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/param.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 *);
81 static int ugly_split(HTAB *, u_int32_t, BUFHEAD *, BUFHEAD *, int, int);
83 #define PAGE_INIT(P) { \
84 ((u_int16_t *)(P))[0] = 0; \
85 ((u_int16_t *)(P))[1] = hashp->BSIZE - 3 * sizeof(u_int16_t); \
86 ((u_int16_t *)(P))[2] = hashp->BSIZE; \
90 * This is called AFTER we have verified that there is room on the page for
91 * the pair (PAIRFITS has returned true) so we go right ahead and start moving
95 putpair(char *p, const DBT *key, const DBT *val)
97 u_int16_t *bp, n, off;
101 /* Enter the key first. */
104 off = OFFSET(bp) - key->size;
105 memmove(p + off, key->data, key->size);
110 memmove(p + off, val->data, val->size);
113 /* Adjust page info. */
115 bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t));
125 __delpair(HTAB *hashp, BUFHEAD *bufp, int ndx)
127 u_int16_t *bp, newoff, pairlen;
130 bp = (u_int16_t *)bufp->page;
133 if (bp[ndx + 1] < REAL_KEY)
134 return (__big_delete(hashp, bufp));
136 newoff = bp[ndx - 1];
138 newoff = hashp->BSIZE;
139 pairlen = newoff - bp[ndx + 1];
141 if (ndx != (n - 1)) {
142 /* Hard Case -- need to shuffle keys */
144 char *src = bufp->page + (int)OFFSET(bp);
145 char *dst = src + (int)pairlen;
146 memmove(dst, src, bp[ndx + 1] - OFFSET(bp));
148 /* Now adjust the pointers */
149 for (i = ndx + 2; i <= n; i += 2) {
150 if (bp[i + 1] == OVFLPAGE) {
152 bp[i - 1] = bp[i + 1];
154 bp[i - 2] = bp[i] + pairlen;
155 bp[i - 1] = bp[i + 1] + pairlen;
158 if (ndx == hashp->cndx) {
160 * We just removed pair we were "pointing" to.
161 * By moving back the cndx we ensure subsequent
162 * hash_seq() calls won't skip over any entries.
167 /* Finally adjust the page data */
168 bp[n] = OFFSET(bp) + pairlen;
169 bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t);
173 bufp->flags |= BUF_MOD;
182 __split_page(HTAB *hashp, u_int32_t obucket, u_int32_t nbucket)
184 BUFHEAD *new_bufp, *old_bufp;
189 u_int16_t copyto, diff, off, moved;
192 copyto = (u_int16_t)hashp->BSIZE;
193 off = (u_int16_t)hashp->BSIZE;
194 old_bufp = __get_buf(hashp, obucket, NULL, 0);
195 if (old_bufp == NULL)
197 new_bufp = __get_buf(hashp, nbucket, NULL, 0);
198 if (new_bufp == NULL)
201 old_bufp->flags |= (BUF_MOD | BUF_PIN);
202 new_bufp->flags |= (BUF_MOD | BUF_PIN);
204 ino = (u_int16_t *)(op = old_bufp->page);
209 for (n = 1, ndx = 1; n < ino[0]; n += 2) {
210 if (ino[n + 1] < REAL_KEY) {
211 retval = ugly_split(hashp, obucket, old_bufp, new_bufp,
212 (int)copyto, (int)moved);
213 old_bufp->flags &= ~BUF_PIN;
214 new_bufp->flags &= ~BUF_PIN;
218 key.data = (u_char *)op + ino[n];
219 key.size = off - ino[n];
221 if (__call_hash(hashp, key.data, key.size) == obucket) {
222 /* Don't switch page */
225 copyto = ino[n + 1] + diff;
226 memmove(op + copyto, op + ino[n + 1],
228 ino[ndx] = copyto + ino[n] - ino[n + 1];
229 ino[ndx + 1] = copyto;
235 val.data = (u_char *)op + ino[n + 1];
236 val.size = ino[n] - ino[n + 1];
237 putpair(np, &key, &val);
244 /* Now clean up the page */
246 FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3);
247 OFFSET(ino) = copyto;
250 (void)fprintf(stderr, "split %d/%d\n",
251 ((u_int16_t *)np)[0] / 2,
252 ((u_int16_t *)op)[0] / 2);
254 /* unpin both pages */
255 old_bufp->flags &= ~BUF_PIN;
256 new_bufp->flags &= ~BUF_PIN;
261 * Called when we encounter an overflow or big key/data page during split
262 * handling. This is special cased since we have to begin checking whether
263 * the key/data pairs fit on their respective pages and because we may need
264 * overflow pages for both the old and new pages.
266 * The first page might be a page with regular key/data pairs in which case
267 * we have a regular overflow condition and just need to go on to the next
268 * page or it might be a big key/data pair in which case we need to fix the
276 ugly_split(HTAB *hashp,
277 u_int32_t obucket, /* Same as __split_page. */
280 int copyto, /* First byte on page which contains key/data values. */
281 int moved) /* Number of pairs moved to new page. */
283 BUFHEAD *bufp; /* Buffer header for ino */
284 u_int16_t *ino; /* Page keys come off of */
285 u_int16_t *np; /* New page */
286 u_int16_t *op; /* Page keys go on to if they aren't moving */
288 BUFHEAD *last_bfp; /* Last buf header OVFL needing to be freed */
291 u_int16_t n, off, ov_addr, scopyto;
292 char *cino; /* Character value of ino */
295 ino = (u_int16_t *)old_bufp->page;
296 np = (u_int16_t *)new_bufp->page;
297 op = (u_int16_t *)old_bufp->page;
299 scopyto = (u_int16_t)copyto; /* ANSI */
303 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) {
304 if (__big_split(hashp, old_bufp,
305 new_bufp, bufp, bufp->addr, obucket, &ret))
310 op = (u_int16_t *)old_bufp->page;
314 np = (u_int16_t *)new_bufp->page;
318 cino = (char *)bufp->page;
319 ino = (u_int16_t *)cino;
320 last_bfp = ret.nextp;
321 } else if (ino[n + 1] == OVFLPAGE) {
324 * Fix up the old page -- the extra 2 are the fields
325 * which contained the overflow information.
327 ino[0] -= (moved + 2);
329 scopyto - sizeof(u_int16_t) * (ino[0] + 3);
330 OFFSET(ino) = scopyto;
332 bufp = __get_buf(hashp, ov_addr, bufp, 0);
336 ino = (u_int16_t *)bufp->page;
338 scopyto = hashp->BSIZE;
342 __free_ovflpage(hashp, last_bfp);
345 /* Move regular sized pairs of there are any */
347 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) {
349 key.data = (u_char *)cino + ino[n];
350 key.size = off - ino[n];
351 val.data = (u_char *)cino + ino[n + 1];
352 val.size = ino[n] - ino[n + 1];
355 if (__call_hash(hashp, key.data, key.size) == obucket) {
356 /* Keep on old page */
357 if (PAIRFITS(op, (&key), (&val)))
358 putpair((char *)op, &key, &val);
361 __add_ovflpage(hashp, old_bufp);
364 op = (u_int16_t *)old_bufp->page;
365 putpair((char *)op, &key, &val);
367 old_bufp->flags |= BUF_MOD;
369 /* Move to new page */
370 if (PAIRFITS(np, (&key), (&val)))
371 putpair((char *)np, &key, &val);
374 __add_ovflpage(hashp, new_bufp);
377 np = (u_int16_t *)new_bufp->page;
378 putpair((char *)np, &key, &val);
380 new_bufp->flags |= BUF_MOD;
385 __free_ovflpage(hashp, last_bfp);
390 * Add the given pair to the page
397 __addel(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val)
402 bp = (u_int16_t *)bufp->page;
404 while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
406 if (bp[2] == FULL_KEY_DATA && bp[0] == 2)
407 /* This is the last page of a big key/data pair
408 and we need to add another page */
410 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
411 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
414 bp = (u_int16_t *)bufp->page;
415 } else if (bp[bp[0]] != OVFLPAGE) {
416 /* Short key/data pairs, no more pages */
419 /* Try to squeeze key on this page */
420 if (bp[2] >= REAL_KEY &&
421 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;
432 if (PAIRFITS(bp, key, val))
433 putpair(bufp->page, key, val);
436 bufp = __add_ovflpage(hashp, bufp);
439 sop = (u_int16_t *)bufp->page;
441 if (PAIRFITS(sop, key, val))
442 putpair((char *)sop, key, val);
444 if (__big_insert(hashp, bufp, key, val))
448 bufp->flags |= BUF_MOD;
450 * If the average number of keys per bucket exceeds the fill factor,
455 (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
456 return (__expand_table(hashp));
467 __add_ovflpage(HTAB *hashp, BUFHEAD *bufp)
469 u_int16_t *sp, ndx, ovfl_num;
473 sp = (u_int16_t *)bufp->page;
475 /* Check if we are dynamically determining the fill factor */
476 if (hashp->FFACTOR == DEF_FFACTOR) {
477 hashp->FFACTOR = sp[0] >> 1;
478 if (hashp->FFACTOR < MIN_FFACTOR)
479 hashp->FFACTOR = MIN_FFACTOR;
481 bufp->flags |= BUF_MOD;
482 ovfl_num = overflow_page(hashp);
485 tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0;
487 if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1)))
489 bufp->ovfl->flags |= BUF_MOD;
491 (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n",
492 tmp1, tmp2, bufp->ovfl->addr);
496 * Since a pair is allocated on a page only if there's room to add
497 * an overflow page, we know that the OVFL information will fit on
500 sp[ndx + 4] = OFFSET(sp);
501 sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE;
502 sp[ndx + 1] = ovfl_num;
503 sp[ndx + 2] = OVFLPAGE;
505 #ifdef HASH_STATISTICS
513 * 0 indicates SUCCESS
514 * -1 indicates FAILURE
517 __get_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_disk,
520 int fd, page, size, rsize;
526 if ((fd == -1) || !is_disk) {
531 page = BUCKET_TO_PAGE(bucket);
533 page = OADDR_TO_PAGE(bucket);
534 if ((rsize = pread(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
538 bp[0] = 0; /* We hit the EOF, so initialize a new page */
544 if (!is_bitmap && !bp[0]) {
547 if (hashp->LORDER != BYTE_ORDER) {
551 max = hashp->BSIZE >> 2; /* divide by 4 */
552 for (i = 0; i < max; i++)
553 M_32_SWAP(((int *)p)[i]);
557 for (i = 1; i <= max; i++)
565 * Write page p to disk
572 __put_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_bitmap)
574 int fd, page, size, wsize;
577 if ((hashp->fp == -1) && open_temp(hashp))
581 if (hashp->LORDER != BYTE_ORDER) {
585 max = hashp->BSIZE >> 2; /* divide by 4 */
586 for (i = 0; i < max; i++)
587 M_32_SWAP(((int *)p)[i]);
589 max = ((u_int16_t *)p)[0] + 2;
590 for (i = 0; i <= max; i++)
591 M_16_SWAP(((u_int16_t *)p)[i]);
595 page = BUCKET_TO_PAGE(bucket);
597 page = OADDR_TO_PAGE(bucket);
598 if ((wsize = pwrite(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
608 #define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1)
610 * Initialize a new bitmap page. Bitmap pages are left in memory
611 * once they are read in.
614 __ibitmap(HTAB *hashp, int pnum, int nbits, int ndx)
617 int clearbytes, clearints;
619 if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
622 clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1;
623 clearbytes = clearints << INT_TO_BYTE;
624 (void)memset((char *)ip, 0, clearbytes);
625 (void)memset(((char *)ip) + clearbytes, 0xFF,
626 hashp->BSIZE - clearbytes);
627 ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK);
629 hashp->BITMAPS[ndx] = (u_int16_t)pnum;
630 hashp->mapp[ndx] = ip;
635 first_free(u_int32_t map)
640 for (i = 0; i < BITS_PER_MAP; i++) {
649 overflow_page(HTAB *hashp)
652 int max_free, offset, splitnum;
654 int bit, first_page, free_bit, free_page, i, in_use_bits, j;
658 splitnum = hashp->OVFL_POINT;
659 max_free = hashp->SPARES[splitnum];
661 free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
662 free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);
664 /* Look through all the free maps to find the first free block */
665 first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT);
666 for ( i = first_page; i <= free_page; i++ ) {
667 if (!(freep = (u_int32_t *)hashp->mapp[i]) &&
668 !(freep = fetch_bitmap(hashp, i)))
671 in_use_bits = free_bit;
673 in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;
675 if (i == first_page) {
676 bit = hashp->LAST_FREED &
677 ((hashp->BSIZE << BYTE_SHIFT) - 1);
678 j = bit / BITS_PER_MAP;
679 bit = bit & ~(BITS_PER_MAP - 1);
684 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
685 if (freep[j] != ALL_SET)
689 /* No Free Page Found */
690 hashp->LAST_FREED = hashp->SPARES[splitnum];
691 hashp->SPARES[splitnum]++;
692 offset = hashp->SPARES[splitnum] -
693 (splitnum ? hashp->SPARES[splitnum - 1] : 0);
695 #define OVMSG "HASH: Out of overflow pages. Increase page size\n"
696 if (offset > SPLITMASK) {
697 if (++splitnum >= NCACHED) {
698 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
702 hashp->OVFL_POINT = splitnum;
703 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
704 hashp->SPARES[splitnum-1]--;
708 /* Check if we need to allocate a new bitmap page */
709 if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
711 if (free_page >= NCACHED) {
712 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
717 * This is tricky. The 1 indicates that you want the new page
718 * allocated with 1 clear bit. Actually, you are going to
719 * allocate 2 pages from this map. The first is going to be
720 * the map page, the second is the overflow page we were
721 * looking for. The init_bitmap routine automatically, sets
722 * the first bit of itself to indicate that the bitmap itself
723 * is in use. We would explicitly set the second bit, but
724 * don't have to if we tell init_bitmap not to leave it clear
725 * in the first place.
728 (int)OADDR_OF(splitnum, offset), 1, free_page))
730 hashp->SPARES[splitnum]++;
735 if (offset > SPLITMASK) {
736 if (++splitnum >= NCACHED) {
737 (void)_write(STDERR_FILENO, OVMSG,
742 hashp->OVFL_POINT = splitnum;
743 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
744 hashp->SPARES[splitnum-1]--;
749 * Free_bit addresses the last used bit. Bump it to address
750 * the first available bit.
753 SETBIT(freep, free_bit);
756 /* Calculate address of the new overflow page */
757 addr = OADDR_OF(splitnum, offset);
759 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
760 addr, free_bit, free_page);
765 bit = bit + first_free(freep[j]);
772 * Bits are addressed starting with 0, but overflow pages are addressed
773 * beginning at 1. Bit is a bit addressnumber, so we need to increment
774 * it to convert it to a page number.
776 bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT));
777 if (bit >= hashp->LAST_FREED)
778 hashp->LAST_FREED = bit - 1;
780 /* Calculate the split number for this page */
781 for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++);
782 offset = (i ? bit - hashp->SPARES[i - 1] : bit);
783 if (offset >= SPLITMASK) {
784 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
786 return (0); /* Out of overflow pages */
788 addr = OADDR_OF(i, offset);
790 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
794 /* Allocate and return the overflow page */
799 * Mark this overflow page as free.
802 __free_ovflpage(HTAB *hashp, BUFHEAD *obufp)
806 int bit_address, free_page, free_bit;
811 (void)fprintf(stderr, "Freeing %d\n", addr);
813 ndx = (((u_int16_t)addr) >> SPLITSHIFT);
815 (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1;
816 if (bit_address < hashp->LAST_FREED)
817 hashp->LAST_FREED = bit_address;
818 free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT));
819 free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1);
821 if (!(freep = hashp->mapp[free_page]))
822 freep = fetch_bitmap(hashp, free_page);
825 * This had better never happen. It means we tried to read a bitmap
826 * that has already had overflow pages allocated off it, and we
827 * failed to read it from the file.
832 CLRBIT(freep, free_bit);
834 (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
835 obufp->addr, free_bit, free_page);
837 __reclaim_buf(hashp, obufp);
846 open_temp(HTAB *hashp)
851 char path[MAXPATHLEN];
853 if (issetugid() == 0)
854 envtmp = getenv("TMPDIR");
856 sizeof(path), "%s/_hash.XXXXXX", envtmp ? envtmp : "/tmp");
857 if (len < 0 || len >= (int)sizeof(path)) {
858 errno = ENAMETOOLONG;
862 /* Block signals; make sure file goes away at process exit. */
863 (void)sigfillset(&set);
864 (void)_sigprocmask(SIG_BLOCK, &set, &oset);
865 if ((hashp->fp = mkstemp(path)) != -1) {
867 (void)_fcntl(hashp->fp, F_SETFD, 1);
869 (void)_sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL);
870 return (hashp->fp != -1 ? 0 : -1);
874 * We have to know that the key will fit, but the last entry on the page is
875 * an overflow pair, so we need to shift things.
878 squeeze_key(u_int16_t *sp, const DBT *key, const DBT *val)
881 u_int16_t free_space, n, off, pageno;
885 free_space = FREESPACE(sp);
891 memmove(p + off, key->data, key->size);
894 memmove(p + off, val->data, val->size);
897 sp[n + 2] = OVFLPAGE;
898 FREESPACE(sp) = free_space - PAIRSIZE(key, val);
903 fetch_bitmap(HTAB *hashp, int ndx)
905 if (ndx >= hashp->nmaps)
907 if ((hashp->mapp[ndx] = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
909 if (__get_page(hashp,
910 (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) {
911 free(hashp->mapp[ndx]);
914 return (hashp->mapp[ndx]);
919 print_chain(int addr)
924 (void)fprintf(stderr, "%d ", addr);
925 bufp = __get_buf(hashp, addr, NULL, 0);
926 bp = (short *)bufp->page;
927 while (bp[0] && ((bp[bp[0]] == OVFLPAGE) ||
928 ((bp[0] > 2) && bp[2] < REAL_KEY))) {
929 oaddr = bp[bp[0] - 1];
930 (void)fprintf(stderr, "%d ", (int)oaddr);
931 bufp = __get_buf(hashp, (int)oaddr, bufp, 0);
932 bp = (short *)bufp->page;
934 (void)fprintf(stderr, "\n");