Avoid beyond bounds copy while caching ACL
[zen-stable.git] / fs / xfs / xfs_dir2_block.c
blob9245e029b8eaddb08f58bbdbe3f78d972cfd2610
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_mount.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
31 #include "xfs_inode_item.h"
32 #include "xfs_dir2_format.h"
33 #include "xfs_dir2_priv.h"
34 #include "xfs_error.h"
35 #include "xfs_trace.h"
38 * Local function prototypes.
40 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, xfs_dabuf_t *bp, int first,
41 int last);
42 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, xfs_dabuf_t *bp);
43 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **bpp,
44 int *entno);
45 static int xfs_dir2_block_sort(const void *a, const void *b);
47 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
50 * One-time startup routine called from xfs_init().
52 void
53 xfs_dir_startup(void)
55 xfs_dir_hash_dot = xfs_da_hashname((unsigned char *)".", 1);
56 xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
60 * Add an entry to a block directory.
62 int /* error */
63 xfs_dir2_block_addname(
64 xfs_da_args_t *args) /* directory op arguments */
66 xfs_dir2_data_free_t *bf; /* bestfree table in block */
67 xfs_dir2_data_hdr_t *hdr; /* block header */
68 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
69 xfs_dabuf_t *bp; /* buffer for block */
70 xfs_dir2_block_tail_t *btp; /* block tail */
71 int compact; /* need to compact leaf ents */
72 xfs_dir2_data_entry_t *dep; /* block data entry */
73 xfs_inode_t *dp; /* directory inode */
74 xfs_dir2_data_unused_t *dup; /* block unused entry */
75 int error; /* error return value */
76 xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
77 xfs_dahash_t hash; /* hash value of found entry */
78 int high; /* high index for binary srch */
79 int highstale; /* high stale index */
80 int lfloghigh=0; /* last final leaf to log */
81 int lfloglow=0; /* first final leaf to log */
82 int len; /* length of the new entry */
83 int low; /* low index for binary srch */
84 int lowstale; /* low stale index */
85 int mid=0; /* midpoint for binary srch */
86 xfs_mount_t *mp; /* filesystem mount point */
87 int needlog; /* need to log header */
88 int needscan; /* need to rescan freespace */
89 __be16 *tagp; /* pointer to tag value */
90 xfs_trans_t *tp; /* transaction structure */
92 trace_xfs_dir2_block_addname(args);
94 dp = args->dp;
95 tp = args->trans;
96 mp = dp->i_mount;
98 * Read the (one and only) directory block into dabuf bp.
100 if ((error =
101 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
102 return error;
104 ASSERT(bp != NULL);
105 hdr = bp->data;
107 * Check the magic number, corrupted if wrong.
109 if (unlikely(hdr->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))) {
110 XFS_CORRUPTION_ERROR("xfs_dir2_block_addname",
111 XFS_ERRLEVEL_LOW, mp, hdr);
112 xfs_da_brelse(tp, bp);
113 return XFS_ERROR(EFSCORRUPTED);
115 len = xfs_dir2_data_entsize(args->namelen);
117 * Set up pointers to parts of the block.
119 bf = hdr->bestfree;
120 btp = xfs_dir2_block_tail_p(mp, hdr);
121 blp = xfs_dir2_block_leaf_p(btp);
123 * No stale entries? Need space for entry and new leaf.
125 if (!btp->stale) {
127 * Tag just before the first leaf entry.
129 tagp = (__be16 *)blp - 1;
131 * Data object just before the first leaf entry.
133 enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
135 * If it's not free then can't do this add without cleaning up:
136 * the space before the first leaf entry needs to be free so it
137 * can be expanded to hold the pointer to the new entry.
139 if (be16_to_cpu(enddup->freetag) != XFS_DIR2_DATA_FREE_TAG)
140 dup = enddup = NULL;
142 * Check out the biggest freespace and see if it's the same one.
144 else {
145 dup = (xfs_dir2_data_unused_t *)
146 ((char *)hdr + be16_to_cpu(bf[0].offset));
147 if (dup == enddup) {
149 * It is the biggest freespace, is it too small
150 * to hold the new leaf too?
152 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
154 * Yes, we use the second-largest
155 * entry instead if it works.
157 if (be16_to_cpu(bf[1].length) >= len)
158 dup = (xfs_dir2_data_unused_t *)
159 ((char *)hdr +
160 be16_to_cpu(bf[1].offset));
161 else
162 dup = NULL;
164 } else {
166 * Not the same free entry,
167 * just check its length.
169 if (be16_to_cpu(dup->length) < len) {
170 dup = NULL;
174 compact = 0;
177 * If there are stale entries we'll use one for the leaf.
178 * Is the biggest entry enough to avoid compaction?
180 else if (be16_to_cpu(bf[0].length) >= len) {
181 dup = (xfs_dir2_data_unused_t *)
182 ((char *)hdr + be16_to_cpu(bf[0].offset));
183 compact = 0;
186 * Will need to compact to make this work.
188 else {
190 * Tag just before the first leaf entry.
192 tagp = (__be16 *)blp - 1;
194 * Data object just before the first leaf entry.
196 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
198 * If it's not free then the data will go where the
199 * leaf data starts now, if it works at all.
201 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
202 if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
203 (uint)sizeof(*blp) < len)
204 dup = NULL;
205 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
206 dup = NULL;
207 else
208 dup = (xfs_dir2_data_unused_t *)blp;
209 compact = 1;
212 * If this isn't a real add, we're done with the buffer.
214 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
215 xfs_da_brelse(tp, bp);
217 * If we don't have space for the new entry & leaf ...
219 if (!dup) {
221 * Not trying to actually do anything, or don't have
222 * a space reservation: return no-space.
224 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) || args->total == 0)
225 return XFS_ERROR(ENOSPC);
227 * Convert to the next larger format.
228 * Then add the new entry in that format.
230 error = xfs_dir2_block_to_leaf(args, bp);
231 xfs_da_buf_done(bp);
232 if (error)
233 return error;
234 return xfs_dir2_leaf_addname(args);
237 * Just checking, and it would work, so say so.
239 if (args->op_flags & XFS_DA_OP_JUSTCHECK)
240 return 0;
241 needlog = needscan = 0;
243 * If need to compact the leaf entries, do it now.
244 * Leave the highest-numbered stale entry stale.
245 * XXX should be the one closest to mid but mid is not yet computed.
247 if (compact) {
248 int fromidx; /* source leaf index */
249 int toidx; /* target leaf index */
251 for (fromidx = toidx = be32_to_cpu(btp->count) - 1,
252 highstale = lfloghigh = -1;
253 fromidx >= 0;
254 fromidx--) {
255 if (blp[fromidx].address ==
256 cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
257 if (highstale == -1)
258 highstale = toidx;
259 else {
260 if (lfloghigh == -1)
261 lfloghigh = toidx;
262 continue;
265 if (fromidx < toidx)
266 blp[toidx] = blp[fromidx];
267 toidx--;
269 lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
270 lfloghigh -= be32_to_cpu(btp->stale) - 1;
271 be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
272 xfs_dir2_data_make_free(tp, bp,
273 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
274 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
275 &needlog, &needscan);
276 blp += be32_to_cpu(btp->stale) - 1;
277 btp->stale = cpu_to_be32(1);
279 * If we now need to rebuild the bestfree map, do so.
280 * This needs to happen before the next call to use_free.
282 if (needscan) {
283 xfs_dir2_data_freescan(mp, hdr, &needlog);
284 needscan = 0;
288 * Set leaf logging boundaries to impossible state.
289 * For the no-stale case they're set explicitly.
291 else if (btp->stale) {
292 lfloglow = be32_to_cpu(btp->count);
293 lfloghigh = -1;
296 * Find the slot that's first lower than our hash value, -1 if none.
298 for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
299 mid = (low + high) >> 1;
300 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
301 break;
302 if (hash < args->hashval)
303 low = mid + 1;
304 else
305 high = mid - 1;
307 while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
308 mid--;
311 * No stale entries, will use enddup space to hold new leaf.
313 if (!btp->stale) {
315 * Mark the space needed for the new leaf entry, now in use.
317 xfs_dir2_data_use_free(tp, bp, enddup,
318 (xfs_dir2_data_aoff_t)
319 ((char *)enddup - (char *)hdr + be16_to_cpu(enddup->length) -
320 sizeof(*blp)),
321 (xfs_dir2_data_aoff_t)sizeof(*blp),
322 &needlog, &needscan);
324 * Update the tail (entry count).
326 be32_add_cpu(&btp->count, 1);
328 * If we now need to rebuild the bestfree map, do so.
329 * This needs to happen before the next call to use_free.
331 if (needscan) {
332 xfs_dir2_data_freescan(mp, hdr, &needlog);
333 needscan = 0;
336 * Adjust pointer to the first leaf entry, we're about to move
337 * the table up one to open up space for the new leaf entry.
338 * Then adjust our index to match.
340 blp--;
341 mid++;
342 if (mid)
343 memmove(blp, &blp[1], mid * sizeof(*blp));
344 lfloglow = 0;
345 lfloghigh = mid;
348 * Use a stale leaf for our new entry.
350 else {
351 for (lowstale = mid;
352 lowstale >= 0 &&
353 blp[lowstale].address !=
354 cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
355 lowstale--)
356 continue;
357 for (highstale = mid + 1;
358 highstale < be32_to_cpu(btp->count) &&
359 blp[highstale].address !=
360 cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
361 (lowstale < 0 || mid - lowstale > highstale - mid);
362 highstale++)
363 continue;
365 * Move entries toward the low-numbered stale entry.
367 if (lowstale >= 0 &&
368 (highstale == be32_to_cpu(btp->count) ||
369 mid - lowstale <= highstale - mid)) {
370 if (mid - lowstale)
371 memmove(&blp[lowstale], &blp[lowstale + 1],
372 (mid - lowstale) * sizeof(*blp));
373 lfloglow = MIN(lowstale, lfloglow);
374 lfloghigh = MAX(mid, lfloghigh);
377 * Move entries toward the high-numbered stale entry.
379 else {
380 ASSERT(highstale < be32_to_cpu(btp->count));
381 mid++;
382 if (highstale - mid)
383 memmove(&blp[mid + 1], &blp[mid],
384 (highstale - mid) * sizeof(*blp));
385 lfloglow = MIN(mid, lfloglow);
386 lfloghigh = MAX(highstale, lfloghigh);
388 be32_add_cpu(&btp->stale, -1);
391 * Point to the new data entry.
393 dep = (xfs_dir2_data_entry_t *)dup;
395 * Fill in the leaf entry.
397 blp[mid].hashval = cpu_to_be32(args->hashval);
398 blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
399 (char *)dep - (char *)hdr));
400 xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
402 * Mark space for the data entry used.
404 xfs_dir2_data_use_free(tp, bp, dup,
405 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
406 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
408 * Create the new data entry.
410 dep->inumber = cpu_to_be64(args->inumber);
411 dep->namelen = args->namelen;
412 memcpy(dep->name, args->name, args->namelen);
413 tagp = xfs_dir2_data_entry_tag_p(dep);
414 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
416 * Clean up the bestfree array and log the header, tail, and entry.
418 if (needscan)
419 xfs_dir2_data_freescan(mp, hdr, &needlog);
420 if (needlog)
421 xfs_dir2_data_log_header(tp, bp);
422 xfs_dir2_block_log_tail(tp, bp);
423 xfs_dir2_data_log_entry(tp, bp, dep);
424 xfs_dir2_data_check(dp, bp);
425 xfs_da_buf_done(bp);
426 return 0;
430 * Readdir for block directories.
432 int /* error */
433 xfs_dir2_block_getdents(
434 xfs_inode_t *dp, /* incore inode */
435 void *dirent,
436 xfs_off_t *offset,
437 filldir_t filldir)
439 xfs_dir2_data_hdr_t *hdr; /* block header */
440 xfs_dabuf_t *bp; /* buffer for block */
441 xfs_dir2_block_tail_t *btp; /* block tail */
442 xfs_dir2_data_entry_t *dep; /* block data entry */
443 xfs_dir2_data_unused_t *dup; /* block unused entry */
444 char *endptr; /* end of the data entries */
445 int error; /* error return value */
446 xfs_mount_t *mp; /* filesystem mount point */
447 char *ptr; /* current data entry */
448 int wantoff; /* starting block offset */
449 xfs_off_t cook;
451 mp = dp->i_mount;
453 * If the block number in the offset is out of range, we're done.
455 if (xfs_dir2_dataptr_to_db(mp, *offset) > mp->m_dirdatablk) {
456 return 0;
459 * Can't read the block, give up, else get dabuf in bp.
461 error = xfs_da_read_buf(NULL, dp, mp->m_dirdatablk, -1,
462 &bp, XFS_DATA_FORK);
463 if (error)
464 return error;
466 ASSERT(bp != NULL);
468 * Extract the byte offset we start at from the seek pointer.
469 * We'll skip entries before this.
471 wantoff = xfs_dir2_dataptr_to_off(mp, *offset);
472 hdr = bp->data;
473 xfs_dir2_data_check(dp, bp);
475 * Set up values for the loop.
477 btp = xfs_dir2_block_tail_p(mp, hdr);
478 ptr = (char *)(hdr + 1);
479 endptr = (char *)xfs_dir2_block_leaf_p(btp);
482 * Loop over the data portion of the block.
483 * Each object is a real entry (dep) or an unused one (dup).
485 while (ptr < endptr) {
486 dup = (xfs_dir2_data_unused_t *)ptr;
488 * Unused, skip it.
490 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
491 ptr += be16_to_cpu(dup->length);
492 continue;
495 dep = (xfs_dir2_data_entry_t *)ptr;
498 * Bump pointer for the next iteration.
500 ptr += xfs_dir2_data_entsize(dep->namelen);
502 * The entry is before the desired starting point, skip it.
504 if ((char *)dep - (char *)hdr < wantoff)
505 continue;
507 cook = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
508 (char *)dep - (char *)hdr);
511 * If it didn't fit, set the final offset to here & return.
513 if (filldir(dirent, (char *)dep->name, dep->namelen,
514 cook & 0x7fffffff, be64_to_cpu(dep->inumber),
515 DT_UNKNOWN)) {
516 *offset = cook & 0x7fffffff;
517 xfs_da_brelse(NULL, bp);
518 return 0;
523 * Reached the end of the block.
524 * Set the offset to a non-existent block 1 and return.
526 *offset = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk + 1, 0) &
527 0x7fffffff;
528 xfs_da_brelse(NULL, bp);
529 return 0;
533 * Log leaf entries from the block.
535 static void
536 xfs_dir2_block_log_leaf(
537 xfs_trans_t *tp, /* transaction structure */
538 xfs_dabuf_t *bp, /* block buffer */
539 int first, /* index of first logged leaf */
540 int last) /* index of last logged leaf */
542 xfs_dir2_data_hdr_t *hdr = bp->data;
543 xfs_dir2_leaf_entry_t *blp;
544 xfs_dir2_block_tail_t *btp;
546 btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
547 blp = xfs_dir2_block_leaf_p(btp);
548 xfs_da_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
549 (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
553 * Log the block tail.
555 static void
556 xfs_dir2_block_log_tail(
557 xfs_trans_t *tp, /* transaction structure */
558 xfs_dabuf_t *bp) /* block buffer */
560 xfs_dir2_data_hdr_t *hdr = bp->data;
561 xfs_dir2_block_tail_t *btp;
563 btp = xfs_dir2_block_tail_p(tp->t_mountp, hdr);
564 xfs_da_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
565 (uint)((char *)(btp + 1) - (char *)hdr - 1));
569 * Look up an entry in the block. This is the external routine,
570 * xfs_dir2_block_lookup_int does the real work.
572 int /* error */
573 xfs_dir2_block_lookup(
574 xfs_da_args_t *args) /* dir lookup arguments */
576 xfs_dir2_data_hdr_t *hdr; /* block header */
577 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
578 xfs_dabuf_t *bp; /* block buffer */
579 xfs_dir2_block_tail_t *btp; /* block tail */
580 xfs_dir2_data_entry_t *dep; /* block data entry */
581 xfs_inode_t *dp; /* incore inode */
582 int ent; /* entry index */
583 int error; /* error return value */
584 xfs_mount_t *mp; /* filesystem mount point */
586 trace_xfs_dir2_block_lookup(args);
589 * Get the buffer, look up the entry.
590 * If not found (ENOENT) then return, have no buffer.
592 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
593 return error;
594 dp = args->dp;
595 mp = dp->i_mount;
596 hdr = bp->data;
597 xfs_dir2_data_check(dp, bp);
598 btp = xfs_dir2_block_tail_p(mp, hdr);
599 blp = xfs_dir2_block_leaf_p(btp);
601 * Get the offset from the leaf entry, to point to the data.
603 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
604 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
606 * Fill in inode number, CI name if appropriate, release the block.
608 args->inumber = be64_to_cpu(dep->inumber);
609 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
610 xfs_da_brelse(args->trans, bp);
611 return XFS_ERROR(error);
615 * Internal block lookup routine.
617 static int /* error */
618 xfs_dir2_block_lookup_int(
619 xfs_da_args_t *args, /* dir lookup arguments */
620 xfs_dabuf_t **bpp, /* returned block buffer */
621 int *entno) /* returned entry number */
623 xfs_dir2_dataptr_t addr; /* data entry address */
624 xfs_dir2_data_hdr_t *hdr; /* block header */
625 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
626 xfs_dabuf_t *bp; /* block buffer */
627 xfs_dir2_block_tail_t *btp; /* block tail */
628 xfs_dir2_data_entry_t *dep; /* block data entry */
629 xfs_inode_t *dp; /* incore inode */
630 int error; /* error return value */
631 xfs_dahash_t hash; /* found hash value */
632 int high; /* binary search high index */
633 int low; /* binary search low index */
634 int mid; /* binary search current idx */
635 xfs_mount_t *mp; /* filesystem mount point */
636 xfs_trans_t *tp; /* transaction pointer */
637 enum xfs_dacmp cmp; /* comparison result */
639 dp = args->dp;
640 tp = args->trans;
641 mp = dp->i_mount;
643 * Read the buffer, return error if we can't get it.
645 if ((error =
646 xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &bp, XFS_DATA_FORK))) {
647 return error;
649 ASSERT(bp != NULL);
650 hdr = bp->data;
651 xfs_dir2_data_check(dp, bp);
652 btp = xfs_dir2_block_tail_p(mp, hdr);
653 blp = xfs_dir2_block_leaf_p(btp);
655 * Loop doing a binary search for our hash value.
656 * Find our entry, ENOENT if it's not there.
658 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
659 ASSERT(low <= high);
660 mid = (low + high) >> 1;
661 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
662 break;
663 if (hash < args->hashval)
664 low = mid + 1;
665 else
666 high = mid - 1;
667 if (low > high) {
668 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
669 xfs_da_brelse(tp, bp);
670 return XFS_ERROR(ENOENT);
674 * Back up to the first one with the right hash value.
676 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
677 mid--;
680 * Now loop forward through all the entries with the
681 * right hash value looking for our name.
683 do {
684 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
685 continue;
687 * Get pointer to the entry from the leaf.
689 dep = (xfs_dir2_data_entry_t *)
690 ((char *)hdr + xfs_dir2_dataptr_to_off(mp, addr));
692 * Compare name and if it's an exact match, return the index
693 * and buffer. If it's the first case-insensitive match, store
694 * the index and buffer and continue looking for an exact match.
696 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
697 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
698 args->cmpresult = cmp;
699 *bpp = bp;
700 *entno = mid;
701 if (cmp == XFS_CMP_EXACT)
702 return 0;
704 } while (++mid < be32_to_cpu(btp->count) &&
705 be32_to_cpu(blp[mid].hashval) == hash);
707 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
709 * Here, we can only be doing a lookup (not a rename or replace).
710 * If a case-insensitive match was found earlier, return success.
712 if (args->cmpresult == XFS_CMP_CASE)
713 return 0;
715 * No match, release the buffer and return ENOENT.
717 xfs_da_brelse(tp, bp);
718 return XFS_ERROR(ENOENT);
722 * Remove an entry from a block format directory.
723 * If that makes the block small enough to fit in shortform, transform it.
725 int /* error */
726 xfs_dir2_block_removename(
727 xfs_da_args_t *args) /* directory operation args */
729 xfs_dir2_data_hdr_t *hdr; /* block header */
730 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
731 xfs_dabuf_t *bp; /* block buffer */
732 xfs_dir2_block_tail_t *btp; /* block tail */
733 xfs_dir2_data_entry_t *dep; /* block data entry */
734 xfs_inode_t *dp; /* incore inode */
735 int ent; /* block leaf entry index */
736 int error; /* error return value */
737 xfs_mount_t *mp; /* filesystem mount point */
738 int needlog; /* need to log block header */
739 int needscan; /* need to fixup bestfree */
740 xfs_dir2_sf_hdr_t sfh; /* shortform header */
741 int size; /* shortform size */
742 xfs_trans_t *tp; /* transaction pointer */
744 trace_xfs_dir2_block_removename(args);
747 * Look up the entry in the block. Gets the buffer and entry index.
748 * It will always be there, the vnodeops level does a lookup first.
750 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
751 return error;
753 dp = args->dp;
754 tp = args->trans;
755 mp = dp->i_mount;
756 hdr = bp->data;
757 btp = xfs_dir2_block_tail_p(mp, hdr);
758 blp = xfs_dir2_block_leaf_p(btp);
760 * Point to the data entry using the leaf entry.
762 dep = (xfs_dir2_data_entry_t *)
763 ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
765 * Mark the data entry's space free.
767 needlog = needscan = 0;
768 xfs_dir2_data_make_free(tp, bp,
769 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
770 xfs_dir2_data_entsize(dep->namelen), &needlog, &needscan);
772 * Fix up the block tail.
774 be32_add_cpu(&btp->stale, 1);
775 xfs_dir2_block_log_tail(tp, bp);
777 * Remove the leaf entry by marking it stale.
779 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
780 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
782 * Fix up bestfree, log the header if necessary.
784 if (needscan)
785 xfs_dir2_data_freescan(mp, hdr, &needlog);
786 if (needlog)
787 xfs_dir2_data_log_header(tp, bp);
788 xfs_dir2_data_check(dp, bp);
790 * See if the size as a shortform is good enough.
792 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
793 if (size > XFS_IFORK_DSIZE(dp)) {
794 xfs_da_buf_done(bp);
795 return 0;
798 * If it works, do the conversion.
800 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
804 * Replace an entry in a V2 block directory.
805 * Change the inode number to the new value.
807 int /* error */
808 xfs_dir2_block_replace(
809 xfs_da_args_t *args) /* directory operation args */
811 xfs_dir2_data_hdr_t *hdr; /* block header */
812 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
813 xfs_dabuf_t *bp; /* block buffer */
814 xfs_dir2_block_tail_t *btp; /* block tail */
815 xfs_dir2_data_entry_t *dep; /* block data entry */
816 xfs_inode_t *dp; /* incore inode */
817 int ent; /* leaf entry index */
818 int error; /* error return value */
819 xfs_mount_t *mp; /* filesystem mount point */
821 trace_xfs_dir2_block_replace(args);
824 * Lookup the entry in the directory. Get buffer and entry index.
825 * This will always succeed since the caller has already done a lookup.
827 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
828 return error;
830 dp = args->dp;
831 mp = dp->i_mount;
832 hdr = bp->data;
833 btp = xfs_dir2_block_tail_p(mp, hdr);
834 blp = xfs_dir2_block_leaf_p(btp);
836 * Point to the data entry we need to change.
838 dep = (xfs_dir2_data_entry_t *)
839 ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
840 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
842 * Change the inode number to the new value.
844 dep->inumber = cpu_to_be64(args->inumber);
845 xfs_dir2_data_log_entry(args->trans, bp, dep);
846 xfs_dir2_data_check(dp, bp);
847 xfs_da_buf_done(bp);
848 return 0;
852 * Qsort comparison routine for the block leaf entries.
854 static int /* sort order */
855 xfs_dir2_block_sort(
856 const void *a, /* first leaf entry */
857 const void *b) /* second leaf entry */
859 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
860 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
862 la = a;
863 lb = b;
864 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
865 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
869 * Convert a V2 leaf directory to a V2 block directory if possible.
871 int /* error */
872 xfs_dir2_leaf_to_block(
873 xfs_da_args_t *args, /* operation arguments */
874 xfs_dabuf_t *lbp, /* leaf buffer */
875 xfs_dabuf_t *dbp) /* data buffer */
877 __be16 *bestsp; /* leaf bests table */
878 xfs_dir2_data_hdr_t *hdr; /* block header */
879 xfs_dir2_block_tail_t *btp; /* block tail */
880 xfs_inode_t *dp; /* incore directory inode */
881 xfs_dir2_data_unused_t *dup; /* unused data entry */
882 int error; /* error return value */
883 int from; /* leaf from index */
884 xfs_dir2_leaf_t *leaf; /* leaf structure */
885 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
886 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
887 xfs_mount_t *mp; /* file system mount point */
888 int needlog; /* need to log data header */
889 int needscan; /* need to scan for bestfree */
890 xfs_dir2_sf_hdr_t sfh; /* shortform header */
891 int size; /* bytes used */
892 __be16 *tagp; /* end of entry (tag) */
893 int to; /* block/leaf to index */
894 xfs_trans_t *tp; /* transaction pointer */
896 trace_xfs_dir2_leaf_to_block(args);
898 dp = args->dp;
899 tp = args->trans;
900 mp = dp->i_mount;
901 leaf = lbp->data;
902 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC));
903 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
905 * If there are data blocks other than the first one, take this
906 * opportunity to remove trailing empty data blocks that may have
907 * been left behind during no-space-reservation operations.
908 * These will show up in the leaf bests table.
910 while (dp->i_d.di_size > mp->m_dirblksize) {
911 bestsp = xfs_dir2_leaf_bests_p(ltp);
912 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
913 mp->m_dirblksize - (uint)sizeof(*hdr)) {
914 if ((error =
915 xfs_dir2_leaf_trim_data(args, lbp,
916 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
917 goto out;
918 } else {
919 error = 0;
920 goto out;
924 * Read the data block if we don't already have it, give up if it fails.
926 if (dbp == NULL &&
927 (error = xfs_da_read_buf(tp, dp, mp->m_dirdatablk, -1, &dbp,
928 XFS_DATA_FORK))) {
929 goto out;
931 hdr = dbp->data;
932 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC));
934 * Size of the "leaf" area in the block.
936 size = (uint)sizeof(xfs_dir2_block_tail_t) +
937 (uint)sizeof(*lep) * (be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
939 * Look at the last data entry.
941 tagp = (__be16 *)((char *)hdr + mp->m_dirblksize) - 1;
942 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
944 * If it's not free or is too short we can't do it.
946 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
947 be16_to_cpu(dup->length) < size) {
948 error = 0;
949 goto out;
952 * Start converting it to block form.
954 hdr->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
955 needlog = 1;
956 needscan = 0;
958 * Use up the space at the end of the block (blp/btp).
960 xfs_dir2_data_use_free(tp, dbp, dup, mp->m_dirblksize - size, size,
961 &needlog, &needscan);
963 * Initialize the block tail.
965 btp = xfs_dir2_block_tail_p(mp, hdr);
966 btp->count = cpu_to_be32(be16_to_cpu(leaf->hdr.count) - be16_to_cpu(leaf->hdr.stale));
967 btp->stale = 0;
968 xfs_dir2_block_log_tail(tp, dbp);
970 * Initialize the block leaf area. We compact out stale entries.
972 lep = xfs_dir2_block_leaf_p(btp);
973 for (from = to = 0; from < be16_to_cpu(leaf->hdr.count); from++) {
974 if (leaf->ents[from].address ==
975 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
976 continue;
977 lep[to++] = leaf->ents[from];
979 ASSERT(to == be32_to_cpu(btp->count));
980 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
982 * Scan the bestfree if we need it and log the data block header.
984 if (needscan)
985 xfs_dir2_data_freescan(mp, hdr, &needlog);
986 if (needlog)
987 xfs_dir2_data_log_header(tp, dbp);
989 * Pitch the old leaf block.
991 error = xfs_da_shrink_inode(args, mp->m_dirleafblk, lbp);
992 lbp = NULL;
993 if (error) {
994 goto out;
997 * Now see if the resulting block can be shrunken to shortform.
999 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1000 if (size > XFS_IFORK_DSIZE(dp)) {
1001 error = 0;
1002 goto out;
1004 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1005 out:
1006 if (lbp)
1007 xfs_da_buf_done(lbp);
1008 if (dbp)
1009 xfs_da_buf_done(dbp);
1010 return error;
1014 * Convert the shortform directory to block form.
1016 int /* error */
1017 xfs_dir2_sf_to_block(
1018 xfs_da_args_t *args) /* operation arguments */
1020 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1021 xfs_dir2_data_hdr_t *hdr; /* block header */
1022 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1023 xfs_dabuf_t *bp; /* block buffer */
1024 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1025 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1026 xfs_inode_t *dp; /* incore directory inode */
1027 int dummy; /* trash */
1028 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1029 int endoffset; /* end of data objects */
1030 int error; /* error return value */
1031 int i; /* index */
1032 xfs_mount_t *mp; /* filesystem mount point */
1033 int needlog; /* need to log block header */
1034 int needscan; /* need to scan block freespc */
1035 int newoffset; /* offset from current entry */
1036 int offset; /* target block offset */
1037 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1038 xfs_dir2_sf_hdr_t *oldsfp; /* old shortform header */
1039 xfs_dir2_sf_hdr_t *sfp; /* shortform header */
1040 __be16 *tagp; /* end of data entry */
1041 xfs_trans_t *tp; /* transaction pointer */
1042 struct xfs_name name;
1044 trace_xfs_dir2_sf_to_block(args);
1046 dp = args->dp;
1047 tp = args->trans;
1048 mp = dp->i_mount;
1049 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
1051 * Bomb out if the shortform directory is way too short.
1053 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1054 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1055 return XFS_ERROR(EIO);
1058 oldsfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
1060 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
1061 ASSERT(dp->i_df.if_u1.if_data != NULL);
1062 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1065 * Copy the directory into a temporary buffer.
1066 * Then pitch the incore inode data so we can make extents.
1068 sfp = kmem_alloc(dp->i_df.if_bytes, KM_SLEEP);
1069 memcpy(sfp, oldsfp, dp->i_df.if_bytes);
1071 xfs_idata_realloc(dp, -dp->i_df.if_bytes, XFS_DATA_FORK);
1072 dp->i_d.di_size = 0;
1073 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
1076 * Add block 0 to the inode.
1078 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1079 if (error) {
1080 kmem_free(sfp);
1081 return error;
1084 * Initialize the data block.
1086 error = xfs_dir2_data_init(args, blkno, &bp);
1087 if (error) {
1088 kmem_free(sfp);
1089 return error;
1091 hdr = bp->data;
1092 hdr->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
1094 * Compute size of block "tail" area.
1096 i = (uint)sizeof(*btp) +
1097 (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1099 * The whole thing is initialized to free by the init routine.
1100 * Say we're using the leaf and tail area.
1102 dup = (xfs_dir2_data_unused_t *)(hdr + 1);
1103 needlog = needscan = 0;
1104 xfs_dir2_data_use_free(tp, bp, dup, mp->m_dirblksize - i, i, &needlog,
1105 &needscan);
1106 ASSERT(needscan == 0);
1108 * Fill in the tail.
1110 btp = xfs_dir2_block_tail_p(mp, hdr);
1111 btp->count = cpu_to_be32(sfp->count + 2); /* ., .. */
1112 btp->stale = 0;
1113 blp = xfs_dir2_block_leaf_p(btp);
1114 endoffset = (uint)((char *)blp - (char *)hdr);
1116 * Remove the freespace, we'll manage it.
1118 xfs_dir2_data_use_free(tp, bp, dup,
1119 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1120 be16_to_cpu(dup->length), &needlog, &needscan);
1122 * Create entry for .
1124 dep = (xfs_dir2_data_entry_t *)
1125 ((char *)hdr + XFS_DIR2_DATA_DOT_OFFSET);
1126 dep->inumber = cpu_to_be64(dp->i_ino);
1127 dep->namelen = 1;
1128 dep->name[0] = '.';
1129 tagp = xfs_dir2_data_entry_tag_p(dep);
1130 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1131 xfs_dir2_data_log_entry(tp, bp, dep);
1132 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1133 blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1134 (char *)dep - (char *)hdr));
1136 * Create entry for ..
1138 dep = (xfs_dir2_data_entry_t *)
1139 ((char *)hdr + XFS_DIR2_DATA_DOTDOT_OFFSET);
1140 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_parent_ino(sfp));
1141 dep->namelen = 2;
1142 dep->name[0] = dep->name[1] = '.';
1143 tagp = xfs_dir2_data_entry_tag_p(dep);
1144 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1145 xfs_dir2_data_log_entry(tp, bp, dep);
1146 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1147 blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1148 (char *)dep - (char *)hdr));
1149 offset = XFS_DIR2_DATA_FIRST_OFFSET;
1151 * Loop over existing entries, stuff them in.
1153 i = 0;
1154 if (!sfp->count)
1155 sfep = NULL;
1156 else
1157 sfep = xfs_dir2_sf_firstentry(sfp);
1159 * Need to preserve the existing offset values in the sf directory.
1160 * Insert holes (unused entries) where necessary.
1162 while (offset < endoffset) {
1164 * sfep is null when we reach the end of the list.
1166 if (sfep == NULL)
1167 newoffset = endoffset;
1168 else
1169 newoffset = xfs_dir2_sf_get_offset(sfep);
1171 * There should be a hole here, make one.
1173 if (offset < newoffset) {
1174 dup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1175 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1176 dup->length = cpu_to_be16(newoffset - offset);
1177 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1178 ((char *)dup - (char *)hdr));
1179 xfs_dir2_data_log_unused(tp, bp, dup);
1180 xfs_dir2_data_freeinsert(hdr, dup, &dummy);
1181 offset += be16_to_cpu(dup->length);
1182 continue;
1185 * Copy a real entry.
1187 dep = (xfs_dir2_data_entry_t *)((char *)hdr + newoffset);
1188 dep->inumber = cpu_to_be64(xfs_dir2_sfe_get_ino(sfp, sfep));
1189 dep->namelen = sfep->namelen;
1190 memcpy(dep->name, sfep->name, dep->namelen);
1191 tagp = xfs_dir2_data_entry_tag_p(dep);
1192 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1193 xfs_dir2_data_log_entry(tp, bp, dep);
1194 name.name = sfep->name;
1195 name.len = sfep->namelen;
1196 blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1197 hashname(&name));
1198 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(mp,
1199 (char *)dep - (char *)hdr));
1200 offset = (int)((char *)(tagp + 1) - (char *)hdr);
1201 if (++i == sfp->count)
1202 sfep = NULL;
1203 else
1204 sfep = xfs_dir2_sf_nextentry(sfp, sfep);
1206 /* Done with the temporary buffer */
1207 kmem_free(sfp);
1209 * Sort the leaf entries by hash value.
1211 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1213 * Log the leaf entry area and tail.
1214 * Already logged the header in data_init, ignore needlog.
1216 ASSERT(needscan == 0);
1217 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1218 xfs_dir2_block_log_tail(tp, bp);
1219 xfs_dir2_data_check(dp, bp);
1220 xfs_da_buf_done(bp);
1221 return 0;