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1203 xfs_trans_brelse(tp, dbp);
1204 xfs_trans_brelse(tp, lbp);
1209 * Look up name/hash in the leaf block.
1210 * Fill in indexp with the found index, and dbpp with the data buffer.
1211 * If not found dbpp will be NULL, and ENOENT comes back.
1212 * lbpp will always be filled in with the leaf buffer unless there's an error.
1214 static int /* error */
1215 xfs_dir2_leaf_lookup_int(
1216 xfs_da_args_t *args, /* operation arguments */
1217 struct xfs_buf **lbpp, /* out: leaf buffer */
1218 int *indexp, /* out: index in leaf block */
1219 struct xfs_buf **dbpp, /* out: data buffer */
1220 struct xfs_dir3_icleaf_hdr *leafhdr)
1222 xfs_dir2_db_t curdb = -1; /* current data block number */
1223 struct xfs_buf *dbp = NULL; /* data buffer */
1224 xfs_dir2_data_entry_t *dep; /* data entry */
1225 xfs_inode_t *dp; /* incore directory inode */
1226 int error; /* error return code */
1227 int index; /* index in leaf block */
1228 struct xfs_buf *lbp; /* leaf buffer */
1229 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1230 xfs_dir2_leaf_t *leaf; /* leaf structure */
1231 xfs_mount_t *mp; /* filesystem mount point */
1232 xfs_dir2_db_t newdb; /* new data block number */
1233 xfs_trans_t *tp; /* transaction pointer */
1234 xfs_dir2_db_t cidb = -1; /* case match data block no. */
1235 enum xfs_dacmp cmp; /* name compare result */
1241 error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, &lbp);
1247 xfs_dir3_leaf_check(dp, lbp);
1248 xfs_dir2_leaf_hdr_from_disk(mp, leafhdr, leaf);
1251 * Look for the first leaf entry with our hash value.
1253 index = xfs_dir2_leaf_search_hash(args, lbp);
1255 * Loop over all the entries with the right hash value
1256 * looking to match the name.
1258 for (lep = &leafhdr->ents[index];
1259 index < leafhdr->count &&
1260 be32_to_cpu(lep->hashval) == args->hashval;
1263 * Skip over stale leaf entries.
1265 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1268 * Get the new data block number.
1270 newdb = xfs_dir2_dataptr_to_db(args->geo,
1271 be32_to_cpu(lep->address));
1273 * If it's not the same as the old data block number,
1274 * need to pitch the old one and read the new one.
1276 if (newdb != curdb) {
1278 xfs_trans_brelse(tp, dbp);
1279 error = xfs_dir3_data_read(tp, dp,
1280 xfs_dir2_db_to_da(args->geo, newdb),
1283 xfs_trans_brelse(tp, lbp);
1289 * Point to the data entry.
1291 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1292 xfs_dir2_dataptr_to_off(args->geo,
1293 be32_to_cpu(lep->address)));
1295 * Compare name and if it's an exact match, return the index
1296 * and buffer. If it's the first case-insensitive match, store
1297 * the index and buffer and continue looking for an exact match.
1299 cmp = xfs_dir2_compname(args, dep->name, dep->namelen);
1300 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1301 args->cmpresult = cmp;
1303 /* case exact match: return the current buffer. */
1304 if (cmp == XFS_CMP_EXACT) {
1311 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1313 * Here, we can only be doing a lookup (not a rename or remove).
1314 * If a case-insensitive match was found earlier, re-read the
1315 * appropriate data block if required and return it.
1317 if (args->cmpresult == XFS_CMP_CASE) {
1319 if (cidb != curdb) {
1320 xfs_trans_brelse(tp, dbp);
1321 error = xfs_dir3_data_read(tp, dp,
1322 xfs_dir2_db_to_da(args->geo, cidb),
1325 xfs_trans_brelse(tp, lbp);
1333 * No match found, return -ENOENT.
1337 xfs_trans_brelse(tp, dbp);
1338 xfs_trans_brelse(tp, lbp);
1343 * Remove an entry from a leaf format directory.
1346 xfs_dir2_leaf_removename(
1347 xfs_da_args_t *args) /* operation arguments */
1349 struct xfs_da_geometry *geo = args->geo;
1350 __be16 *bestsp; /* leaf block best freespace */
1351 xfs_dir2_data_hdr_t *hdr; /* data block header */
1352 xfs_dir2_db_t db; /* data block number */
1353 struct xfs_buf *dbp; /* data block buffer */
1354 xfs_dir2_data_entry_t *dep; /* data entry structure */
1355 xfs_inode_t *dp; /* incore directory inode */
1356 int error; /* error return code */
1357 xfs_dir2_db_t i; /* temporary data block # */
1358 int index; /* index into leaf entries */
1359 struct xfs_buf *lbp; /* leaf buffer */
1360 xfs_dir2_leaf_t *leaf; /* leaf structure */
1361 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1362 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1363 int needlog; /* need to log data header */
1364 int needscan; /* need to rescan data frees */
1365 xfs_dir2_data_off_t oldbest; /* old value of best free */
1366 struct xfs_dir2_data_free *bf; /* bestfree table */
1367 struct xfs_dir3_icleaf_hdr leafhdr;
1369 trace_xfs_dir2_leaf_removename(args);
1372 * Lookup the leaf entry, get the leaf and data blocks read in.
1374 error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1381 xfs_dir3_data_check(dp, dbp);
1382 bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1385 * Point to the leaf entry, use that to point to the data entry.
1387 lep = &leafhdr.ents[index];
1388 db = xfs_dir2_dataptr_to_db(geo, be32_to_cpu(lep->address));
1389 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1390 xfs_dir2_dataptr_to_off(geo, be32_to_cpu(lep->address)));
1391 needscan = needlog = 0;
1392 oldbest = be16_to_cpu(bf[0].length);
1393 ltp = xfs_dir2_leaf_tail_p(geo, leaf);
1394 bestsp = xfs_dir2_leaf_bests_p(ltp);
1395 if (be16_to_cpu(bestsp[db]) != oldbest) {
1396 xfs_buf_mark_corrupt(lbp);
1397 xfs_da_mark_sick(args);
1398 return -EFSCORRUPTED;
1402 * Mark the former data entry unused.
1404 xfs_dir2_data_make_free(args, dbp,
1405 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1406 xfs_dir2_data_entsize(dp->i_mount, dep->namelen), &needlog,
1409 * We just mark the leaf entry stale by putting a null in it.
1412 xfs_dir2_leaf_hdr_to_disk(dp->i_mount, leaf, &leafhdr);
1413 xfs_dir3_leaf_log_header(args, lbp);
1415 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1416 xfs_dir3_leaf_log_ents(args, &leafhdr, lbp, index, index);
1419 * Scan the freespace in the data block again if necessary,
1420 * log the data block header if necessary.
1423 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
1425 xfs_dir2_data_log_header(args, dbp);
1427 * If the longest freespace in the data block has changed,
1428 * put the new value in the bests table and log that.
1430 if (be16_to_cpu(bf[0].length) != oldbest) {
1431 bestsp[db] = bf[0].length;
1432 xfs_dir3_leaf_log_bests(args, lbp, db, db);
1434 xfs_dir3_data_check(dp, dbp);
1436 * If the data block is now empty then get rid of the data block.
1438 if (be16_to_cpu(bf[0].length) ==
1439 geo->blksize - geo->data_entry_offset) {
1440 ASSERT(db != geo->datablk);
1441 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1443 * Nope, can't get rid of it because it caused
1444 * allocation of a bmap btree block to do so.
1445 * Just go on, returning success, leaving the
1446 * empty block in place.
1448 if (error == -ENOSPC && args->total == 0)
1450 xfs_dir3_leaf_check(dp, lbp);
1455 * If this is the last data block then compact the
1456 * bests table by getting rid of entries.
1458 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1460 * Look for the last active entry (i).
1462 for (i = db - 1; i > 0; i--) {
1463 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1467 * Copy the table down so inactive entries at the
1470 memmove(&bestsp[db - i], bestsp,
1471 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1472 be32_add_cpu(<p->bestcount, -(db - i));
1473 xfs_dir3_leaf_log_tail(args, lbp);
1474 xfs_dir3_leaf_log_bests(args, lbp, 0,
1475 be32_to_cpu(ltp->bestcount) - 1);
1477 bestsp[db] = cpu_to_be16(NULLDATAOFF);
1480 * If the data block was not the first one, drop it.
1482 else if (db != geo->datablk)
1485 xfs_dir3_leaf_check(dp, lbp);
1487 * See if we can convert to block form.
1489 return xfs_dir2_leaf_to_block(args, lbp, dbp);
1493 * Replace the inode number in a leaf format directory entry.
1496 xfs_dir2_leaf_replace(
1497 xfs_da_args_t *args) /* operation arguments */
1499 struct xfs_buf *dbp; /* data block buffer */
1500 xfs_dir2_data_entry_t *dep; /* data block entry */
1501 xfs_inode_t *dp; /* incore directory inode */
1502 int error; /* error return code */
1503 int index; /* index of leaf entry */
1504 struct xfs_buf *lbp; /* leaf buffer */
1505 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1506 xfs_trans_t *tp; /* transaction pointer */
1507 struct xfs_dir3_icleaf_hdr leafhdr;
1509 trace_xfs_dir2_leaf_replace(args);
1512 * Look up the entry.
1514 error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp, &leafhdr);
1520 * Point to the leaf entry, get data address from it.
1522 lep = &leafhdr.ents[index];
1524 * Point to the data entry.
1526 dep = (xfs_dir2_data_entry_t *)
1527 ((char *)dbp->b_addr +
1528 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1529 ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1531 * Put the new inode number in, log it.
1533 dep->inumber = cpu_to_be64(args->inumber);
1534 xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
1536 xfs_dir2_data_log_entry(args, dbp, dep);
1537 xfs_dir3_leaf_check(dp, lbp);
1538 xfs_trans_brelse(tp, lbp);
1543 * Return index in the leaf block (lbp) which is either the first
1544 * one with this hash value, or if there are none, the insert point
1545 * for that hash value.
1547 int /* index value */
1548 xfs_dir2_leaf_search_hash(
1549 xfs_da_args_t *args, /* operation arguments */
1550 struct xfs_buf *lbp) /* leaf buffer */
1552 xfs_dahash_t hash=0; /* hash from this entry */
1553 xfs_dahash_t hashwant; /* hash value looking for */
1554 int high; /* high leaf index */
1555 int low; /* low leaf index */
1556 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1557 int mid=0; /* current leaf index */
1558 struct xfs_dir3_icleaf_hdr leafhdr;
1560 xfs_dir2_leaf_hdr_from_disk(args->dp->i_mount, &leafhdr, lbp->b_addr);
1563 * Note, the table cannot be empty, so we have to go through the loop.
1564 * Binary search the leaf entries looking for our hash value.
1566 for (lep = leafhdr.ents, low = 0, high = leafhdr.count - 1,
1567 hashwant = args->hashval;
1569 mid = (low + high) >> 1;
1570 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1572 if (hash < hashwant)
1578 * Found one, back up through all the equal hash values.
1580 if (hash == hashwant) {
1581 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1586 * Need to point to an entry higher than ours.
1588 else if (hash < hashwant)
1594 * Trim off a trailing data block. We know it's empty since the leaf
1595 * freespace table says so.
1598 xfs_dir2_leaf_trim_data(
1599 xfs_da_args_t *args, /* operation arguments */
1600 struct xfs_buf *lbp, /* leaf buffer */
1601 xfs_dir2_db_t db) /* data block number */
1603 struct xfs_da_geometry *geo = args->geo;
1604 __be16 *bestsp; /* leaf bests table */
1605 struct xfs_buf *dbp; /* data block buffer */
1606 xfs_inode_t *dp; /* incore directory inode */
1607 int error; /* error return value */
1608 xfs_dir2_leaf_t *leaf; /* leaf structure */
1609 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1610 xfs_trans_t *tp; /* transaction pointer */
1615 * Read the offending data block. We need its buffer.
1617 error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(geo, db), 0, &dbp);
1622 ltp = xfs_dir2_leaf_tail_p(geo, leaf);
1626 struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1627 struct xfs_dir2_data_free *bf =
1628 xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
1630 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1631 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1632 ASSERT(be16_to_cpu(bf[0].length) ==
1633 geo->blksize - geo->data_entry_offset);
1634 ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1639 * Get rid of the data block.
1641 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1642 ASSERT(error != -ENOSPC);
1643 xfs_trans_brelse(tp, dbp);
1647 * Eliminate the last bests entry from the table.
1649 bestsp = xfs_dir2_leaf_bests_p(ltp);
1650 be32_add_cpu(<p->bestcount, -1);
1651 memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1652 xfs_dir3_leaf_log_tail(args, lbp);
1653 xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1657 static inline size_t
1659 struct xfs_dir3_icleaf_hdr *hdr,
1665 entries = hdr->count - hdr->stale;
1666 if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1667 hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1668 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1670 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1672 return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1673 + counts * sizeof(xfs_dir2_data_off_t)
1674 + sizeof(xfs_dir2_leaf_tail_t);
1678 * Convert node form directory to leaf form directory.
1679 * The root of the node form dir needs to already be a LEAFN block.
1680 * Just return if we can't do anything.
1683 xfs_dir2_node_to_leaf(
1684 xfs_da_state_t *state) /* directory operation state */
1686 xfs_da_args_t *args; /* operation arguments */
1687 xfs_inode_t *dp; /* incore directory inode */
1688 int error; /* error return code */
1689 struct xfs_buf *fbp; /* buffer for freespace block */
1690 xfs_fileoff_t fo; /* freespace file offset */
1691 struct xfs_buf *lbp; /* buffer for leaf block */
1692 xfs_dir2_leaf_tail_t *ltp; /* tail of leaf structure */
1693 xfs_dir2_leaf_t *leaf; /* leaf structure */
1694 xfs_mount_t *mp; /* filesystem mount point */
1695 int rval; /* successful free trim? */
1696 xfs_trans_t *tp; /* transaction pointer */
1697 struct xfs_dir3_icleaf_hdr leafhdr;
1698 struct xfs_dir3_icfree_hdr freehdr;
1701 * There's more than a leaf level in the btree, so there must
1702 * be multiple leafn blocks. Give up.
1704 if (state->path.active > 1)
1708 trace_xfs_dir2_node_to_leaf(args);
1714 * Get the last offset in the file.
1716 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1719 fo -= args->geo->fsbcount;
1721 * If there are freespace blocks other than the first one,
1722 * take this opportunity to remove trailing empty freespace blocks
1723 * that may have been left behind during no-space-reservation
1726 while (fo > args->geo->freeblk) {
1727 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1731 fo -= args->geo->fsbcount;
1736 * Now find the block just before the freespace block.
1738 if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1742 * If it's not the single leaf block, give up.
1744 if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1746 lbp = state->path.blk[0].bp;
1748 xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, leaf);
1750 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1751 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1754 * Read the freespace block.
1756 error = xfs_dir2_free_read(tp, dp, args->geo->freeblk, &fbp);
1759 xfs_dir2_free_hdr_from_disk(mp, &freehdr, fbp->b_addr);
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) > args->geo->blksize) {
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(args->geo, leaf);
1788 ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1791 * Set up the leaf bests table.
1793 memcpy(xfs_dir2_leaf_bests_p(ltp), freehdr.bests,
1794 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1796 xfs_dir2_leaf_hdr_to_disk(mp, leaf, &leafhdr);
1797 xfs_dir3_leaf_log_header(args, lbp);
1798 xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1799 xfs_dir3_leaf_log_tail(args, 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;