2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
22 #include "xfs_trans.h"
25 #include "xfs_mount.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_dinode.h"
29 #include "xfs_inode.h"
31 #include "xfs_dir2_format.h"
32 #include "xfs_dir2_priv.h"
33 #include "xfs_error.h"
34 #include "xfs_trace.h"
37 * Function declarations.
39 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
42 static void xfs_dir2_leafn_check(struct xfs_inode *dp, struct xfs_buf *bp);
44 #define xfs_dir2_leafn_check(dp, bp)
46 static void xfs_dir2_leafn_moveents(xfs_da_args_t *args, struct xfs_buf *bp_s,
47 int start_s, struct xfs_buf *bp_d,
48 int start_d, int count);
49 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
50 xfs_da_state_blk_t *blk1,
51 xfs_da_state_blk_t *blk2);
52 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
53 int index, xfs_da_state_blk_t *dblk,
55 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
56 xfs_da_state_blk_t *fblk);
62 struct xfs_mount *mp = bp->b_target->bt_mount;
63 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
66 block_ok = hdr->magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC);
68 XFS_CORRUPTION_ERROR("xfs_dir2_free_verify magic",
69 XFS_ERRLEVEL_LOW, mp, hdr);
70 xfs_buf_ioerror(bp, EFSCORRUPTED);
82 xfs_daddr_t mappedbno,
85 return xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
86 XFS_DATA_FORK, xfs_dir2_free_verify);
96 return __xfs_dir2_free_read(tp, dp, fbno, -1, bpp);
100 xfs_dir2_free_try_read(
101 struct xfs_trans *tp,
102 struct xfs_inode *dp,
104 struct xfs_buf **bpp)
106 return __xfs_dir2_free_read(tp, dp, fbno, -2, bpp);
110 * Log entries from a freespace block.
113 xfs_dir2_free_log_bests(
114 struct xfs_trans *tp,
116 int first, /* first entry to log */
117 int last) /* last entry to log */
119 xfs_dir2_free_t *free; /* freespace structure */
122 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
123 xfs_trans_log_buf(tp, bp,
124 (uint)((char *)&free->bests[first] - (char *)free),
125 (uint)((char *)&free->bests[last] - (char *)free +
126 sizeof(free->bests[0]) - 1));
130 * Log header from a freespace block.
133 xfs_dir2_free_log_header(
134 struct xfs_trans *tp,
137 xfs_dir2_free_t *free; /* freespace structure */
140 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
141 xfs_trans_log_buf(tp, bp, (uint)((char *)&free->hdr - (char *)free),
142 (uint)(sizeof(xfs_dir2_free_hdr_t) - 1));
146 * Convert a leaf-format directory to a node-format directory.
147 * We need to change the magic number of the leaf block, and copy
148 * the freespace table out of the leaf block into its own block.
151 xfs_dir2_leaf_to_node(
152 xfs_da_args_t *args, /* operation arguments */
153 struct xfs_buf *lbp) /* leaf buffer */
155 xfs_inode_t *dp; /* incore directory inode */
156 int error; /* error return value */
157 struct xfs_buf *fbp; /* freespace buffer */
158 xfs_dir2_db_t fdb; /* freespace block number */
159 xfs_dir2_free_t *free; /* freespace structure */
160 __be16 *from; /* pointer to freespace entry */
161 int i; /* leaf freespace index */
162 xfs_dir2_leaf_t *leaf; /* leaf structure */
163 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
164 xfs_mount_t *mp; /* filesystem mount point */
165 int n; /* count of live freespc ents */
166 xfs_dir2_data_off_t off; /* freespace entry value */
167 __be16 *to; /* pointer to freespace entry */
168 xfs_trans_t *tp; /* transaction pointer */
170 trace_xfs_dir2_leaf_to_node(args);
176 * Add a freespace block to the directory.
178 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
181 ASSERT(fdb == XFS_DIR2_FREE_FIRSTDB(mp));
183 * Get the buffer for the new freespace block.
185 if ((error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, fdb), -1, &fbp,
192 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
194 * Initialize the freespace block header.
196 free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
197 free->hdr.firstdb = 0;
198 ASSERT(be32_to_cpu(ltp->bestcount) <= (uint)dp->i_d.di_size / mp->m_dirblksize);
199 free->hdr.nvalid = ltp->bestcount;
201 * Copy freespace entries from the leaf block to the new block.
202 * Count active entries.
204 for (i = n = 0, from = xfs_dir2_leaf_bests_p(ltp), to = free->bests;
205 i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
206 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
208 *to = cpu_to_be16(off);
210 free->hdr.nused = cpu_to_be32(n);
211 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
215 xfs_dir2_leaf_log_header(tp, lbp);
216 xfs_dir2_free_log_header(tp, fbp);
217 xfs_dir2_free_log_bests(tp, fbp, 0, be32_to_cpu(free->hdr.nvalid) - 1);
218 xfs_dir2_leafn_check(dp, lbp);
223 * Add a leaf entry to a leaf block in a node-form directory.
224 * The other work necessary is done from the caller.
226 static int /* error */
228 struct xfs_buf *bp, /* leaf buffer */
229 xfs_da_args_t *args, /* operation arguments */
230 int index) /* insertion pt for new entry */
232 int compact; /* compacting stale leaves */
233 xfs_inode_t *dp; /* incore directory inode */
234 int highstale; /* next stale entry */
235 xfs_dir2_leaf_t *leaf; /* leaf structure */
236 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
237 int lfloghigh; /* high leaf entry logging */
238 int lfloglow; /* low leaf entry logging */
239 int lowstale; /* previous stale entry */
240 xfs_mount_t *mp; /* filesystem mount point */
241 xfs_trans_t *tp; /* transaction pointer */
243 trace_xfs_dir2_leafn_add(args, index);
251 * Quick check just to make sure we are not going to index
252 * into other peoples memory
255 return XFS_ERROR(EFSCORRUPTED);
258 * If there are already the maximum number of leaf entries in
259 * the block, if there are no stale entries it won't fit.
260 * Caller will do a split. If there are stale entries we'll do
264 if (be16_to_cpu(leaf->hdr.count) == xfs_dir2_max_leaf_ents(mp)) {
265 if (!leaf->hdr.stale)
266 return XFS_ERROR(ENOSPC);
267 compact = be16_to_cpu(leaf->hdr.stale) > 1;
270 ASSERT(index == 0 || be32_to_cpu(leaf->ents[index - 1].hashval) <= args->hashval);
271 ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
272 be32_to_cpu(leaf->ents[index].hashval) >= args->hashval);
274 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
278 * Compact out all but one stale leaf entry. Leaves behind
279 * the entry closest to index.
282 xfs_dir2_leaf_compact_x1(bp, &index, &lowstale, &highstale,
283 &lfloglow, &lfloghigh);
286 * Set impossible logging indices for this case.
288 else if (leaf->hdr.stale) {
289 lfloglow = be16_to_cpu(leaf->hdr.count);
294 * Insert the new entry, log everything.
296 lep = xfs_dir2_leaf_find_entry(leaf, index, compact, lowstale,
297 highstale, &lfloglow, &lfloghigh);
299 lep->hashval = cpu_to_be32(args->hashval);
300 lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp,
301 args->blkno, args->index));
302 xfs_dir2_leaf_log_header(tp, bp);
303 xfs_dir2_leaf_log_ents(tp, bp, lfloglow, lfloghigh);
304 xfs_dir2_leafn_check(dp, bp);
310 * Check internal consistency of a leafn block.
313 xfs_dir2_leafn_check(
314 struct xfs_inode *dp,
317 int i; /* leaf index */
318 xfs_dir2_leaf_t *leaf; /* leaf structure */
319 xfs_mount_t *mp; /* filesystem mount point */
320 int stale; /* count of stale leaves */
324 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
325 ASSERT(be16_to_cpu(leaf->hdr.count) <= xfs_dir2_max_leaf_ents(mp));
326 for (i = stale = 0; i < be16_to_cpu(leaf->hdr.count); i++) {
327 if (i + 1 < be16_to_cpu(leaf->hdr.count)) {
328 ASSERT(be32_to_cpu(leaf->ents[i].hashval) <=
329 be32_to_cpu(leaf->ents[i + 1].hashval));
331 if (leaf->ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
334 ASSERT(be16_to_cpu(leaf->hdr.stale) == stale);
339 * Return the last hash value in the leaf.
340 * Stale entries are ok.
342 xfs_dahash_t /* hash value */
343 xfs_dir2_leafn_lasthash(
344 struct xfs_buf *bp, /* leaf buffer */
345 int *count) /* count of entries in leaf */
347 xfs_dir2_leaf_t *leaf; /* leaf structure */
350 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
352 *count = be16_to_cpu(leaf->hdr.count);
353 if (!leaf->hdr.count)
355 return be32_to_cpu(leaf->ents[be16_to_cpu(leaf->hdr.count) - 1].hashval);
359 * Look up a leaf entry for space to add a name in a node-format leaf block.
360 * The extrablk in state is a freespace block.
363 xfs_dir2_leafn_lookup_for_addname(
364 struct xfs_buf *bp, /* leaf buffer */
365 xfs_da_args_t *args, /* operation arguments */
366 int *indexp, /* out: leaf entry index */
367 xfs_da_state_t *state) /* state to fill in */
369 struct xfs_buf *curbp = NULL; /* current data/free buffer */
370 xfs_dir2_db_t curdb = -1; /* current data block number */
371 xfs_dir2_db_t curfdb = -1; /* current free block number */
372 xfs_inode_t *dp; /* incore directory inode */
373 int error; /* error return value */
374 int fi; /* free entry index */
375 xfs_dir2_free_t *free = NULL; /* free block structure */
376 int index; /* leaf entry index */
377 xfs_dir2_leaf_t *leaf; /* leaf structure */
378 int length; /* length of new data entry */
379 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
380 xfs_mount_t *mp; /* filesystem mount point */
381 xfs_dir2_db_t newdb; /* new data block number */
382 xfs_dir2_db_t newfdb; /* new free block number */
383 xfs_trans_t *tp; /* transaction pointer */
389 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
391 ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
393 xfs_dir2_leafn_check(dp, bp);
395 * Look up the hash value in the leaf entries.
397 index = xfs_dir2_leaf_search_hash(args, bp);
399 * Do we have a buffer coming in?
401 if (state->extravalid) {
402 /* If so, it's a free block buffer, get the block number. */
403 curbp = state->extrablk.bp;
404 curfdb = state->extrablk.blkno;
405 free = curbp->b_addr;
406 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
408 length = xfs_dir2_data_entsize(args->namelen);
410 * Loop over leaf entries with the right hash value.
412 for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
413 be32_to_cpu(lep->hashval) == args->hashval;
416 * Skip stale leaf entries.
418 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
421 * Pull the data block number from the entry.
423 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
425 * For addname, we're looking for a place to put the new entry.
426 * We want to use a data block with an entry of equal
427 * hash value to ours if there is one with room.
429 * If this block isn't the data block we already have
430 * in hand, take a look at it.
432 if (newdb != curdb) {
435 * Convert the data block to the free block
436 * holding its freespace information.
438 newfdb = xfs_dir2_db_to_fdb(mp, newdb);
440 * If it's not the one we have in hand, read it in.
442 if (newfdb != curfdb) {
444 * If we had one before, drop it.
447 xfs_trans_brelse(tp, curbp);
449 error = xfs_dir2_free_read(tp, dp,
450 xfs_dir2_db_to_da(mp, newfdb),
454 free = curbp->b_addr;
455 ASSERT(be32_to_cpu(free->hdr.magic) ==
456 XFS_DIR2_FREE_MAGIC);
457 ASSERT((be32_to_cpu(free->hdr.firstdb) %
458 xfs_dir2_free_max_bests(mp)) == 0);
459 ASSERT(be32_to_cpu(free->hdr.firstdb) <= curdb);
460 ASSERT(curdb < be32_to_cpu(free->hdr.firstdb) +
461 be32_to_cpu(free->hdr.nvalid));
464 * Get the index for our entry.
466 fi = xfs_dir2_db_to_fdindex(mp, curdb);
468 * If it has room, return it.
470 if (unlikely(free->bests[fi] ==
471 cpu_to_be16(NULLDATAOFF))) {
472 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
473 XFS_ERRLEVEL_LOW, mp);
474 if (curfdb != newfdb)
475 xfs_trans_brelse(tp, curbp);
476 return XFS_ERROR(EFSCORRUPTED);
479 if (be16_to_cpu(free->bests[fi]) >= length)
483 /* Didn't find any space */
486 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
488 /* Giving back a free block. */
489 state->extravalid = 1;
490 state->extrablk.bp = curbp;
491 state->extrablk.index = fi;
492 state->extrablk.blkno = curfdb;
493 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
495 state->extravalid = 0;
498 * Return the index, that will be the insertion point.
501 return XFS_ERROR(ENOENT);
505 * Look up a leaf entry in a node-format leaf block.
506 * The extrablk in state a data block.
509 xfs_dir2_leafn_lookup_for_entry(
510 struct xfs_buf *bp, /* leaf buffer */
511 xfs_da_args_t *args, /* operation arguments */
512 int *indexp, /* out: leaf entry index */
513 xfs_da_state_t *state) /* state to fill in */
515 struct xfs_buf *curbp = NULL; /* current data/free buffer */
516 xfs_dir2_db_t curdb = -1; /* current data block number */
517 xfs_dir2_data_entry_t *dep; /* data block entry */
518 xfs_inode_t *dp; /* incore directory inode */
519 int error; /* error return value */
520 int index; /* leaf entry index */
521 xfs_dir2_leaf_t *leaf; /* leaf structure */
522 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
523 xfs_mount_t *mp; /* filesystem mount point */
524 xfs_dir2_db_t newdb; /* new data block number */
525 xfs_trans_t *tp; /* transaction pointer */
526 enum xfs_dacmp cmp; /* comparison result */
532 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
534 ASSERT(be16_to_cpu(leaf->hdr.count) > 0);
536 xfs_dir2_leafn_check(dp, bp);
538 * Look up the hash value in the leaf entries.
540 index = xfs_dir2_leaf_search_hash(args, bp);
542 * Do we have a buffer coming in?
544 if (state->extravalid) {
545 curbp = state->extrablk.bp;
546 curdb = state->extrablk.blkno;
549 * Loop over leaf entries with the right hash value.
551 for (lep = &leaf->ents[index]; index < be16_to_cpu(leaf->hdr.count) &&
552 be32_to_cpu(lep->hashval) == args->hashval;
555 * Skip stale leaf entries.
557 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
560 * Pull the data block number from the entry.
562 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
564 * Not adding a new entry, so we really want to find
565 * the name given to us.
567 * If it's a different data block, go get it.
569 if (newdb != curdb) {
571 * If we had a block before that we aren't saving
572 * for a CI name, drop it
574 if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
575 curdb != state->extrablk.blkno))
576 xfs_trans_brelse(tp, curbp);
578 * If needing the block that is saved with a CI match,
579 * use it otherwise read in the new data block.
581 if (args->cmpresult != XFS_CMP_DIFFERENT &&
582 newdb == state->extrablk.blkno) {
583 ASSERT(state->extravalid);
584 curbp = state->extrablk.bp;
586 error = xfs_da_read_buf(tp, dp,
587 xfs_dir2_db_to_da(mp, newdb),
588 -1, &curbp, XFS_DATA_FORK, NULL);
592 xfs_dir2_data_check(dp, curbp);
596 * Point to the data entry.
598 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
599 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
601 * Compare the entry and if it's an exact match, return
602 * EEXIST immediately. If it's the first case-insensitive
603 * match, store the block & inode number and continue looking.
605 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
606 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
607 /* If there is a CI match block, drop it */
608 if (args->cmpresult != XFS_CMP_DIFFERENT &&
609 curdb != state->extrablk.blkno)
610 xfs_trans_brelse(tp, state->extrablk.bp);
611 args->cmpresult = cmp;
612 args->inumber = be64_to_cpu(dep->inumber);
614 state->extravalid = 1;
615 state->extrablk.bp = curbp;
616 state->extrablk.blkno = curdb;
617 state->extrablk.index = (int)((char *)dep -
618 (char *)curbp->b_addr);
619 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
620 if (cmp == XFS_CMP_EXACT)
621 return XFS_ERROR(EEXIST);
624 ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
625 (args->op_flags & XFS_DA_OP_OKNOENT));
627 if (args->cmpresult == XFS_CMP_DIFFERENT) {
628 /* Giving back last used data block. */
629 state->extravalid = 1;
630 state->extrablk.bp = curbp;
631 state->extrablk.index = -1;
632 state->extrablk.blkno = curdb;
633 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
635 /* If the curbp is not the CI match block, drop it */
636 if (state->extrablk.bp != curbp)
637 xfs_trans_brelse(tp, curbp);
640 state->extravalid = 0;
643 return XFS_ERROR(ENOENT);
647 * Look up a leaf entry in a node-format leaf block.
648 * If this is an addname then the extrablk in state is a freespace block,
649 * otherwise it's a data block.
652 xfs_dir2_leafn_lookup_int(
653 struct xfs_buf *bp, /* leaf buffer */
654 xfs_da_args_t *args, /* operation arguments */
655 int *indexp, /* out: leaf entry index */
656 xfs_da_state_t *state) /* state to fill in */
658 if (args->op_flags & XFS_DA_OP_ADDNAME)
659 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
661 return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
665 * Move count leaf entries from source to destination leaf.
666 * Log entries and headers. Stale entries are preserved.
669 xfs_dir2_leafn_moveents(
670 xfs_da_args_t *args, /* operation arguments */
671 struct xfs_buf *bp_s, /* source leaf buffer */
672 int start_s, /* source leaf index */
673 struct xfs_buf *bp_d, /* destination leaf buffer */
674 int start_d, /* destination leaf index */
675 int count) /* count of leaves to copy */
677 xfs_dir2_leaf_t *leaf_d; /* destination leaf structure */
678 xfs_dir2_leaf_t *leaf_s; /* source leaf structure */
679 int stale; /* count stale leaves copied */
680 xfs_trans_t *tp; /* transaction pointer */
682 trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
685 * Silently return if nothing to do.
691 leaf_s = bp_s->b_addr;
692 leaf_d = bp_d->b_addr;
694 * If the destination index is not the end of the current
695 * destination leaf entries, open up a hole in the destination
696 * to hold the new entries.
698 if (start_d < be16_to_cpu(leaf_d->hdr.count)) {
699 memmove(&leaf_d->ents[start_d + count], &leaf_d->ents[start_d],
700 (be16_to_cpu(leaf_d->hdr.count) - start_d) *
701 sizeof(xfs_dir2_leaf_entry_t));
702 xfs_dir2_leaf_log_ents(tp, bp_d, start_d + count,
703 count + be16_to_cpu(leaf_d->hdr.count) - 1);
706 * If the source has stale leaves, count the ones in the copy range
707 * so we can update the header correctly.
709 if (leaf_s->hdr.stale) {
710 int i; /* temp leaf index */
712 for (i = start_s, stale = 0; i < start_s + count; i++) {
713 if (leaf_s->ents[i].address ==
714 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
720 * Copy the leaf entries from source to destination.
722 memcpy(&leaf_d->ents[start_d], &leaf_s->ents[start_s],
723 count * sizeof(xfs_dir2_leaf_entry_t));
724 xfs_dir2_leaf_log_ents(tp, bp_d, start_d, start_d + count - 1);
726 * If there are source entries after the ones we copied,
727 * delete the ones we copied by sliding the next ones down.
729 if (start_s + count < be16_to_cpu(leaf_s->hdr.count)) {
730 memmove(&leaf_s->ents[start_s], &leaf_s->ents[start_s + count],
731 count * sizeof(xfs_dir2_leaf_entry_t));
732 xfs_dir2_leaf_log_ents(tp, bp_s, start_s, start_s + count - 1);
735 * Update the headers and log them.
737 be16_add_cpu(&leaf_s->hdr.count, -(count));
738 be16_add_cpu(&leaf_s->hdr.stale, -(stale));
739 be16_add_cpu(&leaf_d->hdr.count, count);
740 be16_add_cpu(&leaf_d->hdr.stale, stale);
741 xfs_dir2_leaf_log_header(tp, bp_s);
742 xfs_dir2_leaf_log_header(tp, bp_d);
743 xfs_dir2_leafn_check(args->dp, bp_s);
744 xfs_dir2_leafn_check(args->dp, bp_d);
748 * Determine the sort order of two leaf blocks.
749 * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
752 xfs_dir2_leafn_order(
753 struct xfs_buf *leaf1_bp, /* leaf1 buffer */
754 struct xfs_buf *leaf2_bp) /* leaf2 buffer */
756 xfs_dir2_leaf_t *leaf1; /* leaf1 structure */
757 xfs_dir2_leaf_t *leaf2; /* leaf2 structure */
759 leaf1 = leaf1_bp->b_addr;
760 leaf2 = leaf2_bp->b_addr;
761 ASSERT(leaf1->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
762 ASSERT(leaf2->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
763 if (be16_to_cpu(leaf1->hdr.count) > 0 &&
764 be16_to_cpu(leaf2->hdr.count) > 0 &&
765 (be32_to_cpu(leaf2->ents[0].hashval) < be32_to_cpu(leaf1->ents[0].hashval) ||
766 be32_to_cpu(leaf2->ents[be16_to_cpu(leaf2->hdr.count) - 1].hashval) <
767 be32_to_cpu(leaf1->ents[be16_to_cpu(leaf1->hdr.count) - 1].hashval)))
773 * Rebalance leaf entries between two leaf blocks.
774 * This is actually only called when the second block is new,
775 * though the code deals with the general case.
776 * A new entry will be inserted in one of the blocks, and that
777 * entry is taken into account when balancing.
780 xfs_dir2_leafn_rebalance(
781 xfs_da_state_t *state, /* btree cursor */
782 xfs_da_state_blk_t *blk1, /* first btree block */
783 xfs_da_state_blk_t *blk2) /* second btree block */
785 xfs_da_args_t *args; /* operation arguments */
786 int count; /* count (& direction) leaves */
787 int isleft; /* new goes in left leaf */
788 xfs_dir2_leaf_t *leaf1; /* first leaf structure */
789 xfs_dir2_leaf_t *leaf2; /* second leaf structure */
790 int mid; /* midpoint leaf index */
792 int oldstale; /* old count of stale leaves */
794 int oldsum; /* old total leaf count */
795 int swap; /* swapped leaf blocks */
799 * If the block order is wrong, swap the arguments.
801 if ((swap = xfs_dir2_leafn_order(blk1->bp, blk2->bp))) {
802 xfs_da_state_blk_t *tmp; /* temp for block swap */
808 leaf1 = blk1->bp->b_addr;
809 leaf2 = blk2->bp->b_addr;
810 oldsum = be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count);
812 oldstale = be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale);
816 * If the old leaf count was odd then the new one will be even,
817 * so we need to divide the new count evenly.
820 xfs_dahash_t midhash; /* middle entry hash value */
822 if (mid >= be16_to_cpu(leaf1->hdr.count))
823 midhash = be32_to_cpu(leaf2->ents[mid - be16_to_cpu(leaf1->hdr.count)].hashval);
825 midhash = be32_to_cpu(leaf1->ents[mid].hashval);
826 isleft = args->hashval <= midhash;
829 * If the old count is even then the new count is odd, so there's
830 * no preferred side for the new entry.
836 * Calculate moved entry count. Positive means left-to-right,
837 * negative means right-to-left. Then move the entries.
839 count = be16_to_cpu(leaf1->hdr.count) - mid + (isleft == 0);
841 xfs_dir2_leafn_moveents(args, blk1->bp,
842 be16_to_cpu(leaf1->hdr.count) - count, blk2->bp, 0, count);
844 xfs_dir2_leafn_moveents(args, blk2->bp, 0, blk1->bp,
845 be16_to_cpu(leaf1->hdr.count), count);
846 ASSERT(be16_to_cpu(leaf1->hdr.count) + be16_to_cpu(leaf2->hdr.count) == oldsum);
847 ASSERT(be16_to_cpu(leaf1->hdr.stale) + be16_to_cpu(leaf2->hdr.stale) == oldstale);
849 * Mark whether we're inserting into the old or new leaf.
851 if (be16_to_cpu(leaf1->hdr.count) < be16_to_cpu(leaf2->hdr.count))
852 state->inleaf = swap;
853 else if (be16_to_cpu(leaf1->hdr.count) > be16_to_cpu(leaf2->hdr.count))
854 state->inleaf = !swap;
857 swap ^ (blk1->index <= be16_to_cpu(leaf1->hdr.count));
859 * Adjust the expected index for insertion.
862 blk2->index = blk1->index - be16_to_cpu(leaf1->hdr.count);
865 * Finally sanity check just to make sure we are not returning a
868 if(blk2->index < 0) {
871 xfs_alert(args->dp->i_mount,
872 "%s: picked the wrong leaf? reverting original leaf: blk1->index %d\n",
873 __func__, blk1->index);
878 xfs_dir2_data_block_free(
880 struct xfs_dir2_data_hdr *hdr,
881 struct xfs_dir2_free *free,
887 struct xfs_trans *tp = args->trans;
891 /* One less used entry in the free table. */
892 be32_add_cpu(&free->hdr.nused, -1);
893 xfs_dir2_free_log_header(tp, fbp);
896 * If this was the last entry in the table, we can trim the
897 * table size back. There might be other entries at the end
898 * referring to non-existent data blocks, get those too.
900 if (findex == be32_to_cpu(free->hdr.nvalid) - 1) {
901 int i; /* free entry index */
903 for (i = findex - 1; i >= 0; i--) {
904 if (free->bests[i] != cpu_to_be16(NULLDATAOFF))
907 free->hdr.nvalid = cpu_to_be32(i + 1);
910 /* Not the last entry, just punch it out. */
911 free->bests[findex] = cpu_to_be16(NULLDATAOFF);
915 * If there are no useful entries left in the block,
916 * get rid of the block if we can.
918 if (!free->hdr.nused) {
921 error = xfs_dir2_shrink_inode(args, fdb, fbp);
925 } else if (error != ENOSPC || args->total != 0)
928 * It's possible to get ENOSPC if there is no
929 * space reservation. In this case some one
930 * else will eventually get rid of this block.
935 * Data block is not empty, just set the free entry to the new
938 free->bests[findex] = cpu_to_be16(longest);
942 /* Log the free entry that changed, unless we got rid of it. */
944 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
949 * Remove an entry from a node directory.
950 * This removes the leaf entry and the data entry,
951 * and updates the free block if necessary.
953 static int /* error */
954 xfs_dir2_leafn_remove(
955 xfs_da_args_t *args, /* operation arguments */
956 struct xfs_buf *bp, /* leaf buffer */
957 int index, /* leaf entry index */
958 xfs_da_state_blk_t *dblk, /* data block */
959 int *rval) /* resulting block needs join */
961 xfs_dir2_data_hdr_t *hdr; /* data block header */
962 xfs_dir2_db_t db; /* data block number */
963 struct xfs_buf *dbp; /* data block buffer */
964 xfs_dir2_data_entry_t *dep; /* data block entry */
965 xfs_inode_t *dp; /* incore directory inode */
966 xfs_dir2_leaf_t *leaf; /* leaf structure */
967 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
968 int longest; /* longest data free entry */
969 int off; /* data block entry offset */
970 xfs_mount_t *mp; /* filesystem mount point */
971 int needlog; /* need to log data header */
972 int needscan; /* need to rescan data frees */
973 xfs_trans_t *tp; /* transaction pointer */
975 trace_xfs_dir2_leafn_remove(args, index);
981 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
983 * Point to the entry we're removing.
985 lep = &leaf->ents[index];
987 * Extract the data block and offset from the entry.
989 db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
990 ASSERT(dblk->blkno == db);
991 off = xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address));
992 ASSERT(dblk->index == off);
994 * Kill the leaf entry by marking it stale.
995 * Log the leaf block changes.
997 be16_add_cpu(&leaf->hdr.stale, 1);
998 xfs_dir2_leaf_log_header(tp, bp);
999 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1000 xfs_dir2_leaf_log_ents(tp, bp, index, index);
1002 * Make the data entry free. Keep track of the longest freespace
1003 * in the data block in case it changes.
1007 dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1008 longest = be16_to_cpu(hdr->bestfree[0].length);
1009 needlog = needscan = 0;
1010 xfs_dir2_data_make_free(tp, dbp, off,
1011 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
1013 * Rescan the data block freespaces for bestfree.
1014 * Log the data block header if needed.
1017 xfs_dir2_data_freescan(mp, hdr, &needlog);
1019 xfs_dir2_data_log_header(tp, dbp);
1020 xfs_dir2_data_check(dp, dbp);
1022 * If the longest data block freespace changes, need to update
1023 * the corresponding freeblock entry.
1025 if (longest < be16_to_cpu(hdr->bestfree[0].length)) {
1026 int error; /* error return value */
1027 struct xfs_buf *fbp; /* freeblock buffer */
1028 xfs_dir2_db_t fdb; /* freeblock block number */
1029 int findex; /* index in freeblock entries */
1030 xfs_dir2_free_t *free; /* freeblock structure */
1033 * Convert the data block number to a free block,
1034 * read in the free block.
1036 fdb = xfs_dir2_db_to_fdb(mp, db);
1037 error = xfs_dir2_free_read(tp, dp, xfs_dir2_db_to_da(mp, fdb),
1042 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1043 ASSERT(be32_to_cpu(free->hdr.firstdb) ==
1044 xfs_dir2_free_max_bests(mp) *
1045 (fdb - XFS_DIR2_FREE_FIRSTDB(mp)));
1047 * Calculate which entry we need to fix.
1049 findex = xfs_dir2_db_to_fdindex(mp, db);
1050 longest = be16_to_cpu(hdr->bestfree[0].length);
1052 * If the data block is now empty we can get rid of it
1055 if (longest == mp->m_dirblksize - (uint)sizeof(*hdr)) {
1057 * Try to punch out the data block.
1059 error = xfs_dir2_shrink_inode(args, db, dbp);
1065 * We can get ENOSPC if there's no space reservation.
1066 * In this case just drop the buffer and some one else
1067 * will eventually get rid of the empty block.
1069 else if (!(error == ENOSPC && args->total == 0))
1073 * If we got rid of the data block, we can eliminate that entry
1074 * in the free block.
1076 error = xfs_dir2_data_block_free(args, hdr, free,
1077 fdb, findex, fbp, longest);
1082 xfs_dir2_leafn_check(dp, bp);
1084 * Return indication of whether this leaf block is empty enough
1085 * to justify trying to join it with a neighbor.
1088 ((uint)sizeof(leaf->hdr) +
1089 (uint)sizeof(leaf->ents[0]) *
1090 (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale))) <
1096 * Split the leaf entries in the old block into old and new blocks.
1099 xfs_dir2_leafn_split(
1100 xfs_da_state_t *state, /* btree cursor */
1101 xfs_da_state_blk_t *oldblk, /* original block */
1102 xfs_da_state_blk_t *newblk) /* newly created block */
1104 xfs_da_args_t *args; /* operation arguments */
1105 xfs_dablk_t blkno; /* new leaf block number */
1106 int error; /* error return value */
1107 xfs_mount_t *mp; /* filesystem mount point */
1110 * Allocate space for a new leaf node.
1113 mp = args->dp->i_mount;
1114 ASSERT(args != NULL);
1115 ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1116 error = xfs_da_grow_inode(args, &blkno);
1121 * Initialize the new leaf block.
1123 error = xfs_dir2_leaf_init(args, xfs_dir2_da_to_db(mp, blkno),
1124 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1128 newblk->blkno = blkno;
1129 newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1131 * Rebalance the entries across the two leaves, link the new
1132 * block into the leaves.
1134 xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1135 error = xfs_da_blk_link(state, oldblk, newblk);
1140 * Insert the new entry in the correct block.
1143 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1145 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1147 * Update last hashval in each block since we added the name.
1149 oldblk->hashval = xfs_dir2_leafn_lasthash(oldblk->bp, NULL);
1150 newblk->hashval = xfs_dir2_leafn_lasthash(newblk->bp, NULL);
1151 xfs_dir2_leafn_check(args->dp, oldblk->bp);
1152 xfs_dir2_leafn_check(args->dp, newblk->bp);
1157 * Check a leaf block and its neighbors to see if the block should be
1158 * collapsed into one or the other neighbor. Always keep the block
1159 * with the smaller block number.
1160 * If the current block is over 50% full, don't try to join it, return 0.
1161 * If the block is empty, fill in the state structure and return 2.
1162 * If it can be collapsed, fill in the state structure and return 1.
1163 * If nothing can be done, return 0.
1166 xfs_dir2_leafn_toosmall(
1167 xfs_da_state_t *state, /* btree cursor */
1168 int *action) /* resulting action to take */
1170 xfs_da_state_blk_t *blk; /* leaf block */
1171 xfs_dablk_t blkno; /* leaf block number */
1172 struct xfs_buf *bp; /* leaf buffer */
1173 int bytes; /* bytes in use */
1174 int count; /* leaf live entry count */
1175 int error; /* error return value */
1176 int forward; /* sibling block direction */
1177 int i; /* sibling counter */
1178 xfs_da_blkinfo_t *info; /* leaf block header */
1179 xfs_dir2_leaf_t *leaf; /* leaf structure */
1180 int rval; /* result from path_shift */
1183 * Check for the degenerate case of the block being over 50% full.
1184 * If so, it's not worth even looking to see if we might be able
1185 * to coalesce with a sibling.
1187 blk = &state->path.blk[state->path.active - 1];
1188 info = blk->bp->b_addr;
1189 ASSERT(info->magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1190 leaf = (xfs_dir2_leaf_t *)info;
1191 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1192 bytes = (uint)sizeof(leaf->hdr) + count * (uint)sizeof(leaf->ents[0]);
1193 if (bytes > (state->blocksize >> 1)) {
1195 * Blk over 50%, don't try to join.
1201 * Check for the degenerate case of the block being empty.
1202 * If the block is empty, we'll simply delete it, no need to
1203 * coalesce it with a sibling block. We choose (arbitrarily)
1204 * to merge with the forward block unless it is NULL.
1208 * Make altpath point to the block we want to keep and
1209 * path point to the block we want to drop (this one).
1211 forward = (info->forw != 0);
1212 memcpy(&state->altpath, &state->path, sizeof(state->path));
1213 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1217 *action = rval ? 2 : 0;
1221 * Examine each sibling block to see if we can coalesce with
1222 * at least 25% free space to spare. We need to figure out
1223 * whether to merge with the forward or the backward block.
1224 * We prefer coalescing with the lower numbered sibling so as
1225 * to shrink a directory over time.
1227 forward = be32_to_cpu(info->forw) < be32_to_cpu(info->back);
1228 for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1229 blkno = forward ? be32_to_cpu(info->forw) : be32_to_cpu(info->back);
1233 * Read the sibling leaf block.
1235 error = xfs_da_read_buf(state->args->trans, state->args->dp,
1236 blkno, -1, &bp, XFS_DATA_FORK, NULL);
1241 * Count bytes in the two blocks combined.
1243 leaf = (xfs_dir2_leaf_t *)info;
1244 count = be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1245 bytes = state->blocksize - (state->blocksize >> 2);
1247 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1248 count += be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale);
1249 bytes -= count * (uint)sizeof(leaf->ents[0]);
1251 * Fits with at least 25% to spare.
1255 xfs_trans_brelse(state->args->trans, bp);
1258 * Didn't like either block, give up.
1266 * Make altpath point to the block we want to keep (the lower
1267 * numbered block) and path point to the block we want to drop.
1269 memcpy(&state->altpath, &state->path, sizeof(state->path));
1270 if (blkno < blk->blkno)
1271 error = xfs_da_path_shift(state, &state->altpath, forward, 0,
1274 error = xfs_da_path_shift(state, &state->path, forward, 0,
1279 *action = rval ? 0 : 1;
1284 * Move all the leaf entries from drop_blk to save_blk.
1285 * This is done as part of a join operation.
1288 xfs_dir2_leafn_unbalance(
1289 xfs_da_state_t *state, /* cursor */
1290 xfs_da_state_blk_t *drop_blk, /* dead block */
1291 xfs_da_state_blk_t *save_blk) /* surviving block */
1293 xfs_da_args_t *args; /* operation arguments */
1294 xfs_dir2_leaf_t *drop_leaf; /* dead leaf structure */
1295 xfs_dir2_leaf_t *save_leaf; /* surviving leaf structure */
1298 ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1299 ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1300 drop_leaf = drop_blk->bp->b_addr;
1301 save_leaf = save_blk->bp->b_addr;
1302 ASSERT(drop_leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1303 ASSERT(save_leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC));
1305 * If there are any stale leaf entries, take this opportunity
1308 if (drop_leaf->hdr.stale)
1309 xfs_dir2_leaf_compact(args, drop_blk->bp);
1310 if (save_leaf->hdr.stale)
1311 xfs_dir2_leaf_compact(args, save_blk->bp);
1313 * Move the entries from drop to the appropriate end of save.
1315 drop_blk->hashval = be32_to_cpu(drop_leaf->ents[be16_to_cpu(drop_leaf->hdr.count) - 1].hashval);
1316 if (xfs_dir2_leafn_order(save_blk->bp, drop_blk->bp))
1317 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp, 0,
1318 be16_to_cpu(drop_leaf->hdr.count));
1320 xfs_dir2_leafn_moveents(args, drop_blk->bp, 0, save_blk->bp,
1321 be16_to_cpu(save_leaf->hdr.count), be16_to_cpu(drop_leaf->hdr.count));
1322 save_blk->hashval = be32_to_cpu(save_leaf->ents[be16_to_cpu(save_leaf->hdr.count) - 1].hashval);
1323 xfs_dir2_leafn_check(args->dp, save_blk->bp);
1327 * Top-level node form directory addname routine.
1330 xfs_dir2_node_addname(
1331 xfs_da_args_t *args) /* operation arguments */
1333 xfs_da_state_blk_t *blk; /* leaf block for insert */
1334 int error; /* error return value */
1335 int rval; /* sub-return value */
1336 xfs_da_state_t *state; /* btree cursor */
1338 trace_xfs_dir2_node_addname(args);
1341 * Allocate and initialize the state (btree cursor).
1343 state = xfs_da_state_alloc();
1345 state->mp = args->dp->i_mount;
1346 state->blocksize = state->mp->m_dirblksize;
1347 state->node_ents = state->mp->m_dir_node_ents;
1349 * Look up the name. We're not supposed to find it, but
1350 * this gives us the insertion point.
1352 error = xfs_da_node_lookup_int(state, &rval);
1355 if (rval != ENOENT) {
1359 * Add the data entry to a data block.
1360 * Extravalid is set to a freeblock found by lookup.
1362 rval = xfs_dir2_node_addname_int(args,
1363 state->extravalid ? &state->extrablk : NULL);
1367 blk = &state->path.blk[state->path.active - 1];
1368 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1370 * Add the new leaf entry.
1372 rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1375 * It worked, fix the hash values up the btree.
1377 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1378 xfs_da_fixhashpath(state, &state->path);
1381 * It didn't work, we need to split the leaf block.
1383 if (args->total == 0) {
1384 ASSERT(rval == ENOSPC);
1388 * Split the leaf block and insert the new entry.
1390 rval = xfs_da_split(state);
1393 xfs_da_state_free(state);
1398 * Add the data entry for a node-format directory name addition.
1399 * The leaf entry is added in xfs_dir2_leafn_add.
1400 * We may enter with a freespace block that the lookup found.
1402 static int /* error */
1403 xfs_dir2_node_addname_int(
1404 xfs_da_args_t *args, /* operation arguments */
1405 xfs_da_state_blk_t *fblk) /* optional freespace block */
1407 xfs_dir2_data_hdr_t *hdr; /* data block header */
1408 xfs_dir2_db_t dbno; /* data block number */
1409 struct xfs_buf *dbp; /* data block buffer */
1410 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1411 xfs_inode_t *dp; /* incore directory inode */
1412 xfs_dir2_data_unused_t *dup; /* data unused entry pointer */
1413 int error; /* error return value */
1414 xfs_dir2_db_t fbno; /* freespace block number */
1415 struct xfs_buf *fbp; /* freespace buffer */
1416 int findex; /* freespace entry index */
1417 xfs_dir2_free_t *free=NULL; /* freespace block structure */
1418 xfs_dir2_db_t ifbno; /* initial freespace block no */
1419 xfs_dir2_db_t lastfbno=0; /* highest freespace block no */
1420 int length; /* length of the new entry */
1421 int logfree; /* need to log free entry */
1422 xfs_mount_t *mp; /* filesystem mount point */
1423 int needlog; /* need to log data header */
1424 int needscan; /* need to rescan data frees */
1425 __be16 *tagp; /* data entry tag pointer */
1426 xfs_trans_t *tp; /* transaction pointer */
1431 length = xfs_dir2_data_entsize(args->namelen);
1433 * If we came in with a freespace block that means that lookup
1434 * found an entry with our hash value. This is the freespace
1435 * block for that data entry.
1440 * Remember initial freespace block number.
1442 ifbno = fblk->blkno;
1444 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1445 findex = fblk->index;
1447 * This means the free entry showed that the data block had
1448 * space for our entry, so we remembered it.
1449 * Use that data block.
1452 ASSERT(findex < be32_to_cpu(free->hdr.nvalid));
1453 ASSERT(be16_to_cpu(free->bests[findex]) != NULLDATAOFF);
1454 ASSERT(be16_to_cpu(free->bests[findex]) >= length);
1455 dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1458 * The data block looked at didn't have enough room.
1459 * We'll start at the beginning of the freespace entries.
1467 * Didn't come in with a freespace block, so don't have a data block.
1475 * If we don't have a data block yet, we're going to scan the
1476 * freespace blocks looking for one. Figure out what the
1477 * highest freespace block number is.
1480 xfs_fileoff_t fo; /* freespace block number */
1482 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK)))
1484 lastfbno = xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo);
1488 * While we haven't identified a data block, search the freeblock
1489 * data for a good data block. If we find a null freeblock entry,
1490 * indicating a hole in the data blocks, remember that.
1492 while (dbno == -1) {
1494 * If we don't have a freeblock in hand, get the next one.
1498 * Happens the first time through unless lookup gave
1499 * us a freespace block to start with.
1502 fbno = XFS_DIR2_FREE_FIRSTDB(mp);
1504 * If it's ifbno we already looked at it.
1509 * If it's off the end we're done.
1511 if (fbno >= lastfbno)
1514 * Read the block. There can be holes in the
1515 * freespace blocks, so this might not succeed.
1516 * This should be really rare, so there's no reason
1519 error = xfs_dir2_free_try_read(tp, dp,
1520 xfs_dir2_db_to_da(mp, fbno),
1527 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1531 * Look at the current free entry. Is it good enough?
1533 if (be16_to_cpu(free->bests[findex]) != NULLDATAOFF &&
1534 be16_to_cpu(free->bests[findex]) >= length)
1535 dbno = be32_to_cpu(free->hdr.firstdb) + findex;
1538 * Are we done with the freeblock?
1540 if (++findex == be32_to_cpu(free->hdr.nvalid)) {
1544 xfs_trans_brelse(tp, fbp);
1546 if (fblk && fblk->bp)
1552 * If we don't have a data block, we need to allocate one and make
1553 * the freespace entries refer to it.
1555 if (unlikely(dbno == -1)) {
1557 * Not allowed to allocate, return failure.
1559 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1560 return XFS_ERROR(ENOSPC);
1563 * Allocate and initialize the new data block.
1565 if (unlikely((error = xfs_dir2_grow_inode(args,
1566 XFS_DIR2_DATA_SPACE,
1568 (error = xfs_dir2_data_init(args, dbno, &dbp))))
1572 * If (somehow) we have a freespace block, get rid of it.
1575 xfs_trans_brelse(tp, fbp);
1576 if (fblk && fblk->bp)
1580 * Get the freespace block corresponding to the data block
1581 * that was just allocated.
1583 fbno = xfs_dir2_db_to_fdb(mp, dbno);
1584 error = xfs_dir2_free_try_read(tp, dp,
1585 xfs_dir2_db_to_da(mp, fbno),
1591 * If there wasn't a freespace block, the read will
1592 * return a NULL fbp. Allocate and initialize a new one.
1595 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1600 if (unlikely(xfs_dir2_db_to_fdb(mp, dbno) != fbno)) {
1602 "%s: dir ino %llu needed freesp block %lld for\n"
1603 " data block %lld, got %lld ifbno %llu lastfbno %d",
1604 __func__, (unsigned long long)dp->i_ino,
1605 (long long)xfs_dir2_db_to_fdb(mp, dbno),
1606 (long long)dbno, (long long)fbno,
1607 (unsigned long long)ifbno, lastfbno);
1610 " fblk 0x%p blkno %llu index %d magic 0x%x",
1612 (unsigned long long)fblk->blkno,
1616 xfs_alert(mp, " ... fblk is NULL");
1618 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1619 XFS_ERRLEVEL_LOW, mp);
1620 return XFS_ERROR(EFSCORRUPTED);
1624 * Get a buffer for the new block.
1626 if ((error = xfs_da_get_buf(tp, dp,
1627 xfs_dir2_db_to_da(mp, fbno),
1628 -1, &fbp, XFS_DATA_FORK))) {
1631 ASSERT(fbp != NULL);
1634 * Initialize the new block to be empty, and remember
1635 * its first slot as our empty slot.
1638 free->hdr.magic = cpu_to_be32(XFS_DIR2_FREE_MAGIC);
1639 free->hdr.firstdb = cpu_to_be32(
1640 (fbno - XFS_DIR2_FREE_FIRSTDB(mp)) *
1641 xfs_dir2_free_max_bests(mp));
1642 free->hdr.nvalid = 0;
1643 free->hdr.nused = 0;
1646 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1650 * Set the freespace block index from the data block number.
1652 findex = xfs_dir2_db_to_fdindex(mp, dbno);
1654 * If it's after the end of the current entries in the
1655 * freespace block, extend that table.
1657 if (findex >= be32_to_cpu(free->hdr.nvalid)) {
1658 ASSERT(findex < xfs_dir2_free_max_bests(mp));
1659 free->hdr.nvalid = cpu_to_be32(findex + 1);
1661 * Tag new entry so nused will go up.
1663 free->bests[findex] = cpu_to_be16(NULLDATAOFF);
1666 * If this entry was for an empty data block
1667 * (this should always be true) then update the header.
1669 if (free->bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1670 be32_add_cpu(&free->hdr.nused, 1);
1671 xfs_dir2_free_log_header(tp, fbp);
1674 * Update the real value in the table.
1675 * We haven't allocated the data entry yet so this will
1679 free->bests[findex] = hdr->bestfree[0].length;
1683 * We had a data block so we don't have to make a new one.
1687 * If just checking, we succeeded.
1689 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1693 * Read the data block in.
1695 error = xfs_da_read_buf(tp, dp, xfs_dir2_db_to_da(mp, dbno),
1696 -1, &dbp, XFS_DATA_FORK, NULL);
1702 ASSERT(be16_to_cpu(hdr->bestfree[0].length) >= length);
1704 * Point to the existing unused space.
1706 dup = (xfs_dir2_data_unused_t *)
1707 ((char *)hdr + be16_to_cpu(hdr->bestfree[0].offset));
1708 needscan = needlog = 0;
1710 * Mark the first part of the unused space, inuse for us.
1712 xfs_dir2_data_use_free(tp, dbp, dup,
1713 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
1714 &needlog, &needscan);
1716 * Fill in the new entry and log it.
1718 dep = (xfs_dir2_data_entry_t *)dup;
1719 dep->inumber = cpu_to_be64(args->inumber);
1720 dep->namelen = args->namelen;
1721 memcpy(dep->name, args->name, dep->namelen);
1722 tagp = xfs_dir2_data_entry_tag_p(dep);
1723 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1724 xfs_dir2_data_log_entry(tp, dbp, dep);
1726 * Rescan the block for bestfree if needed.
1729 xfs_dir2_data_freescan(mp, hdr, &needlog);
1731 * Log the data block header if needed.
1734 xfs_dir2_data_log_header(tp, dbp);
1736 * If the freespace entry is now wrong, update it.
1738 if (be16_to_cpu(free->bests[findex]) != be16_to_cpu(hdr->bestfree[0].length)) {
1739 free->bests[findex] = hdr->bestfree[0].length;
1743 * Log the freespace entry if needed.
1746 xfs_dir2_free_log_bests(tp, fbp, findex, findex);
1748 * Return the data block and offset in args, then drop the data block.
1750 args->blkno = (xfs_dablk_t)dbno;
1751 args->index = be16_to_cpu(*tagp);
1756 * Lookup an entry in a node-format directory.
1757 * All the real work happens in xfs_da_node_lookup_int.
1758 * The only real output is the inode number of the entry.
1761 xfs_dir2_node_lookup(
1762 xfs_da_args_t *args) /* operation arguments */
1764 int error; /* error return value */
1765 int i; /* btree level */
1766 int rval; /* operation return value */
1767 xfs_da_state_t *state; /* btree cursor */
1769 trace_xfs_dir2_node_lookup(args);
1772 * Allocate and initialize the btree cursor.
1774 state = xfs_da_state_alloc();
1776 state->mp = args->dp->i_mount;
1777 state->blocksize = state->mp->m_dirblksize;
1778 state->node_ents = state->mp->m_dir_node_ents;
1780 * Fill in the path to the entry in the cursor.
1782 error = xfs_da_node_lookup_int(state, &rval);
1785 else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
1786 /* If a CI match, dup the actual name and return EEXIST */
1787 xfs_dir2_data_entry_t *dep;
1789 dep = (xfs_dir2_data_entry_t *)
1790 ((char *)state->extrablk.bp->b_addr +
1791 state->extrablk.index);
1792 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1795 * Release the btree blocks and leaf block.
1797 for (i = 0; i < state->path.active; i++) {
1798 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
1799 state->path.blk[i].bp = NULL;
1802 * Release the data block if we have it.
1804 if (state->extravalid && state->extrablk.bp) {
1805 xfs_trans_brelse(args->trans, state->extrablk.bp);
1806 state->extrablk.bp = NULL;
1808 xfs_da_state_free(state);
1813 * Remove an entry from a node-format directory.
1816 xfs_dir2_node_removename(
1817 xfs_da_args_t *args) /* operation arguments */
1819 xfs_da_state_blk_t *blk; /* leaf block */
1820 int error; /* error return value */
1821 int rval; /* operation return value */
1822 xfs_da_state_t *state; /* btree cursor */
1824 trace_xfs_dir2_node_removename(args);
1827 * Allocate and initialize the btree cursor.
1829 state = xfs_da_state_alloc();
1831 state->mp = args->dp->i_mount;
1832 state->blocksize = state->mp->m_dirblksize;
1833 state->node_ents = state->mp->m_dir_node_ents;
1835 * Look up the entry we're deleting, set up the cursor.
1837 error = xfs_da_node_lookup_int(state, &rval);
1841 * Didn't find it, upper layer screwed up.
1843 if (rval != EEXIST) {
1844 xfs_da_state_free(state);
1847 blk = &state->path.blk[state->path.active - 1];
1848 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1849 ASSERT(state->extravalid);
1851 * Remove the leaf and data entries.
1852 * Extrablk refers to the data block.
1854 error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
1855 &state->extrablk, &rval);
1859 * Fix the hash values up the btree.
1861 xfs_da_fixhashpath(state, &state->path);
1863 * If we need to join leaf blocks, do it.
1865 if (rval && state->path.active > 1)
1866 error = xfs_da_join(state);
1868 * If no errors so far, try conversion to leaf format.
1871 error = xfs_dir2_node_to_leaf(state);
1872 xfs_da_state_free(state);
1877 * Replace an entry's inode number in a node-format directory.
1880 xfs_dir2_node_replace(
1881 xfs_da_args_t *args) /* operation arguments */
1883 xfs_da_state_blk_t *blk; /* leaf block */
1884 xfs_dir2_data_hdr_t *hdr; /* data block header */
1885 xfs_dir2_data_entry_t *dep; /* data entry changed */
1886 int error; /* error return value */
1887 int i; /* btree level */
1888 xfs_ino_t inum; /* new inode number */
1889 xfs_dir2_leaf_t *leaf; /* leaf structure */
1890 xfs_dir2_leaf_entry_t *lep; /* leaf entry being changed */
1891 int rval; /* internal return value */
1892 xfs_da_state_t *state; /* btree cursor */
1894 trace_xfs_dir2_node_replace(args);
1897 * Allocate and initialize the btree cursor.
1899 state = xfs_da_state_alloc();
1901 state->mp = args->dp->i_mount;
1902 state->blocksize = state->mp->m_dirblksize;
1903 state->node_ents = state->mp->m_dir_node_ents;
1904 inum = args->inumber;
1906 * Lookup the entry to change in the btree.
1908 error = xfs_da_node_lookup_int(state, &rval);
1913 * It should be found, since the vnodeops layer has looked it up
1914 * and locked it. But paranoia is good.
1916 if (rval == EEXIST) {
1918 * Find the leaf entry.
1920 blk = &state->path.blk[state->path.active - 1];
1921 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1922 leaf = blk->bp->b_addr;
1923 lep = &leaf->ents[blk->index];
1924 ASSERT(state->extravalid);
1926 * Point to the data entry.
1928 hdr = state->extrablk.bp->b_addr;
1929 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
1930 dep = (xfs_dir2_data_entry_t *)
1932 xfs_dir2_dataptr_to_off(state->mp, be32_to_cpu(lep->address)));
1933 ASSERT(inum != be64_to_cpu(dep->inumber));
1935 * Fill in the new inode number and log the entry.
1937 dep->inumber = cpu_to_be64(inum);
1938 xfs_dir2_data_log_entry(args->trans, state->extrablk.bp, dep);
1942 * Didn't find it, and we're holding a data block. Drop it.
1944 else if (state->extravalid) {
1945 xfs_trans_brelse(args->trans, state->extrablk.bp);
1946 state->extrablk.bp = NULL;
1949 * Release all the buffers in the cursor.
1951 for (i = 0; i < state->path.active; i++) {
1952 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
1953 state->path.blk[i].bp = NULL;
1955 xfs_da_state_free(state);
1960 * Trim off a trailing empty freespace block.
1961 * Return (in rvalp) 1 if we did it, 0 if not.
1964 xfs_dir2_node_trim_free(
1965 xfs_da_args_t *args, /* operation arguments */
1966 xfs_fileoff_t fo, /* free block number */
1967 int *rvalp) /* out: did something */
1969 struct xfs_buf *bp; /* freespace buffer */
1970 xfs_inode_t *dp; /* incore directory inode */
1971 int error; /* error return code */
1972 xfs_dir2_free_t *free; /* freespace structure */
1973 xfs_mount_t *mp; /* filesystem mount point */
1974 xfs_trans_t *tp; /* transaction pointer */
1980 * Read the freespace block.
1982 error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
1986 * There can be holes in freespace. If fo is a hole, there's
1992 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC));
1994 * If there are used entries, there's nothing to do.
1996 if (be32_to_cpu(free->hdr.nused) > 0) {
1997 xfs_trans_brelse(tp, bp);
2002 * Blow the block away.
2005 xfs_dir2_shrink_inode(args, xfs_dir2_da_to_db(mp, (xfs_dablk_t)fo),
2008 * Can't fail with ENOSPC since that only happens with no
2009 * space reservation, when breaking up an extent into two
2010 * pieces. This is the last block of an extent.
2012 ASSERT(error != ENOSPC);
2013 xfs_trans_brelse(tp, bp);
2017 * Return that we succeeded.