2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 #include "libxfs_priv.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_da_format.h"
26 #include "xfs_da_btree.h"
27 #include "xfs_inode.h"
28 #include "xfs_trans.h"
31 #include "xfs_dir2_priv.h"
32 #include "xfs_trace.h"
33 #include "xfs_cksum.h"
36 * Local function prototypes.
38 static void xfs_dir2_block_log_leaf(xfs_trans_t
*tp
, struct xfs_buf
*bp
,
40 static void xfs_dir2_block_log_tail(xfs_trans_t
*tp
, struct xfs_buf
*bp
);
41 static int xfs_dir2_block_lookup_int(xfs_da_args_t
*args
, struct xfs_buf
**bpp
,
43 static int xfs_dir2_block_sort(const void *a
, const void *b
);
45 static xfs_dahash_t xfs_dir_hash_dot
, xfs_dir_hash_dotdot
;
48 * One-time startup routine called from xfs_init().
53 xfs_dir_hash_dot
= xfs_da_hashname((unsigned char *)".", 1);
54 xfs_dir_hash_dotdot
= xfs_da_hashname((unsigned char *)"..", 2);
58 xfs_dir3_block_verify(
61 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
62 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
64 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
65 if (hdr3
->magic
!= cpu_to_be32(XFS_DIR3_BLOCK_MAGIC
))
66 return __this_address
;
67 if (!uuid_equal(&hdr3
->uuid
, &mp
->m_sb
.sb_meta_uuid
))
68 return __this_address
;
69 if (be64_to_cpu(hdr3
->blkno
) != bp
->b_bn
)
70 return __this_address
;
71 if (!xfs_log_check_lsn(mp
, be64_to_cpu(hdr3
->lsn
)))
72 return __this_address
;
74 if (hdr3
->magic
!= cpu_to_be32(XFS_DIR2_BLOCK_MAGIC
))
75 return __this_address
;
77 return __xfs_dir3_data_check(NULL
, bp
);
81 xfs_dir3_block_read_verify(
84 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
87 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
88 !xfs_buf_verify_cksum(bp
, XFS_DIR3_DATA_CRC_OFF
))
89 xfs_verifier_error(bp
, -EFSBADCRC
, __this_address
);
91 fa
= xfs_dir3_block_verify(bp
);
93 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
98 xfs_dir3_block_write_verify(
101 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
102 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
103 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
106 fa
= xfs_dir3_block_verify(bp
);
108 xfs_verifier_error(bp
, -EFSCORRUPTED
, fa
);
112 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
116 hdr3
->lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
118 xfs_buf_update_cksum(bp
, XFS_DIR3_DATA_CRC_OFF
);
121 const struct xfs_buf_ops xfs_dir3_block_buf_ops
= {
122 .name
= "xfs_dir3_block",
123 .verify_read
= xfs_dir3_block_read_verify
,
124 .verify_write
= xfs_dir3_block_write_verify
,
125 .verify_struct
= xfs_dir3_block_verify
,
130 struct xfs_trans
*tp
,
131 struct xfs_inode
*dp
,
132 struct xfs_buf
**bpp
)
134 struct xfs_mount
*mp
= dp
->i_mount
;
137 err
= xfs_da_read_buf(tp
, dp
, mp
->m_dir_geo
->datablk
, -1, bpp
,
138 XFS_DATA_FORK
, &xfs_dir3_block_buf_ops
);
139 if (!err
&& tp
&& *bpp
)
140 xfs_trans_buf_set_type(tp
, *bpp
, XFS_BLFT_DIR_BLOCK_BUF
);
146 struct xfs_mount
*mp
,
147 struct xfs_trans
*tp
,
149 struct xfs_inode
*dp
)
151 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
153 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
154 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DIR_BLOCK_BUF
);
156 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
157 memset(hdr3
, 0, sizeof(*hdr3
));
158 hdr3
->magic
= cpu_to_be32(XFS_DIR3_BLOCK_MAGIC
);
159 hdr3
->blkno
= cpu_to_be64(bp
->b_bn
);
160 hdr3
->owner
= cpu_to_be64(dp
->i_ino
);
161 uuid_copy(&hdr3
->uuid
, &mp
->m_sb
.sb_meta_uuid
);
165 hdr3
->magic
= cpu_to_be32(XFS_DIR2_BLOCK_MAGIC
);
169 xfs_dir2_block_need_space(
170 struct xfs_inode
*dp
,
171 struct xfs_dir2_data_hdr
*hdr
,
172 struct xfs_dir2_block_tail
*btp
,
173 struct xfs_dir2_leaf_entry
*blp
,
175 struct xfs_dir2_data_unused
**dupp
,
176 struct xfs_dir2_data_unused
**enddupp
,
180 struct xfs_dir2_data_free
*bf
;
182 struct xfs_dir2_data_unused
*dup
= NULL
;
183 struct xfs_dir2_data_unused
*enddup
= NULL
;
186 bf
= dp
->d_ops
->data_bestfree_p(hdr
);
189 * If there are stale entries we'll use one for the leaf.
192 if (be16_to_cpu(bf
[0].length
) >= len
) {
194 * The biggest entry enough to avoid compaction.
196 dup
= (xfs_dir2_data_unused_t
*)
197 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
202 * Will need to compact to make this work.
203 * Tag just before the first leaf entry.
206 tagp
= (__be16
*)blp
- 1;
208 /* Data object just before the first leaf entry. */
209 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
212 * If it's not free then the data will go where the
213 * leaf data starts now, if it works at all.
215 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
216 if (be16_to_cpu(dup
->length
) + (be32_to_cpu(btp
->stale
) - 1) *
217 (uint
)sizeof(*blp
) < len
)
219 } else if ((be32_to_cpu(btp
->stale
) - 1) * (uint
)sizeof(*blp
) < len
)
222 dup
= (xfs_dir2_data_unused_t
*)blp
;
227 * no stale entries, so just use free space.
228 * Tag just before the first leaf entry.
230 tagp
= (__be16
*)blp
- 1;
232 /* Data object just before the first leaf entry. */
233 enddup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
236 * If it's not free then can't do this add without cleaning up:
237 * the space before the first leaf entry needs to be free so it
238 * can be expanded to hold the pointer to the new entry.
240 if (be16_to_cpu(enddup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
242 * Check out the biggest freespace and see if it's the same one.
244 dup
= (xfs_dir2_data_unused_t
*)
245 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
248 * Not the same free entry, just check its length.
250 if (be16_to_cpu(dup
->length
) < len
)
256 * It is the biggest freespace, can it hold the leaf too?
258 if (be16_to_cpu(dup
->length
) < len
+ (uint
)sizeof(*blp
)) {
260 * Yes, use the second-largest entry instead if it works.
262 if (be16_to_cpu(bf
[1].length
) >= len
)
263 dup
= (xfs_dir2_data_unused_t
*)
264 ((char *)hdr
+ be16_to_cpu(bf
[1].offset
));
276 * compact the leaf entries.
277 * Leave the highest-numbered stale entry stale.
278 * XXX should be the one closest to mid but mid is not yet computed.
281 xfs_dir2_block_compact(
282 struct xfs_da_args
*args
,
284 struct xfs_dir2_data_hdr
*hdr
,
285 struct xfs_dir2_block_tail
*btp
,
286 struct xfs_dir2_leaf_entry
*blp
,
291 int fromidx
; /* source leaf index */
292 int toidx
; /* target leaf index */
294 int highstale
; /* high stale index */
296 fromidx
= toidx
= be32_to_cpu(btp
->count
) - 1;
297 highstale
= *lfloghigh
= -1;
298 for (; fromidx
>= 0; fromidx
--) {
299 if (blp
[fromidx
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
)) {
303 if (*lfloghigh
== -1)
309 blp
[toidx
] = blp
[fromidx
];
312 *lfloglow
= toidx
+ 1 - (be32_to_cpu(btp
->stale
) - 1);
313 *lfloghigh
-= be32_to_cpu(btp
->stale
) - 1;
314 be32_add_cpu(&btp
->count
, -(be32_to_cpu(btp
->stale
) - 1));
315 xfs_dir2_data_make_free(args
, bp
,
316 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)hdr
),
317 (xfs_dir2_data_aoff_t
)((be32_to_cpu(btp
->stale
) - 1) * sizeof(*blp
)),
319 btp
->stale
= cpu_to_be32(1);
321 * If we now need to rebuild the bestfree map, do so.
322 * This needs to happen before the next call to use_free.
325 xfs_dir2_data_freescan(args
->dp
, hdr
, needlog
);
329 * Add an entry to a block directory.
332 xfs_dir2_block_addname(
333 xfs_da_args_t
*args
) /* directory op arguments */
335 xfs_dir2_data_hdr_t
*hdr
; /* block header */
336 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
337 struct xfs_buf
*bp
; /* buffer for block */
338 xfs_dir2_block_tail_t
*btp
; /* block tail */
339 int compact
; /* need to compact leaf ents */
340 xfs_dir2_data_entry_t
*dep
; /* block data entry */
341 xfs_inode_t
*dp
; /* directory inode */
342 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
343 int error
; /* error return value */
344 xfs_dir2_data_unused_t
*enddup
=NULL
; /* unused at end of data */
345 xfs_dahash_t hash
; /* hash value of found entry */
346 int high
; /* high index for binary srch */
347 int highstale
; /* high stale index */
348 int lfloghigh
=0; /* last final leaf to log */
349 int lfloglow
=0; /* first final leaf to log */
350 int len
; /* length of the new entry */
351 int low
; /* low index for binary srch */
352 int lowstale
; /* low stale index */
353 int mid
=0; /* midpoint for binary srch */
354 int needlog
; /* need to log header */
355 int needscan
; /* need to rescan freespace */
356 __be16
*tagp
; /* pointer to tag value */
357 xfs_trans_t
*tp
; /* transaction structure */
359 trace_xfs_dir2_block_addname(args
);
364 /* Read the (one and only) directory block into bp. */
365 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
369 len
= dp
->d_ops
->data_entsize(args
->namelen
);
372 * Set up pointers to parts of the block.
375 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
376 blp
= xfs_dir2_block_leaf_p(btp
);
379 * Find out if we can reuse stale entries or whether we need extra
380 * space for entry and new leaf.
382 xfs_dir2_block_need_space(dp
, hdr
, btp
, blp
, &tagp
, &dup
,
383 &enddup
, &compact
, len
);
386 * Done everything we need for a space check now.
388 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
389 xfs_trans_brelse(tp
, bp
);
396 * If we don't have space for the new entry & leaf ...
399 /* Don't have a space reservation: return no-space. */
400 if (args
->total
== 0)
403 * Convert to the next larger format.
404 * Then add the new entry in that format.
406 error
= xfs_dir2_block_to_leaf(args
, bp
);
409 return xfs_dir2_leaf_addname(args
);
412 needlog
= needscan
= 0;
415 * If need to compact the leaf entries, do it now.
418 xfs_dir2_block_compact(args
, bp
, hdr
, btp
, blp
, &needlog
,
419 &lfloghigh
, &lfloglow
);
420 /* recalculate blp post-compaction */
421 blp
= xfs_dir2_block_leaf_p(btp
);
422 } else if (btp
->stale
) {
424 * Set leaf logging boundaries to impossible state.
425 * For the no-stale case they're set explicitly.
427 lfloglow
= be32_to_cpu(btp
->count
);
432 * Find the slot that's first lower than our hash value, -1 if none.
434 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; low
<= high
; ) {
435 mid
= (low
+ high
) >> 1;
436 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
438 if (hash
< args
->hashval
)
443 while (mid
>= 0 && be32_to_cpu(blp
[mid
].hashval
) >= args
->hashval
) {
447 * No stale entries, will use enddup space to hold new leaf.
451 * Mark the space needed for the new leaf entry, now in use.
453 xfs_dir2_data_use_free(args
, bp
, enddup
,
454 (xfs_dir2_data_aoff_t
)
455 ((char *)enddup
- (char *)hdr
+ be16_to_cpu(enddup
->length
) -
457 (xfs_dir2_data_aoff_t
)sizeof(*blp
),
458 &needlog
, &needscan
);
460 * Update the tail (entry count).
462 be32_add_cpu(&btp
->count
, 1);
464 * If we now need to rebuild the bestfree map, do so.
465 * This needs to happen before the next call to use_free.
468 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
472 * Adjust pointer to the first leaf entry, we're about to move
473 * the table up one to open up space for the new leaf entry.
474 * Then adjust our index to match.
479 memmove(blp
, &blp
[1], mid
* sizeof(*blp
));
484 * Use a stale leaf for our new entry.
489 blp
[lowstale
].address
!=
490 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
493 for (highstale
= mid
+ 1;
494 highstale
< be32_to_cpu(btp
->count
) &&
495 blp
[highstale
].address
!=
496 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
) &&
497 (lowstale
< 0 || mid
- lowstale
> highstale
- mid
);
501 * Move entries toward the low-numbered stale entry.
504 (highstale
== be32_to_cpu(btp
->count
) ||
505 mid
- lowstale
<= highstale
- mid
)) {
507 memmove(&blp
[lowstale
], &blp
[lowstale
+ 1],
508 (mid
- lowstale
) * sizeof(*blp
));
509 lfloglow
= MIN(lowstale
, lfloglow
);
510 lfloghigh
= MAX(mid
, lfloghigh
);
513 * Move entries toward the high-numbered stale entry.
516 ASSERT(highstale
< be32_to_cpu(btp
->count
));
519 memmove(&blp
[mid
+ 1], &blp
[mid
],
520 (highstale
- mid
) * sizeof(*blp
));
521 lfloglow
= MIN(mid
, lfloglow
);
522 lfloghigh
= MAX(highstale
, lfloghigh
);
524 be32_add_cpu(&btp
->stale
, -1);
527 * Point to the new data entry.
529 dep
= (xfs_dir2_data_entry_t
*)dup
;
531 * Fill in the leaf entry.
533 blp
[mid
].hashval
= cpu_to_be32(args
->hashval
);
534 blp
[mid
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(
535 (char *)dep
- (char *)hdr
));
536 xfs_dir2_block_log_leaf(tp
, bp
, lfloglow
, lfloghigh
);
538 * Mark space for the data entry used.
540 xfs_dir2_data_use_free(args
, bp
, dup
,
541 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
542 (xfs_dir2_data_aoff_t
)len
, &needlog
, &needscan
);
544 * Create the new data entry.
546 dep
->inumber
= cpu_to_be64(args
->inumber
);
547 dep
->namelen
= args
->namelen
;
548 memcpy(dep
->name
, args
->name
, args
->namelen
);
549 dp
->d_ops
->data_put_ftype(dep
, args
->filetype
);
550 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
551 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
553 * Clean up the bestfree array and log the header, tail, and entry.
556 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
558 xfs_dir2_data_log_header(args
, bp
);
559 xfs_dir2_block_log_tail(tp
, bp
);
560 xfs_dir2_data_log_entry(args
, bp
, dep
);
561 xfs_dir3_data_check(dp
, bp
);
566 * Log leaf entries from the block.
569 xfs_dir2_block_log_leaf(
570 xfs_trans_t
*tp
, /* transaction structure */
571 struct xfs_buf
*bp
, /* block buffer */
572 int first
, /* index of first logged leaf */
573 int last
) /* index of last logged leaf */
575 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
576 xfs_dir2_leaf_entry_t
*blp
;
577 xfs_dir2_block_tail_t
*btp
;
579 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
->m_dir_geo
, hdr
);
580 blp
= xfs_dir2_block_leaf_p(btp
);
581 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)&blp
[first
] - (char *)hdr
),
582 (uint
)((char *)&blp
[last
+ 1] - (char *)hdr
- 1));
586 * Log the block tail.
589 xfs_dir2_block_log_tail(
590 xfs_trans_t
*tp
, /* transaction structure */
591 struct xfs_buf
*bp
) /* block buffer */
593 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
594 xfs_dir2_block_tail_t
*btp
;
596 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
->m_dir_geo
, hdr
);
597 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)btp
- (char *)hdr
),
598 (uint
)((char *)(btp
+ 1) - (char *)hdr
- 1));
602 * Look up an entry in the block. This is the external routine,
603 * xfs_dir2_block_lookup_int does the real work.
606 xfs_dir2_block_lookup(
607 xfs_da_args_t
*args
) /* dir lookup arguments */
609 xfs_dir2_data_hdr_t
*hdr
; /* block header */
610 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
611 struct xfs_buf
*bp
; /* block buffer */
612 xfs_dir2_block_tail_t
*btp
; /* block tail */
613 xfs_dir2_data_entry_t
*dep
; /* block data entry */
614 xfs_inode_t
*dp
; /* incore inode */
615 int ent
; /* entry index */
616 int error
; /* error return value */
618 trace_xfs_dir2_block_lookup(args
);
621 * Get the buffer, look up the entry.
622 * If not found (ENOENT) then return, have no buffer.
624 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
)))
628 xfs_dir3_data_check(dp
, bp
);
629 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
630 blp
= xfs_dir2_block_leaf_p(btp
);
632 * Get the offset from the leaf entry, to point to the data.
634 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
635 xfs_dir2_dataptr_to_off(args
->geo
,
636 be32_to_cpu(blp
[ent
].address
)));
638 * Fill in inode number, CI name if appropriate, release the block.
640 args
->inumber
= be64_to_cpu(dep
->inumber
);
641 args
->filetype
= dp
->d_ops
->data_get_ftype(dep
);
642 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
643 xfs_trans_brelse(args
->trans
, bp
);
648 * Internal block lookup routine.
650 static int /* error */
651 xfs_dir2_block_lookup_int(
652 xfs_da_args_t
*args
, /* dir lookup arguments */
653 struct xfs_buf
**bpp
, /* returned block buffer */
654 int *entno
) /* returned entry number */
656 xfs_dir2_dataptr_t addr
; /* data entry address */
657 xfs_dir2_data_hdr_t
*hdr
; /* block header */
658 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
659 struct xfs_buf
*bp
; /* block buffer */
660 xfs_dir2_block_tail_t
*btp
; /* block tail */
661 xfs_dir2_data_entry_t
*dep
; /* block data entry */
662 xfs_inode_t
*dp
; /* incore inode */
663 int error
; /* error return value */
664 xfs_dahash_t hash
; /* found hash value */
665 int high
; /* binary search high index */
666 int low
; /* binary search low index */
667 int mid
; /* binary search current idx */
668 xfs_mount_t
*mp
; /* filesystem mount point */
669 xfs_trans_t
*tp
; /* transaction pointer */
670 enum xfs_dacmp cmp
; /* comparison result */
676 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
681 xfs_dir3_data_check(dp
, bp
);
682 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
683 blp
= xfs_dir2_block_leaf_p(btp
);
685 * Loop doing a binary search for our hash value.
686 * Find our entry, ENOENT if it's not there.
688 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; ; ) {
690 mid
= (low
+ high
) >> 1;
691 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
693 if (hash
< args
->hashval
)
698 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
699 xfs_trans_brelse(tp
, bp
);
704 * Back up to the first one with the right hash value.
706 while (mid
> 0 && be32_to_cpu(blp
[mid
- 1].hashval
) == args
->hashval
) {
710 * Now loop forward through all the entries with the
711 * right hash value looking for our name.
714 if ((addr
= be32_to_cpu(blp
[mid
].address
)) == XFS_DIR2_NULL_DATAPTR
)
717 * Get pointer to the entry from the leaf.
719 dep
= (xfs_dir2_data_entry_t
*)
720 ((char *)hdr
+ xfs_dir2_dataptr_to_off(args
->geo
, addr
));
722 * Compare name and if it's an exact match, return the index
723 * and buffer. If it's the first case-insensitive match, store
724 * the index and buffer and continue looking for an exact match.
726 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
727 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
728 args
->cmpresult
= cmp
;
731 if (cmp
== XFS_CMP_EXACT
)
734 } while (++mid
< be32_to_cpu(btp
->count
) &&
735 be32_to_cpu(blp
[mid
].hashval
) == hash
);
737 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
739 * Here, we can only be doing a lookup (not a rename or replace).
740 * If a case-insensitive match was found earlier, return success.
742 if (args
->cmpresult
== XFS_CMP_CASE
)
745 * No match, release the buffer and return ENOENT.
747 xfs_trans_brelse(tp
, bp
);
752 * Remove an entry from a block format directory.
753 * If that makes the block small enough to fit in shortform, transform it.
756 xfs_dir2_block_removename(
757 xfs_da_args_t
*args
) /* directory operation args */
759 xfs_dir2_data_hdr_t
*hdr
; /* block header */
760 xfs_dir2_leaf_entry_t
*blp
; /* block leaf pointer */
761 struct xfs_buf
*bp
; /* block buffer */
762 xfs_dir2_block_tail_t
*btp
; /* block tail */
763 xfs_dir2_data_entry_t
*dep
; /* block data entry */
764 xfs_inode_t
*dp
; /* incore inode */
765 int ent
; /* block leaf entry index */
766 int error
; /* error return value */
767 int needlog
; /* need to log block header */
768 int needscan
; /* need to fixup bestfree */
769 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
770 int size
; /* shortform size */
771 xfs_trans_t
*tp
; /* transaction pointer */
773 trace_xfs_dir2_block_removename(args
);
776 * Look up the entry in the block. Gets the buffer and entry index.
777 * It will always be there, the vnodeops level does a lookup first.
779 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
785 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
786 blp
= xfs_dir2_block_leaf_p(btp
);
788 * Point to the data entry using the leaf entry.
790 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
791 xfs_dir2_dataptr_to_off(args
->geo
,
792 be32_to_cpu(blp
[ent
].address
)));
794 * Mark the data entry's space free.
796 needlog
= needscan
= 0;
797 xfs_dir2_data_make_free(args
, bp
,
798 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)hdr
),
799 dp
->d_ops
->data_entsize(dep
->namelen
), &needlog
, &needscan
);
801 * Fix up the block tail.
803 be32_add_cpu(&btp
->stale
, 1);
804 xfs_dir2_block_log_tail(tp
, bp
);
806 * Remove the leaf entry by marking it stale.
808 blp
[ent
].address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
809 xfs_dir2_block_log_leaf(tp
, bp
, ent
, ent
);
811 * Fix up bestfree, log the header if necessary.
814 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
816 xfs_dir2_data_log_header(args
, bp
);
817 xfs_dir3_data_check(dp
, bp
);
819 * See if the size as a shortform is good enough.
821 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
822 if (size
> XFS_IFORK_DSIZE(dp
))
826 * If it works, do the conversion.
828 return xfs_dir2_block_to_sf(args
, bp
, size
, &sfh
);
832 * Replace an entry in a V2 block directory.
833 * Change the inode number to the new value.
836 xfs_dir2_block_replace(
837 xfs_da_args_t
*args
) /* directory operation args */
839 xfs_dir2_data_hdr_t
*hdr
; /* block header */
840 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
841 struct xfs_buf
*bp
; /* block buffer */
842 xfs_dir2_block_tail_t
*btp
; /* block tail */
843 xfs_dir2_data_entry_t
*dep
; /* block data entry */
844 xfs_inode_t
*dp
; /* incore inode */
845 int ent
; /* leaf entry index */
846 int error
; /* error return value */
848 trace_xfs_dir2_block_replace(args
);
851 * Lookup the entry in the directory. Get buffer and entry index.
852 * This will always succeed since the caller has already done a lookup.
854 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
859 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
860 blp
= xfs_dir2_block_leaf_p(btp
);
862 * Point to the data entry we need to change.
864 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
865 xfs_dir2_dataptr_to_off(args
->geo
,
866 be32_to_cpu(blp
[ent
].address
)));
867 ASSERT(be64_to_cpu(dep
->inumber
) != args
->inumber
);
869 * Change the inode number to the new value.
871 dep
->inumber
= cpu_to_be64(args
->inumber
);
872 dp
->d_ops
->data_put_ftype(dep
, args
->filetype
);
873 xfs_dir2_data_log_entry(args
, bp
, dep
);
874 xfs_dir3_data_check(dp
, bp
);
879 * Qsort comparison routine for the block leaf entries.
881 static int /* sort order */
883 const void *a
, /* first leaf entry */
884 const void *b
) /* second leaf entry */
886 const xfs_dir2_leaf_entry_t
*la
; /* first leaf entry */
887 const xfs_dir2_leaf_entry_t
*lb
; /* second leaf entry */
891 return be32_to_cpu(la
->hashval
) < be32_to_cpu(lb
->hashval
) ? -1 :
892 (be32_to_cpu(la
->hashval
) > be32_to_cpu(lb
->hashval
) ? 1 : 0);
896 * Convert a V2 leaf directory to a V2 block directory if possible.
899 xfs_dir2_leaf_to_block(
900 xfs_da_args_t
*args
, /* operation arguments */
901 struct xfs_buf
*lbp
, /* leaf buffer */
902 struct xfs_buf
*dbp
) /* data buffer */
904 __be16
*bestsp
; /* leaf bests table */
905 xfs_dir2_data_hdr_t
*hdr
; /* block header */
906 xfs_dir2_block_tail_t
*btp
; /* block tail */
907 xfs_inode_t
*dp
; /* incore directory inode */
908 xfs_dir2_data_unused_t
*dup
; /* unused data entry */
909 int error
; /* error return value */
910 int from
; /* leaf from index */
911 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
912 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
913 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
914 xfs_mount_t
*mp
; /* file system mount point */
915 int needlog
; /* need to log data header */
916 int needscan
; /* need to scan for bestfree */
917 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
918 int size
; /* bytes used */
919 __be16
*tagp
; /* end of entry (tag) */
920 int to
; /* block/leaf to index */
921 xfs_trans_t
*tp
; /* transaction pointer */
922 struct xfs_dir2_leaf_entry
*ents
;
923 struct xfs_dir3_icleaf_hdr leafhdr
;
925 trace_xfs_dir2_leaf_to_block(args
);
931 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
932 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
933 ltp
= xfs_dir2_leaf_tail_p(args
->geo
, leaf
);
935 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAF1_MAGIC
||
936 leafhdr
.magic
== XFS_DIR3_LEAF1_MAGIC
);
938 * If there are data blocks other than the first one, take this
939 * opportunity to remove trailing empty data blocks that may have
940 * been left behind during no-space-reservation operations.
941 * These will show up in the leaf bests table.
943 while (dp
->i_d
.di_size
> args
->geo
->blksize
) {
946 hdrsz
= dp
->d_ops
->data_entry_offset
;
947 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
948 if (be16_to_cpu(bestsp
[be32_to_cpu(ltp
->bestcount
) - 1]) ==
949 args
->geo
->blksize
- hdrsz
) {
951 xfs_dir2_leaf_trim_data(args
, lbp
,
952 (xfs_dir2_db_t
)(be32_to_cpu(ltp
->bestcount
) - 1))))
958 * Read the data block if we don't already have it, give up if it fails.
961 error
= xfs_dir3_data_read(tp
, dp
, args
->geo
->datablk
, -1, &dbp
);
966 ASSERT(hdr
->magic
== cpu_to_be32(XFS_DIR2_DATA_MAGIC
) ||
967 hdr
->magic
== cpu_to_be32(XFS_DIR3_DATA_MAGIC
));
970 * Size of the "leaf" area in the block.
972 size
= (uint
)sizeof(xfs_dir2_block_tail_t
) +
973 (uint
)sizeof(*lep
) * (leafhdr
.count
- leafhdr
.stale
);
975 * Look at the last data entry.
977 tagp
= (__be16
*)((char *)hdr
+ args
->geo
->blksize
) - 1;
978 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
980 * If it's not free or is too short we can't do it.
982 if (be16_to_cpu(dup
->freetag
) != XFS_DIR2_DATA_FREE_TAG
||
983 be16_to_cpu(dup
->length
) < size
)
987 * Start converting it to block form.
989 xfs_dir3_block_init(mp
, tp
, dbp
, dp
);
994 * Use up the space at the end of the block (blp/btp).
996 xfs_dir2_data_use_free(args
, dbp
, dup
, args
->geo
->blksize
- size
, size
,
997 &needlog
, &needscan
);
999 * Initialize the block tail.
1001 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
1002 btp
->count
= cpu_to_be32(leafhdr
.count
- leafhdr
.stale
);
1004 xfs_dir2_block_log_tail(tp
, dbp
);
1006 * Initialize the block leaf area. We compact out stale entries.
1008 lep
= xfs_dir2_block_leaf_p(btp
);
1009 for (from
= to
= 0; from
< leafhdr
.count
; from
++) {
1010 if (ents
[from
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
1012 lep
[to
++] = ents
[from
];
1014 ASSERT(to
== be32_to_cpu(btp
->count
));
1015 xfs_dir2_block_log_leaf(tp
, dbp
, 0, be32_to_cpu(btp
->count
) - 1);
1017 * Scan the bestfree if we need it and log the data block header.
1020 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
1022 xfs_dir2_data_log_header(args
, dbp
);
1024 * Pitch the old leaf block.
1026 error
= xfs_da_shrink_inode(args
, args
->geo
->leafblk
, lbp
);
1031 * Now see if the resulting block can be shrunken to shortform.
1033 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
1034 if (size
> XFS_IFORK_DSIZE(dp
))
1037 return xfs_dir2_block_to_sf(args
, dbp
, size
, &sfh
);
1041 * Convert the shortform directory to block form.
1044 xfs_dir2_sf_to_block(
1045 xfs_da_args_t
*args
) /* operation arguments */
1047 xfs_dir2_db_t blkno
; /* dir-relative block # (0) */
1048 xfs_dir2_data_hdr_t
*hdr
; /* block header */
1049 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
1050 struct xfs_buf
*bp
; /* block buffer */
1051 xfs_dir2_block_tail_t
*btp
; /* block tail pointer */
1052 xfs_dir2_data_entry_t
*dep
; /* data entry pointer */
1053 xfs_inode_t
*dp
; /* incore directory inode */
1054 int dummy
; /* trash */
1055 xfs_dir2_data_unused_t
*dup
; /* unused entry pointer */
1056 int endoffset
; /* end of data objects */
1057 int error
; /* error return value */
1059 xfs_mount_t
*mp
; /* filesystem mount point */
1060 int needlog
; /* need to log block header */
1061 int needscan
; /* need to scan block freespc */
1062 int newoffset
; /* offset from current entry */
1063 int offset
; /* target block offset */
1064 xfs_dir2_sf_entry_t
*sfep
; /* sf entry pointer */
1065 xfs_dir2_sf_hdr_t
*oldsfp
; /* old shortform header */
1066 xfs_dir2_sf_hdr_t
*sfp
; /* shortform header */
1067 __be16
*tagp
; /* end of data entry */
1068 xfs_trans_t
*tp
; /* transaction pointer */
1069 struct xfs_name name
;
1070 struct xfs_ifork
*ifp
;
1072 trace_xfs_dir2_sf_to_block(args
);
1077 ifp
= XFS_IFORK_PTR(dp
, XFS_DATA_FORK
);
1078 ASSERT(ifp
->if_flags
& XFS_IFINLINE
);
1080 * Bomb out if the shortform directory is way too short.
1082 if (dp
->i_d
.di_size
< offsetof(xfs_dir2_sf_hdr_t
, parent
)) {
1083 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1087 oldsfp
= (xfs_dir2_sf_hdr_t
*)ifp
->if_u1
.if_data
;
1089 ASSERT(ifp
->if_bytes
== dp
->i_d
.di_size
);
1090 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
1091 ASSERT(dp
->i_d
.di_size
>= xfs_dir2_sf_hdr_size(oldsfp
->i8count
));
1092 ASSERT(dp
->i_d
.di_nextents
== 0);
1095 * Copy the directory into a temporary buffer.
1096 * Then pitch the incore inode data so we can make extents.
1098 sfp
= kmem_alloc(ifp
->if_bytes
, KM_SLEEP
);
1099 memcpy(sfp
, oldsfp
, ifp
->if_bytes
);
1101 xfs_idata_realloc(dp
, -ifp
->if_bytes
, XFS_DATA_FORK
);
1102 xfs_bmap_local_to_extents_empty(dp
, XFS_DATA_FORK
);
1103 dp
->i_d
.di_size
= 0;
1106 * Add block 0 to the inode.
1108 error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
, &blkno
);
1114 * Initialize the data block, then convert it to block format.
1116 error
= xfs_dir3_data_init(args
, blkno
, &bp
);
1121 xfs_dir3_block_init(mp
, tp
, bp
, dp
);
1125 * Compute size of block "tail" area.
1127 i
= (uint
)sizeof(*btp
) +
1128 (sfp
->count
+ 2) * (uint
)sizeof(xfs_dir2_leaf_entry_t
);
1130 * The whole thing is initialized to free by the init routine.
1131 * Say we're using the leaf and tail area.
1133 dup
= dp
->d_ops
->data_unused_p(hdr
);
1134 needlog
= needscan
= 0;
1135 xfs_dir2_data_use_free(args
, bp
, dup
, args
->geo
->blksize
- i
,
1136 i
, &needlog
, &needscan
);
1137 ASSERT(needscan
== 0);
1141 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
1142 btp
->count
= cpu_to_be32(sfp
->count
+ 2); /* ., .. */
1144 blp
= xfs_dir2_block_leaf_p(btp
);
1145 endoffset
= (uint
)((char *)blp
- (char *)hdr
);
1147 * Remove the freespace, we'll manage it.
1149 xfs_dir2_data_use_free(args
, bp
, dup
,
1150 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
1151 be16_to_cpu(dup
->length
), &needlog
, &needscan
);
1153 * Create entry for .
1155 dep
= dp
->d_ops
->data_dot_entry_p(hdr
);
1156 dep
->inumber
= cpu_to_be64(dp
->i_ino
);
1159 dp
->d_ops
->data_put_ftype(dep
, XFS_DIR3_FT_DIR
);
1160 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
1161 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1162 xfs_dir2_data_log_entry(args
, bp
, dep
);
1163 blp
[0].hashval
= cpu_to_be32(xfs_dir_hash_dot
);
1164 blp
[0].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(
1165 (char *)dep
- (char *)hdr
));
1167 * Create entry for ..
1169 dep
= dp
->d_ops
->data_dotdot_entry_p(hdr
);
1170 dep
->inumber
= cpu_to_be64(dp
->d_ops
->sf_get_parent_ino(sfp
));
1172 dep
->name
[0] = dep
->name
[1] = '.';
1173 dp
->d_ops
->data_put_ftype(dep
, XFS_DIR3_FT_DIR
);
1174 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
1175 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1176 xfs_dir2_data_log_entry(args
, bp
, dep
);
1177 blp
[1].hashval
= cpu_to_be32(xfs_dir_hash_dotdot
);
1178 blp
[1].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(
1179 (char *)dep
- (char *)hdr
));
1180 offset
= dp
->d_ops
->data_first_offset
;
1182 * Loop over existing entries, stuff them in.
1188 sfep
= xfs_dir2_sf_firstentry(sfp
);
1190 * Need to preserve the existing offset values in the sf directory.
1191 * Insert holes (unused entries) where necessary.
1193 while (offset
< endoffset
) {
1195 * sfep is null when we reach the end of the list.
1198 newoffset
= endoffset
;
1200 newoffset
= xfs_dir2_sf_get_offset(sfep
);
1202 * There should be a hole here, make one.
1204 if (offset
< newoffset
) {
1205 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ offset
);
1206 dup
->freetag
= cpu_to_be16(XFS_DIR2_DATA_FREE_TAG
);
1207 dup
->length
= cpu_to_be16(newoffset
- offset
);
1208 *xfs_dir2_data_unused_tag_p(dup
) = cpu_to_be16(
1209 ((char *)dup
- (char *)hdr
));
1210 xfs_dir2_data_log_unused(args
, bp
, dup
);
1211 xfs_dir2_data_freeinsert(hdr
,
1212 dp
->d_ops
->data_bestfree_p(hdr
),
1214 offset
+= be16_to_cpu(dup
->length
);
1218 * Copy a real entry.
1220 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+ newoffset
);
1221 dep
->inumber
= cpu_to_be64(dp
->d_ops
->sf_get_ino(sfp
, sfep
));
1222 dep
->namelen
= sfep
->namelen
;
1223 dp
->d_ops
->data_put_ftype(dep
, dp
->d_ops
->sf_get_ftype(sfep
));
1224 memcpy(dep
->name
, sfep
->name
, dep
->namelen
);
1225 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
1226 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1227 xfs_dir2_data_log_entry(args
, bp
, dep
);
1228 name
.name
= sfep
->name
;
1229 name
.len
= sfep
->namelen
;
1230 blp
[2 + i
].hashval
= cpu_to_be32(mp
->m_dirnameops
->
1232 blp
[2 + i
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(
1233 (char *)dep
- (char *)hdr
));
1234 offset
= (int)((char *)(tagp
+ 1) - (char *)hdr
);
1235 if (++i
== sfp
->count
)
1238 sfep
= dp
->d_ops
->sf_nextentry(sfp
, sfep
);
1240 /* Done with the temporary buffer */
1243 * Sort the leaf entries by hash value.
1245 xfs_sort(blp
, be32_to_cpu(btp
->count
), sizeof(*blp
), xfs_dir2_block_sort
);
1247 * Log the leaf entry area and tail.
1248 * Already logged the header in data_init, ignore needlog.
1250 ASSERT(needscan
== 0);
1251 xfs_dir2_block_log_leaf(tp
, bp
, 0, be32_to_cpu(btp
->count
) - 1);
1252 xfs_dir2_block_log_tail(tp
, bp
);
1253 xfs_dir3_data_check(dp
, bp
);