Fix mdoc(7)/man(7) mix up.
[netbsd-mini2440.git] / lib / libc / db / hash / hash_bigkey.c
blob66498352c4c74e50ee785563a5963701460a2f94
1 /* $NetBSD: hash_bigkey.c,v 1.22 2008/09/10 17:52:35 joerg Exp $ */
3 /*-
4 * Copyright (c) 1990, 1993, 1994
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * Margo Seltzer.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
35 #if HAVE_NBTOOL_CONFIG_H
36 #include "nbtool_config.h"
37 #endif
39 #include <sys/cdefs.h>
40 __RCSID("$NetBSD: hash_bigkey.c,v 1.22 2008/09/10 17:52:35 joerg Exp $");
43 * PACKAGE: hash
44 * DESCRIPTION:
45 * Big key/data handling for the hashing package.
47 * ROUTINES:
48 * External
49 * __big_keydata
50 * __big_split
51 * __big_insert
52 * __big_return
53 * __big_delete
54 * __find_last_page
55 * Internal
56 * collect_key
57 * collect_data
60 #include <sys/param.h>
62 #include <errno.h>
63 #include <stdio.h>
64 #include <stdlib.h>
65 #include <string.h>
66 #include <assert.h>
68 #include <db.h>
69 #include "hash.h"
70 #include "page.h"
71 #include "extern.h"
73 static int collect_key(HTAB *, BUFHEAD *, int, DBT *, int);
74 static int collect_data(HTAB *, BUFHEAD *, int, int);
77 * Big_insert
79 * You need to do an insert and the key/data pair is too big
81 * Returns:
82 * 0 ==> OK
83 *-1 ==> ERROR
85 int
86 __big_insert(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val)
88 uint16_t *p, n;
89 size_t key_size, val_size;
90 uint16_t space, move_bytes, off;
91 char *cp, *key_data, *val_data;
92 size_t temp;
94 cp = bufp->page; /* Character pointer of p. */
95 p = (uint16_t *)(void *)cp;
97 key_data = (char *)key->data;
98 _DBFIT(key->size, int);
99 key_size = key->size;
100 val_data = (char *)val->data;
101 _DBFIT(val->size, int);
102 val_size = val->size;
104 /* First move the Key */
106 temp = FREESPACE(p) - BIGOVERHEAD;
107 _DBFIT(temp, uint16_t);
108 space = (uint16_t)temp;
109 while (key_size) {
110 move_bytes = MIN(space, key_size);
111 off = OFFSET(p) - move_bytes;
112 memmove(cp + off, key_data, (size_t)move_bytes);
113 key_size -= move_bytes;
114 key_data += move_bytes;
115 n = p[0];
116 p[++n] = off;
117 p[0] = ++n;
118 temp = off - PAGE_META(n);
119 _DBFIT(temp, uint16_t);
120 FREESPACE(p) = (uint16_t)temp;
121 OFFSET(p) = off;
122 p[n] = PARTIAL_KEY;
123 bufp = __add_ovflpage(hashp, bufp);
124 if (!bufp)
125 return (-1);
126 n = p[0];
127 if (!key_size) {
128 space = FREESPACE(p);
129 if (space) {
130 move_bytes = MIN(space, val_size);
132 * If the data would fit exactly in the
133 * remaining space, we must overflow it to the
134 * next page; otherwise the invariant that the
135 * data must end on a page with FREESPACE
136 * non-zero would fail.
138 if (space == val_size && val_size == val->size)
139 goto toolarge;
140 off = OFFSET(p) - move_bytes;
141 memmove(cp + off, val_data, (size_t)move_bytes);
142 val_data += move_bytes;
143 val_size -= move_bytes;
144 p[n] = off;
145 p[n - 2] = FULL_KEY_DATA;
146 FREESPACE(p) = FREESPACE(p) - move_bytes;
147 OFFSET(p) = off;
148 } else {
149 toolarge:
150 p[n - 2] = FULL_KEY;
153 p = (uint16_t *)(void *)bufp->page;
154 cp = bufp->page;
155 bufp->flags |= BUF_MOD;
156 temp = FREESPACE(p) - BIGOVERHEAD;
157 _DBFIT(temp, uint16_t);
158 space = (uint16_t)temp;
161 /* Now move the data */
162 temp = FREESPACE(p) - BIGOVERHEAD;
163 _DBFIT(temp, uint16_t);
164 space = (uint16_t)temp;
165 while (val_size) {
166 move_bytes = MIN(space, val_size);
168 * Here's the hack to make sure that if the data ends on the
169 * same page as the key ends, FREESPACE is at least one.
171 if (space == val_size && val_size == val->size)
172 move_bytes--;
173 off = OFFSET(p) - move_bytes;
174 memmove(cp + off, val_data, (size_t)move_bytes);
175 val_size -= move_bytes;
176 val_data += move_bytes;
177 n = p[0];
178 p[++n] = off;
179 p[0] = ++n;
180 temp = off - PAGE_META(n);
181 _DBFIT(temp, uint16_t);
182 FREESPACE(p) = (uint16_t)temp;
183 OFFSET(p) = off;
184 if (val_size) {
185 p[n] = FULL_KEY;
186 bufp = __add_ovflpage(hashp, bufp);
187 if (!bufp)
188 return (-1);
189 cp = bufp->page;
190 p = (uint16_t *)(void *)cp;
191 } else
192 p[n] = FULL_KEY_DATA;
193 bufp->flags |= BUF_MOD;
194 temp = FREESPACE(p) - BIGOVERHEAD;
195 _DBFIT(temp, uint16_t);
196 space = (uint16_t)temp;
198 return (0);
202 * Called when bufp's page contains a partial key (index should be 1)
204 * All pages in the big key/data pair except bufp are freed. We cannot
205 * free bufp because the page pointing to it is lost and we can't get rid
206 * of its pointer.
208 * Returns:
209 * 0 => OK
210 *-1 => ERROR
213 __big_delete(HTAB *hashp, BUFHEAD *bufp)
215 BUFHEAD *last_bfp, *rbufp;
216 uint16_t *bp, pageno;
217 int key_done, n;
218 size_t temp;
220 rbufp = bufp;
221 last_bfp = NULL;
222 bp = (uint16_t *)(void *)bufp->page;
223 pageno = 0;
224 key_done = 0;
226 while (!key_done || (bp[2] != FULL_KEY_DATA)) {
227 if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA)
228 key_done = 1;
231 * If there is freespace left on a FULL_KEY_DATA page, then
232 * the data is short and fits entirely on this page, and this
233 * is the last page.
235 if (bp[2] == FULL_KEY_DATA && FREESPACE(bp))
236 break;
237 pageno = bp[bp[0] - 1];
238 rbufp->flags |= BUF_MOD;
239 rbufp = __get_buf(hashp, (uint32_t)pageno, rbufp, 0);
240 if (last_bfp)
241 __free_ovflpage(hashp, last_bfp);
242 last_bfp = rbufp;
243 if (!rbufp)
244 return (-1); /* Error. */
245 bp = (uint16_t *)(void *)rbufp->page;
249 * If we get here then rbufp points to the last page of the big
250 * key/data pair. Bufp points to the first one -- it should now be
251 * empty pointing to the next page after this pair. Can't free it
252 * because we don't have the page pointing to it.
255 /* This is information from the last page of the pair. */
256 n = bp[0];
257 pageno = bp[n - 1];
259 /* Now, bp is the first page of the pair. */
260 bp = (uint16_t *)(void *)bufp->page;
261 if (n > 2) {
262 /* There is an overflow page. */
263 bp[1] = pageno;
264 bp[2] = OVFLPAGE;
265 bufp->ovfl = rbufp->ovfl;
266 } else
267 /* This is the last page. */
268 bufp->ovfl = NULL;
269 n -= 2;
270 bp[0] = n;
271 temp = hashp->BSIZE - PAGE_META(n);
272 _DBFIT(temp, uint16_t);
273 FREESPACE(bp) = (uint16_t)temp;
274 OFFSET(bp) = hashp->BSIZE;
276 bufp->flags |= BUF_MOD;
277 if (rbufp)
278 __free_ovflpage(hashp, rbufp);
279 if (last_bfp && last_bfp != rbufp)
280 __free_ovflpage(hashp, last_bfp);
282 hashp->NKEYS--;
283 return (0);
286 * Returns:
287 * 0 = key not found
288 * -1 = get next overflow page
289 * -2 means key not found and this is big key/data
290 * -3 error
293 __find_bigpair(HTAB *hashp, BUFHEAD *bufp, int ndx, char *key, int size)
295 uint16_t *bp;
296 char *p;
297 int ksize;
298 uint16_t bytes;
299 char *kkey;
301 bp = (uint16_t *)(void *)bufp->page;
302 p = bufp->page;
303 ksize = size;
304 kkey = key;
306 for (bytes = hashp->BSIZE - bp[ndx];
307 bytes <= size && bp[ndx + 1] == PARTIAL_KEY;
308 bytes = hashp->BSIZE - bp[ndx]) {
309 if (memcmp(p + bp[ndx], kkey, (size_t)bytes))
310 return (-2);
311 kkey += bytes;
312 ksize -= bytes;
313 bufp = __get_buf(hashp, (uint32_t)bp[ndx + 2], bufp, 0);
314 if (!bufp)
315 return (-3);
316 p = bufp->page;
317 bp = (uint16_t *)(void *)p;
318 ndx = 1;
321 if (bytes != ksize || memcmp(p + bp[ndx], kkey, (size_t)bytes)) {
322 #ifdef HASH_STATISTICS
323 ++hash_collisions;
324 #endif
325 return (-2);
326 } else
327 return (ndx);
331 * Given the buffer pointer of the first overflow page of a big pair,
332 * find the end of the big pair
334 * This will set bpp to the buffer header of the last page of the big pair.
335 * It will return the pageno of the overflow page following the last page
336 * of the pair; 0 if there isn't any (i.e. big pair is the last key in the
337 * bucket)
339 uint16_t
340 __find_last_page(HTAB *hashp, BUFHEAD **bpp)
342 BUFHEAD *bufp;
343 uint16_t *bp, pageno;
344 int n;
346 bufp = *bpp;
347 bp = (uint16_t *)(void *)bufp->page;
348 for (;;) {
349 n = bp[0];
352 * This is the last page if: the tag is FULL_KEY_DATA and
353 * either only 2 entries OVFLPAGE marker is explicit there
354 * is freespace on the page.
356 if (bp[2] == FULL_KEY_DATA &&
357 ((n == 2) || (bp[n] == OVFLPAGE) || (FREESPACE(bp))))
358 break;
360 pageno = bp[n - 1];
361 bufp = __get_buf(hashp, (uint32_t)pageno, bufp, 0);
362 if (!bufp)
363 return (0); /* Need to indicate an error! */
364 bp = (uint16_t *)(void *)bufp->page;
367 *bpp = bufp;
368 if (bp[0] > 2)
369 return (bp[3]);
370 else
371 return (0);
375 * Return the data for the key/data pair that begins on this page at this
376 * index (index should always be 1).
379 __big_return(HTAB *hashp, BUFHEAD *bufp, int ndx, DBT *val, int set_current)
381 BUFHEAD *save_p;
382 uint16_t *bp, len, off, save_addr;
383 char *tp;
385 bp = (uint16_t *)(void *)bufp->page;
386 while (bp[ndx + 1] == PARTIAL_KEY) {
387 bufp = __get_buf(hashp, (uint32_t)bp[bp[0] - 1], bufp, 0);
388 if (!bufp)
389 return (-1);
390 bp = (uint16_t *)(void *)bufp->page;
391 ndx = 1;
394 if (bp[ndx + 1] == FULL_KEY) {
395 bufp = __get_buf(hashp, (uint32_t)bp[bp[0] - 1], bufp, 0);
396 if (!bufp)
397 return (-1);
398 bp = (uint16_t *)(void *)bufp->page;
399 save_p = bufp;
400 save_addr = save_p->addr;
401 off = bp[1];
402 len = 0;
403 } else
404 if (!FREESPACE(bp)) {
406 * This is a hack. We can't distinguish between
407 * FULL_KEY_DATA that contains complete data or
408 * incomplete data, so we require that if the data
409 * is complete, there is at least 1 byte of free
410 * space left.
412 off = bp[bp[0]];
413 len = bp[1] - off;
414 save_p = bufp;
415 save_addr = bufp->addr;
416 bufp = __get_buf(hashp, (uint32_t)bp[bp[0] - 1], bufp,
418 if (!bufp)
419 return (-1);
420 bp = (uint16_t *)(void *)bufp->page;
421 } else {
422 /* The data is all on one page. */
423 tp = (char *)(void *)bp;
424 off = bp[bp[0]];
425 val->data = (uint8_t *)tp + off;
426 val->size = bp[1] - off;
427 if (set_current) {
428 if (bp[0] == 2) { /* No more buckets in
429 * chain */
430 hashp->cpage = NULL;
431 hashp->cbucket++;
432 hashp->cndx = 1;
433 } else {
434 hashp->cpage = __get_buf(hashp,
435 (uint32_t)bp[bp[0] - 1], bufp, 0);
436 if (!hashp->cpage)
437 return (-1);
438 hashp->cndx = 1;
439 if (!((uint16_t *)(void *)
440 hashp->cpage->page)[0]) {
441 hashp->cbucket++;
442 hashp->cpage = NULL;
446 return (0);
449 val->size = collect_data(hashp, bufp, (int)len, set_current);
450 if (val->size == (size_t)-1)
451 return (-1);
452 if (save_p->addr != save_addr) {
453 /* We are pretty short on buffers. */
454 errno = EINVAL; /* OUT OF BUFFERS */
455 return (-1);
457 memmove(hashp->tmp_buf, (save_p->page) + off, (size_t)len);
458 val->data = (uint8_t *)hashp->tmp_buf;
459 return (0);
462 * Count how big the total datasize is by recursing through the pages. Then
463 * allocate a buffer and copy the data as you recurse up.
465 static int
466 collect_data(HTAB *hashp, BUFHEAD *bufp, int len, int set)
468 uint16_t *bp;
469 char *p;
470 BUFHEAD *xbp;
471 uint16_t save_addr;
472 int mylen, totlen;
474 p = bufp->page;
475 bp = (uint16_t *)(void *)p;
476 mylen = hashp->BSIZE - bp[1];
477 save_addr = bufp->addr;
479 if (bp[2] == FULL_KEY_DATA) { /* End of Data */
480 totlen = len + mylen;
481 if (hashp->tmp_buf)
482 free(hashp->tmp_buf);
483 if ((hashp->tmp_buf = calloc(1, (size_t)totlen)) == NULL)
484 return (-1);
485 if (set) {
486 hashp->cndx = 1;
487 if (bp[0] == 2) { /* No more buckets in chain */
488 hashp->cpage = NULL;
489 hashp->cbucket++;
490 } else {
491 hashp->cpage =
492 __get_buf(hashp, (uint32_t)bp[bp[0] - 1],
493 bufp, 0);
494 if (!hashp->cpage)
495 return (-1);
496 else if (!((uint16_t *)(void *)hashp->cpage->page)[0]) {
497 hashp->cbucket++;
498 hashp->cpage = NULL;
502 } else {
503 xbp = __get_buf(hashp, (uint32_t)bp[bp[0] - 1], bufp, 0);
504 if (!xbp || ((totlen =
505 collect_data(hashp, xbp, len + mylen, set)) < 1))
506 return (-1);
508 if (bufp->addr != save_addr) {
509 errno = EINVAL; /* Out of buffers. */
510 return (-1);
512 memmove(&hashp->tmp_buf[len], (bufp->page) + bp[1], (size_t)mylen);
513 return (totlen);
517 * Fill in the key and data for this big pair.
520 __big_keydata(HTAB *hashp, BUFHEAD *bufp, DBT *key, DBT *val, int set)
522 key->size = collect_key(hashp, bufp, 0, val, set);
523 if (key->size == (size_t)-1)
524 return (-1);
525 key->data = (uint8_t *)hashp->tmp_key;
526 return (0);
530 * Count how big the total key size is by recursing through the pages. Then
531 * collect the data, allocate a buffer and copy the key as you recurse up.
533 static int
534 collect_key(HTAB *hashp, BUFHEAD *bufp, int len, DBT *val, int set)
536 BUFHEAD *xbp;
537 char *p;
538 int mylen, totlen;
539 uint16_t *bp, save_addr;
541 p = bufp->page;
542 bp = (uint16_t *)(void *)p;
543 mylen = hashp->BSIZE - bp[1];
545 save_addr = bufp->addr;
546 totlen = len + mylen;
547 if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA) { /* End of Key. */
548 if (hashp->tmp_key != NULL)
549 free(hashp->tmp_key);
550 if ((hashp->tmp_key = calloc(1, (size_t)totlen)) == NULL)
551 return (-1);
552 if (__big_return(hashp, bufp, 1, val, set))
553 return (-1);
554 } else {
555 xbp = __get_buf(hashp, (uint32_t)bp[bp[0] - 1], bufp, 0);
556 if (!xbp || ((totlen =
557 collect_key(hashp, xbp, totlen, val, set)) < 1))
558 return (-1);
560 if (bufp->addr != save_addr) {
561 errno = EINVAL; /* MIS -- OUT OF BUFFERS */
562 return (-1);
564 memmove(&hashp->tmp_key[len], (bufp->page) + bp[1], (size_t)mylen);
565 return (totlen);
569 * Returns:
570 * 0 => OK
571 * -1 => error
574 __big_split(
575 HTAB *hashp,
576 BUFHEAD *op, /* Pointer to where to put keys that go in old bucket */
577 BUFHEAD *np, /* Pointer to new bucket page */
578 /* Pointer to first page containing the big key/data */
579 BUFHEAD *big_keyp,
580 int addr, /* Address of big_keyp */
581 uint32_t obucket,/* Old Bucket */
582 SPLIT_RETURN *ret
585 BUFHEAD *tmpp;
586 uint16_t *tp;
587 BUFHEAD *bp;
588 DBT key, val;
589 uint32_t change;
590 uint16_t free_space, n, off;
591 size_t temp;
593 bp = big_keyp;
595 /* Now figure out where the big key/data goes */
596 if (__big_keydata(hashp, big_keyp, &key, &val, 0))
597 return (-1);
598 change = (__call_hash(hashp, key.data, (int)key.size) != obucket);
600 if ((ret->next_addr = __find_last_page(hashp, &big_keyp)) != 0) {
601 if (!(ret->nextp =
602 __get_buf(hashp, (uint32_t)ret->next_addr, big_keyp, 0)))
603 return (-1);
604 } else
605 ret->nextp = NULL;
607 /* Now make one of np/op point to the big key/data pair */
608 _DIAGASSERT(np->ovfl == NULL);
609 if (change)
610 tmpp = np;
611 else
612 tmpp = op;
614 tmpp->flags |= BUF_MOD;
615 #ifdef DEBUG1
616 (void)fprintf(stderr,
617 "BIG_SPLIT: %d->ovfl was %d is now %d\n", tmpp->addr,
618 (tmpp->ovfl ? tmpp->ovfl->addr : 0), (bp ? bp->addr : 0));
619 #endif
620 tmpp->ovfl = bp; /* one of op/np point to big_keyp */
621 tp = (uint16_t *)(void *)tmpp->page;
622 _DIAGASSERT(FREESPACE(tp) >= OVFLSIZE);
623 n = tp[0];
624 off = OFFSET(tp);
625 free_space = FREESPACE(tp);
626 tp[++n] = (uint16_t)addr;
627 tp[++n] = OVFLPAGE;
628 tp[0] = n;
629 OFFSET(tp) = off;
630 temp = free_space - OVFLSIZE;
631 _DBFIT(temp, uint16_t);
632 FREESPACE(tp) = (uint16_t)temp;
635 * Finally, set the new and old return values. BIG_KEYP contains a
636 * pointer to the last page of the big key_data pair. Make sure that
637 * big_keyp has no following page (2 elements) or create an empty
638 * following page.
641 ret->newp = np;
642 ret->oldp = op;
644 tp = (uint16_t *)(void *)big_keyp->page;
645 big_keyp->flags |= BUF_MOD;
646 if (tp[0] > 2) {
648 * There may be either one or two offsets on this page. If
649 * there is one, then the overflow page is linked on normally
650 * and tp[4] is OVFLPAGE. If there are two, tp[4] contains
651 * the second offset and needs to get stuffed in after the
652 * next overflow page is added.
654 n = tp[4];
655 free_space = FREESPACE(tp);
656 off = OFFSET(tp);
657 tp[0] -= 2;
658 temp = free_space + OVFLSIZE;
659 _DBFIT(temp, uint16_t);
660 FREESPACE(tp) = (uint16_t)temp;
661 OFFSET(tp) = off;
662 tmpp = __add_ovflpage(hashp, big_keyp);
663 if (!tmpp)
664 return (-1);
665 tp[4] = n;
666 } else
667 tmpp = big_keyp;
669 if (change)
670 ret->newp = tmpp;
671 else
672 ret->oldp = tmpp;
673 return (0);