Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
[linux-2.6/suspend2-2.6.18.git] / fs / xfs / xfs_dir2_block.c
blob972ded595476982dbb960e54245e01bcd86b477a
1 /*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
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
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_log.h"
22 #include "xfs_inum.h"
23 #include "xfs_trans.h"
24 #include "xfs_sb.h"
25 #include "xfs_dir.h"
26 #include "xfs_dir2.h"
27 #include "xfs_dmapi.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_dir_sf.h"
32 #include "xfs_dir2_sf.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_inode_item.h"
37 #include "xfs_dir_leaf.h"
38 #include "xfs_dir2_data.h"
39 #include "xfs_dir2_leaf.h"
40 #include "xfs_dir2_block.h"
41 #include "xfs_dir2_trace.h"
42 #include "xfs_error.h"
45 * Local function prototypes.
47 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, xfs_dabuf_t *bp, int first,
48 int last);
49 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, xfs_dabuf_t *bp);
50 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **bpp,
51 int *entno);
52 static int xfs_dir2_block_sort(const void *a, const void *b);
55 * Add an entry to a block directory.
57 int /* error */
58 xfs_dir2_block_addname(
59 xfs_da_args_t *args) /* directory op arguments */
61 xfs_dir2_data_free_t *bf; /* bestfree table in block */
62 xfs_dir2_block_t *block; /* directory block structure */
63 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
64 xfs_dabuf_t *bp; /* buffer for block */
65 xfs_dir2_block_tail_t *btp; /* block tail */
66 int compact; /* need to compact leaf ents */
67 xfs_dir2_data_entry_t *dep; /* block data entry */
68 xfs_inode_t *dp; /* directory inode */
69 xfs_dir2_data_unused_t *dup; /* block unused entry */
70 int error; /* error return value */
71 xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
72 xfs_dahash_t hash; /* hash value of found entry */
73 int high; /* high index for binary srch */
74 int highstale; /* high stale index */
75 int lfloghigh=0; /* last final leaf to log */
76 int lfloglow=0; /* first final leaf to log */
77 int len; /* length of the new entry */
78 int low; /* low index for binary srch */
79 int lowstale; /* low stale index */
80 int mid=0; /* midpoint for binary srch */
81 xfs_mount_t *mp; /* filesystem mount point */
82 int needlog; /* need to log header */
83 int needscan; /* need to rescan freespace */
84 __be16 *tagp; /* pointer to tag value */
85 xfs_trans_t *tp; /* transaction structure */
87 xfs_dir2_trace_args("block_addname", args);
88 dp = args->dp;
89 tp = args->trans;
90 mp = dp->i_mount;
92 * Read the (one and only) directory block into dabuf bp.
94 if ((error =
95 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
96 return error;
98 ASSERT(bp != NULL);
99 block = bp->data;
101 * Check the magic number, corrupted if wrong.
103 if (unlikely(be32_to_cpu(block->hdr.magic) != XFS_DIR2_BLOCK_MAGIC)) {
104 XFS_CORRUPTION_ERROR("xfs_dir2_block_addname",
105 XFS_ERRLEVEL_LOW, mp, block);
106 xfs_da_brelse(tp, bp);
107 return XFS_ERROR(EFSCORRUPTED);
109 len = XFS_DIR2_DATA_ENTSIZE(args->namelen);
111 * Set up pointers to parts of the block.
113 bf = block->hdr.bestfree;
114 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
115 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
117 * No stale entries? Need space for entry and new leaf.
119 if (!btp->stale) {
121 * Tag just before the first leaf entry.
123 tagp = (__be16 *)blp - 1;
125 * Data object just before the first leaf entry.
127 enddup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
129 * If it's not free then can't do this add without cleaning up:
130 * the space before the first leaf entry needs to be free so it
131 * can be expanded to hold the pointer to the new entry.
133 if (be16_to_cpu(enddup->freetag) != XFS_DIR2_DATA_FREE_TAG)
134 dup = enddup = NULL;
136 * Check out the biggest freespace and see if it's the same one.
138 else {
139 dup = (xfs_dir2_data_unused_t *)
140 ((char *)block + be16_to_cpu(bf[0].offset));
141 if (dup == enddup) {
143 * It is the biggest freespace, is it too small
144 * to hold the new leaf too?
146 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
148 * Yes, we use the second-largest
149 * entry instead if it works.
151 if (be16_to_cpu(bf[1].length) >= len)
152 dup = (xfs_dir2_data_unused_t *)
153 ((char *)block +
154 be16_to_cpu(bf[1].offset));
155 else
156 dup = NULL;
158 } else {
160 * Not the same free entry,
161 * just check its length.
163 if (be16_to_cpu(dup->length) < len) {
164 dup = NULL;
168 compact = 0;
171 * If there are stale entries we'll use one for the leaf.
172 * Is the biggest entry enough to avoid compaction?
174 else if (be16_to_cpu(bf[0].length) >= len) {
175 dup = (xfs_dir2_data_unused_t *)
176 ((char *)block + be16_to_cpu(bf[0].offset));
177 compact = 0;
180 * Will need to compact to make this work.
182 else {
184 * Tag just before the first leaf entry.
186 tagp = (__be16 *)blp - 1;
188 * Data object just before the first leaf entry.
190 dup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
192 * If it's not free then the data will go where the
193 * leaf data starts now, if it works at all.
195 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
196 if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
197 (uint)sizeof(*blp) < len)
198 dup = NULL;
199 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
200 dup = NULL;
201 else
202 dup = (xfs_dir2_data_unused_t *)blp;
203 compact = 1;
206 * If this isn't a real add, we're done with the buffer.
208 if (args->justcheck)
209 xfs_da_brelse(tp, bp);
211 * If we don't have space for the new entry & leaf ...
213 if (!dup) {
215 * Not trying to actually do anything, or don't have
216 * a space reservation: return no-space.
218 if (args->justcheck || args->total == 0)
219 return XFS_ERROR(ENOSPC);
221 * Convert to the next larger format.
222 * Then add the new entry in that format.
224 error = xfs_dir2_block_to_leaf(args, bp);
225 xfs_da_buf_done(bp);
226 if (error)
227 return error;
228 return xfs_dir2_leaf_addname(args);
231 * Just checking, and it would work, so say so.
233 if (args->justcheck)
234 return 0;
235 needlog = needscan = 0;
237 * If need to compact the leaf entries, do it now.
238 * Leave the highest-numbered stale entry stale.
239 * XXX should be the one closest to mid but mid is not yet computed.
241 if (compact) {
242 int fromidx; /* source leaf index */
243 int toidx; /* target leaf index */
245 for (fromidx = toidx = be32_to_cpu(btp->count) - 1,
246 highstale = lfloghigh = -1;
247 fromidx >= 0;
248 fromidx--) {
249 if (be32_to_cpu(blp[fromidx].address) == XFS_DIR2_NULL_DATAPTR) {
250 if (highstale == -1)
251 highstale = toidx;
252 else {
253 if (lfloghigh == -1)
254 lfloghigh = toidx;
255 continue;
258 if (fromidx < toidx)
259 blp[toidx] = blp[fromidx];
260 toidx--;
262 lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
263 lfloghigh -= be32_to_cpu(btp->stale) - 1;
264 be32_add(&btp->count, -(be32_to_cpu(btp->stale) - 1));
265 xfs_dir2_data_make_free(tp, bp,
266 (xfs_dir2_data_aoff_t)((char *)blp - (char *)block),
267 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
268 &needlog, &needscan);
269 blp += be32_to_cpu(btp->stale) - 1;
270 btp->stale = cpu_to_be32(1);
272 * If we now need to rebuild the bestfree map, do so.
273 * This needs to happen before the next call to use_free.
275 if (needscan) {
276 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block,
277 &needlog, NULL);
278 needscan = 0;
282 * Set leaf logging boundaries to impossible state.
283 * For the no-stale case they're set explicitly.
285 else if (btp->stale) {
286 lfloglow = be32_to_cpu(btp->count);
287 lfloghigh = -1;
290 * Find the slot that's first lower than our hash value, -1 if none.
292 for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
293 mid = (low + high) >> 1;
294 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
295 break;
296 if (hash < args->hashval)
297 low = mid + 1;
298 else
299 high = mid - 1;
301 while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
302 mid--;
305 * No stale entries, will use enddup space to hold new leaf.
307 if (!btp->stale) {
309 * Mark the space needed for the new leaf entry, now in use.
311 xfs_dir2_data_use_free(tp, bp, enddup,
312 (xfs_dir2_data_aoff_t)
313 ((char *)enddup - (char *)block + be16_to_cpu(enddup->length) -
314 sizeof(*blp)),
315 (xfs_dir2_data_aoff_t)sizeof(*blp),
316 &needlog, &needscan);
318 * Update the tail (entry count).
320 be32_add(&btp->count, 1);
322 * If we now need to rebuild the bestfree map, do so.
323 * This needs to happen before the next call to use_free.
325 if (needscan) {
326 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block,
327 &needlog, NULL);
328 needscan = 0;
331 * Adjust pointer to the first leaf entry, we're about to move
332 * the table up one to open up space for the new leaf entry.
333 * Then adjust our index to match.
335 blp--;
336 mid++;
337 if (mid)
338 memmove(blp, &blp[1], mid * sizeof(*blp));
339 lfloglow = 0;
340 lfloghigh = mid;
343 * Use a stale leaf for our new entry.
345 else {
346 for (lowstale = mid;
347 lowstale >= 0 &&
348 be32_to_cpu(blp[lowstale].address) != XFS_DIR2_NULL_DATAPTR;
349 lowstale--)
350 continue;
351 for (highstale = mid + 1;
352 highstale < be32_to_cpu(btp->count) &&
353 be32_to_cpu(blp[highstale].address) != XFS_DIR2_NULL_DATAPTR &&
354 (lowstale < 0 || mid - lowstale > highstale - mid);
355 highstale++)
356 continue;
358 * Move entries toward the low-numbered stale entry.
360 if (lowstale >= 0 &&
361 (highstale == be32_to_cpu(btp->count) ||
362 mid - lowstale <= highstale - mid)) {
363 if (mid - lowstale)
364 memmove(&blp[lowstale], &blp[lowstale + 1],
365 (mid - lowstale) * sizeof(*blp));
366 lfloglow = MIN(lowstale, lfloglow);
367 lfloghigh = MAX(mid, lfloghigh);
370 * Move entries toward the high-numbered stale entry.
372 else {
373 ASSERT(highstale < be32_to_cpu(btp->count));
374 mid++;
375 if (highstale - mid)
376 memmove(&blp[mid + 1], &blp[mid],
377 (highstale - mid) * sizeof(*blp));
378 lfloglow = MIN(mid, lfloglow);
379 lfloghigh = MAX(highstale, lfloghigh);
381 be32_add(&btp->stale, -1);
384 * Point to the new data entry.
386 dep = (xfs_dir2_data_entry_t *)dup;
388 * Fill in the leaf entry.
390 blp[mid].hashval = cpu_to_be32(args->hashval);
391 blp[mid].address = cpu_to_be32(XFS_DIR2_BYTE_TO_DATAPTR(mp,
392 (char *)dep - (char *)block));
393 xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
395 * Mark space for the data entry used.
397 xfs_dir2_data_use_free(tp, bp, dup,
398 (xfs_dir2_data_aoff_t)((char *)dup - (char *)block),
399 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
401 * Create the new data entry.
403 INT_SET(dep->inumber, ARCH_CONVERT, args->inumber);
404 dep->namelen = args->namelen;
405 memcpy(dep->name, args->name, args->namelen);
406 tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
407 *tagp = cpu_to_be16((char *)dep - (char *)block);
409 * Clean up the bestfree array and log the header, tail, and entry.
411 if (needscan)
412 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
413 NULL);
414 if (needlog)
415 xfs_dir2_data_log_header(tp, bp);
416 xfs_dir2_block_log_tail(tp, bp);
417 xfs_dir2_data_log_entry(tp, bp, dep);
418 xfs_dir2_data_check(dp, bp);
419 xfs_da_buf_done(bp);
420 return 0;
424 * Readdir for block directories.
426 int /* error */
427 xfs_dir2_block_getdents(
428 xfs_trans_t *tp, /* transaction (NULL) */
429 xfs_inode_t *dp, /* incore inode */
430 uio_t *uio, /* caller's buffer control */
431 int *eofp, /* eof reached? (out) */
432 xfs_dirent_t *dbp, /* caller's buffer */
433 xfs_dir2_put_t put) /* abi's formatting function */
435 xfs_dir2_block_t *block; /* directory block structure */
436 xfs_dabuf_t *bp; /* buffer for block */
437 xfs_dir2_block_tail_t *btp; /* block tail */
438 xfs_dir2_data_entry_t *dep; /* block data entry */
439 xfs_dir2_data_unused_t *dup; /* block unused entry */
440 char *endptr; /* end of the data entries */
441 int error; /* error return value */
442 xfs_mount_t *mp; /* filesystem mount point */
443 xfs_dir2_put_args_t p; /* arg package for put rtn */
444 char *ptr; /* current data entry */
445 int wantoff; /* starting block offset */
447 mp = dp->i_mount;
449 * If the block number in the offset is out of range, we're done.
451 if (XFS_DIR2_DATAPTR_TO_DB(mp, uio->uio_offset) > mp->m_dirdatablk) {
452 *eofp = 1;
453 return 0;
456 * Can't read the block, give up, else get dabuf in bp.
458 if ((error =
459 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
460 return error;
462 ASSERT(bp != NULL);
464 * Extract the byte offset we start at from the seek pointer.
465 * We'll skip entries before this.
467 wantoff = XFS_DIR2_DATAPTR_TO_OFF(mp, uio->uio_offset);
468 block = bp->data;
469 xfs_dir2_data_check(dp, bp);
471 * Set up values for the loop.
473 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
474 ptr = (char *)block->u;
475 endptr = (char *)XFS_DIR2_BLOCK_LEAF_P(btp);
476 p.dbp = dbp;
477 p.put = put;
478 p.uio = uio;
480 * Loop over the data portion of the block.
481 * Each object is a real entry (dep) or an unused one (dup).
483 while (ptr < endptr) {
484 dup = (xfs_dir2_data_unused_t *)ptr;
486 * Unused, skip it.
488 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
489 ptr += be16_to_cpu(dup->length);
490 continue;
493 dep = (xfs_dir2_data_entry_t *)ptr;
496 * Bump pointer for the next iteration.
498 ptr += XFS_DIR2_DATA_ENTSIZE(dep->namelen);
500 * The entry is before the desired starting point, skip it.
502 if ((char *)dep - (char *)block < wantoff)
503 continue;
505 * Set up argument structure for put routine.
507 p.namelen = dep->namelen;
509 p.cook = XFS_DIR2_DB_OFF_TO_DATAPTR(mp, mp->m_dirdatablk,
510 ptr - (char *)block);
511 p.ino = INT_GET(dep->inumber, ARCH_CONVERT);
512 #if XFS_BIG_INUMS
513 p.ino += mp->m_inoadd;
514 #endif
515 p.name = (char *)dep->name;
518 * Put the entry in the caller's buffer.
520 error = p.put(&p);
523 * If it didn't fit, set the final offset to here & return.
525 if (!p.done) {
526 uio->uio_offset =
527 XFS_DIR2_DB_OFF_TO_DATAPTR(mp, mp->m_dirdatablk,
528 (char *)dep - (char *)block);
529 xfs_da_brelse(tp, bp);
530 return error;
535 * Reached the end of the block.
536 * Set the offset to a non-existent block 1 and return.
538 *eofp = 1;
540 uio->uio_offset =
541 XFS_DIR2_DB_OFF_TO_DATAPTR(mp, mp->m_dirdatablk + 1, 0);
543 xfs_da_brelse(tp, bp);
545 return 0;
549 * Log leaf entries from the block.
551 static void
552 xfs_dir2_block_log_leaf(
553 xfs_trans_t *tp, /* transaction structure */
554 xfs_dabuf_t *bp, /* block buffer */
555 int first, /* index of first logged leaf */
556 int last) /* index of last logged leaf */
558 xfs_dir2_block_t *block; /* directory block structure */
559 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
560 xfs_dir2_block_tail_t *btp; /* block tail */
561 xfs_mount_t *mp; /* filesystem mount point */
563 mp = tp->t_mountp;
564 block = bp->data;
565 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
566 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
567 xfs_da_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)block),
568 (uint)((char *)&blp[last + 1] - (char *)block - 1));
572 * Log the block tail.
574 static void
575 xfs_dir2_block_log_tail(
576 xfs_trans_t *tp, /* transaction structure */
577 xfs_dabuf_t *bp) /* block buffer */
579 xfs_dir2_block_t *block; /* directory block structure */
580 xfs_dir2_block_tail_t *btp; /* block tail */
581 xfs_mount_t *mp; /* filesystem mount point */
583 mp = tp->t_mountp;
584 block = bp->data;
585 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
586 xfs_da_log_buf(tp, bp, (uint)((char *)btp - (char *)block),
587 (uint)((char *)(btp + 1) - (char *)block - 1));
591 * Look up an entry in the block. This is the external routine,
592 * xfs_dir2_block_lookup_int does the real work.
594 int /* error */
595 xfs_dir2_block_lookup(
596 xfs_da_args_t *args) /* dir lookup arguments */
598 xfs_dir2_block_t *block; /* block structure */
599 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
600 xfs_dabuf_t *bp; /* block buffer */
601 xfs_dir2_block_tail_t *btp; /* block tail */
602 xfs_dir2_data_entry_t *dep; /* block data entry */
603 xfs_inode_t *dp; /* incore inode */
604 int ent; /* entry index */
605 int error; /* error return value */
606 xfs_mount_t *mp; /* filesystem mount point */
608 xfs_dir2_trace_args("block_lookup", args);
610 * Get the buffer, look up the entry.
611 * If not found (ENOENT) then return, have no buffer.
613 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
614 return error;
615 dp = args->dp;
616 mp = dp->i_mount;
617 block = bp->data;
618 xfs_dir2_data_check(dp, bp);
619 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
620 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
622 * Get the offset from the leaf entry, to point to the data.
624 dep = (xfs_dir2_data_entry_t *)
625 ((char *)block + XFS_DIR2_DATAPTR_TO_OFF(mp, be32_to_cpu(blp[ent].address)));
627 * Fill in inode number, release the block.
629 args->inumber = INT_GET(dep->inumber, ARCH_CONVERT);
630 xfs_da_brelse(args->trans, bp);
631 return XFS_ERROR(EEXIST);
635 * Internal block lookup routine.
637 static int /* error */
638 xfs_dir2_block_lookup_int(
639 xfs_da_args_t *args, /* dir lookup arguments */
640 xfs_dabuf_t **bpp, /* returned block buffer */
641 int *entno) /* returned entry number */
643 xfs_dir2_dataptr_t addr; /* data entry address */
644 xfs_dir2_block_t *block; /* block structure */
645 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
646 xfs_dabuf_t *bp; /* block buffer */
647 xfs_dir2_block_tail_t *btp; /* block tail */
648 xfs_dir2_data_entry_t *dep; /* block data entry */
649 xfs_inode_t *dp; /* incore inode */
650 int error; /* error return value */
651 xfs_dahash_t hash; /* found hash value */
652 int high; /* binary search high index */
653 int low; /* binary search low index */
654 int mid; /* binary search current idx */
655 xfs_mount_t *mp; /* filesystem mount point */
656 xfs_trans_t *tp; /* transaction pointer */
658 dp = args->dp;
659 tp = args->trans;
660 mp = dp->i_mount;
662 * Read the buffer, return error if we can't get it.
664 if ((error =
665 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
666 return error;
668 ASSERT(bp != NULL);
669 block = bp->data;
670 xfs_dir2_data_check(dp, bp);
671 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
672 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
674 * Loop doing a binary search for our hash value.
675 * Find our entry, ENOENT if it's not there.
677 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
678 ASSERT(low <= high);
679 mid = (low + high) >> 1;
680 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
681 break;
682 if (hash < args->hashval)
683 low = mid + 1;
684 else
685 high = mid - 1;
686 if (low > high) {
687 ASSERT(args->oknoent);
688 xfs_da_brelse(tp, bp);
689 return XFS_ERROR(ENOENT);
693 * Back up to the first one with the right hash value.
695 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
696 mid--;
699 * Now loop forward through all the entries with the
700 * right hash value looking for our name.
702 do {
703 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
704 continue;
706 * Get pointer to the entry from the leaf.
708 dep = (xfs_dir2_data_entry_t *)
709 ((char *)block + XFS_DIR2_DATAPTR_TO_OFF(mp, addr));
711 * Compare, if it's right give back buffer & entry number.
713 if (dep->namelen == args->namelen &&
714 dep->name[0] == args->name[0] &&
715 memcmp(dep->name, args->name, args->namelen) == 0) {
716 *bpp = bp;
717 *entno = mid;
718 return 0;
720 } while (++mid < be32_to_cpu(btp->count) && be32_to_cpu(blp[mid].hashval) == hash);
722 * No match, release the buffer and return ENOENT.
724 ASSERT(args->oknoent);
725 xfs_da_brelse(tp, bp);
726 return XFS_ERROR(ENOENT);
730 * Remove an entry from a block format directory.
731 * If that makes the block small enough to fit in shortform, transform it.
733 int /* error */
734 xfs_dir2_block_removename(
735 xfs_da_args_t *args) /* directory operation args */
737 xfs_dir2_block_t *block; /* block structure */
738 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
739 xfs_dabuf_t *bp; /* block buffer */
740 xfs_dir2_block_tail_t *btp; /* block tail */
741 xfs_dir2_data_entry_t *dep; /* block data entry */
742 xfs_inode_t *dp; /* incore inode */
743 int ent; /* block leaf entry index */
744 int error; /* error return value */
745 xfs_mount_t *mp; /* filesystem mount point */
746 int needlog; /* need to log block header */
747 int needscan; /* need to fixup bestfree */
748 xfs_dir2_sf_hdr_t sfh; /* shortform header */
749 int size; /* shortform size */
750 xfs_trans_t *tp; /* transaction pointer */
752 xfs_dir2_trace_args("block_removename", args);
754 * Look up the entry in the block. Gets the buffer and entry index.
755 * It will always be there, the vnodeops level does a lookup first.
757 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
758 return error;
760 dp = args->dp;
761 tp = args->trans;
762 mp = dp->i_mount;
763 block = bp->data;
764 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
765 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
767 * Point to the data entry using the leaf entry.
769 dep = (xfs_dir2_data_entry_t *)
770 ((char *)block + XFS_DIR2_DATAPTR_TO_OFF(mp, be32_to_cpu(blp[ent].address)));
772 * Mark the data entry's space free.
774 needlog = needscan = 0;
775 xfs_dir2_data_make_free(tp, bp,
776 (xfs_dir2_data_aoff_t)((char *)dep - (char *)block),
777 XFS_DIR2_DATA_ENTSIZE(dep->namelen), &needlog, &needscan);
779 * Fix up the block tail.
781 be32_add(&btp->stale, 1);
782 xfs_dir2_block_log_tail(tp, bp);
784 * Remove the leaf entry by marking it stale.
786 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
787 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
789 * Fix up bestfree, log the header if necessary.
791 if (needscan)
792 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
793 NULL);
794 if (needlog)
795 xfs_dir2_data_log_header(tp, bp);
796 xfs_dir2_data_check(dp, bp);
798 * See if the size as a shortform is good enough.
800 if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
801 XFS_IFORK_DSIZE(dp)) {
802 xfs_da_buf_done(bp);
803 return 0;
806 * If it works, do the conversion.
808 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
812 * Replace an entry in a V2 block directory.
813 * Change the inode number to the new value.
815 int /* error */
816 xfs_dir2_block_replace(
817 xfs_da_args_t *args) /* directory operation args */
819 xfs_dir2_block_t *block; /* block structure */
820 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
821 xfs_dabuf_t *bp; /* block buffer */
822 xfs_dir2_block_tail_t *btp; /* block tail */
823 xfs_dir2_data_entry_t *dep; /* block data entry */
824 xfs_inode_t *dp; /* incore inode */
825 int ent; /* leaf entry index */
826 int error; /* error return value */
827 xfs_mount_t *mp; /* filesystem mount point */
829 xfs_dir2_trace_args("block_replace", args);
831 * Lookup the entry in the directory. Get buffer and entry index.
832 * This will always succeed since the caller has already done a lookup.
834 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
835 return error;
837 dp = args->dp;
838 mp = dp->i_mount;
839 block = bp->data;
840 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
841 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
843 * Point to the data entry we need to change.
845 dep = (xfs_dir2_data_entry_t *)
846 ((char *)block + XFS_DIR2_DATAPTR_TO_OFF(mp, be32_to_cpu(blp[ent].address)));
847 ASSERT(INT_GET(dep->inumber, ARCH_CONVERT) != args->inumber);
849 * Change the inode number to the new value.
851 INT_SET(dep->inumber, ARCH_CONVERT, args->inumber);
852 xfs_dir2_data_log_entry(args->trans, bp, dep);
853 xfs_dir2_data_check(dp, bp);
854 xfs_da_buf_done(bp);
855 return 0;
859 * Qsort comparison routine for the block leaf entries.
861 static int /* sort order */
862 xfs_dir2_block_sort(
863 const void *a, /* first leaf entry */
864 const void *b) /* second leaf entry */
866 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
867 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
869 la = a;
870 lb = b;
871 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
872 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
876 * Convert a V2 leaf directory to a V2 block directory if possible.
878 int /* error */
879 xfs_dir2_leaf_to_block(
880 xfs_da_args_t *args, /* operation arguments */
881 xfs_dabuf_t *lbp, /* leaf buffer */
882 xfs_dabuf_t *dbp) /* data buffer */
884 __be16 *bestsp; /* leaf bests table */
885 xfs_dir2_block_t *block; /* block structure */
886 xfs_dir2_block_tail_t *btp; /* block tail */
887 xfs_inode_t *dp; /* incore directory inode */
888 xfs_dir2_data_unused_t *dup; /* unused data entry */
889 int error; /* error return value */
890 int from; /* leaf from index */
891 xfs_dir2_leaf_t *leaf; /* leaf structure */
892 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
893 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
894 xfs_mount_t *mp; /* file system mount point */
895 int needlog; /* need to log data header */
896 int needscan; /* need to scan for bestfree */
897 xfs_dir2_sf_hdr_t sfh; /* shortform header */
898 int size; /* bytes used */
899 __be16 *tagp; /* end of entry (tag) */
900 int to; /* block/leaf to index */
901 xfs_trans_t *tp; /* transaction pointer */
903 xfs_dir2_trace_args_bb("leaf_to_block", args, lbp, dbp);
904 dp = args->dp;
905 tp = args->trans;
906 mp = dp->i_mount;
907 leaf = lbp->data;
908 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
909 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
911 * If there are data blocks other than the first one, take this
912 * opportunity to remove trailing empty data blocks that may have
913 * been left behind during no-space-reservation operations.
914 * These will show up in the leaf bests table.
916 while (dp->i_d.di_size > mp->m_dirblksize) {
917 bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
918 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
919 mp->m_dirblksize - (uint)sizeof(block->hdr)) {
920 if ((error =
921 xfs_dir2_leaf_trim_data(args, lbp,
922 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
923 goto out;
924 } else {
925 error = 0;
926 goto out;
930 * Read the data block if we don't already have it, give up if it fails.
932 if (dbp == NULL &&
933 (error = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &dbp,
934 XFS_DATA_FORK))) {
935 goto out;
937 block = dbp->data;
938 ASSERT(be32_to_cpu(block->hdr.magic) == XFS_DIR2_DATA_MAGIC);
940 * Size of the "leaf" area in the block.
942 size = (uint)sizeof(block->tail) +
943 (uint)sizeof(*lep) * (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
945 * Look at the last data entry.
947 tagp = (__be16 *)((char *)block + mp->m_dirblksize) - 1;
948 dup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
950 * If it's not free or is too short we can't do it.
952 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
953 be16_to_cpu(dup->length) < size) {
954 error = 0;
955 goto out;
958 * Start converting it to block form.
960 block->hdr.magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
961 needlog = 1;
962 needscan = 0;
964 * Use up the space at the end of the block (blp/btp).
966 xfs_dir2_data_use_free(tp, dbp, dup, mp->m_dirblksize - size, size,
967 &needlog, &needscan);
969 * Initialize the block tail.
971 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
972 btp->count = cpu_to_be32(be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
973 btp->stale = 0;
974 xfs_dir2_block_log_tail(tp, dbp);
976 * Initialize the block leaf area. We compact out stale entries.
978 lep = XFS_DIR2_BLOCK_LEAF_P(btp);
979 for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
980 if (be32_to_cpu(leaf->ents[from].address) == XFS_DIR2_NULL_DATAPTR)
981 continue;
982 lep[to++] = leaf->ents[from];
984 ASSERT(to == be32_to_cpu(btp->count));
985 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
987 * Scan the bestfree if we need it and log the data block header.
989 if (needscan)
990 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
991 NULL);
992 if (needlog)
993 xfs_dir2_data_log_header(tp, dbp);
995 * Pitch the old leaf block.
997 error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
998 lbp = NULL;
999 if (error) {
1000 goto out;
1003 * Now see if the resulting block can be shrunken to shortform.
1005 if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
1006 XFS_IFORK_DSIZE(dp)) {
1007 error = 0;
1008 goto out;
1010 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1011 out:
1012 if (lbp)
1013 xfs_da_buf_done(lbp);
1014 if (dbp)
1015 xfs_da_buf_done(dbp);
1016 return error;
1020 * Convert the shortform directory to block form.
1022 int /* error */
1023 xfs_dir2_sf_to_block(
1024 xfs_da_args_t *args) /* operation arguments */
1026 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1027 xfs_dir2_block_t *block; /* block structure */
1028 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1029 xfs_dabuf_t *bp; /* block buffer */
1030 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1031 char *buf; /* sf buffer */
1032 int buf_len;
1033 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1034 xfs_inode_t *dp; /* incore directory inode */
1035 int dummy; /* trash */
1036 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1037 int endoffset; /* end of data objects */
1038 int error; /* error return value */
1039 int i; /* index */
1040 xfs_mount_t *mp; /* filesystem mount point */
1041 int needlog; /* need to log block header */
1042 int needscan; /* need to scan block freespc */
1043 int newoffset; /* offset from current entry */
1044 int offset; /* target block offset */
1045 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1046 xfs_dir2_sf_t *sfp; /* shortform structure */
1047 __be16 *tagp; /* end of data entry */
1048 xfs_trans_t *tp; /* transaction pointer */
1050 xfs_dir2_trace_args("sf_to_block", args);
1051 dp = args->dp;
1052 tp = args->trans;
1053 mp = dp->i_mount;
1054 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
1056 * Bomb out if the shortform directory is way too short.
1058 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1059 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1060 return XFS_ERROR(EIO);
1062 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
1063 ASSERT(dp->i_df.if_u1.if_data != NULL);
1064 sfp = (xfs_dir2_sf_t *)dp->i_df.if_u1.if_data;
1065 ASSERT(dp->i_d.di_size >= XFS_DIR2_SF_HDR_SIZE(sfp->hdr.i8count));
1067 * Copy the directory into the stack buffer.
1068 * Then pitch the incore inode data so we can make extents.
1071 buf_len = dp->i_df.if_bytes;
1072 buf = kmem_alloc(dp->i_df.if_bytes, KM_SLEEP);
1074 memcpy(buf, sfp, dp->i_df.if_bytes);
1075 xfs_idata_realloc(dp, -dp->i_df.if_bytes, XFS_DATA_FORK);
1076 dp->i_d.di_size = 0;
1077 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1079 * Reset pointer - old sfp is gone.
1081 sfp = (xfs_dir2_sf_t *)buf;
1083 * Add block 0 to the inode.
1085 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1086 if (error) {
1087 kmem_free(buf, buf_len);
1088 return error;
1091 * Initialize the data block.
1093 error = xfs_dir2_data_init(args, blkno, &bp);
1094 if (error) {
1095 kmem_free(buf, buf_len);
1096 return error;
1098 block = bp->data;
1099 block->hdr.magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
1101 * Compute size of block "tail" area.
1103 i = (uint)sizeof(*btp) +
1104 (sfp->hdr.count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1106 * The whole thing is initialized to free by the init routine.
1107 * Say we're using the leaf and tail area.
1109 dup = (xfs_dir2_data_unused_t *)block->u;
1110 needlog = needscan = 0;
1111 xfs_dir2_data_use_free(tp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1112 &needscan);
1113 ASSERT(needscan == 0);
1115 * Fill in the tail.
1117 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
1118 btp->count = cpu_to_be32(sfp->hdr.count + 2); /* ., .. */
1119 btp->stale = 0;
1120 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
1121 endoffset = (uint)((char *)blp - (char *)block);
1123 * Remove the freespace, we'll manage it.
1125 xfs_dir2_data_use_free(tp, bp, dup,
1126 (xfs_dir2_data_aoff_t)((char *)dup - (char *)block),
1127 be16_to_cpu(dup->length), &needlog, &needscan);
1129 * Create entry for .
1131 dep = (xfs_dir2_data_entry_t *)
1132 ((char *)block + XFS_DIR2_DATA_DOT_OFFSET);
1133 INT_SET(dep->inumber, ARCH_CONVERT, dp->i_ino);
1134 dep->namelen = 1;
1135 dep->name[0] = '.';
1136 tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
1137 *tagp = cpu_to_be16((char *)dep - (char *)block);
1138 xfs_dir2_data_log_entry(tp, bp, dep);
1139 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1140 blp[0].address = cpu_to_be32(XFS_DIR2_BYTE_TO_DATAPTR(mp,
1141 (char *)dep - (char *)block));
1143 * Create entry for ..
1145 dep = (xfs_dir2_data_entry_t *)
1146 ((char *)block + XFS_DIR2_DATA_DOTDOT_OFFSET);
1147 INT_SET(dep->inumber, ARCH_CONVERT, XFS_DIR2_SF_GET_INUMBER(sfp, &sfp->hdr.parent));
1148 dep->namelen = 2;
1149 dep->name[0] = dep->name[1] = '.';
1150 tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
1151 *tagp = cpu_to_be16((char *)dep - (char *)block);
1152 xfs_dir2_data_log_entry(tp, bp, dep);
1153 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1154 blp[1].address = cpu_to_be32(XFS_DIR2_BYTE_TO_DATAPTR(mp,
1155 (char *)dep - (char *)block));
1156 offset = XFS_DIR2_DATA_FIRST_OFFSET;
1158 * Loop over existing entries, stuff them in.
1160 if ((i = 0) == sfp->hdr.count)
1161 sfep = NULL;
1162 else
1163 sfep = XFS_DIR2_SF_FIRSTENTRY(sfp);
1165 * Need to preserve the existing offset values in the sf directory.
1166 * Insert holes (unused entries) where necessary.
1168 while (offset < endoffset) {
1170 * sfep is null when we reach the end of the list.
1172 if (sfep == NULL)
1173 newoffset = endoffset;
1174 else
1175 newoffset = XFS_DIR2_SF_GET_OFFSET(sfep);
1177 * There should be a hole here, make one.
1179 if (offset < newoffset) {
1180 dup = (xfs_dir2_data_unused_t *)
1181 ((char *)block + offset);
1182 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1183 dup->length = cpu_to_be16(newoffset - offset);
1184 *XFS_DIR2_DATA_UNUSED_TAG_P(dup) = cpu_to_be16(
1185 ((char *)dup - (char *)block));
1186 xfs_dir2_data_log_unused(tp, bp, dup);
1187 (void)xfs_dir2_data_freeinsert((xfs_dir2_data_t *)block,
1188 dup, &dummy);
1189 offset += be16_to_cpu(dup->length);
1190 continue;
1193 * Copy a real entry.
1195 dep = (xfs_dir2_data_entry_t *)((char *)block + newoffset);
1196 INT_SET(dep->inumber, ARCH_CONVERT, XFS_DIR2_SF_GET_INUMBER(sfp,
1197 XFS_DIR2_SF_INUMBERP(sfep)));
1198 dep->namelen = sfep->namelen;
1199 memcpy(dep->name, sfep->name, dep->namelen);
1200 tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
1201 *tagp = cpu_to_be16((char *)dep - (char *)block);
1202 xfs_dir2_data_log_entry(tp, bp, dep);
1203 blp[2 + i].hashval = cpu_to_be32(xfs_da_hashname(
1204 (char *)sfep->name, sfep->namelen));
1205 blp[2 + i].address = cpu_to_be32(XFS_DIR2_BYTE_TO_DATAPTR(mp,
1206 (char *)dep - (char *)block));
1207 offset = (int)((char *)(tagp + 1) - (char *)block);
1208 if (++i == sfp->hdr.count)
1209 sfep = NULL;
1210 else
1211 sfep = XFS_DIR2_SF_NEXTENTRY(sfp, sfep);
1213 /* Done with the temporary buffer */
1214 kmem_free(buf, buf_len);
1216 * Sort the leaf entries by hash value.
1218 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1220 * Log the leaf entry area and tail.
1221 * Already logged the header in data_init, ignore needlog.
1223 ASSERT(needscan == 0);
1224 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1225 xfs_dir2_block_log_tail(tp, bp);
1226 xfs_dir2_data_check(dp, bp);
1227 xfs_da_buf_done(bp);
1228 return 0;