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
36 * @(#)hash_bigkey.c 8.3 (Berkeley) 5/31/94
37 * $DragonFly: src/lib/libc/db/hash/hash_bigkey.c,v 1.4 2003/11/12 20:21:23 eirikn Exp $
43 * Big key/data handling for the hashing package.
58 #include <sys/param.h>
74 static int collect_key (HTAB *, BUFHEAD *, int, DBT *, int);
75 static int collect_data (HTAB *, BUFHEAD *, int, int);
80 * You need to do an insert and the key/data pair is too big
87 __big_insert(hashp, bufp, key, val)
92 register u_int16_t *p;
93 int key_size, n, val_size;
94 u_int16_t space, move_bytes, off;
95 char *cp, *key_data, *val_data;
97 cp = bufp->page; /* Character pointer of p. */
100 key_data = (char *)key->data;
101 key_size = key->size;
102 val_data = (char *)val->data;
103 val_size = val->size;
105 /* First move the Key */
106 for (space = FREESPACE(p) - BIGOVERHEAD; key_size;
107 space = FREESPACE(p) - BIGOVERHEAD) {
108 move_bytes = MIN(space, key_size);
109 off = OFFSET(p) - move_bytes;
110 memmove(cp + off, key_data, move_bytes);
111 key_size -= move_bytes;
112 key_data += move_bytes;
116 FREESPACE(p) = off - PAGE_META(n);
119 bufp = __add_ovflpage(hashp, bufp);
125 move_bytes = MIN(FREESPACE(p), val_size);
126 off = OFFSET(p) - move_bytes;
128 memmove(cp + off, val_data, move_bytes);
129 val_data += move_bytes;
130 val_size -= move_bytes;
131 p[n - 2] = FULL_KEY_DATA;
132 FREESPACE(p) = FREESPACE(p) - move_bytes;
136 p = (u_int16_t *)bufp->page;
138 bufp->flags |= BUF_MOD;
141 /* Now move the data */
142 for (space = FREESPACE(p) - BIGOVERHEAD; val_size;
143 space = FREESPACE(p) - BIGOVERHEAD) {
144 move_bytes = MIN(space, val_size);
146 * Here's the hack to make sure that if the data ends on the
147 * same page as the key ends, FREESPACE is at least one.
149 if (space == val_size && val_size == val->size)
151 off = OFFSET(p) - move_bytes;
152 memmove(cp + off, val_data, move_bytes);
153 val_size -= move_bytes;
154 val_data += move_bytes;
158 FREESPACE(p) = off - PAGE_META(n);
162 bufp = __add_ovflpage(hashp, bufp);
168 p[n] = FULL_KEY_DATA;
169 bufp->flags |= BUF_MOD;
175 * Called when bufp's page contains a partial key (index should be 1)
177 * All pages in the big key/data pair except bufp are freed. We cannot
178 * free bufp because the page pointing to it is lost and we can't get rid
186 __big_delete(hashp, bufp)
190 register BUFHEAD *last_bfp, *rbufp;
191 u_int16_t *bp, pageno;
196 bp = (u_int16_t *)bufp->page;
200 while (!key_done || (bp[2] != FULL_KEY_DATA)) {
201 if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA)
205 * If there is freespace left on a FULL_KEY_DATA page, then
206 * the data is short and fits entirely on this page, and this
209 if (bp[2] == FULL_KEY_DATA && FREESPACE(bp))
211 pageno = bp[bp[0] - 1];
212 rbufp->flags |= BUF_MOD;
213 rbufp = __get_buf(hashp, pageno, rbufp, 0);
215 __free_ovflpage(hashp, last_bfp);
218 return (-1); /* Error. */
219 bp = (u_int16_t *)rbufp->page;
223 * If we get here then rbufp points to the last page of the big
224 * key/data pair. Bufp points to the first one -- it should now be
225 * empty pointing to the next page after this pair. Can't free it
226 * because we don't have the page pointing to it.
229 /* This is information from the last page of the pair. */
233 /* Now, bp is the first page of the pair. */
234 bp = (u_int16_t *)bufp->page;
236 /* There is an overflow page. */
239 bufp->ovfl = rbufp->ovfl;
241 /* This is the last page. */
245 FREESPACE(bp) = hashp->BSIZE - PAGE_META(n);
246 OFFSET(bp) = hashp->BSIZE - 1;
248 bufp->flags |= BUF_MOD;
250 __free_ovflpage(hashp, rbufp);
251 if (last_bfp != rbufp)
252 __free_ovflpage(hashp, last_bfp);
260 * -1 = get next overflow page
261 * -2 means key not found and this is big key/data
265 __find_bigpair(hashp, bufp, ndx, key, size)
272 register u_int16_t *bp;
278 bp = (u_int16_t *)bufp->page;
283 for (bytes = hashp->BSIZE - bp[ndx];
284 bytes <= size && bp[ndx + 1] == PARTIAL_KEY;
285 bytes = hashp->BSIZE - bp[ndx]) {
286 if (memcmp(p + bp[ndx], kkey, bytes))
290 bufp = __get_buf(hashp, bp[ndx + 2], bufp, 0);
298 if (bytes != ksize || memcmp(p + bp[ndx], kkey, bytes)) {
299 #ifdef HASH_STATISTICS
308 * Given the buffer pointer of the first overflow page of a big pair,
309 * find the end of the big pair
311 * This will set bpp to the buffer header of the last page of the big pair.
312 * It will return the pageno of the overflow page following the last page
313 * of the pair; 0 if there isn't any (i.e. big pair is the last key in the
317 __find_last_page(hashp, bpp)
322 u_int16_t *bp, pageno;
326 bp = (u_int16_t *)bufp->page;
331 * This is the last page if: the tag is FULL_KEY_DATA and
332 * either only 2 entries OVFLPAGE marker is explicit there
333 * is freespace on the page.
335 if (bp[2] == FULL_KEY_DATA &&
336 ((n == 2) || (bp[n] == OVFLPAGE) || (FREESPACE(bp))))
340 bufp = __get_buf(hashp, pageno, bufp, 0);
342 return (0); /* Need to indicate an error! */
343 bp = (u_int16_t *)bufp->page;
354 * Return the data for the key/data pair that begins on this page at this
355 * index (index should always be 1).
358 __big_return(hashp, bufp, ndx, val, set_current)
366 u_int16_t *bp, len, off, save_addr;
369 bp = (u_int16_t *)bufp->page;
370 while (bp[ndx + 1] == PARTIAL_KEY) {
371 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
374 bp = (u_int16_t *)bufp->page;
378 if (bp[ndx + 1] == FULL_KEY) {
379 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
382 bp = (u_int16_t *)bufp->page;
384 save_addr = save_p->addr;
388 if (!FREESPACE(bp)) {
390 * This is a hack. We can't distinguish between
391 * FULL_KEY_DATA that contains complete data or
392 * incomplete data, so we require that if the data
393 * is complete, there is at least 1 byte of free
399 save_addr = bufp->addr;
400 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
403 bp = (u_int16_t *)bufp->page;
405 /* The data is all on one page. */
408 val->data = (u_char *)tp + off;
409 val->size = bp[1] - off;
411 if (bp[0] == 2) { /* No more buckets in
417 hashp->cpage = __get_buf(hashp,
418 bp[bp[0] - 1], bufp, 0);
423 hashp->cpage->page)[0]) {
432 val->size = collect_data(hashp, bufp, (int)len, set_current);
435 if (save_p->addr != save_addr) {
436 /* We are pretty short on buffers. */
437 errno = EINVAL; /* OUT OF BUFFERS */
440 memmove(hashp->tmp_buf, (save_p->page) + off, len);
441 val->data = (u_char *)hashp->tmp_buf;
445 * Count how big the total datasize is by recursing through the pages. Then
446 * allocate a buffer and copy the data as you recurse up.
449 collect_data(hashp, bufp, len, set)
454 register u_int16_t *bp;
462 mylen = hashp->BSIZE - bp[1];
463 save_addr = bufp->addr;
465 if (bp[2] == FULL_KEY_DATA) { /* End of Data */
466 totlen = len + mylen;
468 free(hashp->tmp_buf);
469 if ((hashp->tmp_buf = (char *)malloc(totlen)) == NULL)
473 if (bp[0] == 2) { /* No more buckets in chain */
478 __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
481 else if (!((u_int16_t *)hashp->cpage->page)[0]) {
488 xbp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
489 if (!xbp || ((totlen =
490 collect_data(hashp, xbp, len + mylen, set)) < 1))
493 if (bufp->addr != save_addr) {
494 errno = EINVAL; /* Out of buffers. */
497 memmove(&hashp->tmp_buf[len], (bufp->page) + bp[1], mylen);
502 * Fill in the key and data for this big pair.
505 __big_keydata(hashp, bufp, key, val, set)
511 key->size = collect_key(hashp, bufp, 0, val, set);
514 key->data = (u_char *)hashp->tmp_key;
519 * Count how big the total key size is by recursing through the pages. Then
520 * collect the data, allocate a buffer and copy the key as you recurse up.
523 collect_key(hashp, bufp, len, val, set)
533 u_int16_t *bp, save_addr;
537 mylen = hashp->BSIZE - bp[1];
539 save_addr = bufp->addr;
540 totlen = len + mylen;
541 if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA) { /* End of Key. */
542 if (hashp->tmp_key != NULL)
543 free(hashp->tmp_key);
544 if ((hashp->tmp_key = (char *)malloc(totlen)) == NULL)
546 if (__big_return(hashp, bufp, 1, val, set))
549 xbp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
550 if (!xbp || ((totlen =
551 collect_key(hashp, xbp, totlen, val, set)) < 1))
554 if (bufp->addr != save_addr) {
555 errno = EINVAL; /* MIS -- OUT OF BUFFERS */
558 memmove(&hashp->tmp_key[len], (bufp->page) + bp[1], mylen);
568 __big_split(hashp, op, np, big_keyp, addr, obucket, ret)
570 BUFHEAD *op; /* Pointer to where to put keys that go in old bucket */
571 BUFHEAD *np; /* Pointer to new bucket page */
572 /* Pointer to first page containing the big key/data */
574 int addr; /* Address of big_keyp */
575 u_int32_t obucket;/* Old Bucket */
578 register BUFHEAD *tmpp;
579 register u_int16_t *tp;
583 u_int16_t free_space, n, off;
587 /* Now figure out where the big key/data goes */
588 if (__big_keydata(hashp, big_keyp, &key, &val, 0))
590 change = (__call_hash(hashp, key.data, key.size) != obucket);
592 if ( (ret->next_addr = __find_last_page(hashp, &big_keyp)) ) {
594 __get_buf(hashp, ret->next_addr, big_keyp, 0)))
599 /* Now make one of np/op point to the big key/data pair */
601 assert(np->ovfl == NULL);
608 tmpp->flags |= BUF_MOD;
610 (void)fprintf(stderr,
611 "BIG_SPLIT: %d->ovfl was %d is now %d\n", tmpp->addr,
612 (tmpp->ovfl ? tmpp->ovfl->addr : 0), (bp ? bp->addr : 0));
614 tmpp->ovfl = bp; /* one of op/np point to big_keyp */
615 tp = (u_int16_t *)tmpp->page;
617 assert(FREESPACE(tp) >= OVFLSIZE);
621 free_space = FREESPACE(tp);
622 tp[++n] = (u_int16_t)addr;
626 FREESPACE(tp) = free_space - OVFLSIZE;
629 * Finally, set the new and old return values. BIG_KEYP contains a
630 * pointer to the last page of the big key_data pair. Make sure that
631 * big_keyp has no following page (2 elements) or create an empty
638 tp = (u_int16_t *)big_keyp->page;
639 big_keyp->flags |= BUF_MOD;
642 * There may be either one or two offsets on this page. If
643 * there is one, then the overflow page is linked on normally
644 * and tp[4] is OVFLPAGE. If there are two, tp[4] contains
645 * the second offset and needs to get stuffed in after the
646 * next overflow page is added.
649 free_space = FREESPACE(tp);
652 FREESPACE(tp) = free_space + OVFLSIZE;
654 tmpp = __add_ovflpage(hashp, big_keyp);