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_dir3_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
;
137 err
= xfs_da_read_buf(tp
, dp
, mp
->m_dirdatablk
, -1, bpp
,
138 XFS_DATA_FORK
, &xfs_dir3_block_buf_ops
);
140 xfs_trans_buf_set_type(tp
, *bpp
, XFS_BLF_DIR_BLOCK_BUF
);
146 struct xfs_mount
*mp
,
147 struct xfs_trans
*tp
,
149 struct xfs_inode
*dp
)
151 struct xfs_dir3_blk_hdr
*hdr3
= bp
->b_addr
;
153 bp
->b_ops
= &xfs_dir3_block_buf_ops
;
154 xfs_trans_buf_set_type(tp
, bp
, XFS_BLF_DIR_BLOCK_BUF
);
156 if (xfs_sb_version_hascrc(&mp
->m_sb
)) {
157 memset(hdr3
, 0, sizeof(*hdr3
));
158 hdr3
->magic
= cpu_to_be32(XFS_DIR3_BLOCK_MAGIC
);
159 hdr3
->blkno
= cpu_to_be64(bp
->b_bn
);
160 hdr3
->owner
= cpu_to_be64(dp
->i_ino
);
161 uuid_copy(&hdr3
->uuid
, &mp
->m_sb
.sb_uuid
);
165 hdr3
->magic
= cpu_to_be32(XFS_DIR2_BLOCK_MAGIC
);
169 xfs_dir2_block_need_space(
170 struct xfs_dir2_data_hdr
*hdr
,
171 struct xfs_dir2_block_tail
*btp
,
172 struct xfs_dir2_leaf_entry
*blp
,
174 struct xfs_dir2_data_unused
**dupp
,
175 struct xfs_dir2_data_unused
**enddupp
,
179 struct xfs_dir2_data_free
*bf
;
181 struct xfs_dir2_data_unused
*dup
= NULL
;
182 struct xfs_dir2_data_unused
*enddup
= NULL
;
185 bf
= xfs_dir3_data_bestfree_p(hdr
);
188 * If there are stale entries we'll use one for the leaf.
191 if (be16_to_cpu(bf
[0].length
) >= len
) {
193 * The biggest entry enough to avoid compaction.
195 dup
= (xfs_dir2_data_unused_t
*)
196 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
201 * Will need to compact to make this work.
202 * Tag just before the first leaf entry.
205 tagp
= (__be16
*)blp
- 1;
207 /* Data object just before the first leaf entry. */
208 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
211 * If it's not free then the data will go where the
212 * leaf data starts now, if it works at all.
214 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
215 if (be16_to_cpu(dup
->length
) + (be32_to_cpu(btp
->stale
) - 1) *
216 (uint
)sizeof(*blp
) < len
)
218 } else if ((be32_to_cpu(btp
->stale
) - 1) * (uint
)sizeof(*blp
) < len
)
221 dup
= (xfs_dir2_data_unused_t
*)blp
;
226 * no stale entries, so just use free space.
227 * Tag just before the first leaf entry.
229 tagp
= (__be16
*)blp
- 1;
231 /* Data object just before the first leaf entry. */
232 enddup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
235 * If it's not free then can't do this add without cleaning up:
236 * the space before the first leaf entry needs to be free so it
237 * can be expanded to hold the pointer to the new entry.
239 if (be16_to_cpu(enddup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
241 * Check out the biggest freespace and see if it's the same one.
243 dup
= (xfs_dir2_data_unused_t
*)
244 ((char *)hdr
+ be16_to_cpu(bf
[0].offset
));
247 * Not the same free entry, just check its length.
249 if (be16_to_cpu(dup
->length
) < len
)
255 * It is the biggest freespace, can it hold the leaf too?
257 if (be16_to_cpu(dup
->length
) < len
+ (uint
)sizeof(*blp
)) {
259 * Yes, use the second-largest entry instead if it works.
261 if (be16_to_cpu(bf
[1].length
) >= len
)
262 dup
= (xfs_dir2_data_unused_t
*)
263 ((char *)hdr
+ be16_to_cpu(bf
[1].offset
));
275 * compact the leaf entries.
276 * Leave the highest-numbered stale entry stale.
277 * XXX should be the one closest to mid but mid is not yet computed.
280 xfs_dir2_block_compact(
281 struct xfs_trans
*tp
,
283 struct xfs_dir2_data_hdr
*hdr
,
284 struct xfs_dir2_block_tail
*btp
,
285 struct xfs_dir2_leaf_entry
*blp
,
290 int fromidx
; /* source leaf index */
291 int toidx
; /* target leaf index */
293 int highstale
; /* high stale index */
295 fromidx
= toidx
= be32_to_cpu(btp
->count
) - 1;
296 highstale
= *lfloghigh
= -1;
297 for (; fromidx
>= 0; fromidx
--) {
298 if (blp
[fromidx
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
)) {
302 if (*lfloghigh
== -1)
308 blp
[toidx
] = blp
[fromidx
];
311 *lfloglow
= toidx
+ 1 - (be32_to_cpu(btp
->stale
) - 1);
312 *lfloghigh
-= be32_to_cpu(btp
->stale
) - 1;
313 be32_add_cpu(&btp
->count
, -(be32_to_cpu(btp
->stale
) - 1));
314 xfs_dir2_data_make_free(tp
, bp
,
315 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)hdr
),
316 (xfs_dir2_data_aoff_t
)((be32_to_cpu(btp
->stale
) - 1) * sizeof(*blp
)),
318 blp
+= be32_to_cpu(btp
->stale
) - 1;
319 btp
->stale
= cpu_to_be32(1);
321 * If we now need to rebuild the bestfree map, do so.
322 * This needs to happen before the next call to use_free.
325 xfs_dir2_data_freescan(tp
->t_mountp
, hdr
, needlog
);
329 * Add an entry to a block directory.
332 xfs_dir2_block_addname(
333 xfs_da_args_t
*args
) /* directory op arguments */
335 xfs_dir2_data_hdr_t
*hdr
; /* block header */
336 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
337 struct xfs_buf
*bp
; /* buffer for block */
338 xfs_dir2_block_tail_t
*btp
; /* block tail */
339 int compact
; /* need to compact leaf ents */
340 xfs_dir2_data_entry_t
*dep
; /* block data entry */
341 xfs_inode_t
*dp
; /* directory inode */
342 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
343 int error
; /* error return value */
344 xfs_dir2_data_unused_t
*enddup
=NULL
; /* unused at end of data */
345 xfs_dahash_t hash
; /* hash value of found entry */
346 int high
; /* high index for binary srch */
347 int highstale
; /* high stale index */
348 int lfloghigh
=0; /* last final leaf to log */
349 int lfloglow
=0; /* first final leaf to log */
350 int len
; /* length of the new entry */
351 int low
; /* low index for binary srch */
352 int lowstale
; /* low stale index */
353 int mid
=0; /* midpoint for binary srch */
354 xfs_mount_t
*mp
; /* filesystem mount point */
355 int needlog
; /* need to log header */
356 int needscan
; /* need to rescan freespace */
357 __be16
*tagp
; /* pointer to tag value */
358 xfs_trans_t
*tp
; /* transaction structure */
360 trace_xfs_dir2_block_addname(args
);
366 /* Read the (one and only) directory block into bp. */
367 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
371 len
= xfs_dir2_data_entsize(args
->namelen
);
374 * Set up pointers to parts of the block.
377 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
378 blp
= xfs_dir2_block_leaf_p(btp
);
381 * Find out if we can reuse stale entries or whether we need extra
382 * space for entry and new leaf.
384 xfs_dir2_block_need_space(hdr
, btp
, blp
, &tagp
, &dup
,
385 &enddup
, &compact
, len
);
388 * Done everything we need for a space check now.
390 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
391 xfs_trans_brelse(tp
, bp
);
393 return XFS_ERROR(ENOSPC
);
398 * If we don't have space for the new entry & leaf ...
401 /* Don't have a space reservation: return no-space. */
402 if (args
->total
== 0)
403 return XFS_ERROR(ENOSPC
);
405 * Convert to the next larger format.
406 * Then add the new entry in that format.
408 error
= xfs_dir2_block_to_leaf(args
, bp
);
411 return xfs_dir2_leaf_addname(args
);
414 needlog
= needscan
= 0;
417 * If need to compact the leaf entries, do it now.
420 xfs_dir2_block_compact(tp
, bp
, hdr
, btp
, blp
, &needlog
,
421 &lfloghigh
, &lfloglow
);
422 /* recalculate blp post-compaction */
423 blp
= xfs_dir2_block_leaf_p(btp
);
424 } else if (btp
->stale
) {
426 * Set leaf logging boundaries to impossible state.
427 * For the no-stale case they're set explicitly.
429 lfloglow
= be32_to_cpu(btp
->count
);
434 * Find the slot that's first lower than our hash value, -1 if none.
436 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; low
<= high
; ) {
437 mid
= (low
+ high
) >> 1;
438 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
440 if (hash
< args
->hashval
)
445 while (mid
>= 0 && be32_to_cpu(blp
[mid
].hashval
) >= args
->hashval
) {
449 * No stale entries, will use enddup space to hold new leaf.
453 * Mark the space needed for the new leaf entry, now in use.
455 xfs_dir2_data_use_free(tp
, bp
, enddup
,
456 (xfs_dir2_data_aoff_t
)
457 ((char *)enddup
- (char *)hdr
+ be16_to_cpu(enddup
->length
) -
459 (xfs_dir2_data_aoff_t
)sizeof(*blp
),
460 &needlog
, &needscan
);
462 * Update the tail (entry count).
464 be32_add_cpu(&btp
->count
, 1);
466 * If we now need to rebuild the bestfree map, do so.
467 * This needs to happen before the next call to use_free.
470 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
474 * Adjust pointer to the first leaf entry, we're about to move
475 * the table up one to open up space for the new leaf entry.
476 * Then adjust our index to match.
481 memmove(blp
, &blp
[1], mid
* sizeof(*blp
));
486 * Use a stale leaf for our new entry.
491 blp
[lowstale
].address
!=
492 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
495 for (highstale
= mid
+ 1;
496 highstale
< be32_to_cpu(btp
->count
) &&
497 blp
[highstale
].address
!=
498 cpu_to_be32(XFS_DIR2_NULL_DATAPTR
) &&
499 (lowstale
< 0 || mid
- lowstale
> highstale
- mid
);
503 * Move entries toward the low-numbered stale entry.
506 (highstale
== be32_to_cpu(btp
->count
) ||
507 mid
- lowstale
<= highstale
- mid
)) {
509 memmove(&blp
[lowstale
], &blp
[lowstale
+ 1],
510 (mid
- lowstale
) * sizeof(*blp
));
511 lfloglow
= MIN(lowstale
, lfloglow
);
512 lfloghigh
= MAX(mid
, lfloghigh
);
515 * Move entries toward the high-numbered stale entry.
518 ASSERT(highstale
< be32_to_cpu(btp
->count
));
521 memmove(&blp
[mid
+ 1], &blp
[mid
],
522 (highstale
- mid
) * sizeof(*blp
));
523 lfloglow
= MIN(mid
, lfloglow
);
524 lfloghigh
= MAX(highstale
, lfloghigh
);
526 be32_add_cpu(&btp
->stale
, -1);
529 * Point to the new data entry.
531 dep
= (xfs_dir2_data_entry_t
*)dup
;
533 * Fill in the leaf entry.
535 blp
[mid
].hashval
= cpu_to_be32(args
->hashval
);
536 blp
[mid
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
537 (char *)dep
- (char *)hdr
));
538 xfs_dir2_block_log_leaf(tp
, bp
, lfloglow
, lfloghigh
);
540 * Mark space for the data entry used.
542 xfs_dir2_data_use_free(tp
, bp
, dup
,
543 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
544 (xfs_dir2_data_aoff_t
)len
, &needlog
, &needscan
);
546 * Create the new data entry.
548 dep
->inumber
= cpu_to_be64(args
->inumber
);
549 dep
->namelen
= args
->namelen
;
550 memcpy(dep
->name
, args
->name
, args
->namelen
);
551 tagp
= xfs_dir2_data_entry_tag_p(dep
);
552 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
554 * Clean up the bestfree array and log the header, tail, and entry.
557 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
559 xfs_dir2_data_log_header(tp
, bp
);
560 xfs_dir2_block_log_tail(tp
, bp
);
561 xfs_dir2_data_log_entry(tp
, bp
, dep
);
562 xfs_dir3_data_check(dp
, bp
);
567 * Readdir for block directories.
570 xfs_dir2_block_getdents(
571 xfs_inode_t
*dp
, /* incore inode */
576 xfs_dir2_data_hdr_t
*hdr
; /* block header */
577 struct xfs_buf
*bp
; /* buffer for block */
578 xfs_dir2_block_tail_t
*btp
; /* block tail */
579 xfs_dir2_data_entry_t
*dep
; /* block data entry */
580 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
581 char *endptr
; /* end of the data entries */
582 int error
; /* error return value */
583 xfs_mount_t
*mp
; /* filesystem mount point */
584 char *ptr
; /* current data entry */
585 int wantoff
; /* starting block offset */
590 * If the block number in the offset is out of range, we're done.
592 if (xfs_dir2_dataptr_to_db(mp
, *offset
) > mp
->m_dirdatablk
)
595 error
= xfs_dir3_block_read(NULL
, dp
, &bp
);
600 * Extract the byte offset we start at from the seek pointer.
601 * We'll skip entries before this.
603 wantoff
= xfs_dir2_dataptr_to_off(mp
, *offset
);
605 xfs_dir3_data_check(dp
, bp
);
607 * Set up values for the loop.
609 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
610 ptr
= (char *)xfs_dir3_data_entry_p(hdr
);
611 endptr
= (char *)xfs_dir2_block_leaf_p(btp
);
614 * Loop over the data portion of the block.
615 * Each object is a real entry (dep) or an unused one (dup).
617 while (ptr
< endptr
) {
618 dup
= (xfs_dir2_data_unused_t
*)ptr
;
622 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
623 ptr
+= be16_to_cpu(dup
->length
);
627 dep
= (xfs_dir2_data_entry_t
*)ptr
;
630 * Bump pointer for the next iteration.
632 ptr
+= xfs_dir2_data_entsize(dep
->namelen
);
634 * The entry is before the desired starting point, skip it.
636 if ((char *)dep
- (char *)hdr
< wantoff
)
639 cook
= xfs_dir2_db_off_to_dataptr(mp
, mp
->m_dirdatablk
,
640 (char *)dep
- (char *)hdr
);
643 * If it didn't fit, set the final offset to here & return.
645 if (filldir(dirent
, (char *)dep
->name
, dep
->namelen
,
646 cook
& 0x7fffffff, be64_to_cpu(dep
->inumber
),
648 *offset
= cook
& 0x7fffffff;
649 xfs_trans_brelse(NULL
, bp
);
655 * Reached the end of the block.
656 * Set the offset to a non-existent block 1 and return.
658 *offset
= xfs_dir2_db_off_to_dataptr(mp
, mp
->m_dirdatablk
+ 1, 0) &
660 xfs_trans_brelse(NULL
, bp
);
665 * Log leaf entries from the block.
668 xfs_dir2_block_log_leaf(
669 xfs_trans_t
*tp
, /* transaction structure */
670 struct xfs_buf
*bp
, /* block buffer */
671 int first
, /* index of first logged leaf */
672 int last
) /* index of last logged leaf */
674 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
675 xfs_dir2_leaf_entry_t
*blp
;
676 xfs_dir2_block_tail_t
*btp
;
678 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
, hdr
);
679 blp
= xfs_dir2_block_leaf_p(btp
);
680 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)&blp
[first
] - (char *)hdr
),
681 (uint
)((char *)&blp
[last
+ 1] - (char *)hdr
- 1));
685 * Log the block tail.
688 xfs_dir2_block_log_tail(
689 xfs_trans_t
*tp
, /* transaction structure */
690 struct xfs_buf
*bp
) /* block buffer */
692 xfs_dir2_data_hdr_t
*hdr
= bp
->b_addr
;
693 xfs_dir2_block_tail_t
*btp
;
695 btp
= xfs_dir2_block_tail_p(tp
->t_mountp
, hdr
);
696 xfs_trans_log_buf(tp
, bp
, (uint
)((char *)btp
- (char *)hdr
),
697 (uint
)((char *)(btp
+ 1) - (char *)hdr
- 1));
701 * Look up an entry in the block. This is the external routine,
702 * xfs_dir2_block_lookup_int does the real work.
705 xfs_dir2_block_lookup(
706 xfs_da_args_t
*args
) /* dir lookup arguments */
708 xfs_dir2_data_hdr_t
*hdr
; /* block header */
709 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
710 struct xfs_buf
*bp
; /* block buffer */
711 xfs_dir2_block_tail_t
*btp
; /* block tail */
712 xfs_dir2_data_entry_t
*dep
; /* block data entry */
713 xfs_inode_t
*dp
; /* incore inode */
714 int ent
; /* entry index */
715 int error
; /* error return value */
716 xfs_mount_t
*mp
; /* filesystem mount point */
718 trace_xfs_dir2_block_lookup(args
);
721 * Get the buffer, look up the entry.
722 * If not found (ENOENT) then return, have no buffer.
724 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
)))
729 xfs_dir3_data_check(dp
, bp
);
730 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
731 blp
= xfs_dir2_block_leaf_p(btp
);
733 * Get the offset from the leaf entry, to point to the data.
735 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+
736 xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(blp
[ent
].address
)));
738 * Fill in inode number, CI name if appropriate, release the block.
740 args
->inumber
= be64_to_cpu(dep
->inumber
);
741 error
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
742 xfs_trans_brelse(args
->trans
, bp
);
743 return XFS_ERROR(error
);
747 * Internal block lookup routine.
749 static int /* error */
750 xfs_dir2_block_lookup_int(
751 xfs_da_args_t
*args
, /* dir lookup arguments */
752 struct xfs_buf
**bpp
, /* returned block buffer */
753 int *entno
) /* returned entry number */
755 xfs_dir2_dataptr_t addr
; /* data entry address */
756 xfs_dir2_data_hdr_t
*hdr
; /* block header */
757 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
758 struct xfs_buf
*bp
; /* block buffer */
759 xfs_dir2_block_tail_t
*btp
; /* block tail */
760 xfs_dir2_data_entry_t
*dep
; /* block data entry */
761 xfs_inode_t
*dp
; /* incore inode */
762 int error
; /* error return value */
763 xfs_dahash_t hash
; /* found hash value */
764 int high
; /* binary search high index */
765 int low
; /* binary search low index */
766 int mid
; /* binary search current idx */
767 xfs_mount_t
*mp
; /* filesystem mount point */
768 xfs_trans_t
*tp
; /* transaction pointer */
769 enum xfs_dacmp cmp
; /* comparison result */
775 error
= xfs_dir3_block_read(tp
, dp
, &bp
);
780 xfs_dir3_data_check(dp
, bp
);
781 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
782 blp
= xfs_dir2_block_leaf_p(btp
);
784 * Loop doing a binary search for our hash value.
785 * Find our entry, ENOENT if it's not there.
787 for (low
= 0, high
= be32_to_cpu(btp
->count
) - 1; ; ) {
789 mid
= (low
+ high
) >> 1;
790 if ((hash
= be32_to_cpu(blp
[mid
].hashval
)) == args
->hashval
)
792 if (hash
< args
->hashval
)
797 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
798 xfs_trans_brelse(tp
, bp
);
799 return XFS_ERROR(ENOENT
);
803 * Back up to the first one with the right hash value.
805 while (mid
> 0 && be32_to_cpu(blp
[mid
- 1].hashval
) == args
->hashval
) {
809 * Now loop forward through all the entries with the
810 * right hash value looking for our name.
813 if ((addr
= be32_to_cpu(blp
[mid
].address
)) == XFS_DIR2_NULL_DATAPTR
)
816 * Get pointer to the entry from the leaf.
818 dep
= (xfs_dir2_data_entry_t
*)
819 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, addr
));
821 * Compare name and if it's an exact match, return the index
822 * and buffer. If it's the first case-insensitive match, store
823 * the index and buffer and continue looking for an exact match.
825 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
826 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
827 args
->cmpresult
= cmp
;
830 if (cmp
== XFS_CMP_EXACT
)
833 } while (++mid
< be32_to_cpu(btp
->count
) &&
834 be32_to_cpu(blp
[mid
].hashval
) == hash
);
836 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
838 * Here, we can only be doing a lookup (not a rename or replace).
839 * If a case-insensitive match was found earlier, return success.
841 if (args
->cmpresult
== XFS_CMP_CASE
)
844 * No match, release the buffer and return ENOENT.
846 xfs_trans_brelse(tp
, bp
);
847 return XFS_ERROR(ENOENT
);
851 * Remove an entry from a block format directory.
852 * If that makes the block small enough to fit in shortform, transform it.
855 xfs_dir2_block_removename(
856 xfs_da_args_t
*args
) /* directory operation args */
858 xfs_dir2_data_hdr_t
*hdr
; /* block header */
859 xfs_dir2_leaf_entry_t
*blp
; /* block leaf pointer */
860 struct xfs_buf
*bp
; /* block buffer */
861 xfs_dir2_block_tail_t
*btp
; /* block tail */
862 xfs_dir2_data_entry_t
*dep
; /* block data entry */
863 xfs_inode_t
*dp
; /* incore inode */
864 int ent
; /* block leaf entry index */
865 int error
; /* error return value */
866 xfs_mount_t
*mp
; /* filesystem mount point */
867 int needlog
; /* need to log block header */
868 int needscan
; /* need to fixup bestfree */
869 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
870 int size
; /* shortform size */
871 xfs_trans_t
*tp
; /* transaction pointer */
873 trace_xfs_dir2_block_removename(args
);
876 * Look up the entry in the block. Gets the buffer and entry index.
877 * It will always be there, the vnodeops level does a lookup first.
879 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
886 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
887 blp
= xfs_dir2_block_leaf_p(btp
);
889 * Point to the data entry using the leaf entry.
891 dep
= (xfs_dir2_data_entry_t
*)
892 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(blp
[ent
].address
)));
894 * Mark the data entry's space free.
896 needlog
= needscan
= 0;
897 xfs_dir2_data_make_free(tp
, bp
,
898 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)hdr
),
899 xfs_dir2_data_entsize(dep
->namelen
), &needlog
, &needscan
);
901 * Fix up the block tail.
903 be32_add_cpu(&btp
->stale
, 1);
904 xfs_dir2_block_log_tail(tp
, bp
);
906 * Remove the leaf entry by marking it stale.
908 blp
[ent
].address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
909 xfs_dir2_block_log_leaf(tp
, bp
, ent
, ent
);
911 * Fix up bestfree, log the header if necessary.
914 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
916 xfs_dir2_data_log_header(tp
, bp
);
917 xfs_dir3_data_check(dp
, bp
);
919 * See if the size as a shortform is good enough.
921 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
922 if (size
> XFS_IFORK_DSIZE(dp
))
926 * If it works, do the conversion.
928 return xfs_dir2_block_to_sf(args
, bp
, size
, &sfh
);
932 * Replace an entry in a V2 block directory.
933 * Change the inode number to the new value.
936 xfs_dir2_block_replace(
937 xfs_da_args_t
*args
) /* directory operation args */
939 xfs_dir2_data_hdr_t
*hdr
; /* block header */
940 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
941 struct xfs_buf
*bp
; /* block buffer */
942 xfs_dir2_block_tail_t
*btp
; /* block tail */
943 xfs_dir2_data_entry_t
*dep
; /* block data entry */
944 xfs_inode_t
*dp
; /* incore inode */
945 int ent
; /* leaf entry index */
946 int error
; /* error return value */
947 xfs_mount_t
*mp
; /* filesystem mount point */
949 trace_xfs_dir2_block_replace(args
);
952 * Lookup the entry in the directory. Get buffer and entry index.
953 * This will always succeed since the caller has already done a lookup.
955 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
961 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
962 blp
= xfs_dir2_block_leaf_p(btp
);
964 * Point to the data entry we need to change.
966 dep
= (xfs_dir2_data_entry_t
*)
967 ((char *)hdr
+ xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(blp
[ent
].address
)));
968 ASSERT(be64_to_cpu(dep
->inumber
) != args
->inumber
);
970 * Change the inode number to the new value.
972 dep
->inumber
= cpu_to_be64(args
->inumber
);
973 xfs_dir2_data_log_entry(args
->trans
, bp
, dep
);
974 xfs_dir3_data_check(dp
, bp
);
979 * Qsort comparison routine for the block leaf entries.
981 static int /* sort order */
983 const void *a
, /* first leaf entry */
984 const void *b
) /* second leaf entry */
986 const xfs_dir2_leaf_entry_t
*la
; /* first leaf entry */
987 const xfs_dir2_leaf_entry_t
*lb
; /* second leaf entry */
991 return be32_to_cpu(la
->hashval
) < be32_to_cpu(lb
->hashval
) ? -1 :
992 (be32_to_cpu(la
->hashval
) > be32_to_cpu(lb
->hashval
) ? 1 : 0);
996 * Convert a V2 leaf directory to a V2 block directory if possible.
999 xfs_dir2_leaf_to_block(
1000 xfs_da_args_t
*args
, /* operation arguments */
1001 struct xfs_buf
*lbp
, /* leaf buffer */
1002 struct xfs_buf
*dbp
) /* data buffer */
1004 __be16
*bestsp
; /* leaf bests table */
1005 xfs_dir2_data_hdr_t
*hdr
; /* block header */
1006 xfs_dir2_block_tail_t
*btp
; /* block tail */
1007 xfs_inode_t
*dp
; /* incore directory inode */
1008 xfs_dir2_data_unused_t
*dup
; /* unused data entry */
1009 int error
; /* error return value */
1010 int from
; /* leaf from index */
1011 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1012 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
1013 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
1014 xfs_mount_t
*mp
; /* file system mount point */
1015 int needlog
; /* need to log data header */
1016 int needscan
; /* need to scan for bestfree */
1017 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
1018 int size
; /* bytes used */
1019 __be16
*tagp
; /* end of entry (tag) */
1020 int to
; /* block/leaf to index */
1021 xfs_trans_t
*tp
; /* transaction pointer */
1022 struct xfs_dir2_leaf_entry
*ents
;
1023 struct xfs_dir3_icleaf_hdr leafhdr
;
1025 trace_xfs_dir2_leaf_to_block(args
);
1031 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
1032 ents
= xfs_dir3_leaf_ents_p(leaf
);
1033 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1035 ASSERT(leafhdr
.magic
== XFS_DIR2_LEAF1_MAGIC
||
1036 leafhdr
.magic
== XFS_DIR3_LEAF1_MAGIC
);
1038 * If there are data blocks other than the first one, take this
1039 * opportunity to remove trailing empty data blocks that may have
1040 * been left behind during no-space-reservation operations.
1041 * These will show up in the leaf bests table.
1043 while (dp
->i_d
.di_size
> mp
->m_dirblksize
) {
1046 hdrsz
= xfs_dir3_data_hdr_size(xfs_sb_version_hascrc(&mp
->m_sb
));
1047 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1048 if (be16_to_cpu(bestsp
[be32_to_cpu(ltp
->bestcount
) - 1]) ==
1049 mp
->m_dirblksize
- hdrsz
) {
1051 xfs_dir2_leaf_trim_data(args
, lbp
,
1052 (xfs_dir2_db_t
)(be32_to_cpu(ltp
->bestcount
) - 1))))
1058 * Read the data block if we don't already have it, give up if it fails.
1061 error
= xfs_dir3_data_read(tp
, dp
, mp
->m_dirdatablk
, -1, &dbp
);
1066 ASSERT(hdr
->magic
== cpu_to_be32(XFS_DIR2_DATA_MAGIC
) ||
1067 hdr
->magic
== cpu_to_be32(XFS_DIR3_DATA_MAGIC
));
1070 * Size of the "leaf" area in the block.
1072 size
= (uint
)sizeof(xfs_dir2_block_tail_t
) +
1073 (uint
)sizeof(*lep
) * (leafhdr
.count
- leafhdr
.stale
);
1075 * Look at the last data entry.
1077 tagp
= (__be16
*)((char *)hdr
+ mp
->m_dirblksize
) - 1;
1078 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ be16_to_cpu(*tagp
));
1080 * If it's not free or is too short we can't do it.
1082 if (be16_to_cpu(dup
->freetag
) != XFS_DIR2_DATA_FREE_TAG
||
1083 be16_to_cpu(dup
->length
) < size
)
1087 * Start converting it to block form.
1089 xfs_dir3_block_init(mp
, tp
, dbp
, dp
);
1094 * Use up the space at the end of the block (blp/btp).
1096 xfs_dir2_data_use_free(tp
, dbp
, dup
, mp
->m_dirblksize
- size
, size
,
1097 &needlog
, &needscan
);
1099 * Initialize the block tail.
1101 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
1102 btp
->count
= cpu_to_be32(leafhdr
.count
- leafhdr
.stale
);
1104 xfs_dir2_block_log_tail(tp
, dbp
);
1106 * Initialize the block leaf area. We compact out stale entries.
1108 lep
= xfs_dir2_block_leaf_p(btp
);
1109 for (from
= to
= 0; from
< leafhdr
.count
; from
++) {
1110 if (ents
[from
].address
== cpu_to_be32(XFS_DIR2_NULL_DATAPTR
))
1112 lep
[to
++] = ents
[from
];
1114 ASSERT(to
== be32_to_cpu(btp
->count
));
1115 xfs_dir2_block_log_leaf(tp
, dbp
, 0, be32_to_cpu(btp
->count
) - 1);
1117 * Scan the bestfree if we need it and log the data block header.
1120 xfs_dir2_data_freescan(mp
, hdr
, &needlog
);
1122 xfs_dir2_data_log_header(tp
, dbp
);
1124 * Pitch the old leaf block.
1126 error
= xfs_da_shrink_inode(args
, mp
->m_dirleafblk
, lbp
);
1131 * Now see if the resulting block can be shrunken to shortform.
1133 size
= xfs_dir2_block_sfsize(dp
, hdr
, &sfh
);
1134 if (size
> XFS_IFORK_DSIZE(dp
))
1137 return xfs_dir2_block_to_sf(args
, dbp
, size
, &sfh
);
1141 * Convert the shortform directory to block form.
1144 xfs_dir2_sf_to_block(
1145 xfs_da_args_t
*args
) /* operation arguments */
1147 xfs_dir2_db_t blkno
; /* dir-relative block # (0) */
1148 xfs_dir2_data_hdr_t
*hdr
; /* block header */
1149 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
1150 struct xfs_buf
*bp
; /* block buffer */
1151 xfs_dir2_block_tail_t
*btp
; /* block tail pointer */
1152 xfs_dir2_data_entry_t
*dep
; /* data entry pointer */
1153 xfs_inode_t
*dp
; /* incore directory inode */
1154 int dummy
; /* trash */
1155 xfs_dir2_data_unused_t
*dup
; /* unused entry pointer */
1156 int endoffset
; /* end of data objects */
1157 int error
; /* error return value */
1159 xfs_mount_t
*mp
; /* filesystem mount point */
1160 int needlog
; /* need to log block header */
1161 int needscan
; /* need to scan block freespc */
1162 int newoffset
; /* offset from current entry */
1163 int offset
; /* target block offset */
1164 xfs_dir2_sf_entry_t
*sfep
; /* sf entry pointer */
1165 xfs_dir2_sf_hdr_t
*oldsfp
; /* old shortform header */
1166 xfs_dir2_sf_hdr_t
*sfp
; /* shortform header */
1167 __be16
*tagp
; /* end of data entry */
1168 xfs_trans_t
*tp
; /* transaction pointer */
1169 struct xfs_name name
;
1171 trace_xfs_dir2_sf_to_block(args
);
1176 ASSERT(dp
->i_df
.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
*)dp
->i_df
.if_u1
.if_data
;
1187 ASSERT(dp
->i_df
.if_bytes
== dp
->i_d
.di_size
);
1188 ASSERT(dp
->i_df
.if_u1
.if_data
!= NULL
);
1189 ASSERT(dp
->i_d
.di_size
>= xfs_dir2_sf_hdr_size(oldsfp
->i8count
));
1192 * Copy the directory into a temporary buffer.
1193 * Then pitch the incore inode data so we can make extents.
1195 sfp
= kmem_alloc(dp
->i_df
.if_bytes
, KM_SLEEP
);
1196 memcpy(sfp
, oldsfp
, dp
->i_df
.if_bytes
);
1198 xfs_idata_realloc(dp
, -dp
->i_df
.if_bytes
, XFS_DATA_FORK
);
1199 dp
->i_d
.di_size
= 0;
1200 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1203 * Add block 0 to the inode.
1205 error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
, &blkno
);
1211 * Initialize the data block, then convert it to block format.
1213 error
= xfs_dir3_data_init(args
, blkno
, &bp
);
1218 xfs_dir3_block_init(mp
, tp
, bp
, dp
);
1222 * Compute size of block "tail" area.
1224 i
= (uint
)sizeof(*btp
) +
1225 (sfp
->count
+ 2) * (uint
)sizeof(xfs_dir2_leaf_entry_t
);
1227 * The whole thing is initialized to free by the init routine.
1228 * Say we're using the leaf and tail area.
1230 dup
= xfs_dir3_data_unused_p(hdr
);
1231 needlog
= needscan
= 0;
1232 xfs_dir2_data_use_free(tp
, bp
, dup
, mp
->m_dirblksize
- i
, i
, &needlog
,
1234 ASSERT(needscan
== 0);
1238 btp
= xfs_dir2_block_tail_p(mp
, hdr
);
1239 btp
->count
= cpu_to_be32(sfp
->count
+ 2); /* ., .. */
1241 blp
= xfs_dir2_block_leaf_p(btp
);
1242 endoffset
= (uint
)((char *)blp
- (char *)hdr
);
1244 * Remove the freespace, we'll manage it.
1246 xfs_dir2_data_use_free(tp
, bp
, dup
,
1247 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)hdr
),
1248 be16_to_cpu(dup
->length
), &needlog
, &needscan
);
1250 * Create entry for .
1252 dep
= xfs_dir3_data_dot_entry_p(hdr
);
1253 dep
->inumber
= cpu_to_be64(dp
->i_ino
);
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
[0].hashval
= cpu_to_be32(xfs_dir_hash_dot
);
1260 blp
[0].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
1261 (char *)dep
- (char *)hdr
));
1263 * Create entry for ..
1265 dep
= xfs_dir3_data_dotdot_entry_p(hdr
);
1266 dep
->inumber
= cpu_to_be64(xfs_dir2_sf_get_parent_ino(sfp
));
1268 dep
->name
[0] = dep
->name
[1] = '.';
1269 tagp
= xfs_dir2_data_entry_tag_p(dep
);
1270 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1271 xfs_dir2_data_log_entry(tp
, bp
, dep
);
1272 blp
[1].hashval
= cpu_to_be32(xfs_dir_hash_dotdot
);
1273 blp
[1].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
1274 (char *)dep
- (char *)hdr
));
1275 offset
= xfs_dir3_data_first_offset(hdr
);
1277 * Loop over existing entries, stuff them in.
1283 sfep
= xfs_dir2_sf_firstentry(sfp
);
1285 * Need to preserve the existing offset values in the sf directory.
1286 * Insert holes (unused entries) where necessary.
1288 while (offset
< endoffset
) {
1290 * sfep is null when we reach the end of the list.
1293 newoffset
= endoffset
;
1295 newoffset
= xfs_dir2_sf_get_offset(sfep
);
1297 * There should be a hole here, make one.
1299 if (offset
< newoffset
) {
1300 dup
= (xfs_dir2_data_unused_t
*)((char *)hdr
+ offset
);
1301 dup
->freetag
= cpu_to_be16(XFS_DIR2_DATA_FREE_TAG
);
1302 dup
->length
= cpu_to_be16(newoffset
- offset
);
1303 *xfs_dir2_data_unused_tag_p(dup
) = cpu_to_be16(
1304 ((char *)dup
- (char *)hdr
));
1305 xfs_dir2_data_log_unused(tp
, bp
, dup
);
1306 xfs_dir2_data_freeinsert(hdr
, dup
, &dummy
);
1307 offset
+= be16_to_cpu(dup
->length
);
1311 * Copy a real entry.
1313 dep
= (xfs_dir2_data_entry_t
*)((char *)hdr
+ newoffset
);
1314 dep
->inumber
= cpu_to_be64(xfs_dir2_sfe_get_ino(sfp
, sfep
));
1315 dep
->namelen
= sfep
->namelen
;
1316 memcpy(dep
->name
, sfep
->name
, dep
->namelen
);
1317 tagp
= xfs_dir2_data_entry_tag_p(dep
);
1318 *tagp
= cpu_to_be16((char *)dep
- (char *)hdr
);
1319 xfs_dir2_data_log_entry(tp
, bp
, dep
);
1320 name
.name
= sfep
->name
;
1321 name
.len
= sfep
->namelen
;
1322 blp
[2 + i
].hashval
= cpu_to_be32(mp
->m_dirnameops
->
1324 blp
[2 + i
].address
= cpu_to_be32(xfs_dir2_byte_to_dataptr(mp
,
1325 (char *)dep
- (char *)hdr
));
1326 offset
= (int)((char *)(tagp
+ 1) - (char *)hdr
);
1327 if (++i
== sfp
->count
)
1330 sfep
= xfs_dir2_sf_nextentry(sfp
, sfep
);
1332 /* Done with the temporary buffer */
1335 * Sort the leaf entries by hash value.
1337 xfs_sort(blp
, be32_to_cpu(btp
->count
), sizeof(*blp
), xfs_dir2_block_sort
);
1339 * Log the leaf entry area and tail.
1340 * Already logged the header in data_init, ignore needlog.
1342 ASSERT(needscan
== 0);
1343 xfs_dir2_block_log_leaf(tp
, bp
, 0, be32_to_cpu(btp
->count
) - 1);
1344 xfs_dir2_block_log_tail(tp
, bp
);
1345 xfs_dir3_data_check(dp
, bp
);