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"
69 #include "libc_private.h"
76 static u_int32_t *fetch_bitmap(HTAB *, int);
77 static u_int32_t first_free(u_int32_t);
78 static int open_temp(HTAB *);
79 static u_int16_t overflow_page(HTAB *);
80 static void putpair(char *, const DBT *, const DBT *);
81 static void squeeze_key(u_int16_t *, const DBT *, const DBT *);
82 static int ugly_split(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
96 putpair(char *p, const DBT *key, const DBT *val)
98 u_int16_t *bp, n, off;
102 /* Enter the key first. */
105 off = OFFSET(bp) - key->size;
106 memmove(p + off, key->data, key->size);
111 memmove(p + off, val->data, val->size);
114 /* Adjust page info. */
116 bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t));
126 __delpair(HTAB *hashp, BUFHEAD *bufp, int ndx)
128 u_int16_t *bp, newoff, pairlen;
131 bp = (u_int16_t *)bufp->page;
134 if (bp[ndx + 1] < REAL_KEY)
135 return (__big_delete(hashp, bufp));
137 newoff = bp[ndx - 1];
139 newoff = hashp->BSIZE;
140 pairlen = newoff - bp[ndx + 1];
142 if (ndx != (n - 1)) {
143 /* Hard Case -- need to shuffle keys */
145 char *src = bufp->page + (int)OFFSET(bp);
146 char *dst = src + (int)pairlen;
147 memmove(dst, src, bp[ndx + 1] - OFFSET(bp));
149 /* Now adjust the pointers */
150 for (i = ndx + 2; i <= n; i += 2) {
151 if (bp[i + 1] == OVFLPAGE) {
153 bp[i - 1] = bp[i + 1];
155 bp[i - 2] = bp[i] + pairlen;
156 bp[i - 1] = bp[i + 1] + pairlen;
159 if (ndx == hashp->cndx) {
161 * We just removed pair we were "pointing" to.
162 * By moving back the cndx we ensure subsequent
163 * hash_seq() calls won't skip over any entries.
168 /* Finally adjust the page data */
169 bp[n] = OFFSET(bp) + pairlen;
170 bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t);
174 bufp->flags |= BUF_MOD;
183 __split_page(HTAB *hashp, u_int32_t obucket, u_int32_t 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(HTAB *hashp,
278 u_int32_t obucket, /* Same as __split_page. */
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(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val)
403 bp = (u_int16_t *)bufp->page;
405 while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
407 if (bp[2] == FULL_KEY_DATA && bp[0] == 2)
408 /* This is the last page of a big key/data pair
409 and we need to add another page */
411 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
412 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
415 bp = (u_int16_t *)bufp->page;
416 } else if (bp[bp[0]] != OVFLPAGE) {
417 /* Short key/data pairs, no more pages */
420 /* Try to squeeze key on this page */
421 if (bp[2] >= REAL_KEY &&
422 FREESPACE(bp) >= PAIRSIZE(key, val)) {
423 squeeze_key(bp, key, val);
426 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
429 bp = (u_int16_t *)bufp->page;
433 if (PAIRFITS(bp, key, val))
434 putpair(bufp->page, key, val);
437 bufp = __add_ovflpage(hashp, bufp);
440 sop = (u_int16_t *)bufp->page;
442 if (PAIRFITS(sop, key, val))
443 putpair((char *)sop, key, val);
445 if (__big_insert(hashp, bufp, key, val))
449 bufp->flags |= BUF_MOD;
451 * If the average number of keys per bucket exceeds the fill factor,
456 (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
457 return (__expand_table(hashp));
468 __add_ovflpage(HTAB *hashp, BUFHEAD *bufp)
470 u_int16_t *sp, 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(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_disk,
521 int fd, page, size, rsize;
527 if ((fd == -1) || !is_disk) {
532 page = BUCKET_TO_PAGE(bucket);
534 page = OADDR_TO_PAGE(bucket);
535 if ((rsize = pread(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
539 bp[0] = 0; /* We hit the EOF, so initialize a new page */
545 if (!is_bitmap && !bp[0]) {
548 if (hashp->LORDER != BYTE_ORDER) {
552 max = hashp->BSIZE >> 2; /* divide by 4 */
553 for (i = 0; i < max; i++)
554 M_32_SWAP(((int *)p)[i]);
558 for (i = 1; i <= max; i++)
566 * Write page p to disk
573 __put_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_bitmap)
575 int fd, page, size, wsize;
578 if ((hashp->fp == -1) && open_temp(hashp))
582 if (hashp->LORDER != BYTE_ORDER) {
586 max = hashp->BSIZE >> 2; /* divide by 4 */
587 for (i = 0; i < max; i++)
588 M_32_SWAP(((int *)p)[i]);
590 max = ((u_int16_t *)p)[0] + 2;
591 for (i = 0; i <= max; i++)
592 M_16_SWAP(((u_int16_t *)p)[i]);
596 page = BUCKET_TO_PAGE(bucket);
598 page = OADDR_TO_PAGE(bucket);
599 if ((wsize = pwrite(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
609 #define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1)
611 * Initialize a new bitmap page. Bitmap pages are left in memory
612 * once they are read in.
615 __ibitmap(HTAB *hashp, int pnum, int nbits, int ndx)
618 int clearbytes, clearints;
620 if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
623 clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1;
624 clearbytes = clearints << INT_TO_BYTE;
625 (void)memset((char *)ip, 0, clearbytes);
626 (void)memset(((char *)ip) + clearbytes, 0xFF,
627 hashp->BSIZE - clearbytes);
628 ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK);
630 hashp->BITMAPS[ndx] = (u_int16_t)pnum;
631 hashp->mapp[ndx] = ip;
636 first_free(u_int32_t map)
641 for (i = 0; i < BITS_PER_MAP; i++) {
650 overflow_page(HTAB *hashp)
653 int max_free, offset, splitnum;
655 int bit, first_page, free_bit, free_page, i, in_use_bits, j;
659 splitnum = hashp->OVFL_POINT;
660 max_free = hashp->SPARES[splitnum];
662 free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
663 free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);
665 /* Look through all the free maps to find the first free block */
666 first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT);
667 for ( i = first_page; i <= free_page; i++ ) {
668 if (!(freep = (u_int32_t *)hashp->mapp[i]) &&
669 !(freep = fetch_bitmap(hashp, i)))
672 in_use_bits = free_bit;
674 in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;
676 if (i == first_page) {
677 bit = hashp->LAST_FREED &
678 ((hashp->BSIZE << BYTE_SHIFT) - 1);
679 j = bit / BITS_PER_MAP;
680 bit = bit & ~(BITS_PER_MAP - 1);
685 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
686 if (freep[j] != ALL_SET)
690 /* No Free Page Found */
691 hashp->LAST_FREED = hashp->SPARES[splitnum];
692 hashp->SPARES[splitnum]++;
693 offset = hashp->SPARES[splitnum] -
694 (splitnum ? hashp->SPARES[splitnum - 1] : 0);
696 #define OVMSG "HASH: Out of overflow pages. Increase page size\n"
697 if (offset > SPLITMASK) {
698 if (++splitnum >= NCACHED) {
699 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
703 hashp->OVFL_POINT = splitnum;
704 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
705 hashp->SPARES[splitnum-1]--;
709 /* Check if we need to allocate a new bitmap page */
710 if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
712 if (free_page >= NCACHED) {
713 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
718 * This is tricky. The 1 indicates that you want the new page
719 * allocated with 1 clear bit. Actually, you are going to
720 * allocate 2 pages from this map. The first is going to be
721 * the map page, the second is the overflow page we were
722 * looking for. The init_bitmap routine automatically, sets
723 * the first bit of itself to indicate that the bitmap itself
724 * is in use. We would explicitly set the second bit, but
725 * don't have to if we tell init_bitmap not to leave it clear
726 * in the first place.
729 (int)OADDR_OF(splitnum, offset), 1, free_page))
731 hashp->SPARES[splitnum]++;
736 if (offset > SPLITMASK) {
737 if (++splitnum >= NCACHED) {
738 (void)_write(STDERR_FILENO, OVMSG,
743 hashp->OVFL_POINT = splitnum;
744 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
745 hashp->SPARES[splitnum-1]--;
750 * Free_bit addresses the last used bit. Bump it to address
751 * the first available bit.
754 SETBIT(freep, free_bit);
757 /* Calculate address of the new overflow page */
758 addr = OADDR_OF(splitnum, offset);
760 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
761 addr, free_bit, free_page);
766 bit = bit + first_free(freep[j]);
773 * Bits are addressed starting with 0, but overflow pages are addressed
774 * beginning at 1. Bit is a bit addressnumber, so we need to increment
775 * it to convert it to a page number.
777 bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT));
778 if (bit >= hashp->LAST_FREED)
779 hashp->LAST_FREED = bit - 1;
781 /* Calculate the split number for this page */
782 for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++);
783 offset = (i ? bit - hashp->SPARES[i - 1] : bit);
784 if (offset >= SPLITMASK) {
785 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
787 return (0); /* Out of overflow pages */
789 addr = OADDR_OF(i, offset);
791 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
795 /* Allocate and return the overflow page */
800 * Mark this overflow page as free.
803 __free_ovflpage(HTAB *hashp, BUFHEAD *obufp)
807 int bit_address, free_page, free_bit;
812 (void)fprintf(stderr, "Freeing %d\n", addr);
814 ndx = (((u_int16_t)addr) >> SPLITSHIFT);
816 (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1;
817 if (bit_address < hashp->LAST_FREED)
818 hashp->LAST_FREED = bit_address;
819 free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT));
820 free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1);
822 if (!(freep = hashp->mapp[free_page]))
823 freep = fetch_bitmap(hashp, free_page);
826 * This had better never happen. It means we tried to read a bitmap
827 * that has already had overflow pages allocated off it, and we
828 * failed to read it from the file.
833 CLRBIT(freep, free_bit);
835 (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
836 obufp->addr, free_bit, free_page);
838 __reclaim_buf(hashp, obufp);
847 open_temp(HTAB *hashp)
852 char path[MAXPATHLEN];
854 if (issetugid() == 0)
855 envtmp = getenv("TMPDIR");
857 sizeof(path), "%s/_hash.XXXXXX", envtmp ? envtmp : "/tmp");
858 if (len < 0 || len >= (int)sizeof(path)) {
859 errno = ENAMETOOLONG;
863 /* Block signals; make sure file goes away at process exit. */
864 (void)sigfillset(&set);
865 (void)__libc_sigprocmask(SIG_BLOCK, &set, &oset);
866 if ((hashp->fp = mkostemp(path, O_CLOEXEC)) != -1)
868 (void)__libc_sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL);
869 return (hashp->fp != -1 ? 0 : -1);
873 * We have to know that the key will fit, but the last entry on the page is
874 * an overflow pair, so we need to shift things.
877 squeeze_key(u_int16_t *sp, const DBT *key, const DBT *val)
880 u_int16_t free_space, n, off, pageno;
884 free_space = FREESPACE(sp);
890 memmove(p + off, key->data, key->size);
893 memmove(p + off, val->data, val->size);
896 sp[n + 2] = OVFLPAGE;
897 FREESPACE(sp) = free_space - PAIRSIZE(key, val);
902 fetch_bitmap(HTAB *hashp, int ndx)
904 if (ndx >= hashp->nmaps)
906 if ((hashp->mapp[ndx] = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
908 if (__get_page(hashp,
909 (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) {
910 free(hashp->mapp[ndx]);
913 return (hashp->mapp[ndx]);
918 print_chain(int addr)
923 (void)fprintf(stderr, "%d ", addr);
924 bufp = __get_buf(hashp, addr, NULL, 0);
925 bp = (short *)bufp->page;
926 while (bp[0] && ((bp[bp[0]] == OVFLPAGE) ||
927 ((bp[0] > 2) && bp[2] < REAL_KEY))) {
928 oaddr = bp[bp[0] - 1];
929 (void)fprintf(stderr, "%d ", (int)oaddr);
930 bufp = __get_buf(hashp, (int)oaddr, bufp, 0);
931 bp = (short *)bufp->page;
933 (void)fprintf(stderr, "\n");