[TCP]: Left out sync->verify (the new meaning of it) & definify
[linux-2.6/mini2440.git] / fs / xfs / xfs_dir2_block.c
blobe4df1aaae2a2083c911d864d102641e6da80644f
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_dir2.h"
26 #include "xfs_dmapi.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_dir2_sf.h"
31 #include "xfs_attr_sf.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_inode_item.h"
35 #include "xfs_dir2_data.h"
36 #include "xfs_dir2_leaf.h"
37 #include "xfs_dir2_block.h"
38 #include "xfs_dir2_trace.h"
39 #include "xfs_error.h"
42 * Local function prototypes.
44 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, xfs_dabuf_t *bp, int first,
45 int last);
46 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, xfs_dabuf_t *bp);
47 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **bpp,
48 int *entno);
49 static int xfs_dir2_block_sort(const void *a, const void *b);
51 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
54 * One-time startup routine called from xfs_init().
56 void
57 xfs_dir_startup(void)
59 xfs_dir_hash_dot = xfs_da_hashname(".", 1);
60 xfs_dir_hash_dotdot = xfs_da_hashname("..", 2);
64 * Add an entry to a block directory.
66 int /* error */
67 xfs_dir2_block_addname(
68 xfs_da_args_t *args) /* directory op arguments */
70 xfs_dir2_data_free_t *bf; /* bestfree table in block */
71 xfs_dir2_block_t *block; /* directory block structure */
72 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
73 xfs_dabuf_t *bp; /* buffer for block */
74 xfs_dir2_block_tail_t *btp; /* block tail */
75 int compact; /* need to compact leaf ents */
76 xfs_dir2_data_entry_t *dep; /* block data entry */
77 xfs_inode_t *dp; /* directory inode */
78 xfs_dir2_data_unused_t *dup; /* block unused entry */
79 int error; /* error return value */
80 xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
81 xfs_dahash_t hash; /* hash value of found entry */
82 int high; /* high index for binary srch */
83 int highstale; /* high stale index */
84 int lfloghigh=0; /* last final leaf to log */
85 int lfloglow=0; /* first final leaf to log */
86 int len; /* length of the new entry */
87 int low; /* low index for binary srch */
88 int lowstale; /* low stale index */
89 int mid=0; /* midpoint for binary srch */
90 xfs_mount_t *mp; /* filesystem mount point */
91 int needlog; /* need to log header */
92 int needscan; /* need to rescan freespace */
93 __be16 *tagp; /* pointer to tag value */
94 xfs_trans_t *tp; /* transaction structure */
96 xfs_dir2_trace_args("block_addname", args);
97 dp = args->dp;
98 tp = args->trans;
99 mp = dp->i_mount;
101 * Read the (one and only) directory block into dabuf bp.
103 if ((error =
104 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
105 return error;
107 ASSERT(bp != NULL);
108 block = bp->data;
110 * Check the magic number, corrupted if wrong.
112 if (unlikely(be32_to_cpu(block->hdr.magic) != XFS_DIR2_BLOCK_MAGIC)) {
113 XFS_CORRUPTION_ERROR("xfs_dir2_block_addname",
114 XFS_ERRLEVEL_LOW, mp, block);
115 xfs_da_brelse(tp, bp);
116 return XFS_ERROR(EFSCORRUPTED);
118 len = xfs_dir2_data_entsize(args->namelen);
120 * Set up pointers to parts of the block.
122 bf = block->hdr.bestfree;
123 btp = xfs_dir2_block_tail_p(mp, block);
124 blp = xfs_dir2_block_leaf_p(btp);
126 * No stale entries? Need space for entry and new leaf.
128 if (!btp->stale) {
130 * Tag just before the first leaf entry.
132 tagp = (__be16 *)blp - 1;
134 * Data object just before the first leaf entry.
136 enddup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
138 * If it's not free then can't do this add without cleaning up:
139 * the space before the first leaf entry needs to be free so it
140 * can be expanded to hold the pointer to the new entry.
142 if (be16_to_cpu(enddup->freetag) != XFS_DIR2_DATA_FREE_TAG)
143 dup = enddup = NULL;
145 * Check out the biggest freespace and see if it's the same one.
147 else {
148 dup = (xfs_dir2_data_unused_t *)
149 ((char *)block + be16_to_cpu(bf[0].offset));
150 if (dup == enddup) {
152 * It is the biggest freespace, is it too small
153 * to hold the new leaf too?
155 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
157 * Yes, we use the second-largest
158 * entry instead if it works.
160 if (be16_to_cpu(bf[1].length) >= len)
161 dup = (xfs_dir2_data_unused_t *)
162 ((char *)block +
163 be16_to_cpu(bf[1].offset));
164 else
165 dup = NULL;
167 } else {
169 * Not the same free entry,
170 * just check its length.
172 if (be16_to_cpu(dup->length) < len) {
173 dup = NULL;
177 compact = 0;
180 * If there are stale entries we'll use one for the leaf.
181 * Is the biggest entry enough to avoid compaction?
183 else if (be16_to_cpu(bf[0].length) >= len) {
184 dup = (xfs_dir2_data_unused_t *)
185 ((char *)block + be16_to_cpu(bf[0].offset));
186 compact = 0;
189 * Will need to compact to make this work.
191 else {
193 * Tag just before the first leaf entry.
195 tagp = (__be16 *)blp - 1;
197 * Data object just before the first leaf entry.
199 dup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
201 * If it's not free then the data will go where the
202 * leaf data starts now, if it works at all.
204 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
205 if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
206 (uint)sizeof(*blp) < len)
207 dup = NULL;
208 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
209 dup = NULL;
210 else
211 dup = (xfs_dir2_data_unused_t *)blp;
212 compact = 1;
215 * If this isn't a real add, we're done with the buffer.
217 if (args->justcheck)
218 xfs_da_brelse(tp, bp);
220 * If we don't have space for the new entry & leaf ...
222 if (!dup) {
224 * Not trying to actually do anything, or don't have
225 * a space reservation: return no-space.
227 if (args->justcheck || args->total == 0)
228 return XFS_ERROR(ENOSPC);
230 * Convert to the next larger format.
231 * Then add the new entry in that format.
233 error = xfs_dir2_block_to_leaf(args, bp);
234 xfs_da_buf_done(bp);
235 if (error)
236 return error;
237 return xfs_dir2_leaf_addname(args);
240 * Just checking, and it would work, so say so.
242 if (args->justcheck)
243 return 0;
244 needlog = needscan = 0;
246 * If need to compact the leaf entries, do it now.
247 * Leave the highest-numbered stale entry stale.
248 * XXX should be the one closest to mid but mid is not yet computed.
250 if (compact) {
251 int fromidx; /* source leaf index */
252 int toidx; /* target leaf index */
254 for (fromidx = toidx = be32_to_cpu(btp->count) - 1,
255 highstale = lfloghigh = -1;
256 fromidx >= 0;
257 fromidx--) {
258 if (be32_to_cpu(blp[fromidx].address) == XFS_DIR2_NULL_DATAPTR) {
259 if (highstale == -1)
260 highstale = toidx;
261 else {
262 if (lfloghigh == -1)
263 lfloghigh = toidx;
264 continue;
267 if (fromidx < toidx)
268 blp[toidx] = blp[fromidx];
269 toidx--;
271 lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
272 lfloghigh -= be32_to_cpu(btp->stale) - 1;
273 be32_add(&btp->count, -(be32_to_cpu(btp->stale) - 1));
274 xfs_dir2_data_make_free(tp, bp,
275 (xfs_dir2_data_aoff_t)((char *)blp - (char *)block),
276 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
277 &needlog, &needscan);
278 blp += be32_to_cpu(btp->stale) - 1;
279 btp->stale = cpu_to_be32(1);
281 * If we now need to rebuild the bestfree map, do so.
282 * This needs to happen before the next call to use_free.
284 if (needscan) {
285 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
286 needscan = 0;
290 * Set leaf logging boundaries to impossible state.
291 * For the no-stale case they're set explicitly.
293 else if (btp->stale) {
294 lfloglow = be32_to_cpu(btp->count);
295 lfloghigh = -1;
298 * Find the slot that's first lower than our hash value, -1 if none.
300 for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
301 mid = (low + high) >> 1;
302 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
303 break;
304 if (hash < args->hashval)
305 low = mid + 1;
306 else
307 high = mid - 1;
309 while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
310 mid--;
313 * No stale entries, will use enddup space to hold new leaf.
315 if (!btp->stale) {
317 * Mark the space needed for the new leaf entry, now in use.
319 xfs_dir2_data_use_free(tp, bp, enddup,
320 (xfs_dir2_data_aoff_t)
321 ((char *)enddup - (char *)block + be16_to_cpu(enddup->length) -
322 sizeof(*blp)),
323 (xfs_dir2_data_aoff_t)sizeof(*blp),
324 &needlog, &needscan);
326 * Update the tail (entry count).
328 be32_add(&btp->count, 1);
330 * If we now need to rebuild the bestfree map, do so.
331 * This needs to happen before the next call to use_free.
333 if (needscan) {
334 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block,
335 &needlog);
336 needscan = 0;
339 * Adjust pointer to the first leaf entry, we're about to move
340 * the table up one to open up space for the new leaf entry.
341 * Then adjust our index to match.
343 blp--;
344 mid++;
345 if (mid)
346 memmove(blp, &blp[1], mid * sizeof(*blp));
347 lfloglow = 0;
348 lfloghigh = mid;
351 * Use a stale leaf for our new entry.
353 else {
354 for (lowstale = mid;
355 lowstale >= 0 &&
356 be32_to_cpu(blp[lowstale].address) != XFS_DIR2_NULL_DATAPTR;
357 lowstale--)
358 continue;
359 for (highstale = mid + 1;
360 highstale < be32_to_cpu(btp->count) &&
361 be32_to_cpu(blp[highstale].address) != XFS_DIR2_NULL_DATAPTR &&
362 (lowstale < 0 || mid - lowstale > highstale - mid);
363 highstale++)
364 continue;
366 * Move entries toward the low-numbered stale entry.
368 if (lowstale >= 0 &&
369 (highstale == be32_to_cpu(btp->count) ||
370 mid - lowstale <= highstale - mid)) {
371 if (mid - lowstale)
372 memmove(&blp[lowstale], &blp[lowstale + 1],
373 (mid - lowstale) * sizeof(*blp));
374 lfloglow = MIN(lowstale, lfloglow);
375 lfloghigh = MAX(mid, lfloghigh);
378 * Move entries toward the high-numbered stale entry.
380 else {
381 ASSERT(highstale < be32_to_cpu(btp->count));
382 mid++;
383 if (highstale - mid)
384 memmove(&blp[mid + 1], &blp[mid],
385 (highstale - mid) * sizeof(*blp));
386 lfloglow = MIN(mid, lfloglow);
387 lfloghigh = MAX(highstale, lfloghigh);
389 be32_add(&btp->stale, -1);
392 * Point to the new data entry.
394 dep = (xfs_dir2_data_entry_t *)dup;
396 * Fill in the leaf entry.
398 blp[mid].hashval = cpu_to_be32(args->hashval);
399 blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
400 (char *)dep - (char *)block));
401 xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
403 * Mark space for the data entry used.
405 xfs_dir2_data_use_free(tp, bp, dup,
406 (xfs_dir2_data_aoff_t)((char *)dup - (char *)block),
407 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
409 * Create the new data entry.
411 dep->inumber = cpu_to_be64(args->inumber);
412 dep->namelen = args->namelen;
413 memcpy(dep->name, args->name, args->namelen);
414 tagp = xfs_dir2_data_entry_tag_p(dep);
415 *tagp = cpu_to_be16((char *)dep - (char *)block);
417 * Clean up the bestfree array and log the header, tail, and entry.
419 if (needscan)
420 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
421 if (needlog)
422 xfs_dir2_data_log_header(tp, bp);
423 xfs_dir2_block_log_tail(tp, bp);
424 xfs_dir2_data_log_entry(tp, bp, dep);
425 xfs_dir2_data_check(dp, bp);
426 xfs_da_buf_done(bp);
427 return 0;
431 * Readdir for block directories.
433 int /* error */
434 xfs_dir2_block_getdents(
435 xfs_trans_t *tp, /* transaction (NULL) */
436 xfs_inode_t *dp, /* incore inode */
437 uio_t *uio, /* caller's buffer control */
438 int *eofp, /* eof reached? (out) */
439 xfs_dirent_t *dbp, /* caller's buffer */
440 xfs_dir2_put_t put) /* abi's formatting function */
442 xfs_dir2_block_t *block; /* directory block structure */
443 xfs_dabuf_t *bp; /* buffer for block */
444 xfs_dir2_block_tail_t *btp; /* block tail */
445 xfs_dir2_data_entry_t *dep; /* block data entry */
446 xfs_dir2_data_unused_t *dup; /* block unused entry */
447 char *endptr; /* end of the data entries */
448 int error; /* error return value */
449 xfs_mount_t *mp; /* filesystem mount point */
450 xfs_dir2_put_args_t p; /* arg package for put rtn */
451 char *ptr; /* current data entry */
452 int wantoff; /* starting block offset */
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, uio->uio_offset) > mp->m_dirdatablk) {
459 *eofp = 1;
460 return 0;
463 * Can't read the block, give up, else get dabuf in bp.
465 if ((error =
466 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
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, uio->uio_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);
483 p.dbp = dbp;
484 p.put = put;
485 p.uio = uio;
487 * Loop over the data portion of the block.
488 * Each object is a real entry (dep) or an unused one (dup).
490 while (ptr < endptr) {
491 dup = (xfs_dir2_data_unused_t *)ptr;
493 * Unused, skip it.
495 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
496 ptr += be16_to_cpu(dup->length);
497 continue;
500 dep = (xfs_dir2_data_entry_t *)ptr;
503 * Bump pointer for the next iteration.
505 ptr += xfs_dir2_data_entsize(dep->namelen);
507 * The entry is before the desired starting point, skip it.
509 if ((char *)dep - (char *)block < wantoff)
510 continue;
512 * Set up argument structure for put routine.
514 p.namelen = dep->namelen;
516 p.cook = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
517 ptr - (char *)block);
518 p.ino = be64_to_cpu(dep->inumber);
519 #if XFS_BIG_INUMS
520 p.ino += mp->m_inoadd;
521 #endif
522 p.name = (char *)dep->name;
525 * Put the entry in the caller's buffer.
527 error = p.put(&p);
530 * If it didn't fit, set the final offset to here & return.
532 if (!p.done) {
533 uio->uio_offset =
534 xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
535 (char *)dep - (char *)block);
536 xfs_da_brelse(tp, bp);
537 return error;
542 * Reached the end of the block.
543 * Set the offset to a non-existent block 1 and return.
545 *eofp = 1;
547 uio->uio_offset =
548 xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk + 1, 0);
550 xfs_da_brelse(tp, bp);
552 return 0;
556 * Log leaf entries from the block.
558 static void
559 xfs_dir2_block_log_leaf(
560 xfs_trans_t *tp, /* transaction structure */
561 xfs_dabuf_t *bp, /* block buffer */
562 int first, /* index of first logged leaf */
563 int last) /* index of last logged leaf */
565 xfs_dir2_block_t *block; /* directory block structure */
566 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
567 xfs_dir2_block_tail_t *btp; /* block tail */
568 xfs_mount_t *mp; /* filesystem mount point */
570 mp = tp->t_mountp;
571 block = bp->data;
572 btp = xfs_dir2_block_tail_p(mp, block);
573 blp = xfs_dir2_block_leaf_p(btp);
574 xfs_da_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)block),
575 (uint)((char *)&blp[last + 1] - (char *)block - 1));
579 * Log the block tail.
581 static void
582 xfs_dir2_block_log_tail(
583 xfs_trans_t *tp, /* transaction structure */
584 xfs_dabuf_t *bp) /* block buffer */
586 xfs_dir2_block_t *block; /* directory block structure */
587 xfs_dir2_block_tail_t *btp; /* block tail */
588 xfs_mount_t *mp; /* filesystem mount point */
590 mp = tp->t_mountp;
591 block = bp->data;
592 btp = xfs_dir2_block_tail_p(mp, block);
593 xfs_da_log_buf(tp, bp, (uint)((char *)btp - (char *)block),
594 (uint)((char *)(btp + 1) - (char *)block - 1));
598 * Look up an entry in the block. This is the external routine,
599 * xfs_dir2_block_lookup_int does the real work.
601 int /* error */
602 xfs_dir2_block_lookup(
603 xfs_da_args_t *args) /* dir lookup arguments */
605 xfs_dir2_block_t *block; /* block structure */
606 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
607 xfs_dabuf_t *bp; /* block buffer */
608 xfs_dir2_block_tail_t *btp; /* block tail */
609 xfs_dir2_data_entry_t *dep; /* block data entry */
610 xfs_inode_t *dp; /* incore inode */
611 int ent; /* entry index */
612 int error; /* error return value */
613 xfs_mount_t *mp; /* filesystem mount point */
615 xfs_dir2_trace_args("block_lookup", args);
617 * Get the buffer, look up the entry.
618 * If not found (ENOENT) then return, have no buffer.
620 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
621 return error;
622 dp = args->dp;
623 mp = dp->i_mount;
624 block = bp->data;
625 xfs_dir2_data_check(dp, bp);
626 btp = xfs_dir2_block_tail_p(mp, block);
627 blp = xfs_dir2_block_leaf_p(btp);
629 * Get the offset from the leaf entry, to point to the data.
631 dep = (xfs_dir2_data_entry_t *)
632 ((char *)block + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
634 * Fill in inode number, release the block.
636 args->inumber = be64_to_cpu(dep->inumber);
637 xfs_da_brelse(args->trans, bp);
638 return XFS_ERROR(EEXIST);
642 * Internal block lookup routine.
644 static int /* error */
645 xfs_dir2_block_lookup_int(
646 xfs_da_args_t *args, /* dir lookup arguments */
647 xfs_dabuf_t **bpp, /* returned block buffer */
648 int *entno) /* returned entry number */
650 xfs_dir2_dataptr_t addr; /* data entry address */
651 xfs_dir2_block_t *block; /* block structure */
652 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
653 xfs_dabuf_t *bp; /* block buffer */
654 xfs_dir2_block_tail_t *btp; /* block tail */
655 xfs_dir2_data_entry_t *dep; /* block data entry */
656 xfs_inode_t *dp; /* incore inode */
657 int error; /* error return value */
658 xfs_dahash_t hash; /* found hash value */
659 int high; /* binary search high index */
660 int low; /* binary search low index */
661 int mid; /* binary search current idx */
662 xfs_mount_t *mp; /* filesystem mount point */
663 xfs_trans_t *tp; /* transaction pointer */
665 dp = args->dp;
666 tp = args->trans;
667 mp = dp->i_mount;
669 * Read the buffer, return error if we can't get it.
671 if ((error =
672 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
673 return error;
675 ASSERT(bp != NULL);
676 block = bp->data;
677 xfs_dir2_data_check(dp, bp);
678 btp = xfs_dir2_block_tail_p(mp, block);
679 blp = xfs_dir2_block_leaf_p(btp);
681 * Loop doing a binary search for our hash value.
682 * Find our entry, ENOENT if it's not there.
684 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
685 ASSERT(low <= high);
686 mid = (low + high) >> 1;
687 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
688 break;
689 if (hash < args->hashval)
690 low = mid + 1;
691 else
692 high = mid - 1;
693 if (low > high) {
694 ASSERT(args->oknoent);
695 xfs_da_brelse(tp, bp);
696 return XFS_ERROR(ENOENT);
700 * Back up to the first one with the right hash value.
702 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
703 mid--;
706 * Now loop forward through all the entries with the
707 * right hash value looking for our name.
709 do {
710 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
711 continue;
713 * Get pointer to the entry from the leaf.
715 dep = (xfs_dir2_data_entry_t *)
716 ((char *)block + xfs_dir2_dataptr_to_off(mp, addr));
718 * Compare, if it's right give back buffer & entry number.
720 if (dep->namelen == args->namelen &&
721 dep->name[0] == args->name[0] &&
722 memcmp(dep->name, args->name, args->namelen) == 0) {
723 *bpp = bp;
724 *entno = mid;
725 return 0;
727 } while (++mid < be32_to_cpu(btp->count) && be32_to_cpu(blp[mid].hashval) == hash);
729 * No match, release the buffer and return ENOENT.
731 ASSERT(args->oknoent);
732 xfs_da_brelse(tp, bp);
733 return XFS_ERROR(ENOENT);
737 * Remove an entry from a block format directory.
738 * If that makes the block small enough to fit in shortform, transform it.
740 int /* error */
741 xfs_dir2_block_removename(
742 xfs_da_args_t *args) /* directory operation args */
744 xfs_dir2_block_t *block; /* block structure */
745 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
746 xfs_dabuf_t *bp; /* block buffer */
747 xfs_dir2_block_tail_t *btp; /* block tail */
748 xfs_dir2_data_entry_t *dep; /* block data entry */
749 xfs_inode_t *dp; /* incore inode */
750 int ent; /* block leaf entry index */
751 int error; /* error return value */
752 xfs_mount_t *mp; /* filesystem mount point */
753 int needlog; /* need to log block header */
754 int needscan; /* need to fixup bestfree */
755 xfs_dir2_sf_hdr_t sfh; /* shortform header */
756 int size; /* shortform size */
757 xfs_trans_t *tp; /* transaction pointer */
759 xfs_dir2_trace_args("block_removename", args);
761 * Look up the entry in the block. Gets the buffer and entry index.
762 * It will always be there, the vnodeops level does a lookup first.
764 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
765 return error;
767 dp = args->dp;
768 tp = args->trans;
769 mp = dp->i_mount;
770 block = bp->data;
771 btp = xfs_dir2_block_tail_p(mp, block);
772 blp = xfs_dir2_block_leaf_p(btp);
774 * Point to the data entry using the leaf entry.
776 dep = (xfs_dir2_data_entry_t *)
777 ((char *)block + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
779 * Mark the data entry's space free.
781 needlog = needscan = 0;
782 xfs_dir2_data_make_free(tp, bp,
783 (xfs_dir2_data_aoff_t)((char *)dep - (char *)block),
784 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
786 * Fix up the block tail.
788 be32_add(&btp->stale, 1);
789 xfs_dir2_block_log_tail(tp, bp);
791 * Remove the leaf entry by marking it stale.
793 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
794 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
796 * Fix up bestfree, log the header if necessary.
798 if (needscan)
799 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
800 if (needlog)
801 xfs_dir2_data_log_header(tp, bp);
802 xfs_dir2_data_check(dp, bp);
804 * See if the size as a shortform is good enough.
806 if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
807 XFS_IFORK_DSIZE(dp)) {
808 xfs_da_buf_done(bp);
809 return 0;
812 * If it works, do the conversion.
814 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
818 * Replace an entry in a V2 block directory.
819 * Change the inode number to the new value.
821 int /* error */
822 xfs_dir2_block_replace(
823 xfs_da_args_t *args) /* directory operation args */
825 xfs_dir2_block_t *block; /* block structure */
826 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
827 xfs_dabuf_t *bp; /* block buffer */
828 xfs_dir2_block_tail_t *btp; /* block tail */
829 xfs_dir2_data_entry_t *dep; /* block data entry */
830 xfs_inode_t *dp; /* incore inode */
831 int ent; /* leaf entry index */
832 int error; /* error return value */
833 xfs_mount_t *mp; /* filesystem mount point */
835 xfs_dir2_trace_args("block_replace", args);
837 * Lookup the entry in the directory. Get buffer and entry index.
838 * This will always succeed since the caller has already done a lookup.
840 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
841 return error;
843 dp = args->dp;
844 mp = dp->i_mount;
845 block = bp->data;
846 btp = xfs_dir2_block_tail_p(mp, block);
847 blp = xfs_dir2_block_leaf_p(btp);
849 * Point to the data entry we need to change.
851 dep = (xfs_dir2_data_entry_t *)
852 ((char *)block + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
853 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
855 * Change the inode number to the new value.
857 dep->inumber = cpu_to_be64(args->inumber);
858 xfs_dir2_data_log_entry(args->trans, bp, dep);
859 xfs_dir2_data_check(dp, bp);
860 xfs_da_buf_done(bp);
861 return 0;
865 * Qsort comparison routine for the block leaf entries.
867 static int /* sort order */
868 xfs_dir2_block_sort(
869 const void *a, /* first leaf entry */
870 const void *b) /* second leaf entry */
872 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
873 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
875 la = a;
876 lb = b;
877 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
878 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
882 * Convert a V2 leaf directory to a V2 block directory if possible.
884 int /* error */
885 xfs_dir2_leaf_to_block(
886 xfs_da_args_t *args, /* operation arguments */
887 xfs_dabuf_t *lbp, /* leaf buffer */
888 xfs_dabuf_t *dbp) /* data buffer */
890 __be16 *bestsp; /* leaf bests table */
891 xfs_dir2_block_t *block; /* block structure */
892 xfs_dir2_block_tail_t *btp; /* block tail */
893 xfs_inode_t *dp; /* incore directory inode */
894 xfs_dir2_data_unused_t *dup; /* unused data entry */
895 int error; /* error return value */
896 int from; /* leaf from index */
897 xfs_dir2_leaf_t *leaf; /* leaf structure */
898 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
899 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
900 xfs_mount_t *mp; /* file system mount point */
901 int needlog; /* need to log data header */
902 int needscan; /* need to scan for bestfree */
903 xfs_dir2_sf_hdr_t sfh; /* shortform header */
904 int size; /* bytes used */
905 __be16 *tagp; /* end of entry (tag) */
906 int to; /* block/leaf to index */
907 xfs_trans_t *tp; /* transaction pointer */
909 xfs_dir2_trace_args_bb("leaf_to_block", args, lbp, dbp);
910 dp = args->dp;
911 tp = args->trans;
912 mp = dp->i_mount;
913 leaf = lbp->data;
914 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
915 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
917 * If there are data blocks other than the first one, take this
918 * opportunity to remove trailing empty data blocks that may have
919 * been left behind during no-space-reservation operations.
920 * These will show up in the leaf bests table.
922 while (dp->i_d.di_size > mp->m_dirblksize) {
923 bestsp = xfs_dir2_leaf_bests_p(ltp);
924 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
925 mp->m_dirblksize - (uint)sizeof(block->hdr)) {
926 if ((error =
927 xfs_dir2_leaf_trim_data(args, lbp,
928 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
929 goto out;
930 } else {
931 error = 0;
932 goto out;
936 * Read the data block if we don't already have it, give up if it fails.
938 if (dbp == NULL &&
939 (error = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &dbp,
940 XFS_DATA_FORK))) {
941 goto out;
943 block = dbp->data;
944 ASSERT(be32_to_cpu(block->hdr.magic) == XFS_DIR2_DATA_MAGIC);
946 * Size of the "leaf" area in the block.
948 size = (uint)sizeof(block->tail) +
949 (uint)sizeof(*lep) * (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
951 * Look at the last data entry.
953 tagp = (__be16 *)((char *)block + mp->m_dirblksize) - 1;
954 dup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
956 * If it's not free or is too short we can't do it.
958 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
959 be16_to_cpu(dup->length) < size) {
960 error = 0;
961 goto out;
964 * Start converting it to block form.
966 block->hdr.magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
967 needlog = 1;
968 needscan = 0;
970 * Use up the space at the end of the block (blp/btp).
972 xfs_dir2_data_use_free(tp, dbp, dup, mp->m_dirblksize - size, size,
973 &needlog, &needscan);
975 * Initialize the block tail.
977 btp = xfs_dir2_block_tail_p(mp, block);
978 btp->count = cpu_to_be32(be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
979 btp->stale = 0;
980 xfs_dir2_block_log_tail(tp, dbp);
982 * Initialize the block leaf area. We compact out stale entries.
984 lep = xfs_dir2_block_leaf_p(btp);
985 for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
986 if (be32_to_cpu(leaf->ents[from].address) == XFS_DIR2_NULL_DATAPTR)
987 continue;
988 lep[to++] = leaf->ents[from];
990 ASSERT(to == be32_to_cpu(btp->count));
991 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
993 * Scan the bestfree if we need it and log the data block header.
995 if (needscan)
996 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
997 if (needlog)
998 xfs_dir2_data_log_header(tp, dbp);
1000 * Pitch the old leaf block.
1002 error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
1003 lbp = NULL;
1004 if (error) {
1005 goto out;
1008 * Now see if the resulting block can be shrunken to shortform.
1010 if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
1011 XFS_IFORK_DSIZE(dp)) {
1012 error = 0;
1013 goto out;
1015 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1016 out:
1017 if (lbp)
1018 xfs_da_buf_done(lbp);
1019 if (dbp)
1020 xfs_da_buf_done(dbp);
1021 return error;
1025 * Convert the shortform directory to block form.
1027 int /* error */
1028 xfs_dir2_sf_to_block(
1029 xfs_da_args_t *args) /* operation arguments */
1031 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1032 xfs_dir2_block_t *block; /* block structure */
1033 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1034 xfs_dabuf_t *bp; /* block buffer */
1035 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1036 char *buf; /* sf buffer */
1037 int buf_len;
1038 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1039 xfs_inode_t *dp; /* incore directory inode */
1040 int dummy; /* trash */
1041 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1042 int endoffset; /* end of data objects */
1043 int error; /* error return value */
1044 int i; /* index */
1045 xfs_mount_t *mp; /* filesystem mount point */
1046 int needlog; /* need to log block header */
1047 int needscan; /* need to scan block freespc */
1048 int newoffset; /* offset from current entry */
1049 int offset; /* target block offset */
1050 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1051 xfs_dir2_sf_t *sfp; /* shortform structure */
1052 __be16 *tagp; /* end of data entry */
1053 xfs_trans_t *tp; /* transaction pointer */
1055 xfs_dir2_trace_args("sf_to_block", args);
1056 dp = args->dp;
1057 tp = args->trans;
1058 mp = dp->i_mount;
1059 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
1061 * Bomb out if the shortform directory is way too short.
1063 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1064 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1065 return XFS_ERROR(EIO);
1067 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
1068 ASSERT(dp->i_df.if_u1.if_data != NULL);
1069 sfp = (xfs_dir2_sf_t *)dp->i_df.if_u1.if_data;
1070 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->hdr.i8count));
1072 * Copy the directory into the stack buffer.
1073 * Then pitch the incore inode data so we can make extents.
1076 buf_len = dp->i_df.if_bytes;
1077 buf = kmem_alloc(dp->i_df.if_bytes, KM_SLEEP);
1079 memcpy(buf, sfp, dp->i_df.if_bytes);
1080 xfs_idata_realloc(dp, -dp->i_df.if_bytes, XFS_DATA_FORK);
1081 dp->i_d.di_size = 0;
1082 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1084 * Reset pointer - old sfp is gone.
1086 sfp = (xfs_dir2_sf_t *)buf;
1088 * Add block 0 to the inode.
1090 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1091 if (error) {
1092 kmem_free(buf, buf_len);
1093 return error;
1096 * Initialize the data block.
1098 error = xfs_dir2_data_init(args, blkno, &bp);
1099 if (error) {
1100 kmem_free(buf, buf_len);
1101 return error;
1103 block = bp->data;
1104 block->hdr.magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
1106 * Compute size of block "tail" area.
1108 i = (uint)sizeof(*btp) +
1109 (sfp->hdr.count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1111 * The whole thing is initialized to free by the init routine.
1112 * Say we're using the leaf and tail area.
1114 dup = (xfs_dir2_data_unused_t *)block->u;
1115 needlog = needscan = 0;
1116 xfs_dir2_data_use_free(tp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1117 &needscan);
1118 ASSERT(needscan == 0);
1120 * Fill in the tail.
1122 btp = xfs_dir2_block_tail_p(mp, block);
1123 btp->count = cpu_to_be32(sfp->hdr.count + 2); /* ., .. */
1124 btp->stale = 0;
1125 blp = xfs_dir2_block_leaf_p(btp);
1126 endoffset = (uint)((char *)blp - (char *)block);
1128 * Remove the freespace, we'll manage it.
1130 xfs_dir2_data_use_free(tp, bp, dup,
1131 (xfs_dir2_data_aoff_t)((char *)dup - (char *)block),
1132 be16_to_cpu(dup->length), &needlog, &needscan);
1134 * Create entry for .
1136 dep = (xfs_dir2_data_entry_t *)
1137 ((char *)block + XFS_DIR2_DATA_DOT_OFFSET);
1138 dep->inumber = cpu_to_be64(dp->i_ino);
1139 dep->namelen = 1;
1140 dep->name[0] = '.';
1141 tagp = xfs_dir2_data_entry_tag_p(dep);
1142 *tagp = cpu_to_be16((char *)dep - (char *)block);
1143 xfs_dir2_data_log_entry(tp, bp, dep);
1144 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1145 blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1146 (char *)dep - (char *)block));
1148 * Create entry for ..
1150 dep = (xfs_dir2_data_entry_t *)
1151 ((char *)block + XFS_DIR2_DATA_DOTDOT_OFFSET);
1152 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_inumber(sfp, &sfp->hdr.parent));
1153 dep->namelen = 2;
1154 dep->name[0] = dep->name[1] = '.';
1155 tagp = xfs_dir2_data_entry_tag_p(dep);
1156 *tagp = cpu_to_be16((char *)dep - (char *)block);
1157 xfs_dir2_data_log_entry(tp, bp, dep);
1158 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1159 blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1160 (char *)dep - (char *)block));
1161 offset = XFS_DIR2_DATA_FIRST_OFFSET;
1163 * Loop over existing entries, stuff them in.
1165 if ((i = 0) == sfp->hdr.count)
1166 sfep = NULL;
1167 else
1168 sfep = xfs_dir2_sf_firstentry(sfp);
1170 * Need to preserve the existing offset values in the sf directory.
1171 * Insert holes (unused entries) where necessary.
1173 while (offset < endoffset) {
1175 * sfep is null when we reach the end of the list.
1177 if (sfep == NULL)
1178 newoffset = endoffset;
1179 else
1180 newoffset = xfs_dir2_sf_get_offset(sfep);
1182 * There should be a hole here, make one.
1184 if (offset < newoffset) {
1185 dup = (xfs_dir2_data_unused_t *)
1186 ((char *)block + offset);
1187 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1188 dup->length = cpu_to_be16(newoffset - offset);
1189 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1190 ((char *)dup - (char *)block));
1191 xfs_dir2_data_log_unused(tp, bp, dup);
1192 (void)xfs_dir2_data_freeinsert((xfs_dir2_data_t *)block,
1193 dup, &dummy);
1194 offset += be16_to_cpu(dup->length);
1195 continue;
1198 * Copy a real entry.
1200 dep = (xfs_dir2_data_entry_t *)((char *)block + newoffset);
1201 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_inumber(sfp,
1202 xfs_dir2_sf_inumberp(sfep)));
1203 dep->namelen = sfep->namelen;
1204 memcpy(dep->name, sfep->name, dep->namelen);
1205 tagp = xfs_dir2_data_entry_tag_p(dep);
1206 *tagp = cpu_to_be16((char *)dep - (char *)block);
1207 xfs_dir2_data_log_entry(tp, bp, dep);
1208 blp[2 + i].hashval = cpu_to_be32(xfs_da_hashname(
1209 (char *)sfep->name, sfep->namelen));
1210 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1211 (char *)dep - (char *)block));
1212 offset = (int)((char *)(tagp + 1) - (char *)block);
1213 if (++i == sfp->hdr.count)
1214 sfep = NULL;
1215 else
1216 sfep = xfs_dir2_sf_nextentry(sfp, sfep);
1218 /* Done with the temporary buffer */
1219 kmem_free(buf, buf_len);
1221 * Sort the leaf entries by hash value.
1223 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1225 * Log the leaf entry area and tail.
1226 * Already logged the header in data_init, ignore needlog.
1228 ASSERT(needscan == 0);
1229 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1230 xfs_dir2_block_log_tail(tp, bp);
1231 xfs_dir2_data_check(dp, bp);
1232 xfs_da_buf_done(bp);
1233 return 0;