[SCSI] aic79xx: bump version to 3.0
[linux-2.6/mini2440.git] / fs / xfs / xfs_dir2_leaf.c
blobd342b6b552395e885e167b82f825750dd9bc025d
1 /*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir.h"
28 #include "xfs_dir2.h"
29 #include "xfs_dmapi.h"
30 #include "xfs_mount.h"
31 #include "xfs_da_btree.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dir_sf.h"
35 #include "xfs_dir2_sf.h"
36 #include "xfs_dinode.h"
37 #include "xfs_inode.h"
38 #include "xfs_bmap.h"
39 #include "xfs_dir2_data.h"
40 #include "xfs_dir2_leaf.h"
41 #include "xfs_dir2_block.h"
42 #include "xfs_dir2_node.h"
43 #include "xfs_dir2_trace.h"
44 #include "xfs_error.h"
47 * Local function declarations.
49 #ifdef DEBUG
50 static void xfs_dir2_leaf_check(xfs_inode_t *dp, xfs_dabuf_t *bp);
51 #else
52 #define xfs_dir2_leaf_check(dp, bp)
53 #endif
54 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, xfs_dabuf_t **lbpp,
55 int *indexp, xfs_dabuf_t **dbpp);
56 static void xfs_dir2_leaf_log_bests(struct xfs_trans *tp, struct xfs_dabuf *bp,
57 int first, int last);
58 static void xfs_dir2_leaf_log_tail(struct xfs_trans *tp, struct xfs_dabuf *bp);
62 * Convert a block form directory to a leaf form directory.
64 int /* error */
65 xfs_dir2_block_to_leaf(
66 xfs_da_args_t *args, /* operation arguments */
67 xfs_dabuf_t *dbp) /* input block's buffer */
69 xfs_dir2_data_off_t *bestsp; /* leaf's bestsp entries */
70 xfs_dablk_t blkno; /* leaf block's bno */
71 xfs_dir2_block_t *block; /* block structure */
72 xfs_dir2_leaf_entry_t *blp; /* block's leaf entries */
73 xfs_dir2_block_tail_t *btp; /* block's tail */
74 xfs_inode_t *dp; /* incore directory inode */
75 int error; /* error return code */
76 xfs_dabuf_t *lbp; /* leaf block's buffer */
77 xfs_dir2_db_t ldb; /* leaf block's bno */
78 xfs_dir2_leaf_t *leaf; /* leaf structure */
79 xfs_dir2_leaf_tail_t *ltp; /* leaf's tail */
80 xfs_mount_t *mp; /* filesystem mount point */
81 int needlog; /* need to log block header */
82 int needscan; /* need to rescan bestfree */
83 xfs_trans_t *tp; /* transaction pointer */
85 xfs_dir2_trace_args_b("block_to_leaf", args, dbp);
86 dp = args->dp;
87 mp = dp->i_mount;
88 tp = args->trans;
90 * Add the leaf block to the inode.
91 * This interface will only put blocks in the leaf/node range.
92 * Since that's empty now, we'll get the root (block 0 in range).
94 if ((error = xfs_da_grow_inode(args, &blkno))) {
95 return error;
97 ldb = XFS_DIR2_DA_TO_DB(mp, blkno);
98 ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
100 * Initialize the leaf block, get a buffer for it.
102 if ((error = xfs_dir2_leaf_init(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC))) {
103 return error;
105 ASSERT(lbp != NULL);
106 leaf = lbp->data;
107 block = dbp->data;
108 xfs_dir2_data_check(dp, dbp);
109 btp = XFS_DIR2_BLOCK_TAIL_P(mp, block);
110 blp = XFS_DIR2_BLOCK_LEAF_P(btp);
112 * Set the counts in the leaf header.
114 INT_COPY(leaf->hdr.count, btp->count, ARCH_CONVERT); /* INT_: type change */
115 INT_COPY(leaf->hdr.stale, btp->stale, ARCH_CONVERT); /* INT_: type change */
117 * Could compact these but I think we always do the conversion
118 * after squeezing out stale entries.
120 memcpy(leaf->ents, blp, INT_GET(btp->count, ARCH_CONVERT) * sizeof(xfs_dir2_leaf_entry_t));
121 xfs_dir2_leaf_log_ents(tp, lbp, 0, INT_GET(leaf->hdr.count, ARCH_CONVERT) - 1);
122 needscan = 0;
123 needlog = 1;
125 * Make the space formerly occupied by the leaf entries and block
126 * tail be free.
128 xfs_dir2_data_make_free(tp, dbp,
129 (xfs_dir2_data_aoff_t)((char *)blp - (char *)block),
130 (xfs_dir2_data_aoff_t)((char *)block + mp->m_dirblksize -
131 (char *)blp),
132 &needlog, &needscan);
134 * Fix up the block header, make it a data block.
136 INT_SET(block->hdr.magic, ARCH_CONVERT, XFS_DIR2_DATA_MAGIC);
137 if (needscan)
138 xfs_dir2_data_freescan(mp, (xfs_dir2_data_t *)block, &needlog,
139 NULL);
141 * Set up leaf tail and bests table.
143 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
144 INT_SET(ltp->bestcount, ARCH_CONVERT, 1);
145 bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
146 INT_COPY(bestsp[0], block->hdr.bestfree[0].length, ARCH_CONVERT);
148 * Log the data header and leaf bests table.
150 if (needlog)
151 xfs_dir2_data_log_header(tp, dbp);
152 xfs_dir2_leaf_check(dp, lbp);
153 xfs_dir2_data_check(dp, dbp);
154 xfs_dir2_leaf_log_bests(tp, lbp, 0, 0);
155 xfs_da_buf_done(lbp);
156 return 0;
160 * Add an entry to a leaf form directory.
162 int /* error */
163 xfs_dir2_leaf_addname(
164 xfs_da_args_t *args) /* operation arguments */
166 xfs_dir2_data_off_t *bestsp; /* freespace table in leaf */
167 int compact; /* need to compact leaves */
168 xfs_dir2_data_t *data; /* data block structure */
169 xfs_dabuf_t *dbp; /* data block buffer */
170 xfs_dir2_data_entry_t *dep; /* data block entry */
171 xfs_inode_t *dp; /* incore directory inode */
172 xfs_dir2_data_unused_t *dup; /* data unused entry */
173 int error; /* error return value */
174 int grown; /* allocated new data block */
175 int highstale; /* index of next stale leaf */
176 int i; /* temporary, index */
177 int index; /* leaf table position */
178 xfs_dabuf_t *lbp; /* leaf's buffer */
179 xfs_dir2_leaf_t *leaf; /* leaf structure */
180 int length; /* length of new entry */
181 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
182 int lfloglow; /* low leaf logging index */
183 int lfloghigh; /* high leaf logging index */
184 int lowstale; /* index of prev stale leaf */
185 xfs_dir2_leaf_tail_t *ltp; /* leaf tail pointer */
186 xfs_mount_t *mp; /* filesystem mount point */
187 int needbytes; /* leaf block bytes needed */
188 int needlog; /* need to log data header */
189 int needscan; /* need to rescan data free */
190 xfs_dir2_data_off_t *tagp; /* end of data entry */
191 xfs_trans_t *tp; /* transaction pointer */
192 xfs_dir2_db_t use_block; /* data block number */
194 xfs_dir2_trace_args("leaf_addname", args);
195 dp = args->dp;
196 tp = args->trans;
197 mp = dp->i_mount;
199 * Read the leaf block.
201 error = xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
202 XFS_DATA_FORK);
203 if (error) {
204 return error;
206 ASSERT(lbp != NULL);
208 * Look up the entry by hash value and name.
209 * We know it's not there, our caller has already done a lookup.
210 * So the index is of the entry to insert in front of.
211 * But if there are dup hash values the index is of the first of those.
213 index = xfs_dir2_leaf_search_hash(args, lbp);
214 leaf = lbp->data;
215 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
216 bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
217 length = XFS_DIR2_DATA_ENTSIZE(args->namelen);
219 * See if there are any entries with the same hash value
220 * and space in their block for the new entry.
221 * This is good because it puts multiple same-hash value entries
222 * in a data block, improving the lookup of those entries.
224 for (use_block = -1, lep = &leaf->ents[index];
225 index < INT_GET(leaf->hdr.count, ARCH_CONVERT) && INT_GET(lep->hashval, ARCH_CONVERT) == args->hashval;
226 index++, lep++) {
227 if (INT_GET(lep->address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR)
228 continue;
229 i = XFS_DIR2_DATAPTR_TO_DB(mp, INT_GET(lep->address, ARCH_CONVERT));
230 ASSERT(i < INT_GET(ltp->bestcount, ARCH_CONVERT));
231 ASSERT(INT_GET(bestsp[i], ARCH_CONVERT) != NULLDATAOFF);
232 if (INT_GET(bestsp[i], ARCH_CONVERT) >= length) {
233 use_block = i;
234 break;
238 * Didn't find a block yet, linear search all the data blocks.
240 if (use_block == -1) {
241 for (i = 0; i < INT_GET(ltp->bestcount, ARCH_CONVERT); i++) {
243 * Remember a block we see that's missing.
245 if (INT_GET(bestsp[i], ARCH_CONVERT) == NULLDATAOFF && use_block == -1)
246 use_block = i;
247 else if (INT_GET(bestsp[i], ARCH_CONVERT) >= length) {
248 use_block = i;
249 break;
254 * How many bytes do we need in the leaf block?
256 needbytes =
257 (leaf->hdr.stale ? 0 : (uint)sizeof(leaf->ents[0])) +
258 (use_block != -1 ? 0 : (uint)sizeof(leaf->bests[0]));
260 * Now kill use_block if it refers to a missing block, so we
261 * can use it as an indication of allocation needed.
263 if (use_block != -1 && INT_GET(bestsp[use_block], ARCH_CONVERT) == NULLDATAOFF)
264 use_block = -1;
266 * If we don't have enough free bytes but we can make enough
267 * by compacting out stale entries, we'll do that.
269 if ((char *)bestsp - (char *)&leaf->ents[INT_GET(leaf->hdr.count, ARCH_CONVERT)] < needbytes &&
270 INT_GET(leaf->hdr.stale, ARCH_CONVERT) > 1) {
271 compact = 1;
274 * Otherwise if we don't have enough free bytes we need to
275 * convert to node form.
277 else if ((char *)bestsp - (char *)&leaf->ents[INT_GET(leaf->hdr.count, ARCH_CONVERT)] <
278 needbytes) {
280 * Just checking or no space reservation, give up.
282 if (args->justcheck || args->total == 0) {
283 xfs_da_brelse(tp, lbp);
284 return XFS_ERROR(ENOSPC);
287 * Convert to node form.
289 error = xfs_dir2_leaf_to_node(args, lbp);
290 xfs_da_buf_done(lbp);
291 if (error)
292 return error;
294 * Then add the new entry.
296 return xfs_dir2_node_addname(args);
299 * Otherwise it will fit without compaction.
301 else
302 compact = 0;
304 * If just checking, then it will fit unless we needed to allocate
305 * a new data block.
307 if (args->justcheck) {
308 xfs_da_brelse(tp, lbp);
309 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
312 * If no allocations are allowed, return now before we've
313 * changed anything.
315 if (args->total == 0 && use_block == -1) {
316 xfs_da_brelse(tp, lbp);
317 return XFS_ERROR(ENOSPC);
320 * Need to compact the leaf entries, removing stale ones.
321 * Leave one stale entry behind - the one closest to our
322 * insertion index - and we'll shift that one to our insertion
323 * point later.
325 if (compact) {
326 xfs_dir2_leaf_compact_x1(lbp, &index, &lowstale, &highstale,
327 &lfloglow, &lfloghigh);
330 * There are stale entries, so we'll need log-low and log-high
331 * impossibly bad values later.
333 else if (INT_GET(leaf->hdr.stale, ARCH_CONVERT)) {
334 lfloglow = INT_GET(leaf->hdr.count, ARCH_CONVERT);
335 lfloghigh = -1;
338 * If there was no data block space found, we need to allocate
339 * a new one.
341 if (use_block == -1) {
343 * Add the new data block.
345 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
346 &use_block))) {
347 xfs_da_brelse(tp, lbp);
348 return error;
351 * Initialize the block.
353 if ((error = xfs_dir2_data_init(args, use_block, &dbp))) {
354 xfs_da_brelse(tp, lbp);
355 return error;
358 * If we're adding a new data block on the end we need to
359 * extend the bests table. Copy it up one entry.
361 if (use_block >= INT_GET(ltp->bestcount, ARCH_CONVERT)) {
362 bestsp--;
363 memmove(&bestsp[0], &bestsp[1],
364 INT_GET(ltp->bestcount, ARCH_CONVERT) * sizeof(bestsp[0]));
365 INT_MOD(ltp->bestcount, ARCH_CONVERT, +1);
366 xfs_dir2_leaf_log_tail(tp, lbp);
367 xfs_dir2_leaf_log_bests(tp, lbp, 0, INT_GET(ltp->bestcount, ARCH_CONVERT) - 1);
370 * If we're filling in a previously empty block just log it.
372 else
373 xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
374 data = dbp->data;
375 INT_COPY(bestsp[use_block], data->hdr.bestfree[0].length, ARCH_CONVERT);
376 grown = 1;
379 * Already had space in some data block.
380 * Just read that one in.
382 else {
383 if ((error =
384 xfs_da_read_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, use_block),
385 -1, &dbp, XFS_DATA_FORK))) {
386 xfs_da_brelse(tp, lbp);
387 return error;
389 data = dbp->data;
390 grown = 0;
392 xfs_dir2_data_check(dp, dbp);
394 * Point to the biggest freespace in our data block.
396 dup = (xfs_dir2_data_unused_t *)
397 ((char *)data + INT_GET(data->hdr.bestfree[0].offset, ARCH_CONVERT));
398 ASSERT(INT_GET(dup->length, ARCH_CONVERT) >= length);
399 needscan = needlog = 0;
401 * Mark the initial part of our freespace in use for the new entry.
403 xfs_dir2_data_use_free(tp, dbp, dup,
404 (xfs_dir2_data_aoff_t)((char *)dup - (char *)data), length,
405 &needlog, &needscan);
407 * Initialize our new entry (at last).
409 dep = (xfs_dir2_data_entry_t *)dup;
410 INT_SET(dep->inumber, ARCH_CONVERT, args->inumber);
411 dep->namelen = args->namelen;
412 memcpy(dep->name, args->name, dep->namelen);
413 tagp = XFS_DIR2_DATA_ENTRY_TAG_P(dep);
414 INT_SET(*tagp, ARCH_CONVERT, (xfs_dir2_data_off_t)((char *)dep - (char *)data));
416 * Need to scan fix up the bestfree table.
418 if (needscan)
419 xfs_dir2_data_freescan(mp, data, &needlog, NULL);
421 * Need to log the data block's header.
423 if (needlog)
424 xfs_dir2_data_log_header(tp, dbp);
425 xfs_dir2_data_log_entry(tp, dbp, dep);
427 * If the bests table needs to be changed, do it.
428 * Log the change unless we've already done that.
430 if (INT_GET(bestsp[use_block], ARCH_CONVERT) != INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT)) {
431 INT_COPY(bestsp[use_block], data->hdr.bestfree[0].length, ARCH_CONVERT);
432 if (!grown)
433 xfs_dir2_leaf_log_bests(tp, lbp, use_block, use_block);
436 * Now we need to make room to insert the leaf entry.
437 * If there are no stale entries, we just insert a hole at index.
439 if (!leaf->hdr.stale) {
441 * lep is still good as the index leaf entry.
443 if (index < INT_GET(leaf->hdr.count, ARCH_CONVERT))
444 memmove(lep + 1, lep,
445 (INT_GET(leaf->hdr.count, ARCH_CONVERT) - index) * sizeof(*lep));
447 * Record low and high logging indices for the leaf.
449 lfloglow = index;
450 lfloghigh = INT_GET(leaf->hdr.count, ARCH_CONVERT);
451 INT_MOD(leaf->hdr.count, ARCH_CONVERT, +1);
454 * There are stale entries.
455 * We will use one of them for the new entry.
456 * It's probably not at the right location, so we'll have to
457 * shift some up or down first.
459 else {
461 * If we didn't compact before, we need to find the nearest
462 * stale entries before and after our insertion point.
464 if (compact == 0) {
466 * Find the first stale entry before the insertion
467 * point, if any.
469 for (lowstale = index - 1;
470 lowstale >= 0 &&
471 INT_GET(leaf->ents[lowstale].address, ARCH_CONVERT) !=
472 XFS_DIR2_NULL_DATAPTR;
473 lowstale--)
474 continue;
476 * Find the next stale entry at or after the insertion
477 * point, if any. Stop if we go so far that the
478 * lowstale entry would be better.
480 for (highstale = index;
481 highstale < INT_GET(leaf->hdr.count, ARCH_CONVERT) &&
482 INT_GET(leaf->ents[highstale].address, ARCH_CONVERT) !=
483 XFS_DIR2_NULL_DATAPTR &&
484 (lowstale < 0 ||
485 index - lowstale - 1 >= highstale - index);
486 highstale++)
487 continue;
490 * If the low one is better, use it.
492 if (lowstale >= 0 &&
493 (highstale == INT_GET(leaf->hdr.count, ARCH_CONVERT) ||
494 index - lowstale - 1 < highstale - index)) {
495 ASSERT(index - lowstale - 1 >= 0);
496 ASSERT(INT_GET(leaf->ents[lowstale].address, ARCH_CONVERT) ==
497 XFS_DIR2_NULL_DATAPTR);
499 * Copy entries up to cover the stale entry
500 * and make room for the new entry.
502 if (index - lowstale - 1 > 0)
503 memmove(&leaf->ents[lowstale],
504 &leaf->ents[lowstale + 1],
505 (index - lowstale - 1) * sizeof(*lep));
506 lep = &leaf->ents[index - 1];
507 lfloglow = MIN(lowstale, lfloglow);
508 lfloghigh = MAX(index - 1, lfloghigh);
511 * The high one is better, so use that one.
513 else {
514 ASSERT(highstale - index >= 0);
515 ASSERT(INT_GET(leaf->ents[highstale].address, ARCH_CONVERT) ==
516 XFS_DIR2_NULL_DATAPTR);
518 * Copy entries down to copver the stale entry
519 * and make room for the new entry.
521 if (highstale - index > 0)
522 memmove(&leaf->ents[index + 1],
523 &leaf->ents[index],
524 (highstale - index) * sizeof(*lep));
525 lep = &leaf->ents[index];
526 lfloglow = MIN(index, lfloglow);
527 lfloghigh = MAX(highstale, lfloghigh);
529 INT_MOD(leaf->hdr.stale, ARCH_CONVERT, -1);
532 * Fill in the new leaf entry.
534 INT_SET(lep->hashval, ARCH_CONVERT, args->hashval);
535 INT_SET(lep->address, ARCH_CONVERT, XFS_DIR2_DB_OFF_TO_DATAPTR(mp, use_block, INT_GET(*tagp, ARCH_CONVERT)));
537 * Log the leaf fields and give up the buffers.
539 xfs_dir2_leaf_log_header(tp, lbp);
540 xfs_dir2_leaf_log_ents(tp, lbp, lfloglow, lfloghigh);
541 xfs_dir2_leaf_check(dp, lbp);
542 xfs_da_buf_done(lbp);
543 xfs_dir2_data_check(dp, dbp);
544 xfs_da_buf_done(dbp);
545 return 0;
548 #ifdef DEBUG
550 * Check the internal consistency of a leaf1 block.
551 * Pop an assert if something is wrong.
553 void
554 xfs_dir2_leaf_check(
555 xfs_inode_t *dp, /* incore directory inode */
556 xfs_dabuf_t *bp) /* leaf's buffer */
558 int i; /* leaf index */
559 xfs_dir2_leaf_t *leaf; /* leaf structure */
560 xfs_dir2_leaf_tail_t *ltp; /* leaf tail pointer */
561 xfs_mount_t *mp; /* filesystem mount point */
562 int stale; /* count of stale leaves */
564 leaf = bp->data;
565 mp = dp->i_mount;
566 ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAF1_MAGIC);
568 * This value is not restrictive enough.
569 * Should factor in the size of the bests table as well.
570 * We can deduce a value for that from di_size.
572 ASSERT(INT_GET(leaf->hdr.count, ARCH_CONVERT) <= XFS_DIR2_MAX_LEAF_ENTS(mp));
573 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
575 * Leaves and bests don't overlap.
577 ASSERT((char *)&leaf->ents[INT_GET(leaf->hdr.count, ARCH_CONVERT)] <=
578 (char *)XFS_DIR2_LEAF_BESTS_P(ltp));
580 * Check hash value order, count stale entries.
582 for (i = stale = 0; i < INT_GET(leaf->hdr.count, ARCH_CONVERT); i++) {
583 if (i + 1 < INT_GET(leaf->hdr.count, ARCH_CONVERT))
584 ASSERT(INT_GET(leaf->ents[i].hashval, ARCH_CONVERT) <=
585 INT_GET(leaf->ents[i + 1].hashval, ARCH_CONVERT));
586 if (INT_GET(leaf->ents[i].address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR)
587 stale++;
589 ASSERT(INT_GET(leaf->hdr.stale, ARCH_CONVERT) == stale);
591 #endif /* DEBUG */
594 * Compact out any stale entries in the leaf.
595 * Log the header and changed leaf entries, if any.
597 void
598 xfs_dir2_leaf_compact(
599 xfs_da_args_t *args, /* operation arguments */
600 xfs_dabuf_t *bp) /* leaf buffer */
602 int from; /* source leaf index */
603 xfs_dir2_leaf_t *leaf; /* leaf structure */
604 int loglow; /* first leaf entry to log */
605 int to; /* target leaf index */
607 leaf = bp->data;
608 if (!leaf->hdr.stale) {
609 return;
612 * Compress out the stale entries in place.
614 for (from = to = 0, loglow = -1; from < INT_GET(leaf->hdr.count, ARCH_CONVERT); from++) {
615 if (INT_GET(leaf->ents[from].address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR)
616 continue;
618 * Only actually copy the entries that are different.
620 if (from > to) {
621 if (loglow == -1)
622 loglow = to;
623 leaf->ents[to] = leaf->ents[from];
625 to++;
628 * Update and log the header, log the leaf entries.
630 ASSERT(INT_GET(leaf->hdr.stale, ARCH_CONVERT) == from - to);
631 INT_MOD(leaf->hdr.count, ARCH_CONVERT, -(INT_GET(leaf->hdr.stale, ARCH_CONVERT)));
632 leaf->hdr.stale = 0;
633 xfs_dir2_leaf_log_header(args->trans, bp);
634 if (loglow != -1)
635 xfs_dir2_leaf_log_ents(args->trans, bp, loglow, to - 1);
639 * Compact the leaf entries, removing stale ones.
640 * Leave one stale entry behind - the one closest to our
641 * insertion index - and the caller will shift that one to our insertion
642 * point later.
643 * Return new insertion index, where the remaining stale entry is,
644 * and leaf logging indices.
646 void
647 xfs_dir2_leaf_compact_x1(
648 xfs_dabuf_t *bp, /* leaf buffer */
649 int *indexp, /* insertion index */
650 int *lowstalep, /* out: stale entry before us */
651 int *highstalep, /* out: stale entry after us */
652 int *lowlogp, /* out: low log index */
653 int *highlogp) /* out: high log index */
655 int from; /* source copy index */
656 int highstale; /* stale entry at/after index */
657 int index; /* insertion index */
658 int keepstale; /* source index of kept stale */
659 xfs_dir2_leaf_t *leaf; /* leaf structure */
660 int lowstale; /* stale entry before index */
661 int newindex=0; /* new insertion index */
662 int to; /* destination copy index */
664 leaf = bp->data;
665 ASSERT(INT_GET(leaf->hdr.stale, ARCH_CONVERT) > 1);
666 index = *indexp;
668 * Find the first stale entry before our index, if any.
670 for (lowstale = index - 1;
671 lowstale >= 0 &&
672 INT_GET(leaf->ents[lowstale].address, ARCH_CONVERT) != XFS_DIR2_NULL_DATAPTR;
673 lowstale--)
674 continue;
676 * Find the first stale entry at or after our index, if any.
677 * Stop if the answer would be worse than lowstale.
679 for (highstale = index;
680 highstale < INT_GET(leaf->hdr.count, ARCH_CONVERT) &&
681 INT_GET(leaf->ents[highstale].address, ARCH_CONVERT) != XFS_DIR2_NULL_DATAPTR &&
682 (lowstale < 0 || index - lowstale > highstale - index);
683 highstale++)
684 continue;
686 * Pick the better of lowstale and highstale.
688 if (lowstale >= 0 &&
689 (highstale == INT_GET(leaf->hdr.count, ARCH_CONVERT) ||
690 index - lowstale <= highstale - index))
691 keepstale = lowstale;
692 else
693 keepstale = highstale;
695 * Copy the entries in place, removing all the stale entries
696 * except keepstale.
698 for (from = to = 0; from < INT_GET(leaf->hdr.count, ARCH_CONVERT); from++) {
700 * Notice the new value of index.
702 if (index == from)
703 newindex = to;
704 if (from != keepstale &&
705 INT_GET(leaf->ents[from].address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR) {
706 if (from == to)
707 *lowlogp = to;
708 continue;
711 * Record the new keepstale value for the insertion.
713 if (from == keepstale)
714 lowstale = highstale = to;
716 * Copy only the entries that have moved.
718 if (from > to)
719 leaf->ents[to] = leaf->ents[from];
720 to++;
722 ASSERT(from > to);
724 * If the insertion point was past the last entry,
725 * set the new insertion point accordingly.
727 if (index == from)
728 newindex = to;
729 *indexp = newindex;
731 * Adjust the leaf header values.
733 INT_MOD(leaf->hdr.count, ARCH_CONVERT, -(from - to));
734 INT_SET(leaf->hdr.stale, ARCH_CONVERT, 1);
736 * Remember the low/high stale value only in the "right"
737 * direction.
739 if (lowstale >= newindex)
740 lowstale = -1;
741 else
742 highstale = INT_GET(leaf->hdr.count, ARCH_CONVERT);
743 *highlogp = INT_GET(leaf->hdr.count, ARCH_CONVERT) - 1;
744 *lowstalep = lowstale;
745 *highstalep = highstale;
749 * Getdents (readdir) for leaf and node directories.
750 * This reads the data blocks only, so is the same for both forms.
752 int /* error */
753 xfs_dir2_leaf_getdents(
754 xfs_trans_t *tp, /* transaction pointer */
755 xfs_inode_t *dp, /* incore directory inode */
756 uio_t *uio, /* I/O control & vectors */
757 int *eofp, /* out: reached end of dir */
758 xfs_dirent_t *dbp, /* caller's buffer */
759 xfs_dir2_put_t put) /* ABI formatting routine */
761 xfs_dabuf_t *bp; /* data block buffer */
762 int byteoff; /* offset in current block */
763 xfs_dir2_db_t curdb; /* db for current block */
764 xfs_dir2_off_t curoff; /* current overall offset */
765 xfs_dir2_data_t *data; /* data block structure */
766 xfs_dir2_data_entry_t *dep; /* data entry */
767 xfs_dir2_data_unused_t *dup; /* unused entry */
768 int eof; /* reached end of directory */
769 int error=0; /* error return value */
770 int i; /* temporary loop index */
771 int j; /* temporary loop index */
772 int length; /* temporary length value */
773 xfs_bmbt_irec_t *map; /* map vector for blocks */
774 xfs_extlen_t map_blocks; /* number of fsbs in map */
775 xfs_dablk_t map_off; /* last mapped file offset */
776 int map_size; /* total entries in *map */
777 int map_valid; /* valid entries in *map */
778 xfs_mount_t *mp; /* filesystem mount point */
779 xfs_dir2_off_t newoff; /* new curoff after new blk */
780 int nmap; /* mappings to ask xfs_bmapi */
781 xfs_dir2_put_args_t p; /* formatting arg bundle */
782 char *ptr=NULL; /* pointer to current data */
783 int ra_current; /* number of read-ahead blks */
784 int ra_index; /* *map index for read-ahead */
785 int ra_offset; /* map entry offset for ra */
786 int ra_want; /* readahead count wanted */
789 * If the offset is at or past the largest allowed value,
790 * give up right away, return eof.
792 if (uio->uio_offset >= XFS_DIR2_MAX_DATAPTR) {
793 *eofp = 1;
794 return 0;
796 mp = dp->i_mount;
798 * Setup formatting arguments.
800 p.dbp = dbp;
801 p.put = put;
802 p.uio = uio;
804 * Set up to bmap a number of blocks based on the caller's
805 * buffer size, the directory block size, and the filesystem
806 * block size.
808 map_size =
809 howmany(uio->uio_resid + mp->m_dirblksize,
810 mp->m_sb.sb_blocksize);
811 map = kmem_alloc(map_size * sizeof(*map), KM_SLEEP);
812 map_valid = ra_index = ra_offset = ra_current = map_blocks = 0;
813 bp = NULL;
814 eof = 1;
816 * Inside the loop we keep the main offset value as a byte offset
817 * in the directory file.
819 curoff = XFS_DIR2_DATAPTR_TO_BYTE(mp, uio->uio_offset);
821 * Force this conversion through db so we truncate the offset
822 * down to get the start of the data block.
824 map_off = XFS_DIR2_DB_TO_DA(mp, XFS_DIR2_BYTE_TO_DB(mp, curoff));
826 * Loop over directory entries until we reach the end offset.
827 * Get more blocks and readahead as necessary.
829 while (curoff < XFS_DIR2_LEAF_OFFSET) {
831 * If we have no buffer, or we're off the end of the
832 * current buffer, need to get another one.
834 if (!bp || ptr >= (char *)bp->data + mp->m_dirblksize) {
836 * If we have a buffer, we need to release it and
837 * take it out of the mapping.
839 if (bp) {
840 xfs_da_brelse(tp, bp);
841 bp = NULL;
842 map_blocks -= mp->m_dirblkfsbs;
844 * Loop to get rid of the extents for the
845 * directory block.
847 for (i = mp->m_dirblkfsbs; i > 0; ) {
848 j = MIN((int)map->br_blockcount, i);
849 map->br_blockcount -= j;
850 map->br_startblock += j;
851 map->br_startoff += j;
853 * If mapping is done, pitch it from
854 * the table.
856 if (!map->br_blockcount && --map_valid)
857 memmove(&map[0], &map[1],
858 sizeof(map[0]) *
859 map_valid);
860 i -= j;
864 * Recalculate the readahead blocks wanted.
866 ra_want = howmany(uio->uio_resid + mp->m_dirblksize,
867 mp->m_sb.sb_blocksize) - 1;
869 * If we don't have as many as we want, and we haven't
870 * run out of data blocks, get some more mappings.
872 if (1 + ra_want > map_blocks &&
873 map_off <
874 XFS_DIR2_BYTE_TO_DA(mp, XFS_DIR2_LEAF_OFFSET)) {
876 * Get more bmaps, fill in after the ones
877 * we already have in the table.
879 nmap = map_size - map_valid;
880 error = xfs_bmapi(tp, dp,
881 map_off,
882 XFS_DIR2_BYTE_TO_DA(mp,
883 XFS_DIR2_LEAF_OFFSET) - map_off,
884 XFS_BMAPI_METADATA, NULL, 0,
885 &map[map_valid], &nmap, NULL);
887 * Don't know if we should ignore this or
888 * try to return an error.
889 * The trouble with returning errors
890 * is that readdir will just stop without
891 * actually passing the error through.
893 if (error)
894 break; /* XXX */
896 * If we got all the mappings we asked for,
897 * set the final map offset based on the
898 * last bmap value received.
899 * Otherwise, we've reached the end.
901 if (nmap == map_size - map_valid)
902 map_off =
903 map[map_valid + nmap - 1].br_startoff +
904 map[map_valid + nmap - 1].br_blockcount;
905 else
906 map_off =
907 XFS_DIR2_BYTE_TO_DA(mp,
908 XFS_DIR2_LEAF_OFFSET);
910 * Look for holes in the mapping, and
911 * eliminate them. Count up the valid blocks.
913 for (i = map_valid; i < map_valid + nmap; ) {
914 if (map[i].br_startblock ==
915 HOLESTARTBLOCK) {
916 nmap--;
917 length = map_valid + nmap - i;
918 if (length)
919 memmove(&map[i],
920 &map[i + 1],
921 sizeof(map[i]) *
922 length);
923 } else {
924 map_blocks +=
925 map[i].br_blockcount;
926 i++;
929 map_valid += nmap;
932 * No valid mappings, so no more data blocks.
934 if (!map_valid) {
935 curoff = XFS_DIR2_DA_TO_BYTE(mp, map_off);
936 break;
939 * Read the directory block starting at the first
940 * mapping.
942 curdb = XFS_DIR2_DA_TO_DB(mp, map->br_startoff);
943 error = xfs_da_read_buf(tp, dp, map->br_startoff,
944 map->br_blockcount >= mp->m_dirblkfsbs ?
945 XFS_FSB_TO_DADDR(mp, map->br_startblock) :
947 &bp, XFS_DATA_FORK);
949 * Should just skip over the data block instead
950 * of giving up.
952 if (error)
953 break; /* XXX */
955 * Adjust the current amount of read-ahead: we just
956 * read a block that was previously ra.
958 if (ra_current)
959 ra_current -= mp->m_dirblkfsbs;
961 * Do we need more readahead?
963 for (ra_index = ra_offset = i = 0;
964 ra_want > ra_current && i < map_blocks;
965 i += mp->m_dirblkfsbs) {
966 ASSERT(ra_index < map_valid);
968 * Read-ahead a contiguous directory block.
970 if (i > ra_current &&
971 map[ra_index].br_blockcount >=
972 mp->m_dirblkfsbs) {
973 xfs_baread(mp->m_ddev_targp,
974 XFS_FSB_TO_DADDR(mp,
975 map[ra_index].br_startblock +
976 ra_offset),
977 (int)BTOBB(mp->m_dirblksize));
978 ra_current = i;
981 * Read-ahead a non-contiguous directory block.
982 * This doesn't use our mapping, but this
983 * is a very rare case.
985 else if (i > ra_current) {
986 (void)xfs_da_reada_buf(tp, dp,
987 map[ra_index].br_startoff +
988 ra_offset, XFS_DATA_FORK);
989 ra_current = i;
992 * Advance offset through the mapping table.
994 for (j = 0; j < mp->m_dirblkfsbs; j++) {
996 * The rest of this extent but not
997 * more than a dir block.
999 length = MIN(mp->m_dirblkfsbs,
1000 (int)(map[ra_index].br_blockcount -
1001 ra_offset));
1002 j += length;
1003 ra_offset += length;
1005 * Advance to the next mapping if
1006 * this one is used up.
1008 if (ra_offset ==
1009 map[ra_index].br_blockcount) {
1010 ra_offset = 0;
1011 ra_index++;
1016 * Having done a read, we need to set a new offset.
1018 newoff = XFS_DIR2_DB_OFF_TO_BYTE(mp, curdb, 0);
1020 * Start of the current block.
1022 if (curoff < newoff)
1023 curoff = newoff;
1025 * Make sure we're in the right block.
1027 else if (curoff > newoff)
1028 ASSERT(XFS_DIR2_BYTE_TO_DB(mp, curoff) ==
1029 curdb);
1030 data = bp->data;
1031 xfs_dir2_data_check(dp, bp);
1033 * Find our position in the block.
1035 ptr = (char *)&data->u;
1036 byteoff = XFS_DIR2_BYTE_TO_OFF(mp, curoff);
1038 * Skip past the header.
1040 if (byteoff == 0)
1041 curoff += (uint)sizeof(data->hdr);
1043 * Skip past entries until we reach our offset.
1045 else {
1046 while ((char *)ptr - (char *)data < byteoff) {
1047 dup = (xfs_dir2_data_unused_t *)ptr;
1049 if (INT_GET(dup->freetag, ARCH_CONVERT)
1050 == XFS_DIR2_DATA_FREE_TAG) {
1052 length = INT_GET(dup->length,
1053 ARCH_CONVERT);
1054 ptr += length;
1055 continue;
1057 dep = (xfs_dir2_data_entry_t *)ptr;
1058 length =
1059 XFS_DIR2_DATA_ENTSIZE(dep->namelen);
1060 ptr += length;
1063 * Now set our real offset.
1065 curoff =
1066 XFS_DIR2_DB_OFF_TO_BYTE(mp,
1067 XFS_DIR2_BYTE_TO_DB(mp, curoff),
1068 (char *)ptr - (char *)data);
1069 if (ptr >= (char *)data + mp->m_dirblksize) {
1070 continue;
1075 * We have a pointer to an entry.
1076 * Is it a live one?
1078 dup = (xfs_dir2_data_unused_t *)ptr;
1080 * No, it's unused, skip over it.
1082 if (INT_GET(dup->freetag, ARCH_CONVERT)
1083 == XFS_DIR2_DATA_FREE_TAG) {
1084 length = INT_GET(dup->length, ARCH_CONVERT);
1085 ptr += length;
1086 curoff += length;
1087 continue;
1091 * Copy the entry into the putargs, and try formatting it.
1093 dep = (xfs_dir2_data_entry_t *)ptr;
1095 p.namelen = dep->namelen;
1097 length = XFS_DIR2_DATA_ENTSIZE(p.namelen);
1099 p.cook = XFS_DIR2_BYTE_TO_DATAPTR(mp, curoff + length);
1101 p.ino = INT_GET(dep->inumber, ARCH_CONVERT);
1102 #if XFS_BIG_INUMS
1103 p.ino += mp->m_inoadd;
1104 #endif
1105 p.name = (char *)dep->name;
1107 error = p.put(&p);
1110 * Won't fit. Return to caller.
1112 if (!p.done) {
1113 eof = 0;
1114 break;
1117 * Advance to next entry in the block.
1119 ptr += length;
1120 curoff += length;
1124 * All done. Set output offset value to current offset.
1126 *eofp = eof;
1127 if (curoff > XFS_DIR2_DATAPTR_TO_BYTE(mp, XFS_DIR2_MAX_DATAPTR))
1128 uio->uio_offset = XFS_DIR2_MAX_DATAPTR;
1129 else
1130 uio->uio_offset = XFS_DIR2_BYTE_TO_DATAPTR(mp, curoff);
1131 kmem_free(map, map_size * sizeof(*map));
1132 if (bp)
1133 xfs_da_brelse(tp, bp);
1134 return error;
1138 * Initialize a new leaf block, leaf1 or leafn magic accepted.
1141 xfs_dir2_leaf_init(
1142 xfs_da_args_t *args, /* operation arguments */
1143 xfs_dir2_db_t bno, /* directory block number */
1144 xfs_dabuf_t **bpp, /* out: leaf buffer */
1145 int magic) /* magic number for block */
1147 xfs_dabuf_t *bp; /* leaf buffer */
1148 xfs_inode_t *dp; /* incore directory inode */
1149 int error; /* error return code */
1150 xfs_dir2_leaf_t *leaf; /* leaf structure */
1151 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1152 xfs_mount_t *mp; /* filesystem mount point */
1153 xfs_trans_t *tp; /* transaction pointer */
1155 dp = args->dp;
1156 ASSERT(dp != NULL);
1157 tp = args->trans;
1158 mp = dp->i_mount;
1159 ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
1160 bno < XFS_DIR2_FREE_FIRSTDB(mp));
1162 * Get the buffer for the block.
1164 error = xfs_da_get_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, bno), -1, &bp,
1165 XFS_DATA_FORK);
1166 if (error) {
1167 return error;
1169 ASSERT(bp != NULL);
1170 leaf = bp->data;
1172 * Initialize the header.
1174 INT_SET(leaf->hdr.info.magic, ARCH_CONVERT, magic);
1175 leaf->hdr.info.forw = 0;
1176 leaf->hdr.info.back = 0;
1177 leaf->hdr.count = 0;
1178 leaf->hdr.stale = 0;
1179 xfs_dir2_leaf_log_header(tp, bp);
1181 * If it's a leaf-format directory initialize the tail.
1182 * In this case our caller has the real bests table to copy into
1183 * the block.
1185 if (magic == XFS_DIR2_LEAF1_MAGIC) {
1186 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1187 ltp->bestcount = 0;
1188 xfs_dir2_leaf_log_tail(tp, bp);
1190 *bpp = bp;
1191 return 0;
1195 * Log the bests entries indicated from a leaf1 block.
1197 static void
1198 xfs_dir2_leaf_log_bests(
1199 xfs_trans_t *tp, /* transaction pointer */
1200 xfs_dabuf_t *bp, /* leaf buffer */
1201 int first, /* first entry to log */
1202 int last) /* last entry to log */
1204 xfs_dir2_data_off_t *firstb; /* pointer to first entry */
1205 xfs_dir2_data_off_t *lastb; /* pointer to last entry */
1206 xfs_dir2_leaf_t *leaf; /* leaf structure */
1207 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1209 leaf = bp->data;
1210 ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAF1_MAGIC);
1211 ltp = XFS_DIR2_LEAF_TAIL_P(tp->t_mountp, leaf);
1212 firstb = XFS_DIR2_LEAF_BESTS_P(ltp) + first;
1213 lastb = XFS_DIR2_LEAF_BESTS_P(ltp) + last;
1214 xfs_da_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1215 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1219 * Log the leaf entries indicated from a leaf1 or leafn block.
1221 void
1222 xfs_dir2_leaf_log_ents(
1223 xfs_trans_t *tp, /* transaction pointer */
1224 xfs_dabuf_t *bp, /* leaf buffer */
1225 int first, /* first entry to log */
1226 int last) /* last entry to log */
1228 xfs_dir2_leaf_entry_t *firstlep; /* pointer to first entry */
1229 xfs_dir2_leaf_entry_t *lastlep; /* pointer to last entry */
1230 xfs_dir2_leaf_t *leaf; /* leaf structure */
1232 leaf = bp->data;
1233 ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAF1_MAGIC ||
1234 INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
1235 firstlep = &leaf->ents[first];
1236 lastlep = &leaf->ents[last];
1237 xfs_da_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1238 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1242 * Log the header of the leaf1 or leafn block.
1244 void
1245 xfs_dir2_leaf_log_header(
1246 xfs_trans_t *tp, /* transaction pointer */
1247 xfs_dabuf_t *bp) /* leaf buffer */
1249 xfs_dir2_leaf_t *leaf; /* leaf structure */
1251 leaf = bp->data;
1252 ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAF1_MAGIC ||
1253 INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
1254 xfs_da_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1255 (uint)(sizeof(leaf->hdr) - 1));
1259 * Log the tail of the leaf1 block.
1261 STATIC void
1262 xfs_dir2_leaf_log_tail(
1263 xfs_trans_t *tp, /* transaction pointer */
1264 xfs_dabuf_t *bp) /* leaf buffer */
1266 xfs_dir2_leaf_t *leaf; /* leaf structure */
1267 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1268 xfs_mount_t *mp; /* filesystem mount point */
1270 mp = tp->t_mountp;
1271 leaf = bp->data;
1272 ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAF1_MAGIC);
1273 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1274 xfs_da_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1275 (uint)(mp->m_dirblksize - 1));
1279 * Look up the entry referred to by args in the leaf format directory.
1280 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1281 * is also used by the node-format code.
1284 xfs_dir2_leaf_lookup(
1285 xfs_da_args_t *args) /* operation arguments */
1287 xfs_dabuf_t *dbp; /* data block buffer */
1288 xfs_dir2_data_entry_t *dep; /* data block entry */
1289 xfs_inode_t *dp; /* incore directory inode */
1290 int error; /* error return code */
1291 int index; /* found entry index */
1292 xfs_dabuf_t *lbp; /* leaf buffer */
1293 xfs_dir2_leaf_t *leaf; /* leaf structure */
1294 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1295 xfs_trans_t *tp; /* transaction pointer */
1297 xfs_dir2_trace_args("leaf_lookup", args);
1299 * Look up name in the leaf block, returning both buffers and index.
1301 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1302 return error;
1304 tp = args->trans;
1305 dp = args->dp;
1306 xfs_dir2_leaf_check(dp, lbp);
1307 leaf = lbp->data;
1309 * Get to the leaf entry and contained data entry address.
1311 lep = &leaf->ents[index];
1313 * Point to the data entry.
1315 dep = (xfs_dir2_data_entry_t *)
1316 ((char *)dbp->data +
1317 XFS_DIR2_DATAPTR_TO_OFF(dp->i_mount, INT_GET(lep->address, ARCH_CONVERT)));
1319 * Return the found inode number.
1321 args->inumber = INT_GET(dep->inumber, ARCH_CONVERT);
1322 xfs_da_brelse(tp, dbp);
1323 xfs_da_brelse(tp, lbp);
1324 return XFS_ERROR(EEXIST);
1328 * Look up name/hash in the leaf block.
1329 * Fill in indexp with the found index, and dbpp with the data buffer.
1330 * If not found dbpp will be NULL, and ENOENT comes back.
1331 * lbpp will always be filled in with the leaf buffer unless there's an error.
1333 static int /* error */
1334 xfs_dir2_leaf_lookup_int(
1335 xfs_da_args_t *args, /* operation arguments */
1336 xfs_dabuf_t **lbpp, /* out: leaf buffer */
1337 int *indexp, /* out: index in leaf block */
1338 xfs_dabuf_t **dbpp) /* out: data buffer */
1340 xfs_dir2_db_t curdb; /* current data block number */
1341 xfs_dabuf_t *dbp; /* data buffer */
1342 xfs_dir2_data_entry_t *dep; /* data entry */
1343 xfs_inode_t *dp; /* incore directory inode */
1344 int error; /* error return code */
1345 int index; /* index in leaf block */
1346 xfs_dabuf_t *lbp; /* leaf buffer */
1347 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1348 xfs_dir2_leaf_t *leaf; /* leaf structure */
1349 xfs_mount_t *mp; /* filesystem mount point */
1350 xfs_dir2_db_t newdb; /* new data block number */
1351 xfs_trans_t *tp; /* transaction pointer */
1353 dp = args->dp;
1354 tp = args->trans;
1355 mp = dp->i_mount;
1357 * Read the leaf block into the buffer.
1359 if ((error =
1360 xfs_da_read_buf(tp, dp, mp->m_dirleafblk, -1, &lbp,
1361 XFS_DATA_FORK))) {
1362 return error;
1364 *lbpp = lbp;
1365 leaf = lbp->data;
1366 xfs_dir2_leaf_check(dp, lbp);
1368 * Look for the first leaf entry with our hash value.
1370 index = xfs_dir2_leaf_search_hash(args, lbp);
1372 * Loop over all the entries with the right hash value
1373 * looking to match the name.
1375 for (lep = &leaf->ents[index], dbp = NULL, curdb = -1;
1376 index < INT_GET(leaf->hdr.count, ARCH_CONVERT) && INT_GET(lep->hashval, ARCH_CONVERT) == args->hashval;
1377 lep++, index++) {
1379 * Skip over stale leaf entries.
1381 if (INT_GET(lep->address, ARCH_CONVERT) == XFS_DIR2_NULL_DATAPTR)
1382 continue;
1384 * Get the new data block number.
1386 newdb = XFS_DIR2_DATAPTR_TO_DB(mp, INT_GET(lep->address, ARCH_CONVERT));
1388 * If it's not the same as the old data block number,
1389 * need to pitch the old one and read the new one.
1391 if (newdb != curdb) {
1392 if (dbp)
1393 xfs_da_brelse(tp, dbp);
1394 if ((error =
1395 xfs_da_read_buf(tp, dp,
1396 XFS_DIR2_DB_TO_DA(mp, newdb), -1, &dbp,
1397 XFS_DATA_FORK))) {
1398 xfs_da_brelse(tp, lbp);
1399 return error;
1401 xfs_dir2_data_check(dp, dbp);
1402 curdb = newdb;
1405 * Point to the data entry.
1407 dep = (xfs_dir2_data_entry_t *)
1408 ((char *)dbp->data +
1409 XFS_DIR2_DATAPTR_TO_OFF(mp, INT_GET(lep->address, ARCH_CONVERT)));
1411 * If it matches then return it.
1413 if (dep->namelen == args->namelen &&
1414 dep->name[0] == args->name[0] &&
1415 memcmp(dep->name, args->name, args->namelen) == 0) {
1416 *dbpp = dbp;
1417 *indexp = index;
1418 return 0;
1422 * No match found, return ENOENT.
1424 ASSERT(args->oknoent);
1425 if (dbp)
1426 xfs_da_brelse(tp, dbp);
1427 xfs_da_brelse(tp, lbp);
1428 return XFS_ERROR(ENOENT);
1432 * Remove an entry from a leaf format directory.
1434 int /* error */
1435 xfs_dir2_leaf_removename(
1436 xfs_da_args_t *args) /* operation arguments */
1438 xfs_dir2_data_off_t *bestsp; /* leaf block best freespace */
1439 xfs_dir2_data_t *data; /* data block structure */
1440 xfs_dir2_db_t db; /* data block number */
1441 xfs_dabuf_t *dbp; /* data block buffer */
1442 xfs_dir2_data_entry_t *dep; /* data entry structure */
1443 xfs_inode_t *dp; /* incore directory inode */
1444 int error; /* error return code */
1445 xfs_dir2_db_t i; /* temporary data block # */
1446 int index; /* index into leaf entries */
1447 xfs_dabuf_t *lbp; /* leaf buffer */
1448 xfs_dir2_leaf_t *leaf; /* leaf structure */
1449 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1450 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1451 xfs_mount_t *mp; /* filesystem mount point */
1452 int needlog; /* need to log data header */
1453 int needscan; /* need to rescan data frees */
1454 xfs_dir2_data_off_t oldbest; /* old value of best free */
1455 xfs_trans_t *tp; /* transaction pointer */
1457 xfs_dir2_trace_args("leaf_removename", args);
1459 * Lookup the leaf entry, get the leaf and data blocks read in.
1461 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1462 return error;
1464 dp = args->dp;
1465 tp = args->trans;
1466 mp = dp->i_mount;
1467 leaf = lbp->data;
1468 data = dbp->data;
1469 xfs_dir2_data_check(dp, dbp);
1471 * Point to the leaf entry, use that to point to the data entry.
1473 lep = &leaf->ents[index];
1474 db = XFS_DIR2_DATAPTR_TO_DB(mp, INT_GET(lep->address, ARCH_CONVERT));
1475 dep = (xfs_dir2_data_entry_t *)
1476 ((char *)data + XFS_DIR2_DATAPTR_TO_OFF(mp, INT_GET(lep->address, ARCH_CONVERT)));
1477 needscan = needlog = 0;
1478 oldbest = INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT);
1479 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1480 bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
1481 ASSERT(INT_GET(bestsp[db], ARCH_CONVERT) == oldbest);
1483 * Mark the former data entry unused.
1485 xfs_dir2_data_make_free(tp, dbp,
1486 (xfs_dir2_data_aoff_t)((char *)dep - (char *)data),
1487 XFS_DIR2_DATA_ENTSIZE(dep->namelen), &needlog, &needscan);
1489 * We just mark the leaf entry stale by putting a null in it.
1491 INT_MOD(leaf->hdr.stale, ARCH_CONVERT, +1);
1492 xfs_dir2_leaf_log_header(tp, lbp);
1493 INT_SET(lep->address, ARCH_CONVERT, XFS_DIR2_NULL_DATAPTR);
1494 xfs_dir2_leaf_log_ents(tp, lbp, index, index);
1496 * Scan the freespace in the data block again if necessary,
1497 * log the data block header if necessary.
1499 if (needscan)
1500 xfs_dir2_data_freescan(mp, data, &needlog, NULL);
1501 if (needlog)
1502 xfs_dir2_data_log_header(tp, dbp);
1504 * If the longest freespace in the data block has changed,
1505 * put the new value in the bests table and log that.
1507 if (INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT) != oldbest) {
1508 INT_COPY(bestsp[db], data->hdr.bestfree[0].length, ARCH_CONVERT);
1509 xfs_dir2_leaf_log_bests(tp, lbp, db, db);
1511 xfs_dir2_data_check(dp, dbp);
1513 * If the data block is now empty then get rid of the data block.
1515 if (INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT) ==
1516 mp->m_dirblksize - (uint)sizeof(data->hdr)) {
1517 ASSERT(db != mp->m_dirdatablk);
1518 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1520 * Nope, can't get rid of it because it caused
1521 * allocation of a bmap btree block to do so.
1522 * Just go on, returning success, leaving the
1523 * empty block in place.
1525 if (error == ENOSPC && args->total == 0) {
1526 xfs_da_buf_done(dbp);
1527 error = 0;
1529 xfs_dir2_leaf_check(dp, lbp);
1530 xfs_da_buf_done(lbp);
1531 return error;
1533 dbp = NULL;
1535 * If this is the last data block then compact the
1536 * bests table by getting rid of entries.
1538 if (db == INT_GET(ltp->bestcount, ARCH_CONVERT) - 1) {
1540 * Look for the last active entry (i).
1542 for (i = db - 1; i > 0; i--) {
1543 if (INT_GET(bestsp[i], ARCH_CONVERT) != NULLDATAOFF)
1544 break;
1547 * Copy the table down so inactive entries at the
1548 * end are removed.
1550 memmove(&bestsp[db - i], bestsp,
1551 (INT_GET(ltp->bestcount, ARCH_CONVERT) - (db - i)) * sizeof(*bestsp));
1552 INT_MOD(ltp->bestcount, ARCH_CONVERT, -(db - i));
1553 xfs_dir2_leaf_log_tail(tp, lbp);
1554 xfs_dir2_leaf_log_bests(tp, lbp, 0, INT_GET(ltp->bestcount, ARCH_CONVERT) - 1);
1555 } else
1556 INT_SET(bestsp[db], ARCH_CONVERT, NULLDATAOFF);
1559 * If the data block was not the first one, drop it.
1561 else if (db != mp->m_dirdatablk && dbp != NULL) {
1562 xfs_da_buf_done(dbp);
1563 dbp = NULL;
1565 xfs_dir2_leaf_check(dp, lbp);
1567 * See if we can convert to block form.
1569 return xfs_dir2_leaf_to_block(args, lbp, dbp);
1573 * Replace the inode number in a leaf format directory entry.
1575 int /* error */
1576 xfs_dir2_leaf_replace(
1577 xfs_da_args_t *args) /* operation arguments */
1579 xfs_dabuf_t *dbp; /* data block buffer */
1580 xfs_dir2_data_entry_t *dep; /* data block entry */
1581 xfs_inode_t *dp; /* incore directory inode */
1582 int error; /* error return code */
1583 int index; /* index of leaf entry */
1584 xfs_dabuf_t *lbp; /* leaf buffer */
1585 xfs_dir2_leaf_t *leaf; /* leaf structure */
1586 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1587 xfs_trans_t *tp; /* transaction pointer */
1589 xfs_dir2_trace_args("leaf_replace", args);
1591 * Look up the entry.
1593 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1594 return error;
1596 dp = args->dp;
1597 leaf = lbp->data;
1599 * Point to the leaf entry, get data address from it.
1601 lep = &leaf->ents[index];
1603 * Point to the data entry.
1605 dep = (xfs_dir2_data_entry_t *)
1606 ((char *)dbp->data +
1607 XFS_DIR2_DATAPTR_TO_OFF(dp->i_mount, INT_GET(lep->address, ARCH_CONVERT)));
1608 ASSERT(args->inumber != INT_GET(dep->inumber, ARCH_CONVERT));
1610 * Put the new inode number in, log it.
1612 INT_SET(dep->inumber, ARCH_CONVERT, args->inumber);
1613 tp = args->trans;
1614 xfs_dir2_data_log_entry(tp, dbp, dep);
1615 xfs_da_buf_done(dbp);
1616 xfs_dir2_leaf_check(dp, lbp);
1617 xfs_da_brelse(tp, lbp);
1618 return 0;
1622 * Return index in the leaf block (lbp) which is either the first
1623 * one with this hash value, or if there are none, the insert point
1624 * for that hash value.
1626 int /* index value */
1627 xfs_dir2_leaf_search_hash(
1628 xfs_da_args_t *args, /* operation arguments */
1629 xfs_dabuf_t *lbp) /* leaf buffer */
1631 xfs_dahash_t hash=0; /* hash from this entry */
1632 xfs_dahash_t hashwant; /* hash value looking for */
1633 int high; /* high leaf index */
1634 int low; /* low leaf index */
1635 xfs_dir2_leaf_t *leaf; /* leaf structure */
1636 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1637 int mid=0; /* current leaf index */
1639 leaf = lbp->data;
1640 #ifndef __KERNEL__
1641 if (!leaf->hdr.count)
1642 return 0;
1643 #endif
1645 * Note, the table cannot be empty, so we have to go through the loop.
1646 * Binary search the leaf entries looking for our hash value.
1648 for (lep = leaf->ents, low = 0, high = INT_GET(leaf->hdr.count, ARCH_CONVERT) - 1,
1649 hashwant = args->hashval;
1650 low <= high; ) {
1651 mid = (low + high) >> 1;
1652 if ((hash = INT_GET(lep[mid].hashval, ARCH_CONVERT)) == hashwant)
1653 break;
1654 if (hash < hashwant)
1655 low = mid + 1;
1656 else
1657 high = mid - 1;
1660 * Found one, back up through all the equal hash values.
1662 if (hash == hashwant) {
1663 while (mid > 0 && INT_GET(lep[mid - 1].hashval, ARCH_CONVERT) == hashwant) {
1664 mid--;
1668 * Need to point to an entry higher than ours.
1670 else if (hash < hashwant)
1671 mid++;
1672 return mid;
1676 * Trim off a trailing data block. We know it's empty since the leaf
1677 * freespace table says so.
1679 int /* error */
1680 xfs_dir2_leaf_trim_data(
1681 xfs_da_args_t *args, /* operation arguments */
1682 xfs_dabuf_t *lbp, /* leaf buffer */
1683 xfs_dir2_db_t db) /* data block number */
1685 xfs_dir2_data_off_t *bestsp; /* leaf bests table */
1686 #ifdef DEBUG
1687 xfs_dir2_data_t *data; /* data block structure */
1688 #endif
1689 xfs_dabuf_t *dbp; /* data block buffer */
1690 xfs_inode_t *dp; /* incore directory inode */
1691 int error; /* error return value */
1692 xfs_dir2_leaf_t *leaf; /* leaf structure */
1693 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1694 xfs_mount_t *mp; /* filesystem mount point */
1695 xfs_trans_t *tp; /* transaction pointer */
1697 dp = args->dp;
1698 mp = dp->i_mount;
1699 tp = args->trans;
1701 * Read the offending data block. We need its buffer.
1703 if ((error = xfs_da_read_buf(tp, dp, XFS_DIR2_DB_TO_DA(mp, db), -1, &dbp,
1704 XFS_DATA_FORK))) {
1705 return error;
1707 #ifdef DEBUG
1708 data = dbp->data;
1709 ASSERT(INT_GET(data->hdr.magic, ARCH_CONVERT) == XFS_DIR2_DATA_MAGIC);
1710 #endif
1711 /* this seems to be an error
1712 * data is only valid if DEBUG is defined?
1713 * RMC 09/08/1999
1716 leaf = lbp->data;
1717 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1718 ASSERT(INT_GET(data->hdr.bestfree[0].length, ARCH_CONVERT) ==
1719 mp->m_dirblksize - (uint)sizeof(data->hdr));
1720 ASSERT(db == INT_GET(ltp->bestcount, ARCH_CONVERT) - 1);
1722 * Get rid of the data block.
1724 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1725 ASSERT(error != ENOSPC);
1726 xfs_da_brelse(tp, dbp);
1727 return error;
1730 * Eliminate the last bests entry from the table.
1732 bestsp = XFS_DIR2_LEAF_BESTS_P(ltp);
1733 INT_MOD(ltp->bestcount, ARCH_CONVERT, -1);
1734 memmove(&bestsp[1], &bestsp[0], INT_GET(ltp->bestcount, ARCH_CONVERT) * sizeof(*bestsp));
1735 xfs_dir2_leaf_log_tail(tp, lbp);
1736 xfs_dir2_leaf_log_bests(tp, lbp, 0, INT_GET(ltp->bestcount, ARCH_CONVERT) - 1);
1737 return 0;
1741 * Convert node form directory to leaf form directory.
1742 * The root of the node form dir needs to already be a LEAFN block.
1743 * Just return if we can't do anything.
1745 int /* error */
1746 xfs_dir2_node_to_leaf(
1747 xfs_da_state_t *state) /* directory operation state */
1749 xfs_da_args_t *args; /* operation arguments */
1750 xfs_inode_t *dp; /* incore directory inode */
1751 int error; /* error return code */
1752 xfs_dabuf_t *fbp; /* buffer for freespace block */
1753 xfs_fileoff_t fo; /* freespace file offset */
1754 xfs_dir2_free_t *free; /* freespace structure */
1755 xfs_dabuf_t *lbp; /* buffer for leaf block */
1756 xfs_dir2_leaf_tail_t *ltp; /* tail of leaf structure */
1757 xfs_dir2_leaf_t *leaf; /* leaf structure */
1758 xfs_mount_t *mp; /* filesystem mount point */
1759 int rval; /* successful free trim? */
1760 xfs_trans_t *tp; /* transaction pointer */
1763 * There's more than a leaf level in the btree, so there must
1764 * be multiple leafn blocks. Give up.
1766 if (state->path.active > 1)
1767 return 0;
1768 args = state->args;
1769 xfs_dir2_trace_args("node_to_leaf", args);
1770 mp = state->mp;
1771 dp = args->dp;
1772 tp = args->trans;
1774 * Get the last offset in the file.
1776 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1777 return error;
1779 fo -= mp->m_dirblkfsbs;
1781 * If there are freespace blocks other than the first one,
1782 * take this opportunity to remove trailing empty freespace blocks
1783 * that may have been left behind during no-space-reservation
1784 * operations.
1786 while (fo > mp->m_dirfreeblk) {
1787 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1788 return error;
1790 if (rval)
1791 fo -= mp->m_dirblkfsbs;
1792 else
1793 return 0;
1796 * Now find the block just before the freespace block.
1798 if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1799 return error;
1802 * If it's not the single leaf block, give up.
1804 if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1805 return 0;
1806 lbp = state->path.blk[0].bp;
1807 leaf = lbp->data;
1808 ASSERT(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT) == XFS_DIR2_LEAFN_MAGIC);
1810 * Read the freespace block.
1812 if ((error = xfs_da_read_buf(tp, dp, mp->m_dirfreeblk, -1, &fbp,
1813 XFS_DATA_FORK))) {
1814 return error;
1816 free = fbp->data;
1817 ASSERT(INT_GET(free->hdr.magic, ARCH_CONVERT) == XFS_DIR2_FREE_MAGIC);
1818 ASSERT(!free->hdr.firstdb);
1820 * Now see if the leafn and free data will fit in a leaf1.
1821 * If not, release the buffer and give up.
1823 if ((uint)sizeof(leaf->hdr) +
1824 (INT_GET(leaf->hdr.count, ARCH_CONVERT) - INT_GET(leaf->hdr.stale, ARCH_CONVERT)) * (uint)sizeof(leaf->ents[0]) +
1825 INT_GET(free->hdr.nvalid, ARCH_CONVERT) * (uint)sizeof(leaf->bests[0]) +
1826 (uint)sizeof(leaf->tail) >
1827 mp->m_dirblksize) {
1828 xfs_da_brelse(tp, fbp);
1829 return 0;
1832 * If the leaf has any stale entries in it, compress them out.
1833 * The compact routine will log the header.
1835 if (INT_GET(leaf->hdr.stale, ARCH_CONVERT))
1836 xfs_dir2_leaf_compact(args, lbp);
1837 else
1838 xfs_dir2_leaf_log_header(tp, lbp);
1839 INT_SET(leaf->hdr.info.magic, ARCH_CONVERT, XFS_DIR2_LEAF1_MAGIC);
1841 * Set up the leaf tail from the freespace block.
1843 ltp = XFS_DIR2_LEAF_TAIL_P(mp, leaf);
1844 INT_COPY(ltp->bestcount, free->hdr.nvalid, ARCH_CONVERT);
1846 * Set up the leaf bests table.
1848 memcpy(XFS_DIR2_LEAF_BESTS_P(ltp), free->bests,
1849 INT_GET(ltp->bestcount, ARCH_CONVERT) * sizeof(leaf->bests[0]));
1850 xfs_dir2_leaf_log_bests(tp, lbp, 0, INT_GET(ltp->bestcount, ARCH_CONVERT) - 1);
1851 xfs_dir2_leaf_log_tail(tp, lbp);
1852 xfs_dir2_leaf_check(dp, lbp);
1854 * Get rid of the freespace block.
1856 error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1857 if (error) {
1859 * This can't fail here because it can only happen when
1860 * punching out the middle of an extent, and this is an
1861 * isolated block.
1863 ASSERT(error != ENOSPC);
1864 return error;
1866 fbp = NULL;
1868 * Now see if we can convert the single-leaf directory
1869 * down to a block form directory.
1870 * This routine always kills the dabuf for the leaf, so
1871 * eliminate it from the path.
1873 error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1874 state->path.blk[0].bp = NULL;
1875 return error;