2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
28 #include "xfs_trans.h"
29 #include "xfs_inode_item.h"
31 #include "xfs_buf_item.h"
33 #include "xfs_dir2_priv.h"
34 #include "xfs_error.h"
35 #include "xfs_trace.h"
36 #include "xfs_cksum.h"
39 * Local function prototypes.
41 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, struct xfs_buf *bp,
43 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, struct xfs_buf *bp);
44 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, struct xfs_buf **bpp,
46 static int xfs_dir2_block_sort(const void *a, const void *b);
48 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
51 * One-time startup routine called from xfs_init().
56 xfs_dir_hash_dot = xfs_da_hashname((unsigned char *)".", 1);
57 xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
61 xfs_dir3_block_verify(
64 struct xfs_mount *mp = bp->b_target->bt_mount;
65 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
67 if (xfs_sb_version_hascrc(&mp->m_sb)) {
68 if (hdr3->magic != cpu_to_be32(XFS_DIR3_BLOCK_MAGIC))
70 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
72 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
75 if (hdr3->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
78 if (__xfs_dir3_data_check(NULL, bp))
84 xfs_dir3_block_read_verify(
87 struct xfs_mount *mp = bp->b_target->bt_mount;
89 if (xfs_sb_version_hascrc(&mp->m_sb) &&
90 !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
91 xfs_buf_ioerror(bp, -EFSBADCRC);
92 else if (!xfs_dir3_block_verify(bp))
93 xfs_buf_ioerror(bp, -EFSCORRUPTED);
96 xfs_verifier_error(bp);
100 xfs_dir3_block_write_verify(
103 struct xfs_mount *mp = bp->b_target->bt_mount;
104 struct xfs_buf_log_item *bip = bp->b_fspriv;
105 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
107 if (!xfs_dir3_block_verify(bp)) {
108 xfs_buf_ioerror(bp, -EFSCORRUPTED);
109 xfs_verifier_error(bp);
113 if (!xfs_sb_version_hascrc(&mp->m_sb))
117 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
119 xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
122 const struct xfs_buf_ops xfs_dir3_block_buf_ops = {
123 .verify_read = xfs_dir3_block_read_verify,
124 .verify_write = xfs_dir3_block_write_verify,
129 struct xfs_trans *tp,
130 struct xfs_inode *dp,
131 struct xfs_buf **bpp)
133 struct xfs_mount *mp = dp->i_mount;
136 err = xfs_da_read_buf(tp, dp, mp->m_dir_geo->datablk, -1, bpp,
137 XFS_DATA_FORK, &xfs_dir3_block_buf_ops);
139 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_BLOCK_BUF);
145 struct xfs_mount *mp,
146 struct xfs_trans *tp,
148 struct xfs_inode *dp)
150 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
152 bp->b_ops = &xfs_dir3_block_buf_ops;
153 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_BLOCK_BUF);
155 if (xfs_sb_version_hascrc(&mp->m_sb)) {
156 memset(hdr3, 0, sizeof(*hdr3));
157 hdr3->magic = cpu_to_be32(XFS_DIR3_BLOCK_MAGIC);
158 hdr3->blkno = cpu_to_be64(bp->b_bn);
159 hdr3->owner = cpu_to_be64(dp->i_ino);
160 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
164 hdr3->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
168 xfs_dir2_block_need_space(
169 struct xfs_inode *dp,
170 struct xfs_dir2_data_hdr *hdr,
171 struct xfs_dir2_block_tail *btp,
172 struct xfs_dir2_leaf_entry *blp,
174 struct xfs_dir2_data_unused **dupp,
175 struct xfs_dir2_data_unused **enddupp,
179 struct xfs_dir2_data_free *bf;
181 struct xfs_dir2_data_unused *dup = NULL;
182 struct xfs_dir2_data_unused *enddup = NULL;
185 bf = dp->d_ops->data_bestfree_p(hdr);
188 * If there are stale entries we'll use one for the leaf.
191 if (be16_to_cpu(bf[0].length) >= len) {
193 * The biggest entry enough to avoid compaction.
195 dup = (xfs_dir2_data_unused_t *)
196 ((char *)hdr + be16_to_cpu(bf[0].offset));
201 * Will need to compact to make this work.
202 * Tag just before the first leaf entry.
205 tagp = (__be16 *)blp - 1;
207 /* Data object just before the first leaf entry. */
208 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
211 * If it's not free then the data will go where the
212 * leaf data starts now, if it works at all.
214 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
215 if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
216 (uint)sizeof(*blp) < len)
218 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
221 dup = (xfs_dir2_data_unused_t *)blp;
226 * no stale entries, so just use free space.
227 * Tag just before the first leaf entry.
229 tagp = (__be16 *)blp - 1;
231 /* Data object just before the first leaf entry. */
232 enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
235 * If it's not free then can't do this add without cleaning up:
236 * the space before the first leaf entry needs to be free so it
237 * can be expanded to hold the pointer to the new entry.
239 if (be16_to_cpu(enddup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
241 * Check out the biggest freespace and see if it's the same one.
243 dup = (xfs_dir2_data_unused_t *)
244 ((char *)hdr + be16_to_cpu(bf[0].offset));
247 * Not the same free entry, just check its length.
249 if (be16_to_cpu(dup->length) < len)
255 * It is the biggest freespace, can it hold the leaf too?
257 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
259 * Yes, use the second-largest entry instead if it works.
261 if (be16_to_cpu(bf[1].length) >= len)
262 dup = (xfs_dir2_data_unused_t *)
263 ((char *)hdr + be16_to_cpu(bf[1].offset));
275 * compact the leaf entries.
276 * Leave the highest-numbered stale entry stale.
277 * XXX should be the one closest to mid but mid is not yet computed.
280 xfs_dir2_block_compact(
281 struct xfs_da_args *args,
283 struct xfs_dir2_data_hdr *hdr,
284 struct xfs_dir2_block_tail *btp,
285 struct xfs_dir2_leaf_entry *blp,
290 int fromidx; /* source leaf index */
291 int toidx; /* target leaf index */
293 int highstale; /* high stale index */
295 fromidx = toidx = be32_to_cpu(btp->count) - 1;
296 highstale = *lfloghigh = -1;
297 for (; fromidx >= 0; fromidx--) {
298 if (blp[fromidx].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
302 if (*lfloghigh == -1)
308 blp[toidx] = blp[fromidx];
311 *lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
312 *lfloghigh -= be32_to_cpu(btp->stale) - 1;
313 be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
314 xfs_dir2_data_make_free(args, bp,
315 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
316 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
318 btp->stale = cpu_to_be32(1);
320 * If we now need to rebuild the bestfree map, do so.
321 * This needs to happen before the next call to use_free.
324 xfs_dir2_data_freescan(args->dp, hdr, needlog);
328 * Add an entry to a block directory.
331 xfs_dir2_block_addname(
332 xfs_da_args_t *args) /* directory op arguments */
334 xfs_dir2_data_hdr_t *hdr; /* block header */
335 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
336 struct xfs_buf *bp; /* buffer for block */
337 xfs_dir2_block_tail_t *btp; /* block tail */
338 int compact; /* need to compact leaf ents */
339 xfs_dir2_data_entry_t *dep; /* block data entry */
340 xfs_inode_t *dp; /* directory inode */
341 xfs_dir2_data_unused_t *dup; /* block unused entry */
342 int error; /* error return value */
343 xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
344 xfs_dahash_t hash; /* hash value of found entry */
345 int high; /* high index for binary srch */
346 int highstale; /* high stale index */
347 int lfloghigh=0; /* last final leaf to log */
348 int lfloglow=0; /* first final leaf to log */
349 int len; /* length of the new entry */
350 int low; /* low index for binary srch */
351 int lowstale; /* low stale index */
352 int mid=0; /* midpoint for binary srch */
353 int needlog; /* need to log header */
354 int needscan; /* need to rescan freespace */
355 __be16 *tagp; /* pointer to tag value */
356 xfs_trans_t *tp; /* transaction structure */
358 trace_xfs_dir2_block_addname(args);
363 /* Read the (one and only) directory block into bp. */
364 error = xfs_dir3_block_read(tp, dp, &bp);
368 len = dp->d_ops->data_entsize(args->namelen);
371 * Set up pointers to parts of the block.
374 btp = xfs_dir2_block_tail_p(args->geo, hdr);
375 blp = xfs_dir2_block_leaf_p(btp);
378 * Find out if we can reuse stale entries or whether we need extra
379 * space for entry and new leaf.
381 xfs_dir2_block_need_space(dp, hdr, btp, blp, &tagp, &dup,
382 &enddup, &compact, len);
385 * Done everything we need for a space check now.
387 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
388 xfs_trans_brelse(tp, bp);
395 * If we don't have space for the new entry & leaf ...
398 /* Don't have a space reservation: return no-space. */
399 if (args->total == 0)
402 * Convert to the next larger format.
403 * Then add the new entry in that format.
405 error = xfs_dir2_block_to_leaf(args, bp);
408 return xfs_dir2_leaf_addname(args);
411 needlog = needscan = 0;
414 * If need to compact the leaf entries, do it now.
417 xfs_dir2_block_compact(args, bp, hdr, btp, blp, &needlog,
418 &lfloghigh, &lfloglow);
419 /* recalculate blp post-compaction */
420 blp = xfs_dir2_block_leaf_p(btp);
421 } else if (btp->stale) {
423 * Set leaf logging boundaries to impossible state.
424 * For the no-stale case they're set explicitly.
426 lfloglow = be32_to_cpu(btp->count);
431 * Find the slot that's first lower than our hash value, -1 if none.
433 for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
434 mid = (low + high) >> 1;
435 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
437 if (hash < args->hashval)
442 while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
446 * No stale entries, will use enddup space to hold new leaf.
450 * Mark the space needed for the new leaf entry, now in use.
452 xfs_dir2_data_use_free(args, bp, enddup,
453 (xfs_dir2_data_aoff_t)
454 ((char *)enddup - (char *)hdr + be16_to_cpu(enddup->length) -
456 (xfs_dir2_data_aoff_t)sizeof(*blp),
457 &needlog, &needscan);
459 * Update the tail (entry count).
461 be32_add_cpu(&btp->count, 1);
463 * If we now need to rebuild the bestfree map, do so.
464 * This needs to happen before the next call to use_free.
467 xfs_dir2_data_freescan(dp, hdr, &needlog);
471 * Adjust pointer to the first leaf entry, we're about to move
472 * the table up one to open up space for the new leaf entry.
473 * Then adjust our index to match.
478 memmove(blp, &blp[1], mid * sizeof(*blp));
483 * Use a stale leaf for our new entry.
488 blp[lowstale].address !=
489 cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
492 for (highstale = mid + 1;
493 highstale < be32_to_cpu(btp->count) &&
494 blp[highstale].address !=
495 cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
496 (lowstale < 0 || mid - lowstale > highstale - mid);
500 * Move entries toward the low-numbered stale entry.
503 (highstale == be32_to_cpu(btp->count) ||
504 mid - lowstale <= highstale - mid)) {
506 memmove(&blp[lowstale], &blp[lowstale + 1],
507 (mid - lowstale) * sizeof(*blp));
508 lfloglow = MIN(lowstale, lfloglow);
509 lfloghigh = MAX(mid, lfloghigh);
512 * Move entries toward the high-numbered stale entry.
515 ASSERT(highstale < be32_to_cpu(btp->count));
518 memmove(&blp[mid + 1], &blp[mid],
519 (highstale - mid) * sizeof(*blp));
520 lfloglow = MIN(mid, lfloglow);
521 lfloghigh = MAX(highstale, lfloghigh);
523 be32_add_cpu(&btp->stale, -1);
526 * Point to the new data entry.
528 dep = (xfs_dir2_data_entry_t *)dup;
530 * Fill in the leaf entry.
532 blp[mid].hashval = cpu_to_be32(args->hashval);
533 blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
534 (char *)dep - (char *)hdr));
535 xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
537 * Mark space for the data entry used.
539 xfs_dir2_data_use_free(args, bp, dup,
540 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
541 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
543 * Create the new data entry.
545 dep->inumber = cpu_to_be64(args->inumber);
546 dep->namelen = args->namelen;
547 memcpy(dep->name, args->name, args->namelen);
548 dp->d_ops->data_put_ftype(dep, args->filetype);
549 tagp = dp->d_ops->data_entry_tag_p(dep);
550 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
552 * Clean up the bestfree array and log the header, tail, and entry.
555 xfs_dir2_data_freescan(dp, hdr, &needlog);
557 xfs_dir2_data_log_header(args, bp);
558 xfs_dir2_block_log_tail(tp, bp);
559 xfs_dir2_data_log_entry(args, bp, dep);
560 xfs_dir3_data_check(dp, bp);
565 * Log leaf entries from the block.
568 xfs_dir2_block_log_leaf(
569 xfs_trans_t *tp, /* transaction structure */
570 struct xfs_buf *bp, /* block buffer */
571 int first, /* index of first logged leaf */
572 int last) /* index of last logged leaf */
574 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
575 xfs_dir2_leaf_entry_t *blp;
576 xfs_dir2_block_tail_t *btp;
578 btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
579 blp = xfs_dir2_block_leaf_p(btp);
580 xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
581 (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
585 * Log the block tail.
588 xfs_dir2_block_log_tail(
589 xfs_trans_t *tp, /* transaction structure */
590 struct xfs_buf *bp) /* block buffer */
592 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
593 xfs_dir2_block_tail_t *btp;
595 btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
596 xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
597 (uint)((char *)(btp + 1) - (char *)hdr - 1));
601 * Look up an entry in the block. This is the external routine,
602 * xfs_dir2_block_lookup_int does the real work.
605 xfs_dir2_block_lookup(
606 xfs_da_args_t *args) /* dir lookup arguments */
608 xfs_dir2_data_hdr_t *hdr; /* block header */
609 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
610 struct xfs_buf *bp; /* block buffer */
611 xfs_dir2_block_tail_t *btp; /* block tail */
612 xfs_dir2_data_entry_t *dep; /* block data entry */
613 xfs_inode_t *dp; /* incore inode */
614 int ent; /* entry index */
615 int error; /* error return value */
617 trace_xfs_dir2_block_lookup(args);
620 * Get the buffer, look up the entry.
621 * If not found (ENOENT) then return, have no buffer.
623 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
627 xfs_dir3_data_check(dp, bp);
628 btp = xfs_dir2_block_tail_p(args->geo, hdr);
629 blp = xfs_dir2_block_leaf_p(btp);
631 * Get the offset from the leaf entry, to point to the data.
633 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
634 xfs_dir2_dataptr_to_off(args->geo,
635 be32_to_cpu(blp[ent].address)));
637 * Fill in inode number, CI name if appropriate, release the block.
639 args->inumber = be64_to_cpu(dep->inumber);
640 args->filetype = dp->d_ops->data_get_ftype(dep);
641 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
642 xfs_trans_brelse(args->trans, bp);
647 * Internal block lookup routine.
649 static int /* error */
650 xfs_dir2_block_lookup_int(
651 xfs_da_args_t *args, /* dir lookup arguments */
652 struct xfs_buf **bpp, /* returned block buffer */
653 int *entno) /* returned entry number */
655 xfs_dir2_dataptr_t addr; /* data entry address */
656 xfs_dir2_data_hdr_t *hdr; /* block header */
657 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
658 struct xfs_buf *bp; /* block buffer */
659 xfs_dir2_block_tail_t *btp; /* block tail */
660 xfs_dir2_data_entry_t *dep; /* block data entry */
661 xfs_inode_t *dp; /* incore inode */
662 int error; /* error return value */
663 xfs_dahash_t hash; /* found hash value */
664 int high; /* binary search high index */
665 int low; /* binary search low index */
666 int mid; /* binary search current idx */
667 xfs_mount_t *mp; /* filesystem mount point */
668 xfs_trans_t *tp; /* transaction pointer */
669 enum xfs_dacmp cmp; /* comparison result */
675 error = xfs_dir3_block_read(tp, dp, &bp);
680 xfs_dir3_data_check(dp, bp);
681 btp = xfs_dir2_block_tail_p(args->geo, hdr);
682 blp = xfs_dir2_block_leaf_p(btp);
684 * Loop doing a binary search for our hash value.
685 * Find our entry, ENOENT if it's not there.
687 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
689 mid = (low + high) >> 1;
690 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
692 if (hash < args->hashval)
697 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
698 xfs_trans_brelse(tp, bp);
703 * Back up to the first one with the right hash value.
705 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
709 * Now loop forward through all the entries with the
710 * right hash value looking for our name.
713 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
716 * Get pointer to the entry from the leaf.
718 dep = (xfs_dir2_data_entry_t *)
719 ((char *)hdr + xfs_dir2_dataptr_to_off(args->geo, addr));
721 * Compare name and if it's an exact match, return the index
722 * and buffer. If it's the first case-insensitive match, store
723 * the index and buffer and continue looking for an exact match.
725 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
726 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
727 args->cmpresult = cmp;
730 if (cmp == XFS_CMP_EXACT)
733 } while (++mid < be32_to_cpu(btp->count) &&
734 be32_to_cpu(blp[mid].hashval) == hash);
736 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
738 * Here, we can only be doing a lookup (not a rename or replace).
739 * If a case-insensitive match was found earlier, return success.
741 if (args->cmpresult == XFS_CMP_CASE)
744 * No match, release the buffer and return ENOENT.
746 xfs_trans_brelse(tp, bp);
751 * Remove an entry from a block format directory.
752 * If that makes the block small enough to fit in shortform, transform it.
755 xfs_dir2_block_removename(
756 xfs_da_args_t *args) /* directory operation args */
758 xfs_dir2_data_hdr_t *hdr; /* block header */
759 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
760 struct xfs_buf *bp; /* block buffer */
761 xfs_dir2_block_tail_t *btp; /* block tail */
762 xfs_dir2_data_entry_t *dep; /* block data entry */
763 xfs_inode_t *dp; /* incore inode */
764 int ent; /* block leaf entry index */
765 int error; /* error return value */
766 int needlog; /* need to log block header */
767 int needscan; /* need to fixup bestfree */
768 xfs_dir2_sf_hdr_t sfh; /* shortform header */
769 int size; /* shortform size */
770 xfs_trans_t *tp; /* transaction pointer */
772 trace_xfs_dir2_block_removename(args);
775 * Look up the entry in the block. Gets the buffer and entry index.
776 * It will always be there, the vnodeops level does a lookup first.
778 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
784 btp = xfs_dir2_block_tail_p(args->geo, hdr);
785 blp = xfs_dir2_block_leaf_p(btp);
787 * Point to the data entry using the leaf entry.
789 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
790 xfs_dir2_dataptr_to_off(args->geo,
791 be32_to_cpu(blp[ent].address)));
793 * Mark the data entry's space free.
795 needlog = needscan = 0;
796 xfs_dir2_data_make_free(args, bp,
797 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
798 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
800 * Fix up the block tail.
802 be32_add_cpu(&btp->stale, 1);
803 xfs_dir2_block_log_tail(tp, bp);
805 * Remove the leaf entry by marking it stale.
807 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
808 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
810 * Fix up bestfree, log the header if necessary.
813 xfs_dir2_data_freescan(dp, hdr, &needlog);
815 xfs_dir2_data_log_header(args, bp);
816 xfs_dir3_data_check(dp, bp);
818 * See if the size as a shortform is good enough.
820 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
821 if (size > XFS_IFORK_DSIZE(dp))
825 * If it works, do the conversion.
827 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
831 * Replace an entry in a V2 block directory.
832 * Change the inode number to the new value.
835 xfs_dir2_block_replace(
836 xfs_da_args_t *args) /* directory operation args */
838 xfs_dir2_data_hdr_t *hdr; /* block header */
839 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
840 struct xfs_buf *bp; /* block buffer */
841 xfs_dir2_block_tail_t *btp; /* block tail */
842 xfs_dir2_data_entry_t *dep; /* block data entry */
843 xfs_inode_t *dp; /* incore inode */
844 int ent; /* leaf entry index */
845 int error; /* error return value */
847 trace_xfs_dir2_block_replace(args);
850 * Lookup the entry in the directory. Get buffer and entry index.
851 * This will always succeed since the caller has already done a lookup.
853 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
858 btp = xfs_dir2_block_tail_p(args->geo, hdr);
859 blp = xfs_dir2_block_leaf_p(btp);
861 * Point to the data entry we need to change.
863 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
864 xfs_dir2_dataptr_to_off(args->geo,
865 be32_to_cpu(blp[ent].address)));
866 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
868 * Change the inode number to the new value.
870 dep->inumber = cpu_to_be64(args->inumber);
871 dp->d_ops->data_put_ftype(dep, args->filetype);
872 xfs_dir2_data_log_entry(args, bp, dep);
873 xfs_dir3_data_check(dp, bp);
878 * Qsort comparison routine for the block leaf entries.
880 static int /* sort order */
882 const void *a, /* first leaf entry */
883 const void *b) /* second leaf entry */
885 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
886 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
890 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
891 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
895 * Convert a V2 leaf directory to a V2 block directory if possible.
898 xfs_dir2_leaf_to_block(
899 xfs_da_args_t *args, /* operation arguments */
900 struct xfs_buf *lbp, /* leaf buffer */
901 struct xfs_buf *dbp) /* data buffer */
903 __be16 *bestsp; /* leaf bests table */
904 xfs_dir2_data_hdr_t *hdr; /* block header */
905 xfs_dir2_block_tail_t *btp; /* block tail */
906 xfs_inode_t *dp; /* incore directory inode */
907 xfs_dir2_data_unused_t *dup; /* unused data entry */
908 int error; /* error return value */
909 int from; /* leaf from index */
910 xfs_dir2_leaf_t *leaf; /* leaf structure */
911 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
912 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
913 xfs_mount_t *mp; /* file system mount point */
914 int needlog; /* need to log data header */
915 int needscan; /* need to scan for bestfree */
916 xfs_dir2_sf_hdr_t sfh; /* shortform header */
917 int size; /* bytes used */
918 __be16 *tagp; /* end of entry (tag) */
919 int to; /* block/leaf to index */
920 xfs_trans_t *tp; /* transaction pointer */
921 struct xfs_dir2_leaf_entry *ents;
922 struct xfs_dir3_icleaf_hdr leafhdr;
924 trace_xfs_dir2_leaf_to_block(args);
930 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
931 ents = dp->d_ops->leaf_ents_p(leaf);
932 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
934 ASSERT(leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
935 leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
937 * If there are data blocks other than the first one, take this
938 * opportunity to remove trailing empty data blocks that may have
939 * been left behind during no-space-reservation operations.
940 * These will show up in the leaf bests table.
942 while (dp->i_d.di_size > args->geo->blksize) {
945 hdrsz = dp->d_ops->data_entry_offset;
946 bestsp = xfs_dir2_leaf_bests_p(ltp);
947 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
948 args->geo->blksize - hdrsz) {
950 xfs_dir2_leaf_trim_data(args, lbp,
951 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
957 * Read the data block if we don't already have it, give up if it fails.
960 error = xfs_dir3_data_read(tp, dp, args->geo->datablk, -1, &dbp);
965 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
966 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
969 * Size of the "leaf" area in the block.
971 size = (uint)sizeof(xfs_dir2_block_tail_t) +
972 (uint)sizeof(*lep) * (leafhdr.count - leafhdr.stale);
974 * Look at the last data entry.
976 tagp = (__be16 *)((char *)hdr + args->geo->blksize) - 1;
977 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
979 * If it's not free or is too short we can't do it.
981 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
982 be16_to_cpu(dup->length) < size)
986 * Start converting it to block form.
988 xfs_dir3_block_init(mp, tp, dbp, dp);
993 * Use up the space at the end of the block (blp/btp).
995 xfs_dir2_data_use_free(args, dbp, dup, args->geo->blksize - size, size,
996 &needlog, &needscan);
998 * Initialize the block tail.
1000 btp = xfs_dir2_block_tail_p(args->geo, hdr);
1001 btp->count = cpu_to_be32(leafhdr.count - leafhdr.stale);
1003 xfs_dir2_block_log_tail(tp, dbp);
1005 * Initialize the block leaf area. We compact out stale entries.
1007 lep = xfs_dir2_block_leaf_p(btp);
1008 for (from = to = 0; from < leafhdr.count; from++) {
1009 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
1011 lep[to++] = ents[from];
1013 ASSERT(to == be32_to_cpu(btp->count));
1014 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
1016 * Scan the bestfree if we need it and log the data block header.
1019 xfs_dir2_data_freescan(dp, hdr, &needlog);
1021 xfs_dir2_data_log_header(args, dbp);
1023 * Pitch the old leaf block.
1025 error = xfs_da_shrink_inode(args, args->geo->leafblk, lbp);
1030 * Now see if the resulting block can be shrunken to shortform.
1032 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1033 if (size > XFS_IFORK_DSIZE(dp))
1036 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1040 * Convert the shortform directory to block form.
1043 xfs_dir2_sf_to_block(
1044 xfs_da_args_t *args) /* operation arguments */
1046 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1047 xfs_dir2_data_hdr_t *hdr; /* block header */
1048 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1049 struct xfs_buf *bp; /* block buffer */
1050 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1051 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1052 xfs_inode_t *dp; /* incore directory inode */
1053 int dummy; /* trash */
1054 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1055 int endoffset; /* end of data objects */
1056 int error; /* error return value */
1058 xfs_mount_t *mp; /* filesystem mount point */
1059 int needlog; /* need to log block header */
1060 int needscan; /* need to scan block freespc */
1061 int newoffset; /* offset from current entry */
1062 int offset; /* target block offset */
1063 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1064 xfs_dir2_sf_hdr_t *oldsfp; /* old shortform header */
1065 xfs_dir2_sf_hdr_t *sfp; /* shortform header */
1066 __be16 *tagp; /* end of data entry */
1067 xfs_trans_t *tp; /* transaction pointer */
1068 struct xfs_name name;
1069 struct xfs_ifork *ifp;
1071 trace_xfs_dir2_sf_to_block(args);
1076 ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
1077 ASSERT(ifp->if_flags & XFS_IFINLINE);
1079 * Bomb out if the shortform directory is way too short.
1081 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1082 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1086 oldsfp = (xfs_dir2_sf_hdr_t *)ifp->if_u1.if_data;
1088 ASSERT(ifp->if_bytes == dp->i_d.di_size);
1089 ASSERT(ifp->if_u1.if_data != NULL);
1090 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1091 ASSERT(dp->i_d.di_nextents == 0);
1094 * Copy the directory into a temporary buffer.
1095 * Then pitch the incore inode data so we can make extents.
1097 sfp = kmem_alloc(ifp->if_bytes, KM_SLEEP);
1098 memcpy(sfp, oldsfp, ifp->if_bytes);
1100 xfs_idata_realloc(dp, -ifp->if_bytes, XFS_DATA_FORK);
1101 xfs_bmap_local_to_extents_empty(dp, XFS_DATA_FORK);
1102 dp->i_d.di_size = 0;
1105 * Add block 0 to the inode.
1107 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1113 * Initialize the data block, then convert it to block format.
1115 error = xfs_dir3_data_init(args, blkno, &bp);
1120 xfs_dir3_block_init(mp, tp, bp, dp);
1124 * Compute size of block "tail" area.
1126 i = (uint)sizeof(*btp) +
1127 (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1129 * The whole thing is initialized to free by the init routine.
1130 * Say we're using the leaf and tail area.
1132 dup = dp->d_ops->data_unused_p(hdr);
1133 needlog = needscan = 0;
1134 xfs_dir2_data_use_free(args, bp, dup, args->geo->blksize - i,
1135 i, &needlog, &needscan);
1136 ASSERT(needscan == 0);
1140 btp = xfs_dir2_block_tail_p(args->geo, hdr);
1141 btp->count = cpu_to_be32(sfp->count + 2); /* ., .. */
1143 blp = xfs_dir2_block_leaf_p(btp);
1144 endoffset = (uint)((char *)blp - (char *)hdr);
1146 * Remove the freespace, we'll manage it.
1148 xfs_dir2_data_use_free(args, bp, dup,
1149 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1150 be16_to_cpu(dup->length), &needlog, &needscan);
1152 * Create entry for .
1154 dep = dp->d_ops->data_dot_entry_p(hdr);
1155 dep->inumber = cpu_to_be64(dp->i_ino);
1158 dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1159 tagp = dp->d_ops->data_entry_tag_p(dep);
1160 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1161 xfs_dir2_data_log_entry(args, bp, dep);
1162 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1163 blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1164 (char *)dep - (char *)hdr));
1166 * Create entry for ..
1168 dep = dp->d_ops->data_dotdot_entry_p(hdr);
1169 dep->inumber = cpu_to_be64(dp->d_ops->sf_get_parent_ino(sfp));
1171 dep->name[0] = dep->name[1] = '.';
1172 dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1173 tagp = dp->d_ops->data_entry_tag_p(dep);
1174 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1175 xfs_dir2_data_log_entry(args, bp, dep);
1176 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1177 blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1178 (char *)dep - (char *)hdr));
1179 offset = dp->d_ops->data_first_offset;
1181 * Loop over existing entries, stuff them in.
1187 sfep = xfs_dir2_sf_firstentry(sfp);
1189 * Need to preserve the existing offset values in the sf directory.
1190 * Insert holes (unused entries) where necessary.
1192 while (offset < endoffset) {
1194 * sfep is null when we reach the end of the list.
1197 newoffset = endoffset;
1199 newoffset = xfs_dir2_sf_get_offset(sfep);
1201 * There should be a hole here, make one.
1203 if (offset < newoffset) {
1204 dup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1205 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1206 dup->length = cpu_to_be16(newoffset - offset);
1207 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1208 ((char *)dup - (char *)hdr));
1209 xfs_dir2_data_log_unused(args, bp, dup);
1210 xfs_dir2_data_freeinsert(hdr,
1211 dp->d_ops->data_bestfree_p(hdr),
1213 offset += be16_to_cpu(dup->length);
1217 * Copy a real entry.
1219 dep = (xfs_dir2_data_entry_t *)((char *)hdr + newoffset);
1220 dep->inumber = cpu_to_be64(dp->d_ops->sf_get_ino(sfp, sfep));
1221 dep->namelen = sfep->namelen;
1222 dp->d_ops->data_put_ftype(dep, dp->d_ops->sf_get_ftype(sfep));
1223 memcpy(dep->name, sfep->name, dep->namelen);
1224 tagp = dp->d_ops->data_entry_tag_p(dep);
1225 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1226 xfs_dir2_data_log_entry(args, bp, dep);
1227 name.name = sfep->name;
1228 name.len = sfep->namelen;
1229 blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1231 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1232 (char *)dep - (char *)hdr));
1233 offset = (int)((char *)(tagp + 1) - (char *)hdr);
1234 if (++i == sfp->count)
1237 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
1239 /* Done with the temporary buffer */
1242 * Sort the leaf entries by hash value.
1244 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1246 * Log the leaf entry area and tail.
1247 * Already logged the header in data_init, ignore needlog.
1249 ASSERT(needscan == 0);
1250 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1251 xfs_dir2_block_log_tail(tp, bp);
1252 xfs_dir3_data_check(dp, bp);