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
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"
29 #include "xfs_inode_item.h"
31 #include "xfs_buf_item.h"
33 #include "xfs_dir2_priv.h"
34 #include "xfs_error.h"
35 #include "xfs_trace.h"
36 #include "xfs_cksum.h"
40 * Local function prototypes.
42 static void xfs_dir2_block_log_leaf(xfs_trans_t
*tp
, struct xfs_buf
*bp
,
44 static void xfs_dir2_block_log_tail(xfs_trans_t
*tp
, struct xfs_buf
*bp
);
45 static int xfs_dir2_block_lookup_int(xfs_da_args_t
*args
, struct xfs_buf
**bpp
,
47 static int xfs_dir2_block_sort(const void *a
, const void *b
);
49 static xfs_dahash_t xfs_dir_hash_dot
, xfs_dir_hash_dotdot
;
52 * One-time startup routine called from xfs_init().
57 xfs_dir_hash_dot
= xfs_da_hashname((unsigned char *)".", 1);
58 xfs_dir_hash_dotdot
= xfs_da_hashname((unsigned char *)"..", 2);
62 xfs_dir3_block_verify(
65 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
66 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
68 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
69 if (hdr3
->magic
!= cpu_to_be32(XFS_DIR3_BLOCK_MAGIC
))
71 if (!uuid_equal(&hdr3
->uuid
, &mp
->m_sb
.sb_meta_uuid
))
73 if (be64_to_cpu(hdr3
->blkno
) != bp
->b_bn
)
75 if (!xfs_log_check_lsn(mp
, be64_to_cpu(hdr3
->lsn
)))
78 if (hdr3
->magic
!= cpu_to_be32(XFS_DIR2_BLOCK_MAGIC
))
81 if (__xfs_dir3_data_check(NULL
, bp
))
87 xfs_dir3_block_read_verify(
90 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
92 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
93 !xfs_buf_verify_cksum(bp
, XFS_DIR3_DATA_CRC_OFF
))
94 xfs_buf_ioerror(bp
, -EFSBADCRC
);
95 else if (!xfs_dir3_block_verify(bp
))
96 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
99 xfs_verifier_error(bp
);
103 xfs_dir3_block_write_verify(
106 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
107 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
108 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
110 if (!xfs_dir3_block_verify(bp
)) {
111 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
112 xfs_verifier_error(bp
);
116 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
120 hdr3
->lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
122 xfs_buf_update_cksum(bp
, XFS_DIR3_DATA_CRC_OFF
);
125 const struct xfs_buf_ops xfs_dir3_block_buf_ops
= {
126 .verify_read
= xfs_dir3_block_read_verify
,
127 .verify_write
= xfs_dir3_block_write_verify
,
132 struct xfs_trans
*tp
,
133 struct xfs_inode
*dp
,
134 struct xfs_buf
**bpp
)
136 struct xfs_mount
*mp
= dp
->i_mount
;
139 err
= xfs_da_read_buf(tp
, dp
, mp
->m_dir_geo
->datablk
, -1, bpp
,
140 XFS_DATA_FORK
, &xfs_dir3_block_buf_ops
);
142 xfs_trans_buf_set_type(tp
, *bpp
, XFS_BLFT_DIR_BLOCK_BUF
);
148 struct xfs_mount
*mp
,
149 struct xfs_trans
*tp
,
151 struct xfs_inode
*dp
)
153 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
155 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
156 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DIR_BLOCK_BUF
);
158 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
159 memset(hdr3
, 0, sizeof(*hdr3
));
160 hdr3
->magic
= cpu_to_be32(XFS_DIR3_BLOCK_MAGIC
);
161 hdr3
->blkno
= cpu_to_be64(bp
->b_bn
);
162 hdr3
->owner
= cpu_to_be64(dp
->i_ino
);
163 uuid_copy(&hdr3
->uuid
, &mp
->m_sb
.sb_meta_uuid
);
167 hdr3
->magic
= cpu_to_be32(XFS_DIR2_BLOCK_MAGIC
);
171 xfs_dir2_block_need_space(
172 struct xfs_inode
*dp
,
173 struct xfs_dir2_data_hdr
*hdr
,
174 struct xfs_dir2_block_tail
*btp
,
175 struct xfs_dir2_leaf_entry
*blp
,
177 struct xfs_dir2_data_unused
**dupp
,
178 struct xfs_dir2_data_unused
**enddupp
,
182 struct xfs_dir2_data_free
*bf
;
184 struct xfs_dir2_data_unused
*dup
= NULL
;
185 struct xfs_dir2_data_unused
*enddup
= NULL
;
188 bf
= dp
->d_ops
->data_bestfree_p(hdr
);
191 * If there are stale entries we'll use one for the leaf.
194 if (be16_to_cpu(bf
[0].length
) >= len
) {
196 * The biggest entry enough to avoid compaction.
198 dup
= (xfs_dir2_data_unused_t
*)
199 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
204 * Will need to compact to make this work.
205 * Tag just before the first leaf entry.
208 tagp
= (__be16
*)blp
- 1;
210 /* Data object just before the first leaf entry. */
211 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
214 * If it's not free then the data will go where the
215 * leaf data starts now, if it works at all.
217 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
218 if (be16_to_cpu(dup
->length
) + (be32_to_cpu(btp
->stale
) - 1) *
219 (uint
)sizeof(*blp
) < len
)
221 } else if ((be32_to_cpu(btp
->stale
) - 1) * (uint
)sizeof(*blp
) < len
)
224 dup
= (xfs_dir2_data_unused_t
*)blp
;
229 * no stale entries, so just use free space.
230 * Tag just before the first leaf entry.
232 tagp
= (__be16
*)blp
- 1;
234 /* Data object just before the first leaf entry. */
235 enddup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
238 * If it's not free then can't do this add without cleaning up:
239 * the space before the first leaf entry needs to be free so it
240 * can be expanded to hold the pointer to the new entry.
242 if (be16_to_cpu(enddup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
244 * Check out the biggest freespace and see if it's the same one.
246 dup
= (xfs_dir2_data_unused_t
*)
247 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
250 * Not the same free entry, just check its length.
252 if (be16_to_cpu(dup
->length
) < len
)
258 * It is the biggest freespace, can it hold the leaf too?
260 if (be16_to_cpu(dup
->length
) < len
+ (uint
)sizeof(*blp
)) {
262 * Yes, use the second-largest entry instead if it works.
264 if (be16_to_cpu(bf
[1].length
) >= len
)
265 dup
= (xfs_dir2_data_unused_t
*)
266 ((char *)hdr
+ be16_to_cpu(bf
[1].offset
));
278 * compact the leaf entries.
279 * Leave the highest-numbered stale entry stale.
280 * XXX should be the one closest to mid but mid is not yet computed.
283 xfs_dir2_block_compact(
284 struct xfs_da_args
*args
,
286 struct xfs_dir2_data_hdr
*hdr
,
287 struct xfs_dir2_block_tail
*btp
,
288 struct xfs_dir2_leaf_entry
*blp
,
293 int fromidx
; /* source leaf index */
294 int toidx
; /* target leaf index */
296 int highstale
; /* high stale index */
298 fromidx
= toidx
= be32_to_cpu(btp
->count
) - 1;
299 highstale
= *lfloghigh
= -1;
300 for (; fromidx
>= 0; fromidx
--) {
301 if (blp
[fromidx
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
)) {
305 if (*lfloghigh
== -1)
311 blp
[toidx
] = blp
[fromidx
];
314 *lfloglow
= toidx
+ 1 - (be32_to_cpu(btp
->stale
) - 1);
315 *lfloghigh
-= be32_to_cpu(btp
->stale
) - 1;
316 be32_add_cpu(&btp
->count
, -(be32_to_cpu(btp
->stale
) - 1));
317 xfs_dir2_data_make_free(args
, bp
,
318 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)hdr
),
319 (xfs_dir2_data_aoff_t
)((be32_to_cpu(btp
->stale
) - 1) * sizeof(*blp
)),
321 btp
->stale
= cpu_to_be32(1);
323 * If we now need to rebuild the bestfree map, do so.
324 * This needs to happen before the next call to use_free.
327 xfs_dir2_data_freescan(args
->dp
, hdr
, needlog
);
331 * Add an entry to a block directory.
334 xfs_dir2_block_addname(
335 xfs_da_args_t
*args
) /* directory op arguments */
337 xfs_dir2_data_hdr_t
*hdr
; /* block header */
338 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
339 struct xfs_buf
*bp
; /* buffer for block */
340 xfs_dir2_block_tail_t
*btp
; /* block tail */
341 int compact
; /* need to compact leaf ents */
342 xfs_dir2_data_entry_t
*dep
; /* block data entry */
343 xfs_inode_t
*dp
; /* directory inode */
344 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
345 int error
; /* error return value */
346 xfs_dir2_data_unused_t
*enddup
=NULL
; /* unused at end of data */
347 xfs_dahash_t hash
; /* hash value of found entry */
348 int high
; /* high index for binary srch */
349 int highstale
; /* high stale index */
350 int lfloghigh
=0; /* last final leaf to log */
351 int lfloglow
=0; /* first final leaf to log */
352 int len
; /* length of the new entry */
353 int low
; /* low index for binary srch */
354 int lowstale
; /* low stale index */
355 int mid
=0; /* midpoint for binary srch */
356 int needlog
; /* need to log header */
357 int needscan
; /* need to rescan freespace */
358 __be16
*tagp
; /* pointer to tag value */
359 xfs_trans_t
*tp
; /* transaction structure */
361 trace_xfs_dir2_block_addname(args
);
366 /* Read the (one and only) directory block into bp. */
367 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
371 len
= dp
->d_ops
->data_entsize(args
->namelen
);
374 * Set up pointers to parts of the block.
377 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
378 blp
= xfs_dir2_block_leaf_p(btp
);
381 * Find out if we can reuse stale entries or whether we need extra
382 * space for entry and new leaf.
384 xfs_dir2_block_need_space(dp
, hdr
, btp
, blp
, &tagp
, &dup
,
385 &enddup
, &compact
, len
);
388 * Done everything we need for a space check now.
390 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
391 xfs_trans_brelse(tp
, bp
);
398 * If we don't have space for the new entry & leaf ...
401 /* Don't have a space reservation: return no-space. */
402 if (args
->total
== 0)
405 * Convert to the next larger format.
406 * Then add the new entry in that format.
408 error
= xfs_dir2_block_to_leaf(args
, bp
);
411 return xfs_dir2_leaf_addname(args
);
414 needlog
= needscan
= 0;
417 * If need to compact the leaf entries, do it now.
420 xfs_dir2_block_compact(args
, bp
, hdr
, btp
, blp
, &needlog
,
421 &lfloghigh
, &lfloglow
);
422 /* recalculate blp post-compaction */
423 blp
= xfs_dir2_block_leaf_p(btp
);
424 } else if (btp
->stale
) {
426 * Set leaf logging boundaries to impossible state.
427 * For the no-stale case they're set explicitly.
429 lfloglow
= be32_to_cpu(btp
->count
);
434 * Find the slot that's first lower than our hash value, -1 if none.
436 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; low
<= high
; ) {
437 mid
= (low
+ high
) >> 1;
438 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
440 if (hash
< args
->hashval
)
445 while (mid
>= 0 && be32_to_cpu(blp
[mid
].hashval
) >= args
->hashval
) {
449 * No stale entries, will use enddup space to hold new leaf.
453 * Mark the space needed for the new leaf entry, now in use.
455 xfs_dir2_data_use_free(args
, bp
, enddup
,
456 (xfs_dir2_data_aoff_t
)
457 ((char *)enddup
- (char *)hdr
+ be16_to_cpu(enddup
->length
) -
459 (xfs_dir2_data_aoff_t
)sizeof(*blp
),
460 &needlog
, &needscan
);
462 * Update the tail (entry count).
464 be32_add_cpu(&btp
->count
, 1);
466 * If we now need to rebuild the bestfree map, do so.
467 * This needs to happen before the next call to use_free.
470 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
474 * Adjust pointer to the first leaf entry, we're about to move
475 * the table up one to open up space for the new leaf entry.
476 * Then adjust our index to match.
481 memmove(blp
, &blp
[1], mid
* sizeof(*blp
));
486 * Use a stale leaf for our new entry.
491 blp
[lowstale
].address
!=
492 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
495 for (highstale
= mid
+ 1;
496 highstale
< be32_to_cpu(btp
->count
) &&
497 blp
[highstale
].address
!=
498 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
) &&
499 (lowstale
< 0 || mid
- lowstale
> highstale
- mid
);
503 * Move entries toward the low-numbered stale entry.
506 (highstale
== be32_to_cpu(btp
->count
) ||
507 mid
- lowstale
<= highstale
- mid
)) {
509 memmove(&blp
[lowstale
], &blp
[lowstale
+ 1],
510 (mid
- lowstale
) * sizeof(*blp
));
511 lfloglow
= MIN(lowstale
, lfloglow
);
512 lfloghigh
= MAX(mid
, lfloghigh
);
515 * Move entries toward the high-numbered stale entry.
518 ASSERT(highstale
< be32_to_cpu(btp
->count
));
521 memmove(&blp
[mid
+ 1], &blp
[mid
],
522 (highstale
- mid
) * sizeof(*blp
));
523 lfloglow
= MIN(mid
, lfloglow
);
524 lfloghigh
= MAX(highstale
, lfloghigh
);
526 be32_add_cpu(&btp
->stale
, -1);
529 * Point to the new data entry.
531 dep
= (xfs_dir2_data_entry_t
*)dup
;
533 * Fill in the leaf entry.
535 blp
[mid
].hashval
= cpu_to_be32(args
->hashval
);
536 blp
[mid
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(
537 (char *)dep
- (char *)hdr
));
538 xfs_dir2_block_log_leaf(tp
, bp
, lfloglow
, lfloghigh
);
540 * Mark space for the data entry used.
542 xfs_dir2_data_use_free(args
, bp
, dup
,
543 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
544 (xfs_dir2_data_aoff_t
)len
, &needlog
, &needscan
);
546 * Create the new data entry.
548 dep
->inumber
= cpu_to_be64(args
->inumber
);
549 dep
->namelen
= args
->namelen
;
550 memcpy(dep
->name
, args
->name
, args
->namelen
);
551 dp
->d_ops
->data_put_ftype(dep
, args
->filetype
);
552 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
553 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
555 * Clean up the bestfree array and log the header, tail, and entry.
558 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
560 xfs_dir2_data_log_header(args
, bp
);
561 xfs_dir2_block_log_tail(tp
, bp
);
562 xfs_dir2_data_log_entry(args
, bp
, dep
);
563 xfs_dir3_data_check(dp
, bp
);
568 * Log leaf entries from the block.
571 xfs_dir2_block_log_leaf(
572 xfs_trans_t
*tp
, /* transaction structure */
573 struct xfs_buf
*bp
, /* block buffer */
574 int first
, /* index of first logged leaf */
575 int last
) /* index of last logged leaf */
577 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
578 xfs_dir2_leaf_entry_t
*blp
;
579 xfs_dir2_block_tail_t
*btp
;
581 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
->m_dir_geo
, hdr
);
582 blp
= xfs_dir2_block_leaf_p(btp
);
583 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)&blp
[first
] - (char *)hdr
),
584 (uint
)((char *)&blp
[last
+ 1] - (char *)hdr
- 1));
588 * Log the block tail.
591 xfs_dir2_block_log_tail(
592 xfs_trans_t
*tp
, /* transaction structure */
593 struct xfs_buf
*bp
) /* block buffer */
595 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
596 xfs_dir2_block_tail_t
*btp
;
598 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
->m_dir_geo
, hdr
);
599 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)btp
- (char *)hdr
),
600 (uint
)((char *)(btp
+ 1) - (char *)hdr
- 1));
604 * Look up an entry in the block. This is the external routine,
605 * xfs_dir2_block_lookup_int does the real work.
608 xfs_dir2_block_lookup(
609 xfs_da_args_t
*args
) /* dir lookup arguments */
611 xfs_dir2_data_hdr_t
*hdr
; /* block header */
612 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
613 struct xfs_buf
*bp
; /* block buffer */
614 xfs_dir2_block_tail_t
*btp
; /* block tail */
615 xfs_dir2_data_entry_t
*dep
; /* block data entry */
616 xfs_inode_t
*dp
; /* incore inode */
617 int ent
; /* entry index */
618 int error
; /* error return value */
620 trace_xfs_dir2_block_lookup(args
);
623 * Get the buffer, look up the entry.
624 * If not found (ENOENT) then return, have no buffer.
626 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
)))
630 xfs_dir3_data_check(dp
, bp
);
631 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
632 blp
= xfs_dir2_block_leaf_p(btp
);
634 * Get the offset from the leaf entry, to point to the data.
636 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
637 xfs_dir2_dataptr_to_off(args
->geo
,
638 be32_to_cpu(blp
[ent
].address
)));
640 * Fill in inode number, CI name if appropriate, release the block.
642 args
->inumber
= be64_to_cpu(dep
->inumber
);
643 args
->filetype
= dp
->d_ops
->data_get_ftype(dep
);
644 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
645 xfs_trans_brelse(args
->trans
, bp
);
650 * Internal block lookup routine.
652 static int /* error */
653 xfs_dir2_block_lookup_int(
654 xfs_da_args_t
*args
, /* dir lookup arguments */
655 struct xfs_buf
**bpp
, /* returned block buffer */
656 int *entno
) /* returned entry number */
658 xfs_dir2_dataptr_t addr
; /* data entry address */
659 xfs_dir2_data_hdr_t
*hdr
; /* block header */
660 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
661 struct xfs_buf
*bp
; /* block buffer */
662 xfs_dir2_block_tail_t
*btp
; /* block tail */
663 xfs_dir2_data_entry_t
*dep
; /* block data entry */
664 xfs_inode_t
*dp
; /* incore inode */
665 int error
; /* error return value */
666 xfs_dahash_t hash
; /* found hash value */
667 int high
; /* binary search high index */
668 int low
; /* binary search low index */
669 int mid
; /* binary search current idx */
670 xfs_mount_t
*mp
; /* filesystem mount point */
671 xfs_trans_t
*tp
; /* transaction pointer */
672 enum xfs_dacmp cmp
; /* comparison result */
678 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
683 xfs_dir3_data_check(dp
, bp
);
684 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
685 blp
= xfs_dir2_block_leaf_p(btp
);
687 * Loop doing a binary search for our hash value.
688 * Find our entry, ENOENT if it's not there.
690 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; ; ) {
692 mid
= (low
+ high
) >> 1;
693 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
695 if (hash
< args
->hashval
)
700 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
701 xfs_trans_brelse(tp
, bp
);
706 * Back up to the first one with the right hash value.
708 while (mid
> 0 && be32_to_cpu(blp
[mid
- 1].hashval
) == args
->hashval
) {
712 * Now loop forward through all the entries with the
713 * right hash value looking for our name.
716 if ((addr
= be32_to_cpu(blp
[mid
].address
)) == XFS_DIR2_NULL_DATAPTR
)
719 * Get pointer to the entry from the leaf.
721 dep
= (xfs_dir2_data_entry_t
*)
722 ((char *)hdr
+ xfs_dir2_dataptr_to_off(args
->geo
, addr
));
724 * Compare name and if it's an exact match, return the index
725 * and buffer. If it's the first case-insensitive match, store
726 * the index and buffer and continue looking for an exact match.
728 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
729 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
730 args
->cmpresult
= cmp
;
733 if (cmp
== XFS_CMP_EXACT
)
736 } while (++mid
< be32_to_cpu(btp
->count
) &&
737 be32_to_cpu(blp
[mid
].hashval
) == hash
);
739 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
741 * Here, we can only be doing a lookup (not a rename or replace).
742 * If a case-insensitive match was found earlier, return success.
744 if (args
->cmpresult
== XFS_CMP_CASE
)
747 * No match, release the buffer and return ENOENT.
749 xfs_trans_brelse(tp
, bp
);
754 * Remove an entry from a block format directory.
755 * If that makes the block small enough to fit in shortform, transform it.
758 xfs_dir2_block_removename(
759 xfs_da_args_t
*args
) /* directory operation args */
761 xfs_dir2_data_hdr_t
*hdr
; /* block header */
762 xfs_dir2_leaf_entry_t
*blp
; /* block leaf pointer */
763 struct xfs_buf
*bp
; /* block buffer */
764 xfs_dir2_block_tail_t
*btp
; /* block tail */
765 xfs_dir2_data_entry_t
*dep
; /* block data entry */
766 xfs_inode_t
*dp
; /* incore inode */
767 int ent
; /* block leaf entry index */
768 int error
; /* error return value */
769 int needlog
; /* need to log block header */
770 int needscan
; /* need to fixup bestfree */
771 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
772 int size
; /* shortform size */
773 xfs_trans_t
*tp
; /* transaction pointer */
775 trace_xfs_dir2_block_removename(args
);
778 * Look up the entry in the block. Gets the buffer and entry index.
779 * It will always be there, the vnodeops level does a lookup first.
781 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
787 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
788 blp
= xfs_dir2_block_leaf_p(btp
);
790 * Point to the data entry using the leaf entry.
792 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
793 xfs_dir2_dataptr_to_off(args
->geo
,
794 be32_to_cpu(blp
[ent
].address
)));
796 * Mark the data entry's space free.
798 needlog
= needscan
= 0;
799 xfs_dir2_data_make_free(args
, bp
,
800 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)hdr
),
801 dp
->d_ops
->data_entsize(dep
->namelen
), &needlog
, &needscan
);
803 * Fix up the block tail.
805 be32_add_cpu(&btp
->stale
, 1);
806 xfs_dir2_block_log_tail(tp
, bp
);
808 * Remove the leaf entry by marking it stale.
810 blp
[ent
].address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
811 xfs_dir2_block_log_leaf(tp
, bp
, ent
, ent
);
813 * Fix up bestfree, log the header if necessary.
816 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
818 xfs_dir2_data_log_header(args
, bp
);
819 xfs_dir3_data_check(dp
, bp
);
821 * See if the size as a shortform is good enough.
823 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
824 if (size
> XFS_IFORK_DSIZE(dp
))
828 * If it works, do the conversion.
830 return xfs_dir2_block_to_sf(args
, bp
, size
, &sfh
);
834 * Replace an entry in a V2 block directory.
835 * Change the inode number to the new value.
838 xfs_dir2_block_replace(
839 xfs_da_args_t
*args
) /* directory operation args */
841 xfs_dir2_data_hdr_t
*hdr
; /* block header */
842 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
843 struct xfs_buf
*bp
; /* block buffer */
844 xfs_dir2_block_tail_t
*btp
; /* block tail */
845 xfs_dir2_data_entry_t
*dep
; /* block data entry */
846 xfs_inode_t
*dp
; /* incore inode */
847 int ent
; /* leaf entry index */
848 int error
; /* error return value */
850 trace_xfs_dir2_block_replace(args
);
853 * Lookup the entry in the directory. Get buffer and entry index.
854 * This will always succeed since the caller has already done a lookup.
856 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
861 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
862 blp
= xfs_dir2_block_leaf_p(btp
);
864 * Point to the data entry we need to change.
866 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
867 xfs_dir2_dataptr_to_off(args
->geo
,
868 be32_to_cpu(blp
[ent
].address
)));
869 ASSERT(be64_to_cpu(dep
->inumber
) != args
->inumber
);
871 * Change the inode number to the new value.
873 dep
->inumber
= cpu_to_be64(args
->inumber
);
874 dp
->d_ops
->data_put_ftype(dep
, args
->filetype
);
875 xfs_dir2_data_log_entry(args
, bp
, dep
);
876 xfs_dir3_data_check(dp
, bp
);
881 * Qsort comparison routine for the block leaf entries.
883 static int /* sort order */
885 const void *a
, /* first leaf entry */
886 const void *b
) /* second leaf entry */
888 const xfs_dir2_leaf_entry_t
*la
; /* first leaf entry */
889 const xfs_dir2_leaf_entry_t
*lb
; /* second leaf entry */
893 return be32_to_cpu(la
->hashval
) < be32_to_cpu(lb
->hashval
) ? -1 :
894 (be32_to_cpu(la
->hashval
) > be32_to_cpu(lb
->hashval
) ? 1 : 0);
898 * Convert a V2 leaf directory to a V2 block directory if possible.
901 xfs_dir2_leaf_to_block(
902 xfs_da_args_t
*args
, /* operation arguments */
903 struct xfs_buf
*lbp
, /* leaf buffer */
904 struct xfs_buf
*dbp
) /* data buffer */
906 __be16
*bestsp
; /* leaf bests table */
907 xfs_dir2_data_hdr_t
*hdr
; /* block header */
908 xfs_dir2_block_tail_t
*btp
; /* block tail */
909 xfs_inode_t
*dp
; /* incore directory inode */
910 xfs_dir2_data_unused_t
*dup
; /* unused data entry */
911 int error
; /* error return value */
912 int from
; /* leaf from index */
913 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
914 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
915 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
916 xfs_mount_t
*mp
; /* file system mount point */
917 int needlog
; /* need to log data header */
918 int needscan
; /* need to scan for bestfree */
919 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
920 int size
; /* bytes used */
921 __be16
*tagp
; /* end of entry (tag) */
922 int to
; /* block/leaf to index */
923 xfs_trans_t
*tp
; /* transaction pointer */
924 struct xfs_dir2_leaf_entry
*ents
;
925 struct xfs_dir3_icleaf_hdr leafhdr
;
927 trace_xfs_dir2_leaf_to_block(args
);
933 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
934 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
935 ltp
= xfs_dir2_leaf_tail_p(args
->geo
, leaf
);
937 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAF1_MAGIC
||
938 leafhdr
.magic
== XFS_DIR3_LEAF1_MAGIC
);
940 * If there are data blocks other than the first one, take this
941 * opportunity to remove trailing empty data blocks that may have
942 * been left behind during no-space-reservation operations.
943 * These will show up in the leaf bests table.
945 while (dp
->i_d
.di_size
> args
->geo
->blksize
) {
948 hdrsz
= dp
->d_ops
->data_entry_offset
;
949 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
950 if (be16_to_cpu(bestsp
[be32_to_cpu(ltp
->bestcount
) - 1]) ==
951 args
->geo
->blksize
- hdrsz
) {
953 xfs_dir2_leaf_trim_data(args
, lbp
,
954 (xfs_dir2_db_t
)(be32_to_cpu(ltp
->bestcount
) - 1))))
960 * Read the data block if we don't already have it, give up if it fails.
963 error
= xfs_dir3_data_read(tp
, dp
, args
->geo
->datablk
, -1, &dbp
);
968 ASSERT(hdr
->magic
== cpu_to_be32(XFS_DIR2_DATA_MAGIC
) ||
969 hdr
->magic
== cpu_to_be32(XFS_DIR3_DATA_MAGIC
));
972 * Size of the "leaf" area in the block.
974 size
= (uint
)sizeof(xfs_dir2_block_tail_t
) +
975 (uint
)sizeof(*lep
) * (leafhdr
.count
- leafhdr
.stale
);
977 * Look at the last data entry.
979 tagp
= (__be16
*)((char *)hdr
+ args
->geo
->blksize
) - 1;
980 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
982 * If it's not free or is too short we can't do it.
984 if (be16_to_cpu(dup
->freetag
) != XFS_DIR2_DATA_FREE_TAG
||
985 be16_to_cpu(dup
->length
) < size
)
989 * Start converting it to block form.
991 xfs_dir3_block_init(mp
, tp
, dbp
, dp
);
996 * Use up the space at the end of the block (blp/btp).
998 xfs_dir2_data_use_free(args
, dbp
, dup
, args
->geo
->blksize
- size
, size
,
999 &needlog
, &needscan
);
1001 * Initialize the block tail.
1003 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
1004 btp
->count
= cpu_to_be32(leafhdr
.count
- leafhdr
.stale
);
1006 xfs_dir2_block_log_tail(tp
, dbp
);
1008 * Initialize the block leaf area. We compact out stale entries.
1010 lep
= xfs_dir2_block_leaf_p(btp
);
1011 for (from
= to
= 0; from
< leafhdr
.count
; from
++) {
1012 if (ents
[from
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
1014 lep
[to
++] = ents
[from
];
1016 ASSERT(to
== be32_to_cpu(btp
->count
));
1017 xfs_dir2_block_log_leaf(tp
, dbp
, 0, be32_to_cpu(btp
->count
) - 1);
1019 * Scan the bestfree if we need it and log the data block header.
1022 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
1024 xfs_dir2_data_log_header(args
, dbp
);
1026 * Pitch the old leaf block.
1028 error
= xfs_da_shrink_inode(args
, args
->geo
->leafblk
, lbp
);
1033 * Now see if the resulting block can be shrunken to shortform.
1035 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
1036 if (size
> XFS_IFORK_DSIZE(dp
))
1039 return xfs_dir2_block_to_sf(args
, dbp
, size
, &sfh
);
1043 * Convert the shortform directory to block form.
1046 xfs_dir2_sf_to_block(
1047 xfs_da_args_t
*args
) /* operation arguments */
1049 xfs_dir2_db_t blkno
; /* dir-relative block # (0) */
1050 xfs_dir2_data_hdr_t
*hdr
; /* block header */
1051 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
1052 struct xfs_buf
*bp
; /* block buffer */
1053 xfs_dir2_block_tail_t
*btp
; /* block tail pointer */
1054 xfs_dir2_data_entry_t
*dep
; /* data entry pointer */
1055 xfs_inode_t
*dp
; /* incore directory inode */
1056 int dummy
; /* trash */
1057 xfs_dir2_data_unused_t
*dup
; /* unused entry pointer */
1058 int endoffset
; /* end of data objects */
1059 int error
; /* error return value */
1061 xfs_mount_t
*mp
; /* filesystem mount point */
1062 int needlog
; /* need to log block header */
1063 int needscan
; /* need to scan block freespc */
1064 int newoffset
; /* offset from current entry */
1065 int offset
; /* target block offset */
1066 xfs_dir2_sf_entry_t
*sfep
; /* sf entry pointer */
1067 xfs_dir2_sf_hdr_t
*oldsfp
; /* old shortform header */
1068 xfs_dir2_sf_hdr_t
*sfp
; /* shortform header */
1069 __be16
*tagp
; /* end of data entry */
1070 xfs_trans_t
*tp
; /* transaction pointer */
1071 struct xfs_name name
;
1072 struct xfs_ifork
*ifp
;
1074 trace_xfs_dir2_sf_to_block(args
);
1079 ifp
= XFS_IFORK_PTR(dp
, XFS_DATA_FORK
);
1080 ASSERT(ifp
->if_flags
& XFS_IFINLINE
);
1082 * Bomb out if the shortform directory is way too short.
1084 if (dp
->i_d
.di_size
< offsetof(xfs_dir2_sf_hdr_t
, parent
)) {
1085 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1089 oldsfp
= (xfs_dir2_sf_hdr_t
*)ifp
->if_u1
.if_data
;
1091 ASSERT(ifp
->if_bytes
== dp
->i_d
.di_size
);
1092 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
1093 ASSERT(dp
->i_d
.di_size
>= xfs_dir2_sf_hdr_size(oldsfp
->i8count
));
1094 ASSERT(dp
->i_d
.di_nextents
== 0);
1097 * Copy the directory into a temporary buffer.
1098 * Then pitch the incore inode data so we can make extents.
1100 sfp
= kmem_alloc(ifp
->if_bytes
, KM_SLEEP
);
1101 memcpy(sfp
, oldsfp
, ifp
->if_bytes
);
1103 xfs_idata_realloc(dp
, -ifp
->if_bytes
, XFS_DATA_FORK
);
1104 xfs_bmap_local_to_extents_empty(dp
, XFS_DATA_FORK
);
1105 dp
->i_d
.di_size
= 0;
1108 * Add block 0 to the inode.
1110 error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
, &blkno
);
1116 * Initialize the data block, then convert it to block format.
1118 error
= xfs_dir3_data_init(args
, blkno
, &bp
);
1123 xfs_dir3_block_init(mp
, tp
, bp
, dp
);
1127 * Compute size of block "tail" area.
1129 i
= (uint
)sizeof(*btp
) +
1130 (sfp
->count
+ 2) * (uint
)sizeof(xfs_dir2_leaf_entry_t
);
1132 * The whole thing is initialized to free by the init routine.
1133 * Say we're using the leaf and tail area.
1135 dup
= dp
->d_ops
->data_unused_p(hdr
);
1136 needlog
= needscan
= 0;
1137 xfs_dir2_data_use_free(args
, bp
, dup
, args
->geo
->blksize
- i
,
1138 i
, &needlog
, &needscan
);
1139 ASSERT(needscan
== 0);
1143 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
1144 btp
->count
= cpu_to_be32(sfp
->count
+ 2); /* ., .. */
1146 blp
= xfs_dir2_block_leaf_p(btp
);
1147 endoffset
= (uint
)((char *)blp
- (char *)hdr
);
1149 * Remove the freespace, we'll manage it.
1151 xfs_dir2_data_use_free(args
, bp
, dup
,
1152 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
1153 be16_to_cpu(dup
->length
), &needlog
, &needscan
);
1155 * Create entry for .
1157 dep
= dp
->d_ops
->data_dot_entry_p(hdr
);
1158 dep
->inumber
= cpu_to_be64(dp
->i_ino
);
1161 dp
->d_ops
->data_put_ftype(dep
, XFS_DIR3_FT_DIR
);
1162 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
1163 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1164 xfs_dir2_data_log_entry(args
, bp
, dep
);
1165 blp
[0].hashval
= cpu_to_be32(xfs_dir_hash_dot
);
1166 blp
[0].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(
1167 (char *)dep
- (char *)hdr
));
1169 * Create entry for ..
1171 dep
= dp
->d_ops
->data_dotdot_entry_p(hdr
);
1172 dep
->inumber
= cpu_to_be64(dp
->d_ops
->sf_get_parent_ino(sfp
));
1174 dep
->name
[0] = dep
->name
[1] = '.';
1175 dp
->d_ops
->data_put_ftype(dep
, XFS_DIR3_FT_DIR
);
1176 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
1177 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1178 xfs_dir2_data_log_entry(args
, bp
, dep
);
1179 blp
[1].hashval
= cpu_to_be32(xfs_dir_hash_dotdot
);
1180 blp
[1].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(
1181 (char *)dep
- (char *)hdr
));
1182 offset
= dp
->d_ops
->data_first_offset
;
1184 * Loop over existing entries, stuff them in.
1190 sfep
= xfs_dir2_sf_firstentry(sfp
);
1192 * Need to preserve the existing offset values in the sf directory.
1193 * Insert holes (unused entries) where necessary.
1195 while (offset
< endoffset
) {
1197 * sfep is null when we reach the end of the list.
1200 newoffset
= endoffset
;
1202 newoffset
= xfs_dir2_sf_get_offset(sfep
);
1204 * There should be a hole here, make one.
1206 if (offset
< newoffset
) {
1207 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ offset
);
1208 dup
->freetag
= cpu_to_be16(XFS_DIR2_DATA_FREE_TAG
);
1209 dup
->length
= cpu_to_be16(newoffset
- offset
);
1210 *xfs_dir2_data_unused_tag_p(dup
) = cpu_to_be16(
1211 ((char *)dup
- (char *)hdr
));
1212 xfs_dir2_data_log_unused(args
, bp
, dup
);
1213 xfs_dir2_data_freeinsert(hdr
,
1214 dp
->d_ops
->data_bestfree_p(hdr
),
1216 offset
+= be16_to_cpu(dup
->length
);
1220 * Copy a real entry.
1222 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+ newoffset
);
1223 dep
->inumber
= cpu_to_be64(dp
->d_ops
->sf_get_ino(sfp
, sfep
));
1224 dep
->namelen
= sfep
->namelen
;
1225 dp
->d_ops
->data_put_ftype(dep
, dp
->d_ops
->sf_get_ftype(sfep
));
1226 memcpy(dep
->name
, sfep
->name
, dep
->namelen
);
1227 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
1228 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1229 xfs_dir2_data_log_entry(args
, bp
, dep
);
1230 name
.name
= sfep
->name
;
1231 name
.len
= sfep
->namelen
;
1232 blp
[2 + i
].hashval
= cpu_to_be32(mp
->m_dirnameops
->
1234 blp
[2 + i
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(
1235 (char *)dep
- (char *)hdr
));
1236 offset
= (int)((char *)(tagp
+ 1) - (char *)hdr
);
1237 if (++i
== sfp
->count
)
1240 sfep
= dp
->d_ops
->sf_nextentry(sfp
, sfep
);
1242 /* Done with the temporary buffer */
1245 * Sort the leaf entries by hash value.
1247 xfs_sort(blp
, be32_to_cpu(btp
->count
), sizeof(*blp
), xfs_dir2_block_sort
);
1249 * Log the leaf entry area and tail.
1250 * Already logged the header in data_init, ignore needlog.
1252 ASSERT(needscan
== 0);
1253 xfs_dir2_block_log_leaf(tp
, bp
, 0, be32_to_cpu(btp
->count
) - 1);
1254 xfs_dir2_block_log_tail(tp
, bp
);
1255 xfs_dir3_data_check(dp
, bp
);