2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_types.h"
23 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
32 #include "xfs_dir2_format.h"
34 #include "xfs_dir2_priv.h"
35 #include "xfs_error.h"
36 #include "xfs_trace.h"
37 #include "xfs_buf_item.h"
38 #include "xfs_cksum.h"
41 * Function declarations.
43 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
45 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
46 xfs_da_state_blk_t *blk1,
47 xfs_da_state_blk_t *blk2);
48 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
49 int index, xfs_da_state_blk_t *dblk,
51 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
52 xfs_da_state_blk_t *fblk);
55 * Check internal consistency of a leafn block.
58 #define xfs_dir3_leaf_check(mp, bp) \
60 if (!xfs_dir3_leafn_check((mp), (bp))) \
69 struct xfs_dir2_leaf *leaf = bp->b_addr;
70 struct xfs_dir3_icleaf_hdr leafhdr;
72 xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
74 if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
75 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
76 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
78 } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
81 return xfs_dir3_leaf_check_int(mp, &leafhdr, leaf);
84 #define xfs_dir3_leaf_check(mp, bp)
91 struct xfs_mount *mp = bp->b_target->bt_mount;
92 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
94 if (xfs_sb_version_hascrc(&mp->m_sb)) {
95 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
97 if (hdr3->magic != cpu_to_be32(XFS_DIR3_FREE_MAGIC))
99 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
101 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
104 if (hdr->magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC))
108 /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
114 xfs_dir3_free_read_verify(
117 struct xfs_mount *mp = bp->b_target->bt_mount;
119 if ((xfs_sb_version_hascrc(&mp->m_sb) &&
120 !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
121 XFS_DIR3_FREE_CRC_OFF)) ||
122 !xfs_dir3_free_verify(bp)) {
123 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
124 xfs_buf_ioerror(bp, EFSCORRUPTED);
129 xfs_dir3_free_write_verify(
132 struct xfs_mount *mp = bp->b_target->bt_mount;
133 struct xfs_buf_log_item *bip = bp->b_fspriv;
134 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
136 if (!xfs_dir3_free_verify(bp)) {
137 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
138 xfs_buf_ioerror(bp, EFSCORRUPTED);
142 if (!xfs_sb_version_hascrc(&mp->m_sb))
146 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
148 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_FREE_CRC_OFF);
151 const struct xfs_buf_ops xfs_dir3_free_buf_ops = {
152 .verify_read = xfs_dir3_free_read_verify,
153 .verify_write = xfs_dir3_free_write_verify,
158 __xfs_dir3_free_read(
159 struct xfs_trans *tp,
160 struct xfs_inode *dp,
162 xfs_daddr_t mappedbno,
163 struct xfs_buf **bpp)
167 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
168 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
170 /* try read returns without an error or *bpp if it lands in a hole */
171 if (!err && tp && *bpp)
172 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
178 struct xfs_trans *tp,
179 struct xfs_inode *dp,
181 struct xfs_buf **bpp)
183 return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
187 xfs_dir2_free_try_read(
188 struct xfs_trans *tp,
189 struct xfs_inode *dp,
191 struct xfs_buf **bpp)
193 return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
198 xfs_dir3_free_hdr_from_disk(
199 struct xfs_dir3_icfree_hdr *to,
200 struct xfs_dir2_free *from)
202 if (from->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC)) {
203 to->magic = be32_to_cpu(from->hdr.magic);
204 to->firstdb = be32_to_cpu(from->hdr.firstdb);
205 to->nvalid = be32_to_cpu(from->hdr.nvalid);
206 to->nused = be32_to_cpu(from->hdr.nused);
208 struct xfs_dir3_free_hdr *hdr3 = (struct xfs_dir3_free_hdr *)from;
210 to->magic = be32_to_cpu(hdr3->hdr.magic);
211 to->firstdb = be32_to_cpu(hdr3->firstdb);
212 to->nvalid = be32_to_cpu(hdr3->nvalid);
213 to->nused = be32_to_cpu(hdr3->nused);
216 ASSERT(to->magic == XFS_DIR2_FREE_MAGIC ||
217 to->magic == XFS_DIR3_FREE_MAGIC);
221 xfs_dir3_free_hdr_to_disk(
222 struct xfs_dir2_free *to,
223 struct xfs_dir3_icfree_hdr *from)
225 ASSERT(from->magic == XFS_DIR2_FREE_MAGIC ||
226 from->magic == XFS_DIR3_FREE_MAGIC);
228 if (from->magic == XFS_DIR2_FREE_MAGIC) {
229 to->hdr.magic = cpu_to_be32(from->magic);
230 to->hdr.firstdb = cpu_to_be32(from->firstdb);
231 to->hdr.nvalid = cpu_to_be32(from->nvalid);
232 to->hdr.nused = cpu_to_be32(from->nused);
234 struct xfs_dir3_free_hdr *hdr3 = (struct xfs_dir3_free_hdr *)to;
236 hdr3->hdr.magic = cpu_to_be32(from->magic);
237 hdr3->firstdb = cpu_to_be32(from->firstdb);
238 hdr3->nvalid = cpu_to_be32(from->nvalid);
239 hdr3->nused = cpu_to_be32(from->nused);
244 xfs_dir3_free_get_buf(
245 struct xfs_trans *tp,
246 struct xfs_inode *dp,
248 struct xfs_buf **bpp)
250 struct xfs_mount *mp = dp->i_mount;
253 struct xfs_dir3_icfree_hdr hdr;
255 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, fbno),
256 -1, &bp, XFS_DATA_FORK);
260 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
261 bp->b_ops = &xfs_dir3_free_buf_ops;
264 * Initialize the new block to be empty, and remember
265 * its first slot as our empty slot.
267 memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
268 memset(&hdr, 0, sizeof(hdr));
270 if (xfs_sb_version_hascrc(&mp->m_sb)) {
271 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
273 hdr.magic = XFS_DIR3_FREE_MAGIC;
275 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
276 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
277 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_uuid);
279 hdr.magic = XFS_DIR2_FREE_MAGIC;
280 xfs_dir3_free_hdr_to_disk(bp->b_addr, &hdr);
286 * Log entries from a freespace block.
289 xfs_dir2_free_log_bests(
290 struct xfs_trans *tp,
292 int first, /* first entry to log */
293 int last) /* last entry to log */
295 xfs_dir2_free_t *free; /* freespace structure */
299 bests = xfs_dir3_free_bests_p(tp->t_mountp, free);
300 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
301 free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
302 xfs_trans_log_buf(tp, bp,
303 (uint)((char *)&bests[first] - (char *)free),
304 (uint)((char *)&bests[last] - (char *)free +
305 sizeof(bests[0]) - 1));
309 * Log header from a freespace block.
312 xfs_dir2_free_log_header(
313 struct xfs_trans *tp,
316 xfs_dir2_free_t *free; /* freespace structure */
319 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
320 free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
321 xfs_trans_log_buf(tp, bp, 0, xfs_dir3_free_hdr_size(tp->t_mountp) - 1);
325 * Convert a leaf-format directory to a node-format directory.
326 * We need to change the magic number of the leaf block, and copy
327 * the freespace table out of the leaf block into its own block.
330 xfs_dir2_leaf_to_node(
331 xfs_da_args_t *args, /* operation arguments */
332 struct xfs_buf *lbp) /* leaf buffer */
334 xfs_inode_t *dp; /* incore directory inode */
335 int error; /* error return value */
336 struct xfs_buf *fbp; /* freespace buffer */
337 xfs_dir2_db_t fdb; /* freespace block number */
338 xfs_dir2_free_t *free; /* freespace structure */
339 __be16 *from; /* pointer to freespace entry */
340 int i; /* leaf freespace index */
341 xfs_dir2_leaf_t *leaf; /* leaf structure */
342 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
343 xfs_mount_t *mp; /* filesystem mount point */
344 int n; /* count of live freespc ents */
345 xfs_dir2_data_off_t off; /* freespace entry value */
346 __be16 *to; /* pointer to freespace entry */
347 xfs_trans_t *tp; /* transaction pointer */
348 struct xfs_dir3_icfree_hdr freehdr;
350 trace_xfs_dir2_leaf_to_node(args);
356 * Add a freespace block to the directory.
358 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
361 ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
363 * Get the buffer for the new freespace block.
365 error = xfs_dir3_free_get_buf(tp, dp, fdb, &fbp);
370 xfs_dir3_free_hdr_from_disk(&freehdr, free);
372 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
373 ASSERT(be32_to_cpu(ltp->bestcount) <=
374 (uint)dp->i_d.di_size / mp->m_dirblksize);
377 * Copy freespace entries from the leaf block to the new block.
378 * Count active entries.
380 from = xfs_dir2_leaf_bests_p(ltp);
381 to = xfs_dir3_free_bests_p(mp, free);
382 for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
383 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
385 *to = cpu_to_be16(off);
389 * Now initialize the freespace block header.
392 freehdr.nvalid = be32_to_cpu(ltp->bestcount);
394 xfs_dir3_free_hdr_to_disk(fbp->b_addr, &freehdr);
395 xfs_dir2_free_log_bests(tp, fbp, 0, freehdr.nvalid - 1);
396 xfs_dir2_free_log_header(tp, fbp);
399 * Converting the leaf to a leafnode is just a matter of changing the
400 * magic number and the ops. Do the change directly to the buffer as
401 * it's less work (and less code) than decoding the header to host
402 * format and back again.
404 if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
405 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
407 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
408 lbp->b_ops = &xfs_dir3_leafn_buf_ops;
409 xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
410 xfs_dir3_leaf_log_header(tp, lbp);
411 xfs_dir3_leaf_check(mp, lbp);
416 * Add a leaf entry to a leaf block in a node-form directory.
417 * The other work necessary is done from the caller.
419 static int /* error */
421 struct xfs_buf *bp, /* leaf buffer */
422 xfs_da_args_t *args, /* operation arguments */
423 int index) /* insertion pt for new entry */
425 int compact; /* compacting stale leaves */
426 xfs_inode_t *dp; /* incore directory inode */
427 int highstale; /* next stale entry */
428 xfs_dir2_leaf_t *leaf; /* leaf structure */
429 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
430 int lfloghigh; /* high leaf entry logging */
431 int lfloglow; /* low leaf entry logging */
432 int lowstale; /* previous stale entry */
433 xfs_mount_t *mp; /* filesystem mount point */
434 xfs_trans_t *tp; /* transaction pointer */
435 struct xfs_dir3_icleaf_hdr leafhdr;
436 struct xfs_dir2_leaf_entry *ents;
438 trace_xfs_dir2_leafn_add(args, index);
444 xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
445 ents = xfs_dir3_leaf_ents_p(leaf);
448 * Quick check just to make sure we are not going to index
449 * into other peoples memory
452 return XFS_ERROR(EFSCORRUPTED);
455 * If there are already the maximum number of leaf entries in
456 * the block, if there are no stale entries it won't fit.
457 * Caller will do a split. If there are stale entries we'll do
461 if (leafhdr.count == xfs_dir3_max_leaf_ents(mp, leaf)) {
463 return XFS_ERROR(ENOSPC);
464 compact = leafhdr.stale > 1;
467 ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
468 ASSERT(index == leafhdr.count ||
469 be32_to_cpu(ents[index].hashval) >= args->hashval);
471 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
475 * Compact out all but one stale leaf entry. Leaves behind
476 * the entry closest to index.
479 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
480 &highstale, &lfloglow, &lfloghigh);
481 else if (leafhdr.stale) {
483 * Set impossible logging indices for this case.
485 lfloglow = leafhdr.count;
490 * Insert the new entry, log everything.
492 lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
493 highstale, &lfloglow, &lfloghigh);
495 lep->hashval = cpu_to_be32(args->hashval);
496 lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
497 args->blkno, args->index));
499 xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
500 xfs_dir3_leaf_log_header(tp, bp);
501 xfs_dir3_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
502 xfs_dir3_leaf_check(mp, bp);
508 xfs_dir2_free_hdr_check(
509 struct xfs_mount *mp,
513 struct xfs_dir3_icfree_hdr hdr;
515 xfs_dir3_free_hdr_from_disk(&hdr, bp->b_addr);
517 ASSERT((hdr.firstdb % xfs_dir3_free_max_bests(mp)) == 0);
518 ASSERT(hdr.firstdb <= db);
519 ASSERT(db < hdr.firstdb + hdr.nvalid);
522 #define xfs_dir2_free_hdr_check(mp, dp, db)
526 * Return the last hash value in the leaf.
527 * Stale entries are ok.
529 xfs_dahash_t /* hash value */
530 xfs_dir2_leafn_lasthash(
531 struct xfs_buf *bp, /* leaf buffer */
532 int *count) /* count of entries in leaf */
534 struct xfs_dir2_leaf *leaf = bp->b_addr;
535 struct xfs_dir2_leaf_entry *ents;
536 struct xfs_dir3_icleaf_hdr leafhdr;
538 xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
540 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
541 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
544 *count = leafhdr.count;
548 ents = xfs_dir3_leaf_ents_p(leaf);
549 return be32_to_cpu(ents[leafhdr.count - 1].hashval);
553 * Look up a leaf entry for space to add a name in a node-format leaf block.
554 * The extrablk in state is a freespace block.
557 xfs_dir2_leafn_lookup_for_addname(
558 struct xfs_buf *bp, /* leaf buffer */
559 xfs_da_args_t *args, /* operation arguments */
560 int *indexp, /* out: leaf entry index */
561 xfs_da_state_t *state) /* state to fill in */
563 struct xfs_buf *curbp = NULL; /* current data/free buffer */
564 xfs_dir2_db_t curdb = -1; /* current data block number */
565 xfs_dir2_db_t curfdb = -1; /* current free block number */
566 xfs_inode_t *dp; /* incore directory inode */
567 int error; /* error return value */
568 int fi; /* free entry index */
569 xfs_dir2_free_t *free = NULL; /* free block structure */
570 int index; /* leaf entry index */
571 xfs_dir2_leaf_t *leaf; /* leaf structure */
572 int length; /* length of new data entry */
573 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
574 xfs_mount_t *mp; /* filesystem mount point */
575 xfs_dir2_db_t newdb; /* new data block number */
576 xfs_dir2_db_t newfdb; /* new free block number */
577 xfs_trans_t *tp; /* transaction pointer */
578 struct xfs_dir2_leaf_entry *ents;
579 struct xfs_dir3_icleaf_hdr leafhdr;
585 xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
586 ents = xfs_dir3_leaf_ents_p(leaf);
588 xfs_dir3_leaf_check(mp, bp);
589 ASSERT(leafhdr.count > 0);
592 * Look up the hash value in the leaf entries.
594 index = xfs_dir2_leaf_search_hash(args, bp);
596 * Do we have a buffer coming in?
598 if (state->extravalid) {
599 /* If so, it's a free block buffer, get the block number. */
600 curbp = state->extrablk.bp;
601 curfdb = state->extrablk.blkno;
602 free = curbp->b_addr;
603 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
604 free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
606 length = xfs_dir2_data_entsize(args->namelen);
608 * Loop over leaf entries with the right hash value.
610 for (lep = &ents[index];
611 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
614 * Skip stale leaf entries.
616 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
619 * Pull the data block number from the entry.
621 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
623 * For addname, we're looking for a place to put the new entry.
624 * We want to use a data block with an entry of equal
625 * hash value to ours if there is one with room.
627 * If this block isn't the data block we already have
628 * in hand, take a look at it.
630 if (newdb != curdb) {
635 * Convert the data block to the free block
636 * holding its freespace information.
638 newfdb = xfs_dir2_db_to_fdb(mp, newdb);
640 * If it's not the one we have in hand, read it in.
642 if (newfdb != curfdb) {
644 * If we had one before, drop it.
647 xfs_trans_brelse(tp, curbp);
649 error = xfs_dir2_free_read(tp, dp,
650 xfs_dir2_db_to_da(mp, newfdb),
654 free = curbp->b_addr;
656 xfs_dir2_free_hdr_check(mp, curbp, curdb);
659 * Get the index for our entry.
661 fi = xfs_dir2_db_to_fdindex(mp, curdb);
663 * If it has room, return it.
665 bests = xfs_dir3_free_bests_p(mp, free);
666 if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
667 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
668 XFS_ERRLEVEL_LOW, mp);
669 if (curfdb != newfdb)
670 xfs_trans_brelse(tp, curbp);
671 return XFS_ERROR(EFSCORRUPTED);
674 if (be16_to_cpu(bests[fi]) >= length)
678 /* Didn't find any space */
681 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
683 /* Giving back a free block. */
684 state->extravalid = 1;
685 state->extrablk.bp = curbp;
686 state->extrablk.index = fi;
687 state->extrablk.blkno = curfdb;
690 * Important: this magic number is not in the buffer - it's for
691 * buffer type information and therefore only the free/data type
692 * matters here, not whether CRCs are enabled or not.
694 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
696 state->extravalid = 0;
699 * Return the index, that will be the insertion point.
702 return XFS_ERROR(ENOENT);
706 * Look up a leaf entry in a node-format leaf block.
707 * The extrablk in state a data block.
710 xfs_dir2_leafn_lookup_for_entry(
711 struct xfs_buf *bp, /* leaf buffer */
712 xfs_da_args_t *args, /* operation arguments */
713 int *indexp, /* out: leaf entry index */
714 xfs_da_state_t *state) /* state to fill in */
716 struct xfs_buf *curbp = NULL; /* current data/free buffer */
717 xfs_dir2_db_t curdb = -1; /* current data block number */
718 xfs_dir2_data_entry_t *dep; /* data block entry */
719 xfs_inode_t *dp; /* incore directory inode */
720 int error; /* error return value */
721 int index; /* leaf entry index */
722 xfs_dir2_leaf_t *leaf; /* leaf structure */
723 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
724 xfs_mount_t *mp; /* filesystem mount point */
725 xfs_dir2_db_t newdb; /* new data block number */
726 xfs_trans_t *tp; /* transaction pointer */
727 enum xfs_dacmp cmp; /* comparison result */
728 struct xfs_dir2_leaf_entry *ents;
729 struct xfs_dir3_icleaf_hdr leafhdr;
735 xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
736 ents = xfs_dir3_leaf_ents_p(leaf);
738 xfs_dir3_leaf_check(mp, bp);
739 ASSERT(leafhdr.count > 0);
742 * Look up the hash value in the leaf entries.
744 index = xfs_dir2_leaf_search_hash(args, bp);
746 * Do we have a buffer coming in?
748 if (state->extravalid) {
749 curbp = state->extrablk.bp;
750 curdb = state->extrablk.blkno;
753 * Loop over leaf entries with the right hash value.
755 for (lep = &ents[index];
756 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
759 * Skip stale leaf entries.
761 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
764 * Pull the data block number from the entry.
766 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
768 * Not adding a new entry, so we really want to find
769 * the name given to us.
771 * If it's a different data block, go get it.
773 if (newdb != curdb) {
775 * If we had a block before that we aren't saving
776 * for a CI name, drop it
778 if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
779 curdb != state->extrablk.blkno))
780 xfs_trans_brelse(tp, curbp);
782 * If needing the block that is saved with a CI match,
783 * use it otherwise read in the new data block.
785 if (args->cmpresult != XFS_CMP_DIFFERENT &&
786 newdb == state->extrablk.blkno) {
787 ASSERT(state->extravalid);
788 curbp = state->extrablk.bp;
790 error = xfs_dir3_data_read(tp, dp,
791 xfs_dir2_db_to_da(mp, newdb),
796 xfs_dir3_data_check(dp, curbp);
800 * Point to the data entry.
802 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
803 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
805 * Compare the entry and if it's an exact match, return
806 * EEXIST immediately. If it's the first case-insensitive
807 * match, store the block & inode number and continue looking.
809 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
810 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
811 /* If there is a CI match block, drop it */
812 if (args->cmpresult != XFS_CMP_DIFFERENT &&
813 curdb != state->extrablk.blkno)
814 xfs_trans_brelse(tp, state->extrablk.bp);
815 args->cmpresult = cmp;
816 args->inumber = be64_to_cpu(dep->inumber);
818 state->extravalid = 1;
819 state->extrablk.bp = curbp;
820 state->extrablk.blkno = curdb;
821 state->extrablk.index = (int)((char *)dep -
822 (char *)curbp->b_addr);
823 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
824 curbp->b_ops = &xfs_dir3_data_buf_ops;
825 xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
826 if (cmp == XFS_CMP_EXACT)
827 return XFS_ERROR(EEXIST);
830 ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
832 if (args->cmpresult == XFS_CMP_DIFFERENT) {
833 /* Giving back last used data block. */
834 state->extravalid = 1;
835 state->extrablk.bp = curbp;
836 state->extrablk.index = -1;
837 state->extrablk.blkno = curdb;
838 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
839 curbp->b_ops = &xfs_dir3_data_buf_ops;
840 xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
842 /* If the curbp is not the CI match block, drop it */
843 if (state->extrablk.bp != curbp)
844 xfs_trans_brelse(tp, curbp);
847 state->extravalid = 0;
850 return XFS_ERROR(ENOENT);
854 * Look up a leaf entry in a node-format leaf block.
855 * If this is an addname then the extrablk in state is a freespace block,
856 * otherwise it's a data block.
859 xfs_dir2_leafn_lookup_int(
860 struct xfs_buf *bp, /* leaf buffer */
861 xfs_da_args_t *args, /* operation arguments */
862 int *indexp, /* out: leaf entry index */
863 xfs_da_state_t *state) /* state to fill in */
865 if (args->op_flags & XFS_DA_OP_ADDNAME)
866 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
868 return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
872 * Move count leaf entries from source to destination leaf.
873 * Log entries and headers. Stale entries are preserved.
876 xfs_dir3_leafn_moveents(
877 xfs_da_args_t *args, /* operation arguments */
878 struct xfs_buf *bp_s, /* source */
879 struct xfs_dir3_icleaf_hdr *shdr,
880 struct xfs_dir2_leaf_entry *sents,
881 int start_s,/* source leaf index */
882 struct xfs_buf *bp_d, /* destination */
883 struct xfs_dir3_icleaf_hdr *dhdr,
884 struct xfs_dir2_leaf_entry *dents,
885 int start_d,/* destination leaf index */
886 int count) /* count of leaves to copy */
888 struct xfs_trans *tp = args->trans;
889 int stale; /* count stale leaves copied */
891 trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
894 * Silently return if nothing to do.
900 * If the destination index is not the end of the current
901 * destination leaf entries, open up a hole in the destination
902 * to hold the new entries.
904 if (start_d < dhdr->count) {
905 memmove(&dents[start_d + count], &dents[start_d],
906 (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
907 xfs_dir3_leaf_log_ents(tp, bp_d, start_d + count,
908 count + dhdr->count - 1);
911 * If the source has stale leaves, count the ones in the copy range
912 * so we can update the header correctly.
915 int i; /* temp leaf index */
917 for (i = start_s, stale = 0; i < start_s + count; i++) {
918 if (sents[i].address ==
919 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
925 * Copy the leaf entries from source to destination.
927 memcpy(&dents[start_d], &sents[start_s],
928 count * sizeof(xfs_dir2_leaf_entry_t));
929 xfs_dir3_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
932 * If there are source entries after the ones we copied,
933 * delete the ones we copied by sliding the next ones down.
935 if (start_s + count < shdr->count) {
936 memmove(&sents[start_s], &sents[start_s + count],
937 count * sizeof(xfs_dir2_leaf_entry_t));
938 xfs_dir3_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
942 * Update the headers and log them.
944 shdr->count -= count;
945 shdr->stale -= stale;
946 dhdr->count += count;
947 dhdr->stale += stale;
951 * Determine the sort order of two leaf blocks.
952 * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
955 xfs_dir2_leafn_order(
956 struct xfs_buf *leaf1_bp, /* leaf1 buffer */
957 struct xfs_buf *leaf2_bp) /* leaf2 buffer */
959 struct xfs_dir2_leaf *leaf1 = leaf1_bp->b_addr;
960 struct xfs_dir2_leaf *leaf2 = leaf2_bp->b_addr;
961 struct xfs_dir2_leaf_entry *ents1;
962 struct xfs_dir2_leaf_entry *ents2;
963 struct xfs_dir3_icleaf_hdr hdr1;
964 struct xfs_dir3_icleaf_hdr hdr2;
966 xfs_dir3_leaf_hdr_from_disk(&hdr1, leaf1);
967 xfs_dir3_leaf_hdr_from_disk(&hdr2, leaf2);
968 ents1 = xfs_dir3_leaf_ents_p(leaf1);
969 ents2 = xfs_dir3_leaf_ents_p(leaf2);
971 if (hdr1.count > 0 && hdr2.count > 0 &&
972 (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
973 be32_to_cpu(ents2[hdr2.count - 1].hashval) <
974 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
980 * Rebalance leaf entries between two leaf blocks.
981 * This is actually only called when the second block is new,
982 * though the code deals with the general case.
983 * A new entry will be inserted in one of the blocks, and that
984 * entry is taken into account when balancing.
987 xfs_dir2_leafn_rebalance(
988 xfs_da_state_t *state, /* btree cursor */
989 xfs_da_state_blk_t *blk1, /* first btree block */
990 xfs_da_state_blk_t *blk2) /* second btree block */
992 xfs_da_args_t *args; /* operation arguments */
993 int count; /* count (& direction) leaves */
994 int isleft; /* new goes in left leaf */
995 xfs_dir2_leaf_t *leaf1; /* first leaf structure */
996 xfs_dir2_leaf_t *leaf2; /* second leaf structure */
997 int mid; /* midpoint leaf index */
998 #if defined(DEBUG) || defined(XFS_WARN)
999 int oldstale; /* old count of stale leaves */
1001 int oldsum; /* old total leaf count */
1002 int swap; /* swapped leaf blocks */
1003 struct xfs_dir2_leaf_entry *ents1;
1004 struct xfs_dir2_leaf_entry *ents2;
1005 struct xfs_dir3_icleaf_hdr hdr1;
1006 struct xfs_dir3_icleaf_hdr hdr2;
1010 * If the block order is wrong, swap the arguments.
1012 if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
1013 xfs_da_state_blk_t *tmp; /* temp for block swap */
1019 leaf1 = blk1->bp->b_addr;
1020 leaf2 = blk2->bp->b_addr;
1021 xfs_dir3_leaf_hdr_from_disk(&hdr1, leaf1);
1022 xfs_dir3_leaf_hdr_from_disk(&hdr2, leaf2);
1023 ents1 = xfs_dir3_leaf_ents_p(leaf1);
1024 ents2 = xfs_dir3_leaf_ents_p(leaf2);
1026 oldsum = hdr1.count + hdr2.count;
1027 #if defined(DEBUG) || defined(XFS_WARN)
1028 oldstale = hdr1.stale + hdr2.stale;
1033 * If the old leaf count was odd then the new one will be even,
1034 * so we need to divide the new count evenly.
1037 xfs_dahash_t midhash; /* middle entry hash value */
1039 if (mid >= hdr1.count)
1040 midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1042 midhash = be32_to_cpu(ents1[mid].hashval);
1043 isleft = args->hashval <= midhash;
1046 * If the old count is even then the new count is odd, so there's
1047 * no preferred side for the new entry.
1048 * Pick the left one.
1053 * Calculate moved entry count. Positive means left-to-right,
1054 * negative means right-to-left. Then move the entries.
1056 count = hdr1.count - mid + (isleft == 0);
1058 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1059 hdr1.count - count, blk2->bp,
1060 &hdr2, ents2, 0, count);
1062 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1063 blk1->bp, &hdr1, ents1,
1066 ASSERT(hdr1.count + hdr2.count == oldsum);
1067 ASSERT(hdr1.stale + hdr2.stale == oldstale);
1069 /* log the changes made when moving the entries */
1070 xfs_dir3_leaf_hdr_to_disk(leaf1, &hdr1);
1071 xfs_dir3_leaf_hdr_to_disk(leaf2, &hdr2);
1072 xfs_dir3_leaf_log_header(args->trans, blk1->bp);
1073 xfs_dir3_leaf_log_header(args->trans, blk2->bp);
1075 xfs_dir3_leaf_check(args->dp->i_mount, blk1->bp);
1076 xfs_dir3_leaf_check(args->dp->i_mount, blk2->bp);
1079 * Mark whether we're inserting into the old or new leaf.
1081 if (hdr1.count < hdr2.count)
1082 state->inleaf = swap;
1083 else if (hdr1.count > hdr2.count)
1084 state->inleaf = !swap;
1086 state->inleaf = swap ^ (blk1->index <= hdr1.count);
1088 * Adjust the expected index for insertion.
1091 blk2->index = blk1->index - hdr1.count;
1094 * Finally sanity check just to make sure we are not returning a
1097 if(blk2->index < 0) {
1100 xfs_alert(args->dp->i_mount,
1101 "%s: picked the wrong leaf? reverting original leaf: blk1->index %d\n",
1102 __func__, blk1->index);
1107 xfs_dir3_data_block_free(
1108 xfs_da_args_t *args,
1109 struct xfs_dir2_data_hdr *hdr,
1110 struct xfs_dir2_free *free,
1113 struct xfs_buf *fbp,
1116 struct xfs_trans *tp = args->trans;
1119 struct xfs_dir3_icfree_hdr freehdr;
1121 xfs_dir3_free_hdr_from_disk(&freehdr, free);
1123 bests = xfs_dir3_free_bests_p(tp->t_mountp, free);
1126 * Data block is not empty, just set the free entry to the new
1129 bests[findex] = cpu_to_be16(longest);
1130 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1134 /* One less used entry in the free table. */
1138 * If this was the last entry in the table, we can trim the table size
1139 * back. There might be other entries at the end referring to
1140 * non-existent data blocks, get those too.
1142 if (findex == freehdr.nvalid - 1) {
1143 int i; /* free entry index */
1145 for (i = findex - 1; i >= 0; i--) {
1146 if (bests[i] != cpu_to_be16(NULLDATAOFF))
1149 freehdr.nvalid = i + 1;
1152 /* Not the last entry, just punch it out. */
1153 bests[findex] = cpu_to_be16(NULLDATAOFF);
1157 xfs_dir3_free_hdr_to_disk(free, &freehdr);
1158 xfs_dir2_free_log_header(tp, fbp);
1161 * If there are no useful entries left in the block, get rid of the
1164 if (!freehdr.nused) {
1167 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1171 } else if (error != ENOSPC || args->total != 0)
1174 * It's possible to get ENOSPC if there is no
1175 * space reservation. In this case some one
1176 * else will eventually get rid of this block.
1180 /* Log the free entry that changed, unless we got rid of it. */
1182 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1187 * Remove an entry from a node directory.
1188 * This removes the leaf entry and the data entry,
1189 * and updates the free block if necessary.
1191 static int /* error */
1192 xfs_dir2_leafn_remove(
1193 xfs_da_args_t *args, /* operation arguments */
1194 struct xfs_buf *bp, /* leaf buffer */
1195 int index, /* leaf entry index */
1196 xfs_da_state_blk_t *dblk, /* data block */
1197 int *rval) /* resulting block needs join */
1199 xfs_dir2_data_hdr_t *hdr; /* data block header */
1200 xfs_dir2_db_t db; /* data block number */
1201 struct xfs_buf *dbp; /* data block buffer */
1202 xfs_dir2_data_entry_t *dep; /* data block entry */
1203 xfs_inode_t *dp; /* incore directory inode */
1204 xfs_dir2_leaf_t *leaf; /* leaf structure */
1205 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1206 int longest; /* longest data free entry */
1207 int off; /* data block entry offset */
1208 xfs_mount_t *mp; /* filesystem mount point */
1209 int needlog; /* need to log data header */
1210 int needscan; /* need to rescan data frees */
1211 xfs_trans_t *tp; /* transaction pointer */
1212 struct xfs_dir2_data_free *bf; /* bestfree table */
1213 struct xfs_dir3_icleaf_hdr leafhdr;
1214 struct xfs_dir2_leaf_entry *ents;
1216 trace_xfs_dir2_leafn_remove(args, index);
1222 xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1223 ents = xfs_dir3_leaf_ents_p(leaf);
1226 * Point to the entry we're removing.
1231 * Extract the data block and offset from the entry.
1233 db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1234 ASSERT(dblk->blkno == db);
1235 off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
1236 ASSERT(dblk->index == off);
1239 * Kill the leaf entry by marking it stale.
1240 * Log the leaf block changes.
1243 xfs_dir3_leaf_hdr_to_disk(leaf, &leafhdr);
1244 xfs_dir3_leaf_log_header(tp, bp);
1246 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1247 xfs_dir3_leaf_log_ents(tp, bp, index, index);
1250 * Make the data entry free. Keep track of the longest freespace
1251 * in the data block in case it changes.
1255 dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1256 bf = xfs_dir3_data_bestfree_p(hdr);
1257 longest = be16_to_cpu(bf[0].length);
1258 needlog = needscan = 0;
1259 xfs_dir2_data_make_free(tp, dbp, off,
1260 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1262 * Rescan the data block freespaces for bestfree.
1263 * Log the data block header if needed.
1266 xfs_dir2_data_freescan(mp, hdr, &needlog);
1268 xfs_dir2_data_log_header(tp, dbp);
1269 xfs_dir3_data_check(dp, dbp);
1271 * If the longest data block freespace changes, need to update
1272 * the corresponding freeblock entry.
1274 if (longest < be16_to_cpu(bf[0].length)) {
1275 int error; /* error return value */
1276 struct xfs_buf *fbp; /* freeblock buffer */
1277 xfs_dir2_db_t fdb; /* freeblock block number */
1278 int findex; /* index in freeblock entries */
1279 xfs_dir2_free_t *free; /* freeblock structure */
1282 * Convert the data block number to a free block,
1283 * read in the free block.
1285 fdb = xfs_dir2_db_to_fdb(mp, db);
1286 error = xfs_dir2_free_read(tp, dp, xfs_dir2_db_to_da(mp, fdb),
1293 struct xfs_dir3_icfree_hdr freehdr;
1294 xfs_dir3_free_hdr_from_disk(&freehdr, free);
1295 ASSERT(freehdr.firstdb == xfs_dir3_free_max_bests(mp) *
1296 (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
1300 * Calculate which entry we need to fix.
1302 findex = xfs_dir2_db_to_fdindex(mp, db);
1303 longest = be16_to_cpu(bf[0].length);
1305 * If the data block is now empty we can get rid of it
1308 if (longest == mp->m_dirblksize -
1309 xfs_dir3_data_entry_offset(hdr)) {
1311 * Try to punch out the data block.
1313 error = xfs_dir2_shrink_inode(args, db, dbp);
1319 * We can get ENOSPC if there's no space reservation.
1320 * In this case just drop the buffer and some one else
1321 * will eventually get rid of the empty block.
1323 else if (!(error == ENOSPC && args->total == 0))
1327 * If we got rid of the data block, we can eliminate that entry
1328 * in the free block.
1330 error = xfs_dir3_data_block_free(args, hdr, free,
1331 fdb, findex, fbp, longest);
1336 xfs_dir3_leaf_check(mp, bp);
1338 * Return indication of whether this leaf block is empty enough
1339 * to justify trying to join it with a neighbor.
1341 *rval = (xfs_dir3_leaf_hdr_size(leaf) +
1342 (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1348 * Split the leaf entries in the old block into old and new blocks.
1351 xfs_dir2_leafn_split(
1352 xfs_da_state_t *state, /* btree cursor */
1353 xfs_da_state_blk_t *oldblk, /* original block */
1354 xfs_da_state_blk_t *newblk) /* newly created block */
1356 xfs_da_args_t *args; /* operation arguments */
1357 xfs_dablk_t blkno; /* new leaf block number */
1358 int error; /* error return value */
1359 xfs_mount_t *mp; /* filesystem mount point */
1362 * Allocate space for a new leaf node.
1365 mp = args->dp->i_mount;
1366 ASSERT(args != NULL);
1367 ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1368 error = xfs_da_grow_inode(args, &blkno);
1373 * Initialize the new leaf block.
1375 error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(mp, blkno),
1376 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1380 newblk->blkno = blkno;
1381 newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1383 * Rebalance the entries across the two leaves, link the new
1384 * block into the leaves.
1386 xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1387 error = xfs_da3_blk_link(state, oldblk, newblk);
1392 * Insert the new entry in the correct block.
1395 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1397 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1399 * Update last hashval in each block since we added the name.
1401 oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1402 newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1403 xfs_dir3_leaf_check(mp, oldblk->bp);
1404 xfs_dir3_leaf_check(mp, newblk->bp);
1409 * Check a leaf block and its neighbors to see if the block should be
1410 * collapsed into one or the other neighbor. Always keep the block
1411 * with the smaller block number.
1412 * If the current block is over 50% full, don't try to join it, return 0.
1413 * If the block is empty, fill in the state structure and return 2.
1414 * If it can be collapsed, fill in the state structure and return 1.
1415 * If nothing can be done, return 0.
1418 xfs_dir2_leafn_toosmall(
1419 xfs_da_state_t *state, /* btree cursor */
1420 int *action) /* resulting action to take */
1422 xfs_da_state_blk_t *blk; /* leaf block */
1423 xfs_dablk_t blkno; /* leaf block number */
1424 struct xfs_buf *bp; /* leaf buffer */
1425 int bytes; /* bytes in use */
1426 int count; /* leaf live entry count */
1427 int error; /* error return value */
1428 int forward; /* sibling block direction */
1429 int i; /* sibling counter */
1430 xfs_dir2_leaf_t *leaf; /* leaf structure */
1431 int rval; /* result from path_shift */
1432 struct xfs_dir3_icleaf_hdr leafhdr;
1433 struct xfs_dir2_leaf_entry *ents;
1436 * Check for the degenerate case of the block being over 50% full.
1437 * If so, it's not worth even looking to see if we might be able
1438 * to coalesce with a sibling.
1440 blk = &state->path.blk[state->path.active - 1];
1441 leaf = blk->bp->b_addr;
1442 xfs_dir3_leaf_hdr_from_disk(&leafhdr, leaf);
1443 ents = xfs_dir3_leaf_ents_p(leaf);
1444 xfs_dir3_leaf_check(state->args->dp->i_mount, blk->bp);
1446 count = leafhdr.count - leafhdr.stale;
1447 bytes = xfs_dir3_leaf_hdr_size(leaf) + count * sizeof(ents[0]);
1448 if (bytes > (state->blocksize >> 1)) {
1450 * Blk over 50%, don't try to join.
1456 * Check for the degenerate case of the block being empty.
1457 * If the block is empty, we'll simply delete it, no need to
1458 * coalesce it with a sibling block. We choose (arbitrarily)
1459 * to merge with the forward block unless it is NULL.
1463 * Make altpath point to the block we want to keep and
1464 * path point to the block we want to drop (this one).
1466 forward = (leafhdr.forw != 0);
1467 memcpy(&state->altpath, &state->path, sizeof(state->path));
1468 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1472 *action = rval ? 2 : 0;
1476 * Examine each sibling block to see if we can coalesce with
1477 * at least 25% free space to spare. We need to figure out
1478 * whether to merge with the forward or the backward block.
1479 * We prefer coalescing with the lower numbered sibling so as
1480 * to shrink a directory over time.
1482 forward = leafhdr.forw < leafhdr.back;
1483 for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1484 struct xfs_dir3_icleaf_hdr hdr2;
1486 blkno = forward ? leafhdr.forw : leafhdr.back;
1490 * Read the sibling leaf block.
1492 error = xfs_dir3_leafn_read(state->args->trans, state->args->dp,
1498 * Count bytes in the two blocks combined.
1500 count = leafhdr.count - leafhdr.stale;
1501 bytes = state->blocksize - (state->blocksize >> 2);
1504 xfs_dir3_leaf_hdr_from_disk(&hdr2, leaf);
1505 ents = xfs_dir3_leaf_ents_p(leaf);
1506 count += hdr2.count - hdr2.stale;
1507 bytes -= count * sizeof(ents[0]);
1510 * Fits with at least 25% to spare.
1514 xfs_trans_brelse(state->args->trans, bp);
1517 * Didn't like either block, give up.
1525 * Make altpath point to the block we want to keep (the lower
1526 * numbered block) and path point to the block we want to drop.
1528 memcpy(&state->altpath, &state->path, sizeof(state->path));
1529 if (blkno < blk->blkno)
1530 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1533 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1538 *action = rval ? 0 : 1;
1543 * Move all the leaf entries from drop_blk to save_blk.
1544 * This is done as part of a join operation.
1547 xfs_dir2_leafn_unbalance(
1548 xfs_da_state_t *state, /* cursor */
1549 xfs_da_state_blk_t *drop_blk, /* dead block */
1550 xfs_da_state_blk_t *save_blk) /* surviving block */
1552 xfs_da_args_t *args; /* operation arguments */
1553 xfs_dir2_leaf_t *drop_leaf; /* dead leaf structure */
1554 xfs_dir2_leaf_t *save_leaf; /* surviving leaf structure */
1555 struct xfs_dir3_icleaf_hdr savehdr;
1556 struct xfs_dir3_icleaf_hdr drophdr;
1557 struct xfs_dir2_leaf_entry *sents;
1558 struct xfs_dir2_leaf_entry *dents;
1561 ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1562 ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1563 drop_leaf = drop_blk->bp->b_addr;
1564 save_leaf = save_blk->bp->b_addr;
1566 xfs_dir3_leaf_hdr_from_disk(&savehdr, save_leaf);
1567 xfs_dir3_leaf_hdr_from_disk(&drophdr, drop_leaf);
1568 sents = xfs_dir3_leaf_ents_p(save_leaf);
1569 dents = xfs_dir3_leaf_ents_p(drop_leaf);
1572 * If there are any stale leaf entries, take this opportunity
1576 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1578 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1581 * Move the entries from drop to the appropriate end of save.
1583 drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1584 if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1585 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1586 save_blk->bp, &savehdr, sents, 0,
1589 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1590 save_blk->bp, &savehdr, sents,
1591 savehdr.count, drophdr.count);
1592 save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1594 /* log the changes made when moving the entries */
1595 xfs_dir3_leaf_hdr_to_disk(save_leaf, &savehdr);
1596 xfs_dir3_leaf_hdr_to_disk(drop_leaf, &drophdr);
1597 xfs_dir3_leaf_log_header(args->trans, save_blk->bp);
1598 xfs_dir3_leaf_log_header(args->trans, drop_blk->bp);
1600 xfs_dir3_leaf_check(args->dp->i_mount, save_blk->bp);
1601 xfs_dir3_leaf_check(args->dp->i_mount, drop_blk->bp);
1605 * Top-level node form directory addname routine.
1608 xfs_dir2_node_addname(
1609 xfs_da_args_t *args) /* operation arguments */
1611 xfs_da_state_blk_t *blk; /* leaf block for insert */
1612 int error; /* error return value */
1613 int rval; /* sub-return value */
1614 xfs_da_state_t *state; /* btree cursor */
1616 trace_xfs_dir2_node_addname(args);
1619 * Allocate and initialize the state (btree cursor).
1621 state = xfs_da_state_alloc();
1623 state->mp = args->dp->i_mount;
1624 state->blocksize = state->mp->m_dirblksize;
1625 state->node_ents = state->mp->m_dir_node_ents;
1627 * Look up the name. We're not supposed to find it, but
1628 * this gives us the insertion point.
1630 error = xfs_da3_node_lookup_int(state, &rval);
1633 if (rval != ENOENT) {
1637 * Add the data entry to a data block.
1638 * Extravalid is set to a freeblock found by lookup.
1640 rval = xfs_dir2_node_addname_int(args,
1641 state->extravalid ? &state->extrablk : NULL);
1645 blk = &state->path.blk[state->path.active - 1];
1646 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1648 * Add the new leaf entry.
1650 rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1653 * It worked, fix the hash values up the btree.
1655 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1656 xfs_da3_fixhashpath(state, &state->path);
1659 * It didn't work, we need to split the leaf block.
1661 if (args->total == 0) {
1662 ASSERT(rval == ENOSPC);
1666 * Split the leaf block and insert the new entry.
1668 rval = xfs_da3_split(state);
1671 xfs_da_state_free(state);
1676 * Add the data entry for a node-format directory name addition.
1677 * The leaf entry is added in xfs_dir2_leafn_add.
1678 * We may enter with a freespace block that the lookup found.
1680 static int /* error */
1681 xfs_dir2_node_addname_int(
1682 xfs_da_args_t *args, /* operation arguments */
1683 xfs_da_state_blk_t *fblk) /* optional freespace block */
1685 xfs_dir2_data_hdr_t *hdr; /* data block header */
1686 xfs_dir2_db_t dbno; /* data block number */
1687 struct xfs_buf *dbp; /* data block buffer */
1688 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1689 xfs_inode_t *dp; /* incore directory inode */
1690 xfs_dir2_data_unused_t *dup; /* data unused entry pointer */
1691 int error; /* error return value */
1692 xfs_dir2_db_t fbno; /* freespace block number */
1693 struct xfs_buf *fbp; /* freespace buffer */
1694 int findex; /* freespace entry index */
1695 xfs_dir2_free_t *free=NULL; /* freespace block structure */
1696 xfs_dir2_db_t ifbno; /* initial freespace block no */
1697 xfs_dir2_db_t lastfbno=0; /* highest freespace block no */
1698 int length; /* length of the new entry */
1699 int logfree; /* need to log free entry */
1700 xfs_mount_t *mp; /* filesystem mount point */
1701 int needlog; /* need to log data header */
1702 int needscan; /* need to rescan data frees */
1703 __be16 *tagp; /* data entry tag pointer */
1704 xfs_trans_t *tp; /* transaction pointer */
1706 struct xfs_dir3_icfree_hdr freehdr;
1707 struct xfs_dir2_data_free *bf;
1712 length = xfs_dir2_data_entsize(args->namelen);
1714 * If we came in with a freespace block that means that lookup
1715 * found an entry with our hash value. This is the freespace
1716 * block for that data entry.
1721 * Remember initial freespace block number.
1723 ifbno = fblk->blkno;
1725 findex = fblk->index;
1726 bests = xfs_dir3_free_bests_p(mp, free);
1727 xfs_dir3_free_hdr_from_disk(&freehdr, free);
1730 * This means the free entry showed that the data block had
1731 * space for our entry, so we remembered it.
1732 * Use that data block.
1735 ASSERT(findex < freehdr.nvalid);
1736 ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1737 ASSERT(be16_to_cpu(bests[findex]) >= length);
1738 dbno = freehdr.firstdb + findex;
1741 * The data block looked at didn't have enough room.
1742 * We'll start at the beginning of the freespace entries.
1749 * Didn't come in with a freespace block, so no data block.
1757 * If we don't have a data block yet, we're going to scan the
1758 * freespace blocks looking for one. Figure out what the
1759 * highest freespace block number is.
1762 xfs_fileoff_t fo; /* freespace block number */
1764 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1766 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1770 * While we haven't identified a data block, search the freeblock
1771 * data for a good data block. If we find a null freeblock entry,
1772 * indicating a hole in the data blocks, remember that.
1774 while (dbno == -1) {
1776 * If we don't have a freeblock in hand, get the next one.
1780 * Happens the first time through unless lookup gave
1781 * us a freespace block to start with.
1784 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1786 * If it's ifbno we already looked at it.
1791 * If it's off the end we're done.
1793 if (fbno >= lastfbno)
1796 * Read the block. There can be holes in the
1797 * freespace blocks, so this might not succeed.
1798 * This should be really rare, so there's no reason
1801 error = xfs_dir2_free_try_read(tp, dp,
1802 xfs_dir2_db_to_da(mp, fbno),
1812 * Look at the current free entry. Is it good enough?
1814 * The bests initialisation should be where the bufer is read in
1815 * the above branch. But gcc is too stupid to realise that bests
1816 * and the freehdr are actually initialised if they are placed
1817 * there, so we have to do it here to avoid warnings. Blech.
1819 bests = xfs_dir3_free_bests_p(mp, free);
1820 xfs_dir3_free_hdr_from_disk(&freehdr, free);
1821 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1822 be16_to_cpu(bests[findex]) >= length)
1823 dbno = freehdr.firstdb + findex;
1826 * Are we done with the freeblock?
1828 if (++findex == freehdr.nvalid) {
1832 xfs_trans_brelse(tp, fbp);
1834 if (fblk && fblk->bp)
1840 * If we don't have a data block, we need to allocate one and make
1841 * the freespace entries refer to it.
1843 if (unlikely(dbno == -1)) {
1845 * Not allowed to allocate, return failure.
1847 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1848 return XFS_ERROR(ENOSPC);
1851 * Allocate and initialize the new data block.
1853 if (unlikely((error = xfs_dir2_grow_inode(args,
1854 XFS_DIR2_DATA_SPACE,
1856 (error = xfs_dir3_data_init(args, dbno, &dbp))))
1860 * If (somehow) we have a freespace block, get rid of it.
1863 xfs_trans_brelse(tp, fbp);
1864 if (fblk && fblk->bp)
1868 * Get the freespace block corresponding to the data block
1869 * that was just allocated.
1871 fbno = xfs_dir2_db_to_fdb(mp, dbno);
1872 error = xfs_dir2_free_try_read(tp, dp,
1873 xfs_dir2_db_to_da(mp, fbno),
1879 * If there wasn't a freespace block, the read will
1880 * return a NULL fbp. Allocate and initialize a new one.
1883 error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1888 if (unlikely(xfs_dir2_db_to_fdb(mp, dbno) != fbno)) {
1890 "%s: dir ino %llu needed freesp block %lld for\n"
1891 " data block %lld, got %lld ifbno %llu lastfbno %d",
1892 __func__, (unsigned long long)dp->i_ino,
1893 (long long)xfs_dir2_db_to_fdb(mp, dbno),
1894 (long long)dbno, (long long)fbno,
1895 (unsigned long long)ifbno, lastfbno);
1898 " fblk 0x%p blkno %llu index %d magic 0x%x",
1900 (unsigned long long)fblk->blkno,
1904 xfs_alert(mp, " ... fblk is NULL");
1906 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1907 XFS_ERRLEVEL_LOW, mp);
1908 return XFS_ERROR(EFSCORRUPTED);
1912 * Get a buffer for the new block.
1914 error = xfs_dir3_free_get_buf(tp, dp, fbno, &fbp);
1918 bests = xfs_dir3_free_bests_p(mp, free);
1919 xfs_dir3_free_hdr_from_disk(&freehdr, free);
1922 * Remember the first slot as our empty slot.
1924 freehdr.firstdb = (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1925 xfs_dir3_free_max_bests(mp);
1928 bests = xfs_dir3_free_bests_p(mp, free);
1929 xfs_dir3_free_hdr_from_disk(&freehdr, free);
1933 * Set the freespace block index from the data block number.
1935 findex = xfs_dir2_db_to_fdindex(mp, dbno);
1937 * If it's after the end of the current entries in the
1938 * freespace block, extend that table.
1940 if (findex >= freehdr.nvalid) {
1941 ASSERT(findex < xfs_dir3_free_max_bests(mp));
1942 freehdr.nvalid = findex + 1;
1944 * Tag new entry so nused will go up.
1946 bests[findex] = cpu_to_be16(NULLDATAOFF);
1949 * If this entry was for an empty data block
1950 * (this should always be true) then update the header.
1952 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1954 xfs_dir3_free_hdr_to_disk(fbp->b_addr, &freehdr);
1955 xfs_dir2_free_log_header(tp, fbp);
1958 * Update the real value in the table.
1959 * We haven't allocated the data entry yet so this will
1963 bf = xfs_dir3_data_bestfree_p(hdr);
1964 bests[findex] = bf[0].length;
1968 * We had a data block so we don't have to make a new one.
1972 * If just checking, we succeeded.
1974 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1978 * Read the data block in.
1980 error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1985 bf = xfs_dir3_data_bestfree_p(hdr);
1988 ASSERT(be16_to_cpu(bf[0].length) >= length);
1990 * Point to the existing unused space.
1992 dup = (xfs_dir2_data_unused_t *)
1993 ((char *)hdr + be16_to_cpu(bf[0].offset));
1994 needscan = needlog = 0;
1996 * Mark the first part of the unused space, inuse for us.
1998 xfs_dir2_data_use_free(tp, dbp, dup,
1999 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
2000 &needlog, &needscan);
2002 * Fill in the new entry and log it.
2004 dep = (xfs_dir2_data_entry_t *)dup;
2005 dep->inumber = cpu_to_be64(args->inumber);
2006 dep->namelen = args->namelen;
2007 memcpy(dep->name, args->name, dep->namelen);
2008 tagp = xfs_dir2_data_entry_tag_p(dep);
2009 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
2010 xfs_dir2_data_log_entry(tp, dbp, dep);
2012 * Rescan the block for bestfree if needed.
2015 xfs_dir2_data_freescan(mp, hdr, &needlog);
2017 * Log the data block header if needed.
2020 xfs_dir2_data_log_header(tp, dbp);
2022 * If the freespace entry is now wrong, update it.
2024 bests = xfs_dir3_free_bests_p(mp, free); /* gcc is so stupid */
2025 if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
2026 bests[findex] = bf[0].length;
2030 * Log the freespace entry if needed.
2033 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
2035 * Return the data block and offset in args, then drop the data block.
2037 args->blkno = (xfs_dablk_t)dbno;
2038 args->index = be16_to_cpu(*tagp);
2043 * Lookup an entry in a node-format directory.
2044 * All the real work happens in xfs_da3_node_lookup_int.
2045 * The only real output is the inode number of the entry.
2048 xfs_dir2_node_lookup(
2049 xfs_da_args_t *args) /* operation arguments */
2051 int error; /* error return value */
2052 int i; /* btree level */
2053 int rval; /* operation return value */
2054 xfs_da_state_t *state; /* btree cursor */
2056 trace_xfs_dir2_node_lookup(args);
2059 * Allocate and initialize the btree cursor.
2061 state = xfs_da_state_alloc();
2063 state->mp = args->dp->i_mount;
2064 state->blocksize = state->mp->m_dirblksize;
2065 state->node_ents = state->mp->m_dir_node_ents;
2067 * Fill in the path to the entry in the cursor.
2069 error = xfs_da3_node_lookup_int(state, &rval);
2072 else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
2073 /* If a CI match, dup the actual name and return EEXIST */
2074 xfs_dir2_data_entry_t *dep;
2076 dep = (xfs_dir2_data_entry_t *)
2077 ((char *)state->extrablk.bp->b_addr +
2078 state->extrablk.index);
2079 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2082 * Release the btree blocks and leaf block.
2084 for (i = 0; i < state->path.active; i++) {
2085 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2086 state->path.blk[i].bp = NULL;
2089 * Release the data block if we have it.
2091 if (state->extravalid && state->extrablk.bp) {
2092 xfs_trans_brelse(args->trans, state->extrablk.bp);
2093 state->extrablk.bp = NULL;
2095 xfs_da_state_free(state);
2100 * Remove an entry from a node-format directory.
2103 xfs_dir2_node_removename(
2104 xfs_da_args_t *args) /* operation arguments */
2106 xfs_da_state_blk_t *blk; /* leaf block */
2107 int error; /* error return value */
2108 int rval; /* operation return value */
2109 xfs_da_state_t *state; /* btree cursor */
2111 trace_xfs_dir2_node_removename(args);
2114 * Allocate and initialize the btree cursor.
2116 state = xfs_da_state_alloc();
2118 state->mp = args->dp->i_mount;
2119 state->blocksize = state->mp->m_dirblksize;
2120 state->node_ents = state->mp->m_dir_node_ents;
2122 * Look up the entry we're deleting, set up the cursor.
2124 error = xfs_da3_node_lookup_int(state, &rval);
2128 * Didn't find it, upper layer screwed up.
2130 if (rval != EEXIST) {
2131 xfs_da_state_free(state);
2134 blk = &state->path.blk[state->path.active - 1];
2135 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2136 ASSERT(state->extravalid);
2138 * Remove the leaf and data entries.
2139 * Extrablk refers to the data block.
2141 error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2142 &state->extrablk, &rval);
2146 * Fix the hash values up the btree.
2148 xfs_da3_fixhashpath(state, &state->path);
2150 * If we need to join leaf blocks, do it.
2152 if (rval && state->path.active > 1)
2153 error = xfs_da3_join(state);
2155 * If no errors so far, try conversion to leaf format.
2158 error = xfs_dir2_node_to_leaf(state);
2159 xfs_da_state_free(state);
2164 * Replace an entry's inode number in a node-format directory.
2167 xfs_dir2_node_replace(
2168 xfs_da_args_t *args) /* operation arguments */
2170 xfs_da_state_blk_t *blk; /* leaf block */
2171 xfs_dir2_data_hdr_t *hdr; /* data block header */
2172 xfs_dir2_data_entry_t *dep; /* data entry changed */
2173 int error; /* error return value */
2174 int i; /* btree level */
2175 xfs_ino_t inum; /* new inode number */
2176 xfs_dir2_leaf_t *leaf; /* leaf structure */
2177 xfs_dir2_leaf_entry_t *lep; /* leaf entry being changed */
2178 int rval; /* internal return value */
2179 xfs_da_state_t *state; /* btree cursor */
2181 trace_xfs_dir2_node_replace(args);
2184 * Allocate and initialize the btree cursor.
2186 state = xfs_da_state_alloc();
2188 state->mp = args->dp->i_mount;
2189 state->blocksize = state->mp->m_dirblksize;
2190 state->node_ents = state->mp->m_dir_node_ents;
2191 inum = args->inumber;
2193 * Lookup the entry to change in the btree.
2195 error = xfs_da3_node_lookup_int(state, &rval);
2200 * It should be found, since the vnodeops layer has looked it up
2201 * and locked it. But paranoia is good.
2203 if (rval == EEXIST) {
2204 struct xfs_dir2_leaf_entry *ents;
2206 * Find the leaf entry.
2208 blk = &state->path.blk[state->path.active - 1];
2209 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2210 leaf = blk->bp->b_addr;
2211 ents = xfs_dir3_leaf_ents_p(leaf);
2212 lep = &ents[blk->index];
2213 ASSERT(state->extravalid);
2215 * Point to the data entry.
2217 hdr = state->extrablk.bp->b_addr;
2218 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2219 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2220 dep = (xfs_dir2_data_entry_t *)
2222 xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
2223 ASSERT(inum != be64_to_cpu(dep->inumber));
2225 * Fill in the new inode number and log the entry.
2227 dep->inumber = cpu_to_be64(inum);
2228 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
2232 * Didn't find it, and we're holding a data block. Drop it.
2234 else if (state->extravalid) {
2235 xfs_trans_brelse(args->trans, state->extrablk.bp);
2236 state->extrablk.bp = NULL;
2239 * Release all the buffers in the cursor.
2241 for (i = 0; i < state->path.active; i++) {
2242 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2243 state->path.blk[i].bp = NULL;
2245 xfs_da_state_free(state);
2250 * Trim off a trailing empty freespace block.
2251 * Return (in rvalp) 1 if we did it, 0 if not.
2254 xfs_dir2_node_trim_free(
2255 xfs_da_args_t *args, /* operation arguments */
2256 xfs_fileoff_t fo, /* free block number */
2257 int *rvalp) /* out: did something */
2259 struct xfs_buf *bp; /* freespace buffer */
2260 xfs_inode_t *dp; /* incore directory inode */
2261 int error; /* error return code */
2262 xfs_dir2_free_t *free; /* freespace structure */
2263 xfs_mount_t *mp; /* filesystem mount point */
2264 xfs_trans_t *tp; /* transaction pointer */
2265 struct xfs_dir3_icfree_hdr freehdr;
2271 * Read the freespace block.
2273 error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2277 * There can be holes in freespace. If fo is a hole, there's
2283 xfs_dir3_free_hdr_from_disk(&freehdr, free);
2286 * If there are used entries, there's nothing to do.
2288 if (freehdr.nused > 0) {
2289 xfs_trans_brelse(tp, bp);
2294 * Blow the block away.
2297 xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
2300 * Can't fail with ENOSPC since that only happens with no
2301 * space reservation, when breaking up an extent into two
2302 * pieces. This is the last block of an extent.
2304 ASSERT(error != ENOSPC);
2305 xfs_trans_brelse(tp, bp);
2309 * Return that we succeeded.