Linux 3.16-rc2
[linux/fpc-iii.git] / fs / xfs / xfs_dir2_block.c
blobc7cd3154026a2eac267af117ad6be5f7230953d5
1 /*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
4 * All Rights Reserved.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_sb.h"
25 #include "xfs_ag.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
31 #include "xfs_inode_item.h"
32 #include "xfs_bmap.h"
33 #include "xfs_buf_item.h"
34 #include "xfs_dir2.h"
35 #include "xfs_dir2_priv.h"
36 #include "xfs_error.h"
37 #include "xfs_trace.h"
38 #include "xfs_cksum.h"
39 #include "xfs_dinode.h"
42 * Local function prototypes.
44 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, struct xfs_buf *bp,
45 int first, int last);
46 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, struct xfs_buf *bp);
47 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, struct xfs_buf **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((unsigned char *)".", 1);
60 xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
63 static bool
64 xfs_dir3_block_verify(
65 struct xfs_buf *bp)
67 struct xfs_mount *mp = bp->b_target->bt_mount;
68 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
70 if (xfs_sb_version_hascrc(&mp->m_sb)) {
71 if (hdr3->magic != cpu_to_be32(XFS_DIR3_BLOCK_MAGIC))
72 return false;
73 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_uuid))
74 return false;
75 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
76 return false;
77 } else {
78 if (hdr3->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
79 return false;
81 if (__xfs_dir3_data_check(NULL, bp))
82 return false;
83 return true;
86 static void
87 xfs_dir3_block_read_verify(
88 struct xfs_buf *bp)
90 struct xfs_mount *mp = bp->b_target->bt_mount;
92 if (xfs_sb_version_hascrc(&mp->m_sb) &&
93 !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
94 xfs_buf_ioerror(bp, EFSBADCRC);
95 else if (!xfs_dir3_block_verify(bp))
96 xfs_buf_ioerror(bp, EFSCORRUPTED);
98 if (bp->b_error)
99 xfs_verifier_error(bp);
102 static void
103 xfs_dir3_block_write_verify(
104 struct xfs_buf *bp)
106 struct xfs_mount *mp = bp->b_target->bt_mount;
107 struct xfs_buf_log_item *bip = bp->b_fspriv;
108 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
110 if (!xfs_dir3_block_verify(bp)) {
111 xfs_buf_ioerror(bp, EFSCORRUPTED);
112 xfs_verifier_error(bp);
113 return;
116 if (!xfs_sb_version_hascrc(&mp->m_sb))
117 return;
119 if (bip)
120 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
122 xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
125 const struct xfs_buf_ops xfs_dir3_block_buf_ops = {
126 .verify_read = xfs_dir3_block_read_verify,
127 .verify_write = xfs_dir3_block_write_verify,
131 xfs_dir3_block_read(
132 struct xfs_trans *tp,
133 struct xfs_inode *dp,
134 struct xfs_buf **bpp)
136 struct xfs_mount *mp = dp->i_mount;
137 int err;
139 err = xfs_da_read_buf(tp, dp, mp->m_dir_geo->datablk, -1, bpp,
140 XFS_DATA_FORK, &xfs_dir3_block_buf_ops);
141 if (!err && tp)
142 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_BLOCK_BUF);
143 return err;
146 static void
147 xfs_dir3_block_init(
148 struct xfs_mount *mp,
149 struct xfs_trans *tp,
150 struct xfs_buf *bp,
151 struct xfs_inode *dp)
153 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
155 bp->b_ops = &xfs_dir3_block_buf_ops;
156 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_BLOCK_BUF);
158 if (xfs_sb_version_hascrc(&mp->m_sb)) {
159 memset(hdr3, 0, sizeof(*hdr3));
160 hdr3->magic = cpu_to_be32(XFS_DIR3_BLOCK_MAGIC);
161 hdr3->blkno = cpu_to_be64(bp->b_bn);
162 hdr3->owner = cpu_to_be64(dp->i_ino);
163 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_uuid);
164 return;
167 hdr3->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
170 static void
171 xfs_dir2_block_need_space(
172 struct xfs_inode *dp,
173 struct xfs_dir2_data_hdr *hdr,
174 struct xfs_dir2_block_tail *btp,
175 struct xfs_dir2_leaf_entry *blp,
176 __be16 **tagpp,
177 struct xfs_dir2_data_unused **dupp,
178 struct xfs_dir2_data_unused **enddupp,
179 int *compact,
180 int len)
182 struct xfs_dir2_data_free *bf;
183 __be16 *tagp = NULL;
184 struct xfs_dir2_data_unused *dup = NULL;
185 struct xfs_dir2_data_unused *enddup = NULL;
187 *compact = 0;
188 bf = dp->d_ops->data_bestfree_p(hdr);
191 * If there are stale entries we'll use one for the leaf.
193 if (btp->stale) {
194 if (be16_to_cpu(bf[0].length) >= len) {
196 * The biggest entry enough to avoid compaction.
198 dup = (xfs_dir2_data_unused_t *)
199 ((char *)hdr + be16_to_cpu(bf[0].offset));
200 goto out;
204 * Will need to compact to make this work.
205 * Tag just before the first leaf entry.
207 *compact = 1;
208 tagp = (__be16 *)blp - 1;
210 /* Data object just before the first leaf entry. */
211 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
214 * If it's not free then the data will go where the
215 * leaf data starts now, if it works at all.
217 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
218 if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
219 (uint)sizeof(*blp) < len)
220 dup = NULL;
221 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
222 dup = NULL;
223 else
224 dup = (xfs_dir2_data_unused_t *)blp;
225 goto out;
229 * no stale entries, so just use free space.
230 * Tag just before the first leaf entry.
232 tagp = (__be16 *)blp - 1;
234 /* Data object just before the first leaf entry. */
235 enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
238 * If it's not free then can't do this add without cleaning up:
239 * the space before the first leaf entry needs to be free so it
240 * can be expanded to hold the pointer to the new entry.
242 if (be16_to_cpu(enddup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
244 * Check out the biggest freespace and see if it's the same one.
246 dup = (xfs_dir2_data_unused_t *)
247 ((char *)hdr + be16_to_cpu(bf[0].offset));
248 if (dup != enddup) {
250 * Not the same free entry, just check its length.
252 if (be16_to_cpu(dup->length) < len)
253 dup = NULL;
254 goto out;
258 * It is the biggest freespace, can it hold the leaf too?
260 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
262 * Yes, use the second-largest entry instead if it works.
264 if (be16_to_cpu(bf[1].length) >= len)
265 dup = (xfs_dir2_data_unused_t *)
266 ((char *)hdr + be16_to_cpu(bf[1].offset));
267 else
268 dup = NULL;
271 out:
272 *tagpp = tagp;
273 *dupp = dup;
274 *enddupp = enddup;
278 * compact the leaf entries.
279 * Leave the highest-numbered stale entry stale.
280 * XXX should be the one closest to mid but mid is not yet computed.
282 static void
283 xfs_dir2_block_compact(
284 struct xfs_da_args *args,
285 struct xfs_buf *bp,
286 struct xfs_dir2_data_hdr *hdr,
287 struct xfs_dir2_block_tail *btp,
288 struct xfs_dir2_leaf_entry *blp,
289 int *needlog,
290 int *lfloghigh,
291 int *lfloglow)
293 int fromidx; /* source leaf index */
294 int toidx; /* target leaf index */
295 int needscan = 0;
296 int highstale; /* high stale index */
298 fromidx = toidx = be32_to_cpu(btp->count) - 1;
299 highstale = *lfloghigh = -1;
300 for (; fromidx >= 0; fromidx--) {
301 if (blp[fromidx].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
302 if (highstale == -1)
303 highstale = toidx;
304 else {
305 if (*lfloghigh == -1)
306 *lfloghigh = toidx;
307 continue;
310 if (fromidx < toidx)
311 blp[toidx] = blp[fromidx];
312 toidx--;
314 *lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
315 *lfloghigh -= be32_to_cpu(btp->stale) - 1;
316 be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
317 xfs_dir2_data_make_free(args, bp,
318 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
319 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
320 needlog, &needscan);
321 btp->stale = cpu_to_be32(1);
323 * If we now need to rebuild the bestfree map, do so.
324 * This needs to happen before the next call to use_free.
326 if (needscan)
327 xfs_dir2_data_freescan(args->dp, hdr, needlog);
331 * Add an entry to a block directory.
333 int /* error */
334 xfs_dir2_block_addname(
335 xfs_da_args_t *args) /* directory op arguments */
337 xfs_dir2_data_hdr_t *hdr; /* block header */
338 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
339 struct xfs_buf *bp; /* buffer for block */
340 xfs_dir2_block_tail_t *btp; /* block tail */
341 int compact; /* need to compact leaf ents */
342 xfs_dir2_data_entry_t *dep; /* block data entry */
343 xfs_inode_t *dp; /* directory inode */
344 xfs_dir2_data_unused_t *dup; /* block unused entry */
345 int error; /* error return value */
346 xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
347 xfs_dahash_t hash; /* hash value of found entry */
348 int high; /* high index for binary srch */
349 int highstale; /* high stale index */
350 int lfloghigh=0; /* last final leaf to log */
351 int lfloglow=0; /* first final leaf to log */
352 int len; /* length of the new entry */
353 int low; /* low index for binary srch */
354 int lowstale; /* low stale index */
355 int mid=0; /* midpoint for binary srch */
356 xfs_mount_t *mp; /* filesystem mount point */
357 int needlog; /* need to log header */
358 int needscan; /* need to rescan freespace */
359 __be16 *tagp; /* pointer to tag value */
360 xfs_trans_t *tp; /* transaction structure */
362 trace_xfs_dir2_block_addname(args);
364 dp = args->dp;
365 tp = args->trans;
366 mp = dp->i_mount;
368 /* Read the (one and only) directory block into bp. */
369 error = xfs_dir3_block_read(tp, dp, &bp);
370 if (error)
371 return error;
373 len = dp->d_ops->data_entsize(args->namelen);
376 * Set up pointers to parts of the block.
378 hdr = bp->b_addr;
379 btp = xfs_dir2_block_tail_p(args->geo, hdr);
380 blp = xfs_dir2_block_leaf_p(btp);
383 * Find out if we can reuse stale entries or whether we need extra
384 * space for entry and new leaf.
386 xfs_dir2_block_need_space(dp, hdr, btp, blp, &tagp, &dup,
387 &enddup, &compact, len);
390 * Done everything we need for a space check now.
392 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
393 xfs_trans_brelse(tp, bp);
394 if (!dup)
395 return XFS_ERROR(ENOSPC);
396 return 0;
400 * If we don't have space for the new entry & leaf ...
402 if (!dup) {
403 /* Don't have a space reservation: return no-space. */
404 if (args->total == 0)
405 return XFS_ERROR(ENOSPC);
407 * Convert to the next larger format.
408 * Then add the new entry in that format.
410 error = xfs_dir2_block_to_leaf(args, bp);
411 if (error)
412 return error;
413 return xfs_dir2_leaf_addname(args);
416 needlog = needscan = 0;
419 * If need to compact the leaf entries, do it now.
421 if (compact) {
422 xfs_dir2_block_compact(args, bp, hdr, btp, blp, &needlog,
423 &lfloghigh, &lfloglow);
424 /* recalculate blp post-compaction */
425 blp = xfs_dir2_block_leaf_p(btp);
426 } else if (btp->stale) {
428 * Set leaf logging boundaries to impossible state.
429 * For the no-stale case they're set explicitly.
431 lfloglow = be32_to_cpu(btp->count);
432 lfloghigh = -1;
436 * Find the slot that's first lower than our hash value, -1 if none.
438 for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
439 mid = (low + high) >> 1;
440 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
441 break;
442 if (hash < args->hashval)
443 low = mid + 1;
444 else
445 high = mid - 1;
447 while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
448 mid--;
451 * No stale entries, will use enddup space to hold new leaf.
453 if (!btp->stale) {
455 * Mark the space needed for the new leaf entry, now in use.
457 xfs_dir2_data_use_free(args, bp, enddup,
458 (xfs_dir2_data_aoff_t)
459 ((char *)enddup - (char *)hdr + be16_to_cpu(enddup->length) -
460 sizeof(*blp)),
461 (xfs_dir2_data_aoff_t)sizeof(*blp),
462 &needlog, &needscan);
464 * Update the tail (entry count).
466 be32_add_cpu(&btp->count, 1);
468 * If we now need to rebuild the bestfree map, do so.
469 * This needs to happen before the next call to use_free.
471 if (needscan) {
472 xfs_dir2_data_freescan(dp, hdr, &needlog);
473 needscan = 0;
476 * Adjust pointer to the first leaf entry, we're about to move
477 * the table up one to open up space for the new leaf entry.
478 * Then adjust our index to match.
480 blp--;
481 mid++;
482 if (mid)
483 memmove(blp, &blp[1], mid * sizeof(*blp));
484 lfloglow = 0;
485 lfloghigh = mid;
488 * Use a stale leaf for our new entry.
490 else {
491 for (lowstale = mid;
492 lowstale >= 0 &&
493 blp[lowstale].address !=
494 cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
495 lowstale--)
496 continue;
497 for (highstale = mid + 1;
498 highstale < be32_to_cpu(btp->count) &&
499 blp[highstale].address !=
500 cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
501 (lowstale < 0 || mid - lowstale > highstale - mid);
502 highstale++)
503 continue;
505 * Move entries toward the low-numbered stale entry.
507 if (lowstale >= 0 &&
508 (highstale == be32_to_cpu(btp->count) ||
509 mid - lowstale <= highstale - mid)) {
510 if (mid - lowstale)
511 memmove(&blp[lowstale], &blp[lowstale + 1],
512 (mid - lowstale) * sizeof(*blp));
513 lfloglow = MIN(lowstale, lfloglow);
514 lfloghigh = MAX(mid, lfloghigh);
517 * Move entries toward the high-numbered stale entry.
519 else {
520 ASSERT(highstale < be32_to_cpu(btp->count));
521 mid++;
522 if (highstale - mid)
523 memmove(&blp[mid + 1], &blp[mid],
524 (highstale - mid) * sizeof(*blp));
525 lfloglow = MIN(mid, lfloglow);
526 lfloghigh = MAX(highstale, lfloghigh);
528 be32_add_cpu(&btp->stale, -1);
531 * Point to the new data entry.
533 dep = (xfs_dir2_data_entry_t *)dup;
535 * Fill in the leaf entry.
537 blp[mid].hashval = cpu_to_be32(args->hashval);
538 blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
539 (char *)dep - (char *)hdr));
540 xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
542 * Mark space for the data entry used.
544 xfs_dir2_data_use_free(args, bp, dup,
545 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
546 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
548 * Create the new data entry.
550 dep->inumber = cpu_to_be64(args->inumber);
551 dep->namelen = args->namelen;
552 memcpy(dep->name, args->name, args->namelen);
553 dp->d_ops->data_put_ftype(dep, args->filetype);
554 tagp = dp->d_ops->data_entry_tag_p(dep);
555 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
557 * Clean up the bestfree array and log the header, tail, and entry.
559 if (needscan)
560 xfs_dir2_data_freescan(dp, hdr, &needlog);
561 if (needlog)
562 xfs_dir2_data_log_header(args, bp);
563 xfs_dir2_block_log_tail(tp, bp);
564 xfs_dir2_data_log_entry(args, bp, dep);
565 xfs_dir3_data_check(dp, bp);
566 return 0;
570 * Log leaf entries from the block.
572 static void
573 xfs_dir2_block_log_leaf(
574 xfs_trans_t *tp, /* transaction structure */
575 struct xfs_buf *bp, /* block buffer */
576 int first, /* index of first logged leaf */
577 int last) /* index of last logged leaf */
579 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
580 xfs_dir2_leaf_entry_t *blp;
581 xfs_dir2_block_tail_t *btp;
583 btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
584 blp = xfs_dir2_block_leaf_p(btp);
585 xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
586 (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
590 * Log the block tail.
592 static void
593 xfs_dir2_block_log_tail(
594 xfs_trans_t *tp, /* transaction structure */
595 struct xfs_buf *bp) /* block buffer */
597 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
598 xfs_dir2_block_tail_t *btp;
600 btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
601 xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
602 (uint)((char *)(btp + 1) - (char *)hdr - 1));
606 * Look up an entry in the block. This is the external routine,
607 * xfs_dir2_block_lookup_int does the real work.
609 int /* error */
610 xfs_dir2_block_lookup(
611 xfs_da_args_t *args) /* dir lookup arguments */
613 xfs_dir2_data_hdr_t *hdr; /* block header */
614 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
615 struct xfs_buf *bp; /* block buffer */
616 xfs_dir2_block_tail_t *btp; /* block tail */
617 xfs_dir2_data_entry_t *dep; /* block data entry */
618 xfs_inode_t *dp; /* incore inode */
619 int ent; /* entry index */
620 int error; /* error return value */
621 xfs_mount_t *mp; /* filesystem mount point */
623 trace_xfs_dir2_block_lookup(args);
626 * Get the buffer, look up the entry.
627 * If not found (ENOENT) then return, have no buffer.
629 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
630 return error;
631 dp = args->dp;
632 mp = dp->i_mount;
633 hdr = bp->b_addr;
634 xfs_dir3_data_check(dp, bp);
635 btp = xfs_dir2_block_tail_p(args->geo, hdr);
636 blp = xfs_dir2_block_leaf_p(btp);
638 * Get the offset from the leaf entry, to point to the data.
640 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
641 xfs_dir2_dataptr_to_off(args->geo,
642 be32_to_cpu(blp[ent].address)));
644 * Fill in inode number, CI name if appropriate, release the block.
646 args->inumber = be64_to_cpu(dep->inumber);
647 args->filetype = dp->d_ops->data_get_ftype(dep);
648 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
649 xfs_trans_brelse(args->trans, bp);
650 return XFS_ERROR(error);
654 * Internal block lookup routine.
656 static int /* error */
657 xfs_dir2_block_lookup_int(
658 xfs_da_args_t *args, /* dir lookup arguments */
659 struct xfs_buf **bpp, /* returned block buffer */
660 int *entno) /* returned entry number */
662 xfs_dir2_dataptr_t addr; /* data entry address */
663 xfs_dir2_data_hdr_t *hdr; /* block header */
664 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
665 struct xfs_buf *bp; /* block buffer */
666 xfs_dir2_block_tail_t *btp; /* block tail */
667 xfs_dir2_data_entry_t *dep; /* block data entry */
668 xfs_inode_t *dp; /* incore inode */
669 int error; /* error return value */
670 xfs_dahash_t hash; /* found hash value */
671 int high; /* binary search high index */
672 int low; /* binary search low index */
673 int mid; /* binary search current idx */
674 xfs_mount_t *mp; /* filesystem mount point */
675 xfs_trans_t *tp; /* transaction pointer */
676 enum xfs_dacmp cmp; /* comparison result */
678 dp = args->dp;
679 tp = args->trans;
680 mp = dp->i_mount;
682 error = xfs_dir3_block_read(tp, dp, &bp);
683 if (error)
684 return error;
686 hdr = bp->b_addr;
687 xfs_dir3_data_check(dp, bp);
688 btp = xfs_dir2_block_tail_p(args->geo, hdr);
689 blp = xfs_dir2_block_leaf_p(btp);
691 * Loop doing a binary search for our hash value.
692 * Find our entry, ENOENT if it's not there.
694 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
695 ASSERT(low <= high);
696 mid = (low + high) >> 1;
697 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
698 break;
699 if (hash < args->hashval)
700 low = mid + 1;
701 else
702 high = mid - 1;
703 if (low > high) {
704 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
705 xfs_trans_brelse(tp, bp);
706 return XFS_ERROR(ENOENT);
710 * Back up to the first one with the right hash value.
712 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
713 mid--;
716 * Now loop forward through all the entries with the
717 * right hash value looking for our name.
719 do {
720 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
721 continue;
723 * Get pointer to the entry from the leaf.
725 dep = (xfs_dir2_data_entry_t *)
726 ((char *)hdr + xfs_dir2_dataptr_to_off(args->geo, addr));
728 * Compare name and if it's an exact match, return the index
729 * and buffer. If it's the first case-insensitive match, store
730 * the index and buffer and continue looking for an exact match.
732 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
733 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
734 args->cmpresult = cmp;
735 *bpp = bp;
736 *entno = mid;
737 if (cmp == XFS_CMP_EXACT)
738 return 0;
740 } while (++mid < be32_to_cpu(btp->count) &&
741 be32_to_cpu(blp[mid].hashval) == hash);
743 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
745 * Here, we can only be doing a lookup (not a rename or replace).
746 * If a case-insensitive match was found earlier, return success.
748 if (args->cmpresult == XFS_CMP_CASE)
749 return 0;
751 * No match, release the buffer and return ENOENT.
753 xfs_trans_brelse(tp, bp);
754 return XFS_ERROR(ENOENT);
758 * Remove an entry from a block format directory.
759 * If that makes the block small enough to fit in shortform, transform it.
761 int /* error */
762 xfs_dir2_block_removename(
763 xfs_da_args_t *args) /* directory operation args */
765 xfs_dir2_data_hdr_t *hdr; /* block header */
766 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
767 struct xfs_buf *bp; /* block buffer */
768 xfs_dir2_block_tail_t *btp; /* block tail */
769 xfs_dir2_data_entry_t *dep; /* block data entry */
770 xfs_inode_t *dp; /* incore inode */
771 int ent; /* block leaf entry index */
772 int error; /* error return value */
773 xfs_mount_t *mp; /* filesystem mount point */
774 int needlog; /* need to log block header */
775 int needscan; /* need to fixup bestfree */
776 xfs_dir2_sf_hdr_t sfh; /* shortform header */
777 int size; /* shortform size */
778 xfs_trans_t *tp; /* transaction pointer */
780 trace_xfs_dir2_block_removename(args);
783 * Look up the entry in the block. Gets the buffer and entry index.
784 * It will always be there, the vnodeops level does a lookup first.
786 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
787 return error;
789 dp = args->dp;
790 tp = args->trans;
791 mp = dp->i_mount;
792 hdr = bp->b_addr;
793 btp = xfs_dir2_block_tail_p(args->geo, hdr);
794 blp = xfs_dir2_block_leaf_p(btp);
796 * Point to the data entry using the leaf entry.
798 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
799 xfs_dir2_dataptr_to_off(args->geo,
800 be32_to_cpu(blp[ent].address)));
802 * Mark the data entry's space free.
804 needlog = needscan = 0;
805 xfs_dir2_data_make_free(args, bp,
806 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
807 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
809 * Fix up the block tail.
811 be32_add_cpu(&btp->stale, 1);
812 xfs_dir2_block_log_tail(tp, bp);
814 * Remove the leaf entry by marking it stale.
816 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
817 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
819 * Fix up bestfree, log the header if necessary.
821 if (needscan)
822 xfs_dir2_data_freescan(dp, hdr, &needlog);
823 if (needlog)
824 xfs_dir2_data_log_header(args, bp);
825 xfs_dir3_data_check(dp, bp);
827 * See if the size as a shortform is good enough.
829 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
830 if (size > XFS_IFORK_DSIZE(dp))
831 return 0;
834 * If it works, do the conversion.
836 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
840 * Replace an entry in a V2 block directory.
841 * Change the inode number to the new value.
843 int /* error */
844 xfs_dir2_block_replace(
845 xfs_da_args_t *args) /* directory operation args */
847 xfs_dir2_data_hdr_t *hdr; /* block header */
848 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
849 struct xfs_buf *bp; /* block buffer */
850 xfs_dir2_block_tail_t *btp; /* block tail */
851 xfs_dir2_data_entry_t *dep; /* block data entry */
852 xfs_inode_t *dp; /* incore inode */
853 int ent; /* leaf entry index */
854 int error; /* error return value */
855 xfs_mount_t *mp; /* filesystem mount point */
857 trace_xfs_dir2_block_replace(args);
860 * Lookup the entry in the directory. Get buffer and entry index.
861 * This will always succeed since the caller has already done a lookup.
863 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
864 return error;
866 dp = args->dp;
867 mp = dp->i_mount;
868 hdr = bp->b_addr;
869 btp = xfs_dir2_block_tail_p(args->geo, hdr);
870 blp = xfs_dir2_block_leaf_p(btp);
872 * Point to the data entry we need to change.
874 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
875 xfs_dir2_dataptr_to_off(args->geo,
876 be32_to_cpu(blp[ent].address)));
877 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
879 * Change the inode number to the new value.
881 dep->inumber = cpu_to_be64(args->inumber);
882 dp->d_ops->data_put_ftype(dep, args->filetype);
883 xfs_dir2_data_log_entry(args, bp, dep);
884 xfs_dir3_data_check(dp, bp);
885 return 0;
889 * Qsort comparison routine for the block leaf entries.
891 static int /* sort order */
892 xfs_dir2_block_sort(
893 const void *a, /* first leaf entry */
894 const void *b) /* second leaf entry */
896 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
897 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
899 la = a;
900 lb = b;
901 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
902 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
906 * Convert a V2 leaf directory to a V2 block directory if possible.
908 int /* error */
909 xfs_dir2_leaf_to_block(
910 xfs_da_args_t *args, /* operation arguments */
911 struct xfs_buf *lbp, /* leaf buffer */
912 struct xfs_buf *dbp) /* data buffer */
914 __be16 *bestsp; /* leaf bests table */
915 xfs_dir2_data_hdr_t *hdr; /* block header */
916 xfs_dir2_block_tail_t *btp; /* block tail */
917 xfs_inode_t *dp; /* incore directory inode */
918 xfs_dir2_data_unused_t *dup; /* unused data entry */
919 int error; /* error return value */
920 int from; /* leaf from index */
921 xfs_dir2_leaf_t *leaf; /* leaf structure */
922 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
923 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
924 xfs_mount_t *mp; /* file system mount point */
925 int needlog; /* need to log data header */
926 int needscan; /* need to scan for bestfree */
927 xfs_dir2_sf_hdr_t sfh; /* shortform header */
928 int size; /* bytes used */
929 __be16 *tagp; /* end of entry (tag) */
930 int to; /* block/leaf to index */
931 xfs_trans_t *tp; /* transaction pointer */
932 struct xfs_dir2_leaf_entry *ents;
933 struct xfs_dir3_icleaf_hdr leafhdr;
935 trace_xfs_dir2_leaf_to_block(args);
937 dp = args->dp;
938 tp = args->trans;
939 mp = dp->i_mount;
940 leaf = lbp->b_addr;
941 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
942 ents = dp->d_ops->leaf_ents_p(leaf);
943 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
945 ASSERT(leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
946 leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
948 * If there are data blocks other than the first one, take this
949 * opportunity to remove trailing empty data blocks that may have
950 * been left behind during no-space-reservation operations.
951 * These will show up in the leaf bests table.
953 while (dp->i_d.di_size > args->geo->blksize) {
954 int hdrsz;
956 hdrsz = dp->d_ops->data_entry_offset;
957 bestsp = xfs_dir2_leaf_bests_p(ltp);
958 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
959 args->geo->blksize - hdrsz) {
960 if ((error =
961 xfs_dir2_leaf_trim_data(args, lbp,
962 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
963 return error;
964 } else
965 return 0;
968 * Read the data block if we don't already have it, give up if it fails.
970 if (!dbp) {
971 error = xfs_dir3_data_read(tp, dp, args->geo->datablk, -1, &dbp);
972 if (error)
973 return error;
975 hdr = dbp->b_addr;
976 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
977 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
980 * Size of the "leaf" area in the block.
982 size = (uint)sizeof(xfs_dir2_block_tail_t) +
983 (uint)sizeof(*lep) * (leafhdr.count - leafhdr.stale);
985 * Look at the last data entry.
987 tagp = (__be16 *)((char *)hdr + args->geo->blksize) - 1;
988 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
990 * If it's not free or is too short we can't do it.
992 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
993 be16_to_cpu(dup->length) < size)
994 return 0;
997 * Start converting it to block form.
999 xfs_dir3_block_init(mp, tp, dbp, dp);
1001 needlog = 1;
1002 needscan = 0;
1004 * Use up the space at the end of the block (blp/btp).
1006 xfs_dir2_data_use_free(args, dbp, dup, args->geo->blksize - size, size,
1007 &needlog, &needscan);
1009 * Initialize the block tail.
1011 btp = xfs_dir2_block_tail_p(args->geo, hdr);
1012 btp->count = cpu_to_be32(leafhdr.count - leafhdr.stale);
1013 btp->stale = 0;
1014 xfs_dir2_block_log_tail(tp, dbp);
1016 * Initialize the block leaf area. We compact out stale entries.
1018 lep = xfs_dir2_block_leaf_p(btp);
1019 for (from = to = 0; from < leafhdr.count; from++) {
1020 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
1021 continue;
1022 lep[to++] = ents[from];
1024 ASSERT(to == be32_to_cpu(btp->count));
1025 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
1027 * Scan the bestfree if we need it and log the data block header.
1029 if (needscan)
1030 xfs_dir2_data_freescan(dp, hdr, &needlog);
1031 if (needlog)
1032 xfs_dir2_data_log_header(args, dbp);
1034 * Pitch the old leaf block.
1036 error = xfs_da_shrink_inode(args, args->geo->leafblk, lbp);
1037 if (error)
1038 return error;
1041 * Now see if the resulting block can be shrunken to shortform.
1043 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1044 if (size > XFS_IFORK_DSIZE(dp))
1045 return 0;
1047 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1051 * Convert the shortform directory to block form.
1053 int /* error */
1054 xfs_dir2_sf_to_block(
1055 xfs_da_args_t *args) /* operation arguments */
1057 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1058 xfs_dir2_data_hdr_t *hdr; /* block header */
1059 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1060 struct xfs_buf *bp; /* block buffer */
1061 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1062 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1063 xfs_inode_t *dp; /* incore directory inode */
1064 int dummy; /* trash */
1065 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1066 int endoffset; /* end of data objects */
1067 int error; /* error return value */
1068 int i; /* index */
1069 xfs_mount_t *mp; /* filesystem mount point */
1070 int needlog; /* need to log block header */
1071 int needscan; /* need to scan block freespc */
1072 int newoffset; /* offset from current entry */
1073 int offset; /* target block offset */
1074 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1075 xfs_dir2_sf_hdr_t *oldsfp; /* old shortform header */
1076 xfs_dir2_sf_hdr_t *sfp; /* shortform header */
1077 __be16 *tagp; /* end of data entry */
1078 xfs_trans_t *tp; /* transaction pointer */
1079 struct xfs_name name;
1080 struct xfs_ifork *ifp;
1082 trace_xfs_dir2_sf_to_block(args);
1084 dp = args->dp;
1085 tp = args->trans;
1086 mp = dp->i_mount;
1087 ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
1088 ASSERT(ifp->if_flags & XFS_IFINLINE);
1090 * Bomb out if the shortform directory is way too short.
1092 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
1093 ASSERT(XFS_FORCED_SHUTDOWN(mp));
1094 return XFS_ERROR(EIO);
1097 oldsfp = (xfs_dir2_sf_hdr_t *)ifp->if_u1.if_data;
1099 ASSERT(ifp->if_bytes == dp->i_d.di_size);
1100 ASSERT(ifp->if_u1.if_data != NULL);
1101 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1102 ASSERT(dp->i_d.di_nextents == 0);
1105 * Copy the directory into a temporary buffer.
1106 * Then pitch the incore inode data so we can make extents.
1108 sfp = kmem_alloc(ifp->if_bytes, KM_SLEEP);
1109 memcpy(sfp, oldsfp, ifp->if_bytes);
1111 xfs_idata_realloc(dp, -ifp->if_bytes, XFS_DATA_FORK);
1112 xfs_bmap_local_to_extents_empty(dp, XFS_DATA_FORK);
1113 dp->i_d.di_size = 0;
1116 * Add block 0 to the inode.
1118 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1119 if (error) {
1120 kmem_free(sfp);
1121 return error;
1124 * Initialize the data block, then convert it to block format.
1126 error = xfs_dir3_data_init(args, blkno, &bp);
1127 if (error) {
1128 kmem_free(sfp);
1129 return error;
1131 xfs_dir3_block_init(mp, tp, bp, dp);
1132 hdr = bp->b_addr;
1135 * Compute size of block "tail" area.
1137 i = (uint)sizeof(*btp) +
1138 (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1140 * The whole thing is initialized to free by the init routine.
1141 * Say we're using the leaf and tail area.
1143 dup = dp->d_ops->data_unused_p(hdr);
1144 needlog = needscan = 0;
1145 xfs_dir2_data_use_free(args, bp, dup, args->geo->blksize - i,
1146 i, &needlog, &needscan);
1147 ASSERT(needscan == 0);
1149 * Fill in the tail.
1151 btp = xfs_dir2_block_tail_p(args->geo, hdr);
1152 btp->count = cpu_to_be32(sfp->count + 2); /* ., .. */
1153 btp->stale = 0;
1154 blp = xfs_dir2_block_leaf_p(btp);
1155 endoffset = (uint)((char *)blp - (char *)hdr);
1157 * Remove the freespace, we'll manage it.
1159 xfs_dir2_data_use_free(args, bp, dup,
1160 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1161 be16_to_cpu(dup->length), &needlog, &needscan);
1163 * Create entry for .
1165 dep = dp->d_ops->data_dot_entry_p(hdr);
1166 dep->inumber = cpu_to_be64(dp->i_ino);
1167 dep->namelen = 1;
1168 dep->name[0] = '.';
1169 dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1170 tagp = dp->d_ops->data_entry_tag_p(dep);
1171 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1172 xfs_dir2_data_log_entry(args, bp, dep);
1173 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1174 blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1175 (char *)dep - (char *)hdr));
1177 * Create entry for ..
1179 dep = dp->d_ops->data_dotdot_entry_p(hdr);
1180 dep->inumber = cpu_to_be64(dp->d_ops->sf_get_parent_ino(sfp));
1181 dep->namelen = 2;
1182 dep->name[0] = dep->name[1] = '.';
1183 dp->d_ops->data_put_ftype(dep, XFS_DIR3_FT_DIR);
1184 tagp = dp->d_ops->data_entry_tag_p(dep);
1185 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1186 xfs_dir2_data_log_entry(args, bp, dep);
1187 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1188 blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1189 (char *)dep - (char *)hdr));
1190 offset = dp->d_ops->data_first_offset;
1192 * Loop over existing entries, stuff them in.
1194 i = 0;
1195 if (!sfp->count)
1196 sfep = NULL;
1197 else
1198 sfep = xfs_dir2_sf_firstentry(sfp);
1200 * Need to preserve the existing offset values in the sf directory.
1201 * Insert holes (unused entries) where necessary.
1203 while (offset < endoffset) {
1205 * sfep is null when we reach the end of the list.
1207 if (sfep == NULL)
1208 newoffset = endoffset;
1209 else
1210 newoffset = xfs_dir2_sf_get_offset(sfep);
1212 * There should be a hole here, make one.
1214 if (offset < newoffset) {
1215 dup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1216 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1217 dup->length = cpu_to_be16(newoffset - offset);
1218 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(
1219 ((char *)dup - (char *)hdr));
1220 xfs_dir2_data_log_unused(args, bp, dup);
1221 xfs_dir2_data_freeinsert(hdr,
1222 dp->d_ops->data_bestfree_p(hdr),
1223 dup, &dummy);
1224 offset += be16_to_cpu(dup->length);
1225 continue;
1228 * Copy a real entry.
1230 dep = (xfs_dir2_data_entry_t *)((char *)hdr + newoffset);
1231 dep->inumber = cpu_to_be64(dp->d_ops->sf_get_ino(sfp, sfep));
1232 dep->namelen = sfep->namelen;
1233 dp->d_ops->data_put_ftype(dep, dp->d_ops->sf_get_ftype(sfep));
1234 memcpy(dep->name, sfep->name, dep->namelen);
1235 tagp = dp->d_ops->data_entry_tag_p(dep);
1236 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
1237 xfs_dir2_data_log_entry(args, bp, dep);
1238 name.name = sfep->name;
1239 name.len = sfep->namelen;
1240 blp[2 + i].hashval = cpu_to_be32(mp->m_dirnameops->
1241 hashname(&name));
1242 blp[2 + i].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
1243 (char *)dep - (char *)hdr));
1244 offset = (int)((char *)(tagp + 1) - (char *)hdr);
1245 if (++i == sfp->count)
1246 sfep = NULL;
1247 else
1248 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
1250 /* Done with the temporary buffer */
1251 kmem_free(sfp);
1253 * Sort the leaf entries by hash value.
1255 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1257 * Log the leaf entry area and tail.
1258 * Already logged the header in data_init, ignore needlog.
1260 ASSERT(needscan == 0);
1261 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1262 xfs_dir2_block_log_tail(tp, bp);
1263 xfs_dir3_data_check(dp, bp);
1264 return 0;