]>
git.saurik.com Git - apple/libc.git/blob - db/hash/FreeBSD/hash_bigkey.c
2 * Copyright (c) 1990, 1993, 1994
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 #if defined(LIBC_SCCS) && !defined(lint)
34 static char sccsid
[] = "@(#)hash_bigkey.c 8.3 (Berkeley) 5/31/94";
35 #endif /* LIBC_SCCS and not lint */
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD: src/lib/libc/db/hash/hash_bigkey.c,v 1.10 2009/03/28 06:47:05 delphij Exp $");
42 * Big key/data handling for the hashing package.
57 #include <sys/param.h>
71 #include "hash_extern.h"
73 static int collect_key(HTAB
*, BUFHEAD
*, int, DBT
*, int);
74 static int collect_data(HTAB
*, BUFHEAD
*, int, int);
79 * You need to do an insert and the key/data pair is too big
86 __big_insert(HTAB
*hashp
, BUFHEAD
*bufp
, const DBT
*key
, const DBT
*val
)
90 unsigned int val_size
;
91 u_int16_t space
, move_bytes
, off
;
92 char *cp
, *key_data
, *val_data
;
94 cp
= bufp
->page
; /* Character pointer of p. */
97 key_data
= (char *)key
->data
;
99 val_data
= (char *)val
->data
;
100 val_size
= val
->size
;
102 /* First move the Key */
103 for (space
= FREESPACE(p
) - BIGOVERHEAD
; key_size
;
104 space
= FREESPACE(p
) - BIGOVERHEAD
) {
105 move_bytes
= MIN(space
, key_size
);
106 off
= OFFSET(p
) - move_bytes
;
107 memmove(cp
+ off
, key_data
, move_bytes
);
108 key_size
-= move_bytes
;
109 key_data
+= move_bytes
;
113 FREESPACE(p
) = off
- PAGE_META(n
);
116 bufp
= __add_ovflpage(hashp
, bufp
);
121 space
= FREESPACE(p
);
123 move_bytes
= MIN(space
, val_size
);
125 * If the data would fit exactly in the
126 * remaining space, we must overflow it to the
127 * next page; otherwise the invariant that the
128 * data must end on a page with FREESPACE
129 * non-zero would fail.
131 if (space
== val_size
&& val_size
== val
->size
)
133 off
= OFFSET(p
) - move_bytes
;
134 memmove(cp
+ off
, val_data
, move_bytes
);
135 val_data
+= move_bytes
;
136 val_size
-= move_bytes
;
138 p
[n
- 2] = FULL_KEY_DATA
;
139 FREESPACE(p
) = FREESPACE(p
) - move_bytes
;
146 p
= (u_int16_t
*)bufp
->page
;
148 bufp
->flags
|= BUF_MOD
;
151 /* Now move the data */
152 for (space
= FREESPACE(p
) - BIGOVERHEAD
; val_size
;
153 space
= FREESPACE(p
) - BIGOVERHEAD
) {
154 move_bytes
= MIN(space
, val_size
);
156 * Here's the hack to make sure that if the data ends on the
157 * same page as the key ends, FREESPACE is at least one.
159 if (space
== val_size
&& val_size
== val
->size
)
161 off
= OFFSET(p
) - move_bytes
;
162 memmove(cp
+ off
, val_data
, move_bytes
);
163 val_size
-= move_bytes
;
164 val_data
+= move_bytes
;
168 FREESPACE(p
) = off
- PAGE_META(n
);
172 bufp
= __add_ovflpage(hashp
, bufp
);
178 p
[n
] = FULL_KEY_DATA
;
179 bufp
->flags
|= BUF_MOD
;
185 * Called when bufp's page contains a partial key (index should be 1)
187 * All pages in the big key/data pair except bufp are freed. We cannot
188 * free bufp because the page pointing to it is lost and we can't get rid
196 __big_delete(HTAB
*hashp
, BUFHEAD
*bufp
)
198 BUFHEAD
*last_bfp
, *rbufp
;
199 u_int16_t
*bp
, pageno
;
204 bp
= (u_int16_t
*)bufp
->page
;
208 while (!key_done
|| (bp
[2] != FULL_KEY_DATA
)) {
209 if (bp
[2] == FULL_KEY
|| bp
[2] == FULL_KEY_DATA
)
213 * If there is freespace left on a FULL_KEY_DATA page, then
214 * the data is short and fits entirely on this page, and this
217 if (bp
[2] == FULL_KEY_DATA
&& FREESPACE(bp
))
219 pageno
= bp
[bp
[0] - 1];
220 rbufp
->flags
|= BUF_MOD
;
221 rbufp
= __get_buf(hashp
, pageno
, rbufp
, 0);
223 __free_ovflpage(hashp
, last_bfp
);
226 return (-1); /* Error. */
227 bp
= (u_int16_t
*)rbufp
->page
;
231 * If we get here then rbufp points to the last page of the big
232 * key/data pair. Bufp points to the first one -- it should now be
233 * empty pointing to the next page after this pair. Can't free it
234 * because we don't have the page pointing to it.
237 /* This is information from the last page of the pair. */
241 /* Now, bp is the first page of the pair. */
242 bp
= (u_int16_t
*)bufp
->page
;
244 /* There is an overflow page. */
247 bufp
->ovfl
= rbufp
->ovfl
;
249 /* This is the last page. */
253 FREESPACE(bp
) = hashp
->BSIZE
- PAGE_META(n
);
254 OFFSET(bp
) = hashp
->BSIZE
;
256 bufp
->flags
|= BUF_MOD
;
258 __free_ovflpage(hashp
, rbufp
);
259 if (last_bfp
&& last_bfp
!= rbufp
)
260 __free_ovflpage(hashp
, last_bfp
);
268 * -1 = get next overflow page
269 * -2 means key not found and this is big key/data
273 __find_bigpair(HTAB
*hashp
, BUFHEAD
*bufp
, int ndx
, char *key
, int size
)
281 bp
= (u_int16_t
*)bufp
->page
;
286 for (bytes
= hashp
->BSIZE
- bp
[ndx
];
287 bytes
<= size
&& bp
[ndx
+ 1] == PARTIAL_KEY
;
288 bytes
= hashp
->BSIZE
- bp
[ndx
]) {
289 if (memcmp(p
+ bp
[ndx
], kkey
, bytes
))
293 bufp
= __get_buf(hashp
, bp
[ndx
+ 2], bufp
, 0);
301 if (bytes
!= ksize
|| memcmp(p
+ bp
[ndx
], kkey
, bytes
)) {
302 #ifdef HASH_STATISTICS
311 * Given the buffer pointer of the first overflow page of a big pair,
312 * find the end of the big pair
314 * This will set bpp to the buffer header of the last page of the big pair.
315 * It will return the pageno of the overflow page following the last page
316 * of the pair; 0 if there isn't any (i.e. big pair is the last key in the
320 __find_last_page(HTAB
*hashp
, BUFHEAD
**bpp
)
323 u_int16_t
*bp
, pageno
;
327 bp
= (u_int16_t
*)bufp
->page
;
332 * This is the last page if: the tag is FULL_KEY_DATA and
333 * either only 2 entries OVFLPAGE marker is explicit there
334 * is freespace on the page.
336 if (bp
[2] == FULL_KEY_DATA
&&
337 ((n
== 2) || (bp
[n
] == OVFLPAGE
) || (FREESPACE(bp
))))
341 bufp
= __get_buf(hashp
, pageno
, bufp
, 0);
343 return (0); /* Need to indicate an error! */
344 bp
= (u_int16_t
*)bufp
->page
;
355 * Return the data for the key/data pair that begins on this page at this
356 * index (index should always be 1).
359 __big_return(HTAB
*hashp
, BUFHEAD
*bufp
, int ndx
, DBT
*val
, int set_current
)
362 u_int16_t
*bp
, len
, off
, save_addr
;
365 bp
= (u_int16_t
*)bufp
->page
;
366 while (bp
[ndx
+ 1] == PARTIAL_KEY
) {
367 bufp
= __get_buf(hashp
, bp
[bp
[0] - 1], bufp
, 0);
370 bp
= (u_int16_t
*)bufp
->page
;
374 if (bp
[ndx
+ 1] == FULL_KEY
) {
375 bufp
= __get_buf(hashp
, bp
[bp
[0] - 1], bufp
, 0);
378 bp
= (u_int16_t
*)bufp
->page
;
380 save_addr
= save_p
->addr
;
384 if (!FREESPACE(bp
)) {
386 * This is a hack. We can't distinguish between
387 * FULL_KEY_DATA that contains complete data or
388 * incomplete data, so we require that if the data
389 * is complete, there is at least 1 byte of free
395 save_addr
= bufp
->addr
;
396 bufp
= __get_buf(hashp
, bp
[bp
[0] - 1], bufp
, 0);
399 bp
= (u_int16_t
*)bufp
->page
;
401 /* The data is all on one page. */
404 val
->data
= (u_char
*)tp
+ off
;
405 val
->size
= bp
[1] - off
;
407 if (bp
[0] == 2) { /* No more buckets in
413 hashp
->cpage
= __get_buf(hashp
,
414 bp
[bp
[0] - 1], bufp
, 0);
419 hashp
->cpage
->page
)[0]) {
428 val
->size
= (size_t)collect_data(hashp
, bufp
, (int)len
, set_current
);
429 if (val
->size
== (size_t)-1)
431 if (save_p
->addr
!= save_addr
) {
432 /* We are pretty short on buffers. */
433 errno
= EINVAL
; /* OUT OF BUFFERS */
436 memmove(hashp
->tmp_buf
, (save_p
->page
) + off
, len
);
437 val
->data
= (u_char
*)hashp
->tmp_buf
;
441 * Count how big the total datasize is by recursing through the pages. Then
442 * allocate a buffer and copy the data as you recurse up.
445 collect_data(HTAB
*hashp
, BUFHEAD
*bufp
, int len
, int set
)
455 mylen
= hashp
->BSIZE
- bp
[1];
456 save_addr
= bufp
->addr
;
458 if (bp
[2] == FULL_KEY_DATA
) { /* End of Data */
459 totlen
= len
+ mylen
;
461 free(hashp
->tmp_buf
);
462 if ((hashp
->tmp_buf
= (char *)malloc(totlen
)) == NULL
)
466 if (bp
[0] == 2) { /* No more buckets in chain */
471 __get_buf(hashp
, bp
[bp
[0] - 1], bufp
, 0);
474 else if (!((u_int16_t
*)hashp
->cpage
->page
)[0]) {
481 xbp
= __get_buf(hashp
, bp
[bp
[0] - 1], bufp
, 0);
482 if (!xbp
|| ((totlen
=
483 collect_data(hashp
, xbp
, len
+ mylen
, set
)) < 1))
486 if (bufp
->addr
!= save_addr
) {
487 errno
= EINVAL
; /* Out of buffers. */
490 memmove(&hashp
->tmp_buf
[len
], (bufp
->page
) + bp
[1], mylen
);
495 * Fill in the key and data for this big pair.
498 __big_keydata(HTAB
*hashp
, BUFHEAD
*bufp
, DBT
*key
, DBT
*val
, int set
)
500 key
->size
= (size_t)collect_key(hashp
, bufp
, 0, val
, set
);
501 if (key
->size
== (size_t)-1)
503 key
->data
= (u_char
*)hashp
->tmp_key
;
508 * Count how big the total key size is by recursing through the pages. Then
509 * collect the data, allocate a buffer and copy the key as you recurse up.
512 collect_key(HTAB
*hashp
, BUFHEAD
*bufp
, int len
, DBT
*val
, int set
)
517 u_int16_t
*bp
, save_addr
;
521 mylen
= hashp
->BSIZE
- bp
[1];
523 save_addr
= bufp
->addr
;
524 totlen
= len
+ mylen
;
525 if (bp
[2] == FULL_KEY
|| bp
[2] == FULL_KEY_DATA
) { /* End of Key. */
526 if (hashp
->tmp_key
!= NULL
)
527 free(hashp
->tmp_key
);
528 if ((hashp
->tmp_key
= (char *)malloc(totlen
)) == NULL
)
530 if (__big_return(hashp
, bufp
, 1, val
, set
))
533 xbp
= __get_buf(hashp
, bp
[bp
[0] - 1], bufp
, 0);
534 if (!xbp
|| ((totlen
=
535 collect_key(hashp
, xbp
, totlen
, val
, set
)) < 1))
538 if (bufp
->addr
!= save_addr
) {
539 errno
= EINVAL
; /* MIS -- OUT OF BUFFERS */
542 memmove(&hashp
->tmp_key
[len
], (bufp
->page
) + bp
[1], mylen
);
552 __big_split(HTAB
*hashp
,
553 BUFHEAD
*op
, /* Pointer to where to put keys that go in old bucket */
554 BUFHEAD
*np
, /* Pointer to new bucket page */
555 BUFHEAD
*big_keyp
, /* Pointer to first page containing the big key/data */
556 int addr
, /* Address of big_keyp */
557 u_int32_t obucket
, /* Old Bucket */
563 u_int16_t free_space
, n
, off
, *tp
;
567 /* Now figure out where the big key/data goes */
568 if (__big_keydata(hashp
, big_keyp
, &key
, &val
, 0))
570 change
= (__call_hash(hashp
, key
.data
, key
.size
) != obucket
);
572 if ( (ret
->next_addr
= __find_last_page(hashp
, &big_keyp
)) ) {
574 __get_buf(hashp
, ret
->next_addr
, big_keyp
, 0)))
579 /* Now make one of np/op point to the big key/data pair */
581 assert(np
->ovfl
== NULL
);
588 tmpp
->flags
|= BUF_MOD
;
590 (void)fprintf(stderr
,
591 "BIG_SPLIT: %d->ovfl was %d is now %d\n", tmpp
->addr
,
592 (tmpp
->ovfl
? tmpp
->ovfl
->addr
: 0), (bp
? bp
->addr
: 0));
594 tmpp
->ovfl
= bp
; /* one of op/np point to big_keyp */
595 tp
= (u_int16_t
*)tmpp
->page
;
597 assert(FREESPACE(tp
) >= OVFLSIZE
);
601 free_space
= FREESPACE(tp
);
602 tp
[++n
] = (u_int16_t
)addr
;
606 FREESPACE(tp
) = free_space
- OVFLSIZE
;
609 * Finally, set the new and old return values. BIG_KEYP contains a
610 * pointer to the last page of the big key_data pair. Make sure that
611 * big_keyp has no following page (2 elements) or create an empty
618 tp
= (u_int16_t
*)big_keyp
->page
;
619 big_keyp
->flags
|= BUF_MOD
;
622 * There may be either one or two offsets on this page. If
623 * there is one, then the overflow page is linked on normally
624 * and tp[4] is OVFLPAGE. If there are two, tp[4] contains
625 * the second offset and needs to get stuffed in after the
626 * next overflow page is added.
629 free_space
= FREESPACE(tp
);
632 FREESPACE(tp
) = free_space
+ OVFLSIZE
;
634 tmpp
= __add_ovflpage(hashp
, big_keyp
);