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"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
31 #include "xfs_inode_item.h"
33 #include "xfs_buf_item.h"
35 #include "xfs_dir2_priv.h"
36 #include "xfs_error.h"
37 #include "xfs_trace.h"
38 #include "xfs_cksum.h"
39 #include "xfs_dinode.h"
42 * Local function prototypes.
44 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, struct xfs_buf *bp,
46 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, struct xfs_buf *bp);
47 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, struct xfs_buf **bpp,
49 static int xfs_dir2_block_sort(const void *a, const void *b);
51 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
54 * One-time startup routine called from xfs_init().
59 xfs_dir_hash_dot = xfs_da_hashname((unsigned char *)".", 1);
60 xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
64 xfs_dir3_block_verify(
67 struct xfs_mount *mp = bp->b_target->bt_mount;
68 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
70 if (xfs_sb_version_hascrc(&mp->m_sb)) {
71 if (hdr3->magic != cpu_to_be32(XFS_DIR3_BLOCK_MAGIC))
73 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
75 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
78 if (hdr3->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
81 if (__xfs_dir3_data_check(NULL, bp))
87 xfs_dir3_block_read_verify(
90 struct xfs_mount *mp = bp->b_target->bt_mount;
92 if (xfs_sb_version_hascrc(&mp->m_sb) &&
93 !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
94 xfs_buf_ioerror(bp, EFSBADCRC);
95 else if (!xfs_dir3_block_verify(bp))
96 xfs_buf_ioerror(bp, EFSCORRUPTED);
99 xfs_verifier_error(bp);
103 xfs_dir3_block_write_verify(
106 struct xfs_mount *mp = bp->b_target->bt_mount;
107 struct xfs_buf_log_item *bip = bp->b_fspriv;
108 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
110 if (!xfs_dir3_block_verify(bp)) {
111 xfs_buf_ioerror(bp, EFSCORRUPTED);
112 xfs_verifier_error(bp);
116 if (!xfs_sb_version_hascrc(&mp->m_sb))
120 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
122 xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
125 const struct xfs_buf_ops xfs_dir3_block_buf_ops = {
126 .verify_read = xfs_dir3_block_read_verify,
127 .verify_write = xfs_dir3_block_write_verify,
132 struct xfs_trans *tp,
133 struct xfs_inode *dp,
134 struct xfs_buf **bpp)
136 struct xfs_mount *mp = dp->i_mount;
139 err = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, bpp,
140 XFS_DATA_FORK, &xfs_dir3_block_buf_ops);
142 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_BLOCK_BUF);
148 struct xfs_mount *mp,
149 struct xfs_trans *tp,
151 struct xfs_inode *dp)
153 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
155 bp->b_ops = &xfs_dir3_block_buf_ops;
156 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_BLOCK_BUF);
158 if (xfs_sb_version_hascrc(&mp->m_sb)) {
159 memset(hdr3, 0, sizeof(*hdr3));
160 hdr3->magic = cpu_to_be32(XFS_DIR3_BLOCK_MAGIC);
161 hdr3->blkno = cpu_to_be64(bp->b_bn);
162 hdr3->owner = cpu_to_be64(dp->i_ino);
163 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
167 hdr3->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
171 xfs_dir2_block_need_space(
172 struct xfs_inode *dp,
173 struct xfs_dir2_data_hdr *hdr,
174 struct xfs_dir2_block_tail *btp,
175 struct xfs_dir2_leaf_entry *blp,
177 struct xfs_dir2_data_unused **dupp,
178 struct xfs_dir2_data_unused **enddupp,
182 struct xfs_dir2_data_free *bf;
184 struct xfs_dir2_data_unused *dup = NULL;
185 struct xfs_dir2_data_unused *enddup = NULL;
188 bf = dp->d_ops->data_bestfree_p(hdr);
191 * If there are stale entries we'll use one for the leaf.
194 if (be16_to_cpu(bf[0].length) >= len) {
196 * The biggest entry enough to avoid compaction.
198 dup = (xfs_dir2_data_unused_t *)
199 ((char *)hdr + be16_to_cpu(bf[0].offset));
204 * Will need to compact to make this work.
205 * Tag just before the first leaf entry.
208 tagp = (__be16 *)blp - 1;
210 /* Data object just before the first leaf entry. */
211 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
214 * If it's not free then the data will go where the
215 * leaf data starts now, if it works at all.
217 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
218 if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
219 (uint)sizeof(*blp) < len)
221 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
224 dup = (xfs_dir2_data_unused_t *)blp;
229 * no stale entries, so just use free space.
230 * Tag just before the first leaf entry.
232 tagp = (__be16 *)blp - 1;
234 /* Data object just before the first leaf entry. */
235 enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
238 * If it's not free then can't do this add without cleaning up:
239 * the space before the first leaf entry needs to be free so it
240 * can be expanded to hold the pointer to the new entry.
242 if (be16_to_cpu(enddup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
244 * Check out the biggest freespace and see if it's the same one.
246 dup = (xfs_dir2_data_unused_t *)
247 ((char *)hdr + be16_to_cpu(bf[0].offset));
250 * Not the same free entry, just check its length.
252 if (be16_to_cpu(dup->length) < len)
258 * It is the biggest freespace, can it hold the leaf too?
260 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
262 * Yes, use the second-largest entry instead if it works.
264 if (be16_to_cpu(bf[1].length) >= len)
265 dup = (xfs_dir2_data_unused_t *)
266 ((char *)hdr + be16_to_cpu(bf[1].offset));
278 * compact the leaf entries.
279 * Leave the highest-numbered stale entry stale.
280 * XXX should be the one closest to mid but mid is not yet computed.
283 xfs_dir2_block_compact(
284 struct xfs_trans *tp,
285 struct xfs_inode *dp,
287 struct xfs_dir2_data_hdr *hdr,
288 struct xfs_dir2_block_tail *btp,
289 struct xfs_dir2_leaf_entry *blp,
294 int fromidx; /* source leaf index */
295 int toidx; /* target leaf index */
297 int highstale; /* high stale index */
299 fromidx = toidx = be32_to_cpu(btp->count) - 1;
300 highstale = *lfloghigh = -1;
301 for (; fromidx >= 0; fromidx--) {
302 if (blp[fromidx].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
306 if (*lfloghigh == -1)
312 blp[toidx] = blp[fromidx];
315 *lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
316 *lfloghigh -= be32_to_cpu(btp->stale) - 1;
317 be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
318 xfs_dir2_data_make_free(tp, dp, bp,
319 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
320 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
322 btp->stale = cpu_to_be32(1);
324 * If we now need to rebuild the bestfree map, do so.
325 * This needs to happen before the next call to use_free.
328 xfs_dir2_data_freescan(dp, hdr, needlog);
332 * Add an entry to a block directory.
335 xfs_dir2_block_addname(
336 xfs_da_args_t *args) /* directory op arguments */
338 xfs_dir2_data_hdr_t *hdr; /* block header */
339 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
340 struct xfs_buf *bp; /* buffer for block */
341 xfs_dir2_block_tail_t *btp; /* block tail */
342 int compact; /* need to compact leaf ents */
343 xfs_dir2_data_entry_t *dep; /* block data entry */
344 xfs_inode_t *dp; /* directory inode */
345 xfs_dir2_data_unused_t *dup; /* block unused entry */
346 int error; /* error return value */
347 xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
348 xfs_dahash_t hash; /* hash value of found entry */
349 int high; /* high index for binary srch */
350 int highstale; /* high stale index */
351 int lfloghigh=0; /* last final leaf to log */
352 int lfloglow=0; /* first final leaf to log */
353 int len; /* length of the new entry */
354 int low; /* low index for binary srch */
355 int lowstale; /* low stale index */
356 int mid=0; /* midpoint for binary srch */
357 xfs_mount_t *mp; /* filesystem mount point */
358 int needlog; /* need to log header */
359 int needscan; /* need to rescan freespace */
360 __be16 *tagp; /* pointer to tag value */
361 xfs_trans_t *tp; /* transaction structure */
363 trace_xfs_dir2_block_addname(args);
369 /* Read the (one and only) directory block into bp. */
370 error = xfs_dir3_block_read(tp, dp, &bp);
374 len = dp->d_ops->data_entsize(args->namelen);
377 * Set up pointers to parts of the block.
380 btp = xfs_dir2_block_tail_p(mp, hdr);
381 blp = xfs_dir2_block_leaf_p(btp);
384 * Find out if we can reuse stale entries or whether we need extra
385 * space for entry and new leaf.
387 xfs_dir2_block_need_space(dp, hdr, btp, blp, &tagp, &dup,
388 &enddup, &compact, len);
391 * Done everything we need for a space check now.
393 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
394 xfs_trans_brelse(tp, bp);
396 return XFS_ERROR(ENOSPC);
401 * If we don't have space for the new entry & leaf ...
404 /* Don't have a space reservation: return no-space. */
405 if (args->total == 0)
406 return XFS_ERROR(ENOSPC);
408 * Convert to the next larger format.
409 * Then add the new entry in that format.
411 error = xfs_dir2_block_to_leaf(args, bp);
414 return xfs_dir2_leaf_addname(args);
417 needlog = needscan = 0;
420 * If need to compact the leaf entries, do it now.
423 xfs_dir2_block_compact(tp, dp, bp, hdr, btp, blp, &needlog,
424 &lfloghigh, &lfloglow);
425 /* recalculate blp post-compaction */
426 blp = xfs_dir2_block_leaf_p(btp);
427 } else if (btp->stale) {
429 * Set leaf logging boundaries to impossible state.
430 * For the no-stale case they're set explicitly.
432 lfloglow = be32_to_cpu(btp->count);
437 * Find the slot that's first lower than our hash value, -1 if none.
439 for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
440 mid = (low + high) >> 1;
441 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
443 if (hash < args->hashval)
448 while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
452 * No stale entries, will use enddup space to hold new leaf.
456 * Mark the space needed for the new leaf entry, now in use.
458 xfs_dir2_data_use_free(tp, dp, bp, enddup,
459 (xfs_dir2_data_aoff_t)
460 ((char *)enddup - (char *)hdr + be16_to_cpu(enddup->length) -
462 (xfs_dir2_data_aoff_t)sizeof(*blp),
463 &needlog, &needscan);
465 * Update the tail (entry count).
467 be32_add_cpu(&btp->count, 1);
469 * If we now need to rebuild the bestfree map, do so.
470 * This needs to happen before the next call to use_free.
473 xfs_dir2_data_freescan(dp, hdr, &needlog);
477 * Adjust pointer to the first leaf entry, we're about to move
478 * the table up one to open up space for the new leaf entry.
479 * Then adjust our index to match.
484 memmove(blp, &blp[1], mid * sizeof(*blp));
489 * Use a stale leaf for our new entry.
494 blp[lowstale].address !=
495 cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
498 for (highstale = mid + 1;
499 highstale < be32_to_cpu(btp->count) &&
500 blp[highstale].address !=
501 cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
502 (lowstale < 0 || mid - lowstale > highstale - mid);
506 * Move entries toward the low-numbered stale entry.
509 (highstale == be32_to_cpu(btp->count) ||
510 mid - lowstale <= highstale - mid)) {
512 memmove(&blp[lowstale], &blp[lowstale + 1],
513 (mid - lowstale) * sizeof(*blp));
514 lfloglow = MIN(lowstale, lfloglow);
515 lfloghigh = MAX(mid, lfloghigh);
518 * Move entries toward the high-numbered stale entry.
521 ASSERT(highstale < be32_to_cpu(btp->count));
524 memmove(&blp[mid + 1], &blp[mid],
525 (highstale - mid) * sizeof(*blp));
526 lfloglow = MIN(mid, lfloglow);
527 lfloghigh = MAX(highstale, lfloghigh);
529 be32_add_cpu(&btp->stale, -1);
532 * Point to the new data entry.
534 dep = (xfs_dir2_data_entry_t *)dup;
536 * Fill in the leaf entry.
538 blp[mid].hashval = cpu_to_be32(args->hashval);
539 blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
540 (char *)dep - (char *)hdr));
541 xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
543 * Mark space for the data entry used.
545 xfs_dir2_data_use_free(tp, dp, bp, dup,
546 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
547 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
549 * Create the new data entry.
551 dep->inumber = cpu_to_be64(args->inumber);
552 dep->namelen = args->namelen;
553 memcpy(dep->name, args->name, args->namelen);
554 dp->d_ops->data_put_ftype(dep, args->filetype);
555 tagp = dp->d_ops->data_entry_tag_p(dep);
556 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
558 * Clean up the bestfree array and log the header, tail, and entry.
561 xfs_dir2_data_freescan(dp, hdr, &needlog);
563 xfs_dir2_data_log_header(tp, dp, bp);
564 xfs_dir2_block_log_tail(tp, bp);
565 xfs_dir2_data_log_entry(tp, dp, bp, dep);
566 xfs_dir3_data_check(dp, bp);
571 * Log leaf entries from the block.
574 xfs_dir2_block_log_leaf(
575 xfs_trans_t *tp, /* transaction structure */
576 struct xfs_buf *bp, /* block buffer */
577 int first, /* index of first logged leaf */
578 int last) /* index of last logged leaf */
580 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
581 xfs_dir2_leaf_entry_t *blp;
582 xfs_dir2_block_tail_t *btp;
584 btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
585 blp = xfs_dir2_block_leaf_p(btp);
586 xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
587 (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
591 * Log the block tail.
594 xfs_dir2_block_log_tail(
595 xfs_trans_t *tp, /* transaction structure */
596 struct xfs_buf *bp) /* block buffer */
598 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
599 xfs_dir2_block_tail_t *btp;
601 btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
602 xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
603 (uint)((char *)(btp + 1) - (char *)hdr - 1));
607 * Look up an entry in the block. This is the external routine,
608 * xfs_dir2_block_lookup_int does the real work.
611 xfs_dir2_block_lookup(
612 xfs_da_args_t *args) /* dir lookup arguments */
614 xfs_dir2_data_hdr_t *hdr; /* block header */
615 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
616 struct xfs_buf *bp; /* block buffer */
617 xfs_dir2_block_tail_t *btp; /* block tail */
618 xfs_dir2_data_entry_t *dep; /* block data entry */
619 xfs_inode_t *dp; /* incore inode */
620 int ent; /* entry index */
621 int error; /* error return value */
622 xfs_mount_t *mp; /* filesystem mount point */
624 trace_xfs_dir2_block_lookup(args);
627 * Get the buffer, look up the entry.
628 * If not found (ENOENT) then return, have no buffer.
630 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
635 xfs_dir3_data_check(dp, bp);
636 btp = xfs_dir2_block_tail_p(mp, hdr);
637 blp = xfs_dir2_block_leaf_p(btp);
639 * Get the offset from the leaf entry, to point to the data.
641 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
642 xfs_dir2_dataptr_to_off(args->geo,
643 be32_to_cpu(blp[ent].address)));
645 * Fill in inode number, CI name if appropriate, release the block.
647 args->inumber = be64_to_cpu(dep->inumber);
648 args->filetype = dp->d_ops->data_get_ftype(dep);
649 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
650 xfs_trans_brelse(args->trans, bp);
651 return XFS_ERROR(error);
655 * Internal block lookup routine.
657 static int /* error */
658 xfs_dir2_block_lookup_int(
659 xfs_da_args_t *args, /* dir lookup arguments */
660 struct xfs_buf **bpp, /* returned block buffer */
661 int *entno) /* returned entry number */
663 xfs_dir2_dataptr_t addr; /* data entry address */
664 xfs_dir2_data_hdr_t *hdr; /* block header */
665 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
666 struct xfs_buf *bp; /* block buffer */
667 xfs_dir2_block_tail_t *btp; /* block tail */
668 xfs_dir2_data_entry_t *dep; /* block data entry */
669 xfs_inode_t *dp; /* incore inode */
670 int error; /* error return value */
671 xfs_dahash_t hash; /* found hash value */
672 int high; /* binary search high index */
673 int low; /* binary search low index */
674 int mid; /* binary search current idx */
675 xfs_mount_t *mp; /* filesystem mount point */
676 xfs_trans_t *tp; /* transaction pointer */
677 enum xfs_dacmp cmp; /* comparison result */
683 error = xfs_dir3_block_read(tp, dp, &bp);
688 xfs_dir3_data_check(dp, bp);
689 btp = xfs_dir2_block_tail_p(mp, hdr);
690 blp = xfs_dir2_block_leaf_p(btp);
692 * Loop doing a binary search for our hash value.
693 * Find our entry, ENOENT if it's not there.
695 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
697 mid = (low + high) >> 1;
698 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
700 if (hash < args->hashval)
705 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
706 xfs_trans_brelse(tp, bp);
707 return XFS_ERROR(ENOENT);
711 * Back up to the first one with the right hash value.
713 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
717 * Now loop forward through all the entries with the
718 * right hash value looking for our name.
721 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
724 * Get pointer to the entry from the leaf.
726 dep = (xfs_dir2_data_entry_t *)
727 ((char *)hdr + xfs_dir2_dataptr_to_off(args->geo, addr));
729 * Compare name and if it's an exact match, return the index
730 * and buffer. If it's the first case-insensitive match, store
731 * the index and buffer and continue looking for an exact match.
733 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
734 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
735 args->cmpresult = cmp;
738 if (cmp == XFS_CMP_EXACT)
741 } while (++mid < be32_to_cpu(btp->count) &&
742 be32_to_cpu(blp[mid].hashval) == hash);
744 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
746 * Here, we can only be doing a lookup (not a rename or replace).
747 * If a case-insensitive match was found earlier, return success.
749 if (args->cmpresult == XFS_CMP_CASE)
752 * No match, release the buffer and return ENOENT.
754 xfs_trans_brelse(tp, bp);
755 return XFS_ERROR(ENOENT);
759 * Remove an entry from a block format directory.
760 * If that makes the block small enough to fit in shortform, transform it.
763 xfs_dir2_block_removename(
764 xfs_da_args_t *args) /* directory operation args */
766 xfs_dir2_data_hdr_t *hdr; /* block header */
767 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
768 struct xfs_buf *bp; /* block buffer */
769 xfs_dir2_block_tail_t *btp; /* block tail */
770 xfs_dir2_data_entry_t *dep; /* block data entry */
771 xfs_inode_t *dp; /* incore inode */
772 int ent; /* block leaf entry index */
773 int error; /* error return value */
774 xfs_mount_t *mp; /* filesystem mount point */
775 int needlog; /* need to log block header */
776 int needscan; /* need to fixup bestfree */
777 xfs_dir2_sf_hdr_t sfh; /* shortform header */
778 int size; /* shortform size */
779 xfs_trans_t *tp; /* transaction pointer */
781 trace_xfs_dir2_block_removename(args);
784 * Look up the entry in the block. Gets the buffer and entry index.
785 * It will always be there, the vnodeops level does a lookup first.
787 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
794 btp = xfs_dir2_block_tail_p(mp, hdr);
795 blp = xfs_dir2_block_leaf_p(btp);
797 * Point to the data entry using the leaf entry.
799 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
800 xfs_dir2_dataptr_to_off(args->geo,
801 be32_to_cpu(blp[ent].address)));
803 * Mark the data entry's space free.
805 needlog = needscan = 0;
806 xfs_dir2_data_make_free(tp, dp, bp,
807 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
808 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
810 * Fix up the block tail.
812 be32_add_cpu(&btp->stale, 1);
813 xfs_dir2_block_log_tail(tp, bp);
815 * Remove the leaf entry by marking it stale.
817 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
818 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
820 * Fix up bestfree, log the header if necessary.
823 xfs_dir2_data_freescan(dp, hdr, &needlog);
825 xfs_dir2_data_log_header(tp, dp, bp);
826 xfs_dir3_data_check(dp, bp);
828 * See if the size as a shortform is good enough.
830 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
831 if (size > XFS_IFORK_DSIZE(dp))
835 * If it works, do the conversion.
837 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
841 * Replace an entry in a V2 block directory.
842 * Change the inode number to the new value.
845 xfs_dir2_block_replace(
846 xfs_da_args_t *args) /* directory operation args */
848 xfs_dir2_data_hdr_t *hdr; /* block header */
849 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
850 struct xfs_buf *bp; /* block buffer */
851 xfs_dir2_block_tail_t *btp; /* block tail */
852 xfs_dir2_data_entry_t *dep; /* block data entry */
853 xfs_inode_t *dp; /* incore inode */
854 int ent; /* leaf entry index */
855 int error; /* error return value */
856 xfs_mount_t *mp; /* filesystem mount point */
858 trace_xfs_dir2_block_replace(args);
861 * Lookup the entry in the directory. Get buffer and entry index.
862 * This will always succeed since the caller has already done a lookup.
864 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
870 btp = xfs_dir2_block_tail_p(mp, hdr);
871 blp = xfs_dir2_block_leaf_p(btp);
873 * Point to the data entry we need to change.
875 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
876 xfs_dir2_dataptr_to_off(args->geo,
877 be32_to_cpu(blp[ent].address)));
878 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
880 * Change the inode number to the new value.
882 dep->inumber = cpu_to_be64(args->inumber);
883 dp->d_ops->data_put_ftype(dep, args->filetype);
884 xfs_dir2_data_log_entry(args->trans, dp, bp, dep);
885 xfs_dir3_data_check(dp, bp);
890 * Qsort comparison routine for the block leaf entries.
892 static int /* sort order */
894 const void *a, /* first leaf entry */
895 const void *b) /* second leaf entry */
897 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
898 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
902 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
903 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
907 * Convert a V2 leaf directory to a V2 block directory if possible.
910 xfs_dir2_leaf_to_block(
911 xfs_da_args_t *args, /* operation arguments */
912 struct xfs_buf *lbp, /* leaf buffer */
913 struct xfs_buf *dbp) /* data buffer */
915 __be16 *bestsp; /* leaf bests table */
916 xfs_dir2_data_hdr_t *hdr; /* block header */
917 xfs_dir2_block_tail_t *btp; /* block tail */
918 xfs_inode_t *dp; /* incore directory inode */
919 xfs_dir2_data_unused_t *dup; /* unused data entry */
920 int error; /* error return value */
921 int from; /* leaf from index */
922 xfs_dir2_leaf_t *leaf; /* leaf structure */
923 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
924 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
925 xfs_mount_t *mp; /* file system mount point */
926 int needlog; /* need to log data header */
927 int needscan; /* need to scan for bestfree */
928 xfs_dir2_sf_hdr_t sfh; /* shortform header */
929 int size; /* bytes used */
930 __be16 *tagp; /* end of entry (tag) */
931 int to; /* block/leaf to index */
932 xfs_trans_t *tp; /* transaction pointer */
933 struct xfs_dir2_leaf_entry *ents;
934 struct xfs_dir3_icleaf_hdr leafhdr;
936 trace_xfs_dir2_leaf_to_block(args);
942 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
943 ents = dp->d_ops->leaf_ents_p(leaf);
944 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
946 ASSERT(leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
947 leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
949 * If there are data blocks other than the first one, take this
950 * opportunity to remove trailing empty data blocks that may have
951 * been left behind during no-space-reservation operations.
952 * These will show up in the leaf bests table.
954 while (dp->i_d.di_size > mp->m_dirblksize) {
957 hdrsz = dp->d_ops->data_entry_offset;
958 bestsp = xfs_dir2_leaf_bests_p(ltp);
959 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
960 mp->m_dirblksize - hdrsz) {
962 xfs_dir2_leaf_trim_data(args, lbp,
963 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
969 * Read the data block if we don't already have it, give up if it fails.
972 error = xfs_dir3_data_read(tp, dp, mp->m_dirdatablk, -1, &dbp);
977 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
978 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
981 * Size of the "leaf" area in the block.
983 size = (uint)sizeof(xfs_dir2_block_tail_t) +
984 (uint)sizeof(*lep) * (leafhdr.count - leafhdr.stale);
986 * Look at the last data entry.
988 tagp = (__be16 *)((char *)hdr + mp->m_dirblksize) - 1;
989 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
991 * If it's not free or is too short we can't do it.
993 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
994 be16_to_cpu(dup->length) < size)
998 * Start converting it to block form.
1000 xfs_dir3_block_init(mp, tp, dbp, dp);
1005 * Use up the space at the end of the block (blp/btp).
1007 xfs_dir2_data_use_free(tp, dp, dbp, dup, mp->m_dirblksize - size, size,
1008 &needlog, &needscan);
1010 * Initialize the block tail.
1012 btp = xfs_dir2_block_tail_p(mp, hdr);
1013 btp->count = cpu_to_be32(leafhdr.count - leafhdr.stale);
1015 xfs_dir2_block_log_tail(tp, dbp);
1017 * Initialize the block leaf area. We compact out stale entries.
1019 lep = xfs_dir2_block_leaf_p(btp);
1020 for (from = to = 0; from < leafhdr.count; from++) {
1021 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
1023 lep[to++] = ents[from];
1025 ASSERT(to == be32_to_cpu(btp->count));
1026 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
1028 * Scan the bestfree if we need it and log the data block header.
1031 xfs_dir2_data_freescan(dp, hdr, &needlog);
1033 xfs_dir2_data_log_header(tp, dp, dbp);
1035 * Pitch the old leaf block.
1037 error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
1042 * Now see if the resulting block can be shrunken to shortform.
1044 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1045 if (size > XFS_IFORK_DSIZE(dp))
1048 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1052 * Convert the shortform directory to block form.
1055 xfs_dir2_sf_to_block(
1056 xfs_da_args_t *args) /* operation arguments */
1058 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1059 xfs_dir2_data_hdr_t *hdr; /* block header */
1060 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1061 struct xfs_buf *bp; /* block buffer */
1062 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1063 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1064 xfs_inode_t *dp; /* incore directory inode */
1065 int dummy; /* trash */
1066 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1067 int endoffset; /* end of data objects */
1068 int error; /* error return value */
1070 xfs_mount_t *mp; /* filesystem mount point */
1071 int needlog; /* need to log block header */
1072 int needscan; /* need to scan block freespc */
1073 int newoffset; /* offset from current entry */
1074 int offset; /* target block offset */
1075 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1076 xfs_dir2_sf_hdr_t *oldsfp; /* old shortform header */
1077 xfs_dir2_sf_hdr_t *sfp; /* shortform header */
1078 __be16 *tagp; /* end of data entry */
1079 xfs_trans_t *tp; /* transaction pointer */
1080 struct xfs_name name;
1081 struct xfs_ifork *ifp;
1083 trace_xfs_dir2_sf_to_block(args);
1088 ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
1089 ASSERT(ifp->if_flags & XFS_IFINLINE);
1091 * Bomb out if the shortform directory is way too short.
1093 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1094 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1095 return XFS_ERROR(EIO);
1098 oldsfp = (xfs_dir2_sf_hdr_t *)ifp->if_u1.if_data;
1100 ASSERT(ifp->if_bytes == dp->i_d.di_size);
1101 ASSERT(ifp->if_u1.if_data != NULL);
1102 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1103 ASSERT(dp->i_d.di_nextents == 0);
1106 * Copy the directory into a temporary buffer.
1107 * Then pitch the incore inode data so we can make extents.
1109 sfp = kmem_alloc(ifp->if_bytes, KM_SLEEP);
1110 memcpy(sfp, oldsfp, ifp->if_bytes);
1112 xfs_idata_realloc(dp, -ifp->if_bytes, XFS_DATA_FORK);
1113 xfs_bmap_local_to_extents_empty(dp, XFS_DATA_FORK);
1114 dp->i_d.di_size = 0;
1117 * Add block 0 to the inode.
1119 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1125 * Initialize the data block, then convert it to block format.
1127 error = xfs_dir3_data_init(args, blkno, &bp);
1132 xfs_dir3_block_init(mp, tp, bp, dp);
1136 * Compute size of block "tail" area.
1138 i = (uint)sizeof(*btp) +
1139 (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1141 * The whole thing is initialized to free by the init routine.
1142 * Say we're using the leaf and tail area.
1144 dup = dp->d_ops->data_unused_p(hdr);
1145 needlog = needscan = 0;
1146 xfs_dir2_data_use_free(tp, dp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1148 ASSERT(needscan == 0);
1152 btp = xfs_dir2_block_tail_p(mp, hdr);
1153 btp->count = cpu_to_be32(sfp->count + 2); /* ., .. */
1155 blp = xfs_dir2_block_leaf_p(btp);
1156 endoffset = (uint)((char *)blp - (char *)hdr);
1158 * Remove the freespace, we'll manage it.
1160 xfs_dir2_data_use_free(tp, dp, bp, dup,
1161 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1162 be16_to_cpu(dup->length), &needlog, &needscan);
1164 * Create entry for .
1166 dep = dp->d_ops->data_dot_entry_p(hdr);
1167 dep->inumber = cpu_to_be64(dp->i_ino);
1170 dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1171 tagp = dp->d_ops->data_entry_tag_p(dep);
1172 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1173 xfs_dir2_data_log_entry(tp, dp, bp, dep);
1174 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1175 blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1176 (char *)dep - (char *)hdr));
1178 * Create entry for ..
1180 dep = dp->d_ops->data_dotdot_entry_p(hdr);
1181 dep->inumber = cpu_to_be64(dp->d_ops->sf_get_parent_ino(sfp));
1183 dep->name[0] = dep->name[1] = '.';
1184 dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1185 tagp = dp->d_ops->data_entry_tag_p(dep);
1186 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1187 xfs_dir2_data_log_entry(tp, dp, bp, dep);
1188 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1189 blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1190 (char *)dep - (char *)hdr));
1191 offset = dp->d_ops->data_first_offset;
1193 * Loop over existing entries, stuff them in.
1199 sfep = xfs_dir2_sf_firstentry(sfp);
1201 * Need to preserve the existing offset values in the sf directory.
1202 * Insert holes (unused entries) where necessary.
1204 while (offset < endoffset) {
1206 * sfep is null when we reach the end of the list.
1209 newoffset = endoffset;
1211 newoffset = xfs_dir2_sf_get_offset(sfep);
1213 * There should be a hole here, make one.
1215 if (offset < newoffset) {
1216 dup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1217 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1218 dup->length = cpu_to_be16(newoffset - offset);
1219 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1220 ((char *)dup - (char *)hdr));
1221 xfs_dir2_data_log_unused(tp, bp, dup);
1222 xfs_dir2_data_freeinsert(hdr,
1223 dp->d_ops->data_bestfree_p(hdr),
1225 offset += be16_to_cpu(dup->length);
1229 * Copy a real entry.
1231 dep = (xfs_dir2_data_entry_t *)((char *)hdr + newoffset);
1232 dep->inumber = cpu_to_be64(dp->d_ops->sf_get_ino(sfp, sfep));
1233 dep->namelen = sfep->namelen;
1234 dp->d_ops->data_put_ftype(dep, dp->d_ops->sf_get_ftype(sfep));
1235 memcpy(dep->name, sfep->name, dep->namelen);
1236 tagp = dp->d_ops->data_entry_tag_p(dep);
1237 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1238 xfs_dir2_data_log_entry(tp, dp, bp, dep);
1239 name.name = sfep->name;
1240 name.len = sfep->namelen;
1241 blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1243 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1244 (char *)dep - (char *)hdr));
1245 offset = (int)((char *)(tagp + 1) - (char *)hdr);
1246 if (++i == sfp->count)
1249 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
1251 /* Done with the temporary buffer */
1254 * Sort the leaf entries by hash value.
1256 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1258 * Log the leaf entry area and tail.
1259 * Already logged the header in data_init, ignore needlog.
1261 ASSERT(needscan == 0);
1262 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1263 xfs_dir2_block_log_tail(tp, bp);
1264 xfs_dir3_data_check(dp, bp);