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 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 #if defined(LIBC_SCCS) && !defined(lint)
38 static char sccsid
[] = "@(#)hash_bigkey.c 8.3 (Berkeley) 5/31/94";
39 #endif /* LIBC_SCCS and not lint */
44 * Big key/data handling for the hashing package.
59 #include <sys/param.h>
75 static int collect_key
__P((HTAB
*, BUFHEAD
*, int, DBT
*, int));
76 static int collect_data
__P((HTAB
*, BUFHEAD
*, int, int));
81 * You need to do an insert and the key/data pair is too big
88 __big_insert(hashp
, bufp
, key
, val
)
93 register u_int16_t
*p
;
94 int key_size
, n
, val_size
;
95 u_int16_t space
, move_bytes
, off
;
96 char *cp
, *key_data
, *val_data
;
98 cp
= bufp
->page
; /* Character pointer of p. */
101 key_data
= (char *)key
->data
;
102 key_size
= key
->size
;
103 val_data
= (char *)val
->data
;
104 val_size
= val
->size
;
106 /* First move the Key */
107 for (space
= FREESPACE(p
) - BIGOVERHEAD
; key_size
;
108 space
= FREESPACE(p
) - BIGOVERHEAD
) {
109 move_bytes
= MIN(space
, key_size
);
110 off
= OFFSET(p
) - move_bytes
;
111 memmove(cp
+ off
, key_data
, move_bytes
);
112 key_size
-= move_bytes
;
113 key_data
+= move_bytes
;
117 FREESPACE(p
) = off
- PAGE_META(n
);
120 bufp
= __add_ovflpage(hashp
, bufp
);
126 move_bytes
= MIN(FREESPACE(p
), val_size
);
127 off
= OFFSET(p
) - move_bytes
;
129 memmove(cp
+ off
, val_data
, move_bytes
);
130 val_data
+= move_bytes
;
131 val_size
-= move_bytes
;
132 p
[n
- 2] = FULL_KEY_DATA
;
133 FREESPACE(p
) = FREESPACE(p
) - move_bytes
;
138 p
= (u_int16_t
*)bufp
->page
;
140 bufp
->flags
|= BUF_MOD
;
143 /* Now move the data */
144 for (space
= FREESPACE(p
) - BIGOVERHEAD
; val_size
;
145 space
= FREESPACE(p
) - BIGOVERHEAD
) {
146 move_bytes
= MIN(space
, val_size
);
148 * Here's the hack to make sure that if the data ends on the
149 * same page as the key ends, FREESPACE is at least one.
151 if ((int) space
== val_size
&& (size_t) val_size
== val
->size
)
153 off
= OFFSET(p
) - move_bytes
;
154 memmove(cp
+ off
, val_data
, move_bytes
);
155 val_size
-= move_bytes
;
156 val_data
+= move_bytes
;
160 FREESPACE(p
) = off
- PAGE_META(n
);
164 bufp
= __add_ovflpage(hashp
, bufp
);
170 p
[n
] = FULL_KEY_DATA
;
171 bufp
->flags
|= BUF_MOD
;
177 * Called when bufp's page contains a partial key (index should be 1)
179 * All pages in the big key/data pair except bufp are freed. We cannot
180 * free bufp because the page pointing to it is lost and we can't get rid
188 __big_delete(hashp
, bufp
)
192 register BUFHEAD
*last_bfp
, *rbufp
;
193 u_int16_t
*bp
, pageno
;
198 bp
= (u_int16_t
*)bufp
->page
;
202 while (!key_done
|| (bp
[2] != FULL_KEY_DATA
)) {
203 if (bp
[2] == FULL_KEY
|| bp
[2] == FULL_KEY_DATA
)
207 * If there is freespace left on a FULL_KEY_DATA page, then
208 * the data is short and fits entirely on this page, and this
211 if (bp
[2] == FULL_KEY_DATA
&& FREESPACE(bp
))
213 pageno
= bp
[bp
[0] - 1];
214 rbufp
->flags
|= BUF_MOD
;
215 rbufp
= __get_buf(hashp
, pageno
, rbufp
, 0);
217 __free_ovflpage(hashp
, last_bfp
);
220 return (-1); /* Error. */
221 bp
= (u_int16_t
*)rbufp
->page
;
225 * If we get here then rbufp points to the last page of the big
226 * key/data pair. Bufp points to the first one -- it should now be
227 * empty pointing to the next page after this pair. Can't free it
228 * because we don't have the page pointing to it.
231 /* This is information from the last page of the pair. */
235 /* Now, bp is the first page of the pair. */
236 bp
= (u_int16_t
*)bufp
->page
;
238 /* There is an overflow page. */
241 bufp
->ovfl
= rbufp
->ovfl
;
243 /* This is the last page. */
247 FREESPACE(bp
) = hashp
->BSIZE
- PAGE_META(n
);
248 OFFSET(bp
) = hashp
->BSIZE
- 1;
250 bufp
->flags
|= BUF_MOD
;
252 __free_ovflpage(hashp
, rbufp
);
253 if (last_bfp
&& last_bfp
!= rbufp
)
254 __free_ovflpage(hashp
, last_bfp
);
262 * -1 = get next overflow page
263 * -2 means key not found and this is big key/data
267 __find_bigpair(hashp
, bufp
, ndx
, key
, size
)
274 register u_int16_t
*bp
;
280 bp
= (u_int16_t
*)bufp
->page
;
285 for (bytes
= hashp
->BSIZE
- bp
[ndx
];
286 bytes
<= size
&& bp
[ndx
+ 1] == PARTIAL_KEY
;
287 bytes
= hashp
->BSIZE
- bp
[ndx
]) {
288 if (memcmp(p
+ bp
[ndx
], kkey
, bytes
))
292 bufp
= __get_buf(hashp
, bp
[ndx
+ 2], bufp
, 0);
300 if (bytes
!= ksize
|| memcmp(p
+ bp
[ndx
], kkey
, bytes
)) {
301 #ifdef HASH_STATISTICS
310 * Given the buffer pointer of the first overflow page of a big pair,
311 * find the end of the big pair
313 * This will set bpp to the buffer header of the last page of the big pair.
314 * It will return the pageno of the overflow page following the last page
315 * of the pair; 0 if there isn't any (i.e. big pair is the last key in the
319 __find_last_page(hashp
, bpp
)
324 u_int16_t
*bp
, pageno
;
328 bp
= (u_int16_t
*)bufp
->page
;
333 * This is the last page if: the tag is FULL_KEY_DATA and
334 * either only 2 entries OVFLPAGE marker is explicit there
335 * is freespace on the page.
337 if (bp
[2] == FULL_KEY_DATA
&&
338 ((n
== 2) || (bp
[n
] == OVFLPAGE
) || (FREESPACE(bp
))))
342 bufp
= __get_buf(hashp
, pageno
, bufp
, 0);
344 return (0); /* Need to indicate an error! */
345 bp
= (u_int16_t
*)bufp
->page
;
356 * Return the data for the key/data pair that begins on this page at this
357 * index (index should always be 1).
360 __big_return(hashp
, bufp
, ndx
, val
, set_current
)
368 u_int16_t
*bp
, len
, off
, save_addr
;
371 bp
= (u_int16_t
*)bufp
->page
;
372 while (bp
[ndx
+ 1] == PARTIAL_KEY
) {
373 bufp
= __get_buf(hashp
, bp
[bp
[0] - 1], bufp
, 0);
376 bp
= (u_int16_t
*)bufp
->page
;
380 if (bp
[ndx
+ 1] == FULL_KEY
) {
381 bufp
= __get_buf(hashp
, bp
[bp
[0] - 1], bufp
, 0);
384 bp
= (u_int16_t
*)bufp
->page
;
386 save_addr
= save_p
->addr
;
390 if (!FREESPACE(bp
)) {
392 * This is a hack. We can't distinguish between
393 * FULL_KEY_DATA that contains complete data or
394 * incomplete data, so we require that if the data
395 * is complete, there is at least 1 byte of free
401 save_addr
= bufp
->addr
;
402 bufp
= __get_buf(hashp
, bp
[bp
[0] - 1], bufp
, 0);
405 bp
= (u_int16_t
*)bufp
->page
;
407 /* The data is all on one page. */
410 val
->data
= (u_char
*)tp
+ off
;
411 val
->size
= bp
[1] - off
;
413 if (bp
[0] == 2) { /* No more buckets in
419 hashp
->cpage
= __get_buf(hashp
,
420 bp
[bp
[0] - 1], bufp
, 0);
425 hashp
->cpage
->page
)[0]) {
434 val
->size
= collect_data(hashp
, bufp
, (int)len
, set_current
);
435 if (val
->size
== (size_t) -1)
437 if (save_p
->addr
!= save_addr
) {
438 /* We are pretty short on buffers. */
439 errno
= EINVAL
; /* OUT OF BUFFERS */
442 memmove(hashp
->tmp_buf
, (save_p
->page
) + off
, len
);
443 val
->data
= (u_char
*)hashp
->tmp_buf
;
447 * Count how big the total datasize is by recursing through the pages. Then
448 * allocate a buffer and copy the data as you recurse up.
451 collect_data(hashp
, bufp
, len
, set
)
456 register u_int16_t
*bp
;
464 mylen
= hashp
->BSIZE
- bp
[1];
465 save_addr
= bufp
->addr
;
467 if (bp
[2] == FULL_KEY_DATA
) { /* End of Data */
468 totlen
= len
+ mylen
;
470 free(hashp
->tmp_buf
);
471 if ((hashp
->tmp_buf
= (char *)malloc(totlen
)) == NULL
)
475 if (bp
[0] == 2) { /* No more buckets in chain */
480 __get_buf(hashp
, bp
[bp
[0] - 1], bufp
, 0);
483 else if (!((u_int16_t
*)hashp
->cpage
->page
)[0]) {
490 xbp
= __get_buf(hashp
, bp
[bp
[0] - 1], bufp
, 0);
491 if (!xbp
|| ((totlen
=
492 collect_data(hashp
, xbp
, len
+ mylen
, set
)) < 1))
495 if (bufp
->addr
!= save_addr
) {
496 errno
= EINVAL
; /* Out of buffers. */
499 memmove(&hashp
->tmp_buf
[len
], (bufp
->page
) + bp
[1], mylen
);
504 * Fill in the key and data for this big pair.
507 __big_keydata(hashp
, bufp
, key
, val
, set
)
513 key
->size
= collect_key(hashp
, bufp
, 0, val
, set
);
514 if (key
->size
== (size_t) -1)
516 key
->data
= (u_char
*)hashp
->tmp_key
;
521 * Count how big the total key size is by recursing through the pages. Then
522 * collect the data, allocate a buffer and copy the key as you recurse up.
525 collect_key(hashp
, bufp
, len
, val
, set
)
535 u_int16_t
*bp
, save_addr
;
539 mylen
= hashp
->BSIZE
- bp
[1];
541 save_addr
= bufp
->addr
;
542 totlen
= len
+ mylen
;
543 if (bp
[2] == FULL_KEY
|| bp
[2] == FULL_KEY_DATA
) { /* End of Key. */
544 if (hashp
->tmp_key
!= NULL
)
545 free(hashp
->tmp_key
);
546 if ((hashp
->tmp_key
= (char *)malloc(totlen
)) == NULL
)
548 if (__big_return(hashp
, bufp
, 1, val
, set
))
551 xbp
= __get_buf(hashp
, bp
[bp
[0] - 1], bufp
, 0);
552 if (!xbp
|| ((totlen
=
553 collect_key(hashp
, xbp
, totlen
, val
, set
)) < 1))
556 if (bufp
->addr
!= save_addr
) {
557 errno
= EINVAL
; /* MIS -- OUT OF BUFFERS */
560 memmove(&hashp
->tmp_key
[len
], (bufp
->page
) + bp
[1], mylen
);
570 __big_split(hashp
, op
, np
, big_keyp
, addr
, obucket
, ret
)
572 BUFHEAD
*op
; /* Pointer to where to put keys that go in old bucket */
573 BUFHEAD
*np
; /* Pointer to new bucket page */
574 /* Pointer to first page containing the big key/data */
576 int addr
; /* Address of big_keyp */
577 u_int32_t obucket
;/* Old Bucket */
580 register BUFHEAD
*tmpp
;
581 register u_int16_t
*tp
;
585 u_int16_t free_space
, n
, off
;
589 /* Now figure out where the big key/data goes */
590 if (__big_keydata(hashp
, big_keyp
, &key
, &val
, 0))
592 change
= (__call_hash(hashp
, key
.data
, key
.size
) != obucket
);
594 if ((ret
->next_addr
= __find_last_page(hashp
, &big_keyp
))) {
596 __get_buf(hashp
, ret
->next_addr
, big_keyp
, 0)))
601 /* Now make one of np/op point to the big key/data pair */
603 assert(np
->ovfl
== NULL
);
610 tmpp
->flags
|= BUF_MOD
;
612 (void)fprintf(stderr
,
613 "BIG_SPLIT: %d->ovfl was %d is now %d\n", tmpp
->addr
,
614 (tmpp
->ovfl
? tmpp
->ovfl
->addr
: 0), (bp
? bp
->addr
: 0));
616 tmpp
->ovfl
= bp
; /* one of op/np point to big_keyp */
617 tp
= (u_int16_t
*)tmpp
->page
;
619 assert(FREESPACE(tp
) >= OVFLSIZE
);
623 free_space
= FREESPACE(tp
);
624 tp
[++n
] = (u_int16_t
)addr
;
628 FREESPACE(tp
) = free_space
- OVFLSIZE
;
631 * Finally, set the new and old return values. BIG_KEYP contains a
632 * pointer to the last page of the big key_data pair. Make sure that
633 * big_keyp has no following page (2 elements) or create an empty
640 tp
= (u_int16_t
*)big_keyp
->page
;
641 big_keyp
->flags
|= BUF_MOD
;
644 * There may be either one or two offsets on this page. If
645 * there is one, then the overflow page is linked on normally
646 * and tp[4] is OVFLPAGE. If there are two, tp[4] contains
647 * the second offset and needs to get stuffed in after the
648 * next overflow page is added.
651 free_space
= FREESPACE(tp
);
654 FREESPACE(tp
) = free_space
+ OVFLSIZE
;
656 tmpp
= __add_ovflpage(hashp
, big_keyp
);