projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xfs: convert m_dirblksize to xfs_da_geometry
[deliverable/linux.git] / fs / xfs / 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 "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 */
42 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
43 int *indexp, struct xfs_buf **dbpp);
44 static void xfs_dir3_leaf_log_bests(struct xfs_trans *tp, struct xfs_buf *bp,
45 int first, int last);
46 static void xfs_dir3_leaf_log_tail(struct xfs_trans *tp, struct xfs_buf *bp);
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) \
54 do { \
55 if (!xfs_dir3_leaf1_check((dp), (bp))) \
56 ASSERT(0); \
57 } while (0);
58
59 STATIC bool
60 xfs_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
82 bool
83 xfs_dir3_leaf_check_int(
84 struct xfs_mount *mp,
85 struct xfs_inode *dp,
86 struct xfs_dir3_icleaf_hdr *hdr,
87 struct xfs_dir2_leaf *leaf)
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 struct xfs_da_geometry *geo = mp->m_dir_geo;
96
97 /*
98 * we can be passed a null dp here from a verifier, so we need to go the
99 * hard way to get them.
100 */
101 ops = xfs_dir_get_ops(mp, dp);
102
103 if (!hdr) {
104 ops->leaf_hdr_from_disk(&leafhdr, leaf);
105 hdr = &leafhdr;
106 }
107
108 ents = ops->leaf_ents_p(leaf);
109 ltp = xfs_dir2_leaf_tail_p(geo, leaf);
110
111 /*
112 * XXX (dgc): This value is not restrictive enough.
113 * Should factor in the size of the bests table as well.
114 * We can deduce a value for that from di_size.
115 */
116 if (hdr->count > ops->leaf_max_ents(geo))
117 return false;
118
119 /* Leaves and bests don't overlap in leaf format. */
120 if ((hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
121 hdr->magic == XFS_DIR3_LEAF1_MAGIC) &&
122 (char *)&ents[hdr->count] > (char *)xfs_dir2_leaf_bests_p(ltp))
123 return false;
124
125 /* Check hash value order, count stale entries. */
126 for (i = stale = 0; i < hdr->count; i++) {
127 if (i + 1 < hdr->count) {
128 if (be32_to_cpu(ents[i].hashval) >
129 be32_to_cpu(ents[i + 1].hashval))
130 return false;
131 }
132 if (ents[i].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
133 stale++;
134 }
135 if (hdr->stale != stale)
136 return false;
137 return true;
138 }
139
140 /*
141 * We verify the magic numbers before decoding the leaf header so that on debug
142 * kernels we don't get assertion failures in xfs_dir3_leaf_hdr_from_disk() due
143 * to incorrect magic numbers.
144 */
145 static bool
146 xfs_dir3_leaf_verify(
147 struct xfs_buf *bp,
148 __uint16_t magic)
149 {
150 struct xfs_mount *mp = bp->b_target->bt_mount;
151 struct xfs_dir2_leaf *leaf = bp->b_addr;
152
153 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
154
155 if (xfs_sb_version_hascrc(&mp->m_sb)) {
156 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
157 __uint16_t magic3;
158
159 magic3 = (magic == XFS_DIR2_LEAF1_MAGIC) ? XFS_DIR3_LEAF1_MAGIC
160 : XFS_DIR3_LEAFN_MAGIC;
161
162 if (leaf3->info.hdr.magic != cpu_to_be16(magic3))
163 return false;
164 if (!uuid_equal(&leaf3->info.uuid, &mp->m_sb.sb_uuid))
165 return false;
166 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
167 return false;
168 } else {
169 if (leaf->hdr.info.magic != cpu_to_be16(magic))
170 return false;
171 }
172
173 return xfs_dir3_leaf_check_int(mp, NULL, NULL, leaf);
174 }
175
176 static void
177 __read_verify(
178 struct xfs_buf *bp,
179 __uint16_t magic)
180 {
181 struct xfs_mount *mp = bp->b_target->bt_mount;
182
183 if (xfs_sb_version_hascrc(&mp->m_sb) &&
184 !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
185 xfs_buf_ioerror(bp, EFSBADCRC);
186 else if (!xfs_dir3_leaf_verify(bp, magic))
187 xfs_buf_ioerror(bp, EFSCORRUPTED);
188
189 if (bp->b_error)
190 xfs_verifier_error(bp);
191 }
192
193 static void
194 __write_verify(
195 struct xfs_buf *bp,
196 __uint16_t magic)
197 {
198 struct xfs_mount *mp = bp->b_target->bt_mount;
199 struct xfs_buf_log_item *bip = bp->b_fspriv;
200 struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
201
202 if (!xfs_dir3_leaf_verify(bp, magic)) {
203 xfs_buf_ioerror(bp, EFSCORRUPTED);
204 xfs_verifier_error(bp);
205 return;
206 }
207
208 if (!xfs_sb_version_hascrc(&mp->m_sb))
209 return;
210
211 if (bip)
212 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
213
214 xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
215 }
216
217 static void
218 xfs_dir3_leaf1_read_verify(
219 struct xfs_buf *bp)
220 {
221 __read_verify(bp, XFS_DIR2_LEAF1_MAGIC);
222 }
223
224 static void
225 xfs_dir3_leaf1_write_verify(
226 struct xfs_buf *bp)
227 {
228 __write_verify(bp, XFS_DIR2_LEAF1_MAGIC);
229 }
230
231 static void
232 xfs_dir3_leafn_read_verify(
233 struct xfs_buf *bp)
234 {
235 __read_verify(bp, XFS_DIR2_LEAFN_MAGIC);
236 }
237
238 static void
239 xfs_dir3_leafn_write_verify(
240 struct xfs_buf *bp)
241 {
242 __write_verify(bp, XFS_DIR2_LEAFN_MAGIC);
243 }
244
245 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
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 .verify_read = xfs_dir3_leafn_read_verify,
252 .verify_write = xfs_dir3_leafn_write_verify,
253 };
254
255 static int
256 xfs_dir3_leaf_read(
257 struct xfs_trans *tp,
258 struct xfs_inode *dp,
259 xfs_dablk_t fbno,
260 xfs_daddr_t mappedbno,
261 struct xfs_buf **bpp)
262 {
263 int err;
264
265 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
266 XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
267 if (!err && tp)
268 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
269 return err;
270 }
271
272 int
273 xfs_dir3_leafn_read(
274 struct xfs_trans *tp,
275 struct xfs_inode *dp,
276 xfs_dablk_t fbno,
277 xfs_daddr_t mappedbno,
278 struct xfs_buf **bpp)
279 {
280 int err;
281
282 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
283 XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
284 if (!err && tp)
285 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
286 return err;
287 }
288
289 /*
290 * Initialize a new leaf block, leaf1 or leafn magic accepted.
291 */
292 static void
293 xfs_dir3_leaf_init(
294 struct xfs_mount *mp,
295 struct xfs_trans *tp,
296 struct xfs_buf *bp,
297 xfs_ino_t owner,
298 __uint16_t type)
299 {
300 struct xfs_dir2_leaf *leaf = bp->b_addr;
301
302 ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
303
304 if (xfs_sb_version_hascrc(&mp->m_sb)) {
305 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
306
307 memset(leaf3, 0, sizeof(*leaf3));
308
309 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
310 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
311 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
312 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
313 leaf3->info.owner = cpu_to_be64(owner);
314 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_uuid);
315 } else {
316 memset(leaf, 0, sizeof(*leaf));
317 leaf->hdr.info.magic = cpu_to_be16(type);
318 }
319
320 /*
321 * If it's a leaf-format directory initialize the tail.
322 * Caller is responsible for initialising the bests table.
323 */
324 if (type == XFS_DIR2_LEAF1_MAGIC) {
325 struct xfs_dir2_leaf_tail *ltp;
326
327 ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
328 ltp->bestcount = 0;
329 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
330 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
331 } else {
332 bp->b_ops = &xfs_dir3_leafn_buf_ops;
333 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
334 }
335 }
336
337 int
338 xfs_dir3_leaf_get_buf(
339 xfs_da_args_t *args,
340 xfs_dir2_db_t bno,
341 struct xfs_buf **bpp,
342 __uint16_t magic)
343 {
344 struct xfs_inode *dp = args->dp;
345 struct xfs_trans *tp = args->trans;
346 struct xfs_mount *mp = dp->i_mount;
347 struct xfs_buf *bp;
348 int error;
349
350 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
351 ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
352 bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
353
354 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
355 -1, &bp, XFS_DATA_FORK);
356 if (error)
357 return error;
358
359 xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
360 xfs_dir3_leaf_log_header(tp, dp, bp);
361 if (magic == XFS_DIR2_LEAF1_MAGIC)
362 xfs_dir3_leaf_log_tail(tp, bp);
363 *bpp = bp;
364 return 0;
365 }
366
367 /*
368 * Convert a block form directory to a leaf form directory.
369 */
370 int /* error */
371 xfs_dir2_block_to_leaf(
372 xfs_da_args_t *args, /* operation arguments */
373 struct xfs_buf *dbp) /* input block's buffer */
374 {
375 __be16 *bestsp; /* leaf's bestsp entries */
376 xfs_dablk_t blkno; /* leaf block's bno */
377 xfs_dir2_data_hdr_t *hdr; /* block header */
378 xfs_dir2_leaf_entry_t *blp; /* block's leaf entries */
379 xfs_dir2_block_tail_t *btp; /* block's tail */
380 xfs_inode_t *dp; /* incore directory inode */
381 int error; /* error return code */
382 struct xfs_buf *lbp; /* leaf block's buffer */
383 xfs_dir2_db_t ldb; /* leaf block's bno */
384 xfs_dir2_leaf_t *leaf; /* leaf structure */
385 xfs_dir2_leaf_tail_t *ltp; /* leaf's tail */
386 xfs_mount_t *mp; /* filesystem mount point */
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 mp = dp->i_mount;
398 tp = args->trans;
399 /*
400 * Add the leaf block to the inode.
401 * This interface will only put blocks in the leaf/node range.
402 * Since that's empty now, we'll get the root (block 0 in range).
403 */
404 if ((error = xfs_da_grow_inode(args, &blkno))) {
405 return error;
406 }
407 ldb = xfs_dir2_da_to_db(args->geo, blkno);
408 ASSERT(ldb == xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET));
409 /*
410 * Initialize the leaf block, get a buffer for it.
411 */
412 error = xfs_dir3_leaf_get_buf(args, ldb, &lbp, XFS_DIR2_LEAF1_MAGIC);
413 if (error)
414 return error;
415
416 leaf = lbp->b_addr;
417 hdr = dbp->b_addr;
418 xfs_dir3_data_check(dp, dbp);
419 btp = xfs_dir2_block_tail_p(args->geo, hdr);
420 blp = xfs_dir2_block_leaf_p(btp);
421 bf = dp->d_ops->data_bestfree_p(hdr);
422 ents = dp->d_ops->leaf_ents_p(leaf);
423
424 /*
425 * Set the counts in the leaf header.
426 */
427 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
428 leafhdr.count = be32_to_cpu(btp->count);
429 leafhdr.stale = be32_to_cpu(btp->stale);
430 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
431 xfs_dir3_leaf_log_header(tp, dp, lbp);
432
433 /*
434 * Could compact these but I think we always do the conversion
435 * after squeezing out stale entries.
436 */
437 memcpy(ents, blp, be32_to_cpu(btp->count) * sizeof(xfs_dir2_leaf_entry_t));
438 xfs_dir3_leaf_log_ents(tp, dp, lbp, 0, leafhdr.count - 1);
439 needscan = 0;
440 needlog = 1;
441 /*
442 * Make the space formerly occupied by the leaf entries and block
443 * tail be free.
444 */
445 xfs_dir2_data_make_free(tp, dp, dbp,
446 (xfs_dir2_data_aoff_t)((char *)blp - (char *)hdr),
447 (xfs_dir2_data_aoff_t)((char *)hdr + args->geo->blksize -
448 (char *)blp),
449 &needlog, &needscan);
450 /*
451 * Fix up the block header, make it a data block.
452 */
453 dbp->b_ops = &xfs_dir3_data_buf_ops;
454 xfs_trans_buf_set_type(tp, dbp, XFS_BLFT_DIR_DATA_BUF);
455 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC))
456 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
457 else
458 hdr->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
459
460 if (needscan)
461 xfs_dir2_data_freescan(dp, hdr, &needlog);
462 /*
463 * Set up leaf tail and bests table.
464 */
465 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
466 ltp->bestcount = cpu_to_be32(1);
467 bestsp = xfs_dir2_leaf_bests_p(ltp);
468 bestsp[0] = bf[0].length;
469 /*
470 * Log the data header and leaf bests table.
471 */
472 if (needlog)
473 xfs_dir2_data_log_header(tp, dp, dbp);
474 xfs_dir3_leaf_check(dp, lbp);
475 xfs_dir3_data_check(dp, dbp);
476 xfs_dir3_leaf_log_bests(tp, lbp, 0, 0);
477 return 0;
478 }
479
480 STATIC void
481 xfs_dir3_leaf_find_stale(
482 struct xfs_dir3_icleaf_hdr *leafhdr,
483 struct xfs_dir2_leaf_entry *ents,
484 int index,
485 int *lowstale,
486 int *highstale)
487 {
488 /*
489 * Find the first stale entry before our index, if any.
490 */
491 for (*lowstale = index - 1; *lowstale >= 0; --*lowstale) {
492 if (ents[*lowstale].address ==
493 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
494 break;
495 }
496
497 /*
498 * Find the first stale entry at or after our index, if any.
499 * Stop if the result would require moving more entries than using
500 * lowstale.
501 */
502 for (*highstale = index; *highstale < leafhdr->count; ++*highstale) {
503 if (ents[*highstale].address ==
504 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
505 break;
506 if (*lowstale >= 0 && index - *lowstale <= *highstale - index)
507 break;
508 }
509 }
510
511 struct xfs_dir2_leaf_entry *
512 xfs_dir3_leaf_find_entry(
513 struct xfs_dir3_icleaf_hdr *leafhdr,
514 struct xfs_dir2_leaf_entry *ents,
515 int index, /* leaf table position */
516 int compact, /* need to compact leaves */
517 int lowstale, /* index of prev stale leaf */
518 int highstale, /* index of next stale leaf */
519 int *lfloglow, /* low leaf logging index */
520 int *lfloghigh) /* high leaf logging index */
521 {
522 if (!leafhdr->stale) {
523 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
524
525 /*
526 * Now we need to make room to insert the leaf entry.
527 *
528 * If there are no stale entries, just insert a hole at index.
529 */
530 lep = &ents[index];
531 if (index < leafhdr->count)
532 memmove(lep + 1, lep,
533 (leafhdr->count - index) * sizeof(*lep));
534
535 /*
536 * Record low and high logging indices for the leaf.
537 */
538 *lfloglow = index;
539 *lfloghigh = leafhdr->count++;
540 return lep;
541 }
542
543 /*
544 * There are stale entries.
545 *
546 * We will use one of them for the new entry. It's probably not at
547 * the right location, so we'll have to shift some up or down first.
548 *
549 * If we didn't compact before, we need to find the nearest stale
550 * entries before and after our insertion point.
551 */
552 if (compact == 0)
553 xfs_dir3_leaf_find_stale(leafhdr, ents, index,
554 &lowstale, &highstale);
555
556 /*
557 * If the low one is better, use it.
558 */
559 if (lowstale >= 0 &&
560 (highstale == leafhdr->count ||
561 index - lowstale - 1 < highstale - index)) {
562 ASSERT(index - lowstale - 1 >= 0);
563 ASSERT(ents[lowstale].address ==
564 cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
565
566 /*
567 * Copy entries up to cover the stale entry and make room
568 * for the new entry.
569 */
570 if (index - lowstale - 1 > 0) {
571 memmove(&ents[lowstale], &ents[lowstale + 1],
572 (index - lowstale - 1) *
573 sizeof(xfs_dir2_leaf_entry_t));
574 }
575 *lfloglow = MIN(lowstale, *lfloglow);
576 *lfloghigh = MAX(index - 1, *lfloghigh);
577 leafhdr->stale--;
578 return &ents[index - 1];
579 }
580
581 /*
582 * The high one is better, so use that one.
583 */
584 ASSERT(highstale - index >= 0);
585 ASSERT(ents[highstale].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR));
586
587 /*
588 * Copy entries down to cover the stale entry and make room for the
589 * new entry.
590 */
591 if (highstale - index > 0) {
592 memmove(&ents[index + 1], &ents[index],
593 (highstale - index) * sizeof(xfs_dir2_leaf_entry_t));
594 }
595 *lfloglow = MIN(index, *lfloglow);
596 *lfloghigh = MAX(highstale, *lfloghigh);
597 leafhdr->stale--;
598 return &ents[index];
599 }
600
601 /*
602 * Add an entry to a leaf form directory.
603 */
604 int /* error */
605 xfs_dir2_leaf_addname(
606 xfs_da_args_t *args) /* operation arguments */
607 {
608 __be16 *bestsp; /* freespace table in leaf */
609 int compact; /* need to compact leaves */
610 xfs_dir2_data_hdr_t *hdr; /* data block header */
611 struct xfs_buf *dbp; /* data block buffer */
612 xfs_dir2_data_entry_t *dep; /* data block entry */
613 xfs_inode_t *dp; /* incore directory inode */
614 xfs_dir2_data_unused_t *dup; /* data unused entry */
615 int error; /* error return value */
616 int grown; /* allocated new data block */
617 int highstale; /* index of next stale leaf */
618 int i; /* temporary, index */
619 int index; /* leaf table position */
620 struct xfs_buf *lbp; /* leaf's buffer */
621 xfs_dir2_leaf_t *leaf; /* leaf structure */
622 int length; /* length of new entry */
623 xfs_dir2_leaf_entry_t *lep; /* leaf entry table pointer */
624 int lfloglow; /* low leaf logging index */
625 int lfloghigh; /* high leaf logging index */
626 int lowstale; /* index of prev stale leaf */
627 xfs_dir2_leaf_tail_t *ltp; /* leaf tail pointer */
628 xfs_mount_t *mp; /* filesystem mount point */
629 int needbytes; /* leaf block bytes needed */
630 int needlog; /* need to log data header */
631 int needscan; /* need to rescan data free */
632 __be16 *tagp; /* end of data entry */
633 xfs_trans_t *tp; /* transaction pointer */
634 xfs_dir2_db_t use_block; /* data block number */
635 struct xfs_dir2_data_free *bf; /* bestfree table */
636 struct xfs_dir2_leaf_entry *ents;
637 struct xfs_dir3_icleaf_hdr leafhdr;
638
639 trace_xfs_dir2_leaf_addname(args);
640
641 dp = args->dp;
642 tp = args->trans;
643 mp = dp->i_mount;
644
645 error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
646 if (error)
647 return error;
648
649 /*
650 * Look up the entry by hash value and name.
651 * We know it's not there, our caller has already done a lookup.
652 * So the index is of the entry to insert in front of.
653 * But if there are dup hash values the index is of the first of those.
654 */
655 index = xfs_dir2_leaf_search_hash(args, lbp);
656 leaf = lbp->b_addr;
657 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
658 ents = dp->d_ops->leaf_ents_p(leaf);
659 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
660 bestsp = xfs_dir2_leaf_bests_p(ltp);
661 length = dp->d_ops->data_entsize(args->namelen);
662
663 /*
664 * See if there are any entries with the same hash value
665 * and space in their block for the new entry.
666 * This is good because it puts multiple same-hash value entries
667 * in a data block, improving the lookup of those entries.
668 */
669 for (use_block = -1, lep = &ents[index];
670 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
671 index++, lep++) {
672 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
673 continue;
674 i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
675 ASSERT(i < be32_to_cpu(ltp->bestcount));
676 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
677 if (be16_to_cpu(bestsp[i]) >= length) {
678 use_block = i;
679 break;
680 }
681 }
682 /*
683 * Didn't find a block yet, linear search all the data blocks.
684 */
685 if (use_block == -1) {
686 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
687 /*
688 * Remember a block we see that's missing.
689 */
690 if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
691 use_block == -1)
692 use_block = i;
693 else if (be16_to_cpu(bestsp[i]) >= length) {
694 use_block = i;
695 break;
696 }
697 }
698 }
699 /*
700 * How many bytes do we need in the leaf block?
701 */
702 needbytes = 0;
703 if (!leafhdr.stale)
704 needbytes += sizeof(xfs_dir2_leaf_entry_t);
705 if (use_block == -1)
706 needbytes += sizeof(xfs_dir2_data_off_t);
707
708 /*
709 * Now kill use_block if it refers to a missing block, so we
710 * can use it as an indication of allocation needed.
711 */
712 if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
713 use_block = -1;
714 /*
715 * If we don't have enough free bytes but we can make enough
716 * by compacting out stale entries, we'll do that.
717 */
718 if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
719 leafhdr.stale > 1)
720 compact = 1;
721
722 /*
723 * Otherwise if we don't have enough free bytes we need to
724 * convert to node form.
725 */
726 else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
727 /*
728 * Just checking or no space reservation, give up.
729 */
730 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
731 args->total == 0) {
732 xfs_trans_brelse(tp, lbp);
733 return XFS_ERROR(ENOSPC);
734 }
735 /*
736 * Convert to node form.
737 */
738 error = xfs_dir2_leaf_to_node(args, lbp);
739 if (error)
740 return error;
741 /*
742 * Then add the new entry.
743 */
744 return xfs_dir2_node_addname(args);
745 }
746 /*
747 * Otherwise it will fit without compaction.
748 */
749 else
750 compact = 0;
751 /*
752 * If just checking, then it will fit unless we needed to allocate
753 * a new data block.
754 */
755 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
756 xfs_trans_brelse(tp, lbp);
757 return use_block == -1 ? XFS_ERROR(ENOSPC) : 0;
758 }
759 /*
760 * If no allocations are allowed, return now before we've
761 * changed anything.
762 */
763 if (args->total == 0 && use_block == -1) {
764 xfs_trans_brelse(tp, lbp);
765 return XFS_ERROR(ENOSPC);
766 }
767 /*
768 * Need to compact the leaf entries, removing stale ones.
769 * Leave one stale entry behind - the one closest to our
770 * insertion index - and we'll shift that one to our insertion
771 * point later.
772 */
773 if (compact) {
774 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
775 &highstale, &lfloglow, &lfloghigh);
776 }
777 /*
778 * There are stale entries, so we'll need log-low and log-high
779 * impossibly bad values later.
780 */
781 else if (leafhdr.stale) {
782 lfloglow = leafhdr.count;
783 lfloghigh = -1;
784 }
785 /*
786 * If there was no data block space found, we need to allocate
787 * a new one.
788 */
789 if (use_block == -1) {
790 /*
791 * Add the new data block.
792 */
793 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
794 &use_block))) {
795 xfs_trans_brelse(tp, lbp);
796 return error;
797 }
798 /*
799 * Initialize the block.
800 */
801 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
802 xfs_trans_brelse(tp, lbp);
803 return error;
804 }
805 /*
806 * If we're adding a new data block on the end we need to
807 * extend the bests table. Copy it up one entry.
808 */
809 if (use_block >= be32_to_cpu(ltp->bestcount)) {
810 bestsp--;
811 memmove(&bestsp[0], &bestsp[1],
812 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
813 be32_add_cpu(&ltp->bestcount, 1);
814 xfs_dir3_leaf_log_tail(tp, lbp);
815 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
816 }
817 /*
818 * If we're filling in a previously empty block just log it.
819 */
820 else
821 xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
822 hdr = dbp->b_addr;
823 bf = dp->d_ops->data_bestfree_p(hdr);
824 bestsp[use_block] = bf[0].length;
825 grown = 1;
826 } else {
827 /*
828 * Already had space in some data block.
829 * Just read that one in.
830 */
831 error = xfs_dir3_data_read(tp, dp,
832 xfs_dir2_db_to_da(args->geo, use_block),
833 -1, &dbp);
834 if (error) {
835 xfs_trans_brelse(tp, lbp);
836 return error;
837 }
838 hdr = dbp->b_addr;
839 bf = dp->d_ops->data_bestfree_p(hdr);
840 grown = 0;
841 }
842 /*
843 * Point to the biggest freespace in our data block.
844 */
845 dup = (xfs_dir2_data_unused_t *)
846 ((char *)hdr + be16_to_cpu(bf[0].offset));
847 ASSERT(be16_to_cpu(dup->length) >= length);
848 needscan = needlog = 0;
849 /*
850 * Mark the initial part of our freespace in use for the new entry.
851 */
852 xfs_dir2_data_use_free(tp, dp, dbp, dup,
853 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
854 &needlog, &needscan);
855 /*
856 * Initialize our new entry (at last).
857 */
858 dep = (xfs_dir2_data_entry_t *)dup;
859 dep->inumber = cpu_to_be64(args->inumber);
860 dep->namelen = args->namelen;
861 memcpy(dep->name, args->name, dep->namelen);
862 dp->d_ops->data_put_ftype(dep, args->filetype);
863 tagp = dp->d_ops->data_entry_tag_p(dep);
864 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
865 /*
866 * Need to scan fix up the bestfree table.
867 */
868 if (needscan)
869 xfs_dir2_data_freescan(dp, hdr, &needlog);
870 /*
871 * Need to log the data block's header.
872 */
873 if (needlog)
874 xfs_dir2_data_log_header(tp, dp, dbp);
875 xfs_dir2_data_log_entry(tp, dp, dbp, dep);
876 /*
877 * If the bests table needs to be changed, do it.
878 * Log the change unless we've already done that.
879 */
880 if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
881 bestsp[use_block] = bf[0].length;
882 if (!grown)
883 xfs_dir3_leaf_log_bests(tp, lbp, use_block, use_block);
884 }
885
886 lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
887 highstale, &lfloglow, &lfloghigh);
888
889 /*
890 * Fill in the new leaf entry.
891 */
892 lep->hashval = cpu_to_be32(args->hashval);
893 lep->address = cpu_to_be32(
894 xfs_dir2_db_off_to_dataptr(args->geo, use_block,
895 be16_to_cpu(*tagp)));
896 /*
897 * Log the leaf fields and give up the buffers.
898 */
899 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
900 xfs_dir3_leaf_log_header(tp, dp, lbp);
901 xfs_dir3_leaf_log_ents(tp, dp, lbp, lfloglow, lfloghigh);
902 xfs_dir3_leaf_check(dp, lbp);
903 xfs_dir3_data_check(dp, dbp);
904 return 0;
905 }
906
907 /*
908 * Compact out any stale entries in the leaf.
909 * Log the header and changed leaf entries, if any.
910 */
911 void
912 xfs_dir3_leaf_compact(
913 xfs_da_args_t *args, /* operation arguments */
914 struct xfs_dir3_icleaf_hdr *leafhdr,
915 struct xfs_buf *bp) /* leaf buffer */
916 {
917 int from; /* source leaf index */
918 xfs_dir2_leaf_t *leaf; /* leaf structure */
919 int loglow; /* first leaf entry to log */
920 int to; /* target leaf index */
921 struct xfs_dir2_leaf_entry *ents;
922 struct xfs_inode *dp = args->dp;
923
924 leaf = bp->b_addr;
925 if (!leafhdr->stale)
926 return;
927
928 /*
929 * Compress out the stale entries in place.
930 */
931 ents = dp->d_ops->leaf_ents_p(leaf);
932 for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
933 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
934 continue;
935 /*
936 * Only actually copy the entries that are different.
937 */
938 if (from > to) {
939 if (loglow == -1)
940 loglow = to;
941 ents[to] = ents[from];
942 }
943 to++;
944 }
945 /*
946 * Update and log the header, log the leaf entries.
947 */
948 ASSERT(leafhdr->stale == from - to);
949 leafhdr->count -= leafhdr->stale;
950 leafhdr->stale = 0;
951
952 dp->d_ops->leaf_hdr_to_disk(leaf, leafhdr);
953 xfs_dir3_leaf_log_header(args->trans, dp, bp);
954 if (loglow != -1)
955 xfs_dir3_leaf_log_ents(args->trans, dp, bp, loglow, to - 1);
956 }
957
958 /*
959 * Compact the leaf entries, removing stale ones.
960 * Leave one stale entry behind - the one closest to our
961 * insertion index - and the caller will shift that one to our insertion
962 * point later.
963 * Return new insertion index, where the remaining stale entry is,
964 * and leaf logging indices.
965 */
966 void
967 xfs_dir3_leaf_compact_x1(
968 struct xfs_dir3_icleaf_hdr *leafhdr,
969 struct xfs_dir2_leaf_entry *ents,
970 int *indexp, /* insertion index */
971 int *lowstalep, /* out: stale entry before us */
972 int *highstalep, /* out: stale entry after us */
973 int *lowlogp, /* out: low log index */
974 int *highlogp) /* out: high log index */
975 {
976 int from; /* source copy index */
977 int highstale; /* stale entry at/after index */
978 int index; /* insertion index */
979 int keepstale; /* source index of kept stale */
980 int lowstale; /* stale entry before index */
981 int newindex=0; /* new insertion index */
982 int to; /* destination copy index */
983
984 ASSERT(leafhdr->stale > 1);
985 index = *indexp;
986
987 xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
988
989 /*
990 * Pick the better of lowstale and highstale.
991 */
992 if (lowstale >= 0 &&
993 (highstale == leafhdr->count ||
994 index - lowstale <= highstale - index))
995 keepstale = lowstale;
996 else
997 keepstale = highstale;
998 /*
999 * Copy the entries in place, removing all the stale entries
1000 * except keepstale.
1001 */
1002 for (from = to = 0; from < leafhdr->count; from++) {
1003 /*
1004 * Notice the new value of index.
1005 */
1006 if (index == from)
1007 newindex = to;
1008 if (from != keepstale &&
1009 ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1010 if (from == to)
1011 *lowlogp = to;
1012 continue;
1013 }
1014 /*
1015 * Record the new keepstale value for the insertion.
1016 */
1017 if (from == keepstale)
1018 lowstale = highstale = to;
1019 /*
1020 * Copy only the entries that have moved.
1021 */
1022 if (from > to)
1023 ents[to] = ents[from];
1024 to++;
1025 }
1026 ASSERT(from > to);
1027 /*
1028 * If the insertion point was past the last entry,
1029 * set the new insertion point accordingly.
1030 */
1031 if (index == from)
1032 newindex = to;
1033 *indexp = newindex;
1034 /*
1035 * Adjust the leaf header values.
1036 */
1037 leafhdr->count -= from - to;
1038 leafhdr->stale = 1;
1039 /*
1040 * Remember the low/high stale value only in the "right"
1041 * direction.
1042 */
1043 if (lowstale >= newindex)
1044 lowstale = -1;
1045 else
1046 highstale = leafhdr->count;
1047 *highlogp = leafhdr->count - 1;
1048 *lowstalep = lowstale;
1049 *highstalep = highstale;
1050 }
1051
1052 /*
1053 * Log the bests entries indicated from a leaf1 block.
1054 */
1055 static void
1056 xfs_dir3_leaf_log_bests(
1057 xfs_trans_t *tp, /* transaction pointer */
1058 struct xfs_buf *bp, /* leaf buffer */
1059 int first, /* first entry to log */
1060 int last) /* last entry to log */
1061 {
1062 __be16 *firstb; /* pointer to first entry */
1063 __be16 *lastb; /* pointer to last entry */
1064 struct xfs_dir2_leaf *leaf = bp->b_addr;
1065 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1066
1067 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1068 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1069
1070 ltp = xfs_dir2_leaf_tail_p(tp->t_mountp->m_dir_geo, leaf);
1071 firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1072 lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1073 xfs_trans_log_buf(tp, bp, (uint)((char *)firstb - (char *)leaf),
1074 (uint)((char *)lastb - (char *)leaf + sizeof(*lastb) - 1));
1075 }
1076
1077 /*
1078 * Log the leaf entries indicated from a leaf1 or leafn block.
1079 */
1080 void
1081 xfs_dir3_leaf_log_ents(
1082 struct xfs_trans *tp,
1083 struct xfs_inode *dp,
1084 struct xfs_buf *bp,
1085 int first,
1086 int last)
1087 {
1088 xfs_dir2_leaf_entry_t *firstlep; /* pointer to first entry */
1089 xfs_dir2_leaf_entry_t *lastlep; /* pointer to last entry */
1090 struct xfs_dir2_leaf *leaf = bp->b_addr;
1091 struct xfs_dir2_leaf_entry *ents;
1092
1093 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1094 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1095 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1096 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1097
1098 ents = dp->d_ops->leaf_ents_p(leaf);
1099 firstlep = &ents[first];
1100 lastlep = &ents[last];
1101 xfs_trans_log_buf(tp, bp, (uint)((char *)firstlep - (char *)leaf),
1102 (uint)((char *)lastlep - (char *)leaf + sizeof(*lastlep) - 1));
1103 }
1104
1105 /*
1106 * Log the header of the leaf1 or leafn block.
1107 */
1108 void
1109 xfs_dir3_leaf_log_header(
1110 struct xfs_trans *tp,
1111 struct xfs_inode *dp,
1112 struct xfs_buf *bp)
1113 {
1114 struct xfs_dir2_leaf *leaf = bp->b_addr;
1115
1116 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1117 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1118 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1119 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1120
1121 xfs_trans_log_buf(tp, bp, (uint)((char *)&leaf->hdr - (char *)leaf),
1122 dp->d_ops->leaf_hdr_size - 1);
1123 }
1124
1125 /*
1126 * Log the tail of the leaf1 block.
1127 */
1128 STATIC void
1129 xfs_dir3_leaf_log_tail(
1130 struct xfs_trans *tp,
1131 struct xfs_buf *bp)
1132 {
1133 struct xfs_dir2_leaf *leaf = bp->b_addr;
1134 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1135 struct xfs_mount *mp = tp->t_mountp;
1136
1137 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1138 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1139 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1140 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1141
1142 ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
1143 xfs_trans_log_buf(tp, bp, (uint)((char *)ltp - (char *)leaf),
1144 (uint)(mp->m_dir_geo->blksize - 1));
1145 }
1146
1147 /*
1148 * Look up the entry referred to by args in the leaf format directory.
1149 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1150 * is also used by the node-format code.
1151 */
1152 int
1153 xfs_dir2_leaf_lookup(
1154 xfs_da_args_t *args) /* operation arguments */
1155 {
1156 struct xfs_buf *dbp; /* data block buffer */
1157 xfs_dir2_data_entry_t *dep; /* data block entry */
1158 xfs_inode_t *dp; /* incore directory inode */
1159 int error; /* error return code */
1160 int index; /* found entry index */
1161 struct xfs_buf *lbp; /* leaf buffer */
1162 xfs_dir2_leaf_t *leaf; /* leaf structure */
1163 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1164 xfs_trans_t *tp; /* transaction pointer */
1165 struct xfs_dir2_leaf_entry *ents;
1166
1167 trace_xfs_dir2_leaf_lookup(args);
1168
1169 /*
1170 * Look up name in the leaf block, returning both buffers and index.
1171 */
1172 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1173 return error;
1174 }
1175 tp = args->trans;
1176 dp = args->dp;
1177 xfs_dir3_leaf_check(dp, lbp);
1178 leaf = lbp->b_addr;
1179 ents = dp->d_ops->leaf_ents_p(leaf);
1180 /*
1181 * Get to the leaf entry and contained data entry address.
1182 */
1183 lep = &ents[index];
1184
1185 /*
1186 * Point to the data entry.
1187 */
1188 dep = (xfs_dir2_data_entry_t *)
1189 ((char *)dbp->b_addr +
1190 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1191 /*
1192 * Return the found inode number & CI name if appropriate
1193 */
1194 args->inumber = be64_to_cpu(dep->inumber);
1195 args->filetype = dp->d_ops->data_get_ftype(dep);
1196 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1197 xfs_trans_brelse(tp, dbp);
1198 xfs_trans_brelse(tp, lbp);
1199 return XFS_ERROR(error);
1200 }
1201
1202 /*
1203 * Look up name/hash in the leaf block.
1204 * Fill in indexp with the found index, and dbpp with the data buffer.
1205 * If not found dbpp will be NULL, and ENOENT comes back.
1206 * lbpp will always be filled in with the leaf buffer unless there's an error.
1207 */
1208 static int /* error */
1209 xfs_dir2_leaf_lookup_int(
1210 xfs_da_args_t *args, /* operation arguments */
1211 struct xfs_buf **lbpp, /* out: leaf buffer */
1212 int *indexp, /* out: index in leaf block */
1213 struct xfs_buf **dbpp) /* out: data buffer */
1214 {
1215 xfs_dir2_db_t curdb = -1; /* current data block number */
1216 struct xfs_buf *dbp = NULL; /* data buffer */
1217 xfs_dir2_data_entry_t *dep; /* data entry */
1218 xfs_inode_t *dp; /* incore directory inode */
1219 int error; /* error return code */
1220 int index; /* index in leaf block */
1221 struct xfs_buf *lbp; /* leaf buffer */
1222 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1223 xfs_dir2_leaf_t *leaf; /* leaf structure */
1224 xfs_mount_t *mp; /* filesystem mount point */
1225 xfs_dir2_db_t newdb; /* new data block number */
1226 xfs_trans_t *tp; /* transaction pointer */
1227 xfs_dir2_db_t cidb = -1; /* case match data block no. */
1228 enum xfs_dacmp cmp; /* name compare result */
1229 struct xfs_dir2_leaf_entry *ents;
1230 struct xfs_dir3_icleaf_hdr leafhdr;
1231
1232 dp = args->dp;
1233 tp = args->trans;
1234 mp = dp->i_mount;
1235
1236 error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
1237 if (error)
1238 return error;
1239
1240 *lbpp = lbp;
1241 leaf = lbp->b_addr;
1242 xfs_dir3_leaf_check(dp, lbp);
1243 ents = dp->d_ops->leaf_ents_p(leaf);
1244 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1245
1246 /*
1247 * Look for the first leaf entry with our hash value.
1248 */
1249 index = xfs_dir2_leaf_search_hash(args, lbp);
1250 /*
1251 * Loop over all the entries with the right hash value
1252 * looking to match the name.
1253 */
1254 for (lep = &ents[index];
1255 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1256 lep++, index++) {
1257 /*
1258 * Skip over stale leaf entries.
1259 */
1260 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1261 continue;
1262 /*
1263 * Get the new data block number.
1264 */
1265 newdb = xfs_dir2_dataptr_to_db(args->geo,
1266 be32_to_cpu(lep->address));
1267 /*
1268 * If it's not the same as the old data block number,
1269 * need to pitch the old one and read the new one.
1270 */
1271 if (newdb != curdb) {
1272 if (dbp)
1273 xfs_trans_brelse(tp, dbp);
1274 error = xfs_dir3_data_read(tp, dp,
1275 xfs_dir2_db_to_da(args->geo, newdb),
1276 -1, &dbp);
1277 if (error) {
1278 xfs_trans_brelse(tp, lbp);
1279 return error;
1280 }
1281 curdb = newdb;
1282 }
1283 /*
1284 * Point to the data entry.
1285 */
1286 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1287 xfs_dir2_dataptr_to_off(args->geo,
1288 be32_to_cpu(lep->address)));
1289 /*
1290 * Compare name and if it's an exact match, return the index
1291 * and buffer. If it's the first case-insensitive match, store
1292 * the index and buffer and continue looking for an exact match.
1293 */
1294 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1295 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1296 args->cmpresult = cmp;
1297 *indexp = index;
1298 /* case exact match: return the current buffer. */
1299 if (cmp == XFS_CMP_EXACT) {
1300 *dbpp = dbp;
1301 return 0;
1302 }
1303 cidb = curdb;
1304 }
1305 }
1306 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1307 /*
1308 * Here, we can only be doing a lookup (not a rename or remove).
1309 * If a case-insensitive match was found earlier, re-read the
1310 * appropriate data block if required and return it.
1311 */
1312 if (args->cmpresult == XFS_CMP_CASE) {
1313 ASSERT(cidb != -1);
1314 if (cidb != curdb) {
1315 xfs_trans_brelse(tp, dbp);
1316 error = xfs_dir3_data_read(tp, dp,
1317 xfs_dir2_db_to_da(args->geo, cidb),
1318 -1, &dbp);
1319 if (error) {
1320 xfs_trans_brelse(tp, lbp);
1321 return error;
1322 }
1323 }
1324 *dbpp = dbp;
1325 return 0;
1326 }
1327 /*
1328 * No match found, return ENOENT.
1329 */
1330 ASSERT(cidb == -1);
1331 if (dbp)
1332 xfs_trans_brelse(tp, dbp);
1333 xfs_trans_brelse(tp, lbp);
1334 return XFS_ERROR(ENOENT);
1335 }
1336
1337 /*
1338 * Remove an entry from a leaf format directory.
1339 */
1340 int /* error */
1341 xfs_dir2_leaf_removename(
1342 xfs_da_args_t *args) /* operation arguments */
1343 {
1344 __be16 *bestsp; /* leaf block best freespace */
1345 xfs_dir2_data_hdr_t *hdr; /* data block header */
1346 xfs_dir2_db_t db; /* data block number */
1347 struct xfs_buf *dbp; /* data block buffer */
1348 xfs_dir2_data_entry_t *dep; /* data entry structure */
1349 xfs_inode_t *dp; /* incore directory inode */
1350 int error; /* error return code */
1351 xfs_dir2_db_t i; /* temporary data block # */
1352 int index; /* index into leaf entries */
1353 struct xfs_buf *lbp; /* leaf buffer */
1354 xfs_dir2_leaf_t *leaf; /* leaf structure */
1355 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1356 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1357 xfs_mount_t *mp; /* filesystem mount point */
1358 int needlog; /* need to log data header */
1359 int needscan; /* need to rescan data frees */
1360 xfs_dir2_data_off_t oldbest; /* old value of best free */
1361 xfs_trans_t *tp; /* transaction pointer */
1362 struct xfs_dir2_data_free *bf; /* bestfree table */
1363 struct xfs_dir2_leaf_entry *ents;
1364 struct xfs_dir3_icleaf_hdr leafhdr;
1365
1366 trace_xfs_dir2_leaf_removename(args);
1367
1368 /*
1369 * Lookup the leaf entry, get the leaf and data blocks read in.
1370 */
1371 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1372 return error;
1373 }
1374 dp = args->dp;
1375 tp = args->trans;
1376 mp = dp->i_mount;
1377 leaf = lbp->b_addr;
1378 hdr = dbp->b_addr;
1379 xfs_dir3_data_check(dp, dbp);
1380 bf = dp->d_ops->data_bestfree_p(hdr);
1381 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1382 ents = dp->d_ops->leaf_ents_p(leaf);
1383 /*
1384 * Point to the leaf entry, use that to point to the data entry.
1385 */
1386 lep = &ents[index];
1387 db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1388 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1389 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1390 needscan = needlog = 0;
1391 oldbest = be16_to_cpu(bf[0].length);
1392 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1393 bestsp = xfs_dir2_leaf_bests_p(ltp);
1394 ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1395 /*
1396 * Mark the former data entry unused.
1397 */
1398 xfs_dir2_data_make_free(tp, dp, dbp,
1399 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1400 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1401 /*
1402 * We just mark the leaf entry stale by putting a null in it.
1403 */
1404 leafhdr.stale++;
1405 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1406 xfs_dir3_leaf_log_header(tp, dp, lbp);
1407
1408 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1409 xfs_dir3_leaf_log_ents(tp, dp, lbp, index, index);
1410
1411 /*
1412 * Scan the freespace in the data block again if necessary,
1413 * log the data block header if necessary.
1414 */
1415 if (needscan)
1416 xfs_dir2_data_freescan(dp, hdr, &needlog);
1417 if (needlog)
1418 xfs_dir2_data_log_header(tp, dp, dbp);
1419 /*
1420 * If the longest freespace in the data block has changed,
1421 * put the new value in the bests table and log that.
1422 */
1423 if (be16_to_cpu(bf[0].length) != oldbest) {
1424 bestsp[db] = bf[0].length;
1425 xfs_dir3_leaf_log_bests(tp, lbp, db, db);
1426 }
1427 xfs_dir3_data_check(dp, dbp);
1428 /*
1429 * If the data block is now empty then get rid of the data block.
1430 */
1431 if (be16_to_cpu(bf[0].length) ==
1432 args->geo->blksize - dp->d_ops->data_entry_offset) {
1433 ASSERT(db != args->geo->datablk);
1434 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1435 /*
1436 * Nope, can't get rid of it because it caused
1437 * allocation of a bmap btree block to do so.
1438 * Just go on, returning success, leaving the
1439 * empty block in place.
1440 */
1441 if (error == ENOSPC && args->total == 0)
1442 error = 0;
1443 xfs_dir3_leaf_check(dp, lbp);
1444 return error;
1445 }
1446 dbp = NULL;
1447 /*
1448 * If this is the last data block then compact the
1449 * bests table by getting rid of entries.
1450 */
1451 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1452 /*
1453 * Look for the last active entry (i).
1454 */
1455 for (i = db - 1; i > 0; i--) {
1456 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1457 break;
1458 }
1459 /*
1460 * Copy the table down so inactive entries at the
1461 * end are removed.
1462 */
1463 memmove(&bestsp[db - i], bestsp,
1464 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1465 be32_add_cpu(&ltp->bestcount, -(db - i));
1466 xfs_dir3_leaf_log_tail(tp, lbp);
1467 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1468 } else
1469 bestsp[db] = cpu_to_be16(NULLDATAOFF);
1470 }
1471 /*
1472 * If the data block was not the first one, drop it.
1473 */
1474 else if (db != args->geo->datablk)
1475 dbp = NULL;
1476
1477 xfs_dir3_leaf_check(dp, lbp);
1478 /*
1479 * See if we can convert to block form.
1480 */
1481 return xfs_dir2_leaf_to_block(args, lbp, dbp);
1482 }
1483
1484 /*
1485 * Replace the inode number in a leaf format directory entry.
1486 */
1487 int /* error */
1488 xfs_dir2_leaf_replace(
1489 xfs_da_args_t *args) /* operation arguments */
1490 {
1491 struct xfs_buf *dbp; /* data block buffer */
1492 xfs_dir2_data_entry_t *dep; /* data block entry */
1493 xfs_inode_t *dp; /* incore directory inode */
1494 int error; /* error return code */
1495 int index; /* index of leaf entry */
1496 struct xfs_buf *lbp; /* leaf buffer */
1497 xfs_dir2_leaf_t *leaf; /* leaf structure */
1498 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1499 xfs_trans_t *tp; /* transaction pointer */
1500 struct xfs_dir2_leaf_entry *ents;
1501
1502 trace_xfs_dir2_leaf_replace(args);
1503
1504 /*
1505 * Look up the entry.
1506 */
1507 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1508 return error;
1509 }
1510 dp = args->dp;
1511 leaf = lbp->b_addr;
1512 ents = dp->d_ops->leaf_ents_p(leaf);
1513 /*
1514 * Point to the leaf entry, get data address from it.
1515 */
1516 lep = &ents[index];
1517 /*
1518 * Point to the data entry.
1519 */
1520 dep = (xfs_dir2_data_entry_t *)
1521 ((char *)dbp->b_addr +
1522 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1523 ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1524 /*
1525 * Put the new inode number in, log it.
1526 */
1527 dep->inumber = cpu_to_be64(args->inumber);
1528 dp->d_ops->data_put_ftype(dep, args->filetype);
1529 tp = args->trans;
1530 xfs_dir2_data_log_entry(tp, dp, dbp, dep);
1531 xfs_dir3_leaf_check(dp, lbp);
1532 xfs_trans_brelse(tp, lbp);
1533 return 0;
1534 }
1535
1536 /*
1537 * Return index in the leaf block (lbp) which is either the first
1538 * one with this hash value, or if there are none, the insert point
1539 * for that hash value.
1540 */
1541 int /* index value */
1542 xfs_dir2_leaf_search_hash(
1543 xfs_da_args_t *args, /* operation arguments */
1544 struct xfs_buf *lbp) /* leaf buffer */
1545 {
1546 xfs_dahash_t hash=0; /* hash from this entry */
1547 xfs_dahash_t hashwant; /* hash value looking for */
1548 int high; /* high leaf index */
1549 int low; /* low leaf index */
1550 xfs_dir2_leaf_t *leaf; /* leaf structure */
1551 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1552 int mid=0; /* current leaf index */
1553 struct xfs_dir2_leaf_entry *ents;
1554 struct xfs_dir3_icleaf_hdr leafhdr;
1555
1556 leaf = lbp->b_addr;
1557 ents = args->dp->d_ops->leaf_ents_p(leaf);
1558 args->dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1559
1560 /*
1561 * Note, the table cannot be empty, so we have to go through the loop.
1562 * Binary search the leaf entries looking for our hash value.
1563 */
1564 for (lep = ents, low = 0, high = leafhdr.count - 1,
1565 hashwant = args->hashval;
1566 low <= high; ) {
1567 mid = (low + high) >> 1;
1568 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1569 break;
1570 if (hash < hashwant)
1571 low = mid + 1;
1572 else
1573 high = mid - 1;
1574 }
1575 /*
1576 * Found one, back up through all the equal hash values.
1577 */
1578 if (hash == hashwant) {
1579 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1580 mid--;
1581 }
1582 }
1583 /*
1584 * Need to point to an entry higher than ours.
1585 */
1586 else if (hash < hashwant)
1587 mid++;
1588 return mid;
1589 }
1590
1591 /*
1592 * Trim off a trailing data block. We know it's empty since the leaf
1593 * freespace table says so.
1594 */
1595 int /* error */
1596 xfs_dir2_leaf_trim_data(
1597 xfs_da_args_t *args, /* operation arguments */
1598 struct xfs_buf *lbp, /* leaf buffer */
1599 xfs_dir2_db_t db) /* data block number */
1600 {
1601 __be16 *bestsp; /* leaf bests table */
1602 struct xfs_buf *dbp; /* data block buffer */
1603 xfs_inode_t *dp; /* incore directory inode */
1604 int error; /* error return value */
1605 xfs_dir2_leaf_t *leaf; /* leaf structure */
1606 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1607 xfs_mount_t *mp; /* filesystem mount point */
1608 xfs_trans_t *tp; /* transaction pointer */
1609
1610 dp = args->dp;
1611 mp = dp->i_mount;
1612 tp = args->trans;
1613 /*
1614 * Read the offending data block. We need its buffer.
1615 */
1616 error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(args->geo, db),
1617 -1, &dbp);
1618 if (error)
1619 return error;
1620
1621 leaf = lbp->b_addr;
1622 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1623
1624 #ifdef DEBUG
1625 {
1626 struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1627 struct xfs_dir2_data_free *bf = dp->d_ops->data_bestfree_p(hdr);
1628
1629 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1630 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1631 ASSERT(be16_to_cpu(bf[0].length) ==
1632 args->geo->blksize - dp->d_ops->data_entry_offset);
1633 ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1634 }
1635 #endif
1636
1637 /*
1638 * Get rid of the data block.
1639 */
1640 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1641 ASSERT(error != ENOSPC);
1642 xfs_trans_brelse(tp, dbp);
1643 return error;
1644 }
1645 /*
1646 * Eliminate the last bests entry from the table.
1647 */
1648 bestsp = xfs_dir2_leaf_bests_p(ltp);
1649 be32_add_cpu(&ltp->bestcount, -1);
1650 memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1651 xfs_dir3_leaf_log_tail(tp, lbp);
1652 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1653 return 0;
1654 }
1655
1656 static inline size_t
1657 xfs_dir3_leaf_size(
1658 struct xfs_dir3_icleaf_hdr *hdr,
1659 int counts)
1660 {
1661 int entries;
1662 int hdrsize;
1663
1664 entries = hdr->count - hdr->stale;
1665 if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1666 hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1667 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1668 else
1669 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1670
1671 return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1672 + counts * sizeof(xfs_dir2_data_off_t)
1673 + sizeof(xfs_dir2_leaf_tail_t);
1674 }
1675
1676 /*
1677 * Convert node form directory to leaf form directory.
1678 * The root of the node form dir needs to already be a LEAFN block.
1679 * Just return if we can't do anything.
1680 */
1681 int /* error */
1682 xfs_dir2_node_to_leaf(
1683 xfs_da_state_t *state) /* directory operation state */
1684 {
1685 xfs_da_args_t *args; /* operation arguments */
1686 xfs_inode_t *dp; /* incore directory inode */
1687 int error; /* error return code */
1688 struct xfs_buf *fbp; /* buffer for freespace block */
1689 xfs_fileoff_t fo; /* freespace file offset */
1690 xfs_dir2_free_t *free; /* freespace structure */
1691 struct xfs_buf *lbp; /* buffer for leaf block */
1692 xfs_dir2_leaf_tail_t *ltp; /* tail of leaf structure */
1693 xfs_dir2_leaf_t *leaf; /* leaf structure */
1694 xfs_mount_t *mp; /* filesystem mount point */
1695 int rval; /* successful free trim? */
1696 xfs_trans_t *tp; /* transaction pointer */
1697 struct xfs_dir3_icleaf_hdr leafhdr;
1698 struct xfs_dir3_icfree_hdr freehdr;
1699
1700 /*
1701 * There's more than a leaf level in the btree, so there must
1702 * be multiple leafn blocks. Give up.
1703 */
1704 if (state->path.active > 1)
1705 return 0;
1706 args = state->args;
1707
1708 trace_xfs_dir2_node_to_leaf(args);
1709
1710 mp = state->mp;
1711 dp = args->dp;
1712 tp = args->trans;
1713 /*
1714 * Get the last offset in the file.
1715 */
1716 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1717 return error;
1718 }
1719 fo -= args->geo->fsbcount;
1720 /*
1721 * If there are freespace blocks other than the first one,
1722 * take this opportunity to remove trailing empty freespace blocks
1723 * that may have been left behind during no-space-reservation
1724 * operations.
1725 */
1726 while (fo > args->geo->freeblk) {
1727 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1728 return error;
1729 }
1730 if (rval)
1731 fo -= args->geo->fsbcount;
1732 else
1733 return 0;
1734 }
1735 /*
1736 * Now find the block just before the freespace block.
1737 */
1738 if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1739 return error;
1740 }
1741 /*
1742 * If it's not the single leaf block, give up.
1743 */
1744 if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1745 return 0;
1746 lbp = state->path.blk[0].bp;
1747 leaf = lbp->b_addr;
1748 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1749
1750 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1751 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1752
1753 /*
1754 * Read the freespace block.
1755 */
1756 error = xfs_dir2_free_read(tp, dp, args->geo->freeblk, &fbp);
1757 if (error)
1758 return error;
1759 free = fbp->b_addr;
1760 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1761
1762 ASSERT(!freehdr.firstdb);
1763
1764 /*
1765 * Now see if the leafn and free data will fit in a leaf1.
1766 * If not, release the buffer and give up.
1767 */
1768 if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > args->geo->blksize) {
1769 xfs_trans_brelse(tp, fbp);
1770 return 0;
1771 }
1772
1773 /*
1774 * If the leaf has any stale entries in it, compress them out.
1775 */
1776 if (leafhdr.stale)
1777 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1778
1779 lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1780 xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1781 leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1782 ? XFS_DIR2_LEAF1_MAGIC
1783 : XFS_DIR3_LEAF1_MAGIC;
1784
1785 /*
1786 * Set up the leaf tail from the freespace block.
1787 */
1788 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1789 ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1790
1791 /*
1792 * Set up the leaf bests table.
1793 */
1794 memcpy(xfs_dir2_leaf_bests_p(ltp), dp->d_ops->free_bests_p(free),
1795 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1796
1797 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1798 xfs_dir3_leaf_log_header(tp, dp, lbp);
1799 xfs_dir3_leaf_log_bests(tp, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1800 xfs_dir3_leaf_log_tail(tp, lbp);
1801 xfs_dir3_leaf_check(dp, lbp);
1802
1803 /*
1804 * Get rid of the freespace block.
1805 */
1806 error = xfs_dir2_shrink_inode(args,
1807 xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET),
1808 fbp);
1809 if (error) {
1810 /*
1811 * This can't fail here because it can only happen when
1812 * punching out the middle of an extent, and this is an
1813 * isolated block.
1814 */
1815 ASSERT(error != ENOSPC);
1816 return error;
1817 }
1818 fbp = NULL;
1819 /*
1820 * Now see if we can convert the single-leaf directory
1821 * down to a block form directory.
1822 * This routine always kills the dabuf for the leaf, so
1823 * eliminate it from the path.
1824 */
1825 error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1826 state->path.blk[0].bp = NULL;
1827 return error;
1828 }
Linux kernel - Deliverables
This page took 0.068038 seconds and 5 git commands to generate.