Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
5 * All Rights Reserved.
6 */
7#include "xfs.h"
8#include "xfs_fs.h"
9#include "xfs_shared.h"
10#include "xfs_format.h"
11#include "xfs_log_format.h"
12#include "xfs_trans_resv.h"
13#include "xfs_mount.h"
14#include "xfs_inode.h"
15#include "xfs_dir2.h"
16#include "xfs_dir2_priv.h"
17#include "xfs_error.h"
18#include "xfs_trans.h"
19#include "xfs_buf_item.h"
20#include "xfs_log.h"
21#include "xfs_health.h"
22
23static xfs_failaddr_t xfs_dir2_data_freefind_verify(
24 struct xfs_dir2_data_hdr *hdr, struct xfs_dir2_data_free *bf,
25 struct xfs_dir2_data_unused *dup,
26 struct xfs_dir2_data_free **bf_ent);
27
28struct xfs_dir2_data_free *
29xfs_dir2_data_bestfree_p(
30 struct xfs_mount *mp,
31 struct xfs_dir2_data_hdr *hdr)
32{
33 if (xfs_has_crc(mp))
34 return ((struct xfs_dir3_data_hdr *)hdr)->best_free;
35 return hdr->bestfree;
36}
37
38/*
39 * Pointer to an entry's tag word.
40 */
41__be16 *
42xfs_dir2_data_entry_tag_p(
43 struct xfs_mount *mp,
44 struct xfs_dir2_data_entry *dep)
45{
46 return (__be16 *)((char *)dep +
47 xfs_dir2_data_entsize(mp, dep->namelen) - sizeof(__be16));
48}
49
50uint8_t
51xfs_dir2_data_get_ftype(
52 struct xfs_mount *mp,
53 struct xfs_dir2_data_entry *dep)
54{
55 if (xfs_has_ftype(mp)) {
56 uint8_t ftype = dep->name[dep->namelen];
57
58 if (likely(ftype < XFS_DIR3_FT_MAX))
59 return ftype;
60 }
61
62 return XFS_DIR3_FT_UNKNOWN;
63}
64
65void
66xfs_dir2_data_put_ftype(
67 struct xfs_mount *mp,
68 struct xfs_dir2_data_entry *dep,
69 uint8_t ftype)
70{
71 ASSERT(ftype < XFS_DIR3_FT_MAX);
72 ASSERT(dep->namelen != 0);
73
74 if (xfs_has_ftype(mp))
75 dep->name[dep->namelen] = ftype;
76}
77
78/*
79 * The number of leaf entries is limited by the size of the block and the amount
80 * of space used by the data entries. We don't know how much space is used by
81 * the data entries yet, so just ensure that the count falls somewhere inside
82 * the block right now.
83 */
84static inline unsigned int
85xfs_dir2_data_max_leaf_entries(
86 struct xfs_da_geometry *geo)
87{
88 return (geo->blksize - sizeof(struct xfs_dir2_block_tail) -
89 geo->data_entry_offset) /
90 sizeof(struct xfs_dir2_leaf_entry);
91}
92
93/*
94 * Check the consistency of the data block.
95 * The input can also be a block-format directory.
96 * Return NULL if the buffer is good, otherwise the address of the error.
97 */
98xfs_failaddr_t
99__xfs_dir3_data_check(
100 struct xfs_inode *dp, /* incore inode pointer */
101 struct xfs_buf *bp) /* data block's buffer */
102{
103 xfs_dir2_dataptr_t addr; /* addr for leaf lookup */
104 xfs_dir2_data_free_t *bf; /* bestfree table */
105 xfs_dir2_block_tail_t *btp=NULL; /* block tail */
106 int count; /* count of entries found */
107 xfs_dir2_data_hdr_t *hdr; /* data block header */
108 xfs_dir2_data_free_t *dfp; /* bestfree entry */
109 int freeseen; /* mask of bestfrees seen */
110 xfs_dahash_t hash; /* hash of current name */
111 int i; /* leaf index */
112 int lastfree; /* last entry was unused */
113 xfs_dir2_leaf_entry_t *lep=NULL; /* block leaf entries */
114 struct xfs_mount *mp = bp->b_mount;
115 int stale; /* count of stale leaves */
116 struct xfs_name name;
117 unsigned int offset;
118 unsigned int end;
119 struct xfs_da_geometry *geo = mp->m_dir_geo;
120
121 /*
122 * If this isn't a directory, something is seriously wrong. Bail out.
123 */
124 if (dp && !S_ISDIR(VFS_I(dp)->i_mode))
125 return __this_address;
126
127 hdr = bp->b_addr;
128 offset = geo->data_entry_offset;
129
130 switch (hdr->magic) {
131 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
132 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
133 btp = xfs_dir2_block_tail_p(geo, hdr);
134 lep = xfs_dir2_block_leaf_p(btp);
135
136 if (be32_to_cpu(btp->count) >=
137 xfs_dir2_data_max_leaf_entries(geo))
138 return __this_address;
139 break;
140 case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
141 case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
142 break;
143 default:
144 return __this_address;
145 }
146 end = xfs_dir3_data_end_offset(geo, hdr);
147 if (!end)
148 return __this_address;
149
150 /*
151 * Account for zero bestfree entries.
152 */
153 bf = xfs_dir2_data_bestfree_p(mp, hdr);
154 count = lastfree = freeseen = 0;
155 if (!bf[0].length) {
156 if (bf[0].offset)
157 return __this_address;
158 freeseen |= 1 << 0;
159 }
160 if (!bf[1].length) {
161 if (bf[1].offset)
162 return __this_address;
163 freeseen |= 1 << 1;
164 }
165 if (!bf[2].length) {
166 if (bf[2].offset)
167 return __this_address;
168 freeseen |= 1 << 2;
169 }
170
171 if (be16_to_cpu(bf[0].length) < be16_to_cpu(bf[1].length))
172 return __this_address;
173 if (be16_to_cpu(bf[1].length) < be16_to_cpu(bf[2].length))
174 return __this_address;
175 /*
176 * Loop over the data/unused entries.
177 */
178 while (offset < end) {
179 struct xfs_dir2_data_unused *dup = bp->b_addr + offset;
180 struct xfs_dir2_data_entry *dep = bp->b_addr + offset;
181 unsigned int reclen;
182
183 /*
184 * Are the remaining bytes large enough to hold an
185 * unused entry?
186 */
187 if (offset > end - xfs_dir2_data_unusedsize(1))
188 return __this_address;
189
190 /*
191 * If it's unused, look for the space in the bestfree table.
192 * If we find it, account for that, else make sure it
193 * doesn't need to be there.
194 */
195 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
196 xfs_failaddr_t fa;
197
198 reclen = xfs_dir2_data_unusedsize(
199 be16_to_cpu(dup->length));
200 if (lastfree != 0)
201 return __this_address;
202 if (be16_to_cpu(dup->length) != reclen)
203 return __this_address;
204 if (offset + reclen > end)
205 return __this_address;
206 if (be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) !=
207 offset)
208 return __this_address;
209 fa = xfs_dir2_data_freefind_verify(hdr, bf, dup, &dfp);
210 if (fa)
211 return fa;
212 if (dfp) {
213 i = (int)(dfp - bf);
214 if ((freeseen & (1 << i)) != 0)
215 return __this_address;
216 freeseen |= 1 << i;
217 } else {
218 if (be16_to_cpu(dup->length) >
219 be16_to_cpu(bf[2].length))
220 return __this_address;
221 }
222 offset += reclen;
223 lastfree = 1;
224 continue;
225 }
226
227 /*
228 * This is not an unused entry. Are the remaining bytes
229 * large enough for a dirent with a single-byte name?
230 */
231 if (offset > end - xfs_dir2_data_entsize(mp, 1))
232 return __this_address;
233
234 /*
235 * It's a real entry. Validate the fields.
236 * If this is a block directory then make sure it's
237 * in the leaf section of the block.
238 * The linear search is crude but this is DEBUG code.
239 */
240 if (dep->namelen == 0)
241 return __this_address;
242 reclen = xfs_dir2_data_entsize(mp, dep->namelen);
243 if (offset + reclen > end)
244 return __this_address;
245 if (!xfs_verify_dir_ino(mp, be64_to_cpu(dep->inumber)))
246 return __this_address;
247 if (be16_to_cpu(*xfs_dir2_data_entry_tag_p(mp, dep)) != offset)
248 return __this_address;
249 if (xfs_dir2_data_get_ftype(mp, dep) >= XFS_DIR3_FT_MAX)
250 return __this_address;
251 count++;
252 lastfree = 0;
253 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
254 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
255 addr = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
256 (xfs_dir2_data_aoff_t)
257 ((char *)dep - (char *)hdr));
258 name.name = dep->name;
259 name.len = dep->namelen;
260 hash = xfs_dir2_hashname(mp, &name);
261 for (i = 0; i < be32_to_cpu(btp->count); i++) {
262 if (be32_to_cpu(lep[i].address) == addr &&
263 be32_to_cpu(lep[i].hashval) == hash)
264 break;
265 }
266 if (i >= be32_to_cpu(btp->count))
267 return __this_address;
268 }
269 offset += reclen;
270 }
271 /*
272 * Need to have seen all the entries and all the bestfree slots.
273 */
274 if (freeseen != 7)
275 return __this_address;
276 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
277 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
278 for (i = stale = 0; i < be32_to_cpu(btp->count); i++) {
279 if (lep[i].address ==
280 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
281 stale++;
282 if (i > 0 && be32_to_cpu(lep[i].hashval) <
283 be32_to_cpu(lep[i - 1].hashval))
284 return __this_address;
285 }
286 if (count != be32_to_cpu(btp->count) - be32_to_cpu(btp->stale))
287 return __this_address;
288 if (stale != be32_to_cpu(btp->stale))
289 return __this_address;
290 }
291 return NULL;
292}
293
294#ifdef DEBUG
295void
296xfs_dir3_data_check(
297 struct xfs_inode *dp,
298 struct xfs_buf *bp)
299{
300 xfs_failaddr_t fa;
301
302 fa = __xfs_dir3_data_check(dp, bp);
303 if (!fa)
304 return;
305 xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
306 bp->b_addr, BBTOB(bp->b_length), __FILE__, __LINE__,
307 fa);
308 ASSERT(0);
309}
310#endif
311
312static xfs_failaddr_t
313xfs_dir3_data_verify(
314 struct xfs_buf *bp)
315{
316 struct xfs_mount *mp = bp->b_mount;
317 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
318
319 if (!xfs_verify_magic(bp, hdr3->magic))
320 return __this_address;
321
322 if (xfs_has_crc(mp)) {
323 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
324 return __this_address;
325 if (be64_to_cpu(hdr3->blkno) != xfs_buf_daddr(bp))
326 return __this_address;
327 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
328 return __this_address;
329 }
330 return __xfs_dir3_data_check(NULL, bp);
331}
332
333/*
334 * Readahead of the first block of the directory when it is opened is completely
335 * oblivious to the format of the directory. Hence we can either get a block
336 * format buffer or a data format buffer on readahead.
337 */
338static void
339xfs_dir3_data_reada_verify(
340 struct xfs_buf *bp)
341{
342 struct xfs_dir2_data_hdr *hdr = bp->b_addr;
343
344 switch (hdr->magic) {
345 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
346 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
347 bp->b_ops = &xfs_dir3_block_buf_ops;
348 bp->b_ops->verify_read(bp);
349 return;
350 case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
351 case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
352 bp->b_ops = &xfs_dir3_data_buf_ops;
353 bp->b_ops->verify_read(bp);
354 return;
355 default:
356 xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
357 break;
358 }
359}
360
361static void
362xfs_dir3_data_read_verify(
363 struct xfs_buf *bp)
364{
365 struct xfs_mount *mp = bp->b_mount;
366 xfs_failaddr_t fa;
367
368 if (xfs_has_crc(mp) &&
369 !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
370 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
371 else {
372 fa = xfs_dir3_data_verify(bp);
373 if (fa)
374 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
375 }
376}
377
378static void
379xfs_dir3_data_write_verify(
380 struct xfs_buf *bp)
381{
382 struct xfs_mount *mp = bp->b_mount;
383 struct xfs_buf_log_item *bip = bp->b_log_item;
384 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
385 xfs_failaddr_t fa;
386
387 fa = xfs_dir3_data_verify(bp);
388 if (fa) {
389 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
390 return;
391 }
392
393 if (!xfs_has_crc(mp))
394 return;
395
396 if (bip)
397 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
398
399 xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
400}
401
402const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
403 .name = "xfs_dir3_data",
404 .magic = { cpu_to_be32(XFS_DIR2_DATA_MAGIC),
405 cpu_to_be32(XFS_DIR3_DATA_MAGIC) },
406 .verify_read = xfs_dir3_data_read_verify,
407 .verify_write = xfs_dir3_data_write_verify,
408 .verify_struct = xfs_dir3_data_verify,
409};
410
411static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
412 .name = "xfs_dir3_data_reada",
413 .magic = { cpu_to_be32(XFS_DIR2_DATA_MAGIC),
414 cpu_to_be32(XFS_DIR3_DATA_MAGIC) },
415 .verify_read = xfs_dir3_data_reada_verify,
416 .verify_write = xfs_dir3_data_write_verify,
417};
418
419xfs_failaddr_t
420xfs_dir3_data_header_check(
421 struct xfs_buf *bp,
422 xfs_ino_t owner)
423{
424 struct xfs_mount *mp = bp->b_mount;
425
426 if (xfs_has_crc(mp)) {
427 struct xfs_dir3_data_hdr *hdr3 = bp->b_addr;
428
429 if (hdr3->hdr.magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC))
430 return __this_address;
431
432 if (be64_to_cpu(hdr3->hdr.owner) != owner)
433 return __this_address;
434 }
435
436 return NULL;
437}
438
439int
440xfs_dir3_data_read(
441 struct xfs_trans *tp,
442 struct xfs_inode *dp,
443 xfs_ino_t owner,
444 xfs_dablk_t bno,
445 unsigned int flags,
446 struct xfs_buf **bpp)
447{
448 xfs_failaddr_t fa;
449 int err;
450
451 err = xfs_da_read_buf(tp, dp, bno, flags, bpp, XFS_DATA_FORK,
452 &xfs_dir3_data_buf_ops);
453 if (err || !*bpp)
454 return err;
455
456 /* Check things that we can't do in the verifier. */
457 fa = xfs_dir3_data_header_check(*bpp, owner);
458 if (fa) {
459 __xfs_buf_mark_corrupt(*bpp, fa);
460 xfs_trans_brelse(tp, *bpp);
461 *bpp = NULL;
462 xfs_dirattr_mark_sick(dp, XFS_DATA_FORK);
463 return -EFSCORRUPTED;
464 }
465
466 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
467 return err;
468}
469
470int
471xfs_dir3_data_readahead(
472 struct xfs_inode *dp,
473 xfs_dablk_t bno,
474 unsigned int flags)
475{
476 return xfs_da_reada_buf(dp, bno, flags, XFS_DATA_FORK,
477 &xfs_dir3_data_reada_buf_ops);
478}
479
480/*
481 * Find the bestfree entry that exactly coincides with unused directory space
482 * or a verifier error because the bestfree data are bad.
483 */
484static xfs_failaddr_t
485xfs_dir2_data_freefind_verify(
486 struct xfs_dir2_data_hdr *hdr,
487 struct xfs_dir2_data_free *bf,
488 struct xfs_dir2_data_unused *dup,
489 struct xfs_dir2_data_free **bf_ent)
490{
491 struct xfs_dir2_data_free *dfp;
492 xfs_dir2_data_aoff_t off;
493 bool matched = false;
494 bool seenzero = false;
495
496 *bf_ent = NULL;
497 off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
498
499 /*
500 * Validate some consistency in the bestfree table.
501 * Check order, non-overlapping entries, and if we find the
502 * one we're looking for it has to be exact.
503 */
504 for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
505 if (!dfp->offset) {
506 if (dfp->length)
507 return __this_address;
508 seenzero = true;
509 continue;
510 }
511 if (seenzero)
512 return __this_address;
513 if (be16_to_cpu(dfp->offset) == off) {
514 matched = true;
515 if (dfp->length != dup->length)
516 return __this_address;
517 } else if (be16_to_cpu(dfp->offset) > off) {
518 if (off + be16_to_cpu(dup->length) >
519 be16_to_cpu(dfp->offset))
520 return __this_address;
521 } else {
522 if (be16_to_cpu(dfp->offset) +
523 be16_to_cpu(dfp->length) > off)
524 return __this_address;
525 }
526 if (!matched &&
527 be16_to_cpu(dfp->length) < be16_to_cpu(dup->length))
528 return __this_address;
529 if (dfp > &bf[0] &&
530 be16_to_cpu(dfp[-1].length) < be16_to_cpu(dfp[0].length))
531 return __this_address;
532 }
533
534 /* Looks ok so far; now try to match up with a bestfree entry. */
535 *bf_ent = xfs_dir2_data_freefind(hdr, bf, dup);
536 return NULL;
537}
538
539/*
540 * Given a data block and an unused entry from that block,
541 * return the bestfree entry if any that corresponds to it.
542 */
543xfs_dir2_data_free_t *
544xfs_dir2_data_freefind(
545 struct xfs_dir2_data_hdr *hdr, /* data block header */
546 struct xfs_dir2_data_free *bf, /* bestfree table pointer */
547 struct xfs_dir2_data_unused *dup) /* unused space */
548{
549 xfs_dir2_data_free_t *dfp; /* bestfree entry */
550 xfs_dir2_data_aoff_t off; /* offset value needed */
551
552 off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
553
554 /*
555 * If this is smaller than the smallest bestfree entry,
556 * it can't be there since they're sorted.
557 */
558 if (be16_to_cpu(dup->length) <
559 be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
560 return NULL;
561 /*
562 * Look at the three bestfree entries for our guy.
563 */
564 for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
565 if (!dfp->offset)
566 return NULL;
567 if (be16_to_cpu(dfp->offset) == off)
568 return dfp;
569 }
570 /*
571 * Didn't find it. This only happens if there are duplicate lengths.
572 */
573 return NULL;
574}
575
576/*
577 * Insert an unused-space entry into the bestfree table.
578 */
579xfs_dir2_data_free_t * /* entry inserted */
580xfs_dir2_data_freeinsert(
581 struct xfs_dir2_data_hdr *hdr, /* data block pointer */
582 struct xfs_dir2_data_free *dfp, /* bestfree table pointer */
583 struct xfs_dir2_data_unused *dup, /* unused space */
584 int *loghead) /* log the data header (out) */
585{
586 xfs_dir2_data_free_t new; /* new bestfree entry */
587
588 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
589 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
590 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
591 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
592
593 new.length = dup->length;
594 new.offset = cpu_to_be16((char *)dup - (char *)hdr);
595
596 /*
597 * Insert at position 0, 1, or 2; or not at all.
598 */
599 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
600 dfp[2] = dfp[1];
601 dfp[1] = dfp[0];
602 dfp[0] = new;
603 *loghead = 1;
604 return &dfp[0];
605 }
606 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
607 dfp[2] = dfp[1];
608 dfp[1] = new;
609 *loghead = 1;
610 return &dfp[1];
611 }
612 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
613 dfp[2] = new;
614 *loghead = 1;
615 return &dfp[2];
616 }
617 return NULL;
618}
619
620/*
621 * Remove a bestfree entry from the table.
622 */
623STATIC void
624xfs_dir2_data_freeremove(
625 struct xfs_dir2_data_hdr *hdr, /* data block header */
626 struct xfs_dir2_data_free *bf, /* bestfree table pointer */
627 struct xfs_dir2_data_free *dfp, /* bestfree entry pointer */
628 int *loghead) /* out: log data header */
629{
630
631 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
632 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
633 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
634 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
635
636 /*
637 * It's the first entry, slide the next 2 up.
638 */
639 if (dfp == &bf[0]) {
640 bf[0] = bf[1];
641 bf[1] = bf[2];
642 }
643 /*
644 * It's the second entry, slide the 3rd entry up.
645 */
646 else if (dfp == &bf[1])
647 bf[1] = bf[2];
648 /*
649 * Must be the last entry.
650 */
651 else
652 ASSERT(dfp == &bf[2]);
653 /*
654 * Clear the 3rd entry, must be zero now.
655 */
656 bf[2].length = 0;
657 bf[2].offset = 0;
658 *loghead = 1;
659}
660
661/*
662 * Given a data block, reconstruct its bestfree map.
663 */
664void
665xfs_dir2_data_freescan(
666 struct xfs_mount *mp,
667 struct xfs_dir2_data_hdr *hdr,
668 int *loghead)
669{
670 struct xfs_da_geometry *geo = mp->m_dir_geo;
671 struct xfs_dir2_data_free *bf = xfs_dir2_data_bestfree_p(mp, hdr);
672 void *addr = hdr;
673 unsigned int offset = geo->data_entry_offset;
674 unsigned int end;
675
676 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
677 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
678 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
679 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
680
681 /*
682 * Start by clearing the table.
683 */
684 memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
685 *loghead = 1;
686
687 end = xfs_dir3_data_end_offset(geo, addr);
688 while (offset < end) {
689 struct xfs_dir2_data_unused *dup = addr + offset;
690 struct xfs_dir2_data_entry *dep = addr + offset;
691
692 /*
693 * If it's a free entry, insert it.
694 */
695 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
696 ASSERT(offset ==
697 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
698 xfs_dir2_data_freeinsert(hdr, bf, dup, loghead);
699 offset += be16_to_cpu(dup->length);
700 continue;
701 }
702
703 /*
704 * For active entries, check their tags and skip them.
705 */
706 ASSERT(offset ==
707 be16_to_cpu(*xfs_dir2_data_entry_tag_p(mp, dep)));
708 offset += xfs_dir2_data_entsize(mp, dep->namelen);
709 }
710}
711
712/*
713 * Initialize a data block at the given block number in the directory.
714 * Give back the buffer for the created block.
715 */
716int /* error */
717xfs_dir3_data_init(
718 struct xfs_da_args *args, /* directory operation args */
719 xfs_dir2_db_t blkno, /* logical dir block number */
720 struct xfs_buf **bpp) /* output block buffer */
721{
722 struct xfs_trans *tp = args->trans;
723 struct xfs_inode *dp = args->dp;
724 struct xfs_mount *mp = dp->i_mount;
725 struct xfs_da_geometry *geo = args->geo;
726 struct xfs_buf *bp;
727 struct xfs_dir2_data_hdr *hdr;
728 struct xfs_dir2_data_unused *dup;
729 struct xfs_dir2_data_free *bf;
730 int error;
731 int i;
732
733 /*
734 * Get the buffer set up for the block.
735 */
736 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, blkno),
737 &bp, XFS_DATA_FORK);
738 if (error)
739 return error;
740 bp->b_ops = &xfs_dir3_data_buf_ops;
741 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
742
743 /*
744 * Initialize the header.
745 */
746 hdr = bp->b_addr;
747 if (xfs_has_crc(mp)) {
748 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
749
750 memset(hdr3, 0, sizeof(*hdr3));
751 hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
752 hdr3->blkno = cpu_to_be64(xfs_buf_daddr(bp));
753 hdr3->owner = cpu_to_be64(args->owner);
754 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
755
756 } else
757 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
758
759 bf = xfs_dir2_data_bestfree_p(mp, hdr);
760 bf[0].offset = cpu_to_be16(geo->data_entry_offset);
761 bf[0].length = cpu_to_be16(geo->blksize - geo->data_entry_offset);
762 for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
763 bf[i].length = 0;
764 bf[i].offset = 0;
765 }
766
767 /*
768 * Set up an unused entry for the block's body.
769 */
770 dup = bp->b_addr + geo->data_entry_offset;
771 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
772 dup->length = bf[0].length;
773 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
774
775 /*
776 * Log it and return it.
777 */
778 xfs_dir2_data_log_header(args, bp);
779 xfs_dir2_data_log_unused(args, bp, dup);
780 *bpp = bp;
781 return 0;
782}
783
784/*
785 * Log an active data entry from the block.
786 */
787void
788xfs_dir2_data_log_entry(
789 struct xfs_da_args *args,
790 struct xfs_buf *bp,
791 xfs_dir2_data_entry_t *dep) /* data entry pointer */
792{
793 struct xfs_mount *mp = bp->b_mount;
794 struct xfs_dir2_data_hdr *hdr = bp->b_addr;
795
796 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
797 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
798 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
799 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
800
801 xfs_trans_log_buf(args->trans, bp, (uint)((char *)dep - (char *)hdr),
802 (uint)((char *)(xfs_dir2_data_entry_tag_p(mp, dep) + 1) -
803 (char *)hdr - 1));
804}
805
806/*
807 * Log a data block header.
808 */
809void
810xfs_dir2_data_log_header(
811 struct xfs_da_args *args,
812 struct xfs_buf *bp)
813{
814#ifdef DEBUG
815 struct xfs_dir2_data_hdr *hdr = bp->b_addr;
816
817 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
818 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
819 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
820 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
821#endif
822
823 xfs_trans_log_buf(args->trans, bp, 0, args->geo->data_entry_offset - 1);
824}
825
826/*
827 * Log a data unused entry.
828 */
829void
830xfs_dir2_data_log_unused(
831 struct xfs_da_args *args,
832 struct xfs_buf *bp,
833 xfs_dir2_data_unused_t *dup) /* data unused pointer */
834{
835 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
836
837 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
838 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
839 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
840 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
841
842 /*
843 * Log the first part of the unused entry.
844 */
845 xfs_trans_log_buf(args->trans, bp, (uint)((char *)dup - (char *)hdr),
846 (uint)((char *)&dup->length + sizeof(dup->length) -
847 1 - (char *)hdr));
848 /*
849 * Log the end (tag) of the unused entry.
850 */
851 xfs_trans_log_buf(args->trans, bp,
852 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
853 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
854 sizeof(xfs_dir2_data_off_t) - 1));
855}
856
857/*
858 * Make a byte range in the data block unused.
859 * Its current contents are unimportant.
860 */
861void
862xfs_dir2_data_make_free(
863 struct xfs_da_args *args,
864 struct xfs_buf *bp,
865 xfs_dir2_data_aoff_t offset, /* starting byte offset */
866 xfs_dir2_data_aoff_t len, /* length in bytes */
867 int *needlogp, /* out: log header */
868 int *needscanp) /* out: regen bestfree */
869{
870 xfs_dir2_data_hdr_t *hdr; /* data block pointer */
871 xfs_dir2_data_free_t *dfp; /* bestfree pointer */
872 int needscan; /* need to regen bestfree */
873 xfs_dir2_data_unused_t *newdup; /* new unused entry */
874 xfs_dir2_data_unused_t *postdup; /* unused entry after us */
875 xfs_dir2_data_unused_t *prevdup; /* unused entry before us */
876 unsigned int end;
877 struct xfs_dir2_data_free *bf;
878
879 hdr = bp->b_addr;
880
881 /*
882 * Figure out where the end of the data area is.
883 */
884 end = xfs_dir3_data_end_offset(args->geo, hdr);
885 ASSERT(end != 0);
886
887 /*
888 * If this isn't the start of the block, then back up to
889 * the previous entry and see if it's free.
890 */
891 if (offset > args->geo->data_entry_offset) {
892 __be16 *tagp; /* tag just before us */
893
894 tagp = (__be16 *)((char *)hdr + offset) - 1;
895 prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
896 if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
897 prevdup = NULL;
898 } else
899 prevdup = NULL;
900 /*
901 * If this isn't the end of the block, see if the entry after
902 * us is free.
903 */
904 if (offset + len < end) {
905 postdup =
906 (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
907 if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
908 postdup = NULL;
909 } else
910 postdup = NULL;
911 ASSERT(*needscanp == 0);
912 needscan = 0;
913 /*
914 * Previous and following entries are both free,
915 * merge everything into a single free entry.
916 */
917 bf = xfs_dir2_data_bestfree_p(args->dp->i_mount, hdr);
918 if (prevdup && postdup) {
919 xfs_dir2_data_free_t *dfp2; /* another bestfree pointer */
920
921 /*
922 * See if prevdup and/or postdup are in bestfree table.
923 */
924 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
925 dfp2 = xfs_dir2_data_freefind(hdr, bf, postdup);
926 /*
927 * We need a rescan unless there are exactly 2 free entries
928 * namely our two. Then we know what's happening, otherwise
929 * since the third bestfree is there, there might be more
930 * entries.
931 */
932 needscan = (bf[2].length != 0);
933 /*
934 * Fix up the new big freespace.
935 */
936 be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
937 *xfs_dir2_data_unused_tag_p(prevdup) =
938 cpu_to_be16((char *)prevdup - (char *)hdr);
939 xfs_dir2_data_log_unused(args, bp, prevdup);
940 if (!needscan) {
941 /*
942 * Has to be the case that entries 0 and 1 are
943 * dfp and dfp2 (don't know which is which), and
944 * entry 2 is empty.
945 * Remove entry 1 first then entry 0.
946 */
947 ASSERT(dfp && dfp2);
948 if (dfp == &bf[1]) {
949 dfp = &bf[0];
950 ASSERT(dfp2 == dfp);
951 dfp2 = &bf[1];
952 }
953 xfs_dir2_data_freeremove(hdr, bf, dfp2, needlogp);
954 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
955 /*
956 * Now insert the new entry.
957 */
958 dfp = xfs_dir2_data_freeinsert(hdr, bf, prevdup,
959 needlogp);
960 ASSERT(dfp == &bf[0]);
961 ASSERT(dfp->length == prevdup->length);
962 ASSERT(!dfp[1].length);
963 ASSERT(!dfp[2].length);
964 }
965 }
966 /*
967 * The entry before us is free, merge with it.
968 */
969 else if (prevdup) {
970 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
971 be16_add_cpu(&prevdup->length, len);
972 *xfs_dir2_data_unused_tag_p(prevdup) =
973 cpu_to_be16((char *)prevdup - (char *)hdr);
974 xfs_dir2_data_log_unused(args, bp, prevdup);
975 /*
976 * If the previous entry was in the table, the new entry
977 * is longer, so it will be in the table too. Remove
978 * the old one and add the new one.
979 */
980 if (dfp) {
981 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
982 xfs_dir2_data_freeinsert(hdr, bf, prevdup, needlogp);
983 }
984 /*
985 * Otherwise we need a scan if the new entry is big enough.
986 */
987 else {
988 needscan = be16_to_cpu(prevdup->length) >
989 be16_to_cpu(bf[2].length);
990 }
991 }
992 /*
993 * The following entry is free, merge with it.
994 */
995 else if (postdup) {
996 dfp = xfs_dir2_data_freefind(hdr, bf, postdup);
997 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
998 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
999 newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
1000 *xfs_dir2_data_unused_tag_p(newdup) =
1001 cpu_to_be16((char *)newdup - (char *)hdr);
1002 xfs_dir2_data_log_unused(args, bp, newdup);
1003 /*
1004 * If the following entry was in the table, the new entry
1005 * is longer, so it will be in the table too. Remove
1006 * the old one and add the new one.
1007 */
1008 if (dfp) {
1009 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
1010 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
1011 }
1012 /*
1013 * Otherwise we need a scan if the new entry is big enough.
1014 */
1015 else {
1016 needscan = be16_to_cpu(newdup->length) >
1017 be16_to_cpu(bf[2].length);
1018 }
1019 }
1020 /*
1021 * Neither neighbor is free. Make a new entry.
1022 */
1023 else {
1024 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
1025 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1026 newdup->length = cpu_to_be16(len);
1027 *xfs_dir2_data_unused_tag_p(newdup) =
1028 cpu_to_be16((char *)newdup - (char *)hdr);
1029 xfs_dir2_data_log_unused(args, bp, newdup);
1030 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
1031 }
1032 *needscanp = needscan;
1033}
1034
1035/* Check our free data for obvious signs of corruption. */
1036static inline xfs_failaddr_t
1037xfs_dir2_data_check_free(
1038 struct xfs_dir2_data_hdr *hdr,
1039 struct xfs_dir2_data_unused *dup,
1040 xfs_dir2_data_aoff_t offset,
1041 xfs_dir2_data_aoff_t len)
1042{
1043 if (hdr->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC) &&
1044 hdr->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC) &&
1045 hdr->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) &&
1046 hdr->magic != cpu_to_be32(XFS_DIR3_BLOCK_MAGIC))
1047 return __this_address;
1048 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG)
1049 return __this_address;
1050 if (offset < (char *)dup - (char *)hdr)
1051 return __this_address;
1052 if (offset + len > (char *)dup + be16_to_cpu(dup->length) - (char *)hdr)
1053 return __this_address;
1054 if ((char *)dup - (char *)hdr !=
1055 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)))
1056 return __this_address;
1057 return NULL;
1058}
1059
1060/* Sanity-check a new bestfree entry. */
1061static inline xfs_failaddr_t
1062xfs_dir2_data_check_new_free(
1063 struct xfs_dir2_data_hdr *hdr,
1064 struct xfs_dir2_data_free *dfp,
1065 struct xfs_dir2_data_unused *newdup)
1066{
1067 if (dfp == NULL)
1068 return __this_address;
1069 if (dfp->length != newdup->length)
1070 return __this_address;
1071 if (be16_to_cpu(dfp->offset) != (char *)newdup - (char *)hdr)
1072 return __this_address;
1073 return NULL;
1074}
1075
1076/*
1077 * Take a byte range out of an existing unused space and make it un-free.
1078 */
1079int
1080xfs_dir2_data_use_free(
1081 struct xfs_da_args *args,
1082 struct xfs_buf *bp,
1083 xfs_dir2_data_unused_t *dup, /* unused entry */
1084 xfs_dir2_data_aoff_t offset, /* starting offset to use */
1085 xfs_dir2_data_aoff_t len, /* length to use */
1086 int *needlogp, /* out: need to log header */
1087 int *needscanp) /* out: need regen bestfree */
1088{
1089 xfs_dir2_data_hdr_t *hdr; /* data block header */
1090 xfs_dir2_data_free_t *dfp; /* bestfree pointer */
1091 xfs_dir2_data_unused_t *newdup; /* new unused entry */
1092 xfs_dir2_data_unused_t *newdup2; /* another new unused entry */
1093 struct xfs_dir2_data_free *bf;
1094 xfs_failaddr_t fa;
1095 int matchback; /* matches end of freespace */
1096 int matchfront; /* matches start of freespace */
1097 int needscan; /* need to regen bestfree */
1098 int oldlen; /* old unused entry's length */
1099
1100 hdr = bp->b_addr;
1101 fa = xfs_dir2_data_check_free(hdr, dup, offset, len);
1102 if (fa)
1103 goto corrupt;
1104 /*
1105 * Look up the entry in the bestfree table.
1106 */
1107 oldlen = be16_to_cpu(dup->length);
1108 bf = xfs_dir2_data_bestfree_p(args->dp->i_mount, hdr);
1109 dfp = xfs_dir2_data_freefind(hdr, bf, dup);
1110 ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
1111 /*
1112 * Check for alignment with front and back of the entry.
1113 */
1114 matchfront = (char *)dup - (char *)hdr == offset;
1115 matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
1116 ASSERT(*needscanp == 0);
1117 needscan = 0;
1118 /*
1119 * If we matched it exactly we just need to get rid of it from
1120 * the bestfree table.
1121 */
1122 if (matchfront && matchback) {
1123 if (dfp) {
1124 needscan = (bf[2].offset != 0);
1125 if (!needscan)
1126 xfs_dir2_data_freeremove(hdr, bf, dfp,
1127 needlogp);
1128 }
1129 }
1130 /*
1131 * We match the first part of the entry.
1132 * Make a new entry with the remaining freespace.
1133 */
1134 else if (matchfront) {
1135 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1136 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1137 newdup->length = cpu_to_be16(oldlen - len);
1138 *xfs_dir2_data_unused_tag_p(newdup) =
1139 cpu_to_be16((char *)newdup - (char *)hdr);
1140 xfs_dir2_data_log_unused(args, bp, newdup);
1141 /*
1142 * If it was in the table, remove it and add the new one.
1143 */
1144 if (dfp) {
1145 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
1146 dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
1147 needlogp);
1148 fa = xfs_dir2_data_check_new_free(hdr, dfp, newdup);
1149 if (fa)
1150 goto corrupt;
1151 /*
1152 * If we got inserted at the last slot,
1153 * that means we don't know if there was a better
1154 * choice for the last slot, or not. Rescan.
1155 */
1156 needscan = dfp == &bf[2];
1157 }
1158 }
1159 /*
1160 * We match the last part of the entry.
1161 * Trim the allocated space off the tail of the entry.
1162 */
1163 else if (matchback) {
1164 newdup = dup;
1165 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1166 *xfs_dir2_data_unused_tag_p(newdup) =
1167 cpu_to_be16((char *)newdup - (char *)hdr);
1168 xfs_dir2_data_log_unused(args, bp, newdup);
1169 /*
1170 * If it was in the table, remove it and add the new one.
1171 */
1172 if (dfp) {
1173 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
1174 dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
1175 needlogp);
1176 fa = xfs_dir2_data_check_new_free(hdr, dfp, newdup);
1177 if (fa)
1178 goto corrupt;
1179 /*
1180 * If we got inserted at the last slot,
1181 * that means we don't know if there was a better
1182 * choice for the last slot, or not. Rescan.
1183 */
1184 needscan = dfp == &bf[2];
1185 }
1186 }
1187 /*
1188 * Poking out the middle of an entry.
1189 * Make two new entries.
1190 */
1191 else {
1192 newdup = dup;
1193 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1194 *xfs_dir2_data_unused_tag_p(newdup) =
1195 cpu_to_be16((char *)newdup - (char *)hdr);
1196 xfs_dir2_data_log_unused(args, bp, newdup);
1197 newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1198 newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1199 newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1200 *xfs_dir2_data_unused_tag_p(newdup2) =
1201 cpu_to_be16((char *)newdup2 - (char *)hdr);
1202 xfs_dir2_data_log_unused(args, bp, newdup2);
1203 /*
1204 * If the old entry was in the table, we need to scan
1205 * if the 3rd entry was valid, since these entries
1206 * are smaller than the old one.
1207 * If we don't need to scan that means there were 1 or 2
1208 * entries in the table, and removing the old and adding
1209 * the 2 new will work.
1210 */
1211 if (dfp) {
1212 needscan = (bf[2].length != 0);
1213 if (!needscan) {
1214 xfs_dir2_data_freeremove(hdr, bf, dfp,
1215 needlogp);
1216 xfs_dir2_data_freeinsert(hdr, bf, newdup,
1217 needlogp);
1218 xfs_dir2_data_freeinsert(hdr, bf, newdup2,
1219 needlogp);
1220 }
1221 }
1222 }
1223 *needscanp = needscan;
1224 return 0;
1225corrupt:
1226 xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, args->dp->i_mount,
1227 hdr, sizeof(*hdr), __FILE__, __LINE__, fa);
1228 xfs_da_mark_sick(args);
1229 return -EFSCORRUPTED;
1230}
1231
1232/* Find the end of the entry data in a data/block format dir block. */
1233unsigned int
1234xfs_dir3_data_end_offset(
1235 struct xfs_da_geometry *geo,
1236 struct xfs_dir2_data_hdr *hdr)
1237{
1238 void *p;
1239
1240 switch (hdr->magic) {
1241 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
1242 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
1243 p = xfs_dir2_block_leaf_p(xfs_dir2_block_tail_p(geo, hdr));
1244 return p - (void *)hdr;
1245 case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
1246 case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
1247 return geo->blksize;
1248 default:
1249 return 0;
1250 }
1251}
1/*
2 * Copyright (c) 2000-2002,2005 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_dir2.h"
29#include "xfs_dir2_priv.h"
30#include "xfs_error.h"
31#include "xfs_trans.h"
32#include "xfs_buf_item.h"
33#include "xfs_cksum.h"
34#include "xfs_log.h"
35
36/*
37 * Check the consistency of the data block.
38 * The input can also be a block-format directory.
39 * Return NULL if the buffer is good, otherwise the address of the error.
40 */
41xfs_failaddr_t
42__xfs_dir3_data_check(
43 struct xfs_inode *dp, /* incore inode pointer */
44 struct xfs_buf *bp) /* data block's buffer */
45{
46 xfs_dir2_dataptr_t addr; /* addr for leaf lookup */
47 xfs_dir2_data_free_t *bf; /* bestfree table */
48 xfs_dir2_block_tail_t *btp=NULL; /* block tail */
49 int count; /* count of entries found */
50 xfs_dir2_data_hdr_t *hdr; /* data block header */
51 xfs_dir2_data_entry_t *dep; /* data entry */
52 xfs_dir2_data_free_t *dfp; /* bestfree entry */
53 xfs_dir2_data_unused_t *dup; /* unused entry */
54 char *endp; /* end of useful data */
55 int freeseen; /* mask of bestfrees seen */
56 xfs_dahash_t hash; /* hash of current name */
57 int i; /* leaf index */
58 int lastfree; /* last entry was unused */
59 xfs_dir2_leaf_entry_t *lep=NULL; /* block leaf entries */
60 xfs_mount_t *mp; /* filesystem mount point */
61 char *p; /* current data position */
62 int stale; /* count of stale leaves */
63 struct xfs_name name;
64 const struct xfs_dir_ops *ops;
65 struct xfs_da_geometry *geo;
66
67 mp = bp->b_target->bt_mount;
68 geo = mp->m_dir_geo;
69
70 /*
71 * We can be passed a null dp here from a verifier, so we need to go the
72 * hard way to get them.
73 */
74 ops = xfs_dir_get_ops(mp, dp);
75
76 /*
77 * If this isn't a directory, or we don't get handed the dir ops,
78 * something is seriously wrong. Bail out.
79 */
80 if ((dp && !S_ISDIR(VFS_I(dp)->i_mode)) ||
81 ops != xfs_dir_get_ops(mp, NULL))
82 return __this_address;
83
84 hdr = bp->b_addr;
85 p = (char *)ops->data_entry_p(hdr);
86
87 switch (hdr->magic) {
88 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
89 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
90 btp = xfs_dir2_block_tail_p(geo, hdr);
91 lep = xfs_dir2_block_leaf_p(btp);
92
93 /*
94 * The number of leaf entries is limited by the size of the
95 * block and the amount of space used by the data entries.
96 * We don't know how much space is used by the data entries yet,
97 * so just ensure that the count falls somewhere inside the
98 * block right now.
99 */
100 if (be32_to_cpu(btp->count) >=
101 ((char *)btp - p) / sizeof(struct xfs_dir2_leaf_entry))
102 return __this_address;
103 break;
104 case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
105 case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
106 break;
107 default:
108 return __this_address;
109 }
110 endp = xfs_dir3_data_endp(geo, hdr);
111 if (!endp)
112 return __this_address;
113
114 /*
115 * Account for zero bestfree entries.
116 */
117 bf = ops->data_bestfree_p(hdr);
118 count = lastfree = freeseen = 0;
119 if (!bf[0].length) {
120 if (bf[0].offset)
121 return __this_address;
122 freeseen |= 1 << 0;
123 }
124 if (!bf[1].length) {
125 if (bf[1].offset)
126 return __this_address;
127 freeseen |= 1 << 1;
128 }
129 if (!bf[2].length) {
130 if (bf[2].offset)
131 return __this_address;
132 freeseen |= 1 << 2;
133 }
134
135 if (be16_to_cpu(bf[0].length) < be16_to_cpu(bf[1].length))
136 return __this_address;
137 if (be16_to_cpu(bf[1].length) < be16_to_cpu(bf[2].length))
138 return __this_address;
139 /*
140 * Loop over the data/unused entries.
141 */
142 while (p < endp) {
143 dup = (xfs_dir2_data_unused_t *)p;
144 /*
145 * If it's unused, look for the space in the bestfree table.
146 * If we find it, account for that, else make sure it
147 * doesn't need to be there.
148 */
149 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
150 if (lastfree != 0)
151 return __this_address;
152 if (endp < p + be16_to_cpu(dup->length))
153 return __this_address;
154 if (be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)) !=
155 (char *)dup - (char *)hdr)
156 return __this_address;
157 dfp = xfs_dir2_data_freefind(hdr, bf, dup);
158 if (dfp) {
159 i = (int)(dfp - bf);
160 if ((freeseen & (1 << i)) != 0)
161 return __this_address;
162 freeseen |= 1 << i;
163 } else {
164 if (be16_to_cpu(dup->length) >
165 be16_to_cpu(bf[2].length))
166 return __this_address;
167 }
168 p += be16_to_cpu(dup->length);
169 lastfree = 1;
170 continue;
171 }
172 /*
173 * It's a real entry. Validate the fields.
174 * If this is a block directory then make sure it's
175 * in the leaf section of the block.
176 * The linear search is crude but this is DEBUG code.
177 */
178 dep = (xfs_dir2_data_entry_t *)p;
179 if (dep->namelen == 0)
180 return __this_address;
181 if (xfs_dir_ino_validate(mp, be64_to_cpu(dep->inumber)))
182 return __this_address;
183 if (endp < p + ops->data_entsize(dep->namelen))
184 return __this_address;
185 if (be16_to_cpu(*ops->data_entry_tag_p(dep)) !=
186 (char *)dep - (char *)hdr)
187 return __this_address;
188 if (ops->data_get_ftype(dep) >= XFS_DIR3_FT_MAX)
189 return __this_address;
190 count++;
191 lastfree = 0;
192 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
193 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
194 addr = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
195 (xfs_dir2_data_aoff_t)
196 ((char *)dep - (char *)hdr));
197 name.name = dep->name;
198 name.len = dep->namelen;
199 hash = mp->m_dirnameops->hashname(&name);
200 for (i = 0; i < be32_to_cpu(btp->count); i++) {
201 if (be32_to_cpu(lep[i].address) == addr &&
202 be32_to_cpu(lep[i].hashval) == hash)
203 break;
204 }
205 if (i >= be32_to_cpu(btp->count))
206 return __this_address;
207 }
208 p += ops->data_entsize(dep->namelen);
209 }
210 /*
211 * Need to have seen all the entries and all the bestfree slots.
212 */
213 if (freeseen != 7)
214 return __this_address;
215 if (hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
216 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC)) {
217 for (i = stale = 0; i < be32_to_cpu(btp->count); i++) {
218 if (lep[i].address ==
219 cpu_to_be32(XFS_DIR2_NULL_DATAPTR))
220 stale++;
221 if (i > 0 && be32_to_cpu(lep[i].hashval) <
222 be32_to_cpu(lep[i - 1].hashval))
223 return __this_address;
224 }
225 if (count != be32_to_cpu(btp->count) - be32_to_cpu(btp->stale))
226 return __this_address;
227 if (stale != be32_to_cpu(btp->stale))
228 return __this_address;
229 }
230 return NULL;
231}
232
233#ifdef DEBUG
234void
235xfs_dir3_data_check(
236 struct xfs_inode *dp,
237 struct xfs_buf *bp)
238{
239 xfs_failaddr_t fa;
240
241 fa = __xfs_dir3_data_check(dp, bp);
242 if (!fa)
243 return;
244 xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, dp->i_mount,
245 bp->b_addr, __FILE__, __LINE__, fa);
246 ASSERT(0);
247}
248#endif
249
250static xfs_failaddr_t
251xfs_dir3_data_verify(
252 struct xfs_buf *bp)
253{
254 struct xfs_mount *mp = bp->b_target->bt_mount;
255 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
256
257 if (xfs_sb_version_hascrc(&mp->m_sb)) {
258 if (hdr3->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC))
259 return __this_address;
260 if (!uuid_equal(&hdr3->uuid, &mp->m_sb.sb_meta_uuid))
261 return __this_address;
262 if (be64_to_cpu(hdr3->blkno) != bp->b_bn)
263 return __this_address;
264 if (!xfs_log_check_lsn(mp, be64_to_cpu(hdr3->lsn)))
265 return __this_address;
266 } else {
267 if (hdr3->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC))
268 return __this_address;
269 }
270 return __xfs_dir3_data_check(NULL, bp);
271}
272
273/*
274 * Readahead of the first block of the directory when it is opened is completely
275 * oblivious to the format of the directory. Hence we can either get a block
276 * format buffer or a data format buffer on readahead.
277 */
278static void
279xfs_dir3_data_reada_verify(
280 struct xfs_buf *bp)
281{
282 struct xfs_dir2_data_hdr *hdr = bp->b_addr;
283
284 switch (hdr->magic) {
285 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
286 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
287 bp->b_ops = &xfs_dir3_block_buf_ops;
288 bp->b_ops->verify_read(bp);
289 return;
290 case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
291 case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
292 bp->b_ops = &xfs_dir3_data_buf_ops;
293 bp->b_ops->verify_read(bp);
294 return;
295 default:
296 xfs_verifier_error(bp, -EFSCORRUPTED, __this_address);
297 break;
298 }
299}
300
301static void
302xfs_dir3_data_read_verify(
303 struct xfs_buf *bp)
304{
305 struct xfs_mount *mp = bp->b_target->bt_mount;
306 xfs_failaddr_t fa;
307
308 if (xfs_sb_version_hascrc(&mp->m_sb) &&
309 !xfs_buf_verify_cksum(bp, XFS_DIR3_DATA_CRC_OFF))
310 xfs_verifier_error(bp, -EFSBADCRC, __this_address);
311 else {
312 fa = xfs_dir3_data_verify(bp);
313 if (fa)
314 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
315 }
316}
317
318static void
319xfs_dir3_data_write_verify(
320 struct xfs_buf *bp)
321{
322 struct xfs_mount *mp = bp->b_target->bt_mount;
323 struct xfs_buf_log_item *bip = bp->b_log_item;
324 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
325 xfs_failaddr_t fa;
326
327 fa = xfs_dir3_data_verify(bp);
328 if (fa) {
329 xfs_verifier_error(bp, -EFSCORRUPTED, fa);
330 return;
331 }
332
333 if (!xfs_sb_version_hascrc(&mp->m_sb))
334 return;
335
336 if (bip)
337 hdr3->lsn = cpu_to_be64(bip->bli_item.li_lsn);
338
339 xfs_buf_update_cksum(bp, XFS_DIR3_DATA_CRC_OFF);
340}
341
342const struct xfs_buf_ops xfs_dir3_data_buf_ops = {
343 .name = "xfs_dir3_data",
344 .verify_read = xfs_dir3_data_read_verify,
345 .verify_write = xfs_dir3_data_write_verify,
346 .verify_struct = xfs_dir3_data_verify,
347};
348
349static const struct xfs_buf_ops xfs_dir3_data_reada_buf_ops = {
350 .name = "xfs_dir3_data_reada",
351 .verify_read = xfs_dir3_data_reada_verify,
352 .verify_write = xfs_dir3_data_write_verify,
353};
354
355
356int
357xfs_dir3_data_read(
358 struct xfs_trans *tp,
359 struct xfs_inode *dp,
360 xfs_dablk_t bno,
361 xfs_daddr_t mapped_bno,
362 struct xfs_buf **bpp)
363{
364 int err;
365
366 err = xfs_da_read_buf(tp, dp, bno, mapped_bno, bpp,
367 XFS_DATA_FORK, &xfs_dir3_data_buf_ops);
368 if (!err && tp && *bpp)
369 xfs_trans_buf_set_type(tp, *bpp, XFS_BLFT_DIR_DATA_BUF);
370 return err;
371}
372
373int
374xfs_dir3_data_readahead(
375 struct xfs_inode *dp,
376 xfs_dablk_t bno,
377 xfs_daddr_t mapped_bno)
378{
379 return xfs_da_reada_buf(dp, bno, mapped_bno,
380 XFS_DATA_FORK, &xfs_dir3_data_reada_buf_ops);
381}
382
383/*
384 * Given a data block and an unused entry from that block,
385 * return the bestfree entry if any that corresponds to it.
386 */
387xfs_dir2_data_free_t *
388xfs_dir2_data_freefind(
389 struct xfs_dir2_data_hdr *hdr, /* data block header */
390 struct xfs_dir2_data_free *bf, /* bestfree table pointer */
391 struct xfs_dir2_data_unused *dup) /* unused space */
392{
393 xfs_dir2_data_free_t *dfp; /* bestfree entry */
394 xfs_dir2_data_aoff_t off; /* offset value needed */
395#ifdef DEBUG
396 int matched; /* matched the value */
397 int seenzero; /* saw a 0 bestfree entry */
398#endif
399
400 off = (xfs_dir2_data_aoff_t)((char *)dup - (char *)hdr);
401
402#ifdef DEBUG
403 /*
404 * Validate some consistency in the bestfree table.
405 * Check order, non-overlapping entries, and if we find the
406 * one we're looking for it has to be exact.
407 */
408 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
409 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
410 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
411 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
412 for (dfp = &bf[0], seenzero = matched = 0;
413 dfp < &bf[XFS_DIR2_DATA_FD_COUNT];
414 dfp++) {
415 if (!dfp->offset) {
416 ASSERT(!dfp->length);
417 seenzero = 1;
418 continue;
419 }
420 ASSERT(seenzero == 0);
421 if (be16_to_cpu(dfp->offset) == off) {
422 matched = 1;
423 ASSERT(dfp->length == dup->length);
424 } else if (off < be16_to_cpu(dfp->offset))
425 ASSERT(off + be16_to_cpu(dup->length) <= be16_to_cpu(dfp->offset));
426 else
427 ASSERT(be16_to_cpu(dfp->offset) + be16_to_cpu(dfp->length) <= off);
428 ASSERT(matched || be16_to_cpu(dfp->length) >= be16_to_cpu(dup->length));
429 if (dfp > &bf[0])
430 ASSERT(be16_to_cpu(dfp[-1].length) >= be16_to_cpu(dfp[0].length));
431 }
432#endif
433 /*
434 * If this is smaller than the smallest bestfree entry,
435 * it can't be there since they're sorted.
436 */
437 if (be16_to_cpu(dup->length) <
438 be16_to_cpu(bf[XFS_DIR2_DATA_FD_COUNT - 1].length))
439 return NULL;
440 /*
441 * Look at the three bestfree entries for our guy.
442 */
443 for (dfp = &bf[0]; dfp < &bf[XFS_DIR2_DATA_FD_COUNT]; dfp++) {
444 if (!dfp->offset)
445 return NULL;
446 if (be16_to_cpu(dfp->offset) == off)
447 return dfp;
448 }
449 /*
450 * Didn't find it. This only happens if there are duplicate lengths.
451 */
452 return NULL;
453}
454
455/*
456 * Insert an unused-space entry into the bestfree table.
457 */
458xfs_dir2_data_free_t * /* entry inserted */
459xfs_dir2_data_freeinsert(
460 struct xfs_dir2_data_hdr *hdr, /* data block pointer */
461 struct xfs_dir2_data_free *dfp, /* bestfree table pointer */
462 struct xfs_dir2_data_unused *dup, /* unused space */
463 int *loghead) /* log the data header (out) */
464{
465 xfs_dir2_data_free_t new; /* new bestfree entry */
466
467 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
468 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
469 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
470 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
471
472 new.length = dup->length;
473 new.offset = cpu_to_be16((char *)dup - (char *)hdr);
474
475 /*
476 * Insert at position 0, 1, or 2; or not at all.
477 */
478 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[0].length)) {
479 dfp[2] = dfp[1];
480 dfp[1] = dfp[0];
481 dfp[0] = new;
482 *loghead = 1;
483 return &dfp[0];
484 }
485 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[1].length)) {
486 dfp[2] = dfp[1];
487 dfp[1] = new;
488 *loghead = 1;
489 return &dfp[1];
490 }
491 if (be16_to_cpu(new.length) > be16_to_cpu(dfp[2].length)) {
492 dfp[2] = new;
493 *loghead = 1;
494 return &dfp[2];
495 }
496 return NULL;
497}
498
499/*
500 * Remove a bestfree entry from the table.
501 */
502STATIC void
503xfs_dir2_data_freeremove(
504 struct xfs_dir2_data_hdr *hdr, /* data block header */
505 struct xfs_dir2_data_free *bf, /* bestfree table pointer */
506 struct xfs_dir2_data_free *dfp, /* bestfree entry pointer */
507 int *loghead) /* out: log data header */
508{
509
510 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
511 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
512 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
513 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
514
515 /*
516 * It's the first entry, slide the next 2 up.
517 */
518 if (dfp == &bf[0]) {
519 bf[0] = bf[1];
520 bf[1] = bf[2];
521 }
522 /*
523 * It's the second entry, slide the 3rd entry up.
524 */
525 else if (dfp == &bf[1])
526 bf[1] = bf[2];
527 /*
528 * Must be the last entry.
529 */
530 else
531 ASSERT(dfp == &bf[2]);
532 /*
533 * Clear the 3rd entry, must be zero now.
534 */
535 bf[2].length = 0;
536 bf[2].offset = 0;
537 *loghead = 1;
538}
539
540/*
541 * Given a data block, reconstruct its bestfree map.
542 */
543void
544xfs_dir2_data_freescan_int(
545 struct xfs_da_geometry *geo,
546 const struct xfs_dir_ops *ops,
547 struct xfs_dir2_data_hdr *hdr,
548 int *loghead)
549{
550 xfs_dir2_data_entry_t *dep; /* active data entry */
551 xfs_dir2_data_unused_t *dup; /* unused data entry */
552 struct xfs_dir2_data_free *bf;
553 char *endp; /* end of block's data */
554 char *p; /* current entry pointer */
555
556 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
557 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
558 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
559 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
560
561 /*
562 * Start by clearing the table.
563 */
564 bf = ops->data_bestfree_p(hdr);
565 memset(bf, 0, sizeof(*bf) * XFS_DIR2_DATA_FD_COUNT);
566 *loghead = 1;
567 /*
568 * Set up pointers.
569 */
570 p = (char *)ops->data_entry_p(hdr);
571 endp = xfs_dir3_data_endp(geo, hdr);
572 /*
573 * Loop over the block's entries.
574 */
575 while (p < endp) {
576 dup = (xfs_dir2_data_unused_t *)p;
577 /*
578 * If it's a free entry, insert it.
579 */
580 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
581 ASSERT((char *)dup - (char *)hdr ==
582 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)));
583 xfs_dir2_data_freeinsert(hdr, bf, dup, loghead);
584 p += be16_to_cpu(dup->length);
585 }
586 /*
587 * For active entries, check their tags and skip them.
588 */
589 else {
590 dep = (xfs_dir2_data_entry_t *)p;
591 ASSERT((char *)dep - (char *)hdr ==
592 be16_to_cpu(*ops->data_entry_tag_p(dep)));
593 p += ops->data_entsize(dep->namelen);
594 }
595 }
596}
597
598void
599xfs_dir2_data_freescan(
600 struct xfs_inode *dp,
601 struct xfs_dir2_data_hdr *hdr,
602 int *loghead)
603{
604 return xfs_dir2_data_freescan_int(dp->i_mount->m_dir_geo, dp->d_ops,
605 hdr, loghead);
606}
607
608/*
609 * Initialize a data block at the given block number in the directory.
610 * Give back the buffer for the created block.
611 */
612int /* error */
613xfs_dir3_data_init(
614 xfs_da_args_t *args, /* directory operation args */
615 xfs_dir2_db_t blkno, /* logical dir block number */
616 struct xfs_buf **bpp) /* output block buffer */
617{
618 struct xfs_buf *bp; /* block buffer */
619 xfs_dir2_data_hdr_t *hdr; /* data block header */
620 xfs_inode_t *dp; /* incore directory inode */
621 xfs_dir2_data_unused_t *dup; /* unused entry pointer */
622 struct xfs_dir2_data_free *bf;
623 int error; /* error return value */
624 int i; /* bestfree index */
625 xfs_mount_t *mp; /* filesystem mount point */
626 xfs_trans_t *tp; /* transaction pointer */
627 int t; /* temp */
628
629 dp = args->dp;
630 mp = dp->i_mount;
631 tp = args->trans;
632 /*
633 * Get the buffer set up for the block.
634 */
635 error = xfs_da_get_buf(tp, dp, xfs_dir2_db_to_da(args->geo, blkno),
636 -1, &bp, XFS_DATA_FORK);
637 if (error)
638 return error;
639 bp->b_ops = &xfs_dir3_data_buf_ops;
640 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DIR_DATA_BUF);
641
642 /*
643 * Initialize the header.
644 */
645 hdr = bp->b_addr;
646 if (xfs_sb_version_hascrc(&mp->m_sb)) {
647 struct xfs_dir3_blk_hdr *hdr3 = bp->b_addr;
648
649 memset(hdr3, 0, sizeof(*hdr3));
650 hdr3->magic = cpu_to_be32(XFS_DIR3_DATA_MAGIC);
651 hdr3->blkno = cpu_to_be64(bp->b_bn);
652 hdr3->owner = cpu_to_be64(dp->i_ino);
653 uuid_copy(&hdr3->uuid, &mp->m_sb.sb_meta_uuid);
654
655 } else
656 hdr->magic = cpu_to_be32(XFS_DIR2_DATA_MAGIC);
657
658 bf = dp->d_ops->data_bestfree_p(hdr);
659 bf[0].offset = cpu_to_be16(dp->d_ops->data_entry_offset);
660 for (i = 1; i < XFS_DIR2_DATA_FD_COUNT; i++) {
661 bf[i].length = 0;
662 bf[i].offset = 0;
663 }
664
665 /*
666 * Set up an unused entry for the block's body.
667 */
668 dup = dp->d_ops->data_unused_p(hdr);
669 dup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
670
671 t = args->geo->blksize - (uint)dp->d_ops->data_entry_offset;
672 bf[0].length = cpu_to_be16(t);
673 dup->length = cpu_to_be16(t);
674 *xfs_dir2_data_unused_tag_p(dup) = cpu_to_be16((char *)dup - (char *)hdr);
675 /*
676 * Log it and return it.
677 */
678 xfs_dir2_data_log_header(args, bp);
679 xfs_dir2_data_log_unused(args, bp, dup);
680 *bpp = bp;
681 return 0;
682}
683
684/*
685 * Log an active data entry from the block.
686 */
687void
688xfs_dir2_data_log_entry(
689 struct xfs_da_args *args,
690 struct xfs_buf *bp,
691 xfs_dir2_data_entry_t *dep) /* data entry pointer */
692{
693 struct xfs_dir2_data_hdr *hdr = bp->b_addr;
694
695 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
696 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
697 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
698 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
699
700 xfs_trans_log_buf(args->trans, bp, (uint)((char *)dep - (char *)hdr),
701 (uint)((char *)(args->dp->d_ops->data_entry_tag_p(dep) + 1) -
702 (char *)hdr - 1));
703}
704
705/*
706 * Log a data block header.
707 */
708void
709xfs_dir2_data_log_header(
710 struct xfs_da_args *args,
711 struct xfs_buf *bp)
712{
713#ifdef DEBUG
714 struct xfs_dir2_data_hdr *hdr = bp->b_addr;
715
716 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
717 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
718 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
719 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
720#endif
721
722 xfs_trans_log_buf(args->trans, bp, 0,
723 args->dp->d_ops->data_entry_offset - 1);
724}
725
726/*
727 * Log a data unused entry.
728 */
729void
730xfs_dir2_data_log_unused(
731 struct xfs_da_args *args,
732 struct xfs_buf *bp,
733 xfs_dir2_data_unused_t *dup) /* data unused pointer */
734{
735 xfs_dir2_data_hdr_t *hdr = bp->b_addr;
736
737 ASSERT(hdr->magic == cpu_to_be32(XFS_DIR2_DATA_MAGIC) ||
738 hdr->magic == cpu_to_be32(XFS_DIR3_DATA_MAGIC) ||
739 hdr->magic == cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) ||
740 hdr->magic == cpu_to_be32(XFS_DIR3_BLOCK_MAGIC));
741
742 /*
743 * Log the first part of the unused entry.
744 */
745 xfs_trans_log_buf(args->trans, bp, (uint)((char *)dup - (char *)hdr),
746 (uint)((char *)&dup->length + sizeof(dup->length) -
747 1 - (char *)hdr));
748 /*
749 * Log the end (tag) of the unused entry.
750 */
751 xfs_trans_log_buf(args->trans, bp,
752 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr),
753 (uint)((char *)xfs_dir2_data_unused_tag_p(dup) - (char *)hdr +
754 sizeof(xfs_dir2_data_off_t) - 1));
755}
756
757/*
758 * Make a byte range in the data block unused.
759 * Its current contents are unimportant.
760 */
761void
762xfs_dir2_data_make_free(
763 struct xfs_da_args *args,
764 struct xfs_buf *bp,
765 xfs_dir2_data_aoff_t offset, /* starting byte offset */
766 xfs_dir2_data_aoff_t len, /* length in bytes */
767 int *needlogp, /* out: log header */
768 int *needscanp) /* out: regen bestfree */
769{
770 xfs_dir2_data_hdr_t *hdr; /* data block pointer */
771 xfs_dir2_data_free_t *dfp; /* bestfree pointer */
772 char *endptr; /* end of data area */
773 int needscan; /* need to regen bestfree */
774 xfs_dir2_data_unused_t *newdup; /* new unused entry */
775 xfs_dir2_data_unused_t *postdup; /* unused entry after us */
776 xfs_dir2_data_unused_t *prevdup; /* unused entry before us */
777 struct xfs_dir2_data_free *bf;
778
779 hdr = bp->b_addr;
780
781 /*
782 * Figure out where the end of the data area is.
783 */
784 endptr = xfs_dir3_data_endp(args->geo, hdr);
785 ASSERT(endptr != NULL);
786
787 /*
788 * If this isn't the start of the block, then back up to
789 * the previous entry and see if it's free.
790 */
791 if (offset > args->dp->d_ops->data_entry_offset) {
792 __be16 *tagp; /* tag just before us */
793
794 tagp = (__be16 *)((char *)hdr + offset) - 1;
795 prevdup = (xfs_dir2_data_unused_t *)((char *)hdr + be16_to_cpu(*tagp));
796 if (be16_to_cpu(prevdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
797 prevdup = NULL;
798 } else
799 prevdup = NULL;
800 /*
801 * If this isn't the end of the block, see if the entry after
802 * us is free.
803 */
804 if ((char *)hdr + offset + len < endptr) {
805 postdup =
806 (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
807 if (be16_to_cpu(postdup->freetag) != XFS_DIR2_DATA_FREE_TAG)
808 postdup = NULL;
809 } else
810 postdup = NULL;
811 ASSERT(*needscanp == 0);
812 needscan = 0;
813 /*
814 * Previous and following entries are both free,
815 * merge everything into a single free entry.
816 */
817 bf = args->dp->d_ops->data_bestfree_p(hdr);
818 if (prevdup && postdup) {
819 xfs_dir2_data_free_t *dfp2; /* another bestfree pointer */
820
821 /*
822 * See if prevdup and/or postdup are in bestfree table.
823 */
824 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
825 dfp2 = xfs_dir2_data_freefind(hdr, bf, postdup);
826 /*
827 * We need a rescan unless there are exactly 2 free entries
828 * namely our two. Then we know what's happening, otherwise
829 * since the third bestfree is there, there might be more
830 * entries.
831 */
832 needscan = (bf[2].length != 0);
833 /*
834 * Fix up the new big freespace.
835 */
836 be16_add_cpu(&prevdup->length, len + be16_to_cpu(postdup->length));
837 *xfs_dir2_data_unused_tag_p(prevdup) =
838 cpu_to_be16((char *)prevdup - (char *)hdr);
839 xfs_dir2_data_log_unused(args, bp, prevdup);
840 if (!needscan) {
841 /*
842 * Has to be the case that entries 0 and 1 are
843 * dfp and dfp2 (don't know which is which), and
844 * entry 2 is empty.
845 * Remove entry 1 first then entry 0.
846 */
847 ASSERT(dfp && dfp2);
848 if (dfp == &bf[1]) {
849 dfp = &bf[0];
850 ASSERT(dfp2 == dfp);
851 dfp2 = &bf[1];
852 }
853 xfs_dir2_data_freeremove(hdr, bf, dfp2, needlogp);
854 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
855 /*
856 * Now insert the new entry.
857 */
858 dfp = xfs_dir2_data_freeinsert(hdr, bf, prevdup,
859 needlogp);
860 ASSERT(dfp == &bf[0]);
861 ASSERT(dfp->length == prevdup->length);
862 ASSERT(!dfp[1].length);
863 ASSERT(!dfp[2].length);
864 }
865 }
866 /*
867 * The entry before us is free, merge with it.
868 */
869 else if (prevdup) {
870 dfp = xfs_dir2_data_freefind(hdr, bf, prevdup);
871 be16_add_cpu(&prevdup->length, len);
872 *xfs_dir2_data_unused_tag_p(prevdup) =
873 cpu_to_be16((char *)prevdup - (char *)hdr);
874 xfs_dir2_data_log_unused(args, bp, prevdup);
875 /*
876 * If the previous entry was in the table, the new entry
877 * is longer, so it will be in the table too. Remove
878 * the old one and add the new one.
879 */
880 if (dfp) {
881 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
882 xfs_dir2_data_freeinsert(hdr, bf, prevdup, needlogp);
883 }
884 /*
885 * Otherwise we need a scan if the new entry is big enough.
886 */
887 else {
888 needscan = be16_to_cpu(prevdup->length) >
889 be16_to_cpu(bf[2].length);
890 }
891 }
892 /*
893 * The following entry is free, merge with it.
894 */
895 else if (postdup) {
896 dfp = xfs_dir2_data_freefind(hdr, bf, postdup);
897 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
898 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
899 newdup->length = cpu_to_be16(len + be16_to_cpu(postdup->length));
900 *xfs_dir2_data_unused_tag_p(newdup) =
901 cpu_to_be16((char *)newdup - (char *)hdr);
902 xfs_dir2_data_log_unused(args, bp, newdup);
903 /*
904 * If the following entry was in the table, the new entry
905 * is longer, so it will be in the table too. Remove
906 * the old one and add the new one.
907 */
908 if (dfp) {
909 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
910 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
911 }
912 /*
913 * Otherwise we need a scan if the new entry is big enough.
914 */
915 else {
916 needscan = be16_to_cpu(newdup->length) >
917 be16_to_cpu(bf[2].length);
918 }
919 }
920 /*
921 * Neither neighbor is free. Make a new entry.
922 */
923 else {
924 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset);
925 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
926 newdup->length = cpu_to_be16(len);
927 *xfs_dir2_data_unused_tag_p(newdup) =
928 cpu_to_be16((char *)newdup - (char *)hdr);
929 xfs_dir2_data_log_unused(args, bp, newdup);
930 xfs_dir2_data_freeinsert(hdr, bf, newdup, needlogp);
931 }
932 *needscanp = needscan;
933}
934
935/* Check our free data for obvious signs of corruption. */
936static inline xfs_failaddr_t
937xfs_dir2_data_check_free(
938 struct xfs_dir2_data_hdr *hdr,
939 struct xfs_dir2_data_unused *dup,
940 xfs_dir2_data_aoff_t offset,
941 xfs_dir2_data_aoff_t len)
942{
943 if (hdr->magic != cpu_to_be32(XFS_DIR2_DATA_MAGIC) &&
944 hdr->magic != cpu_to_be32(XFS_DIR3_DATA_MAGIC) &&
945 hdr->magic != cpu_to_be32(XFS_DIR2_BLOCK_MAGIC) &&
946 hdr->magic != cpu_to_be32(XFS_DIR3_BLOCK_MAGIC))
947 return __this_address;
948 if (be16_to_cpu(dup->freetag) != XFS_DIR2_DATA_FREE_TAG)
949 return __this_address;
950 if (offset < (char *)dup - (char *)hdr)
951 return __this_address;
952 if (offset + len > (char *)dup + be16_to_cpu(dup->length) - (char *)hdr)
953 return __this_address;
954 if ((char *)dup - (char *)hdr !=
955 be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup)))
956 return __this_address;
957 return NULL;
958}
959
960/* Sanity-check a new bestfree entry. */
961static inline xfs_failaddr_t
962xfs_dir2_data_check_new_free(
963 struct xfs_dir2_data_hdr *hdr,
964 struct xfs_dir2_data_free *dfp,
965 struct xfs_dir2_data_unused *newdup)
966{
967 if (dfp == NULL)
968 return __this_address;
969 if (dfp->length != newdup->length)
970 return __this_address;
971 if (be16_to_cpu(dfp->offset) != (char *)newdup - (char *)hdr)
972 return __this_address;
973 return NULL;
974}
975
976/*
977 * Take a byte range out of an existing unused space and make it un-free.
978 */
979int
980xfs_dir2_data_use_free(
981 struct xfs_da_args *args,
982 struct xfs_buf *bp,
983 xfs_dir2_data_unused_t *dup, /* unused entry */
984 xfs_dir2_data_aoff_t offset, /* starting offset to use */
985 xfs_dir2_data_aoff_t len, /* length to use */
986 int *needlogp, /* out: need to log header */
987 int *needscanp) /* out: need regen bestfree */
988{
989 xfs_dir2_data_hdr_t *hdr; /* data block header */
990 xfs_dir2_data_free_t *dfp; /* bestfree pointer */
991 xfs_dir2_data_unused_t *newdup; /* new unused entry */
992 xfs_dir2_data_unused_t *newdup2; /* another new unused entry */
993 struct xfs_dir2_data_free *bf;
994 xfs_failaddr_t fa;
995 int matchback; /* matches end of freespace */
996 int matchfront; /* matches start of freespace */
997 int needscan; /* need to regen bestfree */
998 int oldlen; /* old unused entry's length */
999
1000 hdr = bp->b_addr;
1001 fa = xfs_dir2_data_check_free(hdr, dup, offset, len);
1002 if (fa)
1003 goto corrupt;
1004 /*
1005 * Look up the entry in the bestfree table.
1006 */
1007 oldlen = be16_to_cpu(dup->length);
1008 bf = args->dp->d_ops->data_bestfree_p(hdr);
1009 dfp = xfs_dir2_data_freefind(hdr, bf, dup);
1010 ASSERT(dfp || oldlen <= be16_to_cpu(bf[2].length));
1011 /*
1012 * Check for alignment with front and back of the entry.
1013 */
1014 matchfront = (char *)dup - (char *)hdr == offset;
1015 matchback = (char *)dup + oldlen - (char *)hdr == offset + len;
1016 ASSERT(*needscanp == 0);
1017 needscan = 0;
1018 /*
1019 * If we matched it exactly we just need to get rid of it from
1020 * the bestfree table.
1021 */
1022 if (matchfront && matchback) {
1023 if (dfp) {
1024 needscan = (bf[2].offset != 0);
1025 if (!needscan)
1026 xfs_dir2_data_freeremove(hdr, bf, dfp,
1027 needlogp);
1028 }
1029 }
1030 /*
1031 * We match the first part of the entry.
1032 * Make a new entry with the remaining freespace.
1033 */
1034 else if (matchfront) {
1035 newdup = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1036 newdup->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1037 newdup->length = cpu_to_be16(oldlen - len);
1038 *xfs_dir2_data_unused_tag_p(newdup) =
1039 cpu_to_be16((char *)newdup - (char *)hdr);
1040 xfs_dir2_data_log_unused(args, bp, newdup);
1041 /*
1042 * If it was in the table, remove it and add the new one.
1043 */
1044 if (dfp) {
1045 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
1046 dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
1047 needlogp);
1048 fa = xfs_dir2_data_check_new_free(hdr, dfp, newdup);
1049 if (fa)
1050 goto corrupt;
1051 /*
1052 * If we got inserted at the last slot,
1053 * that means we don't know if there was a better
1054 * choice for the last slot, or not. Rescan.
1055 */
1056 needscan = dfp == &bf[2];
1057 }
1058 }
1059 /*
1060 * We match the last part of the entry.
1061 * Trim the allocated space off the tail of the entry.
1062 */
1063 else if (matchback) {
1064 newdup = dup;
1065 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1066 *xfs_dir2_data_unused_tag_p(newdup) =
1067 cpu_to_be16((char *)newdup - (char *)hdr);
1068 xfs_dir2_data_log_unused(args, bp, newdup);
1069 /*
1070 * If it was in the table, remove it and add the new one.
1071 */
1072 if (dfp) {
1073 xfs_dir2_data_freeremove(hdr, bf, dfp, needlogp);
1074 dfp = xfs_dir2_data_freeinsert(hdr, bf, newdup,
1075 needlogp);
1076 fa = xfs_dir2_data_check_new_free(hdr, dfp, newdup);
1077 if (fa)
1078 goto corrupt;
1079 /*
1080 * If we got inserted at the last slot,
1081 * that means we don't know if there was a better
1082 * choice for the last slot, or not. Rescan.
1083 */
1084 needscan = dfp == &bf[2];
1085 }
1086 }
1087 /*
1088 * Poking out the middle of an entry.
1089 * Make two new entries.
1090 */
1091 else {
1092 newdup = dup;
1093 newdup->length = cpu_to_be16(((char *)hdr + offset) - (char *)newdup);
1094 *xfs_dir2_data_unused_tag_p(newdup) =
1095 cpu_to_be16((char *)newdup - (char *)hdr);
1096 xfs_dir2_data_log_unused(args, bp, newdup);
1097 newdup2 = (xfs_dir2_data_unused_t *)((char *)hdr + offset + len);
1098 newdup2->freetag = cpu_to_be16(XFS_DIR2_DATA_FREE_TAG);
1099 newdup2->length = cpu_to_be16(oldlen - len - be16_to_cpu(newdup->length));
1100 *xfs_dir2_data_unused_tag_p(newdup2) =
1101 cpu_to_be16((char *)newdup2 - (char *)hdr);
1102 xfs_dir2_data_log_unused(args, bp, newdup2);
1103 /*
1104 * If the old entry was in the table, we need to scan
1105 * if the 3rd entry was valid, since these entries
1106 * are smaller than the old one.
1107 * If we don't need to scan that means there were 1 or 2
1108 * entries in the table, and removing the old and adding
1109 * the 2 new will work.
1110 */
1111 if (dfp) {
1112 needscan = (bf[2].length != 0);
1113 if (!needscan) {
1114 xfs_dir2_data_freeremove(hdr, bf, dfp,
1115 needlogp);
1116 xfs_dir2_data_freeinsert(hdr, bf, newdup,
1117 needlogp);
1118 xfs_dir2_data_freeinsert(hdr, bf, newdup2,
1119 needlogp);
1120 }
1121 }
1122 }
1123 *needscanp = needscan;
1124 return 0;
1125corrupt:
1126 xfs_corruption_error(__func__, XFS_ERRLEVEL_LOW, args->dp->i_mount,
1127 hdr, __FILE__, __LINE__, fa);
1128 return -EFSCORRUPTED;
1129}
1130
1131/* Find the end of the entry data in a data/block format dir block. */
1132void *
1133xfs_dir3_data_endp(
1134 struct xfs_da_geometry *geo,
1135 struct xfs_dir2_data_hdr *hdr)
1136{
1137 switch (hdr->magic) {
1138 case cpu_to_be32(XFS_DIR3_BLOCK_MAGIC):
1139 case cpu_to_be32(XFS_DIR2_BLOCK_MAGIC):
1140 return xfs_dir2_block_leaf_p(xfs_dir2_block_tail_p(geo, hdr));
1141 case cpu_to_be32(XFS_DIR3_DATA_MAGIC):
1142 case cpu_to_be32(XFS_DIR2_DATA_MAGIC):
1143 return (char *)hdr + geo->blksize;
1144 default:
1145 return NULL;
1146 }
1147}