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