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"
32 #include "xfs_buf_item.h"
34 #include "xfs_dir2_format.h"
35 #include "xfs_dir2_priv.h"
36 #include "xfs_error.h"
37 #include "xfs_trace.h"
38 #include "xfs_cksum.h"
41 * Local function prototypes.
43 static void xfs_dir2_block_log_leaf(xfs_trans_t
*tp
, struct xfs_buf
*bp
,
45 static void xfs_dir2_block_log_tail(xfs_trans_t
*tp
, struct xfs_buf
*bp
);
46 static int xfs_dir2_block_lookup_int(xfs_da_args_t
*args
, struct xfs_buf
**bpp
,
48 static int xfs_dir2_block_sort(const void *a
, const void *b
);
50 static xfs_dahash_t xfs_dir_hash_dot
, xfs_dir_hash_dotdot
;
53 * One-time startup routine called from xfs_init().
58 xfs_dir_hash_dot
= xfs_da_hashname((unsigned char *)".", 1);
59 xfs_dir_hash_dotdot
= xfs_da_hashname((unsigned char *)"..", 2);
63 xfs_dir3_block_verify(
66 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
67 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
69 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
70 if (hdr3
->magic
!= cpu_to_be32(XFS_DIR3_BLOCK_MAGIC
))
72 if (!uuid_equal(&hdr3
->uuid
, &mp
->m_sb
.sb_uuid
))
74 if (be64_to_cpu(hdr3
->blkno
) != bp
->b_bn
)
77 if (hdr3
->magic
!= cpu_to_be32(XFS_DIR2_BLOCK_MAGIC
))
80 if (__xfs_dir2_data_check(NULL
, bp
))
86 xfs_dir3_block_read_verify(
89 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
91 if ((xfs_sb_version_hascrc(&mp
->m_sb
) &&
92 !xfs_verify_cksum(bp
->b_addr
, BBTOB(bp
->b_length
),
93 XFS_DIR3_DATA_CRC_OFF
)) ||
94 !xfs_dir3_block_verify(bp
)) {
95 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, bp
->b_addr
);
96 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
101 xfs_dir3_block_write_verify(
104 struct xfs_mount
*mp
= bp
->b_target
->bt_mount
;
105 struct xfs_buf_log_item
*bip
= bp
->b_fspriv
;
106 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
108 if (!xfs_dir3_block_verify(bp
)) {
109 XFS_CORRUPTION_ERROR(__func__
, XFS_ERRLEVEL_LOW
, mp
, bp
->b_addr
);
110 xfs_buf_ioerror(bp
, EFSCORRUPTED
);
114 if (!xfs_sb_version_hascrc(&mp
->m_sb
))
118 hdr3
->lsn
= cpu_to_be64(bip
->bli_item
.li_lsn
);
120 xfs_update_cksum(bp
->b_addr
, BBTOB(bp
->b_length
), XFS_DIR3_DATA_CRC_OFF
);
123 const struct xfs_buf_ops xfs_dir3_block_buf_ops
= {
124 .verify_read
= xfs_dir3_block_read_verify
,
125 .verify_write
= xfs_dir3_block_write_verify
,
130 struct xfs_trans
*tp
,
131 struct xfs_inode
*dp
,
132 struct xfs_buf
**bpp
)
134 struct xfs_mount
*mp
= dp
->i_mount
;
136 return xfs_da_read_buf(tp
, dp
, mp
->m_dirdatablk
, -1, bpp
,
137 XFS_DATA_FORK
, &xfs_dir3_block_buf_ops
);
142 struct xfs_mount
*mp
,
144 struct xfs_inode
*dp
)
146 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
148 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
150 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
151 memset(hdr3
, 0, sizeof(*hdr3
));
152 hdr3
->magic
= cpu_to_be32(XFS_DIR3_BLOCK_MAGIC
);
153 hdr3
->blkno
= cpu_to_be64(bp
->b_bn
);
154 hdr3
->owner
= cpu_to_be64(dp
->i_ino
);
155 uuid_copy(&hdr3
->uuid
, &mp
->m_sb
.sb_uuid
);
159 hdr3
->magic
= cpu_to_be32(XFS_DIR2_BLOCK_MAGIC
);
163 xfs_dir2_block_need_space(
164 struct xfs_dir2_data_hdr
*hdr
,
165 struct xfs_dir2_block_tail
*btp
,
166 struct xfs_dir2_leaf_entry
*blp
,
168 struct xfs_dir2_data_unused
**dupp
,
169 struct xfs_dir2_data_unused
**enddupp
,
173 struct xfs_dir2_data_free
*bf
;
175 struct xfs_dir2_data_unused
*dup
= NULL
;
176 struct xfs_dir2_data_unused
*enddup
= NULL
;
179 bf
= xfs_dir3_data_bestfree_p(hdr
);
182 * If there are stale entries we'll use one for the leaf.
185 if (be16_to_cpu(bf
[0].length
) >= len
) {
187 * The biggest entry enough to avoid compaction.
189 dup
= (xfs_dir2_data_unused_t
*)
190 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
195 * Will need to compact to make this work.
196 * Tag just before the first leaf entry.
199 tagp
= (__be16
*)blp
- 1;
201 /* Data object just before the first leaf entry. */
202 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
205 * If it's not free then the data will go where the
206 * leaf data starts now, if it works at all.
208 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
209 if (be16_to_cpu(dup
->length
) + (be32_to_cpu(btp
->stale
) - 1) *
210 (uint
)sizeof(*blp
) < len
)
212 } else if ((be32_to_cpu(btp
->stale
) - 1) * (uint
)sizeof(*blp
) < len
)
215 dup
= (xfs_dir2_data_unused_t
*)blp
;
220 * no stale entries, so just use free space.
221 * Tag just before the first leaf entry.
223 tagp
= (__be16
*)blp
- 1;
225 /* Data object just before the first leaf entry. */
226 enddup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
229 * If it's not free then can't do this add without cleaning up:
230 * the space before the first leaf entry needs to be free so it
231 * can be expanded to hold the pointer to the new entry.
233 if (be16_to_cpu(enddup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
235 * Check out the biggest freespace and see if it's the same one.
237 dup
= (xfs_dir2_data_unused_t
*)
238 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
241 * Not the same free entry, just check its length.
243 if (be16_to_cpu(dup
->length
) < len
)
249 * It is the biggest freespace, can it hold the leaf too?
251 if (be16_to_cpu(dup
->length
) < len
+ (uint
)sizeof(*blp
)) {
253 * Yes, use the second-largest entry instead if it works.
255 if (be16_to_cpu(bf
[1].length
) >= len
)
256 dup
= (xfs_dir2_data_unused_t
*)
257 ((char *)hdr
+ be16_to_cpu(bf
[1].offset
));
269 * compact the leaf entries.
270 * Leave the highest-numbered stale entry stale.
271 * XXX should be the one closest to mid but mid is not yet computed.
274 xfs_dir2_block_compact(
275 struct xfs_trans
*tp
,
277 struct xfs_dir2_data_hdr
*hdr
,
278 struct xfs_dir2_block_tail
*btp
,
279 struct xfs_dir2_leaf_entry
*blp
,
284 int fromidx
; /* source leaf index */
285 int toidx
; /* target leaf index */
287 int highstale
; /* high stale index */
289 fromidx
= toidx
= be32_to_cpu(btp
->count
) - 1;
290 highstale
= *lfloghigh
= -1;
291 for (; fromidx
>= 0; fromidx
--) {
292 if (blp
[fromidx
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
)) {
296 if (*lfloghigh
== -1)
302 blp
[toidx
] = blp
[fromidx
];
305 *lfloglow
= toidx
+ 1 - (be32_to_cpu(btp
->stale
) - 1);
306 *lfloghigh
-= be32_to_cpu(btp
->stale
) - 1;
307 be32_add_cpu(&btp
->count
, -(be32_to_cpu(btp
->stale
) - 1));
308 xfs_dir2_data_make_free(tp
, bp
,
309 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)hdr
),
310 (xfs_dir2_data_aoff_t
)((be32_to_cpu(btp
->stale
) - 1) * sizeof(*blp
)),
312 blp
+= be32_to_cpu(btp
->stale
) - 1;
313 btp
->stale
= cpu_to_be32(1);
315 * If we now need to rebuild the bestfree map, do so.
316 * This needs to happen before the next call to use_free.
319 xfs_dir2_data_freescan(tp
->t_mountp
, hdr
, needlog
);
323 * Add an entry to a block directory.
326 xfs_dir2_block_addname(
327 xfs_da_args_t
*args
) /* directory op arguments */
329 xfs_dir2_data_hdr_t
*hdr
; /* block header */
330 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
331 struct xfs_buf
*bp
; /* buffer for block */
332 xfs_dir2_block_tail_t
*btp
; /* block tail */
333 int compact
; /* need to compact leaf ents */
334 xfs_dir2_data_entry_t
*dep
; /* block data entry */
335 xfs_inode_t
*dp
; /* directory inode */
336 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
337 int error
; /* error return value */
338 xfs_dir2_data_unused_t
*enddup
=NULL
; /* unused at end of data */
339 xfs_dahash_t hash
; /* hash value of found entry */
340 int high
; /* high index for binary srch */
341 int highstale
; /* high stale index */
342 int lfloghigh
=0; /* last final leaf to log */
343 int lfloglow
=0; /* first final leaf to log */
344 int len
; /* length of the new entry */
345 int low
; /* low index for binary srch */
346 int lowstale
; /* low stale index */
347 int mid
=0; /* midpoint for binary srch */
348 xfs_mount_t
*mp
; /* filesystem mount point */
349 int needlog
; /* need to log header */
350 int needscan
; /* need to rescan freespace */
351 __be16
*tagp
; /* pointer to tag value */
352 xfs_trans_t
*tp
; /* transaction structure */
354 trace_xfs_dir2_block_addname(args
);
360 /* Read the (one and only) directory block into bp. */
361 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
365 len
= xfs_dir2_data_entsize(args
->namelen
);
368 * Set up pointers to parts of the block.
371 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
372 blp
= xfs_dir2_block_leaf_p(btp
);
375 * Find out if we can reuse stale entries or whether we need extra
376 * space for entry and new leaf.
378 xfs_dir2_block_need_space(hdr
, btp
, blp
, &tagp
, &dup
,
379 &enddup
, &compact
, len
);
382 * Done everything we need for a space check now.
384 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
385 xfs_trans_brelse(tp
, bp
);
387 return XFS_ERROR(ENOSPC
);
392 * If we don't have space for the new entry & leaf ...
395 /* Don't have a space reservation: return no-space. */
396 if (args
->total
== 0)
397 return XFS_ERROR(ENOSPC
);
399 * Convert to the next larger format.
400 * Then add the new entry in that format.
402 error
= xfs_dir2_block_to_leaf(args
, bp
);
405 return xfs_dir2_leaf_addname(args
);
408 needlog
= needscan
= 0;
411 * If need to compact the leaf entries, do it now.
414 xfs_dir2_block_compact(tp
, bp
, hdr
, btp
, blp
, &needlog
,
415 &lfloghigh
, &lfloglow
);
416 /* recalculate blp post-compaction */
417 blp
= xfs_dir2_block_leaf_p(btp
);
418 } else if (btp
->stale
) {
420 * Set leaf logging boundaries to impossible state.
421 * For the no-stale case they're set explicitly.
423 lfloglow
= be32_to_cpu(btp
->count
);
428 * Find the slot that's first lower than our hash value, -1 if none.
430 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; low
<= high
; ) {
431 mid
= (low
+ high
) >> 1;
432 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
434 if (hash
< args
->hashval
)
439 while (mid
>= 0 && be32_to_cpu(blp
[mid
].hashval
) >= args
->hashval
) {
443 * No stale entries, will use enddup space to hold new leaf.
447 * Mark the space needed for the new leaf entry, now in use.
449 xfs_dir2_data_use_free(tp
, bp
, enddup
,
450 (xfs_dir2_data_aoff_t
)
451 ((char *)enddup
- (char *)hdr
+ be16_to_cpu(enddup
->length
) -
453 (xfs_dir2_data_aoff_t
)sizeof(*blp
),
454 &needlog
, &needscan
);
456 * Update the tail (entry count).
458 be32_add_cpu(&btp
->count
, 1);
460 * If we now need to rebuild the bestfree map, do so.
461 * This needs to happen before the next call to use_free.
464 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
468 * Adjust pointer to the first leaf entry, we're about to move
469 * the table up one to open up space for the new leaf entry.
470 * Then adjust our index to match.
475 memmove(blp
, &blp
[1], mid
* sizeof(*blp
));
480 * Use a stale leaf for our new entry.
485 blp
[lowstale
].address
!=
486 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
489 for (highstale
= mid
+ 1;
490 highstale
< be32_to_cpu(btp
->count
) &&
491 blp
[highstale
].address
!=
492 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
) &&
493 (lowstale
< 0 || mid
- lowstale
> highstale
- mid
);
497 * Move entries toward the low-numbered stale entry.
500 (highstale
== be32_to_cpu(btp
->count
) ||
501 mid
- lowstale
<= highstale
- mid
)) {
503 memmove(&blp
[lowstale
], &blp
[lowstale
+ 1],
504 (mid
- lowstale
) * sizeof(*blp
));
505 lfloglow
= MIN(lowstale
, lfloglow
);
506 lfloghigh
= MAX(mid
, lfloghigh
);
509 * Move entries toward the high-numbered stale entry.
512 ASSERT(highstale
< be32_to_cpu(btp
->count
));
515 memmove(&blp
[mid
+ 1], &blp
[mid
],
516 (highstale
- mid
) * sizeof(*blp
));
517 lfloglow
= MIN(mid
, lfloglow
);
518 lfloghigh
= MAX(highstale
, lfloghigh
);
520 be32_add_cpu(&btp
->stale
, -1);
523 * Point to the new data entry.
525 dep
= (xfs_dir2_data_entry_t
*)dup
;
527 * Fill in the leaf entry.
529 blp
[mid
].hashval
= cpu_to_be32(args
->hashval
);
530 blp
[mid
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
531 (char *)dep
- (char *)hdr
));
532 xfs_dir2_block_log_leaf(tp
, bp
, lfloglow
, lfloghigh
);
534 * Mark space for the data entry used.
536 xfs_dir2_data_use_free(tp
, bp
, dup
,
537 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
538 (xfs_dir2_data_aoff_t
)len
, &needlog
, &needscan
);
540 * Create the new data entry.
542 dep
->inumber
= cpu_to_be64(args
->inumber
);
543 dep
->namelen
= args
->namelen
;
544 memcpy(dep
->name
, args
->name
, args
->namelen
);
545 tagp
= xfs_dir2_data_entry_tag_p(dep
);
546 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
548 * Clean up the bestfree array and log the header, tail, and entry.
551 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
553 xfs_dir2_data_log_header(tp
, bp
);
554 xfs_dir2_block_log_tail(tp
, bp
);
555 xfs_dir2_data_log_entry(tp
, bp
, dep
);
556 xfs_dir2_data_check(dp
, bp
);
561 * Readdir for block directories.
564 xfs_dir2_block_getdents(
565 xfs_inode_t
*dp
, /* incore inode */
570 xfs_dir2_data_hdr_t
*hdr
; /* block header */
571 struct xfs_buf
*bp
; /* buffer for block */
572 xfs_dir2_block_tail_t
*btp
; /* block tail */
573 xfs_dir2_data_entry_t
*dep
; /* block data entry */
574 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
575 char *endptr
; /* end of the data entries */
576 int error
; /* error return value */
577 xfs_mount_t
*mp
; /* filesystem mount point */
578 char *ptr
; /* current data entry */
579 int wantoff
; /* starting block offset */
584 * If the block number in the offset is out of range, we're done.
586 if (xfs_dir2_dataptr_to_db(mp
, *offset
) > mp
->m_dirdatablk
)
589 error
= xfs_dir3_block_read(NULL
, dp
, &bp
);
594 * Extract the byte offset we start at from the seek pointer.
595 * We'll skip entries before this.
597 wantoff
= xfs_dir2_dataptr_to_off(mp
, *offset
);
599 xfs_dir2_data_check(dp
, bp
);
601 * Set up values for the loop.
603 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
604 ptr
= (char *)xfs_dir3_data_entry_p(hdr
);
605 endptr
= (char *)xfs_dir2_block_leaf_p(btp
);
608 * Loop over the data portion of the block.
609 * Each object is a real entry (dep) or an unused one (dup).
611 while (ptr
< endptr
) {
612 dup
= (xfs_dir2_data_unused_t
*)ptr
;
616 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
617 ptr
+= be16_to_cpu(dup
->length
);
621 dep
= (xfs_dir2_data_entry_t
*)ptr
;
624 * Bump pointer for the next iteration.
626 ptr
+= xfs_dir2_data_entsize(dep
->namelen
);
628 * The entry is before the desired starting point, skip it.
630 if ((char *)dep
- (char *)hdr
< wantoff
)
633 cook
= xfs_dir2_db_off_to_dataptr(mp
, mp
->m_dirdatablk
,
634 (char *)dep
- (char *)hdr
);
637 * If it didn't fit, set the final offset to here & return.
639 if (filldir(dirent
, (char *)dep
->name
, dep
->namelen
,
640 cook
& 0x7fffffff, be64_to_cpu(dep
->inumber
),
642 *offset
= cook
& 0x7fffffff;
643 xfs_trans_brelse(NULL
, bp
);
649 * Reached the end of the block.
650 * Set the offset to a non-existent block 1 and return.
652 *offset
= xfs_dir2_db_off_to_dataptr(mp
, mp
->m_dirdatablk
+ 1, 0) &
654 xfs_trans_brelse(NULL
, bp
);
659 * Log leaf entries from the block.
662 xfs_dir2_block_log_leaf(
663 xfs_trans_t
*tp
, /* transaction structure */
664 struct xfs_buf
*bp
, /* block buffer */
665 int first
, /* index of first logged leaf */
666 int last
) /* index of last logged leaf */
668 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
669 xfs_dir2_leaf_entry_t
*blp
;
670 xfs_dir2_block_tail_t
*btp
;
672 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
, hdr
);
673 blp
= xfs_dir2_block_leaf_p(btp
);
674 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)&blp
[first
] - (char *)hdr
),
675 (uint
)((char *)&blp
[last
+ 1] - (char *)hdr
- 1));
679 * Log the block tail.
682 xfs_dir2_block_log_tail(
683 xfs_trans_t
*tp
, /* transaction structure */
684 struct xfs_buf
*bp
) /* block buffer */
686 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
687 xfs_dir2_block_tail_t
*btp
;
689 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
, hdr
);
690 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)btp
- (char *)hdr
),
691 (uint
)((char *)(btp
+ 1) - (char *)hdr
- 1));
695 * Look up an entry in the block. This is the external routine,
696 * xfs_dir2_block_lookup_int does the real work.
699 xfs_dir2_block_lookup(
700 xfs_da_args_t
*args
) /* dir lookup arguments */
702 xfs_dir2_data_hdr_t
*hdr
; /* block header */
703 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
704 struct xfs_buf
*bp
; /* block buffer */
705 xfs_dir2_block_tail_t
*btp
; /* block tail */
706 xfs_dir2_data_entry_t
*dep
; /* block data entry */
707 xfs_inode_t
*dp
; /* incore inode */
708 int ent
; /* entry index */
709 int error
; /* error return value */
710 xfs_mount_t
*mp
; /* filesystem mount point */
712 trace_xfs_dir2_block_lookup(args
);
715 * Get the buffer, look up the entry.
716 * If not found (ENOENT) then return, have no buffer.
718 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
)))
723 xfs_dir2_data_check(dp
, bp
);
724 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
725 blp
= xfs_dir2_block_leaf_p(btp
);
727 * Get the offset from the leaf entry, to point to the data.
729 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
730 xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(blp
[ent
].address
)));
732 * Fill in inode number, CI name if appropriate, release the block.
734 args
->inumber
= be64_to_cpu(dep
->inumber
);
735 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
736 xfs_trans_brelse(args
->trans
, bp
);
737 return XFS_ERROR(error
);
741 * Internal block lookup routine.
743 static int /* error */
744 xfs_dir2_block_lookup_int(
745 xfs_da_args_t
*args
, /* dir lookup arguments */
746 struct xfs_buf
**bpp
, /* returned block buffer */
747 int *entno
) /* returned entry number */
749 xfs_dir2_dataptr_t addr
; /* data entry address */
750 xfs_dir2_data_hdr_t
*hdr
; /* block header */
751 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
752 struct xfs_buf
*bp
; /* block buffer */
753 xfs_dir2_block_tail_t
*btp
; /* block tail */
754 xfs_dir2_data_entry_t
*dep
; /* block data entry */
755 xfs_inode_t
*dp
; /* incore inode */
756 int error
; /* error return value */
757 xfs_dahash_t hash
; /* found hash value */
758 int high
; /* binary search high index */
759 int low
; /* binary search low index */
760 int mid
; /* binary search current idx */
761 xfs_mount_t
*mp
; /* filesystem mount point */
762 xfs_trans_t
*tp
; /* transaction pointer */
763 enum xfs_dacmp cmp
; /* comparison result */
769 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
774 xfs_dir2_data_check(dp
, bp
);
775 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
776 blp
= xfs_dir2_block_leaf_p(btp
);
778 * Loop doing a binary search for our hash value.
779 * Find our entry, ENOENT if it's not there.
781 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; ; ) {
783 mid
= (low
+ high
) >> 1;
784 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
786 if (hash
< args
->hashval
)
791 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
792 xfs_trans_brelse(tp
, bp
);
793 return XFS_ERROR(ENOENT
);
797 * Back up to the first one with the right hash value.
799 while (mid
> 0 && be32_to_cpu(blp
[mid
- 1].hashval
) == args
->hashval
) {
803 * Now loop forward through all the entries with the
804 * right hash value looking for our name.
807 if ((addr
= be32_to_cpu(blp
[mid
].address
)) == XFS_DIR2_NULL_DATAPTR
)
810 * Get pointer to the entry from the leaf.
812 dep
= (xfs_dir2_data_entry_t
*)
813 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, addr
));
815 * Compare name and if it's an exact match, return the index
816 * and buffer. If it's the first case-insensitive match, store
817 * the index and buffer and continue looking for an exact match.
819 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
820 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
821 args
->cmpresult
= cmp
;
824 if (cmp
== XFS_CMP_EXACT
)
827 } while (++mid
< be32_to_cpu(btp
->count
) &&
828 be32_to_cpu(blp
[mid
].hashval
) == hash
);
830 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
832 * Here, we can only be doing a lookup (not a rename or replace).
833 * If a case-insensitive match was found earlier, return success.
835 if (args
->cmpresult
== XFS_CMP_CASE
)
838 * No match, release the buffer and return ENOENT.
840 xfs_trans_brelse(tp
, bp
);
841 return XFS_ERROR(ENOENT
);
845 * Remove an entry from a block format directory.
846 * If that makes the block small enough to fit in shortform, transform it.
849 xfs_dir2_block_removename(
850 xfs_da_args_t
*args
) /* directory operation args */
852 xfs_dir2_data_hdr_t
*hdr
; /* block header */
853 xfs_dir2_leaf_entry_t
*blp
; /* block leaf pointer */
854 struct xfs_buf
*bp
; /* block buffer */
855 xfs_dir2_block_tail_t
*btp
; /* block tail */
856 xfs_dir2_data_entry_t
*dep
; /* block data entry */
857 xfs_inode_t
*dp
; /* incore inode */
858 int ent
; /* block leaf entry index */
859 int error
; /* error return value */
860 xfs_mount_t
*mp
; /* filesystem mount point */
861 int needlog
; /* need to log block header */
862 int needscan
; /* need to fixup bestfree */
863 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
864 int size
; /* shortform size */
865 xfs_trans_t
*tp
; /* transaction pointer */
867 trace_xfs_dir2_block_removename(args
);
870 * Look up the entry in the block. Gets the buffer and entry index.
871 * It will always be there, the vnodeops level does a lookup first.
873 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
880 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
881 blp
= xfs_dir2_block_leaf_p(btp
);
883 * Point to the data entry using the leaf entry.
885 dep
= (xfs_dir2_data_entry_t
*)
886 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(blp
[ent
].address
)));
888 * Mark the data entry's space free.
890 needlog
= needscan
= 0;
891 xfs_dir2_data_make_free(tp
, bp
,
892 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)hdr
),
893 xfs_dir2_data_entsize(dep
->namelen
), &needlog
, &needscan
);
895 * Fix up the block tail.
897 be32_add_cpu(&btp
->stale
, 1);
898 xfs_dir2_block_log_tail(tp
, bp
);
900 * Remove the leaf entry by marking it stale.
902 blp
[ent
].address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
903 xfs_dir2_block_log_leaf(tp
, bp
, ent
, ent
);
905 * Fix up bestfree, log the header if necessary.
908 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
910 xfs_dir2_data_log_header(tp
, bp
);
911 xfs_dir2_data_check(dp
, bp
);
913 * See if the size as a shortform is good enough.
915 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
916 if (size
> XFS_IFORK_DSIZE(dp
))
920 * If it works, do the conversion.
922 return xfs_dir2_block_to_sf(args
, bp
, size
, &sfh
);
926 * Replace an entry in a V2 block directory.
927 * Change the inode number to the new value.
930 xfs_dir2_block_replace(
931 xfs_da_args_t
*args
) /* directory operation args */
933 xfs_dir2_data_hdr_t
*hdr
; /* block header */
934 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
935 struct xfs_buf
*bp
; /* block buffer */
936 xfs_dir2_block_tail_t
*btp
; /* block tail */
937 xfs_dir2_data_entry_t
*dep
; /* block data entry */
938 xfs_inode_t
*dp
; /* incore inode */
939 int ent
; /* leaf entry index */
940 int error
; /* error return value */
941 xfs_mount_t
*mp
; /* filesystem mount point */
943 trace_xfs_dir2_block_replace(args
);
946 * Lookup the entry in the directory. Get buffer and entry index.
947 * This will always succeed since the caller has already done a lookup.
949 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
955 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
956 blp
= xfs_dir2_block_leaf_p(btp
);
958 * Point to the data entry we need to change.
960 dep
= (xfs_dir2_data_entry_t
*)
961 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(blp
[ent
].address
)));
962 ASSERT(be64_to_cpu(dep
->inumber
) != args
->inumber
);
964 * Change the inode number to the new value.
966 dep
->inumber
= cpu_to_be64(args
->inumber
);
967 xfs_dir2_data_log_entry(args
->trans
, bp
, dep
);
968 xfs_dir2_data_check(dp
, bp
);
973 * Qsort comparison routine for the block leaf entries.
975 static int /* sort order */
977 const void *a
, /* first leaf entry */
978 const void *b
) /* second leaf entry */
980 const xfs_dir2_leaf_entry_t
*la
; /* first leaf entry */
981 const xfs_dir2_leaf_entry_t
*lb
; /* second leaf entry */
985 return be32_to_cpu(la
->hashval
) < be32_to_cpu(lb
->hashval
) ? -1 :
986 (be32_to_cpu(la
->hashval
) > be32_to_cpu(lb
->hashval
) ? 1 : 0);
990 * Convert a V2 leaf directory to a V2 block directory if possible.
993 xfs_dir2_leaf_to_block(
994 xfs_da_args_t
*args
, /* operation arguments */
995 struct xfs_buf
*lbp
, /* leaf buffer */
996 struct xfs_buf
*dbp
) /* data buffer */
998 __be16
*bestsp
; /* leaf bests table */
999 xfs_dir2_data_hdr_t
*hdr
; /* block header */
1000 xfs_dir2_block_tail_t
*btp
; /* block tail */
1001 xfs_inode_t
*dp
; /* incore directory inode */
1002 xfs_dir2_data_unused_t
*dup
; /* unused data entry */
1003 int error
; /* error return value */
1004 int from
; /* leaf from index */
1005 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1006 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1007 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1008 xfs_mount_t
*mp
; /* file system mount point */
1009 int needlog
; /* need to log data header */
1010 int needscan
; /* need to scan for bestfree */
1011 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
1012 int size
; /* bytes used */
1013 __be16
*tagp
; /* end of entry (tag) */
1014 int to
; /* block/leaf to index */
1015 xfs_trans_t
*tp
; /* transaction pointer */
1017 trace_xfs_dir2_leaf_to_block(args
);
1023 ASSERT(leaf
->hdr
.info
.magic
== cpu_to_be16(XFS_DIR2_LEAF1_MAGIC
));
1024 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1026 * If there are data blocks other than the first one, take this
1027 * opportunity to remove trailing empty data blocks that may have
1028 * been left behind during no-space-reservation operations.
1029 * These will show up in the leaf bests table.
1031 while (dp
->i_d
.di_size
> mp
->m_dirblksize
) {
1034 hdrsz
= xfs_dir3_data_hdr_size(xfs_sb_version_hascrc(&mp
->m_sb
));
1035 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1036 if (be16_to_cpu(bestsp
[be32_to_cpu(ltp
->bestcount
) - 1]) ==
1037 mp
->m_dirblksize
- hdrsz
) {
1039 xfs_dir2_leaf_trim_data(args
, lbp
,
1040 (xfs_dir2_db_t
)(be32_to_cpu(ltp
->bestcount
) - 1))))
1046 * Read the data block if we don't already have it, give up if it fails.
1049 error
= xfs_dir2_data_read(tp
, dp
, mp
->m_dirdatablk
, -1, &dbp
);
1054 ASSERT(hdr
->magic
== cpu_to_be32(XFS_DIR2_DATA_MAGIC
));
1056 * Size of the "leaf" area in the block.
1058 size
= (uint
)sizeof(xfs_dir2_block_tail_t
) +
1059 (uint
)sizeof(*lep
) * (be16_to_cpu(leaf
->hdr
.count
) - be16_to_cpu(leaf
->hdr
.stale
));
1061 * Look at the last data entry.
1063 tagp
= (__be16
*)((char *)hdr
+ mp
->m_dirblksize
) - 1;
1064 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
1066 * If it's not free or is too short we can't do it.
1068 if (be16_to_cpu(dup
->freetag
) != XFS_DIR2_DATA_FREE_TAG
||
1069 be16_to_cpu(dup
->length
) < size
)
1073 * Start converting it to block form.
1075 xfs_dir3_block_init(mp
, dbp
, dp
);
1080 * Use up the space at the end of the block (blp/btp).
1082 xfs_dir2_data_use_free(tp
, dbp
, dup
, mp
->m_dirblksize
- size
, size
,
1083 &needlog
, &needscan
);
1085 * Initialize the block tail.
1087 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
1088 btp
->count
= cpu_to_be32(be16_to_cpu(leaf
->hdr
.count
) - be16_to_cpu(leaf
->hdr
.stale
));
1090 xfs_dir2_block_log_tail(tp
, dbp
);
1092 * Initialize the block leaf area. We compact out stale entries.
1094 lep
= xfs_dir2_block_leaf_p(btp
);
1095 for (from
= to
= 0; from
< be16_to_cpu(leaf
->hdr
.count
); from
++) {
1096 if (leaf
->ents
[from
].address
==
1097 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
1099 lep
[to
++] = leaf
->ents
[from
];
1101 ASSERT(to
== be32_to_cpu(btp
->count
));
1102 xfs_dir2_block_log_leaf(tp
, dbp
, 0, be32_to_cpu(btp
->count
) - 1);
1104 * Scan the bestfree if we need it and log the data block header.
1107 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
1109 xfs_dir2_data_log_header(tp
, dbp
);
1111 * Pitch the old leaf block.
1113 error
= xfs_da_shrink_inode(args
, mp
->m_dirleafblk
, lbp
);
1118 * Now see if the resulting block can be shrunken to shortform.
1120 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
1121 if (size
> XFS_IFORK_DSIZE(dp
))
1124 return xfs_dir2_block_to_sf(args
, dbp
, size
, &sfh
);
1128 * Convert the shortform directory to block form.
1131 xfs_dir2_sf_to_block(
1132 xfs_da_args_t
*args
) /* operation arguments */
1134 xfs_dir2_db_t blkno
; /* dir-relative block # (0) */
1135 xfs_dir2_data_hdr_t
*hdr
; /* block header */
1136 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
1137 struct xfs_buf
*bp
; /* block buffer */
1138 xfs_dir2_block_tail_t
*btp
; /* block tail pointer */
1139 xfs_dir2_data_entry_t
*dep
; /* data entry pointer */
1140 xfs_inode_t
*dp
; /* incore directory inode */
1141 int dummy
; /* trash */
1142 xfs_dir2_data_unused_t
*dup
; /* unused entry pointer */
1143 int endoffset
; /* end of data objects */
1144 int error
; /* error return value */
1146 xfs_mount_t
*mp
; /* filesystem mount point */
1147 int needlog
; /* need to log block header */
1148 int needscan
; /* need to scan block freespc */
1149 int newoffset
; /* offset from current entry */
1150 int offset
; /* target block offset */
1151 xfs_dir2_sf_entry_t
*sfep
; /* sf entry pointer */
1152 xfs_dir2_sf_hdr_t
*oldsfp
; /* old shortform header */
1153 xfs_dir2_sf_hdr_t
*sfp
; /* shortform header */
1154 __be16
*tagp
; /* end of data entry */
1155 xfs_trans_t
*tp
; /* transaction pointer */
1156 struct xfs_name name
;
1158 trace_xfs_dir2_sf_to_block(args
);
1163 ASSERT(dp
->i_df
.if_flags
& XFS_IFINLINE
);
1165 * Bomb out if the shortform directory is way too short.
1167 if (dp
->i_d
.di_size
< offsetof(xfs_dir2_sf_hdr_t
, parent
)) {
1168 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1169 return XFS_ERROR(EIO
);
1172 oldsfp
= (xfs_dir2_sf_hdr_t
*)dp
->i_df
.if_u1
.if_data
;
1174 ASSERT(dp
->i_df
.if_bytes
== dp
->i_d
.di_size
);
1175 ASSERT(dp
->i_df
.if_u1
.if_data
!= NULL
);
1176 ASSERT(dp
->i_d
.di_size
>= xfs_dir2_sf_hdr_size(oldsfp
->i8count
));
1179 * Copy the directory into a temporary buffer.
1180 * Then pitch the incore inode data so we can make extents.
1182 sfp
= kmem_alloc(dp
->i_df
.if_bytes
, KM_SLEEP
);
1183 memcpy(sfp
, oldsfp
, dp
->i_df
.if_bytes
);
1185 xfs_idata_realloc(dp
, -dp
->i_df
.if_bytes
, XFS_DATA_FORK
);
1186 dp
->i_d
.di_size
= 0;
1187 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1190 * Add block 0 to the inode.
1192 error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
, &blkno
);
1198 * Initialize the data block, then convert it to block format.
1200 error
= xfs_dir3_data_init(args
, blkno
, &bp
);
1205 xfs_dir3_block_init(mp
, bp
, dp
);
1209 * Compute size of block "tail" area.
1211 i
= (uint
)sizeof(*btp
) +
1212 (sfp
->count
+ 2) * (uint
)sizeof(xfs_dir2_leaf_entry_t
);
1214 * The whole thing is initialized to free by the init routine.
1215 * Say we're using the leaf and tail area.
1217 dup
= xfs_dir3_data_unused_p(hdr
);
1218 needlog
= needscan
= 0;
1219 xfs_dir2_data_use_free(tp
, bp
, dup
, mp
->m_dirblksize
- i
, i
, &needlog
,
1221 ASSERT(needscan
== 0);
1225 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
1226 btp
->count
= cpu_to_be32(sfp
->count
+ 2); /* ., .. */
1228 blp
= xfs_dir2_block_leaf_p(btp
);
1229 endoffset
= (uint
)((char *)blp
- (char *)hdr
);
1231 * Remove the freespace, we'll manage it.
1233 xfs_dir2_data_use_free(tp
, bp
, dup
,
1234 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
1235 be16_to_cpu(dup
->length
), &needlog
, &needscan
);
1237 * Create entry for .
1239 dep
= xfs_dir3_data_dot_entry_p(hdr
);
1240 dep
->inumber
= cpu_to_be64(dp
->i_ino
);
1243 tagp
= xfs_dir2_data_entry_tag_p(dep
);
1244 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1245 xfs_dir2_data_log_entry(tp
, bp
, dep
);
1246 blp
[0].hashval
= cpu_to_be32(xfs_dir_hash_dot
);
1247 blp
[0].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
1248 (char *)dep
- (char *)hdr
));
1250 * Create entry for ..
1252 dep
= xfs_dir3_data_dotdot_entry_p(hdr
);
1253 dep
->inumber
= cpu_to_be64(xfs_dir2_sf_get_parent_ino(sfp
));
1255 dep
->name
[0] = dep
->name
[1] = '.';
1256 tagp
= xfs_dir2_data_entry_tag_p(dep
);
1257 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1258 xfs_dir2_data_log_entry(tp
, bp
, dep
);
1259 blp
[1].hashval
= cpu_to_be32(xfs_dir_hash_dotdot
);
1260 blp
[1].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
1261 (char *)dep
- (char *)hdr
));
1262 offset
= xfs_dir3_data_first_offset(hdr
);
1264 * Loop over existing entries, stuff them in.
1270 sfep
= xfs_dir2_sf_firstentry(sfp
);
1272 * Need to preserve the existing offset values in the sf directory.
1273 * Insert holes (unused entries) where necessary.
1275 while (offset
< endoffset
) {
1277 * sfep is null when we reach the end of the list.
1280 newoffset
= endoffset
;
1282 newoffset
= xfs_dir2_sf_get_offset(sfep
);
1284 * There should be a hole here, make one.
1286 if (offset
< newoffset
) {
1287 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ offset
);
1288 dup
->freetag
= cpu_to_be16(XFS_DIR2_DATA_FREE_TAG
);
1289 dup
->length
= cpu_to_be16(newoffset
- offset
);
1290 *xfs_dir2_data_unused_tag_p(dup
) = cpu_to_be16(
1291 ((char *)dup
- (char *)hdr
));
1292 xfs_dir2_data_log_unused(tp
, bp
, dup
);
1293 xfs_dir2_data_freeinsert(hdr
, dup
, &dummy
);
1294 offset
+= be16_to_cpu(dup
->length
);
1298 * Copy a real entry.
1300 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+ newoffset
);
1301 dep
->inumber
= cpu_to_be64(xfs_dir2_sfe_get_ino(sfp
, sfep
));
1302 dep
->namelen
= sfep
->namelen
;
1303 memcpy(dep
->name
, sfep
->name
, dep
->namelen
);
1304 tagp
= xfs_dir2_data_entry_tag_p(dep
);
1305 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1306 xfs_dir2_data_log_entry(tp
, bp
, dep
);
1307 name
.name
= sfep
->name
;
1308 name
.len
= sfep
->namelen
;
1309 blp
[2 + i
].hashval
= cpu_to_be32(mp
->m_dirnameops
->
1311 blp
[2 + i
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
1312 (char *)dep
- (char *)hdr
));
1313 offset
= (int)((char *)(tagp
+ 1) - (char *)hdr
);
1314 if (++i
== sfp
->count
)
1317 sfep
= xfs_dir2_sf_nextentry(sfp
, sfep
);
1319 /* Done with the temporary buffer */
1322 * Sort the leaf entries by hash value.
1324 xfs_sort(blp
, be32_to_cpu(btp
->count
), sizeof(*blp
), xfs_dir2_block_sort
);
1326 * Log the leaf entry area and tail.
1327 * Already logged the header in data_init, ignore needlog.
1329 ASSERT(needscan
== 0);
1330 xfs_dir2_block_log_leaf(tp
, bp
, 0, be32_to_cpu(btp
->count
) - 1);
1331 xfs_dir2_block_log_tail(tp
, bp
);
1332 xfs_dir2_data_check(dp
, bp
);