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
))
67 if (!uuid_equal(&hdr3
->uuid
, &mp
->m_sb
.sb_meta_uuid
))
69 if (be64_to_cpu(hdr3
->blkno
) != bp
->b_bn
)
71 if (!xfs_log_check_lsn(mp
, be64_to_cpu(hdr3
->lsn
)))
74 if (hdr3
->magic
!= cpu_to_be32(XFS_DIR2_BLOCK_MAGIC
))
77 if (__xfs_dir3_data_check(NULL
, bp
))
83 xfs_dir3_block_read_verify(
86 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
88 if (xfs_sb_version_hascrc(&mp
->m_sb
) &&
89 !xfs_buf_verify_cksum(bp
, XFS_DIR3_DATA_CRC_OFF
))
90 xfs_buf_ioerror(bp
, -EFSBADCRC
);
91 else if (!xfs_dir3_block_verify(bp
))
92 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
95 xfs_verifier_error(bp
);
99 xfs_dir3_block_write_verify(
102 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
103 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
104 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
106 if (!xfs_dir3_block_verify(bp
)) {
107 xfs_buf_ioerror(bp
, -EFSCORRUPTED
);
108 xfs_verifier_error(bp
);
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 .verify_read
= xfs_dir3_block_read_verify
,
123 .verify_write
= xfs_dir3_block_write_verify
,
128 struct xfs_trans
*tp
,
129 struct xfs_inode
*dp
,
130 struct xfs_buf
**bpp
)
132 struct xfs_mount
*mp
= dp
->i_mount
;
135 err
= xfs_da_read_buf(tp
, dp
, mp
->m_dir_geo
->datablk
, -1, bpp
,
136 XFS_DATA_FORK
, &xfs_dir3_block_buf_ops
);
138 xfs_trans_buf_set_type(tp
, *bpp
, XFS_BLFT_DIR_BLOCK_BUF
);
144 struct xfs_mount
*mp
,
145 struct xfs_trans
*tp
,
147 struct xfs_inode
*dp
)
149 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
151 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
152 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DIR_BLOCK_BUF
);
154 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
155 memset(hdr3
, 0, sizeof(*hdr3
));
156 hdr3
->magic
= cpu_to_be32(XFS_DIR3_BLOCK_MAGIC
);
157 hdr3
->blkno
= cpu_to_be64(bp
->b_bn
);
158 hdr3
->owner
= cpu_to_be64(dp
->i_ino
);
159 uuid_copy(&hdr3
->uuid
, &mp
->m_sb
.sb_meta_uuid
);
163 hdr3
->magic
= cpu_to_be32(XFS_DIR2_BLOCK_MAGIC
);
167 xfs_dir2_block_need_space(
168 struct xfs_inode
*dp
,
169 struct xfs_dir2_data_hdr
*hdr
,
170 struct xfs_dir2_block_tail
*btp
,
171 struct xfs_dir2_leaf_entry
*blp
,
173 struct xfs_dir2_data_unused
**dupp
,
174 struct xfs_dir2_data_unused
**enddupp
,
178 struct xfs_dir2_data_free
*bf
;
180 struct xfs_dir2_data_unused
*dup
= NULL
;
181 struct xfs_dir2_data_unused
*enddup
= NULL
;
184 bf
= dp
->d_ops
->data_bestfree_p(hdr
);
187 * If there are stale entries we'll use one for the leaf.
190 if (be16_to_cpu(bf
[0].length
) >= len
) {
192 * The biggest entry enough to avoid compaction.
194 dup
= (xfs_dir2_data_unused_t
*)
195 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
200 * Will need to compact to make this work.
201 * Tag just before the first leaf entry.
204 tagp
= (__be16
*)blp
- 1;
206 /* Data object just before the first leaf entry. */
207 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
210 * If it's not free then the data will go where the
211 * leaf data starts now, if it works at all.
213 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
214 if (be16_to_cpu(dup
->length
) + (be32_to_cpu(btp
->stale
) - 1) *
215 (uint
)sizeof(*blp
) < len
)
217 } else if ((be32_to_cpu(btp
->stale
) - 1) * (uint
)sizeof(*blp
) < len
)
220 dup
= (xfs_dir2_data_unused_t
*)blp
;
225 * no stale entries, so just use free space.
226 * Tag just before the first leaf entry.
228 tagp
= (__be16
*)blp
- 1;
230 /* Data object just before the first leaf entry. */
231 enddup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
234 * If it's not free then can't do this add without cleaning up:
235 * the space before the first leaf entry needs to be free so it
236 * can be expanded to hold the pointer to the new entry.
238 if (be16_to_cpu(enddup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
240 * Check out the biggest freespace and see if it's the same one.
242 dup
= (xfs_dir2_data_unused_t
*)
243 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
246 * Not the same free entry, just check its length.
248 if (be16_to_cpu(dup
->length
) < len
)
254 * It is the biggest freespace, can it hold the leaf too?
256 if (be16_to_cpu(dup
->length
) < len
+ (uint
)sizeof(*blp
)) {
258 * Yes, use the second-largest entry instead if it works.
260 if (be16_to_cpu(bf
[1].length
) >= len
)
261 dup
= (xfs_dir2_data_unused_t
*)
262 ((char *)hdr
+ be16_to_cpu(bf
[1].offset
));
274 * compact the leaf entries.
275 * Leave the highest-numbered stale entry stale.
276 * XXX should be the one closest to mid but mid is not yet computed.
279 xfs_dir2_block_compact(
280 struct xfs_da_args
*args
,
282 struct xfs_dir2_data_hdr
*hdr
,
283 struct xfs_dir2_block_tail
*btp
,
284 struct xfs_dir2_leaf_entry
*blp
,
289 int fromidx
; /* source leaf index */
290 int toidx
; /* target leaf index */
292 int highstale
; /* high stale index */
294 fromidx
= toidx
= be32_to_cpu(btp
->count
) - 1;
295 highstale
= *lfloghigh
= -1;
296 for (; fromidx
>= 0; fromidx
--) {
297 if (blp
[fromidx
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
)) {
301 if (*lfloghigh
== -1)
307 blp
[toidx
] = blp
[fromidx
];
310 *lfloglow
= toidx
+ 1 - (be32_to_cpu(btp
->stale
) - 1);
311 *lfloghigh
-= be32_to_cpu(btp
->stale
) - 1;
312 be32_add_cpu(&btp
->count
, -(be32_to_cpu(btp
->stale
) - 1));
313 xfs_dir2_data_make_free(args
, bp
,
314 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)hdr
),
315 (xfs_dir2_data_aoff_t
)((be32_to_cpu(btp
->stale
) - 1) * sizeof(*blp
)),
317 btp
->stale
= cpu_to_be32(1);
319 * If we now need to rebuild the bestfree map, do so.
320 * This needs to happen before the next call to use_free.
323 xfs_dir2_data_freescan(args
->dp
, hdr
, needlog
);
327 * Add an entry to a block directory.
330 xfs_dir2_block_addname(
331 xfs_da_args_t
*args
) /* directory op arguments */
333 xfs_dir2_data_hdr_t
*hdr
; /* block header */
334 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
335 struct xfs_buf
*bp
; /* buffer for block */
336 xfs_dir2_block_tail_t
*btp
; /* block tail */
337 int compact
; /* need to compact leaf ents */
338 xfs_dir2_data_entry_t
*dep
; /* block data entry */
339 xfs_inode_t
*dp
; /* directory inode */
340 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
341 int error
; /* error return value */
342 xfs_dir2_data_unused_t
*enddup
=NULL
; /* unused at end of data */
343 xfs_dahash_t hash
; /* hash value of found entry */
344 int high
; /* high index for binary srch */
345 int highstale
; /* high stale index */
346 int lfloghigh
=0; /* last final leaf to log */
347 int lfloglow
=0; /* first final leaf to log */
348 int len
; /* length of the new entry */
349 int low
; /* low index for binary srch */
350 int lowstale
; /* low stale index */
351 int mid
=0; /* midpoint for binary srch */
352 int needlog
; /* need to log header */
353 int needscan
; /* need to rescan freespace */
354 __be16
*tagp
; /* pointer to tag value */
355 xfs_trans_t
*tp
; /* transaction structure */
357 trace_xfs_dir2_block_addname(args
);
362 /* Read the (one and only) directory block into bp. */
363 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
367 len
= dp
->d_ops
->data_entsize(args
->namelen
);
370 * Set up pointers to parts of the block.
373 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
374 blp
= xfs_dir2_block_leaf_p(btp
);
377 * Find out if we can reuse stale entries or whether we need extra
378 * space for entry and new leaf.
380 xfs_dir2_block_need_space(dp
, hdr
, btp
, blp
, &tagp
, &dup
,
381 &enddup
, &compact
, len
);
384 * Done everything we need for a space check now.
386 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
387 xfs_trans_brelse(tp
, bp
);
394 * If we don't have space for the new entry & leaf ...
397 /* Don't have a space reservation: return no-space. */
398 if (args
->total
== 0)
401 * Convert to the next larger format.
402 * Then add the new entry in that format.
404 error
= xfs_dir2_block_to_leaf(args
, bp
);
407 return xfs_dir2_leaf_addname(args
);
410 needlog
= needscan
= 0;
413 * If need to compact the leaf entries, do it now.
416 xfs_dir2_block_compact(args
, bp
, hdr
, btp
, blp
, &needlog
,
417 &lfloghigh
, &lfloglow
);
418 /* recalculate blp post-compaction */
419 blp
= xfs_dir2_block_leaf_p(btp
);
420 } else if (btp
->stale
) {
422 * Set leaf logging boundaries to impossible state.
423 * For the no-stale case they're set explicitly.
425 lfloglow
= be32_to_cpu(btp
->count
);
430 * Find the slot that's first lower than our hash value, -1 if none.
432 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; low
<= high
; ) {
433 mid
= (low
+ high
) >> 1;
434 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
436 if (hash
< args
->hashval
)
441 while (mid
>= 0 && be32_to_cpu(blp
[mid
].hashval
) >= args
->hashval
) {
445 * No stale entries, will use enddup space to hold new leaf.
449 * Mark the space needed for the new leaf entry, now in use.
451 xfs_dir2_data_use_free(args
, bp
, enddup
,
452 (xfs_dir2_data_aoff_t
)
453 ((char *)enddup
- (char *)hdr
+ be16_to_cpu(enddup
->length
) -
455 (xfs_dir2_data_aoff_t
)sizeof(*blp
),
456 &needlog
, &needscan
);
458 * Update the tail (entry count).
460 be32_add_cpu(&btp
->count
, 1);
462 * If we now need to rebuild the bestfree map, do so.
463 * This needs to happen before the next call to use_free.
466 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
470 * Adjust pointer to the first leaf entry, we're about to move
471 * the table up one to open up space for the new leaf entry.
472 * Then adjust our index to match.
477 memmove(blp
, &blp
[1], mid
* sizeof(*blp
));
482 * Use a stale leaf for our new entry.
487 blp
[lowstale
].address
!=
488 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
491 for (highstale
= mid
+ 1;
492 highstale
< be32_to_cpu(btp
->count
) &&
493 blp
[highstale
].address
!=
494 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
) &&
495 (lowstale
< 0 || mid
- lowstale
> highstale
- mid
);
499 * Move entries toward the low-numbered stale entry.
502 (highstale
== be32_to_cpu(btp
->count
) ||
503 mid
- lowstale
<= highstale
- mid
)) {
505 memmove(&blp
[lowstale
], &blp
[lowstale
+ 1],
506 (mid
- lowstale
) * sizeof(*blp
));
507 lfloglow
= MIN(lowstale
, lfloglow
);
508 lfloghigh
= MAX(mid
, lfloghigh
);
511 * Move entries toward the high-numbered stale entry.
514 ASSERT(highstale
< be32_to_cpu(btp
->count
));
517 memmove(&blp
[mid
+ 1], &blp
[mid
],
518 (highstale
- mid
) * sizeof(*blp
));
519 lfloglow
= MIN(mid
, lfloglow
);
520 lfloghigh
= MAX(highstale
, lfloghigh
);
522 be32_add_cpu(&btp
->stale
, -1);
525 * Point to the new data entry.
527 dep
= (xfs_dir2_data_entry_t
*)dup
;
529 * Fill in the leaf entry.
531 blp
[mid
].hashval
= cpu_to_be32(args
->hashval
);
532 blp
[mid
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(
533 (char *)dep
- (char *)hdr
));
534 xfs_dir2_block_log_leaf(tp
, bp
, lfloglow
, lfloghigh
);
536 * Mark space for the data entry used.
538 xfs_dir2_data_use_free(args
, bp
, dup
,
539 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
540 (xfs_dir2_data_aoff_t
)len
, &needlog
, &needscan
);
542 * Create the new data entry.
544 dep
->inumber
= cpu_to_be64(args
->inumber
);
545 dep
->namelen
= args
->namelen
;
546 memcpy(dep
->name
, args
->name
, args
->namelen
);
547 dp
->d_ops
->data_put_ftype(dep
, args
->filetype
);
548 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
549 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
551 * Clean up the bestfree array and log the header, tail, and entry.
554 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
556 xfs_dir2_data_log_header(args
, bp
);
557 xfs_dir2_block_log_tail(tp
, bp
);
558 xfs_dir2_data_log_entry(args
, bp
, dep
);
559 xfs_dir3_data_check(dp
, bp
);
564 * Log leaf entries from the block.
567 xfs_dir2_block_log_leaf(
568 xfs_trans_t
*tp
, /* transaction structure */
569 struct xfs_buf
*bp
, /* block buffer */
570 int first
, /* index of first logged leaf */
571 int last
) /* index of last logged leaf */
573 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
574 xfs_dir2_leaf_entry_t
*blp
;
575 xfs_dir2_block_tail_t
*btp
;
577 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
->m_dir_geo
, hdr
);
578 blp
= xfs_dir2_block_leaf_p(btp
);
579 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)&blp
[first
] - (char *)hdr
),
580 (uint
)((char *)&blp
[last
+ 1] - (char *)hdr
- 1));
584 * Log the block tail.
587 xfs_dir2_block_log_tail(
588 xfs_trans_t
*tp
, /* transaction structure */
589 struct xfs_buf
*bp
) /* block buffer */
591 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
592 xfs_dir2_block_tail_t
*btp
;
594 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
->m_dir_geo
, hdr
);
595 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)btp
- (char *)hdr
),
596 (uint
)((char *)(btp
+ 1) - (char *)hdr
- 1));
600 * Look up an entry in the block. This is the external routine,
601 * xfs_dir2_block_lookup_int does the real work.
604 xfs_dir2_block_lookup(
605 xfs_da_args_t
*args
) /* dir lookup arguments */
607 xfs_dir2_data_hdr_t
*hdr
; /* block header */
608 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
609 struct xfs_buf
*bp
; /* block buffer */
610 xfs_dir2_block_tail_t
*btp
; /* block tail */
611 xfs_dir2_data_entry_t
*dep
; /* block data entry */
612 xfs_inode_t
*dp
; /* incore inode */
613 int ent
; /* entry index */
614 int error
; /* error return value */
616 trace_xfs_dir2_block_lookup(args
);
619 * Get the buffer, look up the entry.
620 * If not found (ENOENT) then return, have no buffer.
622 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
)))
626 xfs_dir3_data_check(dp
, bp
);
627 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
628 blp
= xfs_dir2_block_leaf_p(btp
);
630 * Get the offset from the leaf entry, to point to the data.
632 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
633 xfs_dir2_dataptr_to_off(args
->geo
,
634 be32_to_cpu(blp
[ent
].address
)));
636 * Fill in inode number, CI name if appropriate, release the block.
638 args
->inumber
= be64_to_cpu(dep
->inumber
);
639 args
->filetype
= dp
->d_ops
->data_get_ftype(dep
);
640 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
641 xfs_trans_brelse(args
->trans
, bp
);
646 * Internal block lookup routine.
648 static int /* error */
649 xfs_dir2_block_lookup_int(
650 xfs_da_args_t
*args
, /* dir lookup arguments */
651 struct xfs_buf
**bpp
, /* returned block buffer */
652 int *entno
) /* returned entry number */
654 xfs_dir2_dataptr_t addr
; /* data entry address */
655 xfs_dir2_data_hdr_t
*hdr
; /* block header */
656 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
657 struct xfs_buf
*bp
; /* block buffer */
658 xfs_dir2_block_tail_t
*btp
; /* block tail */
659 xfs_dir2_data_entry_t
*dep
; /* block data entry */
660 xfs_inode_t
*dp
; /* incore inode */
661 int error
; /* error return value */
662 xfs_dahash_t hash
; /* found hash value */
663 int high
; /* binary search high index */
664 int low
; /* binary search low index */
665 int mid
; /* binary search current idx */
666 xfs_mount_t
*mp
; /* filesystem mount point */
667 xfs_trans_t
*tp
; /* transaction pointer */
668 enum xfs_dacmp cmp
; /* comparison result */
674 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
679 xfs_dir3_data_check(dp
, bp
);
680 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
681 blp
= xfs_dir2_block_leaf_p(btp
);
683 * Loop doing a binary search for our hash value.
684 * Find our entry, ENOENT if it's not there.
686 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; ; ) {
688 mid
= (low
+ high
) >> 1;
689 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
691 if (hash
< args
->hashval
)
696 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
697 xfs_trans_brelse(tp
, bp
);
702 * Back up to the first one with the right hash value.
704 while (mid
> 0 && be32_to_cpu(blp
[mid
- 1].hashval
) == args
->hashval
) {
708 * Now loop forward through all the entries with the
709 * right hash value looking for our name.
712 if ((addr
= be32_to_cpu(blp
[mid
].address
)) == XFS_DIR2_NULL_DATAPTR
)
715 * Get pointer to the entry from the leaf.
717 dep
= (xfs_dir2_data_entry_t
*)
718 ((char *)hdr
+ xfs_dir2_dataptr_to_off(args
->geo
, addr
));
720 * Compare name and if it's an exact match, return the index
721 * and buffer. If it's the first case-insensitive match, store
722 * the index and buffer and continue looking for an exact match.
724 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
725 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
726 args
->cmpresult
= cmp
;
729 if (cmp
== XFS_CMP_EXACT
)
732 } while (++mid
< be32_to_cpu(btp
->count
) &&
733 be32_to_cpu(blp
[mid
].hashval
) == hash
);
735 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
737 * Here, we can only be doing a lookup (not a rename or replace).
738 * If a case-insensitive match was found earlier, return success.
740 if (args
->cmpresult
== XFS_CMP_CASE
)
743 * No match, release the buffer and return ENOENT.
745 xfs_trans_brelse(tp
, bp
);
750 * Remove an entry from a block format directory.
751 * If that makes the block small enough to fit in shortform, transform it.
754 xfs_dir2_block_removename(
755 xfs_da_args_t
*args
) /* directory operation args */
757 xfs_dir2_data_hdr_t
*hdr
; /* block header */
758 xfs_dir2_leaf_entry_t
*blp
; /* block leaf pointer */
759 struct xfs_buf
*bp
; /* block buffer */
760 xfs_dir2_block_tail_t
*btp
; /* block tail */
761 xfs_dir2_data_entry_t
*dep
; /* block data entry */
762 xfs_inode_t
*dp
; /* incore inode */
763 int ent
; /* block leaf entry index */
764 int error
; /* error return value */
765 int needlog
; /* need to log block header */
766 int needscan
; /* need to fixup bestfree */
767 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
768 int size
; /* shortform size */
769 xfs_trans_t
*tp
; /* transaction pointer */
771 trace_xfs_dir2_block_removename(args
);
774 * Look up the entry in the block. Gets the buffer and entry index.
775 * It will always be there, the vnodeops level does a lookup first.
777 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
783 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
784 blp
= xfs_dir2_block_leaf_p(btp
);
786 * Point to the data entry using the leaf entry.
788 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
789 xfs_dir2_dataptr_to_off(args
->geo
,
790 be32_to_cpu(blp
[ent
].address
)));
792 * Mark the data entry's space free.
794 needlog
= needscan
= 0;
795 xfs_dir2_data_make_free(args
, bp
,
796 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)hdr
),
797 dp
->d_ops
->data_entsize(dep
->namelen
), &needlog
, &needscan
);
799 * Fix up the block tail.
801 be32_add_cpu(&btp
->stale
, 1);
802 xfs_dir2_block_log_tail(tp
, bp
);
804 * Remove the leaf entry by marking it stale.
806 blp
[ent
].address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
807 xfs_dir2_block_log_leaf(tp
, bp
, ent
, ent
);
809 * Fix up bestfree, log the header if necessary.
812 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
814 xfs_dir2_data_log_header(args
, bp
);
815 xfs_dir3_data_check(dp
, bp
);
817 * See if the size as a shortform is good enough.
819 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
820 if (size
> XFS_IFORK_DSIZE(dp
))
824 * If it works, do the conversion.
826 return xfs_dir2_block_to_sf(args
, bp
, size
, &sfh
);
830 * Replace an entry in a V2 block directory.
831 * Change the inode number to the new value.
834 xfs_dir2_block_replace(
835 xfs_da_args_t
*args
) /* directory operation args */
837 xfs_dir2_data_hdr_t
*hdr
; /* block header */
838 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
839 struct xfs_buf
*bp
; /* block buffer */
840 xfs_dir2_block_tail_t
*btp
; /* block tail */
841 xfs_dir2_data_entry_t
*dep
; /* block data entry */
842 xfs_inode_t
*dp
; /* incore inode */
843 int ent
; /* leaf entry index */
844 int error
; /* error return value */
846 trace_xfs_dir2_block_replace(args
);
849 * Lookup the entry in the directory. Get buffer and entry index.
850 * This will always succeed since the caller has already done a lookup.
852 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
857 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
858 blp
= xfs_dir2_block_leaf_p(btp
);
860 * Point to the data entry we need to change.
862 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
863 xfs_dir2_dataptr_to_off(args
->geo
,
864 be32_to_cpu(blp
[ent
].address
)));
865 ASSERT(be64_to_cpu(dep
->inumber
) != args
->inumber
);
867 * Change the inode number to the new value.
869 dep
->inumber
= cpu_to_be64(args
->inumber
);
870 dp
->d_ops
->data_put_ftype(dep
, args
->filetype
);
871 xfs_dir2_data_log_entry(args
, bp
, dep
);
872 xfs_dir3_data_check(dp
, bp
);
877 * Qsort comparison routine for the block leaf entries.
879 static int /* sort order */
881 const void *a
, /* first leaf entry */
882 const void *b
) /* second leaf entry */
884 const xfs_dir2_leaf_entry_t
*la
; /* first leaf entry */
885 const xfs_dir2_leaf_entry_t
*lb
; /* second leaf entry */
889 return be32_to_cpu(la
->hashval
) < be32_to_cpu(lb
->hashval
) ? -1 :
890 (be32_to_cpu(la
->hashval
) > be32_to_cpu(lb
->hashval
) ? 1 : 0);
894 * Convert a V2 leaf directory to a V2 block directory if possible.
897 xfs_dir2_leaf_to_block(
898 xfs_da_args_t
*args
, /* operation arguments */
899 struct xfs_buf
*lbp
, /* leaf buffer */
900 struct xfs_buf
*dbp
) /* data buffer */
902 __be16
*bestsp
; /* leaf bests table */
903 xfs_dir2_data_hdr_t
*hdr
; /* block header */
904 xfs_dir2_block_tail_t
*btp
; /* block tail */
905 xfs_inode_t
*dp
; /* incore directory inode */
906 xfs_dir2_data_unused_t
*dup
; /* unused data entry */
907 int error
; /* error return value */
908 int from
; /* leaf from index */
909 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
910 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
911 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
912 xfs_mount_t
*mp
; /* file system mount point */
913 int needlog
; /* need to log data header */
914 int needscan
; /* need to scan for bestfree */
915 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
916 int size
; /* bytes used */
917 __be16
*tagp
; /* end of entry (tag) */
918 int to
; /* block/leaf to index */
919 xfs_trans_t
*tp
; /* transaction pointer */
920 struct xfs_dir2_leaf_entry
*ents
;
921 struct xfs_dir3_icleaf_hdr leafhdr
;
923 trace_xfs_dir2_leaf_to_block(args
);
929 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
930 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
931 ltp
= xfs_dir2_leaf_tail_p(args
->geo
, leaf
);
933 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAF1_MAGIC
||
934 leafhdr
.magic
== XFS_DIR3_LEAF1_MAGIC
);
936 * If there are data blocks other than the first one, take this
937 * opportunity to remove trailing empty data blocks that may have
938 * been left behind during no-space-reservation operations.
939 * These will show up in the leaf bests table.
941 while (dp
->i_d
.di_size
> args
->geo
->blksize
) {
944 hdrsz
= dp
->d_ops
->data_entry_offset
;
945 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
946 if (be16_to_cpu(bestsp
[be32_to_cpu(ltp
->bestcount
) - 1]) ==
947 args
->geo
->blksize
- hdrsz
) {
949 xfs_dir2_leaf_trim_data(args
, lbp
,
950 (xfs_dir2_db_t
)(be32_to_cpu(ltp
->bestcount
) - 1))))
956 * Read the data block if we don't already have it, give up if it fails.
959 error
= xfs_dir3_data_read(tp
, dp
, args
->geo
->datablk
, -1, &dbp
);
964 ASSERT(hdr
->magic
== cpu_to_be32(XFS_DIR2_DATA_MAGIC
) ||
965 hdr
->magic
== cpu_to_be32(XFS_DIR3_DATA_MAGIC
));
968 * Size of the "leaf" area in the block.
970 size
= (uint
)sizeof(xfs_dir2_block_tail_t
) +
971 (uint
)sizeof(*lep
) * (leafhdr
.count
- leafhdr
.stale
);
973 * Look at the last data entry.
975 tagp
= (__be16
*)((char *)hdr
+ args
->geo
->blksize
) - 1;
976 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
978 * If it's not free or is too short we can't do it.
980 if (be16_to_cpu(dup
->freetag
) != XFS_DIR2_DATA_FREE_TAG
||
981 be16_to_cpu(dup
->length
) < size
)
985 * Start converting it to block form.
987 xfs_dir3_block_init(mp
, tp
, dbp
, dp
);
992 * Use up the space at the end of the block (blp/btp).
994 xfs_dir2_data_use_free(args
, dbp
, dup
, args
->geo
->blksize
- size
, size
,
995 &needlog
, &needscan
);
997 * Initialize the block tail.
999 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
1000 btp
->count
= cpu_to_be32(leafhdr
.count
- leafhdr
.stale
);
1002 xfs_dir2_block_log_tail(tp
, dbp
);
1004 * Initialize the block leaf area. We compact out stale entries.
1006 lep
= xfs_dir2_block_leaf_p(btp
);
1007 for (from
= to
= 0; from
< leafhdr
.count
; from
++) {
1008 if (ents
[from
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
1010 lep
[to
++] = ents
[from
];
1012 ASSERT(to
== be32_to_cpu(btp
->count
));
1013 xfs_dir2_block_log_leaf(tp
, dbp
, 0, be32_to_cpu(btp
->count
) - 1);
1015 * Scan the bestfree if we need it and log the data block header.
1018 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
1020 xfs_dir2_data_log_header(args
, dbp
);
1022 * Pitch the old leaf block.
1024 error
= xfs_da_shrink_inode(args
, args
->geo
->leafblk
, lbp
);
1029 * Now see if the resulting block can be shrunken to shortform.
1031 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
1032 if (size
> XFS_IFORK_DSIZE(dp
))
1035 return xfs_dir2_block_to_sf(args
, dbp
, size
, &sfh
);
1039 * Convert the shortform directory to block form.
1042 xfs_dir2_sf_to_block(
1043 xfs_da_args_t
*args
) /* operation arguments */
1045 xfs_dir2_db_t blkno
; /* dir-relative block # (0) */
1046 xfs_dir2_data_hdr_t
*hdr
; /* block header */
1047 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
1048 struct xfs_buf
*bp
; /* block buffer */
1049 xfs_dir2_block_tail_t
*btp
; /* block tail pointer */
1050 xfs_dir2_data_entry_t
*dep
; /* data entry pointer */
1051 xfs_inode_t
*dp
; /* incore directory inode */
1052 int dummy
; /* trash */
1053 xfs_dir2_data_unused_t
*dup
; /* unused entry pointer */
1054 int endoffset
; /* end of data objects */
1055 int error
; /* error return value */
1057 xfs_mount_t
*mp
; /* filesystem mount point */
1058 int needlog
; /* need to log block header */
1059 int needscan
; /* need to scan block freespc */
1060 int newoffset
; /* offset from current entry */
1061 int offset
; /* target block offset */
1062 xfs_dir2_sf_entry_t
*sfep
; /* sf entry pointer */
1063 xfs_dir2_sf_hdr_t
*oldsfp
; /* old shortform header */
1064 xfs_dir2_sf_hdr_t
*sfp
; /* shortform header */
1065 __be16
*tagp
; /* end of data entry */
1066 xfs_trans_t
*tp
; /* transaction pointer */
1067 struct xfs_name name
;
1068 struct xfs_ifork
*ifp
;
1070 trace_xfs_dir2_sf_to_block(args
);
1075 ifp
= XFS_IFORK_PTR(dp
, XFS_DATA_FORK
);
1076 ASSERT(ifp
->if_flags
& XFS_IFINLINE
);
1078 * Bomb out if the shortform directory is way too short.
1080 if (dp
->i_d
.di_size
< offsetof(xfs_dir2_sf_hdr_t
, parent
)) {
1081 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1085 oldsfp
= (xfs_dir2_sf_hdr_t
*)ifp
->if_u1
.if_data
;
1087 ASSERT(ifp
->if_bytes
== dp
->i_d
.di_size
);
1088 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
1089 ASSERT(dp
->i_d
.di_size
>= xfs_dir2_sf_hdr_size(oldsfp
->i8count
));
1090 ASSERT(dp
->i_d
.di_nextents
== 0);
1093 * Copy the directory into a temporary buffer.
1094 * Then pitch the incore inode data so we can make extents.
1096 sfp
= kmem_alloc(ifp
->if_bytes
, KM_SLEEP
);
1097 memcpy(sfp
, oldsfp
, ifp
->if_bytes
);
1099 xfs_idata_realloc(dp
, -ifp
->if_bytes
, XFS_DATA_FORK
);
1100 xfs_bmap_local_to_extents_empty(dp
, XFS_DATA_FORK
);
1101 dp
->i_d
.di_size
= 0;
1104 * Add block 0 to the inode.
1106 error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
, &blkno
);
1112 * Initialize the data block, then convert it to block format.
1114 error
= xfs_dir3_data_init(args
, blkno
, &bp
);
1119 xfs_dir3_block_init(mp
, tp
, bp
, dp
);
1123 * Compute size of block "tail" area.
1125 i
= (uint
)sizeof(*btp
) +
1126 (sfp
->count
+ 2) * (uint
)sizeof(xfs_dir2_leaf_entry_t
);
1128 * The whole thing is initialized to free by the init routine.
1129 * Say we're using the leaf and tail area.
1131 dup
= dp
->d_ops
->data_unused_p(hdr
);
1132 needlog
= needscan
= 0;
1133 xfs_dir2_data_use_free(args
, bp
, dup
, args
->geo
->blksize
- i
,
1134 i
, &needlog
, &needscan
);
1135 ASSERT(needscan
== 0);
1139 btp
= xfs_dir2_block_tail_p(args
->geo
, hdr
);
1140 btp
->count
= cpu_to_be32(sfp
->count
+ 2); /* ., .. */
1142 blp
= xfs_dir2_block_leaf_p(btp
);
1143 endoffset
= (uint
)((char *)blp
- (char *)hdr
);
1145 * Remove the freespace, we'll manage it.
1147 xfs_dir2_data_use_free(args
, bp
, dup
,
1148 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
1149 be16_to_cpu(dup
->length
), &needlog
, &needscan
);
1151 * Create entry for .
1153 dep
= dp
->d_ops
->data_dot_entry_p(hdr
);
1154 dep
->inumber
= cpu_to_be64(dp
->i_ino
);
1157 dp
->d_ops
->data_put_ftype(dep
, XFS_DIR3_FT_DIR
);
1158 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
1159 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1160 xfs_dir2_data_log_entry(args
, bp
, dep
);
1161 blp
[0].hashval
= cpu_to_be32(xfs_dir_hash_dot
);
1162 blp
[0].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(
1163 (char *)dep
- (char *)hdr
));
1165 * Create entry for ..
1167 dep
= dp
->d_ops
->data_dotdot_entry_p(hdr
);
1168 dep
->inumber
= cpu_to_be64(dp
->d_ops
->sf_get_parent_ino(sfp
));
1170 dep
->name
[0] = dep
->name
[1] = '.';
1171 dp
->d_ops
->data_put_ftype(dep
, XFS_DIR3_FT_DIR
);
1172 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
1173 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1174 xfs_dir2_data_log_entry(args
, bp
, dep
);
1175 blp
[1].hashval
= cpu_to_be32(xfs_dir_hash_dotdot
);
1176 blp
[1].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(
1177 (char *)dep
- (char *)hdr
));
1178 offset
= dp
->d_ops
->data_first_offset
;
1180 * Loop over existing entries, stuff them in.
1186 sfep
= xfs_dir2_sf_firstentry(sfp
);
1188 * Need to preserve the existing offset values in the sf directory.
1189 * Insert holes (unused entries) where necessary.
1191 while (offset
< endoffset
) {
1193 * sfep is null when we reach the end of the list.
1196 newoffset
= endoffset
;
1198 newoffset
= xfs_dir2_sf_get_offset(sfep
);
1200 * There should be a hole here, make one.
1202 if (offset
< newoffset
) {
1203 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ offset
);
1204 dup
->freetag
= cpu_to_be16(XFS_DIR2_DATA_FREE_TAG
);
1205 dup
->length
= cpu_to_be16(newoffset
- offset
);
1206 *xfs_dir2_data_unused_tag_p(dup
) = cpu_to_be16(
1207 ((char *)dup
- (char *)hdr
));
1208 xfs_dir2_data_log_unused(args
, bp
, dup
);
1209 xfs_dir2_data_freeinsert(hdr
,
1210 dp
->d_ops
->data_bestfree_p(hdr
),
1212 offset
+= be16_to_cpu(dup
->length
);
1216 * Copy a real entry.
1218 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+ newoffset
);
1219 dep
->inumber
= cpu_to_be64(dp
->d_ops
->sf_get_ino(sfp
, sfep
));
1220 dep
->namelen
= sfep
->namelen
;
1221 dp
->d_ops
->data_put_ftype(dep
, dp
->d_ops
->sf_get_ftype(sfep
));
1222 memcpy(dep
->name
, sfep
->name
, dep
->namelen
);
1223 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
1224 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1225 xfs_dir2_data_log_entry(args
, bp
, dep
);
1226 name
.name
= sfep
->name
;
1227 name
.len
= sfep
->namelen
;
1228 blp
[2 + i
].hashval
= cpu_to_be32(mp
->m_dirnameops
->
1230 blp
[2 + i
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(
1231 (char *)dep
- (char *)hdr
));
1232 offset
= (int)((char *)(tagp
+ 1) - (char *)hdr
);
1233 if (++i
== sfp
->count
)
1236 sfep
= dp
->d_ops
->sf_nextentry(sfp
, sfep
);
1238 /* Done with the temporary buffer */
1241 * Sort the leaf entries by hash value.
1243 xfs_sort(blp
, be32_to_cpu(btp
->count
), sizeof(*blp
), xfs_dir2_block_sort
);
1245 * Log the leaf entry area and tail.
1246 * Already logged the header in data_init, ignore needlog.
1248 ASSERT(needscan
== 0);
1249 xfs_dir2_block_log_leaf(tp
, bp
, 0, be32_to_cpu(btp
->count
) - 1);
1250 xfs_dir2_block_log_tail(tp
, bp
);
1251 xfs_dir3_data_check(dp
, bp
);