signals: fix sigqueue_free() vs __exit_signal() race
[linux-2.6/openmoko-kernel/knife-kernel.git] / fs / xfs / xfs_dir2_block.c
blobfb5a556725b344904090769fb37c78f6e970179a
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_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,
521 ino, DT_UNKNOWN)) {
522 *offset = cook;
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 xfs_da_brelse(NULL, bp);
534 return 0;
538 * Log leaf entries from the block.
540 static void
541 xfs_dir2_block_log_leaf(
542 xfs_trans_t *tp, /* transaction structure */
543 xfs_dabuf_t *bp, /* block buffer */
544 int first, /* index of first logged leaf */
545 int last) /* index of last logged leaf */
547 xfs_dir2_block_t *block; /* directory block structure */
548 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
549 xfs_dir2_block_tail_t *btp; /* block tail */
550 xfs_mount_t *mp; /* filesystem mount point */
552 mp = tp->t_mountp;
553 block = bp->data;
554 btp = xfs_dir2_block_tail_p(mp, block);
555 blp = xfs_dir2_block_leaf_p(btp);
556 xfs_da_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)block),
557 (uint)((char *)&blp[last + 1] - (char *)block - 1));
561 * Log the block tail.
563 static void
564 xfs_dir2_block_log_tail(
565 xfs_trans_t *tp, /* transaction structure */
566 xfs_dabuf_t *bp) /* block buffer */
568 xfs_dir2_block_t *block; /* directory block structure */
569 xfs_dir2_block_tail_t *btp; /* block tail */
570 xfs_mount_t *mp; /* filesystem mount point */
572 mp = tp->t_mountp;
573 block = bp->data;
574 btp = xfs_dir2_block_tail_p(mp, block);
575 xfs_da_log_buf(tp, bp, (uint)((char *)btp - (char *)block),
576 (uint)((char *)(btp + 1) - (char *)block - 1));
580 * Look up an entry in the block. This is the external routine,
581 * xfs_dir2_block_lookup_int does the real work.
583 int /* error */
584 xfs_dir2_block_lookup(
585 xfs_da_args_t *args) /* dir lookup arguments */
587 xfs_dir2_block_t *block; /* block structure */
588 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
589 xfs_dabuf_t *bp; /* block buffer */
590 xfs_dir2_block_tail_t *btp; /* block tail */
591 xfs_dir2_data_entry_t *dep; /* block data entry */
592 xfs_inode_t *dp; /* incore inode */
593 int ent; /* entry index */
594 int error; /* error return value */
595 xfs_mount_t *mp; /* filesystem mount point */
597 xfs_dir2_trace_args("block_lookup", args);
599 * Get the buffer, look up the entry.
600 * If not found (ENOENT) then return, have no buffer.
602 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
603 return error;
604 dp = args->dp;
605 mp = dp->i_mount;
606 block = bp->data;
607 xfs_dir2_data_check(dp, bp);
608 btp = xfs_dir2_block_tail_p(mp, block);
609 blp = xfs_dir2_block_leaf_p(btp);
611 * Get the offset from the leaf entry, to point to the data.
613 dep = (xfs_dir2_data_entry_t *)
614 ((char *)block + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
616 * Fill in inode number, release the block.
618 args->inumber = be64_to_cpu(dep->inumber);
619 xfs_da_brelse(args->trans, bp);
620 return XFS_ERROR(EEXIST);
624 * Internal block lookup routine.
626 static int /* error */
627 xfs_dir2_block_lookup_int(
628 xfs_da_args_t *args, /* dir lookup arguments */
629 xfs_dabuf_t **bpp, /* returned block buffer */
630 int *entno) /* returned entry number */
632 xfs_dir2_dataptr_t addr; /* data entry address */
633 xfs_dir2_block_t *block; /* block structure */
634 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
635 xfs_dabuf_t *bp; /* block buffer */
636 xfs_dir2_block_tail_t *btp; /* block tail */
637 xfs_dir2_data_entry_t *dep; /* block data entry */
638 xfs_inode_t *dp; /* incore inode */
639 int error; /* error return value */
640 xfs_dahash_t hash; /* found hash value */
641 int high; /* binary search high index */
642 int low; /* binary search low index */
643 int mid; /* binary search current idx */
644 xfs_mount_t *mp; /* filesystem mount point */
645 xfs_trans_t *tp; /* transaction pointer */
647 dp = args->dp;
648 tp = args->trans;
649 mp = dp->i_mount;
651 * Read the buffer, return error if we can't get it.
653 if ((error =
654 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
655 return error;
657 ASSERT(bp != NULL);
658 block = bp->data;
659 xfs_dir2_data_check(dp, bp);
660 btp = xfs_dir2_block_tail_p(mp, block);
661 blp = xfs_dir2_block_leaf_p(btp);
663 * Loop doing a binary search for our hash value.
664 * Find our entry, ENOENT if it's not there.
666 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
667 ASSERT(low <= high);
668 mid = (low + high) >> 1;
669 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
670 break;
671 if (hash < args->hashval)
672 low = mid + 1;
673 else
674 high = mid - 1;
675 if (low > high) {
676 ASSERT(args->oknoent);
677 xfs_da_brelse(tp, bp);
678 return XFS_ERROR(ENOENT);
682 * Back up to the first one with the right hash value.
684 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
685 mid--;
688 * Now loop forward through all the entries with the
689 * right hash value looking for our name.
691 do {
692 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
693 continue;
695 * Get pointer to the entry from the leaf.
697 dep = (xfs_dir2_data_entry_t *)
698 ((char *)block + xfs_dir2_dataptr_to_off(mp, addr));
700 * Compare, if it's right give back buffer & entry number.
702 if (dep->namelen == args->namelen &&
703 dep->name[0] == args->name[0] &&
704 memcmp(dep->name, args->name, args->namelen) == 0) {
705 *bpp = bp;
706 *entno = mid;
707 return 0;
709 } while (++mid < be32_to_cpu(btp->count) && be32_to_cpu(blp[mid].hashval) == hash);
711 * No match, release the buffer and return ENOENT.
713 ASSERT(args->oknoent);
714 xfs_da_brelse(tp, bp);
715 return XFS_ERROR(ENOENT);
719 * Remove an entry from a block format directory.
720 * If that makes the block small enough to fit in shortform, transform it.
722 int /* error */
723 xfs_dir2_block_removename(
724 xfs_da_args_t *args) /* directory operation args */
726 xfs_dir2_block_t *block; /* block structure */
727 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
728 xfs_dabuf_t *bp; /* block buffer */
729 xfs_dir2_block_tail_t *btp; /* block tail */
730 xfs_dir2_data_entry_t *dep; /* block data entry */
731 xfs_inode_t *dp; /* incore inode */
732 int ent; /* block leaf entry index */
733 int error; /* error return value */
734 xfs_mount_t *mp; /* filesystem mount point */
735 int needlog; /* need to log block header */
736 int needscan; /* need to fixup bestfree */
737 xfs_dir2_sf_hdr_t sfh; /* shortform header */
738 int size; /* shortform size */
739 xfs_trans_t *tp; /* transaction pointer */
741 xfs_dir2_trace_args("block_removename", args);
743 * Look up the entry in the block. Gets the buffer and entry index.
744 * It will always be there, the vnodeops level does a lookup first.
746 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
747 return error;
749 dp = args->dp;
750 tp = args->trans;
751 mp = dp->i_mount;
752 block = bp->data;
753 btp = xfs_dir2_block_tail_p(mp, block);
754 blp = xfs_dir2_block_leaf_p(btp);
756 * Point to the data entry using the leaf entry.
758 dep = (xfs_dir2_data_entry_t *)
759 ((char *)block + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
761 * Mark the data entry's space free.
763 needlog = needscan = 0;
764 xfs_dir2_data_make_free(tp, bp,
765 (xfs_dir2_data_aoff_t)((char *)dep - (char *)block),
766 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
768 * Fix up the block tail.
770 be32_add_cpu(&btp->stale, 1);
771 xfs_dir2_block_log_tail(tp, bp);
773 * Remove the leaf entry by marking it stale.
775 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
776 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
778 * Fix up bestfree, log the header if necessary.
780 if (needscan)
781 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
782 if (needlog)
783 xfs_dir2_data_log_header(tp, bp);
784 xfs_dir2_data_check(dp, bp);
786 * See if the size as a shortform is good enough.
788 if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
789 XFS_IFORK_DSIZE(dp)) {
790 xfs_da_buf_done(bp);
791 return 0;
794 * If it works, do the conversion.
796 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
800 * Replace an entry in a V2 block directory.
801 * Change the inode number to the new value.
803 int /* error */
804 xfs_dir2_block_replace(
805 xfs_da_args_t *args) /* directory operation args */
807 xfs_dir2_block_t *block; /* block structure */
808 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
809 xfs_dabuf_t *bp; /* block buffer */
810 xfs_dir2_block_tail_t *btp; /* block tail */
811 xfs_dir2_data_entry_t *dep; /* block data entry */
812 xfs_inode_t *dp; /* incore inode */
813 int ent; /* leaf entry index */
814 int error; /* error return value */
815 xfs_mount_t *mp; /* filesystem mount point */
817 xfs_dir2_trace_args("block_replace", args);
819 * Lookup the entry in the directory. Get buffer and entry index.
820 * This will always succeed since the caller has already done a lookup.
822 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
823 return error;
825 dp = args->dp;
826 mp = dp->i_mount;
827 block = bp->data;
828 btp = xfs_dir2_block_tail_p(mp, block);
829 blp = xfs_dir2_block_leaf_p(btp);
831 * Point to the data entry we need to change.
833 dep = (xfs_dir2_data_entry_t *)
834 ((char *)block + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
835 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
837 * Change the inode number to the new value.
839 dep->inumber = cpu_to_be64(args->inumber);
840 xfs_dir2_data_log_entry(args->trans, bp, dep);
841 xfs_dir2_data_check(dp, bp);
842 xfs_da_buf_done(bp);
843 return 0;
847 * Qsort comparison routine for the block leaf entries.
849 static int /* sort order */
850 xfs_dir2_block_sort(
851 const void *a, /* first leaf entry */
852 const void *b) /* second leaf entry */
854 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
855 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
857 la = a;
858 lb = b;
859 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
860 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
864 * Convert a V2 leaf directory to a V2 block directory if possible.
866 int /* error */
867 xfs_dir2_leaf_to_block(
868 xfs_da_args_t *args, /* operation arguments */
869 xfs_dabuf_t *lbp, /* leaf buffer */
870 xfs_dabuf_t *dbp) /* data buffer */
872 __be16 *bestsp; /* leaf bests table */
873 xfs_dir2_block_t *block; /* block structure */
874 xfs_dir2_block_tail_t *btp; /* block tail */
875 xfs_inode_t *dp; /* incore directory inode */
876 xfs_dir2_data_unused_t *dup; /* unused data entry */
877 int error; /* error return value */
878 int from; /* leaf from index */
879 xfs_dir2_leaf_t *leaf; /* leaf structure */
880 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
881 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
882 xfs_mount_t *mp; /* file system mount point */
883 int needlog; /* need to log data header */
884 int needscan; /* need to scan for bestfree */
885 xfs_dir2_sf_hdr_t sfh; /* shortform header */
886 int size; /* bytes used */
887 __be16 *tagp; /* end of entry (tag) */
888 int to; /* block/leaf to index */
889 xfs_trans_t *tp; /* transaction pointer */
891 xfs_dir2_trace_args_bb("leaf_to_block", args, lbp, dbp);
892 dp = args->dp;
893 tp = args->trans;
894 mp = dp->i_mount;
895 leaf = lbp->data;
896 ASSERT(be16_to_cpu(leaf->hdr.info.magic) == XFS_DIR2_LEAF1_MAGIC);
897 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
899 * If there are data blocks other than the first one, take this
900 * opportunity to remove trailing empty data blocks that may have
901 * been left behind during no-space-reservation operations.
902 * These will show up in the leaf bests table.
904 while (dp->i_d.di_size > mp->m_dirblksize) {
905 bestsp = xfs_dir2_leaf_bests_p(ltp);
906 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
907 mp->m_dirblksize - (uint)sizeof(block->hdr)) {
908 if ((error =
909 xfs_dir2_leaf_trim_data(args, lbp,
910 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
911 goto out;
912 } else {
913 error = 0;
914 goto out;
918 * Read the data block if we don't already have it, give up if it fails.
920 if (dbp == NULL &&
921 (error = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &dbp,
922 XFS_DATA_FORK))) {
923 goto out;
925 block = dbp->data;
926 ASSERT(be32_to_cpu(block->hdr.magic) == XFS_DIR2_DATA_MAGIC);
928 * Size of the "leaf" area in the block.
930 size = (uint)sizeof(block->tail) +
931 (uint)sizeof(*lep) * (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
933 * Look at the last data entry.
935 tagp = (__be16 *)((char *)block + mp->m_dirblksize) - 1;
936 dup = (xfs_dir2_data_unused_t *)((char *)block + be16_to_cpu(*tagp));
938 * If it's not free or is too short we can't do it.
940 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
941 be16_to_cpu(dup->length) < size) {
942 error = 0;
943 goto out;
946 * Start converting it to block form.
948 block->hdr.magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
949 needlog = 1;
950 needscan = 0;
952 * Use up the space at the end of the block (blp/btp).
954 xfs_dir2_data_use_free(tp, dbp, dup, mp->m_dirblksize - size, size,
955 &needlog, &needscan);
957 * Initialize the block tail.
959 btp = xfs_dir2_block_tail_p(mp, block);
960 btp->count = cpu_to_be32(be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
961 btp->stale = 0;
962 xfs_dir2_block_log_tail(tp, dbp);
964 * Initialize the block leaf area. We compact out stale entries.
966 lep = xfs_dir2_block_leaf_p(btp);
967 for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
968 if (be32_to_cpu(leaf->ents[from].address) == XFS_DIR2_NULL_DATAPTR)
969 continue;
970 lep[to++] = leaf->ents[from];
972 ASSERT(to == be32_to_cpu(btp->count));
973 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
975 * Scan the bestfree if we need it and log the data block header.
977 if (needscan)
978 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog);
979 if (needlog)
980 xfs_dir2_data_log_header(tp, dbp);
982 * Pitch the old leaf block.
984 error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
985 lbp = NULL;
986 if (error) {
987 goto out;
990 * Now see if the resulting block can be shrunken to shortform.
992 if ((size = xfs_dir2_block_sfsize(dp, block, &sfh)) >
993 XFS_IFORK_DSIZE(dp)) {
994 error = 0;
995 goto out;
997 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
998 out:
999 if (lbp)
1000 xfs_da_buf_done(lbp);
1001 if (dbp)
1002 xfs_da_buf_done(dbp);
1003 return error;
1007 * Convert the shortform directory to block form.
1009 int /* error */
1010 xfs_dir2_sf_to_block(
1011 xfs_da_args_t *args) /* operation arguments */
1013 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1014 xfs_dir2_block_t *block; /* block structure */
1015 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1016 xfs_dabuf_t *bp; /* block buffer */
1017 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1018 char *buf; /* sf buffer */
1019 int buf_len;
1020 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1021 xfs_inode_t *dp; /* incore directory inode */
1022 int dummy; /* trash */
1023 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1024 int endoffset; /* end of data objects */
1025 int error; /* error return value */
1026 int i; /* index */
1027 xfs_mount_t *mp; /* filesystem mount point */
1028 int needlog; /* need to log block header */
1029 int needscan; /* need to scan block freespc */
1030 int newoffset; /* offset from current entry */
1031 int offset; /* target block offset */
1032 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1033 xfs_dir2_sf_t *sfp; /* shortform structure */
1034 __be16 *tagp; /* end of data entry */
1035 xfs_trans_t *tp; /* transaction pointer */
1037 xfs_dir2_trace_args("sf_to_block", args);
1038 dp = args->dp;
1039 tp = args->trans;
1040 mp = dp->i_mount;
1041 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
1043 * Bomb out if the shortform directory is way too short.
1045 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1046 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1047 return XFS_ERROR(EIO);
1049 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
1050 ASSERT(dp->i_df.if_u1.if_data != NULL);
1051 sfp = (xfs_dir2_sf_t *)dp->i_df.if_u1.if_data;
1052 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->hdr.i8count));
1054 * Copy the directory into the stack buffer.
1055 * Then pitch the incore inode data so we can make extents.
1058 buf_len = dp->i_df.if_bytes;
1059 buf = kmem_alloc(dp->i_df.if_bytes, KM_SLEEP);
1061 memcpy(buf, sfp, dp->i_df.if_bytes);
1062 xfs_idata_realloc(dp, -dp->i_df.if_bytes, XFS_DATA_FORK);
1063 dp->i_d.di_size = 0;
1064 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1066 * Reset pointer - old sfp is gone.
1068 sfp = (xfs_dir2_sf_t *)buf;
1070 * Add block 0 to the inode.
1072 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1073 if (error) {
1074 kmem_free(buf, buf_len);
1075 return error;
1078 * Initialize the data block.
1080 error = xfs_dir2_data_init(args, blkno, &bp);
1081 if (error) {
1082 kmem_free(buf, buf_len);
1083 return error;
1085 block = bp->data;
1086 block->hdr.magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
1088 * Compute size of block "tail" area.
1090 i = (uint)sizeof(*btp) +
1091 (sfp->hdr.count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1093 * The whole thing is initialized to free by the init routine.
1094 * Say we're using the leaf and tail area.
1096 dup = (xfs_dir2_data_unused_t *)block->u;
1097 needlog = needscan = 0;
1098 xfs_dir2_data_use_free(tp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1099 &needscan);
1100 ASSERT(needscan == 0);
1102 * Fill in the tail.
1104 btp = xfs_dir2_block_tail_p(mp, block);
1105 btp->count = cpu_to_be32(sfp->hdr.count + 2); /* ., .. */
1106 btp->stale = 0;
1107 blp = xfs_dir2_block_leaf_p(btp);
1108 endoffset = (uint)((char *)blp - (char *)block);
1110 * Remove the freespace, we'll manage it.
1112 xfs_dir2_data_use_free(tp, bp, dup,
1113 (xfs_dir2_data_aoff_t)((char *)dup - (char *)block),
1114 be16_to_cpu(dup->length), &needlog, &needscan);
1116 * Create entry for .
1118 dep = (xfs_dir2_data_entry_t *)
1119 ((char *)block + XFS_DIR2_DATA_DOT_OFFSET);
1120 dep->inumber = cpu_to_be64(dp->i_ino);
1121 dep->namelen = 1;
1122 dep->name[0] = '.';
1123 tagp = xfs_dir2_data_entry_tag_p(dep);
1124 *tagp = cpu_to_be16((char *)dep - (char *)block);
1125 xfs_dir2_data_log_entry(tp, bp, dep);
1126 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1127 blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1128 (char *)dep - (char *)block));
1130 * Create entry for ..
1132 dep = (xfs_dir2_data_entry_t *)
1133 ((char *)block + XFS_DIR2_DATA_DOTDOT_OFFSET);
1134 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_inumber(sfp, &sfp->hdr.parent));
1135 dep->namelen = 2;
1136 dep->name[0] = dep->name[1] = '.';
1137 tagp = xfs_dir2_data_entry_tag_p(dep);
1138 *tagp = cpu_to_be16((char *)dep - (char *)block);
1139 xfs_dir2_data_log_entry(tp, bp, dep);
1140 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1141 blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1142 (char *)dep - (char *)block));
1143 offset = XFS_DIR2_DATA_FIRST_OFFSET;
1145 * Loop over existing entries, stuff them in.
1147 if ((i = 0) == sfp->hdr.count)
1148 sfep = NULL;
1149 else
1150 sfep = xfs_dir2_sf_firstentry(sfp);
1152 * Need to preserve the existing offset values in the sf directory.
1153 * Insert holes (unused entries) where necessary.
1155 while (offset < endoffset) {
1157 * sfep is null when we reach the end of the list.
1159 if (sfep == NULL)
1160 newoffset = endoffset;
1161 else
1162 newoffset = xfs_dir2_sf_get_offset(sfep);
1164 * There should be a hole here, make one.
1166 if (offset < newoffset) {
1167 dup = (xfs_dir2_data_unused_t *)
1168 ((char *)block + offset);
1169 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1170 dup->length = cpu_to_be16(newoffset - offset);
1171 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1172 ((char *)dup - (char *)block));
1173 xfs_dir2_data_log_unused(tp, bp, dup);
1174 (void)xfs_dir2_data_freeinsert((xfs_dir2_data_t *)block,
1175 dup, &dummy);
1176 offset += be16_to_cpu(dup->length);
1177 continue;
1180 * Copy a real entry.
1182 dep = (xfs_dir2_data_entry_t *)((char *)block + newoffset);
1183 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_inumber(sfp,
1184 xfs_dir2_sf_inumberp(sfep)));
1185 dep->namelen = sfep->namelen;
1186 memcpy(dep->name, sfep->name, dep->namelen);
1187 tagp = xfs_dir2_data_entry_tag_p(dep);
1188 *tagp = cpu_to_be16((char *)dep - (char *)block);
1189 xfs_dir2_data_log_entry(tp, bp, dep);
1190 blp[2 + i].hashval = cpu_to_be32(xfs_da_hashname(
1191 (char *)sfep->name, sfep->namelen));
1192 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1193 (char *)dep - (char *)block));
1194 offset = (int)((char *)(tagp + 1) - (char *)block);
1195 if (++i == sfp->hdr.count)
1196 sfep = NULL;
1197 else
1198 sfep = xfs_dir2_sf_nextentry(sfp, sfep);
1200 /* Done with the temporary buffer */
1201 kmem_free(buf, buf_len);
1203 * Sort the leaf entries by hash value.
1205 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1207 * Log the leaf entry area and tail.
1208 * Already logged the header in data_init, ignore needlog.
1210 ASSERT(needscan == 0);
1211 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1212 xfs_dir2_block_log_tail(tp, bp);
1213 xfs_dir2_data_check(dp, bp);
1214 xfs_da_buf_done(bp);
1215 return 0;