]> git.ipfire.org Git - thirdparty/xfsprogs-dev.git/blob - libxfs/xfs_dir2_leaf.c
xfs: queue deferred rmap ops for cow staging extent alloc/free in the right order
[thirdparty/xfsprogs-dev.git] / libxfs / xfs_dir2_leaf.c
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 "libxfs_priv.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_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
28 #include "xfs_bmap.h"
29 #include "xfs_dir2.h"
30 #include "xfs_dir2_priv.h"
31 #include "xfs_trace.h"
32 #include "xfs_trans.h"
33 #include "xfs_cksum.h"
34
35 /*
36 * Local function declarations.
37 */
38 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
39 int *indexp, struct xfs_buf **dbpp);
40 static void xfs_dir3_leaf_log_bests(struct xfs_da_args *args,
41 struct xfs_buf *bp, int first, int last);
42 static void xfs_dir3_leaf_log_tail(struct xfs_da_args *args,
43 struct xfs_buf *bp);
44
45 /*
46 * Check the internal consistency of a leaf1 block.
47 * Pop an assert if something is wrong.
48 */
49 #ifdef DEBUG
50 #define xfs_dir3_leaf_check(dp, bp) \
51 do { \
52 if (!xfs_dir3_leaf1_check((dp), (bp))) \
53 ASSERT(0); \
54 } while (0);
55
56 STATIC bool
57 xfs_dir3_leaf1_check(
58 struct xfs_inode *dp,
59 struct xfs_buf *bp)
60 {
61 struct xfs_dir2_leaf *leaf = bp->b_addr;
62 struct xfs_dir3_icleaf_hdr leafhdr;
63
64 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
65
66 if (leafhdr.magic == XFS_DIR3_LEAF1_MAGIC) {
67 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
68 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
69 return false;
70 } else if (leafhdr.magic != XFS_DIR2_LEAF1_MAGIC)
71 return false;
72
73 return xfs_dir3_leaf_check_int(dp->i_mount, dp, &leafhdr, leaf);
74 }
75 #else
76 #define xfs_dir3_leaf_check(dp, bp)
77 #endif
78
79 bool
80 xfs_dir3_leaf_check_int(
81 struct xfs_mount *mp,
82 struct xfs_inode *dp,
83 struct xfs_dir3_icleaf_hdr *hdr,
84 struct xfs_dir2_leaf *leaf)
85 {
86 struct xfs_dir2_leaf_entry *ents;
87 xfs_dir2_leaf_tail_t *ltp;
88 int stale;
89 int i;
90 const struct xfs_dir_ops *ops;
91 struct xfs_dir3_icleaf_hdr leafhdr;
92 struct xfs_da_geometry *geo = mp->m_dir_geo;
93
94 /*
95 * we can be passed a null dp here from a verifier, so we need to go the
96 * hard way to get them.
97 */
98 ops = xfs_dir_get_ops(mp, dp);
99
100 if (!hdr) {
101 ops->leaf_hdr_from_disk(&leafhdr, leaf);
102 hdr = &leafhdr;
103 }
104
105 ents = ops->leaf_ents_p(leaf);
106 ltp = xfs_dir2_leaf_tail_p(geo, leaf);
107
108 /*
109 * XXX (dgc): This value is not restrictive enough.
110 * Should factor in the size of the bests table as well.
111 * We can deduce a value for that from di_size.
112 */
113 if (hdr->count > ops->leaf_max_ents(geo))
114 return false;
115
116 /* Leaves and bests don't overlap in leaf format. */
117 if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
118 hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
119 (char *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
120 return false;
121
122 /* Check hash value order, count stale entries. */
123 for (i = stale = 0; i < hdr->count; i++) {
124 if (i + 1 < hdr->count) {
125 if (be32_to_cpu(ents[i].hashval) >
126 be32_to_cpu(ents[i + 1].hashval))
127 return false;
128 }
129 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
130 stale++;
131 }
132 if (hdr->stale != stale)
133 return false;
134 return true;
135 }
136
137 /*
138 * We verify the magic numbers before decoding the leaf header so that on debug
139 * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
140 * to incorrect magic numbers.
141 */
142 static bool
143 xfs_dir3_leaf_verify(
144 struct xfs_buf *bp,
145 uint16_t magic)
146 {
147 struct xfs_mount *mp = bp->b_target->bt_mount;
148 struct xfs_dir2_leaf *leaf = bp->b_addr;
149
150 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
151
152 if (xfs_sb_version_hascrc(&mp->m_sb)) {
153 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
154 uint16_t magic3;
155
156 magic3 = (magic == XFS_DIR2_LEAF1_MAGIC) ? XFS_DIR3_LEAF1_MAGIC
157 : XFS_DIR3_LEAFN_MAGIC;
158
159 if (leaf3->info.hdr.magic != cpu_to_be16(magic3))
160 return false;
161 if (!uuid_equal(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid))
162 return false;
163 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
164 return false;
165 if (!xfs_log_check_lsn(mp, be64_to_cpu(leaf3->info.lsn)))
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
175 static 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
192 static 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
216 static void
217 xfs_dir3_leaf1_read_verify(
218 struct xfs_buf *bp)
219 {
220 __read_verify(bp, XFS_DIR2_LEAF1_MAGIC);
221 }
222
223 static void
224 xfs_dir3_leaf1_write_verify(
225 struct xfs_buf *bp)
226 {
227 __write_verify(bp, XFS_DIR2_LEAF1_MAGIC);
228 }
229
230 static void
231 xfs_dir3_leafn_read_verify(
232 struct xfs_buf *bp)
233 {
234 __read_verify(bp, XFS_DIR2_LEAFN_MAGIC);
235 }
236
237 static void
238 xfs_dir3_leafn_write_verify(
239 struct xfs_buf *bp)
240 {
241 __write_verify(bp, XFS_DIR2_LEAFN_MAGIC);
242 }
243
244 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
245 .name = "xfs_dir3_leaf1",
246 .verify_read = xfs_dir3_leaf1_read_verify,
247 .verify_write = xfs_dir3_leaf1_write_verify,
248 };
249
250 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
251 .name = "xfs_dir3_leafn",
252 .verify_read = xfs_dir3_leafn_read_verify,
253 .verify_write = xfs_dir3_leafn_write_verify,
254 };
255
256 int
257 xfs_dir3_leaf_read(
258 struct xfs_trans *tp,
259 struct xfs_inode *dp,
260 xfs_dablk_t fbno,
261 xfs_daddr_t mappedbno,
262 struct xfs_buf **bpp)
263 {
264 int err;
265
266 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
267 XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
268 if (!err && tp && *bpp)
269 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
270 return err;
271 }
272
273 int
274 xfs_dir3_leafn_read(
275 struct xfs_trans *tp,
276 struct xfs_inode *dp,
277 xfs_dablk_t fbno,
278 xfs_daddr_t mappedbno,
279 struct xfs_buf **bpp)
280 {
281 int err;
282
283 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
284 XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
285 if (!err && tp && *bpp)
286 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
287 return err;
288 }
289
290 /*
291 * Initialize a new leaf block, leaf1 or leafn magic accepted.
292 */
293 static void
294 xfs_dir3_leaf_init(
295 struct xfs_mount *mp,
296 struct xfs_trans *tp,
297 struct xfs_buf *bp,
298 xfs_ino_t owner,
299 uint16_t type)
300 {
301 struct xfs_dir2_leaf *leaf = bp->b_addr;
302
303 ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
304
305 if (xfs_sb_version_hascrc(&mp->m_sb)) {
306 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
307
308 memset(leaf3, 0, sizeof(*leaf3));
309
310 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
311 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
312 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
313 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
314 leaf3->info.owner = cpu_to_be64(owner);
315 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid);
316 } else {
317 memset(leaf, 0, sizeof(*leaf));
318 leaf->hdr.info.magic = cpu_to_be16(type);
319 }
320
321 /*
322 * If it's a leaf-format directory initialize the tail.
323 * Caller is responsible for initialising the bests table.
324 */
325 if (type == XFS_DIR2_LEAF1_MAGIC) {
326 struct xfs_dir2_leaf_tail *ltp;
327
328 ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
329 ltp->bestcount = 0;
330 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
331 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
332 } else {
333 bp->b_ops = &xfs_dir3_leafn_buf_ops;
334 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
335 }
336 }
337
338 int
339 xfs_dir3_leaf_get_buf(
340 xfs_da_args_t *args,
341 xfs_dir2_db_t bno,
342 struct xfs_buf **bpp,
343 uint16_t magic)
344 {
345 struct xfs_inode *dp = args->dp;
346 struct xfs_trans *tp = args->trans;
347 struct xfs_mount *mp = dp->i_mount;
348 struct xfs_buf *bp;
349 int error;
350
351 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
352 ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
353 bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
354
355 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
356 -1, &bp, XFS_DATA_FORK);
357 if (error)
358 return error;
359
360 xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
361 xfs_dir3_leaf_log_header(args, bp);
362 if (magic == XFS_DIR2_LEAF1_MAGIC)
363 xfs_dir3_leaf_log_tail(args, bp);
364 *bpp = bp;
365 return 0;
366 }
367
368 /*
369 * Convert a block form directory to a leaf form directory.
370 */
371 int /* error */
372 xfs_dir2_block_to_leaf(
373 xfs_da_args_t *args, /* operation arguments */
374 struct xfs_buf *dbp) /* input block's buffer */
375 {
376 __be16 *bestsp; /* leaf's bestsp entries */
377 xfs_dablk_t blkno; /* leaf block's bno */
378 xfs_dir2_data_hdr_t *hdr; /* block header */
379 xfs_dir2_leaf_entry_t *blp; /* block's leaf entries */
380 xfs_dir2_block_tail_t *btp; /* block's tail */
381 xfs_inode_t *dp; /* incore directory inode */
382 int error; /* error return code */
383 struct xfs_buf *lbp; /* leaf block's buffer */
384 xfs_dir2_db_t ldb; /* leaf block's bno */
385 xfs_dir2_leaf_t *leaf; /* leaf structure */
386 xfs_dir2_leaf_tail_t *ltp; /* leaf's tail */
387 int needlog; /* need to log block header */
388 int needscan; /* need to rescan bestfree */
389 xfs_trans_t *tp; /* transaction pointer */
390 struct xfs_dir2_data_free *bf;
391 struct xfs_dir2_leaf_entry *ents;
392 struct xfs_dir3_icleaf_hdr leafhdr;
393
394 trace_xfs_dir2_block_to_leaf(args);
395
396 dp = args->dp;
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(args->geo, blkno);
407 ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
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(args->geo, 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(args, 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(args, 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(args, dbp,
445 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
446 (xfs_dir2_data_aoff_t)((char *)hdr + args->geo->blksize -
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(args->geo, 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(args, dbp);
473 xfs_dir3_leaf_check(dp, lbp);
474 xfs_dir3_data_check(dp, dbp);
475 xfs_dir3_leaf_log_bests(args, lbp, 0, 0);
476 return 0;
477 }
478
479 STATIC void
480 xfs_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
510 struct xfs_dir2_leaf_entry *
511 xfs_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 */
603 int /* error */
604 xfs_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 int needbytes; /* leaf block bytes needed */
628 int needlog; /* need to log data header */
629 int needscan; /* need to rescan data free */
630 __be16 *tagp; /* end of data entry */
631 xfs_trans_t *tp; /* transaction pointer */
632 xfs_dir2_db_t use_block; /* data block number */
633 struct xfs_dir2_data_free *bf; /* bestfree table */
634 struct xfs_dir2_leaf_entry *ents;
635 struct xfs_dir3_icleaf_hdr leafhdr;
636
637 trace_xfs_dir2_leaf_addname(args);
638
639 dp = args->dp;
640 tp = args->trans;
641
642 error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
643 if (error)
644 return error;
645
646 /*
647 * Look up the entry by hash value and name.
648 * We know it's not there, our caller has already done a lookup.
649 * So the index is of the entry to insert in front of.
650 * But if there are dup hash values the index is of the first of those.
651 */
652 index = xfs_dir2_leaf_search_hash(args, lbp);
653 leaf = lbp->b_addr;
654 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
655 ents = dp->d_ops->leaf_ents_p(leaf);
656 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
657 bestsp = xfs_dir2_leaf_bests_p(ltp);
658 length = dp->d_ops->data_entsize(args->namelen);
659
660 /*
661 * See if there are any entries with the same hash value
662 * and space in their block for the new entry.
663 * This is good because it puts multiple same-hash value entries
664 * in a data block, improving the lookup of those entries.
665 */
666 for (use_block = -1, lep = &ents[index];
667 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
668 index++, lep++) {
669 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
670 continue;
671 i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
672 ASSERT(i < be32_to_cpu(ltp->bestcount));
673 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
674 if (be16_to_cpu(bestsp[i]) >= length) {
675 use_block = i;
676 break;
677 }
678 }
679 /*
680 * Didn't find a block yet, linear search all the data blocks.
681 */
682 if (use_block == -1) {
683 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
684 /*
685 * Remember a block we see that's missing.
686 */
687 if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
688 use_block == -1)
689 use_block = i;
690 else if (be16_to_cpu(bestsp[i]) >= length) {
691 use_block = i;
692 break;
693 }
694 }
695 }
696 /*
697 * How many bytes do we need in the leaf block?
698 */
699 needbytes = 0;
700 if (!leafhdr.stale)
701 needbytes += sizeof(xfs_dir2_leaf_entry_t);
702 if (use_block == -1)
703 needbytes += sizeof(xfs_dir2_data_off_t);
704
705 /*
706 * Now kill use_block if it refers to a missing block, so we
707 * can use it as an indication of allocation needed.
708 */
709 if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
710 use_block = -1;
711 /*
712 * If we don't have enough free bytes but we can make enough
713 * by compacting out stale entries, we'll do that.
714 */
715 if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
716 leafhdr.stale > 1)
717 compact = 1;
718
719 /*
720 * Otherwise if we don't have enough free bytes we need to
721 * convert to node form.
722 */
723 else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
724 /*
725 * Just checking or no space reservation, give up.
726 */
727 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
728 args->total == 0) {
729 xfs_trans_brelse(tp, lbp);
730 return -ENOSPC;
731 }
732 /*
733 * Convert to node form.
734 */
735 error = xfs_dir2_leaf_to_node(args, lbp);
736 if (error)
737 return error;
738 /*
739 * Then add the new entry.
740 */
741 return xfs_dir2_node_addname(args);
742 }
743 /*
744 * Otherwise it will fit without compaction.
745 */
746 else
747 compact = 0;
748 /*
749 * If just checking, then it will fit unless we needed to allocate
750 * a new data block.
751 */
752 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
753 xfs_trans_brelse(tp, lbp);
754 return use_block == -1 ? -ENOSPC : 0;
755 }
756 /*
757 * If no allocations are allowed, return now before we've
758 * changed anything.
759 */
760 if (args->total == 0 && use_block == -1) {
761 xfs_trans_brelse(tp, lbp);
762 return -ENOSPC;
763 }
764 /*
765 * Need to compact the leaf entries, removing stale ones.
766 * Leave one stale entry behind - the one closest to our
767 * insertion index - and we'll shift that one to our insertion
768 * point later.
769 */
770 if (compact) {
771 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
772 &highstale, &lfloglow, &lfloghigh);
773 }
774 /*
775 * There are stale entries, so we'll need log-low and log-high
776 * impossibly bad values later.
777 */
778 else if (leafhdr.stale) {
779 lfloglow = leafhdr.count;
780 lfloghigh = -1;
781 }
782 /*
783 * If there was no data block space found, we need to allocate
784 * a new one.
785 */
786 if (use_block == -1) {
787 /*
788 * Add the new data block.
789 */
790 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
791 &use_block))) {
792 xfs_trans_brelse(tp, lbp);
793 return error;
794 }
795 /*
796 * Initialize the block.
797 */
798 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
799 xfs_trans_brelse(tp, lbp);
800 return error;
801 }
802 /*
803 * If we're adding a new data block on the end we need to
804 * extend the bests table. Copy it up one entry.
805 */
806 if (use_block >= be32_to_cpu(ltp->bestcount)) {
807 bestsp--;
808 memmove(&bestsp[0], &bestsp[1],
809 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
810 be32_add_cpu(&ltp->bestcount, 1);
811 xfs_dir3_leaf_log_tail(args, lbp);
812 xfs_dir3_leaf_log_bests(args, lbp, 0,
813 be32_to_cpu(ltp->bestcount) - 1);
814 }
815 /*
816 * If we're filling in a previously empty block just log it.
817 */
818 else
819 xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
820 hdr = dbp->b_addr;
821 bf = dp->d_ops->data_bestfree_p(hdr);
822 bestsp[use_block] = bf[0].length;
823 grown = 1;
824 } else {
825 /*
826 * Already had space in some data block.
827 * Just read that one in.
828 */
829 error = xfs_dir3_data_read(tp, dp,
830 xfs_dir2_db_to_da(args->geo, use_block),
831 -1, &dbp);
832 if (error) {
833 xfs_trans_brelse(tp, lbp);
834 return error;
835 }
836 hdr = dbp->b_addr;
837 bf = dp->d_ops->data_bestfree_p(hdr);
838 grown = 0;
839 }
840 /*
841 * Point to the biggest freespace in our data block.
842 */
843 dup = (xfs_dir2_data_unused_t *)
844 ((char *)hdr + be16_to_cpu(bf[0].offset));
845 ASSERT(be16_to_cpu(dup->length) >= length);
846 needscan = needlog = 0;
847 /*
848 * Mark the initial part of our freespace in use for the new entry.
849 */
850 xfs_dir2_data_use_free(args, dbp, dup,
851 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
852 &needlog, &needscan);
853 /*
854 * Initialize our new entry (at last).
855 */
856 dep = (xfs_dir2_data_entry_t *)dup;
857 dep->inumber = cpu_to_be64(args->inumber);
858 dep->namelen = args->namelen;
859 memcpy(dep->name, args->name, dep->namelen);
860 dp->d_ops->data_put_ftype(dep, args->filetype);
861 tagp = dp->d_ops->data_entry_tag_p(dep);
862 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
863 /*
864 * Need to scan fix up the bestfree table.
865 */
866 if (needscan)
867 xfs_dir2_data_freescan(dp, hdr, &needlog);
868 /*
869 * Need to log the data block's header.
870 */
871 if (needlog)
872 xfs_dir2_data_log_header(args, dbp);
873 xfs_dir2_data_log_entry(args, dbp, dep);
874 /*
875 * If the bests table needs to be changed, do it.
876 * Log the change unless we've already done that.
877 */
878 if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
879 bestsp[use_block] = bf[0].length;
880 if (!grown)
881 xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
882 }
883
884 lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
885 highstale, &lfloglow, &lfloghigh);
886
887 /*
888 * Fill in the new leaf entry.
889 */
890 lep->hashval = cpu_to_be32(args->hashval);
891 lep->address = cpu_to_be32(
892 xfs_dir2_db_off_to_dataptr(args->geo, 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(args, lbp);
899 xfs_dir3_leaf_log_ents(args, 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 */
909 void
910 xfs_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, bp);
952 if (loglow != -1)
953 xfs_dir3_leaf_log_ents(args, 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 */
964 void
965 xfs_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 */
1053 static void
1054 xfs_dir3_leaf_log_bests(
1055 struct xfs_da_args *args,
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(args->geo, leaf);
1069 firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1070 lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1071 xfs_trans_log_buf(args->trans, bp,
1072 (uint)((char *)firstb - (char *)leaf),
1073 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1074 }
1075
1076 /*
1077 * Log the leaf entries indicated from a leaf1 or leafn block.
1078 */
1079 void
1080 xfs_dir3_leaf_log_ents(
1081 struct xfs_da_args *args,
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 = args->dp->d_ops->leaf_ents_p(leaf);
1097 firstlep = &ents[first];
1098 lastlep = &ents[last];
1099 xfs_trans_log_buf(args->trans, bp,
1100 (uint)((char *)firstlep - (char *)leaf),
1101 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1102 }
1103
1104 /*
1105 * Log the header of the leaf1 or leafn block.
1106 */
1107 void
1108 xfs_dir3_leaf_log_header(
1109 struct xfs_da_args *args,
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(args->trans, bp,
1120 (uint)((char *)&leaf->hdr - (char *)leaf),
1121 args->dp->d_ops->leaf_hdr_size - 1);
1122 }
1123
1124 /*
1125 * Log the tail of the leaf1 block.
1126 */
1127 STATIC void
1128 xfs_dir3_leaf_log_tail(
1129 struct xfs_da_args *args,
1130 struct xfs_buf *bp)
1131 {
1132 struct xfs_dir2_leaf *leaf = bp->b_addr;
1133 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
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(args->geo, leaf);
1141 xfs_trans_log_buf(args->trans, bp, (uint)((char *)ltp - (char *)leaf),
1142 (uint)(args->geo->blksize - 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 */
1150 int
1151 xfs_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(args->geo, 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 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 */
1206 static int /* error */
1207 xfs_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, args->geo->leafblk, -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(args->geo,
1264 be32_to_cpu(lep->address));
1265 /*
1266 * If it's not the same as the old data block number,
1267 * need to pitch the old one and read the new one.
1268 */
1269 if (newdb != curdb) {
1270 if (dbp)
1271 xfs_trans_brelse(tp, dbp);
1272 error = xfs_dir3_data_read(tp, dp,
1273 xfs_dir2_db_to_da(args->geo, newdb),
1274 -1, &dbp);
1275 if (error) {
1276 xfs_trans_brelse(tp, lbp);
1277 return error;
1278 }
1279 curdb = newdb;
1280 }
1281 /*
1282 * Point to the data entry.
1283 */
1284 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1285 xfs_dir2_dataptr_to_off(args->geo,
1286 be32_to_cpu(lep->address)));
1287 /*
1288 * Compare name and if it's an exact match, return the index
1289 * and buffer. If it's the first case-insensitive match, store
1290 * the index and buffer and continue looking for an exact match.
1291 */
1292 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1293 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1294 args->cmpresult = cmp;
1295 *indexp = index;
1296 /* case exact match: return the current buffer. */
1297 if (cmp == XFS_CMP_EXACT) {
1298 *dbpp = dbp;
1299 return 0;
1300 }
1301 cidb = curdb;
1302 }
1303 }
1304 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1305 /*
1306 * Here, we can only be doing a lookup (not a rename or remove).
1307 * If a case-insensitive match was found earlier, re-read the
1308 * appropriate data block if required and return it.
1309 */
1310 if (args->cmpresult == XFS_CMP_CASE) {
1311 ASSERT(cidb != -1);
1312 if (cidb != curdb) {
1313 xfs_trans_brelse(tp, dbp);
1314 error = xfs_dir3_data_read(tp, dp,
1315 xfs_dir2_db_to_da(args->geo, cidb),
1316 -1, &dbp);
1317 if (error) {
1318 xfs_trans_brelse(tp, lbp);
1319 return error;
1320 }
1321 }
1322 *dbpp = dbp;
1323 return 0;
1324 }
1325 /*
1326 * No match found, return -ENOENT.
1327 */
1328 ASSERT(cidb == -1);
1329 if (dbp)
1330 xfs_trans_brelse(tp, dbp);
1331 xfs_trans_brelse(tp, lbp);
1332 return -ENOENT;
1333 }
1334
1335 /*
1336 * Remove an entry from a leaf format directory.
1337 */
1338 int /* error */
1339 xfs_dir2_leaf_removename(
1340 xfs_da_args_t *args) /* operation arguments */
1341 {
1342 __be16 *bestsp; /* leaf block best freespace */
1343 xfs_dir2_data_hdr_t *hdr; /* data block header */
1344 xfs_dir2_db_t db; /* data block number */
1345 struct xfs_buf *dbp; /* data block buffer */
1346 xfs_dir2_data_entry_t *dep; /* data entry structure */
1347 xfs_inode_t *dp; /* incore directory inode */
1348 int error; /* error return code */
1349 xfs_dir2_db_t i; /* temporary data block # */
1350 int index; /* index into leaf entries */
1351 struct xfs_buf *lbp; /* leaf buffer */
1352 xfs_dir2_leaf_t *leaf; /* leaf structure */
1353 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1354 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1355 int needlog; /* need to log data header */
1356 int needscan; /* need to rescan data frees */
1357 xfs_dir2_data_off_t oldbest; /* old value of best free */
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 leaf = lbp->b_addr;
1372 hdr = dbp->b_addr;
1373 xfs_dir3_data_check(dp, dbp);
1374 bf = dp->d_ops->data_bestfree_p(hdr);
1375 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1376 ents = dp->d_ops->leaf_ents_p(leaf);
1377 /*
1378 * Point to the leaf entry, use that to point to the data entry.
1379 */
1380 lep = &ents[index];
1381 db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1382 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1383 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1384 needscan = needlog = 0;
1385 oldbest = be16_to_cpu(bf[0].length);
1386 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1387 bestsp = xfs_dir2_leaf_bests_p(ltp);
1388 ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1389 /*
1390 * Mark the former data entry unused.
1391 */
1392 xfs_dir2_data_make_free(args, dbp,
1393 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1394 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1395 /*
1396 * We just mark the leaf entry stale by putting a null in it.
1397 */
1398 leafhdr.stale++;
1399 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1400 xfs_dir3_leaf_log_header(args, lbp);
1401
1402 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1403 xfs_dir3_leaf_log_ents(args, lbp, index, index);
1404
1405 /*
1406 * Scan the freespace in the data block again if necessary,
1407 * log the data block header if necessary.
1408 */
1409 if (needscan)
1410 xfs_dir2_data_freescan(dp, hdr, &needlog);
1411 if (needlog)
1412 xfs_dir2_data_log_header(args, dbp);
1413 /*
1414 * If the longest freespace in the data block has changed,
1415 * put the new value in the bests table and log that.
1416 */
1417 if (be16_to_cpu(bf[0].length) != oldbest) {
1418 bestsp[db] = bf[0].length;
1419 xfs_dir3_leaf_log_bests(args, lbp, db, db);
1420 }
1421 xfs_dir3_data_check(dp, dbp);
1422 /*
1423 * If the data block is now empty then get rid of the data block.
1424 */
1425 if (be16_to_cpu(bf[0].length) ==
1426 args->geo->blksize - dp->d_ops->data_entry_offset) {
1427 ASSERT(db != args->geo->datablk);
1428 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1429 /*
1430 * Nope, can't get rid of it because it caused
1431 * allocation of a bmap btree block to do so.
1432 * Just go on, returning success, leaving the
1433 * empty block in place.
1434 */
1435 if (error == -ENOSPC && args->total == 0)
1436 error = 0;
1437 xfs_dir3_leaf_check(dp, lbp);
1438 return error;
1439 }
1440 dbp = NULL;
1441 /*
1442 * If this is the last data block then compact the
1443 * bests table by getting rid of entries.
1444 */
1445 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1446 /*
1447 * Look for the last active entry (i).
1448 */
1449 for (i = db - 1; i > 0; i--) {
1450 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1451 break;
1452 }
1453 /*
1454 * Copy the table down so inactive entries at the
1455 * end are removed.
1456 */
1457 memmove(&bestsp[db - i], bestsp,
1458 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1459 be32_add_cpu(&ltp->bestcount, -(db - i));
1460 xfs_dir3_leaf_log_tail(args, lbp);
1461 xfs_dir3_leaf_log_bests(args, lbp, 0,
1462 be32_to_cpu(ltp->bestcount) - 1);
1463 } else
1464 bestsp[db] = cpu_to_be16(NULLDATAOFF);
1465 }
1466 /*
1467 * If the data block was not the first one, drop it.
1468 */
1469 else if (db != args->geo->datablk)
1470 dbp = NULL;
1471
1472 xfs_dir3_leaf_check(dp, lbp);
1473 /*
1474 * See if we can convert to block form.
1475 */
1476 return xfs_dir2_leaf_to_block(args, lbp, dbp);
1477 }
1478
1479 /*
1480 * Replace the inode number in a leaf format directory entry.
1481 */
1482 int /* error */
1483 xfs_dir2_leaf_replace(
1484 xfs_da_args_t *args) /* operation arguments */
1485 {
1486 struct xfs_buf *dbp; /* data block buffer */
1487 xfs_dir2_data_entry_t *dep; /* data block entry */
1488 xfs_inode_t *dp; /* incore directory inode */
1489 int error; /* error return code */
1490 int index; /* index of leaf entry */
1491 struct xfs_buf *lbp; /* leaf buffer */
1492 xfs_dir2_leaf_t *leaf; /* leaf structure */
1493 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1494 xfs_trans_t *tp; /* transaction pointer */
1495 struct xfs_dir2_leaf_entry *ents;
1496
1497 trace_xfs_dir2_leaf_replace(args);
1498
1499 /*
1500 * Look up the entry.
1501 */
1502 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1503 return error;
1504 }
1505 dp = args->dp;
1506 leaf = lbp->b_addr;
1507 ents = dp->d_ops->leaf_ents_p(leaf);
1508 /*
1509 * Point to the leaf entry, get data address from it.
1510 */
1511 lep = &ents[index];
1512 /*
1513 * Point to the data entry.
1514 */
1515 dep = (xfs_dir2_data_entry_t *)
1516 ((char *)dbp->b_addr +
1517 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1518 ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1519 /*
1520 * Put the new inode number in, log it.
1521 */
1522 dep->inumber = cpu_to_be64(args->inumber);
1523 dp->d_ops->data_put_ftype(dep, args->filetype);
1524 tp = args->trans;
1525 xfs_dir2_data_log_entry(args, dbp, dep);
1526 xfs_dir3_leaf_check(dp, lbp);
1527 xfs_trans_brelse(tp, lbp);
1528 return 0;
1529 }
1530
1531 /*
1532 * Return index in the leaf block (lbp) which is either the first
1533 * one with this hash value, or if there are none, the insert point
1534 * for that hash value.
1535 */
1536 int /* index value */
1537 xfs_dir2_leaf_search_hash(
1538 xfs_da_args_t *args, /* operation arguments */
1539 struct xfs_buf *lbp) /* leaf buffer */
1540 {
1541 xfs_dahash_t hash=0; /* hash from this entry */
1542 xfs_dahash_t hashwant; /* hash value looking for */
1543 int high; /* high leaf index */
1544 int low; /* low leaf index */
1545 xfs_dir2_leaf_t *leaf; /* leaf structure */
1546 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1547 int mid=0; /* current leaf index */
1548 struct xfs_dir2_leaf_entry *ents;
1549 struct xfs_dir3_icleaf_hdr leafhdr;
1550
1551 leaf = lbp->b_addr;
1552 ents = args->dp->d_ops->leaf_ents_p(leaf);
1553 args->dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1554
1555 /*
1556 * Note, the table cannot be empty, so we have to go through the loop.
1557 * Binary search the leaf entries looking for our hash value.
1558 */
1559 for (lep = ents, low = 0, high = leafhdr.count - 1,
1560 hashwant = args->hashval;
1561 low <= high; ) {
1562 mid = (low + high) >> 1;
1563 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1564 break;
1565 if (hash < hashwant)
1566 low = mid + 1;
1567 else
1568 high = mid - 1;
1569 }
1570 /*
1571 * Found one, back up through all the equal hash values.
1572 */
1573 if (hash == hashwant) {
1574 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1575 mid--;
1576 }
1577 }
1578 /*
1579 * Need to point to an entry higher than ours.
1580 */
1581 else if (hash < hashwant)
1582 mid++;
1583 return mid;
1584 }
1585
1586 /*
1587 * Trim off a trailing data block. We know it's empty since the leaf
1588 * freespace table says so.
1589 */
1590 int /* error */
1591 xfs_dir2_leaf_trim_data(
1592 xfs_da_args_t *args, /* operation arguments */
1593 struct xfs_buf *lbp, /* leaf buffer */
1594 xfs_dir2_db_t db) /* data block number */
1595 {
1596 __be16 *bestsp; /* leaf bests table */
1597 struct xfs_buf *dbp; /* data block buffer */
1598 xfs_inode_t *dp; /* incore directory inode */
1599 int error; /* error return value */
1600 xfs_dir2_leaf_t *leaf; /* leaf structure */
1601 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1602 xfs_trans_t *tp; /* transaction pointer */
1603
1604 dp = args->dp;
1605 tp = args->trans;
1606 /*
1607 * Read the offending data block. We need its buffer.
1608 */
1609 error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(args->geo, db),
1610 -1, &dbp);
1611 if (error)
1612 return error;
1613
1614 leaf = lbp->b_addr;
1615 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1616
1617 #ifdef DEBUG
1618 {
1619 struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1620 struct xfs_dir2_data_free *bf = dp->d_ops->data_bestfree_p(hdr);
1621
1622 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1623 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1624 ASSERT(be16_to_cpu(bf[0].length) ==
1625 args->geo->blksize - dp->d_ops->data_entry_offset);
1626 ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1627 }
1628 #endif
1629
1630 /*
1631 * Get rid of the data block.
1632 */
1633 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1634 ASSERT(error != -ENOSPC);
1635 xfs_trans_brelse(tp, dbp);
1636 return error;
1637 }
1638 /*
1639 * Eliminate the last bests entry from the table.
1640 */
1641 bestsp = xfs_dir2_leaf_bests_p(ltp);
1642 be32_add_cpu(&ltp->bestcount, -1);
1643 memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1644 xfs_dir3_leaf_log_tail(args, lbp);
1645 xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1646 return 0;
1647 }
1648
1649 static inline size_t
1650 xfs_dir3_leaf_size(
1651 struct xfs_dir3_icleaf_hdr *hdr,
1652 int counts)
1653 {
1654 int entries;
1655 int hdrsize;
1656
1657 entries = hdr->count - hdr->stale;
1658 if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1659 hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1660 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1661 else
1662 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1663
1664 return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1665 + counts * sizeof(xfs_dir2_data_off_t)
1666 + sizeof(xfs_dir2_leaf_tail_t);
1667 }
1668
1669 /*
1670 * Convert node form directory to leaf form directory.
1671 * The root of the node form dir needs to already be a LEAFN block.
1672 * Just return if we can't do anything.
1673 */
1674 int /* error */
1675 xfs_dir2_node_to_leaf(
1676 xfs_da_state_t *state) /* directory operation state */
1677 {
1678 xfs_da_args_t *args; /* operation arguments */
1679 xfs_inode_t *dp; /* incore directory inode */
1680 int error; /* error return code */
1681 struct xfs_buf *fbp; /* buffer for freespace block */
1682 xfs_fileoff_t fo; /* freespace file offset */
1683 xfs_dir2_free_t *free; /* freespace structure */
1684 struct xfs_buf *lbp; /* buffer for leaf block */
1685 xfs_dir2_leaf_tail_t *ltp; /* tail of leaf structure */
1686 xfs_dir2_leaf_t *leaf; /* leaf structure */
1687 xfs_mount_t *mp; /* filesystem mount point */
1688 int rval; /* successful free trim? */
1689 xfs_trans_t *tp; /* transaction pointer */
1690 struct xfs_dir3_icleaf_hdr leafhdr;
1691 struct xfs_dir3_icfree_hdr freehdr;
1692
1693 /*
1694 * There's more than a leaf level in the btree, so there must
1695 * be multiple leafn blocks. Give up.
1696 */
1697 if (state->path.active > 1)
1698 return 0;
1699 args = state->args;
1700
1701 trace_xfs_dir2_node_to_leaf(args);
1702
1703 mp = state->mp;
1704 dp = args->dp;
1705 tp = args->trans;
1706 /*
1707 * Get the last offset in the file.
1708 */
1709 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1710 return error;
1711 }
1712 fo -= args->geo->fsbcount;
1713 /*
1714 * If there are freespace blocks other than the first one,
1715 * take this opportunity to remove trailing empty freespace blocks
1716 * that may have been left behind during no-space-reservation
1717 * operations.
1718 */
1719 while (fo > args->geo->freeblk) {
1720 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1721 return error;
1722 }
1723 if (rval)
1724 fo -= args->geo->fsbcount;
1725 else
1726 return 0;
1727 }
1728 /*
1729 * Now find the block just before the freespace block.
1730 */
1731 if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1732 return error;
1733 }
1734 /*
1735 * If it's not the single leaf block, give up.
1736 */
1737 if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1738 return 0;
1739 lbp = state->path.blk[0].bp;
1740 leaf = lbp->b_addr;
1741 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1742
1743 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1744 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1745
1746 /*
1747 * Read the freespace block.
1748 */
1749 error = xfs_dir2_free_read(tp, dp, args->geo->freeblk, &fbp);
1750 if (error)
1751 return error;
1752 free = fbp->b_addr;
1753 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1754
1755 ASSERT(!freehdr.firstdb);
1756
1757 /*
1758 * Now see if the leafn and free data will fit in a leaf1.
1759 * If not, release the buffer and give up.
1760 */
1761 if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > args->geo->blksize) {
1762 xfs_trans_brelse(tp, fbp);
1763 return 0;
1764 }
1765
1766 /*
1767 * If the leaf has any stale entries in it, compress them out.
1768 */
1769 if (leafhdr.stale)
1770 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1771
1772 lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1773 xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1774 leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1775 ? XFS_DIR2_LEAF1_MAGIC
1776 : XFS_DIR3_LEAF1_MAGIC;
1777
1778 /*
1779 * Set up the leaf tail from the freespace block.
1780 */
1781 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1782 ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1783
1784 /*
1785 * Set up the leaf bests table.
1786 */
1787 memcpy(xfs_dir2_leaf_bests_p(ltp), dp->d_ops->free_bests_p(free),
1788 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1789
1790 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1791 xfs_dir3_leaf_log_header(args, lbp);
1792 xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1793 xfs_dir3_leaf_log_tail(args, lbp);
1794 xfs_dir3_leaf_check(dp, lbp);
1795
1796 /*
1797 * Get rid of the freespace block.
1798 */
1799 error = xfs_dir2_shrink_inode(args,
1800 xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET),
1801 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 }