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"
23 #include "xfs_health.h"
26 * Local function prototypes.
28 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, struct xfs_buf *bp,
30 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, struct xfs_buf *bp);
31 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, struct xfs_buf **bpp,
33 static int xfs_dir2_block_sort(const void *a, const void *b);
35 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
38 * One-time startup routine called from xfs_init().
43 xfs_dir_hash_dot = xfs_da_hashname((unsigned char *)".", 1);
44 xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
48 xfs_dir3_block_verify(
51 struct xfs_mount *mp = bp->b_mount;
52 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
54 if (!xfs_verify_magic(bp, hdr3->magic))
55 return __this_address;
57 if (xfs_has_crc(mp)) {
58 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
59 return __this_address;
60 if (be64_to_cpu(hdr3->blkno) != xfs_buf_daddr(bp))
61 return __this_address;
62 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
63 return __this_address;
65 return __xfs_dir3_data_check(NULL, bp);
69 xfs_dir3_block_read_verify(
72 struct xfs_mount *mp = bp->b_mount;
75 if (xfs_has_crc(mp) &&
76 !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
77 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
79 fa = xfs_dir3_block_verify(bp);
81 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
86 xfs_dir3_block_write_verify(
89 struct xfs_mount *mp = bp->b_mount;
90 struct xfs_buf_log_item *bip = bp->b_log_item;
91 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
94 fa = xfs_dir3_block_verify(bp);
96 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
100 if (!xfs_has_crc(mp))
104 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
106 xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
109 const struct xfs_buf_ops xfs_dir3_block_buf_ops = {
110 .name = "xfs_dir3_block",
111 .magic = { cpu_to_be32(XFS_DIR2_BLOCK_MAGIC),
112 cpu_to_be32(XFS_DIR3_BLOCK_MAGIC) },
113 .verify_read = xfs_dir3_block_read_verify,
114 .verify_write = xfs_dir3_block_write_verify,
115 .verify_struct = xfs_dir3_block_verify,
118 static xfs_failaddr_t
119 xfs_dir3_block_header_check(
120 struct xfs_inode *dp,
123 struct xfs_mount *mp = dp->i_mount;
125 if (xfs_has_crc(mp)) {
126 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
128 if (be64_to_cpu(hdr3->owner) != dp->i_ino)
129 return __this_address;
137 struct xfs_trans *tp,
138 struct xfs_inode *dp,
139 struct xfs_buf **bpp)
141 struct xfs_mount *mp = dp->i_mount;
145 err = xfs_da_read_buf(tp, dp, mp->m_dir_geo->datablk, 0, bpp,
146 XFS_DATA_FORK, &xfs_dir3_block_buf_ops);
150 /* Check things that we can't do in the verifier. */
151 fa = xfs_dir3_block_header_check(dp, *bpp);
153 __xfs_buf_mark_corrupt(*bpp, fa);
154 xfs_trans_brelse(tp, *bpp);
156 xfs_dirattr_mark_sick(dp, XFS_DATA_FORK);
157 return -EFSCORRUPTED;
160 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_BLOCK_BUF);
166 struct xfs_mount *mp,
167 struct xfs_trans *tp,
169 struct xfs_inode *dp)
171 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
173 bp->b_ops = &xfs_dir3_block_buf_ops;
174 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_BLOCK_BUF);
176 if (xfs_has_crc(mp)) {
177 memset(hdr3, 0, sizeof(*hdr3));
178 hdr3->magic = cpu_to_be32(XFS_DIR3_BLOCK_MAGIC);
179 hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
180 hdr3->owner = cpu_to_be64(dp->i_ino);
181 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
185 hdr3->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
189 xfs_dir2_block_need_space(
190 struct xfs_inode *dp,
191 struct xfs_dir2_data_hdr *hdr,
192 struct xfs_dir2_block_tail *btp,
193 struct xfs_dir2_leaf_entry *blp,
195 struct xfs_dir2_data_unused **dupp,
196 struct xfs_dir2_data_unused **enddupp,
200 struct xfs_dir2_data_free *bf;
202 struct xfs_dir2_data_unused *dup = NULL;
203 struct xfs_dir2_data_unused *enddup = NULL;
206 bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
209 * If there are stale entries we'll use one for the leaf.
212 if (be16_to_cpu(bf[0].length) >= len) {
214 * The biggest entry enough to avoid compaction.
216 dup = (xfs_dir2_data_unused_t *)
217 ((char *)hdr + be16_to_cpu(bf[0].offset));
222 * Will need to compact to make this work.
223 * Tag just before the first leaf entry.
226 tagp = (__be16 *)blp - 1;
228 /* Data object just before the first leaf entry. */
229 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
232 * If it's not free then the data will go where the
233 * leaf data starts now, if it works at all.
235 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
236 if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
237 (uint)sizeof(*blp) < len)
239 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
242 dup = (xfs_dir2_data_unused_t *)blp;
247 * no stale entries, so just use free space.
248 * Tag just before the first leaf entry.
250 tagp = (__be16 *)blp - 1;
252 /* Data object just before the first leaf entry. */
253 enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
256 * If it's not free then can't do this add without cleaning up:
257 * the space before the first leaf entry needs to be free so it
258 * can be expanded to hold the pointer to the new entry.
260 if (be16_to_cpu(enddup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
262 * Check out the biggest freespace and see if it's the same one.
264 dup = (xfs_dir2_data_unused_t *)
265 ((char *)hdr + be16_to_cpu(bf[0].offset));
268 * Not the same free entry, just check its length.
270 if (be16_to_cpu(dup->length) < len)
276 * It is the biggest freespace, can it hold the leaf too?
278 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
280 * Yes, use the second-largest entry instead if it works.
282 if (be16_to_cpu(bf[1].length) >= len)
283 dup = (xfs_dir2_data_unused_t *)
284 ((char *)hdr + be16_to_cpu(bf[1].offset));
296 * compact the leaf entries.
297 * Leave the highest-numbered stale entry stale.
298 * XXX should be the one closest to mid but mid is not yet computed.
301 xfs_dir2_block_compact(
302 struct xfs_da_args *args,
304 struct xfs_dir2_data_hdr *hdr,
305 struct xfs_dir2_block_tail *btp,
306 struct xfs_dir2_leaf_entry *blp,
311 int fromidx; /* source leaf index */
312 int toidx; /* target leaf index */
314 int highstale; /* high stale index */
316 fromidx = toidx = be32_to_cpu(btp->count) - 1;
317 highstale = *lfloghigh = -1;
318 for (; fromidx >= 0; fromidx--) {
319 if (blp[fromidx].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
323 if (*lfloghigh == -1)
329 blp[toidx] = blp[fromidx];
332 *lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
333 *lfloghigh -= be32_to_cpu(btp->stale) - 1;
334 be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
335 xfs_dir2_data_make_free(args, bp,
336 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
337 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
339 btp->stale = cpu_to_be32(1);
341 * If we now need to rebuild the bestfree map, do so.
342 * This needs to happen before the next call to use_free.
345 xfs_dir2_data_freescan(args->dp->i_mount, hdr, needlog);
349 * Add an entry to a block directory.
352 xfs_dir2_block_addname(
353 xfs_da_args_t *args) /* directory op arguments */
355 xfs_dir2_data_hdr_t *hdr; /* block header */
356 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
357 struct xfs_buf *bp; /* buffer for block */
358 xfs_dir2_block_tail_t *btp; /* block tail */
359 int compact; /* need to compact leaf ents */
360 xfs_dir2_data_entry_t *dep; /* block data entry */
361 xfs_inode_t *dp; /* directory inode */
362 xfs_dir2_data_unused_t *dup; /* block unused entry */
363 int error; /* error return value */
364 xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
365 xfs_dahash_t hash; /* hash value of found entry */
366 int high; /* high index for binary srch */
367 int highstale; /* high stale index */
368 int lfloghigh=0; /* last final leaf to log */
369 int lfloglow=0; /* first final leaf to log */
370 int len; /* length of the new entry */
371 int low; /* low index for binary srch */
372 int lowstale; /* low stale index */
373 int mid=0; /* midpoint for binary srch */
374 int needlog; /* need to log header */
375 int needscan; /* need to rescan freespace */
376 __be16 *tagp; /* pointer to tag value */
377 xfs_trans_t *tp; /* transaction structure */
379 trace_xfs_dir2_block_addname(args);
384 /* Read the (one and only) directory block into bp. */
385 error = xfs_dir3_block_read(tp, dp, &bp);
389 len = xfs_dir2_data_entsize(dp->i_mount, args->namelen);
392 * Set up pointers to parts of the block.
395 btp = xfs_dir2_block_tail_p(args->geo, hdr);
396 blp = xfs_dir2_block_leaf_p(btp);
399 * Find out if we can reuse stale entries or whether we need extra
400 * space for entry and new leaf.
402 xfs_dir2_block_need_space(dp, hdr, btp, blp, &tagp, &dup,
403 &enddup, &compact, len);
406 * Done everything we need for a space check now.
408 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
409 xfs_trans_brelse(tp, bp);
416 * If we don't have space for the new entry & leaf ...
419 /* Don't have a space reservation: return no-space. */
420 if (args->total == 0)
423 * Convert to the next larger format.
424 * Then add the new entry in that format.
426 error = xfs_dir2_block_to_leaf(args, bp);
429 return xfs_dir2_leaf_addname(args);
432 needlog = needscan = 0;
435 * If need to compact the leaf entries, do it now.
438 xfs_dir2_block_compact(args, bp, hdr, btp, blp, &needlog,
439 &lfloghigh, &lfloglow);
440 /* recalculate blp post-compaction */
441 blp = xfs_dir2_block_leaf_p(btp);
442 } else if (btp->stale) {
444 * Set leaf logging boundaries to impossible state.
445 * For the no-stale case they're set explicitly.
447 lfloglow = be32_to_cpu(btp->count);
452 * Find the slot that's first lower than our hash value, -1 if none.
454 for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
455 mid = (low + high) >> 1;
456 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
458 if (hash < args->hashval)
463 while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
467 * No stale entries, will use enddup space to hold new leaf.
470 xfs_dir2_data_aoff_t aoff;
473 * Mark the space needed for the new leaf entry, now in use.
475 aoff = (xfs_dir2_data_aoff_t)((char *)enddup - (char *)hdr +
476 be16_to_cpu(enddup->length) - sizeof(*blp));
477 error = xfs_dir2_data_use_free(args, bp, enddup, aoff,
478 (xfs_dir2_data_aoff_t)sizeof(*blp), &needlog,
484 * Update the tail (entry count).
486 be32_add_cpu(&btp->count, 1);
488 * If we now need to rebuild the bestfree map, do so.
489 * This needs to happen before the next call to use_free.
492 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
496 * Adjust pointer to the first leaf entry, we're about to move
497 * the table up one to open up space for the new leaf entry.
498 * Then adjust our index to match.
503 memmove(blp, &blp[1], mid * sizeof(*blp));
508 * Use a stale leaf for our new entry.
513 blp[lowstale].address !=
514 cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
517 for (highstale = mid + 1;
518 highstale < be32_to_cpu(btp->count) &&
519 blp[highstale].address !=
520 cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
521 (lowstale < 0 || mid - lowstale > highstale - mid);
525 * Move entries toward the low-numbered stale entry.
528 (highstale == be32_to_cpu(btp->count) ||
529 mid - lowstale <= highstale - mid)) {
531 memmove(&blp[lowstale], &blp[lowstale + 1],
532 (mid - lowstale) * sizeof(*blp));
533 lfloglow = min(lowstale, lfloglow);
534 lfloghigh = max(mid, lfloghigh);
537 * Move entries toward the high-numbered stale entry.
540 ASSERT(highstale < be32_to_cpu(btp->count));
543 memmove(&blp[mid + 1], &blp[mid],
544 (highstale - mid) * sizeof(*blp));
545 lfloglow = min(mid, lfloglow);
546 lfloghigh = max(highstale, lfloghigh);
548 be32_add_cpu(&btp->stale, -1);
551 * Point to the new data entry.
553 dep = (xfs_dir2_data_entry_t *)dup;
555 * Fill in the leaf entry.
557 blp[mid].hashval = cpu_to_be32(args->hashval);
558 blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
559 (char *)dep - (char *)hdr));
560 xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
562 * Mark space for the data entry used.
564 error = xfs_dir2_data_use_free(args, bp, dup,
565 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
566 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
570 * Create the new data entry.
572 dep->inumber = cpu_to_be64(args->inumber);
573 dep->namelen = args->namelen;
574 memcpy(dep->name, args->name, args->namelen);
575 xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
576 tagp = xfs_dir2_data_entry_tag_p(dp->i_mount, dep);
577 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
579 * Clean up the bestfree array and log the header, tail, and entry.
582 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
584 xfs_dir2_data_log_header(args, bp);
585 xfs_dir2_block_log_tail(tp, bp);
586 xfs_dir2_data_log_entry(args, bp, dep);
587 xfs_dir3_data_check(dp, bp);
592 * Log leaf entries from the block.
595 xfs_dir2_block_log_leaf(
596 xfs_trans_t *tp, /* transaction structure */
597 struct xfs_buf *bp, /* block buffer */
598 int first, /* index of first logged leaf */
599 int last) /* index of last logged leaf */
601 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
602 xfs_dir2_leaf_entry_t *blp;
603 xfs_dir2_block_tail_t *btp;
605 btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
606 blp = xfs_dir2_block_leaf_p(btp);
607 xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
608 (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
612 * Log the block tail.
615 xfs_dir2_block_log_tail(
616 xfs_trans_t *tp, /* transaction structure */
617 struct xfs_buf *bp) /* block buffer */
619 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
620 xfs_dir2_block_tail_t *btp;
622 btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
623 xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
624 (uint)((char *)(btp + 1) - (char *)hdr - 1));
628 * Look up an entry in the block. This is the external routine,
629 * xfs_dir2_block_lookup_int does the real work.
632 xfs_dir2_block_lookup(
633 xfs_da_args_t *args) /* dir lookup arguments */
635 xfs_dir2_data_hdr_t *hdr; /* block header */
636 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
637 struct xfs_buf *bp; /* block buffer */
638 xfs_dir2_block_tail_t *btp; /* block tail */
639 xfs_dir2_data_entry_t *dep; /* block data entry */
640 xfs_inode_t *dp; /* incore inode */
641 int ent; /* entry index */
642 int error; /* error return value */
644 trace_xfs_dir2_block_lookup(args);
647 * Get the buffer, look up the entry.
648 * If not found (ENOENT) then return, have no buffer.
650 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
654 xfs_dir3_data_check(dp, bp);
655 btp = xfs_dir2_block_tail_p(args->geo, hdr);
656 blp = xfs_dir2_block_leaf_p(btp);
658 * Get the offset from the leaf entry, to point to the data.
660 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
661 xfs_dir2_dataptr_to_off(args->geo,
662 be32_to_cpu(blp[ent].address)));
664 * Fill in inode number, CI name if appropriate, release the block.
666 args->inumber = be64_to_cpu(dep->inumber);
667 args->filetype = xfs_dir2_data_get_ftype(dp->i_mount, dep);
668 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
669 xfs_trans_brelse(args->trans, bp);
674 * Internal block lookup routine.
676 static int /* error */
677 xfs_dir2_block_lookup_int(
678 xfs_da_args_t *args, /* dir lookup arguments */
679 struct xfs_buf **bpp, /* returned block buffer */
680 int *entno) /* returned entry number */
682 xfs_dir2_dataptr_t addr; /* data entry address */
683 xfs_dir2_data_hdr_t *hdr; /* block header */
684 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
685 struct xfs_buf *bp; /* block buffer */
686 xfs_dir2_block_tail_t *btp; /* block tail */
687 xfs_dir2_data_entry_t *dep; /* block data entry */
688 xfs_inode_t *dp; /* incore inode */
689 int error; /* error return value */
690 xfs_dahash_t hash; /* found hash value */
691 int high; /* binary search high index */
692 int low; /* binary search low index */
693 int mid; /* binary search current idx */
694 xfs_trans_t *tp; /* transaction pointer */
695 enum xfs_dacmp cmp; /* comparison result */
700 error = xfs_dir3_block_read(tp, dp, &bp);
705 xfs_dir3_data_check(dp, bp);
706 btp = xfs_dir2_block_tail_p(args->geo, hdr);
707 blp = xfs_dir2_block_leaf_p(btp);
709 * Loop doing a binary search for our hash value.
710 * Find our entry, ENOENT if it's not there.
712 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
714 mid = (low + high) >> 1;
715 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
717 if (hash < args->hashval)
722 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
723 xfs_trans_brelse(tp, bp);
728 * Back up to the first one with the right hash value.
730 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
734 * Now loop forward through all the entries with the
735 * right hash value looking for our name.
738 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
741 * Get pointer to the entry from the leaf.
743 dep = (xfs_dir2_data_entry_t *)
744 ((char *)hdr + xfs_dir2_dataptr_to_off(args->geo, addr));
746 * Compare name and if it's an exact match, return the index
747 * and buffer. If it's the first case-insensitive match, store
748 * the index and buffer and continue looking for an exact match.
750 cmp = xfs_dir2_compname(args, dep->name, dep->namelen);
751 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
752 args->cmpresult = cmp;
755 if (cmp == XFS_CMP_EXACT)
758 } while (++mid < be32_to_cpu(btp->count) &&
759 be32_to_cpu(blp[mid].hashval) == hash);
761 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
763 * Here, we can only be doing a lookup (not a rename or replace).
764 * If a case-insensitive match was found earlier, return success.
766 if (args->cmpresult == XFS_CMP_CASE)
769 * No match, release the buffer and return ENOENT.
771 xfs_trans_brelse(tp, bp);
776 * Remove an entry from a block format directory.
777 * If that makes the block small enough to fit in shortform, transform it.
780 xfs_dir2_block_removename(
781 xfs_da_args_t *args) /* directory operation args */
783 xfs_dir2_data_hdr_t *hdr; /* block header */
784 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
785 struct xfs_buf *bp; /* block buffer */
786 xfs_dir2_block_tail_t *btp; /* block tail */
787 xfs_dir2_data_entry_t *dep; /* block data entry */
788 xfs_inode_t *dp; /* incore inode */
789 int ent; /* block leaf entry index */
790 int error; /* error return value */
791 int needlog; /* need to log block header */
792 int needscan; /* need to fixup bestfree */
793 xfs_dir2_sf_hdr_t sfh; /* shortform header */
794 int size; /* shortform size */
795 xfs_trans_t *tp; /* transaction pointer */
797 trace_xfs_dir2_block_removename(args);
800 * Look up the entry in the block. Gets the buffer and entry index.
801 * It will always be there, the vnodeops level does a lookup first.
803 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
809 btp = xfs_dir2_block_tail_p(args->geo, hdr);
810 blp = xfs_dir2_block_leaf_p(btp);
812 * Point to the data entry using the leaf entry.
814 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
815 xfs_dir2_dataptr_to_off(args->geo,
816 be32_to_cpu(blp[ent].address)));
818 * Mark the data entry's space free.
820 needlog = needscan = 0;
821 xfs_dir2_data_make_free(args, bp,
822 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
823 xfs_dir2_data_entsize(dp->i_mount, dep->namelen), &needlog,
826 * Fix up the block tail.
828 be32_add_cpu(&btp->stale, 1);
829 xfs_dir2_block_log_tail(tp, bp);
831 * Remove the leaf entry by marking it stale.
833 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
834 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
836 * Fix up bestfree, log the header if necessary.
839 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
841 xfs_dir2_data_log_header(args, bp);
842 xfs_dir3_data_check(dp, bp);
844 * See if the size as a shortform is good enough.
846 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
847 if (size > xfs_inode_data_fork_size(dp))
851 * If it works, do the conversion.
853 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
857 * Replace an entry in a V2 block directory.
858 * Change the inode number to the new value.
861 xfs_dir2_block_replace(
862 xfs_da_args_t *args) /* directory operation args */
864 xfs_dir2_data_hdr_t *hdr; /* block header */
865 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
866 struct xfs_buf *bp; /* block buffer */
867 xfs_dir2_block_tail_t *btp; /* block tail */
868 xfs_dir2_data_entry_t *dep; /* block data entry */
869 xfs_inode_t *dp; /* incore inode */
870 int ent; /* leaf entry index */
871 int error; /* error return value */
873 trace_xfs_dir2_block_replace(args);
876 * Lookup the entry in the directory. Get buffer and entry index.
877 * This will always succeed since the caller has already done a lookup.
879 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
884 btp = xfs_dir2_block_tail_p(args->geo, hdr);
885 blp = xfs_dir2_block_leaf_p(btp);
887 * Point to the data entry we need to change.
889 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
890 xfs_dir2_dataptr_to_off(args->geo,
891 be32_to_cpu(blp[ent].address)));
892 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
894 * Change the inode number to the new value.
896 dep->inumber = cpu_to_be64(args->inumber);
897 xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
898 xfs_dir2_data_log_entry(args, bp, dep);
899 xfs_dir3_data_check(dp, bp);
904 * Qsort comparison routine for the block leaf entries.
906 static int /* sort order */
908 const void *a, /* first leaf entry */
909 const void *b) /* second leaf entry */
911 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
912 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
916 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
917 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
921 * Convert a V2 leaf directory to a V2 block directory if possible.
924 xfs_dir2_leaf_to_block(
925 xfs_da_args_t *args, /* operation arguments */
926 struct xfs_buf *lbp, /* leaf buffer */
927 struct xfs_buf *dbp) /* data buffer */
929 __be16 *bestsp; /* leaf bests table */
930 xfs_dir2_data_hdr_t *hdr; /* block header */
931 xfs_dir2_block_tail_t *btp; /* block tail */
932 xfs_inode_t *dp; /* incore directory inode */
933 xfs_dir2_data_unused_t *dup; /* unused data entry */
934 int error; /* error return value */
935 int from; /* leaf from index */
936 xfs_dir2_leaf_t *leaf; /* leaf structure */
937 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
938 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
939 xfs_mount_t *mp; /* file system mount point */
940 int needlog; /* need to log data header */
941 int needscan; /* need to scan for bestfree */
942 xfs_dir2_sf_hdr_t sfh; /* shortform header */
943 int size; /* bytes used */
944 __be16 *tagp; /* end of entry (tag) */
945 int to; /* block/leaf to index */
946 xfs_trans_t *tp; /* transaction pointer */
947 struct xfs_dir3_icleaf_hdr leafhdr;
949 trace_xfs_dir2_leaf_to_block(args);
955 xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, leaf);
956 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
958 ASSERT(leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
959 leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
961 * If there are data blocks other than the first one, take this
962 * opportunity to remove trailing empty data blocks that may have
963 * been left behind during no-space-reservation operations.
964 * These will show up in the leaf bests table.
966 while (dp->i_disk_size > args->geo->blksize) {
969 hdrsz = args->geo->data_entry_offset;
970 bestsp = xfs_dir2_leaf_bests_p(ltp);
971 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
972 args->geo->blksize - hdrsz) {
974 xfs_dir2_leaf_trim_data(args, lbp,
975 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
981 * Read the data block if we don't already have it, give up if it fails.
984 error = xfs_dir3_data_read(tp, dp, args->geo->datablk, 0, &dbp);
989 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
990 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
993 * Size of the "leaf" area in the block.
995 size = (uint)sizeof(xfs_dir2_block_tail_t) +
996 (uint)sizeof(*lep) * (leafhdr.count - leafhdr.stale);
998 * Look at the last data entry.
1000 tagp = (__be16 *)((char *)hdr + args->geo->blksize) - 1;
1001 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
1003 * If it's not free or is too short we can't do it.
1005 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
1006 be16_to_cpu(dup->length) < size)
1010 * Start converting it to block form.
1012 xfs_dir3_block_init(mp, tp, dbp, dp);
1017 * Use up the space at the end of the block (blp/btp).
1019 error = xfs_dir2_data_use_free(args, dbp, dup,
1020 args->geo->blksize - size, size, &needlog, &needscan);
1024 * Initialize the block tail.
1026 btp = xfs_dir2_block_tail_p(args->geo, hdr);
1027 btp->count = cpu_to_be32(leafhdr.count - leafhdr.stale);
1029 xfs_dir2_block_log_tail(tp, dbp);
1031 * Initialize the block leaf area. We compact out stale entries.
1033 lep = xfs_dir2_block_leaf_p(btp);
1034 for (from = to = 0; from < leafhdr.count; from++) {
1035 if (leafhdr.ents[from].address ==
1036 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
1038 lep[to++] = leafhdr.ents[from];
1040 ASSERT(to == be32_to_cpu(btp->count));
1041 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
1043 * Scan the bestfree if we need it and log the data block header.
1046 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
1048 xfs_dir2_data_log_header(args, dbp);
1050 * Pitch the old leaf block.
1052 error = xfs_da_shrink_inode(args, args->geo->leafblk, lbp);
1057 * Now see if the resulting block can be shrunken to shortform.
1059 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1060 if (size > xfs_inode_data_fork_size(dp))
1063 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1067 * Convert the shortform directory to block form.
1070 xfs_dir2_sf_to_block(
1071 struct xfs_da_args *args)
1073 struct xfs_trans *tp = args->trans;
1074 struct xfs_inode *dp = args->dp;
1075 struct xfs_mount *mp = dp->i_mount;
1076 struct xfs_ifork *ifp = xfs_ifork_ptr(dp, XFS_DATA_FORK);
1077 struct xfs_da_geometry *geo = args->geo;
1078 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1079 xfs_dir2_data_hdr_t *hdr; /* block header */
1080 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1081 struct xfs_buf *bp; /* block buffer */
1082 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1083 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1084 int dummy; /* trash */
1085 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1086 int endoffset; /* end of data objects */
1087 int error; /* error return value */
1089 int needlog; /* need to log block header */
1090 int needscan; /* need to scan block freespc */
1091 int newoffset; /* offset from current entry */
1092 unsigned int offset = geo->data_entry_offset;
1093 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1094 struct xfs_dir2_sf_hdr *oldsfp = ifp->if_data;
1095 xfs_dir2_sf_hdr_t *sfp; /* shortform header */
1096 __be16 *tagp; /* end of data entry */
1097 struct xfs_name name;
1099 trace_xfs_dir2_sf_to_block(args);
1101 ASSERT(ifp->if_format == XFS_DINODE_FMT_LOCAL);
1102 ASSERT(dp->i_disk_size >= offsetof(struct xfs_dir2_sf_hdr, parent));
1104 ASSERT(ifp->if_bytes == dp->i_disk_size);
1105 ASSERT(oldsfp != NULL);
1106 ASSERT(dp->i_disk_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1107 ASSERT(dp->i_df.if_nextents == 0);
1110 * Copy the directory into a temporary buffer.
1111 * Then pitch the incore inode data so we can make extents.
1113 sfp = kmalloc(ifp->if_bytes, GFP_KERNEL | __GFP_NOFAIL);
1114 memcpy(sfp, oldsfp, ifp->if_bytes);
1116 xfs_idata_realloc(dp, -ifp->if_bytes, XFS_DATA_FORK);
1117 xfs_bmap_local_to_extents_empty(tp, dp, XFS_DATA_FORK);
1118 dp->i_disk_size = 0;
1121 * Add block 0 to the inode.
1123 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1127 * Initialize the data block, then convert it to block format.
1129 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 = bp->b_addr + offset;
1145 needlog = needscan = 0;
1146 error = xfs_dir2_data_use_free(args, bp, dup, args->geo->blksize - i,
1147 i, &needlog, &needscan);
1150 ASSERT(needscan == 0);
1154 btp = xfs_dir2_block_tail_p(args->geo, hdr);
1155 btp->count = cpu_to_be32(sfp->count + 2); /* ., .. */
1157 blp = xfs_dir2_block_leaf_p(btp);
1158 endoffset = (uint)((char *)blp - (char *)hdr);
1160 * Remove the freespace, we'll manage it.
1162 error = xfs_dir2_data_use_free(args, bp, dup,
1163 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1164 be16_to_cpu(dup->length), &needlog, &needscan);
1169 * Create entry for .
1171 dep = bp->b_addr + offset;
1172 dep->inumber = cpu_to_be64(dp->i_ino);
1175 xfs_dir2_data_put_ftype(mp, dep, XFS_DIR3_FT_DIR);
1176 tagp = xfs_dir2_data_entry_tag_p(mp, dep);
1177 *tagp = cpu_to_be16(offset);
1178 xfs_dir2_data_log_entry(args, bp, dep);
1179 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1180 blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(offset));
1181 offset += xfs_dir2_data_entsize(mp, dep->namelen);
1184 * Create entry for ..
1186 dep = bp->b_addr + offset;
1187 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_parent_ino(sfp));
1189 dep->name[0] = dep->name[1] = '.';
1190 xfs_dir2_data_put_ftype(mp, dep, XFS_DIR3_FT_DIR);
1191 tagp = xfs_dir2_data_entry_tag_p(mp, dep);
1192 *tagp = cpu_to_be16(offset);
1193 xfs_dir2_data_log_entry(args, bp, dep);
1194 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1195 blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(offset));
1196 offset += xfs_dir2_data_entsize(mp, dep->namelen);
1199 * Loop over existing entries, stuff them in.
1205 sfep = xfs_dir2_sf_firstentry(sfp);
1208 * Need to preserve the existing offset values in the sf directory.
1209 * Insert holes (unused entries) where necessary.
1211 while (offset < endoffset) {
1213 * sfep is null when we reach the end of the list.
1216 newoffset = endoffset;
1218 newoffset = xfs_dir2_sf_get_offset(sfep);
1220 * There should be a hole here, make one.
1222 if (offset < newoffset) {
1223 dup = bp->b_addr + offset;
1224 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1225 dup->length = cpu_to_be16(newoffset - offset);
1226 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(offset);
1227 xfs_dir2_data_log_unused(args, bp, dup);
1228 xfs_dir2_data_freeinsert(hdr,
1229 xfs_dir2_data_bestfree_p(mp, hdr),
1231 offset += be16_to_cpu(dup->length);
1235 * Copy a real entry.
1237 dep = bp->b_addr + newoffset;
1238 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_ino(mp, sfp, sfep));
1239 dep->namelen = sfep->namelen;
1240 xfs_dir2_data_put_ftype(mp, dep,
1241 xfs_dir2_sf_get_ftype(mp, sfep));
1242 memcpy(dep->name, sfep->name, dep->namelen);
1243 tagp = xfs_dir2_data_entry_tag_p(mp, dep);
1244 *tagp = cpu_to_be16(newoffset);
1245 xfs_dir2_data_log_entry(args, bp, dep);
1246 name.name = sfep->name;
1247 name.len = sfep->namelen;
1248 blp[2 + i].hashval = cpu_to_be32(xfs_dir2_hashname(mp, &name));
1249 blp[2 + i].address =
1250 cpu_to_be32(xfs_dir2_byte_to_dataptr(newoffset));
1251 offset = (int)((char *)(tagp + 1) - (char *)hdr);
1252 if (++i == sfp->count)
1255 sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
1257 /* Done with the temporary buffer */
1260 * Sort the leaf entries by hash value.
1262 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1264 * Log the leaf entry area and tail.
1265 * Already logged the header in data_init, ignore needlog.
1267 ASSERT(needscan == 0);
1268 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1269 xfs_dir2_block_log_tail(tp, bp);
1270 xfs_dir3_data_check(dp, bp);