fix O_SYNC|O_APPEND syncing the wrong range on write()
[linux/fpc-iii.git] / fs / xfs / xfs_dir2_leaf.c
blobae47ec6e16c4031e50ba9ef63884ce216757d41e
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_bmap.h"
31 #include "xfs_dir2.h"
32 #include "xfs_dir2_priv.h"
33 #include "xfs_error.h"
34 #include "xfs_trace.h"
35 #include "xfs_trans.h"
36 #include "xfs_buf_item.h"
37 #include "xfs_cksum.h"
40 * Local function declarations.
42 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
43 int *indexp, struct xfs_buf **dbpp);
44 static void xfs_dir3_leaf_log_bests(struct xfs_trans *tp, struct xfs_buf *bp,
45 int first, int last);
46 static void xfs_dir3_leaf_log_tail(struct xfs_trans *tp, struct xfs_buf *bp);
49 * Check the internal consistency of a leaf1 block.
50 * Pop an assert if something is wrong.
52 #ifdef DEBUG
53 #define xfs_dir3_leaf_check(dp, bp) \
54 do { \
55 if (!xfs_dir3_leaf1_check((dp), (bp))) \
56 ASSERT(0); \
57 } while (0);
59 STATIC bool
60 xfs_dir3_leaf1_check(
61 struct xfs_inode *dp,
62 struct xfs_buf *bp)
64 struct xfs_dir2_leaf *leaf = bp->b_addr;
65 struct xfs_dir3_icleaf_hdr leafhdr;
67 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
69 if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
70 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
71 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
72 return false;
73 } else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
74 return false;
76 return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
78 #else
79 #define xfs_dir3_leaf_check(dp, bp)
80 #endif
82 bool
83 xfs_dir3_leaf_check_int(
84 struct xfs_mount *mp,
85 struct xfs_inode *dp,
86 struct xfs_dir3_icleaf_hdr *hdr,
87 struct xfs_dir2_leaf *leaf)
89 struct xfs_dir2_leaf_entry *ents;
90 xfs_dir2_leaf_tail_t *ltp;
91 int stale;
92 int i;
93 const struct xfs_dir_ops *ops;
94 struct xfs_dir3_icleaf_hdr leafhdr;
97 * we can be passed a null dp here from a verifier, so we need to go the
98 * hard way to get them.
100 ops = xfs_dir_get_ops(mp, dp);
102 if (!hdr) {
103 ops->leaf_hdr_from_disk(&leafhdr, leaf);
104 hdr = &leafhdr;
107 ents = ops->leaf_ents_p(leaf);
108 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
111 * XXX (dgc): This value is not restrictive enough.
112 * Should factor in the size of the bests table as well.
113 * We can deduce a value for that from di_size.
115 if (hdr->count > ops->leaf_max_ents(mp))
116 return false;
118 /* Leaves and bests don't overlap in leaf format. */
119 if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
120 hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
121 (char *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
122 return false;
124 /* Check hash value order, count stale entries. */
125 for (i = stale = 0; i < hdr->count; i++) {
126 if (i + 1 < hdr->count) {
127 if (be32_to_cpu(ents[i].hashval) >
128 be32_to_cpu(ents[i + 1].hashval))
129 return false;
131 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
132 stale++;
134 if (hdr->stale != stale)
135 return false;
136 return true;
140 * We verify the magic numbers before decoding the leaf header so that on debug
141 * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
142 * to incorrect magic numbers.
144 static bool
145 xfs_dir3_leaf_verify(
146 struct xfs_buf *bp,
147 __uint16_t magic)
149 struct xfs_mount *mp = bp->b_target->bt_mount;
150 struct xfs_dir2_leaf *leaf = bp->b_addr;
152 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
154 if (xfs_sb_version_hascrc(&mp->m_sb)) {
155 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
156 __uint16_t magic3;
158 magic3 = (magic == XFS_DIR2_LEAF1_MAGIC) ? XFS_DIR3_LEAF1_MAGIC
159 : XFS_DIR3_LEAFN_MAGIC;
161 if (leaf3->info.hdr.magic != cpu_to_be16(magic3))
162 return false;
163 if (!uuid_equal(&leaf3->info.uuid, &mp->m_sb.sb_uuid))
164 return false;
165 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
166 return false;
167 } else {
168 if (leaf->hdr.info.magic != cpu_to_be16(magic))
169 return false;
172 return xfs_dir3_leaf_check_int(mp, NULL, NULL, leaf);
175 static void
176 __read_verify(
177 struct xfs_buf *bp,
178 __uint16_t magic)
180 struct xfs_mount *mp = bp->b_target->bt_mount;
182 if ((xfs_sb_version_hascrc(&mp->m_sb) &&
183 !xfs_verify_cksum(bp->b_addr, BBTOB(bp->b_length),
184 XFS_DIR3_LEAF_CRC_OFF)) ||
185 !xfs_dir3_leaf_verify(bp, magic)) {
186 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
187 xfs_buf_ioerror(bp, EFSCORRUPTED);
191 static void
192 __write_verify(
193 struct xfs_buf *bp,
194 __uint16_t magic)
196 struct xfs_mount *mp = bp->b_target->bt_mount;
197 struct xfs_buf_log_item *bip = bp->b_fspriv;
198 struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
200 if (!xfs_dir3_leaf_verify(bp, magic)) {
201 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, bp->b_addr);
202 xfs_buf_ioerror(bp, EFSCORRUPTED);
203 return;
206 if (!xfs_sb_version_hascrc(&mp->m_sb))
207 return;
209 if (bip)
210 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
212 xfs_update_cksum(bp->b_addr, BBTOB(bp->b_length), XFS_DIR3_LEAF_CRC_OFF);
215 static void
216 xfs_dir3_leaf1_read_verify(
217 struct xfs_buf *bp)
219 __read_verify(bp, XFS_DIR2_LEAF1_MAGIC);
222 static void
223 xfs_dir3_leaf1_write_verify(
224 struct xfs_buf *bp)
226 __write_verify(bp, XFS_DIR2_LEAF1_MAGIC);
229 static void
230 xfs_dir3_leafn_read_verify(
231 struct xfs_buf *bp)
233 __read_verify(bp, XFS_DIR2_LEAFN_MAGIC);
236 static void
237 xfs_dir3_leafn_write_verify(
238 struct xfs_buf *bp)
240 __write_verify(bp, XFS_DIR2_LEAFN_MAGIC);
243 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
244 .verify_read = xfs_dir3_leaf1_read_verify,
245 .verify_write = xfs_dir3_leaf1_write_verify,
248 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
249 .verify_read = xfs_dir3_leafn_read_verify,
250 .verify_write = xfs_dir3_leafn_write_verify,
253 static int
254 xfs_dir3_leaf_read(
255 struct xfs_trans *tp,
256 struct xfs_inode *dp,
257 xfs_dablk_t fbno,
258 xfs_daddr_t mappedbno,
259 struct xfs_buf **bpp)
261 int err;
263 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
264 XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
265 if (!err && tp)
266 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
267 return err;
271 xfs_dir3_leafn_read(
272 struct xfs_trans *tp,
273 struct xfs_inode *dp,
274 xfs_dablk_t fbno,
275 xfs_daddr_t mappedbno,
276 struct xfs_buf **bpp)
278 int err;
280 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
281 XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
282 if (!err && tp)
283 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
284 return err;
288 * Initialize a new leaf block, leaf1 or leafn magic accepted.
290 static void
291 xfs_dir3_leaf_init(
292 struct xfs_mount *mp,
293 struct xfs_trans *tp,
294 struct xfs_buf *bp,
295 xfs_ino_t owner,
296 __uint16_t type)
298 struct xfs_dir2_leaf *leaf = bp->b_addr;
300 ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
302 if (xfs_sb_version_hascrc(&mp->m_sb)) {
303 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
305 memset(leaf3, 0, sizeof(*leaf3));
307 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
308 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
309 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
310 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
311 leaf3->info.owner = cpu_to_be64(owner);
312 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_uuid);
313 } else {
314 memset(leaf, 0, sizeof(*leaf));
315 leaf->hdr.info.magic = cpu_to_be16(type);
319 * If it's a leaf-format directory initialize the tail.
320 * Caller is responsible for initialising the bests table.
322 if (type == XFS_DIR2_LEAF1_MAGIC) {
323 struct xfs_dir2_leaf_tail *ltp;
325 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
326 ltp->bestcount = 0;
327 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
328 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
329 } else {
330 bp->b_ops = &xfs_dir3_leafn_buf_ops;
331 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
336 xfs_dir3_leaf_get_buf(
337 xfs_da_args_t *args,
338 xfs_dir2_db_t bno,
339 struct xfs_buf **bpp,
340 __uint16_t magic)
342 struct xfs_inode *dp = args->dp;
343 struct xfs_trans *tp = args->trans;
344 struct xfs_mount *mp = dp->i_mount;
345 struct xfs_buf *bp;
346 int error;
348 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
349 ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
350 bno < XFS_DIR2_FREE_FIRSTDB(mp));
352 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
353 XFS_DATA_FORK);
354 if (error)
355 return error;
357 xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
358 xfs_dir3_leaf_log_header(tp, dp, bp);
359 if (magic == XFS_DIR2_LEAF1_MAGIC)
360 xfs_dir3_leaf_log_tail(tp, bp);
361 *bpp = bp;
362 return 0;
366 * Convert a block form directory to a leaf form directory.
368 int /* error */
369 xfs_dir2_block_to_leaf(
370 xfs_da_args_t *args, /* operation arguments */
371 struct xfs_buf *dbp) /* input block's buffer */
373 __be16 *bestsp; /* leaf's bestsp entries */
374 xfs_dablk_t blkno; /* leaf block's bno */
375 xfs_dir2_data_hdr_t *hdr; /* block header */
376 xfs_dir2_leaf_entry_t *blp; /* block's leaf entries */
377 xfs_dir2_block_tail_t *btp; /* block's tail */
378 xfs_inode_t *dp; /* incore directory inode */
379 int error; /* error return code */
380 struct xfs_buf *lbp; /* leaf block's buffer */
381 xfs_dir2_db_t ldb; /* leaf block's bno */
382 xfs_dir2_leaf_t *leaf; /* leaf structure */
383 xfs_dir2_leaf_tail_t *ltp; /* leaf's tail */
384 xfs_mount_t *mp; /* filesystem mount point */
385 int needlog; /* need to log block header */
386 int needscan; /* need to rescan bestfree */
387 xfs_trans_t *tp; /* transaction pointer */
388 struct xfs_dir2_data_free *bf;
389 struct xfs_dir2_leaf_entry *ents;
390 struct xfs_dir3_icleaf_hdr leafhdr;
392 trace_xfs_dir2_block_to_leaf(args);
394 dp = args->dp;
395 mp = dp->i_mount;
396 tp = args->trans;
398 * Add the leaf block to the inode.
399 * This interface will only put blocks in the leaf/node range.
400 * Since that's empty now, we'll get the root (block 0 in range).
402 if ((error = xfs_da_grow_inode(args, &blkno))) {
403 return error;
405 ldb = xfs_dir2_da_to_db(mp, blkno);
406 ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
408 * Initialize the leaf block, get a buffer for it.
410 error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
411 if (error)
412 return error;
414 leaf = lbp->b_addr;
415 hdr = dbp->b_addr;
416 xfs_dir3_data_check(dp, dbp);
417 btp = xfs_dir2_block_tail_p(mp, hdr);
418 blp = xfs_dir2_block_leaf_p(btp);
419 bf = dp->d_ops->data_bestfree_p(hdr);
420 ents = dp->d_ops->leaf_ents_p(leaf);
423 * Set the counts in the leaf header.
425 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
426 leafhdr.count = be32_to_cpu(btp->count);
427 leafhdr.stale = be32_to_cpu(btp->stale);
428 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
429 xfs_dir3_leaf_log_header(tp, dp, lbp);
432 * Could compact these but I think we always do the conversion
433 * after squeezing out stale entries.
435 memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
436 xfs_dir3_leaf_log_ents(tp, dp, lbp, 0, leafhdr.count - 1);
437 needscan = 0;
438 needlog = 1;
440 * Make the space formerly occupied by the leaf entries and block
441 * tail be free.
443 xfs_dir2_data_make_free(tp, dp, dbp,
444 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
445 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
446 (char *)blp),
447 &needlog, &needscan);
449 * Fix up the block header, make it a data block.
451 dbp->b_ops = &xfs_dir3_data_buf_ops;
452 xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
453 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
454 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
455 else
456 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
458 if (needscan)
459 xfs_dir2_data_freescan(dp, hdr, &needlog);
461 * Set up leaf tail and bests table.
463 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
464 ltp->bestcount = cpu_to_be32(1);
465 bestsp = xfs_dir2_leaf_bests_p(ltp);
466 bestsp[0] = bf[0].length;
468 * Log the data header and leaf bests table.
470 if (needlog)
471 xfs_dir2_data_log_header(tp, dp, dbp);
472 xfs_dir3_leaf_check(dp, lbp);
473 xfs_dir3_data_check(dp, dbp);
474 xfs_dir3_leaf_log_bests(tp, lbp, 0, 0);
475 return 0;
478 STATIC void
479 xfs_dir3_leaf_find_stale(
480 struct xfs_dir3_icleaf_hdr *leafhdr,
481 struct xfs_dir2_leaf_entry *ents,
482 int index,
483 int *lowstale,
484 int *highstale)
487 * Find the first stale entry before our index, if any.
489 for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
490 if (ents[*lowstale].address ==
491 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
492 break;
496 * Find the first stale entry at or after our index, if any.
497 * Stop if the result would require moving more entries than using
498 * lowstale.
500 for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
501 if (ents[*highstale].address ==
502 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
503 break;
504 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
505 break;
509 struct xfs_dir2_leaf_entry *
510 xfs_dir3_leaf_find_entry(
511 struct xfs_dir3_icleaf_hdr *leafhdr,
512 struct xfs_dir2_leaf_entry *ents,
513 int index, /* leaf table position */
514 int compact, /* need to compact leaves */
515 int lowstale, /* index of prev stale leaf */
516 int highstale, /* index of next stale leaf */
517 int *lfloglow, /* low leaf logging index */
518 int *lfloghigh) /* high leaf logging index */
520 if (!leafhdr->stale) {
521 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
524 * Now we need to make room to insert the leaf entry.
526 * If there are no stale entries, just insert a hole at index.
528 lep = &ents[index];
529 if (index < leafhdr->count)
530 memmove(lep + 1, lep,
531 (leafhdr->count - index) * sizeof(*lep));
534 * Record low and high logging indices for the leaf.
536 *lfloglow = index;
537 *lfloghigh = leafhdr->count++;
538 return lep;
542 * There are stale entries.
544 * We will use one of them for the new entry. It's probably not at
545 * the right location, so we'll have to shift some up or down first.
547 * If we didn't compact before, we need to find the nearest stale
548 * entries before and after our insertion point.
550 if (compact == 0)
551 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
552 &lowstale, &highstale);
555 * If the low one is better, use it.
557 if (lowstale >= 0 &&
558 (highstale == leafhdr->count ||
559 index - lowstale - 1 < highstale - index)) {
560 ASSERT(index - lowstale - 1 >= 0);
561 ASSERT(ents[lowstale].address ==
562 cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
565 * Copy entries up to cover the stale entry and make room
566 * for the new entry.
568 if (index - lowstale - 1 > 0) {
569 memmove(&ents[lowstale], &ents[lowstale + 1],
570 (index - lowstale - 1) *
571 sizeof(xfs_dir2_leaf_entry_t));
573 *lfloglow = MIN(lowstale, *lfloglow);
574 *lfloghigh = MAX(index - 1, *lfloghigh);
575 leafhdr->stale--;
576 return &ents[index - 1];
580 * The high one is better, so use that one.
582 ASSERT(highstale - index >= 0);
583 ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
586 * Copy entries down to cover the stale entry and make room for the
587 * new entry.
589 if (highstale - index > 0) {
590 memmove(&ents[index + 1], &ents[index],
591 (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
593 *lfloglow = MIN(index, *lfloglow);
594 *lfloghigh = MAX(highstale, *lfloghigh);
595 leafhdr->stale--;
596 return &ents[index];
600 * Add an entry to a leaf form directory.
602 int /* error */
603 xfs_dir2_leaf_addname(
604 xfs_da_args_t *args) /* operation arguments */
606 __be16 *bestsp; /* freespace table in leaf */
607 int compact; /* need to compact leaves */
608 xfs_dir2_data_hdr_t *hdr; /* data block header */
609 struct xfs_buf *dbp; /* data block buffer */
610 xfs_dir2_data_entry_t *dep; /* data block entry */
611 xfs_inode_t *dp; /* incore directory inode */
612 xfs_dir2_data_unused_t *dup; /* data unused entry */
613 int error; /* error return value */
614 int grown; /* allocated new data block */
615 int highstale; /* index of next stale leaf */
616 int i; /* temporary, index */
617 int index; /* leaf table position */
618 struct xfs_buf *lbp; /* leaf's buffer */
619 xfs_dir2_leaf_t *leaf; /* leaf structure */
620 int length; /* length of new entry */
621 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
622 int lfloglow; /* low leaf logging index */
623 int lfloghigh; /* high leaf logging index */
624 int lowstale; /* index of prev stale leaf */
625 xfs_dir2_leaf_tail_t *ltp; /* leaf tail pointer */
626 xfs_mount_t *mp; /* filesystem mount point */
627 int needbytes; /* leaf block bytes needed */
628 int needlog; /* need to log data header */
629 int needscan; /* need to rescan data free */
630 __be16 *tagp; /* end of data entry */
631 xfs_trans_t *tp; /* transaction pointer */
632 xfs_dir2_db_t use_block; /* data block number */
633 struct xfs_dir2_data_free *bf; /* bestfree table */
634 struct xfs_dir2_leaf_entry *ents;
635 struct xfs_dir3_icleaf_hdr leafhdr;
637 trace_xfs_dir2_leaf_addname(args);
639 dp = args->dp;
640 tp = args->trans;
641 mp = dp->i_mount;
643 error = xfs_dir3_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
644 if (error)
645 return error;
648 * Look up the entry by hash value and name.
649 * We know it's not there, our caller has already done a lookup.
650 * So the index is of the entry to insert in front of.
651 * But if there are dup hash values the index is of the first of those.
653 index = xfs_dir2_leaf_search_hash(args, lbp);
654 leaf = lbp->b_addr;
655 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
656 ents = dp->d_ops->leaf_ents_p(leaf);
657 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
658 bestsp = xfs_dir2_leaf_bests_p(ltp);
659 length = dp->d_ops->data_entsize(args->namelen);
662 * See if there are any entries with the same hash value
663 * and space in their block for the new entry.
664 * This is good because it puts multiple same-hash value entries
665 * in a data block, improving the lookup of those entries.
667 for (use_block = -1, lep = &ents[index];
668 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
669 index++, lep++) {
670 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
671 continue;
672 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
673 ASSERT(i < be32_to_cpu(ltp->bestcount));
674 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
675 if (be16_to_cpu(bestsp[i]) >= length) {
676 use_block = i;
677 break;
681 * Didn't find a block yet, linear search all the data blocks.
683 if (use_block == -1) {
684 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
686 * Remember a block we see that's missing.
688 if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
689 use_block == -1)
690 use_block = i;
691 else if (be16_to_cpu(bestsp[i]) >= length) {
692 use_block = i;
693 break;
698 * How many bytes do we need in the leaf block?
700 needbytes = 0;
701 if (!leafhdr.stale)
702 needbytes += sizeof(xfs_dir2_leaf_entry_t);
703 if (use_block == -1)
704 needbytes += sizeof(xfs_dir2_data_off_t);
707 * Now kill use_block if it refers to a missing block, so we
708 * can use it as an indication of allocation needed.
710 if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
711 use_block = -1;
713 * If we don't have enough free bytes but we can make enough
714 * by compacting out stale entries, we'll do that.
716 if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
717 leafhdr.stale > 1)
718 compact = 1;
721 * Otherwise if we don't have enough free bytes we need to
722 * convert to node form.
724 else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
726 * Just checking or no space reservation, give up.
728 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
729 args->total == 0) {
730 xfs_trans_brelse(tp, lbp);
731 return XFS_ERROR(ENOSPC);
734 * Convert to node form.
736 error = xfs_dir2_leaf_to_node(args, lbp);
737 if (error)
738 return error;
740 * Then add the new entry.
742 return xfs_dir2_node_addname(args);
745 * Otherwise it will fit without compaction.
747 else
748 compact = 0;
750 * If just checking, then it will fit unless we needed to allocate
751 * a new data block.
753 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
754 xfs_trans_brelse(tp, lbp);
755 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
758 * If no allocations are allowed, return now before we've
759 * changed anything.
761 if (args->total == 0 && use_block == -1) {
762 xfs_trans_brelse(tp, lbp);
763 return XFS_ERROR(ENOSPC);
766 * Need to compact the leaf entries, removing stale ones.
767 * Leave one stale entry behind - the one closest to our
768 * insertion index - and we'll shift that one to our insertion
769 * point later.
771 if (compact) {
772 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
773 &highstale, &lfloglow, &lfloghigh);
776 * There are stale entries, so we'll need log-low and log-high
777 * impossibly bad values later.
779 else if (leafhdr.stale) {
780 lfloglow = leafhdr.count;
781 lfloghigh = -1;
784 * If there was no data block space found, we need to allocate
785 * a new one.
787 if (use_block == -1) {
789 * Add the new data block.
791 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
792 &use_block))) {
793 xfs_trans_brelse(tp, lbp);
794 return error;
797 * Initialize the block.
799 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
800 xfs_trans_brelse(tp, lbp);
801 return error;
804 * If we're adding a new data block on the end we need to
805 * extend the bests table. Copy it up one entry.
807 if (use_block >= be32_to_cpu(ltp->bestcount)) {
808 bestsp--;
809 memmove(&bestsp[0], &bestsp[1],
810 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
811 be32_add_cpu(&ltp->bestcount, 1);
812 xfs_dir3_leaf_log_tail(tp, lbp);
813 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
816 * If we're filling in a previously empty block just log it.
818 else
819 xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
820 hdr = dbp->b_addr;
821 bf = dp->d_ops->data_bestfree_p(hdr);
822 bestsp[use_block] = bf[0].length;
823 grown = 1;
824 } else {
826 * Already had space in some data block.
827 * Just read that one in.
829 error = xfs_dir3_data_read(tp, dp,
830 xfs_dir2_db_to_da(mp, use_block),
831 -1, &dbp);
832 if (error) {
833 xfs_trans_brelse(tp, lbp);
834 return error;
836 hdr = dbp->b_addr;
837 bf = dp->d_ops->data_bestfree_p(hdr);
838 grown = 0;
841 * Point to the biggest freespace in our data block.
843 dup = (xfs_dir2_data_unused_t *)
844 ((char *)hdr + be16_to_cpu(bf[0].offset));
845 ASSERT(be16_to_cpu(dup->length) >= length);
846 needscan = needlog = 0;
848 * Mark the initial part of our freespace in use for the new entry.
850 xfs_dir2_data_use_free(tp, dp, dbp, dup,
851 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
852 &needlog, &needscan);
854 * Initialize our new entry (at last).
856 dep = (xfs_dir2_data_entry_t *)dup;
857 dep->inumber = cpu_to_be64(args->inumber);
858 dep->namelen = args->namelen;
859 memcpy(dep->name, args->name, dep->namelen);
860 dp->d_ops->data_put_ftype(dep, args->filetype);
861 tagp = dp->d_ops->data_entry_tag_p(dep);
862 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
864 * Need to scan fix up the bestfree table.
866 if (needscan)
867 xfs_dir2_data_freescan(dp, hdr, &needlog);
869 * Need to log the data block's header.
871 if (needlog)
872 xfs_dir2_data_log_header(tp, dp, dbp);
873 xfs_dir2_data_log_entry(tp, dp, dbp, dep);
875 * If the bests table needs to be changed, do it.
876 * Log the change unless we've already done that.
878 if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
879 bestsp[use_block] = bf[0].length;
880 if (!grown)
881 xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
884 lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
885 highstale, &lfloglow, &lfloghigh);
888 * Fill in the new leaf entry.
890 lep->hashval = cpu_to_be32(args->hashval);
891 lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
892 be16_to_cpu(*tagp)));
894 * Log the leaf fields and give up the buffers.
896 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
897 xfs_dir3_leaf_log_header(tp, dp, lbp);
898 xfs_dir3_leaf_log_ents(tp, dp, lbp, lfloglow, lfloghigh);
899 xfs_dir3_leaf_check(dp, lbp);
900 xfs_dir3_data_check(dp, dbp);
901 return 0;
905 * Compact out any stale entries in the leaf.
906 * Log the header and changed leaf entries, if any.
908 void
909 xfs_dir3_leaf_compact(
910 xfs_da_args_t *args, /* operation arguments */
911 struct xfs_dir3_icleaf_hdr *leafhdr,
912 struct xfs_buf *bp) /* leaf buffer */
914 int from; /* source leaf index */
915 xfs_dir2_leaf_t *leaf; /* leaf structure */
916 int loglow; /* first leaf entry to log */
917 int to; /* target leaf index */
918 struct xfs_dir2_leaf_entry *ents;
919 struct xfs_inode *dp = args->dp;
921 leaf = bp->b_addr;
922 if (!leafhdr->stale)
923 return;
926 * Compress out the stale entries in place.
928 ents = dp->d_ops->leaf_ents_p(leaf);
929 for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
930 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
931 continue;
933 * Only actually copy the entries that are different.
935 if (from > to) {
936 if (loglow == -1)
937 loglow = to;
938 ents[to] = ents[from];
940 to++;
943 * Update and log the header, log the leaf entries.
945 ASSERT(leafhdr->stale == from - to);
946 leafhdr->count -= leafhdr->stale;
947 leafhdr->stale = 0;
949 dp->d_ops->leaf_hdr_to_disk(leaf, leafhdr);
950 xfs_dir3_leaf_log_header(args->trans, dp, bp);
951 if (loglow != -1)
952 xfs_dir3_leaf_log_ents(args->trans, dp, bp, loglow, to - 1);
956 * Compact the leaf entries, removing stale ones.
957 * Leave one stale entry behind - the one closest to our
958 * insertion index - and the caller will shift that one to our insertion
959 * point later.
960 * Return new insertion index, where the remaining stale entry is,
961 * and leaf logging indices.
963 void
964 xfs_dir3_leaf_compact_x1(
965 struct xfs_dir3_icleaf_hdr *leafhdr,
966 struct xfs_dir2_leaf_entry *ents,
967 int *indexp, /* insertion index */
968 int *lowstalep, /* out: stale entry before us */
969 int *highstalep, /* out: stale entry after us */
970 int *lowlogp, /* out: low log index */
971 int *highlogp) /* out: high log index */
973 int from; /* source copy index */
974 int highstale; /* stale entry at/after index */
975 int index; /* insertion index */
976 int keepstale; /* source index of kept stale */
977 int lowstale; /* stale entry before index */
978 int newindex=0; /* new insertion index */
979 int to; /* destination copy index */
981 ASSERT(leafhdr->stale > 1);
982 index = *indexp;
984 xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
987 * Pick the better of lowstale and highstale.
989 if (lowstale >= 0 &&
990 (highstale == leafhdr->count ||
991 index - lowstale <= highstale - index))
992 keepstale = lowstale;
993 else
994 keepstale = highstale;
996 * Copy the entries in place, removing all the stale entries
997 * except keepstale.
999 for (from = to = 0; from < leafhdr->count; from++) {
1001 * Notice the new value of index.
1003 if (index == from)
1004 newindex = to;
1005 if (from != keepstale &&
1006 ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1007 if (from == to)
1008 *lowlogp = to;
1009 continue;
1012 * Record the new keepstale value for the insertion.
1014 if (from == keepstale)
1015 lowstale = highstale = to;
1017 * Copy only the entries that have moved.
1019 if (from > to)
1020 ents[to] = ents[from];
1021 to++;
1023 ASSERT(from > to);
1025 * If the insertion point was past the last entry,
1026 * set the new insertion point accordingly.
1028 if (index == from)
1029 newindex = to;
1030 *indexp = newindex;
1032 * Adjust the leaf header values.
1034 leafhdr->count -= from - to;
1035 leafhdr->stale = 1;
1037 * Remember the low/high stale value only in the "right"
1038 * direction.
1040 if (lowstale >= newindex)
1041 lowstale = -1;
1042 else
1043 highstale = leafhdr->count;
1044 *highlogp = leafhdr->count - 1;
1045 *lowstalep = lowstale;
1046 *highstalep = highstale;
1050 * Log the bests entries indicated from a leaf1 block.
1052 static void
1053 xfs_dir3_leaf_log_bests(
1054 xfs_trans_t *tp, /* transaction pointer */
1055 struct xfs_buf *bp, /* leaf buffer */
1056 int first, /* first entry to log */
1057 int last) /* last entry to log */
1059 __be16 *firstb; /* pointer to first entry */
1060 __be16 *lastb; /* pointer to last entry */
1061 struct xfs_dir2_leaf *leaf = bp->b_addr;
1062 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1064 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1065 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1067 ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1068 firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1069 lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1070 xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1071 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1075 * Log the leaf entries indicated from a leaf1 or leafn block.
1077 void
1078 xfs_dir3_leaf_log_ents(
1079 struct xfs_trans *tp,
1080 struct xfs_inode *dp,
1081 struct xfs_buf *bp,
1082 int first,
1083 int last)
1085 xfs_dir2_leaf_entry_t *firstlep; /* pointer to first entry */
1086 xfs_dir2_leaf_entry_t *lastlep; /* pointer to last entry */
1087 struct xfs_dir2_leaf *leaf = bp->b_addr;
1088 struct xfs_dir2_leaf_entry *ents;
1090 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1091 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1092 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1093 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1095 ents = dp->d_ops->leaf_ents_p(leaf);
1096 firstlep = &ents[first];
1097 lastlep = &ents[last];
1098 xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1099 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1103 * Log the header of the leaf1 or leafn block.
1105 void
1106 xfs_dir3_leaf_log_header(
1107 struct xfs_trans *tp,
1108 struct xfs_inode *dp,
1109 struct xfs_buf *bp)
1111 struct xfs_dir2_leaf *leaf = bp->b_addr;
1113 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1114 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1115 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1116 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1118 xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1119 dp->d_ops->leaf_hdr_size - 1);
1123 * Log the tail of the leaf1 block.
1125 STATIC void
1126 xfs_dir3_leaf_log_tail(
1127 struct xfs_trans *tp,
1128 struct xfs_buf *bp)
1130 struct xfs_dir2_leaf *leaf = bp->b_addr;
1131 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1132 struct xfs_mount *mp = tp->t_mountp;
1134 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1135 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1136 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1137 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1139 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1140 xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1141 (uint)(mp->m_dirblksize - 1));
1145 * Look up the entry referred to by args in the leaf format directory.
1146 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1147 * is also used by the node-format code.
1150 xfs_dir2_leaf_lookup(
1151 xfs_da_args_t *args) /* operation arguments */
1153 struct xfs_buf *dbp; /* data block buffer */
1154 xfs_dir2_data_entry_t *dep; /* data block entry */
1155 xfs_inode_t *dp; /* incore directory inode */
1156 int error; /* error return code */
1157 int index; /* found entry index */
1158 struct xfs_buf *lbp; /* leaf buffer */
1159 xfs_dir2_leaf_t *leaf; /* leaf structure */
1160 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1161 xfs_trans_t *tp; /* transaction pointer */
1162 struct xfs_dir2_leaf_entry *ents;
1164 trace_xfs_dir2_leaf_lookup(args);
1167 * Look up name in the leaf block, returning both buffers and index.
1169 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1170 return error;
1172 tp = args->trans;
1173 dp = args->dp;
1174 xfs_dir3_leaf_check(dp, lbp);
1175 leaf = lbp->b_addr;
1176 ents = dp->d_ops->leaf_ents_p(leaf);
1178 * Get to the leaf entry and contained data entry address.
1180 lep = &ents[index];
1183 * Point to the data entry.
1185 dep = (xfs_dir2_data_entry_t *)
1186 ((char *)dbp->b_addr +
1187 xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1189 * Return the found inode number & CI name if appropriate
1191 args->inumber = be64_to_cpu(dep->inumber);
1192 args->filetype = dp->d_ops->data_get_ftype(dep);
1193 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1194 xfs_trans_brelse(tp, dbp);
1195 xfs_trans_brelse(tp, lbp);
1196 return XFS_ERROR(error);
1200 * Look up name/hash in the leaf block.
1201 * Fill in indexp with the found index, and dbpp with the data buffer.
1202 * If not found dbpp will be NULL, and ENOENT comes back.
1203 * lbpp will always be filled in with the leaf buffer unless there's an error.
1205 static int /* error */
1206 xfs_dir2_leaf_lookup_int(
1207 xfs_da_args_t *args, /* operation arguments */
1208 struct xfs_buf **lbpp, /* out: leaf buffer */
1209 int *indexp, /* out: index in leaf block */
1210 struct xfs_buf **dbpp) /* out: data buffer */
1212 xfs_dir2_db_t curdb = -1; /* current data block number */
1213 struct xfs_buf *dbp = NULL; /* data buffer */
1214 xfs_dir2_data_entry_t *dep; /* data entry */
1215 xfs_inode_t *dp; /* incore directory inode */
1216 int error; /* error return code */
1217 int index; /* index in leaf block */
1218 struct xfs_buf *lbp; /* leaf buffer */
1219 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1220 xfs_dir2_leaf_t *leaf; /* leaf structure */
1221 xfs_mount_t *mp; /* filesystem mount point */
1222 xfs_dir2_db_t newdb; /* new data block number */
1223 xfs_trans_t *tp; /* transaction pointer */
1224 xfs_dir2_db_t cidb = -1; /* case match data block no. */
1225 enum xfs_dacmp cmp; /* name compare result */
1226 struct xfs_dir2_leaf_entry *ents;
1227 struct xfs_dir3_icleaf_hdr leafhdr;
1229 dp = args->dp;
1230 tp = args->trans;
1231 mp = dp->i_mount;
1233 error = xfs_dir3_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
1234 if (error)
1235 return error;
1237 *lbpp = lbp;
1238 leaf = lbp->b_addr;
1239 xfs_dir3_leaf_check(dp, lbp);
1240 ents = dp->d_ops->leaf_ents_p(leaf);
1241 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1244 * Look for the first leaf entry with our hash value.
1246 index = xfs_dir2_leaf_search_hash(args, lbp);
1248 * Loop over all the entries with the right hash value
1249 * looking to match the name.
1251 for (lep = &ents[index];
1252 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1253 lep++, index++) {
1255 * Skip over stale leaf entries.
1257 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1258 continue;
1260 * Get the new data block number.
1262 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1264 * If it's not the same as the old data block number,
1265 * need to pitch the old one and read the new one.
1267 if (newdb != curdb) {
1268 if (dbp)
1269 xfs_trans_brelse(tp, dbp);
1270 error = xfs_dir3_data_read(tp, dp,
1271 xfs_dir2_db_to_da(mp, newdb),
1272 -1, &dbp);
1273 if (error) {
1274 xfs_trans_brelse(tp, lbp);
1275 return error;
1277 curdb = newdb;
1280 * Point to the data entry.
1282 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1283 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1285 * Compare name and if it's an exact match, return the index
1286 * and buffer. If it's the first case-insensitive match, store
1287 * the index and buffer and continue looking for an exact match.
1289 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1290 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1291 args->cmpresult = cmp;
1292 *indexp = index;
1293 /* case exact match: return the current buffer. */
1294 if (cmp == XFS_CMP_EXACT) {
1295 *dbpp = dbp;
1296 return 0;
1298 cidb = curdb;
1301 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1303 * Here, we can only be doing a lookup (not a rename or remove).
1304 * If a case-insensitive match was found earlier, re-read the
1305 * appropriate data block if required and return it.
1307 if (args->cmpresult == XFS_CMP_CASE) {
1308 ASSERT(cidb != -1);
1309 if (cidb != curdb) {
1310 xfs_trans_brelse(tp, dbp);
1311 error = xfs_dir3_data_read(tp, dp,
1312 xfs_dir2_db_to_da(mp, cidb),
1313 -1, &dbp);
1314 if (error) {
1315 xfs_trans_brelse(tp, lbp);
1316 return error;
1319 *dbpp = dbp;
1320 return 0;
1323 * No match found, return ENOENT.
1325 ASSERT(cidb == -1);
1326 if (dbp)
1327 xfs_trans_brelse(tp, dbp);
1328 xfs_trans_brelse(tp, lbp);
1329 return XFS_ERROR(ENOENT);
1333 * Remove an entry from a leaf format directory.
1335 int /* error */
1336 xfs_dir2_leaf_removename(
1337 xfs_da_args_t *args) /* operation arguments */
1339 __be16 *bestsp; /* leaf block best freespace */
1340 xfs_dir2_data_hdr_t *hdr; /* data block header */
1341 xfs_dir2_db_t db; /* data block number */
1342 struct xfs_buf *dbp; /* data block buffer */
1343 xfs_dir2_data_entry_t *dep; /* data entry structure */
1344 xfs_inode_t *dp; /* incore directory inode */
1345 int error; /* error return code */
1346 xfs_dir2_db_t i; /* temporary data block # */
1347 int index; /* index into leaf entries */
1348 struct xfs_buf *lbp; /* leaf buffer */
1349 xfs_dir2_leaf_t *leaf; /* leaf structure */
1350 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1351 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1352 xfs_mount_t *mp; /* filesystem mount point */
1353 int needlog; /* need to log data header */
1354 int needscan; /* need to rescan data frees */
1355 xfs_dir2_data_off_t oldbest; /* old value of best free */
1356 xfs_trans_t *tp; /* transaction pointer */
1357 struct xfs_dir2_data_free *bf; /* bestfree table */
1358 struct xfs_dir2_leaf_entry *ents;
1359 struct xfs_dir3_icleaf_hdr leafhdr;
1361 trace_xfs_dir2_leaf_removename(args);
1364 * Lookup the leaf entry, get the leaf and data blocks read in.
1366 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1367 return error;
1369 dp = args->dp;
1370 tp = args->trans;
1371 mp = dp->i_mount;
1372 leaf = lbp->b_addr;
1373 hdr = dbp->b_addr;
1374 xfs_dir3_data_check(dp, dbp);
1375 bf = dp->d_ops->data_bestfree_p(hdr);
1376 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1377 ents = dp->d_ops->leaf_ents_p(leaf);
1379 * Point to the leaf entry, use that to point to the data entry.
1381 lep = &ents[index];
1382 db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1383 dep = (xfs_dir2_data_entry_t *)
1384 ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1385 needscan = needlog = 0;
1386 oldbest = be16_to_cpu(bf[0].length);
1387 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1388 bestsp = xfs_dir2_leaf_bests_p(ltp);
1389 ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1391 * Mark the former data entry unused.
1393 xfs_dir2_data_make_free(tp, dp, dbp,
1394 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1395 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1397 * We just mark the leaf entry stale by putting a null in it.
1399 leafhdr.stale++;
1400 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1401 xfs_dir3_leaf_log_header(tp, dp, lbp);
1403 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1404 xfs_dir3_leaf_log_ents(tp, dp, lbp, index, index);
1407 * Scan the freespace in the data block again if necessary,
1408 * log the data block header if necessary.
1410 if (needscan)
1411 xfs_dir2_data_freescan(dp, hdr, &needlog);
1412 if (needlog)
1413 xfs_dir2_data_log_header(tp, dp, dbp);
1415 * If the longest freespace in the data block has changed,
1416 * put the new value in the bests table and log that.
1418 if (be16_to_cpu(bf[0].length) != oldbest) {
1419 bestsp[db] = bf[0].length;
1420 xfs_dir3_leaf_log_bests(tp, lbp, db, db);
1422 xfs_dir3_data_check(dp, dbp);
1424 * If the data block is now empty then get rid of the data block.
1426 if (be16_to_cpu(bf[0].length) ==
1427 mp->m_dirblksize - dp->d_ops->data_entry_offset) {
1428 ASSERT(db != mp->m_dirdatablk);
1429 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1431 * Nope, can't get rid of it because it caused
1432 * allocation of a bmap btree block to do so.
1433 * Just go on, returning success, leaving the
1434 * empty block in place.
1436 if (error == ENOSPC && args->total == 0)
1437 error = 0;
1438 xfs_dir3_leaf_check(dp, lbp);
1439 return error;
1441 dbp = NULL;
1443 * If this is the last data block then compact the
1444 * bests table by getting rid of entries.
1446 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1448 * Look for the last active entry (i).
1450 for (i = db - 1; i > 0; i--) {
1451 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1452 break;
1455 * Copy the table down so inactive entries at the
1456 * end are removed.
1458 memmove(&bestsp[db - i], bestsp,
1459 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1460 be32_add_cpu(&ltp->bestcount, -(db - i));
1461 xfs_dir3_leaf_log_tail(tp, lbp);
1462 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1463 } else
1464 bestsp[db] = cpu_to_be16(NULLDATAOFF);
1467 * If the data block was not the first one, drop it.
1469 else if (db != mp->m_dirdatablk)
1470 dbp = NULL;
1472 xfs_dir3_leaf_check(dp, lbp);
1474 * See if we can convert to block form.
1476 return xfs_dir2_leaf_to_block(args, lbp, dbp);
1480 * Replace the inode number in a leaf format directory entry.
1482 int /* error */
1483 xfs_dir2_leaf_replace(
1484 xfs_da_args_t *args) /* operation arguments */
1486 struct xfs_buf *dbp; /* data block buffer */
1487 xfs_dir2_data_entry_t *dep; /* data block entry */
1488 xfs_inode_t *dp; /* incore directory inode */
1489 int error; /* error return code */
1490 int index; /* index of leaf entry */
1491 struct xfs_buf *lbp; /* leaf buffer */
1492 xfs_dir2_leaf_t *leaf; /* leaf structure */
1493 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1494 xfs_trans_t *tp; /* transaction pointer */
1495 struct xfs_dir2_leaf_entry *ents;
1497 trace_xfs_dir2_leaf_replace(args);
1500 * Look up the entry.
1502 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1503 return error;
1505 dp = args->dp;
1506 leaf = lbp->b_addr;
1507 ents = dp->d_ops->leaf_ents_p(leaf);
1509 * Point to the leaf entry, get data address from it.
1511 lep = &ents[index];
1513 * Point to the data entry.
1515 dep = (xfs_dir2_data_entry_t *)
1516 ((char *)dbp->b_addr +
1517 xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1518 ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1520 * Put the new inode number in, log it.
1522 dep->inumber = cpu_to_be64(args->inumber);
1523 dp->d_ops->data_put_ftype(dep, args->filetype);
1524 tp = args->trans;
1525 xfs_dir2_data_log_entry(tp, dp, dbp, dep);
1526 xfs_dir3_leaf_check(dp, lbp);
1527 xfs_trans_brelse(tp, lbp);
1528 return 0;
1532 * Return index in the leaf block (lbp) which is either the first
1533 * one with this hash value, or if there are none, the insert point
1534 * for that hash value.
1536 int /* index value */
1537 xfs_dir2_leaf_search_hash(
1538 xfs_da_args_t *args, /* operation arguments */
1539 struct xfs_buf *lbp) /* leaf buffer */
1541 xfs_dahash_t hash=0; /* hash from this entry */
1542 xfs_dahash_t hashwant; /* hash value looking for */
1543 int high; /* high leaf index */
1544 int low; /* low leaf index */
1545 xfs_dir2_leaf_t *leaf; /* leaf structure */
1546 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1547 int mid=0; /* current leaf index */
1548 struct xfs_dir2_leaf_entry *ents;
1549 struct xfs_dir3_icleaf_hdr leafhdr;
1551 leaf = lbp->b_addr;
1552 ents = args->dp->d_ops->leaf_ents_p(leaf);
1553 args->dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1556 * Note, the table cannot be empty, so we have to go through the loop.
1557 * Binary search the leaf entries looking for our hash value.
1559 for (lep = ents, low = 0, high = leafhdr.count - 1,
1560 hashwant = args->hashval;
1561 low <= high; ) {
1562 mid = (low + high) >> 1;
1563 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1564 break;
1565 if (hash < hashwant)
1566 low = mid + 1;
1567 else
1568 high = mid - 1;
1571 * Found one, back up through all the equal hash values.
1573 if (hash == hashwant) {
1574 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1575 mid--;
1579 * Need to point to an entry higher than ours.
1581 else if (hash < hashwant)
1582 mid++;
1583 return mid;
1587 * Trim off a trailing data block. We know it's empty since the leaf
1588 * freespace table says so.
1590 int /* error */
1591 xfs_dir2_leaf_trim_data(
1592 xfs_da_args_t *args, /* operation arguments */
1593 struct xfs_buf *lbp, /* leaf buffer */
1594 xfs_dir2_db_t db) /* data block number */
1596 __be16 *bestsp; /* leaf bests table */
1597 struct xfs_buf *dbp; /* data block buffer */
1598 xfs_inode_t *dp; /* incore directory inode */
1599 int error; /* error return value */
1600 xfs_dir2_leaf_t *leaf; /* leaf structure */
1601 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1602 xfs_mount_t *mp; /* filesystem mount point */
1603 xfs_trans_t *tp; /* transaction pointer */
1605 dp = args->dp;
1606 mp = dp->i_mount;
1607 tp = args->trans;
1609 * Read the offending data block. We need its buffer.
1611 error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp);
1612 if (error)
1613 return error;
1615 leaf = lbp->b_addr;
1616 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1618 #ifdef DEBUG
1620 struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1621 struct xfs_dir2_data_free *bf = dp->d_ops->data_bestfree_p(hdr);
1623 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1624 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1625 ASSERT(be16_to_cpu(bf[0].length) ==
1626 mp->m_dirblksize - dp->d_ops->data_entry_offset);
1627 ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1629 #endif
1632 * Get rid of the data block.
1634 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1635 ASSERT(error != ENOSPC);
1636 xfs_trans_brelse(tp, dbp);
1637 return error;
1640 * Eliminate the last bests entry from the table.
1642 bestsp = xfs_dir2_leaf_bests_p(ltp);
1643 be32_add_cpu(&ltp->bestcount, -1);
1644 memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1645 xfs_dir3_leaf_log_tail(tp, lbp);
1646 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1647 return 0;
1650 static inline size_t
1651 xfs_dir3_leaf_size(
1652 struct xfs_dir3_icleaf_hdr *hdr,
1653 int counts)
1655 int entries;
1656 int hdrsize;
1658 entries = hdr->count - hdr->stale;
1659 if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1660 hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1661 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1662 else
1663 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1665 return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1666 + counts * sizeof(xfs_dir2_data_off_t)
1667 + sizeof(xfs_dir2_leaf_tail_t);
1671 * Convert node form directory to leaf form directory.
1672 * The root of the node form dir needs to already be a LEAFN block.
1673 * Just return if we can't do anything.
1675 int /* error */
1676 xfs_dir2_node_to_leaf(
1677 xfs_da_state_t *state) /* directory operation state */
1679 xfs_da_args_t *args; /* operation arguments */
1680 xfs_inode_t *dp; /* incore directory inode */
1681 int error; /* error return code */
1682 struct xfs_buf *fbp; /* buffer for freespace block */
1683 xfs_fileoff_t fo; /* freespace file offset */
1684 xfs_dir2_free_t *free; /* freespace structure */
1685 struct xfs_buf *lbp; /* buffer for leaf block */
1686 xfs_dir2_leaf_tail_t *ltp; /* tail of leaf structure */
1687 xfs_dir2_leaf_t *leaf; /* leaf structure */
1688 xfs_mount_t *mp; /* filesystem mount point */
1689 int rval; /* successful free trim? */
1690 xfs_trans_t *tp; /* transaction pointer */
1691 struct xfs_dir3_icleaf_hdr leafhdr;
1692 struct xfs_dir3_icfree_hdr freehdr;
1695 * There's more than a leaf level in the btree, so there must
1696 * be multiple leafn blocks. Give up.
1698 if (state->path.active > 1)
1699 return 0;
1700 args = state->args;
1702 trace_xfs_dir2_node_to_leaf(args);
1704 mp = state->mp;
1705 dp = args->dp;
1706 tp = args->trans;
1708 * Get the last offset in the file.
1710 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1711 return error;
1713 fo -= mp->m_dirblkfsbs;
1715 * If there are freespace blocks other than the first one,
1716 * take this opportunity to remove trailing empty freespace blocks
1717 * that may have been left behind during no-space-reservation
1718 * operations.
1720 while (fo > mp->m_dirfreeblk) {
1721 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1722 return error;
1724 if (rval)
1725 fo -= mp->m_dirblkfsbs;
1726 else
1727 return 0;
1730 * Now find the block just before the freespace block.
1732 if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1733 return error;
1736 * If it's not the single leaf block, give up.
1738 if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1739 return 0;
1740 lbp = state->path.blk[0].bp;
1741 leaf = lbp->b_addr;
1742 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1744 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1745 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1748 * Read the freespace block.
1750 error = xfs_dir2_free_read(tp, dp, mp->m_dirfreeblk, &fbp);
1751 if (error)
1752 return error;
1753 free = fbp->b_addr;
1754 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1756 ASSERT(!freehdr.firstdb);
1759 * Now see if the leafn and free data will fit in a leaf1.
1760 * If not, release the buffer and give up.
1762 if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > mp->m_dirblksize) {
1763 xfs_trans_brelse(tp, fbp);
1764 return 0;
1768 * If the leaf has any stale entries in it, compress them out.
1770 if (leafhdr.stale)
1771 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1773 lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1774 xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1775 leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1776 ? XFS_DIR2_LEAF1_MAGIC
1777 : XFS_DIR3_LEAF1_MAGIC;
1780 * Set up the leaf tail from the freespace block.
1782 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1783 ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1786 * Set up the leaf bests table.
1788 memcpy(xfs_dir2_leaf_bests_p(ltp), dp->d_ops->free_bests_p(free),
1789 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1791 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1792 xfs_dir3_leaf_log_header(tp, dp, lbp);
1793 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1794 xfs_dir3_leaf_log_tail(tp, lbp);
1795 xfs_dir3_leaf_check(dp, lbp);
1798 * Get rid of the freespace block.
1800 error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1801 if (error) {
1803 * This can't fail here because it can only happen when
1804 * punching out the middle of an extent, and this is an
1805 * isolated block.
1807 ASSERT(error != ENOSPC);
1808 return error;
1810 fbp = NULL;
1812 * Now see if we can convert the single-leaf directory
1813 * down to a block form directory.
1814 * This routine always kills the dabuf for the leaf, so
1815 * eliminate it from the path.
1817 error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1818 state->path.blk[0].bp = NULL;
1819 return error;