2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
22 #include "xfs_trans.h"
25 #include "xfs_mount.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_dinode.h"
29 #include "xfs_inode.h"
31 #include "xfs_dir2_format.h"
32 #include "xfs_dir2_priv.h"
33 #include "xfs_error.h"
34 #include "xfs_trace.h"
37 * Function declarations.
39 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
42 static void xfs_dir2_leafn_check(struct xfs_inode *dp, struct xfs_buf *bp);
44 #define xfs_dir2_leafn_check(dp, bp)
46 static void xfs_dir2_leafn_moveents(xfs_da_args_t *args, struct xfs_buf *bp_s,
47 int start_s, struct xfs_buf *bp_d,
48 int start_d, int count);
49 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
50 xfs_da_state_blk_t *blk1,
51 xfs_da_state_blk_t *blk2);
52 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
53 int index, xfs_da_state_blk_t *dblk,
55 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
56 xfs_da_state_blk_t *fblk);
62 struct xfs_mount *mp = bp->b_target->bt_mount;
63 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
66 block_ok = hdr->magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC);
68 XFS_CORRUPTION_ERROR("xfs_dir2_free_verify magic",
69 XFS_ERRLEVEL_LOW, mp, hdr);
70 xfs_buf_ioerror(bp, EFSCORRUPTED);
75 xfs_dir2_free_read_verify(
78 xfs_dir2_free_verify(bp);
82 xfs_dir2_free_write_verify(
85 xfs_dir2_free_verify(bp);
88 static const struct xfs_buf_ops xfs_dir2_free_buf_ops = {
89 .verify_read = xfs_dir2_free_read_verify,
90 .verify_write = xfs_dir2_free_write_verify,
99 xfs_daddr_t mappedbno,
100 struct xfs_buf **bpp)
102 return xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
103 XFS_DATA_FORK, &xfs_dir2_free_buf_ops);
108 struct xfs_trans *tp,
109 struct xfs_inode *dp,
111 struct xfs_buf **bpp)
113 return __xfs_dir2_free_read(tp, dp, fbno, -1, bpp);
117 xfs_dir2_free_try_read(
118 struct xfs_trans *tp,
119 struct xfs_inode *dp,
121 struct xfs_buf **bpp)
123 return __xfs_dir2_free_read(tp, dp, fbno, -2, bpp);
127 * Log entries from a freespace block.
130 xfs_dir2_free_log_bests(
131 struct xfs_trans *tp,
133 int first, /* first entry to log */
134 int last) /* last entry to log */
136 xfs_dir2_free_t *free; /* freespace structure */
139 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
140 xfs_trans_log_buf(tp, bp,
141 (uint)((char *)&free->bests[first] - (char *)free),
142 (uint)((char *)&free->bests[last] - (char *)free +
143 sizeof(free->bests[0]) - 1));
147 * Log header from a freespace block.
150 xfs_dir2_free_log_header(
151 struct xfs_trans *tp,
154 xfs_dir2_free_t *free; /* freespace structure */
157 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
158 xfs_trans_log_buf(tp, bp, (uint)((char *)&free->hdr - (char *)free),
159 (uint)(sizeof(xfs_dir2_free_hdr_t) - 1));
163 * Convert a leaf-format directory to a node-format directory.
164 * We need to change the magic number of the leaf block, and copy
165 * the freespace table out of the leaf block into its own block.
168 xfs_dir2_leaf_to_node(
169 xfs_da_args_t *args, /* operation arguments */
170 struct xfs_buf *lbp) /* leaf buffer */
172 xfs_inode_t *dp; /* incore directory inode */
173 int error; /* error return value */
174 struct xfs_buf *fbp; /* freespace buffer */
175 xfs_dir2_db_t fdb; /* freespace block number */
176 xfs_dir2_free_t *free; /* freespace structure */
177 __be16 *from; /* pointer to freespace entry */
178 int i; /* leaf freespace index */
179 xfs_dir2_leaf_t *leaf; /* leaf structure */
180 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
181 xfs_mount_t *mp; /* filesystem mount point */
182 int n; /* count of live freespc ents */
183 xfs_dir2_data_off_t off; /* freespace entry value */
184 __be16 *to; /* pointer to freespace entry */
185 xfs_trans_t *tp; /* transaction pointer */
187 trace_xfs_dir2_leaf_to_node(args);
193 * Add a freespace block to the directory.
195 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
198 ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
200 * Get the buffer for the new freespace block.
202 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb), -1, &fbp,
206 fbp->b_ops = &xfs_dir2_free_buf_ops;
210 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
212 * Initialize the freespace block header.
214 free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
215 free->hdr.firstdb = 0;
216 ASSERT(be32_to_cpu(ltp->bestcount) <= (uint)dp->i_d.di_size / mp->m_dirblksize);
217 free->hdr.nvalid = ltp->bestcount;
219 * Copy freespace entries from the leaf block to the new block.
220 * Count active entries.
222 for (i = n = 0, from = xfs_dir2_leaf_bests_p(ltp), to = free->bests;
223 i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
224 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
226 *to = cpu_to_be16(off);
228 free->hdr.nused = cpu_to_be32(n);
230 lbp->b_ops = &xfs_dir2_leafn_buf_ops;
231 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
236 xfs_dir2_leaf_log_header(tp, lbp);
237 xfs_dir2_free_log_header(tp, fbp);
238 xfs_dir2_free_log_bests(tp, fbp, 0, be32_to_cpu(free->hdr.nvalid) - 1);
239 xfs_dir2_leafn_check(dp, lbp);
244 * Add a leaf entry to a leaf block in a node-form directory.
245 * The other work necessary is done from the caller.
247 static int /* error */
249 struct xfs_buf *bp, /* leaf buffer */
250 xfs_da_args_t *args, /* operation arguments */
251 int index) /* insertion pt for new entry */
253 int compact; /* compacting stale leaves */
254 xfs_inode_t *dp; /* incore directory inode */
255 int highstale; /* next stale entry */
256 xfs_dir2_leaf_t *leaf; /* leaf structure */
257 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
258 int lfloghigh; /* high leaf entry logging */
259 int lfloglow; /* low leaf entry logging */
260 int lowstale; /* previous stale entry */
261 xfs_mount_t *mp; /* filesystem mount point */
262 xfs_trans_t *tp; /* transaction pointer */
264 trace_xfs_dir2_leafn_add(args, index);
272 * Quick check just to make sure we are not going to index
273 * into other peoples memory
276 return XFS_ERROR(EFSCORRUPTED);
279 * If there are already the maximum number of leaf entries in
280 * the block, if there are no stale entries it won't fit.
281 * Caller will do a split. If there are stale entries we'll do
285 if (be16_to_cpu(leaf->hdr.count) == xfs_dir2_max_leaf_ents(mp)) {
286 if (!leaf->hdr.stale)
287 return XFS_ERROR(ENOSPC);
288 compact = be16_to_cpu(leaf->hdr.stale) > 1;
291 ASSERT(index == 0 || be32_to_cpu(leaf->ents[index - 1].hashval) <= args->hashval);
292 ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
293 be32_to_cpu(leaf->ents[index].hashval) >= args->hashval);
295 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
299 * Compact out all but one stale leaf entry. Leaves behind
300 * the entry closest to index.
303 xfs_dir2_leaf_compact_x1(bp, &index, &lowstale, &highstale,
304 &lfloglow, &lfloghigh);
307 * Set impossible logging indices for this case.
309 else if (leaf->hdr.stale) {
310 lfloglow = be16_to_cpu(leaf->hdr.count);
315 * Insert the new entry, log everything.
317 lep = xfs_dir2_leaf_find_entry(leaf, index, compact, lowstale,
318 highstale, &lfloglow, &lfloghigh);
320 lep->hashval = cpu_to_be32(args->hashval);
321 lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
322 args->blkno, args->index));
323 xfs_dir2_leaf_log_header(tp, bp);
324 xfs_dir2_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
325 xfs_dir2_leafn_check(dp, bp);
331 * Check internal consistency of a leafn block.
334 xfs_dir2_leafn_check(
335 struct xfs_inode *dp,
338 int i; /* leaf index */
339 xfs_dir2_leaf_t *leaf; /* leaf structure */
340 xfs_mount_t *mp; /* filesystem mount point */
341 int stale; /* count of stale leaves */
345 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
346 ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
347 for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
348 if (i + 1 < be16_to_cpu(leaf->hdr.count)) {
349 ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
350 be32_to_cpu(leaf->ents[i + 1].hashval));
352 if (leaf->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
355 ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
360 * Return the last hash value in the leaf.
361 * Stale entries are ok.
363 xfs_dahash_t /* hash value */
364 xfs_dir2_leafn_lasthash(
365 struct xfs_buf *bp, /* leaf buffer */
366 int *count) /* count of entries in leaf */
368 xfs_dir2_leaf_t *leaf; /* leaf structure */
371 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
373 *count = be16_to_cpu(leaf->hdr.count);
374 if (!leaf->hdr.count)
376 return be32_to_cpu(leaf->ents[be16_to_cpu(leaf->hdr.count) - 1].hashval);
380 * Look up a leaf entry for space to add a name in a node-format leaf block.
381 * The extrablk in state is a freespace block.
384 xfs_dir2_leafn_lookup_for_addname(
385 struct xfs_buf *bp, /* leaf buffer */
386 xfs_da_args_t *args, /* operation arguments */
387 int *indexp, /* out: leaf entry index */
388 xfs_da_state_t *state) /* state to fill in */
390 struct xfs_buf *curbp = NULL; /* current data/free buffer */
391 xfs_dir2_db_t curdb = -1; /* current data block number */
392 xfs_dir2_db_t curfdb = -1; /* current free block number */
393 xfs_inode_t *dp; /* incore directory inode */
394 int error; /* error return value */
395 int fi; /* free entry index */
396 xfs_dir2_free_t *free = NULL; /* free block structure */
397 int index; /* leaf entry index */
398 xfs_dir2_leaf_t *leaf; /* leaf structure */
399 int length; /* length of new data entry */
400 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
401 xfs_mount_t *mp; /* filesystem mount point */
402 xfs_dir2_db_t newdb; /* new data block number */
403 xfs_dir2_db_t newfdb; /* new free block number */
404 xfs_trans_t *tp; /* transaction pointer */
410 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
412 ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
414 xfs_dir2_leafn_check(dp, bp);
416 * Look up the hash value in the leaf entries.
418 index = xfs_dir2_leaf_search_hash(args, bp);
420 * Do we have a buffer coming in?
422 if (state->extravalid) {
423 /* If so, it's a free block buffer, get the block number. */
424 curbp = state->extrablk.bp;
425 curfdb = state->extrablk.blkno;
426 free = curbp->b_addr;
427 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
429 length = xfs_dir2_data_entsize(args->namelen);
431 * Loop over leaf entries with the right hash value.
433 for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
434 be32_to_cpu(lep->hashval) == args->hashval;
437 * Skip stale leaf entries.
439 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
442 * Pull the data block number from the entry.
444 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
446 * For addname, we're looking for a place to put the new entry.
447 * We want to use a data block with an entry of equal
448 * hash value to ours if there is one with room.
450 * If this block isn't the data block we already have
451 * in hand, take a look at it.
453 if (newdb != curdb) {
456 * Convert the data block to the free block
457 * holding its freespace information.
459 newfdb = xfs_dir2_db_to_fdb(mp, newdb);
461 * If it's not the one we have in hand, read it in.
463 if (newfdb != curfdb) {
465 * If we had one before, drop it.
468 xfs_trans_brelse(tp, curbp);
470 error = xfs_dir2_free_read(tp, dp,
471 xfs_dir2_db_to_da(mp, newfdb),
475 free = curbp->b_addr;
476 ASSERT(be32_to_cpu(free->hdr.magic) ==
477 XFS_DIR2_FREE_MAGIC);
478 ASSERT((be32_to_cpu(free->hdr.firstdb) %
479 xfs_dir2_free_max_bests(mp)) == 0);
480 ASSERT(be32_to_cpu(free->hdr.firstdb) <= curdb);
481 ASSERT(curdb < be32_to_cpu(free->hdr.firstdb) +
482 be32_to_cpu(free->hdr.nvalid));
485 * Get the index for our entry.
487 fi = xfs_dir2_db_to_fdindex(mp, curdb);
489 * If it has room, return it.
491 if (unlikely(free->bests[fi] ==
492 cpu_to_be16(NULLDATAOFF))) {
493 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
494 XFS_ERRLEVEL_LOW, mp);
495 if (curfdb != newfdb)
496 xfs_trans_brelse(tp, curbp);
497 return XFS_ERROR(EFSCORRUPTED);
500 if (be16_to_cpu(free->bests[fi]) >= length)
504 /* Didn't find any space */
507 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
509 /* Giving back a free block. */
510 state->extravalid = 1;
511 state->extrablk.bp = curbp;
512 state->extrablk.index = fi;
513 state->extrablk.blkno = curfdb;
514 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
516 state->extravalid = 0;
519 * Return the index, that will be the insertion point.
522 return XFS_ERROR(ENOENT);
526 * Look up a leaf entry in a node-format leaf block.
527 * The extrablk in state a data block.
530 xfs_dir2_leafn_lookup_for_entry(
531 struct xfs_buf *bp, /* leaf buffer */
532 xfs_da_args_t *args, /* operation arguments */
533 int *indexp, /* out: leaf entry index */
534 xfs_da_state_t *state) /* state to fill in */
536 struct xfs_buf *curbp = NULL; /* current data/free buffer */
537 xfs_dir2_db_t curdb = -1; /* current data block number */
538 xfs_dir2_data_entry_t *dep; /* data block entry */
539 xfs_inode_t *dp; /* incore directory inode */
540 int error; /* error return value */
541 int index; /* leaf entry index */
542 xfs_dir2_leaf_t *leaf; /* leaf structure */
543 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
544 xfs_mount_t *mp; /* filesystem mount point */
545 xfs_dir2_db_t newdb; /* new data block number */
546 xfs_trans_t *tp; /* transaction pointer */
547 enum xfs_dacmp cmp; /* comparison result */
553 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
555 ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
557 xfs_dir2_leafn_check(dp, bp);
559 * Look up the hash value in the leaf entries.
561 index = xfs_dir2_leaf_search_hash(args, bp);
563 * Do we have a buffer coming in?
565 if (state->extravalid) {
566 curbp = state->extrablk.bp;
567 curdb = state->extrablk.blkno;
570 * Loop over leaf entries with the right hash value.
572 for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
573 be32_to_cpu(lep->hashval) == args->hashval;
576 * Skip stale leaf entries.
578 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
581 * Pull the data block number from the entry.
583 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
585 * Not adding a new entry, so we really want to find
586 * the name given to us.
588 * If it's a different data block, go get it.
590 if (newdb != curdb) {
592 * If we had a block before that we aren't saving
593 * for a CI name, drop it
595 if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
596 curdb != state->extrablk.blkno))
597 xfs_trans_brelse(tp, curbp);
599 * If needing the block that is saved with a CI match,
600 * use it otherwise read in the new data block.
602 if (args->cmpresult != XFS_CMP_DIFFERENT &&
603 newdb == state->extrablk.blkno) {
604 ASSERT(state->extravalid);
605 curbp = state->extrablk.bp;
607 error = xfs_dir2_data_read(tp, dp,
608 xfs_dir2_db_to_da(mp, newdb),
613 xfs_dir2_data_check(dp, curbp);
617 * Point to the data entry.
619 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
620 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
622 * Compare the entry and if it's an exact match, return
623 * EEXIST immediately. If it's the first case-insensitive
624 * match, store the block & inode number and continue looking.
626 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
627 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
628 /* If there is a CI match block, drop it */
629 if (args->cmpresult != XFS_CMP_DIFFERENT &&
630 curdb != state->extrablk.blkno)
631 xfs_trans_brelse(tp, state->extrablk.bp);
632 args->cmpresult = cmp;
633 args->inumber = be64_to_cpu(dep->inumber);
635 state->extravalid = 1;
636 state->extrablk.bp = curbp;
637 state->extrablk.blkno = curdb;
638 state->extrablk.index = (int)((char *)dep -
639 (char *)curbp->b_addr);
640 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
641 curbp->b_ops = &xfs_dir2_data_buf_ops;
642 if (cmp == XFS_CMP_EXACT)
643 return XFS_ERROR(EEXIST);
646 ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
647 (args->op_flags & XFS_DA_OP_OKNOENT));
649 if (args->cmpresult == XFS_CMP_DIFFERENT) {
650 /* Giving back last used data block. */
651 state->extravalid = 1;
652 state->extrablk.bp = curbp;
653 state->extrablk.index = -1;
654 state->extrablk.blkno = curdb;
655 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
656 curbp->b_ops = &xfs_dir2_data_buf_ops;
658 /* If the curbp is not the CI match block, drop it */
659 if (state->extrablk.bp != curbp)
660 xfs_trans_brelse(tp, curbp);
663 state->extravalid = 0;
666 return XFS_ERROR(ENOENT);
670 * Look up a leaf entry in a node-format leaf block.
671 * If this is an addname then the extrablk in state is a freespace block,
672 * otherwise it's a data block.
675 xfs_dir2_leafn_lookup_int(
676 struct xfs_buf *bp, /* leaf buffer */
677 xfs_da_args_t *args, /* operation arguments */
678 int *indexp, /* out: leaf entry index */
679 xfs_da_state_t *state) /* state to fill in */
681 if (args->op_flags & XFS_DA_OP_ADDNAME)
682 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
684 return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
688 * Move count leaf entries from source to destination leaf.
689 * Log entries and headers. Stale entries are preserved.
692 xfs_dir2_leafn_moveents(
693 xfs_da_args_t *args, /* operation arguments */
694 struct xfs_buf *bp_s, /* source leaf buffer */
695 int start_s, /* source leaf index */
696 struct xfs_buf *bp_d, /* destination leaf buffer */
697 int start_d, /* destination leaf index */
698 int count) /* count of leaves to copy */
700 xfs_dir2_leaf_t *leaf_d; /* destination leaf structure */
701 xfs_dir2_leaf_t *leaf_s; /* source leaf structure */
702 int stale; /* count stale leaves copied */
703 xfs_trans_t *tp; /* transaction pointer */
705 trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
708 * Silently return if nothing to do.
714 leaf_s = bp_s->b_addr;
715 leaf_d = bp_d->b_addr;
717 * If the destination index is not the end of the current
718 * destination leaf entries, open up a hole in the destination
719 * to hold the new entries.
721 if (start_d < be16_to_cpu(leaf_d->hdr.count)) {
722 memmove(&leaf_d->ents[start_d + count], &leaf_d->ents[start_d],
723 (be16_to_cpu(leaf_d->hdr.count) - start_d) *
724 sizeof(xfs_dir2_leaf_entry_t));
725 xfs_dir2_leaf_log_ents(tp, bp_d, start_d + count,
726 count + be16_to_cpu(leaf_d->hdr.count) - 1);
729 * If the source has stale leaves, count the ones in the copy range
730 * so we can update the header correctly.
732 if (leaf_s->hdr.stale) {
733 int i; /* temp leaf index */
735 for (i = start_s, stale = 0; i < start_s + count; i++) {
736 if (leaf_s->ents[i].address ==
737 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
743 * Copy the leaf entries from source to destination.
745 memcpy(&leaf_d->ents[start_d], &leaf_s->ents[start_s],
746 count * sizeof(xfs_dir2_leaf_entry_t));
747 xfs_dir2_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
749 * If there are source entries after the ones we copied,
750 * delete the ones we copied by sliding the next ones down.
752 if (start_s + count < be16_to_cpu(leaf_s->hdr.count)) {
753 memmove(&leaf_s->ents[start_s], &leaf_s->ents[start_s + count],
754 count * sizeof(xfs_dir2_leaf_entry_t));
755 xfs_dir2_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
758 * Update the headers and log them.
760 be16_add_cpu(&leaf_s->hdr.count, -(count));
761 be16_add_cpu(&leaf_s->hdr.stale, -(stale));
762 be16_add_cpu(&leaf_d->hdr.count, count);
763 be16_add_cpu(&leaf_d->hdr.stale, stale);
764 xfs_dir2_leaf_log_header(tp, bp_s);
765 xfs_dir2_leaf_log_header(tp, bp_d);
766 xfs_dir2_leafn_check(args->dp, bp_s);
767 xfs_dir2_leafn_check(args->dp, bp_d);
771 * Determine the sort order of two leaf blocks.
772 * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
775 xfs_dir2_leafn_order(
776 struct xfs_buf *leaf1_bp, /* leaf1 buffer */
777 struct xfs_buf *leaf2_bp) /* leaf2 buffer */
779 xfs_dir2_leaf_t *leaf1; /* leaf1 structure */
780 xfs_dir2_leaf_t *leaf2; /* leaf2 structure */
782 leaf1 = leaf1_bp->b_addr;
783 leaf2 = leaf2_bp->b_addr;
784 ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
785 ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
786 if (be16_to_cpu(leaf1->hdr.count) > 0 &&
787 be16_to_cpu(leaf2->hdr.count) > 0 &&
788 (be32_to_cpu(leaf2->ents[0].hashval) < be32_to_cpu(leaf1->ents[0].hashval) ||
789 be32_to_cpu(leaf2->ents[be16_to_cpu(leaf2->hdr.count) - 1].hashval) <
790 be32_to_cpu(leaf1->ents[be16_to_cpu(leaf1->hdr.count) - 1].hashval)))
796 * Rebalance leaf entries between two leaf blocks.
797 * This is actually only called when the second block is new,
798 * though the code deals with the general case.
799 * A new entry will be inserted in one of the blocks, and that
800 * entry is taken into account when balancing.
803 xfs_dir2_leafn_rebalance(
804 xfs_da_state_t *state, /* btree cursor */
805 xfs_da_state_blk_t *blk1, /* first btree block */
806 xfs_da_state_blk_t *blk2) /* second btree block */
808 xfs_da_args_t *args; /* operation arguments */
809 int count; /* count (& direction) leaves */
810 int isleft; /* new goes in left leaf */
811 xfs_dir2_leaf_t *leaf1; /* first leaf structure */
812 xfs_dir2_leaf_t *leaf2; /* second leaf structure */
813 int mid; /* midpoint leaf index */
815 int oldstale; /* old count of stale leaves */
817 int oldsum; /* old total leaf count */
818 int swap; /* swapped leaf blocks */
822 * If the block order is wrong, swap the arguments.
824 if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
825 xfs_da_state_blk_t *tmp; /* temp for block swap */
831 leaf1 = blk1->bp->b_addr;
832 leaf2 = blk2->bp->b_addr;
833 oldsum = be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count);
835 oldstale = be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale);
839 * If the old leaf count was odd then the new one will be even,
840 * so we need to divide the new count evenly.
843 xfs_dahash_t midhash; /* middle entry hash value */
845 if (mid >= be16_to_cpu(leaf1->hdr.count))
846 midhash = be32_to_cpu(leaf2->ents[mid - be16_to_cpu(leaf1->hdr.count)].hashval);
848 midhash = be32_to_cpu(leaf1->ents[mid].hashval);
849 isleft = args->hashval <= midhash;
852 * If the old count is even then the new count is odd, so there's
853 * no preferred side for the new entry.
859 * Calculate moved entry count. Positive means left-to-right,
860 * negative means right-to-left. Then move the entries.
862 count = be16_to_cpu(leaf1->hdr.count) - mid + (isleft == 0);
864 xfs_dir2_leafn_moveents(args, blk1->bp,
865 be16_to_cpu(leaf1->hdr.count) - count, blk2->bp, 0, count);
867 xfs_dir2_leafn_moveents(args, blk2->bp, 0, blk1->bp,
868 be16_to_cpu(leaf1->hdr.count), count);
869 ASSERT(be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count) == oldsum);
870 ASSERT(be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale) == oldstale);
872 * Mark whether we're inserting into the old or new leaf.
874 if (be16_to_cpu(leaf1->hdr.count) < be16_to_cpu(leaf2->hdr.count))
875 state->inleaf = swap;
876 else if (be16_to_cpu(leaf1->hdr.count) > be16_to_cpu(leaf2->hdr.count))
877 state->inleaf = !swap;
880 swap ^ (blk1->index <= be16_to_cpu(leaf1->hdr.count));
882 * Adjust the expected index for insertion.
885 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
888 * Finally sanity check just to make sure we are not returning a
891 if(blk2->index < 0) {
894 xfs_alert(args->dp->i_mount,
895 "%s: picked the wrong leaf? reverting original leaf: blk1->index %d\n",
896 __func__, blk1->index);
901 xfs_dir2_data_block_free(
903 struct xfs_dir2_data_hdr *hdr,
904 struct xfs_dir2_free *free,
910 struct xfs_trans *tp = args->trans;
914 /* One less used entry in the free table. */
915 be32_add_cpu(&free->hdr.nused, -1);
916 xfs_dir2_free_log_header(tp, fbp);
919 * If this was the last entry in the table, we can trim the
920 * table size back. There might be other entries at the end
921 * referring to non-existent data blocks, get those too.
923 if (findex == be32_to_cpu(free->hdr.nvalid) - 1) {
924 int i; /* free entry index */
926 for (i = findex - 1; i >= 0; i--) {
927 if (free->bests[i] != cpu_to_be16(NULLDATAOFF))
930 free->hdr.nvalid = cpu_to_be32(i + 1);
933 /* Not the last entry, just punch it out. */
934 free->bests[findex] = cpu_to_be16(NULLDATAOFF);
938 * If there are no useful entries left in the block,
939 * get rid of the block if we can.
941 if (!free->hdr.nused) {
944 error = xfs_dir2_shrink_inode(args, fdb, fbp);
948 } else if (error != ENOSPC || args->total != 0)
951 * It's possible to get ENOSPC if there is no
952 * space reservation. In this case some one
953 * else will eventually get rid of this block.
958 * Data block is not empty, just set the free entry to the new
961 free->bests[findex] = cpu_to_be16(longest);
965 /* Log the free entry that changed, unless we got rid of it. */
967 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
972 * Remove an entry from a node directory.
973 * This removes the leaf entry and the data entry,
974 * and updates the free block if necessary.
976 static int /* error */
977 xfs_dir2_leafn_remove(
978 xfs_da_args_t *args, /* operation arguments */
979 struct xfs_buf *bp, /* leaf buffer */
980 int index, /* leaf entry index */
981 xfs_da_state_blk_t *dblk, /* data block */
982 int *rval) /* resulting block needs join */
984 xfs_dir2_data_hdr_t *hdr; /* data block header */
985 xfs_dir2_db_t db; /* data block number */
986 struct xfs_buf *dbp; /* data block buffer */
987 xfs_dir2_data_entry_t *dep; /* data block entry */
988 xfs_inode_t *dp; /* incore directory inode */
989 xfs_dir2_leaf_t *leaf; /* leaf structure */
990 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
991 int longest; /* longest data free entry */
992 int off; /* data block entry offset */
993 xfs_mount_t *mp; /* filesystem mount point */
994 int needlog; /* need to log data header */
995 int needscan; /* need to rescan data frees */
996 xfs_trans_t *tp; /* transaction pointer */
998 trace_xfs_dir2_leafn_remove(args, index);
1004 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1006 * Point to the entry we're removing.
1008 lep = &leaf->ents[index];
1010 * Extract the data block and offset from the entry.
1012 db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1013 ASSERT(dblk->blkno == db);
1014 off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
1015 ASSERT(dblk->index == off);
1017 * Kill the leaf entry by marking it stale.
1018 * Log the leaf block changes.
1020 be16_add_cpu(&leaf->hdr.stale, 1);
1021 xfs_dir2_leaf_log_header(tp, bp);
1022 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1023 xfs_dir2_leaf_log_ents(tp, bp, index, index);
1025 * Make the data entry free. Keep track of the longest freespace
1026 * in the data block in case it changes.
1030 dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1031 longest = be16_to_cpu(hdr->bestfree[0].length);
1032 needlog = needscan = 0;
1033 xfs_dir2_data_make_free(tp, dbp, off,
1034 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1036 * Rescan the data block freespaces for bestfree.
1037 * Log the data block header if needed.
1040 xfs_dir2_data_freescan(mp, hdr, &needlog);
1042 xfs_dir2_data_log_header(tp, dbp);
1043 xfs_dir2_data_check(dp, dbp);
1045 * If the longest data block freespace changes, need to update
1046 * the corresponding freeblock entry.
1048 if (longest < be16_to_cpu(hdr->bestfree[0].length)) {
1049 int error; /* error return value */
1050 struct xfs_buf *fbp; /* freeblock buffer */
1051 xfs_dir2_db_t fdb; /* freeblock block number */
1052 int findex; /* index in freeblock entries */
1053 xfs_dir2_free_t *free; /* freeblock structure */
1056 * Convert the data block number to a free block,
1057 * read in the free block.
1059 fdb = xfs_dir2_db_to_fdb(mp, db);
1060 error = xfs_dir2_free_read(tp, dp, xfs_dir2_db_to_da(mp, fdb),
1065 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1066 ASSERT(be32_to_cpu(free->hdr.firstdb) ==
1067 xfs_dir2_free_max_bests(mp) *
1068 (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
1070 * Calculate which entry we need to fix.
1072 findex = xfs_dir2_db_to_fdindex(mp, db);
1073 longest = be16_to_cpu(hdr->bestfree[0].length);
1075 * If the data block is now empty we can get rid of it
1078 if (longest == mp->m_dirblksize - (uint)sizeof(*hdr)) {
1080 * Try to punch out the data block.
1082 error = xfs_dir2_shrink_inode(args, db, dbp);
1088 * We can get ENOSPC if there's no space reservation.
1089 * In this case just drop the buffer and some one else
1090 * will eventually get rid of the empty block.
1092 else if (!(error == ENOSPC && args->total == 0))
1096 * If we got rid of the data block, we can eliminate that entry
1097 * in the free block.
1099 error = xfs_dir2_data_block_free(args, hdr, free,
1100 fdb, findex, fbp, longest);
1105 xfs_dir2_leafn_check(dp, bp);
1107 * Return indication of whether this leaf block is empty enough
1108 * to justify trying to join it with a neighbor.
1111 ((uint)sizeof(leaf->hdr) +
1112 (uint)sizeof(leaf->ents[0]) *
1113 (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale))) <
1119 * Split the leaf entries in the old block into old and new blocks.
1122 xfs_dir2_leafn_split(
1123 xfs_da_state_t *state, /* btree cursor */
1124 xfs_da_state_blk_t *oldblk, /* original block */
1125 xfs_da_state_blk_t *newblk) /* newly created block */
1127 xfs_da_args_t *args; /* operation arguments */
1128 xfs_dablk_t blkno; /* new leaf block number */
1129 int error; /* error return value */
1130 xfs_mount_t *mp; /* filesystem mount point */
1133 * Allocate space for a new leaf node.
1136 mp = args->dp->i_mount;
1137 ASSERT(args != NULL);
1138 ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1139 error = xfs_da_grow_inode(args, &blkno);
1144 * Initialize the new leaf block.
1146 error = xfs_dir2_leaf_init(args, xfs_dir2_da_to_db(mp, blkno),
1147 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1151 newblk->blkno = blkno;
1152 newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1154 * Rebalance the entries across the two leaves, link the new
1155 * block into the leaves.
1157 xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1158 error = xfs_da_blk_link(state, oldblk, newblk);
1163 * Insert the new entry in the correct block.
1166 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1168 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1170 * Update last hashval in each block since we added the name.
1172 oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1173 newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1174 xfs_dir2_leafn_check(args->dp, oldblk->bp);
1175 xfs_dir2_leafn_check(args->dp, newblk->bp);
1180 * Check a leaf block and its neighbors to see if the block should be
1181 * collapsed into one or the other neighbor. Always keep the block
1182 * with the smaller block number.
1183 * If the current block is over 50% full, don't try to join it, return 0.
1184 * If the block is empty, fill in the state structure and return 2.
1185 * If it can be collapsed, fill in the state structure and return 1.
1186 * If nothing can be done, return 0.
1189 xfs_dir2_leafn_toosmall(
1190 xfs_da_state_t *state, /* btree cursor */
1191 int *action) /* resulting action to take */
1193 xfs_da_state_blk_t *blk; /* leaf block */
1194 xfs_dablk_t blkno; /* leaf block number */
1195 struct xfs_buf *bp; /* leaf buffer */
1196 int bytes; /* bytes in use */
1197 int count; /* leaf live entry count */
1198 int error; /* error return value */
1199 int forward; /* sibling block direction */
1200 int i; /* sibling counter */
1201 xfs_da_blkinfo_t *info; /* leaf block header */
1202 xfs_dir2_leaf_t *leaf; /* leaf structure */
1203 int rval; /* result from path_shift */
1206 * Check for the degenerate case of the block being over 50% full.
1207 * If so, it's not worth even looking to see if we might be able
1208 * to coalesce with a sibling.
1210 blk = &state->path.blk[state->path.active - 1];
1211 info = blk->bp->b_addr;
1212 ASSERT(info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1213 leaf = (xfs_dir2_leaf_t *)info;
1214 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1215 bytes = (uint)sizeof(leaf->hdr) + count * (uint)sizeof(leaf->ents[0]);
1216 if (bytes > (state->blocksize >> 1)) {
1218 * Blk over 50%, don't try to join.
1224 * Check for the degenerate case of the block being empty.
1225 * If the block is empty, we'll simply delete it, no need to
1226 * coalesce it with a sibling block. We choose (arbitrarily)
1227 * to merge with the forward block unless it is NULL.
1231 * Make altpath point to the block we want to keep and
1232 * path point to the block we want to drop (this one).
1234 forward = (info->forw != 0);
1235 memcpy(&state->altpath, &state->path, sizeof(state->path));
1236 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1240 *action = rval ? 2 : 0;
1244 * Examine each sibling block to see if we can coalesce with
1245 * at least 25% free space to spare. We need to figure out
1246 * whether to merge with the forward or the backward block.
1247 * We prefer coalescing with the lower numbered sibling so as
1248 * to shrink a directory over time.
1250 forward = be32_to_cpu(info->forw) < be32_to_cpu(info->back);
1251 for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1252 blkno = forward ? be32_to_cpu(info->forw) : be32_to_cpu(info->back);
1256 * Read the sibling leaf block.
1258 error = xfs_dir2_leafn_read(state->args->trans, state->args->dp,
1264 * Count bytes in the two blocks combined.
1266 leaf = (xfs_dir2_leaf_t *)info;
1267 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1268 bytes = state->blocksize - (state->blocksize >> 2);
1270 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1271 count += be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1272 bytes -= count * (uint)sizeof(leaf->ents[0]);
1274 * Fits with at least 25% to spare.
1278 xfs_trans_brelse(state->args->trans, bp);
1281 * Didn't like either block, give up.
1289 * Make altpath point to the block we want to keep (the lower
1290 * numbered block) and path point to the block we want to drop.
1292 memcpy(&state->altpath, &state->path, sizeof(state->path));
1293 if (blkno < blk->blkno)
1294 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1297 error = xfs_da_path_shift(state, &state->path, forward, 0,
1302 *action = rval ? 0 : 1;
1307 * Move all the leaf entries from drop_blk to save_blk.
1308 * This is done as part of a join operation.
1311 xfs_dir2_leafn_unbalance(
1312 xfs_da_state_t *state, /* cursor */
1313 xfs_da_state_blk_t *drop_blk, /* dead block */
1314 xfs_da_state_blk_t *save_blk) /* surviving block */
1316 xfs_da_args_t *args; /* operation arguments */
1317 xfs_dir2_leaf_t *drop_leaf; /* dead leaf structure */
1318 xfs_dir2_leaf_t *save_leaf; /* surviving leaf structure */
1321 ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1322 ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1323 drop_leaf = drop_blk->bp->b_addr;
1324 save_leaf = save_blk->bp->b_addr;
1325 ASSERT(drop_leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1326 ASSERT(save_leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1328 * If there are any stale leaf entries, take this opportunity
1331 if (drop_leaf->hdr.stale)
1332 xfs_dir2_leaf_compact(args, drop_blk->bp);
1333 if (save_leaf->hdr.stale)
1334 xfs_dir2_leaf_compact(args, save_blk->bp);
1336 * Move the entries from drop to the appropriate end of save.
1338 drop_blk->hashval = be32_to_cpu(drop_leaf->ents[be16_to_cpu(drop_leaf->hdr.count) - 1].hashval);
1339 if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1340 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp, 0,
1341 be16_to_cpu(drop_leaf->hdr.count));
1343 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp,
1344 be16_to_cpu(save_leaf->hdr.count), be16_to_cpu(drop_leaf->hdr.count));
1345 save_blk->hashval = be32_to_cpu(save_leaf->ents[be16_to_cpu(save_leaf->hdr.count) - 1].hashval);
1346 xfs_dir2_leafn_check(args->dp, save_blk->bp);
1350 * Top-level node form directory addname routine.
1353 xfs_dir2_node_addname(
1354 xfs_da_args_t *args) /* operation arguments */
1356 xfs_da_state_blk_t *blk; /* leaf block for insert */
1357 int error; /* error return value */
1358 int rval; /* sub-return value */
1359 xfs_da_state_t *state; /* btree cursor */
1361 trace_xfs_dir2_node_addname(args);
1364 * Allocate and initialize the state (btree cursor).
1366 state = xfs_da_state_alloc();
1368 state->mp = args->dp->i_mount;
1369 state->blocksize = state->mp->m_dirblksize;
1370 state->node_ents = state->mp->m_dir_node_ents;
1372 * Look up the name. We're not supposed to find it, but
1373 * this gives us the insertion point.
1375 error = xfs_da_node_lookup_int(state, &rval);
1378 if (rval != ENOENT) {
1382 * Add the data entry to a data block.
1383 * Extravalid is set to a freeblock found by lookup.
1385 rval = xfs_dir2_node_addname_int(args,
1386 state->extravalid ? &state->extrablk : NULL);
1390 blk = &state->path.blk[state->path.active - 1];
1391 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1393 * Add the new leaf entry.
1395 rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1398 * It worked, fix the hash values up the btree.
1400 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1401 xfs_da_fixhashpath(state, &state->path);
1404 * It didn't work, we need to split the leaf block.
1406 if (args->total == 0) {
1407 ASSERT(rval == ENOSPC);
1411 * Split the leaf block and insert the new entry.
1413 rval = xfs_da_split(state);
1416 xfs_da_state_free(state);
1421 * Add the data entry for a node-format directory name addition.
1422 * The leaf entry is added in xfs_dir2_leafn_add.
1423 * We may enter with a freespace block that the lookup found.
1425 static int /* error */
1426 xfs_dir2_node_addname_int(
1427 xfs_da_args_t *args, /* operation arguments */
1428 xfs_da_state_blk_t *fblk) /* optional freespace block */
1430 xfs_dir2_data_hdr_t *hdr; /* data block header */
1431 xfs_dir2_db_t dbno; /* data block number */
1432 struct xfs_buf *dbp; /* data block buffer */
1433 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1434 xfs_inode_t *dp; /* incore directory inode */
1435 xfs_dir2_data_unused_t *dup; /* data unused entry pointer */
1436 int error; /* error return value */
1437 xfs_dir2_db_t fbno; /* freespace block number */
1438 struct xfs_buf *fbp; /* freespace buffer */
1439 int findex; /* freespace entry index */
1440 xfs_dir2_free_t *free=NULL; /* freespace block structure */
1441 xfs_dir2_db_t ifbno; /* initial freespace block no */
1442 xfs_dir2_db_t lastfbno=0; /* highest freespace block no */
1443 int length; /* length of the new entry */
1444 int logfree; /* need to log free entry */
1445 xfs_mount_t *mp; /* filesystem mount point */
1446 int needlog; /* need to log data header */
1447 int needscan; /* need to rescan data frees */
1448 __be16 *tagp; /* data entry tag pointer */
1449 xfs_trans_t *tp; /* transaction pointer */
1454 length = xfs_dir2_data_entsize(args->namelen);
1456 * If we came in with a freespace block that means that lookup
1457 * found an entry with our hash value. This is the freespace
1458 * block for that data entry.
1463 * Remember initial freespace block number.
1465 ifbno = fblk->blkno;
1467 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1468 findex = fblk->index;
1470 * This means the free entry showed that the data block had
1471 * space for our entry, so we remembered it.
1472 * Use that data block.
1475 ASSERT(findex < be32_to_cpu(free->hdr.nvalid));
1476 ASSERT(be16_to_cpu(free->bests[findex]) != NULLDATAOFF);
1477 ASSERT(be16_to_cpu(free->bests[findex]) >= length);
1478 dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1481 * The data block looked at didn't have enough room.
1482 * We'll start at the beginning of the freespace entries.
1490 * Didn't come in with a freespace block, so don't have a data block.
1498 * If we don't have a data block yet, we're going to scan the
1499 * freespace blocks looking for one. Figure out what the
1500 * highest freespace block number is.
1503 xfs_fileoff_t fo; /* freespace block number */
1505 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1507 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1511 * While we haven't identified a data block, search the freeblock
1512 * data for a good data block. If we find a null freeblock entry,
1513 * indicating a hole in the data blocks, remember that.
1515 while (dbno == -1) {
1517 * If we don't have a freeblock in hand, get the next one.
1521 * Happens the first time through unless lookup gave
1522 * us a freespace block to start with.
1525 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1527 * If it's ifbno we already looked at it.
1532 * If it's off the end we're done.
1534 if (fbno >= lastfbno)
1537 * Read the block. There can be holes in the
1538 * freespace blocks, so this might not succeed.
1539 * This should be really rare, so there's no reason
1542 error = xfs_dir2_free_try_read(tp, dp,
1543 xfs_dir2_db_to_da(mp, fbno),
1550 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1554 * Look at the current free entry. Is it good enough?
1556 if (be16_to_cpu(free->bests[findex]) != NULLDATAOFF &&
1557 be16_to_cpu(free->bests[findex]) >= length)
1558 dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1561 * Are we done with the freeblock?
1563 if (++findex == be32_to_cpu(free->hdr.nvalid)) {
1567 xfs_trans_brelse(tp, fbp);
1569 if (fblk && fblk->bp)
1575 * If we don't have a data block, we need to allocate one and make
1576 * the freespace entries refer to it.
1578 if (unlikely(dbno == -1)) {
1580 * Not allowed to allocate, return failure.
1582 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1583 return XFS_ERROR(ENOSPC);
1586 * Allocate and initialize the new data block.
1588 if (unlikely((error = xfs_dir2_grow_inode(args,
1589 XFS_DIR2_DATA_SPACE,
1591 (error = xfs_dir2_data_init(args, dbno, &dbp))))
1595 * If (somehow) we have a freespace block, get rid of it.
1598 xfs_trans_brelse(tp, fbp);
1599 if (fblk && fblk->bp)
1603 * Get the freespace block corresponding to the data block
1604 * that was just allocated.
1606 fbno = xfs_dir2_db_to_fdb(mp, dbno);
1607 error = xfs_dir2_free_try_read(tp, dp,
1608 xfs_dir2_db_to_da(mp, fbno),
1614 * If there wasn't a freespace block, the read will
1615 * return a NULL fbp. Allocate and initialize a new one.
1618 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1623 if (unlikely(xfs_dir2_db_to_fdb(mp, dbno) != fbno)) {
1625 "%s: dir ino %llu needed freesp block %lld for\n"
1626 " data block %lld, got %lld ifbno %llu lastfbno %d",
1627 __func__, (unsigned long long)dp->i_ino,
1628 (long long)xfs_dir2_db_to_fdb(mp, dbno),
1629 (long long)dbno, (long long)fbno,
1630 (unsigned long long)ifbno, lastfbno);
1633 " fblk 0x%p blkno %llu index %d magic 0x%x",
1635 (unsigned long long)fblk->blkno,
1639 xfs_alert(mp, " ... fblk is NULL");
1641 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1642 XFS_ERRLEVEL_LOW, mp);
1643 return XFS_ERROR(EFSCORRUPTED);
1647 * Get a buffer for the new block.
1649 error = xfs_da_get_buf(tp, dp,
1650 xfs_dir2_db_to_da(mp, fbno),
1651 -1, &fbp, XFS_DATA_FORK);
1654 fbp->b_ops = &xfs_dir2_free_buf_ops;
1657 * Initialize the new block to be empty, and remember
1658 * its first slot as our empty slot.
1661 free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
1662 free->hdr.firstdb = cpu_to_be32(
1663 (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1664 xfs_dir2_free_max_bests(mp));
1665 free->hdr.nvalid = 0;
1666 free->hdr.nused = 0;
1669 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1673 * Set the freespace block index from the data block number.
1675 findex = xfs_dir2_db_to_fdindex(mp, dbno);
1677 * If it's after the end of the current entries in the
1678 * freespace block, extend that table.
1680 if (findex >= be32_to_cpu(free->hdr.nvalid)) {
1681 ASSERT(findex < xfs_dir2_free_max_bests(mp));
1682 free->hdr.nvalid = cpu_to_be32(findex + 1);
1684 * Tag new entry so nused will go up.
1686 free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1689 * If this entry was for an empty data block
1690 * (this should always be true) then update the header.
1692 if (free->bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1693 be32_add_cpu(&free->hdr.nused, 1);
1694 xfs_dir2_free_log_header(tp, fbp);
1697 * Update the real value in the table.
1698 * We haven't allocated the data entry yet so this will
1702 free->bests[findex] = hdr->bestfree[0].length;
1706 * We had a data block so we don't have to make a new one.
1710 * If just checking, we succeeded.
1712 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1716 * Read the data block in.
1718 error = xfs_dir2_data_read(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1725 ASSERT(be16_to_cpu(hdr->bestfree[0].length) >= length);
1727 * Point to the existing unused space.
1729 dup = (xfs_dir2_data_unused_t *)
1730 ((char *)hdr + be16_to_cpu(hdr->bestfree[0].offset));
1731 needscan = needlog = 0;
1733 * Mark the first part of the unused space, inuse for us.
1735 xfs_dir2_data_use_free(tp, dbp, dup,
1736 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
1737 &needlog, &needscan);
1739 * Fill in the new entry and log it.
1741 dep = (xfs_dir2_data_entry_t *)dup;
1742 dep->inumber = cpu_to_be64(args->inumber);
1743 dep->namelen = args->namelen;
1744 memcpy(dep->name, args->name, dep->namelen);
1745 tagp = xfs_dir2_data_entry_tag_p(dep);
1746 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1747 xfs_dir2_data_log_entry(tp, dbp, dep);
1749 * Rescan the block for bestfree if needed.
1752 xfs_dir2_data_freescan(mp, hdr, &needlog);
1754 * Log the data block header if needed.
1757 xfs_dir2_data_log_header(tp, dbp);
1759 * If the freespace entry is now wrong, update it.
1761 if (be16_to_cpu(free->bests[findex]) != be16_to_cpu(hdr->bestfree[0].length)) {
1762 free->bests[findex] = hdr->bestfree[0].length;
1766 * Log the freespace entry if needed.
1769 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1771 * Return the data block and offset in args, then drop the data block.
1773 args->blkno = (xfs_dablk_t)dbno;
1774 args->index = be16_to_cpu(*tagp);
1779 * Lookup an entry in a node-format directory.
1780 * All the real work happens in xfs_da_node_lookup_int.
1781 * The only real output is the inode number of the entry.
1784 xfs_dir2_node_lookup(
1785 xfs_da_args_t *args) /* operation arguments */
1787 int error; /* error return value */
1788 int i; /* btree level */
1789 int rval; /* operation return value */
1790 xfs_da_state_t *state; /* btree cursor */
1792 trace_xfs_dir2_node_lookup(args);
1795 * Allocate and initialize the btree cursor.
1797 state = xfs_da_state_alloc();
1799 state->mp = args->dp->i_mount;
1800 state->blocksize = state->mp->m_dirblksize;
1801 state->node_ents = state->mp->m_dir_node_ents;
1803 * Fill in the path to the entry in the cursor.
1805 error = xfs_da_node_lookup_int(state, &rval);
1808 else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
1809 /* If a CI match, dup the actual name and return EEXIST */
1810 xfs_dir2_data_entry_t *dep;
1812 dep = (xfs_dir2_data_entry_t *)
1813 ((char *)state->extrablk.bp->b_addr +
1814 state->extrablk.index);
1815 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1818 * Release the btree blocks and leaf block.
1820 for (i = 0; i < state->path.active; i++) {
1821 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
1822 state->path.blk[i].bp = NULL;
1825 * Release the data block if we have it.
1827 if (state->extravalid && state->extrablk.bp) {
1828 xfs_trans_brelse(args->trans, state->extrablk.bp);
1829 state->extrablk.bp = NULL;
1831 xfs_da_state_free(state);
1836 * Remove an entry from a node-format directory.
1839 xfs_dir2_node_removename(
1840 xfs_da_args_t *args) /* operation arguments */
1842 xfs_da_state_blk_t *blk; /* leaf block */
1843 int error; /* error return value */
1844 int rval; /* operation return value */
1845 xfs_da_state_t *state; /* btree cursor */
1847 trace_xfs_dir2_node_removename(args);
1850 * Allocate and initialize the btree cursor.
1852 state = xfs_da_state_alloc();
1854 state->mp = args->dp->i_mount;
1855 state->blocksize = state->mp->m_dirblksize;
1856 state->node_ents = state->mp->m_dir_node_ents;
1858 * Look up the entry we're deleting, set up the cursor.
1860 error = xfs_da_node_lookup_int(state, &rval);
1864 * Didn't find it, upper layer screwed up.
1866 if (rval != EEXIST) {
1867 xfs_da_state_free(state);
1870 blk = &state->path.blk[state->path.active - 1];
1871 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1872 ASSERT(state->extravalid);
1874 * Remove the leaf and data entries.
1875 * Extrablk refers to the data block.
1877 error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
1878 &state->extrablk, &rval);
1882 * Fix the hash values up the btree.
1884 xfs_da_fixhashpath(state, &state->path);
1886 * If we need to join leaf blocks, do it.
1888 if (rval && state->path.active > 1)
1889 error = xfs_da_join(state);
1891 * If no errors so far, try conversion to leaf format.
1894 error = xfs_dir2_node_to_leaf(state);
1895 xfs_da_state_free(state);
1900 * Replace an entry's inode number in a node-format directory.
1903 xfs_dir2_node_replace(
1904 xfs_da_args_t *args) /* operation arguments */
1906 xfs_da_state_blk_t *blk; /* leaf block */
1907 xfs_dir2_data_hdr_t *hdr; /* data block header */
1908 xfs_dir2_data_entry_t *dep; /* data entry changed */
1909 int error; /* error return value */
1910 int i; /* btree level */
1911 xfs_ino_t inum; /* new inode number */
1912 xfs_dir2_leaf_t *leaf; /* leaf structure */
1913 xfs_dir2_leaf_entry_t *lep; /* leaf entry being changed */
1914 int rval; /* internal return value */
1915 xfs_da_state_t *state; /* btree cursor */
1917 trace_xfs_dir2_node_replace(args);
1920 * Allocate and initialize the btree cursor.
1922 state = xfs_da_state_alloc();
1924 state->mp = args->dp->i_mount;
1925 state->blocksize = state->mp->m_dirblksize;
1926 state->node_ents = state->mp->m_dir_node_ents;
1927 inum = args->inumber;
1929 * Lookup the entry to change in the btree.
1931 error = xfs_da_node_lookup_int(state, &rval);
1936 * It should be found, since the vnodeops layer has looked it up
1937 * and locked it. But paranoia is good.
1939 if (rval == EEXIST) {
1941 * Find the leaf entry.
1943 blk = &state->path.blk[state->path.active - 1];
1944 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1945 leaf = blk->bp->b_addr;
1946 lep = &leaf->ents[blk->index];
1947 ASSERT(state->extravalid);
1949 * Point to the data entry.
1951 hdr = state->extrablk.bp->b_addr;
1952 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
1953 dep = (xfs_dir2_data_entry_t *)
1955 xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
1956 ASSERT(inum != be64_to_cpu(dep->inumber));
1958 * Fill in the new inode number and log the entry.
1960 dep->inumber = cpu_to_be64(inum);
1961 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
1965 * Didn't find it, and we're holding a data block. Drop it.
1967 else if (state->extravalid) {
1968 xfs_trans_brelse(args->trans, state->extrablk.bp);
1969 state->extrablk.bp = NULL;
1972 * Release all the buffers in the cursor.
1974 for (i = 0; i < state->path.active; i++) {
1975 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
1976 state->path.blk[i].bp = NULL;
1978 xfs_da_state_free(state);
1983 * Trim off a trailing empty freespace block.
1984 * Return (in rvalp) 1 if we did it, 0 if not.
1987 xfs_dir2_node_trim_free(
1988 xfs_da_args_t *args, /* operation arguments */
1989 xfs_fileoff_t fo, /* free block number */
1990 int *rvalp) /* out: did something */
1992 struct xfs_buf *bp; /* freespace buffer */
1993 xfs_inode_t *dp; /* incore directory inode */
1994 int error; /* error return code */
1995 xfs_dir2_free_t *free; /* freespace structure */
1996 xfs_mount_t *mp; /* filesystem mount point */
1997 xfs_trans_t *tp; /* transaction pointer */
2003 * Read the freespace block.
2005 error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2009 * There can be holes in freespace. If fo is a hole, there's
2015 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
2017 * If there are used entries, there's nothing to do.
2019 if (be32_to_cpu(free->hdr.nused) > 0) {
2020 xfs_trans_brelse(tp, bp);
2025 * Blow the block away.
2028 xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
2031 * Can't fail with ENOSPC since that only happens with no
2032 * space reservation, when breaking up an extent into two
2033 * pieces. This is the last block of an extent.
2035 ASSERT(error != ENOSPC);
2036 xfs_trans_brelse(tp, bp);
2040 * Return that we succeeded.