2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
23 #include "xfs_trans.h"
26 #include "xfs_dmapi.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_dir2_sf.h"
31 #include "xfs_attr_sf.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_dir2_data.h"
36 #include "xfs_dir2_leaf.h"
37 #include "xfs_dir2_block.h"
38 #include "xfs_dir2_trace.h"
39 #include "xfs_error.h"
42 * Local function prototypes.
44 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, xfs_dabuf_t *bp, int first,
46 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, xfs_dabuf_t *bp);
47 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **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(".", 1);
60 xfs_dir_hash_dotdot = xfs_da_hashname("..", 2);
64 * Add an entry to a block directory.
67 xfs_dir2_block_addname(
68 xfs_da_args_t *args) /* directory op arguments */
70 xfs_dir2_data_free_t *bf; /* bestfree table in block */
71 xfs_dir2_block_t *block; /* directory block structure */
72 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
73 xfs_dabuf_t *bp; /* buffer for block */
74 xfs_dir2_block_tail_t *btp; /* block tail */
75 int compact; /* need to compact leaf ents */
76 xfs_dir2_data_entry_t *dep; /* block data entry */
77 xfs_inode_t *dp; /* directory inode */
78 xfs_dir2_data_unused_t *dup; /* block unused entry */
79 int error; /* error return value */
80 xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
81 xfs_dahash_t hash; /* hash value of found entry */
82 int high; /* high index for binary srch */
83 int highstale; /* high stale index */
84 int lfloghigh=0; /* last final leaf to log */
85 int lfloglow=0; /* first final leaf to log */
86 int len; /* length of the new entry */
87 int low; /* low index for binary srch */
88 int lowstale; /* low stale index */
89 int mid=0; /* midpoint for binary srch */
90 xfs_mount_t *mp; /* filesystem mount point */
91 int needlog; /* need to log header */
92 int needscan; /* need to rescan freespace */
93 __be16 *tagp; /* pointer to tag value */
94 xfs_trans_t *tp; /* transaction structure */
96 xfs_dir2_trace_args("block_addname", args);
101 * Read the (one and only) directory block into dabuf bp.
104 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
110 * Check the magic number, corrupted if wrong.
112 if (unlikely(be32_to_cpu(block->hdr.magic) != XFS_DIR2_BLOCK_MAGIC)) {
113 XFS_CORRUPTION_ERROR("xfs_dir2_block_addname",
114 XFS_ERRLEVEL_LOW, mp, block);
115 xfs_da_brelse(tp, bp);
116 return XFS_ERROR(EFSCORRUPTED);
118 len = XFS_DIR2_DATA_ENTSIZE(args->namelen);
120 * Set up pointers to parts of the block.
122 bf = block->hdr.bestfree;
123 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
124 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
126 * No stale entries? Need space for entry and new leaf.
130 * Tag just before the first leaf entry.
132 tagp = (__be16 *)blp - 1;
134 * Data object just before the first leaf entry.
136 enddup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
138 * If it's not free then can't do this add without cleaning up:
139 * the space before the first leaf entry needs to be free so it
140 * can be expanded to hold the pointer to the new entry.
142 if (be16_to_cpu(enddup->freetag) != XFS_DIR2_DATA_FREE_TAG)
145 * Check out the biggest freespace and see if it's the same one.
148 dup = (xfs_dir2_data_unused_t *)
149 ((char *)block + be16_to_cpu(bf[0].offset));
152 * It is the biggest freespace, is it too small
153 * to hold the new leaf too?
155 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
157 * Yes, we use the second-largest
158 * entry instead if it works.
160 if (be16_to_cpu(bf[1].length) >= len)
161 dup = (xfs_dir2_data_unused_t *)
163 be16_to_cpu(bf[1].offset));
169 * Not the same free entry,
170 * just check its length.
172 if (be16_to_cpu(dup->length) < len) {
180 * If there are stale entries we'll use one for the leaf.
181 * Is the biggest entry enough to avoid compaction?
183 else if (be16_to_cpu(bf[0].length) >= len) {
184 dup = (xfs_dir2_data_unused_t *)
185 ((char *)block + be16_to_cpu(bf[0].offset));
189 * Will need to compact to make this work.
193 * Tag just before the first leaf entry.
195 tagp = (__be16 *)blp - 1;
197 * Data object just before the first leaf entry.
199 dup = (xfs_dir2_data_unused_t *)((char *)block + 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;
215 * If this isn't a real add, we're done with the buffer.
218 xfs_da_brelse(tp, bp);
220 * If we don't have space for the new entry & leaf ...
224 * Not trying to actually do anything, or don't have
225 * a space reservation: return no-space.
227 if (args->justcheck || args->total == 0)
228 return XFS_ERROR(ENOSPC);
230 * Convert to the next larger format.
231 * Then add the new entry in that format.
233 error = xfs_dir2_block_to_leaf(args, bp);
237 return xfs_dir2_leaf_addname(args);
240 * Just checking, and it would work, so say so.
244 needlog = needscan = 0;
246 * If need to compact the leaf entries, do it now.
247 * Leave the highest-numbered stale entry stale.
248 * XXX should be the one closest to mid but mid is not yet computed.
251 int fromidx; /* source leaf index */
252 int toidx; /* target leaf index */
254 for (fromidx = toidx = be32_to_cpu(btp->count) - 1,
255 highstale = lfloghigh = -1;
258 if (be32_to_cpu(blp[fromidx].address) == XFS_DIR2_NULL_DATAPTR) {
268 blp[toidx] = blp[fromidx];
271 lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
272 lfloghigh -= be32_to_cpu(btp->stale) - 1;
273 be32_add(&btp->count, -(be32_to_cpu(btp->stale) - 1));
274 xfs_dir2_data_make_free(tp, bp,
275 (xfs_dir2_data_aoff_t)((char *)blp - (char *)block),
276 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
277 &needlog, &needscan);
278 blp += be32_to_cpu(btp->stale) - 1;
279 btp->stale = cpu_to_be32(1);
281 * If we now need to rebuild the bestfree map, do so.
282 * This needs to happen before the next call to use_free.
285 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block,
291 * Set leaf logging boundaries to impossible state.
292 * For the no-stale case they're set explicitly.
294 else if (btp->stale) {
295 lfloglow = be32_to_cpu(btp->count);
299 * Find the slot that's first lower than our hash value, -1 if none.
301 for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
302 mid = (low + high) >> 1;
303 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
305 if (hash < args->hashval)
310 while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
314 * No stale entries, will use enddup space to hold new leaf.
318 * Mark the space needed for the new leaf entry, now in use.
320 xfs_dir2_data_use_free(tp, bp, enddup,
321 (xfs_dir2_data_aoff_t)
322 ((char *)enddup - (char *)block + be16_to_cpu(enddup->length) -
324 (xfs_dir2_data_aoff_t)sizeof(*blp),
325 &needlog, &needscan);
327 * Update the tail (entry count).
329 be32_add(&btp->count, 1);
331 * If we now need to rebuild the bestfree map, do so.
332 * This needs to happen before the next call to use_free.
335 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block,
340 * Adjust pointer to the first leaf entry, we're about to move
341 * the table up one to open up space for the new leaf entry.
342 * Then adjust our index to match.
347 memmove(blp, &blp[1], mid * sizeof(*blp));
352 * Use a stale leaf for our new entry.
357 be32_to_cpu(blp[lowstale].address) != XFS_DIR2_NULL_DATAPTR;
360 for (highstale = mid + 1;
361 highstale < be32_to_cpu(btp->count) &&
362 be32_to_cpu(blp[highstale].address) != XFS_DIR2_NULL_DATAPTR &&
363 (lowstale < 0 || mid - lowstale > highstale - mid);
367 * Move entries toward the low-numbered stale entry.
370 (highstale == be32_to_cpu(btp->count) ||
371 mid - lowstale <= highstale - mid)) {
373 memmove(&blp[lowstale], &blp[lowstale + 1],
374 (mid - lowstale) * sizeof(*blp));
375 lfloglow = MIN(lowstale, lfloglow);
376 lfloghigh = MAX(mid, lfloghigh);
379 * Move entries toward the high-numbered stale entry.
382 ASSERT(highstale < be32_to_cpu(btp->count));
385 memmove(&blp[mid + 1], &blp[mid],
386 (highstale - mid) * sizeof(*blp));
387 lfloglow = MIN(mid, lfloglow);
388 lfloghigh = MAX(highstale, lfloghigh);
390 be32_add(&btp->stale, -1);
393 * Point to the new data entry.
395 dep = (xfs_dir2_data_entry_t *)dup;
397 * Fill in the leaf entry.
399 blp[mid].hashval = cpu_to_be32(args->hashval);
400 blp[mid].address = cpu_to_be32(XFS_DIR2_BYTE_TO_DATAPTR(mp,
401 (char *)dep - (char *)block));
402 xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
404 * Mark space for the data entry used.
406 xfs_dir2_data_use_free(tp, bp, dup,
407 (xfs_dir2_data_aoff_t)((char *)dup - (char *)block),
408 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
410 * Create the new data entry.
412 dep->inumber = cpu_to_be64(args->inumber);
413 dep->namelen = args->namelen;
414 memcpy(dep->name, args->name, args->namelen);
415 tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
416 *tagp = cpu_to_be16((char *)dep - (char *)block);
418 * Clean up the bestfree array and log the header, tail, and entry.
421 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
424 xfs_dir2_data_log_header(tp, bp);
425 xfs_dir2_block_log_tail(tp, bp);
426 xfs_dir2_data_log_entry(tp, bp, dep);
427 xfs_dir2_data_check(dp, bp);
433 * Readdir for block directories.
436 xfs_dir2_block_getdents(
437 xfs_trans_t *tp, /* transaction (NULL) */
438 xfs_inode_t *dp, /* incore inode */
439 uio_t *uio, /* caller's buffer control */
440 int *eofp, /* eof reached? (out) */
441 xfs_dirent_t *dbp, /* caller's buffer */
442 xfs_dir2_put_t put) /* abi's formatting function */
444 xfs_dir2_block_t *block; /* directory block structure */
445 xfs_dabuf_t *bp; /* buffer for block */
446 xfs_dir2_block_tail_t *btp; /* block tail */
447 xfs_dir2_data_entry_t *dep; /* block data entry */
448 xfs_dir2_data_unused_t *dup; /* block unused entry */
449 char *endptr; /* end of the data entries */
450 int error; /* error return value */
451 xfs_mount_t *mp; /* filesystem mount point */
452 xfs_dir2_put_args_t p; /* arg package for put rtn */
453 char *ptr; /* current data entry */
454 int wantoff; /* starting block offset */
458 * If the block number in the offset is out of range, we're done.
460 if (XFS_DIR2_DATAPTR_TO_DB(mp, uio->uio_offset) > mp->m_dirdatablk) {
465 * Can't read the block, give up, else get dabuf in bp.
468 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
473 * Extract the byte offset we start at from the seek pointer.
474 * We'll skip entries before this.
476 wantoff = XFS_DIR2_DATAPTR_TO_OFF(mp, uio->uio_offset);
478 xfs_dir2_data_check(dp, bp);
480 * Set up values for the loop.
482 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
483 ptr = (char *)block->u;
484 endptr = (char *)XFS_DIR2_BLOCK_LEAF_P(btp);
489 * Loop over the data portion of the block.
490 * Each object is a real entry (dep) or an unused one (dup).
492 while (ptr < endptr) {
493 dup = (xfs_dir2_data_unused_t *)ptr;
497 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
498 ptr += be16_to_cpu(dup->length);
502 dep = (xfs_dir2_data_entry_t *)ptr;
505 * Bump pointer for the next iteration.
507 ptr += XFS_DIR2_DATA_ENTSIZE(dep->namelen);
509 * The entry is before the desired starting point, skip it.
511 if ((char *)dep - (char *)block < wantoff)
514 * Set up argument structure for put routine.
516 p.namelen = dep->namelen;
518 p.cook = XFS_DIR2_DB_OFF_TO_DATAPTR(mp, mp->m_dirdatablk,
519 ptr - (char *)block);
520 p.ino = be64_to_cpu(dep->inumber);
522 p.ino += mp->m_inoadd;
524 p.name = (char *)dep->name;
527 * Put the entry in the caller's buffer.
532 * If it didn't fit, set the final offset to here & return.
536 XFS_DIR2_DB_OFF_TO_DATAPTR(mp, mp->m_dirdatablk,
537 (char *)dep - (char *)block);
538 xfs_da_brelse(tp, bp);
544 * Reached the end of the block.
545 * Set the offset to a non-existent block 1 and return.
550 XFS_DIR2_DB_OFF_TO_DATAPTR(mp, mp->m_dirdatablk + 1, 0);
552 xfs_da_brelse(tp, bp);
558 * Log leaf entries from the block.
561 xfs_dir2_block_log_leaf(
562 xfs_trans_t *tp, /* transaction structure */
563 xfs_dabuf_t *bp, /* block buffer */
564 int first, /* index of first logged leaf */
565 int last) /* index of last logged leaf */
567 xfs_dir2_block_t *block; /* directory block structure */
568 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
569 xfs_dir2_block_tail_t *btp; /* block tail */
570 xfs_mount_t *mp; /* filesystem mount point */
574 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
575 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
576 xfs_da_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)block),
577 (uint)((char *)&blp[last + 1] - (char *)block - 1));
581 * Log the block tail.
584 xfs_dir2_block_log_tail(
585 xfs_trans_t *tp, /* transaction structure */
586 xfs_dabuf_t *bp) /* block buffer */
588 xfs_dir2_block_t *block; /* directory block structure */
589 xfs_dir2_block_tail_t *btp; /* block tail */
590 xfs_mount_t *mp; /* filesystem mount point */
594 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
595 xfs_da_log_buf(tp, bp, (uint)((char *)btp - (char *)block),
596 (uint)((char *)(btp + 1) - (char *)block - 1));
600 * Look up an entry in the block. This is the external routine,
601 * xfs_dir2_block_lookup_int does the real work.
604 xfs_dir2_block_lookup(
605 xfs_da_args_t *args) /* dir lookup arguments */
607 xfs_dir2_block_t *block; /* block structure */
608 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
609 xfs_dabuf_t *bp; /* block buffer */
610 xfs_dir2_block_tail_t *btp; /* block tail */
611 xfs_dir2_data_entry_t *dep; /* block data entry */
612 xfs_inode_t *dp; /* incore inode */
613 int ent; /* entry index */
614 int error; /* error return value */
615 xfs_mount_t *mp; /* filesystem mount point */
617 xfs_dir2_trace_args("block_lookup", args);
619 * Get the buffer, look up the entry.
620 * If not found (ENOENT) then return, have no buffer.
622 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
627 xfs_dir2_data_check(dp, bp);
628 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
629 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
631 * Get the offset from the leaf entry, to point to the data.
633 dep = (xfs_dir2_data_entry_t *)
634 ((char *)block + XFS_DIR2_DATAPTR_TO_OFF(mp, be32_to_cpu(blp[ent].address)));
636 * Fill in inode number, release the block.
638 args->inumber = be64_to_cpu(dep->inumber);
639 xfs_da_brelse(args->trans, bp);
640 return XFS_ERROR(EEXIST);
644 * Internal block lookup routine.
646 static int /* error */
647 xfs_dir2_block_lookup_int(
648 xfs_da_args_t *args, /* dir lookup arguments */
649 xfs_dabuf_t **bpp, /* returned block buffer */
650 int *entno) /* returned entry number */
652 xfs_dir2_dataptr_t addr; /* data entry address */
653 xfs_dir2_block_t *block; /* block structure */
654 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
655 xfs_dabuf_t *bp; /* block buffer */
656 xfs_dir2_block_tail_t *btp; /* block tail */
657 xfs_dir2_data_entry_t *dep; /* block data entry */
658 xfs_inode_t *dp; /* incore inode */
659 int error; /* error return value */
660 xfs_dahash_t hash; /* found hash value */
661 int high; /* binary search high index */
662 int low; /* binary search low index */
663 int mid; /* binary search current idx */
664 xfs_mount_t *mp; /* filesystem mount point */
665 xfs_trans_t *tp; /* transaction pointer */
671 * Read the buffer, return error if we can't get it.
674 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
679 xfs_dir2_data_check(dp, bp);
680 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
681 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
683 * Loop doing a binary search for our hash value.
684 * Find our entry, ENOENT if it's not there.
686 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
688 mid = (low + high) >> 1;
689 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
691 if (hash < args->hashval)
696 ASSERT(args->oknoent);
697 xfs_da_brelse(tp, bp);
698 return XFS_ERROR(ENOENT);
702 * Back up to the first one with the right hash value.
704 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
708 * Now loop forward through all the entries with the
709 * right hash value looking for our name.
712 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
715 * Get pointer to the entry from the leaf.
717 dep = (xfs_dir2_data_entry_t *)
718 ((char *)block + XFS_DIR2_DATAPTR_TO_OFF(mp, addr));
720 * Compare, if it's right give back buffer & entry number.
722 if (dep->namelen == args->namelen &&
723 dep->name[0] == args->name[0] &&
724 memcmp(dep->name, args->name, args->namelen) == 0) {
729 } while (++mid < be32_to_cpu(btp->count) && be32_to_cpu(blp[mid].hashval) == hash);
731 * No match, release the buffer and return ENOENT.
733 ASSERT(args->oknoent);
734 xfs_da_brelse(tp, bp);
735 return XFS_ERROR(ENOENT);
739 * Remove an entry from a block format directory.
740 * If that makes the block small enough to fit in shortform, transform it.
743 xfs_dir2_block_removename(
744 xfs_da_args_t *args) /* directory operation args */
746 xfs_dir2_block_t *block; /* block structure */
747 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
748 xfs_dabuf_t *bp; /* block buffer */
749 xfs_dir2_block_tail_t *btp; /* block tail */
750 xfs_dir2_data_entry_t *dep; /* block data entry */
751 xfs_inode_t *dp; /* incore inode */
752 int ent; /* block leaf entry index */
753 int error; /* error return value */
754 xfs_mount_t *mp; /* filesystem mount point */
755 int needlog; /* need to log block header */
756 int needscan; /* need to fixup bestfree */
757 xfs_dir2_sf_hdr_t sfh; /* shortform header */
758 int size; /* shortform size */
759 xfs_trans_t *tp; /* transaction pointer */
761 xfs_dir2_trace_args("block_removename", args);
763 * Look up the entry in the block. Gets the buffer and entry index.
764 * It will always be there, the vnodeops level does a lookup first.
766 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
773 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
774 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
776 * Point to the data entry using the leaf entry.
778 dep = (xfs_dir2_data_entry_t *)
779 ((char *)block + XFS_DIR2_DATAPTR_TO_OFF(mp, be32_to_cpu(blp[ent].address)));
781 * Mark the data entry's space free.
783 needlog = needscan = 0;
784 xfs_dir2_data_make_free(tp, bp,
785 (xfs_dir2_data_aoff_t)((char *)dep - (char *)block),
786 XFS_DIR2_DATA_ENTSIZE(dep->namelen), &needlog, &needscan);
788 * Fix up the block tail.
790 be32_add(&btp->stale, 1);
791 xfs_dir2_block_log_tail(tp, bp);
793 * Remove the leaf entry by marking it stale.
795 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
796 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
798 * Fix up bestfree, log the header if necessary.
801 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
804 xfs_dir2_data_log_header(tp, bp);
805 xfs_dir2_data_check(dp, bp);
807 * See if the size as a shortform is good enough.
809 if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
810 XFS_IFORK_DSIZE(dp)) {
815 * If it works, do the conversion.
817 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
821 * Replace an entry in a V2 block directory.
822 * Change the inode number to the new value.
825 xfs_dir2_block_replace(
826 xfs_da_args_t *args) /* directory operation args */
828 xfs_dir2_block_t *block; /* block structure */
829 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
830 xfs_dabuf_t *bp; /* block buffer */
831 xfs_dir2_block_tail_t *btp; /* block tail */
832 xfs_dir2_data_entry_t *dep; /* block data entry */
833 xfs_inode_t *dp; /* incore inode */
834 int ent; /* leaf entry index */
835 int error; /* error return value */
836 xfs_mount_t *mp; /* filesystem mount point */
838 xfs_dir2_trace_args("block_replace", args);
840 * Lookup the entry in the directory. Get buffer and entry index.
841 * This will always succeed since the caller has already done a lookup.
843 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
849 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
850 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
852 * Point to the data entry we need to change.
854 dep = (xfs_dir2_data_entry_t *)
855 ((char *)block + XFS_DIR2_DATAPTR_TO_OFF(mp, be32_to_cpu(blp[ent].address)));
856 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
858 * Change the inode number to the new value.
860 dep->inumber = cpu_to_be64(args->inumber);
861 xfs_dir2_data_log_entry(args->trans, bp, dep);
862 xfs_dir2_data_check(dp, bp);
868 * Qsort comparison routine for the block leaf entries.
870 static int /* sort order */
872 const void *a, /* first leaf entry */
873 const void *b) /* second leaf entry */
875 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
876 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
880 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
881 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
885 * Convert a V2 leaf directory to a V2 block directory if possible.
888 xfs_dir2_leaf_to_block(
889 xfs_da_args_t *args, /* operation arguments */
890 xfs_dabuf_t *lbp, /* leaf buffer */
891 xfs_dabuf_t *dbp) /* data buffer */
893 __be16 *bestsp; /* leaf bests table */
894 xfs_dir2_block_t *block; /* block structure */
895 xfs_dir2_block_tail_t *btp; /* block tail */
896 xfs_inode_t *dp; /* incore directory inode */
897 xfs_dir2_data_unused_t *dup; /* unused data entry */
898 int error; /* error return value */
899 int from; /* leaf from index */
900 xfs_dir2_leaf_t *leaf; /* leaf structure */
901 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
902 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
903 xfs_mount_t *mp; /* file system mount point */
904 int needlog; /* need to log data header */
905 int needscan; /* need to scan for bestfree */
906 xfs_dir2_sf_hdr_t sfh; /* shortform header */
907 int size; /* bytes used */
908 __be16 *tagp; /* end of entry (tag) */
909 int to; /* block/leaf to index */
910 xfs_trans_t *tp; /* transaction pointer */
912 xfs_dir2_trace_args_bb("leaf_to_block", args, lbp, dbp);
917 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
918 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
920 * If there are data blocks other than the first one, take this
921 * opportunity to remove trailing empty data blocks that may have
922 * been left behind during no-space-reservation operations.
923 * These will show up in the leaf bests table.
925 while (dp->i_d.di_size > mp->m_dirblksize) {
926 bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
927 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
928 mp->m_dirblksize - (uint)sizeof(block->hdr)) {
930 xfs_dir2_leaf_trim_data(args, lbp,
931 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
939 * Read the data block if we don't already have it, give up if it fails.
942 (error = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &dbp,
947 ASSERT(be32_to_cpu(block->hdr.magic) == XFS_DIR2_DATA_MAGIC);
949 * Size of the "leaf" area in the block.
951 size = (uint)sizeof(block->tail) +
952 (uint)sizeof(*lep) * (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
954 * Look at the last data entry.
956 tagp = (__be16 *)((char *)block + mp->m_dirblksize) - 1;
957 dup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
959 * If it's not free or is too short we can't do it.
961 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
962 be16_to_cpu(dup->length) < size) {
967 * Start converting it to block form.
969 block->hdr.magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
973 * Use up the space at the end of the block (blp/btp).
975 xfs_dir2_data_use_free(tp, dbp, dup, mp->m_dirblksize - size, size,
976 &needlog, &needscan);
978 * Initialize the block tail.
980 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
981 btp->count = cpu_to_be32(be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
983 xfs_dir2_block_log_tail(tp, dbp);
985 * Initialize the block leaf area. We compact out stale entries.
987 lep = XFS_DIR2_BLOCK_LEAF_P(btp);
988 for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
989 if (be32_to_cpu(leaf->ents[from].address) == XFS_DIR2_NULL_DATAPTR)
991 lep[to++] = leaf->ents[from];
993 ASSERT(to == be32_to_cpu(btp->count));
994 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
996 * Scan the bestfree if we need it and log the data block header.
999 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
1002 xfs_dir2_data_log_header(tp, dbp);
1004 * Pitch the old leaf block.
1006 error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
1012 * Now see if the resulting block can be shrunken to shortform.
1014 if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
1015 XFS_IFORK_DSIZE(dp)) {
1019 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1022 xfs_da_buf_done(lbp);
1024 xfs_da_buf_done(dbp);
1029 * Convert the shortform directory to block form.
1032 xfs_dir2_sf_to_block(
1033 xfs_da_args_t *args) /* operation arguments */
1035 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1036 xfs_dir2_block_t *block; /* block structure */
1037 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1038 xfs_dabuf_t *bp; /* block buffer */
1039 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1040 char *buf; /* sf buffer */
1042 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1043 xfs_inode_t *dp; /* incore directory inode */
1044 int dummy; /* trash */
1045 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1046 int endoffset; /* end of data objects */
1047 int error; /* error return value */
1049 xfs_mount_t *mp; /* filesystem mount point */
1050 int needlog; /* need to log block header */
1051 int needscan; /* need to scan block freespc */
1052 int newoffset; /* offset from current entry */
1053 int offset; /* target block offset */
1054 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1055 xfs_dir2_sf_t *sfp; /* shortform structure */
1056 __be16 *tagp; /* end of data entry */
1057 xfs_trans_t *tp; /* transaction pointer */
1059 xfs_dir2_trace_args("sf_to_block", args);
1063 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
1065 * Bomb out if the shortform directory is way too short.
1067 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1068 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1069 return XFS_ERROR(EIO);
1071 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
1072 ASSERT(dp->i_df.if_u1.if_data != NULL);
1073 sfp = (xfs_dir2_sf_t *)dp->i_df.if_u1.if_data;
1074 ASSERT(dp->i_d.di_size >= XFS_DIR2_SF_HDR_SIZE(sfp->hdr.i8count));
1076 * Copy the directory into the stack buffer.
1077 * Then pitch the incore inode data so we can make extents.
1080 buf_len = dp->i_df.if_bytes;
1081 buf = kmem_alloc(dp->i_df.if_bytes, KM_SLEEP);
1083 memcpy(buf, sfp, dp->i_df.if_bytes);
1084 xfs_idata_realloc(dp, -dp->i_df.if_bytes, XFS_DATA_FORK);
1085 dp->i_d.di_size = 0;
1086 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1088 * Reset pointer - old sfp is gone.
1090 sfp = (xfs_dir2_sf_t *)buf;
1092 * Add block 0 to the inode.
1094 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1096 kmem_free(buf, buf_len);
1100 * Initialize the data block.
1102 error = xfs_dir2_data_init(args, blkno, &bp);
1104 kmem_free(buf, buf_len);
1108 block->hdr.magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
1110 * Compute size of block "tail" area.
1112 i = (uint)sizeof(*btp) +
1113 (sfp->hdr.count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1115 * The whole thing is initialized to free by the init routine.
1116 * Say we're using the leaf and tail area.
1118 dup = (xfs_dir2_data_unused_t *)block->u;
1119 needlog = needscan = 0;
1120 xfs_dir2_data_use_free(tp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1122 ASSERT(needscan == 0);
1126 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
1127 btp->count = cpu_to_be32(sfp->hdr.count + 2); /* ., .. */
1129 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
1130 endoffset = (uint)((char *)blp - (char *)block);
1132 * Remove the freespace, we'll manage it.
1134 xfs_dir2_data_use_free(tp, bp, dup,
1135 (xfs_dir2_data_aoff_t)((char *)dup - (char *)block),
1136 be16_to_cpu(dup->length), &needlog, &needscan);
1138 * Create entry for .
1140 dep = (xfs_dir2_data_entry_t *)
1141 ((char *)block + XFS_DIR2_DATA_DOT_OFFSET);
1142 dep->inumber = cpu_to_be64(dp->i_ino);
1145 tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
1146 *tagp = cpu_to_be16((char *)dep - (char *)block);
1147 xfs_dir2_data_log_entry(tp, bp, dep);
1148 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1149 blp[0].address = cpu_to_be32(XFS_DIR2_BYTE_TO_DATAPTR(mp,
1150 (char *)dep - (char *)block));
1152 * Create entry for ..
1154 dep = (xfs_dir2_data_entry_t *)
1155 ((char *)block + XFS_DIR2_DATA_DOTDOT_OFFSET);
1156 dep->inumber = cpu_to_be64(XFS_DIR2_SF_GET_INUMBER(sfp, &sfp->hdr.parent));
1158 dep->name[0] = dep->name[1] = '.';
1159 tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
1160 *tagp = cpu_to_be16((char *)dep - (char *)block);
1161 xfs_dir2_data_log_entry(tp, bp, dep);
1162 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1163 blp[1].address = cpu_to_be32(XFS_DIR2_BYTE_TO_DATAPTR(mp,
1164 (char *)dep - (char *)block));
1165 offset = XFS_DIR2_DATA_FIRST_OFFSET;
1167 * Loop over existing entries, stuff them in.
1169 if ((i = 0) == sfp->hdr.count)
1172 sfep = XFS_DIR2_SF_FIRSTENTRY(sfp);
1174 * Need to preserve the existing offset values in the sf directory.
1175 * Insert holes (unused entries) where necessary.
1177 while (offset < endoffset) {
1179 * sfep is null when we reach the end of the list.
1182 newoffset = endoffset;
1184 newoffset = XFS_DIR2_SF_GET_OFFSET(sfep);
1186 * There should be a hole here, make one.
1188 if (offset < newoffset) {
1189 dup = (xfs_dir2_data_unused_t *)
1190 ((char *)block + offset);
1191 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1192 dup->length = cpu_to_be16(newoffset - offset);
1193 *XFS_DIR2_DATA_UNUSED_TAG_P(dup) = cpu_to_be16(
1194 ((char *)dup - (char *)block));
1195 xfs_dir2_data_log_unused(tp, bp, dup);
1196 (void)xfs_dir2_data_freeinsert((xfs_dir2_data_t *)block,
1198 offset += be16_to_cpu(dup->length);
1202 * Copy a real entry.
1204 dep = (xfs_dir2_data_entry_t *)((char *)block + newoffset);
1205 dep->inumber = cpu_to_be64(XFS_DIR2_SF_GET_INUMBER(sfp,
1206 XFS_DIR2_SF_INUMBERP(sfep)));
1207 dep->namelen = sfep->namelen;
1208 memcpy(dep->name, sfep->name, dep->namelen);
1209 tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
1210 *tagp = cpu_to_be16((char *)dep - (char *)block);
1211 xfs_dir2_data_log_entry(tp, bp, dep);
1212 blp[2 + i].hashval = cpu_to_be32(xfs_da_hashname(
1213 (char *)sfep->name, sfep->namelen));
1214 blp[2 + i].address = cpu_to_be32(XFS_DIR2_BYTE_TO_DATAPTR(mp,
1215 (char *)dep - (char *)block));
1216 offset = (int)((char *)(tagp + 1) - (char *)block);
1217 if (++i == sfp->hdr.count)
1220 sfep = XFS_DIR2_SF_NEXTENTRY(sfp, sfep);
1222 /* Done with the temporary buffer */
1223 kmem_free(buf, buf_len);
1225 * Sort the leaf entries by hash value.
1227 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1229 * Log the leaf entry area and tail.
1230 * Already logged the header in data_init, ignore needlog.
1232 ASSERT(needscan == 0);
1233 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1234 xfs_dir2_block_log_tail(tp, bp);
1235 xfs_dir2_data_check(dp, bp);
1236 xfs_da_buf_done(bp);