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(args->geo, 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(args->geo,
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(args->geo,
581 be32_to_cpu(lep->address));
583 * For addname, we're looking for a place to put the new entry.
584 * We want to use a data block with an entry of equal
585 * hash value to ours if there is one with room.
587 * If this block isn't the data block we already have
588 * in hand, take a look at it.
590 if (newdb != curdb) {
595 * Convert the data block to the free block
596 * holding its freespace information.
598 newfdb = dp->d_ops->db_to_fdb(mp, newdb);
600 * If it's not the one we have in hand, read it in.
602 if (newfdb != curfdb) {
604 * If we had one before, drop it.
607 xfs_trans_brelse(tp, curbp);
609 error = xfs_dir2_free_read(tp, dp,
610 xfs_dir2_db_to_da(args->geo,
615 free = curbp->b_addr;
617 xfs_dir2_free_hdr_check(dp, curbp, curdb);
620 * Get the index for our entry.
622 fi = dp->d_ops->db_to_fdindex(mp, curdb);
624 * If it has room, return it.
626 bests = dp->d_ops->free_bests_p(free);
627 if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
628 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
629 XFS_ERRLEVEL_LOW, mp);
630 if (curfdb != newfdb)
631 xfs_trans_brelse(tp, curbp);
632 return XFS_ERROR(EFSCORRUPTED);
635 if (be16_to_cpu(bests[fi]) >= length)
639 /* Didn't find any space */
642 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
644 /* Giving back a free block. */
645 state->extravalid = 1;
646 state->extrablk.bp = curbp;
647 state->extrablk.index = fi;
648 state->extrablk.blkno = curfdb;
651 * Important: this magic number is not in the buffer - it's for
652 * buffer type information and therefore only the free/data type
653 * matters here, not whether CRCs are enabled or not.
655 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
657 state->extravalid = 0;
660 * Return the index, that will be the insertion point.
663 return XFS_ERROR(ENOENT);
667 * Look up a leaf entry in a node-format leaf block.
668 * The extrablk in state a data block.
671 xfs_dir2_leafn_lookup_for_entry(
672 struct xfs_buf *bp, /* leaf buffer */
673 xfs_da_args_t *args, /* operation arguments */
674 int *indexp, /* out: leaf entry index */
675 xfs_da_state_t *state) /* state to fill in */
677 struct xfs_buf *curbp = NULL; /* current data/free buffer */
678 xfs_dir2_db_t curdb = -1; /* current data block number */
679 xfs_dir2_data_entry_t *dep; /* data block entry */
680 xfs_inode_t *dp; /* incore directory inode */
681 int error; /* error return value */
682 int index; /* leaf entry index */
683 xfs_dir2_leaf_t *leaf; /* leaf structure */
684 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
685 xfs_mount_t *mp; /* filesystem mount point */
686 xfs_dir2_db_t newdb; /* new data block number */
687 xfs_trans_t *tp; /* transaction pointer */
688 enum xfs_dacmp cmp; /* comparison result */
689 struct xfs_dir2_leaf_entry *ents;
690 struct xfs_dir3_icleaf_hdr leafhdr;
696 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
697 ents = dp->d_ops->leaf_ents_p(leaf);
699 xfs_dir3_leaf_check(dp, bp);
700 ASSERT(leafhdr.count > 0);
703 * Look up the hash value in the leaf entries.
705 index = xfs_dir2_leaf_search_hash(args, bp);
707 * Do we have a buffer coming in?
709 if (state->extravalid) {
710 curbp = state->extrablk.bp;
711 curdb = state->extrablk.blkno;
714 * Loop over leaf entries with the right hash value.
716 for (lep = &ents[index];
717 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
720 * Skip stale leaf entries.
722 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
725 * Pull the data block number from the entry.
727 newdb = xfs_dir2_dataptr_to_db(args->geo,
728 be32_to_cpu(lep->address));
730 * Not adding a new entry, so we really want to find
731 * the name given to us.
733 * If it's a different data block, go get it.
735 if (newdb != curdb) {
737 * If we had a block before that we aren't saving
738 * for a CI name, drop it
740 if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
741 curdb != state->extrablk.blkno))
742 xfs_trans_brelse(tp, curbp);
744 * If needing the block that is saved with a CI match,
745 * use it otherwise read in the new data block.
747 if (args->cmpresult != XFS_CMP_DIFFERENT &&
748 newdb == state->extrablk.blkno) {
749 ASSERT(state->extravalid);
750 curbp = state->extrablk.bp;
752 error = xfs_dir3_data_read(tp, dp,
753 xfs_dir2_db_to_da(args->geo,
759 xfs_dir3_data_check(dp, curbp);
763 * Point to the data entry.
765 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
766 xfs_dir2_dataptr_to_off(args->geo,
767 be32_to_cpu(lep->address)));
769 * Compare the entry and if it's an exact match, return
770 * EEXIST immediately. If it's the first case-insensitive
771 * match, store the block & inode number and continue looking.
773 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
774 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
775 /* If there is a CI match block, drop it */
776 if (args->cmpresult != XFS_CMP_DIFFERENT &&
777 curdb != state->extrablk.blkno)
778 xfs_trans_brelse(tp, state->extrablk.bp);
779 args->cmpresult = cmp;
780 args->inumber = be64_to_cpu(dep->inumber);
781 args->filetype = dp->d_ops->data_get_ftype(dep);
783 state->extravalid = 1;
784 state->extrablk.bp = curbp;
785 state->extrablk.blkno = curdb;
786 state->extrablk.index = (int)((char *)dep -
787 (char *)curbp->b_addr);
788 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
789 curbp->b_ops = &xfs_dir3_data_buf_ops;
790 xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
791 if (cmp == XFS_CMP_EXACT)
792 return XFS_ERROR(EEXIST);
795 ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
797 if (args->cmpresult == XFS_CMP_DIFFERENT) {
798 /* Giving back last used data block. */
799 state->extravalid = 1;
800 state->extrablk.bp = curbp;
801 state->extrablk.index = -1;
802 state->extrablk.blkno = curdb;
803 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
804 curbp->b_ops = &xfs_dir3_data_buf_ops;
805 xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
807 /* If the curbp is not the CI match block, drop it */
808 if (state->extrablk.bp != curbp)
809 xfs_trans_brelse(tp, curbp);
812 state->extravalid = 0;
815 return XFS_ERROR(ENOENT);
819 * Look up a leaf entry in a node-format leaf block.
820 * If this is an addname then the extrablk in state is a freespace block,
821 * otherwise it's a data block.
824 xfs_dir2_leafn_lookup_int(
825 struct xfs_buf *bp, /* leaf buffer */
826 xfs_da_args_t *args, /* operation arguments */
827 int *indexp, /* out: leaf entry index */
828 xfs_da_state_t *state) /* state to fill in */
830 if (args->op_flags & XFS_DA_OP_ADDNAME)
831 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
833 return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
837 * Move count leaf entries from source to destination leaf.
838 * Log entries and headers. Stale entries are preserved.
841 xfs_dir3_leafn_moveents(
842 xfs_da_args_t *args, /* operation arguments */
843 struct xfs_buf *bp_s, /* source */
844 struct xfs_dir3_icleaf_hdr *shdr,
845 struct xfs_dir2_leaf_entry *sents,
846 int start_s,/* source leaf index */
847 struct xfs_buf *bp_d, /* destination */
848 struct xfs_dir3_icleaf_hdr *dhdr,
849 struct xfs_dir2_leaf_entry *dents,
850 int start_d,/* destination leaf index */
851 int count) /* count of leaves to copy */
853 struct xfs_trans *tp = args->trans;
854 int stale; /* count stale leaves copied */
856 trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
859 * Silently return if nothing to do.
865 * If the destination index is not the end of the current
866 * destination leaf entries, open up a hole in the destination
867 * to hold the new entries.
869 if (start_d < dhdr->count) {
870 memmove(&dents[start_d + count], &dents[start_d],
871 (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
872 xfs_dir3_leaf_log_ents(tp, args->dp, bp_d, start_d + count,
873 count + dhdr->count - 1);
876 * If the source has stale leaves, count the ones in the copy range
877 * so we can update the header correctly.
880 int i; /* temp leaf index */
882 for (i = start_s, stale = 0; i < start_s + count; i++) {
883 if (sents[i].address ==
884 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
890 * Copy the leaf entries from source to destination.
892 memcpy(&dents[start_d], &sents[start_s],
893 count * sizeof(xfs_dir2_leaf_entry_t));
894 xfs_dir3_leaf_log_ents(tp, args->dp, bp_d,
895 start_d, start_d + count - 1);
898 * If there are source entries after the ones we copied,
899 * delete the ones we copied by sliding the next ones down.
901 if (start_s + count < shdr->count) {
902 memmove(&sents[start_s], &sents[start_s + count],
903 count * sizeof(xfs_dir2_leaf_entry_t));
904 xfs_dir3_leaf_log_ents(tp, args->dp, bp_s,
905 start_s, start_s + count - 1);
909 * Update the headers and log them.
911 shdr->count -= count;
912 shdr->stale -= stale;
913 dhdr->count += count;
914 dhdr->stale += stale;
918 * Determine the sort order of two leaf blocks.
919 * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
922 xfs_dir2_leafn_order(
923 struct xfs_inode *dp,
924 struct xfs_buf *leaf1_bp, /* leaf1 buffer */
925 struct xfs_buf *leaf2_bp) /* leaf2 buffer */
927 struct xfs_dir2_leaf *leaf1 = leaf1_bp->b_addr;
928 struct xfs_dir2_leaf *leaf2 = leaf2_bp->b_addr;
929 struct xfs_dir2_leaf_entry *ents1;
930 struct xfs_dir2_leaf_entry *ents2;
931 struct xfs_dir3_icleaf_hdr hdr1;
932 struct xfs_dir3_icleaf_hdr hdr2;
934 dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
935 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
936 ents1 = dp->d_ops->leaf_ents_p(leaf1);
937 ents2 = dp->d_ops->leaf_ents_p(leaf2);
939 if (hdr1.count > 0 && hdr2.count > 0 &&
940 (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
941 be32_to_cpu(ents2[hdr2.count - 1].hashval) <
942 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
948 * Rebalance leaf entries between two leaf blocks.
949 * This is actually only called when the second block is new,
950 * though the code deals with the general case.
951 * A new entry will be inserted in one of the blocks, and that
952 * entry is taken into account when balancing.
955 xfs_dir2_leafn_rebalance(
956 xfs_da_state_t *state, /* btree cursor */
957 xfs_da_state_blk_t *blk1, /* first btree block */
958 xfs_da_state_blk_t *blk2) /* second btree block */
960 xfs_da_args_t *args; /* operation arguments */
961 int count; /* count (& direction) leaves */
962 int isleft; /* new goes in left leaf */
963 xfs_dir2_leaf_t *leaf1; /* first leaf structure */
964 xfs_dir2_leaf_t *leaf2; /* second leaf structure */
965 int mid; /* midpoint leaf index */
966 #if defined(DEBUG) || defined(XFS_WARN)
967 int oldstale; /* old count of stale leaves */
969 int oldsum; /* old total leaf count */
970 int swap; /* swapped leaf blocks */
971 struct xfs_dir2_leaf_entry *ents1;
972 struct xfs_dir2_leaf_entry *ents2;
973 struct xfs_dir3_icleaf_hdr hdr1;
974 struct xfs_dir3_icleaf_hdr hdr2;
975 struct xfs_inode *dp = state->args->dp;
979 * If the block order is wrong, swap the arguments.
981 if ((swap = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp))) {
982 xfs_da_state_blk_t *tmp; /* temp for block swap */
988 leaf1 = blk1->bp->b_addr;
989 leaf2 = blk2->bp->b_addr;
990 dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
991 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
992 ents1 = dp->d_ops->leaf_ents_p(leaf1);
993 ents2 = dp->d_ops->leaf_ents_p(leaf2);
995 oldsum = hdr1.count + hdr2.count;
996 #if defined(DEBUG) || defined(XFS_WARN)
997 oldstale = hdr1.stale + hdr2.stale;
1002 * If the old leaf count was odd then the new one will be even,
1003 * so we need to divide the new count evenly.
1006 xfs_dahash_t midhash; /* middle entry hash value */
1008 if (mid >= hdr1.count)
1009 midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1011 midhash = be32_to_cpu(ents1[mid].hashval);
1012 isleft = args->hashval <= midhash;
1015 * If the old count is even then the new count is odd, so there's
1016 * no preferred side for the new entry.
1017 * Pick the left one.
1022 * Calculate moved entry count. Positive means left-to-right,
1023 * negative means right-to-left. Then move the entries.
1025 count = hdr1.count - mid + (isleft == 0);
1027 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1028 hdr1.count - count, blk2->bp,
1029 &hdr2, ents2, 0, count);
1031 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1032 blk1->bp, &hdr1, ents1,
1035 ASSERT(hdr1.count + hdr2.count == oldsum);
1036 ASSERT(hdr1.stale + hdr2.stale == oldstale);
1038 /* log the changes made when moving the entries */
1039 dp->d_ops->leaf_hdr_to_disk(leaf1, &hdr1);
1040 dp->d_ops->leaf_hdr_to_disk(leaf2, &hdr2);
1041 xfs_dir3_leaf_log_header(args->trans, dp, blk1->bp);
1042 xfs_dir3_leaf_log_header(args->trans, dp, blk2->bp);
1044 xfs_dir3_leaf_check(dp, blk1->bp);
1045 xfs_dir3_leaf_check(dp, blk2->bp);
1048 * Mark whether we're inserting into the old or new leaf.
1050 if (hdr1.count < hdr2.count)
1051 state->inleaf = swap;
1052 else if (hdr1.count > hdr2.count)
1053 state->inleaf = !swap;
1055 state->inleaf = swap ^ (blk1->index <= hdr1.count);
1057 * Adjust the expected index for insertion.
1060 blk2->index = blk1->index - hdr1.count;
1063 * Finally sanity check just to make sure we are not returning a
1066 if (blk2->index < 0) {
1069 xfs_alert(dp->i_mount,
1070 "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1071 __func__, blk1->index);
1076 xfs_dir3_data_block_free(
1077 xfs_da_args_t *args,
1078 struct xfs_dir2_data_hdr *hdr,
1079 struct xfs_dir2_free *free,
1082 struct xfs_buf *fbp,
1085 struct xfs_trans *tp = args->trans;
1088 struct xfs_dir3_icfree_hdr freehdr;
1089 struct xfs_inode *dp = args->dp;
1091 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1092 bests = dp->d_ops->free_bests_p(free);
1095 * Data block is not empty, just set the free entry to the new
1098 bests[findex] = cpu_to_be16(longest);
1099 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
1103 /* One less used entry in the free table. */
1107 * If this was the last entry in the table, we can trim the table size
1108 * back. There might be other entries at the end referring to
1109 * non-existent data blocks, get those too.
1111 if (findex == freehdr.nvalid - 1) {
1112 int i; /* free entry index */
1114 for (i = findex - 1; i >= 0; i--) {
1115 if (bests[i] != cpu_to_be16(NULLDATAOFF))
1118 freehdr.nvalid = i + 1;
1121 /* Not the last entry, just punch it out. */
1122 bests[findex] = cpu_to_be16(NULLDATAOFF);
1126 dp->d_ops->free_hdr_to_disk(free, &freehdr);
1127 xfs_dir2_free_log_header(tp, dp, fbp);
1130 * If there are no useful entries left in the block, get rid of the
1133 if (!freehdr.nused) {
1136 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1140 } else if (error != ENOSPC || args->total != 0)
1143 * It's possible to get ENOSPC if there is no
1144 * space reservation. In this case some one
1145 * else will eventually get rid of this block.
1149 /* Log the free entry that changed, unless we got rid of it. */
1151 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
1156 * Remove an entry from a node directory.
1157 * This removes the leaf entry and the data entry,
1158 * and updates the free block if necessary.
1160 static int /* error */
1161 xfs_dir2_leafn_remove(
1162 xfs_da_args_t *args, /* operation arguments */
1163 struct xfs_buf *bp, /* leaf buffer */
1164 int index, /* leaf entry index */
1165 xfs_da_state_blk_t *dblk, /* data block */
1166 int *rval) /* resulting block needs join */
1168 xfs_dir2_data_hdr_t *hdr; /* data block header */
1169 xfs_dir2_db_t db; /* data block number */
1170 struct xfs_buf *dbp; /* data block buffer */
1171 xfs_dir2_data_entry_t *dep; /* data block entry */
1172 xfs_inode_t *dp; /* incore directory inode */
1173 xfs_dir2_leaf_t *leaf; /* leaf structure */
1174 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1175 int longest; /* longest data free entry */
1176 int off; /* data block entry offset */
1177 xfs_mount_t *mp; /* filesystem mount point */
1178 int needlog; /* need to log data header */
1179 int needscan; /* need to rescan data frees */
1180 xfs_trans_t *tp; /* transaction pointer */
1181 struct xfs_dir2_data_free *bf; /* bestfree table */
1182 struct xfs_dir3_icleaf_hdr leafhdr;
1183 struct xfs_dir2_leaf_entry *ents;
1185 trace_xfs_dir2_leafn_remove(args, index);
1191 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1192 ents = dp->d_ops->leaf_ents_p(leaf);
1195 * Point to the entry we're removing.
1200 * Extract the data block and offset from the entry.
1202 db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1203 ASSERT(dblk->blkno == db);
1204 off = xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address));
1205 ASSERT(dblk->index == off);
1208 * Kill the leaf entry by marking it stale.
1209 * Log the leaf block changes.
1212 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1213 xfs_dir3_leaf_log_header(tp, dp, bp);
1215 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1216 xfs_dir3_leaf_log_ents(tp, dp, bp, index, index);
1219 * Make the data entry free. Keep track of the longest freespace
1220 * in the data block in case it changes.
1224 dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1225 bf = dp->d_ops->data_bestfree_p(hdr);
1226 longest = be16_to_cpu(bf[0].length);
1227 needlog = needscan = 0;
1228 xfs_dir2_data_make_free(tp, dp, dbp, off,
1229 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1231 * Rescan the data block freespaces for bestfree.
1232 * Log the data block header if needed.
1235 xfs_dir2_data_freescan(dp, hdr, &needlog);
1237 xfs_dir2_data_log_header(tp, dp, dbp);
1238 xfs_dir3_data_check(dp, dbp);
1240 * If the longest data block freespace changes, need to update
1241 * the corresponding freeblock entry.
1243 if (longest < be16_to_cpu(bf[0].length)) {
1244 int error; /* error return value */
1245 struct xfs_buf *fbp; /* freeblock buffer */
1246 xfs_dir2_db_t fdb; /* freeblock block number */
1247 int findex; /* index in freeblock entries */
1248 xfs_dir2_free_t *free; /* freeblock structure */
1251 * Convert the data block number to a free block,
1252 * read in the free block.
1254 fdb = dp->d_ops->db_to_fdb(mp, db);
1255 error = xfs_dir2_free_read(tp, dp,
1256 xfs_dir2_db_to_da(args->geo, fdb),
1263 struct xfs_dir3_icfree_hdr freehdr;
1264 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1265 ASSERT(freehdr.firstdb == dp->d_ops->free_max_bests(mp) *
1266 (fdb - xfs_dir2_byte_to_db(args->geo,
1267 XFS_DIR2_FREE_OFFSET)));
1271 * Calculate which entry we need to fix.
1273 findex = dp->d_ops->db_to_fdindex(mp, db);
1274 longest = be16_to_cpu(bf[0].length);
1276 * If the data block is now empty we can get rid of it
1279 if (longest == mp->m_dirblksize -
1280 dp->d_ops->data_entry_offset) {
1282 * Try to punch out the data block.
1284 error = xfs_dir2_shrink_inode(args, db, dbp);
1290 * We can get ENOSPC if there's no space reservation.
1291 * In this case just drop the buffer and some one else
1292 * will eventually get rid of the empty block.
1294 else if (!(error == ENOSPC && args->total == 0))
1298 * If we got rid of the data block, we can eliminate that entry
1299 * in the free block.
1301 error = xfs_dir3_data_block_free(args, hdr, free,
1302 fdb, findex, fbp, longest);
1307 xfs_dir3_leaf_check(dp, bp);
1309 * Return indication of whether this leaf block is empty enough
1310 * to justify trying to join it with a neighbor.
1312 *rval = (dp->d_ops->leaf_hdr_size +
1313 (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1319 * Split the leaf entries in the old block into old and new blocks.
1322 xfs_dir2_leafn_split(
1323 xfs_da_state_t *state, /* btree cursor */
1324 xfs_da_state_blk_t *oldblk, /* original block */
1325 xfs_da_state_blk_t *newblk) /* newly created block */
1327 xfs_da_args_t *args; /* operation arguments */
1328 xfs_dablk_t blkno; /* new leaf block number */
1329 int error; /* error return value */
1330 xfs_mount_t *mp; /* filesystem mount point */
1331 struct xfs_inode *dp;
1334 * Allocate space for a new leaf node.
1339 ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1340 error = xfs_da_grow_inode(args, &blkno);
1345 * Initialize the new leaf block.
1347 error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(args->geo, blkno),
1348 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1352 newblk->blkno = blkno;
1353 newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1355 * Rebalance the entries across the two leaves, link the new
1356 * block into the leaves.
1358 xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1359 error = xfs_da3_blk_link(state, oldblk, newblk);
1364 * Insert the new entry in the correct block.
1367 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1369 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1371 * Update last hashval in each block since we added the name.
1373 oldblk->hashval = xfs_dir2_leafn_lasthash(dp, oldblk->bp, NULL);
1374 newblk->hashval = xfs_dir2_leafn_lasthash(dp, newblk->bp, NULL);
1375 xfs_dir3_leaf_check(dp, oldblk->bp);
1376 xfs_dir3_leaf_check(dp, newblk->bp);
1381 * Check a leaf block and its neighbors to see if the block should be
1382 * collapsed into one or the other neighbor. Always keep the block
1383 * with the smaller block number.
1384 * If the current block is over 50% full, don't try to join it, return 0.
1385 * If the block is empty, fill in the state structure and return 2.
1386 * If it can be collapsed, fill in the state structure and return 1.
1387 * If nothing can be done, return 0.
1390 xfs_dir2_leafn_toosmall(
1391 xfs_da_state_t *state, /* btree cursor */
1392 int *action) /* resulting action to take */
1394 xfs_da_state_blk_t *blk; /* leaf block */
1395 xfs_dablk_t blkno; /* leaf block number */
1396 struct xfs_buf *bp; /* leaf buffer */
1397 int bytes; /* bytes in use */
1398 int count; /* leaf live entry count */
1399 int error; /* error return value */
1400 int forward; /* sibling block direction */
1401 int i; /* sibling counter */
1402 xfs_dir2_leaf_t *leaf; /* leaf structure */
1403 int rval; /* result from path_shift */
1404 struct xfs_dir3_icleaf_hdr leafhdr;
1405 struct xfs_dir2_leaf_entry *ents;
1406 struct xfs_inode *dp = state->args->dp;
1409 * Check for the degenerate case of the block being over 50% full.
1410 * If so, it's not worth even looking to see if we might be able
1411 * to coalesce with a sibling.
1413 blk = &state->path.blk[state->path.active - 1];
1414 leaf = blk->bp->b_addr;
1415 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1416 ents = dp->d_ops->leaf_ents_p(leaf);
1417 xfs_dir3_leaf_check(dp, blk->bp);
1419 count = leafhdr.count - leafhdr.stale;
1420 bytes = dp->d_ops->leaf_hdr_size + count * sizeof(ents[0]);
1421 if (bytes > (state->blocksize >> 1)) {
1423 * Blk over 50%, don't try to join.
1429 * Check for the degenerate case of the block being empty.
1430 * If the block is empty, we'll simply delete it, no need to
1431 * coalesce it with a sibling block. We choose (arbitrarily)
1432 * to merge with the forward block unless it is NULL.
1436 * Make altpath point to the block we want to keep and
1437 * path point to the block we want to drop (this one).
1439 forward = (leafhdr.forw != 0);
1440 memcpy(&state->altpath, &state->path, sizeof(state->path));
1441 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1445 *action = rval ? 2 : 0;
1449 * Examine each sibling block to see if we can coalesce with
1450 * at least 25% free space to spare. We need to figure out
1451 * whether to merge with the forward or the backward block.
1452 * We prefer coalescing with the lower numbered sibling so as
1453 * to shrink a directory over time.
1455 forward = leafhdr.forw < leafhdr.back;
1456 for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1457 struct xfs_dir3_icleaf_hdr hdr2;
1459 blkno = forward ? leafhdr.forw : leafhdr.back;
1463 * Read the sibling leaf block.
1465 error = xfs_dir3_leafn_read(state->args->trans, dp,
1471 * Count bytes in the two blocks combined.
1473 count = leafhdr.count - leafhdr.stale;
1474 bytes = state->blocksize - (state->blocksize >> 2);
1477 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf);
1478 ents = dp->d_ops->leaf_ents_p(leaf);
1479 count += hdr2.count - hdr2.stale;
1480 bytes -= count * sizeof(ents[0]);
1483 * Fits with at least 25% to spare.
1487 xfs_trans_brelse(state->args->trans, bp);
1490 * Didn't like either block, give up.
1498 * Make altpath point to the block we want to keep (the lower
1499 * numbered block) and path point to the block we want to drop.
1501 memcpy(&state->altpath, &state->path, sizeof(state->path));
1502 if (blkno < blk->blkno)
1503 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1506 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1511 *action = rval ? 0 : 1;
1516 * Move all the leaf entries from drop_blk to save_blk.
1517 * This is done as part of a join operation.
1520 xfs_dir2_leafn_unbalance(
1521 xfs_da_state_t *state, /* cursor */
1522 xfs_da_state_blk_t *drop_blk, /* dead block */
1523 xfs_da_state_blk_t *save_blk) /* surviving block */
1525 xfs_da_args_t *args; /* operation arguments */
1526 xfs_dir2_leaf_t *drop_leaf; /* dead leaf structure */
1527 xfs_dir2_leaf_t *save_leaf; /* surviving leaf structure */
1528 struct xfs_dir3_icleaf_hdr savehdr;
1529 struct xfs_dir3_icleaf_hdr drophdr;
1530 struct xfs_dir2_leaf_entry *sents;
1531 struct xfs_dir2_leaf_entry *dents;
1532 struct xfs_inode *dp = state->args->dp;
1535 ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1536 ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1537 drop_leaf = drop_blk->bp->b_addr;
1538 save_leaf = save_blk->bp->b_addr;
1540 dp->d_ops->leaf_hdr_from_disk(&savehdr, save_leaf);
1541 dp->d_ops->leaf_hdr_from_disk(&drophdr, drop_leaf);
1542 sents = dp->d_ops->leaf_ents_p(save_leaf);
1543 dents = dp->d_ops->leaf_ents_p(drop_leaf);
1546 * If there are any stale leaf entries, take this opportunity
1550 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1552 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1555 * Move the entries from drop to the appropriate end of save.
1557 drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1558 if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1559 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1560 save_blk->bp, &savehdr, sents, 0,
1563 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1564 save_blk->bp, &savehdr, sents,
1565 savehdr.count, drophdr.count);
1566 save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1568 /* log the changes made when moving the entries */
1569 dp->d_ops->leaf_hdr_to_disk(save_leaf, &savehdr);
1570 dp->d_ops->leaf_hdr_to_disk(drop_leaf, &drophdr);
1571 xfs_dir3_leaf_log_header(args->trans, dp, save_blk->bp);
1572 xfs_dir3_leaf_log_header(args->trans, dp, drop_blk->bp);
1574 xfs_dir3_leaf_check(dp, save_blk->bp);
1575 xfs_dir3_leaf_check(dp, drop_blk->bp);
1579 * Top-level node form directory addname routine.
1582 xfs_dir2_node_addname(
1583 xfs_da_args_t *args) /* operation arguments */
1585 xfs_da_state_blk_t *blk; /* leaf block for insert */
1586 int error; /* error return value */
1587 int rval; /* sub-return value */
1588 xfs_da_state_t *state; /* btree cursor */
1590 trace_xfs_dir2_node_addname(args);
1593 * Allocate and initialize the state (btree cursor).
1595 state = xfs_da_state_alloc();
1597 state->mp = args->dp->i_mount;
1598 state->blocksize = state->mp->m_dirblksize;
1599 state->node_ents = state->mp->m_dir_node_ents;
1601 * Look up the name. We're not supposed to find it, but
1602 * this gives us the insertion point.
1604 error = xfs_da3_node_lookup_int(state, &rval);
1607 if (rval != ENOENT) {
1611 * Add the data entry to a data block.
1612 * Extravalid is set to a freeblock found by lookup.
1614 rval = xfs_dir2_node_addname_int(args,
1615 state->extravalid ? &state->extrablk : NULL);
1619 blk = &state->path.blk[state->path.active - 1];
1620 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1622 * Add the new leaf entry.
1624 rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1627 * It worked, fix the hash values up the btree.
1629 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1630 xfs_da3_fixhashpath(state, &state->path);
1633 * It didn't work, we need to split the leaf block.
1635 if (args->total == 0) {
1636 ASSERT(rval == ENOSPC);
1640 * Split the leaf block and insert the new entry.
1642 rval = xfs_da3_split(state);
1645 xfs_da_state_free(state);
1650 * Add the data entry for a node-format directory name addition.
1651 * The leaf entry is added in xfs_dir2_leafn_add.
1652 * We may enter with a freespace block that the lookup found.
1654 static int /* error */
1655 xfs_dir2_node_addname_int(
1656 xfs_da_args_t *args, /* operation arguments */
1657 xfs_da_state_blk_t *fblk) /* optional freespace block */
1659 xfs_dir2_data_hdr_t *hdr; /* data block header */
1660 xfs_dir2_db_t dbno; /* data block number */
1661 struct xfs_buf *dbp; /* data block buffer */
1662 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1663 xfs_inode_t *dp; /* incore directory inode */
1664 xfs_dir2_data_unused_t *dup; /* data unused entry pointer */
1665 int error; /* error return value */
1666 xfs_dir2_db_t fbno; /* freespace block number */
1667 struct xfs_buf *fbp; /* freespace buffer */
1668 int findex; /* freespace entry index */
1669 xfs_dir2_free_t *free=NULL; /* freespace block structure */
1670 xfs_dir2_db_t ifbno; /* initial freespace block no */
1671 xfs_dir2_db_t lastfbno=0; /* highest freespace block no */
1672 int length; /* length of the new entry */
1673 int logfree; /* need to log free entry */
1674 xfs_mount_t *mp; /* filesystem mount point */
1675 int needlog; /* need to log data header */
1676 int needscan; /* need to rescan data frees */
1677 __be16 *tagp; /* data entry tag pointer */
1678 xfs_trans_t *tp; /* transaction pointer */
1680 struct xfs_dir3_icfree_hdr freehdr;
1681 struct xfs_dir2_data_free *bf;
1686 length = dp->d_ops->data_entsize(args->namelen);
1688 * If we came in with a freespace block that means that lookup
1689 * found an entry with our hash value. This is the freespace
1690 * block for that data entry.
1695 * Remember initial freespace block number.
1697 ifbno = fblk->blkno;
1699 findex = fblk->index;
1700 bests = dp->d_ops->free_bests_p(free);
1701 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1704 * This means the free entry showed that the data block had
1705 * space for our entry, so we remembered it.
1706 * Use that data block.
1709 ASSERT(findex < freehdr.nvalid);
1710 ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1711 ASSERT(be16_to_cpu(bests[findex]) >= length);
1712 dbno = freehdr.firstdb + findex;
1715 * The data block looked at didn't have enough room.
1716 * We'll start at the beginning of the freespace entries.
1723 * Didn't come in with a freespace block, so no data block.
1731 * If we don't have a data block yet, we're going to scan the
1732 * freespace blocks looking for one. Figure out what the
1733 * highest freespace block number is.
1736 xfs_fileoff_t fo; /* freespace block number */
1738 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK)))
1740 lastfbno = xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo);
1744 * While we haven't identified a data block, search the freeblock
1745 * data for a good data block. If we find a null freeblock entry,
1746 * indicating a hole in the data blocks, remember that.
1748 while (dbno == -1) {
1750 * If we don't have a freeblock in hand, get the next one.
1754 * Happens the first time through unless lookup gave
1755 * us a freespace block to start with.
1758 fbno = xfs_dir2_byte_to_db(args->geo,
1759 XFS_DIR2_FREE_OFFSET);
1761 * If it's ifbno we already looked at it.
1766 * If it's off the end we're done.
1768 if (fbno >= lastfbno)
1771 * Read the block. There can be holes in the
1772 * freespace blocks, so this might not succeed.
1773 * This should be really rare, so there's no reason
1776 error = xfs_dir2_free_try_read(tp, dp,
1777 xfs_dir2_db_to_da(args->geo, fbno),
1787 * Look at the current free entry. Is it good enough?
1789 * The bests initialisation should be where the bufer is read in
1790 * the above branch. But gcc is too stupid to realise that bests
1791 * and the freehdr are actually initialised if they are placed
1792 * there, so we have to do it here to avoid warnings. Blech.
1794 bests = dp->d_ops->free_bests_p(free);
1795 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1796 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1797 be16_to_cpu(bests[findex]) >= length)
1798 dbno = freehdr.firstdb + findex;
1801 * Are we done with the freeblock?
1803 if (++findex == freehdr.nvalid) {
1807 xfs_trans_brelse(tp, fbp);
1809 if (fblk && fblk->bp)
1815 * If we don't have a data block, we need to allocate one and make
1816 * the freespace entries refer to it.
1818 if (unlikely(dbno == -1)) {
1820 * Not allowed to allocate, return failure.
1822 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1823 return XFS_ERROR(ENOSPC);
1826 * Allocate and initialize the new data block.
1828 if (unlikely((error = xfs_dir2_grow_inode(args,
1829 XFS_DIR2_DATA_SPACE,
1831 (error = xfs_dir3_data_init(args, dbno, &dbp))))
1835 * If (somehow) we have a freespace block, get rid of it.
1838 xfs_trans_brelse(tp, fbp);
1839 if (fblk && fblk->bp)
1843 * Get the freespace block corresponding to the data block
1844 * that was just allocated.
1846 fbno = dp->d_ops->db_to_fdb(mp, dbno);
1847 error = xfs_dir2_free_try_read(tp, dp,
1848 xfs_dir2_db_to_da(args->geo, fbno),
1854 * If there wasn't a freespace block, the read will
1855 * return a NULL fbp. Allocate and initialize a new one.
1858 error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1863 if (unlikely(dp->d_ops->db_to_fdb(mp, dbno) != fbno)) {
1865 "%s: dir ino %llu needed freesp block %lld for\n"
1866 " data block %lld, got %lld ifbno %llu lastfbno %d",
1867 __func__, (unsigned long long)dp->i_ino,
1868 (long long)dp->d_ops->db_to_fdb(mp, dbno),
1869 (long long)dbno, (long long)fbno,
1870 (unsigned long long)ifbno, lastfbno);
1873 " fblk 0x%p blkno %llu index %d magic 0x%x",
1875 (unsigned long long)fblk->blkno,
1879 xfs_alert(mp, " ... fblk is NULL");
1881 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1882 XFS_ERRLEVEL_LOW, mp);
1883 return XFS_ERROR(EFSCORRUPTED);
1887 * Get a buffer for the new block.
1889 error = xfs_dir3_free_get_buf(args, fbno, &fbp);
1893 bests = dp->d_ops->free_bests_p(free);
1894 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1897 * Remember the first slot as our empty slot.
1900 (fbno - xfs_dir2_byte_to_db(args->geo,
1901 XFS_DIR2_FREE_OFFSET)) *
1902 dp->d_ops->free_max_bests(mp);
1905 bests = dp->d_ops->free_bests_p(free);
1906 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1910 * Set the freespace block index from the data block number.
1912 findex = dp->d_ops->db_to_fdindex(mp, dbno);
1914 * If it's after the end of the current entries in the
1915 * freespace block, extend that table.
1917 if (findex >= freehdr.nvalid) {
1918 ASSERT(findex < dp->d_ops->free_max_bests(mp));
1919 freehdr.nvalid = findex + 1;
1921 * Tag new entry so nused will go up.
1923 bests[findex] = cpu_to_be16(NULLDATAOFF);
1926 * If this entry was for an empty data block
1927 * (this should always be true) then update the header.
1929 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1931 dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
1932 xfs_dir2_free_log_header(tp, dp, fbp);
1935 * Update the real value in the table.
1936 * We haven't allocated the data entry yet so this will
1940 bf = dp->d_ops->data_bestfree_p(hdr);
1941 bests[findex] = bf[0].length;
1945 * We had a data block so we don't have to make a new one.
1949 * If just checking, we succeeded.
1951 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1955 * Read the data block in.
1957 error = xfs_dir3_data_read(tp, dp,
1958 xfs_dir2_db_to_da(args->geo, dbno),
1963 bf = dp->d_ops->data_bestfree_p(hdr);
1966 ASSERT(be16_to_cpu(bf[0].length) >= length);
1968 * Point to the existing unused space.
1970 dup = (xfs_dir2_data_unused_t *)
1971 ((char *)hdr + be16_to_cpu(bf[0].offset));
1972 needscan = needlog = 0;
1974 * Mark the first part of the unused space, inuse for us.
1976 xfs_dir2_data_use_free(tp, dp, dbp, dup,
1977 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
1978 &needlog, &needscan);
1980 * Fill in the new entry and log it.
1982 dep = (xfs_dir2_data_entry_t *)dup;
1983 dep->inumber = cpu_to_be64(args->inumber);
1984 dep->namelen = args->namelen;
1985 memcpy(dep->name, args->name, dep->namelen);
1986 dp->d_ops->data_put_ftype(dep, args->filetype);
1987 tagp = dp->d_ops->data_entry_tag_p(dep);
1988 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1989 xfs_dir2_data_log_entry(tp, dp, dbp, dep);
1991 * Rescan the block for bestfree if needed.
1994 xfs_dir2_data_freescan(dp, hdr, &needlog);
1996 * Log the data block header if needed.
1999 xfs_dir2_data_log_header(tp, dp, dbp);
2001 * If the freespace entry is now wrong, update it.
2003 bests = dp->d_ops->free_bests_p(free); /* gcc is so stupid */
2004 if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
2005 bests[findex] = bf[0].length;
2009 * Log the freespace entry if needed.
2012 xfs_dir2_free_log_bests(tp, dp, fbp, findex, findex);
2014 * Return the data block and offset in args, then drop the data block.
2016 args->blkno = (xfs_dablk_t)dbno;
2017 args->index = be16_to_cpu(*tagp);
2022 * Lookup an entry in a node-format directory.
2023 * All the real work happens in xfs_da3_node_lookup_int.
2024 * The only real output is the inode number of the entry.
2027 xfs_dir2_node_lookup(
2028 xfs_da_args_t *args) /* operation arguments */
2030 int error; /* error return value */
2031 int i; /* btree level */
2032 int rval; /* operation return value */
2033 xfs_da_state_t *state; /* btree cursor */
2035 trace_xfs_dir2_node_lookup(args);
2038 * Allocate and initialize the btree cursor.
2040 state = xfs_da_state_alloc();
2042 state->mp = args->dp->i_mount;
2043 state->blocksize = state->mp->m_dirblksize;
2044 state->node_ents = state->mp->m_dir_node_ents;
2046 * Fill in the path to the entry in the cursor.
2048 error = xfs_da3_node_lookup_int(state, &rval);
2051 else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
2052 /* If a CI match, dup the actual name and return EEXIST */
2053 xfs_dir2_data_entry_t *dep;
2055 dep = (xfs_dir2_data_entry_t *)
2056 ((char *)state->extrablk.bp->b_addr +
2057 state->extrablk.index);
2058 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2061 * Release the btree blocks and leaf block.
2063 for (i = 0; i < state->path.active; i++) {
2064 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2065 state->path.blk[i].bp = NULL;
2068 * Release the data block if we have it.
2070 if (state->extravalid && state->extrablk.bp) {
2071 xfs_trans_brelse(args->trans, state->extrablk.bp);
2072 state->extrablk.bp = NULL;
2074 xfs_da_state_free(state);
2079 * Remove an entry from a node-format directory.
2082 xfs_dir2_node_removename(
2083 struct xfs_da_args *args) /* operation arguments */
2085 struct xfs_da_state_blk *blk; /* leaf block */
2086 int error; /* error return value */
2087 int rval; /* operation return value */
2088 struct xfs_da_state *state; /* btree cursor */
2090 trace_xfs_dir2_node_removename(args);
2093 * Allocate and initialize the btree cursor.
2095 state = xfs_da_state_alloc();
2097 state->mp = args->dp->i_mount;
2098 state->blocksize = state->mp->m_dirblksize;
2099 state->node_ents = state->mp->m_dir_node_ents;
2101 /* Look up the entry we're deleting, set up the cursor. */
2102 error = xfs_da3_node_lookup_int(state, &rval);
2106 /* Didn't find it, upper layer screwed up. */
2107 if (rval != EEXIST) {
2112 blk = &state->path.blk[state->path.active - 1];
2113 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2114 ASSERT(state->extravalid);
2116 * Remove the leaf and data entries.
2117 * Extrablk refers to the data block.
2119 error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2120 &state->extrablk, &rval);
2124 * Fix the hash values up the btree.
2126 xfs_da3_fixhashpath(state, &state->path);
2128 * If we need to join leaf blocks, do it.
2130 if (rval && state->path.active > 1)
2131 error = xfs_da3_join(state);
2133 * If no errors so far, try conversion to leaf format.
2136 error = xfs_dir2_node_to_leaf(state);
2138 xfs_da_state_free(state);
2143 * Replace an entry's inode number in a node-format directory.
2146 xfs_dir2_node_replace(
2147 xfs_da_args_t *args) /* operation arguments */
2149 xfs_da_state_blk_t *blk; /* leaf block */
2150 xfs_dir2_data_hdr_t *hdr; /* data block header */
2151 xfs_dir2_data_entry_t *dep; /* data entry changed */
2152 int error; /* error return value */
2153 int i; /* btree level */
2154 xfs_ino_t inum; /* new inode number */
2155 xfs_dir2_leaf_t *leaf; /* leaf structure */
2156 xfs_dir2_leaf_entry_t *lep; /* leaf entry being changed */
2157 int rval; /* internal return value */
2158 xfs_da_state_t *state; /* btree cursor */
2160 trace_xfs_dir2_node_replace(args);
2163 * Allocate and initialize the btree cursor.
2165 state = xfs_da_state_alloc();
2167 state->mp = args->dp->i_mount;
2168 state->blocksize = state->mp->m_dirblksize;
2169 state->node_ents = state->mp->m_dir_node_ents;
2170 inum = args->inumber;
2172 * Lookup the entry to change in the btree.
2174 error = xfs_da3_node_lookup_int(state, &rval);
2179 * It should be found, since the vnodeops layer has looked it up
2180 * and locked it. But paranoia is good.
2182 if (rval == EEXIST) {
2183 struct xfs_dir2_leaf_entry *ents;
2185 * Find the leaf entry.
2187 blk = &state->path.blk[state->path.active - 1];
2188 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2189 leaf = blk->bp->b_addr;
2190 ents = args->dp->d_ops->leaf_ents_p(leaf);
2191 lep = &ents[blk->index];
2192 ASSERT(state->extravalid);
2194 * Point to the data entry.
2196 hdr = state->extrablk.bp->b_addr;
2197 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2198 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2199 dep = (xfs_dir2_data_entry_t *)
2201 xfs_dir2_dataptr_to_off(args->geo,
2202 be32_to_cpu(lep->address)));
2203 ASSERT(inum != be64_to_cpu(dep->inumber));
2205 * Fill in the new inode number and log the entry.
2207 dep->inumber = cpu_to_be64(inum);
2208 args->dp->d_ops->data_put_ftype(dep, args->filetype);
2209 xfs_dir2_data_log_entry(args->trans, args->dp,
2210 state->extrablk.bp, dep);
2214 * Didn't find it, and we're holding a data block. Drop it.
2216 else if (state->extravalid) {
2217 xfs_trans_brelse(args->trans, state->extrablk.bp);
2218 state->extrablk.bp = NULL;
2221 * Release all the buffers in the cursor.
2223 for (i = 0; i < state->path.active; i++) {
2224 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2225 state->path.blk[i].bp = NULL;
2227 xfs_da_state_free(state);
2232 * Trim off a trailing empty freespace block.
2233 * Return (in rvalp) 1 if we did it, 0 if not.
2236 xfs_dir2_node_trim_free(
2237 xfs_da_args_t *args, /* operation arguments */
2238 xfs_fileoff_t fo, /* free block number */
2239 int *rvalp) /* out: did something */
2241 struct xfs_buf *bp; /* freespace buffer */
2242 xfs_inode_t *dp; /* incore directory inode */
2243 int error; /* error return code */
2244 xfs_dir2_free_t *free; /* freespace structure */
2245 xfs_mount_t *mp; /* filesystem mount point */
2246 xfs_trans_t *tp; /* transaction pointer */
2247 struct xfs_dir3_icfree_hdr freehdr;
2253 * Read the freespace block.
2255 error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2259 * There can be holes in freespace. If fo is a hole, there's
2265 dp->d_ops->free_hdr_from_disk(&freehdr, free);
2268 * If there are used entries, there's nothing to do.
2270 if (freehdr.nused > 0) {
2271 xfs_trans_brelse(tp, bp);
2276 * Blow the block away.
2278 error = xfs_dir2_shrink_inode(args,
2279 xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo), bp);
2282 * Can't fail with ENOSPC since that only happens with no
2283 * space reservation, when breaking up an extent into two
2284 * pieces. This is the last block of an extent.
2286 ASSERT(error != ENOSPC);
2287 xfs_trans_brelse(tp, bp);
2291 * Return that we succeeded.