]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/xfs/libxfs/xfs_dir2_leaf.c
Merge branch 'xfs-misc-fixes-for-4.4-2' into for-next
[mirror_ubuntu-artful-kernel.git] / fs / xfs / 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 "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_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_error.h"
32 #include "xfs_trace.h"
33 #include "xfs_trans.h"
34 #include "xfs_buf_item.h"
35 #include "xfs_cksum.h"
36 #include "xfs_log.h"
37
38 /*
39 * Local function declarations.
40 */
41 static int xfs_dir2_leaf_lookup_int(xfs_da_args_t *args, struct xfs_buf **lbpp,
42 int *indexp, struct xfs_buf **dbpp);
43 static void xfs_dir3_leaf_log_bests(struct xfs_da_args *args,
44 struct xfs_buf *bp, int first, int last);
45 static void xfs_dir3_leaf_log_tail(struct xfs_da_args *args,
46 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_meta_uuid))
165 return false;
166 if (be64_to_cpu(leaf3->info.blkno) != bp->b_bn)
167 return false;
168 if (!xfs_log_check_lsn(mp, be64_to_cpu(leaf3->info.lsn)))
169 return false;
170 } else {
171 if (leaf->hdr.info.magic != cpu_to_be16(magic))
172 return false;
173 }
174
175 return xfs_dir3_leaf_check_int(mp, NULL, NULL, leaf);
176 }
177
178 static void
179 __read_verify(
180 struct xfs_buf *bp,
181 __uint16_t magic)
182 {
183 struct xfs_mount *mp = bp->b_target->bt_mount;
184
185 if (xfs_sb_version_hascrc(&mp->m_sb) &&
186 !xfs_buf_verify_cksum(bp, XFS_DIR3_LEAF_CRC_OFF))
187 xfs_buf_ioerror(bp, -EFSBADCRC);
188 else if (!xfs_dir3_leaf_verify(bp, magic))
189 xfs_buf_ioerror(bp, -EFSCORRUPTED);
190
191 if (bp->b_error)
192 xfs_verifier_error(bp);
193 }
194
195 static void
196 __write_verify(
197 struct xfs_buf *bp,
198 __uint16_t magic)
199 {
200 struct xfs_mount *mp = bp->b_target->bt_mount;
201 struct xfs_buf_log_item *bip = bp->b_fspriv;
202 struct xfs_dir3_leaf_hdr *hdr3 = bp->b_addr;
203
204 if (!xfs_dir3_leaf_verify(bp, magic)) {
205 xfs_buf_ioerror(bp, -EFSCORRUPTED);
206 xfs_verifier_error(bp);
207 return;
208 }
209
210 if (!xfs_sb_version_hascrc(&mp->m_sb))
211 return;
212
213 if (bip)
214 hdr3->info.lsn = cpu_to_be64(bip->bli_item.li_lsn);
215
216 xfs_buf_update_cksum(bp, XFS_DIR3_LEAF_CRC_OFF);
217 }
218
219 static void
220 xfs_dir3_leaf1_read_verify(
221 struct xfs_buf *bp)
222 {
223 __read_verify(bp, XFS_DIR2_LEAF1_MAGIC);
224 }
225
226 static void
227 xfs_dir3_leaf1_write_verify(
228 struct xfs_buf *bp)
229 {
230 __write_verify(bp, XFS_DIR2_LEAF1_MAGIC);
231 }
232
233 static void
234 xfs_dir3_leafn_read_verify(
235 struct xfs_buf *bp)
236 {
237 __read_verify(bp, XFS_DIR2_LEAFN_MAGIC);
238 }
239
240 static void
241 xfs_dir3_leafn_write_verify(
242 struct xfs_buf *bp)
243 {
244 __write_verify(bp, XFS_DIR2_LEAFN_MAGIC);
245 }
246
247 const struct xfs_buf_ops xfs_dir3_leaf1_buf_ops = {
248 .verify_read = xfs_dir3_leaf1_read_verify,
249 .verify_write = xfs_dir3_leaf1_write_verify,
250 };
251
252 const struct xfs_buf_ops xfs_dir3_leafn_buf_ops = {
253 .verify_read = xfs_dir3_leafn_read_verify,
254 .verify_write = xfs_dir3_leafn_write_verify,
255 };
256
257 static int
258 xfs_dir3_leaf_read(
259 struct xfs_trans *tp,
260 struct xfs_inode *dp,
261 xfs_dablk_t fbno,
262 xfs_daddr_t mappedbno,
263 struct xfs_buf **bpp)
264 {
265 int err;
266
267 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
268 XFS_DATA_FORK, &xfs_dir3_leaf1_buf_ops);
269 if (!err && tp)
270 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAF1_BUF);
271 return err;
272 }
273
274 int
275 xfs_dir3_leafn_read(
276 struct xfs_trans *tp,
277 struct xfs_inode *dp,
278 xfs_dablk_t fbno,
279 xfs_daddr_t mappedbno,
280 struct xfs_buf **bpp)
281 {
282 int err;
283
284 err = xfs_da_read_buf(tp, dp, fbno, mappedbno, bpp,
285 XFS_DATA_FORK, &xfs_dir3_leafn_buf_ops);
286 if (!err && tp)
287 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_LEAFN_BUF);
288 return err;
289 }
290
291 /*
292 * Initialize a new leaf block, leaf1 or leafn magic accepted.
293 */
294 static void
295 xfs_dir3_leaf_init(
296 struct xfs_mount *mp,
297 struct xfs_trans *tp,
298 struct xfs_buf *bp,
299 xfs_ino_t owner,
300 __uint16_t type)
301 {
302 struct xfs_dir2_leaf *leaf = bp->b_addr;
303
304 ASSERT(type == XFS_DIR2_LEAF1_MAGIC || type == XFS_DIR2_LEAFN_MAGIC);
305
306 if (xfs_sb_version_hascrc(&mp->m_sb)) {
307 struct xfs_dir3_leaf_hdr *leaf3 = bp->b_addr;
308
309 memset(leaf3, 0, sizeof(*leaf3));
310
311 leaf3->info.hdr.magic = (type == XFS_DIR2_LEAF1_MAGIC)
312 ? cpu_to_be16(XFS_DIR3_LEAF1_MAGIC)
313 : cpu_to_be16(XFS_DIR3_LEAFN_MAGIC);
314 leaf3->info.blkno = cpu_to_be64(bp->b_bn);
315 leaf3->info.owner = cpu_to_be64(owner);
316 uuid_copy(&leaf3->info.uuid, &mp->m_sb.sb_meta_uuid);
317 } else {
318 memset(leaf, 0, sizeof(*leaf));
319 leaf->hdr.info.magic = cpu_to_be16(type);
320 }
321
322 /*
323 * If it's a leaf-format directory initialize the tail.
324 * Caller is responsible for initialising the bests table.
325 */
326 if (type == XFS_DIR2_LEAF1_MAGIC) {
327 struct xfs_dir2_leaf_tail *ltp;
328
329 ltp = xfs_dir2_leaf_tail_p(mp->m_dir_geo, leaf);
330 ltp->bestcount = 0;
331 bp->b_ops = &xfs_dir3_leaf1_buf_ops;
332 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAF1_BUF);
333 } else {
334 bp->b_ops = &xfs_dir3_leafn_buf_ops;
335 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_LEAFN_BUF);
336 }
337 }
338
339 int
340 xfs_dir3_leaf_get_buf(
341 xfs_da_args_t *args,
342 xfs_dir2_db_t bno,
343 struct xfs_buf **bpp,
344 __uint16_t magic)
345 {
346 struct xfs_inode *dp = args->dp;
347 struct xfs_trans *tp = args->trans;
348 struct xfs_mount *mp = dp->i_mount;
349 struct xfs_buf *bp;
350 int error;
351
352 ASSERT(magic == XFS_DIR2_LEAF1_MAGIC || magic == XFS_DIR2_LEAFN_MAGIC);
353 ASSERT(bno >= xfs_dir2_byte_to_db(args->geo, XFS_DIR2_LEAF_OFFSET) &&
354 bno < xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET));
355
356 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, bno),
357 -1, &bp, XFS_DATA_FORK);
358 if (error)
359 return error;
360
361 xfs_dir3_leaf_init(mp, tp, bp, dp->i_ino, magic);
362 xfs_dir3_leaf_log_header(args, bp);
363 if (magic == XFS_DIR2_LEAF1_MAGIC)
364 xfs_dir3_leaf_log_tail(args, bp);
365 *bpp = bp;
366 return 0;
367 }
368
369 /*
370 * Convert a block form directory to a leaf form directory.
371 */
372 int /* error */
373 xfs_dir2_block_to_leaf(
374 xfs_da_args_t *args, /* operation arguments */
375 struct xfs_buf *dbp) /* input block's buffer */
376 {
377 __be16 *bestsp; /* leaf's bestsp entries */
378 xfs_dablk_t blkno; /* leaf block's bno */
379 xfs_dir2_data_hdr_t *hdr; /* block header */
380 xfs_dir2_leaf_entry_t *blp; /* block's leaf entries */
381 xfs_dir2_block_tail_t *btp; /* block's tail */
382 xfs_inode_t *dp; /* incore directory inode */
383 int error; /* error return code */
384 struct xfs_buf *lbp; /* leaf block's buffer */
385 xfs_dir2_db_t ldb; /* leaf block's bno */
386 xfs_dir2_leaf_t *leaf; /* leaf structure */
387 xfs_dir2_leaf_tail_t *ltp; /* leaf's tail */
388 int needlog; /* need to log block header */
389 int needscan; /* need to rescan bestfree */
390 xfs_trans_t *tp; /* transaction pointer */
391 struct xfs_dir2_data_free *bf;
392 struct xfs_dir2_leaf_entry *ents;
393 struct xfs_dir3_icleaf_hdr leafhdr;
394
395 trace_xfs_dir2_block_to_leaf(args);
396
397 dp = args->dp;
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(args, 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(args, 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(args, 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(args, dbp);
474 xfs_dir3_leaf_check(dp, lbp);
475 xfs_dir3_data_check(dp, dbp);
476 xfs_dir3_leaf_log_bests(args, 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 int needbytes; /* leaf block bytes needed */
629 int needlog; /* need to log data header */
630 int needscan; /* need to rescan data free */
631 __be16 *tagp; /* end of data entry */
632 xfs_trans_t *tp; /* transaction pointer */
633 xfs_dir2_db_t use_block; /* data block number */
634 struct xfs_dir2_data_free *bf; /* bestfree table */
635 struct xfs_dir2_leaf_entry *ents;
636 struct xfs_dir3_icleaf_hdr leafhdr;
637
638 trace_xfs_dir2_leaf_addname(args);
639
640 dp = args->dp;
641 tp = args->trans;
642
643 error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
644 if (error)
645 return error;
646
647 /*
648 * Look up the entry by hash value and name.
649 * We know it's not there, our caller has already done a lookup.
650 * So the index is of the entry to insert in front of.
651 * But if there are dup hash values the index is of the first of those.
652 */
653 index = xfs_dir2_leaf_search_hash(args, lbp);
654 leaf = lbp->b_addr;
655 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
656 ents = dp->d_ops->leaf_ents_p(leaf);
657 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
658 bestsp = xfs_dir2_leaf_bests_p(ltp);
659 length = dp->d_ops->data_entsize(args->namelen);
660
661 /*
662 * See if there are any entries with the same hash value
663 * and space in their block for the new entry.
664 * This is good because it puts multiple same-hash value entries
665 * in a data block, improving the lookup of those entries.
666 */
667 for (use_block = -1, lep = &ents[index];
668 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
669 index++, lep++) {
670 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
671 continue;
672 i = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
673 ASSERT(i < be32_to_cpu(ltp->bestcount));
674 ASSERT(bestsp[i] != cpu_to_be16(NULLDATAOFF));
675 if (be16_to_cpu(bestsp[i]) >= length) {
676 use_block = i;
677 break;
678 }
679 }
680 /*
681 * Didn't find a block yet, linear search all the data blocks.
682 */
683 if (use_block == -1) {
684 for (i = 0; i < be32_to_cpu(ltp->bestcount); i++) {
685 /*
686 * Remember a block we see that's missing.
687 */
688 if (bestsp[i] == cpu_to_be16(NULLDATAOFF) &&
689 use_block == -1)
690 use_block = i;
691 else if (be16_to_cpu(bestsp[i]) >= length) {
692 use_block = i;
693 break;
694 }
695 }
696 }
697 /*
698 * How many bytes do we need in the leaf block?
699 */
700 needbytes = 0;
701 if (!leafhdr.stale)
702 needbytes += sizeof(xfs_dir2_leaf_entry_t);
703 if (use_block == -1)
704 needbytes += sizeof(xfs_dir2_data_off_t);
705
706 /*
707 * Now kill use_block if it refers to a missing block, so we
708 * can use it as an indication of allocation needed.
709 */
710 if (use_block != -1 && bestsp[use_block] == cpu_to_be16(NULLDATAOFF))
711 use_block = -1;
712 /*
713 * If we don't have enough free bytes but we can make enough
714 * by compacting out stale entries, we'll do that.
715 */
716 if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes &&
717 leafhdr.stale > 1)
718 compact = 1;
719
720 /*
721 * Otherwise if we don't have enough free bytes we need to
722 * convert to node form.
723 */
724 else if ((char *)bestsp - (char *)&ents[leafhdr.count] < needbytes) {
725 /*
726 * Just checking or no space reservation, give up.
727 */
728 if ((args->op_flags & XFS_DA_OP_JUSTCHECK) ||
729 args->total == 0) {
730 xfs_trans_brelse(tp, lbp);
731 return -ENOSPC;
732 }
733 /*
734 * Convert to node form.
735 */
736 error = xfs_dir2_leaf_to_node(args, lbp);
737 if (error)
738 return error;
739 /*
740 * Then add the new entry.
741 */
742 return xfs_dir2_node_addname(args);
743 }
744 /*
745 * Otherwise it will fit without compaction.
746 */
747 else
748 compact = 0;
749 /*
750 * If just checking, then it will fit unless we needed to allocate
751 * a new data block.
752 */
753 if (args->op_flags & XFS_DA_OP_JUSTCHECK) {
754 xfs_trans_brelse(tp, lbp);
755 return use_block == -1 ? -ENOSPC : 0;
756 }
757 /*
758 * If no allocations are allowed, return now before we've
759 * changed anything.
760 */
761 if (args->total == 0 && use_block == -1) {
762 xfs_trans_brelse(tp, lbp);
763 return -ENOSPC;
764 }
765 /*
766 * Need to compact the leaf entries, removing stale ones.
767 * Leave one stale entry behind - the one closest to our
768 * insertion index - and we'll shift that one to our insertion
769 * point later.
770 */
771 if (compact) {
772 xfs_dir3_leaf_compact_x1(&leafhdr, ents, &index, &lowstale,
773 &highstale, &lfloglow, &lfloghigh);
774 }
775 /*
776 * There are stale entries, so we'll need log-low and log-high
777 * impossibly bad values later.
778 */
779 else if (leafhdr.stale) {
780 lfloglow = leafhdr.count;
781 lfloghigh = -1;
782 }
783 /*
784 * If there was no data block space found, we need to allocate
785 * a new one.
786 */
787 if (use_block == -1) {
788 /*
789 * Add the new data block.
790 */
791 if ((error = xfs_dir2_grow_inode(args, XFS_DIR2_DATA_SPACE,
792 &use_block))) {
793 xfs_trans_brelse(tp, lbp);
794 return error;
795 }
796 /*
797 * Initialize the block.
798 */
799 if ((error = xfs_dir3_data_init(args, use_block, &dbp))) {
800 xfs_trans_brelse(tp, lbp);
801 return error;
802 }
803 /*
804 * If we're adding a new data block on the end we need to
805 * extend the bests table. Copy it up one entry.
806 */
807 if (use_block >= be32_to_cpu(ltp->bestcount)) {
808 bestsp--;
809 memmove(&bestsp[0], &bestsp[1],
810 be32_to_cpu(ltp->bestcount) * sizeof(bestsp[0]));
811 be32_add_cpu(&ltp->bestcount, 1);
812 xfs_dir3_leaf_log_tail(args, lbp);
813 xfs_dir3_leaf_log_bests(args, lbp, 0,
814 be32_to_cpu(ltp->bestcount) - 1);
815 }
816 /*
817 * If we're filling in a previously empty block just log it.
818 */
819 else
820 xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
821 hdr = dbp->b_addr;
822 bf = dp->d_ops->data_bestfree_p(hdr);
823 bestsp[use_block] = bf[0].length;
824 grown = 1;
825 } else {
826 /*
827 * Already had space in some data block.
828 * Just read that one in.
829 */
830 error = xfs_dir3_data_read(tp, dp,
831 xfs_dir2_db_to_da(args->geo, use_block),
832 -1, &dbp);
833 if (error) {
834 xfs_trans_brelse(tp, lbp);
835 return error;
836 }
837 hdr = dbp->b_addr;
838 bf = dp->d_ops->data_bestfree_p(hdr);
839 grown = 0;
840 }
841 /*
842 * Point to the biggest freespace in our data block.
843 */
844 dup = (xfs_dir2_data_unused_t *)
845 ((char *)hdr + be16_to_cpu(bf[0].offset));
846 ASSERT(be16_to_cpu(dup->length) >= length);
847 needscan = needlog = 0;
848 /*
849 * Mark the initial part of our freespace in use for the new entry.
850 */
851 xfs_dir2_data_use_free(args, dbp, dup,
852 (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr), length,
853 &needlog, &needscan);
854 /*
855 * Initialize our new entry (at last).
856 */
857 dep = (xfs_dir2_data_entry_t *)dup;
858 dep->inumber = cpu_to_be64(args->inumber);
859 dep->namelen = args->namelen;
860 memcpy(dep->name, args->name, dep->namelen);
861 dp->d_ops->data_put_ftype(dep, args->filetype);
862 tagp = dp->d_ops->data_entry_tag_p(dep);
863 *tagp = cpu_to_be16((char *)dep - (char *)hdr);
864 /*
865 * Need to scan fix up the bestfree table.
866 */
867 if (needscan)
868 xfs_dir2_data_freescan(dp, hdr, &needlog);
869 /*
870 * Need to log the data block's header.
871 */
872 if (needlog)
873 xfs_dir2_data_log_header(args, dbp);
874 xfs_dir2_data_log_entry(args, dbp, dep);
875 /*
876 * If the bests table needs to be changed, do it.
877 * Log the change unless we've already done that.
878 */
879 if (be16_to_cpu(bestsp[use_block]) != be16_to_cpu(bf[0].length)) {
880 bestsp[use_block] = bf[0].length;
881 if (!grown)
882 xfs_dir3_leaf_log_bests(args, lbp, use_block, use_block);
883 }
884
885 lep = xfs_dir3_leaf_find_entry(&leafhdr, ents, index, compact, lowstale,
886 highstale, &lfloglow, &lfloghigh);
887
888 /*
889 * Fill in the new leaf entry.
890 */
891 lep->hashval = cpu_to_be32(args->hashval);
892 lep->address = cpu_to_be32(
893 xfs_dir2_db_off_to_dataptr(args->geo, use_block,
894 be16_to_cpu(*tagp)));
895 /*
896 * Log the leaf fields and give up the buffers.
897 */
898 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
899 xfs_dir3_leaf_log_header(args, lbp);
900 xfs_dir3_leaf_log_ents(args, lbp, lfloglow, lfloghigh);
901 xfs_dir3_leaf_check(dp, lbp);
902 xfs_dir3_data_check(dp, dbp);
903 return 0;
904 }
905
906 /*
907 * Compact out any stale entries in the leaf.
908 * Log the header and changed leaf entries, if any.
909 */
910 void
911 xfs_dir3_leaf_compact(
912 xfs_da_args_t *args, /* operation arguments */
913 struct xfs_dir3_icleaf_hdr *leafhdr,
914 struct xfs_buf *bp) /* leaf buffer */
915 {
916 int from; /* source leaf index */
917 xfs_dir2_leaf_t *leaf; /* leaf structure */
918 int loglow; /* first leaf entry to log */
919 int to; /* target leaf index */
920 struct xfs_dir2_leaf_entry *ents;
921 struct xfs_inode *dp = args->dp;
922
923 leaf = bp->b_addr;
924 if (!leafhdr->stale)
925 return;
926
927 /*
928 * Compress out the stale entries in place.
929 */
930 ents = dp->d_ops->leaf_ents_p(leaf);
931 for (from = to = 0, loglow = -1; from < leafhdr->count; from++) {
932 if (ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
933 continue;
934 /*
935 * Only actually copy the entries that are different.
936 */
937 if (from > to) {
938 if (loglow == -1)
939 loglow = to;
940 ents[to] = ents[from];
941 }
942 to++;
943 }
944 /*
945 * Update and log the header, log the leaf entries.
946 */
947 ASSERT(leafhdr->stale == from - to);
948 leafhdr->count -= leafhdr->stale;
949 leafhdr->stale = 0;
950
951 dp->d_ops->leaf_hdr_to_disk(leaf, leafhdr);
952 xfs_dir3_leaf_log_header(args, bp);
953 if (loglow != -1)
954 xfs_dir3_leaf_log_ents(args, bp, loglow, to - 1);
955 }
956
957 /*
958 * Compact the leaf entries, removing stale ones.
959 * Leave one stale entry behind - the one closest to our
960 * insertion index - and the caller will shift that one to our insertion
961 * point later.
962 * Return new insertion index, where the remaining stale entry is,
963 * and leaf logging indices.
964 */
965 void
966 xfs_dir3_leaf_compact_x1(
967 struct xfs_dir3_icleaf_hdr *leafhdr,
968 struct xfs_dir2_leaf_entry *ents,
969 int *indexp, /* insertion index */
970 int *lowstalep, /* out: stale entry before us */
971 int *highstalep, /* out: stale entry after us */
972 int *lowlogp, /* out: low log index */
973 int *highlogp) /* out: high log index */
974 {
975 int from; /* source copy index */
976 int highstale; /* stale entry at/after index */
977 int index; /* insertion index */
978 int keepstale; /* source index of kept stale */
979 int lowstale; /* stale entry before index */
980 int newindex=0; /* new insertion index */
981 int to; /* destination copy index */
982
983 ASSERT(leafhdr->stale > 1);
984 index = *indexp;
985
986 xfs_dir3_leaf_find_stale(leafhdr, ents, index, &lowstale, &highstale);
987
988 /*
989 * Pick the better of lowstale and highstale.
990 */
991 if (lowstale >= 0 &&
992 (highstale == leafhdr->count ||
993 index - lowstale <= highstale - index))
994 keepstale = lowstale;
995 else
996 keepstale = highstale;
997 /*
998 * Copy the entries in place, removing all the stale entries
999 * except keepstale.
1000 */
1001 for (from = to = 0; from < leafhdr->count; from++) {
1002 /*
1003 * Notice the new value of index.
1004 */
1005 if (index == from)
1006 newindex = to;
1007 if (from != keepstale &&
1008 ents[from].address == cpu_to_be32(XFS_DIR2_NULL_DATAPTR)) {
1009 if (from == to)
1010 *lowlogp = to;
1011 continue;
1012 }
1013 /*
1014 * Record the new keepstale value for the insertion.
1015 */
1016 if (from == keepstale)
1017 lowstale = highstale = to;
1018 /*
1019 * Copy only the entries that have moved.
1020 */
1021 if (from > to)
1022 ents[to] = ents[from];
1023 to++;
1024 }
1025 ASSERT(from > to);
1026 /*
1027 * If the insertion point was past the last entry,
1028 * set the new insertion point accordingly.
1029 */
1030 if (index == from)
1031 newindex = to;
1032 *indexp = newindex;
1033 /*
1034 * Adjust the leaf header values.
1035 */
1036 leafhdr->count -= from - to;
1037 leafhdr->stale = 1;
1038 /*
1039 * Remember the low/high stale value only in the "right"
1040 * direction.
1041 */
1042 if (lowstale >= newindex)
1043 lowstale = -1;
1044 else
1045 highstale = leafhdr->count;
1046 *highlogp = leafhdr->count - 1;
1047 *lowstalep = lowstale;
1048 *highstalep = highstale;
1049 }
1050
1051 /*
1052 * Log the bests entries indicated from a leaf1 block.
1053 */
1054 static void
1055 xfs_dir3_leaf_log_bests(
1056 struct xfs_da_args *args,
1057 struct xfs_buf *bp, /* leaf buffer */
1058 int first, /* first entry to log */
1059 int last) /* last entry to log */
1060 {
1061 __be16 *firstb; /* pointer to first entry */
1062 __be16 *lastb; /* pointer to last entry */
1063 struct xfs_dir2_leaf *leaf = bp->b_addr;
1064 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1065
1066 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1067 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC));
1068
1069 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1070 firstb = xfs_dir2_leaf_bests_p(ltp) + first;
1071 lastb = xfs_dir2_leaf_bests_p(ltp) + last;
1072 xfs_trans_log_buf(args->trans, bp,
1073 (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_da_args *args,
1083 struct xfs_buf *bp,
1084 int first,
1085 int last)
1086 {
1087 xfs_dir2_leaf_entry_t *firstlep; /* pointer to first entry */
1088 xfs_dir2_leaf_entry_t *lastlep; /* pointer to last entry */
1089 struct xfs_dir2_leaf *leaf = bp->b_addr;
1090 struct xfs_dir2_leaf_entry *ents;
1091
1092 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1093 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1094 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1095 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1096
1097 ents = args->dp->d_ops->leaf_ents_p(leaf);
1098 firstlep = &ents[first];
1099 lastlep = &ents[last];
1100 xfs_trans_log_buf(args->trans, bp,
1101 (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_da_args *args,
1111 struct xfs_buf *bp)
1112 {
1113 struct xfs_dir2_leaf *leaf = bp->b_addr;
1114
1115 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1116 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1117 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1118 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1119
1120 xfs_trans_log_buf(args->trans, bp,
1121 (uint)((char *)&leaf->hdr - (char *)leaf),
1122 args->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_da_args *args,
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
1136 ASSERT(leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAF1_MAGIC) ||
1137 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAF1_MAGIC) ||
1138 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR2_LEAFN_MAGIC) ||
1139 leaf->hdr.info.magic == cpu_to_be16(XFS_DIR3_LEAFN_MAGIC));
1140
1141 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1142 xfs_trans_log_buf(args->trans, bp, (uint)((char *)ltp - (char *)leaf),
1143 (uint)(args->geo->blksize - 1));
1144 }
1145
1146 /*
1147 * Look up the entry referred to by args in the leaf format directory.
1148 * Most of the work is done by the xfs_dir2_leaf_lookup_int routine which
1149 * is also used by the node-format code.
1150 */
1151 int
1152 xfs_dir2_leaf_lookup(
1153 xfs_da_args_t *args) /* operation arguments */
1154 {
1155 struct xfs_buf *dbp; /* data block buffer */
1156 xfs_dir2_data_entry_t *dep; /* data block entry */
1157 xfs_inode_t *dp; /* incore directory inode */
1158 int error; /* error return code */
1159 int index; /* found entry index */
1160 struct xfs_buf *lbp; /* leaf buffer */
1161 xfs_dir2_leaf_t *leaf; /* leaf structure */
1162 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1163 xfs_trans_t *tp; /* transaction pointer */
1164 struct xfs_dir2_leaf_entry *ents;
1165
1166 trace_xfs_dir2_leaf_lookup(args);
1167
1168 /*
1169 * Look up name in the leaf block, returning both buffers and index.
1170 */
1171 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1172 return error;
1173 }
1174 tp = args->trans;
1175 dp = args->dp;
1176 xfs_dir3_leaf_check(dp, lbp);
1177 leaf = lbp->b_addr;
1178 ents = dp->d_ops->leaf_ents_p(leaf);
1179 /*
1180 * Get to the leaf entry and contained data entry address.
1181 */
1182 lep = &ents[index];
1183
1184 /*
1185 * Point to the data entry.
1186 */
1187 dep = (xfs_dir2_data_entry_t *)
1188 ((char *)dbp->b_addr +
1189 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1190 /*
1191 * Return the found inode number & CI name if appropriate
1192 */
1193 args->inumber = be64_to_cpu(dep->inumber);
1194 args->filetype = dp->d_ops->data_get_ftype(dep);
1195 error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
1196 xfs_trans_brelse(tp, dbp);
1197 xfs_trans_brelse(tp, lbp);
1198 return error;
1199 }
1200
1201 /*
1202 * Look up name/hash in the leaf block.
1203 * Fill in indexp with the found index, and dbpp with the data buffer.
1204 * If not found dbpp will be NULL, and ENOENT comes back.
1205 * lbpp will always be filled in with the leaf buffer unless there's an error.
1206 */
1207 static int /* error */
1208 xfs_dir2_leaf_lookup_int(
1209 xfs_da_args_t *args, /* operation arguments */
1210 struct xfs_buf **lbpp, /* out: leaf buffer */
1211 int *indexp, /* out: index in leaf block */
1212 struct xfs_buf **dbpp) /* out: data buffer */
1213 {
1214 xfs_dir2_db_t curdb = -1; /* current data block number */
1215 struct xfs_buf *dbp = NULL; /* data buffer */
1216 xfs_dir2_data_entry_t *dep; /* data entry */
1217 xfs_inode_t *dp; /* incore directory inode */
1218 int error; /* error return code */
1219 int index; /* index in leaf block */
1220 struct xfs_buf *lbp; /* leaf buffer */
1221 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1222 xfs_dir2_leaf_t *leaf; /* leaf structure */
1223 xfs_mount_t *mp; /* filesystem mount point */
1224 xfs_dir2_db_t newdb; /* new data block number */
1225 xfs_trans_t *tp; /* transaction pointer */
1226 xfs_dir2_db_t cidb = -1; /* case match data block no. */
1227 enum xfs_dacmp cmp; /* name compare result */
1228 struct xfs_dir2_leaf_entry *ents;
1229 struct xfs_dir3_icleaf_hdr leafhdr;
1230
1231 dp = args->dp;
1232 tp = args->trans;
1233 mp = dp->i_mount;
1234
1235 error = xfs_dir3_leaf_read(tp, dp, args->geo->leafblk, -1, &lbp);
1236 if (error)
1237 return error;
1238
1239 *lbpp = lbp;
1240 leaf = lbp->b_addr;
1241 xfs_dir3_leaf_check(dp, lbp);
1242 ents = dp->d_ops->leaf_ents_p(leaf);
1243 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1244
1245 /*
1246 * Look for the first leaf entry with our hash value.
1247 */
1248 index = xfs_dir2_leaf_search_hash(args, lbp);
1249 /*
1250 * Loop over all the entries with the right hash value
1251 * looking to match the name.
1252 */
1253 for (lep = &ents[index];
1254 index < leafhdr.count && be32_to_cpu(lep->hashval) == args->hashval;
1255 lep++, index++) {
1256 /*
1257 * Skip over stale leaf entries.
1258 */
1259 if (be32_to_cpu(lep->address) == XFS_DIR2_NULL_DATAPTR)
1260 continue;
1261 /*
1262 * Get the new data block number.
1263 */
1264 newdb = xfs_dir2_dataptr_to_db(args->geo,
1265 be32_to_cpu(lep->address));
1266 /*
1267 * If it's not the same as the old data block number,
1268 * need to pitch the old one and read the new one.
1269 */
1270 if (newdb != curdb) {
1271 if (dbp)
1272 xfs_trans_brelse(tp, dbp);
1273 error = xfs_dir3_data_read(tp, dp,
1274 xfs_dir2_db_to_da(args->geo, newdb),
1275 -1, &dbp);
1276 if (error) {
1277 xfs_trans_brelse(tp, lbp);
1278 return error;
1279 }
1280 curdb = newdb;
1281 }
1282 /*
1283 * Point to the data entry.
1284 */
1285 dep = (xfs_dir2_data_entry_t *)((char *)dbp->b_addr +
1286 xfs_dir2_dataptr_to_off(args->geo,
1287 be32_to_cpu(lep->address)));
1288 /*
1289 * Compare name and if it's an exact match, return the index
1290 * and buffer. If it's the first case-insensitive match, store
1291 * the index and buffer and continue looking for an exact match.
1292 */
1293 cmp = mp->m_dirnameops->compname(args, dep->name, dep->namelen);
1294 if (cmp != XFS_CMP_DIFFERENT && cmp != args->cmpresult) {
1295 args->cmpresult = cmp;
1296 *indexp = index;
1297 /* case exact match: return the current buffer. */
1298 if (cmp == XFS_CMP_EXACT) {
1299 *dbpp = dbp;
1300 return 0;
1301 }
1302 cidb = curdb;
1303 }
1304 }
1305 ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
1306 /*
1307 * Here, we can only be doing a lookup (not a rename or remove).
1308 * If a case-insensitive match was found earlier, re-read the
1309 * appropriate data block if required and return it.
1310 */
1311 if (args->cmpresult == XFS_CMP_CASE) {
1312 ASSERT(cidb != -1);
1313 if (cidb != curdb) {
1314 xfs_trans_brelse(tp, dbp);
1315 error = xfs_dir3_data_read(tp, dp,
1316 xfs_dir2_db_to_da(args->geo, cidb),
1317 -1, &dbp);
1318 if (error) {
1319 xfs_trans_brelse(tp, lbp);
1320 return error;
1321 }
1322 }
1323 *dbpp = dbp;
1324 return 0;
1325 }
1326 /*
1327 * No match found, return -ENOENT.
1328 */
1329 ASSERT(cidb == -1);
1330 if (dbp)
1331 xfs_trans_brelse(tp, dbp);
1332 xfs_trans_brelse(tp, lbp);
1333 return -ENOENT;
1334 }
1335
1336 /*
1337 * Remove an entry from a leaf format directory.
1338 */
1339 int /* error */
1340 xfs_dir2_leaf_removename(
1341 xfs_da_args_t *args) /* operation arguments */
1342 {
1343 __be16 *bestsp; /* leaf block best freespace */
1344 xfs_dir2_data_hdr_t *hdr; /* data block header */
1345 xfs_dir2_db_t db; /* data block number */
1346 struct xfs_buf *dbp; /* data block buffer */
1347 xfs_dir2_data_entry_t *dep; /* data entry structure */
1348 xfs_inode_t *dp; /* incore directory inode */
1349 int error; /* error return code */
1350 xfs_dir2_db_t i; /* temporary data block # */
1351 int index; /* index into leaf entries */
1352 struct xfs_buf *lbp; /* leaf buffer */
1353 xfs_dir2_leaf_t *leaf; /* leaf structure */
1354 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1355 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1356 int needlog; /* need to log data header */
1357 int needscan; /* need to rescan data frees */
1358 xfs_dir2_data_off_t oldbest; /* old value of best free */
1359 struct xfs_dir2_data_free *bf; /* bestfree table */
1360 struct xfs_dir2_leaf_entry *ents;
1361 struct xfs_dir3_icleaf_hdr leafhdr;
1362
1363 trace_xfs_dir2_leaf_removename(args);
1364
1365 /*
1366 * Lookup the leaf entry, get the leaf and data blocks read in.
1367 */
1368 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1369 return error;
1370 }
1371 dp = args->dp;
1372 leaf = lbp->b_addr;
1373 hdr = dbp->b_addr;
1374 xfs_dir3_data_check(dp, dbp);
1375 bf = dp->d_ops->data_bestfree_p(hdr);
1376 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1377 ents = dp->d_ops->leaf_ents_p(leaf);
1378 /*
1379 * Point to the leaf entry, use that to point to the data entry.
1380 */
1381 lep = &ents[index];
1382 db = xfs_dir2_dataptr_to_db(args->geo, be32_to_cpu(lep->address));
1383 dep = (xfs_dir2_data_entry_t *)((char *)hdr +
1384 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1385 needscan = needlog = 0;
1386 oldbest = be16_to_cpu(bf[0].length);
1387 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1388 bestsp = xfs_dir2_leaf_bests_p(ltp);
1389 ASSERT(be16_to_cpu(bestsp[db]) == oldbest);
1390 /*
1391 * Mark the former data entry unused.
1392 */
1393 xfs_dir2_data_make_free(args, dbp,
1394 (xfs_dir2_data_aoff_t)((char *)dep - (char *)hdr),
1395 dp->d_ops->data_entsize(dep->namelen), &needlog, &needscan);
1396 /*
1397 * We just mark the leaf entry stale by putting a null in it.
1398 */
1399 leafhdr.stale++;
1400 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1401 xfs_dir3_leaf_log_header(args, lbp);
1402
1403 lep->address = cpu_to_be32(XFS_DIR2_NULL_DATAPTR);
1404 xfs_dir3_leaf_log_ents(args, lbp, index, index);
1405
1406 /*
1407 * Scan the freespace in the data block again if necessary,
1408 * log the data block header if necessary.
1409 */
1410 if (needscan)
1411 xfs_dir2_data_freescan(dp, hdr, &needlog);
1412 if (needlog)
1413 xfs_dir2_data_log_header(args, dbp);
1414 /*
1415 * If the longest freespace in the data block has changed,
1416 * put the new value in the bests table and log that.
1417 */
1418 if (be16_to_cpu(bf[0].length) != oldbest) {
1419 bestsp[db] = bf[0].length;
1420 xfs_dir3_leaf_log_bests(args, lbp, db, db);
1421 }
1422 xfs_dir3_data_check(dp, dbp);
1423 /*
1424 * If the data block is now empty then get rid of the data block.
1425 */
1426 if (be16_to_cpu(bf[0].length) ==
1427 args->geo->blksize - dp->d_ops->data_entry_offset) {
1428 ASSERT(db != args->geo->datablk);
1429 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1430 /*
1431 * Nope, can't get rid of it because it caused
1432 * allocation of a bmap btree block to do so.
1433 * Just go on, returning success, leaving the
1434 * empty block in place.
1435 */
1436 if (error == -ENOSPC && args->total == 0)
1437 error = 0;
1438 xfs_dir3_leaf_check(dp, lbp);
1439 return error;
1440 }
1441 dbp = NULL;
1442 /*
1443 * If this is the last data block then compact the
1444 * bests table by getting rid of entries.
1445 */
1446 if (db == be32_to_cpu(ltp->bestcount) - 1) {
1447 /*
1448 * Look for the last active entry (i).
1449 */
1450 for (i = db - 1; i > 0; i--) {
1451 if (bestsp[i] != cpu_to_be16(NULLDATAOFF))
1452 break;
1453 }
1454 /*
1455 * Copy the table down so inactive entries at the
1456 * end are removed.
1457 */
1458 memmove(&bestsp[db - i], bestsp,
1459 (be32_to_cpu(ltp->bestcount) - (db - i)) * sizeof(*bestsp));
1460 be32_add_cpu(&ltp->bestcount, -(db - i));
1461 xfs_dir3_leaf_log_tail(args, lbp);
1462 xfs_dir3_leaf_log_bests(args, lbp, 0,
1463 be32_to_cpu(ltp->bestcount) - 1);
1464 } else
1465 bestsp[db] = cpu_to_be16(NULLDATAOFF);
1466 }
1467 /*
1468 * If the data block was not the first one, drop it.
1469 */
1470 else if (db != args->geo->datablk)
1471 dbp = NULL;
1472
1473 xfs_dir3_leaf_check(dp, lbp);
1474 /*
1475 * See if we can convert to block form.
1476 */
1477 return xfs_dir2_leaf_to_block(args, lbp, dbp);
1478 }
1479
1480 /*
1481 * Replace the inode number in a leaf format directory entry.
1482 */
1483 int /* error */
1484 xfs_dir2_leaf_replace(
1485 xfs_da_args_t *args) /* operation arguments */
1486 {
1487 struct xfs_buf *dbp; /* data block buffer */
1488 xfs_dir2_data_entry_t *dep; /* data block entry */
1489 xfs_inode_t *dp; /* incore directory inode */
1490 int error; /* error return code */
1491 int index; /* index of leaf entry */
1492 struct xfs_buf *lbp; /* leaf buffer */
1493 xfs_dir2_leaf_t *leaf; /* leaf structure */
1494 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1495 xfs_trans_t *tp; /* transaction pointer */
1496 struct xfs_dir2_leaf_entry *ents;
1497
1498 trace_xfs_dir2_leaf_replace(args);
1499
1500 /*
1501 * Look up the entry.
1502 */
1503 if ((error = xfs_dir2_leaf_lookup_int(args, &lbp, &index, &dbp))) {
1504 return error;
1505 }
1506 dp = args->dp;
1507 leaf = lbp->b_addr;
1508 ents = dp->d_ops->leaf_ents_p(leaf);
1509 /*
1510 * Point to the leaf entry, get data address from it.
1511 */
1512 lep = &ents[index];
1513 /*
1514 * Point to the data entry.
1515 */
1516 dep = (xfs_dir2_data_entry_t *)
1517 ((char *)dbp->b_addr +
1518 xfs_dir2_dataptr_to_off(args->geo, be32_to_cpu(lep->address)));
1519 ASSERT(args->inumber != be64_to_cpu(dep->inumber));
1520 /*
1521 * Put the new inode number in, log it.
1522 */
1523 dep->inumber = cpu_to_be64(args->inumber);
1524 dp->d_ops->data_put_ftype(dep, args->filetype);
1525 tp = args->trans;
1526 xfs_dir2_data_log_entry(args, dbp, dep);
1527 xfs_dir3_leaf_check(dp, lbp);
1528 xfs_trans_brelse(tp, lbp);
1529 return 0;
1530 }
1531
1532 /*
1533 * Return index in the leaf block (lbp) which is either the first
1534 * one with this hash value, or if there are none, the insert point
1535 * for that hash value.
1536 */
1537 int /* index value */
1538 xfs_dir2_leaf_search_hash(
1539 xfs_da_args_t *args, /* operation arguments */
1540 struct xfs_buf *lbp) /* leaf buffer */
1541 {
1542 xfs_dahash_t hash=0; /* hash from this entry */
1543 xfs_dahash_t hashwant; /* hash value looking for */
1544 int high; /* high leaf index */
1545 int low; /* low leaf index */
1546 xfs_dir2_leaf_t *leaf; /* leaf structure */
1547 xfs_dir2_leaf_entry_t *lep; /* leaf entry */
1548 int mid=0; /* current leaf index */
1549 struct xfs_dir2_leaf_entry *ents;
1550 struct xfs_dir3_icleaf_hdr leafhdr;
1551
1552 leaf = lbp->b_addr;
1553 ents = args->dp->d_ops->leaf_ents_p(leaf);
1554 args->dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1555
1556 /*
1557 * Note, the table cannot be empty, so we have to go through the loop.
1558 * Binary search the leaf entries looking for our hash value.
1559 */
1560 for (lep = ents, low = 0, high = leafhdr.count - 1,
1561 hashwant = args->hashval;
1562 low <= high; ) {
1563 mid = (low + high) >> 1;
1564 if ((hash = be32_to_cpu(lep[mid].hashval)) == hashwant)
1565 break;
1566 if (hash < hashwant)
1567 low = mid + 1;
1568 else
1569 high = mid - 1;
1570 }
1571 /*
1572 * Found one, back up through all the equal hash values.
1573 */
1574 if (hash == hashwant) {
1575 while (mid > 0 && be32_to_cpu(lep[mid - 1].hashval) == hashwant) {
1576 mid--;
1577 }
1578 }
1579 /*
1580 * Need to point to an entry higher than ours.
1581 */
1582 else if (hash < hashwant)
1583 mid++;
1584 return mid;
1585 }
1586
1587 /*
1588 * Trim off a trailing data block. We know it's empty since the leaf
1589 * freespace table says so.
1590 */
1591 int /* error */
1592 xfs_dir2_leaf_trim_data(
1593 xfs_da_args_t *args, /* operation arguments */
1594 struct xfs_buf *lbp, /* leaf buffer */
1595 xfs_dir2_db_t db) /* data block number */
1596 {
1597 __be16 *bestsp; /* leaf bests table */
1598 struct xfs_buf *dbp; /* data block buffer */
1599 xfs_inode_t *dp; /* incore directory inode */
1600 int error; /* error return value */
1601 xfs_dir2_leaf_t *leaf; /* leaf structure */
1602 xfs_dir2_leaf_tail_t *ltp; /* leaf tail structure */
1603 xfs_trans_t *tp; /* transaction pointer */
1604
1605 dp = args->dp;
1606 tp = args->trans;
1607 /*
1608 * Read the offending data block. We need its buffer.
1609 */
1610 error = xfs_dir3_data_read(tp, dp, xfs_dir2_db_to_da(args->geo, db),
1611 -1, &dbp);
1612 if (error)
1613 return error;
1614
1615 leaf = lbp->b_addr;
1616 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1617
1618 #ifdef DEBUG
1619 {
1620 struct xfs_dir2_data_hdr *hdr = dbp->b_addr;
1621 struct xfs_dir2_data_free *bf = dp->d_ops->data_bestfree_p(hdr);
1622
1623 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
1624 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC));
1625 ASSERT(be16_to_cpu(bf[0].length) ==
1626 args->geo->blksize - dp->d_ops->data_entry_offset);
1627 ASSERT(db == be32_to_cpu(ltp->bestcount) - 1);
1628 }
1629 #endif
1630
1631 /*
1632 * Get rid of the data block.
1633 */
1634 if ((error = xfs_dir2_shrink_inode(args, db, dbp))) {
1635 ASSERT(error != -ENOSPC);
1636 xfs_trans_brelse(tp, dbp);
1637 return error;
1638 }
1639 /*
1640 * Eliminate the last bests entry from the table.
1641 */
1642 bestsp = xfs_dir2_leaf_bests_p(ltp);
1643 be32_add_cpu(&ltp->bestcount, -1);
1644 memmove(&bestsp[1], &bestsp[0], be32_to_cpu(ltp->bestcount) * sizeof(*bestsp));
1645 xfs_dir3_leaf_log_tail(args, lbp);
1646 xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1647 return 0;
1648 }
1649
1650 static inline size_t
1651 xfs_dir3_leaf_size(
1652 struct xfs_dir3_icleaf_hdr *hdr,
1653 int counts)
1654 {
1655 int entries;
1656 int hdrsize;
1657
1658 entries = hdr->count - hdr->stale;
1659 if (hdr->magic == XFS_DIR2_LEAF1_MAGIC ||
1660 hdr->magic == XFS_DIR2_LEAFN_MAGIC)
1661 hdrsize = sizeof(struct xfs_dir2_leaf_hdr);
1662 else
1663 hdrsize = sizeof(struct xfs_dir3_leaf_hdr);
1664
1665 return hdrsize + entries * sizeof(xfs_dir2_leaf_entry_t)
1666 + counts * sizeof(xfs_dir2_data_off_t)
1667 + sizeof(xfs_dir2_leaf_tail_t);
1668 }
1669
1670 /*
1671 * Convert node form directory to leaf form directory.
1672 * The root of the node form dir needs to already be a LEAFN block.
1673 * Just return if we can't do anything.
1674 */
1675 int /* error */
1676 xfs_dir2_node_to_leaf(
1677 xfs_da_state_t *state) /* directory operation state */
1678 {
1679 xfs_da_args_t *args; /* operation arguments */
1680 xfs_inode_t *dp; /* incore directory inode */
1681 int error; /* error return code */
1682 struct xfs_buf *fbp; /* buffer for freespace block */
1683 xfs_fileoff_t fo; /* freespace file offset */
1684 xfs_dir2_free_t *free; /* freespace structure */
1685 struct xfs_buf *lbp; /* buffer for leaf block */
1686 xfs_dir2_leaf_tail_t *ltp; /* tail of leaf structure */
1687 xfs_dir2_leaf_t *leaf; /* leaf structure */
1688 xfs_mount_t *mp; /* filesystem mount point */
1689 int rval; /* successful free trim? */
1690 xfs_trans_t *tp; /* transaction pointer */
1691 struct xfs_dir3_icleaf_hdr leafhdr;
1692 struct xfs_dir3_icfree_hdr freehdr;
1693
1694 /*
1695 * There's more than a leaf level in the btree, so there must
1696 * be multiple leafn blocks. Give up.
1697 */
1698 if (state->path.active > 1)
1699 return 0;
1700 args = state->args;
1701
1702 trace_xfs_dir2_node_to_leaf(args);
1703
1704 mp = state->mp;
1705 dp = args->dp;
1706 tp = args->trans;
1707 /*
1708 * Get the last offset in the file.
1709 */
1710 if ((error = xfs_bmap_last_offset(dp, &fo, XFS_DATA_FORK))) {
1711 return error;
1712 }
1713 fo -= args->geo->fsbcount;
1714 /*
1715 * If there are freespace blocks other than the first one,
1716 * take this opportunity to remove trailing empty freespace blocks
1717 * that may have been left behind during no-space-reservation
1718 * operations.
1719 */
1720 while (fo > args->geo->freeblk) {
1721 if ((error = xfs_dir2_node_trim_free(args, fo, &rval))) {
1722 return error;
1723 }
1724 if (rval)
1725 fo -= args->geo->fsbcount;
1726 else
1727 return 0;
1728 }
1729 /*
1730 * Now find the block just before the freespace block.
1731 */
1732 if ((error = xfs_bmap_last_before(tp, dp, &fo, XFS_DATA_FORK))) {
1733 return error;
1734 }
1735 /*
1736 * If it's not the single leaf block, give up.
1737 */
1738 if (XFS_FSB_TO_B(mp, fo) > XFS_DIR2_LEAF_OFFSET + args->geo->blksize)
1739 return 0;
1740 lbp = state->path.blk[0].bp;
1741 leaf = lbp->b_addr;
1742 dp->d_ops->leaf_hdr_from_disk(&leafhdr, leaf);
1743
1744 ASSERT(leafhdr.magic == XFS_DIR2_LEAFN_MAGIC ||
1745 leafhdr.magic == XFS_DIR3_LEAFN_MAGIC);
1746
1747 /*
1748 * Read the freespace block.
1749 */
1750 error = xfs_dir2_free_read(tp, dp, args->geo->freeblk, &fbp);
1751 if (error)
1752 return error;
1753 free = fbp->b_addr;
1754 dp->d_ops->free_hdr_from_disk(&freehdr, free);
1755
1756 ASSERT(!freehdr.firstdb);
1757
1758 /*
1759 * Now see if the leafn and free data will fit in a leaf1.
1760 * If not, release the buffer and give up.
1761 */
1762 if (xfs_dir3_leaf_size(&leafhdr, freehdr.nvalid) > args->geo->blksize) {
1763 xfs_trans_brelse(tp, fbp);
1764 return 0;
1765 }
1766
1767 /*
1768 * If the leaf has any stale entries in it, compress them out.
1769 */
1770 if (leafhdr.stale)
1771 xfs_dir3_leaf_compact(args, &leafhdr, lbp);
1772
1773 lbp->b_ops = &xfs_dir3_leaf1_buf_ops;
1774 xfs_trans_buf_set_type(tp, lbp, XFS_BLFT_DIR_LEAF1_BUF);
1775 leafhdr.magic = (leafhdr.magic == XFS_DIR2_LEAFN_MAGIC)
1776 ? XFS_DIR2_LEAF1_MAGIC
1777 : XFS_DIR3_LEAF1_MAGIC;
1778
1779 /*
1780 * Set up the leaf tail from the freespace block.
1781 */
1782 ltp = xfs_dir2_leaf_tail_p(args->geo, leaf);
1783 ltp->bestcount = cpu_to_be32(freehdr.nvalid);
1784
1785 /*
1786 * Set up the leaf bests table.
1787 */
1788 memcpy(xfs_dir2_leaf_bests_p(ltp), dp->d_ops->free_bests_p(free),
1789 freehdr.nvalid * sizeof(xfs_dir2_data_off_t));
1790
1791 dp->d_ops->leaf_hdr_to_disk(leaf, &leafhdr);
1792 xfs_dir3_leaf_log_header(args, lbp);
1793 xfs_dir3_leaf_log_bests(args, lbp, 0, be32_to_cpu(ltp->bestcount) - 1);
1794 xfs_dir3_leaf_log_tail(args, lbp);
1795 xfs_dir3_leaf_check(dp, lbp);
1796
1797 /*
1798 * Get rid of the freespace block.
1799 */
1800 error = xfs_dir2_shrink_inode(args,
1801 xfs_dir2_byte_to_db(args->geo, XFS_DIR2_FREE_OFFSET),
1802 fbp);
1803 if (error) {
1804 /*
1805 * This can't fail here because it can only happen when
1806 * punching out the middle of an extent, and this is an
1807 * isolated block.
1808 */
1809 ASSERT(error != -ENOSPC);
1810 return error;
1811 }
1812 fbp = NULL;
1813 /*
1814 * Now see if we can convert the single-leaf directory
1815 * down to a block form directory.
1816 * This routine always kills the dabuf for the leaf, so
1817 * eliminate it from the path.
1818 */
1819 error = xfs_dir2_leaf_to_block(args, lbp, NULL);
1820 state->path.blk[0].bp = NULL;
1821 return error;
1822 }