Loading...
Note: File does not exist in v6.13.7.
1/*
2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
4 * All Rights Reserved.
5 *
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.
9 *
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.
14 *
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
18 */
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"
38
39/*
40 * Local function declarations.
41 */
42static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
43 int *indexp, struct xfs_buf **dbpp);
44static void xfs_dir3_leaf_log_bests(struct xfs_trans *tp, struct xfs_buf *bp,
45 int first, int last);
46static void xfs_dir3_leaf_log_tail(struct xfs_trans *tp, struct xfs_buf *bp);
47
48/*
49 * Check the internal consistency of a leaf1 block.
50 * Pop an assert if something is wrong.
51 */
52#ifdef DEBUG
53#define xfs_dir3_leaf_check(dp, bp) \
54do { \
55 if (!xfs_dir3_leaf1_check((dp), (bp))) \
56 ASSERT(0); \
57} while (0);
58
59STATIC bool
60xfs_dir3_leaf1_check(
61 struct xfs_inode *dp,
62 struct xfs_buf *bp)
63{
64 struct xfs_dir2_leaf *leaf = bp->b_addr;
65 struct xfs_dir3_icleaf_hdr leafhdr;
66
67 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
68
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;
75
76 return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
77}
78#else
79#define xfs_dir3_leaf_check(dp, bp)
80#endif
81
82bool
83xfs_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)
88{
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;
95
96 /*
97 * we can be passed a null dp here from a verifier, so we need to go the
98 * hard way to get them.
99 */
100 ops = xfs_dir_get_ops(mp, dp);
101
102 if (!hdr) {
103 ops->leaf_hdr_from_disk(&leafhdr, leaf);
104 hdr = &leafhdr;
105 }
106
107 ents = ops->leaf_ents_p(leaf);
108 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
109
110 /*
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.
114 */
115 if (hdr->count > ops->leaf_max_ents(mp))
116 return false;
117
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;
123
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;
130 }
131 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
132 stale++;
133 }
134 if (hdr->stale != stale)
135 return false;
136 return true;
137}
138
139/*
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.
143 */
144static bool
145xfs_dir3_leaf_verify(
146 struct xfs_buf *bp,
147 __uint16_t magic)
148{
149 struct xfs_mount *mp = bp->b_target->bt_mount;
150 struct xfs_dir2_leaf *leaf = bp->b_addr;
151
152 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
153
154 if (xfs_sb_version_hascrc(&mp->m_sb)) {
155 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
156 __uint16_t magic3;
157
158 magic3 = (magic == XFS_DIR2_LEAF1_MAGIC) ? XFS_DIR3_LEAF1_MAGIC
159 : XFS_DIR3_LEAFN_MAGIC;
160
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;
170 }
171
172 return xfs_dir3_leaf_check_int(mp, NULL, NULL, leaf);
173}
174
175static void
176__read_verify(
177 struct xfs_buf *bp,
178 __uint16_t magic)
179{
180 struct xfs_mount *mp = bp->b_target->bt_mount;
181
182 if (xfs_sb_version_hascrc(&mp->m_sb) &&
183 !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
184 xfs_buf_ioerror(bp, EFSBADCRC);
185 else if (!xfs_dir3_leaf_verify(bp, magic))
186 xfs_buf_ioerror(bp, EFSCORRUPTED);
187
188 if (bp->b_error)
189 xfs_verifier_error(bp);
190}
191
192static void
193__write_verify(
194 struct xfs_buf *bp,
195 __uint16_t magic)
196{
197 struct xfs_mount *mp = bp->b_target->bt_mount;
198 struct xfs_buf_log_item *bip = bp->b_fspriv;
199 struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
200
201 if (!xfs_dir3_leaf_verify(bp, magic)) {
202 xfs_buf_ioerror(bp, EFSCORRUPTED);
203 xfs_verifier_error(bp);
204 return;
205 }
206
207 if (!xfs_sb_version_hascrc(&mp->m_sb))
208 return;
209
210 if (bip)
211 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
212
213 xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
214}
215
216static void
217xfs_dir3_leaf1_read_verify(
218 struct xfs_buf *bp)
219{
220 __read_verify(bp, XFS_DIR2_LEAF1_MAGIC);
221}
222
223static void
224xfs_dir3_leaf1_write_verify(
225 struct xfs_buf *bp)
226{
227 __write_verify(bp, XFS_DIR2_LEAF1_MAGIC);
228}
229
230static void
231xfs_dir3_leafn_read_verify(
232 struct xfs_buf *bp)
233{
234 __read_verify(bp, XFS_DIR2_LEAFN_MAGIC);
235}
236
237static void
238xfs_dir3_leafn_write_verify(
239 struct xfs_buf *bp)
240{
241 __write_verify(bp, XFS_DIR2_LEAFN_MAGIC);
242}
243
244const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
245 .verify_read = xfs_dir3_leaf1_read_verify,
246 .verify_write = xfs_dir3_leaf1_write_verify,
247};
248
249const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
250 .verify_read = xfs_dir3_leafn_read_verify,
251 .verify_write = xfs_dir3_leafn_write_verify,
252};
253
254static int
255xfs_dir3_leaf_read(
256 struct xfs_trans *tp,
257 struct xfs_inode *dp,
258 xfs_dablk_t fbno,
259 xfs_daddr_t mappedbno,
260 struct xfs_buf **bpp)
261{
262 int err;
263
264 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
265 XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
266 if (!err && tp)
267 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
268 return err;
269}
270
271int
272xfs_dir3_leafn_read(
273 struct xfs_trans *tp,
274 struct xfs_inode *dp,
275 xfs_dablk_t fbno,
276 xfs_daddr_t mappedbno,
277 struct xfs_buf **bpp)
278{
279 int err;
280
281 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
282 XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
283 if (!err && tp)
284 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
285 return err;
286}
287
288/*
289 * Initialize a new leaf block, leaf1 or leafn magic accepted.
290 */
291static void
292xfs_dir3_leaf_init(
293 struct xfs_mount *mp,
294 struct xfs_trans *tp,
295 struct xfs_buf *bp,
296 xfs_ino_t owner,
297 __uint16_t type)
298{
299 struct xfs_dir2_leaf *leaf = bp->b_addr;
300
301 ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
302
303 if (xfs_sb_version_hascrc(&mp->m_sb)) {
304 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
305
306 memset(leaf3, 0, sizeof(*leaf3));
307
308 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
309 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
310 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
311 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
312 leaf3->info.owner = cpu_to_be64(owner);
313 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_uuid);
314 } else {
315 memset(leaf, 0, sizeof(*leaf));
316 leaf->hdr.info.magic = cpu_to_be16(type);
317 }
318
319 /*
320 * If it's a leaf-format directory initialize the tail.
321 * Caller is responsible for initialising the bests table.
322 */
323 if (type == XFS_DIR2_LEAF1_MAGIC) {
324 struct xfs_dir2_leaf_tail *ltp;
325
326 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
327 ltp->bestcount = 0;
328 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
329 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
330 } else {
331 bp->b_ops = &xfs_dir3_leafn_buf_ops;
332 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
333 }
334}
335
336int
337xfs_dir3_leaf_get_buf(
338 xfs_da_args_t *args,
339 xfs_dir2_db_t bno,
340 struct xfs_buf **bpp,
341 __uint16_t magic)
342{
343 struct xfs_inode *dp = args->dp;
344 struct xfs_trans *tp = args->trans;
345 struct xfs_mount *mp = dp->i_mount;
346 struct xfs_buf *bp;
347 int error;
348
349 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
350 ASSERT(bno >= XFS_DIR2_LEAF_FIRSTDB(mp) &&
351 bno < XFS_DIR2_FREE_FIRSTDB(mp));
352
353 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(mp, bno), -1, &bp,
354 XFS_DATA_FORK);
355 if (error)
356 return error;
357
358 xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
359 xfs_dir3_leaf_log_header(tp, dp, bp);
360 if (magic == XFS_DIR2_LEAF1_MAGIC)
361 xfs_dir3_leaf_log_tail(tp, bp);
362 *bpp = bp;
363 return 0;
364}
365
366/*
367 * Convert a block form directory to a leaf form directory.
368 */
369int /* error */
370xfs_dir2_block_to_leaf(
371 xfs_da_args_t *args, /* operation arguments */
372 struct xfs_buf *dbp) /* input block's buffer */
373{
374 __be16 *bestsp; /* leaf's bestsp entries */
375 xfs_dablk_t blkno; /* leaf block's bno */
376 xfs_dir2_data_hdr_t *hdr; /* block header */
377 xfs_dir2_leaf_entry_t *blp; /* block's leaf entries */
378 xfs_dir2_block_tail_t *btp; /* block's tail */
379 xfs_inode_t *dp; /* incore directory inode */
380 int error; /* error return code */
381 struct xfs_buf *lbp; /* leaf block's buffer */
382 xfs_dir2_db_t ldb; /* leaf block's bno */
383 xfs_dir2_leaf_t *leaf; /* leaf structure */
384 xfs_dir2_leaf_tail_t *ltp; /* leaf's tail */
385 xfs_mount_t *mp; /* filesystem mount point */
386 int needlog; /* need to log block header */
387 int needscan; /* need to rescan bestfree */
388 xfs_trans_t *tp; /* transaction pointer */
389 struct xfs_dir2_data_free *bf;
390 struct xfs_dir2_leaf_entry *ents;
391 struct xfs_dir3_icleaf_hdr leafhdr;
392
393 trace_xfs_dir2_block_to_leaf(args);
394
395 dp = args->dp;
396 mp = dp->i_mount;
397 tp = args->trans;
398 /*
399 * Add the leaf block to the inode.
400 * This interface will only put blocks in the leaf/node range.
401 * Since that's empty now, we'll get the root (block 0 in range).
402 */
403 if ((error = xfs_da_grow_inode(args, &blkno))) {
404 return error;
405 }
406 ldb = xfs_dir2_da_to_db(mp, blkno);
407 ASSERT(ldb == XFS_DIR2_LEAF_FIRSTDB(mp));
408 /*
409 * Initialize the leaf block, get a buffer for it.
410 */
411 error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
412 if (error)
413 return error;
414
415 leaf = lbp->b_addr;
416 hdr = dbp->b_addr;
417 xfs_dir3_data_check(dp, dbp);
418 btp = xfs_dir2_block_tail_p(mp, hdr);
419 blp = xfs_dir2_block_leaf_p(btp);
420 bf = dp->d_ops->data_bestfree_p(hdr);
421 ents = dp->d_ops->leaf_ents_p(leaf);
422
423 /*
424 * Set the counts in the leaf header.
425 */
426 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
427 leafhdr.count = be32_to_cpu(btp->count);
428 leafhdr.stale = be32_to_cpu(btp->stale);
429 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
430 xfs_dir3_leaf_log_header(tp, dp, lbp);
431
432 /*
433 * Could compact these but I think we always do the conversion
434 * after squeezing out stale entries.
435 */
436 memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
437 xfs_dir3_leaf_log_ents(tp, dp, lbp, 0, leafhdr.count - 1);
438 needscan = 0;
439 needlog = 1;
440 /*
441 * Make the space formerly occupied by the leaf entries and block
442 * tail be free.
443 */
444 xfs_dir2_data_make_free(tp, dp, dbp,
445 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
446 (xfs_dir2_data_aoff_t)((char *)hdr + mp->m_dirblksize -
447 (char *)blp),
448 &needlog, &needscan);
449 /*
450 * Fix up the block header, make it a data block.
451 */
452 dbp->b_ops = &xfs_dir3_data_buf_ops;
453 xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
454 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
455 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
456 else
457 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
458
459 if (needscan)
460 xfs_dir2_data_freescan(dp, hdr, &needlog);
461 /*
462 * Set up leaf tail and bests table.
463 */
464 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
465 ltp->bestcount = cpu_to_be32(1);
466 bestsp = xfs_dir2_leaf_bests_p(ltp);
467 bestsp[0] = bf[0].length;
468 /*
469 * Log the data header and leaf bests table.
470 */
471 if (needlog)
472 xfs_dir2_data_log_header(tp, dp, dbp);
473 xfs_dir3_leaf_check(dp, lbp);
474 xfs_dir3_data_check(dp, dbp);
475 xfs_dir3_leaf_log_bests(tp, lbp, 0, 0);
476 return 0;
477}
478
479STATIC void
480xfs_dir3_leaf_find_stale(
481 struct xfs_dir3_icleaf_hdr *leafhdr,
482 struct xfs_dir2_leaf_entry *ents,
483 int index,
484 int *lowstale,
485 int *highstale)
486{
487 /*
488 * Find the first stale entry before our index, if any.
489 */
490 for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
491 if (ents[*lowstale].address ==
492 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
493 break;
494 }
495
496 /*
497 * Find the first stale entry at or after our index, if any.
498 * Stop if the result would require moving more entries than using
499 * lowstale.
500 */
501 for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
502 if (ents[*highstale].address ==
503 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
504 break;
505 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
506 break;
507 }
508}
509
510struct xfs_dir2_leaf_entry *
511xfs_dir3_leaf_find_entry(
512 struct xfs_dir3_icleaf_hdr *leafhdr,
513 struct xfs_dir2_leaf_entry *ents,
514 int index, /* leaf table position */
515 int compact, /* need to compact leaves */
516 int lowstale, /* index of prev stale leaf */
517 int highstale, /* index of next stale leaf */
518 int *lfloglow, /* low leaf logging index */
519 int *lfloghigh) /* high leaf logging index */
520{
521 if (!leafhdr->stale) {
522 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
523
524 /*
525 * Now we need to make room to insert the leaf entry.
526 *
527 * If there are no stale entries, just insert a hole at index.
528 */
529 lep = &ents[index];
530 if (index < leafhdr->count)
531 memmove(lep + 1, lep,
532 (leafhdr->count - index) * sizeof(*lep));
533
534 /*
535 * Record low and high logging indices for the leaf.
536 */
537 *lfloglow = index;
538 *lfloghigh = leafhdr->count++;
539 return lep;
540 }
541
542 /*
543 * There are stale entries.
544 *
545 * We will use one of them for the new entry. It's probably not at
546 * the right location, so we'll have to shift some up or down first.
547 *
548 * If we didn't compact before, we need to find the nearest stale
549 * entries before and after our insertion point.
550 */
551 if (compact == 0)
552 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
553 &lowstale, &highstale);
554
555 /*
556 * If the low one is better, use it.
557 */
558 if (lowstale >= 0 &&
559 (highstale == leafhdr->count ||
560 index - lowstale - 1 < highstale - index)) {
561 ASSERT(index - lowstale - 1 >= 0);
562 ASSERT(ents[lowstale].address ==
563 cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
564
565 /*
566 * Copy entries up to cover the stale entry and make room
567 * for the new entry.
568 */
569 if (index - lowstale - 1 > 0) {
570 memmove(&ents[lowstale], &ents[lowstale + 1],
571 (index - lowstale - 1) *
572 sizeof(xfs_dir2_leaf_entry_t));
573 }
574 *lfloglow = MIN(lowstale, *lfloglow);
575 *lfloghigh = MAX(index - 1, *lfloghigh);
576 leafhdr->stale--;
577 return &ents[index - 1];
578 }
579
580 /*
581 * The high one is better, so use that one.
582 */
583 ASSERT(highstale - index >= 0);
584 ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
585
586 /*
587 * Copy entries down to cover the stale entry and make room for the
588 * new entry.
589 */
590 if (highstale - index > 0) {
591 memmove(&ents[index + 1], &ents[index],
592 (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
593 }
594 *lfloglow = MIN(index, *lfloglow);
595 *lfloghigh = MAX(highstale, *lfloghigh);
596 leafhdr->stale--;
597 return &ents[index];
598}
599
600/*
601 * Add an entry to a leaf form directory.
602 */
603int /* error */
604xfs_dir2_leaf_addname(
605 xfs_da_args_t *args) /* operation arguments */
606{
607 __be16 *bestsp; /* freespace table in leaf */
608 int compact; /* need to compact leaves */
609 xfs_dir2_data_hdr_t *hdr; /* data block header */
610 struct xfs_buf *dbp; /* data block buffer */
611 xfs_dir2_data_entry_t *dep; /* data block entry */
612 xfs_inode_t *dp; /* incore directory inode */
613 xfs_dir2_data_unused_t *dup; /* data unused entry */
614 int error; /* error return value */
615 int grown; /* allocated new data block */
616 int highstale; /* index of next stale leaf */
617 int i; /* temporary, index */
618 int index; /* leaf table position */
619 struct xfs_buf *lbp; /* leaf's buffer */
620 xfs_dir2_leaf_t *leaf; /* leaf structure */
621 int length; /* length of new entry */
622 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
623 int lfloglow; /* low leaf logging index */
624 int lfloghigh; /* high leaf logging index */
625 int lowstale; /* index of prev stale leaf */
626 xfs_dir2_leaf_tail_t *ltp; /* leaf tail pointer */
627 xfs_mount_t *mp; /* filesystem mount point */
628 int needbytes; /* leaf block bytes needed */
629 int needlog; /* need to log data header */
630 int needscan; /* need to rescan data free */
631 __be16 *tagp; /* end of data entry */
632 xfs_trans_t *tp; /* transaction pointer */
633 xfs_dir2_db_t use_block; /* data block number */
634 struct xfs_dir2_data_free *bf; /* bestfree table */
635 struct xfs_dir2_leaf_entry *ents;
636 struct xfs_dir3_icleaf_hdr leafhdr;
637
638 trace_xfs_dir2_leaf_addname(args);
639
640 dp = args->dp;
641 tp = args->trans;
642 mp = dp->i_mount;
643
644 error = xfs_dir3_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
645 if (error)
646 return error;
647
648 /*
649 * Look up the entry by hash value and name.
650 * We know it's not there, our caller has already done a lookup.
651 * So the index is of the entry to insert in front of.
652 * But if there are dup hash values the index is of the first of those.
653 */
654 index = xfs_dir2_leaf_search_hash(args, lbp);
655 leaf = lbp->b_addr;
656 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
657 ents = dp->d_ops->leaf_ents_p(leaf);
658 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
659 bestsp = xfs_dir2_leaf_bests_p(ltp);
660 length = dp->d_ops->data_entsize(args->namelen);
661
662 /*
663 * See if there are any entries with the same hash value
664 * and space in their block for the new entry.
665 * This is good because it puts multiple same-hash value entries
666 * in a data block, improving the lookup of those entries.
667 */
668 for (use_block = -1, lep = &ents[index];
669 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
670 index++, lep++) {
671 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
672 continue;
673 i = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
674 ASSERT(i < be32_to_cpu(ltp->bestcount));
675 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
676 if (be16_to_cpu(bestsp[i]) >= length) {
677 use_block = i;
678 break;
679 }
680 }
681 /*
682 * Didn't find a block yet, linear search all the data blocks.
683 */
684 if (use_block == -1) {
685 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
686 /*
687 * Remember a block we see that's missing.
688 */
689 if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
690 use_block == -1)
691 use_block = i;
692 else if (be16_to_cpu(bestsp[i]) >= length) {
693 use_block = i;
694 break;
695 }
696 }
697 }
698 /*
699 * How many bytes do we need in the leaf block?
700 */
701 needbytes = 0;
702 if (!leafhdr.stale)
703 needbytes += sizeof(xfs_dir2_leaf_entry_t);
704 if (use_block == -1)
705 needbytes += sizeof(xfs_dir2_data_off_t);
706
707 /*
708 * Now kill use_block if it refers to a missing block, so we
709 * can use it as an indication of allocation needed.
710 */
711 if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
712 use_block = -1;
713 /*
714 * If we don't have enough free bytes but we can make enough
715 * by compacting out stale entries, we'll do that.
716 */
717 if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
718 leafhdr.stale > 1)
719 compact = 1;
720
721 /*
722 * Otherwise if we don't have enough free bytes we need to
723 * convert to node form.
724 */
725 else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
726 /*
727 * Just checking or no space reservation, give up.
728 */
729 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
730 args->total == 0) {
731 xfs_trans_brelse(tp, lbp);
732 return XFS_ERROR(ENOSPC);
733 }
734 /*
735 * Convert to node form.
736 */
737 error = xfs_dir2_leaf_to_node(args, lbp);
738 if (error)
739 return error;
740 /*
741 * Then add the new entry.
742 */
743 return xfs_dir2_node_addname(args);
744 }
745 /*
746 * Otherwise it will fit without compaction.
747 */
748 else
749 compact = 0;
750 /*
751 * If just checking, then it will fit unless we needed to allocate
752 * a new data block.
753 */
754 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
755 xfs_trans_brelse(tp, lbp);
756 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
757 }
758 /*
759 * If no allocations are allowed, return now before we've
760 * changed anything.
761 */
762 if (args->total == 0 && use_block == -1) {
763 xfs_trans_brelse(tp, lbp);
764 return XFS_ERROR(ENOSPC);
765 }
766 /*
767 * Need to compact the leaf entries, removing stale ones.
768 * Leave one stale entry behind - the one closest to our
769 * insertion index - and we'll shift that one to our insertion
770 * point later.
771 */
772 if (compact) {
773 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
774 &highstale, &lfloglow, &lfloghigh);
775 }
776 /*
777 * There are stale entries, so we'll need log-low and log-high
778 * impossibly bad values later.
779 */
780 else if (leafhdr.stale) {
781 lfloglow = leafhdr.count;
782 lfloghigh = -1;
783 }
784 /*
785 * If there was no data block space found, we need to allocate
786 * a new one.
787 */
788 if (use_block == -1) {
789 /*
790 * Add the new data block.
791 */
792 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
793 &use_block))) {
794 xfs_trans_brelse(tp, lbp);
795 return error;
796 }
797 /*
798 * Initialize the block.
799 */
800 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
801 xfs_trans_brelse(tp, lbp);
802 return error;
803 }
804 /*
805 * If we're adding a new data block on the end we need to
806 * extend the bests table. Copy it up one entry.
807 */
808 if (use_block >= be32_to_cpu(ltp->bestcount)) {
809 bestsp--;
810 memmove(&bestsp[0], &bestsp[1],
811 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
812 be32_add_cpu(<p->bestcount, 1);
813 xfs_dir3_leaf_log_tail(tp, lbp);
814 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
815 }
816 /*
817 * If we're filling in a previously empty block just log it.
818 */
819 else
820 xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
821 hdr = dbp->b_addr;
822 bf = dp->d_ops->data_bestfree_p(hdr);
823 bestsp[use_block] = bf[0].length;
824 grown = 1;
825 } else {
826 /*
827 * Already had space in some data block.
828 * Just read that one in.
829 */
830 error = xfs_dir3_data_read(tp, dp,
831 xfs_dir2_db_to_da(mp, use_block),
832 -1, &dbp);
833 if (error) {
834 xfs_trans_brelse(tp, lbp);
835 return error;
836 }
837 hdr = dbp->b_addr;
838 bf = dp->d_ops->data_bestfree_p(hdr);
839 grown = 0;
840 }
841 /*
842 * Point to the biggest freespace in our data block.
843 */
844 dup = (xfs_dir2_data_unused_t *)
845 ((char *)hdr + be16_to_cpu(bf[0].offset));
846 ASSERT(be16_to_cpu(dup->length) >= length);
847 needscan = needlog = 0;
848 /*
849 * Mark the initial part of our freespace in use for the new entry.
850 */
851 xfs_dir2_data_use_free(tp, dp, dbp, dup,
852 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
853 &needlog, &needscan);
854 /*
855 * Initialize our new entry (at last).
856 */
857 dep = (xfs_dir2_data_entry_t *)dup;
858 dep->inumber = cpu_to_be64(args->inumber);
859 dep->namelen = args->namelen;
860 memcpy(dep->name, args->name, dep->namelen);
861 dp->d_ops->data_put_ftype(dep, args->filetype);
862 tagp = dp->d_ops->data_entry_tag_p(dep);
863 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
864 /*
865 * Need to scan fix up the bestfree table.
866 */
867 if (needscan)
868 xfs_dir2_data_freescan(dp, hdr, &needlog);
869 /*
870 * Need to log the data block's header.
871 */
872 if (needlog)
873 xfs_dir2_data_log_header(tp, dp, dbp);
874 xfs_dir2_data_log_entry(tp, dp, dbp, dep);
875 /*
876 * If the bests table needs to be changed, do it.
877 * Log the change unless we've already done that.
878 */
879 if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
880 bestsp[use_block] = bf[0].length;
881 if (!grown)
882 xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
883 }
884
885 lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
886 highstale, &lfloglow, &lfloghigh);
887
888 /*
889 * Fill in the new leaf entry.
890 */
891 lep->hashval = cpu_to_be32(args->hashval);
892 lep->address = cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp, use_block,
893 be16_to_cpu(*tagp)));
894 /*
895 * Log the leaf fields and give up the buffers.
896 */
897 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
898 xfs_dir3_leaf_log_header(tp, dp, lbp);
899 xfs_dir3_leaf_log_ents(tp, dp, lbp, lfloglow, lfloghigh);
900 xfs_dir3_leaf_check(dp, lbp);
901 xfs_dir3_data_check(dp, dbp);
902 return 0;
903}
904
905/*
906 * Compact out any stale entries in the leaf.
907 * Log the header and changed leaf entries, if any.
908 */
909void
910xfs_dir3_leaf_compact(
911 xfs_da_args_t *args, /* operation arguments */
912 struct xfs_dir3_icleaf_hdr *leafhdr,
913 struct xfs_buf *bp) /* leaf buffer */
914{
915 int from; /* source leaf index */
916 xfs_dir2_leaf_t *leaf; /* leaf structure */
917 int loglow; /* first leaf entry to log */
918 int to; /* target leaf index */
919 struct xfs_dir2_leaf_entry *ents;
920 struct xfs_inode *dp = args->dp;
921
922 leaf = bp->b_addr;
923 if (!leafhdr->stale)
924 return;
925
926 /*
927 * Compress out the stale entries in place.
928 */
929 ents = dp->d_ops->leaf_ents_p(leaf);
930 for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
931 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
932 continue;
933 /*
934 * Only actually copy the entries that are different.
935 */
936 if (from > to) {
937 if (loglow == -1)
938 loglow = to;
939 ents[to] = ents[from];
940 }
941 to++;
942 }
943 /*
944 * Update and log the header, log the leaf entries.
945 */
946 ASSERT(leafhdr->stale == from - to);
947 leafhdr->count -= leafhdr->stale;
948 leafhdr->stale = 0;
949
950 dp->d_ops->leaf_hdr_to_disk(leaf, leafhdr);
951 xfs_dir3_leaf_log_header(args->trans, dp, bp);
952 if (loglow != -1)
953 xfs_dir3_leaf_log_ents(args->trans, dp, bp, loglow, to - 1);
954}
955
956/*
957 * Compact the leaf entries, removing stale ones.
958 * Leave one stale entry behind - the one closest to our
959 * insertion index - and the caller will shift that one to our insertion
960 * point later.
961 * Return new insertion index, where the remaining stale entry is,
962 * and leaf logging indices.
963 */
964void
965xfs_dir3_leaf_compact_x1(
966 struct xfs_dir3_icleaf_hdr *leafhdr,
967 struct xfs_dir2_leaf_entry *ents,
968 int *indexp, /* insertion index */
969 int *lowstalep, /* out: stale entry before us */
970 int *highstalep, /* out: stale entry after us */
971 int *lowlogp, /* out: low log index */
972 int *highlogp) /* out: high log index */
973{
974 int from; /* source copy index */
975 int highstale; /* stale entry at/after index */
976 int index; /* insertion index */
977 int keepstale; /* source index of kept stale */
978 int lowstale; /* stale entry before index */
979 int newindex=0; /* new insertion index */
980 int to; /* destination copy index */
981
982 ASSERT(leafhdr->stale > 1);
983 index = *indexp;
984
985 xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
986
987 /*
988 * Pick the better of lowstale and highstale.
989 */
990 if (lowstale >= 0 &&
991 (highstale == leafhdr->count ||
992 index - lowstale <= highstale - index))
993 keepstale = lowstale;
994 else
995 keepstale = highstale;
996 /*
997 * Copy the entries in place, removing all the stale entries
998 * except keepstale.
999 */
1000 for (from = to = 0; from < leafhdr->count; from++) {
1001 /*
1002 * Notice the new value of index.
1003 */
1004 if (index == from)
1005 newindex = to;
1006 if (from != keepstale &&
1007 ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1008 if (from == to)
1009 *lowlogp = to;
1010 continue;
1011 }
1012 /*
1013 * Record the new keepstale value for the insertion.
1014 */
1015 if (from == keepstale)
1016 lowstale = highstale = to;
1017 /*
1018 * Copy only the entries that have moved.
1019 */
1020 if (from > to)
1021 ents[to] = ents[from];
1022 to++;
1023 }
1024 ASSERT(from > to);
1025 /*
1026 * If the insertion point was past the last entry,
1027 * set the new insertion point accordingly.
1028 */
1029 if (index == from)
1030 newindex = to;
1031 *indexp = newindex;
1032 /*
1033 * Adjust the leaf header values.
1034 */
1035 leafhdr->count -= from - to;
1036 leafhdr->stale = 1;
1037 /*
1038 * Remember the low/high stale value only in the "right"
1039 * direction.
1040 */
1041 if (lowstale >= newindex)
1042 lowstale = -1;
1043 else
1044 highstale = leafhdr->count;
1045 *highlogp = leafhdr->count - 1;
1046 *lowstalep = lowstale;
1047 *highstalep = highstale;
1048}
1049
1050/*
1051 * Log the bests entries indicated from a leaf1 block.
1052 */
1053static void
1054xfs_dir3_leaf_log_bests(
1055 xfs_trans_t *tp, /* transaction pointer */
1056 struct xfs_buf *bp, /* leaf buffer */
1057 int first, /* first entry to log */
1058 int last) /* last entry to log */
1059{
1060 __be16 *firstb; /* pointer to first entry */
1061 __be16 *lastb; /* pointer to last entry */
1062 struct xfs_dir2_leaf *leaf = bp->b_addr;
1063 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1064
1065 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1066 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1067
1068 ltp = xfs_dir2_leaf_tail_p(tp->t_mountp, leaf);
1069 firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1070 lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1071 xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1072 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1073}
1074
1075/*
1076 * Log the leaf entries indicated from a leaf1 or leafn block.
1077 */
1078void
1079xfs_dir3_leaf_log_ents(
1080 struct xfs_trans *tp,
1081 struct xfs_inode *dp,
1082 struct xfs_buf *bp,
1083 int first,
1084 int last)
1085{
1086 xfs_dir2_leaf_entry_t *firstlep; /* pointer to first entry */
1087 xfs_dir2_leaf_entry_t *lastlep; /* pointer to last entry */
1088 struct xfs_dir2_leaf *leaf = bp->b_addr;
1089 struct xfs_dir2_leaf_entry *ents;
1090
1091 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1092 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1093 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1094 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1095
1096 ents = dp->d_ops->leaf_ents_p(leaf);
1097 firstlep = &ents[first];
1098 lastlep = &ents[last];
1099 xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1100 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1101}
1102
1103/*
1104 * Log the header of the leaf1 or leafn block.
1105 */
1106void
1107xfs_dir3_leaf_log_header(
1108 struct xfs_trans *tp,
1109 struct xfs_inode *dp,
1110 struct xfs_buf *bp)
1111{
1112 struct xfs_dir2_leaf *leaf = bp->b_addr;
1113
1114 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1115 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1116 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1117 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1118
1119 xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1120 dp->d_ops->leaf_hdr_size - 1);
1121}
1122
1123/*
1124 * Log the tail of the leaf1 block.
1125 */
1126STATIC void
1127xfs_dir3_leaf_log_tail(
1128 struct xfs_trans *tp,
1129 struct xfs_buf *bp)
1130{
1131 struct xfs_dir2_leaf *leaf = bp->b_addr;
1132 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1133 struct xfs_mount *mp = tp->t_mountp;
1134
1135 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1136 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1137 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1138 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1139
1140 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1141 xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1142 (uint)(mp->m_dirblksize - 1));
1143}
1144
1145/*
1146 * Look up the entry referred to by args in the leaf format directory.
1147 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1148 * is also used by the node-format code.
1149 */
1150int
1151xfs_dir2_leaf_lookup(
1152 xfs_da_args_t *args) /* operation arguments */
1153{
1154 struct xfs_buf *dbp; /* data block buffer */
1155 xfs_dir2_data_entry_t *dep; /* data block entry */
1156 xfs_inode_t *dp; /* incore directory inode */
1157 int error; /* error return code */
1158 int index; /* found entry index */
1159 struct xfs_buf *lbp; /* leaf buffer */
1160 xfs_dir2_leaf_t *leaf; /* leaf structure */
1161 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1162 xfs_trans_t *tp; /* transaction pointer */
1163 struct xfs_dir2_leaf_entry *ents;
1164
1165 trace_xfs_dir2_leaf_lookup(args);
1166
1167 /*
1168 * Look up name in the leaf block, returning both buffers and index.
1169 */
1170 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1171 return error;
1172 }
1173 tp = args->trans;
1174 dp = args->dp;
1175 xfs_dir3_leaf_check(dp, lbp);
1176 leaf = lbp->b_addr;
1177 ents = dp->d_ops->leaf_ents_p(leaf);
1178 /*
1179 * Get to the leaf entry and contained data entry address.
1180 */
1181 lep = &ents[index];
1182
1183 /*
1184 * Point to the data entry.
1185 */
1186 dep = (xfs_dir2_data_entry_t *)
1187 ((char *)dbp->b_addr +
1188 xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1189 /*
1190 * Return the found inode number & CI name if appropriate
1191 */
1192 args->inumber = be64_to_cpu(dep->inumber);
1193 args->filetype = dp->d_ops->data_get_ftype(dep);
1194 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1195 xfs_trans_brelse(tp, dbp);
1196 xfs_trans_brelse(tp, lbp);
1197 return XFS_ERROR(error);
1198}
1199
1200/*
1201 * Look up name/hash in the leaf block.
1202 * Fill in indexp with the found index, and dbpp with the data buffer.
1203 * If not found dbpp will be NULL, and ENOENT comes back.
1204 * lbpp will always be filled in with the leaf buffer unless there's an error.
1205 */
1206static int /* error */
1207xfs_dir2_leaf_lookup_int(
1208 xfs_da_args_t *args, /* operation arguments */
1209 struct xfs_buf **lbpp, /* out: leaf buffer */
1210 int *indexp, /* out: index in leaf block */
1211 struct xfs_buf **dbpp) /* out: data buffer */
1212{
1213 xfs_dir2_db_t curdb = -1; /* current data block number */
1214 struct xfs_buf *dbp = NULL; /* data buffer */
1215 xfs_dir2_data_entry_t *dep; /* data entry */
1216 xfs_inode_t *dp; /* incore directory inode */
1217 int error; /* error return code */
1218 int index; /* index in leaf block */
1219 struct xfs_buf *lbp; /* leaf buffer */
1220 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1221 xfs_dir2_leaf_t *leaf; /* leaf structure */
1222 xfs_mount_t *mp; /* filesystem mount point */
1223 xfs_dir2_db_t newdb; /* new data block number */
1224 xfs_trans_t *tp; /* transaction pointer */
1225 xfs_dir2_db_t cidb = -1; /* case match data block no. */
1226 enum xfs_dacmp cmp; /* name compare result */
1227 struct xfs_dir2_leaf_entry *ents;
1228 struct xfs_dir3_icleaf_hdr leafhdr;
1229
1230 dp = args->dp;
1231 tp = args->trans;
1232 mp = dp->i_mount;
1233
1234 error = xfs_dir3_leaf_read(tp, dp, mp->m_dirleafblk, -1, &lbp);
1235 if (error)
1236 return error;
1237
1238 *lbpp = lbp;
1239 leaf = lbp->b_addr;
1240 xfs_dir3_leaf_check(dp, lbp);
1241 ents = dp->d_ops->leaf_ents_p(leaf);
1242 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1243
1244 /*
1245 * Look for the first leaf entry with our hash value.
1246 */
1247 index = xfs_dir2_leaf_search_hash(args, lbp);
1248 /*
1249 * Loop over all the entries with the right hash value
1250 * looking to match the name.
1251 */
1252 for (lep = &ents[index];
1253 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1254 lep++, index++) {
1255 /*
1256 * Skip over stale leaf entries.
1257 */
1258 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1259 continue;
1260 /*
1261 * Get the new data block number.
1262 */
1263 newdb = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1264 /*
1265 * If it's not the same as the old data block number,
1266 * need to pitch the old one and read the new one.
1267 */
1268 if (newdb != curdb) {
1269 if (dbp)
1270 xfs_trans_brelse(tp, dbp);
1271 error = xfs_dir3_data_read(tp, dp,
1272 xfs_dir2_db_to_da(mp, newdb),
1273 -1, &dbp);
1274 if (error) {
1275 xfs_trans_brelse(tp, lbp);
1276 return error;
1277 }
1278 curdb = newdb;
1279 }
1280 /*
1281 * Point to the data entry.
1282 */
1283 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1284 xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1285 /*
1286 * Compare name and if it's an exact match, return the index
1287 * and buffer. If it's the first case-insensitive match, store
1288 * the index and buffer and continue looking for an exact match.
1289 */
1290 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1291 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1292 args->cmpresult = cmp;
1293 *indexp = index;
1294 /* case exact match: return the current buffer. */
1295 if (cmp == XFS_CMP_EXACT) {
1296 *dbpp = dbp;
1297 return 0;
1298 }
1299 cidb = curdb;
1300 }
1301 }
1302 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1303 /*
1304 * Here, we can only be doing a lookup (not a rename or remove).
1305 * If a case-insensitive match was found earlier, re-read the
1306 * appropriate data block if required and return it.
1307 */
1308 if (args->cmpresult == XFS_CMP_CASE) {
1309 ASSERT(cidb != -1);
1310 if (cidb != curdb) {
1311 xfs_trans_brelse(tp, dbp);
1312 error = xfs_dir3_data_read(tp, dp,
1313 xfs_dir2_db_to_da(mp, cidb),
1314 -1, &dbp);
1315 if (error) {
1316 xfs_trans_brelse(tp, lbp);
1317 return error;
1318 }
1319 }
1320 *dbpp = dbp;
1321 return 0;
1322 }
1323 /*
1324 * No match found, return ENOENT.
1325 */
1326 ASSERT(cidb == -1);
1327 if (dbp)
1328 xfs_trans_brelse(tp, dbp);
1329 xfs_trans_brelse(tp, lbp);
1330 return XFS_ERROR(ENOENT);
1331}
1332
1333/*
1334 * Remove an entry from a leaf format directory.
1335 */
1336int /* error */
1337xfs_dir2_leaf_removename(
1338 xfs_da_args_t *args) /* operation arguments */
1339{
1340 __be16 *bestsp; /* leaf block best freespace */
1341 xfs_dir2_data_hdr_t *hdr; /* data block header */
1342 xfs_dir2_db_t db; /* data block number */
1343 struct xfs_buf *dbp; /* data block buffer */
1344 xfs_dir2_data_entry_t *dep; /* data entry structure */
1345 xfs_inode_t *dp; /* incore directory inode */
1346 int error; /* error return code */
1347 xfs_dir2_db_t i; /* temporary data block # */
1348 int index; /* index into leaf entries */
1349 struct xfs_buf *lbp; /* leaf buffer */
1350 xfs_dir2_leaf_t *leaf; /* leaf structure */
1351 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1352 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1353 xfs_mount_t *mp; /* filesystem mount point */
1354 int needlog; /* need to log data header */
1355 int needscan; /* need to rescan data frees */
1356 xfs_dir2_data_off_t oldbest; /* old value of best free */
1357 xfs_trans_t *tp; /* transaction pointer */
1358 struct xfs_dir2_data_free *bf; /* bestfree table */
1359 struct xfs_dir2_leaf_entry *ents;
1360 struct xfs_dir3_icleaf_hdr leafhdr;
1361
1362 trace_xfs_dir2_leaf_removename(args);
1363
1364 /*
1365 * Lookup the leaf entry, get the leaf and data blocks read in.
1366 */
1367 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1368 return error;
1369 }
1370 dp = args->dp;
1371 tp = args->trans;
1372 mp = dp->i_mount;
1373 leaf = lbp->b_addr;
1374 hdr = dbp->b_addr;
1375 xfs_dir3_data_check(dp, dbp);
1376 bf = dp->d_ops->data_bestfree_p(hdr);
1377 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1378 ents = dp->d_ops->leaf_ents_p(leaf);
1379 /*
1380 * Point to the leaf entry, use that to point to the data entry.
1381 */
1382 lep = &ents[index];
1383 db = xfs_dir2_dataptr_to_db(mp, be32_to_cpu(lep->address));
1384 dep = (xfs_dir2_data_entry_t *)
1385 ((char *)hdr + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(lep->address)));
1386 needscan = needlog = 0;
1387 oldbest = be16_to_cpu(bf[0].length);
1388 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1389 bestsp = xfs_dir2_leaf_bests_p(ltp);
1390 ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1391 /*
1392 * Mark the former data entry unused.
1393 */
1394 xfs_dir2_data_make_free(tp, dp, dbp,
1395 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1396 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1397 /*
1398 * We just mark the leaf entry stale by putting a null in it.
1399 */
1400 leafhdr.stale++;
1401 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1402 xfs_dir3_leaf_log_header(tp, dp, lbp);
1403
1404 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1405 xfs_dir3_leaf_log_ents(tp, dp, lbp, index, index);
1406
1407 /*
1408 * Scan the freespace in the data block again if necessary,
1409 * log the data block header if necessary.
1410 */
1411 if (needscan)
1412 xfs_dir2_data_freescan(dp, hdr, &needlog);
1413 if (needlog)
1414 xfs_dir2_data_log_header(tp, dp, dbp);
1415 /*
1416 * If the longest freespace in the data block has changed,
1417 * put the new value in the bests table and log that.
1418 */
1419 if (be16_to_cpu(bf[0].length) != oldbest) {
1420 bestsp[db] = bf[0].length;
1421 xfs_dir3_leaf_log_bests(tp, lbp, db, db);
1422 }
1423 xfs_dir3_data_check(dp, dbp);
1424 /*
1425 * If the data block is now empty then get rid of the data block.
1426 */
1427 if (be16_to_cpu(bf[0].length) ==
1428 mp->m_dirblksize - dp->d_ops->data_entry_offset) {
1429 ASSERT(db != mp->m_dirdatablk);
1430 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1431 /*
1432 * Nope, can't get rid of it because it caused
1433 * allocation of a bmap btree block to do so.
1434 * Just go on, returning success, leaving the
1435 * empty block in place.
1436 */
1437 if (error == ENOSPC && args->total == 0)
1438 error = 0;
1439 xfs_dir3_leaf_check(dp, lbp);
1440 return error;
1441 }
1442 dbp = NULL;
1443 /*
1444 * If this is the last data block then compact the
1445 * bests table by getting rid of entries.
1446 */
1447 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1448 /*
1449 * Look for the last active entry (i).
1450 */
1451 for (i = db - 1; i > 0; i--) {
1452 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1453 break;
1454 }
1455 /*
1456 * Copy the table down so inactive entries at the
1457 * end are removed.
1458 */
1459 memmove(&bestsp[db - i], bestsp,
1460 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1461 be32_add_cpu(<p->bestcount, -(db - i));
1462 xfs_dir3_leaf_log_tail(tp, lbp);
1463 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1464 } else
1465 bestsp[db] = cpu_to_be16(NULLDATAOFF);
1466 }
1467 /*
1468 * If the data block was not the first one, drop it.
1469 */
1470 else if (db != mp->m_dirdatablk)
1471 dbp = NULL;
1472
1473 xfs_dir3_leaf_check(dp, lbp);
1474 /*
1475 * See if we can convert to block form.
1476 */
1477 return xfs_dir2_leaf_to_block(args, lbp, dbp);
1478}
1479
1480/*
1481 * Replace the inode number in a leaf format directory entry.
1482 */
1483int /* error */
1484xfs_dir2_leaf_replace(
1485 xfs_da_args_t *args) /* operation arguments */
1486{
1487 struct xfs_buf *dbp; /* data block buffer */
1488 xfs_dir2_data_entry_t *dep; /* data block entry */
1489 xfs_inode_t *dp; /* incore directory inode */
1490 int error; /* error return code */
1491 int index; /* index of leaf entry */
1492 struct xfs_buf *lbp; /* leaf buffer */
1493 xfs_dir2_leaf_t *leaf; /* leaf structure */
1494 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1495 xfs_trans_t *tp; /* transaction pointer */
1496 struct xfs_dir2_leaf_entry *ents;
1497
1498 trace_xfs_dir2_leaf_replace(args);
1499
1500 /*
1501 * Look up the entry.
1502 */
1503 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1504 return error;
1505 }
1506 dp = args->dp;
1507 leaf = lbp->b_addr;
1508 ents = dp->d_ops->leaf_ents_p(leaf);
1509 /*
1510 * Point to the leaf entry, get data address from it.
1511 */
1512 lep = &ents[index];
1513 /*
1514 * Point to the data entry.
1515 */
1516 dep = (xfs_dir2_data_entry_t *)
1517 ((char *)dbp->b_addr +
1518 xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
1519 ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1520 /*
1521 * Put the new inode number in, log it.
1522 */
1523 dep->inumber = cpu_to_be64(args->inumber);
1524 dp->d_ops->data_put_ftype(dep, args->filetype);
1525 tp = args->trans;
1526 xfs_dir2_data_log_entry(tp, dp, dbp, dep);
1527 xfs_dir3_leaf_check(dp, lbp);
1528 xfs_trans_brelse(tp, lbp);
1529 return 0;
1530}
1531
1532/*
1533 * Return index in the leaf block (lbp) which is either the first
1534 * one with this hash value, or if there are none, the insert point
1535 * for that hash value.
1536 */
1537int /* index value */
1538xfs_dir2_leaf_search_hash(
1539 xfs_da_args_t *args, /* operation arguments */
1540 struct xfs_buf *lbp) /* leaf buffer */
1541{
1542 xfs_dahash_t hash=0; /* hash from this entry */
1543 xfs_dahash_t hashwant; /* hash value looking for */
1544 int high; /* high leaf index */
1545 int low; /* low leaf index */
1546 xfs_dir2_leaf_t *leaf; /* leaf structure */
1547 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1548 int mid=0; /* current leaf index */
1549 struct xfs_dir2_leaf_entry *ents;
1550 struct xfs_dir3_icleaf_hdr leafhdr;
1551
1552 leaf = lbp->b_addr;
1553 ents = args->dp->d_ops->leaf_ents_p(leaf);
1554 args->dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1555
1556 /*
1557 * Note, the table cannot be empty, so we have to go through the loop.
1558 * Binary search the leaf entries looking for our hash value.
1559 */
1560 for (lep = ents, low = 0, high = leafhdr.count - 1,
1561 hashwant = args->hashval;
1562 low <= high; ) {
1563 mid = (low + high) >> 1;
1564 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1565 break;
1566 if (hash < hashwant)
1567 low = mid + 1;
1568 else
1569 high = mid - 1;
1570 }
1571 /*
1572 * Found one, back up through all the equal hash values.
1573 */
1574 if (hash == hashwant) {
1575 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1576 mid--;
1577 }
1578 }
1579 /*
1580 * Need to point to an entry higher than ours.
1581 */
1582 else if (hash < hashwant)
1583 mid++;
1584 return mid;
1585}
1586
1587/*
1588 * Trim off a trailing data block. We know it's empty since the leaf
1589 * freespace table says so.
1590 */
1591int /* error */
1592xfs_dir2_leaf_trim_data(
1593 xfs_da_args_t *args, /* operation arguments */
1594 struct xfs_buf *lbp, /* leaf buffer */
1595 xfs_dir2_db_t db) /* data block number */
1596{
1597 __be16 *bestsp; /* leaf bests table */
1598 struct xfs_buf *dbp; /* data block buffer */
1599 xfs_inode_t *dp; /* incore directory inode */
1600 int error; /* error return value */
1601 xfs_dir2_leaf_t *leaf; /* leaf structure */
1602 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1603 xfs_mount_t *mp; /* filesystem mount point */
1604 xfs_trans_t *tp; /* transaction pointer */
1605
1606 dp = args->dp;
1607 mp = dp->i_mount;
1608 tp = args->trans;
1609 /*
1610 * Read the offending data block. We need its buffer.
1611 */
1612 error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(mp, db), -1, &dbp);
1613 if (error)
1614 return error;
1615
1616 leaf = lbp->b_addr;
1617 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1618
1619#ifdef DEBUG
1620{
1621 struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1622 struct xfs_dir2_data_free *bf = dp->d_ops->data_bestfree_p(hdr);
1623
1624 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1625 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1626 ASSERT(be16_to_cpu(bf[0].length) ==
1627 mp->m_dirblksize - dp->d_ops->data_entry_offset);
1628 ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1629}
1630#endif
1631
1632 /*
1633 * Get rid of the data block.
1634 */
1635 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1636 ASSERT(error != ENOSPC);
1637 xfs_trans_brelse(tp, dbp);
1638 return error;
1639 }
1640 /*
1641 * Eliminate the last bests entry from the table.
1642 */
1643 bestsp = xfs_dir2_leaf_bests_p(ltp);
1644 be32_add_cpu(<p->bestcount, -1);
1645 memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1646 xfs_dir3_leaf_log_tail(tp, lbp);
1647 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1648 return 0;
1649}
1650
1651static inline size_t
1652xfs_dir3_leaf_size(
1653 struct xfs_dir3_icleaf_hdr *hdr,
1654 int counts)
1655{
1656 int entries;
1657 int hdrsize;
1658
1659 entries = hdr->count - hdr->stale;
1660 if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1661 hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1662 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1663 else
1664 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1665
1666 return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1667 + counts * sizeof(xfs_dir2_data_off_t)
1668 + sizeof(xfs_dir2_leaf_tail_t);
1669}
1670
1671/*
1672 * Convert node form directory to leaf form directory.
1673 * The root of the node form dir needs to already be a LEAFN block.
1674 * Just return if we can't do anything.
1675 */
1676int /* error */
1677xfs_dir2_node_to_leaf(
1678 xfs_da_state_t *state) /* directory operation state */
1679{
1680 xfs_da_args_t *args; /* operation arguments */
1681 xfs_inode_t *dp; /* incore directory inode */
1682 int error; /* error return code */
1683 struct xfs_buf *fbp; /* buffer for freespace block */
1684 xfs_fileoff_t fo; /* freespace file offset */
1685 xfs_dir2_free_t *free; /* freespace structure */
1686 struct xfs_buf *lbp; /* buffer for leaf block */
1687 xfs_dir2_leaf_tail_t *ltp; /* tail of leaf structure */
1688 xfs_dir2_leaf_t *leaf; /* leaf structure */
1689 xfs_mount_t *mp; /* filesystem mount point */
1690 int rval; /* successful free trim? */
1691 xfs_trans_t *tp; /* transaction pointer */
1692 struct xfs_dir3_icleaf_hdr leafhdr;
1693 struct xfs_dir3_icfree_hdr freehdr;
1694
1695 /*
1696 * There's more than a leaf level in the btree, so there must
1697 * be multiple leafn blocks. Give up.
1698 */
1699 if (state->path.active > 1)
1700 return 0;
1701 args = state->args;
1702
1703 trace_xfs_dir2_node_to_leaf(args);
1704
1705 mp = state->mp;
1706 dp = args->dp;
1707 tp = args->trans;
1708 /*
1709 * Get the last offset in the file.
1710 */
1711 if ((error = xfs_bmap_last_offset(tp, dp, &fo, XFS_DATA_FORK))) {
1712 return error;
1713 }
1714 fo -= mp->m_dirblkfsbs;
1715 /*
1716 * If there are freespace blocks other than the first one,
1717 * take this opportunity to remove trailing empty freespace blocks
1718 * that may have been left behind during no-space-reservation
1719 * operations.
1720 */
1721 while (fo > mp->m_dirfreeblk) {
1722 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1723 return error;
1724 }
1725 if (rval)
1726 fo -= mp->m_dirblkfsbs;
1727 else
1728 return 0;
1729 }
1730 /*
1731 * Now find the block just before the freespace block.
1732 */
1733 if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1734 return error;
1735 }
1736 /*
1737 * If it's not the single leaf block, give up.
1738 */
1739 if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + mp->m_dirblksize)
1740 return 0;
1741 lbp = state->path.blk[0].bp;
1742 leaf = lbp->b_addr;
1743 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1744
1745 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1746 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1747
1748 /*
1749 * Read the freespace block.
1750 */
1751 error = xfs_dir2_free_read(tp, dp, mp->m_dirfreeblk, &fbp);
1752 if (error)
1753 return error;
1754 free = fbp->b_addr;
1755 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1756
1757 ASSERT(!freehdr.firstdb);
1758
1759 /*
1760 * Now see if the leafn and free data will fit in a leaf1.
1761 * If not, release the buffer and give up.
1762 */
1763 if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > mp->m_dirblksize) {
1764 xfs_trans_brelse(tp, fbp);
1765 return 0;
1766 }
1767
1768 /*
1769 * If the leaf has any stale entries in it, compress them out.
1770 */
1771 if (leafhdr.stale)
1772 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1773
1774 lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1775 xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1776 leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1777 ? XFS_DIR2_LEAF1_MAGIC
1778 : XFS_DIR3_LEAF1_MAGIC;
1779
1780 /*
1781 * Set up the leaf tail from the freespace block.
1782 */
1783 ltp = xfs_dir2_leaf_tail_p(mp, leaf);
1784 ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1785
1786 /*
1787 * Set up the leaf bests table.
1788 */
1789 memcpy(xfs_dir2_leaf_bests_p(ltp), dp->d_ops->free_bests_p(free),
1790 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1791
1792 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1793 xfs_dir3_leaf_log_header(tp, dp, lbp);
1794 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1795 xfs_dir3_leaf_log_tail(tp, lbp);
1796 xfs_dir3_leaf_check(dp, lbp);
1797
1798 /*
1799 * Get rid of the freespace block.
1800 */
1801 error = xfs_dir2_shrink_inode(args, XFS_DIR2_FREE_FIRSTDB(mp), fbp);
1802 if (error) {
1803 /*
1804 * This can't fail here because it can only happen when
1805 * punching out the middle of an extent, and this is an
1806 * isolated block.
1807 */
1808 ASSERT(error != ENOSPC);
1809 return error;
1810 }
1811 fbp = NULL;
1812 /*
1813 * Now see if we can convert the single-leaf directory
1814 * down to a block form directory.
1815 * This routine always kills the dabuf for the leaf, so
1816 * eliminate it from the path.
1817 */
1818 error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1819 state->path.blk[0].bp = NULL;
1820 return error;
1821}