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"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
31 #include "xfs_inode_item.h"
33 #include "xfs_buf_item.h"
35 #include "xfs_dir2_priv.h"
36 #include "xfs_error.h"
37 #include "xfs_trace.h"
38 #include "xfs_cksum.h"
39 #include "xfs_dinode.h"
42 * Local function prototypes.
44 static void xfs_dir2_block_log_leaf(xfs_trans_t
*tp
, struct xfs_buf
*bp
,
46 static void xfs_dir2_block_log_tail(xfs_trans_t
*tp
, struct xfs_buf
*bp
);
47 static int xfs_dir2_block_lookup_int(xfs_da_args_t
*args
, struct xfs_buf
**bpp
,
49 static int xfs_dir2_block_sort(const void *a
, const void *b
);
51 static xfs_dahash_t xfs_dir_hash_dot
, xfs_dir_hash_dotdot
;
54 * One-time startup routine called from xfs_init().
59 xfs_dir_hash_dot
= xfs_da_hashname((unsigned char *)".", 1);
60 xfs_dir_hash_dotdot
= xfs_da_hashname((unsigned char *)"..", 2);
64 xfs_dir3_block_verify(
67 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
68 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
70 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
71 if (hdr3
->magic
!= cpu_to_be32(XFS_DIR3_BLOCK_MAGIC
))
73 if (!uuid_equal(&hdr3
->uuid
, &mp
->m_sb
.sb_uuid
))
75 if (be64_to_cpu(hdr3
->blkno
) != bp
->b_bn
)
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_verify_cksum(bp
->b_addr
, BBTOB(bp
->b_length
),
94 XFS_DIR3_DATA_CRC_OFF
)) ||
95 !xfs_dir3_block_verify(bp
)) {
96 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, bp
->b_addr
);
97 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
102 xfs_dir3_block_write_verify(
105 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
106 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
107 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
109 if (!xfs_dir3_block_verify(bp
)) {
110 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, bp
->b_addr
);
111 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
115 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
119 hdr3
->lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
121 xfs_update_cksum(bp
->b_addr
, BBTOB(bp
->b_length
), XFS_DIR3_DATA_CRC_OFF
);
124 const struct xfs_buf_ops xfs_dir3_block_buf_ops
= {
125 .verify_read
= xfs_dir3_block_read_verify
,
126 .verify_write
= xfs_dir3_block_write_verify
,
131 struct xfs_trans
*tp
,
132 struct xfs_inode
*dp
,
133 struct xfs_buf
**bpp
)
135 struct xfs_mount
*mp
= dp
->i_mount
;
138 err
= xfs_da_read_buf(tp
, dp
, mp
->m_dirdatablk
, -1, bpp
,
139 XFS_DATA_FORK
, &xfs_dir3_block_buf_ops
);
141 xfs_trans_buf_set_type(tp
, *bpp
, XFS_BLFT_DIR_BLOCK_BUF
);
147 struct xfs_mount
*mp
,
148 struct xfs_trans
*tp
,
150 struct xfs_inode
*dp
)
152 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
154 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
155 xfs_trans_buf_set_type(tp
, bp
, XFS_BLFT_DIR_BLOCK_BUF
);
157 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
158 memset(hdr3
, 0, sizeof(*hdr3
));
159 hdr3
->magic
= cpu_to_be32(XFS_DIR3_BLOCK_MAGIC
);
160 hdr3
->blkno
= cpu_to_be64(bp
->b_bn
);
161 hdr3
->owner
= cpu_to_be64(dp
->i_ino
);
162 uuid_copy(&hdr3
->uuid
, &mp
->m_sb
.sb_uuid
);
166 hdr3
->magic
= cpu_to_be32(XFS_DIR2_BLOCK_MAGIC
);
170 xfs_dir2_block_need_space(
171 struct xfs_inode
*dp
,
172 struct xfs_dir2_data_hdr
*hdr
,
173 struct xfs_dir2_block_tail
*btp
,
174 struct xfs_dir2_leaf_entry
*blp
,
176 struct xfs_dir2_data_unused
**dupp
,
177 struct xfs_dir2_data_unused
**enddupp
,
181 struct xfs_dir2_data_free
*bf
;
183 struct xfs_dir2_data_unused
*dup
= NULL
;
184 struct xfs_dir2_data_unused
*enddup
= NULL
;
187 bf
= dp
->d_ops
->data_bestfree_p(hdr
);
190 * If there are stale entries we'll use one for the leaf.
193 if (be16_to_cpu(bf
[0].length
) >= len
) {
195 * The biggest entry enough to avoid compaction.
197 dup
= (xfs_dir2_data_unused_t
*)
198 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
203 * Will need to compact to make this work.
204 * Tag just before the first leaf entry.
207 tagp
= (__be16
*)blp
- 1;
209 /* Data object just before the first leaf entry. */
210 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
213 * If it's not free then the data will go where the
214 * leaf data starts now, if it works at all.
216 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
217 if (be16_to_cpu(dup
->length
) + (be32_to_cpu(btp
->stale
) - 1) *
218 (uint
)sizeof(*blp
) < len
)
220 } else if ((be32_to_cpu(btp
->stale
) - 1) * (uint
)sizeof(*blp
) < len
)
223 dup
= (xfs_dir2_data_unused_t
*)blp
;
228 * no stale entries, so just use free space.
229 * Tag just before the first leaf entry.
231 tagp
= (__be16
*)blp
- 1;
233 /* Data object just before the first leaf entry. */
234 enddup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
237 * If it's not free then can't do this add without cleaning up:
238 * the space before the first leaf entry needs to be free so it
239 * can be expanded to hold the pointer to the new entry.
241 if (be16_to_cpu(enddup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
243 * Check out the biggest freespace and see if it's the same one.
245 dup
= (xfs_dir2_data_unused_t
*)
246 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
249 * Not the same free entry, just check its length.
251 if (be16_to_cpu(dup
->length
) < len
)
257 * It is the biggest freespace, can it hold the leaf too?
259 if (be16_to_cpu(dup
->length
) < len
+ (uint
)sizeof(*blp
)) {
261 * Yes, use the second-largest entry instead if it works.
263 if (be16_to_cpu(bf
[1].length
) >= len
)
264 dup
= (xfs_dir2_data_unused_t
*)
265 ((char *)hdr
+ be16_to_cpu(bf
[1].offset
));
277 * compact the leaf entries.
278 * Leave the highest-numbered stale entry stale.
279 * XXX should be the one closest to mid but mid is not yet computed.
282 xfs_dir2_block_compact(
283 struct xfs_trans
*tp
,
284 struct xfs_inode
*dp
,
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(tp
, dp
, 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 blp
+= be32_to_cpu(btp
->stale
) - 1;
322 btp
->stale
= cpu_to_be32(1);
324 * If we now need to rebuild the bestfree map, do so.
325 * This needs to happen before the next call to use_free.
328 xfs_dir2_data_freescan(dp
, hdr
, needlog
);
332 * Add an entry to a block directory.
335 xfs_dir2_block_addname(
336 xfs_da_args_t
*args
) /* directory op arguments */
338 xfs_dir2_data_hdr_t
*hdr
; /* block header */
339 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
340 struct xfs_buf
*bp
; /* buffer for block */
341 xfs_dir2_block_tail_t
*btp
; /* block tail */
342 int compact
; /* need to compact leaf ents */
343 xfs_dir2_data_entry_t
*dep
; /* block data entry */
344 xfs_inode_t
*dp
; /* directory inode */
345 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
346 int error
; /* error return value */
347 xfs_dir2_data_unused_t
*enddup
=NULL
; /* unused at end of data */
348 xfs_dahash_t hash
; /* hash value of found entry */
349 int high
; /* high index for binary srch */
350 int highstale
; /* high stale index */
351 int lfloghigh
=0; /* last final leaf to log */
352 int lfloglow
=0; /* first final leaf to log */
353 int len
; /* length of the new entry */
354 int low
; /* low index for binary srch */
355 int lowstale
; /* low stale index */
356 int mid
=0; /* midpoint for binary srch */
357 xfs_mount_t
*mp
; /* filesystem mount point */
358 int needlog
; /* need to log header */
359 int needscan
; /* need to rescan freespace */
360 __be16
*tagp
; /* pointer to tag value */
361 xfs_trans_t
*tp
; /* transaction structure */
363 trace_xfs_dir2_block_addname(args
);
369 /* Read the (one and only) directory block into bp. */
370 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
374 len
= dp
->d_ops
->data_entsize(args
->namelen
);
377 * Set up pointers to parts of the block.
380 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
381 blp
= xfs_dir2_block_leaf_p(btp
);
384 * Find out if we can reuse stale entries or whether we need extra
385 * space for entry and new leaf.
387 xfs_dir2_block_need_space(dp
, hdr
, btp
, blp
, &tagp
, &dup
,
388 &enddup
, &compact
, len
);
391 * Done everything we need for a space check now.
393 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
394 xfs_trans_brelse(tp
, bp
);
396 return XFS_ERROR(ENOSPC
);
401 * If we don't have space for the new entry & leaf ...
404 /* Don't have a space reservation: return no-space. */
405 if (args
->total
== 0)
406 return XFS_ERROR(ENOSPC
);
408 * Convert to the next larger format.
409 * Then add the new entry in that format.
411 error
= xfs_dir2_block_to_leaf(args
, bp
);
414 return xfs_dir2_leaf_addname(args
);
417 needlog
= needscan
= 0;
420 * If need to compact the leaf entries, do it now.
423 xfs_dir2_block_compact(tp
, dp
, bp
, hdr
, btp
, blp
, &needlog
,
424 &lfloghigh
, &lfloglow
);
425 /* recalculate blp post-compaction */
426 blp
= xfs_dir2_block_leaf_p(btp
);
427 } else if (btp
->stale
) {
429 * Set leaf logging boundaries to impossible state.
430 * For the no-stale case they're set explicitly.
432 lfloglow
= be32_to_cpu(btp
->count
);
437 * Find the slot that's first lower than our hash value, -1 if none.
439 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; low
<= high
; ) {
440 mid
= (low
+ high
) >> 1;
441 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
443 if (hash
< args
->hashval
)
448 while (mid
>= 0 && be32_to_cpu(blp
[mid
].hashval
) >= args
->hashval
) {
452 * No stale entries, will use enddup space to hold new leaf.
456 * Mark the space needed for the new leaf entry, now in use.
458 xfs_dir2_data_use_free(tp
, dp
, bp
, enddup
,
459 (xfs_dir2_data_aoff_t
)
460 ((char *)enddup
- (char *)hdr
+ be16_to_cpu(enddup
->length
) -
462 (xfs_dir2_data_aoff_t
)sizeof(*blp
),
463 &needlog
, &needscan
);
465 * Update the tail (entry count).
467 be32_add_cpu(&btp
->count
, 1);
469 * If we now need to rebuild the bestfree map, do so.
470 * This needs to happen before the next call to use_free.
473 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
477 * Adjust pointer to the first leaf entry, we're about to move
478 * the table up one to open up space for the new leaf entry.
479 * Then adjust our index to match.
484 memmove(blp
, &blp
[1], mid
* sizeof(*blp
));
489 * Use a stale leaf for our new entry.
494 blp
[lowstale
].address
!=
495 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
498 for (highstale
= mid
+ 1;
499 highstale
< be32_to_cpu(btp
->count
) &&
500 blp
[highstale
].address
!=
501 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
) &&
502 (lowstale
< 0 || mid
- lowstale
> highstale
- mid
);
506 * Move entries toward the low-numbered stale entry.
509 (highstale
== be32_to_cpu(btp
->count
) ||
510 mid
- lowstale
<= highstale
- mid
)) {
512 memmove(&blp
[lowstale
], &blp
[lowstale
+ 1],
513 (mid
- lowstale
) * sizeof(*blp
));
514 lfloglow
= MIN(lowstale
, lfloglow
);
515 lfloghigh
= MAX(mid
, lfloghigh
);
518 * Move entries toward the high-numbered stale entry.
521 ASSERT(highstale
< be32_to_cpu(btp
->count
));
524 memmove(&blp
[mid
+ 1], &blp
[mid
],
525 (highstale
- mid
) * sizeof(*blp
));
526 lfloglow
= MIN(mid
, lfloglow
);
527 lfloghigh
= MAX(highstale
, lfloghigh
);
529 be32_add_cpu(&btp
->stale
, -1);
532 * Point to the new data entry.
534 dep
= (xfs_dir2_data_entry_t
*)dup
;
536 * Fill in the leaf entry.
538 blp
[mid
].hashval
= cpu_to_be32(args
->hashval
);
539 blp
[mid
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
540 (char *)dep
- (char *)hdr
));
541 xfs_dir2_block_log_leaf(tp
, bp
, lfloglow
, lfloghigh
);
543 * Mark space for the data entry used.
545 xfs_dir2_data_use_free(tp
, dp
, bp
, dup
,
546 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
547 (xfs_dir2_data_aoff_t
)len
, &needlog
, &needscan
);
549 * Create the new data entry.
551 dep
->inumber
= cpu_to_be64(args
->inumber
);
552 dep
->namelen
= args
->namelen
;
553 memcpy(dep
->name
, args
->name
, args
->namelen
);
554 dp
->d_ops
->data_put_ftype(dep
, args
->filetype
);
555 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
556 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
558 * Clean up the bestfree array and log the header, tail, and entry.
561 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
563 xfs_dir2_data_log_header(tp
, dp
, bp
);
564 xfs_dir2_block_log_tail(tp
, bp
);
565 xfs_dir2_data_log_entry(tp
, dp
, bp
, dep
);
566 xfs_dir3_data_check(dp
, bp
);
571 * Log leaf entries from the block.
574 xfs_dir2_block_log_leaf(
575 xfs_trans_t
*tp
, /* transaction structure */
576 struct xfs_buf
*bp
, /* block buffer */
577 int first
, /* index of first logged leaf */
578 int last
) /* index of last logged leaf */
580 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
581 xfs_dir2_leaf_entry_t
*blp
;
582 xfs_dir2_block_tail_t
*btp
;
584 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
, hdr
);
585 blp
= xfs_dir2_block_leaf_p(btp
);
586 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)&blp
[first
] - (char *)hdr
),
587 (uint
)((char *)&blp
[last
+ 1] - (char *)hdr
- 1));
591 * Log the block tail.
594 xfs_dir2_block_log_tail(
595 xfs_trans_t
*tp
, /* transaction structure */
596 struct xfs_buf
*bp
) /* block buffer */
598 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
599 xfs_dir2_block_tail_t
*btp
;
601 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
, hdr
);
602 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)btp
- (char *)hdr
),
603 (uint
)((char *)(btp
+ 1) - (char *)hdr
- 1));
607 * Look up an entry in the block. This is the external routine,
608 * xfs_dir2_block_lookup_int does the real work.
611 xfs_dir2_block_lookup(
612 xfs_da_args_t
*args
) /* dir lookup arguments */
614 xfs_dir2_data_hdr_t
*hdr
; /* block header */
615 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
616 struct xfs_buf
*bp
; /* block buffer */
617 xfs_dir2_block_tail_t
*btp
; /* block tail */
618 xfs_dir2_data_entry_t
*dep
; /* block data entry */
619 xfs_inode_t
*dp
; /* incore inode */
620 int ent
; /* entry index */
621 int error
; /* error return value */
622 xfs_mount_t
*mp
; /* filesystem mount point */
624 trace_xfs_dir2_block_lookup(args
);
627 * Get the buffer, look up the entry.
628 * If not found (ENOENT) then return, have no buffer.
630 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
)))
635 xfs_dir3_data_check(dp
, bp
);
636 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
637 blp
= xfs_dir2_block_leaf_p(btp
);
639 * Get the offset from the leaf entry, to point to the data.
641 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
642 xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(blp
[ent
].address
)));
644 * Fill in inode number, CI name if appropriate, release the block.
646 args
->inumber
= be64_to_cpu(dep
->inumber
);
647 args
->filetype
= dp
->d_ops
->data_get_ftype(dep
);
648 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
649 xfs_trans_brelse(args
->trans
, bp
);
650 return XFS_ERROR(error
);
654 * Internal block lookup routine.
656 static int /* error */
657 xfs_dir2_block_lookup_int(
658 xfs_da_args_t
*args
, /* dir lookup arguments */
659 struct xfs_buf
**bpp
, /* returned block buffer */
660 int *entno
) /* returned entry number */
662 xfs_dir2_dataptr_t addr
; /* data entry address */
663 xfs_dir2_data_hdr_t
*hdr
; /* block header */
664 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
665 struct xfs_buf
*bp
; /* block buffer */
666 xfs_dir2_block_tail_t
*btp
; /* block tail */
667 xfs_dir2_data_entry_t
*dep
; /* block data entry */
668 xfs_inode_t
*dp
; /* incore inode */
669 int error
; /* error return value */
670 xfs_dahash_t hash
; /* found hash value */
671 int high
; /* binary search high index */
672 int low
; /* binary search low index */
673 int mid
; /* binary search current idx */
674 xfs_mount_t
*mp
; /* filesystem mount point */
675 xfs_trans_t
*tp
; /* transaction pointer */
676 enum xfs_dacmp cmp
; /* comparison result */
682 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
687 xfs_dir3_data_check(dp
, bp
);
688 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
689 blp
= xfs_dir2_block_leaf_p(btp
);
691 * Loop doing a binary search for our hash value.
692 * Find our entry, ENOENT if it's not there.
694 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; ; ) {
696 mid
= (low
+ high
) >> 1;
697 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
699 if (hash
< args
->hashval
)
704 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
705 xfs_trans_brelse(tp
, bp
);
706 return XFS_ERROR(ENOENT
);
710 * Back up to the first one with the right hash value.
712 while (mid
> 0 && be32_to_cpu(blp
[mid
- 1].hashval
) == args
->hashval
) {
716 * Now loop forward through all the entries with the
717 * right hash value looking for our name.
720 if ((addr
= be32_to_cpu(blp
[mid
].address
)) == XFS_DIR2_NULL_DATAPTR
)
723 * Get pointer to the entry from the leaf.
725 dep
= (xfs_dir2_data_entry_t
*)
726 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, addr
));
728 * Compare name and if it's an exact match, return the index
729 * and buffer. If it's the first case-insensitive match, store
730 * the index and buffer and continue looking for an exact match.
732 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
733 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
734 args
->cmpresult
= cmp
;
737 if (cmp
== XFS_CMP_EXACT
)
740 } while (++mid
< be32_to_cpu(btp
->count
) &&
741 be32_to_cpu(blp
[mid
].hashval
) == hash
);
743 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
745 * Here, we can only be doing a lookup (not a rename or replace).
746 * If a case-insensitive match was found earlier, return success.
748 if (args
->cmpresult
== XFS_CMP_CASE
)
751 * No match, release the buffer and return ENOENT.
753 xfs_trans_brelse(tp
, bp
);
754 return XFS_ERROR(ENOENT
);
758 * Remove an entry from a block format directory.
759 * If that makes the block small enough to fit in shortform, transform it.
762 xfs_dir2_block_removename(
763 xfs_da_args_t
*args
) /* directory operation args */
765 xfs_dir2_data_hdr_t
*hdr
; /* block header */
766 xfs_dir2_leaf_entry_t
*blp
; /* block leaf pointer */
767 struct xfs_buf
*bp
; /* block buffer */
768 xfs_dir2_block_tail_t
*btp
; /* block tail */
769 xfs_dir2_data_entry_t
*dep
; /* block data entry */
770 xfs_inode_t
*dp
; /* incore inode */
771 int ent
; /* block leaf entry index */
772 int error
; /* error return value */
773 xfs_mount_t
*mp
; /* filesystem mount point */
774 int needlog
; /* need to log block header */
775 int needscan
; /* need to fixup bestfree */
776 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
777 int size
; /* shortform size */
778 xfs_trans_t
*tp
; /* transaction pointer */
780 trace_xfs_dir2_block_removename(args
);
783 * Look up the entry in the block. Gets the buffer and entry index.
784 * It will always be there, the vnodeops level does a lookup first.
786 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
793 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
794 blp
= xfs_dir2_block_leaf_p(btp
);
796 * Point to the data entry using the leaf entry.
798 dep
= (xfs_dir2_data_entry_t
*)
799 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(blp
[ent
].address
)));
801 * Mark the data entry's space free.
803 needlog
= needscan
= 0;
804 xfs_dir2_data_make_free(tp
, dp
, bp
,
805 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)hdr
),
806 dp
->d_ops
->data_entsize(dep
->namelen
), &needlog
, &needscan
);
808 * Fix up the block tail.
810 be32_add_cpu(&btp
->stale
, 1);
811 xfs_dir2_block_log_tail(tp
, bp
);
813 * Remove the leaf entry by marking it stale.
815 blp
[ent
].address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
816 xfs_dir2_block_log_leaf(tp
, bp
, ent
, ent
);
818 * Fix up bestfree, log the header if necessary.
821 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
823 xfs_dir2_data_log_header(tp
, dp
, bp
);
824 xfs_dir3_data_check(dp
, bp
);
826 * See if the size as a shortform is good enough.
828 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
829 if (size
> XFS_IFORK_DSIZE(dp
))
833 * If it works, do the conversion.
835 return xfs_dir2_block_to_sf(args
, bp
, size
, &sfh
);
839 * Replace an entry in a V2 block directory.
840 * Change the inode number to the new value.
843 xfs_dir2_block_replace(
844 xfs_da_args_t
*args
) /* directory operation args */
846 xfs_dir2_data_hdr_t
*hdr
; /* block header */
847 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
848 struct xfs_buf
*bp
; /* block buffer */
849 xfs_dir2_block_tail_t
*btp
; /* block tail */
850 xfs_dir2_data_entry_t
*dep
; /* block data entry */
851 xfs_inode_t
*dp
; /* incore inode */
852 int ent
; /* leaf entry index */
853 int error
; /* error return value */
854 xfs_mount_t
*mp
; /* filesystem mount point */
856 trace_xfs_dir2_block_replace(args
);
859 * Lookup the entry in the directory. Get buffer and entry index.
860 * This will always succeed since the caller has already done a lookup.
862 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
868 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
869 blp
= xfs_dir2_block_leaf_p(btp
);
871 * Point to the data entry we need to change.
873 dep
= (xfs_dir2_data_entry_t
*)
874 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(blp
[ent
].address
)));
875 ASSERT(be64_to_cpu(dep
->inumber
) != args
->inumber
);
877 * Change the inode number to the new value.
879 dep
->inumber
= cpu_to_be64(args
->inumber
);
880 dp
->d_ops
->data_put_ftype(dep
, args
->filetype
);
881 xfs_dir2_data_log_entry(args
->trans
, dp
, bp
, dep
);
882 xfs_dir3_data_check(dp
, bp
);
887 * Qsort comparison routine for the block leaf entries.
889 static int /* sort order */
891 const void *a
, /* first leaf entry */
892 const void *b
) /* second leaf entry */
894 const xfs_dir2_leaf_entry_t
*la
; /* first leaf entry */
895 const xfs_dir2_leaf_entry_t
*lb
; /* second leaf entry */
899 return be32_to_cpu(la
->hashval
) < be32_to_cpu(lb
->hashval
) ? -1 :
900 (be32_to_cpu(la
->hashval
) > be32_to_cpu(lb
->hashval
) ? 1 : 0);
904 * Convert a V2 leaf directory to a V2 block directory if possible.
907 xfs_dir2_leaf_to_block(
908 xfs_da_args_t
*args
, /* operation arguments */
909 struct xfs_buf
*lbp
, /* leaf buffer */
910 struct xfs_buf
*dbp
) /* data buffer */
912 __be16
*bestsp
; /* leaf bests table */
913 xfs_dir2_data_hdr_t
*hdr
; /* block header */
914 xfs_dir2_block_tail_t
*btp
; /* block tail */
915 xfs_inode_t
*dp
; /* incore directory inode */
916 xfs_dir2_data_unused_t
*dup
; /* unused data entry */
917 int error
; /* error return value */
918 int from
; /* leaf from index */
919 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
920 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
921 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
922 xfs_mount_t
*mp
; /* file system mount point */
923 int needlog
; /* need to log data header */
924 int needscan
; /* need to scan for bestfree */
925 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
926 int size
; /* bytes used */
927 __be16
*tagp
; /* end of entry (tag) */
928 int to
; /* block/leaf to index */
929 xfs_trans_t
*tp
; /* transaction pointer */
930 struct xfs_dir2_leaf_entry
*ents
;
931 struct xfs_dir3_icleaf_hdr leafhdr
;
933 trace_xfs_dir2_leaf_to_block(args
);
939 dp
->d_ops
->leaf_hdr_from_disk(&leafhdr
, leaf
);
940 ents
= dp
->d_ops
->leaf_ents_p(leaf
);
941 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
943 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAF1_MAGIC
||
944 leafhdr
.magic
== XFS_DIR3_LEAF1_MAGIC
);
946 * If there are data blocks other than the first one, take this
947 * opportunity to remove trailing empty data blocks that may have
948 * been left behind during no-space-reservation operations.
949 * These will show up in the leaf bests table.
951 while (dp
->i_d
.di_size
> mp
->m_dirblksize
) {
954 hdrsz
= dp
->d_ops
->data_entry_offset
;
955 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
956 if (be16_to_cpu(bestsp
[be32_to_cpu(ltp
->bestcount
) - 1]) ==
957 mp
->m_dirblksize
- hdrsz
) {
959 xfs_dir2_leaf_trim_data(args
, lbp
,
960 (xfs_dir2_db_t
)(be32_to_cpu(ltp
->bestcount
) - 1))))
966 * Read the data block if we don't already have it, give up if it fails.
969 error
= xfs_dir3_data_read(tp
, dp
, mp
->m_dirdatablk
, -1, &dbp
);
974 ASSERT(hdr
->magic
== cpu_to_be32(XFS_DIR2_DATA_MAGIC
) ||
975 hdr
->magic
== cpu_to_be32(XFS_DIR3_DATA_MAGIC
));
978 * Size of the "leaf" area in the block.
980 size
= (uint
)sizeof(xfs_dir2_block_tail_t
) +
981 (uint
)sizeof(*lep
) * (leafhdr
.count
- leafhdr
.stale
);
983 * Look at the last data entry.
985 tagp
= (__be16
*)((char *)hdr
+ mp
->m_dirblksize
) - 1;
986 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
988 * If it's not free or is too short we can't do it.
990 if (be16_to_cpu(dup
->freetag
) != XFS_DIR2_DATA_FREE_TAG
||
991 be16_to_cpu(dup
->length
) < size
)
995 * Start converting it to block form.
997 xfs_dir3_block_init(mp
, tp
, dbp
, dp
);
1002 * Use up the space at the end of the block (blp/btp).
1004 xfs_dir2_data_use_free(tp
, dp
, dbp
, dup
, mp
->m_dirblksize
- size
, size
,
1005 &needlog
, &needscan
);
1007 * Initialize the block tail.
1009 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
1010 btp
->count
= cpu_to_be32(leafhdr
.count
- leafhdr
.stale
);
1012 xfs_dir2_block_log_tail(tp
, dbp
);
1014 * Initialize the block leaf area. We compact out stale entries.
1016 lep
= xfs_dir2_block_leaf_p(btp
);
1017 for (from
= to
= 0; from
< leafhdr
.count
; from
++) {
1018 if (ents
[from
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
1020 lep
[to
++] = ents
[from
];
1022 ASSERT(to
== be32_to_cpu(btp
->count
));
1023 xfs_dir2_block_log_leaf(tp
, dbp
, 0, be32_to_cpu(btp
->count
) - 1);
1025 * Scan the bestfree if we need it and log the data block header.
1028 xfs_dir2_data_freescan(dp
, hdr
, &needlog
);
1030 xfs_dir2_data_log_header(tp
, dp
, dbp
);
1032 * Pitch the old leaf block.
1034 error
= xfs_da_shrink_inode(args
, mp
->m_dirleafblk
, lbp
);
1039 * Now see if the resulting block can be shrunken to shortform.
1041 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
1042 if (size
> XFS_IFORK_DSIZE(dp
))
1045 return xfs_dir2_block_to_sf(args
, dbp
, size
, &sfh
);
1049 * Convert the shortform directory to block form.
1052 xfs_dir2_sf_to_block(
1053 xfs_da_args_t
*args
) /* operation arguments */
1055 xfs_dir2_db_t blkno
; /* dir-relative block # (0) */
1056 xfs_dir2_data_hdr_t
*hdr
; /* block header */
1057 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
1058 struct xfs_buf
*bp
; /* block buffer */
1059 xfs_dir2_block_tail_t
*btp
; /* block tail pointer */
1060 xfs_dir2_data_entry_t
*dep
; /* data entry pointer */
1061 xfs_inode_t
*dp
; /* incore directory inode */
1062 int dummy
; /* trash */
1063 xfs_dir2_data_unused_t
*dup
; /* unused entry pointer */
1064 int endoffset
; /* end of data objects */
1065 int error
; /* error return value */
1067 xfs_mount_t
*mp
; /* filesystem mount point */
1068 int needlog
; /* need to log block header */
1069 int needscan
; /* need to scan block freespc */
1070 int newoffset
; /* offset from current entry */
1071 int offset
; /* target block offset */
1072 xfs_dir2_sf_entry_t
*sfep
; /* sf entry pointer */
1073 xfs_dir2_sf_hdr_t
*oldsfp
; /* old shortform header */
1074 xfs_dir2_sf_hdr_t
*sfp
; /* shortform header */
1075 __be16
*tagp
; /* end of data entry */
1076 xfs_trans_t
*tp
; /* transaction pointer */
1077 struct xfs_name name
;
1078 struct xfs_ifork
*ifp
;
1080 trace_xfs_dir2_sf_to_block(args
);
1085 ifp
= XFS_IFORK_PTR(dp
, XFS_DATA_FORK
);
1086 ASSERT(ifp
->if_flags
& XFS_IFINLINE
);
1088 * Bomb out if the shortform directory is way too short.
1090 if (dp
->i_d
.di_size
< offsetof(xfs_dir2_sf_hdr_t
, parent
)) {
1091 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1092 return XFS_ERROR(EIO
);
1095 oldsfp
= (xfs_dir2_sf_hdr_t
*)ifp
->if_u1
.if_data
;
1097 ASSERT(ifp
->if_bytes
== dp
->i_d
.di_size
);
1098 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
1099 ASSERT(dp
->i_d
.di_size
>= xfs_dir2_sf_hdr_size(oldsfp
->i8count
));
1100 ASSERT(dp
->i_d
.di_nextents
== 0);
1103 * Copy the directory into a temporary buffer.
1104 * Then pitch the incore inode data so we can make extents.
1106 sfp
= kmem_alloc(ifp
->if_bytes
, KM_SLEEP
);
1107 memcpy(sfp
, oldsfp
, ifp
->if_bytes
);
1109 xfs_idata_realloc(dp
, -ifp
->if_bytes
, XFS_DATA_FORK
);
1110 xfs_bmap_local_to_extents_empty(dp
, XFS_DATA_FORK
);
1111 dp
->i_d
.di_size
= 0;
1114 * Add block 0 to the inode.
1116 error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
, &blkno
);
1122 * Initialize the data block, then convert it to block format.
1124 error
= xfs_dir3_data_init(args
, blkno
, &bp
);
1129 xfs_dir3_block_init(mp
, tp
, bp
, dp
);
1133 * Compute size of block "tail" area.
1135 i
= (uint
)sizeof(*btp
) +
1136 (sfp
->count
+ 2) * (uint
)sizeof(xfs_dir2_leaf_entry_t
);
1138 * The whole thing is initialized to free by the init routine.
1139 * Say we're using the leaf and tail area.
1141 dup
= dp
->d_ops
->data_unused_p(hdr
);
1142 needlog
= needscan
= 0;
1143 xfs_dir2_data_use_free(tp
, dp
, bp
, dup
, mp
->m_dirblksize
- i
, i
, &needlog
,
1145 ASSERT(needscan
== 0);
1149 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
1150 btp
->count
= cpu_to_be32(sfp
->count
+ 2); /* ., .. */
1152 blp
= xfs_dir2_block_leaf_p(btp
);
1153 endoffset
= (uint
)((char *)blp
- (char *)hdr
);
1155 * Remove the freespace, we'll manage it.
1157 xfs_dir2_data_use_free(tp
, dp
, bp
, dup
,
1158 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
1159 be16_to_cpu(dup
->length
), &needlog
, &needscan
);
1161 * Create entry for .
1163 dep
= dp
->d_ops
->data_dot_entry_p(hdr
);
1164 dep
->inumber
= cpu_to_be64(dp
->i_ino
);
1167 dp
->d_ops
->data_put_ftype(dep
, XFS_DIR3_FT_DIR
);
1168 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
1169 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1170 xfs_dir2_data_log_entry(tp
, dp
, bp
, dep
);
1171 blp
[0].hashval
= cpu_to_be32(xfs_dir_hash_dot
);
1172 blp
[0].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
1173 (char *)dep
- (char *)hdr
));
1175 * Create entry for ..
1177 dep
= dp
->d_ops
->data_dotdot_entry_p(hdr
);
1178 dep
->inumber
= cpu_to_be64(dp
->d_ops
->sf_get_parent_ino(sfp
));
1180 dep
->name
[0] = dep
->name
[1] = '.';
1181 dp
->d_ops
->data_put_ftype(dep
, XFS_DIR3_FT_DIR
);
1182 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
1183 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1184 xfs_dir2_data_log_entry(tp
, dp
, bp
, dep
);
1185 blp
[1].hashval
= cpu_to_be32(xfs_dir_hash_dotdot
);
1186 blp
[1].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
1187 (char *)dep
- (char *)hdr
));
1188 offset
= dp
->d_ops
->data_first_offset
;
1190 * Loop over existing entries, stuff them in.
1196 sfep
= xfs_dir2_sf_firstentry(sfp
);
1198 * Need to preserve the existing offset values in the sf directory.
1199 * Insert holes (unused entries) where necessary.
1201 while (offset
< endoffset
) {
1203 * sfep is null when we reach the end of the list.
1206 newoffset
= endoffset
;
1208 newoffset
= xfs_dir2_sf_get_offset(sfep
);
1210 * There should be a hole here, make one.
1212 if (offset
< newoffset
) {
1213 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ offset
);
1214 dup
->freetag
= cpu_to_be16(XFS_DIR2_DATA_FREE_TAG
);
1215 dup
->length
= cpu_to_be16(newoffset
- offset
);
1216 *xfs_dir2_data_unused_tag_p(dup
) = cpu_to_be16(
1217 ((char *)dup
- (char *)hdr
));
1218 xfs_dir2_data_log_unused(tp
, bp
, dup
);
1219 xfs_dir2_data_freeinsert(hdr
,
1220 dp
->d_ops
->data_bestfree_p(hdr
),
1222 offset
+= be16_to_cpu(dup
->length
);
1226 * Copy a real entry.
1228 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+ newoffset
);
1229 dep
->inumber
= cpu_to_be64(dp
->d_ops
->sf_get_ino(sfp
, sfep
));
1230 dep
->namelen
= sfep
->namelen
;
1231 dp
->d_ops
->data_put_ftype(dep
, dp
->d_ops
->sf_get_ftype(sfep
));
1232 memcpy(dep
->name
, sfep
->name
, dep
->namelen
);
1233 tagp
= dp
->d_ops
->data_entry_tag_p(dep
);
1234 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1235 xfs_dir2_data_log_entry(tp
, dp
, bp
, dep
);
1236 name
.name
= sfep
->name
;
1237 name
.len
= sfep
->namelen
;
1238 blp
[2 + i
].hashval
= cpu_to_be32(mp
->m_dirnameops
->
1240 blp
[2 + i
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
1241 (char *)dep
- (char *)hdr
));
1242 offset
= (int)((char *)(tagp
+ 1) - (char *)hdr
);
1243 if (++i
== sfp
->count
)
1246 sfep
= dp
->d_ops
->sf_nextentry(sfp
, sfep
);
1248 /* Done with the temporary buffer */
1251 * Sort the leaf entries by hash value.
1253 xfs_sort(blp
, be32_to_cpu(btp
->count
), sizeof(*blp
), xfs_dir2_block_sort
);
1255 * Log the leaf entry area and tail.
1256 * Already logged the header in data_init, ignore needlog.
1258 ASSERT(needscan
== 0);
1259 xfs_dir2_block_log_leaf(tp
, bp
, 0, be32_to_cpu(btp
->count
) - 1);
1260 xfs_dir2_block_log_tail(tp
, bp
);
1261 xfs_dir3_data_check(dp
, bp
);