2 * Copyright (c) 2000-2003,2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_types.h"
23 #include "xfs_trans.h"
27 #include "xfs_dmapi.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_dir_sf.h"
32 #include "xfs_dir2_sf.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_dinode.h"
35 #include "xfs_inode.h"
36 #include "xfs_inode_item.h"
37 #include "xfs_dir_leaf.h"
38 #include "xfs_dir2_data.h"
39 #include "xfs_dir2_leaf.h"
40 #include "xfs_dir2_block.h"
41 #include "xfs_dir2_trace.h"
42 #include "xfs_error.h"
45 * Local function prototypes.
47 static void xfs_dir2_block_log_leaf(xfs_trans_t
*tp
, xfs_dabuf_t
*bp
, int first
,
49 static void xfs_dir2_block_log_tail(xfs_trans_t
*tp
, xfs_dabuf_t
*bp
);
50 static int xfs_dir2_block_lookup_int(xfs_da_args_t
*args
, xfs_dabuf_t
**bpp
,
52 static int xfs_dir2_block_sort(const void *a
, const void *b
);
55 * Add an entry to a block directory.
58 xfs_dir2_block_addname(
59 xfs_da_args_t
*args
) /* directory op arguments */
61 xfs_dir2_data_free_t
*bf
; /* bestfree table in block */
62 xfs_dir2_block_t
*block
; /* directory block structure */
63 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
64 xfs_dabuf_t
*bp
; /* buffer for block */
65 xfs_dir2_block_tail_t
*btp
; /* block tail */
66 int compact
; /* need to compact leaf ents */
67 xfs_dir2_data_entry_t
*dep
; /* block data entry */
68 xfs_inode_t
*dp
; /* directory inode */
69 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
70 int error
; /* error return value */
71 xfs_dir2_data_unused_t
*enddup
=NULL
; /* unused at end of data */
72 xfs_dahash_t hash
; /* hash value of found entry */
73 int high
; /* high index for binary srch */
74 int highstale
; /* high stale index */
75 int lfloghigh
=0; /* last final leaf to log */
76 int lfloglow
=0; /* first final leaf to log */
77 int len
; /* length of the new entry */
78 int low
; /* low index for binary srch */
79 int lowstale
; /* low stale index */
80 int mid
=0; /* midpoint for binary srch */
81 xfs_mount_t
*mp
; /* filesystem mount point */
82 int needlog
; /* need to log header */
83 int needscan
; /* need to rescan freespace */
84 xfs_dir2_data_off_t
*tagp
; /* pointer to tag value */
85 xfs_trans_t
*tp
; /* transaction structure */
87 xfs_dir2_trace_args("block_addname", args
);
92 * Read the (one and only) directory block into dabuf bp.
95 xfs_da_read_buf(tp
, dp
, mp
->m_dirdatablk
, -1, &bp
, XFS_DATA_FORK
))) {
101 * Check the magic number, corrupted if wrong.
103 if (unlikely(INT_GET(block
->hdr
.magic
, ARCH_CONVERT
)
104 != XFS_DIR2_BLOCK_MAGIC
)) {
105 XFS_CORRUPTION_ERROR("xfs_dir2_block_addname",
106 XFS_ERRLEVEL_LOW
, mp
, block
);
107 xfs_da_brelse(tp
, bp
);
108 return XFS_ERROR(EFSCORRUPTED
);
110 len
= XFS_DIR2_DATA_ENTSIZE(args
->namelen
);
112 * Set up pointers to parts of the block.
114 bf
= block
->hdr
.bestfree
;
115 btp
= XFS_DIR2_BLOCK_TAIL_P(mp
, block
);
116 blp
= XFS_DIR2_BLOCK_LEAF_P(btp
);
118 * No stale entries? Need space for entry and new leaf.
122 * Tag just before the first leaf entry.
124 tagp
= (xfs_dir2_data_off_t
*)blp
- 1;
126 * Data object just before the first leaf entry.
128 enddup
= (xfs_dir2_data_unused_t
*)((char *)block
+ INT_GET(*tagp
, ARCH_CONVERT
));
130 * If it's not free then can't do this add without cleaning up:
131 * the space before the first leaf entry needs to be free so it
132 * can be expanded to hold the pointer to the new entry.
134 if (INT_GET(enddup
->freetag
, ARCH_CONVERT
) != XFS_DIR2_DATA_FREE_TAG
)
137 * Check out the biggest freespace and see if it's the same one.
140 dup
= (xfs_dir2_data_unused_t
*)
141 ((char *)block
+ INT_GET(bf
[0].offset
, ARCH_CONVERT
));
144 * It is the biggest freespace, is it too small
145 * to hold the new leaf too?
147 if (INT_GET(dup
->length
, ARCH_CONVERT
) < len
+ (uint
)sizeof(*blp
)) {
149 * Yes, we use the second-largest
150 * entry instead if it works.
152 if (INT_GET(bf
[1].length
, ARCH_CONVERT
) >= len
)
153 dup
= (xfs_dir2_data_unused_t
*)
155 INT_GET(bf
[1].offset
, ARCH_CONVERT
));
161 * Not the same free entry,
162 * just check its length.
164 if (INT_GET(dup
->length
, ARCH_CONVERT
) < len
) {
172 * If there are stale entries we'll use one for the leaf.
173 * Is the biggest entry enough to avoid compaction?
175 else if (INT_GET(bf
[0].length
, ARCH_CONVERT
) >= len
) {
176 dup
= (xfs_dir2_data_unused_t
*)
177 ((char *)block
+ INT_GET(bf
[0].offset
, ARCH_CONVERT
));
181 * Will need to compact to make this work.
185 * Tag just before the first leaf entry.
187 tagp
= (xfs_dir2_data_off_t
*)blp
- 1;
189 * Data object just before the first leaf entry.
191 dup
= (xfs_dir2_data_unused_t
*)((char *)block
+ INT_GET(*tagp
, ARCH_CONVERT
));
193 * If it's not free then the data will go where the
194 * leaf data starts now, if it works at all.
196 if (INT_GET(dup
->freetag
, ARCH_CONVERT
) == XFS_DIR2_DATA_FREE_TAG
) {
197 if (INT_GET(dup
->length
, ARCH_CONVERT
) + (INT_GET(btp
->stale
, ARCH_CONVERT
) - 1) *
198 (uint
)sizeof(*blp
) < len
)
200 } else if ((INT_GET(btp
->stale
, ARCH_CONVERT
) - 1) * (uint
)sizeof(*blp
) < len
)
203 dup
= (xfs_dir2_data_unused_t
*)blp
;
207 * If this isn't a real add, we're done with the buffer.
210 xfs_da_brelse(tp
, bp
);
212 * If we don't have space for the new entry & leaf ...
216 * Not trying to actually do anything, or don't have
217 * a space reservation: return no-space.
219 if (args
->justcheck
|| args
->total
== 0)
220 return XFS_ERROR(ENOSPC
);
222 * Convert to the next larger format.
223 * Then add the new entry in that format.
225 error
= xfs_dir2_block_to_leaf(args
, bp
);
229 return xfs_dir2_leaf_addname(args
);
232 * Just checking, and it would work, so say so.
236 needlog
= needscan
= 0;
238 * If need to compact the leaf entries, do it now.
239 * Leave the highest-numbered stale entry stale.
240 * XXX should be the one closest to mid but mid is not yet computed.
243 int fromidx
; /* source leaf index */
244 int toidx
; /* target leaf index */
246 for (fromidx
= toidx
= INT_GET(btp
->count
, ARCH_CONVERT
) - 1,
247 highstale
= lfloghigh
= -1;
250 if (INT_GET(blp
[fromidx
].address
, ARCH_CONVERT
) == XFS_DIR2_NULL_DATAPTR
) {
260 blp
[toidx
] = blp
[fromidx
];
263 lfloglow
= toidx
+ 1 - (INT_GET(btp
->stale
, ARCH_CONVERT
) - 1);
264 lfloghigh
-= INT_GET(btp
->stale
, ARCH_CONVERT
) - 1;
265 INT_MOD(btp
->count
, ARCH_CONVERT
, -(INT_GET(btp
->stale
, ARCH_CONVERT
) - 1));
266 xfs_dir2_data_make_free(tp
, bp
,
267 (xfs_dir2_data_aoff_t
)((char *)blp
- (char *)block
),
268 (xfs_dir2_data_aoff_t
)((INT_GET(btp
->stale
, ARCH_CONVERT
) - 1) * sizeof(*blp
)),
269 &needlog
, &needscan
);
270 blp
+= INT_GET(btp
->stale
, ARCH_CONVERT
) - 1;
271 INT_SET(btp
->stale
, ARCH_CONVERT
, 1);
273 * If we now need to rebuild the bestfree map, do so.
274 * This needs to happen before the next call to use_free.
277 xfs_dir2_data_freescan(mp
, (xfs_dir2_data_t
*)block
,
283 * Set leaf logging boundaries to impossible state.
284 * For the no-stale case they're set explicitly.
286 else if (INT_GET(btp
->stale
, ARCH_CONVERT
)) {
287 lfloglow
= INT_GET(btp
->count
, ARCH_CONVERT
);
291 * Find the slot that's first lower than our hash value, -1 if none.
293 for (low
= 0, high
= INT_GET(btp
->count
, ARCH_CONVERT
) - 1; low
<= high
; ) {
294 mid
= (low
+ high
) >> 1;
295 if ((hash
= INT_GET(blp
[mid
].hashval
, ARCH_CONVERT
)) == args
->hashval
)
297 if (hash
< args
->hashval
)
302 while (mid
>= 0 && INT_GET(blp
[mid
].hashval
, ARCH_CONVERT
) >= args
->hashval
) {
306 * No stale entries, will use enddup space to hold new leaf.
310 * Mark the space needed for the new leaf entry, now in use.
312 xfs_dir2_data_use_free(tp
, bp
, enddup
,
313 (xfs_dir2_data_aoff_t
)
314 ((char *)enddup
- (char *)block
+ INT_GET(enddup
->length
, ARCH_CONVERT
) -
316 (xfs_dir2_data_aoff_t
)sizeof(*blp
),
317 &needlog
, &needscan
);
319 * Update the tail (entry count).
321 INT_MOD(btp
->count
, ARCH_CONVERT
, +1);
323 * If we now need to rebuild the bestfree map, do so.
324 * This needs to happen before the next call to use_free.
327 xfs_dir2_data_freescan(mp
, (xfs_dir2_data_t
*)block
,
332 * Adjust pointer to the first leaf entry, we're about to move
333 * the table up one to open up space for the new leaf entry.
334 * Then adjust our index to match.
339 memmove(blp
, &blp
[1], mid
* sizeof(*blp
));
344 * Use a stale leaf for our new entry.
349 INT_GET(blp
[lowstale
].address
, ARCH_CONVERT
) != XFS_DIR2_NULL_DATAPTR
;
352 for (highstale
= mid
+ 1;
353 highstale
< INT_GET(btp
->count
, ARCH_CONVERT
) &&
354 INT_GET(blp
[highstale
].address
, ARCH_CONVERT
) != XFS_DIR2_NULL_DATAPTR
&&
355 (lowstale
< 0 || mid
- lowstale
> highstale
- mid
);
359 * Move entries toward the low-numbered stale entry.
362 (highstale
== INT_GET(btp
->count
, ARCH_CONVERT
) ||
363 mid
- lowstale
<= highstale
- mid
)) {
365 memmove(&blp
[lowstale
], &blp
[lowstale
+ 1],
366 (mid
- lowstale
) * sizeof(*blp
));
367 lfloglow
= MIN(lowstale
, lfloglow
);
368 lfloghigh
= MAX(mid
, lfloghigh
);
371 * Move entries toward the high-numbered stale entry.
374 ASSERT(highstale
< INT_GET(btp
->count
, ARCH_CONVERT
));
377 memmove(&blp
[mid
+ 1], &blp
[mid
],
378 (highstale
- mid
) * sizeof(*blp
));
379 lfloglow
= MIN(mid
, lfloglow
);
380 lfloghigh
= MAX(highstale
, lfloghigh
);
382 INT_MOD(btp
->stale
, ARCH_CONVERT
, -1);
385 * Point to the new data entry.
387 dep
= (xfs_dir2_data_entry_t
*)dup
;
389 * Fill in the leaf entry.
391 INT_SET(blp
[mid
].hashval
, ARCH_CONVERT
, args
->hashval
);
392 INT_SET(blp
[mid
].address
, ARCH_CONVERT
, XFS_DIR2_BYTE_TO_DATAPTR(mp
, (char *)dep
- (char *)block
));
393 xfs_dir2_block_log_leaf(tp
, bp
, lfloglow
, lfloghigh
);
395 * Mark space for the data entry used.
397 xfs_dir2_data_use_free(tp
, bp
, dup
,
398 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)block
),
399 (xfs_dir2_data_aoff_t
)len
, &needlog
, &needscan
);
401 * Create the new data entry.
403 INT_SET(dep
->inumber
, ARCH_CONVERT
, args
->inumber
);
404 dep
->namelen
= args
->namelen
;
405 memcpy(dep
->name
, args
->name
, args
->namelen
);
406 tagp
= XFS_DIR2_DATA_ENTRY_TAG_P(dep
);
407 INT_SET(*tagp
, ARCH_CONVERT
, (xfs_dir2_data_off_t
)((char *)dep
- (char *)block
));
409 * Clean up the bestfree array and log the header, tail, and entry.
412 xfs_dir2_data_freescan(mp
, (xfs_dir2_data_t
*)block
, &needlog
,
415 xfs_dir2_data_log_header(tp
, bp
);
416 xfs_dir2_block_log_tail(tp
, bp
);
417 xfs_dir2_data_log_entry(tp
, bp
, dep
);
418 xfs_dir2_data_check(dp
, bp
);
424 * Readdir for block directories.
427 xfs_dir2_block_getdents(
428 xfs_trans_t
*tp
, /* transaction (NULL) */
429 xfs_inode_t
*dp
, /* incore inode */
430 uio_t
*uio
, /* caller's buffer control */
431 int *eofp
, /* eof reached? (out) */
432 xfs_dirent_t
*dbp
, /* caller's buffer */
433 xfs_dir2_put_t put
) /* abi's formatting function */
435 xfs_dir2_block_t
*block
; /* directory block structure */
436 xfs_dabuf_t
*bp
; /* buffer for block */
437 xfs_dir2_block_tail_t
*btp
; /* block tail */
438 xfs_dir2_data_entry_t
*dep
; /* block data entry */
439 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
440 char *endptr
; /* end of the data entries */
441 int error
; /* error return value */
442 xfs_mount_t
*mp
; /* filesystem mount point */
443 xfs_dir2_put_args_t p
; /* arg package for put rtn */
444 char *ptr
; /* current data entry */
445 int wantoff
; /* starting block offset */
449 * If the block number in the offset is out of range, we're done.
451 if (XFS_DIR2_DATAPTR_TO_DB(mp
, uio
->uio_offset
) > mp
->m_dirdatablk
) {
456 * Can't read the block, give up, else get dabuf in bp.
459 xfs_da_read_buf(tp
, dp
, mp
->m_dirdatablk
, -1, &bp
, XFS_DATA_FORK
))) {
464 * Extract the byte offset we start at from the seek pointer.
465 * We'll skip entries before this.
467 wantoff
= XFS_DIR2_DATAPTR_TO_OFF(mp
, uio
->uio_offset
);
469 xfs_dir2_data_check(dp
, bp
);
471 * Set up values for the loop.
473 btp
= XFS_DIR2_BLOCK_TAIL_P(mp
, block
);
474 ptr
= (char *)block
->u
;
475 endptr
= (char *)XFS_DIR2_BLOCK_LEAF_P(btp
);
480 * Loop over the data portion of the block.
481 * Each object is a real entry (dep) or an unused one (dup).
483 while (ptr
< endptr
) {
484 dup
= (xfs_dir2_data_unused_t
*)ptr
;
488 if (INT_GET(dup
->freetag
, ARCH_CONVERT
) == XFS_DIR2_DATA_FREE_TAG
) {
489 ptr
+= INT_GET(dup
->length
, ARCH_CONVERT
);
493 dep
= (xfs_dir2_data_entry_t
*)ptr
;
496 * Bump pointer for the next iteration.
498 ptr
+= XFS_DIR2_DATA_ENTSIZE(dep
->namelen
);
500 * The entry is before the desired starting point, skip it.
502 if ((char *)dep
- (char *)block
< wantoff
)
505 * Set up argument structure for put routine.
507 p
.namelen
= dep
->namelen
;
509 p
.cook
= XFS_DIR2_DB_OFF_TO_DATAPTR(mp
, mp
->m_dirdatablk
,
510 ptr
- (char *)block
);
511 p
.ino
= INT_GET(dep
->inumber
, ARCH_CONVERT
);
513 p
.ino
+= mp
->m_inoadd
;
515 p
.name
= (char *)dep
->name
;
518 * Put the entry in the caller's buffer.
523 * If it didn't fit, set the final offset to here & return.
527 XFS_DIR2_DB_OFF_TO_DATAPTR(mp
, mp
->m_dirdatablk
,
528 (char *)dep
- (char *)block
);
529 xfs_da_brelse(tp
, bp
);
535 * Reached the end of the block.
536 * Set the offset to a nonexistent block 1 and return.
541 XFS_DIR2_DB_OFF_TO_DATAPTR(mp
, mp
->m_dirdatablk
+ 1, 0);
543 xfs_da_brelse(tp
, bp
);
549 * Log leaf entries from the block.
552 xfs_dir2_block_log_leaf(
553 xfs_trans_t
*tp
, /* transaction structure */
554 xfs_dabuf_t
*bp
, /* block buffer */
555 int first
, /* index of first logged leaf */
556 int last
) /* index of last logged leaf */
558 xfs_dir2_block_t
*block
; /* directory block structure */
559 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
560 xfs_dir2_block_tail_t
*btp
; /* block tail */
561 xfs_mount_t
*mp
; /* filesystem mount point */
565 btp
= XFS_DIR2_BLOCK_TAIL_P(mp
, block
);
566 blp
= XFS_DIR2_BLOCK_LEAF_P(btp
);
567 xfs_da_log_buf(tp
, bp
, (uint
)((char *)&blp
[first
] - (char *)block
),
568 (uint
)((char *)&blp
[last
+ 1] - (char *)block
- 1));
572 * Log the block tail.
575 xfs_dir2_block_log_tail(
576 xfs_trans_t
*tp
, /* transaction structure */
577 xfs_dabuf_t
*bp
) /* block buffer */
579 xfs_dir2_block_t
*block
; /* directory block structure */
580 xfs_dir2_block_tail_t
*btp
; /* block tail */
581 xfs_mount_t
*mp
; /* filesystem mount point */
585 btp
= XFS_DIR2_BLOCK_TAIL_P(mp
, block
);
586 xfs_da_log_buf(tp
, bp
, (uint
)((char *)btp
- (char *)block
),
587 (uint
)((char *)(btp
+ 1) - (char *)block
- 1));
591 * Look up an entry in the block. This is the external routine,
592 * xfs_dir2_block_lookup_int does the real work.
595 xfs_dir2_block_lookup(
596 xfs_da_args_t
*args
) /* dir lookup arguments */
598 xfs_dir2_block_t
*block
; /* block structure */
599 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
600 xfs_dabuf_t
*bp
; /* block buffer */
601 xfs_dir2_block_tail_t
*btp
; /* block tail */
602 xfs_dir2_data_entry_t
*dep
; /* block data entry */
603 xfs_inode_t
*dp
; /* incore inode */
604 int ent
; /* entry index */
605 int error
; /* error return value */
606 xfs_mount_t
*mp
; /* filesystem mount point */
608 xfs_dir2_trace_args("block_lookup", args
);
610 * Get the buffer, look up the entry.
611 * If not found (ENOENT) then return, have no buffer.
613 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
)))
618 xfs_dir2_data_check(dp
, bp
);
619 btp
= XFS_DIR2_BLOCK_TAIL_P(mp
, block
);
620 blp
= XFS_DIR2_BLOCK_LEAF_P(btp
);
622 * Get the offset from the leaf entry, to point to the data.
624 dep
= (xfs_dir2_data_entry_t
*)
625 ((char *)block
+ XFS_DIR2_DATAPTR_TO_OFF(mp
, INT_GET(blp
[ent
].address
, ARCH_CONVERT
)));
627 * Fill in inode number, release the block.
629 args
->inumber
= INT_GET(dep
->inumber
, ARCH_CONVERT
);
630 xfs_da_brelse(args
->trans
, bp
);
631 return XFS_ERROR(EEXIST
);
635 * Internal block lookup routine.
637 static int /* error */
638 xfs_dir2_block_lookup_int(
639 xfs_da_args_t
*args
, /* dir lookup arguments */
640 xfs_dabuf_t
**bpp
, /* returned block buffer */
641 int *entno
) /* returned entry number */
643 xfs_dir2_dataptr_t addr
; /* data entry address */
644 xfs_dir2_block_t
*block
; /* block structure */
645 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
646 xfs_dabuf_t
*bp
; /* block buffer */
647 xfs_dir2_block_tail_t
*btp
; /* block tail */
648 xfs_dir2_data_entry_t
*dep
; /* block data entry */
649 xfs_inode_t
*dp
; /* incore inode */
650 int error
; /* error return value */
651 xfs_dahash_t hash
; /* found hash value */
652 int high
; /* binary search high index */
653 int low
; /* binary search low index */
654 int mid
; /* binary search current idx */
655 xfs_mount_t
*mp
; /* filesystem mount point */
656 xfs_trans_t
*tp
; /* transaction pointer */
662 * Read the buffer, return error if we can't get it.
665 xfs_da_read_buf(tp
, dp
, mp
->m_dirdatablk
, -1, &bp
, XFS_DATA_FORK
))) {
670 xfs_dir2_data_check(dp
, bp
);
671 btp
= XFS_DIR2_BLOCK_TAIL_P(mp
, block
);
672 blp
= XFS_DIR2_BLOCK_LEAF_P(btp
);
674 * Loop doing a binary search for our hash value.
675 * Find our entry, ENOENT if it's not there.
677 for (low
= 0, high
= INT_GET(btp
->count
, ARCH_CONVERT
) - 1; ; ) {
679 mid
= (low
+ high
) >> 1;
680 if ((hash
= INT_GET(blp
[mid
].hashval
, ARCH_CONVERT
)) == args
->hashval
)
682 if (hash
< args
->hashval
)
687 ASSERT(args
->oknoent
);
688 xfs_da_brelse(tp
, bp
);
689 return XFS_ERROR(ENOENT
);
693 * Back up to the first one with the right hash value.
695 while (mid
> 0 && INT_GET(blp
[mid
- 1].hashval
, ARCH_CONVERT
) == args
->hashval
) {
699 * Now loop forward through all the entries with the
700 * right hash value looking for our name.
703 if ((addr
= INT_GET(blp
[mid
].address
, ARCH_CONVERT
)) == XFS_DIR2_NULL_DATAPTR
)
706 * Get pointer to the entry from the leaf.
708 dep
= (xfs_dir2_data_entry_t
*)
709 ((char *)block
+ XFS_DIR2_DATAPTR_TO_OFF(mp
, addr
));
711 * Compare, if it's right give back buffer & entry number.
713 if (dep
->namelen
== args
->namelen
&&
714 dep
->name
[0] == args
->name
[0] &&
715 memcmp(dep
->name
, args
->name
, args
->namelen
) == 0) {
720 } while (++mid
< INT_GET(btp
->count
, ARCH_CONVERT
) && INT_GET(blp
[mid
].hashval
, ARCH_CONVERT
) == hash
);
722 * No match, release the buffer and return ENOENT.
724 ASSERT(args
->oknoent
);
725 xfs_da_brelse(tp
, bp
);
726 return XFS_ERROR(ENOENT
);
730 * Remove an entry from a block format directory.
731 * If that makes the block small enough to fit in shortform, transform it.
734 xfs_dir2_block_removename(
735 xfs_da_args_t
*args
) /* directory operation args */
737 xfs_dir2_block_t
*block
; /* block structure */
738 xfs_dir2_leaf_entry_t
*blp
; /* block leaf pointer */
739 xfs_dabuf_t
*bp
; /* block buffer */
740 xfs_dir2_block_tail_t
*btp
; /* block tail */
741 xfs_dir2_data_entry_t
*dep
; /* block data entry */
742 xfs_inode_t
*dp
; /* incore inode */
743 int ent
; /* block leaf entry index */
744 int error
; /* error return value */
745 xfs_mount_t
*mp
; /* filesystem mount point */
746 int needlog
; /* need to log block header */
747 int needscan
; /* need to fixup bestfree */
748 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
749 int size
; /* shortform size */
750 xfs_trans_t
*tp
; /* transaction pointer */
752 xfs_dir2_trace_args("block_removename", args
);
754 * Look up the entry in the block. Gets the buffer and entry index.
755 * It will always be there, the vnodeops level does a lookup first.
757 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
764 btp
= XFS_DIR2_BLOCK_TAIL_P(mp
, block
);
765 blp
= XFS_DIR2_BLOCK_LEAF_P(btp
);
767 * Point to the data entry using the leaf entry.
769 dep
= (xfs_dir2_data_entry_t
*)
770 ((char *)block
+ XFS_DIR2_DATAPTR_TO_OFF(mp
, INT_GET(blp
[ent
].address
, ARCH_CONVERT
)));
772 * Mark the data entry's space free.
774 needlog
= needscan
= 0;
775 xfs_dir2_data_make_free(tp
, bp
,
776 (xfs_dir2_data_aoff_t
)((char *)dep
- (char *)block
),
777 XFS_DIR2_DATA_ENTSIZE(dep
->namelen
), &needlog
, &needscan
);
779 * Fix up the block tail.
781 INT_MOD(btp
->stale
, ARCH_CONVERT
, +1);
782 xfs_dir2_block_log_tail(tp
, bp
);
784 * Remove the leaf entry by marking it stale.
786 INT_SET(blp
[ent
].address
, ARCH_CONVERT
, XFS_DIR2_NULL_DATAPTR
);
787 xfs_dir2_block_log_leaf(tp
, bp
, ent
, ent
);
789 * Fix up bestfree, log the header if necessary.
792 xfs_dir2_data_freescan(mp
, (xfs_dir2_data_t
*)block
, &needlog
,
795 xfs_dir2_data_log_header(tp
, bp
);
796 xfs_dir2_data_check(dp
, bp
);
798 * See if the size as a shortform is good enough.
800 if ((size
= xfs_dir2_block_sfsize(dp
, block
, &sfh
)) >
801 XFS_IFORK_DSIZE(dp
)) {
806 * If it works, do the conversion.
808 return xfs_dir2_block_to_sf(args
, bp
, size
, &sfh
);
812 * Replace an entry in a V2 block directory.
813 * Change the inode number to the new value.
816 xfs_dir2_block_replace(
817 xfs_da_args_t
*args
) /* directory operation args */
819 xfs_dir2_block_t
*block
; /* block structure */
820 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
821 xfs_dabuf_t
*bp
; /* block buffer */
822 xfs_dir2_block_tail_t
*btp
; /* block tail */
823 xfs_dir2_data_entry_t
*dep
; /* block data entry */
824 xfs_inode_t
*dp
; /* incore inode */
825 int ent
; /* leaf entry index */
826 int error
; /* error return value */
827 xfs_mount_t
*mp
; /* filesystem mount point */
829 xfs_dir2_trace_args("block_replace", args
);
831 * Lookup the entry in the directory. Get buffer and entry index.
832 * This will always succeed since the caller has already done a lookup.
834 if ((error
= xfs_dir2_block_lookup_int(args
, &bp
, &ent
))) {
840 btp
= XFS_DIR2_BLOCK_TAIL_P(mp
, block
);
841 blp
= XFS_DIR2_BLOCK_LEAF_P(btp
);
843 * Point to the data entry we need to change.
845 dep
= (xfs_dir2_data_entry_t
*)
846 ((char *)block
+ XFS_DIR2_DATAPTR_TO_OFF(mp
, INT_GET(blp
[ent
].address
, ARCH_CONVERT
)));
847 ASSERT(INT_GET(dep
->inumber
, ARCH_CONVERT
) != args
->inumber
);
849 * Change the inode number to the new value.
851 INT_SET(dep
->inumber
, ARCH_CONVERT
, args
->inumber
);
852 xfs_dir2_data_log_entry(args
->trans
, bp
, dep
);
853 xfs_dir2_data_check(dp
, bp
);
859 * Qsort comparison routine for the block leaf entries.
861 static int /* sort order */
863 const void *a
, /* first leaf entry */
864 const void *b
) /* second leaf entry */
866 const xfs_dir2_leaf_entry_t
*la
; /* first leaf entry */
867 const xfs_dir2_leaf_entry_t
*lb
; /* second leaf entry */
871 return INT_GET(la
->hashval
, ARCH_CONVERT
) < INT_GET(lb
->hashval
, ARCH_CONVERT
) ? -1 :
872 (INT_GET(la
->hashval
, ARCH_CONVERT
) > INT_GET(lb
->hashval
, ARCH_CONVERT
) ? 1 : 0);
876 * Convert a V2 leaf directory to a V2 block directory if possible.
879 xfs_dir2_leaf_to_block(
880 xfs_da_args_t
*args
, /* operation arguments */
881 xfs_dabuf_t
*lbp
, /* leaf buffer */
882 xfs_dabuf_t
*dbp
) /* data buffer */
884 xfs_dir2_data_off_t
*bestsp
; /* leaf bests table */
885 xfs_dir2_block_t
*block
; /* block structure */
886 xfs_dir2_block_tail_t
*btp
; /* block tail */
887 xfs_inode_t
*dp
; /* incore directory inode */
888 xfs_dir2_data_unused_t
*dup
; /* unused data entry */
889 int error
; /* error return value */
890 int from
; /* leaf from index */
891 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
892 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
893 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
894 xfs_mount_t
*mp
; /* file system mount point */
895 int needlog
; /* need to log data header */
896 int needscan
; /* need to scan for bestfree */
897 xfs_dir2_sf_hdr_t sfh
; /* shortform header */
898 int size
; /* bytes used */
899 xfs_dir2_data_off_t
*tagp
; /* end of entry (tag) */
900 int to
; /* block/leaf to index */
901 xfs_trans_t
*tp
; /* transaction pointer */
903 xfs_dir2_trace_args_bb("leaf_to_block", args
, lbp
, dbp
);
908 ASSERT(INT_GET(leaf
->hdr
.info
.magic
, ARCH_CONVERT
) == XFS_DIR2_LEAF1_MAGIC
);
909 ltp
= XFS_DIR2_LEAF_TAIL_P(mp
, leaf
);
911 * If there are data blocks other than the first one, take this
912 * opportunity to remove trailing empty data blocks that may have
913 * been left behind during no-space-reservation operations.
914 * These will show up in the leaf bests table.
916 while (dp
->i_d
.di_size
> mp
->m_dirblksize
) {
917 bestsp
= XFS_DIR2_LEAF_BESTS_P(ltp
);
918 if (INT_GET(bestsp
[INT_GET(ltp
->bestcount
, ARCH_CONVERT
) - 1], ARCH_CONVERT
) ==
919 mp
->m_dirblksize
- (uint
)sizeof(block
->hdr
)) {
921 xfs_dir2_leaf_trim_data(args
, lbp
,
922 (xfs_dir2_db_t
)(INT_GET(ltp
->bestcount
, ARCH_CONVERT
) - 1))))
930 * Read the data block if we don't already have it, give up if it fails.
933 (error
= xfs_da_read_buf(tp
, dp
, mp
->m_dirdatablk
, -1, &dbp
,
938 ASSERT(INT_GET(block
->hdr
.magic
, ARCH_CONVERT
) == XFS_DIR2_DATA_MAGIC
);
940 * Size of the "leaf" area in the block.
942 size
= (uint
)sizeof(block
->tail
) +
943 (uint
)sizeof(*lep
) * (INT_GET(leaf
->hdr
.count
, ARCH_CONVERT
) - INT_GET(leaf
->hdr
.stale
, ARCH_CONVERT
));
945 * Look at the last data entry.
947 tagp
= (xfs_dir2_data_off_t
*)((char *)block
+ mp
->m_dirblksize
) - 1;
948 dup
= (xfs_dir2_data_unused_t
*)((char *)block
+ INT_GET(*tagp
, ARCH_CONVERT
));
950 * If it's not free or is too short we can't do it.
952 if (INT_GET(dup
->freetag
, ARCH_CONVERT
) != XFS_DIR2_DATA_FREE_TAG
|| INT_GET(dup
->length
, ARCH_CONVERT
) < size
) {
957 * Start converting it to block form.
959 INT_SET(block
->hdr
.magic
, ARCH_CONVERT
, XFS_DIR2_BLOCK_MAGIC
);
963 * Use up the space at the end of the block (blp/btp).
965 xfs_dir2_data_use_free(tp
, dbp
, dup
, mp
->m_dirblksize
- size
, size
,
966 &needlog
, &needscan
);
968 * Initialize the block tail.
970 btp
= XFS_DIR2_BLOCK_TAIL_P(mp
, block
);
971 INT_SET(btp
->count
, ARCH_CONVERT
, INT_GET(leaf
->hdr
.count
, ARCH_CONVERT
) - INT_GET(leaf
->hdr
.stale
, ARCH_CONVERT
));
973 xfs_dir2_block_log_tail(tp
, dbp
);
975 * Initialize the block leaf area. We compact out stale entries.
977 lep
= XFS_DIR2_BLOCK_LEAF_P(btp
);
978 for (from
= to
= 0; from
< INT_GET(leaf
->hdr
.count
, ARCH_CONVERT
); from
++) {
979 if (INT_GET(leaf
->ents
[from
].address
, ARCH_CONVERT
) == XFS_DIR2_NULL_DATAPTR
)
981 lep
[to
++] = leaf
->ents
[from
];
983 ASSERT(to
== INT_GET(btp
->count
, ARCH_CONVERT
));
984 xfs_dir2_block_log_leaf(tp
, dbp
, 0, INT_GET(btp
->count
, ARCH_CONVERT
) - 1);
986 * Scan the bestfree if we need it and log the data block header.
989 xfs_dir2_data_freescan(mp
, (xfs_dir2_data_t
*)block
, &needlog
,
992 xfs_dir2_data_log_header(tp
, dbp
);
994 * Pitch the old leaf block.
996 error
= xfs_da_shrink_inode(args
, mp
->m_dirleafblk
, lbp
);
1002 * Now see if the resulting block can be shrunken to shortform.
1004 if ((size
= xfs_dir2_block_sfsize(dp
, block
, &sfh
)) >
1005 XFS_IFORK_DSIZE(dp
)) {
1009 return xfs_dir2_block_to_sf(args
, dbp
, size
, &sfh
);
1012 xfs_da_buf_done(lbp
);
1014 xfs_da_buf_done(dbp
);
1019 * Convert the shortform directory to block form.
1022 xfs_dir2_sf_to_block(
1023 xfs_da_args_t
*args
) /* operation arguments */
1025 xfs_dir2_db_t blkno
; /* dir-relative block # (0) */
1026 xfs_dir2_block_t
*block
; /* block structure */
1027 xfs_dir2_leaf_entry_t
*blp
; /* block leaf entries */
1028 xfs_dabuf_t
*bp
; /* block buffer */
1029 xfs_dir2_block_tail_t
*btp
; /* block tail pointer */
1030 char *buf
; /* sf buffer */
1032 xfs_dir2_data_entry_t
*dep
; /* data entry pointer */
1033 xfs_inode_t
*dp
; /* incore directory inode */
1034 int dummy
; /* trash */
1035 xfs_dir2_data_unused_t
*dup
; /* unused entry pointer */
1036 int endoffset
; /* end of data objects */
1037 int error
; /* error return value */
1039 xfs_mount_t
*mp
; /* filesystem mount point */
1040 int needlog
; /* need to log block header */
1041 int needscan
; /* need to scan block freespc */
1042 int newoffset
; /* offset from current entry */
1043 int offset
; /* target block offset */
1044 xfs_dir2_sf_entry_t
*sfep
; /* sf entry pointer */
1045 xfs_dir2_sf_t
*sfp
; /* shortform structure */
1046 xfs_dir2_data_off_t
*tagp
; /* end of data entry */
1047 xfs_trans_t
*tp
; /* transaction pointer */
1049 xfs_dir2_trace_args("sf_to_block", args
);
1053 ASSERT(dp
->i_df
.if_flags
& XFS_IFINLINE
);
1055 * Bomb out if the shortform directory is way too short.
1057 if (dp
->i_d
.di_size
< offsetof(xfs_dir2_sf_hdr_t
, parent
)) {
1058 ASSERT(XFS_FORCED_SHUTDOWN(mp
));
1059 return XFS_ERROR(EIO
);
1061 ASSERT(dp
->i_df
.if_bytes
== dp
->i_d
.di_size
);
1062 ASSERT(dp
->i_df
.if_u1
.if_data
!= NULL
);
1063 sfp
= (xfs_dir2_sf_t
*)dp
->i_df
.if_u1
.if_data
;
1064 ASSERT(dp
->i_d
.di_size
>= XFS_DIR2_SF_HDR_SIZE(sfp
->hdr
.i8count
));
1066 * Copy the directory into the stack buffer.
1067 * Then pitch the incore inode data so we can make extents.
1070 buf_len
= dp
->i_df
.if_bytes
;
1071 buf
= kmem_alloc(dp
->i_df
.if_bytes
, KM_SLEEP
);
1073 memcpy(buf
, sfp
, dp
->i_df
.if_bytes
);
1074 xfs_idata_realloc(dp
, -dp
->i_df
.if_bytes
, XFS_DATA_FORK
);
1075 dp
->i_d
.di_size
= 0;
1076 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
1078 * Reset pointer - old sfp is gone.
1080 sfp
= (xfs_dir2_sf_t
*)buf
;
1082 * Add block 0 to the inode.
1084 error
= xfs_dir2_grow_inode(args
, XFS_DIR2_DATA_SPACE
, &blkno
);
1086 kmem_free(buf
, buf_len
);
1090 * Initialize the data block.
1092 error
= xfs_dir2_data_init(args
, blkno
, &bp
);
1094 kmem_free(buf
, buf_len
);
1098 INT_SET(block
->hdr
.magic
, ARCH_CONVERT
, XFS_DIR2_BLOCK_MAGIC
);
1100 * Compute size of block "tail" area.
1102 i
= (uint
)sizeof(*btp
) +
1103 (INT_GET(sfp
->hdr
.count
, ARCH_CONVERT
) + 2) * (uint
)sizeof(xfs_dir2_leaf_entry_t
);
1105 * The whole thing is initialized to free by the init routine.
1106 * Say we're using the leaf and tail area.
1108 dup
= (xfs_dir2_data_unused_t
*)block
->u
;
1109 needlog
= needscan
= 0;
1110 xfs_dir2_data_use_free(tp
, bp
, dup
, mp
->m_dirblksize
- i
, i
, &needlog
,
1112 ASSERT(needscan
== 0);
1116 btp
= XFS_DIR2_BLOCK_TAIL_P(mp
, block
);
1117 INT_SET(btp
->count
, ARCH_CONVERT
, INT_GET(sfp
->hdr
.count
, ARCH_CONVERT
) + 2); /* ., .. */
1119 blp
= XFS_DIR2_BLOCK_LEAF_P(btp
);
1120 endoffset
= (uint
)((char *)blp
- (char *)block
);
1122 * Remove the freespace, we'll manage it.
1124 xfs_dir2_data_use_free(tp
, bp
, dup
,
1125 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)block
),
1126 INT_GET(dup
->length
, ARCH_CONVERT
), &needlog
, &needscan
);
1128 * Create entry for .
1130 dep
= (xfs_dir2_data_entry_t
*)
1131 ((char *)block
+ XFS_DIR2_DATA_DOT_OFFSET
);
1132 INT_SET(dep
->inumber
, ARCH_CONVERT
, dp
->i_ino
);
1135 tagp
= XFS_DIR2_DATA_ENTRY_TAG_P(dep
);
1136 INT_SET(*tagp
, ARCH_CONVERT
, (xfs_dir2_data_off_t
)((char *)dep
- (char *)block
));
1137 xfs_dir2_data_log_entry(tp
, bp
, dep
);
1138 INT_SET(blp
[0].hashval
, ARCH_CONVERT
, xfs_dir_hash_dot
);
1139 INT_SET(blp
[0].address
, ARCH_CONVERT
, XFS_DIR2_BYTE_TO_DATAPTR(mp
, (char *)dep
- (char *)block
));
1141 * Create entry for ..
1143 dep
= (xfs_dir2_data_entry_t
*)
1144 ((char *)block
+ XFS_DIR2_DATA_DOTDOT_OFFSET
);
1145 INT_SET(dep
->inumber
, ARCH_CONVERT
, XFS_DIR2_SF_GET_INUMBER(sfp
, &sfp
->hdr
.parent
));
1147 dep
->name
[0] = dep
->name
[1] = '.';
1148 tagp
= XFS_DIR2_DATA_ENTRY_TAG_P(dep
);
1149 INT_SET(*tagp
, ARCH_CONVERT
, (xfs_dir2_data_off_t
)((char *)dep
- (char *)block
));
1150 xfs_dir2_data_log_entry(tp
, bp
, dep
);
1151 INT_SET(blp
[1].hashval
, ARCH_CONVERT
, xfs_dir_hash_dotdot
);
1152 INT_SET(blp
[1].address
, ARCH_CONVERT
, XFS_DIR2_BYTE_TO_DATAPTR(mp
, (char *)dep
- (char *)block
));
1153 offset
= XFS_DIR2_DATA_FIRST_OFFSET
;
1155 * Loop over existing entries, stuff them in.
1157 if ((i
= 0) == INT_GET(sfp
->hdr
.count
, ARCH_CONVERT
))
1160 sfep
= XFS_DIR2_SF_FIRSTENTRY(sfp
);
1162 * Need to preserve the existing offset values in the sf directory.
1163 * Insert holes (unused entries) where necessary.
1165 while (offset
< endoffset
) {
1167 * sfep is null when we reach the end of the list.
1170 newoffset
= endoffset
;
1172 newoffset
= XFS_DIR2_SF_GET_OFFSET(sfep
);
1174 * There should be a hole here, make one.
1176 if (offset
< newoffset
) {
1177 dup
= (xfs_dir2_data_unused_t
*)
1178 ((char *)block
+ offset
);
1179 INT_SET(dup
->freetag
, ARCH_CONVERT
, XFS_DIR2_DATA_FREE_TAG
);
1180 INT_SET(dup
->length
, ARCH_CONVERT
, newoffset
- offset
);
1181 INT_SET(*XFS_DIR2_DATA_UNUSED_TAG_P(dup
), ARCH_CONVERT
,
1182 (xfs_dir2_data_off_t
)
1183 ((char *)dup
- (char *)block
));
1184 xfs_dir2_data_log_unused(tp
, bp
, dup
);
1185 (void)xfs_dir2_data_freeinsert((xfs_dir2_data_t
*)block
,
1187 offset
+= INT_GET(dup
->length
, ARCH_CONVERT
);
1191 * Copy a real entry.
1193 dep
= (xfs_dir2_data_entry_t
*)((char *)block
+ newoffset
);
1194 INT_SET(dep
->inumber
, ARCH_CONVERT
, XFS_DIR2_SF_GET_INUMBER(sfp
,
1195 XFS_DIR2_SF_INUMBERP(sfep
)));
1196 dep
->namelen
= sfep
->namelen
;
1197 memcpy(dep
->name
, sfep
->name
, dep
->namelen
);
1198 tagp
= XFS_DIR2_DATA_ENTRY_TAG_P(dep
);
1199 INT_SET(*tagp
, ARCH_CONVERT
, (xfs_dir2_data_off_t
)((char *)dep
- (char *)block
));
1200 xfs_dir2_data_log_entry(tp
, bp
, dep
);
1201 INT_SET(blp
[2 + i
].hashval
, ARCH_CONVERT
, xfs_da_hashname((char *)sfep
->name
, sfep
->namelen
));
1202 INT_SET(blp
[2 + i
].address
, ARCH_CONVERT
, XFS_DIR2_BYTE_TO_DATAPTR(mp
,
1203 (char *)dep
- (char *)block
));
1204 offset
= (int)((char *)(tagp
+ 1) - (char *)block
);
1205 if (++i
== INT_GET(sfp
->hdr
.count
, ARCH_CONVERT
))
1208 sfep
= XFS_DIR2_SF_NEXTENTRY(sfp
, sfep
);
1210 /* Done with the temporary buffer */
1211 kmem_free(buf
, buf_len
);
1213 * Sort the leaf entries by hash value.
1215 xfs_sort(blp
, INT_GET(btp
->count
, ARCH_CONVERT
), sizeof(*blp
), xfs_dir2_block_sort
);
1217 * Log the leaf entry area and tail.
1218 * Already logged the header in data_init, ignore needlog.
1220 ASSERT(needscan
== 0);
1221 xfs_dir2_block_log_leaf(tp
, bp
, 0, INT_GET(btp
->count
, ARCH_CONVERT
) - 1);
1222 xfs_dir2_block_log_tail(tp
, bp
);
1223 xfs_dir2_data_check(dp
, bp
);
1224 xfs_da_buf_done(bp
);