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_page.c 8.7 (Berkeley) 8/16/94";
35 #endif /* LIBC_SCCS and not lint */
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
43 * Page manipulation for hashing package.
55 #include "namespace.h"
56 #include <sys/types.h>
68 #include "un-namespace.h"
75 static u_int32_t *fetch_bitmap(HTAB *, int);
76 static u_int32_t first_free(u_int32_t);
77 static int open_temp(HTAB *);
78 static u_int16_t overflow_page(HTAB *);
79 static void putpair(char *, const DBT *, const DBT *);
80 static void squeeze_key(u_int16_t *, const DBT *, const DBT *);
82 (HTAB *, u_int32_t, BUFHEAD *, BUFHEAD *, int, int);
84 #define PAGE_INIT(P) { \
85 ((u_int16_t *)(P))[0] = 0; \
86 ((u_int16_t *)(P))[1] = hashp->BSIZE - 3 * sizeof(u_int16_t); \
87 ((u_int16_t *)(P))[2] = hashp->BSIZE; \
91 * This is called AFTER we have verified that there is room on the page for
92 * the pair (PAIRFITS has returned true) so we go right ahead and start moving
100 u_int16_t *bp, n, off;
104 /* Enter the key first. */
107 off = OFFSET(bp) - key->size;
108 memmove(p + off, key->data, key->size);
113 memmove(p + off, val->data, val->size);
116 /* Adjust page info. */
118 bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t));
128 __delpair(hashp, bufp, ndx)
133 u_int16_t *bp, newoff;
137 bp = (u_int16_t *)bufp->page;
140 if (bp[ndx + 1] < REAL_KEY)
141 return (__big_delete(hashp, bufp));
143 newoff = bp[ndx - 1];
145 newoff = hashp->BSIZE;
146 pairlen = newoff - bp[ndx + 1];
148 if (ndx != (n - 1)) {
149 /* Hard Case -- need to shuffle keys */
151 char *src = bufp->page + (int)OFFSET(bp);
152 char *dst = src + (int)pairlen;
153 memmove(dst, src, bp[ndx + 1] - OFFSET(bp));
155 /* Now adjust the pointers */
156 for (i = ndx + 2; i <= n; i += 2) {
157 if (bp[i + 1] == OVFLPAGE) {
159 bp[i - 1] = bp[i + 1];
161 bp[i - 2] = bp[i] + pairlen;
162 bp[i - 1] = bp[i + 1] + pairlen;
166 /* Finally adjust the page data */
167 bp[n] = OFFSET(bp) + pairlen;
168 bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t);
172 bufp->flags |= BUF_MOD;
181 __split_page(hashp, obucket, nbucket)
183 u_int32_t obucket, nbucket;
185 BUFHEAD *new_bufp, *old_bufp;
190 u_int16_t copyto, diff, off, moved;
193 copyto = (u_int16_t)hashp->BSIZE;
194 off = (u_int16_t)hashp->BSIZE;
195 old_bufp = __get_buf(hashp, obucket, NULL, 0);
196 if (old_bufp == NULL)
198 new_bufp = __get_buf(hashp, nbucket, NULL, 0);
199 if (new_bufp == NULL)
202 old_bufp->flags |= (BUF_MOD | BUF_PIN);
203 new_bufp->flags |= (BUF_MOD | BUF_PIN);
205 ino = (u_int16_t *)(op = old_bufp->page);
210 for (n = 1, ndx = 1; n < ino[0]; n += 2) {
211 if (ino[n + 1] < REAL_KEY) {
212 retval = ugly_split(hashp, obucket, old_bufp, new_bufp,
213 (int)copyto, (int)moved);
214 old_bufp->flags &= ~BUF_PIN;
215 new_bufp->flags &= ~BUF_PIN;
219 key.data = (u_char *)op + ino[n];
220 key.size = off - ino[n];
222 if (__call_hash(hashp, key.data, key.size) == obucket) {
223 /* Don't switch page */
226 copyto = ino[n + 1] + diff;
227 memmove(op + copyto, op + ino[n + 1],
229 ino[ndx] = copyto + ino[n] - ino[n + 1];
230 ino[ndx + 1] = copyto;
236 val.data = (u_char *)op + ino[n + 1];
237 val.size = ino[n] - ino[n + 1];
238 putpair(np, &key, &val);
245 /* Now clean up the page */
247 FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3);
248 OFFSET(ino) = copyto;
251 (void)fprintf(stderr, "split %d/%d\n",
252 ((u_int16_t *)np)[0] / 2,
253 ((u_int16_t *)op)[0] / 2);
255 /* unpin both pages */
256 old_bufp->flags &= ~BUF_PIN;
257 new_bufp->flags &= ~BUF_PIN;
262 * Called when we encounter an overflow or big key/data page during split
263 * handling. This is special cased since we have to begin checking whether
264 * the key/data pairs fit on their respective pages and because we may need
265 * overflow pages for both the old and new pages.
267 * The first page might be a page with regular key/data pairs in which case
268 * we have a regular overflow condition and just need to go on to the next
269 * page or it might be a big key/data pair in which case we need to fix the
277 ugly_split(hashp, obucket, old_bufp, new_bufp, copyto, moved)
279 u_int32_t obucket; /* Same as __split_page. */
280 BUFHEAD *old_bufp, *new_bufp;
281 int copyto; /* First byte on page which contains key/data values. */
282 int moved; /* Number of pairs moved to new page. */
284 BUFHEAD *bufp; /* Buffer header for ino */
285 u_int16_t *ino; /* Page keys come off of */
286 u_int16_t *np; /* New page */
287 u_int16_t *op; /* Page keys go on to if they aren't moving */
289 BUFHEAD *last_bfp; /* Last buf header OVFL needing to be freed */
292 u_int16_t n, off, ov_addr, scopyto;
293 char *cino; /* Character value of ino */
296 ino = (u_int16_t *)old_bufp->page;
297 np = (u_int16_t *)new_bufp->page;
298 op = (u_int16_t *)old_bufp->page;
300 scopyto = (u_int16_t)copyto; /* ANSI */
304 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) {
305 if (__big_split(hashp, old_bufp,
306 new_bufp, bufp, bufp->addr, obucket, &ret))
311 op = (u_int16_t *)old_bufp->page;
315 np = (u_int16_t *)new_bufp->page;
319 cino = (char *)bufp->page;
320 ino = (u_int16_t *)cino;
321 last_bfp = ret.nextp;
322 } else if (ino[n + 1] == OVFLPAGE) {
325 * Fix up the old page -- the extra 2 are the fields
326 * which contained the overflow information.
328 ino[0] -= (moved + 2);
330 scopyto - sizeof(u_int16_t) * (ino[0] + 3);
331 OFFSET(ino) = scopyto;
333 bufp = __get_buf(hashp, ov_addr, bufp, 0);
337 ino = (u_int16_t *)bufp->page;
339 scopyto = hashp->BSIZE;
343 __free_ovflpage(hashp, last_bfp);
346 /* Move regular sized pairs of there are any */
348 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) {
350 key.data = (u_char *)cino + ino[n];
351 key.size = off - ino[n];
352 val.data = (u_char *)cino + ino[n + 1];
353 val.size = ino[n] - ino[n + 1];
356 if (__call_hash(hashp, key.data, key.size) == obucket) {
357 /* Keep on old page */
358 if (PAIRFITS(op, (&key), (&val)))
359 putpair((char *)op, &key, &val);
362 __add_ovflpage(hashp, old_bufp);
365 op = (u_int16_t *)old_bufp->page;
366 putpair((char *)op, &key, &val);
368 old_bufp->flags |= BUF_MOD;
370 /* Move to new page */
371 if (PAIRFITS(np, (&key), (&val)))
372 putpair((char *)np, &key, &val);
375 __add_ovflpage(hashp, new_bufp);
378 np = (u_int16_t *)new_bufp->page;
379 putpair((char *)np, &key, &val);
381 new_bufp->flags |= BUF_MOD;
386 __free_ovflpage(hashp, last_bfp);
391 * Add the given pair to the page
398 __addel(hashp, bufp, key, val)
401 const DBT *key, *val;
406 bp = (u_int16_t *)bufp->page;
408 while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
410 if (bp[2] == FULL_KEY_DATA && bp[0] == 2)
411 /* This is the last page of a big key/data pair
412 and we need to add another page */
414 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
415 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
418 bp = (u_int16_t *)bufp->page;
420 /* Try to squeeze key on this page */
421 if (FREESPACE(bp) > PAIRSIZE(key, val)) {
422 squeeze_key(bp, key, val);
425 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
428 bp = (u_int16_t *)bufp->page;
431 if (PAIRFITS(bp, key, val))
432 putpair(bufp->page, key, val);
435 bufp = __add_ovflpage(hashp, bufp);
438 sop = (u_int16_t *)bufp->page;
440 if (PAIRFITS(sop, key, val))
441 putpair((char *)sop, key, val);
443 if (__big_insert(hashp, bufp, key, val))
446 bufp->flags |= BUF_MOD;
448 * If the average number of keys per bucket exceeds the fill factor,
453 (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
454 return (__expand_table(hashp));
465 __add_ovflpage(hashp, bufp)
470 u_int16_t ndx, ovfl_num;
474 sp = (u_int16_t *)bufp->page;
476 /* Check if we are dynamically determining the fill factor */
477 if (hashp->FFACTOR == DEF_FFACTOR) {
478 hashp->FFACTOR = sp[0] >> 1;
479 if (hashp->FFACTOR < MIN_FFACTOR)
480 hashp->FFACTOR = MIN_FFACTOR;
482 bufp->flags |= BUF_MOD;
483 ovfl_num = overflow_page(hashp);
486 tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0;
488 if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1)))
490 bufp->ovfl->flags |= BUF_MOD;
492 (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n",
493 tmp1, tmp2, bufp->ovfl->addr);
497 * Since a pair is allocated on a page only if there's room to add
498 * an overflow page, we know that the OVFL information will fit on
501 sp[ndx + 4] = OFFSET(sp);
502 sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE;
503 sp[ndx + 1] = ovfl_num;
504 sp[ndx + 2] = OVFLPAGE;
506 #ifdef HASH_STATISTICS
514 * 0 indicates SUCCESS
515 * -1 indicates FAILURE
518 __get_page(hashp, p, bucket, is_bucket, is_disk, is_bitmap)
522 int is_bucket, is_disk, is_bitmap;
531 if ((fd == -1) || !is_disk) {
536 page = BUCKET_TO_PAGE(bucket);
538 page = OADDR_TO_PAGE(bucket);
539 if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) ||
540 ((rsize = _read(fd, p, size)) == -1))
544 bp[0] = 0; /* We hit the EOF, so initialize a new page */
550 if (!is_bitmap && !bp[0]) {
553 if (hashp->LORDER != BYTE_ORDER) {
557 max = hashp->BSIZE >> 2; /* divide by 4 */
558 for (i = 0; i < max; i++)
559 M_32_SWAP(((int *)p)[i]);
563 for (i = 1; i <= max; i++)
571 * Write page p to disk
578 __put_page(hashp, p, bucket, is_bucket, is_bitmap)
582 int is_bucket, is_bitmap;
588 if ((hashp->fp == -1) && open_temp(hashp))
592 if (hashp->LORDER != BYTE_ORDER) {
597 max = hashp->BSIZE >> 2; /* divide by 4 */
598 for (i = 0; i < max; i++)
599 M_32_SWAP(((int *)p)[i]);
601 max = ((u_int16_t *)p)[0] + 2;
602 for (i = 0; i <= max; i++)
603 M_16_SWAP(((u_int16_t *)p)[i]);
607 page = BUCKET_TO_PAGE(bucket);
609 page = OADDR_TO_PAGE(bucket);
610 if ((lseek(fd, (off_t)page << hashp->BSHIFT, SEEK_SET) == -1) ||
611 ((wsize = _write(fd, p, size)) == -1))
621 #define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1)
623 * Initialize a new bitmap page. Bitmap pages are left in memory
624 * once they are read in.
627 __ibitmap(hashp, pnum, nbits, ndx)
629 int pnum, nbits, ndx;
632 int clearbytes, clearints;
634 if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
637 clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1;
638 clearbytes = clearints << INT_TO_BYTE;
639 (void)memset((char *)ip, 0, clearbytes);
640 (void)memset(((char *)ip) + clearbytes, 0xFF,
641 hashp->BSIZE - clearbytes);
642 ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK);
644 hashp->BITMAPS[ndx] = (u_int16_t)pnum;
645 hashp->mapp[ndx] = ip;
656 for (i = 0; i < BITS_PER_MAP; i++) {
669 int max_free, offset, splitnum;
671 int bit, first_page, free_bit, free_page, i, in_use_bits, j;
675 splitnum = hashp->OVFL_POINT;
676 max_free = hashp->SPARES[splitnum];
678 free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
679 free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);
681 /* Look through all the free maps to find the first free block */
682 first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT);
683 for ( i = first_page; i <= free_page; i++ ) {
684 if (!(freep = (u_int32_t *)hashp->mapp[i]) &&
685 !(freep = fetch_bitmap(hashp, i)))
688 in_use_bits = free_bit;
690 in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;
692 if (i == first_page) {
693 bit = hashp->LAST_FREED &
694 ((hashp->BSIZE << BYTE_SHIFT) - 1);
695 j = bit / BITS_PER_MAP;
696 bit = bit & ~(BITS_PER_MAP - 1);
701 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
702 if (freep[j] != ALL_SET)
706 /* No Free Page Found */
707 hashp->LAST_FREED = hashp->SPARES[splitnum];
708 hashp->SPARES[splitnum]++;
709 offset = hashp->SPARES[splitnum] -
710 (splitnum ? hashp->SPARES[splitnum - 1] : 0);
712 #define OVMSG "HASH: Out of overflow pages. Increase page size\n"
713 if (offset > SPLITMASK) {
714 if (++splitnum >= NCACHED) {
715 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
718 hashp->OVFL_POINT = splitnum;
719 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
720 hashp->SPARES[splitnum-1]--;
724 /* Check if we need to allocate a new bitmap page */
725 if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
727 if (free_page >= NCACHED) {
728 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
732 * This is tricky. The 1 indicates that you want the new page
733 * allocated with 1 clear bit. Actually, you are going to
734 * allocate 2 pages from this map. The first is going to be
735 * the map page, the second is the overflow page we were
736 * looking for. The init_bitmap routine automatically, sets
737 * the first bit of itself to indicate that the bitmap itself
738 * is in use. We would explicitly set the second bit, but
739 * don't have to if we tell init_bitmap not to leave it clear
740 * in the first place.
743 (int)OADDR_OF(splitnum, offset), 1, free_page))
745 hashp->SPARES[splitnum]++;
750 if (offset > SPLITMASK) {
751 if (++splitnum >= NCACHED) {
752 (void)_write(STDERR_FILENO, OVMSG,
756 hashp->OVFL_POINT = splitnum;
757 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
758 hashp->SPARES[splitnum-1]--;
763 * Free_bit addresses the last used bit. Bump it to address
764 * the first available bit.
767 SETBIT(freep, free_bit);
770 /* Calculate address of the new overflow page */
771 addr = OADDR_OF(splitnum, offset);
773 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
774 addr, free_bit, free_page);
779 bit = bit + first_free(freep[j]);
786 * Bits are addressed starting with 0, but overflow pages are addressed
787 * beginning at 1. Bit is a bit addressnumber, so we need to increment
788 * it to convert it to a page number.
790 bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT));
791 if (bit >= hashp->LAST_FREED)
792 hashp->LAST_FREED = bit - 1;
794 /* Calculate the split number for this page */
795 for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++);
796 offset = (i ? bit - hashp->SPARES[i - 1] : bit);
797 if (offset >= SPLITMASK)
798 return (0); /* Out of overflow pages */
799 addr = OADDR_OF(i, offset);
801 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
805 /* Allocate and return the overflow page */
810 * Mark this overflow page as free.
813 __free_ovflpage(hashp, obufp)
819 int bit_address, free_page, free_bit;
824 (void)fprintf(stderr, "Freeing %d\n", addr);
826 ndx = (((u_int16_t)addr) >> SPLITSHIFT);
828 (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1;
829 if (bit_address < hashp->LAST_FREED)
830 hashp->LAST_FREED = bit_address;
831 free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT));
832 free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1);
834 if (!(freep = hashp->mapp[free_page]))
835 freep = fetch_bitmap(hashp, free_page);
838 * This had better never happen. It means we tried to read a bitmap
839 * that has already had overflow pages allocated off it, and we
840 * failed to read it from the file.
845 CLRBIT(freep, free_bit);
847 (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
848 obufp->addr, free_bit, free_page);
850 __reclaim_buf(hashp, obufp);
863 static char namestr[] = "_hashXXXXXX";
865 /* Block signals; make sure file goes away at process exit. */
866 (void)sigfillset(&set);
867 (void)_sigprocmask(SIG_BLOCK, &set, &oset);
868 if ((hashp->fp = mkstemp(namestr)) != -1) {
869 (void)unlink(namestr);
870 (void)_fcntl(hashp->fp, F_SETFD, 1);
872 (void)_sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL);
873 return (hashp->fp != -1 ? 0 : -1);
877 * We have to know that the key will fit, but the last entry on the page is
878 * an overflow pair, so we need to shift things.
881 squeeze_key(sp, key, val)
883 const DBT *key, *val;
886 u_int16_t free_space, n, off, pageno;
890 free_space = FREESPACE(sp);
896 memmove(p + off, key->data, key->size);
899 memmove(p + off, val->data, val->size);
902 sp[n + 2] = OVFLPAGE;
903 FREESPACE(sp) = free_space - PAIRSIZE(key, val);
908 fetch_bitmap(hashp, ndx)
912 if (ndx >= hashp->nmaps)
914 if ((hashp->mapp[ndx] = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
916 if (__get_page(hashp,
917 (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) {
918 free(hashp->mapp[ndx]);
921 return (hashp->mapp[ndx]);
932 (void)fprintf(stderr, "%d ", addr);
933 bufp = __get_buf(hashp, addr, NULL, 0);
934 bp = (short *)bufp->page;
935 while (bp[0] && ((bp[bp[0]] == OVFLPAGE) ||
936 ((bp[0] > 2) && bp[2] < REAL_KEY))) {
937 oaddr = bp[bp[0] - 1];
938 (void)fprintf(stderr, "%d ", (int)oaddr);
939 bufp = __get_buf(hashp, (int)oaddr, bufp, 0);
940 bp = (short *)bufp->page;
942 (void)fprintf(stderr, "\n");