io_uring: don't use 'fd' for openat/openat2/statx
[linux/fpc-iii.git] / fs / xfs / libxfs / xfs_dir2_block.c
blobd6ced59b95676668eff71e2462727aa1fafb6290
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
5 * All Rights Reserved.
6 */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_trans.h"
16 #include "xfs_bmap.h"
17 #include "xfs_buf_item.h"
18 #include "xfs_dir2.h"
19 #include "xfs_dir2_priv.h"
20 #include "xfs_error.h"
21 #include "xfs_trace.h"
22 #include "xfs_log.h"
25 * Local function prototypes.
27 static void xfs_dir2_block_log_leaf(xfs_trans_t *tp, struct xfs_buf *bp,
28 int first, int last);
29 static void xfs_dir2_block_log_tail(xfs_trans_t *tp, struct xfs_buf *bp);
30 static int xfs_dir2_block_lookup_int(xfs_da_args_t *args, struct xfs_buf **bpp,
31 int *entno);
32 static int xfs_dir2_block_sort(const void *a, const void *b);
34 static xfs_dahash_t xfs_dir_hash_dot, xfs_dir_hash_dotdot;
37 * One-time startup routine called from xfs_init().
39 void
40 xfs_dir_startup(void)
42 xfs_dir_hash_dot = xfs_da_hashname((unsigned char *)".", 1);
43 xfs_dir_hash_dotdot = xfs_da_hashname((unsigned char *)"..", 2);
46 static xfs_failaddr_t
47 xfs_dir3_block_verify(
48 struct xfs_buf *bp)
50 struct xfs_mount *mp = bp->b_mount;
51 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
53 if (!xfs_verify_magic(bp, hdr3->magic))
54 return __this_address;
56 if (xfs_sb_version_hascrc(&mp->m_sb)) {
57 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
58 return __this_address;
59 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
60 return __this_address;
61 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
62 return __this_address;
64 return __xfs_dir3_data_check(NULL, bp);
67 static void
68 xfs_dir3_block_read_verify(
69 struct xfs_buf *bp)
71 struct xfs_mount *mp = bp->b_mount;
72 xfs_failaddr_t fa;
74 if (xfs_sb_version_hascrc(&mp->m_sb) &&
75 !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
76 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
77 else {
78 fa = xfs_dir3_block_verify(bp);
79 if (fa)
80 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
84 static void
85 xfs_dir3_block_write_verify(
86 struct xfs_buf *bp)
88 struct xfs_mount *mp = bp->b_mount;
89 struct xfs_buf_log_item *bip = bp->b_log_item;
90 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
91 xfs_failaddr_t fa;
93 fa = xfs_dir3_block_verify(bp);
94 if (fa) {
95 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
96 return;
99 if (!xfs_sb_version_hascrc(&mp->m_sb))
100 return;
102 if (bip)
103 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
105 xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
108 const struct xfs_buf_ops xfs_dir3_block_buf_ops = {
109 .name = "xfs_dir3_block",
110 .magic = { cpu_to_be32(XFS_DIR2_BLOCK_MAGIC),
111 cpu_to_be32(XFS_DIR3_BLOCK_MAGIC) },
112 .verify_read = xfs_dir3_block_read_verify,
113 .verify_write = xfs_dir3_block_write_verify,
114 .verify_struct = xfs_dir3_block_verify,
118 xfs_dir3_block_read(
119 struct xfs_trans *tp,
120 struct xfs_inode *dp,
121 struct xfs_buf **bpp)
123 struct xfs_mount *mp = dp->i_mount;
124 int err;
126 err = xfs_da_read_buf(tp, dp, mp->m_dir_geo->datablk, 0, bpp,
127 XFS_DATA_FORK, &xfs_dir3_block_buf_ops);
128 if (!err && tp && *bpp)
129 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_BLOCK_BUF);
130 return err;
133 static void
134 xfs_dir3_block_init(
135 struct xfs_mount *mp,
136 struct xfs_trans *tp,
137 struct xfs_buf *bp,
138 struct xfs_inode *dp)
140 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
142 bp->b_ops = &xfs_dir3_block_buf_ops;
143 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_BLOCK_BUF);
145 if (xfs_sb_version_hascrc(&mp->m_sb)) {
146 memset(hdr3, 0, sizeof(*hdr3));
147 hdr3->magic = cpu_to_be32(XFS_DIR3_BLOCK_MAGIC);
148 hdr3->blkno = cpu_to_be64(bp->b_bn);
149 hdr3->owner = cpu_to_be64(dp->i_ino);
150 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
151 return;
154 hdr3->magic = cpu_to_be32(XFS_DIR2_BLOCK_MAGIC);
157 static void
158 xfs_dir2_block_need_space(
159 struct xfs_inode *dp,
160 struct xfs_dir2_data_hdr *hdr,
161 struct xfs_dir2_block_tail *btp,
162 struct xfs_dir2_leaf_entry *blp,
163 __be16 **tagpp,
164 struct xfs_dir2_data_unused **dupp,
165 struct xfs_dir2_data_unused **enddupp,
166 int *compact,
167 int len)
169 struct xfs_dir2_data_free *bf;
170 __be16 *tagp = NULL;
171 struct xfs_dir2_data_unused *dup = NULL;
172 struct xfs_dir2_data_unused *enddup = NULL;
174 *compact = 0;
175 bf = xfs_dir2_data_bestfree_p(dp->i_mount, hdr);
178 * If there are stale entries we'll use one for the leaf.
180 if (btp->stale) {
181 if (be16_to_cpu(bf[0].length) >= len) {
183 * The biggest entry enough to avoid compaction.
185 dup = (xfs_dir2_data_unused_t *)
186 ((char *)hdr + be16_to_cpu(bf[0].offset));
187 goto out;
191 * Will need to compact to make this work.
192 * Tag just before the first leaf entry.
194 *compact = 1;
195 tagp = (__be16 *)blp - 1;
197 /* Data object just before the first leaf entry. */
198 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
201 * If it's not free then the data will go where the
202 * leaf data starts now, if it works at all.
204 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
205 if (be16_to_cpu(dup->length) + (be32_to_cpu(btp->stale) - 1) *
206 (uint)sizeof(*blp) < len)
207 dup = NULL;
208 } else if ((be32_to_cpu(btp->stale) - 1) * (uint)sizeof(*blp) < len)
209 dup = NULL;
210 else
211 dup = (xfs_dir2_data_unused_t *)blp;
212 goto out;
216 * no stale entries, so just use free space.
217 * Tag just before the first leaf entry.
219 tagp = (__be16 *)blp - 1;
221 /* Data object just before the first leaf entry. */
222 enddup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
225 * If it's not free then can't do this add without cleaning up:
226 * the space before the first leaf entry needs to be free so it
227 * can be expanded to hold the pointer to the new entry.
229 if (be16_to_cpu(enddup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
231 * Check out the biggest freespace and see if it's the same one.
233 dup = (xfs_dir2_data_unused_t *)
234 ((char *)hdr + be16_to_cpu(bf[0].offset));
235 if (dup != enddup) {
237 * Not the same free entry, just check its length.
239 if (be16_to_cpu(dup->length) < len)
240 dup = NULL;
241 goto out;
245 * It is the biggest freespace, can it hold the leaf too?
247 if (be16_to_cpu(dup->length) < len + (uint)sizeof(*blp)) {
249 * Yes, use the second-largest entry instead if it works.
251 if (be16_to_cpu(bf[1].length) >= len)
252 dup = (xfs_dir2_data_unused_t *)
253 ((char *)hdr + be16_to_cpu(bf[1].offset));
254 else
255 dup = NULL;
258 out:
259 *tagpp = tagp;
260 *dupp = dup;
261 *enddupp = enddup;
265 * compact the leaf entries.
266 * Leave the highest-numbered stale entry stale.
267 * XXX should be the one closest to mid but mid is not yet computed.
269 static void
270 xfs_dir2_block_compact(
271 struct xfs_da_args *args,
272 struct xfs_buf *bp,
273 struct xfs_dir2_data_hdr *hdr,
274 struct xfs_dir2_block_tail *btp,
275 struct xfs_dir2_leaf_entry *blp,
276 int *needlog,
277 int *lfloghigh,
278 int *lfloglow)
280 int fromidx; /* source leaf index */
281 int toidx; /* target leaf index */
282 int needscan = 0;
283 int highstale; /* high stale index */
285 fromidx = toidx = be32_to_cpu(btp->count) - 1;
286 highstale = *lfloghigh = -1;
287 for (; fromidx >= 0; fromidx--) {
288 if (blp[fromidx].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
289 if (highstale == -1)
290 highstale = toidx;
291 else {
292 if (*lfloghigh == -1)
293 *lfloghigh = toidx;
294 continue;
297 if (fromidx < toidx)
298 blp[toidx] = blp[fromidx];
299 toidx--;
301 *lfloglow = toidx + 1 - (be32_to_cpu(btp->stale) - 1);
302 *lfloghigh -= be32_to_cpu(btp->stale) - 1;
303 be32_add_cpu(&btp->count, -(be32_to_cpu(btp->stale) - 1));
304 xfs_dir2_data_make_free(args, bp,
305 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
306 (xfs_dir2_data_aoff_t)((be32_to_cpu(btp->stale) - 1) * sizeof(*blp)),
307 needlog, &needscan);
308 btp->stale = cpu_to_be32(1);
310 * If we now need to rebuild the bestfree map, do so.
311 * This needs to happen before the next call to use_free.
313 if (needscan)
314 xfs_dir2_data_freescan(args->dp->i_mount, hdr, needlog);
318 * Add an entry to a block directory.
320 int /* error */
321 xfs_dir2_block_addname(
322 xfs_da_args_t *args) /* directory op arguments */
324 xfs_dir2_data_hdr_t *hdr; /* block header */
325 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
326 struct xfs_buf *bp; /* buffer for block */
327 xfs_dir2_block_tail_t *btp; /* block tail */
328 int compact; /* need to compact leaf ents */
329 xfs_dir2_data_entry_t *dep; /* block data entry */
330 xfs_inode_t *dp; /* directory inode */
331 xfs_dir2_data_unused_t *dup; /* block unused entry */
332 int error; /* error return value */
333 xfs_dir2_data_unused_t *enddup=NULL; /* unused at end of data */
334 xfs_dahash_t hash; /* hash value of found entry */
335 int high; /* high index for binary srch */
336 int highstale; /* high stale index */
337 int lfloghigh=0; /* last final leaf to log */
338 int lfloglow=0; /* first final leaf to log */
339 int len; /* length of the new entry */
340 int low; /* low index for binary srch */
341 int lowstale; /* low stale index */
342 int mid=0; /* midpoint for binary srch */
343 int needlog; /* need to log header */
344 int needscan; /* need to rescan freespace */
345 __be16 *tagp; /* pointer to tag value */
346 xfs_trans_t *tp; /* transaction structure */
348 trace_xfs_dir2_block_addname(args);
350 dp = args->dp;
351 tp = args->trans;
353 /* Read the (one and only) directory block into bp. */
354 error = xfs_dir3_block_read(tp, dp, &bp);
355 if (error)
356 return error;
358 len = xfs_dir2_data_entsize(dp->i_mount, args->namelen);
361 * Set up pointers to parts of the block.
363 hdr = bp->b_addr;
364 btp = xfs_dir2_block_tail_p(args->geo, hdr);
365 blp = xfs_dir2_block_leaf_p(btp);
368 * Find out if we can reuse stale entries or whether we need extra
369 * space for entry and new leaf.
371 xfs_dir2_block_need_space(dp, hdr, btp, blp, &tagp, &dup,
372 &enddup, &compact, len);
375 * Done everything we need for a space check now.
377 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
378 xfs_trans_brelse(tp, bp);
379 if (!dup)
380 return -ENOSPC;
381 return 0;
385 * If we don't have space for the new entry & leaf ...
387 if (!dup) {
388 /* Don't have a space reservation: return no-space. */
389 if (args->total == 0)
390 return -ENOSPC;
392 * Convert to the next larger format.
393 * Then add the new entry in that format.
395 error = xfs_dir2_block_to_leaf(args, bp);
396 if (error)
397 return error;
398 return xfs_dir2_leaf_addname(args);
401 needlog = needscan = 0;
404 * If need to compact the leaf entries, do it now.
406 if (compact) {
407 xfs_dir2_block_compact(args, bp, hdr, btp, blp, &needlog,
408 &lfloghigh, &lfloglow);
409 /* recalculate blp post-compaction */
410 blp = xfs_dir2_block_leaf_p(btp);
411 } else if (btp->stale) {
413 * Set leaf logging boundaries to impossible state.
414 * For the no-stale case they're set explicitly.
416 lfloglow = be32_to_cpu(btp->count);
417 lfloghigh = -1;
421 * Find the slot that's first lower than our hash value, -1 if none.
423 for (low = 0, high = be32_to_cpu(btp->count) - 1; low <= high; ) {
424 mid = (low + high) >> 1;
425 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
426 break;
427 if (hash < args->hashval)
428 low = mid + 1;
429 else
430 high = mid - 1;
432 while (mid >= 0 && be32_to_cpu(blp[mid].hashval) >= args->hashval) {
433 mid--;
436 * No stale entries, will use enddup space to hold new leaf.
438 if (!btp->stale) {
439 xfs_dir2_data_aoff_t aoff;
442 * Mark the space needed for the new leaf entry, now in use.
444 aoff = (xfs_dir2_data_aoff_t)((char *)enddup - (char *)hdr +
445 be16_to_cpu(enddup->length) - sizeof(*blp));
446 error = xfs_dir2_data_use_free(args, bp, enddup, aoff,
447 (xfs_dir2_data_aoff_t)sizeof(*blp), &needlog,
448 &needscan);
449 if (error)
450 return error;
453 * Update the tail (entry count).
455 be32_add_cpu(&btp->count, 1);
457 * If we now need to rebuild the bestfree map, do so.
458 * This needs to happen before the next call to use_free.
460 if (needscan) {
461 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
462 needscan = 0;
465 * Adjust pointer to the first leaf entry, we're about to move
466 * the table up one to open up space for the new leaf entry.
467 * Then adjust our index to match.
469 blp--;
470 mid++;
471 if (mid)
472 memmove(blp, &blp[1], mid * sizeof(*blp));
473 lfloglow = 0;
474 lfloghigh = mid;
477 * Use a stale leaf for our new entry.
479 else {
480 for (lowstale = mid;
481 lowstale >= 0 &&
482 blp[lowstale].address !=
483 cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
484 lowstale--)
485 continue;
486 for (highstale = mid + 1;
487 highstale < be32_to_cpu(btp->count) &&
488 blp[highstale].address !=
489 cpu_to_be32(XFS_DIR2_NULL_DATAPTR) &&
490 (lowstale < 0 || mid - lowstale > highstale - mid);
491 highstale++)
492 continue;
494 * Move entries toward the low-numbered stale entry.
496 if (lowstale >= 0 &&
497 (highstale == be32_to_cpu(btp->count) ||
498 mid - lowstale <= highstale - mid)) {
499 if (mid - lowstale)
500 memmove(&blp[lowstale], &blp[lowstale + 1],
501 (mid - lowstale) * sizeof(*blp));
502 lfloglow = min(lowstale, lfloglow);
503 lfloghigh = max(mid, lfloghigh);
506 * Move entries toward the high-numbered stale entry.
508 else {
509 ASSERT(highstale < be32_to_cpu(btp->count));
510 mid++;
511 if (highstale - mid)
512 memmove(&blp[mid + 1], &blp[mid],
513 (highstale - mid) * sizeof(*blp));
514 lfloglow = min(mid, lfloglow);
515 lfloghigh = max(highstale, lfloghigh);
517 be32_add_cpu(&btp->stale, -1);
520 * Point to the new data entry.
522 dep = (xfs_dir2_data_entry_t *)dup;
524 * Fill in the leaf entry.
526 blp[mid].hashval = cpu_to_be32(args->hashval);
527 blp[mid].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(
528 (char *)dep - (char *)hdr));
529 xfs_dir2_block_log_leaf(tp, bp, lfloglow, lfloghigh);
531 * Mark space for the data entry used.
533 error = xfs_dir2_data_use_free(args, bp, dup,
534 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
535 (xfs_dir2_data_aoff_t)len, &needlog, &needscan);
536 if (error)
537 return error;
539 * Create the new data entry.
541 dep->inumber = cpu_to_be64(args->inumber);
542 dep->namelen = args->namelen;
543 memcpy(dep->name, args->name, args->namelen);
544 xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
545 tagp = xfs_dir2_data_entry_tag_p(dp->i_mount, dep);
546 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
548 * Clean up the bestfree array and log the header, tail, and entry.
550 if (needscan)
551 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
552 if (needlog)
553 xfs_dir2_data_log_header(args, bp);
554 xfs_dir2_block_log_tail(tp, bp);
555 xfs_dir2_data_log_entry(args, bp, dep);
556 xfs_dir3_data_check(dp, bp);
557 return 0;
561 * Log leaf entries from the block.
563 static void
564 xfs_dir2_block_log_leaf(
565 xfs_trans_t *tp, /* transaction structure */
566 struct xfs_buf *bp, /* block buffer */
567 int first, /* index of first logged leaf */
568 int last) /* index of last logged leaf */
570 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
571 xfs_dir2_leaf_entry_t *blp;
572 xfs_dir2_block_tail_t *btp;
574 btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
575 blp = xfs_dir2_block_leaf_p(btp);
576 xfs_trans_log_buf(tp, bp, (uint)((char *)&blp[first] - (char *)hdr),
577 (uint)((char *)&blp[last + 1] - (char *)hdr - 1));
581 * Log the block tail.
583 static void
584 xfs_dir2_block_log_tail(
585 xfs_trans_t *tp, /* transaction structure */
586 struct xfs_buf *bp) /* block buffer */
588 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
589 xfs_dir2_block_tail_t *btp;
591 btp = xfs_dir2_block_tail_p(tp->t_mountp->m_dir_geo, hdr);
592 xfs_trans_log_buf(tp, bp, (uint)((char *)btp - (char *)hdr),
593 (uint)((char *)(btp + 1) - (char *)hdr - 1));
597 * Look up an entry in the block. This is the external routine,
598 * xfs_dir2_block_lookup_int does the real work.
600 int /* error */
601 xfs_dir2_block_lookup(
602 xfs_da_args_t *args) /* dir lookup arguments */
604 xfs_dir2_data_hdr_t *hdr; /* block header */
605 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
606 struct xfs_buf *bp; /* block buffer */
607 xfs_dir2_block_tail_t *btp; /* block tail */
608 xfs_dir2_data_entry_t *dep; /* block data entry */
609 xfs_inode_t *dp; /* incore inode */
610 int ent; /* entry index */
611 int error; /* error return value */
613 trace_xfs_dir2_block_lookup(args);
616 * Get the buffer, look up the entry.
617 * If not found (ENOENT) then return, have no buffer.
619 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent)))
620 return error;
621 dp = args->dp;
622 hdr = bp->b_addr;
623 xfs_dir3_data_check(dp, bp);
624 btp = xfs_dir2_block_tail_p(args->geo, hdr);
625 blp = xfs_dir2_block_leaf_p(btp);
627 * Get the offset from the leaf entry, to point to the data.
629 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
630 xfs_dir2_dataptr_to_off(args->geo,
631 be32_to_cpu(blp[ent].address)));
633 * Fill in inode number, CI name if appropriate, release the block.
635 args->inumber = be64_to_cpu(dep->inumber);
636 args->filetype = xfs_dir2_data_get_ftype(dp->i_mount, dep);
637 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
638 xfs_trans_brelse(args->trans, bp);
639 return error;
643 * Internal block lookup routine.
645 static int /* error */
646 xfs_dir2_block_lookup_int(
647 xfs_da_args_t *args, /* dir lookup arguments */
648 struct xfs_buf **bpp, /* returned block buffer */
649 int *entno) /* returned entry number */
651 xfs_dir2_dataptr_t addr; /* data entry address */
652 xfs_dir2_data_hdr_t *hdr; /* block header */
653 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
654 struct xfs_buf *bp; /* block buffer */
655 xfs_dir2_block_tail_t *btp; /* block tail */
656 xfs_dir2_data_entry_t *dep; /* block data entry */
657 xfs_inode_t *dp; /* incore inode */
658 int error; /* error return value */
659 xfs_dahash_t hash; /* found hash value */
660 int high; /* binary search high index */
661 int low; /* binary search low index */
662 int mid; /* binary search current idx */
663 xfs_trans_t *tp; /* transaction pointer */
664 enum xfs_dacmp cmp; /* comparison result */
666 dp = args->dp;
667 tp = args->trans;
669 error = xfs_dir3_block_read(tp, dp, &bp);
670 if (error)
671 return error;
673 hdr = bp->b_addr;
674 xfs_dir3_data_check(dp, bp);
675 btp = xfs_dir2_block_tail_p(args->geo, hdr);
676 blp = xfs_dir2_block_leaf_p(btp);
678 * Loop doing a binary search for our hash value.
679 * Find our entry, ENOENT if it's not there.
681 for (low = 0, high = be32_to_cpu(btp->count) - 1; ; ) {
682 ASSERT(low <= high);
683 mid = (low + high) >> 1;
684 if ((hash = be32_to_cpu(blp[mid].hashval)) == args->hashval)
685 break;
686 if (hash < args->hashval)
687 low = mid + 1;
688 else
689 high = mid - 1;
690 if (low > high) {
691 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
692 xfs_trans_brelse(tp, bp);
693 return -ENOENT;
697 * Back up to the first one with the right hash value.
699 while (mid > 0 && be32_to_cpu(blp[mid - 1].hashval) == args->hashval) {
700 mid--;
703 * Now loop forward through all the entries with the
704 * right hash value looking for our name.
706 do {
707 if ((addr = be32_to_cpu(blp[mid].address)) == XFS_DIR2_NULL_DATAPTR)
708 continue;
710 * Get pointer to the entry from the leaf.
712 dep = (xfs_dir2_data_entry_t *)
713 ((char *)hdr + xfs_dir2_dataptr_to_off(args->geo, addr));
715 * Compare name and if it's an exact match, return the index
716 * and buffer. If it's the first case-insensitive match, store
717 * the index and buffer and continue looking for an exact match.
719 cmp = xfs_dir2_compname(args, dep->name, dep->namelen);
720 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
721 args->cmpresult = cmp;
722 *bpp = bp;
723 *entno = mid;
724 if (cmp == XFS_CMP_EXACT)
725 return 0;
727 } while (++mid < be32_to_cpu(btp->count) &&
728 be32_to_cpu(blp[mid].hashval) == hash);
730 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
732 * Here, we can only be doing a lookup (not a rename or replace).
733 * If a case-insensitive match was found earlier, return success.
735 if (args->cmpresult == XFS_CMP_CASE)
736 return 0;
738 * No match, release the buffer and return ENOENT.
740 xfs_trans_brelse(tp, bp);
741 return -ENOENT;
745 * Remove an entry from a block format directory.
746 * If that makes the block small enough to fit in shortform, transform it.
748 int /* error */
749 xfs_dir2_block_removename(
750 xfs_da_args_t *args) /* directory operation args */
752 xfs_dir2_data_hdr_t *hdr; /* block header */
753 xfs_dir2_leaf_entry_t *blp; /* block leaf pointer */
754 struct xfs_buf *bp; /* block buffer */
755 xfs_dir2_block_tail_t *btp; /* block tail */
756 xfs_dir2_data_entry_t *dep; /* block data entry */
757 xfs_inode_t *dp; /* incore inode */
758 int ent; /* block leaf entry index */
759 int error; /* error return value */
760 int needlog; /* need to log block header */
761 int needscan; /* need to fixup bestfree */
762 xfs_dir2_sf_hdr_t sfh; /* shortform header */
763 int size; /* shortform size */
764 xfs_trans_t *tp; /* transaction pointer */
766 trace_xfs_dir2_block_removename(args);
769 * Look up the entry in the block. Gets the buffer and entry index.
770 * It will always be there, the vnodeops level does a lookup first.
772 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
773 return error;
775 dp = args->dp;
776 tp = args->trans;
777 hdr = bp->b_addr;
778 btp = xfs_dir2_block_tail_p(args->geo, hdr);
779 blp = xfs_dir2_block_leaf_p(btp);
781 * Point to the data entry using the leaf entry.
783 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
784 xfs_dir2_dataptr_to_off(args->geo,
785 be32_to_cpu(blp[ent].address)));
787 * Mark the data entry's space free.
789 needlog = needscan = 0;
790 xfs_dir2_data_make_free(args, bp,
791 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
792 xfs_dir2_data_entsize(dp->i_mount, dep->namelen), &needlog,
793 &needscan);
795 * Fix up the block tail.
797 be32_add_cpu(&btp->stale, 1);
798 xfs_dir2_block_log_tail(tp, bp);
800 * Remove the leaf entry by marking it stale.
802 blp[ent].address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
803 xfs_dir2_block_log_leaf(tp, bp, ent, ent);
805 * Fix up bestfree, log the header if necessary.
807 if (needscan)
808 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
809 if (needlog)
810 xfs_dir2_data_log_header(args, bp);
811 xfs_dir3_data_check(dp, bp);
813 * See if the size as a shortform is good enough.
815 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
816 if (size > XFS_IFORK_DSIZE(dp))
817 return 0;
820 * If it works, do the conversion.
822 return xfs_dir2_block_to_sf(args, bp, size, &sfh);
826 * Replace an entry in a V2 block directory.
827 * Change the inode number to the new value.
829 int /* error */
830 xfs_dir2_block_replace(
831 xfs_da_args_t *args) /* directory operation args */
833 xfs_dir2_data_hdr_t *hdr; /* block header */
834 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
835 struct xfs_buf *bp; /* block buffer */
836 xfs_dir2_block_tail_t *btp; /* block tail */
837 xfs_dir2_data_entry_t *dep; /* block data entry */
838 xfs_inode_t *dp; /* incore inode */
839 int ent; /* leaf entry index */
840 int error; /* error return value */
842 trace_xfs_dir2_block_replace(args);
845 * Lookup the entry in the directory. Get buffer and entry index.
846 * This will always succeed since the caller has already done a lookup.
848 if ((error = xfs_dir2_block_lookup_int(args, &bp, &ent))) {
849 return error;
851 dp = args->dp;
852 hdr = bp->b_addr;
853 btp = xfs_dir2_block_tail_p(args->geo, hdr);
854 blp = xfs_dir2_block_leaf_p(btp);
856 * Point to the data entry we need to change.
858 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
859 xfs_dir2_dataptr_to_off(args->geo,
860 be32_to_cpu(blp[ent].address)));
861 ASSERT(be64_to_cpu(dep->inumber) != args->inumber);
863 * Change the inode number to the new value.
865 dep->inumber = cpu_to_be64(args->inumber);
866 xfs_dir2_data_put_ftype(dp->i_mount, dep, args->filetype);
867 xfs_dir2_data_log_entry(args, bp, dep);
868 xfs_dir3_data_check(dp, bp);
869 return 0;
873 * Qsort comparison routine for the block leaf entries.
875 static int /* sort order */
876 xfs_dir2_block_sort(
877 const void *a, /* first leaf entry */
878 const void *b) /* second leaf entry */
880 const xfs_dir2_leaf_entry_t *la; /* first leaf entry */
881 const xfs_dir2_leaf_entry_t *lb; /* second leaf entry */
883 la = a;
884 lb = b;
885 return be32_to_cpu(la->hashval) < be32_to_cpu(lb->hashval) ? -1 :
886 (be32_to_cpu(la->hashval) > be32_to_cpu(lb->hashval) ? 1 : 0);
890 * Convert a V2 leaf directory to a V2 block directory if possible.
892 int /* error */
893 xfs_dir2_leaf_to_block(
894 xfs_da_args_t *args, /* operation arguments */
895 struct xfs_buf *lbp, /* leaf buffer */
896 struct xfs_buf *dbp) /* data buffer */
898 __be16 *bestsp; /* leaf bests table */
899 xfs_dir2_data_hdr_t *hdr; /* block header */
900 xfs_dir2_block_tail_t *btp; /* block tail */
901 xfs_inode_t *dp; /* incore directory inode */
902 xfs_dir2_data_unused_t *dup; /* unused data entry */
903 int error; /* error return value */
904 int from; /* leaf from index */
905 xfs_dir2_leaf_t *leaf; /* leaf structure */
906 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
907 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
908 xfs_mount_t *mp; /* file system mount point */
909 int needlog; /* need to log data header */
910 int needscan; /* need to scan for bestfree */
911 xfs_dir2_sf_hdr_t sfh; /* shortform header */
912 int size; /* bytes used */
913 __be16 *tagp; /* end of entry (tag) */
914 int to; /* block/leaf to index */
915 xfs_trans_t *tp; /* transaction pointer */
916 struct xfs_dir3_icleaf_hdr leafhdr;
918 trace_xfs_dir2_leaf_to_block(args);
920 dp = args->dp;
921 tp = args->trans;
922 mp = dp->i_mount;
923 leaf = lbp->b_addr;
924 xfs_dir2_leaf_hdr_from_disk(mp, &leafhdr, leaf);
925 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
927 ASSERT(leafhdr.magic == XFS_DIR2_LEAF1_MAGIC ||
928 leafhdr.magic == XFS_DIR3_LEAF1_MAGIC);
930 * If there are data blocks other than the first one, take this
931 * opportunity to remove trailing empty data blocks that may have
932 * been left behind during no-space-reservation operations.
933 * These will show up in the leaf bests table.
935 while (dp->i_d.di_size > args->geo->blksize) {
936 int hdrsz;
938 hdrsz = args->geo->data_entry_offset;
939 bestsp = xfs_dir2_leaf_bests_p(ltp);
940 if (be16_to_cpu(bestsp[be32_to_cpu(ltp->bestcount) - 1]) ==
941 args->geo->blksize - hdrsz) {
942 if ((error =
943 xfs_dir2_leaf_trim_data(args, lbp,
944 (xfs_dir2_db_t)(be32_to_cpu(ltp->bestcount) - 1))))
945 return error;
946 } else
947 return 0;
950 * Read the data block if we don't already have it, give up if it fails.
952 if (!dbp) {
953 error = xfs_dir3_data_read(tp, dp, args->geo->datablk, 0, &dbp);
954 if (error)
955 return error;
957 hdr = dbp->b_addr;
958 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
959 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
962 * Size of the "leaf" area in the block.
964 size = (uint)sizeof(xfs_dir2_block_tail_t) +
965 (uint)sizeof(*lep) * (leafhdr.count - leafhdr.stale);
967 * Look at the last data entry.
969 tagp = (__be16 *)((char *)hdr + args->geo->blksize) - 1;
970 dup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
972 * If it's not free or is too short we can't do it.
974 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG ||
975 be16_to_cpu(dup->length) < size)
976 return 0;
979 * Start converting it to block form.
981 xfs_dir3_block_init(mp, tp, dbp, dp);
983 needlog = 1;
984 needscan = 0;
986 * Use up the space at the end of the block (blp/btp).
988 error = xfs_dir2_data_use_free(args, dbp, dup,
989 args->geo->blksize - size, size, &needlog, &needscan);
990 if (error)
991 return error;
993 * Initialize the block tail.
995 btp = xfs_dir2_block_tail_p(args->geo, hdr);
996 btp->count = cpu_to_be32(leafhdr.count - leafhdr.stale);
997 btp->stale = 0;
998 xfs_dir2_block_log_tail(tp, dbp);
1000 * Initialize the block leaf area. We compact out stale entries.
1002 lep = xfs_dir2_block_leaf_p(btp);
1003 for (from = to = 0; from < leafhdr.count; from++) {
1004 if (leafhdr.ents[from].address ==
1005 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
1006 continue;
1007 lep[to++] = leafhdr.ents[from];
1009 ASSERT(to == be32_to_cpu(btp->count));
1010 xfs_dir2_block_log_leaf(tp, dbp, 0, be32_to_cpu(btp->count) - 1);
1012 * Scan the bestfree if we need it and log the data block header.
1014 if (needscan)
1015 xfs_dir2_data_freescan(dp->i_mount, hdr, &needlog);
1016 if (needlog)
1017 xfs_dir2_data_log_header(args, dbp);
1019 * Pitch the old leaf block.
1021 error = xfs_da_shrink_inode(args, args->geo->leafblk, lbp);
1022 if (error)
1023 return error;
1026 * Now see if the resulting block can be shrunken to shortform.
1028 size = xfs_dir2_block_sfsize(dp, hdr, &sfh);
1029 if (size > XFS_IFORK_DSIZE(dp))
1030 return 0;
1032 return xfs_dir2_block_to_sf(args, dbp, size, &sfh);
1036 * Convert the shortform directory to block form.
1038 int /* error */
1039 xfs_dir2_sf_to_block(
1040 struct xfs_da_args *args)
1042 struct xfs_trans *tp = args->trans;
1043 struct xfs_inode *dp = args->dp;
1044 struct xfs_mount *mp = dp->i_mount;
1045 struct xfs_ifork *ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
1046 struct xfs_da_geometry *geo = args->geo;
1047 xfs_dir2_db_t blkno; /* dir-relative block # (0) */
1048 xfs_dir2_data_hdr_t *hdr; /* block header */
1049 xfs_dir2_leaf_entry_t *blp; /* block leaf entries */
1050 struct xfs_buf *bp; /* block buffer */
1051 xfs_dir2_block_tail_t *btp; /* block tail pointer */
1052 xfs_dir2_data_entry_t *dep; /* data entry pointer */
1053 int dummy; /* trash */
1054 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
1055 int endoffset; /* end of data objects */
1056 int error; /* error return value */
1057 int i; /* index */
1058 int needlog; /* need to log block header */
1059 int needscan; /* need to scan block freespc */
1060 int newoffset; /* offset from current entry */
1061 unsigned int offset = geo->data_entry_offset;
1062 xfs_dir2_sf_entry_t *sfep; /* sf entry pointer */
1063 xfs_dir2_sf_hdr_t *oldsfp; /* old shortform header */
1064 xfs_dir2_sf_hdr_t *sfp; /* shortform header */
1065 __be16 *tagp; /* end of data entry */
1066 struct xfs_name name;
1068 trace_xfs_dir2_sf_to_block(args);
1070 ASSERT(ifp->if_flags & XFS_IFINLINE);
1071 ASSERT(dp->i_d.di_size >= offsetof(struct xfs_dir2_sf_hdr, parent));
1073 oldsfp = (xfs_dir2_sf_hdr_t *)ifp->if_u1.if_data;
1075 ASSERT(ifp->if_bytes == dp->i_d.di_size);
1076 ASSERT(ifp->if_u1.if_data != NULL);
1077 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(oldsfp->i8count));
1078 ASSERT(dp->i_d.di_nextents == 0);
1081 * Copy the directory into a temporary buffer.
1082 * Then pitch the incore inode data so we can make extents.
1084 sfp = kmem_alloc(ifp->if_bytes, 0);
1085 memcpy(sfp, oldsfp, ifp->if_bytes);
1087 xfs_idata_realloc(dp, -ifp->if_bytes, XFS_DATA_FORK);
1088 xfs_bmap_local_to_extents_empty(tp, dp, XFS_DATA_FORK);
1089 dp->i_d.di_size = 0;
1092 * Add block 0 to the inode.
1094 error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE, &blkno);
1095 if (error)
1096 goto out_free;
1098 * Initialize the data block, then convert it to block format.
1100 error = xfs_dir3_data_init(args, blkno, &bp);
1101 if (error)
1102 goto out_free;
1103 xfs_dir3_block_init(mp, tp, bp, dp);
1104 hdr = bp->b_addr;
1107 * Compute size of block "tail" area.
1109 i = (uint)sizeof(*btp) +
1110 (sfp->count + 2) * (uint)sizeof(xfs_dir2_leaf_entry_t);
1112 * The whole thing is initialized to free by the init routine.
1113 * Say we're using the leaf and tail area.
1115 dup = bp->b_addr + offset;
1116 needlog = needscan = 0;
1117 error = xfs_dir2_data_use_free(args, bp, dup, args->geo->blksize - i,
1118 i, &needlog, &needscan);
1119 if (error)
1120 goto out_free;
1121 ASSERT(needscan == 0);
1123 * Fill in the tail.
1125 btp = xfs_dir2_block_tail_p(args->geo, hdr);
1126 btp->count = cpu_to_be32(sfp->count + 2); /* ., .. */
1127 btp->stale = 0;
1128 blp = xfs_dir2_block_leaf_p(btp);
1129 endoffset = (uint)((char *)blp - (char *)hdr);
1131 * Remove the freespace, we'll manage it.
1133 error = xfs_dir2_data_use_free(args, bp, dup,
1134 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
1135 be16_to_cpu(dup->length), &needlog, &needscan);
1136 if (error)
1137 goto out_free;
1140 * Create entry for .
1142 dep = bp->b_addr + offset;
1143 dep->inumber = cpu_to_be64(dp->i_ino);
1144 dep->namelen = 1;
1145 dep->name[0] = '.';
1146 xfs_dir2_data_put_ftype(mp, dep, XFS_DIR3_FT_DIR);
1147 tagp = xfs_dir2_data_entry_tag_p(mp, dep);
1148 *tagp = cpu_to_be16(offset);
1149 xfs_dir2_data_log_entry(args, bp, dep);
1150 blp[0].hashval = cpu_to_be32(xfs_dir_hash_dot);
1151 blp[0].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(offset));
1152 offset += xfs_dir2_data_entsize(mp, dep->namelen);
1155 * Create entry for ..
1157 dep = bp->b_addr + offset;
1158 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_parent_ino(sfp));
1159 dep->namelen = 2;
1160 dep->name[0] = dep->name[1] = '.';
1161 xfs_dir2_data_put_ftype(mp, dep, XFS_DIR3_FT_DIR);
1162 tagp = xfs_dir2_data_entry_tag_p(mp, dep);
1163 *tagp = cpu_to_be16(offset);
1164 xfs_dir2_data_log_entry(args, bp, dep);
1165 blp[1].hashval = cpu_to_be32(xfs_dir_hash_dotdot);
1166 blp[1].address = cpu_to_be32(xfs_dir2_byte_to_dataptr(offset));
1167 offset += xfs_dir2_data_entsize(mp, dep->namelen);
1170 * Loop over existing entries, stuff them in.
1172 i = 0;
1173 if (!sfp->count)
1174 sfep = NULL;
1175 else
1176 sfep = xfs_dir2_sf_firstentry(sfp);
1179 * Need to preserve the existing offset values in the sf directory.
1180 * Insert holes (unused entries) where necessary.
1182 while (offset < endoffset) {
1184 * sfep is null when we reach the end of the list.
1186 if (sfep == NULL)
1187 newoffset = endoffset;
1188 else
1189 newoffset = xfs_dir2_sf_get_offset(sfep);
1191 * There should be a hole here, make one.
1193 if (offset < newoffset) {
1194 dup = bp->b_addr + offset;
1195 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1196 dup->length = cpu_to_be16(newoffset - offset);
1197 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16(offset);
1198 xfs_dir2_data_log_unused(args, bp, dup);
1199 xfs_dir2_data_freeinsert(hdr,
1200 xfs_dir2_data_bestfree_p(mp, hdr),
1201 dup, &dummy);
1202 offset += be16_to_cpu(dup->length);
1203 continue;
1206 * Copy a real entry.
1208 dep = bp->b_addr + newoffset;
1209 dep->inumber = cpu_to_be64(xfs_dir2_sf_get_ino(mp, sfp, sfep));
1210 dep->namelen = sfep->namelen;
1211 xfs_dir2_data_put_ftype(mp, dep,
1212 xfs_dir2_sf_get_ftype(mp, sfep));
1213 memcpy(dep->name, sfep->name, dep->namelen);
1214 tagp = xfs_dir2_data_entry_tag_p(mp, dep);
1215 *tagp = cpu_to_be16(newoffset);
1216 xfs_dir2_data_log_entry(args, bp, dep);
1217 name.name = sfep->name;
1218 name.len = sfep->namelen;
1219 blp[2 + i].hashval = cpu_to_be32(xfs_dir2_hashname(mp, &name));
1220 blp[2 + i].address =
1221 cpu_to_be32(xfs_dir2_byte_to_dataptr(newoffset));
1222 offset = (int)((char *)(tagp + 1) - (char *)hdr);
1223 if (++i == sfp->count)
1224 sfep = NULL;
1225 else
1226 sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
1228 /* Done with the temporary buffer */
1229 kmem_free(sfp);
1231 * Sort the leaf entries by hash value.
1233 xfs_sort(blp, be32_to_cpu(btp->count), sizeof(*blp), xfs_dir2_block_sort);
1235 * Log the leaf entry area and tail.
1236 * Already logged the header in data_init, ignore needlog.
1238 ASSERT(needscan == 0);
1239 xfs_dir2_block_log_leaf(tp, bp, 0, be32_to_cpu(btp->count) - 1);
1240 xfs_dir2_block_log_tail(tp, bp);
1241 xfs_dir3_data_check(dp, bp);
1242 return 0;
1243 out_free:
1244 kmem_free(sfp);
1245 return error;