1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_trans.h"
17 #include "xfs_buf_item.h"
19 #include "xfs_dir2_priv.h"
20 #include "xfs_error.h"
21 #include "xfs_trace.h"
25 * Local function prototypes.
27 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, struct xfs_buf *bp,
29 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, struct xfs_buf *bp);
30 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, struct xfs_buf **bpp,
32 static int xfs_dir2_block_sort(const void *a, const void *b);
34 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
37 * One-time startup routine called from xfs_init().
42 xfs_dir_hash_dot = xfs_da_hashname((unsigned char *)".", 1);
43 xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
47 xfs_dir3_block_verify(
50 struct xfs_mount *mp = bp->b_mount;
51 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
53 if (!xfs_verify_magic(bp, hdr3->magic))
54 return __this_address;
56 if (xfs_sb_version_hascrc(&mp->m_sb)) {
57 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
58 return __this_address;
59 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
60 return __this_address;
61 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
62 return __this_address;
64 return __xfs_dir3_data_check(NULL, bp);
68 xfs_dir3_block_read_verify(
71 struct xfs_mount *mp = bp->b_mount;
74 if (xfs_sb_version_hascrc(&mp->m_sb) &&
75 !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
76 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
78 fa = xfs_dir3_block_verify(bp);
80 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
85 xfs_dir3_block_write_verify(
88 struct xfs_mount *mp = bp->b_mount;
89 struct xfs_buf_log_item *bip = bp->b_log_item;
90 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
93 fa = xfs_dir3_block_verify(bp);
95 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
99 if (!xfs_sb_version_hascrc(&mp->m_sb))
103 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
105 xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
108 const struct xfs_buf_ops xfs_dir3_block_buf_ops = {
109 .name = "xfs_dir3_block",
110 .magic = { cpu_to_be32(XFS_DIR2_BLOCK_MAGIC),
111 cpu_to_be32(XFS_DIR3_BLOCK_MAGIC) },
112 .verify_read = xfs_dir3_block_read_verify,
113 .verify_write = xfs_dir3_block_write_verify,
114 .verify_struct = xfs_dir3_block_verify,
119 struct xfs_trans *tp,
120 struct xfs_inode *dp,
121 struct xfs_buf **bpp)
123 struct xfs_mount *mp = dp->i_mount;
126 err = xfs_da_read_buf(tp, dp, mp->m_dir_geo->datablk, -1, bpp,
127 XFS_DATA_FORK, &xfs_dir3_block_buf_ops);
128 if (!err && tp && *bpp)
129 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_BLOCK_BUF);
135 struct xfs_mount *mp,
136 struct xfs_trans *tp,
138 struct xfs_inode *dp)
140 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
142 bp->b_ops = &xfs_dir3_block_buf_ops;
143 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_BLOCK_BUF);
145 if (xfs_sb_version_hascrc(&mp->m_sb)) {
146 memset(hdr3, 0, sizeof(*hdr3));
147 hdr3->magic = cpu_to_be32(XFS_DIR3_BLOCK_MAGIC);
148 hdr3->blkno = cpu_to_be64(bp->b_bn);
149 hdr3->owner = cpu_to_be64(dp->i_ino);
150 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
154 hdr3->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
158 xfs_dir2_block_need_space(
159 struct xfs_inode *dp,
160 struct xfs_dir2_data_hdr *hdr,
161 struct xfs_dir2_block_tail *btp,
162 struct xfs_dir2_leaf_entry *blp,
164 struct xfs_dir2_data_unused **dupp,
165 struct xfs_dir2_data_unused **enddupp,
169 struct xfs_dir2_data_free *bf;
171 struct xfs_dir2_data_unused *dup = NULL;
172 struct xfs_dir2_data_unused *enddup = NULL;
175 bf = dp->d_ops->data_bestfree_p(hdr);
178 * If there are stale entries we'll use one for the leaf.
181 if (be16_to_cpu(bf[0].length) >= len) {
183 * The biggest entry enough to avoid compaction.
185 dup = (xfs_dir2_data_unused_t *)
186 ((char *)hdr + be16_to_cpu(bf[0].offset));
191 * Will need to compact to make this work.
192 * Tag just before the first leaf entry.
195 tagp = (__be16 *)blp - 1;
197 /* Data object just before the first leaf entry. */
198 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
201 * If it's not free then the data will go where the
202 * leaf data starts now, if it works at all.
204 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
205 if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
206 (uint)sizeof(*blp) < len)
208 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
211 dup = (xfs_dir2_data_unused_t *)blp;
216 * no stale entries, so just use free space.
217 * Tag just before the first leaf entry.
219 tagp = (__be16 *)blp - 1;
221 /* Data object just before the first leaf entry. */
222 enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
225 * If it's not free then can't do this add without cleaning up:
226 * the space before the first leaf entry needs to be free so it
227 * can be expanded to hold the pointer to the new entry.
229 if (be16_to_cpu(enddup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
231 * Check out the biggest freespace and see if it's the same one.
233 dup = (xfs_dir2_data_unused_t *)
234 ((char *)hdr + be16_to_cpu(bf[0].offset));
237 * Not the same free entry, just check its length.
239 if (be16_to_cpu(dup->length) < len)
245 * It is the biggest freespace, can it hold the leaf too?
247 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
249 * Yes, use the second-largest entry instead if it works.
251 if (be16_to_cpu(bf[1].length) >= len)
252 dup = (xfs_dir2_data_unused_t *)
253 ((char *)hdr + be16_to_cpu(bf[1].offset));
265 * compact the leaf entries.
266 * Leave the highest-numbered stale entry stale.
267 * XXX should be the one closest to mid but mid is not yet computed.
270 xfs_dir2_block_compact(
271 struct xfs_da_args *args,
273 struct xfs_dir2_data_hdr *hdr,
274 struct xfs_dir2_block_tail *btp,
275 struct xfs_dir2_leaf_entry *blp,
280 int fromidx; /* source leaf index */
281 int toidx; /* target leaf index */
283 int highstale; /* high stale index */
285 fromidx = toidx = be32_to_cpu(btp->count) - 1;
286 highstale = *lfloghigh = -1;
287 for (; fromidx >= 0; fromidx--) {
288 if (blp[fromidx].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
292 if (*lfloghigh == -1)
298 blp[toidx] = blp[fromidx];
301 *lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
302 *lfloghigh -= be32_to_cpu(btp->stale) - 1;
303 be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
304 xfs_dir2_data_make_free(args, bp,
305 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
306 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
308 btp->stale = cpu_to_be32(1);
310 * If we now need to rebuild the bestfree map, do so.
311 * This needs to happen before the next call to use_free.
314 xfs_dir2_data_freescan(args->dp, hdr, needlog);
318 * Add an entry to a block directory.
321 xfs_dir2_block_addname(
322 xfs_da_args_t *args) /* directory op arguments */
324 xfs_dir2_data_hdr_t *hdr; /* block header */
325 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
326 struct xfs_buf *bp; /* buffer for block */
327 xfs_dir2_block_tail_t *btp; /* block tail */
328 int compact; /* need to compact leaf ents */
329 xfs_dir2_data_entry_t *dep; /* block data entry */
330 xfs_inode_t *dp; /* directory inode */
331 xfs_dir2_data_unused_t *dup; /* block unused entry */
332 int error; /* error return value */
333 xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
334 xfs_dahash_t hash; /* hash value of found entry */
335 int high; /* high index for binary srch */
336 int highstale; /* high stale index */
337 int lfloghigh=0; /* last final leaf to log */
338 int lfloglow=0; /* first final leaf to log */
339 int len; /* length of the new entry */
340 int low; /* low index for binary srch */
341 int lowstale; /* low stale index */
342 int mid=0; /* midpoint for binary srch */
343 int needlog; /* need to log header */
344 int needscan; /* need to rescan freespace */
345 __be16 *tagp; /* pointer to tag value */
346 xfs_trans_t *tp; /* transaction structure */
348 trace_xfs_dir2_block_addname(args);
353 /* Read the (one and only) directory block into bp. */
354 error = xfs_dir3_block_read(tp, dp, &bp);
358 len = dp->d_ops->data_entsize(args->namelen);
361 * Set up pointers to parts of the block.
364 btp = xfs_dir2_block_tail_p(args->geo, hdr);
365 blp = xfs_dir2_block_leaf_p(btp);
368 * Find out if we can reuse stale entries or whether we need extra
369 * space for entry and new leaf.
371 xfs_dir2_block_need_space(dp, hdr, btp, blp, &tagp, &dup,
372 &enddup, &compact, len);
375 * Done everything we need for a space check now.
377 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
378 xfs_trans_brelse(tp, bp);
385 * If we don't have space for the new entry & leaf ...
388 /* Don't have a space reservation: return no-space. */
389 if (args->total == 0)
392 * Convert to the next larger format.
393 * Then add the new entry in that format.
395 error = xfs_dir2_block_to_leaf(args, bp);
398 return xfs_dir2_leaf_addname(args);
401 needlog = needscan = 0;
404 * If need to compact the leaf entries, do it now.
407 xfs_dir2_block_compact(args, bp, hdr, btp, blp, &needlog,
408 &lfloghigh, &lfloglow);
409 /* recalculate blp post-compaction */
410 blp = xfs_dir2_block_leaf_p(btp);
411 } else if (btp->stale) {
413 * Set leaf logging boundaries to impossible state.
414 * For the no-stale case they're set explicitly.
416 lfloglow = be32_to_cpu(btp->count);
421 * Find the slot that's first lower than our hash value, -1 if none.
423 for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
424 mid = (low + high) >> 1;
425 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
427 if (hash < args->hashval)
432 while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
436 * No stale entries, will use enddup space to hold new leaf.
439 xfs_dir2_data_aoff_t aoff;
442 * Mark the space needed for the new leaf entry, now in use.
444 aoff = (xfs_dir2_data_aoff_t)((char *)enddup - (char *)hdr +
445 be16_to_cpu(enddup->length) - sizeof(*blp));
446 error = xfs_dir2_data_use_free(args, bp, enddup, aoff,
447 (xfs_dir2_data_aoff_t)sizeof(*blp), &needlog,
453 * Update the tail (entry count).
455 be32_add_cpu(&btp->count, 1);
457 * If we now need to rebuild the bestfree map, do so.
458 * This needs to happen before the next call to use_free.
461 xfs_dir2_data_freescan(dp, hdr, &needlog);
465 * Adjust pointer to the first leaf entry, we're about to move
466 * the table up one to open up space for the new leaf entry.
467 * Then adjust our index to match.
472 memmove(blp, &blp[1], mid * sizeof(*blp));
477 * Use a stale leaf for our new entry.
482 blp[lowstale].address !=
483 cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
486 for (highstale = mid + 1;
487 highstale < be32_to_cpu(btp->count) &&
488 blp[highstale].address !=
489 cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
490 (lowstale < 0 || mid - lowstale > highstale - mid);
494 * Move entries toward the low-numbered stale entry.
497 (highstale == be32_to_cpu(btp->count) ||
498 mid - lowstale <= highstale - mid)) {
500 memmove(&blp[lowstale], &blp[lowstale + 1],
501 (mid - lowstale) * sizeof(*blp));
502 lfloglow = min(lowstale, lfloglow);
503 lfloghigh = max(mid, lfloghigh);
506 * Move entries toward the high-numbered stale entry.
509 ASSERT(highstale < be32_to_cpu(btp->count));
512 memmove(&blp[mid + 1], &blp[mid],
513 (highstale - mid) * sizeof(*blp));
514 lfloglow = min(mid, lfloglow);
515 lfloghigh = max(highstale, lfloghigh);
517 be32_add_cpu(&btp->stale, -1);
520 * Point to the new data entry.
522 dep = (xfs_dir2_data_entry_t *)dup;
524 * Fill in the leaf entry.
526 blp[mid].hashval = cpu_to_be32(args->hashval);
527 blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
528 (char *)dep - (char *)hdr));
529 xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
531 * Mark space for the data entry used.
533 error = xfs_dir2_data_use_free(args, bp, dup,
534 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
535 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
539 * Create the new data entry.
541 dep->inumber = cpu_to_be64(args->inumber);
542 dep->namelen = args->namelen;
543 memcpy(dep->name, args->name, args->namelen);
544 dp->d_ops->data_put_ftype(dep, args->filetype);
545 tagp = dp->d_ops->data_entry_tag_p(dep);
546 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
548 * Clean up the bestfree array and log the header, tail, and entry.
551 xfs_dir2_data_freescan(dp, hdr, &needlog);
553 xfs_dir2_data_log_header(args, bp);
554 xfs_dir2_block_log_tail(tp, bp);
555 xfs_dir2_data_log_entry(args, bp, dep);
556 xfs_dir3_data_check(dp, bp);
561 * Log leaf entries from the block.
564 xfs_dir2_block_log_leaf(
565 xfs_trans_t *tp, /* transaction structure */
566 struct xfs_buf *bp, /* block buffer */
567 int first, /* index of first logged leaf */
568 int last) /* index of last logged leaf */
570 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
571 xfs_dir2_leaf_entry_t *blp;
572 xfs_dir2_block_tail_t *btp;
574 btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
575 blp = xfs_dir2_block_leaf_p(btp);
576 xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
577 (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
581 * Log the block tail.
584 xfs_dir2_block_log_tail(
585 xfs_trans_t *tp, /* transaction structure */
586 struct xfs_buf *bp) /* block buffer */
588 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
589 xfs_dir2_block_tail_t *btp;
591 btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
592 xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
593 (uint)((char *)(btp + 1) - (char *)hdr - 1));
597 * Look up an entry in the block. This is the external routine,
598 * xfs_dir2_block_lookup_int does the real work.
601 xfs_dir2_block_lookup(
602 xfs_da_args_t *args) /* dir lookup arguments */
604 xfs_dir2_data_hdr_t *hdr; /* block header */
605 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
606 struct xfs_buf *bp; /* block buffer */
607 xfs_dir2_block_tail_t *btp; /* block tail */
608 xfs_dir2_data_entry_t *dep; /* block data entry */
609 xfs_inode_t *dp; /* incore inode */
610 int ent; /* entry index */
611 int error; /* error return value */
613 trace_xfs_dir2_block_lookup(args);
616 * Get the buffer, look up the entry.
617 * If not found (ENOENT) then return, have no buffer.
619 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
623 xfs_dir3_data_check(dp, bp);
624 btp = xfs_dir2_block_tail_p(args->geo, hdr);
625 blp = xfs_dir2_block_leaf_p(btp);
627 * Get the offset from the leaf entry, to point to the data.
629 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
630 xfs_dir2_dataptr_to_off(args->geo,
631 be32_to_cpu(blp[ent].address)));
633 * Fill in inode number, CI name if appropriate, release the block.
635 args->inumber = be64_to_cpu(dep->inumber);
636 args->filetype = dp->d_ops->data_get_ftype(dep);
637 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
638 xfs_trans_brelse(args->trans, bp);
643 * Internal block lookup routine.
645 static int /* error */
646 xfs_dir2_block_lookup_int(
647 xfs_da_args_t *args, /* dir lookup arguments */
648 struct xfs_buf **bpp, /* returned block buffer */
649 int *entno) /* returned entry number */
651 xfs_dir2_dataptr_t addr; /* data entry address */
652 xfs_dir2_data_hdr_t *hdr; /* block header */
653 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
654 struct xfs_buf *bp; /* block buffer */
655 xfs_dir2_block_tail_t *btp; /* block tail */
656 xfs_dir2_data_entry_t *dep; /* block data entry */
657 xfs_inode_t *dp; /* incore inode */
658 int error; /* error return value */
659 xfs_dahash_t hash; /* found hash value */
660 int high; /* binary search high index */
661 int low; /* binary search low index */
662 int mid; /* binary search current idx */
663 xfs_mount_t *mp; /* filesystem mount point */
664 xfs_trans_t *tp; /* transaction pointer */
665 enum xfs_dacmp cmp; /* comparison result */
671 error = xfs_dir3_block_read(tp, dp, &bp);
676 xfs_dir3_data_check(dp, bp);
677 btp = xfs_dir2_block_tail_p(args->geo, hdr);
678 blp = xfs_dir2_block_leaf_p(btp);
680 * Loop doing a binary search for our hash value.
681 * Find our entry, ENOENT if it's not there.
683 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
685 mid = (low + high) >> 1;
686 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
688 if (hash < args->hashval)
693 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
694 xfs_trans_brelse(tp, bp);
699 * Back up to the first one with the right hash value.
701 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
705 * Now loop forward through all the entries with the
706 * right hash value looking for our name.
709 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
712 * Get pointer to the entry from the leaf.
714 dep = (xfs_dir2_data_entry_t *)
715 ((char *)hdr + xfs_dir2_dataptr_to_off(args->geo, addr));
717 * Compare name and if it's an exact match, return the index
718 * and buffer. If it's the first case-insensitive match, store
719 * the index and buffer and continue looking for an exact match.
721 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
722 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
723 args->cmpresult = cmp;
726 if (cmp == XFS_CMP_EXACT)
729 } while (++mid < be32_to_cpu(btp->count) &&
730 be32_to_cpu(blp[mid].hashval) == hash);
732 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
734 * Here, we can only be doing a lookup (not a rename or replace).
735 * If a case-insensitive match was found earlier, return success.
737 if (args->cmpresult == XFS_CMP_CASE)
740 * No match, release the buffer and return ENOENT.
742 xfs_trans_brelse(tp, bp);
747 * Remove an entry from a block format directory.
748 * If that makes the block small enough to fit in shortform, transform it.
751 xfs_dir2_block_removename(
752 xfs_da_args_t *args) /* directory operation args */
754 xfs_dir2_data_hdr_t *hdr; /* block header */
755 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
756 struct xfs_buf *bp; /* block buffer */
757 xfs_dir2_block_tail_t *btp; /* block tail */
758 xfs_dir2_data_entry_t *dep; /* block data entry */
759 xfs_inode_t *dp; /* incore inode */
760 int ent; /* block leaf entry index */
761 int error; /* error return value */
762 int needlog; /* need to log block header */
763 int needscan; /* need to fixup bestfree */
764 xfs_dir2_sf_hdr_t sfh; /* shortform header */
765 int size; /* shortform size */
766 xfs_trans_t *tp; /* transaction pointer */
768 trace_xfs_dir2_block_removename(args);
771 * Look up the entry in the block. Gets the buffer and entry index.
772 * It will always be there, the vnodeops level does a lookup first.
774 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
780 btp = xfs_dir2_block_tail_p(args->geo, hdr);
781 blp = xfs_dir2_block_leaf_p(btp);
783 * Point to the data entry using the leaf entry.
785 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
786 xfs_dir2_dataptr_to_off(args->geo,
787 be32_to_cpu(blp[ent].address)));
789 * Mark the data entry's space free.
791 needlog = needscan = 0;
792 xfs_dir2_data_make_free(args, bp,
793 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
794 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
796 * Fix up the block tail.
798 be32_add_cpu(&btp->stale, 1);
799 xfs_dir2_block_log_tail(tp, bp);
801 * Remove the leaf entry by marking it stale.
803 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
804 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
806 * Fix up bestfree, log the header if necessary.
809 xfs_dir2_data_freescan(dp, hdr, &needlog);
811 xfs_dir2_data_log_header(args, bp);
812 xfs_dir3_data_check(dp, bp);
814 * See if the size as a shortform is good enough.
816 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
817 if (size > XFS_IFORK_DSIZE(dp))
821 * If it works, do the conversion.
823 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
827 * Replace an entry in a V2 block directory.
828 * Change the inode number to the new value.
831 xfs_dir2_block_replace(
832 xfs_da_args_t *args) /* directory operation args */
834 xfs_dir2_data_hdr_t *hdr; /* block header */
835 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
836 struct xfs_buf *bp; /* block buffer */
837 xfs_dir2_block_tail_t *btp; /* block tail */
838 xfs_dir2_data_entry_t *dep; /* block data entry */
839 xfs_inode_t *dp; /* incore inode */
840 int ent; /* leaf entry index */
841 int error; /* error return value */
843 trace_xfs_dir2_block_replace(args);
846 * Lookup the entry in the directory. Get buffer and entry index.
847 * This will always succeed since the caller has already done a lookup.
849 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
854 btp = xfs_dir2_block_tail_p(args->geo, hdr);
855 blp = xfs_dir2_block_leaf_p(btp);
857 * Point to the data entry we need to change.
859 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
860 xfs_dir2_dataptr_to_off(args->geo,
861 be32_to_cpu(blp[ent].address)));
862 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
864 * Change the inode number to the new value.
866 dep->inumber = cpu_to_be64(args->inumber);
867 dp->d_ops->data_put_ftype(dep, args->filetype);
868 xfs_dir2_data_log_entry(args, bp, dep);
869 xfs_dir3_data_check(dp, bp);
874 * Qsort comparison routine for the block leaf entries.
876 static int /* sort order */
878 const void *a, /* first leaf entry */
879 const void *b) /* second leaf entry */
881 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
882 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
886 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
887 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
891 * Convert a V2 leaf directory to a V2 block directory if possible.
894 xfs_dir2_leaf_to_block(
895 xfs_da_args_t *args, /* operation arguments */
896 struct xfs_buf *lbp, /* leaf buffer */
897 struct xfs_buf *dbp) /* data buffer */
899 __be16 *bestsp; /* leaf bests table */
900 xfs_dir2_data_hdr_t *hdr; /* block header */
901 xfs_dir2_block_tail_t *btp; /* block tail */
902 xfs_inode_t *dp; /* incore directory inode */
903 xfs_dir2_data_unused_t *dup; /* unused data entry */
904 int error; /* error return value */
905 int from; /* leaf from index */
906 xfs_dir2_leaf_t *leaf; /* leaf structure */
907 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
908 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
909 xfs_mount_t *mp; /* file system mount point */
910 int needlog; /* need to log data header */
911 int needscan; /* need to scan for bestfree */
912 xfs_dir2_sf_hdr_t sfh; /* shortform header */
913 int size; /* bytes used */
914 __be16 *tagp; /* end of entry (tag) */
915 int to; /* block/leaf to index */
916 xfs_trans_t *tp; /* transaction pointer */
917 struct xfs_dir2_leaf_entry *ents;
918 struct xfs_dir3_icleaf_hdr leafhdr;
920 trace_xfs_dir2_leaf_to_block(args);
926 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
927 ents = dp->d_ops->leaf_ents_p(leaf);
928 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
930 ASSERT(leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
931 leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
933 * If there are data blocks other than the first one, take this
934 * opportunity to remove trailing empty data blocks that may have
935 * been left behind during no-space-reservation operations.
936 * These will show up in the leaf bests table.
938 while (dp->i_d.di_size > args->geo->blksize) {
941 hdrsz = dp->d_ops->data_entry_offset;
942 bestsp = xfs_dir2_leaf_bests_p(ltp);
943 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
944 args->geo->blksize - hdrsz) {
946 xfs_dir2_leaf_trim_data(args, lbp,
947 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
953 * Read the data block if we don't already have it, give up if it fails.
956 error = xfs_dir3_data_read(tp, dp, args->geo->datablk, -1, &dbp);
961 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
962 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
965 * Size of the "leaf" area in the block.
967 size = (uint)sizeof(xfs_dir2_block_tail_t) +
968 (uint)sizeof(*lep) * (leafhdr.count - leafhdr.stale);
970 * Look at the last data entry.
972 tagp = (__be16 *)((char *)hdr + args->geo->blksize) - 1;
973 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
975 * If it's not free or is too short we can't do it.
977 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
978 be16_to_cpu(dup->length) < size)
982 * Start converting it to block form.
984 xfs_dir3_block_init(mp, tp, dbp, dp);
989 * Use up the space at the end of the block (blp/btp).
991 error = xfs_dir2_data_use_free(args, dbp, dup,
992 args->geo->blksize - size, size, &needlog, &needscan);
996 * Initialize the block tail.
998 btp = xfs_dir2_block_tail_p(args->geo, hdr);
999 btp->count = cpu_to_be32(leafhdr.count - leafhdr.stale);
1001 xfs_dir2_block_log_tail(tp, dbp);
1003 * Initialize the block leaf area. We compact out stale entries.
1005 lep = xfs_dir2_block_leaf_p(btp);
1006 for (from = to = 0; from < leafhdr.count; from++) {
1007 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
1009 lep[to++] = ents[from];
1011 ASSERT(to == be32_to_cpu(btp->count));
1012 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
1014 * Scan the bestfree if we need it and log the data block header.
1017 xfs_dir2_data_freescan(dp, hdr, &needlog);
1019 xfs_dir2_data_log_header(args, dbp);
1021 * Pitch the old leaf block.
1023 error = xfs_da_shrink_inode(args, args->geo->leafblk, lbp);
1028 * Now see if the resulting block can be shrunken to shortform.
1030 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1031 if (size > XFS_IFORK_DSIZE(dp))
1034 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1038 * Convert the shortform directory to block form.
1041 xfs_dir2_sf_to_block(
1042 xfs_da_args_t *args) /* operation arguments */
1044 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1045 xfs_dir2_data_hdr_t *hdr; /* block header */
1046 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1047 struct xfs_buf *bp; /* block buffer */
1048 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1049 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1050 xfs_inode_t *dp; /* incore directory inode */
1051 int dummy; /* trash */
1052 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1053 int endoffset; /* end of data objects */
1054 int error; /* error return value */
1056 xfs_mount_t *mp; /* filesystem mount point */
1057 int needlog; /* need to log block header */
1058 int needscan; /* need to scan block freespc */
1059 int newoffset; /* offset from current entry */
1060 int offset; /* target block offset */
1061 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1062 xfs_dir2_sf_hdr_t *oldsfp; /* old shortform header */
1063 xfs_dir2_sf_hdr_t *sfp; /* shortform header */
1064 __be16 *tagp; /* end of data entry */
1065 xfs_trans_t *tp; /* transaction pointer */
1066 struct xfs_name name;
1067 struct xfs_ifork *ifp;
1069 trace_xfs_dir2_sf_to_block(args);
1074 ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
1075 ASSERT(ifp->if_flags & XFS_IFINLINE);
1077 * Bomb out if the shortform directory is way too short.
1079 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1080 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1084 oldsfp = (xfs_dir2_sf_hdr_t *)ifp->if_u1.if_data;
1086 ASSERT(ifp->if_bytes == dp->i_d.di_size);
1087 ASSERT(ifp->if_u1.if_data != NULL);
1088 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1089 ASSERT(dp->i_d.di_nextents == 0);
1092 * Copy the directory into a temporary buffer.
1093 * Then pitch the incore inode data so we can make extents.
1095 sfp = kmem_alloc(ifp->if_bytes, 0);
1096 memcpy(sfp, oldsfp, ifp->if_bytes);
1098 xfs_idata_realloc(dp, -ifp->if_bytes, XFS_DATA_FORK);
1099 xfs_bmap_local_to_extents_empty(dp, XFS_DATA_FORK);
1100 dp->i_d.di_size = 0;
1103 * Add block 0 to the inode.
1105 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1109 * Initialize the data block, then convert it to block format.
1111 error = xfs_dir3_data_init(args, blkno, &bp);
1114 xfs_dir3_block_init(mp, tp, bp, dp);
1118 * Compute size of block "tail" area.
1120 i = (uint)sizeof(*btp) +
1121 (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1123 * The whole thing is initialized to free by the init routine.
1124 * Say we're using the leaf and tail area.
1126 dup = dp->d_ops->data_unused_p(hdr);
1127 needlog = needscan = 0;
1128 error = xfs_dir2_data_use_free(args, bp, dup, args->geo->blksize - i,
1129 i, &needlog, &needscan);
1132 ASSERT(needscan == 0);
1136 btp = xfs_dir2_block_tail_p(args->geo, hdr);
1137 btp->count = cpu_to_be32(sfp->count + 2); /* ., .. */
1139 blp = xfs_dir2_block_leaf_p(btp);
1140 endoffset = (uint)((char *)blp - (char *)hdr);
1142 * Remove the freespace, we'll manage it.
1144 error = xfs_dir2_data_use_free(args, bp, dup,
1145 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1146 be16_to_cpu(dup->length), &needlog, &needscan);
1150 * Create entry for .
1152 dep = dp->d_ops->data_dot_entry_p(hdr);
1153 dep->inumber = cpu_to_be64(dp->i_ino);
1156 dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1157 tagp = dp->d_ops->data_entry_tag_p(dep);
1158 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1159 xfs_dir2_data_log_entry(args, bp, dep);
1160 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1161 blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1162 (char *)dep - (char *)hdr));
1164 * Create entry for ..
1166 dep = dp->d_ops->data_dotdot_entry_p(hdr);
1167 dep->inumber = cpu_to_be64(dp->d_ops->sf_get_parent_ino(sfp));
1169 dep->name[0] = dep->name[1] = '.';
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(args, bp, dep);
1174 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1175 blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1176 (char *)dep - (char *)hdr));
1177 offset = dp->d_ops->data_first_offset;
1179 * Loop over existing entries, stuff them in.
1185 sfep = xfs_dir2_sf_firstentry(sfp);
1187 * Need to preserve the existing offset values in the sf directory.
1188 * Insert holes (unused entries) where necessary.
1190 while (offset < endoffset) {
1192 * sfep is null when we reach the end of the list.
1195 newoffset = endoffset;
1197 newoffset = xfs_dir2_sf_get_offset(sfep);
1199 * There should be a hole here, make one.
1201 if (offset < newoffset) {
1202 dup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1203 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1204 dup->length = cpu_to_be16(newoffset - offset);
1205 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1206 ((char *)dup - (char *)hdr));
1207 xfs_dir2_data_log_unused(args, bp, dup);
1208 xfs_dir2_data_freeinsert(hdr,
1209 dp->d_ops->data_bestfree_p(hdr),
1211 offset += be16_to_cpu(dup->length);
1215 * Copy a real entry.
1217 dep = (xfs_dir2_data_entry_t *)((char *)hdr + newoffset);
1218 dep->inumber = cpu_to_be64(dp->d_ops->sf_get_ino(sfp, sfep));
1219 dep->namelen = sfep->namelen;
1220 dp->d_ops->data_put_ftype(dep, dp->d_ops->sf_get_ftype(sfep));
1221 memcpy(dep->name, sfep->name, dep->namelen);
1222 tagp = dp->d_ops->data_entry_tag_p(dep);
1223 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1224 xfs_dir2_data_log_entry(args, bp, dep);
1225 name.name = sfep->name;
1226 name.len = sfep->namelen;
1227 blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1229 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1230 (char *)dep - (char *)hdr));
1231 offset = (int)((char *)(tagp + 1) - (char *)hdr);
1232 if (++i == sfp->count)
1235 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
1237 /* Done with the temporary buffer */
1240 * Sort the leaf entries by hash value.
1242 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1244 * Log the leaf entry area and tail.
1245 * Already logged the header in data_init, ignore needlog.
1247 ASSERT(needscan == 0);
1248 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1249 xfs_dir2_block_log_tail(tp, bp);
1250 xfs_dir3_data_check(dp, bp);