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_types.h"
23 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_dinode.h"
30 #include "xfs_inode.h"
31 #include "xfs_inode_item.h"
33 #include "xfs_buf_item.h"
35 #include "xfs_dir2_format.h"
36 #include "xfs_dir2_priv.h"
37 #include "xfs_error.h"
38 #include "xfs_trace.h"
39 #include "xfs_cksum.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_dir2_data_hdr
*hdr
,
172 struct xfs_dir2_block_tail
*btp
,
173 struct xfs_dir2_leaf_entry
*blp
,
175 struct xfs_dir2_data_unused
**dupp
,
176 struct xfs_dir2_data_unused
**enddupp
,
180 struct xfs_dir2_data_free
*bf
;
182 struct xfs_dir2_data_unused
*dup
= NULL
;
183 struct xfs_dir2_data_unused
*enddup
= NULL
;
186 bf
= xfs_dir3_data_bestfree_p(hdr
);
189 * If there are stale entries we'll use one for the leaf.
192 if (be16_to_cpu(bf
[0].length
) >= len
) {
194 * The biggest entry enough to avoid compaction.
196 dup
= (xfs_dir2_data_unused_t
*)
197 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
202 * Will need to compact to make this work.
203 * Tag just before the first leaf entry.
206 tagp
= (__be16
*)blp
- 1;
208 /* Data object just before the first leaf entry. */
209 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
212 * If it's not free then the data will go where the
213 * leaf data starts now, if it works at all.
215 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
216 if (be16_to_cpu(dup
->length
) + (be32_to_cpu(btp
->stale
) - 1) *
217 (uint
)sizeof(*blp
) < len
)
219 } else if ((be32_to_cpu(btp
->stale
) - 1) * (uint
)sizeof(*blp
) < len
)
222 dup
= (xfs_dir2_data_unused_t
*)blp
;
227 * no stale entries, so just use free space.
228 * Tag just before the first leaf entry.
230 tagp
= (__be16
*)blp
- 1;
232 /* Data object just before the first leaf entry. */
233 enddup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
236 * If it's not free then can't do this add without cleaning up:
237 * the space before the first leaf entry needs to be free so it
238 * can be expanded to hold the pointer to the new entry.
240 if (be16_to_cpu(enddup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
242 * Check out the biggest freespace and see if it's the same one.
244 dup
= (xfs_dir2_data_unused_t
*)
245 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
248 * Not the same free entry, just check its length.
250 if (be16_to_cpu(dup
->length
) < len
)
256 * It is the biggest freespace, can it hold the leaf too?
258 if (be16_to_cpu(dup
->length
) < len
+ (uint
)sizeof(*blp
)) {
260 * Yes, use the second-largest entry instead if it works.
262 if (be16_to_cpu(bf
[1].length
) >= len
)
263 dup
= (xfs_dir2_data_unused_t
*)
264 ((char *)hdr
+ be16_to_cpu(bf
[1].offset
));
276 * compact the leaf entries.
277 * Leave the highest-numbered stale entry stale.
278 * XXX should be the one closest to mid but mid is not yet computed.
281 xfs_dir2_block_compact(
282 struct xfs_trans
*tp
,
284 struct xfs_dir2_data_hdr
*hdr
,
285 struct xfs_dir2_block_tail
*btp
,
286 struct xfs_dir2_leaf_entry
*blp
,
291 int fromidx
; /* source leaf index */
292 int toidx
; /* target leaf index */
294 int highstale
; /* high stale index */
296 fromidx
= toidx
= be32_to_cpu(btp
->count
) - 1;
297 highstale
= *lfloghigh
= -1;
298 for (; fromidx
>= 0; fromidx
--) {
299 if (blp
[fromidx
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
)) {
303 if (*lfloghigh
== -1)
309 blp
[toidx
] = blp
[fromidx
];
312 *lfloglow
= toidx
+ 1 - (be32_to_cpu(btp
->stale
) - 1);
313 *lfloghigh
-= be32_to_cpu(btp
->stale
) - 1;
314 be32_add_cpu(&btp
->count
, -(be32_to_cpu(btp
->stale
) - 1));
315 xfs_dir2_data_make_free(tp
, bp
,
316 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)hdr
),
317 (xfs_dir2_data_aoff_t
)((be32_to_cpu(btp
->stale
) - 1) * sizeof(*blp
)),
319 blp
+= be32_to_cpu(btp
->stale
) - 1;
320 btp
->stale
= cpu_to_be32(1);
322 * If we now need to rebuild the bestfree map, do so.
323 * This needs to happen before the next call to use_free.
326 xfs_dir2_data_freescan(tp
->t_mountp
, hdr
, needlog
);
330 * Add an entry to a block directory.
333 xfs_dir2_block_addname(
334 xfs_da_args_t
*args
) /* directory op arguments */
336 xfs_dir2_data_hdr_t
*hdr
; /* block header */
337 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
338 struct xfs_buf
*bp
; /* buffer for block */
339 xfs_dir2_block_tail_t
*btp
; /* block tail */
340 int compact
; /* need to compact leaf ents */
341 xfs_dir2_data_entry_t
*dep
; /* block data entry */
342 xfs_inode_t
*dp
; /* directory inode */
343 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
344 int error
; /* error return value */
345 xfs_dir2_data_unused_t
*enddup
=NULL
; /* unused at end of data */
346 xfs_dahash_t hash
; /* hash value of found entry */
347 int high
; /* high index for binary srch */
348 int highstale
; /* high stale index */
349 int lfloghigh
=0; /* last final leaf to log */
350 int lfloglow
=0; /* first final leaf to log */
351 int len
; /* length of the new entry */
352 int low
; /* low index for binary srch */
353 int lowstale
; /* low stale index */
354 int mid
=0; /* midpoint for binary srch */
355 xfs_mount_t
*mp
; /* filesystem mount point */
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
);
367 /* Read the (one and only) directory block into bp. */
368 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
372 len
= xfs_dir2_data_entsize(args
->namelen
);
375 * Set up pointers to parts of the block.
378 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
379 blp
= xfs_dir2_block_leaf_p(btp
);
382 * Find out if we can reuse stale entries or whether we need extra
383 * space for entry and new leaf.
385 xfs_dir2_block_need_space(hdr
, btp
, blp
, &tagp
, &dup
,
386 &enddup
, &compact
, len
);
389 * Done everything we need for a space check now.
391 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
392 xfs_trans_brelse(tp
, bp
);
394 return XFS_ERROR(ENOSPC
);
399 * If we don't have space for the new entry & leaf ...
402 /* Don't have a space reservation: return no-space. */
403 if (args
->total
== 0)
404 return XFS_ERROR(ENOSPC
);
406 * Convert to the next larger format.
407 * Then add the new entry in that format.
409 error
= xfs_dir2_block_to_leaf(args
, bp
);
412 return xfs_dir2_leaf_addname(args
);
415 needlog
= needscan
= 0;
418 * If need to compact the leaf entries, do it now.
421 xfs_dir2_block_compact(tp
, bp
, hdr
, btp
, blp
, &needlog
,
422 &lfloghigh
, &lfloglow
);
423 /* recalculate blp post-compaction */
424 blp
= xfs_dir2_block_leaf_p(btp
);
425 } else if (btp
->stale
) {
427 * Set leaf logging boundaries to impossible state.
428 * For the no-stale case they're set explicitly.
430 lfloglow
= be32_to_cpu(btp
->count
);
435 * Find the slot that's first lower than our hash value, -1 if none.
437 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; low
<= high
; ) {
438 mid
= (low
+ high
) >> 1;
439 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
441 if (hash
< args
->hashval
)
446 while (mid
>= 0 && be32_to_cpu(blp
[mid
].hashval
) >= args
->hashval
) {
450 * No stale entries, will use enddup space to hold new leaf.
454 * Mark the space needed for the new leaf entry, now in use.
456 xfs_dir2_data_use_free(tp
, bp
, enddup
,
457 (xfs_dir2_data_aoff_t
)
458 ((char *)enddup
- (char *)hdr
+ be16_to_cpu(enddup
->length
) -
460 (xfs_dir2_data_aoff_t
)sizeof(*blp
),
461 &needlog
, &needscan
);
463 * Update the tail (entry count).
465 be32_add_cpu(&btp
->count
, 1);
467 * If we now need to rebuild the bestfree map, do so.
468 * This needs to happen before the next call to use_free.
471 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
475 * Adjust pointer to the first leaf entry, we're about to move
476 * the table up one to open up space for the new leaf entry.
477 * Then adjust our index to match.
482 memmove(blp
, &blp
[1], mid
* sizeof(*blp
));
487 * Use a stale leaf for our new entry.
492 blp
[lowstale
].address
!=
493 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
496 for (highstale
= mid
+ 1;
497 highstale
< be32_to_cpu(btp
->count
) &&
498 blp
[highstale
].address
!=
499 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
) &&
500 (lowstale
< 0 || mid
- lowstale
> highstale
- mid
);
504 * Move entries toward the low-numbered stale entry.
507 (highstale
== be32_to_cpu(btp
->count
) ||
508 mid
- lowstale
<= highstale
- mid
)) {
510 memmove(&blp
[lowstale
], &blp
[lowstale
+ 1],
511 (mid
- lowstale
) * sizeof(*blp
));
512 lfloglow
= MIN(lowstale
, lfloglow
);
513 lfloghigh
= MAX(mid
, lfloghigh
);
516 * Move entries toward the high-numbered stale entry.
519 ASSERT(highstale
< be32_to_cpu(btp
->count
));
522 memmove(&blp
[mid
+ 1], &blp
[mid
],
523 (highstale
- mid
) * sizeof(*blp
));
524 lfloglow
= MIN(mid
, lfloglow
);
525 lfloghigh
= MAX(highstale
, lfloghigh
);
527 be32_add_cpu(&btp
->stale
, -1);
530 * Point to the new data entry.
532 dep
= (xfs_dir2_data_entry_t
*)dup
;
534 * Fill in the leaf entry.
536 blp
[mid
].hashval
= cpu_to_be32(args
->hashval
);
537 blp
[mid
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
538 (char *)dep
- (char *)hdr
));
539 xfs_dir2_block_log_leaf(tp
, bp
, lfloglow
, lfloghigh
);
541 * Mark space for the data entry used.
543 xfs_dir2_data_use_free(tp
, bp
, dup
,
544 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
545 (xfs_dir2_data_aoff_t
)len
, &needlog
, &needscan
);
547 * Create the new data entry.
549 dep
->inumber
= cpu_to_be64(args
->inumber
);
550 dep
->namelen
= args
->namelen
;
551 memcpy(dep
->name
, args
->name
, args
->namelen
);
552 tagp
= xfs_dir2_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(mp
, hdr
, &needlog
);
560 xfs_dir2_data_log_header(tp
, bp
);
561 xfs_dir2_block_log_tail(tp
, bp
);
562 xfs_dir2_data_log_entry(tp
, bp
, dep
);
563 xfs_dir3_data_check(dp
, bp
);
568 * Readdir for block directories.
571 xfs_dir2_block_getdents(
572 xfs_inode_t
*dp
, /* incore inode */
573 struct dir_context
*ctx
)
575 xfs_dir2_data_hdr_t
*hdr
; /* block header */
576 struct xfs_buf
*bp
; /* buffer for block */
577 xfs_dir2_block_tail_t
*btp
; /* block tail */
578 xfs_dir2_data_entry_t
*dep
; /* block data entry */
579 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
580 char *endptr
; /* end of the data entries */
581 int error
; /* error return value */
582 xfs_mount_t
*mp
; /* filesystem mount point */
583 char *ptr
; /* current data entry */
584 int wantoff
; /* starting block offset */
589 * If the block number in the offset is out of range, we're done.
591 if (xfs_dir2_dataptr_to_db(mp
, ctx
->pos
) > mp
->m_dirdatablk
)
594 error
= xfs_dir3_block_read(NULL
, dp
, &bp
);
599 * Extract the byte offset we start at from the seek pointer.
600 * We'll skip entries before this.
602 wantoff
= xfs_dir2_dataptr_to_off(mp
, ctx
->pos
);
604 xfs_dir3_data_check(dp
, bp
);
606 * Set up values for the loop.
608 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
609 ptr
= (char *)xfs_dir3_data_entry_p(hdr
);
610 endptr
= (char *)xfs_dir2_block_leaf_p(btp
);
613 * Loop over the data portion of the block.
614 * Each object is a real entry (dep) or an unused one (dup).
616 while (ptr
< endptr
) {
617 dup
= (xfs_dir2_data_unused_t
*)ptr
;
621 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
622 ptr
+= be16_to_cpu(dup
->length
);
626 dep
= (xfs_dir2_data_entry_t
*)ptr
;
629 * Bump pointer for the next iteration.
631 ptr
+= xfs_dir2_data_entsize(dep
->namelen
);
633 * The entry is before the desired starting point, skip it.
635 if ((char *)dep
- (char *)hdr
< wantoff
)
638 cook
= xfs_dir2_db_off_to_dataptr(mp
, mp
->m_dirdatablk
,
639 (char *)dep
- (char *)hdr
);
641 ctx
->pos
= cook
& 0x7fffffff;
643 * If it didn't fit, set the final offset to here & return.
645 if (!dir_emit(ctx
, (char *)dep
->name
, dep
->namelen
,
646 be64_to_cpu(dep
->inumber
), DT_UNKNOWN
)) {
647 xfs_trans_brelse(NULL
, bp
);
653 * Reached the end of the block.
654 * Set the offset to a non-existent block 1 and return.
656 ctx
->pos
= xfs_dir2_db_off_to_dataptr(mp
, mp
->m_dirdatablk
+ 1, 0) &
658 xfs_trans_brelse(NULL
, bp
);
663 * Log leaf entries from the block.
666 xfs_dir2_block_log_leaf(
667 xfs_trans_t
*tp
, /* transaction structure */
668 struct xfs_buf
*bp
, /* block buffer */
669 int first
, /* index of first logged leaf */
670 int last
) /* index of last logged leaf */
672 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
673 xfs_dir2_leaf_entry_t
*blp
;
674 xfs_dir2_block_tail_t
*btp
;
676 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
, hdr
);
677 blp
= xfs_dir2_block_leaf_p(btp
);
678 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)&blp
[first
] - (char *)hdr
),
679 (uint
)((char *)&blp
[last
+ 1] - (char *)hdr
- 1));
683 * Log the block tail.
686 xfs_dir2_block_log_tail(
687 xfs_trans_t
*tp
, /* transaction structure */
688 struct xfs_buf
*bp
) /* block buffer */
690 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
691 xfs_dir2_block_tail_t
*btp
;
693 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
, hdr
);
694 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)btp
- (char *)hdr
),
695 (uint
)((char *)(btp
+ 1) - (char *)hdr
- 1));
699 * Look up an entry in the block. This is the external routine,
700 * xfs_dir2_block_lookup_int does the real work.
703 xfs_dir2_block_lookup(
704 xfs_da_args_t
*args
) /* dir lookup arguments */
706 xfs_dir2_data_hdr_t
*hdr
; /* block header */
707 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
708 struct xfs_buf
*bp
; /* block buffer */
709 xfs_dir2_block_tail_t
*btp
; /* block tail */
710 xfs_dir2_data_entry_t
*dep
; /* block data entry */
711 xfs_inode_t
*dp
; /* incore inode */
712 int ent
; /* entry index */
713 int error
; /* error return value */
714 xfs_mount_t
*mp
; /* filesystem mount point */
716 trace_xfs_dir2_block_lookup(args
);
719 * Get the buffer, look up the entry.
720 * If not found (ENOENT) then return, have no buffer.
722 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
)))
727 xfs_dir3_data_check(dp
, bp
);
728 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
729 blp
= xfs_dir2_block_leaf_p(btp
);
731 * Get the offset from the leaf entry, to point to the data.
733 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
734 xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(blp
[ent
].address
)));
736 * Fill in inode number, CI name if appropriate, release the block.
738 args
->inumber
= be64_to_cpu(dep
->inumber
);
739 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
740 xfs_trans_brelse(args
->trans
, bp
);
741 return XFS_ERROR(error
);
745 * Internal block lookup routine.
747 static int /* error */
748 xfs_dir2_block_lookup_int(
749 xfs_da_args_t
*args
, /* dir lookup arguments */
750 struct xfs_buf
**bpp
, /* returned block buffer */
751 int *entno
) /* returned entry number */
753 xfs_dir2_dataptr_t addr
; /* data entry address */
754 xfs_dir2_data_hdr_t
*hdr
; /* block header */
755 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
756 struct xfs_buf
*bp
; /* block buffer */
757 xfs_dir2_block_tail_t
*btp
; /* block tail */
758 xfs_dir2_data_entry_t
*dep
; /* block data entry */
759 xfs_inode_t
*dp
; /* incore inode */
760 int error
; /* error return value */
761 xfs_dahash_t hash
; /* found hash value */
762 int high
; /* binary search high index */
763 int low
; /* binary search low index */
764 int mid
; /* binary search current idx */
765 xfs_mount_t
*mp
; /* filesystem mount point */
766 xfs_trans_t
*tp
; /* transaction pointer */
767 enum xfs_dacmp cmp
; /* comparison result */
773 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
778 xfs_dir3_data_check(dp
, bp
);
779 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
780 blp
= xfs_dir2_block_leaf_p(btp
);
782 * Loop doing a binary search for our hash value.
783 * Find our entry, ENOENT if it's not there.
785 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; ; ) {
787 mid
= (low
+ high
) >> 1;
788 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
790 if (hash
< args
->hashval
)
795 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
796 xfs_trans_brelse(tp
, bp
);
797 return XFS_ERROR(ENOENT
);
801 * Back up to the first one with the right hash value.
803 while (mid
> 0 && be32_to_cpu(blp
[mid
- 1].hashval
) == args
->hashval
) {
807 * Now loop forward through all the entries with the
808 * right hash value looking for our name.
811 if ((addr
= be32_to_cpu(blp
[mid
].address
)) == XFS_DIR2_NULL_DATAPTR
)
814 * Get pointer to the entry from the leaf.
816 dep
= (xfs_dir2_data_entry_t
*)
817 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, addr
));
819 * Compare name and if it's an exact match, return the index
820 * and buffer. If it's the first case-insensitive match, store
821 * the index and buffer and continue looking for an exact match.
823 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
824 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
825 args
->cmpresult
= cmp
;
828 if (cmp
== XFS_CMP_EXACT
)
831 } while (++mid
< be32_to_cpu(btp
->count
) &&
832 be32_to_cpu(blp
[mid
].hashval
) == hash
);
834 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
836 * Here, we can only be doing a lookup (not a rename or replace).
837 * If a case-insensitive match was found earlier, return success.
839 if (args
->cmpresult
== XFS_CMP_CASE
)
842 * No match, release the buffer and return ENOENT.
844 xfs_trans_brelse(tp
, bp
);
845 return XFS_ERROR(ENOENT
);
849 * Remove an entry from a block format directory.
850 * If that makes the block small enough to fit in shortform, transform it.
853 xfs_dir2_block_removename(
854 xfs_da_args_t
*args
) /* directory operation args */
856 xfs_dir2_data_hdr_t
*hdr
; /* block header */
857 xfs_dir2_leaf_entry_t
*blp
; /* block leaf pointer */
858 struct xfs_buf
*bp
; /* block buffer */
859 xfs_dir2_block_tail_t
*btp
; /* block tail */
860 xfs_dir2_data_entry_t
*dep
; /* block data entry */
861 xfs_inode_t
*dp
; /* incore inode */
862 int ent
; /* block leaf entry index */
863 int error
; /* error return value */
864 xfs_mount_t
*mp
; /* filesystem mount point */
865 int needlog
; /* need to log block header */
866 int needscan
; /* need to fixup bestfree */
867 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
868 int size
; /* shortform size */
869 xfs_trans_t
*tp
; /* transaction pointer */
871 trace_xfs_dir2_block_removename(args
);
874 * Look up the entry in the block. Gets the buffer and entry index.
875 * It will always be there, the vnodeops level does a lookup first.
877 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
884 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
885 blp
= xfs_dir2_block_leaf_p(btp
);
887 * Point to the data entry using the leaf entry.
889 dep
= (xfs_dir2_data_entry_t
*)
890 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(blp
[ent
].address
)));
892 * Mark the data entry's space free.
894 needlog
= needscan
= 0;
895 xfs_dir2_data_make_free(tp
, bp
,
896 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)hdr
),
897 xfs_dir2_data_entsize(dep
->namelen
), &needlog
, &needscan
);
899 * Fix up the block tail.
901 be32_add_cpu(&btp
->stale
, 1);
902 xfs_dir2_block_log_tail(tp
, bp
);
904 * Remove the leaf entry by marking it stale.
906 blp
[ent
].address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
907 xfs_dir2_block_log_leaf(tp
, bp
, ent
, ent
);
909 * Fix up bestfree, log the header if necessary.
912 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
914 xfs_dir2_data_log_header(tp
, bp
);
915 xfs_dir3_data_check(dp
, bp
);
917 * See if the size as a shortform is good enough.
919 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
920 if (size
> XFS_IFORK_DSIZE(dp
))
924 * If it works, do the conversion.
926 return xfs_dir2_block_to_sf(args
, bp
, size
, &sfh
);
930 * Replace an entry in a V2 block directory.
931 * Change the inode number to the new value.
934 xfs_dir2_block_replace(
935 xfs_da_args_t
*args
) /* directory operation args */
937 xfs_dir2_data_hdr_t
*hdr
; /* block header */
938 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
939 struct xfs_buf
*bp
; /* block buffer */
940 xfs_dir2_block_tail_t
*btp
; /* block tail */
941 xfs_dir2_data_entry_t
*dep
; /* block data entry */
942 xfs_inode_t
*dp
; /* incore inode */
943 int ent
; /* leaf entry index */
944 int error
; /* error return value */
945 xfs_mount_t
*mp
; /* filesystem mount point */
947 trace_xfs_dir2_block_replace(args
);
950 * Lookup the entry in the directory. Get buffer and entry index.
951 * This will always succeed since the caller has already done a lookup.
953 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
959 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
960 blp
= xfs_dir2_block_leaf_p(btp
);
962 * Point to the data entry we need to change.
964 dep
= (xfs_dir2_data_entry_t
*)
965 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(blp
[ent
].address
)));
966 ASSERT(be64_to_cpu(dep
->inumber
) != args
->inumber
);
968 * Change the inode number to the new value.
970 dep
->inumber
= cpu_to_be64(args
->inumber
);
971 xfs_dir2_data_log_entry(args
->trans
, bp
, dep
);
972 xfs_dir3_data_check(dp
, bp
);
977 * Qsort comparison routine for the block leaf entries.
979 static int /* sort order */
981 const void *a
, /* first leaf entry */
982 const void *b
) /* second leaf entry */
984 const xfs_dir2_leaf_entry_t
*la
; /* first leaf entry */
985 const xfs_dir2_leaf_entry_t
*lb
; /* second leaf entry */
989 return be32_to_cpu(la
->hashval
) < be32_to_cpu(lb
->hashval
) ? -1 :
990 (be32_to_cpu(la
->hashval
) > be32_to_cpu(lb
->hashval
) ? 1 : 0);
994 * Convert a V2 leaf directory to a V2 block directory if possible.
997 xfs_dir2_leaf_to_block(
998 xfs_da_args_t
*args
, /* operation arguments */
999 struct xfs_buf
*lbp
, /* leaf buffer */
1000 struct xfs_buf
*dbp
) /* data buffer */
1002 __be16
*bestsp
; /* leaf bests table */
1003 xfs_dir2_data_hdr_t
*hdr
; /* block header */
1004 xfs_dir2_block_tail_t
*btp
; /* block tail */
1005 xfs_inode_t
*dp
; /* incore directory inode */
1006 xfs_dir2_data_unused_t
*dup
; /* unused data entry */
1007 int error
; /* error return value */
1008 int from
; /* leaf from index */
1009 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1010 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1011 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1012 xfs_mount_t
*mp
; /* file system mount point */
1013 int needlog
; /* need to log data header */
1014 int needscan
; /* need to scan for bestfree */
1015 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
1016 int size
; /* bytes used */
1017 __be16
*tagp
; /* end of entry (tag) */
1018 int to
; /* block/leaf to index */
1019 xfs_trans_t
*tp
; /* transaction pointer */
1020 struct xfs_dir2_leaf_entry
*ents
;
1021 struct xfs_dir3_icleaf_hdr leafhdr
;
1023 trace_xfs_dir2_leaf_to_block(args
);
1029 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
1030 ents
= xfs_dir3_leaf_ents_p(leaf
);
1031 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1033 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAF1_MAGIC
||
1034 leafhdr
.magic
== XFS_DIR3_LEAF1_MAGIC
);
1036 * If there are data blocks other than the first one, take this
1037 * opportunity to remove trailing empty data blocks that may have
1038 * been left behind during no-space-reservation operations.
1039 * These will show up in the leaf bests table.
1041 while (dp
->i_d
.di_size
> mp
->m_dirblksize
) {
1044 hdrsz
= xfs_dir3_data_hdr_size(xfs_sb_version_hascrc(&mp
->m_sb
));
1045 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1046 if (be16_to_cpu(bestsp
[be32_to_cpu(ltp
->bestcount
) - 1]) ==
1047 mp
->m_dirblksize
- hdrsz
) {
1049 xfs_dir2_leaf_trim_data(args
, lbp
,
1050 (xfs_dir2_db_t
)(be32_to_cpu(ltp
->bestcount
) - 1))))
1056 * Read the data block if we don't already have it, give up if it fails.
1059 error
= xfs_dir3_data_read(tp
, dp
, mp
->m_dirdatablk
, -1, &dbp
);
1064 ASSERT(hdr
->magic
== cpu_to_be32(XFS_DIR2_DATA_MAGIC
) ||
1065 hdr
->magic
== cpu_to_be32(XFS_DIR3_DATA_MAGIC
));
1068 * Size of the "leaf" area in the block.
1070 size
= (uint
)sizeof(xfs_dir2_block_tail_t
) +
1071 (uint
)sizeof(*lep
) * (leafhdr
.count
- leafhdr
.stale
);
1073 * Look at the last data entry.
1075 tagp
= (__be16
*)((char *)hdr
+ mp
->m_dirblksize
) - 1;
1076 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
1078 * If it's not free or is too short we can't do it.
1080 if (be16_to_cpu(dup
->freetag
) != XFS_DIR2_DATA_FREE_TAG
||
1081 be16_to_cpu(dup
->length
) < size
)
1085 * Start converting it to block form.
1087 xfs_dir3_block_init(mp
, tp
, dbp
, dp
);
1092 * Use up the space at the end of the block (blp/btp).
1094 xfs_dir2_data_use_free(tp
, dbp
, dup
, mp
->m_dirblksize
- size
, size
,
1095 &needlog
, &needscan
);
1097 * Initialize the block tail.
1099 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
1100 btp
->count
= cpu_to_be32(leafhdr
.count
- leafhdr
.stale
);
1102 xfs_dir2_block_log_tail(tp
, dbp
);
1104 * Initialize the block leaf area. We compact out stale entries.
1106 lep
= xfs_dir2_block_leaf_p(btp
);
1107 for (from
= to
= 0; from
< leafhdr
.count
; from
++) {
1108 if (ents
[from
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
1110 lep
[to
++] = ents
[from
];
1112 ASSERT(to
== be32_to_cpu(btp
->count
));
1113 xfs_dir2_block_log_leaf(tp
, dbp
, 0, be32_to_cpu(btp
->count
) - 1);
1115 * Scan the bestfree if we need it and log the data block header.
1118 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
1120 xfs_dir2_data_log_header(tp
, dbp
);
1122 * Pitch the old leaf block.
1124 error
= xfs_da_shrink_inode(args
, mp
->m_dirleafblk
, lbp
);
1129 * Now see if the resulting block can be shrunken to shortform.
1131 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
1132 if (size
> XFS_IFORK_DSIZE(dp
))
1135 return xfs_dir2_block_to_sf(args
, dbp
, size
, &sfh
);
1139 * Convert the shortform directory to block form.
1142 xfs_dir2_sf_to_block(
1143 xfs_da_args_t
*args
) /* operation arguments */
1145 xfs_dir2_db_t blkno
; /* dir-relative block # (0) */
1146 xfs_dir2_data_hdr_t
*hdr
; /* block header */
1147 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
1148 struct xfs_buf
*bp
; /* block buffer */
1149 xfs_dir2_block_tail_t
*btp
; /* block tail pointer */
1150 xfs_dir2_data_entry_t
*dep
; /* data entry pointer */
1151 xfs_inode_t
*dp
; /* incore directory inode */
1152 int dummy
; /* trash */
1153 xfs_dir2_data_unused_t
*dup
; /* unused entry pointer */
1154 int endoffset
; /* end of data objects */
1155 int error
; /* error return value */
1157 xfs_mount_t
*mp
; /* filesystem mount point */
1158 int needlog
; /* need to log block header */
1159 int needscan
; /* need to scan block freespc */
1160 int newoffset
; /* offset from current entry */
1161 int offset
; /* target block offset */
1162 xfs_dir2_sf_entry_t
*sfep
; /* sf entry pointer */
1163 xfs_dir2_sf_hdr_t
*oldsfp
; /* old shortform header */
1164 xfs_dir2_sf_hdr_t
*sfp
; /* shortform header */
1165 __be16
*tagp
; /* end of data entry */
1166 xfs_trans_t
*tp
; /* transaction pointer */
1167 struct xfs_name name
;
1168 struct xfs_ifork
*ifp
;
1170 trace_xfs_dir2_sf_to_block(args
);
1175 ifp
= XFS_IFORK_PTR(dp
, XFS_DATA_FORK
);
1176 ASSERT(ifp
->if_flags
& XFS_IFINLINE
);
1178 * Bomb out if the shortform directory is way too short.
1180 if (dp
->i_d
.di_size
< offsetof(xfs_dir2_sf_hdr_t
, parent
)) {
1181 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1182 return XFS_ERROR(EIO
);
1185 oldsfp
= (xfs_dir2_sf_hdr_t
*)ifp
->if_u1
.if_data
;
1187 ASSERT(ifp
->if_bytes
== dp
->i_d
.di_size
);
1188 ASSERT(ifp
->if_u1
.if_data
!= NULL
);
1189 ASSERT(dp
->i_d
.di_size
>= xfs_dir2_sf_hdr_size(oldsfp
->i8count
));
1190 ASSERT(dp
->i_d
.di_nextents
== 0);
1193 * Copy the directory into a temporary buffer.
1194 * Then pitch the incore inode data so we can make extents.
1196 sfp
= kmem_alloc(ifp
->if_bytes
, KM_SLEEP
);
1197 memcpy(sfp
, oldsfp
, ifp
->if_bytes
);
1199 xfs_idata_realloc(dp
, -ifp
->if_bytes
, XFS_DATA_FORK
);
1200 xfs_bmap_local_to_extents_empty(dp
, XFS_DATA_FORK
);
1201 dp
->i_d
.di_size
= 0;
1204 * Add block 0 to the inode.
1206 error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
, &blkno
);
1212 * Initialize the data block, then convert it to block format.
1214 error
= xfs_dir3_data_init(args
, blkno
, &bp
);
1219 xfs_dir3_block_init(mp
, tp
, bp
, dp
);
1223 * Compute size of block "tail" area.
1225 i
= (uint
)sizeof(*btp
) +
1226 (sfp
->count
+ 2) * (uint
)sizeof(xfs_dir2_leaf_entry_t
);
1228 * The whole thing is initialized to free by the init routine.
1229 * Say we're using the leaf and tail area.
1231 dup
= xfs_dir3_data_unused_p(hdr
);
1232 needlog
= needscan
= 0;
1233 xfs_dir2_data_use_free(tp
, bp
, dup
, mp
->m_dirblksize
- i
, i
, &needlog
,
1235 ASSERT(needscan
== 0);
1239 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
1240 btp
->count
= cpu_to_be32(sfp
->count
+ 2); /* ., .. */
1242 blp
= xfs_dir2_block_leaf_p(btp
);
1243 endoffset
= (uint
)((char *)blp
- (char *)hdr
);
1245 * Remove the freespace, we'll manage it.
1247 xfs_dir2_data_use_free(tp
, bp
, dup
,
1248 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
1249 be16_to_cpu(dup
->length
), &needlog
, &needscan
);
1251 * Create entry for .
1253 dep
= xfs_dir3_data_dot_entry_p(hdr
);
1254 dep
->inumber
= cpu_to_be64(dp
->i_ino
);
1257 tagp
= xfs_dir2_data_entry_tag_p(dep
);
1258 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1259 xfs_dir2_data_log_entry(tp
, bp
, dep
);
1260 blp
[0].hashval
= cpu_to_be32(xfs_dir_hash_dot
);
1261 blp
[0].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
1262 (char *)dep
- (char *)hdr
));
1264 * Create entry for ..
1266 dep
= xfs_dir3_data_dotdot_entry_p(hdr
);
1267 dep
->inumber
= cpu_to_be64(xfs_dir2_sf_get_parent_ino(sfp
));
1269 dep
->name
[0] = dep
->name
[1] = '.';
1270 tagp
= xfs_dir2_data_entry_tag_p(dep
);
1271 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1272 xfs_dir2_data_log_entry(tp
, bp
, dep
);
1273 blp
[1].hashval
= cpu_to_be32(xfs_dir_hash_dotdot
);
1274 blp
[1].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
1275 (char *)dep
- (char *)hdr
));
1276 offset
= xfs_dir3_data_first_offset(hdr
);
1278 * Loop over existing entries, stuff them in.
1284 sfep
= xfs_dir2_sf_firstentry(sfp
);
1286 * Need to preserve the existing offset values in the sf directory.
1287 * Insert holes (unused entries) where necessary.
1289 while (offset
< endoffset
) {
1291 * sfep is null when we reach the end of the list.
1294 newoffset
= endoffset
;
1296 newoffset
= xfs_dir2_sf_get_offset(sfep
);
1298 * There should be a hole here, make one.
1300 if (offset
< newoffset
) {
1301 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ offset
);
1302 dup
->freetag
= cpu_to_be16(XFS_DIR2_DATA_FREE_TAG
);
1303 dup
->length
= cpu_to_be16(newoffset
- offset
);
1304 *xfs_dir2_data_unused_tag_p(dup
) = cpu_to_be16(
1305 ((char *)dup
- (char *)hdr
));
1306 xfs_dir2_data_log_unused(tp
, bp
, dup
);
1307 xfs_dir2_data_freeinsert(hdr
, dup
, &dummy
);
1308 offset
+= be16_to_cpu(dup
->length
);
1312 * Copy a real entry.
1314 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+ newoffset
);
1315 dep
->inumber
= cpu_to_be64(xfs_dir2_sfe_get_ino(sfp
, sfep
));
1316 dep
->namelen
= sfep
->namelen
;
1317 memcpy(dep
->name
, sfep
->name
, dep
->namelen
);
1318 tagp
= xfs_dir2_data_entry_tag_p(dep
);
1319 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1320 xfs_dir2_data_log_entry(tp
, bp
, dep
);
1321 name
.name
= sfep
->name
;
1322 name
.len
= sfep
->namelen
;
1323 blp
[2 + i
].hashval
= cpu_to_be32(mp
->m_dirnameops
->
1325 blp
[2 + i
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
1326 (char *)dep
- (char *)hdr
));
1327 offset
= (int)((char *)(tagp
+ 1) - (char *)hdr
);
1328 if (++i
== sfp
->count
)
1331 sfep
= xfs_dir2_sf_nextentry(sfp
, sfep
);
1333 /* Done with the temporary buffer */
1336 * Sort the leaf entries by hash value.
1338 xfs_sort(blp
, be32_to_cpu(btp
->count
), sizeof(*blp
), xfs_dir2_block_sort
);
1340 * Log the leaf entry area and tail.
1341 * Already logged the header in data_init, ignore needlog.
1343 ASSERT(needscan
== 0);
1344 xfs_dir2_block_log_leaf(tp
, bp
, 0, be32_to_cpu(btp
->count
) - 1);
1345 xfs_dir2_block_log_tail(tp
, bp
);
1346 xfs_dir3_data_check(dp
, bp
);