[XFS] kill the vfs_fsid and vfs_altfsid members in struct bhv_vfs
[linux-2.6/mini2440.git] / fs / xfs / xfs_dir2_block.c
blobc171767e242ad3b3d19c8c93ea4651f0e30146ea
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->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->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->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(&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(&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(&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 ptr - (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,
521 ino, DT_UNKNOWN)) {
522 *offset = xfs_dir2_db_off_to_dataptr(mp,
523 mp->m_dirdatablk,
524 (char *)dep - (char *)block);
525 xfs_da_brelse(NULL, bp);
526 return 0;
531 * Reached the end of the block.
532 * Set the offset to a non-existent block 1 and return.
534 *offset = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk + 1, 0);
535 xfs_da_brelse(NULL, bp);
536 return 0;
540 * Log leaf entries from the block.
542 static void
543 xfs_dir2_block_log_leaf(
544 xfs_trans_t *tp, /* transaction structure */
545 xfs_dabuf_t *bp, /* block buffer */
546 int first, /* index of first logged leaf */
547 int last) /* index of last logged leaf */
549 xfs_dir2_block_t *block; /* directory block structure */
550 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
551 xfs_dir2_block_tail_t *btp; /* block tail */
552 xfs_mount_t *mp; /* filesystem mount point */
554 mp = tp->t_mountp;
555 block = bp->data;
556 btp = xfs_dir2_block_tail_p(mp, block);
557 blp = xfs_dir2_block_leaf_p(btp);
558 xfs_da_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)block),
559 (uint)((char *)&blp[last + 1] - (char *)block - 1));
563 * Log the block tail.
565 static void
566 xfs_dir2_block_log_tail(
567 xfs_trans_t *tp, /* transaction structure */
568 xfs_dabuf_t *bp) /* block buffer */
570 xfs_dir2_block_t *block; /* directory block structure */
571 xfs_dir2_block_tail_t *btp; /* block tail */
572 xfs_mount_t *mp; /* filesystem mount point */
574 mp = tp->t_mountp;
575 block = bp->data;
576 btp = xfs_dir2_block_tail_p(mp, block);
577 xfs_da_log_buf(tp, bp, (uint)((char *)btp - (char *)block),
578 (uint)((char *)(btp + 1) - (char *)block - 1));
582 * Look up an entry in the block. This is the external routine,
583 * xfs_dir2_block_lookup_int does the real work.
585 int /* error */
586 xfs_dir2_block_lookup(
587 xfs_da_args_t *args) /* dir lookup arguments */
589 xfs_dir2_block_t *block; /* block structure */
590 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
591 xfs_dabuf_t *bp; /* block buffer */
592 xfs_dir2_block_tail_t *btp; /* block tail */
593 xfs_dir2_data_entry_t *dep; /* block data entry */
594 xfs_inode_t *dp; /* incore inode */
595 int ent; /* entry index */
596 int error; /* error return value */
597 xfs_mount_t *mp; /* filesystem mount point */
599 xfs_dir2_trace_args("block_lookup", args);
601 * Get the buffer, look up the entry.
602 * If not found (ENOENT) then return, have no buffer.
604 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
605 return error;
606 dp = args->dp;
607 mp = dp->i_mount;
608 block = bp->data;
609 xfs_dir2_data_check(dp, bp);
610 btp = xfs_dir2_block_tail_p(mp, block);
611 blp = xfs_dir2_block_leaf_p(btp);
613 * Get the offset from the leaf entry, to point to the data.
615 dep = (xfs_dir2_data_entry_t *)
616 ((char *)block + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
618 * Fill in inode number, release the block.
620 args->inumber = be64_to_cpu(dep->inumber);
621 xfs_da_brelse(args->trans, bp);
622 return XFS_ERROR(EEXIST);
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 */
649 dp = args->dp;
650 tp = args->trans;
651 mp = dp->i_mount;
653 * Read the buffer, return error if we can't get it.
655 if ((error =
656 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
657 return error;
659 ASSERT(bp != NULL);
660 block = bp->data;
661 xfs_dir2_data_check(dp, bp);
662 btp = xfs_dir2_block_tail_p(mp, block);
663 blp = xfs_dir2_block_leaf_p(btp);
665 * Loop doing a binary search for our hash value.
666 * Find our entry, ENOENT if it's not there.
668 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
669 ASSERT(low <= high);
670 mid = (low + high) >> 1;
671 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
672 break;
673 if (hash < args->hashval)
674 low = mid + 1;
675 else
676 high = mid - 1;
677 if (low > high) {
678 ASSERT(args->oknoent);
679 xfs_da_brelse(tp, bp);
680 return XFS_ERROR(ENOENT);
684 * Back up to the first one with the right hash value.
686 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
687 mid--;
690 * Now loop forward through all the entries with the
691 * right hash value looking for our name.
693 do {
694 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
695 continue;
697 * Get pointer to the entry from the leaf.
699 dep = (xfs_dir2_data_entry_t *)
700 ((char *)block + xfs_dir2_dataptr_to_off(mp, addr));
702 * Compare, if it's right give back buffer & entry number.
704 if (dep->namelen == args->namelen &&
705 dep->name[0] == args->name[0] &&
706 memcmp(dep->name, args->name, args->namelen) == 0) {
707 *bpp = bp;
708 *entno = mid;
709 return 0;
711 } while (++mid < be32_to_cpu(btp->count) && be32_to_cpu(blp[mid].hashval) == hash);
713 * No match, release the buffer and return ENOENT.
715 ASSERT(args->oknoent);
716 xfs_da_brelse(tp, bp);
717 return XFS_ERROR(ENOENT);
721 * Remove an entry from a block format directory.
722 * If that makes the block small enough to fit in shortform, transform it.
724 int /* error */
725 xfs_dir2_block_removename(
726 xfs_da_args_t *args) /* directory operation args */
728 xfs_dir2_block_t *block; /* block structure */
729 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
730 xfs_dabuf_t *bp; /* block buffer */
731 xfs_dir2_block_tail_t *btp; /* block tail */
732 xfs_dir2_data_entry_t *dep; /* block data entry */
733 xfs_inode_t *dp; /* incore inode */
734 int ent; /* block leaf entry index */
735 int error; /* error return value */
736 xfs_mount_t *mp; /* filesystem mount point */
737 int needlog; /* need to log block header */
738 int needscan; /* need to fixup bestfree */
739 xfs_dir2_sf_hdr_t sfh; /* shortform header */
740 int size; /* shortform size */
741 xfs_trans_t *tp; /* transaction pointer */
743 xfs_dir2_trace_args("block_removename", args);
745 * Look up the entry in the block. Gets the buffer and entry index.
746 * It will always be there, the vnodeops level does a lookup first.
748 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
749 return error;
751 dp = args->dp;
752 tp = args->trans;
753 mp = dp->i_mount;
754 block = bp->data;
755 btp = xfs_dir2_block_tail_p(mp, block);
756 blp = xfs_dir2_block_leaf_p(btp);
758 * Point to the data entry using the leaf entry.
760 dep = (xfs_dir2_data_entry_t *)
761 ((char *)block + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
763 * Mark the data entry's space free.
765 needlog = needscan = 0;
766 xfs_dir2_data_make_free(tp, bp,
767 (xfs_dir2_data_aoff_t)((char *)dep - (char *)block),
768 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
770 * Fix up the block tail.
772 be32_add(&btp->stale, 1);
773 xfs_dir2_block_log_tail(tp, bp);
775 * Remove the leaf entry by marking it stale.
777 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
778 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
780 * Fix up bestfree, log the header if necessary.
782 if (needscan)
783 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
784 if (needlog)
785 xfs_dir2_data_log_header(tp, bp);
786 xfs_dir2_data_check(dp, bp);
788 * See if the size as a shortform is good enough.
790 if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
791 XFS_IFORK_DSIZE(dp)) {
792 xfs_da_buf_done(bp);
793 return 0;
796 * If it works, do the conversion.
798 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
802 * Replace an entry in a V2 block directory.
803 * Change the inode number to the new value.
805 int /* error */
806 xfs_dir2_block_replace(
807 xfs_da_args_t *args) /* directory operation args */
809 xfs_dir2_block_t *block; /* block structure */
810 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
811 xfs_dabuf_t *bp; /* block buffer */
812 xfs_dir2_block_tail_t *btp; /* block tail */
813 xfs_dir2_data_entry_t *dep; /* block data entry */
814 xfs_inode_t *dp; /* incore inode */
815 int ent; /* leaf entry index */
816 int error; /* error return value */
817 xfs_mount_t *mp; /* filesystem mount point */
819 xfs_dir2_trace_args("block_replace", args);
821 * Lookup the entry in the directory. Get buffer and entry index.
822 * This will always succeed since the caller has already done a lookup.
824 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
825 return error;
827 dp = args->dp;
828 mp = dp->i_mount;
829 block = bp->data;
830 btp = xfs_dir2_block_tail_p(mp, block);
831 blp = xfs_dir2_block_leaf_p(btp);
833 * Point to the data entry we need to change.
835 dep = (xfs_dir2_data_entry_t *)
836 ((char *)block + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
837 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
839 * Change the inode number to the new value.
841 dep->inumber = cpu_to_be64(args->inumber);
842 xfs_dir2_data_log_entry(args->trans, bp, dep);
843 xfs_dir2_data_check(dp, bp);
844 xfs_da_buf_done(bp);
845 return 0;
849 * Qsort comparison routine for the block leaf entries.
851 static int /* sort order */
852 xfs_dir2_block_sort(
853 const void *a, /* first leaf entry */
854 const void *b) /* second leaf entry */
856 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
857 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
859 la = a;
860 lb = b;
861 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
862 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
866 * Convert a V2 leaf directory to a V2 block directory if possible.
868 int /* error */
869 xfs_dir2_leaf_to_block(
870 xfs_da_args_t *args, /* operation arguments */
871 xfs_dabuf_t *lbp, /* leaf buffer */
872 xfs_dabuf_t *dbp) /* data buffer */
874 __be16 *bestsp; /* leaf bests table */
875 xfs_dir2_block_t *block; /* block structure */
876 xfs_dir2_block_tail_t *btp; /* block tail */
877 xfs_inode_t *dp; /* incore directory inode */
878 xfs_dir2_data_unused_t *dup; /* unused data entry */
879 int error; /* error return value */
880 int from; /* leaf from index */
881 xfs_dir2_leaf_t *leaf; /* leaf structure */
882 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
883 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
884 xfs_mount_t *mp; /* file system mount point */
885 int needlog; /* need to log data header */
886 int needscan; /* need to scan for bestfree */
887 xfs_dir2_sf_hdr_t sfh; /* shortform header */
888 int size; /* bytes used */
889 __be16 *tagp; /* end of entry (tag) */
890 int to; /* block/leaf to index */
891 xfs_trans_t *tp; /* transaction pointer */
893 xfs_dir2_trace_args_bb("leaf_to_block", args, lbp, dbp);
894 dp = args->dp;
895 tp = args->trans;
896 mp = dp->i_mount;
897 leaf = lbp->data;
898 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
899 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
901 * If there are data blocks other than the first one, take this
902 * opportunity to remove trailing empty data blocks that may have
903 * been left behind during no-space-reservation operations.
904 * These will show up in the leaf bests table.
906 while (dp->i_d.di_size > mp->m_dirblksize) {
907 bestsp = xfs_dir2_leaf_bests_p(ltp);
908 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
909 mp->m_dirblksize - (uint)sizeof(block->hdr)) {
910 if ((error =
911 xfs_dir2_leaf_trim_data(args, lbp,
912 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
913 goto out;
914 } else {
915 error = 0;
916 goto out;
920 * Read the data block if we don't already have it, give up if it fails.
922 if (dbp == NULL &&
923 (error = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &dbp,
924 XFS_DATA_FORK))) {
925 goto out;
927 block = dbp->data;
928 ASSERT(be32_to_cpu(block->hdr.magic) == XFS_DIR2_DATA_MAGIC);
930 * Size of the "leaf" area in the block.
932 size = (uint)sizeof(block->tail) +
933 (uint)sizeof(*lep) * (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
935 * Look at the last data entry.
937 tagp = (__be16 *)((char *)block + mp->m_dirblksize) - 1;
938 dup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
940 * If it's not free or is too short we can't do it.
942 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
943 be16_to_cpu(dup->length) < size) {
944 error = 0;
945 goto out;
948 * Start converting it to block form.
950 block->hdr.magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
951 needlog = 1;
952 needscan = 0;
954 * Use up the space at the end of the block (blp/btp).
956 xfs_dir2_data_use_free(tp, dbp, dup, mp->m_dirblksize - size, size,
957 &needlog, &needscan);
959 * Initialize the block tail.
961 btp = xfs_dir2_block_tail_p(mp, block);
962 btp->count = cpu_to_be32(be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
963 btp->stale = 0;
964 xfs_dir2_block_log_tail(tp, dbp);
966 * Initialize the block leaf area. We compact out stale entries.
968 lep = xfs_dir2_block_leaf_p(btp);
969 for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
970 if (be32_to_cpu(leaf->ents[from].address) == XFS_DIR2_NULL_DATAPTR)
971 continue;
972 lep[to++] = leaf->ents[from];
974 ASSERT(to == be32_to_cpu(btp->count));
975 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
977 * Scan the bestfree if we need it and log the data block header.
979 if (needscan)
980 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
981 if (needlog)
982 xfs_dir2_data_log_header(tp, dbp);
984 * Pitch the old leaf block.
986 error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
987 lbp = NULL;
988 if (error) {
989 goto out;
992 * Now see if the resulting block can be shrunken to shortform.
994 if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
995 XFS_IFORK_DSIZE(dp)) {
996 error = 0;
997 goto out;
999 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1000 out:
1001 if (lbp)
1002 xfs_da_buf_done(lbp);
1003 if (dbp)
1004 xfs_da_buf_done(dbp);
1005 return error;
1009 * Convert the shortform directory to block form.
1011 int /* error */
1012 xfs_dir2_sf_to_block(
1013 xfs_da_args_t *args) /* operation arguments */
1015 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1016 xfs_dir2_block_t *block; /* block structure */
1017 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1018 xfs_dabuf_t *bp; /* block buffer */
1019 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1020 char *buf; /* sf buffer */
1021 int buf_len;
1022 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1023 xfs_inode_t *dp; /* incore directory inode */
1024 int dummy; /* trash */
1025 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1026 int endoffset; /* end of data objects */
1027 int error; /* error return value */
1028 int i; /* index */
1029 xfs_mount_t *mp; /* filesystem mount point */
1030 int needlog; /* need to log block header */
1031 int needscan; /* need to scan block freespc */
1032 int newoffset; /* offset from current entry */
1033 int offset; /* target block offset */
1034 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1035 xfs_dir2_sf_t *sfp; /* shortform structure */
1036 __be16 *tagp; /* end of data entry */
1037 xfs_trans_t *tp; /* transaction pointer */
1039 xfs_dir2_trace_args("sf_to_block", args);
1040 dp = args->dp;
1041 tp = args->trans;
1042 mp = dp->i_mount;
1043 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
1045 * Bomb out if the shortform directory is way too short.
1047 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1048 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1049 return XFS_ERROR(EIO);
1051 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
1052 ASSERT(dp->i_df.if_u1.if_data != NULL);
1053 sfp = (xfs_dir2_sf_t *)dp->i_df.if_u1.if_data;
1054 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->hdr.i8count));
1056 * Copy the directory into the stack buffer.
1057 * Then pitch the incore inode data so we can make extents.
1060 buf_len = dp->i_df.if_bytes;
1061 buf = kmem_alloc(dp->i_df.if_bytes, KM_SLEEP);
1063 memcpy(buf, sfp, dp->i_df.if_bytes);
1064 xfs_idata_realloc(dp, -dp->i_df.if_bytes, XFS_DATA_FORK);
1065 dp->i_d.di_size = 0;
1066 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1068 * Reset pointer - old sfp is gone.
1070 sfp = (xfs_dir2_sf_t *)buf;
1072 * Add block 0 to the inode.
1074 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1075 if (error) {
1076 kmem_free(buf, buf_len);
1077 return error;
1080 * Initialize the data block.
1082 error = xfs_dir2_data_init(args, blkno, &bp);
1083 if (error) {
1084 kmem_free(buf, buf_len);
1085 return error;
1087 block = bp->data;
1088 block->hdr.magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
1090 * Compute size of block "tail" area.
1092 i = (uint)sizeof(*btp) +
1093 (sfp->hdr.count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1095 * The whole thing is initialized to free by the init routine.
1096 * Say we're using the leaf and tail area.
1098 dup = (xfs_dir2_data_unused_t *)block->u;
1099 needlog = needscan = 0;
1100 xfs_dir2_data_use_free(tp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1101 &needscan);
1102 ASSERT(needscan == 0);
1104 * Fill in the tail.
1106 btp = xfs_dir2_block_tail_p(mp, block);
1107 btp->count = cpu_to_be32(sfp->hdr.count + 2); /* ., .. */
1108 btp->stale = 0;
1109 blp = xfs_dir2_block_leaf_p(btp);
1110 endoffset = (uint)((char *)blp - (char *)block);
1112 * Remove the freespace, we'll manage it.
1114 xfs_dir2_data_use_free(tp, bp, dup,
1115 (xfs_dir2_data_aoff_t)((char *)dup - (char *)block),
1116 be16_to_cpu(dup->length), &needlog, &needscan);
1118 * Create entry for .
1120 dep = (xfs_dir2_data_entry_t *)
1121 ((char *)block + XFS_DIR2_DATA_DOT_OFFSET);
1122 dep->inumber = cpu_to_be64(dp->i_ino);
1123 dep->namelen = 1;
1124 dep->name[0] = '.';
1125 tagp = xfs_dir2_data_entry_tag_p(dep);
1126 *tagp = cpu_to_be16((char *)dep - (char *)block);
1127 xfs_dir2_data_log_entry(tp, bp, dep);
1128 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1129 blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1130 (char *)dep - (char *)block));
1132 * Create entry for ..
1134 dep = (xfs_dir2_data_entry_t *)
1135 ((char *)block + XFS_DIR2_DATA_DOTDOT_OFFSET);
1136 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_inumber(sfp, &sfp->hdr.parent));
1137 dep->namelen = 2;
1138 dep->name[0] = dep->name[1] = '.';
1139 tagp = xfs_dir2_data_entry_tag_p(dep);
1140 *tagp = cpu_to_be16((char *)dep - (char *)block);
1141 xfs_dir2_data_log_entry(tp, bp, dep);
1142 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1143 blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1144 (char *)dep - (char *)block));
1145 offset = XFS_DIR2_DATA_FIRST_OFFSET;
1147 * Loop over existing entries, stuff them in.
1149 if ((i = 0) == sfp->hdr.count)
1150 sfep = NULL;
1151 else
1152 sfep = xfs_dir2_sf_firstentry(sfp);
1154 * Need to preserve the existing offset values in the sf directory.
1155 * Insert holes (unused entries) where necessary.
1157 while (offset < endoffset) {
1159 * sfep is null when we reach the end of the list.
1161 if (sfep == NULL)
1162 newoffset = endoffset;
1163 else
1164 newoffset = xfs_dir2_sf_get_offset(sfep);
1166 * There should be a hole here, make one.
1168 if (offset < newoffset) {
1169 dup = (xfs_dir2_data_unused_t *)
1170 ((char *)block + offset);
1171 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1172 dup->length = cpu_to_be16(newoffset - offset);
1173 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1174 ((char *)dup - (char *)block));
1175 xfs_dir2_data_log_unused(tp, bp, dup);
1176 (void)xfs_dir2_data_freeinsert((xfs_dir2_data_t *)block,
1177 dup, &dummy);
1178 offset += be16_to_cpu(dup->length);
1179 continue;
1182 * Copy a real entry.
1184 dep = (xfs_dir2_data_entry_t *)((char *)block + newoffset);
1185 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_inumber(sfp,
1186 xfs_dir2_sf_inumberp(sfep)));
1187 dep->namelen = sfep->namelen;
1188 memcpy(dep->name, sfep->name, dep->namelen);
1189 tagp = xfs_dir2_data_entry_tag_p(dep);
1190 *tagp = cpu_to_be16((char *)dep - (char *)block);
1191 xfs_dir2_data_log_entry(tp, bp, dep);
1192 blp[2 + i].hashval = cpu_to_be32(xfs_da_hashname(
1193 (char *)sfep->name, sfep->namelen));
1194 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1195 (char *)dep - (char *)block));
1196 offset = (int)((char *)(tagp + 1) - (char *)block);
1197 if (++i == sfp->hdr.count)
1198 sfep = NULL;
1199 else
1200 sfep = xfs_dir2_sf_nextentry(sfp, sfep);
1202 /* Done with the temporary buffer */
1203 kmem_free(buf, buf_len);
1205 * Sort the leaf entries by hash value.
1207 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1209 * Log the leaf entry area and tail.
1210 * Already logged the header in data_init, ignore needlog.
1212 ASSERT(needscan == 0);
1213 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1214 xfs_dir2_block_log_tail(tp, bp);
1215 xfs_dir2_data_check(dp, bp);
1216 xfs_da_buf_done(bp);
1217 return 0;