perf intel-pt: Factor out intel_pt_8b_tsc()
[linux/fpc-iii.git] / fs / xfs / libxfs / xfs_dir2_leaf.c
blob9c2a0a13ed61289b43cb2fbb6b86a29452c0323b
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_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_da_format.h"
14 #include "xfs_da_btree.h"
15 #include "xfs_inode.h"
16 #include "xfs_bmap.h"
17 #include "xfs_dir2.h"
18 #include "xfs_dir2_priv.h"
19 #include "xfs_error.h"
20 #include "xfs_trace.h"
21 #include "xfs_trans.h"
22 #include "xfs_buf_item.h"
23 #include "xfs_cksum.h"
24 #include "xfs_log.h"
27 * Local function declarations.
29 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
30 int *indexp, struct xfs_buf **dbpp);
31 static void xfs_dir3_leaf_log_bests(struct xfs_da_args *args,
32 struct xfs_buf *bp, int first, int last);
33 static void xfs_dir3_leaf_log_tail(struct xfs_da_args *args,
34 struct xfs_buf *bp);
37 * Check the internal consistency of a leaf1 block.
38 * Pop an assert if something is wrong.
40 #ifdef DEBUG
41 static xfs_failaddr_t
42 xfs_dir3_leaf1_check(
43 struct xfs_inode *dp,
44 struct xfs_buf *bp)
46 struct xfs_dir2_leaf *leaf = bp->b_addr;
47 struct xfs_dir3_icleaf_hdr leafhdr;
49 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
51 if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
52 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
53 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
54 return __this_address;
55 } else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
56 return __this_address;
58 return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
61 static inline void
62 xfs_dir3_leaf_check(
63 struct xfs_inode *dp,
64 struct xfs_buf *bp)
66 xfs_failaddr_t fa;
68 fa = xfs_dir3_leaf1_check(dp, bp);
69 if (!fa)
70 return;
71 xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
72 bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
73 fa);
74 ASSERT(0);
76 #else
77 #define xfs_dir3_leaf_check(dp, bp)
78 #endif
80 xfs_failaddr_t
81 xfs_dir3_leaf_check_int(
82 struct xfs_mount *mp,
83 struct xfs_inode *dp,
84 struct xfs_dir3_icleaf_hdr *hdr,
85 struct xfs_dir2_leaf *leaf)
87 struct xfs_dir2_leaf_entry *ents;
88 xfs_dir2_leaf_tail_t *ltp;
89 int stale;
90 int i;
91 const struct xfs_dir_ops *ops;
92 struct xfs_dir3_icleaf_hdr leafhdr;
93 struct xfs_da_geometry *geo = mp->m_dir_geo;
96 * we can be passed a null dp here from a verifier, so we need to go the
97 * hard way to get them.
99 ops = xfs_dir_get_ops(mp, dp);
101 if (!hdr) {
102 ops->leaf_hdr_from_disk(&leafhdr, leaf);
103 hdr = &leafhdr;
106 ents = ops->leaf_ents_p(leaf);
107 ltp = xfs_dir2_leaf_tail_p(geo, leaf);
110 * XXX (dgc): This value is not restrictive enough.
111 * Should factor in the size of the bests table as well.
112 * We can deduce a value for that from di_size.
114 if (hdr->count > ops->leaf_max_ents(geo))
115 return __this_address;
117 /* Leaves and bests don't overlap in leaf format. */
118 if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
119 hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
120 (char *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
121 return __this_address;
123 /* Check hash value order, count stale entries. */
124 for (i = stale = 0; i < hdr->count; i++) {
125 if (i + 1 < hdr->count) {
126 if (be32_to_cpu(ents[i].hashval) >
127 be32_to_cpu(ents[i + 1].hashval))
128 return __this_address;
130 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
131 stale++;
133 if (hdr->stale != stale)
134 return __this_address;
135 return NULL;
139 * We verify the magic numbers before decoding the leaf header so that on debug
140 * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
141 * to incorrect magic numbers.
143 static xfs_failaddr_t
144 xfs_dir3_leaf_verify(
145 struct xfs_buf *bp)
147 struct xfs_mount *mp = bp->b_target->bt_mount;
148 struct xfs_dir2_leaf *leaf = bp->b_addr;
149 xfs_failaddr_t fa;
151 fa = xfs_da3_blkinfo_verify(bp, bp->b_addr);
152 if (fa)
153 return fa;
155 return xfs_dir3_leaf_check_int(mp, NULL, NULL, leaf);
158 static void
159 xfs_dir3_leaf_read_verify(
160 struct xfs_buf *bp)
162 struct xfs_mount *mp = bp->b_target->bt_mount;
163 xfs_failaddr_t fa;
165 if (xfs_sb_version_hascrc(&mp->m_sb) &&
166 !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
167 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
168 else {
169 fa = xfs_dir3_leaf_verify(bp);
170 if (fa)
171 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
175 static void
176 xfs_dir3_leaf_write_verify(
177 struct xfs_buf *bp)
179 struct xfs_mount *mp = bp->b_target->bt_mount;
180 struct xfs_buf_log_item *bip = bp->b_log_item;
181 struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
182 xfs_failaddr_t fa;
184 fa = xfs_dir3_leaf_verify(bp);
185 if (fa) {
186 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
187 return;
190 if (!xfs_sb_version_hascrc(&mp->m_sb))
191 return;
193 if (bip)
194 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
196 xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
199 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
200 .name = "xfs_dir3_leaf1",
201 .magic16 = { cpu_to_be16(XFS_DIR2_LEAF1_MAGIC),
202 cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) },
203 .verify_read = xfs_dir3_leaf_read_verify,
204 .verify_write = xfs_dir3_leaf_write_verify,
205 .verify_struct = xfs_dir3_leaf_verify,
208 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
209 .name = "xfs_dir3_leafn",
210 .magic16 = { cpu_to_be16(XFS_DIR2_LEAFN_MAGIC),
211 cpu_to_be16(XFS_DIR3_LEAFN_MAGIC) },
212 .verify_read = xfs_dir3_leaf_read_verify,
213 .verify_write = xfs_dir3_leaf_write_verify,
214 .verify_struct = xfs_dir3_leaf_verify,
218 xfs_dir3_leaf_read(
219 struct xfs_trans *tp,
220 struct xfs_inode *dp,
221 xfs_dablk_t fbno,
222 xfs_daddr_t mappedbno,
223 struct xfs_buf **bpp)
225 int err;
227 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
228 XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
229 if (!err && tp && *bpp)
230 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
231 return err;
235 xfs_dir3_leafn_read(
236 struct xfs_trans *tp,
237 struct xfs_inode *dp,
238 xfs_dablk_t fbno,
239 xfs_daddr_t mappedbno,
240 struct xfs_buf **bpp)
242 int err;
244 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
245 XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
246 if (!err && tp && *bpp)
247 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
248 return err;
252 * Initialize a new leaf block, leaf1 or leafn magic accepted.
254 static void
255 xfs_dir3_leaf_init(
256 struct xfs_mount *mp,
257 struct xfs_trans *tp,
258 struct xfs_buf *bp,
259 xfs_ino_t owner,
260 uint16_t type)
262 struct xfs_dir2_leaf *leaf = bp->b_addr;
264 ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
266 if (xfs_sb_version_hascrc(&mp->m_sb)) {
267 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
269 memset(leaf3, 0, sizeof(*leaf3));
271 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
272 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
273 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
274 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
275 leaf3->info.owner = cpu_to_be64(owner);
276 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid);
277 } else {
278 memset(leaf, 0, sizeof(*leaf));
279 leaf->hdr.info.magic = cpu_to_be16(type);
283 * If it's a leaf-format directory initialize the tail.
284 * Caller is responsible for initialising the bests table.
286 if (type == XFS_DIR2_LEAF1_MAGIC) {
287 struct xfs_dir2_leaf_tail *ltp;
289 ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
290 ltp->bestcount = 0;
291 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
292 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
293 } else {
294 bp->b_ops = &xfs_dir3_leafn_buf_ops;
295 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
300 xfs_dir3_leaf_get_buf(
301 xfs_da_args_t *args,
302 xfs_dir2_db_t bno,
303 struct xfs_buf **bpp,
304 uint16_t magic)
306 struct xfs_inode *dp = args->dp;
307 struct xfs_trans *tp = args->trans;
308 struct xfs_mount *mp = dp->i_mount;
309 struct xfs_buf *bp;
310 int error;
312 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
313 ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
314 bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
316 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
317 -1, &bp, XFS_DATA_FORK);
318 if (error)
319 return error;
321 xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
322 xfs_dir3_leaf_log_header(args, bp);
323 if (magic == XFS_DIR2_LEAF1_MAGIC)
324 xfs_dir3_leaf_log_tail(args, bp);
325 *bpp = bp;
326 return 0;
330 * Convert a block form directory to a leaf form directory.
332 int /* error */
333 xfs_dir2_block_to_leaf(
334 xfs_da_args_t *args, /* operation arguments */
335 struct xfs_buf *dbp) /* input block's buffer */
337 __be16 *bestsp; /* leaf's bestsp entries */
338 xfs_dablk_t blkno; /* leaf block's bno */
339 xfs_dir2_data_hdr_t *hdr; /* block header */
340 xfs_dir2_leaf_entry_t *blp; /* block's leaf entries */
341 xfs_dir2_block_tail_t *btp; /* block's tail */
342 xfs_inode_t *dp; /* incore directory inode */
343 int error; /* error return code */
344 struct xfs_buf *lbp; /* leaf block's buffer */
345 xfs_dir2_db_t ldb; /* leaf block's bno */
346 xfs_dir2_leaf_t *leaf; /* leaf structure */
347 xfs_dir2_leaf_tail_t *ltp; /* leaf's tail */
348 int needlog; /* need to log block header */
349 int needscan; /* need to rescan bestfree */
350 xfs_trans_t *tp; /* transaction pointer */
351 struct xfs_dir2_data_free *bf;
352 struct xfs_dir2_leaf_entry *ents;
353 struct xfs_dir3_icleaf_hdr leafhdr;
355 trace_xfs_dir2_block_to_leaf(args);
357 dp = args->dp;
358 tp = args->trans;
360 * Add the leaf block to the inode.
361 * This interface will only put blocks in the leaf/node range.
362 * Since that's empty now, we'll get the root (block 0 in range).
364 if ((error = xfs_da_grow_inode(args, &blkno))) {
365 return error;
367 ldb = xfs_dir2_da_to_db(args->geo, blkno);
368 ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
370 * Initialize the leaf block, get a buffer for it.
372 error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
373 if (error)
374 return error;
376 leaf = lbp->b_addr;
377 hdr = dbp->b_addr;
378 xfs_dir3_data_check(dp, dbp);
379 btp = xfs_dir2_block_tail_p(args->geo, hdr);
380 blp = xfs_dir2_block_leaf_p(btp);
381 bf = dp->d_ops->data_bestfree_p(hdr);
382 ents = dp->d_ops->leaf_ents_p(leaf);
385 * Set the counts in the leaf header.
387 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
388 leafhdr.count = be32_to_cpu(btp->count);
389 leafhdr.stale = be32_to_cpu(btp->stale);
390 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
391 xfs_dir3_leaf_log_header(args, lbp);
394 * Could compact these but I think we always do the conversion
395 * after squeezing out stale entries.
397 memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
398 xfs_dir3_leaf_log_ents(args, lbp, 0, leafhdr.count - 1);
399 needscan = 0;
400 needlog = 1;
402 * Make the space formerly occupied by the leaf entries and block
403 * tail be free.
405 xfs_dir2_data_make_free(args, dbp,
406 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
407 (xfs_dir2_data_aoff_t)((char *)hdr + args->geo->blksize -
408 (char *)blp),
409 &needlog, &needscan);
411 * Fix up the block header, make it a data block.
413 dbp->b_ops = &xfs_dir3_data_buf_ops;
414 xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
415 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
416 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
417 else
418 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
420 if (needscan)
421 xfs_dir2_data_freescan(dp, hdr, &needlog);
423 * Set up leaf tail and bests table.
425 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
426 ltp->bestcount = cpu_to_be32(1);
427 bestsp = xfs_dir2_leaf_bests_p(ltp);
428 bestsp[0] = bf[0].length;
430 * Log the data header and leaf bests table.
432 if (needlog)
433 xfs_dir2_data_log_header(args, dbp);
434 xfs_dir3_leaf_check(dp, lbp);
435 xfs_dir3_data_check(dp, dbp);
436 xfs_dir3_leaf_log_bests(args, lbp, 0, 0);
437 return 0;
440 STATIC void
441 xfs_dir3_leaf_find_stale(
442 struct xfs_dir3_icleaf_hdr *leafhdr,
443 struct xfs_dir2_leaf_entry *ents,
444 int index,
445 int *lowstale,
446 int *highstale)
449 * Find the first stale entry before our index, if any.
451 for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
452 if (ents[*lowstale].address ==
453 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
454 break;
458 * Find the first stale entry at or after our index, if any.
459 * Stop if the result would require moving more entries than using
460 * lowstale.
462 for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
463 if (ents[*highstale].address ==
464 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
465 break;
466 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
467 break;
471 struct xfs_dir2_leaf_entry *
472 xfs_dir3_leaf_find_entry(
473 struct xfs_dir3_icleaf_hdr *leafhdr,
474 struct xfs_dir2_leaf_entry *ents,
475 int index, /* leaf table position */
476 int compact, /* need to compact leaves */
477 int lowstale, /* index of prev stale leaf */
478 int highstale, /* index of next stale leaf */
479 int *lfloglow, /* low leaf logging index */
480 int *lfloghigh) /* high leaf logging index */
482 if (!leafhdr->stale) {
483 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
486 * Now we need to make room to insert the leaf entry.
488 * If there are no stale entries, just insert a hole at index.
490 lep = &ents[index];
491 if (index < leafhdr->count)
492 memmove(lep + 1, lep,
493 (leafhdr->count - index) * sizeof(*lep));
496 * Record low and high logging indices for the leaf.
498 *lfloglow = index;
499 *lfloghigh = leafhdr->count++;
500 return lep;
504 * There are stale entries.
506 * We will use one of them for the new entry. It's probably not at
507 * the right location, so we'll have to shift some up or down first.
509 * If we didn't compact before, we need to find the nearest stale
510 * entries before and after our insertion point.
512 if (compact == 0)
513 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
514 &lowstale, &highstale);
517 * If the low one is better, use it.
519 if (lowstale >= 0 &&
520 (highstale == leafhdr->count ||
521 index - lowstale - 1 < highstale - index)) {
522 ASSERT(index - lowstale - 1 >= 0);
523 ASSERT(ents[lowstale].address ==
524 cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
527 * Copy entries up to cover the stale entry and make room
528 * for the new entry.
530 if (index - lowstale - 1 > 0) {
531 memmove(&ents[lowstale], &ents[lowstale + 1],
532 (index - lowstale - 1) *
533 sizeof(xfs_dir2_leaf_entry_t));
535 *lfloglow = min(lowstale, *lfloglow);
536 *lfloghigh = max(index - 1, *lfloghigh);
537 leafhdr->stale--;
538 return &ents[index - 1];
542 * The high one is better, so use that one.
544 ASSERT(highstale - index >= 0);
545 ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
548 * Copy entries down to cover the stale entry and make room for the
549 * new entry.
551 if (highstale - index > 0) {
552 memmove(&ents[index + 1], &ents[index],
553 (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
555 *lfloglow = min(index, *lfloglow);
556 *lfloghigh = max(highstale, *lfloghigh);
557 leafhdr->stale--;
558 return &ents[index];
562 * Add an entry to a leaf form directory.
564 int /* error */
565 xfs_dir2_leaf_addname(
566 struct xfs_da_args *args) /* operation arguments */
568 struct xfs_dir3_icleaf_hdr leafhdr;
569 struct xfs_trans *tp = args->trans;
570 __be16 *bestsp; /* freespace table in leaf */
571 __be16 *tagp; /* end of data entry */
572 struct xfs_buf *dbp; /* data block buffer */
573 struct xfs_buf *lbp; /* leaf's buffer */
574 struct xfs_dir2_leaf *leaf; /* leaf structure */
575 struct xfs_inode *dp = args->dp; /* incore directory inode */
576 struct xfs_dir2_data_hdr *hdr; /* data block header */
577 struct xfs_dir2_data_entry *dep; /* data block entry */
578 struct xfs_dir2_leaf_entry *lep; /* leaf entry table pointer */
579 struct xfs_dir2_leaf_entry *ents;
580 struct xfs_dir2_data_unused *dup; /* data unused entry */
581 struct xfs_dir2_leaf_tail *ltp; /* leaf tail pointer */
582 struct xfs_dir2_data_free *bf; /* bestfree table */
583 int compact; /* need to compact leaves */
584 int error; /* error return value */
585 int grown; /* allocated new data block */
586 int highstale = 0; /* index of next stale leaf */
587 int i; /* temporary, index */
588 int index; /* leaf table position */
589 int length; /* length of new entry */
590 int lfloglow; /* low leaf logging index */
591 int lfloghigh; /* high leaf logging index */
592 int lowstale = 0; /* index of prev stale leaf */
593 int needbytes; /* leaf block bytes needed */
594 int needlog; /* need to log data header */
595 int needscan; /* need to rescan data free */
596 xfs_dir2_db_t use_block; /* data block number */
598 trace_xfs_dir2_leaf_addname(args);
600 error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
601 if (error)
602 return error;
605 * Look up the entry by hash value and name.
606 * We know it's not there, our caller has already done a lookup.
607 * So the index is of the entry to insert in front of.
608 * But if there are dup hash values the index is of the first of those.
610 index = xfs_dir2_leaf_search_hash(args, lbp);
611 leaf = lbp->b_addr;
612 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
613 ents = dp->d_ops->leaf_ents_p(leaf);
614 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
615 bestsp = xfs_dir2_leaf_bests_p(ltp);
616 length = dp->d_ops->data_entsize(args->namelen);
619 * See if there are any entries with the same hash value
620 * and space in their block for the new entry.
621 * This is good because it puts multiple same-hash value entries
622 * in a data block, improving the lookup of those entries.
624 for (use_block = -1, lep = &ents[index];
625 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
626 index++, lep++) {
627 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
628 continue;
629 i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
630 ASSERT(i < be32_to_cpu(ltp->bestcount));
631 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
632 if (be16_to_cpu(bestsp[i]) >= length) {
633 use_block = i;
634 break;
638 * Didn't find a block yet, linear search all the data blocks.
640 if (use_block == -1) {
641 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
643 * Remember a block we see that's missing.
645 if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
646 use_block == -1)
647 use_block = i;
648 else if (be16_to_cpu(bestsp[i]) >= length) {
649 use_block = i;
650 break;
655 * How many bytes do we need in the leaf block?
657 needbytes = 0;
658 if (!leafhdr.stale)
659 needbytes += sizeof(xfs_dir2_leaf_entry_t);
660 if (use_block == -1)
661 needbytes += sizeof(xfs_dir2_data_off_t);
664 * Now kill use_block if it refers to a missing block, so we
665 * can use it as an indication of allocation needed.
667 if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
668 use_block = -1;
670 * If we don't have enough free bytes but we can make enough
671 * by compacting out stale entries, we'll do that.
673 if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
674 leafhdr.stale > 1)
675 compact = 1;
678 * Otherwise if we don't have enough free bytes we need to
679 * convert to node form.
681 else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
683 * Just checking or no space reservation, give up.
685 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
686 args->total == 0) {
687 xfs_trans_brelse(tp, lbp);
688 return -ENOSPC;
691 * Convert to node form.
693 error = xfs_dir2_leaf_to_node(args, lbp);
694 if (error)
695 return error;
697 * Then add the new entry.
699 return xfs_dir2_node_addname(args);
702 * Otherwise it will fit without compaction.
704 else
705 compact = 0;
707 * If just checking, then it will fit unless we needed to allocate
708 * a new data block.
710 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
711 xfs_trans_brelse(tp, lbp);
712 return use_block == -1 ? -ENOSPC : 0;
715 * If no allocations are allowed, return now before we've
716 * changed anything.
718 if (args->total == 0 && use_block == -1) {
719 xfs_trans_brelse(tp, lbp);
720 return -ENOSPC;
723 * Need to compact the leaf entries, removing stale ones.
724 * Leave one stale entry behind - the one closest to our
725 * insertion index - and we'll shift that one to our insertion
726 * point later.
728 if (compact) {
729 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
730 &highstale, &lfloglow, &lfloghigh);
733 * There are stale entries, so we'll need log-low and log-high
734 * impossibly bad values later.
736 else if (leafhdr.stale) {
737 lfloglow = leafhdr.count;
738 lfloghigh = -1;
741 * If there was no data block space found, we need to allocate
742 * a new one.
744 if (use_block == -1) {
746 * Add the new data block.
748 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
749 &use_block))) {
750 xfs_trans_brelse(tp, lbp);
751 return error;
754 * Initialize the block.
756 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
757 xfs_trans_brelse(tp, lbp);
758 return error;
761 * If we're adding a new data block on the end we need to
762 * extend the bests table. Copy it up one entry.
764 if (use_block >= be32_to_cpu(ltp->bestcount)) {
765 bestsp--;
766 memmove(&bestsp[0], &bestsp[1],
767 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
768 be32_add_cpu(&ltp->bestcount, 1);
769 xfs_dir3_leaf_log_tail(args, lbp);
770 xfs_dir3_leaf_log_bests(args, lbp, 0,
771 be32_to_cpu(ltp->bestcount) - 1);
774 * If we're filling in a previously empty block just log it.
776 else
777 xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
778 hdr = dbp->b_addr;
779 bf = dp->d_ops->data_bestfree_p(hdr);
780 bestsp[use_block] = bf[0].length;
781 grown = 1;
782 } else {
784 * Already had space in some data block.
785 * Just read that one in.
787 error = xfs_dir3_data_read(tp, dp,
788 xfs_dir2_db_to_da(args->geo, use_block),
789 -1, &dbp);
790 if (error) {
791 xfs_trans_brelse(tp, lbp);
792 return error;
794 hdr = dbp->b_addr;
795 bf = dp->d_ops->data_bestfree_p(hdr);
796 grown = 0;
799 * Point to the biggest freespace in our data block.
801 dup = (xfs_dir2_data_unused_t *)
802 ((char *)hdr + be16_to_cpu(bf[0].offset));
803 needscan = needlog = 0;
805 * Mark the initial part of our freespace in use for the new entry.
807 error = xfs_dir2_data_use_free(args, dbp, dup,
808 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr),
809 length, &needlog, &needscan);
810 if (error) {
811 xfs_trans_brelse(tp, lbp);
812 return error;
815 * Initialize our new entry (at last).
817 dep = (xfs_dir2_data_entry_t *)dup;
818 dep->inumber = cpu_to_be64(args->inumber);
819 dep->namelen = args->namelen;
820 memcpy(dep->name, args->name, dep->namelen);
821 dp->d_ops->data_put_ftype(dep, args->filetype);
822 tagp = dp->d_ops->data_entry_tag_p(dep);
823 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
825 * Need to scan fix up the bestfree table.
827 if (needscan)
828 xfs_dir2_data_freescan(dp, hdr, &needlog);
830 * Need to log the data block's header.
832 if (needlog)
833 xfs_dir2_data_log_header(args, dbp);
834 xfs_dir2_data_log_entry(args, dbp, dep);
836 * If the bests table needs to be changed, do it.
837 * Log the change unless we've already done that.
839 if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
840 bestsp[use_block] = bf[0].length;
841 if (!grown)
842 xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
845 lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
846 highstale, &lfloglow, &lfloghigh);
849 * Fill in the new leaf entry.
851 lep->hashval = cpu_to_be32(args->hashval);
852 lep->address = cpu_to_be32(
853 xfs_dir2_db_off_to_dataptr(args->geo, use_block,
854 be16_to_cpu(*tagp)));
856 * Log the leaf fields and give up the buffers.
858 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
859 xfs_dir3_leaf_log_header(args, lbp);
860 xfs_dir3_leaf_log_ents(args, lbp, lfloglow, lfloghigh);
861 xfs_dir3_leaf_check(dp, lbp);
862 xfs_dir3_data_check(dp, dbp);
863 return 0;
867 * Compact out any stale entries in the leaf.
868 * Log the header and changed leaf entries, if any.
870 void
871 xfs_dir3_leaf_compact(
872 xfs_da_args_t *args, /* operation arguments */
873 struct xfs_dir3_icleaf_hdr *leafhdr,
874 struct xfs_buf *bp) /* leaf buffer */
876 int from; /* source leaf index */
877 xfs_dir2_leaf_t *leaf; /* leaf structure */
878 int loglow; /* first leaf entry to log */
879 int to; /* target leaf index */
880 struct xfs_dir2_leaf_entry *ents;
881 struct xfs_inode *dp = args->dp;
883 leaf = bp->b_addr;
884 if (!leafhdr->stale)
885 return;
888 * Compress out the stale entries in place.
890 ents = dp->d_ops->leaf_ents_p(leaf);
891 for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
892 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
893 continue;
895 * Only actually copy the entries that are different.
897 if (from > to) {
898 if (loglow == -1)
899 loglow = to;
900 ents[to] = ents[from];
902 to++;
905 * Update and log the header, log the leaf entries.
907 ASSERT(leafhdr->stale == from - to);
908 leafhdr->count -= leafhdr->stale;
909 leafhdr->stale = 0;
911 dp->d_ops->leaf_hdr_to_disk(leaf, leafhdr);
912 xfs_dir3_leaf_log_header(args, bp);
913 if (loglow != -1)
914 xfs_dir3_leaf_log_ents(args, bp, loglow, to - 1);
918 * Compact the leaf entries, removing stale ones.
919 * Leave one stale entry behind - the one closest to our
920 * insertion index - and the caller will shift that one to our insertion
921 * point later.
922 * Return new insertion index, where the remaining stale entry is,
923 * and leaf logging indices.
925 void
926 xfs_dir3_leaf_compact_x1(
927 struct xfs_dir3_icleaf_hdr *leafhdr,
928 struct xfs_dir2_leaf_entry *ents,
929 int *indexp, /* insertion index */
930 int *lowstalep, /* out: stale entry before us */
931 int *highstalep, /* out: stale entry after us */
932 int *lowlogp, /* out: low log index */
933 int *highlogp) /* out: high log index */
935 int from; /* source copy index */
936 int highstale; /* stale entry at/after index */
937 int index; /* insertion index */
938 int keepstale; /* source index of kept stale */
939 int lowstale; /* stale entry before index */
940 int newindex=0; /* new insertion index */
941 int to; /* destination copy index */
943 ASSERT(leafhdr->stale > 1);
944 index = *indexp;
946 xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
949 * Pick the better of lowstale and highstale.
951 if (lowstale >= 0 &&
952 (highstale == leafhdr->count ||
953 index - lowstale <= highstale - index))
954 keepstale = lowstale;
955 else
956 keepstale = highstale;
958 * Copy the entries in place, removing all the stale entries
959 * except keepstale.
961 for (from = to = 0; from < leafhdr->count; from++) {
963 * Notice the new value of index.
965 if (index == from)
966 newindex = to;
967 if (from != keepstale &&
968 ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
969 if (from == to)
970 *lowlogp = to;
971 continue;
974 * Record the new keepstale value for the insertion.
976 if (from == keepstale)
977 lowstale = highstale = to;
979 * Copy only the entries that have moved.
981 if (from > to)
982 ents[to] = ents[from];
983 to++;
985 ASSERT(from > to);
987 * If the insertion point was past the last entry,
988 * set the new insertion point accordingly.
990 if (index == from)
991 newindex = to;
992 *indexp = newindex;
994 * Adjust the leaf header values.
996 leafhdr->count -= from - to;
997 leafhdr->stale = 1;
999 * Remember the low/high stale value only in the "right"
1000 * direction.
1002 if (lowstale >= newindex)
1003 lowstale = -1;
1004 else
1005 highstale = leafhdr->count;
1006 *highlogp = leafhdr->count - 1;
1007 *lowstalep = lowstale;
1008 *highstalep = highstale;
1012 * Log the bests entries indicated from a leaf1 block.
1014 static void
1015 xfs_dir3_leaf_log_bests(
1016 struct xfs_da_args *args,
1017 struct xfs_buf *bp, /* leaf buffer */
1018 int first, /* first entry to log */
1019 int last) /* last entry to log */
1021 __be16 *firstb; /* pointer to first entry */
1022 __be16 *lastb; /* pointer to last entry */
1023 struct xfs_dir2_leaf *leaf = bp->b_addr;
1024 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1026 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1027 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1029 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1030 firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1031 lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1032 xfs_trans_log_buf(args->trans, bp,
1033 (uint)((char *)firstb - (char *)leaf),
1034 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1038 * Log the leaf entries indicated from a leaf1 or leafn block.
1040 void
1041 xfs_dir3_leaf_log_ents(
1042 struct xfs_da_args *args,
1043 struct xfs_buf *bp,
1044 int first,
1045 int last)
1047 xfs_dir2_leaf_entry_t *firstlep; /* pointer to first entry */
1048 xfs_dir2_leaf_entry_t *lastlep; /* pointer to last entry */
1049 struct xfs_dir2_leaf *leaf = bp->b_addr;
1050 struct xfs_dir2_leaf_entry *ents;
1052 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1053 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1054 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1055 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1057 ents = args->dp->d_ops->leaf_ents_p(leaf);
1058 firstlep = &ents[first];
1059 lastlep = &ents[last];
1060 xfs_trans_log_buf(args->trans, bp,
1061 (uint)((char *)firstlep - (char *)leaf),
1062 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1066 * Log the header of the leaf1 or leafn block.
1068 void
1069 xfs_dir3_leaf_log_header(
1070 struct xfs_da_args *args,
1071 struct xfs_buf *bp)
1073 struct xfs_dir2_leaf *leaf = bp->b_addr;
1075 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1076 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1077 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1078 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1080 xfs_trans_log_buf(args->trans, bp,
1081 (uint)((char *)&leaf->hdr - (char *)leaf),
1082 args->dp->d_ops->leaf_hdr_size - 1);
1086 * Log the tail of the leaf1 block.
1088 STATIC void
1089 xfs_dir3_leaf_log_tail(
1090 struct xfs_da_args *args,
1091 struct xfs_buf *bp)
1093 struct xfs_dir2_leaf *leaf = bp->b_addr;
1094 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1096 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1097 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1098 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1099 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1101 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1102 xfs_trans_log_buf(args->trans, bp, (uint)((char *)ltp - (char *)leaf),
1103 (uint)(args->geo->blksize - 1));
1107 * Look up the entry referred to by args in the leaf format directory.
1108 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1109 * is also used by the node-format code.
1112 xfs_dir2_leaf_lookup(
1113 xfs_da_args_t *args) /* operation arguments */
1115 struct xfs_buf *dbp; /* data block buffer */
1116 xfs_dir2_data_entry_t *dep; /* data block entry */
1117 xfs_inode_t *dp; /* incore directory inode */
1118 int error; /* error return code */
1119 int index; /* found entry index */
1120 struct xfs_buf *lbp; /* leaf buffer */
1121 xfs_dir2_leaf_t *leaf; /* leaf structure */
1122 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1123 xfs_trans_t *tp; /* transaction pointer */
1124 struct xfs_dir2_leaf_entry *ents;
1126 trace_xfs_dir2_leaf_lookup(args);
1129 * Look up name in the leaf block, returning both buffers and index.
1131 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1132 return error;
1134 tp = args->trans;
1135 dp = args->dp;
1136 xfs_dir3_leaf_check(dp, lbp);
1137 leaf = lbp->b_addr;
1138 ents = dp->d_ops->leaf_ents_p(leaf);
1140 * Get to the leaf entry and contained data entry address.
1142 lep = &ents[index];
1145 * Point to the data entry.
1147 dep = (xfs_dir2_data_entry_t *)
1148 ((char *)dbp->b_addr +
1149 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1151 * Return the found inode number & CI name if appropriate
1153 args->inumber = be64_to_cpu(dep->inumber);
1154 args->filetype = dp->d_ops->data_get_ftype(dep);
1155 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1156 xfs_trans_brelse(tp, dbp);
1157 xfs_trans_brelse(tp, lbp);
1158 return error;
1162 * Look up name/hash in the leaf block.
1163 * Fill in indexp with the found index, and dbpp with the data buffer.
1164 * If not found dbpp will be NULL, and ENOENT comes back.
1165 * lbpp will always be filled in with the leaf buffer unless there's an error.
1167 static int /* error */
1168 xfs_dir2_leaf_lookup_int(
1169 xfs_da_args_t *args, /* operation arguments */
1170 struct xfs_buf **lbpp, /* out: leaf buffer */
1171 int *indexp, /* out: index in leaf block */
1172 struct xfs_buf **dbpp) /* out: data buffer */
1174 xfs_dir2_db_t curdb = -1; /* current data block number */
1175 struct xfs_buf *dbp = NULL; /* data buffer */
1176 xfs_dir2_data_entry_t *dep; /* data entry */
1177 xfs_inode_t *dp; /* incore directory inode */
1178 int error; /* error return code */
1179 int index; /* index in leaf block */
1180 struct xfs_buf *lbp; /* leaf buffer */
1181 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1182 xfs_dir2_leaf_t *leaf; /* leaf structure */
1183 xfs_mount_t *mp; /* filesystem mount point */
1184 xfs_dir2_db_t newdb; /* new data block number */
1185 xfs_trans_t *tp; /* transaction pointer */
1186 xfs_dir2_db_t cidb = -1; /* case match data block no. */
1187 enum xfs_dacmp cmp; /* name compare result */
1188 struct xfs_dir2_leaf_entry *ents;
1189 struct xfs_dir3_icleaf_hdr leafhdr;
1191 dp = args->dp;
1192 tp = args->trans;
1193 mp = dp->i_mount;
1195 error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
1196 if (error)
1197 return error;
1199 *lbpp = lbp;
1200 leaf = lbp->b_addr;
1201 xfs_dir3_leaf_check(dp, lbp);
1202 ents = dp->d_ops->leaf_ents_p(leaf);
1203 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1206 * Look for the first leaf entry with our hash value.
1208 index = xfs_dir2_leaf_search_hash(args, lbp);
1210 * Loop over all the entries with the right hash value
1211 * looking to match the name.
1213 for (lep = &ents[index];
1214 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1215 lep++, index++) {
1217 * Skip over stale leaf entries.
1219 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1220 continue;
1222 * Get the new data block number.
1224 newdb = xfs_dir2_dataptr_to_db(args->geo,
1225 be32_to_cpu(lep->address));
1227 * If it's not the same as the old data block number,
1228 * need to pitch the old one and read the new one.
1230 if (newdb != curdb) {
1231 if (dbp)
1232 xfs_trans_brelse(tp, dbp);
1233 error = xfs_dir3_data_read(tp, dp,
1234 xfs_dir2_db_to_da(args->geo, newdb),
1235 -1, &dbp);
1236 if (error) {
1237 xfs_trans_brelse(tp, lbp);
1238 return error;
1240 curdb = newdb;
1243 * Point to the data entry.
1245 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1246 xfs_dir2_dataptr_to_off(args->geo,
1247 be32_to_cpu(lep->address)));
1249 * Compare name and if it's an exact match, return the index
1250 * and buffer. If it's the first case-insensitive match, store
1251 * the index and buffer and continue looking for an exact match.
1253 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1254 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1255 args->cmpresult = cmp;
1256 *indexp = index;
1257 /* case exact match: return the current buffer. */
1258 if (cmp == XFS_CMP_EXACT) {
1259 *dbpp = dbp;
1260 return 0;
1262 cidb = curdb;
1265 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1267 * Here, we can only be doing a lookup (not a rename or remove).
1268 * If a case-insensitive match was found earlier, re-read the
1269 * appropriate data block if required and return it.
1271 if (args->cmpresult == XFS_CMP_CASE) {
1272 ASSERT(cidb != -1);
1273 if (cidb != curdb) {
1274 xfs_trans_brelse(tp, dbp);
1275 error = xfs_dir3_data_read(tp, dp,
1276 xfs_dir2_db_to_da(args->geo, cidb),
1277 -1, &dbp);
1278 if (error) {
1279 xfs_trans_brelse(tp, lbp);
1280 return error;
1283 *dbpp = dbp;
1284 return 0;
1287 * No match found, return -ENOENT.
1289 ASSERT(cidb == -1);
1290 if (dbp)
1291 xfs_trans_brelse(tp, dbp);
1292 xfs_trans_brelse(tp, lbp);
1293 return -ENOENT;
1297 * Remove an entry from a leaf format directory.
1299 int /* error */
1300 xfs_dir2_leaf_removename(
1301 xfs_da_args_t *args) /* operation arguments */
1303 __be16 *bestsp; /* leaf block best freespace */
1304 xfs_dir2_data_hdr_t *hdr; /* data block header */
1305 xfs_dir2_db_t db; /* data block number */
1306 struct xfs_buf *dbp; /* data block buffer */
1307 xfs_dir2_data_entry_t *dep; /* data entry structure */
1308 xfs_inode_t *dp; /* incore directory inode */
1309 int error; /* error return code */
1310 xfs_dir2_db_t i; /* temporary data block # */
1311 int index; /* index into leaf entries */
1312 struct xfs_buf *lbp; /* leaf buffer */
1313 xfs_dir2_leaf_t *leaf; /* leaf structure */
1314 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1315 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1316 int needlog; /* need to log data header */
1317 int needscan; /* need to rescan data frees */
1318 xfs_dir2_data_off_t oldbest; /* old value of best free */
1319 struct xfs_dir2_data_free *bf; /* bestfree table */
1320 struct xfs_dir2_leaf_entry *ents;
1321 struct xfs_dir3_icleaf_hdr leafhdr;
1323 trace_xfs_dir2_leaf_removename(args);
1326 * Lookup the leaf entry, get the leaf and data blocks read in.
1328 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1329 return error;
1331 dp = args->dp;
1332 leaf = lbp->b_addr;
1333 hdr = dbp->b_addr;
1334 xfs_dir3_data_check(dp, dbp);
1335 bf = dp->d_ops->data_bestfree_p(hdr);
1336 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1337 ents = dp->d_ops->leaf_ents_p(leaf);
1339 * Point to the leaf entry, use that to point to the data entry.
1341 lep = &ents[index];
1342 db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1343 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1344 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1345 needscan = needlog = 0;
1346 oldbest = be16_to_cpu(bf[0].length);
1347 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1348 bestsp = xfs_dir2_leaf_bests_p(ltp);
1349 if (be16_to_cpu(bestsp[db]) != oldbest)
1350 return -EFSCORRUPTED;
1352 * Mark the former data entry unused.
1354 xfs_dir2_data_make_free(args, dbp,
1355 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1356 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1358 * We just mark the leaf entry stale by putting a null in it.
1360 leafhdr.stale++;
1361 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1362 xfs_dir3_leaf_log_header(args, lbp);
1364 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1365 xfs_dir3_leaf_log_ents(args, lbp, index, index);
1368 * Scan the freespace in the data block again if necessary,
1369 * log the data block header if necessary.
1371 if (needscan)
1372 xfs_dir2_data_freescan(dp, hdr, &needlog);
1373 if (needlog)
1374 xfs_dir2_data_log_header(args, dbp);
1376 * If the longest freespace in the data block has changed,
1377 * put the new value in the bests table and log that.
1379 if (be16_to_cpu(bf[0].length) != oldbest) {
1380 bestsp[db] = bf[0].length;
1381 xfs_dir3_leaf_log_bests(args, lbp, db, db);
1383 xfs_dir3_data_check(dp, dbp);
1385 * If the data block is now empty then get rid of the data block.
1387 if (be16_to_cpu(bf[0].length) ==
1388 args->geo->blksize - dp->d_ops->data_entry_offset) {
1389 ASSERT(db != args->geo->datablk);
1390 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1392 * Nope, can't get rid of it because it caused
1393 * allocation of a bmap btree block to do so.
1394 * Just go on, returning success, leaving the
1395 * empty block in place.
1397 if (error == -ENOSPC && args->total == 0)
1398 error = 0;
1399 xfs_dir3_leaf_check(dp, lbp);
1400 return error;
1402 dbp = NULL;
1404 * If this is the last data block then compact the
1405 * bests table by getting rid of entries.
1407 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1409 * Look for the last active entry (i).
1411 for (i = db - 1; i > 0; i--) {
1412 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1413 break;
1416 * Copy the table down so inactive entries at the
1417 * end are removed.
1419 memmove(&bestsp[db - i], bestsp,
1420 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1421 be32_add_cpu(&ltp->bestcount, -(db - i));
1422 xfs_dir3_leaf_log_tail(args, lbp);
1423 xfs_dir3_leaf_log_bests(args, lbp, 0,
1424 be32_to_cpu(ltp->bestcount) - 1);
1425 } else
1426 bestsp[db] = cpu_to_be16(NULLDATAOFF);
1429 * If the data block was not the first one, drop it.
1431 else if (db != args->geo->datablk)
1432 dbp = NULL;
1434 xfs_dir3_leaf_check(dp, lbp);
1436 * See if we can convert to block form.
1438 return xfs_dir2_leaf_to_block(args, lbp, dbp);
1442 * Replace the inode number in a leaf format directory entry.
1444 int /* error */
1445 xfs_dir2_leaf_replace(
1446 xfs_da_args_t *args) /* operation arguments */
1448 struct xfs_buf *dbp; /* data block buffer */
1449 xfs_dir2_data_entry_t *dep; /* data block entry */
1450 xfs_inode_t *dp; /* incore directory inode */
1451 int error; /* error return code */
1452 int index; /* index of leaf entry */
1453 struct xfs_buf *lbp; /* leaf buffer */
1454 xfs_dir2_leaf_t *leaf; /* leaf structure */
1455 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1456 xfs_trans_t *tp; /* transaction pointer */
1457 struct xfs_dir2_leaf_entry *ents;
1459 trace_xfs_dir2_leaf_replace(args);
1462 * Look up the entry.
1464 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1465 return error;
1467 dp = args->dp;
1468 leaf = lbp->b_addr;
1469 ents = dp->d_ops->leaf_ents_p(leaf);
1471 * Point to the leaf entry, get data address from it.
1473 lep = &ents[index];
1475 * Point to the data entry.
1477 dep = (xfs_dir2_data_entry_t *)
1478 ((char *)dbp->b_addr +
1479 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1480 ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1482 * Put the new inode number in, log it.
1484 dep->inumber = cpu_to_be64(args->inumber);
1485 dp->d_ops->data_put_ftype(dep, args->filetype);
1486 tp = args->trans;
1487 xfs_dir2_data_log_entry(args, dbp, dep);
1488 xfs_dir3_leaf_check(dp, lbp);
1489 xfs_trans_brelse(tp, lbp);
1490 return 0;
1494 * Return index in the leaf block (lbp) which is either the first
1495 * one with this hash value, or if there are none, the insert point
1496 * for that hash value.
1498 int /* index value */
1499 xfs_dir2_leaf_search_hash(
1500 xfs_da_args_t *args, /* operation arguments */
1501 struct xfs_buf *lbp) /* leaf buffer */
1503 xfs_dahash_t hash=0; /* hash from this entry */
1504 xfs_dahash_t hashwant; /* hash value looking for */
1505 int high; /* high leaf index */
1506 int low; /* low leaf index */
1507 xfs_dir2_leaf_t *leaf; /* leaf structure */
1508 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1509 int mid=0; /* current leaf index */
1510 struct xfs_dir2_leaf_entry *ents;
1511 struct xfs_dir3_icleaf_hdr leafhdr;
1513 leaf = lbp->b_addr;
1514 ents = args->dp->d_ops->leaf_ents_p(leaf);
1515 args->dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1518 * Note, the table cannot be empty, so we have to go through the loop.
1519 * Binary search the leaf entries looking for our hash value.
1521 for (lep = ents, low = 0, high = leafhdr.count - 1,
1522 hashwant = args->hashval;
1523 low <= high; ) {
1524 mid = (low + high) >> 1;
1525 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1526 break;
1527 if (hash < hashwant)
1528 low = mid + 1;
1529 else
1530 high = mid - 1;
1533 * Found one, back up through all the equal hash values.
1535 if (hash == hashwant) {
1536 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1537 mid--;
1541 * Need to point to an entry higher than ours.
1543 else if (hash < hashwant)
1544 mid++;
1545 return mid;
1549 * Trim off a trailing data block. We know it's empty since the leaf
1550 * freespace table says so.
1552 int /* error */
1553 xfs_dir2_leaf_trim_data(
1554 xfs_da_args_t *args, /* operation arguments */
1555 struct xfs_buf *lbp, /* leaf buffer */
1556 xfs_dir2_db_t db) /* data block number */
1558 __be16 *bestsp; /* leaf bests table */
1559 struct xfs_buf *dbp; /* data block buffer */
1560 xfs_inode_t *dp; /* incore directory inode */
1561 int error; /* error return value */
1562 xfs_dir2_leaf_t *leaf; /* leaf structure */
1563 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1564 xfs_trans_t *tp; /* transaction pointer */
1566 dp = args->dp;
1567 tp = args->trans;
1569 * Read the offending data block. We need its buffer.
1571 error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(args->geo, db),
1572 -1, &dbp);
1573 if (error)
1574 return error;
1576 leaf = lbp->b_addr;
1577 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1579 #ifdef DEBUG
1581 struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1582 struct xfs_dir2_data_free *bf = dp->d_ops->data_bestfree_p(hdr);
1584 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1585 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1586 ASSERT(be16_to_cpu(bf[0].length) ==
1587 args->geo->blksize - dp->d_ops->data_entry_offset);
1588 ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1590 #endif
1593 * Get rid of the data block.
1595 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1596 ASSERT(error != -ENOSPC);
1597 xfs_trans_brelse(tp, dbp);
1598 return error;
1601 * Eliminate the last bests entry from the table.
1603 bestsp = xfs_dir2_leaf_bests_p(ltp);
1604 be32_add_cpu(&ltp->bestcount, -1);
1605 memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1606 xfs_dir3_leaf_log_tail(args, lbp);
1607 xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1608 return 0;
1611 static inline size_t
1612 xfs_dir3_leaf_size(
1613 struct xfs_dir3_icleaf_hdr *hdr,
1614 int counts)
1616 int entries;
1617 int hdrsize;
1619 entries = hdr->count - hdr->stale;
1620 if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1621 hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1622 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1623 else
1624 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1626 return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1627 + counts * sizeof(xfs_dir2_data_off_t)
1628 + sizeof(xfs_dir2_leaf_tail_t);
1632 * Convert node form directory to leaf form directory.
1633 * The root of the node form dir needs to already be a LEAFN block.
1634 * Just return if we can't do anything.
1636 int /* error */
1637 xfs_dir2_node_to_leaf(
1638 xfs_da_state_t *state) /* directory operation state */
1640 xfs_da_args_t *args; /* operation arguments */
1641 xfs_inode_t *dp; /* incore directory inode */
1642 int error; /* error return code */
1643 struct xfs_buf *fbp; /* buffer for freespace block */
1644 xfs_fileoff_t fo; /* freespace file offset */
1645 xfs_dir2_free_t *free; /* freespace structure */
1646 struct xfs_buf *lbp; /* buffer for leaf block */
1647 xfs_dir2_leaf_tail_t *ltp; /* tail of leaf structure */
1648 xfs_dir2_leaf_t *leaf; /* leaf structure */
1649 xfs_mount_t *mp; /* filesystem mount point */
1650 int rval; /* successful free trim? */
1651 xfs_trans_t *tp; /* transaction pointer */
1652 struct xfs_dir3_icleaf_hdr leafhdr;
1653 struct xfs_dir3_icfree_hdr freehdr;
1656 * There's more than a leaf level in the btree, so there must
1657 * be multiple leafn blocks. Give up.
1659 if (state->path.active > 1)
1660 return 0;
1661 args = state->args;
1663 trace_xfs_dir2_node_to_leaf(args);
1665 mp = state->mp;
1666 dp = args->dp;
1667 tp = args->trans;
1669 * Get the last offset in the file.
1671 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1672 return error;
1674 fo -= args->geo->fsbcount;
1676 * If there are freespace blocks other than the first one,
1677 * take this opportunity to remove trailing empty freespace blocks
1678 * that may have been left behind during no-space-reservation
1679 * operations.
1681 while (fo > args->geo->freeblk) {
1682 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1683 return error;
1685 if (rval)
1686 fo -= args->geo->fsbcount;
1687 else
1688 return 0;
1691 * Now find the block just before the freespace block.
1693 if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1694 return error;
1697 * If it's not the single leaf block, give up.
1699 if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1700 return 0;
1701 lbp = state->path.blk[0].bp;
1702 leaf = lbp->b_addr;
1703 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1705 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1706 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1709 * Read the freespace block.
1711 error = xfs_dir2_free_read(tp, dp, args->geo->freeblk, &fbp);
1712 if (error)
1713 return error;
1714 free = fbp->b_addr;
1715 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1717 ASSERT(!freehdr.firstdb);
1720 * Now see if the leafn and free data will fit in a leaf1.
1721 * If not, release the buffer and give up.
1723 if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > args->geo->blksize) {
1724 xfs_trans_brelse(tp, fbp);
1725 return 0;
1729 * If the leaf has any stale entries in it, compress them out.
1731 if (leafhdr.stale)
1732 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1734 lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1735 xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1736 leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1737 ? XFS_DIR2_LEAF1_MAGIC
1738 : XFS_DIR3_LEAF1_MAGIC;
1741 * Set up the leaf tail from the freespace block.
1743 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1744 ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1747 * Set up the leaf bests table.
1749 memcpy(xfs_dir2_leaf_bests_p(ltp), dp->d_ops->free_bests_p(free),
1750 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1752 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1753 xfs_dir3_leaf_log_header(args, lbp);
1754 xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1755 xfs_dir3_leaf_log_tail(args, lbp);
1756 xfs_dir3_leaf_check(dp, lbp);
1759 * Get rid of the freespace block.
1761 error = xfs_dir2_shrink_inode(args,
1762 xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET),
1763 fbp);
1764 if (error) {
1766 * This can't fail here because it can only happen when
1767 * punching out the middle of an extent, and this is an
1768 * isolated block.
1770 ASSERT(error != -ENOSPC);
1771 return error;
1773 fbp = NULL;
1775 * Now see if we can convert the single-leaf directory
1776 * down to a block form directory.
1777 * This routine always kills the dabuf for the leaf, so
1778 * eliminate it from the path.
1780 error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1781 state->path.blk[0].bp = NULL;
1782 return error;