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