1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_da_format.h"
14 #include "xfs_da_btree.h"
15 #include "xfs_inode.h"
18 #include "xfs_dir2_priv.h"
19 #include "xfs_error.h"
20 #include "xfs_trace.h"
21 #include "xfs_trans.h"
22 #include "xfs_buf_item.h"
23 #include "xfs_cksum.h"
27 * Function declarations.
29 static int xfs_dir2_leafn_add(struct xfs_buf *bp, xfs_da_args_t *args,
31 static void xfs_dir2_leafn_rebalance(xfs_da_state_t *state,
32 xfs_da_state_blk_t *blk1,
33 xfs_da_state_blk_t *blk2);
34 static int xfs_dir2_leafn_remove(xfs_da_args_t *args, struct xfs_buf *bp,
35 int index, xfs_da_state_blk_t *dblk,
37 static int xfs_dir2_node_addname_int(xfs_da_args_t *args,
38 xfs_da_state_blk_t *fblk);
41 * Check internal consistency of a leafn block.
49 struct xfs_dir2_leaf *leaf = bp->b_addr;
50 struct xfs_dir3_icleaf_hdr leafhdr;
52 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
54 if (leafhdr.magic == XFS_DIR3_LEAFN_MAGIC) {
55 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
56 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
57 return __this_address;
58 } else if (leafhdr.magic != XFS_DIR2_LEAFN_MAGIC)
59 return __this_address;
61 return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
71 fa = xfs_dir3_leafn_check(dp, bp);
74 xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
75 bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
80 #define xfs_dir3_leaf_check(dp, bp)
87 struct xfs_mount *mp = bp->b_target->bt_mount;
88 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
90 if (!xfs_verify_magic(bp, hdr->magic))
91 return __this_address;
93 if (xfs_sb_version_hascrc(&mp->m_sb)) {
94 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
96 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
97 return __this_address;
98 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
99 return __this_address;
100 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
101 return __this_address;
104 /* XXX: should bounds check the xfs_dir3_icfree_hdr here */
110 xfs_dir3_free_read_verify(
113 struct xfs_mount *mp = bp->b_target->bt_mount;
116 if (xfs_sb_version_hascrc(&mp->m_sb) &&
117 !xfs_buf_verify_cksum(bp, XFS_DIR3_FREE_CRC_OFF))
118 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
120 fa = xfs_dir3_free_verify(bp);
122 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
127 xfs_dir3_free_write_verify(
130 struct xfs_mount *mp = bp->b_target->bt_mount;
131 struct xfs_buf_log_item *bip = bp->b_log_item;
132 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
135 fa = xfs_dir3_free_verify(bp);
137 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
141 if (!xfs_sb_version_hascrc(&mp->m_sb))
145 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
147 xfs_buf_update_cksum(bp, XFS_DIR3_FREE_CRC_OFF);
150 const struct xfs_buf_ops xfs_dir3_free_buf_ops = {
151 .name = "xfs_dir3_free",
152 .magic = { cpu_to_be32(XFS_DIR2_FREE_MAGIC),
153 cpu_to_be32(XFS_DIR3_FREE_MAGIC) },
154 .verify_read = xfs_dir3_free_read_verify,
155 .verify_write = xfs_dir3_free_write_verify,
156 .verify_struct = xfs_dir3_free_verify,
159 /* Everything ok in the free block header? */
160 static xfs_failaddr_t
161 xfs_dir3_free_header_check(
162 struct xfs_inode *dp,
166 struct xfs_mount *mp = dp->i_mount;
167 unsigned int firstdb;
170 maxbests = dp->d_ops->free_max_bests(mp->m_dir_geo);
171 firstdb = (xfs_dir2_da_to_db(mp->m_dir_geo, fbno) -
172 xfs_dir2_byte_to_db(mp->m_dir_geo, XFS_DIR2_FREE_OFFSET)) *
174 if (xfs_sb_version_hascrc(&mp->m_sb)) {
175 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
177 if (be32_to_cpu(hdr3->firstdb) != firstdb)
178 return __this_address;
179 if (be32_to_cpu(hdr3->nvalid) > maxbests)
180 return __this_address;
181 if (be32_to_cpu(hdr3->nvalid) < be32_to_cpu(hdr3->nused))
182 return __this_address;
184 struct xfs_dir2_free_hdr *hdr = bp->b_addr;
186 if (be32_to_cpu(hdr->firstdb) != firstdb)
187 return __this_address;
188 if (be32_to_cpu(hdr->nvalid) > maxbests)
189 return __this_address;
190 if (be32_to_cpu(hdr->nvalid) < be32_to_cpu(hdr->nused))
191 return __this_address;
197 __xfs_dir3_free_read(
198 struct xfs_trans *tp,
199 struct xfs_inode *dp,
201 xfs_daddr_t mappedbno,
202 struct xfs_buf **bpp)
207 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
208 XFS_DATA_FORK, &xfs_dir3_free_buf_ops);
212 /* Check things that we can't do in the verifier. */
213 fa = xfs_dir3_free_header_check(dp, fbno, *bpp);
215 xfs_verifier_error(*bpp, -EFSCORRUPTED, fa);
216 xfs_trans_brelse(tp, *bpp);
217 return -EFSCORRUPTED;
220 /* try read returns without an error or *bpp if it lands in a hole */
222 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_FREE_BUF);
229 struct xfs_trans *tp,
230 struct xfs_inode *dp,
232 struct xfs_buf **bpp)
234 return __xfs_dir3_free_read(tp, dp, fbno, -1, bpp);
238 xfs_dir2_free_try_read(
239 struct xfs_trans *tp,
240 struct xfs_inode *dp,
242 struct xfs_buf **bpp)
244 return __xfs_dir3_free_read(tp, dp, fbno, -2, bpp);
248 xfs_dir3_free_get_buf(
251 struct xfs_buf **bpp)
253 struct xfs_trans *tp = args->trans;
254 struct xfs_inode *dp = args->dp;
255 struct xfs_mount *mp = dp->i_mount;
258 struct xfs_dir3_icfree_hdr hdr;
260 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, fbno),
261 -1, &bp, XFS_DATA_FORK);
265 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_FREE_BUF);
266 bp->b_ops = &xfs_dir3_free_buf_ops;
269 * Initialize the new block to be empty, and remember
270 * its first slot as our empty slot.
272 memset(bp->b_addr, 0, sizeof(struct xfs_dir3_free_hdr));
273 memset(&hdr, 0, sizeof(hdr));
275 if (xfs_sb_version_hascrc(&mp->m_sb)) {
276 struct xfs_dir3_free_hdr *hdr3 = bp->b_addr;
278 hdr.magic = XFS_DIR3_FREE_MAGIC;
280 hdr3->hdr.blkno = cpu_to_be64(bp->b_bn);
281 hdr3->hdr.owner = cpu_to_be64(dp->i_ino);
282 uuid_copy(&hdr3->hdr.uuid, &mp->m_sb.sb_meta_uuid);
284 hdr.magic = XFS_DIR2_FREE_MAGIC;
285 dp->d_ops->free_hdr_to_disk(bp->b_addr, &hdr);
291 * Log entries from a freespace block.
294 xfs_dir2_free_log_bests(
295 struct xfs_da_args *args,
297 int first, /* first entry to log */
298 int last) /* last entry to log */
300 xfs_dir2_free_t *free; /* freespace structure */
304 bests = args->dp->d_ops->free_bests_p(free);
305 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
306 free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
307 xfs_trans_log_buf(args->trans, bp,
308 (uint)((char *)&bests[first] - (char *)free),
309 (uint)((char *)&bests[last] - (char *)free +
310 sizeof(bests[0]) - 1));
314 * Log header from a freespace block.
317 xfs_dir2_free_log_header(
318 struct xfs_da_args *args,
322 xfs_dir2_free_t *free; /* freespace structure */
325 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
326 free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
328 xfs_trans_log_buf(args->trans, bp, 0,
329 args->dp->d_ops->free_hdr_size - 1);
333 * Convert a leaf-format directory to a node-format directory.
334 * We need to change the magic number of the leaf block, and copy
335 * the freespace table out of the leaf block into its own block.
338 xfs_dir2_leaf_to_node(
339 xfs_da_args_t *args, /* operation arguments */
340 struct xfs_buf *lbp) /* leaf buffer */
342 xfs_inode_t *dp; /* incore directory inode */
343 int error; /* error return value */
344 struct xfs_buf *fbp; /* freespace buffer */
345 xfs_dir2_db_t fdb; /* freespace block number */
346 xfs_dir2_free_t *free; /* freespace structure */
347 __be16 *from; /* pointer to freespace entry */
348 int i; /* leaf freespace index */
349 xfs_dir2_leaf_t *leaf; /* leaf structure */
350 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
351 int n; /* count of live freespc ents */
352 xfs_dir2_data_off_t off; /* freespace entry value */
353 __be16 *to; /* pointer to freespace entry */
354 xfs_trans_t *tp; /* transaction pointer */
355 struct xfs_dir3_icfree_hdr freehdr;
357 trace_xfs_dir2_leaf_to_node(args);
362 * Add a freespace block to the directory.
364 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE, &fdb))) {
367 ASSERT(fdb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
369 * Get the buffer for the new freespace block.
371 error = xfs_dir3_free_get_buf(args, fdb, &fbp);
376 dp->d_ops->free_hdr_from_disk(&freehdr, free);
378 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
379 if (be32_to_cpu(ltp->bestcount) >
380 (uint)dp->i_d.di_size / args->geo->blksize)
381 return -EFSCORRUPTED;
384 * Copy freespace entries from the leaf block to the new block.
385 * Count active entries.
387 from = xfs_dir2_leaf_bests_p(ltp);
388 to = dp->d_ops->free_bests_p(free);
389 for (i = n = 0; i < be32_to_cpu(ltp->bestcount); i++, from++, to++) {
390 if ((off = be16_to_cpu(*from)) != NULLDATAOFF)
392 *to = cpu_to_be16(off);
396 * Now initialize the freespace block header.
399 freehdr.nvalid = be32_to_cpu(ltp->bestcount);
401 dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
402 xfs_dir2_free_log_bests(args, fbp, 0, freehdr.nvalid - 1);
403 xfs_dir2_free_log_header(args, fbp);
406 * Converting the leaf to a leafnode is just a matter of changing the
407 * magic number and the ops. Do the change directly to the buffer as
408 * it's less work (and less code) than decoding the header to host
409 * format and back again.
411 if (leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC))
412 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR2_LEAFN_MAGIC);
414 leaf->hdr.info.magic = cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
415 lbp->b_ops = &xfs_dir3_leafn_buf_ops;
416 xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAFN_BUF);
417 xfs_dir3_leaf_log_header(args, lbp);
418 xfs_dir3_leaf_check(dp, lbp);
423 * Add a leaf entry to a leaf block in a node-form directory.
424 * The other work necessary is done from the caller.
426 static int /* error */
428 struct xfs_buf *bp, /* leaf buffer */
429 struct xfs_da_args *args, /* operation arguments */
430 int index) /* insertion pt for new entry */
432 struct xfs_dir3_icleaf_hdr leafhdr;
433 struct xfs_inode *dp = args->dp;
434 struct xfs_dir2_leaf *leaf = bp->b_addr;
435 struct xfs_dir2_leaf_entry *lep;
436 struct xfs_dir2_leaf_entry *ents;
437 int compact; /* compacting stale leaves */
438 int highstale = 0; /* next stale entry */
439 int lfloghigh; /* high leaf entry logging */
440 int lfloglow; /* low leaf entry logging */
441 int lowstale = 0; /* previous stale entry */
443 trace_xfs_dir2_leafn_add(args, index);
445 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
446 ents = dp->d_ops->leaf_ents_p(leaf);
449 * Quick check just to make sure we are not going to index
450 * into other peoples memory
453 return -EFSCORRUPTED;
456 * If there are already the maximum number of leaf entries in
457 * the block, if there are no stale entries it won't fit.
458 * Caller will do a split. If there are stale entries we'll do
462 if (leafhdr.count == dp->d_ops->leaf_max_ents(args->geo)) {
465 compact = leafhdr.stale > 1;
468 ASSERT(index == 0 || be32_to_cpu(ents[index - 1].hashval) <= args->hashval);
469 ASSERT(index == leafhdr.count ||
470 be32_to_cpu(ents[index].hashval) >= args->hashval);
472 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
476 * Compact out all but one stale leaf entry. Leaves behind
477 * the entry closest to index.
480 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
481 &highstale, &lfloglow, &lfloghigh);
482 else if (leafhdr.stale) {
484 * Set impossible logging indices for this case.
486 lfloglow = leafhdr.count;
491 * Insert the new entry, log everything.
493 lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
494 highstale, &lfloglow, &lfloghigh);
496 lep->hashval = cpu_to_be32(args->hashval);
497 lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(args->geo,
498 args->blkno, args->index));
500 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
501 xfs_dir3_leaf_log_header(args, bp);
502 xfs_dir3_leaf_log_ents(args, bp, lfloglow, lfloghigh);
503 xfs_dir3_leaf_check(dp, bp);
509 xfs_dir2_free_hdr_check(
510 struct xfs_inode *dp,
514 struct xfs_dir3_icfree_hdr hdr;
516 dp->d_ops->free_hdr_from_disk(&hdr, bp->b_addr);
518 ASSERT((hdr.firstdb %
519 dp->d_ops->free_max_bests(dp->i_mount->m_dir_geo)) == 0);
520 ASSERT(hdr.firstdb <= db);
521 ASSERT(db < hdr.firstdb + hdr.nvalid);
524 #define xfs_dir2_free_hdr_check(dp, bp, db)
528 * Return the last hash value in the leaf.
529 * Stale entries are ok.
531 xfs_dahash_t /* hash value */
532 xfs_dir2_leaf_lasthash(
533 struct xfs_inode *dp,
534 struct xfs_buf *bp, /* leaf buffer */
535 int *count) /* count of entries in leaf */
537 struct xfs_dir2_leaf *leaf = bp->b_addr;
538 struct xfs_dir2_leaf_entry *ents;
539 struct xfs_dir3_icleaf_hdr leafhdr;
541 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
543 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
544 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC ||
545 leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
546 leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
549 *count = leafhdr.count;
553 ents = dp->d_ops->leaf_ents_p(leaf);
554 return be32_to_cpu(ents[leafhdr.count - 1].hashval);
558 * Look up a leaf entry for space to add a name in a node-format leaf block.
559 * The extrablk in state is a freespace block.
562 xfs_dir2_leafn_lookup_for_addname(
563 struct xfs_buf *bp, /* leaf buffer */
564 xfs_da_args_t *args, /* operation arguments */
565 int *indexp, /* out: leaf entry index */
566 xfs_da_state_t *state) /* state to fill in */
568 struct xfs_buf *curbp = NULL; /* current data/free buffer */
569 xfs_dir2_db_t curdb = -1; /* current data block number */
570 xfs_dir2_db_t curfdb = -1; /* current free block number */
571 xfs_inode_t *dp; /* incore directory inode */
572 int error; /* error return value */
573 int fi; /* free entry index */
574 xfs_dir2_free_t *free = NULL; /* free block structure */
575 int index; /* leaf entry index */
576 xfs_dir2_leaf_t *leaf; /* leaf structure */
577 int length; /* length of new data entry */
578 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
579 xfs_mount_t *mp; /* filesystem mount point */
580 xfs_dir2_db_t newdb; /* new data block number */
581 xfs_dir2_db_t newfdb; /* new free block number */
582 xfs_trans_t *tp; /* transaction pointer */
583 struct xfs_dir2_leaf_entry *ents;
584 struct xfs_dir3_icleaf_hdr leafhdr;
590 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
591 ents = dp->d_ops->leaf_ents_p(leaf);
593 xfs_dir3_leaf_check(dp, bp);
594 ASSERT(leafhdr.count > 0);
597 * Look up the hash value in the leaf entries.
599 index = xfs_dir2_leaf_search_hash(args, bp);
601 * Do we have a buffer coming in?
603 if (state->extravalid) {
604 /* If so, it's a free block buffer, get the block number. */
605 curbp = state->extrablk.bp;
606 curfdb = state->extrablk.blkno;
607 free = curbp->b_addr;
608 ASSERT(free->hdr.magic == cpu_to_be32(XFS_DIR2_FREE_MAGIC) ||
609 free->hdr.magic == cpu_to_be32(XFS_DIR3_FREE_MAGIC));
611 length = dp->d_ops->data_entsize(args->namelen);
613 * Loop over leaf entries with the right hash value.
615 for (lep = &ents[index];
616 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
619 * Skip stale leaf entries.
621 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
624 * Pull the data block number from the entry.
626 newdb = xfs_dir2_dataptr_to_db(args->geo,
627 be32_to_cpu(lep->address));
629 * For addname, we're looking for a place to put the new entry.
630 * We want to use a data block with an entry of equal
631 * hash value to ours if there is one with room.
633 * If this block isn't the data block we already have
634 * in hand, take a look at it.
636 if (newdb != curdb) {
641 * Convert the data block to the free block
642 * holding its freespace information.
644 newfdb = dp->d_ops->db_to_fdb(args->geo, newdb);
646 * If it's not the one we have in hand, read it in.
648 if (newfdb != curfdb) {
650 * If we had one before, drop it.
653 xfs_trans_brelse(tp, curbp);
655 error = xfs_dir2_free_read(tp, dp,
656 xfs_dir2_db_to_da(args->geo,
661 free = curbp->b_addr;
663 xfs_dir2_free_hdr_check(dp, curbp, curdb);
666 * Get the index for our entry.
668 fi = dp->d_ops->db_to_fdindex(args->geo, curdb);
670 * If it has room, return it.
672 bests = dp->d_ops->free_bests_p(free);
673 if (unlikely(bests[fi] == cpu_to_be16(NULLDATAOFF))) {
674 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
675 XFS_ERRLEVEL_LOW, mp);
676 if (curfdb != newfdb)
677 xfs_trans_brelse(tp, curbp);
678 return -EFSCORRUPTED;
681 if (be16_to_cpu(bests[fi]) >= length)
685 /* Didn't find any space */
688 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
690 /* Giving back a free block. */
691 state->extravalid = 1;
692 state->extrablk.bp = curbp;
693 state->extrablk.index = fi;
694 state->extrablk.blkno = curfdb;
697 * Important: this magic number is not in the buffer - it's for
698 * buffer type information and therefore only the free/data type
699 * matters here, not whether CRCs are enabled or not.
701 state->extrablk.magic = XFS_DIR2_FREE_MAGIC;
703 state->extravalid = 0;
706 * Return the index, that will be the insertion point.
713 * Look up a leaf entry in a node-format leaf block.
714 * The extrablk in state a data block.
717 xfs_dir2_leafn_lookup_for_entry(
718 struct xfs_buf *bp, /* leaf buffer */
719 xfs_da_args_t *args, /* operation arguments */
720 int *indexp, /* out: leaf entry index */
721 xfs_da_state_t *state) /* state to fill in */
723 struct xfs_buf *curbp = NULL; /* current data/free buffer */
724 xfs_dir2_db_t curdb = -1; /* current data block number */
725 xfs_dir2_data_entry_t *dep; /* data block entry */
726 xfs_inode_t *dp; /* incore directory inode */
727 int error; /* error return value */
728 int index; /* leaf entry index */
729 xfs_dir2_leaf_t *leaf; /* leaf structure */
730 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
731 xfs_mount_t *mp; /* filesystem mount point */
732 xfs_dir2_db_t newdb; /* new data block number */
733 xfs_trans_t *tp; /* transaction pointer */
734 enum xfs_dacmp cmp; /* comparison result */
735 struct xfs_dir2_leaf_entry *ents;
736 struct xfs_dir3_icleaf_hdr leafhdr;
742 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
743 ents = dp->d_ops->leaf_ents_p(leaf);
745 xfs_dir3_leaf_check(dp, bp);
746 ASSERT(leafhdr.count > 0);
749 * Look up the hash value in the leaf entries.
751 index = xfs_dir2_leaf_search_hash(args, bp);
753 * Do we have a buffer coming in?
755 if (state->extravalid) {
756 curbp = state->extrablk.bp;
757 curdb = state->extrablk.blkno;
760 * Loop over leaf entries with the right hash value.
762 for (lep = &ents[index];
763 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
766 * Skip stale leaf entries.
768 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
771 * Pull the data block number from the entry.
773 newdb = xfs_dir2_dataptr_to_db(args->geo,
774 be32_to_cpu(lep->address));
776 * Not adding a new entry, so we really want to find
777 * the name given to us.
779 * If it's a different data block, go get it.
781 if (newdb != curdb) {
783 * If we had a block before that we aren't saving
784 * for a CI name, drop it
786 if (curbp && (args->cmpresult == XFS_CMP_DIFFERENT ||
787 curdb != state->extrablk.blkno))
788 xfs_trans_brelse(tp, curbp);
790 * If needing the block that is saved with a CI match,
791 * use it otherwise read in the new data block.
793 if (args->cmpresult != XFS_CMP_DIFFERENT &&
794 newdb == state->extrablk.blkno) {
795 ASSERT(state->extravalid);
796 curbp = state->extrablk.bp;
798 error = xfs_dir3_data_read(tp, dp,
799 xfs_dir2_db_to_da(args->geo,
805 xfs_dir3_data_check(dp, curbp);
809 * Point to the data entry.
811 dep = (xfs_dir2_data_entry_t *)((char *)curbp->b_addr +
812 xfs_dir2_dataptr_to_off(args->geo,
813 be32_to_cpu(lep->address)));
815 * Compare the entry and if it's an exact match, return
816 * EEXIST immediately. If it's the first case-insensitive
817 * match, store the block & inode number and continue looking.
819 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
820 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
821 /* If there is a CI match block, drop it */
822 if (args->cmpresult != XFS_CMP_DIFFERENT &&
823 curdb != state->extrablk.blkno)
824 xfs_trans_brelse(tp, state->extrablk.bp);
825 args->cmpresult = cmp;
826 args->inumber = be64_to_cpu(dep->inumber);
827 args->filetype = dp->d_ops->data_get_ftype(dep);
829 state->extravalid = 1;
830 state->extrablk.bp = curbp;
831 state->extrablk.blkno = curdb;
832 state->extrablk.index = (int)((char *)dep -
833 (char *)curbp->b_addr);
834 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
835 curbp->b_ops = &xfs_dir3_data_buf_ops;
836 xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
837 if (cmp == XFS_CMP_EXACT)
841 ASSERT(index == leafhdr.count || (args->op_flags & XFS_DA_OP_OKNOENT));
843 if (args->cmpresult == XFS_CMP_DIFFERENT) {
844 /* Giving back last used data block. */
845 state->extravalid = 1;
846 state->extrablk.bp = curbp;
847 state->extrablk.index = -1;
848 state->extrablk.blkno = curdb;
849 state->extrablk.magic = XFS_DIR2_DATA_MAGIC;
850 curbp->b_ops = &xfs_dir3_data_buf_ops;
851 xfs_trans_buf_set_type(tp, curbp, XFS_BLFT_DIR_DATA_BUF);
853 /* If the curbp is not the CI match block, drop it */
854 if (state->extrablk.bp != curbp)
855 xfs_trans_brelse(tp, curbp);
858 state->extravalid = 0;
865 * Look up a leaf entry in a node-format leaf block.
866 * If this is an addname then the extrablk in state is a freespace block,
867 * otherwise it's a data block.
870 xfs_dir2_leafn_lookup_int(
871 struct xfs_buf *bp, /* leaf buffer */
872 xfs_da_args_t *args, /* operation arguments */
873 int *indexp, /* out: leaf entry index */
874 xfs_da_state_t *state) /* state to fill in */
876 if (args->op_flags & XFS_DA_OP_ADDNAME)
877 return xfs_dir2_leafn_lookup_for_addname(bp, args, indexp,
879 return xfs_dir2_leafn_lookup_for_entry(bp, args, indexp, state);
883 * Move count leaf entries from source to destination leaf.
884 * Log entries and headers. Stale entries are preserved.
887 xfs_dir3_leafn_moveents(
888 xfs_da_args_t *args, /* operation arguments */
889 struct xfs_buf *bp_s, /* source */
890 struct xfs_dir3_icleaf_hdr *shdr,
891 struct xfs_dir2_leaf_entry *sents,
892 int start_s,/* source leaf index */
893 struct xfs_buf *bp_d, /* destination */
894 struct xfs_dir3_icleaf_hdr *dhdr,
895 struct xfs_dir2_leaf_entry *dents,
896 int start_d,/* destination leaf index */
897 int count) /* count of leaves to copy */
899 int stale; /* count stale leaves copied */
901 trace_xfs_dir2_leafn_moveents(args, start_s, start_d, count);
904 * Silently return if nothing to do.
910 * If the destination index is not the end of the current
911 * destination leaf entries, open up a hole in the destination
912 * to hold the new entries.
914 if (start_d < dhdr->count) {
915 memmove(&dents[start_d + count], &dents[start_d],
916 (dhdr->count - start_d) * sizeof(xfs_dir2_leaf_entry_t));
917 xfs_dir3_leaf_log_ents(args, bp_d, start_d + count,
918 count + dhdr->count - 1);
921 * If the source has stale leaves, count the ones in the copy range
922 * so we can update the header correctly.
925 int i; /* temp leaf index */
927 for (i = start_s, stale = 0; i < start_s + count; i++) {
928 if (sents[i].address ==
929 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
935 * Copy the leaf entries from source to destination.
937 memcpy(&dents[start_d], &sents[start_s],
938 count * sizeof(xfs_dir2_leaf_entry_t));
939 xfs_dir3_leaf_log_ents(args, bp_d, start_d, start_d + count - 1);
942 * If there are source entries after the ones we copied,
943 * delete the ones we copied by sliding the next ones down.
945 if (start_s + count < shdr->count) {
946 memmove(&sents[start_s], &sents[start_s + count],
947 count * sizeof(xfs_dir2_leaf_entry_t));
948 xfs_dir3_leaf_log_ents(args, bp_s, start_s, start_s + count - 1);
952 * Update the headers and log them.
954 shdr->count -= count;
955 shdr->stale -= stale;
956 dhdr->count += count;
957 dhdr->stale += stale;
961 * Determine the sort order of two leaf blocks.
962 * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
965 xfs_dir2_leafn_order(
966 struct xfs_inode *dp,
967 struct xfs_buf *leaf1_bp, /* leaf1 buffer */
968 struct xfs_buf *leaf2_bp) /* leaf2 buffer */
970 struct xfs_dir2_leaf *leaf1 = leaf1_bp->b_addr;
971 struct xfs_dir2_leaf *leaf2 = leaf2_bp->b_addr;
972 struct xfs_dir2_leaf_entry *ents1;
973 struct xfs_dir2_leaf_entry *ents2;
974 struct xfs_dir3_icleaf_hdr hdr1;
975 struct xfs_dir3_icleaf_hdr hdr2;
977 dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
978 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
979 ents1 = dp->d_ops->leaf_ents_p(leaf1);
980 ents2 = dp->d_ops->leaf_ents_p(leaf2);
982 if (hdr1.count > 0 && hdr2.count > 0 &&
983 (be32_to_cpu(ents2[0].hashval) < be32_to_cpu(ents1[0].hashval) ||
984 be32_to_cpu(ents2[hdr2.count - 1].hashval) <
985 be32_to_cpu(ents1[hdr1.count - 1].hashval)))
991 * Rebalance leaf entries between two leaf blocks.
992 * This is actually only called when the second block is new,
993 * though the code deals with the general case.
994 * A new entry will be inserted in one of the blocks, and that
995 * entry is taken into account when balancing.
998 xfs_dir2_leafn_rebalance(
999 xfs_da_state_t *state, /* btree cursor */
1000 xfs_da_state_blk_t *blk1, /* first btree block */
1001 xfs_da_state_blk_t *blk2) /* second btree block */
1003 xfs_da_args_t *args; /* operation arguments */
1004 int count; /* count (& direction) leaves */
1005 int isleft; /* new goes in left leaf */
1006 xfs_dir2_leaf_t *leaf1; /* first leaf structure */
1007 xfs_dir2_leaf_t *leaf2; /* second leaf structure */
1008 int mid; /* midpoint leaf index */
1009 #if defined(DEBUG) || defined(XFS_WARN)
1010 int oldstale; /* old count of stale leaves */
1012 int oldsum; /* old total leaf count */
1013 int swap_blocks; /* swapped leaf blocks */
1014 struct xfs_dir2_leaf_entry *ents1;
1015 struct xfs_dir2_leaf_entry *ents2;
1016 struct xfs_dir3_icleaf_hdr hdr1;
1017 struct xfs_dir3_icleaf_hdr hdr2;
1018 struct xfs_inode *dp = state->args->dp;
1022 * If the block order is wrong, swap the arguments.
1024 swap_blocks = xfs_dir2_leafn_order(dp, blk1->bp, blk2->bp);
1028 leaf1 = blk1->bp->b_addr;
1029 leaf2 = blk2->bp->b_addr;
1030 dp->d_ops->leaf_hdr_from_disk(&hdr1, leaf1);
1031 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf2);
1032 ents1 = dp->d_ops->leaf_ents_p(leaf1);
1033 ents2 = dp->d_ops->leaf_ents_p(leaf2);
1035 oldsum = hdr1.count + hdr2.count;
1036 #if defined(DEBUG) || defined(XFS_WARN)
1037 oldstale = hdr1.stale + hdr2.stale;
1042 * If the old leaf count was odd then the new one will be even,
1043 * so we need to divide the new count evenly.
1046 xfs_dahash_t midhash; /* middle entry hash value */
1048 if (mid >= hdr1.count)
1049 midhash = be32_to_cpu(ents2[mid - hdr1.count].hashval);
1051 midhash = be32_to_cpu(ents1[mid].hashval);
1052 isleft = args->hashval <= midhash;
1055 * If the old count is even then the new count is odd, so there's
1056 * no preferred side for the new entry.
1057 * Pick the left one.
1062 * Calculate moved entry count. Positive means left-to-right,
1063 * negative means right-to-left. Then move the entries.
1065 count = hdr1.count - mid + (isleft == 0);
1067 xfs_dir3_leafn_moveents(args, blk1->bp, &hdr1, ents1,
1068 hdr1.count - count, blk2->bp,
1069 &hdr2, ents2, 0, count);
1071 xfs_dir3_leafn_moveents(args, blk2->bp, &hdr2, ents2, 0,
1072 blk1->bp, &hdr1, ents1,
1075 ASSERT(hdr1.count + hdr2.count == oldsum);
1076 ASSERT(hdr1.stale + hdr2.stale == oldstale);
1078 /* log the changes made when moving the entries */
1079 dp->d_ops->leaf_hdr_to_disk(leaf1, &hdr1);
1080 dp->d_ops->leaf_hdr_to_disk(leaf2, &hdr2);
1081 xfs_dir3_leaf_log_header(args, blk1->bp);
1082 xfs_dir3_leaf_log_header(args, blk2->bp);
1084 xfs_dir3_leaf_check(dp, blk1->bp);
1085 xfs_dir3_leaf_check(dp, blk2->bp);
1088 * Mark whether we're inserting into the old or new leaf.
1090 if (hdr1.count < hdr2.count)
1091 state->inleaf = swap_blocks;
1092 else if (hdr1.count > hdr2.count)
1093 state->inleaf = !swap_blocks;
1095 state->inleaf = swap_blocks ^ (blk1->index <= hdr1.count);
1097 * Adjust the expected index for insertion.
1100 blk2->index = blk1->index - hdr1.count;
1103 * Finally sanity check just to make sure we are not returning a
1106 if (blk2->index < 0) {
1109 xfs_alert(dp->i_mount,
1110 "%s: picked the wrong leaf? reverting original leaf: blk1->index %d",
1111 __func__, blk1->index);
1116 xfs_dir3_data_block_free(
1117 xfs_da_args_t *args,
1118 struct xfs_dir2_data_hdr *hdr,
1119 struct xfs_dir2_free *free,
1122 struct xfs_buf *fbp,
1127 struct xfs_dir3_icfree_hdr freehdr;
1128 struct xfs_inode *dp = args->dp;
1130 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1131 bests = dp->d_ops->free_bests_p(free);
1134 * Data block is not empty, just set the free entry to the new
1137 bests[findex] = cpu_to_be16(longest);
1138 xfs_dir2_free_log_bests(args, fbp, findex, findex);
1142 /* One less used entry in the free table. */
1146 * If this was the last entry in the table, we can trim the table size
1147 * back. There might be other entries at the end referring to
1148 * non-existent data blocks, get those too.
1150 if (findex == freehdr.nvalid - 1) {
1151 int i; /* free entry index */
1153 for (i = findex - 1; i >= 0; i--) {
1154 if (bests[i] != cpu_to_be16(NULLDATAOFF))
1157 freehdr.nvalid = i + 1;
1160 /* Not the last entry, just punch it out. */
1161 bests[findex] = cpu_to_be16(NULLDATAOFF);
1165 dp->d_ops->free_hdr_to_disk(free, &freehdr);
1166 xfs_dir2_free_log_header(args, fbp);
1169 * If there are no useful entries left in the block, get rid of the
1172 if (!freehdr.nused) {
1175 error = xfs_dir2_shrink_inode(args, fdb, fbp);
1179 } else if (error != -ENOSPC || args->total != 0)
1182 * It's possible to get ENOSPC if there is no
1183 * space reservation. In this case some one
1184 * else will eventually get rid of this block.
1188 /* Log the free entry that changed, unless we got rid of it. */
1190 xfs_dir2_free_log_bests(args, fbp, findex, findex);
1195 * Remove an entry from a node directory.
1196 * This removes the leaf entry and the data entry,
1197 * and updates the free block if necessary.
1199 static int /* error */
1200 xfs_dir2_leafn_remove(
1201 xfs_da_args_t *args, /* operation arguments */
1202 struct xfs_buf *bp, /* leaf buffer */
1203 int index, /* leaf entry index */
1204 xfs_da_state_blk_t *dblk, /* data block */
1205 int *rval) /* resulting block needs join */
1207 xfs_dir2_data_hdr_t *hdr; /* data block header */
1208 xfs_dir2_db_t db; /* data block number */
1209 struct xfs_buf *dbp; /* data block buffer */
1210 xfs_dir2_data_entry_t *dep; /* data block entry */
1211 xfs_inode_t *dp; /* incore directory inode */
1212 xfs_dir2_leaf_t *leaf; /* leaf structure */
1213 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1214 int longest; /* longest data free entry */
1215 int off; /* data block entry offset */
1216 int needlog; /* need to log data header */
1217 int needscan; /* need to rescan data frees */
1218 xfs_trans_t *tp; /* transaction pointer */
1219 struct xfs_dir2_data_free *bf; /* bestfree table */
1220 struct xfs_dir3_icleaf_hdr leafhdr;
1221 struct xfs_dir2_leaf_entry *ents;
1223 trace_xfs_dir2_leafn_remove(args, index);
1228 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1229 ents = dp->d_ops->leaf_ents_p(leaf);
1232 * Point to the entry we're removing.
1237 * Extract the data block and offset from the entry.
1239 db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1240 ASSERT(dblk->blkno == db);
1241 off = xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address));
1242 ASSERT(dblk->index == off);
1245 * Kill the leaf entry by marking it stale.
1246 * Log the leaf block changes.
1249 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1250 xfs_dir3_leaf_log_header(args, bp);
1252 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1253 xfs_dir3_leaf_log_ents(args, bp, index, index);
1256 * Make the data entry free. Keep track of the longest freespace
1257 * in the data block in case it changes.
1261 dep = (xfs_dir2_data_entry_t *)((char *)hdr + off);
1262 bf = dp->d_ops->data_bestfree_p(hdr);
1263 longest = be16_to_cpu(bf[0].length);
1264 needlog = needscan = 0;
1265 xfs_dir2_data_make_free(args, dbp, off,
1266 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1268 * Rescan the data block freespaces for bestfree.
1269 * Log the data block header if needed.
1272 xfs_dir2_data_freescan(dp, hdr, &needlog);
1274 xfs_dir2_data_log_header(args, dbp);
1275 xfs_dir3_data_check(dp, dbp);
1277 * If the longest data block freespace changes, need to update
1278 * the corresponding freeblock entry.
1280 if (longest < be16_to_cpu(bf[0].length)) {
1281 int error; /* error return value */
1282 struct xfs_buf *fbp; /* freeblock buffer */
1283 xfs_dir2_db_t fdb; /* freeblock block number */
1284 int findex; /* index in freeblock entries */
1285 xfs_dir2_free_t *free; /* freeblock structure */
1288 * Convert the data block number to a free block,
1289 * read in the free block.
1291 fdb = dp->d_ops->db_to_fdb(args->geo, db);
1292 error = xfs_dir2_free_read(tp, dp,
1293 xfs_dir2_db_to_da(args->geo, fdb),
1300 struct xfs_dir3_icfree_hdr freehdr;
1301 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1302 ASSERT(freehdr.firstdb == dp->d_ops->free_max_bests(args->geo) *
1303 (fdb - xfs_dir2_byte_to_db(args->geo,
1304 XFS_DIR2_FREE_OFFSET)));
1308 * Calculate which entry we need to fix.
1310 findex = dp->d_ops->db_to_fdindex(args->geo, db);
1311 longest = be16_to_cpu(bf[0].length);
1313 * If the data block is now empty we can get rid of it
1316 if (longest == args->geo->blksize -
1317 dp->d_ops->data_entry_offset) {
1319 * Try to punch out the data block.
1321 error = xfs_dir2_shrink_inode(args, db, dbp);
1327 * We can get ENOSPC if there's no space reservation.
1328 * In this case just drop the buffer and some one else
1329 * will eventually get rid of the empty block.
1331 else if (!(error == -ENOSPC && args->total == 0))
1335 * If we got rid of the data block, we can eliminate that entry
1336 * in the free block.
1338 error = xfs_dir3_data_block_free(args, hdr, free,
1339 fdb, findex, fbp, longest);
1344 xfs_dir3_leaf_check(dp, bp);
1346 * Return indication of whether this leaf block is empty enough
1347 * to justify trying to join it with a neighbor.
1349 *rval = (dp->d_ops->leaf_hdr_size +
1350 (uint)sizeof(ents[0]) * (leafhdr.count - leafhdr.stale)) <
1351 args->geo->magicpct;
1356 * Split the leaf entries in the old block into old and new blocks.
1359 xfs_dir2_leafn_split(
1360 xfs_da_state_t *state, /* btree cursor */
1361 xfs_da_state_blk_t *oldblk, /* original block */
1362 xfs_da_state_blk_t *newblk) /* newly created block */
1364 xfs_da_args_t *args; /* operation arguments */
1365 xfs_dablk_t blkno; /* new leaf block number */
1366 int error; /* error return value */
1367 struct xfs_inode *dp;
1370 * Allocate space for a new leaf node.
1374 ASSERT(oldblk->magic == XFS_DIR2_LEAFN_MAGIC);
1375 error = xfs_da_grow_inode(args, &blkno);
1380 * Initialize the new leaf block.
1382 error = xfs_dir3_leaf_get_buf(args, xfs_dir2_da_to_db(args->geo, blkno),
1383 &newblk->bp, XFS_DIR2_LEAFN_MAGIC);
1387 newblk->blkno = blkno;
1388 newblk->magic = XFS_DIR2_LEAFN_MAGIC;
1390 * Rebalance the entries across the two leaves, link the new
1391 * block into the leaves.
1393 xfs_dir2_leafn_rebalance(state, oldblk, newblk);
1394 error = xfs_da3_blk_link(state, oldblk, newblk);
1399 * Insert the new entry in the correct block.
1402 error = xfs_dir2_leafn_add(oldblk->bp, args, oldblk->index);
1404 error = xfs_dir2_leafn_add(newblk->bp, args, newblk->index);
1406 * Update last hashval in each block since we added the name.
1408 oldblk->hashval = xfs_dir2_leaf_lasthash(dp, oldblk->bp, NULL);
1409 newblk->hashval = xfs_dir2_leaf_lasthash(dp, newblk->bp, NULL);
1410 xfs_dir3_leaf_check(dp, oldblk->bp);
1411 xfs_dir3_leaf_check(dp, newblk->bp);
1416 * Check a leaf block and its neighbors to see if the block should be
1417 * collapsed into one or the other neighbor. Always keep the block
1418 * with the smaller block number.
1419 * If the current block is over 50% full, don't try to join it, return 0.
1420 * If the block is empty, fill in the state structure and return 2.
1421 * If it can be collapsed, fill in the state structure and return 1.
1422 * If nothing can be done, return 0.
1425 xfs_dir2_leafn_toosmall(
1426 xfs_da_state_t *state, /* btree cursor */
1427 int *action) /* resulting action to take */
1429 xfs_da_state_blk_t *blk; /* leaf block */
1430 xfs_dablk_t blkno; /* leaf block number */
1431 struct xfs_buf *bp; /* leaf buffer */
1432 int bytes; /* bytes in use */
1433 int count; /* leaf live entry count */
1434 int error; /* error return value */
1435 int forward; /* sibling block direction */
1436 int i; /* sibling counter */
1437 xfs_dir2_leaf_t *leaf; /* leaf structure */
1438 int rval; /* result from path_shift */
1439 struct xfs_dir3_icleaf_hdr leafhdr;
1440 struct xfs_dir2_leaf_entry *ents;
1441 struct xfs_inode *dp = state->args->dp;
1444 * Check for the degenerate case of the block being over 50% full.
1445 * If so, it's not worth even looking to see if we might be able
1446 * to coalesce with a sibling.
1448 blk = &state->path.blk[state->path.active - 1];
1449 leaf = blk->bp->b_addr;
1450 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1451 ents = dp->d_ops->leaf_ents_p(leaf);
1452 xfs_dir3_leaf_check(dp, blk->bp);
1454 count = leafhdr.count - leafhdr.stale;
1455 bytes = dp->d_ops->leaf_hdr_size + count * sizeof(ents[0]);
1456 if (bytes > (state->args->geo->blksize >> 1)) {
1458 * Blk over 50%, don't try to join.
1464 * Check for the degenerate case of the block being empty.
1465 * If the block is empty, we'll simply delete it, no need to
1466 * coalesce it with a sibling block. We choose (arbitrarily)
1467 * to merge with the forward block unless it is NULL.
1471 * Make altpath point to the block we want to keep and
1472 * path point to the block we want to drop (this one).
1474 forward = (leafhdr.forw != 0);
1475 memcpy(&state->altpath, &state->path, sizeof(state->path));
1476 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1480 *action = rval ? 2 : 0;
1484 * Examine each sibling block to see if we can coalesce with
1485 * at least 25% free space to spare. We need to figure out
1486 * whether to merge with the forward or the backward block.
1487 * We prefer coalescing with the lower numbered sibling so as
1488 * to shrink a directory over time.
1490 forward = leafhdr.forw < leafhdr.back;
1491 for (i = 0, bp = NULL; i < 2; forward = !forward, i++) {
1492 struct xfs_dir3_icleaf_hdr hdr2;
1494 blkno = forward ? leafhdr.forw : leafhdr.back;
1498 * Read the sibling leaf block.
1500 error = xfs_dir3_leafn_read(state->args->trans, dp,
1506 * Count bytes in the two blocks combined.
1508 count = leafhdr.count - leafhdr.stale;
1509 bytes = state->args->geo->blksize -
1510 (state->args->geo->blksize >> 2);
1513 dp->d_ops->leaf_hdr_from_disk(&hdr2, leaf);
1514 ents = dp->d_ops->leaf_ents_p(leaf);
1515 count += hdr2.count - hdr2.stale;
1516 bytes -= count * sizeof(ents[0]);
1519 * Fits with at least 25% to spare.
1523 xfs_trans_brelse(state->args->trans, bp);
1526 * Didn't like either block, give up.
1534 * Make altpath point to the block we want to keep (the lower
1535 * numbered block) and path point to the block we want to drop.
1537 memcpy(&state->altpath, &state->path, sizeof(state->path));
1538 if (blkno < blk->blkno)
1539 error = xfs_da3_path_shift(state, &state->altpath, forward, 0,
1542 error = xfs_da3_path_shift(state, &state->path, forward, 0,
1547 *action = rval ? 0 : 1;
1552 * Move all the leaf entries from drop_blk to save_blk.
1553 * This is done as part of a join operation.
1556 xfs_dir2_leafn_unbalance(
1557 xfs_da_state_t *state, /* cursor */
1558 xfs_da_state_blk_t *drop_blk, /* dead block */
1559 xfs_da_state_blk_t *save_blk) /* surviving block */
1561 xfs_da_args_t *args; /* operation arguments */
1562 xfs_dir2_leaf_t *drop_leaf; /* dead leaf structure */
1563 xfs_dir2_leaf_t *save_leaf; /* surviving leaf structure */
1564 struct xfs_dir3_icleaf_hdr savehdr;
1565 struct xfs_dir3_icleaf_hdr drophdr;
1566 struct xfs_dir2_leaf_entry *sents;
1567 struct xfs_dir2_leaf_entry *dents;
1568 struct xfs_inode *dp = state->args->dp;
1571 ASSERT(drop_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1572 ASSERT(save_blk->magic == XFS_DIR2_LEAFN_MAGIC);
1573 drop_leaf = drop_blk->bp->b_addr;
1574 save_leaf = save_blk->bp->b_addr;
1576 dp->d_ops->leaf_hdr_from_disk(&savehdr, save_leaf);
1577 dp->d_ops->leaf_hdr_from_disk(&drophdr, drop_leaf);
1578 sents = dp->d_ops->leaf_ents_p(save_leaf);
1579 dents = dp->d_ops->leaf_ents_p(drop_leaf);
1582 * If there are any stale leaf entries, take this opportunity
1586 xfs_dir3_leaf_compact(args, &drophdr, drop_blk->bp);
1588 xfs_dir3_leaf_compact(args, &savehdr, save_blk->bp);
1591 * Move the entries from drop to the appropriate end of save.
1593 drop_blk->hashval = be32_to_cpu(dents[drophdr.count - 1].hashval);
1594 if (xfs_dir2_leafn_order(dp, save_blk->bp, drop_blk->bp))
1595 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1596 save_blk->bp, &savehdr, sents, 0,
1599 xfs_dir3_leafn_moveents(args, drop_blk->bp, &drophdr, dents, 0,
1600 save_blk->bp, &savehdr, sents,
1601 savehdr.count, drophdr.count);
1602 save_blk->hashval = be32_to_cpu(sents[savehdr.count - 1].hashval);
1604 /* log the changes made when moving the entries */
1605 dp->d_ops->leaf_hdr_to_disk(save_leaf, &savehdr);
1606 dp->d_ops->leaf_hdr_to_disk(drop_leaf, &drophdr);
1607 xfs_dir3_leaf_log_header(args, save_blk->bp);
1608 xfs_dir3_leaf_log_header(args, drop_blk->bp);
1610 xfs_dir3_leaf_check(dp, save_blk->bp);
1611 xfs_dir3_leaf_check(dp, drop_blk->bp);
1615 * Top-level node form directory addname routine.
1618 xfs_dir2_node_addname(
1619 xfs_da_args_t *args) /* operation arguments */
1621 xfs_da_state_blk_t *blk; /* leaf block for insert */
1622 int error; /* error return value */
1623 int rval; /* sub-return value */
1624 xfs_da_state_t *state; /* btree cursor */
1626 trace_xfs_dir2_node_addname(args);
1629 * Allocate and initialize the state (btree cursor).
1631 state = xfs_da_state_alloc();
1633 state->mp = args->dp->i_mount;
1635 * Look up the name. We're not supposed to find it, but
1636 * this gives us the insertion point.
1638 error = xfs_da3_node_lookup_int(state, &rval);
1641 if (rval != -ENOENT) {
1645 * Add the data entry to a data block.
1646 * Extravalid is set to a freeblock found by lookup.
1648 rval = xfs_dir2_node_addname_int(args,
1649 state->extravalid ? &state->extrablk : NULL);
1653 blk = &state->path.blk[state->path.active - 1];
1654 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
1656 * Add the new leaf entry.
1658 rval = xfs_dir2_leafn_add(blk->bp, args, blk->index);
1661 * It worked, fix the hash values up the btree.
1663 if (!(args->op_flags & XFS_DA_OP_JUSTCHECK))
1664 xfs_da3_fixhashpath(state, &state->path);
1667 * It didn't work, we need to split the leaf block.
1669 if (args->total == 0) {
1670 ASSERT(rval == -ENOSPC);
1674 * Split the leaf block and insert the new entry.
1676 rval = xfs_da3_split(state);
1679 xfs_da_state_free(state);
1684 * Add the data entry for a node-format directory name addition.
1685 * The leaf entry is added in xfs_dir2_leafn_add.
1686 * We may enter with a freespace block that the lookup found.
1688 static int /* error */
1689 xfs_dir2_node_addname_int(
1690 xfs_da_args_t *args, /* operation arguments */
1691 xfs_da_state_blk_t *fblk) /* optional freespace block */
1693 xfs_dir2_data_hdr_t *hdr; /* data block header */
1694 xfs_dir2_db_t dbno; /* data block number */
1695 struct xfs_buf *dbp; /* data block buffer */
1696 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1697 xfs_inode_t *dp; /* incore directory inode */
1698 xfs_dir2_data_unused_t *dup; /* data unused entry pointer */
1699 int error; /* error return value */
1700 xfs_dir2_db_t fbno; /* freespace block number */
1701 struct xfs_buf *fbp; /* freespace buffer */
1702 int findex; /* freespace entry index */
1703 xfs_dir2_free_t *free=NULL; /* freespace block structure */
1704 xfs_dir2_db_t ifbno; /* initial freespace block no */
1705 xfs_dir2_db_t lastfbno=0; /* highest freespace block no */
1706 int length; /* length of the new entry */
1707 int logfree; /* need to log free entry */
1708 xfs_mount_t *mp; /* filesystem mount point */
1709 int needlog; /* need to log data header */
1710 int needscan; /* need to rescan data frees */
1711 __be16 *tagp; /* data entry tag pointer */
1712 xfs_trans_t *tp; /* transaction pointer */
1714 struct xfs_dir3_icfree_hdr freehdr;
1715 struct xfs_dir2_data_free *bf;
1716 xfs_dir2_data_aoff_t aoff;
1721 length = dp->d_ops->data_entsize(args->namelen);
1723 * If we came in with a freespace block that means that lookup
1724 * found an entry with our hash value. This is the freespace
1725 * block for that data entry.
1730 * Remember initial freespace block number.
1732 ifbno = fblk->blkno;
1734 findex = fblk->index;
1735 bests = dp->d_ops->free_bests_p(free);
1736 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1739 * This means the free entry showed that the data block had
1740 * space for our entry, so we remembered it.
1741 * Use that data block.
1744 ASSERT(findex < freehdr.nvalid);
1745 ASSERT(be16_to_cpu(bests[findex]) != NULLDATAOFF);
1746 ASSERT(be16_to_cpu(bests[findex]) >= length);
1747 dbno = freehdr.firstdb + findex;
1750 * The data block looked at didn't have enough room.
1751 * We'll start at the beginning of the freespace entries.
1758 * Didn't come in with a freespace block, so no data block.
1766 * If we don't have a data block yet, we're going to scan the
1767 * freespace blocks looking for one. Figure out what the
1768 * highest freespace block number is.
1771 xfs_fileoff_t fo; /* freespace block number */
1773 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK)))
1775 lastfbno = xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo);
1779 * While we haven't identified a data block, search the freeblock
1780 * data for a good data block. If we find a null freeblock entry,
1781 * indicating a hole in the data blocks, remember that.
1783 while (dbno == -1) {
1785 * If we don't have a freeblock in hand, get the next one.
1789 * Happens the first time through unless lookup gave
1790 * us a freespace block to start with.
1793 fbno = xfs_dir2_byte_to_db(args->geo,
1794 XFS_DIR2_FREE_OFFSET);
1796 * If it's ifbno we already looked at it.
1801 * If it's off the end we're done.
1803 if (fbno >= lastfbno)
1806 * Read the block. There can be holes in the
1807 * freespace blocks, so this might not succeed.
1808 * This should be really rare, so there's no reason
1811 error = xfs_dir2_free_try_read(tp, dp,
1812 xfs_dir2_db_to_da(args->geo, fbno),
1822 * Look at the current free entry. Is it good enough?
1824 * The bests initialisation should be where the bufer is read in
1825 * the above branch. But gcc is too stupid to realise that bests
1826 * and the freehdr are actually initialised if they are placed
1827 * there, so we have to do it here to avoid warnings. Blech.
1829 bests = dp->d_ops->free_bests_p(free);
1830 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1831 if (be16_to_cpu(bests[findex]) != NULLDATAOFF &&
1832 be16_to_cpu(bests[findex]) >= length)
1833 dbno = freehdr.firstdb + findex;
1836 * Are we done with the freeblock?
1838 if (++findex == freehdr.nvalid) {
1842 xfs_trans_brelse(tp, fbp);
1844 if (fblk && fblk->bp)
1850 * If we don't have a data block, we need to allocate one and make
1851 * the freespace entries refer to it.
1853 if (unlikely(dbno == -1)) {
1855 * Not allowed to allocate, return failure.
1857 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
1861 * Allocate and initialize the new data block.
1863 if (unlikely((error = xfs_dir2_grow_inode(args,
1864 XFS_DIR2_DATA_SPACE,
1866 (error = xfs_dir3_data_init(args, dbno, &dbp))))
1870 * If (somehow) we have a freespace block, get rid of it.
1873 xfs_trans_brelse(tp, fbp);
1874 if (fblk && fblk->bp)
1878 * Get the freespace block corresponding to the data block
1879 * that was just allocated.
1881 fbno = dp->d_ops->db_to_fdb(args->geo, dbno);
1882 error = xfs_dir2_free_try_read(tp, dp,
1883 xfs_dir2_db_to_da(args->geo, fbno),
1889 * If there wasn't a freespace block, the read will
1890 * return a NULL fbp. Allocate and initialize a new one.
1893 error = xfs_dir2_grow_inode(args, XFS_DIR2_FREE_SPACE,
1898 if (dp->d_ops->db_to_fdb(args->geo, dbno) != fbno) {
1900 "%s: dir ino %llu needed freesp block %lld for data block %lld, got %lld ifbno %llu lastfbno %d",
1901 __func__, (unsigned long long)dp->i_ino,
1902 (long long)dp->d_ops->db_to_fdb(
1904 (long long)dbno, (long long)fbno,
1905 (unsigned long long)ifbno, lastfbno);
1908 " fblk "PTR_FMT" blkno %llu index %d magic 0x%x",
1910 (unsigned long long)fblk->blkno,
1914 xfs_alert(mp, " ... fblk is NULL");
1916 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1917 XFS_ERRLEVEL_LOW, mp);
1918 return -EFSCORRUPTED;
1922 * Get a buffer for the new block.
1924 error = xfs_dir3_free_get_buf(args, fbno, &fbp);
1928 bests = dp->d_ops->free_bests_p(free);
1929 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1932 * Remember the first slot as our empty slot.
1935 (fbno - xfs_dir2_byte_to_db(args->geo,
1936 XFS_DIR2_FREE_OFFSET)) *
1937 dp->d_ops->free_max_bests(args->geo);
1940 bests = dp->d_ops->free_bests_p(free);
1941 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1945 * Set the freespace block index from the data block number.
1947 findex = dp->d_ops->db_to_fdindex(args->geo, dbno);
1949 * If it's after the end of the current entries in the
1950 * freespace block, extend that table.
1952 if (findex >= freehdr.nvalid) {
1953 ASSERT(findex < dp->d_ops->free_max_bests(args->geo));
1954 freehdr.nvalid = findex + 1;
1956 * Tag new entry so nused will go up.
1958 bests[findex] = cpu_to_be16(NULLDATAOFF);
1961 * If this entry was for an empty data block
1962 * (this should always be true) then update the header.
1964 if (bests[findex] == cpu_to_be16(NULLDATAOFF)) {
1966 dp->d_ops->free_hdr_to_disk(fbp->b_addr, &freehdr);
1967 xfs_dir2_free_log_header(args, fbp);
1970 * Update the real value in the table.
1971 * We haven't allocated the data entry yet so this will
1975 bf = dp->d_ops->data_bestfree_p(hdr);
1976 bests[findex] = bf[0].length;
1980 * We had a data block so we don't have to make a new one.
1984 * If just checking, we succeeded.
1986 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
1990 * Read the data block in.
1992 error = xfs_dir3_data_read(tp, dp,
1993 xfs_dir2_db_to_da(args->geo, dbno),
1998 bf = dp->d_ops->data_bestfree_p(hdr);
2001 ASSERT(be16_to_cpu(bf[0].length) >= length);
2003 * Point to the existing unused space.
2005 dup = (xfs_dir2_data_unused_t *)
2006 ((char *)hdr + be16_to_cpu(bf[0].offset));
2007 needscan = needlog = 0;
2009 * Mark the first part of the unused space, inuse for us.
2011 aoff = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
2012 error = xfs_dir2_data_use_free(args, dbp, dup, aoff, length,
2013 &needlog, &needscan);
2015 xfs_trans_brelse(tp, dbp);
2019 * Fill in the new entry and log it.
2021 dep = (xfs_dir2_data_entry_t *)dup;
2022 dep->inumber = cpu_to_be64(args->inumber);
2023 dep->namelen = args->namelen;
2024 memcpy(dep->name, args->name, dep->namelen);
2025 dp->d_ops->data_put_ftype(dep, args->filetype);
2026 tagp = dp->d_ops->data_entry_tag_p(dep);
2027 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
2028 xfs_dir2_data_log_entry(args, dbp, dep);
2030 * Rescan the block for bestfree if needed.
2033 xfs_dir2_data_freescan(dp, hdr, &needlog);
2035 * Log the data block header if needed.
2038 xfs_dir2_data_log_header(args, dbp);
2040 * If the freespace entry is now wrong, update it.
2042 bests = dp->d_ops->free_bests_p(free); /* gcc is so stupid */
2043 if (be16_to_cpu(bests[findex]) != be16_to_cpu(bf[0].length)) {
2044 bests[findex] = bf[0].length;
2048 * Log the freespace entry if needed.
2051 xfs_dir2_free_log_bests(args, fbp, findex, findex);
2053 * Return the data block and offset in args, then drop the data block.
2055 args->blkno = (xfs_dablk_t)dbno;
2056 args->index = be16_to_cpu(*tagp);
2061 * Lookup an entry in a node-format directory.
2062 * All the real work happens in xfs_da3_node_lookup_int.
2063 * The only real output is the inode number of the entry.
2066 xfs_dir2_node_lookup(
2067 xfs_da_args_t *args) /* operation arguments */
2069 int error; /* error return value */
2070 int i; /* btree level */
2071 int rval; /* operation return value */
2072 xfs_da_state_t *state; /* btree cursor */
2074 trace_xfs_dir2_node_lookup(args);
2077 * Allocate and initialize the btree cursor.
2079 state = xfs_da_state_alloc();
2081 state->mp = args->dp->i_mount;
2083 * Fill in the path to the entry in the cursor.
2085 error = xfs_da3_node_lookup_int(state, &rval);
2088 else if (rval == -ENOENT && args->cmpresult == XFS_CMP_CASE) {
2089 /* If a CI match, dup the actual name and return -EEXIST */
2090 xfs_dir2_data_entry_t *dep;
2092 dep = (xfs_dir2_data_entry_t *)
2093 ((char *)state->extrablk.bp->b_addr +
2094 state->extrablk.index);
2095 rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
2098 * Release the btree blocks and leaf block.
2100 for (i = 0; i < state->path.active; i++) {
2101 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2102 state->path.blk[i].bp = NULL;
2105 * Release the data block if we have it.
2107 if (state->extravalid && state->extrablk.bp) {
2108 xfs_trans_brelse(args->trans, state->extrablk.bp);
2109 state->extrablk.bp = NULL;
2111 xfs_da_state_free(state);
2116 * Remove an entry from a node-format directory.
2119 xfs_dir2_node_removename(
2120 struct xfs_da_args *args) /* operation arguments */
2122 struct xfs_da_state_blk *blk; /* leaf block */
2123 int error; /* error return value */
2124 int rval; /* operation return value */
2125 struct xfs_da_state *state; /* btree cursor */
2127 trace_xfs_dir2_node_removename(args);
2130 * Allocate and initialize the btree cursor.
2132 state = xfs_da_state_alloc();
2134 state->mp = args->dp->i_mount;
2136 /* Look up the entry we're deleting, set up the cursor. */
2137 error = xfs_da3_node_lookup_int(state, &rval);
2141 /* Didn't find it, upper layer screwed up. */
2142 if (rval != -EEXIST) {
2147 blk = &state->path.blk[state->path.active - 1];
2148 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2149 ASSERT(state->extravalid);
2151 * Remove the leaf and data entries.
2152 * Extrablk refers to the data block.
2154 error = xfs_dir2_leafn_remove(args, blk->bp, blk->index,
2155 &state->extrablk, &rval);
2159 * Fix the hash values up the btree.
2161 xfs_da3_fixhashpath(state, &state->path);
2163 * If we need to join leaf blocks, do it.
2165 if (rval && state->path.active > 1)
2166 error = xfs_da3_join(state);
2168 * If no errors so far, try conversion to leaf format.
2171 error = xfs_dir2_node_to_leaf(state);
2173 xfs_da_state_free(state);
2178 * Replace an entry's inode number in a node-format directory.
2181 xfs_dir2_node_replace(
2182 xfs_da_args_t *args) /* operation arguments */
2184 xfs_da_state_blk_t *blk; /* leaf block */
2185 xfs_dir2_data_hdr_t *hdr; /* data block header */
2186 xfs_dir2_data_entry_t *dep; /* data entry changed */
2187 int error; /* error return value */
2188 int i; /* btree level */
2189 xfs_ino_t inum; /* new inode number */
2190 int ftype; /* new file type */
2191 xfs_dir2_leaf_t *leaf; /* leaf structure */
2192 xfs_dir2_leaf_entry_t *lep; /* leaf entry being changed */
2193 int rval; /* internal return value */
2194 xfs_da_state_t *state; /* btree cursor */
2196 trace_xfs_dir2_node_replace(args);
2199 * Allocate and initialize the btree cursor.
2201 state = xfs_da_state_alloc();
2203 state->mp = args->dp->i_mount;
2206 * We have to save new inode number and ftype since
2207 * xfs_da3_node_lookup_int() is going to overwrite them
2209 inum = args->inumber;
2210 ftype = args->filetype;
2213 * Lookup the entry to change in the btree.
2215 error = xfs_da3_node_lookup_int(state, &rval);
2220 * It should be found, since the vnodeops layer has looked it up
2221 * and locked it. But paranoia is good.
2223 if (rval == -EEXIST) {
2224 struct xfs_dir2_leaf_entry *ents;
2226 * Find the leaf entry.
2228 blk = &state->path.blk[state->path.active - 1];
2229 ASSERT(blk->magic == XFS_DIR2_LEAFN_MAGIC);
2230 leaf = blk->bp->b_addr;
2231 ents = args->dp->d_ops->leaf_ents_p(leaf);
2232 lep = &ents[blk->index];
2233 ASSERT(state->extravalid);
2235 * Point to the data entry.
2237 hdr = state->extrablk.bp->b_addr;
2238 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
2239 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
2240 dep = (xfs_dir2_data_entry_t *)
2242 xfs_dir2_dataptr_to_off(args->geo,
2243 be32_to_cpu(lep->address)));
2244 ASSERT(inum != be64_to_cpu(dep->inumber));
2246 * Fill in the new inode number and log the entry.
2248 dep->inumber = cpu_to_be64(inum);
2249 args->dp->d_ops->data_put_ftype(dep, ftype);
2250 xfs_dir2_data_log_entry(args, state->extrablk.bp, dep);
2254 * Didn't find it, and we're holding a data block. Drop it.
2256 else if (state->extravalid) {
2257 xfs_trans_brelse(args->trans, state->extrablk.bp);
2258 state->extrablk.bp = NULL;
2261 * Release all the buffers in the cursor.
2263 for (i = 0; i < state->path.active; i++) {
2264 xfs_trans_brelse(args->trans, state->path.blk[i].bp);
2265 state->path.blk[i].bp = NULL;
2267 xfs_da_state_free(state);
2272 * Trim off a trailing empty freespace block.
2273 * Return (in rvalp) 1 if we did it, 0 if not.
2276 xfs_dir2_node_trim_free(
2277 xfs_da_args_t *args, /* operation arguments */
2278 xfs_fileoff_t fo, /* free block number */
2279 int *rvalp) /* out: did something */
2281 struct xfs_buf *bp; /* freespace buffer */
2282 xfs_inode_t *dp; /* incore directory inode */
2283 int error; /* error return code */
2284 xfs_dir2_free_t *free; /* freespace structure */
2285 xfs_trans_t *tp; /* transaction pointer */
2286 struct xfs_dir3_icfree_hdr freehdr;
2294 * Read the freespace block.
2296 error = xfs_dir2_free_try_read(tp, dp, fo, &bp);
2300 * There can be holes in freespace. If fo is a hole, there's
2306 dp->d_ops->free_hdr_from_disk(&freehdr, free);
2309 * If there are used entries, there's nothing to do.
2311 if (freehdr.nused > 0) {
2312 xfs_trans_brelse(tp, bp);
2316 * Blow the block away.
2318 error = xfs_dir2_shrink_inode(args,
2319 xfs_dir2_da_to_db(args->geo, (xfs_dablk_t)fo), bp);
2322 * Can't fail with ENOSPC since that only happens with no
2323 * space reservation, when breaking up an extent into two
2324 * pieces. This is the last block of an extent.
2326 ASSERT(error != -ENOSPC);
2327 xfs_trans_brelse(tp, bp);
2331 * Return that we succeeded.