2 * Copyright (c) 2000-2005 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_mount.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_dir2_sf.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
34 #include "xfs_dir2_data.h"
35 #include "xfs_dir2_leaf.h"
36 #include "xfs_dir2_block.h"
37 #include "xfs_dir2_node.h"
38 #include "xfs_error.h"
39 #include "xfs_trace.h"
42 * Function declarations.
44 static void xfs_dir2_free_log_header(xfs_trans_t
*tp
, xfs_dabuf_t
*bp
);
45 static int xfs_dir2_leafn_add(xfs_dabuf_t
*bp
, xfs_da_args_t
*args
, int index
);
47 static void xfs_dir2_leafn_check(xfs_inode_t
*dp
, xfs_dabuf_t
*bp
);
49 #define xfs_dir2_leafn_check(dp, bp)
51 static void xfs_dir2_leafn_moveents(xfs_da_args_t
*args
, xfs_dabuf_t
*bp_s
,
52 int start_s
, xfs_dabuf_t
*bp_d
, int start_d
,
54 static void xfs_dir2_leafn_rebalance(xfs_da_state_t
*state
,
55 xfs_da_state_blk_t
*blk1
,
56 xfs_da_state_blk_t
*blk2
);
57 static int xfs_dir2_leafn_remove(xfs_da_args_t
*args
, xfs_dabuf_t
*bp
,
58 int index
, xfs_da_state_blk_t
*dblk
,
60 static int xfs_dir2_node_addname_int(xfs_da_args_t
*args
,
61 xfs_da_state_blk_t
*fblk
);
64 * Log entries from a freespace block.
67 xfs_dir2_free_log_bests(
68 xfs_trans_t
*tp
, /* transaction pointer */
69 xfs_dabuf_t
*bp
, /* freespace buffer */
70 int first
, /* first entry to log */
71 int last
) /* last entry to log */
73 xfs_dir2_free_t
*free
; /* freespace structure */
76 ASSERT(be32_to_cpu(free
->hdr
.magic
) == XFS_DIR2_FREE_MAGIC
);
77 xfs_da_log_buf(tp
, bp
,
78 (uint
)((char *)&free
->bests
[first
] - (char *)free
),
79 (uint
)((char *)&free
->bests
[last
] - (char *)free
+
80 sizeof(free
->bests
[0]) - 1));
84 * Log header from a freespace block.
87 xfs_dir2_free_log_header(
88 xfs_trans_t
*tp
, /* transaction pointer */
89 xfs_dabuf_t
*bp
) /* freespace buffer */
91 xfs_dir2_free_t
*free
; /* freespace structure */
94 ASSERT(be32_to_cpu(free
->hdr
.magic
) == XFS_DIR2_FREE_MAGIC
);
95 xfs_da_log_buf(tp
, bp
, (uint
)((char *)&free
->hdr
- (char *)free
),
96 (uint
)(sizeof(xfs_dir2_free_hdr_t
) - 1));
100 * Convert a leaf-format directory to a node-format directory.
101 * We need to change the magic number of the leaf block, and copy
102 * the freespace table out of the leaf block into its own block.
105 xfs_dir2_leaf_to_node(
106 xfs_da_args_t
*args
, /* operation arguments */
107 xfs_dabuf_t
*lbp
) /* leaf buffer */
109 xfs_inode_t
*dp
; /* incore directory inode */
110 int error
; /* error return value */
111 xfs_dabuf_t
*fbp
; /* freespace buffer */
112 xfs_dir2_db_t fdb
; /* freespace block number */
113 xfs_dir2_free_t
*free
; /* freespace structure */
114 __be16
*from
; /* pointer to freespace entry */
115 int i
; /* leaf freespace index */
116 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
117 xfs_dir2_leaf_tail_t
*ltp
; /* leaf tail structure */
118 xfs_mount_t
*mp
; /* filesystem mount point */
119 int n
; /* count of live freespc ents */
120 xfs_dir2_data_off_t off
; /* freespace entry value */
121 __be16
*to
; /* pointer to freespace entry */
122 xfs_trans_t
*tp
; /* transaction pointer */
124 trace_xfs_dir2_leaf_to_node(args
);
130 * Add a freespace block to the directory.
132 if ((error
= xfs_dir2_grow_inode(args
, XFS_DIR2_FREE_SPACE
, &fdb
))) {
135 ASSERT(fdb
== XFS_DIR2_FREE_FIRSTDB(mp
));
137 * Get the buffer for the new freespace block.
139 if ((error
= xfs_da_get_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, fdb
), -1, &fbp
,
146 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
148 * Initialize the freespace block header.
150 free
->hdr
.magic
= cpu_to_be32(XFS_DIR2_FREE_MAGIC
);
151 free
->hdr
.firstdb
= 0;
152 ASSERT(be32_to_cpu(ltp
->bestcount
) <= (uint
)dp
->i_d
.di_size
/ mp
->m_dirblksize
);
153 free
->hdr
.nvalid
= ltp
->bestcount
;
155 * Copy freespace entries from the leaf block to the new block.
156 * Count active entries.
158 for (i
= n
= 0, from
= xfs_dir2_leaf_bests_p(ltp
), to
= free
->bests
;
159 i
< be32_to_cpu(ltp
->bestcount
); i
++, from
++, to
++) {
160 if ((off
= be16_to_cpu(*from
)) != NULLDATAOFF
)
162 *to
= cpu_to_be16(off
);
164 free
->hdr
.nused
= cpu_to_be32(n
);
165 leaf
->hdr
.info
.magic
= cpu_to_be16(XFS_DIR2_LEAFN_MAGIC
);
169 xfs_dir2_leaf_log_header(tp
, lbp
);
170 xfs_dir2_free_log_header(tp
, fbp
);
171 xfs_dir2_free_log_bests(tp
, fbp
, 0, be32_to_cpu(free
->hdr
.nvalid
) - 1);
172 xfs_da_buf_done(fbp
);
173 xfs_dir2_leafn_check(dp
, lbp
);
178 * Add a leaf entry to a leaf block in a node-form directory.
179 * The other work necessary is done from the caller.
181 static int /* error */
183 xfs_dabuf_t
*bp
, /* leaf buffer */
184 xfs_da_args_t
*args
, /* operation arguments */
185 int index
) /* insertion pt for new entry */
187 int compact
; /* compacting stale leaves */
188 xfs_inode_t
*dp
; /* incore directory inode */
189 int highstale
; /* next stale entry */
190 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
191 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
192 int lfloghigh
; /* high leaf entry logging */
193 int lfloglow
; /* low leaf entry logging */
194 int lowstale
; /* previous stale entry */
195 xfs_mount_t
*mp
; /* filesystem mount point */
196 xfs_trans_t
*tp
; /* transaction pointer */
198 trace_xfs_dir2_leafn_add(args
, index
);
206 * Quick check just to make sure we are not going to index
207 * into other peoples memory
210 return XFS_ERROR(EFSCORRUPTED
);
213 * If there are already the maximum number of leaf entries in
214 * the block, if there are no stale entries it won't fit.
215 * Caller will do a split. If there are stale entries we'll do
219 if (be16_to_cpu(leaf
->hdr
.count
) == xfs_dir2_max_leaf_ents(mp
)) {
220 if (!leaf
->hdr
.stale
)
221 return XFS_ERROR(ENOSPC
);
222 compact
= be16_to_cpu(leaf
->hdr
.stale
) > 1;
225 ASSERT(index
== 0 || be32_to_cpu(leaf
->ents
[index
- 1].hashval
) <= args
->hashval
);
226 ASSERT(index
== be16_to_cpu(leaf
->hdr
.count
) ||
227 be32_to_cpu(leaf
->ents
[index
].hashval
) >= args
->hashval
);
229 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
)
233 * Compact out all but one stale leaf entry. Leaves behind
234 * the entry closest to index.
237 xfs_dir2_leaf_compact_x1(bp
, &index
, &lowstale
, &highstale
,
238 &lfloglow
, &lfloghigh
);
241 * Set impossible logging indices for this case.
243 else if (leaf
->hdr
.stale
) {
244 lfloglow
= be16_to_cpu(leaf
->hdr
.count
);
248 * No stale entries, just insert a space for the new entry.
250 if (!leaf
->hdr
.stale
) {
251 lep
= &leaf
->ents
[index
];
252 if (index
< be16_to_cpu(leaf
->hdr
.count
))
253 memmove(lep
+ 1, lep
,
254 (be16_to_cpu(leaf
->hdr
.count
) - index
) * sizeof(*lep
));
256 lfloghigh
= be16_to_cpu(leaf
->hdr
.count
);
257 be16_add_cpu(&leaf
->hdr
.count
, 1);
260 * There are stale entries. We'll use one for the new entry.
264 * If we didn't do a compact then we need to figure out
265 * which stale entry will be used.
269 * Find first stale entry before our insertion point.
271 for (lowstale
= index
- 1;
273 be32_to_cpu(leaf
->ents
[lowstale
].address
) !=
274 XFS_DIR2_NULL_DATAPTR
;
278 * Find next stale entry after insertion point.
279 * Stop looking if the answer would be worse than
280 * lowstale already found.
282 for (highstale
= index
;
283 highstale
< be16_to_cpu(leaf
->hdr
.count
) &&
284 be32_to_cpu(leaf
->ents
[highstale
].address
) !=
285 XFS_DIR2_NULL_DATAPTR
&&
287 index
- lowstale
- 1 >= highstale
- index
);
292 * Using the low stale entry.
293 * Shift entries up toward the stale slot.
296 (highstale
== be16_to_cpu(leaf
->hdr
.count
) ||
297 index
- lowstale
- 1 < highstale
- index
)) {
298 ASSERT(be32_to_cpu(leaf
->ents
[lowstale
].address
) ==
299 XFS_DIR2_NULL_DATAPTR
);
300 ASSERT(index
- lowstale
- 1 >= 0);
301 if (index
- lowstale
- 1 > 0)
302 memmove(&leaf
->ents
[lowstale
],
303 &leaf
->ents
[lowstale
+ 1],
304 (index
- lowstale
- 1) * sizeof(*lep
));
305 lep
= &leaf
->ents
[index
- 1];
306 lfloglow
= MIN(lowstale
, lfloglow
);
307 lfloghigh
= MAX(index
- 1, lfloghigh
);
310 * Using the high stale entry.
311 * Shift entries down toward the stale slot.
314 ASSERT(be32_to_cpu(leaf
->ents
[highstale
].address
) ==
315 XFS_DIR2_NULL_DATAPTR
);
316 ASSERT(highstale
- index
>= 0);
317 if (highstale
- index
> 0)
318 memmove(&leaf
->ents
[index
+ 1],
320 (highstale
- index
) * sizeof(*lep
));
321 lep
= &leaf
->ents
[index
];
322 lfloglow
= MIN(index
, lfloglow
);
323 lfloghigh
= MAX(highstale
, lfloghigh
);
325 be16_add_cpu(&leaf
->hdr
.stale
, -1);
328 * Insert the new entry, log everything.
330 lep
->hashval
= cpu_to_be32(args
->hashval
);
331 lep
->address
= cpu_to_be32(xfs_dir2_db_off_to_dataptr(mp
,
332 args
->blkno
, args
->index
));
333 xfs_dir2_leaf_log_header(tp
, bp
);
334 xfs_dir2_leaf_log_ents(tp
, bp
, lfloglow
, lfloghigh
);
335 xfs_dir2_leafn_check(dp
, bp
);
341 * Check internal consistency of a leafn block.
344 xfs_dir2_leafn_check(
345 xfs_inode_t
*dp
, /* incore directory inode */
346 xfs_dabuf_t
*bp
) /* leaf buffer */
348 int i
; /* leaf index */
349 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
350 xfs_mount_t
*mp
; /* filesystem mount point */
351 int stale
; /* count of stale leaves */
355 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
356 ASSERT(be16_to_cpu(leaf
->hdr
.count
) <= xfs_dir2_max_leaf_ents(mp
));
357 for (i
= stale
= 0; i
< be16_to_cpu(leaf
->hdr
.count
); i
++) {
358 if (i
+ 1 < be16_to_cpu(leaf
->hdr
.count
)) {
359 ASSERT(be32_to_cpu(leaf
->ents
[i
].hashval
) <=
360 be32_to_cpu(leaf
->ents
[i
+ 1].hashval
));
362 if (be32_to_cpu(leaf
->ents
[i
].address
) == XFS_DIR2_NULL_DATAPTR
)
365 ASSERT(be16_to_cpu(leaf
->hdr
.stale
) == stale
);
370 * Return the last hash value in the leaf.
371 * Stale entries are ok.
373 xfs_dahash_t
/* hash value */
374 xfs_dir2_leafn_lasthash(
375 xfs_dabuf_t
*bp
, /* leaf buffer */
376 int *count
) /* count of entries in leaf */
378 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
381 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
383 *count
= be16_to_cpu(leaf
->hdr
.count
);
384 if (!leaf
->hdr
.count
)
386 return be32_to_cpu(leaf
->ents
[be16_to_cpu(leaf
->hdr
.count
) - 1].hashval
);
390 * Look up a leaf entry for space to add a name in a node-format leaf block.
391 * The extrablk in state is a freespace block.
394 xfs_dir2_leafn_lookup_for_addname(
395 xfs_dabuf_t
*bp
, /* leaf buffer */
396 xfs_da_args_t
*args
, /* operation arguments */
397 int *indexp
, /* out: leaf entry index */
398 xfs_da_state_t
*state
) /* state to fill in */
400 xfs_dabuf_t
*curbp
= NULL
; /* current data/free buffer */
401 xfs_dir2_db_t curdb
= -1; /* current data block number */
402 xfs_dir2_db_t curfdb
= -1; /* current free block number */
403 xfs_inode_t
*dp
; /* incore directory inode */
404 int error
; /* error return value */
405 int fi
; /* free entry index */
406 xfs_dir2_free_t
*free
= NULL
; /* free block structure */
407 int index
; /* leaf entry index */
408 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
409 int length
; /* length of new data entry */
410 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
411 xfs_mount_t
*mp
; /* filesystem mount point */
412 xfs_dir2_db_t newdb
; /* new data block number */
413 xfs_dir2_db_t newfdb
; /* new free block number */
414 xfs_trans_t
*tp
; /* transaction pointer */
420 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
422 ASSERT(be16_to_cpu(leaf
->hdr
.count
) > 0);
424 xfs_dir2_leafn_check(dp
, bp
);
426 * Look up the hash value in the leaf entries.
428 index
= xfs_dir2_leaf_search_hash(args
, bp
);
430 * Do we have a buffer coming in?
432 if (state
->extravalid
) {
433 /* If so, it's a free block buffer, get the block number. */
434 curbp
= state
->extrablk
.bp
;
435 curfdb
= state
->extrablk
.blkno
;
437 ASSERT(be32_to_cpu(free
->hdr
.magic
) == XFS_DIR2_FREE_MAGIC
);
439 length
= xfs_dir2_data_entsize(args
->namelen
);
441 * Loop over leaf entries with the right hash value.
443 for (lep
= &leaf
->ents
[index
]; index
< be16_to_cpu(leaf
->hdr
.count
) &&
444 be32_to_cpu(lep
->hashval
) == args
->hashval
;
447 * Skip stale leaf entries.
449 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
452 * Pull the data block number from the entry.
454 newdb
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
456 * For addname, we're looking for a place to put the new entry.
457 * We want to use a data block with an entry of equal
458 * hash value to ours if there is one with room.
460 * If this block isn't the data block we already have
461 * in hand, take a look at it.
463 if (newdb
!= curdb
) {
466 * Convert the data block to the free block
467 * holding its freespace information.
469 newfdb
= xfs_dir2_db_to_fdb(mp
, newdb
);
471 * If it's not the one we have in hand, read it in.
473 if (newfdb
!= curfdb
) {
475 * If we had one before, drop it.
478 xfs_da_brelse(tp
, curbp
);
480 * Read the free block.
482 error
= xfs_da_read_buf(tp
, dp
,
483 xfs_dir2_db_to_da(mp
, newfdb
),
484 -1, &curbp
, XFS_DATA_FORK
);
488 ASSERT(be32_to_cpu(free
->hdr
.magic
) ==
489 XFS_DIR2_FREE_MAGIC
);
490 ASSERT((be32_to_cpu(free
->hdr
.firstdb
) %
491 XFS_DIR2_MAX_FREE_BESTS(mp
)) == 0);
492 ASSERT(be32_to_cpu(free
->hdr
.firstdb
) <= curdb
);
493 ASSERT(curdb
< be32_to_cpu(free
->hdr
.firstdb
) +
494 be32_to_cpu(free
->hdr
.nvalid
));
497 * Get the index for our entry.
499 fi
= xfs_dir2_db_to_fdindex(mp
, curdb
);
501 * If it has room, return it.
503 if (unlikely(be16_to_cpu(free
->bests
[fi
]) == NULLDATAOFF
)) {
504 XFS_ERROR_REPORT("xfs_dir2_leafn_lookup_int",
505 XFS_ERRLEVEL_LOW
, mp
);
506 if (curfdb
!= newfdb
)
507 xfs_da_brelse(tp
, curbp
);
508 return XFS_ERROR(EFSCORRUPTED
);
511 if (be16_to_cpu(free
->bests
[fi
]) >= length
)
515 /* Didn't find any space */
518 ASSERT(args
->op_flags
& XFS_DA_OP_OKNOENT
);
520 /* Giving back a free block. */
521 state
->extravalid
= 1;
522 state
->extrablk
.bp
= curbp
;
523 state
->extrablk
.index
= fi
;
524 state
->extrablk
.blkno
= curfdb
;
525 state
->extrablk
.magic
= XFS_DIR2_FREE_MAGIC
;
527 state
->extravalid
= 0;
530 * Return the index, that will be the insertion point.
533 return XFS_ERROR(ENOENT
);
537 * Look up a leaf entry in a node-format leaf block.
538 * The extrablk in state a data block.
541 xfs_dir2_leafn_lookup_for_entry(
542 xfs_dabuf_t
*bp
, /* leaf buffer */
543 xfs_da_args_t
*args
, /* operation arguments */
544 int *indexp
, /* out: leaf entry index */
545 xfs_da_state_t
*state
) /* state to fill in */
547 xfs_dabuf_t
*curbp
= NULL
; /* current data/free buffer */
548 xfs_dir2_db_t curdb
= -1; /* current data block number */
549 xfs_dir2_data_entry_t
*dep
; /* data block entry */
550 xfs_inode_t
*dp
; /* incore directory inode */
551 int error
; /* error return value */
552 int index
; /* leaf entry index */
553 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
554 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
555 xfs_mount_t
*mp
; /* filesystem mount point */
556 xfs_dir2_db_t newdb
; /* new data block number */
557 xfs_trans_t
*tp
; /* transaction pointer */
558 enum xfs_dacmp cmp
; /* comparison result */
564 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
566 ASSERT(be16_to_cpu(leaf
->hdr
.count
) > 0);
568 xfs_dir2_leafn_check(dp
, bp
);
570 * Look up the hash value in the leaf entries.
572 index
= xfs_dir2_leaf_search_hash(args
, bp
);
574 * Do we have a buffer coming in?
576 if (state
->extravalid
) {
577 curbp
= state
->extrablk
.bp
;
578 curdb
= state
->extrablk
.blkno
;
581 * Loop over leaf entries with the right hash value.
583 for (lep
= &leaf
->ents
[index
]; index
< be16_to_cpu(leaf
->hdr
.count
) &&
584 be32_to_cpu(lep
->hashval
) == args
->hashval
;
587 * Skip stale leaf entries.
589 if (be32_to_cpu(lep
->address
) == XFS_DIR2_NULL_DATAPTR
)
592 * Pull the data block number from the entry.
594 newdb
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
596 * Not adding a new entry, so we really want to find
597 * the name given to us.
599 * If it's a different data block, go get it.
601 if (newdb
!= curdb
) {
603 * If we had a block before that we aren't saving
604 * for a CI name, drop it
606 if (curbp
&& (args
->cmpresult
== XFS_CMP_DIFFERENT
||
607 curdb
!= state
->extrablk
.blkno
))
608 xfs_da_brelse(tp
, curbp
);
610 * If needing the block that is saved with a CI match,
611 * use it otherwise read in the new data block.
613 if (args
->cmpresult
!= XFS_CMP_DIFFERENT
&&
614 newdb
== state
->extrablk
.blkno
) {
615 ASSERT(state
->extravalid
);
616 curbp
= state
->extrablk
.bp
;
618 error
= xfs_da_read_buf(tp
, dp
,
619 xfs_dir2_db_to_da(mp
, newdb
),
620 -1, &curbp
, XFS_DATA_FORK
);
624 xfs_dir2_data_check(dp
, curbp
);
628 * Point to the data entry.
630 dep
= (xfs_dir2_data_entry_t
*)((char *)curbp
->data
+
631 xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
)));
633 * Compare the entry and if it's an exact match, return
634 * EEXIST immediately. If it's the first case-insensitive
635 * match, store the block & inode number and continue looking.
637 cmp
= mp
->m_dirnameops
->compname(args
, dep
->name
, dep
->namelen
);
638 if (cmp
!= XFS_CMP_DIFFERENT
&& cmp
!= args
->cmpresult
) {
639 /* If there is a CI match block, drop it */
640 if (args
->cmpresult
!= XFS_CMP_DIFFERENT
&&
641 curdb
!= state
->extrablk
.blkno
)
642 xfs_da_brelse(tp
, state
->extrablk
.bp
);
643 args
->cmpresult
= cmp
;
644 args
->inumber
= be64_to_cpu(dep
->inumber
);
646 state
->extravalid
= 1;
647 state
->extrablk
.bp
= curbp
;
648 state
->extrablk
.blkno
= curdb
;
649 state
->extrablk
.index
= (int)((char *)dep
-
650 (char *)curbp
->data
);
651 state
->extrablk
.magic
= XFS_DIR2_DATA_MAGIC
;
652 if (cmp
== XFS_CMP_EXACT
)
653 return XFS_ERROR(EEXIST
);
656 ASSERT(index
== be16_to_cpu(leaf
->hdr
.count
) ||
657 (args
->op_flags
& XFS_DA_OP_OKNOENT
));
659 if (args
->cmpresult
== XFS_CMP_DIFFERENT
) {
660 /* Giving back last used data block. */
661 state
->extravalid
= 1;
662 state
->extrablk
.bp
= curbp
;
663 state
->extrablk
.index
= -1;
664 state
->extrablk
.blkno
= curdb
;
665 state
->extrablk
.magic
= XFS_DIR2_DATA_MAGIC
;
667 /* If the curbp is not the CI match block, drop it */
668 if (state
->extrablk
.bp
!= curbp
)
669 xfs_da_brelse(tp
, curbp
);
672 state
->extravalid
= 0;
675 return XFS_ERROR(ENOENT
);
679 * Look up a leaf entry in a node-format leaf block.
680 * If this is an addname then the extrablk in state is a freespace block,
681 * otherwise it's a data block.
684 xfs_dir2_leafn_lookup_int(
685 xfs_dabuf_t
*bp
, /* leaf buffer */
686 xfs_da_args_t
*args
, /* operation arguments */
687 int *indexp
, /* out: leaf entry index */
688 xfs_da_state_t
*state
) /* state to fill in */
690 if (args
->op_flags
& XFS_DA_OP_ADDNAME
)
691 return xfs_dir2_leafn_lookup_for_addname(bp
, args
, indexp
,
693 return xfs_dir2_leafn_lookup_for_entry(bp
, args
, indexp
, state
);
697 * Move count leaf entries from source to destination leaf.
698 * Log entries and headers. Stale entries are preserved.
701 xfs_dir2_leafn_moveents(
702 xfs_da_args_t
*args
, /* operation arguments */
703 xfs_dabuf_t
*bp_s
, /* source leaf buffer */
704 int start_s
, /* source leaf index */
705 xfs_dabuf_t
*bp_d
, /* destination leaf buffer */
706 int start_d
, /* destination leaf index */
707 int count
) /* count of leaves to copy */
709 xfs_dir2_leaf_t
*leaf_d
; /* destination leaf structure */
710 xfs_dir2_leaf_t
*leaf_s
; /* source leaf structure */
711 int stale
; /* count stale leaves copied */
712 xfs_trans_t
*tp
; /* transaction pointer */
714 trace_xfs_dir2_leafn_moveents(args
, start_s
, start_d
, count
);
717 * Silently return if nothing to do.
726 * If the destination index is not the end of the current
727 * destination leaf entries, open up a hole in the destination
728 * to hold the new entries.
730 if (start_d
< be16_to_cpu(leaf_d
->hdr
.count
)) {
731 memmove(&leaf_d
->ents
[start_d
+ count
], &leaf_d
->ents
[start_d
],
732 (be16_to_cpu(leaf_d
->hdr
.count
) - start_d
) *
733 sizeof(xfs_dir2_leaf_entry_t
));
734 xfs_dir2_leaf_log_ents(tp
, bp_d
, start_d
+ count
,
735 count
+ be16_to_cpu(leaf_d
->hdr
.count
) - 1);
738 * If the source has stale leaves, count the ones in the copy range
739 * so we can update the header correctly.
741 if (leaf_s
->hdr
.stale
) {
742 int i
; /* temp leaf index */
744 for (i
= start_s
, stale
= 0; i
< start_s
+ count
; i
++) {
745 if (be32_to_cpu(leaf_s
->ents
[i
].address
) == XFS_DIR2_NULL_DATAPTR
)
751 * Copy the leaf entries from source to destination.
753 memcpy(&leaf_d
->ents
[start_d
], &leaf_s
->ents
[start_s
],
754 count
* sizeof(xfs_dir2_leaf_entry_t
));
755 xfs_dir2_leaf_log_ents(tp
, bp_d
, start_d
, start_d
+ count
- 1);
757 * If there are source entries after the ones we copied,
758 * delete the ones we copied by sliding the next ones down.
760 if (start_s
+ count
< be16_to_cpu(leaf_s
->hdr
.count
)) {
761 memmove(&leaf_s
->ents
[start_s
], &leaf_s
->ents
[start_s
+ count
],
762 count
* sizeof(xfs_dir2_leaf_entry_t
));
763 xfs_dir2_leaf_log_ents(tp
, bp_s
, start_s
, start_s
+ count
- 1);
766 * Update the headers and log them.
768 be16_add_cpu(&leaf_s
->hdr
.count
, -(count
));
769 be16_add_cpu(&leaf_s
->hdr
.stale
, -(stale
));
770 be16_add_cpu(&leaf_d
->hdr
.count
, count
);
771 be16_add_cpu(&leaf_d
->hdr
.stale
, stale
);
772 xfs_dir2_leaf_log_header(tp
, bp_s
);
773 xfs_dir2_leaf_log_header(tp
, bp_d
);
774 xfs_dir2_leafn_check(args
->dp
, bp_s
);
775 xfs_dir2_leafn_check(args
->dp
, bp_d
);
779 * Determine the sort order of two leaf blocks.
780 * Returns 1 if both are valid and leaf2 should be before leaf1, else 0.
783 xfs_dir2_leafn_order(
784 xfs_dabuf_t
*leaf1_bp
, /* leaf1 buffer */
785 xfs_dabuf_t
*leaf2_bp
) /* leaf2 buffer */
787 xfs_dir2_leaf_t
*leaf1
; /* leaf1 structure */
788 xfs_dir2_leaf_t
*leaf2
; /* leaf2 structure */
790 leaf1
= leaf1_bp
->data
;
791 leaf2
= leaf2_bp
->data
;
792 ASSERT(be16_to_cpu(leaf1
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
793 ASSERT(be16_to_cpu(leaf2
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
794 if (be16_to_cpu(leaf1
->hdr
.count
) > 0 &&
795 be16_to_cpu(leaf2
->hdr
.count
) > 0 &&
796 (be32_to_cpu(leaf2
->ents
[0].hashval
) < be32_to_cpu(leaf1
->ents
[0].hashval
) ||
797 be32_to_cpu(leaf2
->ents
[be16_to_cpu(leaf2
->hdr
.count
) - 1].hashval
) <
798 be32_to_cpu(leaf1
->ents
[be16_to_cpu(leaf1
->hdr
.count
) - 1].hashval
)))
804 * Rebalance leaf entries between two leaf blocks.
805 * This is actually only called when the second block is new,
806 * though the code deals with the general case.
807 * A new entry will be inserted in one of the blocks, and that
808 * entry is taken into account when balancing.
811 xfs_dir2_leafn_rebalance(
812 xfs_da_state_t
*state
, /* btree cursor */
813 xfs_da_state_blk_t
*blk1
, /* first btree block */
814 xfs_da_state_blk_t
*blk2
) /* second btree block */
816 xfs_da_args_t
*args
; /* operation arguments */
817 int count
; /* count (& direction) leaves */
818 int isleft
; /* new goes in left leaf */
819 xfs_dir2_leaf_t
*leaf1
; /* first leaf structure */
820 xfs_dir2_leaf_t
*leaf2
; /* second leaf structure */
821 int mid
; /* midpoint leaf index */
823 int oldstale
; /* old count of stale leaves */
825 int oldsum
; /* old total leaf count */
826 int swap
; /* swapped leaf blocks */
830 * If the block order is wrong, swap the arguments.
832 if ((swap
= xfs_dir2_leafn_order(blk1
->bp
, blk2
->bp
))) {
833 xfs_da_state_blk_t
*tmp
; /* temp for block swap */
839 leaf1
= blk1
->bp
->data
;
840 leaf2
= blk2
->bp
->data
;
841 oldsum
= be16_to_cpu(leaf1
->hdr
.count
) + be16_to_cpu(leaf2
->hdr
.count
);
843 oldstale
= be16_to_cpu(leaf1
->hdr
.stale
) + be16_to_cpu(leaf2
->hdr
.stale
);
847 * If the old leaf count was odd then the new one will be even,
848 * so we need to divide the new count evenly.
851 xfs_dahash_t midhash
; /* middle entry hash value */
853 if (mid
>= be16_to_cpu(leaf1
->hdr
.count
))
854 midhash
= be32_to_cpu(leaf2
->ents
[mid
- be16_to_cpu(leaf1
->hdr
.count
)].hashval
);
856 midhash
= be32_to_cpu(leaf1
->ents
[mid
].hashval
);
857 isleft
= args
->hashval
<= midhash
;
860 * If the old count is even then the new count is odd, so there's
861 * no preferred side for the new entry.
867 * Calculate moved entry count. Positive means left-to-right,
868 * negative means right-to-left. Then move the entries.
870 count
= be16_to_cpu(leaf1
->hdr
.count
) - mid
+ (isleft
== 0);
872 xfs_dir2_leafn_moveents(args
, blk1
->bp
,
873 be16_to_cpu(leaf1
->hdr
.count
) - count
, blk2
->bp
, 0, count
);
875 xfs_dir2_leafn_moveents(args
, blk2
->bp
, 0, blk1
->bp
,
876 be16_to_cpu(leaf1
->hdr
.count
), count
);
877 ASSERT(be16_to_cpu(leaf1
->hdr
.count
) + be16_to_cpu(leaf2
->hdr
.count
) == oldsum
);
878 ASSERT(be16_to_cpu(leaf1
->hdr
.stale
) + be16_to_cpu(leaf2
->hdr
.stale
) == oldstale
);
880 * Mark whether we're inserting into the old or new leaf.
882 if (be16_to_cpu(leaf1
->hdr
.count
) < be16_to_cpu(leaf2
->hdr
.count
))
883 state
->inleaf
= swap
;
884 else if (be16_to_cpu(leaf1
->hdr
.count
) > be16_to_cpu(leaf2
->hdr
.count
))
885 state
->inleaf
= !swap
;
888 swap
^ (blk1
->index
<= be16_to_cpu(leaf1
->hdr
.count
));
890 * Adjust the expected index for insertion.
893 blk2
->index
= blk1
->index
- be16_to_cpu(leaf1
->hdr
.count
);
896 * Finally sanity check just to make sure we are not returning a
899 if(blk2
->index
< 0) {
902 xfs_alert(args
->dp
->i_mount
,
903 "%s: picked the wrong leaf? reverting original leaf: blk1->index %d\n",
904 __func__
, blk1
->index
);
909 * Remove an entry from a node directory.
910 * This removes the leaf entry and the data entry,
911 * and updates the free block if necessary.
913 static int /* error */
914 xfs_dir2_leafn_remove(
915 xfs_da_args_t
*args
, /* operation arguments */
916 xfs_dabuf_t
*bp
, /* leaf buffer */
917 int index
, /* leaf entry index */
918 xfs_da_state_blk_t
*dblk
, /* data block */
919 int *rval
) /* resulting block needs join */
921 xfs_dir2_data_t
*data
; /* data block structure */
922 xfs_dir2_db_t db
; /* data block number */
923 xfs_dabuf_t
*dbp
; /* data block buffer */
924 xfs_dir2_data_entry_t
*dep
; /* data block entry */
925 xfs_inode_t
*dp
; /* incore directory inode */
926 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
927 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry */
928 int longest
; /* longest data free entry */
929 int off
; /* data block entry offset */
930 xfs_mount_t
*mp
; /* filesystem mount point */
931 int needlog
; /* need to log data header */
932 int needscan
; /* need to rescan data frees */
933 xfs_trans_t
*tp
; /* transaction pointer */
935 trace_xfs_dir2_leafn_remove(args
, index
);
941 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
943 * Point to the entry we're removing.
945 lep
= &leaf
->ents
[index
];
947 * Extract the data block and offset from the entry.
949 db
= xfs_dir2_dataptr_to_db(mp
, be32_to_cpu(lep
->address
));
950 ASSERT(dblk
->blkno
== db
);
951 off
= xfs_dir2_dataptr_to_off(mp
, be32_to_cpu(lep
->address
));
952 ASSERT(dblk
->index
== off
);
954 * Kill the leaf entry by marking it stale.
955 * Log the leaf block changes.
957 be16_add_cpu(&leaf
->hdr
.stale
, 1);
958 xfs_dir2_leaf_log_header(tp
, bp
);
959 lep
->address
= cpu_to_be32(XFS_DIR2_NULL_DATAPTR
);
960 xfs_dir2_leaf_log_ents(tp
, bp
, index
, index
);
962 * Make the data entry free. Keep track of the longest freespace
963 * in the data block in case it changes.
967 dep
= (xfs_dir2_data_entry_t
*)((char *)data
+ off
);
968 longest
= be16_to_cpu(data
->hdr
.bestfree
[0].length
);
969 needlog
= needscan
= 0;
970 xfs_dir2_data_make_free(tp
, dbp
, off
,
971 xfs_dir2_data_entsize(dep
->namelen
), &needlog
, &needscan
);
973 * Rescan the data block freespaces for bestfree.
974 * Log the data block header if needed.
977 xfs_dir2_data_freescan(mp
, data
, &needlog
);
979 xfs_dir2_data_log_header(tp
, dbp
);
980 xfs_dir2_data_check(dp
, dbp
);
982 * If the longest data block freespace changes, need to update
983 * the corresponding freeblock entry.
985 if (longest
< be16_to_cpu(data
->hdr
.bestfree
[0].length
)) {
986 int error
; /* error return value */
987 xfs_dabuf_t
*fbp
; /* freeblock buffer */
988 xfs_dir2_db_t fdb
; /* freeblock block number */
989 int findex
; /* index in freeblock entries */
990 xfs_dir2_free_t
*free
; /* freeblock structure */
991 int logfree
; /* need to log free entry */
994 * Convert the data block number to a free block,
995 * read in the free block.
997 fdb
= xfs_dir2_db_to_fdb(mp
, db
);
998 if ((error
= xfs_da_read_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, fdb
),
999 -1, &fbp
, XFS_DATA_FORK
))) {
1003 ASSERT(be32_to_cpu(free
->hdr
.magic
) == XFS_DIR2_FREE_MAGIC
);
1004 ASSERT(be32_to_cpu(free
->hdr
.firstdb
) ==
1005 XFS_DIR2_MAX_FREE_BESTS(mp
) *
1006 (fdb
- XFS_DIR2_FREE_FIRSTDB(mp
)));
1008 * Calculate which entry we need to fix.
1010 findex
= xfs_dir2_db_to_fdindex(mp
, db
);
1011 longest
= be16_to_cpu(data
->hdr
.bestfree
[0].length
);
1013 * If the data block is now empty we can get rid of it
1016 if (longest
== mp
->m_dirblksize
- (uint
)sizeof(data
->hdr
)) {
1018 * Try to punch out the data block.
1020 error
= xfs_dir2_shrink_inode(args
, db
, dbp
);
1026 * We can get ENOSPC if there's no space reservation.
1027 * In this case just drop the buffer and some one else
1028 * will eventually get rid of the empty block.
1030 else if (error
== ENOSPC
&& args
->total
== 0)
1031 xfs_da_buf_done(dbp
);
1036 * If we got rid of the data block, we can eliminate that entry
1037 * in the free block.
1041 * One less used entry in the free table.
1043 be32_add_cpu(&free
->hdr
.nused
, -1);
1044 xfs_dir2_free_log_header(tp
, fbp
);
1046 * If this was the last entry in the table, we can
1047 * trim the table size back. There might be other
1048 * entries at the end referring to non-existent
1049 * data blocks, get those too.
1051 if (findex
== be32_to_cpu(free
->hdr
.nvalid
) - 1) {
1052 int i
; /* free entry index */
1054 for (i
= findex
- 1;
1055 i
>= 0 && be16_to_cpu(free
->bests
[i
]) == NULLDATAOFF
;
1058 free
->hdr
.nvalid
= cpu_to_be32(i
+ 1);
1062 * Not the last entry, just punch it out.
1065 free
->bests
[findex
] = cpu_to_be16(NULLDATAOFF
);
1069 * If there are no useful entries left in the block,
1070 * get rid of the block if we can.
1072 if (!free
->hdr
.nused
) {
1073 error
= xfs_dir2_shrink_inode(args
, fdb
, fbp
);
1077 } else if (error
!= ENOSPC
|| args
->total
!= 0)
1080 * It's possible to get ENOSPC if there is no
1081 * space reservation. In this case some one
1082 * else will eventually get rid of this block.
1087 * Data block is not empty, just set the free entry to
1091 free
->bests
[findex
] = cpu_to_be16(longest
);
1095 * Log the free entry that changed, unless we got rid of it.
1098 xfs_dir2_free_log_bests(tp
, fbp
, findex
, findex
);
1100 * Drop the buffer if we still have it.
1103 xfs_da_buf_done(fbp
);
1105 xfs_dir2_leafn_check(dp
, bp
);
1107 * Return indication of whether this leaf block is empty enough
1108 * to justify trying to join it with a neighbor.
1111 ((uint
)sizeof(leaf
->hdr
) +
1112 (uint
)sizeof(leaf
->ents
[0]) *
1113 (be16_to_cpu(leaf
->hdr
.count
) - be16_to_cpu(leaf
->hdr
.stale
))) <
1119 * Split the leaf entries in the old block into old and new blocks.
1122 xfs_dir2_leafn_split(
1123 xfs_da_state_t
*state
, /* btree cursor */
1124 xfs_da_state_blk_t
*oldblk
, /* original block */
1125 xfs_da_state_blk_t
*newblk
) /* newly created block */
1127 xfs_da_args_t
*args
; /* operation arguments */
1128 xfs_dablk_t blkno
; /* new leaf block number */
1129 int error
; /* error return value */
1130 xfs_mount_t
*mp
; /* filesystem mount point */
1133 * Allocate space for a new leaf node.
1136 mp
= args
->dp
->i_mount
;
1137 ASSERT(args
!= NULL
);
1138 ASSERT(oldblk
->magic
== XFS_DIR2_LEAFN_MAGIC
);
1139 error
= xfs_da_grow_inode(args
, &blkno
);
1144 * Initialize the new leaf block.
1146 error
= xfs_dir2_leaf_init(args
, xfs_dir2_da_to_db(mp
, blkno
),
1147 &newblk
->bp
, XFS_DIR2_LEAFN_MAGIC
);
1151 newblk
->blkno
= blkno
;
1152 newblk
->magic
= XFS_DIR2_LEAFN_MAGIC
;
1154 * Rebalance the entries across the two leaves, link the new
1155 * block into the leaves.
1157 xfs_dir2_leafn_rebalance(state
, oldblk
, newblk
);
1158 error
= xfs_da_blk_link(state
, oldblk
, newblk
);
1163 * Insert the new entry in the correct block.
1166 error
= xfs_dir2_leafn_add(oldblk
->bp
, args
, oldblk
->index
);
1168 error
= xfs_dir2_leafn_add(newblk
->bp
, args
, newblk
->index
);
1170 * Update last hashval in each block since we added the name.
1172 oldblk
->hashval
= xfs_dir2_leafn_lasthash(oldblk
->bp
, NULL
);
1173 newblk
->hashval
= xfs_dir2_leafn_lasthash(newblk
->bp
, NULL
);
1174 xfs_dir2_leafn_check(args
->dp
, oldblk
->bp
);
1175 xfs_dir2_leafn_check(args
->dp
, newblk
->bp
);
1180 * Check a leaf block and its neighbors to see if the block should be
1181 * collapsed into one or the other neighbor. Always keep the block
1182 * with the smaller block number.
1183 * If the current block is over 50% full, don't try to join it, return 0.
1184 * If the block is empty, fill in the state structure and return 2.
1185 * If it can be collapsed, fill in the state structure and return 1.
1186 * If nothing can be done, return 0.
1189 xfs_dir2_leafn_toosmall(
1190 xfs_da_state_t
*state
, /* btree cursor */
1191 int *action
) /* resulting action to take */
1193 xfs_da_state_blk_t
*blk
; /* leaf block */
1194 xfs_dablk_t blkno
; /* leaf block number */
1195 xfs_dabuf_t
*bp
; /* leaf buffer */
1196 int bytes
; /* bytes in use */
1197 int count
; /* leaf live entry count */
1198 int error
; /* error return value */
1199 int forward
; /* sibling block direction */
1200 int i
; /* sibling counter */
1201 xfs_da_blkinfo_t
*info
; /* leaf block header */
1202 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1203 int rval
; /* result from path_shift */
1206 * Check for the degenerate case of the block being over 50% full.
1207 * If so, it's not worth even looking to see if we might be able
1208 * to coalesce with a sibling.
1210 blk
= &state
->path
.blk
[state
->path
.active
- 1];
1211 info
= blk
->bp
->data
;
1212 ASSERT(be16_to_cpu(info
->magic
) == XFS_DIR2_LEAFN_MAGIC
);
1213 leaf
= (xfs_dir2_leaf_t
*)info
;
1214 count
= be16_to_cpu(leaf
->hdr
.count
) - be16_to_cpu(leaf
->hdr
.stale
);
1215 bytes
= (uint
)sizeof(leaf
->hdr
) + count
* (uint
)sizeof(leaf
->ents
[0]);
1216 if (bytes
> (state
->blocksize
>> 1)) {
1218 * Blk over 50%, don't try to join.
1224 * Check for the degenerate case of the block being empty.
1225 * If the block is empty, we'll simply delete it, no need to
1226 * coalesce it with a sibling block. We choose (arbitrarily)
1227 * to merge with the forward block unless it is NULL.
1231 * Make altpath point to the block we want to keep and
1232 * path point to the block we want to drop (this one).
1234 forward
= (info
->forw
!= 0);
1235 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1236 error
= xfs_da_path_shift(state
, &state
->altpath
, forward
, 0,
1240 *action
= rval
? 2 : 0;
1244 * Examine each sibling block to see if we can coalesce with
1245 * at least 25% free space to spare. We need to figure out
1246 * whether to merge with the forward or the backward block.
1247 * We prefer coalescing with the lower numbered sibling so as
1248 * to shrink a directory over time.
1250 forward
= be32_to_cpu(info
->forw
) < be32_to_cpu(info
->back
);
1251 for (i
= 0, bp
= NULL
; i
< 2; forward
= !forward
, i
++) {
1252 blkno
= forward
? be32_to_cpu(info
->forw
) : be32_to_cpu(info
->back
);
1256 * Read the sibling leaf block.
1259 xfs_da_read_buf(state
->args
->trans
, state
->args
->dp
, blkno
,
1260 -1, &bp
, XFS_DATA_FORK
))) {
1265 * Count bytes in the two blocks combined.
1267 leaf
= (xfs_dir2_leaf_t
*)info
;
1268 count
= be16_to_cpu(leaf
->hdr
.count
) - be16_to_cpu(leaf
->hdr
.stale
);
1269 bytes
= state
->blocksize
- (state
->blocksize
>> 2);
1271 ASSERT(be16_to_cpu(leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
1272 count
+= be16_to_cpu(leaf
->hdr
.count
) - be16_to_cpu(leaf
->hdr
.stale
);
1273 bytes
-= count
* (uint
)sizeof(leaf
->ents
[0]);
1275 * Fits with at least 25% to spare.
1279 xfs_da_brelse(state
->args
->trans
, bp
);
1282 * Didn't like either block, give up.
1289 * Done with the sibling leaf block here, drop the dabuf
1290 * so path_shift can get it.
1292 xfs_da_buf_done(bp
);
1294 * Make altpath point to the block we want to keep (the lower
1295 * numbered block) and path point to the block we want to drop.
1297 memcpy(&state
->altpath
, &state
->path
, sizeof(state
->path
));
1298 if (blkno
< blk
->blkno
)
1299 error
= xfs_da_path_shift(state
, &state
->altpath
, forward
, 0,
1302 error
= xfs_da_path_shift(state
, &state
->path
, forward
, 0,
1307 *action
= rval
? 0 : 1;
1312 * Move all the leaf entries from drop_blk to save_blk.
1313 * This is done as part of a join operation.
1316 xfs_dir2_leafn_unbalance(
1317 xfs_da_state_t
*state
, /* cursor */
1318 xfs_da_state_blk_t
*drop_blk
, /* dead block */
1319 xfs_da_state_blk_t
*save_blk
) /* surviving block */
1321 xfs_da_args_t
*args
; /* operation arguments */
1322 xfs_dir2_leaf_t
*drop_leaf
; /* dead leaf structure */
1323 xfs_dir2_leaf_t
*save_leaf
; /* surviving leaf structure */
1326 ASSERT(drop_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
);
1327 ASSERT(save_blk
->magic
== XFS_DIR2_LEAFN_MAGIC
);
1328 drop_leaf
= drop_blk
->bp
->data
;
1329 save_leaf
= save_blk
->bp
->data
;
1330 ASSERT(be16_to_cpu(drop_leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
1331 ASSERT(be16_to_cpu(save_leaf
->hdr
.info
.magic
) == XFS_DIR2_LEAFN_MAGIC
);
1333 * If there are any stale leaf entries, take this opportunity
1336 if (drop_leaf
->hdr
.stale
)
1337 xfs_dir2_leaf_compact(args
, drop_blk
->bp
);
1338 if (save_leaf
->hdr
.stale
)
1339 xfs_dir2_leaf_compact(args
, save_blk
->bp
);
1341 * Move the entries from drop to the appropriate end of save.
1343 drop_blk
->hashval
= be32_to_cpu(drop_leaf
->ents
[be16_to_cpu(drop_leaf
->hdr
.count
) - 1].hashval
);
1344 if (xfs_dir2_leafn_order(save_blk
->bp
, drop_blk
->bp
))
1345 xfs_dir2_leafn_moveents(args
, drop_blk
->bp
, 0, save_blk
->bp
, 0,
1346 be16_to_cpu(drop_leaf
->hdr
.count
));
1348 xfs_dir2_leafn_moveents(args
, drop_blk
->bp
, 0, save_blk
->bp
,
1349 be16_to_cpu(save_leaf
->hdr
.count
), be16_to_cpu(drop_leaf
->hdr
.count
));
1350 save_blk
->hashval
= be32_to_cpu(save_leaf
->ents
[be16_to_cpu(save_leaf
->hdr
.count
) - 1].hashval
);
1351 xfs_dir2_leafn_check(args
->dp
, save_blk
->bp
);
1355 * Top-level node form directory addname routine.
1358 xfs_dir2_node_addname(
1359 xfs_da_args_t
*args
) /* operation arguments */
1361 xfs_da_state_blk_t
*blk
; /* leaf block for insert */
1362 int error
; /* error return value */
1363 int rval
; /* sub-return value */
1364 xfs_da_state_t
*state
; /* btree cursor */
1366 trace_xfs_dir2_node_addname(args
);
1369 * Allocate and initialize the state (btree cursor).
1371 state
= xfs_da_state_alloc();
1373 state
->mp
= args
->dp
->i_mount
;
1374 state
->blocksize
= state
->mp
->m_dirblksize
;
1375 state
->node_ents
= state
->mp
->m_dir_node_ents
;
1377 * Look up the name. We're not supposed to find it, but
1378 * this gives us the insertion point.
1380 error
= xfs_da_node_lookup_int(state
, &rval
);
1383 if (rval
!= ENOENT
) {
1387 * Add the data entry to a data block.
1388 * Extravalid is set to a freeblock found by lookup.
1390 rval
= xfs_dir2_node_addname_int(args
,
1391 state
->extravalid
? &state
->extrablk
: NULL
);
1395 blk
= &state
->path
.blk
[state
->path
.active
- 1];
1396 ASSERT(blk
->magic
== XFS_DIR2_LEAFN_MAGIC
);
1398 * Add the new leaf entry.
1400 rval
= xfs_dir2_leafn_add(blk
->bp
, args
, blk
->index
);
1403 * It worked, fix the hash values up the btree.
1405 if (!(args
->op_flags
& XFS_DA_OP_JUSTCHECK
))
1406 xfs_da_fixhashpath(state
, &state
->path
);
1409 * It didn't work, we need to split the leaf block.
1411 if (args
->total
== 0) {
1412 ASSERT(rval
== ENOSPC
);
1416 * Split the leaf block and insert the new entry.
1418 rval
= xfs_da_split(state
);
1421 xfs_da_state_free(state
);
1426 * Add the data entry for a node-format directory name addition.
1427 * The leaf entry is added in xfs_dir2_leafn_add.
1428 * We may enter with a freespace block that the lookup found.
1430 static int /* error */
1431 xfs_dir2_node_addname_int(
1432 xfs_da_args_t
*args
, /* operation arguments */
1433 xfs_da_state_blk_t
*fblk
) /* optional freespace block */
1435 xfs_dir2_data_t
*data
; /* data block structure */
1436 xfs_dir2_db_t dbno
; /* data block number */
1437 xfs_dabuf_t
*dbp
; /* data block buffer */
1438 xfs_dir2_data_entry_t
*dep
; /* data entry pointer */
1439 xfs_inode_t
*dp
; /* incore directory inode */
1440 xfs_dir2_data_unused_t
*dup
; /* data unused entry pointer */
1441 int error
; /* error return value */
1442 xfs_dir2_db_t fbno
; /* freespace block number */
1443 xfs_dabuf_t
*fbp
; /* freespace buffer */
1444 int findex
; /* freespace entry index */
1445 xfs_dir2_free_t
*free
=NULL
; /* freespace block structure */
1446 xfs_dir2_db_t ifbno
; /* initial freespace block no */
1447 xfs_dir2_db_t lastfbno
=0; /* highest freespace block no */
1448 int length
; /* length of the new entry */
1449 int logfree
; /* need to log free entry */
1450 xfs_mount_t
*mp
; /* filesystem mount point */
1451 int needlog
; /* need to log data header */
1452 int needscan
; /* need to rescan data frees */
1453 __be16
*tagp
; /* data entry tag pointer */
1454 xfs_trans_t
*tp
; /* transaction pointer */
1459 length
= xfs_dir2_data_entsize(args
->namelen
);
1461 * If we came in with a freespace block that means that lookup
1462 * found an entry with our hash value. This is the freespace
1463 * block for that data entry.
1468 * Remember initial freespace block number.
1470 ifbno
= fblk
->blkno
;
1472 ASSERT(be32_to_cpu(free
->hdr
.magic
) == XFS_DIR2_FREE_MAGIC
);
1473 findex
= fblk
->index
;
1475 * This means the free entry showed that the data block had
1476 * space for our entry, so we remembered it.
1477 * Use that data block.
1480 ASSERT(findex
< be32_to_cpu(free
->hdr
.nvalid
));
1481 ASSERT(be16_to_cpu(free
->bests
[findex
]) != NULLDATAOFF
);
1482 ASSERT(be16_to_cpu(free
->bests
[findex
]) >= length
);
1483 dbno
= be32_to_cpu(free
->hdr
.firstdb
) + findex
;
1486 * The data block looked at didn't have enough room.
1487 * We'll start at the beginning of the freespace entries.
1495 * Didn't come in with a freespace block, so don't have a data block.
1503 * If we don't have a data block yet, we're going to scan the
1504 * freespace blocks looking for one. Figure out what the
1505 * highest freespace block number is.
1508 xfs_fileoff_t fo
; /* freespace block number */
1510 if ((error
= xfs_bmap_last_offset(tp
, dp
, &fo
, XFS_DATA_FORK
)))
1512 lastfbno
= xfs_dir2_da_to_db(mp
, (xfs_dablk_t
)fo
);
1516 * While we haven't identified a data block, search the freeblock
1517 * data for a good data block. If we find a null freeblock entry,
1518 * indicating a hole in the data blocks, remember that.
1520 while (dbno
== -1) {
1522 * If we don't have a freeblock in hand, get the next one.
1526 * Happens the first time through unless lookup gave
1527 * us a freespace block to start with.
1530 fbno
= XFS_DIR2_FREE_FIRSTDB(mp
);
1532 * If it's ifbno we already looked at it.
1537 * If it's off the end we're done.
1539 if (fbno
>= lastfbno
)
1542 * Read the block. There can be holes in the
1543 * freespace blocks, so this might not succeed.
1544 * This should be really rare, so there's no reason
1547 if ((error
= xfs_da_read_buf(tp
, dp
,
1548 xfs_dir2_db_to_da(mp
, fbno
), -2, &fbp
,
1552 if (unlikely(fbp
== NULL
)) {
1556 ASSERT(be32_to_cpu(free
->hdr
.magic
) == XFS_DIR2_FREE_MAGIC
);
1560 * Look at the current free entry. Is it good enough?
1562 if (be16_to_cpu(free
->bests
[findex
]) != NULLDATAOFF
&&
1563 be16_to_cpu(free
->bests
[findex
]) >= length
)
1564 dbno
= be32_to_cpu(free
->hdr
.firstdb
) + findex
;
1567 * Are we done with the freeblock?
1569 if (++findex
== be32_to_cpu(free
->hdr
.nvalid
)) {
1573 xfs_da_brelse(tp
, fbp
);
1575 if (fblk
&& fblk
->bp
)
1581 * If we don't have a data block, we need to allocate one and make
1582 * the freespace entries refer to it.
1584 if (unlikely(dbno
== -1)) {
1586 * Not allowed to allocate, return failure.
1588 if ((args
->op_flags
& XFS_DA_OP_JUSTCHECK
) ||
1591 * Drop the freespace buffer unless it came from our
1594 if ((fblk
== NULL
|| fblk
->bp
== NULL
) && fbp
!= NULL
)
1595 xfs_da_buf_done(fbp
);
1596 return XFS_ERROR(ENOSPC
);
1599 * Allocate and initialize the new data block.
1601 if (unlikely((error
= xfs_dir2_grow_inode(args
,
1602 XFS_DIR2_DATA_SPACE
,
1604 (error
= xfs_dir2_data_init(args
, dbno
, &dbp
)))) {
1606 * Drop the freespace buffer unless it came from our
1609 if ((fblk
== NULL
|| fblk
->bp
== NULL
) && fbp
!= NULL
)
1610 xfs_da_buf_done(fbp
);
1614 * If (somehow) we have a freespace block, get rid of it.
1617 xfs_da_brelse(tp
, fbp
);
1618 if (fblk
&& fblk
->bp
)
1622 * Get the freespace block corresponding to the data block
1623 * that was just allocated.
1625 fbno
= xfs_dir2_db_to_fdb(mp
, dbno
);
1626 if (unlikely(error
= xfs_da_read_buf(tp
, dp
,
1627 xfs_dir2_db_to_da(mp
, fbno
), -2, &fbp
,
1629 xfs_da_buf_done(dbp
);
1633 * If there wasn't a freespace block, the read will
1634 * return a NULL fbp. Allocate and initialize a new one.
1637 if ((error
= xfs_dir2_grow_inode(args
, XFS_DIR2_FREE_SPACE
,
1642 if (unlikely(xfs_dir2_db_to_fdb(mp
, dbno
) != fbno
)) {
1644 "%s: dir ino " "%llu needed freesp block %lld for\n"
1645 " data block %lld, got %lld ifbno %llu lastfbno %d",
1646 __func__
, (unsigned long long)dp
->i_ino
,
1647 (long long)xfs_dir2_db_to_fdb(mp
, dbno
),
1648 (long long)dbno
, (long long)fbno
,
1649 (unsigned long long)ifbno
, lastfbno
);
1652 " fblk 0x%p blkno %llu index %d magic 0x%x",
1654 (unsigned long long)fblk
->blkno
,
1658 xfs_alert(mp
, " ... fblk is NULL");
1660 XFS_ERROR_REPORT("xfs_dir2_node_addname_int",
1661 XFS_ERRLEVEL_LOW
, mp
);
1662 return XFS_ERROR(EFSCORRUPTED
);
1666 * Get a buffer for the new block.
1668 if ((error
= xfs_da_get_buf(tp
, dp
,
1669 xfs_dir2_db_to_da(mp
, fbno
),
1670 -1, &fbp
, XFS_DATA_FORK
))) {
1673 ASSERT(fbp
!= NULL
);
1676 * Initialize the new block to be empty, and remember
1677 * its first slot as our empty slot.
1680 free
->hdr
.magic
= cpu_to_be32(XFS_DIR2_FREE_MAGIC
);
1681 free
->hdr
.firstdb
= cpu_to_be32(
1682 (fbno
- XFS_DIR2_FREE_FIRSTDB(mp
)) *
1683 XFS_DIR2_MAX_FREE_BESTS(mp
));
1684 free
->hdr
.nvalid
= 0;
1685 free
->hdr
.nused
= 0;
1688 ASSERT(be32_to_cpu(free
->hdr
.magic
) == XFS_DIR2_FREE_MAGIC
);
1692 * Set the freespace block index from the data block number.
1694 findex
= xfs_dir2_db_to_fdindex(mp
, dbno
);
1696 * If it's after the end of the current entries in the
1697 * freespace block, extend that table.
1699 if (findex
>= be32_to_cpu(free
->hdr
.nvalid
)) {
1700 ASSERT(findex
< XFS_DIR2_MAX_FREE_BESTS(mp
));
1701 free
->hdr
.nvalid
= cpu_to_be32(findex
+ 1);
1703 * Tag new entry so nused will go up.
1705 free
->bests
[findex
] = cpu_to_be16(NULLDATAOFF
);
1708 * If this entry was for an empty data block
1709 * (this should always be true) then update the header.
1711 if (be16_to_cpu(free
->bests
[findex
]) == NULLDATAOFF
) {
1712 be32_add_cpu(&free
->hdr
.nused
, 1);
1713 xfs_dir2_free_log_header(tp
, fbp
);
1716 * Update the real value in the table.
1717 * We haven't allocated the data entry yet so this will
1721 free
->bests
[findex
] = data
->hdr
.bestfree
[0].length
;
1725 * We had a data block so we don't have to make a new one.
1729 * If just checking, we succeeded.
1731 if (args
->op_flags
& XFS_DA_OP_JUSTCHECK
) {
1732 if ((fblk
== NULL
|| fblk
->bp
== NULL
) && fbp
!= NULL
)
1733 xfs_da_buf_done(fbp
);
1737 * Read the data block in.
1740 error
= xfs_da_read_buf(tp
, dp
, xfs_dir2_db_to_da(mp
, dbno
),
1741 -1, &dbp
, XFS_DATA_FORK
))) {
1742 if ((fblk
== NULL
|| fblk
->bp
== NULL
) && fbp
!= NULL
)
1743 xfs_da_buf_done(fbp
);
1749 ASSERT(be16_to_cpu(data
->hdr
.bestfree
[0].length
) >= length
);
1751 * Point to the existing unused space.
1753 dup
= (xfs_dir2_data_unused_t
*)
1754 ((char *)data
+ be16_to_cpu(data
->hdr
.bestfree
[0].offset
));
1755 needscan
= needlog
= 0;
1757 * Mark the first part of the unused space, inuse for us.
1759 xfs_dir2_data_use_free(tp
, dbp
, dup
,
1760 (xfs_dir2_data_aoff_t
)((char *)dup
- (char *)data
), length
,
1761 &needlog
, &needscan
);
1763 * Fill in the new entry and log it.
1765 dep
= (xfs_dir2_data_entry_t
*)dup
;
1766 dep
->inumber
= cpu_to_be64(args
->inumber
);
1767 dep
->namelen
= args
->namelen
;
1768 memcpy(dep
->name
, args
->name
, dep
->namelen
);
1769 tagp
= xfs_dir2_data_entry_tag_p(dep
);
1770 *tagp
= cpu_to_be16((char *)dep
- (char *)data
);
1771 xfs_dir2_data_log_entry(tp
, dbp
, dep
);
1773 * Rescan the block for bestfree if needed.
1776 xfs_dir2_data_freescan(mp
, data
, &needlog
);
1778 * Log the data block header if needed.
1781 xfs_dir2_data_log_header(tp
, dbp
);
1783 * If the freespace entry is now wrong, update it.
1785 if (be16_to_cpu(free
->bests
[findex
]) != be16_to_cpu(data
->hdr
.bestfree
[0].length
)) {
1786 free
->bests
[findex
] = data
->hdr
.bestfree
[0].length
;
1790 * Log the freespace entry if needed.
1793 xfs_dir2_free_log_bests(tp
, fbp
, findex
, findex
);
1795 * If the caller didn't hand us the freespace block, drop it.
1797 if ((fblk
== NULL
|| fblk
->bp
== NULL
) && fbp
!= NULL
)
1798 xfs_da_buf_done(fbp
);
1800 * Return the data block and offset in args, then drop the data block.
1802 args
->blkno
= (xfs_dablk_t
)dbno
;
1803 args
->index
= be16_to_cpu(*tagp
);
1804 xfs_da_buf_done(dbp
);
1809 * Lookup an entry in a node-format directory.
1810 * All the real work happens in xfs_da_node_lookup_int.
1811 * The only real output is the inode number of the entry.
1814 xfs_dir2_node_lookup(
1815 xfs_da_args_t
*args
) /* operation arguments */
1817 int error
; /* error return value */
1818 int i
; /* btree level */
1819 int rval
; /* operation return value */
1820 xfs_da_state_t
*state
; /* btree cursor */
1822 trace_xfs_dir2_node_lookup(args
);
1825 * Allocate and initialize the btree cursor.
1827 state
= xfs_da_state_alloc();
1829 state
->mp
= args
->dp
->i_mount
;
1830 state
->blocksize
= state
->mp
->m_dirblksize
;
1831 state
->node_ents
= state
->mp
->m_dir_node_ents
;
1833 * Fill in the path to the entry in the cursor.
1835 error
= xfs_da_node_lookup_int(state
, &rval
);
1838 else if (rval
== ENOENT
&& args
->cmpresult
== XFS_CMP_CASE
) {
1839 /* If a CI match, dup the actual name and return EEXIST */
1840 xfs_dir2_data_entry_t
*dep
;
1842 dep
= (xfs_dir2_data_entry_t
*)((char *)state
->extrablk
.bp
->
1843 data
+ state
->extrablk
.index
);
1844 rval
= xfs_dir_cilookup_result(args
, dep
->name
, dep
->namelen
);
1847 * Release the btree blocks and leaf block.
1849 for (i
= 0; i
< state
->path
.active
; i
++) {
1850 xfs_da_brelse(args
->trans
, state
->path
.blk
[i
].bp
);
1851 state
->path
.blk
[i
].bp
= NULL
;
1854 * Release the data block if we have it.
1856 if (state
->extravalid
&& state
->extrablk
.bp
) {
1857 xfs_da_brelse(args
->trans
, state
->extrablk
.bp
);
1858 state
->extrablk
.bp
= NULL
;
1860 xfs_da_state_free(state
);
1865 * Remove an entry from a node-format directory.
1868 xfs_dir2_node_removename(
1869 xfs_da_args_t
*args
) /* operation arguments */
1871 xfs_da_state_blk_t
*blk
; /* leaf block */
1872 int error
; /* error return value */
1873 int rval
; /* operation return value */
1874 xfs_da_state_t
*state
; /* btree cursor */
1876 trace_xfs_dir2_node_removename(args
);
1879 * Allocate and initialize the btree cursor.
1881 state
= xfs_da_state_alloc();
1883 state
->mp
= args
->dp
->i_mount
;
1884 state
->blocksize
= state
->mp
->m_dirblksize
;
1885 state
->node_ents
= state
->mp
->m_dir_node_ents
;
1887 * Look up the entry we're deleting, set up the cursor.
1889 error
= xfs_da_node_lookup_int(state
, &rval
);
1893 * Didn't find it, upper layer screwed up.
1895 if (rval
!= EEXIST
) {
1896 xfs_da_state_free(state
);
1899 blk
= &state
->path
.blk
[state
->path
.active
- 1];
1900 ASSERT(blk
->magic
== XFS_DIR2_LEAFN_MAGIC
);
1901 ASSERT(state
->extravalid
);
1903 * Remove the leaf and data entries.
1904 * Extrablk refers to the data block.
1906 error
= xfs_dir2_leafn_remove(args
, blk
->bp
, blk
->index
,
1907 &state
->extrablk
, &rval
);
1911 * Fix the hash values up the btree.
1913 xfs_da_fixhashpath(state
, &state
->path
);
1915 * If we need to join leaf blocks, do it.
1917 if (rval
&& state
->path
.active
> 1)
1918 error
= xfs_da_join(state
);
1920 * If no errors so far, try conversion to leaf format.
1923 error
= xfs_dir2_node_to_leaf(state
);
1924 xfs_da_state_free(state
);
1929 * Replace an entry's inode number in a node-format directory.
1932 xfs_dir2_node_replace(
1933 xfs_da_args_t
*args
) /* operation arguments */
1935 xfs_da_state_blk_t
*blk
; /* leaf block */
1936 xfs_dir2_data_t
*data
; /* data block structure */
1937 xfs_dir2_data_entry_t
*dep
; /* data entry changed */
1938 int error
; /* error return value */
1939 int i
; /* btree level */
1940 xfs_ino_t inum
; /* new inode number */
1941 xfs_dir2_leaf_t
*leaf
; /* leaf structure */
1942 xfs_dir2_leaf_entry_t
*lep
; /* leaf entry being changed */
1943 int rval
; /* internal return value */
1944 xfs_da_state_t
*state
; /* btree cursor */
1946 trace_xfs_dir2_node_replace(args
);
1949 * Allocate and initialize the btree cursor.
1951 state
= xfs_da_state_alloc();
1953 state
->mp
= args
->dp
->i_mount
;
1954 state
->blocksize
= state
->mp
->m_dirblksize
;
1955 state
->node_ents
= state
->mp
->m_dir_node_ents
;
1956 inum
= args
->inumber
;
1958 * Lookup the entry to change in the btree.
1960 error
= xfs_da_node_lookup_int(state
, &rval
);
1965 * It should be found, since the vnodeops layer has looked it up
1966 * and locked it. But paranoia is good.
1968 if (rval
== EEXIST
) {
1970 * Find the leaf entry.
1972 blk
= &state
->path
.blk
[state
->path
.active
- 1];
1973 ASSERT(blk
->magic
== XFS_DIR2_LEAFN_MAGIC
);
1974 leaf
= blk
->bp
->data
;
1975 lep
= &leaf
->ents
[blk
->index
];
1976 ASSERT(state
->extravalid
);
1978 * Point to the data entry.
1980 data
= state
->extrablk
.bp
->data
;
1981 ASSERT(be32_to_cpu(data
->hdr
.magic
) == XFS_DIR2_DATA_MAGIC
);
1982 dep
= (xfs_dir2_data_entry_t
*)
1984 xfs_dir2_dataptr_to_off(state
->mp
, be32_to_cpu(lep
->address
)));
1985 ASSERT(inum
!= be64_to_cpu(dep
->inumber
));
1987 * Fill in the new inode number and log the entry.
1989 dep
->inumber
= cpu_to_be64(inum
);
1990 xfs_dir2_data_log_entry(args
->trans
, state
->extrablk
.bp
, dep
);
1994 * Didn't find it, and we're holding a data block. Drop it.
1996 else if (state
->extravalid
) {
1997 xfs_da_brelse(args
->trans
, state
->extrablk
.bp
);
1998 state
->extrablk
.bp
= NULL
;
2001 * Release all the buffers in the cursor.
2003 for (i
= 0; i
< state
->path
.active
; i
++) {
2004 xfs_da_brelse(args
->trans
, state
->path
.blk
[i
].bp
);
2005 state
->path
.blk
[i
].bp
= NULL
;
2007 xfs_da_state_free(state
);
2012 * Trim off a trailing empty freespace block.
2013 * Return (in rvalp) 1 if we did it, 0 if not.
2016 xfs_dir2_node_trim_free(
2017 xfs_da_args_t
*args
, /* operation arguments */
2018 xfs_fileoff_t fo
, /* free block number */
2019 int *rvalp
) /* out: did something */
2021 xfs_dabuf_t
*bp
; /* freespace buffer */
2022 xfs_inode_t
*dp
; /* incore directory inode */
2023 int error
; /* error return code */
2024 xfs_dir2_free_t
*free
; /* freespace structure */
2025 xfs_mount_t
*mp
; /* filesystem mount point */
2026 xfs_trans_t
*tp
; /* transaction pointer */
2032 * Read the freespace block.
2034 if (unlikely(error
= xfs_da_read_buf(tp
, dp
, (xfs_dablk_t
)fo
, -2, &bp
,
2040 * There can be holes in freespace. If fo is a hole, there's
2047 ASSERT(be32_to_cpu(free
->hdr
.magic
) == XFS_DIR2_FREE_MAGIC
);
2049 * If there are used entries, there's nothing to do.
2051 if (be32_to_cpu(free
->hdr
.nused
) > 0) {
2052 xfs_da_brelse(tp
, bp
);
2057 * Blow the block away.
2060 xfs_dir2_shrink_inode(args
, xfs_dir2_da_to_db(mp
, (xfs_dablk_t
)fo
),
2063 * Can't fail with ENOSPC since that only happens with no
2064 * space reservation, when breaking up an extent into two
2065 * pieces. This is the last block of an extent.
2067 ASSERT(error
!= ENOSPC
);
2068 xfs_da_brelse(tp
, bp
);
2072 * Return that we succeeded.