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_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
32 #include "xfs_dir2_priv.h"
33 #include "xfs_error.h"
34 #include "xfs_trace.h"
35 #include "xfs_trans.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_cksum.h"
40 * Function declarations.
42 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
44 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
45 xfs_da_state_blk_t *blk1,
46 xfs_da_state_blk_t *blk2);
47 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
48 int index, xfs_da_state_blk_t *dblk,
50 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
51 xfs_da_state_blk_t *fblk);
54 * Check internal consistency of a leafn block.
57 #define xfs_dir3_leaf_check(dp, bp) \
59 if (!xfs_dir3_leafn_check((dp), (bp))) \
68 struct xfs_dir2_leaf *leaf = bp->b_addr;
69 struct xfs_dir3_icleaf_hdr leafhdr;
71 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
73 if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
74 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
75 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
77 } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
80 return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
83 #define xfs_dir3_leaf_check(dp, bp)
90 struct xfs_mount *mp = bp->b_target->bt_mount;
91 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
93 if (xfs_sb_version_hascrc(&mp->m_sb)) {
94 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
96 if (hdr3->magic != cpu_to_be32(XFS_DIR3_FREE_MAGIC))
98 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
100 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
103 if (hdr->magic != cpu_to_be32(XFS_DIR2_FREE_MAGIC))
107 /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
113 xfs_dir3_free_read_verify(
116 struct xfs_mount *mp = bp->b_target->bt_mount;
118 if (xfs_sb_version_hascrc(&mp->m_sb) &&
119 !xfs_buf_verify_cksum(bp, XFS_DIR3_FREE_CRC_OFF))
120 xfs_buf_ioerror(bp, EFSBADCRC);
121 else if (!xfs_dir3_free_verify(bp))
122 xfs_buf_ioerror(bp, EFSCORRUPTED);
125 xfs_verifier_error(bp);
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_buf_ioerror(bp, EFSCORRUPTED);
138 xfs_verifier_error(bp);
142 if (!xfs_sb_version_hascrc(&mp->m_sb))
146 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
148 xfs_buf_update_cksum(bp, 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);
197 xfs_dir3_free_get_buf(
200 struct xfs_buf **bpp)
202 struct xfs_trans *tp = args->trans;
203 struct xfs_inode *dp = args->dp;
204 struct xfs_mount *mp = dp->i_mount;
207 struct xfs_dir3_icfree_hdr hdr;
209 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, fbno),
210 -1, &bp, XFS_DATA_FORK);
214 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
215 bp->b_ops = &xfs_dir3_free_buf_ops;
218 * Initialize the new block to be empty, and remember
219 * its first slot as our empty slot.
221 memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
222 memset(&hdr, 0, sizeof(hdr));
224 if (xfs_sb_version_hascrc(&mp->m_sb)) {
225 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
227 hdr.magic = XFS_DIR3_FREE_MAGIC;
229 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
230 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
231 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_uuid);
233 hdr.magic = XFS_DIR2_FREE_MAGIC;
234 dp->d_ops->free_hdr_to_disk(bp->b_addr, &hdr);
240 * Log entries from a freespace block.
243 xfs_dir2_free_log_bests(
244 struct xfs_trans *tp,
245 struct xfs_inode *dp,
247 int first, /* first entry to log */
248 int last) /* last entry to log */
250 xfs_dir2_free_t *free; /* freespace structure */
254 bests = dp->d_ops->free_bests_p(free);
255 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
256 free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
257 xfs_trans_log_buf(tp, bp,
258 (uint)((char *)&bests[first] - (char *)free),
259 (uint)((char *)&bests[last] - (char *)free +
260 sizeof(bests[0]) - 1));
264 * Log header from a freespace block.
267 xfs_dir2_free_log_header(
268 struct xfs_trans *tp,
269 struct xfs_inode *dp,
273 xfs_dir2_free_t *free; /* freespace structure */
276 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
277 free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
279 xfs_trans_log_buf(tp, bp, 0, dp->d_ops->free_hdr_size - 1);
283 * Convert a leaf-format directory to a node-format directory.
284 * We need to change the magic number of the leaf block, and copy
285 * the freespace table out of the leaf block into its own block.
288 xfs_dir2_leaf_to_node(
289 xfs_da_args_t *args, /* operation arguments */
290 struct xfs_buf *lbp) /* leaf buffer */
292 xfs_inode_t *dp; /* incore directory inode */
293 int error; /* error return value */
294 struct xfs_buf *fbp; /* freespace buffer */
295 xfs_dir2_db_t fdb; /* freespace block number */
296 xfs_dir2_free_t *free; /* freespace structure */
297 __be16 *from; /* pointer to freespace entry */
298 int i; /* leaf freespace index */
299 xfs_dir2_leaf_t *leaf; /* leaf structure */
300 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
301 xfs_mount_t *mp; /* filesystem mount point */
302 int n; /* count of live freespc ents */
303 xfs_dir2_data_off_t off; /* freespace entry value */
304 __be16 *to; /* pointer to freespace entry */
305 xfs_trans_t *tp; /* transaction pointer */
306 struct xfs_dir3_icfree_hdr freehdr;
308 trace_xfs_dir2_leaf_to_node(args);
314 * Add a freespace block to the directory.
316 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
319 ASSERT(fdb == xfs_dir2_byte_to_db(mp, XFS_DIR2_FREE_OFFSET));
321 * Get the buffer for the new freespace block.
323 error = xfs_dir3_free_get_buf(args, fdb, &fbp);
328 dp->d_ops->free_hdr_from_disk(&freehdr, free);
330 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
331 ASSERT(be32_to_cpu(ltp->bestcount) <=
332 (uint)dp->i_d.di_size / mp->m_dirblksize);
335 * Copy freespace entries from the leaf block to the new block.
336 * Count active entries.
338 from = xfs_dir2_leaf_bests_p(ltp);
339 to = dp->d_ops->free_bests_p(free);
340 for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
341 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
343 *to = cpu_to_be16(off);
347 * Now initialize the freespace block header.
350 freehdr.nvalid = be32_to_cpu(ltp->bestcount);
352 dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
353 xfs_dir2_free_log_bests(tp, dp, fbp, 0, freehdr.nvalid - 1);
354 xfs_dir2_free_log_header(tp, dp, fbp);
357 * Converting the leaf to a leafnode is just a matter of changing the
358 * magic number and the ops. Do the change directly to the buffer as
359 * it's less work (and less code) than decoding the header to host
360 * format and back again.
362 if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
363 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
365 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
366 lbp->b_ops = &xfs_dir3_leafn_buf_ops;
367 xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
368 xfs_dir3_leaf_log_header(tp, dp, lbp);
369 xfs_dir3_leaf_check(dp, lbp);
374 * Add a leaf entry to a leaf block in a node-form directory.
375 * The other work necessary is done from the caller.
377 static int /* error */
379 struct xfs_buf *bp, /* leaf buffer */
380 xfs_da_args_t *args, /* operation arguments */
381 int index) /* insertion pt for new entry */
383 int compact; /* compacting stale leaves */
384 xfs_inode_t *dp; /* incore directory inode */
385 int highstale; /* next stale entry */
386 xfs_dir2_leaf_t *leaf; /* leaf structure */
387 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
388 int lfloghigh; /* high leaf entry logging */
389 int lfloglow; /* low leaf entry logging */
390 int lowstale; /* previous stale entry */
391 xfs_mount_t *mp; /* filesystem mount point */
392 xfs_trans_t *tp; /* transaction pointer */
393 struct xfs_dir3_icleaf_hdr leafhdr;
394 struct xfs_dir2_leaf_entry *ents;
396 trace_xfs_dir2_leafn_add(args, index);
402 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
403 ents = dp->d_ops->leaf_ents_p(leaf);
406 * Quick check just to make sure we are not going to index
407 * into other peoples memory
410 return XFS_ERROR(EFSCORRUPTED);
413 * If there are already the maximum number of leaf entries in
414 * the block, if there are no stale entries it won't fit.
415 * Caller will do a split. If there are stale entries we'll do
419 if (leafhdr.count == dp->d_ops->leaf_max_ents(mp)) {
421 return XFS_ERROR(ENOSPC);
422 compact = leafhdr.stale > 1;
425 ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
426 ASSERT(index == leafhdr.count ||
427 be32_to_cpu(ents[index].hashval) >= args->hashval);
429 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
433 * Compact out all but one stale leaf entry. Leaves behind
434 * the entry closest to index.
437 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
438 &highstale, &lfloglow, &lfloghigh);
439 else if (leafhdr.stale) {
441 * Set impossible logging indices for this case.
443 lfloglow = leafhdr.count;
448 * Insert the new entry, log everything.
450 lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
451 highstale, &lfloglow, &lfloghigh);
453 lep->hashval = cpu_to_be32(args->hashval);
454 lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
455 args->blkno, args->index));
457 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
458 xfs_dir3_leaf_log_header(tp, dp, bp);
459 xfs_dir3_leaf_log_ents(tp, dp, bp, lfloglow, lfloghigh);
460 xfs_dir3_leaf_check(dp, bp);
466 xfs_dir2_free_hdr_check(
467 struct xfs_inode *dp,
471 struct xfs_dir3_icfree_hdr hdr;
473 dp->d_ops->free_hdr_from_disk(&hdr, bp->b_addr);
475 ASSERT((hdr.firstdb % dp->d_ops->free_max_bests(dp->i_mount)) == 0);
476 ASSERT(hdr.firstdb <= db);
477 ASSERT(db < hdr.firstdb + hdr.nvalid);
480 #define xfs_dir2_free_hdr_check(dp, bp, db)
484 * Return the last hash value in the leaf.
485 * Stale entries are ok.
487 xfs_dahash_t /* hash value */
488 xfs_dir2_leafn_lasthash(
489 struct xfs_inode *dp,
490 struct xfs_buf *bp, /* leaf buffer */
491 int *count) /* count of entries in leaf */
493 struct xfs_dir2_leaf *leaf = bp->b_addr;
494 struct xfs_dir2_leaf_entry *ents;
495 struct xfs_dir3_icleaf_hdr leafhdr;
497 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
499 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
500 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
503 *count = leafhdr.count;
507 ents = dp->d_ops->leaf_ents_p(leaf);
508 return be32_to_cpu(ents[leafhdr.count - 1].hashval);
512 * Look up a leaf entry for space to add a name in a node-format leaf block.
513 * The extrablk in state is a freespace block.
516 xfs_dir2_leafn_lookup_for_addname(
517 struct xfs_buf *bp, /* leaf buffer */
518 xfs_da_args_t *args, /* operation arguments */
519 int *indexp, /* out: leaf entry index */
520 xfs_da_state_t *state) /* state to fill in */
522 struct xfs_buf *curbp = NULL; /* current data/free buffer */
523 xfs_dir2_db_t curdb = -1; /* current data block number */
524 xfs_dir2_db_t curfdb = -1; /* current free block number */
525 xfs_inode_t *dp; /* incore directory inode */
526 int error; /* error return value */
527 int fi; /* free entry index */
528 xfs_dir2_free_t *free = NULL; /* free block structure */
529 int index; /* leaf entry index */
530 xfs_dir2_leaf_t *leaf; /* leaf structure */
531 int length; /* length of new data entry */
532 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
533 xfs_mount_t *mp; /* filesystem mount point */
534 xfs_dir2_db_t newdb; /* new data block number */
535 xfs_dir2_db_t newfdb; /* new free block number */
536 xfs_trans_t *tp; /* transaction pointer */
537 struct xfs_dir2_leaf_entry *ents;
538 struct xfs_dir3_icleaf_hdr leafhdr;
544 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
545 ents = dp->d_ops->leaf_ents_p(leaf);
547 xfs_dir3_leaf_check(dp, bp);
548 ASSERT(leafhdr.count > 0);
551 * Look up the hash value in the leaf entries.
553 index = xfs_dir2_leaf_search_hash(args, bp);
555 * Do we have a buffer coming in?
557 if (state->extravalid) {
558 /* If so, it's a free block buffer, get the block number. */
559 curbp = state->extrablk.bp;
560 curfdb = state->extrablk.blkno;
561 free = curbp->b_addr;
562 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
563 free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
565 length = dp->d_ops->data_entsize(args->namelen);
567 * Loop over leaf entries with the right hash value.
569 for (lep = &ents[index];
570 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
573 * Skip stale leaf entries.
575 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
578 * Pull the data block number from the entry.
580 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
582 * For addname, we're looking for a place to put the new entry.
583 * We want to use a data block with an entry of equal
584 * hash value to ours if there is one with room.
586 * If this block isn't the data block we already have
587 * in hand, take a look at it.
589 if (newdb != curdb) {
594 * Convert the data block to the free block
595 * holding its freespace information.
597 newfdb = dp->d_ops->db_to_fdb(mp, newdb);
599 * If it's not the one we have in hand, read it in.
601 if (newfdb != curfdb) {
603 * If we had one before, drop it.
606 xfs_trans_brelse(tp, curbp);
608 error = xfs_dir2_free_read(tp, dp,
609 xfs_dir2_db_to_da(args->geo,
614 free = curbp->b_addr;
616 xfs_dir2_free_hdr_check(dp, curbp, curdb);
619 * Get the index for our entry.
621 fi = dp->d_ops->db_to_fdindex(mp, curdb);
623 * If it has room, return it.
625 bests = dp->d_ops->free_bests_p(free);
626 if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
627 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
628 XFS_ERRLEVEL_LOW, mp);
629 if (curfdb != newfdb)
630 xfs_trans_brelse(tp, curbp);
631 return XFS_ERROR(EFSCORRUPTED);
634 if (be16_to_cpu(bests[fi]) >= length)
638 /* Didn't find any space */
641 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
643 /* Giving back a free block. */
644 state->extravalid = 1;
645 state->extrablk.bp = curbp;
646 state->extrablk.index = fi;
647 state->extrablk.blkno = curfdb;
650 * Important: this magic number is not in the buffer - it's for
651 * buffer type information and therefore only the free/data type
652 * matters here, not whether CRCs are enabled or not.
654 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
656 state->extravalid = 0;
659 * Return the index, that will be the insertion point.
662 return XFS_ERROR(ENOENT);
666 * Look up a leaf entry in a node-format leaf block.
667 * The extrablk in state a data block.
670 xfs_dir2_leafn_lookup_for_entry(
671 struct xfs_buf *bp, /* leaf buffer */
672 xfs_da_args_t *args, /* operation arguments */
673 int *indexp, /* out: leaf entry index */
674 xfs_da_state_t *state) /* state to fill in */
676 struct xfs_buf *curbp = NULL; /* current data/free buffer */
677 xfs_dir2_db_t curdb = -1; /* current data block number */
678 xfs_dir2_data_entry_t *dep; /* data block entry */
679 xfs_inode_t *dp; /* incore directory inode */
680 int error; /* error return value */
681 int index; /* leaf entry index */
682 xfs_dir2_leaf_t *leaf; /* leaf structure */
683 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
684 xfs_mount_t *mp; /* filesystem mount point */
685 xfs_dir2_db_t newdb; /* new data block number */
686 xfs_trans_t *tp; /* transaction pointer */
687 enum xfs_dacmp cmp; /* comparison result */
688 struct xfs_dir2_leaf_entry *ents;
689 struct xfs_dir3_icleaf_hdr leafhdr;
695 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
696 ents = dp->d_ops->leaf_ents_p(leaf);
698 xfs_dir3_leaf_check(dp, bp);
699 ASSERT(leafhdr.count > 0);
702 * Look up the hash value in the leaf entries.
704 index = xfs_dir2_leaf_search_hash(args, bp);
706 * Do we have a buffer coming in?
708 if (state->extravalid) {
709 curbp = state->extrablk.bp;
710 curdb = state->extrablk.blkno;
713 * Loop over leaf entries with the right hash value.
715 for (lep = &ents[index];
716 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
719 * Skip stale leaf entries.
721 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
724 * Pull the data block number from the entry.
726 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
728 * Not adding a new entry, so we really want to find
729 * the name given to us.
731 * If it's a different data block, go get it.
733 if (newdb != curdb) {
735 * If we had a block before that we aren't saving
736 * for a CI name, drop it
738 if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
739 curdb != state->extrablk.blkno))
740 xfs_trans_brelse(tp, curbp);
742 * If needing the block that is saved with a CI match,
743 * use it otherwise read in the new data block.
745 if (args->cmpresult != XFS_CMP_DIFFERENT &&
746 newdb == state->extrablk.blkno) {
747 ASSERT(state->extravalid);
748 curbp = state->extrablk.bp;
750 error = xfs_dir3_data_read(tp, dp,
751 xfs_dir2_db_to_da(args->geo,
757 xfs_dir3_data_check(dp, curbp);
761 * Point to the data entry.
763 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
764 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
766 * Compare the entry and if it's an exact match, return
767 * EEXIST immediately. If it's the first case-insensitive
768 * match, store the block & inode number and continue looking.
770 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
771 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
772 /* If there is a CI match block, drop it */
773 if (args->cmpresult != XFS_CMP_DIFFERENT &&
774 curdb != state->extrablk.blkno)
775 xfs_trans_brelse(tp, state->extrablk.bp);
776 args->cmpresult = cmp;
777 args->inumber = be64_to_cpu(dep->inumber);
778 args->filetype = dp->d_ops->data_get_ftype(dep);
780 state->extravalid = 1;
781 state->extrablk.bp = curbp;
782 state->extrablk.blkno = curdb;
783 state->extrablk.index = (int)((char *)dep -
784 (char *)curbp->b_addr);
785 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
786 curbp->b_ops = &xfs_dir3_data_buf_ops;
787 xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
788 if (cmp == XFS_CMP_EXACT)
789 return XFS_ERROR(EEXIST);
792 ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
794 if (args->cmpresult == XFS_CMP_DIFFERENT) {
795 /* Giving back last used data block. */
796 state->extravalid = 1;
797 state->extrablk.bp = curbp;
798 state->extrablk.index = -1;
799 state->extrablk.blkno = curdb;
800 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
801 curbp->b_ops = &xfs_dir3_data_buf_ops;
802 xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
804 /* If the curbp is not the CI match block, drop it */
805 if (state->extrablk.bp != curbp)
806 xfs_trans_brelse(tp, curbp);
809 state->extravalid = 0;
812 return XFS_ERROR(ENOENT);
816 * Look up a leaf entry in a node-format leaf block.
817 * If this is an addname then the extrablk in state is a freespace block,
818 * otherwise it's a data block.
821 xfs_dir2_leafn_lookup_int(
822 struct xfs_buf *bp, /* leaf buffer */
823 xfs_da_args_t *args, /* operation arguments */
824 int *indexp, /* out: leaf entry index */
825 xfs_da_state_t *state) /* state to fill in */
827 if (args->op_flags & XFS_DA_OP_ADDNAME)
828 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
830 return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
834 * Move count leaf entries from source to destination leaf.
835 * Log entries and headers. Stale entries are preserved.
838 xfs_dir3_leafn_moveents(
839 xfs_da_args_t *args, /* operation arguments */
840 struct xfs_buf *bp_s, /* source */
841 struct xfs_dir3_icleaf_hdr *shdr,
842 struct xfs_dir2_leaf_entry *sents,
843 int start_s,/* source leaf index */
844 struct xfs_buf *bp_d, /* destination */
845 struct xfs_dir3_icleaf_hdr *dhdr,
846 struct xfs_dir2_leaf_entry *dents,
847 int start_d,/* destination leaf index */
848 int count) /* count of leaves to copy */
850 struct xfs_trans *tp = args->trans;
851 int stale; /* count stale leaves copied */
853 trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
856 * Silently return if nothing to do.
862 * If the destination index is not the end of the current
863 * destination leaf entries, open up a hole in the destination
864 * to hold the new entries.
866 if (start_d < dhdr->count) {
867 memmove(&dents[start_d + count], &dents[start_d],
868 (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
869 xfs_dir3_leaf_log_ents(tp, args->dp, bp_d, start_d + count,
870 count + dhdr->count - 1);
873 * If the source has stale leaves, count the ones in the copy range
874 * so we can update the header correctly.
877 int i; /* temp leaf index */
879 for (i = start_s, stale = 0; i < start_s + count; i++) {
880 if (sents[i].address ==
881 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
887 * Copy the leaf entries from source to destination.
889 memcpy(&dents[start_d], &sents[start_s],
890 count * sizeof(xfs_dir2_leaf_entry_t));
891 xfs_dir3_leaf_log_ents(tp, args->dp, bp_d,
892 start_d, start_d + count - 1);
895 * If there are source entries after the ones we copied,
896 * delete the ones we copied by sliding the next ones down.
898 if (start_s + count < shdr->count) {
899 memmove(&sents[start_s], &sents[start_s + count],
900 count * sizeof(xfs_dir2_leaf_entry_t));
901 xfs_dir3_leaf_log_ents(tp, args->dp, bp_s,
902 start_s, start_s + count - 1);
906 * Update the headers and log them.
908 shdr->count -= count;
909 shdr->stale -= stale;
910 dhdr->count += count;
911 dhdr->stale += stale;
915 * Determine the sort order of two leaf blocks.
916 * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
919 xfs_dir2_leafn_order(
920 struct xfs_inode *dp,
921 struct xfs_buf *leaf1_bp, /* leaf1 buffer */
922 struct xfs_buf *leaf2_bp) /* leaf2 buffer */
924 struct xfs_dir2_leaf *leaf1 = leaf1_bp->b_addr;
925 struct xfs_dir2_leaf *leaf2 = leaf2_bp->b_addr;
926 struct xfs_dir2_leaf_entry *ents1;
927 struct xfs_dir2_leaf_entry *ents2;
928 struct xfs_dir3_icleaf_hdr hdr1;
929 struct xfs_dir3_icleaf_hdr hdr2;
931 dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
932 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
933 ents1 = dp->d_ops->leaf_ents_p(leaf1);
934 ents2 = dp->d_ops->leaf_ents_p(leaf2);
936 if (hdr1.count > 0 && hdr2.count > 0 &&
937 (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
938 be32_to_cpu(ents2[hdr2.count - 1].hashval) <
939 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
945 * Rebalance leaf entries between two leaf blocks.
946 * This is actually only called when the second block is new,
947 * though the code deals with the general case.
948 * A new entry will be inserted in one of the blocks, and that
949 * entry is taken into account when balancing.
952 xfs_dir2_leafn_rebalance(
953 xfs_da_state_t *state, /* btree cursor */
954 xfs_da_state_blk_t *blk1, /* first btree block */
955 xfs_da_state_blk_t *blk2) /* second btree block */
957 xfs_da_args_t *args; /* operation arguments */
958 int count; /* count (& direction) leaves */
959 int isleft; /* new goes in left leaf */
960 xfs_dir2_leaf_t *leaf1; /* first leaf structure */
961 xfs_dir2_leaf_t *leaf2; /* second leaf structure */
962 int mid; /* midpoint leaf index */
963 #if defined(DEBUG) || defined(XFS_WARN)
964 int oldstale; /* old count of stale leaves */
966 int oldsum; /* old total leaf count */
967 int swap; /* swapped leaf blocks */
968 struct xfs_dir2_leaf_entry *ents1;
969 struct xfs_dir2_leaf_entry *ents2;
970 struct xfs_dir3_icleaf_hdr hdr1;
971 struct xfs_dir3_icleaf_hdr hdr2;
972 struct xfs_inode *dp = state->args->dp;
976 * If the block order is wrong, swap the arguments.
978 if ((swap = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp))) {
979 xfs_da_state_blk_t *tmp; /* temp for block swap */
985 leaf1 = blk1->bp->b_addr;
986 leaf2 = blk2->bp->b_addr;
987 dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
988 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
989 ents1 = dp->d_ops->leaf_ents_p(leaf1);
990 ents2 = dp->d_ops->leaf_ents_p(leaf2);
992 oldsum = hdr1.count + hdr2.count;
993 #if defined(DEBUG) || defined(XFS_WARN)
994 oldstale = hdr1.stale + hdr2.stale;
999 * If the old leaf count was odd then the new one will be even,
1000 * so we need to divide the new count evenly.
1003 xfs_dahash_t midhash; /* middle entry hash value */
1005 if (mid >= hdr1.count)
1006 midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1008 midhash = be32_to_cpu(ents1[mid].hashval);
1009 isleft = args->hashval <= midhash;
1012 * If the old count is even then the new count is odd, so there's
1013 * no preferred side for the new entry.
1014 * Pick the left one.
1019 * Calculate moved entry count. Positive means left-to-right,
1020 * negative means right-to-left. Then move the entries.
1022 count = hdr1.count - mid + (isleft == 0);
1024 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1025 hdr1.count - count, blk2->bp,
1026 &hdr2, ents2, 0, count);
1028 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1029 blk1->bp, &hdr1, ents1,
1032 ASSERT(hdr1.count + hdr2.count == oldsum);
1033 ASSERT(hdr1.stale + hdr2.stale == oldstale);
1035 /* log the changes made when moving the entries */
1036 dp->d_ops->leaf_hdr_to_disk(leaf1, &hdr1);
1037 dp->d_ops->leaf_hdr_to_disk(leaf2, &hdr2);
1038 xfs_dir3_leaf_log_header(args->trans, dp, blk1->bp);
1039 xfs_dir3_leaf_log_header(args->trans, dp, blk2->bp);
1041 xfs_dir3_leaf_check(dp, blk1->bp);
1042 xfs_dir3_leaf_check(dp, blk2->bp);
1045 * Mark whether we're inserting into the old or new leaf.
1047 if (hdr1.count < hdr2.count)
1048 state->inleaf = swap;
1049 else if (hdr1.count > hdr2.count)
1050 state->inleaf = !swap;
1052 state->inleaf = swap ^ (blk1->index <= hdr1.count);
1054 * Adjust the expected index for insertion.
1057 blk2->index = blk1->index - hdr1.count;
1060 * Finally sanity check just to make sure we are not returning a
1063 if (blk2->index < 0) {
1066 xfs_alert(dp->i_mount,
1067 "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1068 __func__, blk1->index);
1073 xfs_dir3_data_block_free(
1074 xfs_da_args_t *args,
1075 struct xfs_dir2_data_hdr *hdr,
1076 struct xfs_dir2_free *free,
1079 struct xfs_buf *fbp,
1082 struct xfs_trans *tp = args->trans;
1085 struct xfs_dir3_icfree_hdr freehdr;
1086 struct xfs_inode *dp = args->dp;
1088 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1089 bests = dp->d_ops->free_bests_p(free);
1092 * Data block is not empty, just set the free entry to the new
1095 bests[findex] = cpu_to_be16(longest);
1096 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
1100 /* One less used entry in the free table. */
1104 * If this was the last entry in the table, we can trim the table size
1105 * back. There might be other entries at the end referring to
1106 * non-existent data blocks, get those too.
1108 if (findex == freehdr.nvalid - 1) {
1109 int i; /* free entry index */
1111 for (i = findex - 1; i >= 0; i--) {
1112 if (bests[i] != cpu_to_be16(NULLDATAOFF))
1115 freehdr.nvalid = i + 1;
1118 /* Not the last entry, just punch it out. */
1119 bests[findex] = cpu_to_be16(NULLDATAOFF);
1123 dp->d_ops->free_hdr_to_disk(free, &freehdr);
1124 xfs_dir2_free_log_header(tp, dp, fbp);
1127 * If there are no useful entries left in the block, get rid of the
1130 if (!freehdr.nused) {
1133 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1137 } else if (error != ENOSPC || args->total != 0)
1140 * It's possible to get ENOSPC if there is no
1141 * space reservation. In this case some one
1142 * else will eventually get rid of this block.
1146 /* Log the free entry that changed, unless we got rid of it. */
1148 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
1153 * Remove an entry from a node directory.
1154 * This removes the leaf entry and the data entry,
1155 * and updates the free block if necessary.
1157 static int /* error */
1158 xfs_dir2_leafn_remove(
1159 xfs_da_args_t *args, /* operation arguments */
1160 struct xfs_buf *bp, /* leaf buffer */
1161 int index, /* leaf entry index */
1162 xfs_da_state_blk_t *dblk, /* data block */
1163 int *rval) /* resulting block needs join */
1165 xfs_dir2_data_hdr_t *hdr; /* data block header */
1166 xfs_dir2_db_t db; /* data block number */
1167 struct xfs_buf *dbp; /* data block buffer */
1168 xfs_dir2_data_entry_t *dep; /* data block entry */
1169 xfs_inode_t *dp; /* incore directory inode */
1170 xfs_dir2_leaf_t *leaf; /* leaf structure */
1171 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1172 int longest; /* longest data free entry */
1173 int off; /* data block entry offset */
1174 xfs_mount_t *mp; /* filesystem mount point */
1175 int needlog; /* need to log data header */
1176 int needscan; /* need to rescan data frees */
1177 xfs_trans_t *tp; /* transaction pointer */
1178 struct xfs_dir2_data_free *bf; /* bestfree table */
1179 struct xfs_dir3_icleaf_hdr leafhdr;
1180 struct xfs_dir2_leaf_entry *ents;
1182 trace_xfs_dir2_leafn_remove(args, index);
1188 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1189 ents = dp->d_ops->leaf_ents_p(leaf);
1192 * Point to the entry we're removing.
1197 * Extract the data block and offset from the entry.
1199 db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1200 ASSERT(dblk->blkno == db);
1201 off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
1202 ASSERT(dblk->index == off);
1205 * Kill the leaf entry by marking it stale.
1206 * Log the leaf block changes.
1209 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1210 xfs_dir3_leaf_log_header(tp, dp, bp);
1212 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1213 xfs_dir3_leaf_log_ents(tp, dp, bp, index, index);
1216 * Make the data entry free. Keep track of the longest freespace
1217 * in the data block in case it changes.
1221 dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1222 bf = dp->d_ops->data_bestfree_p(hdr);
1223 longest = be16_to_cpu(bf[0].length);
1224 needlog = needscan = 0;
1225 xfs_dir2_data_make_free(tp, dp, dbp, off,
1226 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1228 * Rescan the data block freespaces for bestfree.
1229 * Log the data block header if needed.
1232 xfs_dir2_data_freescan(dp, hdr, &needlog);
1234 xfs_dir2_data_log_header(tp, dp, dbp);
1235 xfs_dir3_data_check(dp, dbp);
1237 * If the longest data block freespace changes, need to update
1238 * the corresponding freeblock entry.
1240 if (longest < be16_to_cpu(bf[0].length)) {
1241 int error; /* error return value */
1242 struct xfs_buf *fbp; /* freeblock buffer */
1243 xfs_dir2_db_t fdb; /* freeblock block number */
1244 int findex; /* index in freeblock entries */
1245 xfs_dir2_free_t *free; /* freeblock structure */
1248 * Convert the data block number to a free block,
1249 * read in the free block.
1251 fdb = dp->d_ops->db_to_fdb(mp, db);
1252 error = xfs_dir2_free_read(tp, dp,
1253 xfs_dir2_db_to_da(args->geo, fdb),
1260 struct xfs_dir3_icfree_hdr freehdr;
1261 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1262 ASSERT(freehdr.firstdb == dp->d_ops->free_max_bests(mp) *
1263 (fdb - xfs_dir2_byte_to_db(mp, XFS_DIR2_FREE_OFFSET)));
1267 * Calculate which entry we need to fix.
1269 findex = dp->d_ops->db_to_fdindex(mp, db);
1270 longest = be16_to_cpu(bf[0].length);
1272 * If the data block is now empty we can get rid of it
1275 if (longest == mp->m_dirblksize -
1276 dp->d_ops->data_entry_offset) {
1278 * Try to punch out the data block.
1280 error = xfs_dir2_shrink_inode(args, db, dbp);
1286 * We can get ENOSPC if there's no space reservation.
1287 * In this case just drop the buffer and some one else
1288 * will eventually get rid of the empty block.
1290 else if (!(error == ENOSPC && args->total == 0))
1294 * If we got rid of the data block, we can eliminate that entry
1295 * in the free block.
1297 error = xfs_dir3_data_block_free(args, hdr, free,
1298 fdb, findex, fbp, longest);
1303 xfs_dir3_leaf_check(dp, bp);
1305 * Return indication of whether this leaf block is empty enough
1306 * to justify trying to join it with a neighbor.
1308 *rval = (dp->d_ops->leaf_hdr_size +
1309 (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1315 * Split the leaf entries in the old block into old and new blocks.
1318 xfs_dir2_leafn_split(
1319 xfs_da_state_t *state, /* btree cursor */
1320 xfs_da_state_blk_t *oldblk, /* original block */
1321 xfs_da_state_blk_t *newblk) /* newly created block */
1323 xfs_da_args_t *args; /* operation arguments */
1324 xfs_dablk_t blkno; /* new leaf block number */
1325 int error; /* error return value */
1326 xfs_mount_t *mp; /* filesystem mount point */
1327 struct xfs_inode *dp;
1330 * Allocate space for a new leaf node.
1335 ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1336 error = xfs_da_grow_inode(args, &blkno);
1341 * Initialize the new leaf block.
1343 error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(args->geo, blkno),
1344 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1348 newblk->blkno = blkno;
1349 newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1351 * Rebalance the entries across the two leaves, link the new
1352 * block into the leaves.
1354 xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1355 error = xfs_da3_blk_link(state, oldblk, newblk);
1360 * Insert the new entry in the correct block.
1363 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1365 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1367 * Update last hashval in each block since we added the name.
1369 oldblk->hashval = xfs_dir2_leafn_lasthash(dp, oldblk->bp, NULL);
1370 newblk->hashval = xfs_dir2_leafn_lasthash(dp, newblk->bp, NULL);
1371 xfs_dir3_leaf_check(dp, oldblk->bp);
1372 xfs_dir3_leaf_check(dp, newblk->bp);
1377 * Check a leaf block and its neighbors to see if the block should be
1378 * collapsed into one or the other neighbor. Always keep the block
1379 * with the smaller block number.
1380 * If the current block is over 50% full, don't try to join it, return 0.
1381 * If the block is empty, fill in the state structure and return 2.
1382 * If it can be collapsed, fill in the state structure and return 1.
1383 * If nothing can be done, return 0.
1386 xfs_dir2_leafn_toosmall(
1387 xfs_da_state_t *state, /* btree cursor */
1388 int *action) /* resulting action to take */
1390 xfs_da_state_blk_t *blk; /* leaf block */
1391 xfs_dablk_t blkno; /* leaf block number */
1392 struct xfs_buf *bp; /* leaf buffer */
1393 int bytes; /* bytes in use */
1394 int count; /* leaf live entry count */
1395 int error; /* error return value */
1396 int forward; /* sibling block direction */
1397 int i; /* sibling counter */
1398 xfs_dir2_leaf_t *leaf; /* leaf structure */
1399 int rval; /* result from path_shift */
1400 struct xfs_dir3_icleaf_hdr leafhdr;
1401 struct xfs_dir2_leaf_entry *ents;
1402 struct xfs_inode *dp = state->args->dp;
1405 * Check for the degenerate case of the block being over 50% full.
1406 * If so, it's not worth even looking to see if we might be able
1407 * to coalesce with a sibling.
1409 blk = &state->path.blk[state->path.active - 1];
1410 leaf = blk->bp->b_addr;
1411 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1412 ents = dp->d_ops->leaf_ents_p(leaf);
1413 xfs_dir3_leaf_check(dp, blk->bp);
1415 count = leafhdr.count - leafhdr.stale;
1416 bytes = dp->d_ops->leaf_hdr_size + count * sizeof(ents[0]);
1417 if (bytes > (state->blocksize >> 1)) {
1419 * Blk over 50%, don't try to join.
1425 * Check for the degenerate case of the block being empty.
1426 * If the block is empty, we'll simply delete it, no need to
1427 * coalesce it with a sibling block. We choose (arbitrarily)
1428 * to merge with the forward block unless it is NULL.
1432 * Make altpath point to the block we want to keep and
1433 * path point to the block we want to drop (this one).
1435 forward = (leafhdr.forw != 0);
1436 memcpy(&state->altpath, &state->path, sizeof(state->path));
1437 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1441 *action = rval ? 2 : 0;
1445 * Examine each sibling block to see if we can coalesce with
1446 * at least 25% free space to spare. We need to figure out
1447 * whether to merge with the forward or the backward block.
1448 * We prefer coalescing with the lower numbered sibling so as
1449 * to shrink a directory over time.
1451 forward = leafhdr.forw < leafhdr.back;
1452 for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1453 struct xfs_dir3_icleaf_hdr hdr2;
1455 blkno = forward ? leafhdr.forw : leafhdr.back;
1459 * Read the sibling leaf block.
1461 error = xfs_dir3_leafn_read(state->args->trans, dp,
1467 * Count bytes in the two blocks combined.
1469 count = leafhdr.count - leafhdr.stale;
1470 bytes = state->blocksize - (state->blocksize >> 2);
1473 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf);
1474 ents = dp->d_ops->leaf_ents_p(leaf);
1475 count += hdr2.count - hdr2.stale;
1476 bytes -= count * sizeof(ents[0]);
1479 * Fits with at least 25% to spare.
1483 xfs_trans_brelse(state->args->trans, bp);
1486 * Didn't like either block, give up.
1494 * Make altpath point to the block we want to keep (the lower
1495 * numbered block) and path point to the block we want to drop.
1497 memcpy(&state->altpath, &state->path, sizeof(state->path));
1498 if (blkno < blk->blkno)
1499 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1502 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1507 *action = rval ? 0 : 1;
1512 * Move all the leaf entries from drop_blk to save_blk.
1513 * This is done as part of a join operation.
1516 xfs_dir2_leafn_unbalance(
1517 xfs_da_state_t *state, /* cursor */
1518 xfs_da_state_blk_t *drop_blk, /* dead block */
1519 xfs_da_state_blk_t *save_blk) /* surviving block */
1521 xfs_da_args_t *args; /* operation arguments */
1522 xfs_dir2_leaf_t *drop_leaf; /* dead leaf structure */
1523 xfs_dir2_leaf_t *save_leaf; /* surviving leaf structure */
1524 struct xfs_dir3_icleaf_hdr savehdr;
1525 struct xfs_dir3_icleaf_hdr drophdr;
1526 struct xfs_dir2_leaf_entry *sents;
1527 struct xfs_dir2_leaf_entry *dents;
1528 struct xfs_inode *dp = state->args->dp;
1531 ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1532 ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1533 drop_leaf = drop_blk->bp->b_addr;
1534 save_leaf = save_blk->bp->b_addr;
1536 dp->d_ops->leaf_hdr_from_disk(&savehdr, save_leaf);
1537 dp->d_ops->leaf_hdr_from_disk(&drophdr, drop_leaf);
1538 sents = dp->d_ops->leaf_ents_p(save_leaf);
1539 dents = dp->d_ops->leaf_ents_p(drop_leaf);
1542 * If there are any stale leaf entries, take this opportunity
1546 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1548 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1551 * Move the entries from drop to the appropriate end of save.
1553 drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1554 if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1555 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1556 save_blk->bp, &savehdr, sents, 0,
1559 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1560 save_blk->bp, &savehdr, sents,
1561 savehdr.count, drophdr.count);
1562 save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1564 /* log the changes made when moving the entries */
1565 dp->d_ops->leaf_hdr_to_disk(save_leaf, &savehdr);
1566 dp->d_ops->leaf_hdr_to_disk(drop_leaf, &drophdr);
1567 xfs_dir3_leaf_log_header(args->trans, dp, save_blk->bp);
1568 xfs_dir3_leaf_log_header(args->trans, dp, drop_blk->bp);
1570 xfs_dir3_leaf_check(dp, save_blk->bp);
1571 xfs_dir3_leaf_check(dp, drop_blk->bp);
1575 * Top-level node form directory addname routine.
1578 xfs_dir2_node_addname(
1579 xfs_da_args_t *args) /* operation arguments */
1581 xfs_da_state_blk_t *blk; /* leaf block for insert */
1582 int error; /* error return value */
1583 int rval; /* sub-return value */
1584 xfs_da_state_t *state; /* btree cursor */
1586 trace_xfs_dir2_node_addname(args);
1589 * Allocate and initialize the state (btree cursor).
1591 state = xfs_da_state_alloc();
1593 state->mp = args->dp->i_mount;
1594 state->blocksize = state->mp->m_dirblksize;
1595 state->node_ents = state->mp->m_dir_node_ents;
1597 * Look up the name. We're not supposed to find it, but
1598 * this gives us the insertion point.
1600 error = xfs_da3_node_lookup_int(state, &rval);
1603 if (rval != ENOENT) {
1607 * Add the data entry to a data block.
1608 * Extravalid is set to a freeblock found by lookup.
1610 rval = xfs_dir2_node_addname_int(args,
1611 state->extravalid ? &state->extrablk : NULL);
1615 blk = &state->path.blk[state->path.active - 1];
1616 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1618 * Add the new leaf entry.
1620 rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1623 * It worked, fix the hash values up the btree.
1625 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1626 xfs_da3_fixhashpath(state, &state->path);
1629 * It didn't work, we need to split the leaf block.
1631 if (args->total == 0) {
1632 ASSERT(rval == ENOSPC);
1636 * Split the leaf block and insert the new entry.
1638 rval = xfs_da3_split(state);
1641 xfs_da_state_free(state);
1646 * Add the data entry for a node-format directory name addition.
1647 * The leaf entry is added in xfs_dir2_leafn_add.
1648 * We may enter with a freespace block that the lookup found.
1650 static int /* error */
1651 xfs_dir2_node_addname_int(
1652 xfs_da_args_t *args, /* operation arguments */
1653 xfs_da_state_blk_t *fblk) /* optional freespace block */
1655 xfs_dir2_data_hdr_t *hdr; /* data block header */
1656 xfs_dir2_db_t dbno; /* data block number */
1657 struct xfs_buf *dbp; /* data block buffer */
1658 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1659 xfs_inode_t *dp; /* incore directory inode */
1660 xfs_dir2_data_unused_t *dup; /* data unused entry pointer */
1661 int error; /* error return value */
1662 xfs_dir2_db_t fbno; /* freespace block number */
1663 struct xfs_buf *fbp; /* freespace buffer */
1664 int findex; /* freespace entry index */
1665 xfs_dir2_free_t *free=NULL; /* freespace block structure */
1666 xfs_dir2_db_t ifbno; /* initial freespace block no */
1667 xfs_dir2_db_t lastfbno=0; /* highest freespace block no */
1668 int length; /* length of the new entry */
1669 int logfree; /* need to log free entry */
1670 xfs_mount_t *mp; /* filesystem mount point */
1671 int needlog; /* need to log data header */
1672 int needscan; /* need to rescan data frees */
1673 __be16 *tagp; /* data entry tag pointer */
1674 xfs_trans_t *tp; /* transaction pointer */
1676 struct xfs_dir3_icfree_hdr freehdr;
1677 struct xfs_dir2_data_free *bf;
1682 length = dp->d_ops->data_entsize(args->namelen);
1684 * If we came in with a freespace block that means that lookup
1685 * found an entry with our hash value. This is the freespace
1686 * block for that data entry.
1691 * Remember initial freespace block number.
1693 ifbno = fblk->blkno;
1695 findex = fblk->index;
1696 bests = dp->d_ops->free_bests_p(free);
1697 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1700 * This means the free entry showed that the data block had
1701 * space for our entry, so we remembered it.
1702 * Use that data block.
1705 ASSERT(findex < freehdr.nvalid);
1706 ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1707 ASSERT(be16_to_cpu(bests[findex]) >= length);
1708 dbno = freehdr.firstdb + findex;
1711 * The data block looked at didn't have enough room.
1712 * We'll start at the beginning of the freespace entries.
1719 * Didn't come in with a freespace block, so no data block.
1727 * If we don't have a data block yet, we're going to scan the
1728 * freespace blocks looking for one. Figure out what the
1729 * highest freespace block number is.
1732 xfs_fileoff_t fo; /* freespace block number */
1734 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK)))
1736 lastfbno = xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo);
1740 * While we haven't identified a data block, search the freeblock
1741 * data for a good data block. If we find a null freeblock entry,
1742 * indicating a hole in the data blocks, remember that.
1744 while (dbno == -1) {
1746 * If we don't have a freeblock in hand, get the next one.
1750 * Happens the first time through unless lookup gave
1751 * us a freespace block to start with.
1754 fbno = xfs_dir2_byte_to_db(mp,
1755 XFS_DIR2_FREE_OFFSET);
1757 * If it's ifbno we already looked at it.
1762 * If it's off the end we're done.
1764 if (fbno >= lastfbno)
1767 * Read the block. There can be holes in the
1768 * freespace blocks, so this might not succeed.
1769 * This should be really rare, so there's no reason
1772 error = xfs_dir2_free_try_read(tp, dp,
1773 xfs_dir2_db_to_da(args->geo, fbno),
1783 * Look at the current free entry. Is it good enough?
1785 * The bests initialisation should be where the bufer is read in
1786 * the above branch. But gcc is too stupid to realise that bests
1787 * and the freehdr are actually initialised if they are placed
1788 * there, so we have to do it here to avoid warnings. Blech.
1790 bests = dp->d_ops->free_bests_p(free);
1791 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1792 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1793 be16_to_cpu(bests[findex]) >= length)
1794 dbno = freehdr.firstdb + findex;
1797 * Are we done with the freeblock?
1799 if (++findex == freehdr.nvalid) {
1803 xfs_trans_brelse(tp, fbp);
1805 if (fblk && fblk->bp)
1811 * If we don't have a data block, we need to allocate one and make
1812 * the freespace entries refer to it.
1814 if (unlikely(dbno == -1)) {
1816 * Not allowed to allocate, return failure.
1818 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1819 return XFS_ERROR(ENOSPC);
1822 * Allocate and initialize the new data block.
1824 if (unlikely((error = xfs_dir2_grow_inode(args,
1825 XFS_DIR2_DATA_SPACE,
1827 (error = xfs_dir3_data_init(args, dbno, &dbp))))
1831 * If (somehow) we have a freespace block, get rid of it.
1834 xfs_trans_brelse(tp, fbp);
1835 if (fblk && fblk->bp)
1839 * Get the freespace block corresponding to the data block
1840 * that was just allocated.
1842 fbno = dp->d_ops->db_to_fdb(mp, dbno);
1843 error = xfs_dir2_free_try_read(tp, dp,
1844 xfs_dir2_db_to_da(args->geo, fbno),
1850 * If there wasn't a freespace block, the read will
1851 * return a NULL fbp. Allocate and initialize a new one.
1854 error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1859 if (unlikely(dp->d_ops->db_to_fdb(mp, dbno) != fbno)) {
1861 "%s: dir ino %llu needed freesp block %lld for\n"
1862 " data block %lld, got %lld ifbno %llu lastfbno %d",
1863 __func__, (unsigned long long)dp->i_ino,
1864 (long long)dp->d_ops->db_to_fdb(mp, dbno),
1865 (long long)dbno, (long long)fbno,
1866 (unsigned long long)ifbno, lastfbno);
1869 " fblk 0x%p blkno %llu index %d magic 0x%x",
1871 (unsigned long long)fblk->blkno,
1875 xfs_alert(mp, " ... fblk is NULL");
1877 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1878 XFS_ERRLEVEL_LOW, mp);
1879 return XFS_ERROR(EFSCORRUPTED);
1883 * Get a buffer for the new block.
1885 error = xfs_dir3_free_get_buf(args, fbno, &fbp);
1889 bests = dp->d_ops->free_bests_p(free);
1890 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1893 * Remember the first slot as our empty slot.
1896 (fbno - xfs_dir2_byte_to_db(mp,
1897 XFS_DIR2_FREE_OFFSET)) *
1898 dp->d_ops->free_max_bests(mp);
1901 bests = dp->d_ops->free_bests_p(free);
1902 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1906 * Set the freespace block index from the data block number.
1908 findex = dp->d_ops->db_to_fdindex(mp, dbno);
1910 * If it's after the end of the current entries in the
1911 * freespace block, extend that table.
1913 if (findex >= freehdr.nvalid) {
1914 ASSERT(findex < dp->d_ops->free_max_bests(mp));
1915 freehdr.nvalid = findex + 1;
1917 * Tag new entry so nused will go up.
1919 bests[findex] = cpu_to_be16(NULLDATAOFF);
1922 * If this entry was for an empty data block
1923 * (this should always be true) then update the header.
1925 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1927 dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
1928 xfs_dir2_free_log_header(tp, dp, fbp);
1931 * Update the real value in the table.
1932 * We haven't allocated the data entry yet so this will
1936 bf = dp->d_ops->data_bestfree_p(hdr);
1937 bests[findex] = bf[0].length;
1941 * We had a data block so we don't have to make a new one.
1945 * If just checking, we succeeded.
1947 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1951 * Read the data block in.
1953 error = xfs_dir3_data_read(tp, dp,
1954 xfs_dir2_db_to_da(args->geo, dbno),
1959 bf = dp->d_ops->data_bestfree_p(hdr);
1962 ASSERT(be16_to_cpu(bf[0].length) >= length);
1964 * Point to the existing unused space.
1966 dup = (xfs_dir2_data_unused_t *)
1967 ((char *)hdr + be16_to_cpu(bf[0].offset));
1968 needscan = needlog = 0;
1970 * Mark the first part of the unused space, inuse for us.
1972 xfs_dir2_data_use_free(tp, dp, dbp, dup,
1973 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
1974 &needlog, &needscan);
1976 * Fill in the new entry and log it.
1978 dep = (xfs_dir2_data_entry_t *)dup;
1979 dep->inumber = cpu_to_be64(args->inumber);
1980 dep->namelen = args->namelen;
1981 memcpy(dep->name, args->name, dep->namelen);
1982 dp->d_ops->data_put_ftype(dep, args->filetype);
1983 tagp = dp->d_ops->data_entry_tag_p(dep);
1984 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1985 xfs_dir2_data_log_entry(tp, dp, dbp, dep);
1987 * Rescan the block for bestfree if needed.
1990 xfs_dir2_data_freescan(dp, hdr, &needlog);
1992 * Log the data block header if needed.
1995 xfs_dir2_data_log_header(tp, dp, dbp);
1997 * If the freespace entry is now wrong, update it.
1999 bests = dp->d_ops->free_bests_p(free); /* gcc is so stupid */
2000 if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
2001 bests[findex] = bf[0].length;
2005 * Log the freespace entry if needed.
2008 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
2010 * Return the data block and offset in args, then drop the data block.
2012 args->blkno = (xfs_dablk_t)dbno;
2013 args->index = be16_to_cpu(*tagp);
2018 * Lookup an entry in a node-format directory.
2019 * All the real work happens in xfs_da3_node_lookup_int.
2020 * The only real output is the inode number of the entry.
2023 xfs_dir2_node_lookup(
2024 xfs_da_args_t *args) /* operation arguments */
2026 int error; /* error return value */
2027 int i; /* btree level */
2028 int rval; /* operation return value */
2029 xfs_da_state_t *state; /* btree cursor */
2031 trace_xfs_dir2_node_lookup(args);
2034 * Allocate and initialize the btree cursor.
2036 state = xfs_da_state_alloc();
2038 state->mp = args->dp->i_mount;
2039 state->blocksize = state->mp->m_dirblksize;
2040 state->node_ents = state->mp->m_dir_node_ents;
2042 * Fill in the path to the entry in the cursor.
2044 error = xfs_da3_node_lookup_int(state, &rval);
2047 else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
2048 /* If a CI match, dup the actual name and return EEXIST */
2049 xfs_dir2_data_entry_t *dep;
2051 dep = (xfs_dir2_data_entry_t *)
2052 ((char *)state->extrablk.bp->b_addr +
2053 state->extrablk.index);
2054 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2057 * Release the btree blocks and leaf block.
2059 for (i = 0; i < state->path.active; i++) {
2060 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2061 state->path.blk[i].bp = NULL;
2064 * Release the data block if we have it.
2066 if (state->extravalid && state->extrablk.bp) {
2067 xfs_trans_brelse(args->trans, state->extrablk.bp);
2068 state->extrablk.bp = NULL;
2070 xfs_da_state_free(state);
2075 * Remove an entry from a node-format directory.
2078 xfs_dir2_node_removename(
2079 struct xfs_da_args *args) /* operation arguments */
2081 struct xfs_da_state_blk *blk; /* leaf block */
2082 int error; /* error return value */
2083 int rval; /* operation return value */
2084 struct xfs_da_state *state; /* btree cursor */
2086 trace_xfs_dir2_node_removename(args);
2089 * Allocate and initialize the btree cursor.
2091 state = xfs_da_state_alloc();
2093 state->mp = args->dp->i_mount;
2094 state->blocksize = state->mp->m_dirblksize;
2095 state->node_ents = state->mp->m_dir_node_ents;
2097 /* Look up the entry we're deleting, set up the cursor. */
2098 error = xfs_da3_node_lookup_int(state, &rval);
2102 /* Didn't find it, upper layer screwed up. */
2103 if (rval != EEXIST) {
2108 blk = &state->path.blk[state->path.active - 1];
2109 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2110 ASSERT(state->extravalid);
2112 * Remove the leaf and data entries.
2113 * Extrablk refers to the data block.
2115 error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2116 &state->extrablk, &rval);
2120 * Fix the hash values up the btree.
2122 xfs_da3_fixhashpath(state, &state->path);
2124 * If we need to join leaf blocks, do it.
2126 if (rval && state->path.active > 1)
2127 error = xfs_da3_join(state);
2129 * If no errors so far, try conversion to leaf format.
2132 error = xfs_dir2_node_to_leaf(state);
2134 xfs_da_state_free(state);
2139 * Replace an entry's inode number in a node-format directory.
2142 xfs_dir2_node_replace(
2143 xfs_da_args_t *args) /* operation arguments */
2145 xfs_da_state_blk_t *blk; /* leaf block */
2146 xfs_dir2_data_hdr_t *hdr; /* data block header */
2147 xfs_dir2_data_entry_t *dep; /* data entry changed */
2148 int error; /* error return value */
2149 int i; /* btree level */
2150 xfs_ino_t inum; /* new inode number */
2151 xfs_dir2_leaf_t *leaf; /* leaf structure */
2152 xfs_dir2_leaf_entry_t *lep; /* leaf entry being changed */
2153 int rval; /* internal return value */
2154 xfs_da_state_t *state; /* btree cursor */
2156 trace_xfs_dir2_node_replace(args);
2159 * Allocate and initialize the btree cursor.
2161 state = xfs_da_state_alloc();
2163 state->mp = args->dp->i_mount;
2164 state->blocksize = state->mp->m_dirblksize;
2165 state->node_ents = state->mp->m_dir_node_ents;
2166 inum = args->inumber;
2168 * Lookup the entry to change in the btree.
2170 error = xfs_da3_node_lookup_int(state, &rval);
2175 * It should be found, since the vnodeops layer has looked it up
2176 * and locked it. But paranoia is good.
2178 if (rval == EEXIST) {
2179 struct xfs_dir2_leaf_entry *ents;
2181 * Find the leaf entry.
2183 blk = &state->path.blk[state->path.active - 1];
2184 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2185 leaf = blk->bp->b_addr;
2186 ents = args->dp->d_ops->leaf_ents_p(leaf);
2187 lep = &ents[blk->index];
2188 ASSERT(state->extravalid);
2190 * Point to the data entry.
2192 hdr = state->extrablk.bp->b_addr;
2193 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2194 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2195 dep = (xfs_dir2_data_entry_t *)
2197 xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
2198 ASSERT(inum != be64_to_cpu(dep->inumber));
2200 * Fill in the new inode number and log the entry.
2202 dep->inumber = cpu_to_be64(inum);
2203 args->dp->d_ops->data_put_ftype(dep, args->filetype);
2204 xfs_dir2_data_log_entry(args->trans, args->dp,
2205 state->extrablk.bp, dep);
2209 * Didn't find it, and we're holding a data block. Drop it.
2211 else if (state->extravalid) {
2212 xfs_trans_brelse(args->trans, state->extrablk.bp);
2213 state->extrablk.bp = NULL;
2216 * Release all the buffers in the cursor.
2218 for (i = 0; i < state->path.active; i++) {
2219 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2220 state->path.blk[i].bp = NULL;
2222 xfs_da_state_free(state);
2227 * Trim off a trailing empty freespace block.
2228 * Return (in rvalp) 1 if we did it, 0 if not.
2231 xfs_dir2_node_trim_free(
2232 xfs_da_args_t *args, /* operation arguments */
2233 xfs_fileoff_t fo, /* free block number */
2234 int *rvalp) /* out: did something */
2236 struct xfs_buf *bp; /* freespace buffer */
2237 xfs_inode_t *dp; /* incore directory inode */
2238 int error; /* error return code */
2239 xfs_dir2_free_t *free; /* freespace structure */
2240 xfs_mount_t *mp; /* filesystem mount point */
2241 xfs_trans_t *tp; /* transaction pointer */
2242 struct xfs_dir3_icfree_hdr freehdr;
2248 * Read the freespace block.
2250 error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2254 * There can be holes in freespace. If fo is a hole, there's
2260 dp->d_ops->free_hdr_from_disk(&freehdr, free);
2263 * If there are used entries, there's nothing to do.
2265 if (freehdr.nused > 0) {
2266 xfs_trans_brelse(tp, bp);
2271 * Blow the block away.
2273 error = xfs_dir2_shrink_inode(args,
2274 xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo), bp);
2277 * Can't fail with ENOSPC since that only happens with no
2278 * space reservation, when breaking up an extent into two
2279 * pieces. This is the last block of an extent.
2281 ASSERT(error != ENOSPC);
2282 xfs_trans_brelse(tp, bp);
2286 * Return that we succeeded.