libata: make sure port is thawed when skipping resets
[linux-2.6/mini2440.git] / fs / xfs / xfs_dir2_block.c
blobe1f0a06aaf042c945e02f07b872190414122a2b7
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_ag.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_dir2_sf.h"
32 #include "xfs_attr_sf.h"
33 #include "xfs_dinode.h"
34 #include "xfs_inode.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_dir2_data.h"
37 #include "xfs_dir2_leaf.h"
38 #include "xfs_dir2_block.h"
39 #include "xfs_dir2_trace.h"
40 #include "xfs_error.h"
43 * Local function prototypes.
45 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, xfs_dabuf_t *bp, int first,
46 int last);
47 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, xfs_dabuf_t *bp);
48 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **bpp,
49 int *entno);
50 static int xfs_dir2_block_sort(const void *a, const void *b);
52 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
55 * One-time startup routine called from xfs_init().
57 void
58 xfs_dir_startup(void)
60 xfs_dir_hash_dot = xfs_da_hashname(".", 1);
61 xfs_dir_hash_dotdot = xfs_da_hashname("..", 2);
65 * Add an entry to a block directory.
67 int /* error */
68 xfs_dir2_block_addname(
69 xfs_da_args_t *args) /* directory op arguments */
71 xfs_dir2_data_free_t *bf; /* bestfree table in block */
72 xfs_dir2_block_t *block; /* directory block structure */
73 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
74 xfs_dabuf_t *bp; /* buffer for block */
75 xfs_dir2_block_tail_t *btp; /* block tail */
76 int compact; /* need to compact leaf ents */
77 xfs_dir2_data_entry_t *dep; /* block data entry */
78 xfs_inode_t *dp; /* directory inode */
79 xfs_dir2_data_unused_t *dup; /* block unused entry */
80 int error; /* error return value */
81 xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
82 xfs_dahash_t hash; /* hash value of found entry */
83 int high; /* high index for binary srch */
84 int highstale; /* high stale index */
85 int lfloghigh=0; /* last final leaf to log */
86 int lfloglow=0; /* first final leaf to log */
87 int len; /* length of the new entry */
88 int low; /* low index for binary srch */
89 int lowstale; /* low stale index */
90 int mid=0; /* midpoint for binary srch */
91 xfs_mount_t *mp; /* filesystem mount point */
92 int needlog; /* need to log header */
93 int needscan; /* need to rescan freespace */
94 __be16 *tagp; /* pointer to tag value */
95 xfs_trans_t *tp; /* transaction structure */
97 xfs_dir2_trace_args("block_addname", args);
98 dp = args->dp;
99 tp = args->trans;
100 mp = dp->i_mount;
102 * Read the (one and only) directory block into dabuf bp.
104 if ((error =
105 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
106 return error;
108 ASSERT(bp != NULL);
109 block = bp->data;
111 * Check the magic number, corrupted if wrong.
113 if (unlikely(be32_to_cpu(block->hdr.magic) != XFS_DIR2_BLOCK_MAGIC)) {
114 XFS_CORRUPTION_ERROR("xfs_dir2_block_addname",
115 XFS_ERRLEVEL_LOW, mp, block);
116 xfs_da_brelse(tp, bp);
117 return XFS_ERROR(EFSCORRUPTED);
119 len = xfs_dir2_data_entsize(args->namelen);
121 * Set up pointers to parts of the block.
123 bf = block->hdr.bestfree;
124 btp = xfs_dir2_block_tail_p(mp, block);
125 blp = xfs_dir2_block_leaf_p(btp);
127 * No stale entries? Need space for entry and new leaf.
129 if (!btp->stale) {
131 * Tag just before the first leaf entry.
133 tagp = (__be16 *)blp - 1;
135 * Data object just before the first leaf entry.
137 enddup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
139 * If it's not free then can't do this add without cleaning up:
140 * the space before the first leaf entry needs to be free so it
141 * can be expanded to hold the pointer to the new entry.
143 if (be16_to_cpu(enddup->freetag) != XFS_DIR2_DATA_FREE_TAG)
144 dup = enddup = NULL;
146 * Check out the biggest freespace and see if it's the same one.
148 else {
149 dup = (xfs_dir2_data_unused_t *)
150 ((char *)block + be16_to_cpu(bf[0].offset));
151 if (dup == enddup) {
153 * It is the biggest freespace, is it too small
154 * to hold the new leaf too?
156 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
158 * Yes, we use the second-largest
159 * entry instead if it works.
161 if (be16_to_cpu(bf[1].length) >= len)
162 dup = (xfs_dir2_data_unused_t *)
163 ((char *)block +
164 be16_to_cpu(bf[1].offset));
165 else
166 dup = NULL;
168 } else {
170 * Not the same free entry,
171 * just check its length.
173 if (be16_to_cpu(dup->length) < len) {
174 dup = NULL;
178 compact = 0;
181 * If there are stale entries we'll use one for the leaf.
182 * Is the biggest entry enough to avoid compaction?
184 else if (be16_to_cpu(bf[0].length) >= len) {
185 dup = (xfs_dir2_data_unused_t *)
186 ((char *)block + be16_to_cpu(bf[0].offset));
187 compact = 0;
190 * Will need to compact to make this work.
192 else {
194 * Tag just before the first leaf entry.
196 tagp = (__be16 *)blp - 1;
198 * Data object just before the first leaf entry.
200 dup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
202 * If it's not free then the data will go where the
203 * leaf data starts now, if it works at all.
205 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
206 if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
207 (uint)sizeof(*blp) < len)
208 dup = NULL;
209 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
210 dup = NULL;
211 else
212 dup = (xfs_dir2_data_unused_t *)blp;
213 compact = 1;
216 * If this isn't a real add, we're done with the buffer.
218 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
219 xfs_da_brelse(tp, bp);
221 * If we don't have space for the new entry & leaf ...
223 if (!dup) {
225 * Not trying to actually do anything, or don't have
226 * a space reservation: return no-space.
228 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
229 return XFS_ERROR(ENOSPC);
231 * Convert to the next larger format.
232 * Then add the new entry in that format.
234 error = xfs_dir2_block_to_leaf(args, bp);
235 xfs_da_buf_done(bp);
236 if (error)
237 return error;
238 return xfs_dir2_leaf_addname(args);
241 * Just checking, and it would work, so say so.
243 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
244 return 0;
245 needlog = needscan = 0;
247 * If need to compact the leaf entries, do it now.
248 * Leave the highest-numbered stale entry stale.
249 * XXX should be the one closest to mid but mid is not yet computed.
251 if (compact) {
252 int fromidx; /* source leaf index */
253 int toidx; /* target leaf index */
255 for (fromidx = toidx = be32_to_cpu(btp->count) - 1,
256 highstale = lfloghigh = -1;
257 fromidx >= 0;
258 fromidx--) {
259 if (be32_to_cpu(blp[fromidx].address) == XFS_DIR2_NULL_DATAPTR) {
260 if (highstale == -1)
261 highstale = toidx;
262 else {
263 if (lfloghigh == -1)
264 lfloghigh = toidx;
265 continue;
268 if (fromidx < toidx)
269 blp[toidx] = blp[fromidx];
270 toidx--;
272 lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
273 lfloghigh -= be32_to_cpu(btp->stale) - 1;
274 be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
275 xfs_dir2_data_make_free(tp, bp,
276 (xfs_dir2_data_aoff_t)((char *)blp - (char *)block),
277 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
278 &needlog, &needscan);
279 blp += be32_to_cpu(btp->stale) - 1;
280 btp->stale = cpu_to_be32(1);
282 * If we now need to rebuild the bestfree map, do so.
283 * This needs to happen before the next call to use_free.
285 if (needscan) {
286 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
287 needscan = 0;
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);
296 lfloghigh = -1;
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)
304 break;
305 if (hash < args->hashval)
306 low = mid + 1;
307 else
308 high = mid - 1;
310 while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
311 mid--;
314 * No stale entries, will use enddup space to hold new leaf.
316 if (!btp->stale) {
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) -
323 sizeof(*blp)),
324 (xfs_dir2_data_aoff_t)sizeof(*blp),
325 &needlog, &needscan);
327 * Update the tail (entry count).
329 be32_add_cpu(&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.
334 if (needscan) {
335 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block,
336 &needlog);
337 needscan = 0;
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.
344 blp--;
345 mid++;
346 if (mid)
347 memmove(blp, &blp[1], mid * sizeof(*blp));
348 lfloglow = 0;
349 lfloghigh = mid;
352 * Use a stale leaf for our new entry.
354 else {
355 for (lowstale = mid;
356 lowstale >= 0 &&
357 be32_to_cpu(blp[lowstale].address) != XFS_DIR2_NULL_DATAPTR;
358 lowstale--)
359 continue;
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);
364 highstale++)
365 continue;
367 * Move entries toward the low-numbered stale entry.
369 if (lowstale >= 0 &&
370 (highstale == be32_to_cpu(btp->count) ||
371 mid - lowstale <= highstale - mid)) {
372 if (mid - lowstale)
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.
381 else {
382 ASSERT(highstale < be32_to_cpu(btp->count));
383 mid++;
384 if (highstale - mid)
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_cpu(&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.
420 if (needscan)
421 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
422 if (needlog)
423 xfs_dir2_data_log_header(tp, bp);
424 xfs_dir2_block_log_tail(tp, bp);
425 xfs_dir2_data_log_entry(tp, bp, dep);
426 xfs_dir2_data_check(dp, bp);
427 xfs_da_buf_done(bp);
428 return 0;
432 * Readdir for block directories.
434 int /* error */
435 xfs_dir2_block_getdents(
436 xfs_inode_t *dp, /* incore inode */
437 void *dirent,
438 xfs_off_t *offset,
439 filldir_t filldir)
441 xfs_dir2_block_t *block; /* directory block structure */
442 xfs_dabuf_t *bp; /* buffer for block */
443 xfs_dir2_block_tail_t *btp; /* block tail */
444 xfs_dir2_data_entry_t *dep; /* block data entry */
445 xfs_dir2_data_unused_t *dup; /* block unused entry */
446 char *endptr; /* end of the data entries */
447 int error; /* error return value */
448 xfs_mount_t *mp; /* filesystem mount point */
449 char *ptr; /* current data entry */
450 int wantoff; /* starting block offset */
451 xfs_ino_t ino;
452 xfs_off_t cook;
454 mp = dp->i_mount;
456 * If the block number in the offset is out of range, we're done.
458 if (xfs_dir2_dataptr_to_db(mp, *offset) > mp->m_dirdatablk) {
459 return 0;
462 * Can't read the block, give up, else get dabuf in bp.
464 error = xfs_da_read_buf(NULL, dp, mp->m_dirdatablk, -1,
465 &bp, XFS_DATA_FORK);
466 if (error)
467 return error;
469 ASSERT(bp != NULL);
471 * Extract the byte offset we start at from the seek pointer.
472 * We'll skip entries before this.
474 wantoff = xfs_dir2_dataptr_to_off(mp, *offset);
475 block = bp->data;
476 xfs_dir2_data_check(dp, bp);
478 * Set up values for the loop.
480 btp = xfs_dir2_block_tail_p(mp, block);
481 ptr = (char *)block->u;
482 endptr = (char *)xfs_dir2_block_leaf_p(btp);
485 * Loop over the data portion of the block.
486 * Each object is a real entry (dep) or an unused one (dup).
488 while (ptr < endptr) {
489 dup = (xfs_dir2_data_unused_t *)ptr;
491 * Unused, skip it.
493 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
494 ptr += be16_to_cpu(dup->length);
495 continue;
498 dep = (xfs_dir2_data_entry_t *)ptr;
501 * Bump pointer for the next iteration.
503 ptr += xfs_dir2_data_entsize(dep->namelen);
505 * The entry is before the desired starting point, skip it.
507 if ((char *)dep - (char *)block < wantoff)
508 continue;
510 cook = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
511 (char *)dep - (char *)block);
512 ino = be64_to_cpu(dep->inumber);
513 #if XFS_BIG_INUMS
514 ino += mp->m_inoadd;
515 #endif
518 * If it didn't fit, set the final offset to here & return.
520 if (filldir(dirent, dep->name, dep->namelen, cook & 0x7fffffff,
521 ino, DT_UNKNOWN)) {
522 *offset = cook & 0x7fffffff;
523 xfs_da_brelse(NULL, bp);
524 return 0;
529 * Reached the end of the block.
530 * Set the offset to a non-existent block 1 and return.
532 *offset = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk + 1, 0) &
533 0x7fffffff;
534 xfs_da_brelse(NULL, bp);
535 return 0;
539 * Log leaf entries from the block.
541 static void
542 xfs_dir2_block_log_leaf(
543 xfs_trans_t *tp, /* transaction structure */
544 xfs_dabuf_t *bp, /* block buffer */
545 int first, /* index of first logged leaf */
546 int last) /* index of last logged leaf */
548 xfs_dir2_block_t *block; /* directory block structure */
549 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
550 xfs_dir2_block_tail_t *btp; /* block tail */
551 xfs_mount_t *mp; /* filesystem mount point */
553 mp = tp->t_mountp;
554 block = bp->data;
555 btp = xfs_dir2_block_tail_p(mp, block);
556 blp = xfs_dir2_block_leaf_p(btp);
557 xfs_da_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)block),
558 (uint)((char *)&blp[last + 1] - (char *)block - 1));
562 * Log the block tail.
564 static void
565 xfs_dir2_block_log_tail(
566 xfs_trans_t *tp, /* transaction structure */
567 xfs_dabuf_t *bp) /* block buffer */
569 xfs_dir2_block_t *block; /* directory block structure */
570 xfs_dir2_block_tail_t *btp; /* block tail */
571 xfs_mount_t *mp; /* filesystem mount point */
573 mp = tp->t_mountp;
574 block = bp->data;
575 btp = xfs_dir2_block_tail_p(mp, block);
576 xfs_da_log_buf(tp, bp, (uint)((char *)btp - (char *)block),
577 (uint)((char *)(btp + 1) - (char *)block - 1));
581 * Look up an entry in the block. This is the external routine,
582 * xfs_dir2_block_lookup_int does the real work.
584 int /* error */
585 xfs_dir2_block_lookup(
586 xfs_da_args_t *args) /* dir lookup arguments */
588 xfs_dir2_block_t *block; /* block structure */
589 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
590 xfs_dabuf_t *bp; /* block buffer */
591 xfs_dir2_block_tail_t *btp; /* block tail */
592 xfs_dir2_data_entry_t *dep; /* block data entry */
593 xfs_inode_t *dp; /* incore inode */
594 int ent; /* entry index */
595 int error; /* error return value */
596 xfs_mount_t *mp; /* filesystem mount point */
598 xfs_dir2_trace_args("block_lookup", args);
600 * Get the buffer, look up the entry.
601 * If not found (ENOENT) then return, have no buffer.
603 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
604 return error;
605 dp = args->dp;
606 mp = dp->i_mount;
607 block = bp->data;
608 xfs_dir2_data_check(dp, bp);
609 btp = xfs_dir2_block_tail_p(mp, block);
610 blp = xfs_dir2_block_leaf_p(btp);
612 * Get the offset from the leaf entry, to point to the data.
614 dep = (xfs_dir2_data_entry_t *)((char *)block +
615 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
617 * Fill in inode number, CI name if appropriate, release the block.
619 args->inumber = be64_to_cpu(dep->inumber);
620 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
621 xfs_da_brelse(args->trans, bp);
622 return XFS_ERROR(error);
626 * Internal block lookup routine.
628 static int /* error */
629 xfs_dir2_block_lookup_int(
630 xfs_da_args_t *args, /* dir lookup arguments */
631 xfs_dabuf_t **bpp, /* returned block buffer */
632 int *entno) /* returned entry number */
634 xfs_dir2_dataptr_t addr; /* data entry address */
635 xfs_dir2_block_t *block; /* block structure */
636 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
637 xfs_dabuf_t *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 error; /* error return value */
642 xfs_dahash_t hash; /* found hash value */
643 int high; /* binary search high index */
644 int low; /* binary search low index */
645 int mid; /* binary search current idx */
646 xfs_mount_t *mp; /* filesystem mount point */
647 xfs_trans_t *tp; /* transaction pointer */
648 enum xfs_dacmp cmp; /* comparison result */
650 dp = args->dp;
651 tp = args->trans;
652 mp = dp->i_mount;
654 * Read the buffer, return error if we can't get it.
656 if ((error =
657 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
658 return error;
660 ASSERT(bp != NULL);
661 block = bp->data;
662 xfs_dir2_data_check(dp, bp);
663 btp = xfs_dir2_block_tail_p(mp, block);
664 blp = xfs_dir2_block_leaf_p(btp);
666 * Loop doing a binary search for our hash value.
667 * Find our entry, ENOENT if it's not there.
669 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
670 ASSERT(low <= high);
671 mid = (low + high) >> 1;
672 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
673 break;
674 if (hash < args->hashval)
675 low = mid + 1;
676 else
677 high = mid - 1;
678 if (low > high) {
679 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
680 xfs_da_brelse(tp, bp);
681 return XFS_ERROR(ENOENT);
685 * Back up to the first one with the right hash value.
687 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
688 mid--;
691 * Now loop forward through all the entries with the
692 * right hash value looking for our name.
694 do {
695 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
696 continue;
698 * Get pointer to the entry from the leaf.
700 dep = (xfs_dir2_data_entry_t *)
701 ((char *)block + xfs_dir2_dataptr_to_off(mp, addr));
703 * Compare name and if it's an exact match, return the index
704 * and buffer. If it's the first case-insensitive match, store
705 * the index and buffer and continue looking for an exact match.
707 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
708 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
709 args->cmpresult = cmp;
710 *bpp = bp;
711 *entno = mid;
712 if (cmp == XFS_CMP_EXACT)
713 return 0;
715 } while (++mid < be32_to_cpu(btp->count) &&
716 be32_to_cpu(blp[mid].hashval) == hash);
718 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
720 * Here, we can only be doing a lookup (not a rename or replace).
721 * If a case-insensitive match was found earlier, return success.
723 if (args->cmpresult == XFS_CMP_CASE)
724 return 0;
726 * No match, release the buffer and return ENOENT.
728 xfs_da_brelse(tp, bp);
729 return XFS_ERROR(ENOENT);
733 * Remove an entry from a block format directory.
734 * If that makes the block small enough to fit in shortform, transform it.
736 int /* error */
737 xfs_dir2_block_removename(
738 xfs_da_args_t *args) /* directory operation args */
740 xfs_dir2_block_t *block; /* block structure */
741 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
742 xfs_dabuf_t *bp; /* block buffer */
743 xfs_dir2_block_tail_t *btp; /* block tail */
744 xfs_dir2_data_entry_t *dep; /* block data entry */
745 xfs_inode_t *dp; /* incore inode */
746 int ent; /* block leaf entry index */
747 int error; /* error return value */
748 xfs_mount_t *mp; /* filesystem mount point */
749 int needlog; /* need to log block header */
750 int needscan; /* need to fixup bestfree */
751 xfs_dir2_sf_hdr_t sfh; /* shortform header */
752 int size; /* shortform size */
753 xfs_trans_t *tp; /* transaction pointer */
755 xfs_dir2_trace_args("block_removename", args);
757 * Look up the entry in the block. Gets the buffer and entry index.
758 * It will always be there, the vnodeops level does a lookup first.
760 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
761 return error;
763 dp = args->dp;
764 tp = args->trans;
765 mp = dp->i_mount;
766 block = bp->data;
767 btp = xfs_dir2_block_tail_p(mp, block);
768 blp = xfs_dir2_block_leaf_p(btp);
770 * Point to the data entry using the leaf entry.
772 dep = (xfs_dir2_data_entry_t *)
773 ((char *)block + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
775 * Mark the data entry's space free.
777 needlog = needscan = 0;
778 xfs_dir2_data_make_free(tp, bp,
779 (xfs_dir2_data_aoff_t)((char *)dep - (char *)block),
780 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
782 * Fix up the block tail.
784 be32_add_cpu(&btp->stale, 1);
785 xfs_dir2_block_log_tail(tp, bp);
787 * Remove the leaf entry by marking it stale.
789 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
790 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
792 * Fix up bestfree, log the header if necessary.
794 if (needscan)
795 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
796 if (needlog)
797 xfs_dir2_data_log_header(tp, bp);
798 xfs_dir2_data_check(dp, bp);
800 * See if the size as a shortform is good enough.
802 if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
803 XFS_IFORK_DSIZE(dp)) {
804 xfs_da_buf_done(bp);
805 return 0;
808 * If it works, do the conversion.
810 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
814 * Replace an entry in a V2 block directory.
815 * Change the inode number to the new value.
817 int /* error */
818 xfs_dir2_block_replace(
819 xfs_da_args_t *args) /* directory operation args */
821 xfs_dir2_block_t *block; /* block structure */
822 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
823 xfs_dabuf_t *bp; /* block buffer */
824 xfs_dir2_block_tail_t *btp; /* block tail */
825 xfs_dir2_data_entry_t *dep; /* block data entry */
826 xfs_inode_t *dp; /* incore inode */
827 int ent; /* leaf entry index */
828 int error; /* error return value */
829 xfs_mount_t *mp; /* filesystem mount point */
831 xfs_dir2_trace_args("block_replace", args);
833 * Lookup the entry in the directory. Get buffer and entry index.
834 * This will always succeed since the caller has already done a lookup.
836 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
837 return error;
839 dp = args->dp;
840 mp = dp->i_mount;
841 block = bp->data;
842 btp = xfs_dir2_block_tail_p(mp, block);
843 blp = xfs_dir2_block_leaf_p(btp);
845 * Point to the data entry we need to change.
847 dep = (xfs_dir2_data_entry_t *)
848 ((char *)block + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
849 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
851 * Change the inode number to the new value.
853 dep->inumber = cpu_to_be64(args->inumber);
854 xfs_dir2_data_log_entry(args->trans, bp, dep);
855 xfs_dir2_data_check(dp, bp);
856 xfs_da_buf_done(bp);
857 return 0;
861 * Qsort comparison routine for the block leaf entries.
863 static int /* sort order */
864 xfs_dir2_block_sort(
865 const void *a, /* first leaf entry */
866 const void *b) /* second leaf entry */
868 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
869 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
871 la = a;
872 lb = b;
873 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
874 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
878 * Convert a V2 leaf directory to a V2 block directory if possible.
880 int /* error */
881 xfs_dir2_leaf_to_block(
882 xfs_da_args_t *args, /* operation arguments */
883 xfs_dabuf_t *lbp, /* leaf buffer */
884 xfs_dabuf_t *dbp) /* data buffer */
886 __be16 *bestsp; /* leaf bests table */
887 xfs_dir2_block_t *block; /* block structure */
888 xfs_dir2_block_tail_t *btp; /* block tail */
889 xfs_inode_t *dp; /* incore directory inode */
890 xfs_dir2_data_unused_t *dup; /* unused data entry */
891 int error; /* error return value */
892 int from; /* leaf from index */
893 xfs_dir2_leaf_t *leaf; /* leaf structure */
894 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
895 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
896 xfs_mount_t *mp; /* file system mount point */
897 int needlog; /* need to log data header */
898 int needscan; /* need to scan for bestfree */
899 xfs_dir2_sf_hdr_t sfh; /* shortform header */
900 int size; /* bytes used */
901 __be16 *tagp; /* end of entry (tag) */
902 int to; /* block/leaf to index */
903 xfs_trans_t *tp; /* transaction pointer */
905 xfs_dir2_trace_args_bb("leaf_to_block", args, lbp, dbp);
906 dp = args->dp;
907 tp = args->trans;
908 mp = dp->i_mount;
909 leaf = lbp->data;
910 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
911 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
913 * If there are data blocks other than the first one, take this
914 * opportunity to remove trailing empty data blocks that may have
915 * been left behind during no-space-reservation operations.
916 * These will show up in the leaf bests table.
918 while (dp->i_d.di_size > mp->m_dirblksize) {
919 bestsp = xfs_dir2_leaf_bests_p(ltp);
920 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
921 mp->m_dirblksize - (uint)sizeof(block->hdr)) {
922 if ((error =
923 xfs_dir2_leaf_trim_data(args, lbp,
924 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
925 goto out;
926 } else {
927 error = 0;
928 goto out;
932 * Read the data block if we don't already have it, give up if it fails.
934 if (dbp == NULL &&
935 (error = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &dbp,
936 XFS_DATA_FORK))) {
937 goto out;
939 block = dbp->data;
940 ASSERT(be32_to_cpu(block->hdr.magic) == XFS_DIR2_DATA_MAGIC);
942 * Size of the "leaf" area in the block.
944 size = (uint)sizeof(block->tail) +
945 (uint)sizeof(*lep) * (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
947 * Look at the last data entry.
949 tagp = (__be16 *)((char *)block + mp->m_dirblksize) - 1;
950 dup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
952 * If it's not free or is too short we can't do it.
954 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
955 be16_to_cpu(dup->length) < size) {
956 error = 0;
957 goto out;
960 * Start converting it to block form.
962 block->hdr.magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
963 needlog = 1;
964 needscan = 0;
966 * Use up the space at the end of the block (blp/btp).
968 xfs_dir2_data_use_free(tp, dbp, dup, mp->m_dirblksize - size, size,
969 &needlog, &needscan);
971 * Initialize the block tail.
973 btp = xfs_dir2_block_tail_p(mp, block);
974 btp->count = cpu_to_be32(be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
975 btp->stale = 0;
976 xfs_dir2_block_log_tail(tp, dbp);
978 * Initialize the block leaf area. We compact out stale entries.
980 lep = xfs_dir2_block_leaf_p(btp);
981 for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
982 if (be32_to_cpu(leaf->ents[from].address) == XFS_DIR2_NULL_DATAPTR)
983 continue;
984 lep[to++] = leaf->ents[from];
986 ASSERT(to == be32_to_cpu(btp->count));
987 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
989 * Scan the bestfree if we need it and log the data block header.
991 if (needscan)
992 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
993 if (needlog)
994 xfs_dir2_data_log_header(tp, dbp);
996 * Pitch the old leaf block.
998 error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
999 lbp = NULL;
1000 if (error) {
1001 goto out;
1004 * Now see if the resulting block can be shrunken to shortform.
1006 if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
1007 XFS_IFORK_DSIZE(dp)) {
1008 error = 0;
1009 goto out;
1011 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1012 out:
1013 if (lbp)
1014 xfs_da_buf_done(lbp);
1015 if (dbp)
1016 xfs_da_buf_done(dbp);
1017 return error;
1021 * Convert the shortform directory to block form.
1023 int /* error */
1024 xfs_dir2_sf_to_block(
1025 xfs_da_args_t *args) /* operation arguments */
1027 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1028 xfs_dir2_block_t *block; /* block structure */
1029 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1030 xfs_dabuf_t *bp; /* block buffer */
1031 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1032 char *buf; /* sf buffer */
1033 int buf_len;
1034 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1035 xfs_inode_t *dp; /* incore directory inode */
1036 int dummy; /* trash */
1037 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1038 int endoffset; /* end of data objects */
1039 int error; /* error return value */
1040 int i; /* index */
1041 xfs_mount_t *mp; /* filesystem mount point */
1042 int needlog; /* need to log block header */
1043 int needscan; /* need to scan block freespc */
1044 int newoffset; /* offset from current entry */
1045 int offset; /* target block offset */
1046 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1047 xfs_dir2_sf_t *sfp; /* shortform structure */
1048 __be16 *tagp; /* end of data entry */
1049 xfs_trans_t *tp; /* transaction pointer */
1050 struct xfs_name name;
1052 xfs_dir2_trace_args("sf_to_block", args);
1053 dp = args->dp;
1054 tp = args->trans;
1055 mp = dp->i_mount;
1056 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
1058 * Bomb out if the shortform directory is way too short.
1060 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1061 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1062 return XFS_ERROR(EIO);
1064 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
1065 ASSERT(dp->i_df.if_u1.if_data != NULL);
1066 sfp = (xfs_dir2_sf_t *)dp->i_df.if_u1.if_data;
1067 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->hdr.i8count));
1069 * Copy the directory into the stack buffer.
1070 * Then pitch the incore inode data so we can make extents.
1073 buf_len = dp->i_df.if_bytes;
1074 buf = kmem_alloc(dp->i_df.if_bytes, KM_SLEEP);
1076 memcpy(buf, sfp, dp->i_df.if_bytes);
1077 xfs_idata_realloc(dp, -dp->i_df.if_bytes, XFS_DATA_FORK);
1078 dp->i_d.di_size = 0;
1079 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1081 * Reset pointer - old sfp is gone.
1083 sfp = (xfs_dir2_sf_t *)buf;
1085 * Add block 0 to the inode.
1087 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1088 if (error) {
1089 kmem_free(buf);
1090 return error;
1093 * Initialize the data block.
1095 error = xfs_dir2_data_init(args, blkno, &bp);
1096 if (error) {
1097 kmem_free(buf);
1098 return error;
1100 block = bp->data;
1101 block->hdr.magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
1103 * Compute size of block "tail" area.
1105 i = (uint)sizeof(*btp) +
1106 (sfp->hdr.count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1108 * The whole thing is initialized to free by the init routine.
1109 * Say we're using the leaf and tail area.
1111 dup = (xfs_dir2_data_unused_t *)block->u;
1112 needlog = needscan = 0;
1113 xfs_dir2_data_use_free(tp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1114 &needscan);
1115 ASSERT(needscan == 0);
1117 * Fill in the tail.
1119 btp = xfs_dir2_block_tail_p(mp, block);
1120 btp->count = cpu_to_be32(sfp->hdr.count + 2); /* ., .. */
1121 btp->stale = 0;
1122 blp = xfs_dir2_block_leaf_p(btp);
1123 endoffset = (uint)((char *)blp - (char *)block);
1125 * Remove the freespace, we'll manage it.
1127 xfs_dir2_data_use_free(tp, bp, dup,
1128 (xfs_dir2_data_aoff_t)((char *)dup - (char *)block),
1129 be16_to_cpu(dup->length), &needlog, &needscan);
1131 * Create entry for .
1133 dep = (xfs_dir2_data_entry_t *)
1134 ((char *)block + XFS_DIR2_DATA_DOT_OFFSET);
1135 dep->inumber = cpu_to_be64(dp->i_ino);
1136 dep->namelen = 1;
1137 dep->name[0] = '.';
1138 tagp = xfs_dir2_data_entry_tag_p(dep);
1139 *tagp = cpu_to_be16((char *)dep - (char *)block);
1140 xfs_dir2_data_log_entry(tp, bp, dep);
1141 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1142 blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1143 (char *)dep - (char *)block));
1145 * Create entry for ..
1147 dep = (xfs_dir2_data_entry_t *)
1148 ((char *)block + XFS_DIR2_DATA_DOTDOT_OFFSET);
1149 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_inumber(sfp, &sfp->hdr.parent));
1150 dep->namelen = 2;
1151 dep->name[0] = dep->name[1] = '.';
1152 tagp = xfs_dir2_data_entry_tag_p(dep);
1153 *tagp = cpu_to_be16((char *)dep - (char *)block);
1154 xfs_dir2_data_log_entry(tp, bp, dep);
1155 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1156 blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1157 (char *)dep - (char *)block));
1158 offset = XFS_DIR2_DATA_FIRST_OFFSET;
1160 * Loop over existing entries, stuff them in.
1162 if ((i = 0) == sfp->hdr.count)
1163 sfep = NULL;
1164 else
1165 sfep = xfs_dir2_sf_firstentry(sfp);
1167 * Need to preserve the existing offset values in the sf directory.
1168 * Insert holes (unused entries) where necessary.
1170 while (offset < endoffset) {
1172 * sfep is null when we reach the end of the list.
1174 if (sfep == NULL)
1175 newoffset = endoffset;
1176 else
1177 newoffset = xfs_dir2_sf_get_offset(sfep);
1179 * There should be a hole here, make one.
1181 if (offset < newoffset) {
1182 dup = (xfs_dir2_data_unused_t *)
1183 ((char *)block + offset);
1184 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1185 dup->length = cpu_to_be16(newoffset - offset);
1186 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1187 ((char *)dup - (char *)block));
1188 xfs_dir2_data_log_unused(tp, bp, dup);
1189 (void)xfs_dir2_data_freeinsert((xfs_dir2_data_t *)block,
1190 dup, &dummy);
1191 offset += be16_to_cpu(dup->length);
1192 continue;
1195 * Copy a real entry.
1197 dep = (xfs_dir2_data_entry_t *)((char *)block + newoffset);
1198 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_inumber(sfp,
1199 xfs_dir2_sf_inumberp(sfep)));
1200 dep->namelen = sfep->namelen;
1201 memcpy(dep->name, sfep->name, dep->namelen);
1202 tagp = xfs_dir2_data_entry_tag_p(dep);
1203 *tagp = cpu_to_be16((char *)dep - (char *)block);
1204 xfs_dir2_data_log_entry(tp, bp, dep);
1205 name.name = sfep->name;
1206 name.len = sfep->namelen;
1207 blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1208 hashname(&name));
1209 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1210 (char *)dep - (char *)block));
1211 offset = (int)((char *)(tagp + 1) - (char *)block);
1212 if (++i == sfp->hdr.count)
1213 sfep = NULL;
1214 else
1215 sfep = xfs_dir2_sf_nextentry(sfp, sfep);
1217 /* Done with the temporary buffer */
1218 kmem_free(buf);
1220 * Sort the leaf entries by hash value.
1222 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1224 * Log the leaf entry area and tail.
1225 * Already logged the header in data_init, ignore needlog.
1227 ASSERT(needscan == 0);
1228 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1229 xfs_dir2_block_log_tail(tp, bp);
1230 xfs_dir2_data_check(dp, bp);
1231 xfs_da_buf_done(bp);
1232 return 0;