2 * SPDX-License-Identifier: BSD-3-Clause
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
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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
35 #if defined(LIBC_SCCS) && !defined(lint)
36 static char sccsid[] = "@(#)hash_page.c 8.7 (Berkeley) 8/16/94";
37 #endif /* LIBC_SCCS and not lint */
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
45 * Page manipulation for hashing package.
57 #include "namespace.h"
58 #include <sys/param.h>
70 #include "un-namespace.h"
71 #include "libc_private.h"
78 static u_int32_t *fetch_bitmap(HTAB *, int);
79 static u_int32_t first_free(u_int32_t);
80 static int open_temp(HTAB *);
81 static u_int16_t overflow_page(HTAB *);
82 static void putpair(char *, const DBT *, const DBT *);
83 static void squeeze_key(u_int16_t *, const DBT *, const DBT *);
84 static int ugly_split(HTAB *, u_int32_t, BUFHEAD *, BUFHEAD *, int, int);
86 #define PAGE_INIT(P) { \
87 ((u_int16_t *)(P))[0] = 0; \
88 ((u_int16_t *)(P))[1] = hashp->BSIZE - 3 * sizeof(u_int16_t); \
89 ((u_int16_t *)(P))[2] = hashp->BSIZE; \
93 * This is called AFTER we have verified that there is room on the page for
94 * the pair (PAIRFITS has returned true) so we go right ahead and start moving
98 putpair(char *p, const DBT *key, const DBT *val)
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(HTAB *hashp, BUFHEAD *bufp, int ndx)
130 u_int16_t *bp, newoff, pairlen;
133 bp = (u_int16_t *)bufp->page;
136 if (bp[ndx + 1] < REAL_KEY)
137 return (__big_delete(hashp, bufp));
139 newoff = bp[ndx - 1];
141 newoff = hashp->BSIZE;
142 pairlen = newoff - bp[ndx + 1];
144 if (ndx != (n - 1)) {
145 /* Hard Case -- need to shuffle keys */
147 char *src = bufp->page + (int)OFFSET(bp);
148 char *dst = src + (int)pairlen;
149 memmove(dst, src, bp[ndx + 1] - OFFSET(bp));
151 /* Now adjust the pointers */
152 for (i = ndx + 2; i <= n; i += 2) {
153 if (bp[i + 1] == OVFLPAGE) {
155 bp[i - 1] = bp[i + 1];
157 bp[i - 2] = bp[i] + pairlen;
158 bp[i - 1] = bp[i + 1] + pairlen;
161 if (ndx == hashp->cndx) {
163 * We just removed pair we were "pointing" to.
164 * By moving back the cndx we ensure subsequent
165 * hash_seq() calls won't skip over any entries.
170 /* Finally adjust the page data */
171 bp[n] = OFFSET(bp) + pairlen;
172 bp[n - 1] = bp[n + 1] + pairlen + 2 * sizeof(u_int16_t);
176 bufp->flags |= BUF_MOD;
185 __split_page(HTAB *hashp, u_int32_t obucket, u_int32_t nbucket)
187 BUFHEAD *new_bufp, *old_bufp;
192 u_int16_t copyto, diff, off, moved;
195 copyto = (u_int16_t)hashp->BSIZE;
196 off = (u_int16_t)hashp->BSIZE;
197 old_bufp = __get_buf(hashp, obucket, NULL, 0);
198 if (old_bufp == NULL)
200 new_bufp = __get_buf(hashp, nbucket, NULL, 0);
201 if (new_bufp == NULL)
204 old_bufp->flags |= (BUF_MOD | BUF_PIN);
205 new_bufp->flags |= (BUF_MOD | BUF_PIN);
207 ino = (u_int16_t *)(op = old_bufp->page);
212 for (n = 1, ndx = 1; n < ino[0]; n += 2) {
213 if (ino[n + 1] < REAL_KEY) {
214 retval = ugly_split(hashp, obucket, old_bufp, new_bufp,
215 (int)copyto, (int)moved);
216 old_bufp->flags &= ~BUF_PIN;
217 new_bufp->flags &= ~BUF_PIN;
221 key.data = (u_char *)op + ino[n];
222 key.size = off - ino[n];
224 if (__call_hash(hashp, key.data, key.size) == obucket) {
225 /* Don't switch page */
228 copyto = ino[n + 1] + diff;
229 memmove(op + copyto, op + ino[n + 1],
231 ino[ndx] = copyto + ino[n] - ino[n + 1];
232 ino[ndx + 1] = copyto;
238 val.data = (u_char *)op + ino[n + 1];
239 val.size = ino[n] - ino[n + 1];
240 putpair(np, &key, &val);
247 /* Now clean up the page */
249 FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3);
250 OFFSET(ino) = copyto;
253 (void)fprintf(stderr, "split %d/%d\n",
254 ((u_int16_t *)np)[0] / 2,
255 ((u_int16_t *)op)[0] / 2);
257 /* unpin both pages */
258 old_bufp->flags &= ~BUF_PIN;
259 new_bufp->flags &= ~BUF_PIN;
264 * Called when we encounter an overflow or big key/data page during split
265 * handling. This is special cased since we have to begin checking whether
266 * the key/data pairs fit on their respective pages and because we may need
267 * overflow pages for both the old and new pages.
269 * The first page might be a page with regular key/data pairs in which case
270 * we have a regular overflow condition and just need to go on to the next
271 * page or it might be a big key/data pair in which case we need to fix the
279 ugly_split(HTAB *hashp,
280 u_int32_t obucket, /* Same as __split_page. */
283 int copyto, /* First byte on page which contains key/data values. */
284 int moved) /* Number of pairs moved to new page. */
286 BUFHEAD *bufp; /* Buffer header for ino */
287 u_int16_t *ino; /* Page keys come off of */
288 u_int16_t *np; /* New page */
289 u_int16_t *op; /* Page keys go on to if they aren't moving */
291 BUFHEAD *last_bfp; /* Last buf header OVFL needing to be freed */
294 u_int16_t n, off, ov_addr, scopyto;
295 char *cino; /* Character value of ino */
298 ino = (u_int16_t *)old_bufp->page;
299 np = (u_int16_t *)new_bufp->page;
300 op = (u_int16_t *)old_bufp->page;
302 scopyto = (u_int16_t)copyto; /* ANSI */
306 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) {
307 if (__big_split(hashp, old_bufp,
308 new_bufp, bufp, bufp->addr, obucket, &ret))
313 op = (u_int16_t *)old_bufp->page;
317 np = (u_int16_t *)new_bufp->page;
321 cino = (char *)bufp->page;
322 ino = (u_int16_t *)cino;
323 last_bfp = ret.nextp;
324 } else if (ino[n + 1] == OVFLPAGE) {
327 * Fix up the old page -- the extra 2 are the fields
328 * which contained the overflow information.
330 ino[0] -= (moved + 2);
332 scopyto - sizeof(u_int16_t) * (ino[0] + 3);
333 OFFSET(ino) = scopyto;
335 bufp = __get_buf(hashp, ov_addr, bufp, 0);
339 ino = (u_int16_t *)bufp->page;
341 scopyto = hashp->BSIZE;
345 __free_ovflpage(hashp, last_bfp);
348 /* Move regular sized pairs of there are any */
350 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) {
352 key.data = (u_char *)cino + ino[n];
353 key.size = off - ino[n];
354 val.data = (u_char *)cino + ino[n + 1];
355 val.size = ino[n] - ino[n + 1];
358 if (__call_hash(hashp, key.data, key.size) == obucket) {
359 /* Keep on old page */
360 if (PAIRFITS(op, (&key), (&val)))
361 putpair((char *)op, &key, &val);
364 __add_ovflpage(hashp, old_bufp);
367 op = (u_int16_t *)old_bufp->page;
368 putpair((char *)op, &key, &val);
370 old_bufp->flags |= BUF_MOD;
372 /* Move to new page */
373 if (PAIRFITS(np, (&key), (&val)))
374 putpair((char *)np, &key, &val);
377 __add_ovflpage(hashp, new_bufp);
380 np = (u_int16_t *)new_bufp->page;
381 putpair((char *)np, &key, &val);
383 new_bufp->flags |= BUF_MOD;
388 __free_ovflpage(hashp, last_bfp);
393 * Add the given pair to the page
400 __addel(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val)
405 bp = (u_int16_t *)bufp->page;
407 while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
409 if (bp[2] == FULL_KEY_DATA && bp[0] == 2)
410 /* This is the last page of a big key/data pair
411 and we need to add another page */
413 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
414 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
417 bp = (u_int16_t *)bufp->page;
418 } else if (bp[bp[0]] != OVFLPAGE) {
419 /* Short key/data pairs, no more pages */
422 /* Try to squeeze key on this page */
423 if (bp[2] >= REAL_KEY &&
424 FREESPACE(bp) >= PAIRSIZE(key, val)) {
425 squeeze_key(bp, key, val);
428 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
431 bp = (u_int16_t *)bufp->page;
435 if (PAIRFITS(bp, key, val))
436 putpair(bufp->page, key, val);
439 bufp = __add_ovflpage(hashp, bufp);
442 sop = (u_int16_t *)bufp->page;
444 if (PAIRFITS(sop, key, val))
445 putpair((char *)sop, key, val);
447 if (__big_insert(hashp, bufp, key, val))
451 bufp->flags |= BUF_MOD;
453 * If the average number of keys per bucket exceeds the fill factor,
458 (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
459 return (__expand_table(hashp));
470 __add_ovflpage(HTAB *hashp, BUFHEAD *bufp)
472 u_int16_t *sp, ndx, ovfl_num;
476 sp = (u_int16_t *)bufp->page;
478 /* Check if we are dynamically determining the fill factor */
479 if (hashp->FFACTOR == DEF_FFACTOR) {
480 hashp->FFACTOR = sp[0] >> 1;
481 if (hashp->FFACTOR < MIN_FFACTOR)
482 hashp->FFACTOR = MIN_FFACTOR;
484 bufp->flags |= BUF_MOD;
485 ovfl_num = overflow_page(hashp);
488 tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0;
490 if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1)))
492 bufp->ovfl->flags |= BUF_MOD;
494 (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n",
495 tmp1, tmp2, bufp->ovfl->addr);
499 * Since a pair is allocated on a page only if there's room to add
500 * an overflow page, we know that the OVFL information will fit on
503 sp[ndx + 4] = OFFSET(sp);
504 sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE;
505 sp[ndx + 1] = ovfl_num;
506 sp[ndx + 2] = OVFLPAGE;
508 #ifdef HASH_STATISTICS
516 * 0 indicates SUCCESS
517 * -1 indicates FAILURE
520 __get_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_disk,
523 int fd, page, size, rsize;
529 if ((fd == -1) || !is_disk) {
534 page = BUCKET_TO_PAGE(bucket);
536 page = OADDR_TO_PAGE(bucket);
537 if ((rsize = pread(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
541 bp[0] = 0; /* We hit the EOF, so initialize a new page */
547 if (!is_bitmap && !bp[0]) {
550 if (hashp->LORDER != BYTE_ORDER) {
554 max = hashp->BSIZE >> 2; /* divide by 4 */
555 for (i = 0; i < max; i++)
556 M_32_SWAP(((int *)p)[i]);
560 for (i = 1; i <= max; i++)
568 * Write page p to disk
575 __put_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_bitmap)
579 char pbuf[MAX_BSIZE];
582 if ((hashp->fp == -1) && open_temp(hashp))
586 if (hashp->LORDER != BYTE_ORDER) {
589 memcpy(pbuf, p, size);
591 max = hashp->BSIZE >> 2; /* divide by 4 */
592 for (i = 0; i < max; i++)
593 M_32_SWAP(((int *)pbuf)[i]);
595 uint16_t *bp = (uint16_t *)(void *)pbuf;
597 for (i = 0; i <= max; i++)
603 page = BUCKET_TO_PAGE(bucket);
605 page = OADDR_TO_PAGE(bucket);
606 if ((wsize = pwrite(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
616 #define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1)
618 * Initialize a new bitmap page. Bitmap pages are left in memory
619 * once they are read in.
622 __ibitmap(HTAB *hashp, int pnum, int nbits, int ndx)
625 int clearbytes, clearints;
627 if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
630 clearints = ((nbits - 1) >> INT_BYTE_SHIFT) + 1;
631 clearbytes = clearints << INT_TO_BYTE;
632 (void)memset((char *)ip, 0, clearbytes);
633 (void)memset(((char *)ip) + clearbytes, 0xFF,
634 hashp->BSIZE - clearbytes);
635 ip[clearints - 1] = ALL_SET << (nbits & BYTE_MASK);
637 hashp->BITMAPS[ndx] = (u_int16_t)pnum;
638 hashp->mapp[ndx] = ip;
643 first_free(u_int32_t map)
648 for (i = 0; i < BITS_PER_MAP; i++) {
657 overflow_page(HTAB *hashp)
660 int max_free, offset, splitnum;
662 int bit, first_page, free_bit, free_page, i, in_use_bits, j;
666 splitnum = hashp->OVFL_POINT;
667 max_free = hashp->SPARES[splitnum];
669 free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
670 free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);
672 /* Look through all the free maps to find the first free block */
673 first_page = hashp->LAST_FREED >>(hashp->BSHIFT + BYTE_SHIFT);
674 for ( i = first_page; i <= free_page; i++ ) {
675 if (!(freep = (u_int32_t *)hashp->mapp[i]) &&
676 !(freep = fetch_bitmap(hashp, i)))
679 in_use_bits = free_bit;
681 in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;
683 if (i == first_page) {
684 bit = hashp->LAST_FREED &
685 ((hashp->BSIZE << BYTE_SHIFT) - 1);
686 j = bit / BITS_PER_MAP;
687 bit = rounddown2(bit, BITS_PER_MAP);
692 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
693 if (freep[j] != ALL_SET)
697 /* No Free Page Found */
698 hashp->LAST_FREED = hashp->SPARES[splitnum];
699 hashp->SPARES[splitnum]++;
700 offset = hashp->SPARES[splitnum] -
701 (splitnum ? hashp->SPARES[splitnum - 1] : 0);
703 #define OVMSG "HASH: Out of overflow pages. Increase page size\n"
704 if (offset > SPLITMASK) {
705 if (++splitnum >= NCACHED) {
706 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
710 hashp->OVFL_POINT = splitnum;
711 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
712 hashp->SPARES[splitnum-1]--;
716 /* Check if we need to allocate a new bitmap page */
717 if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
719 if (free_page >= NCACHED) {
720 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
725 * This is tricky. The 1 indicates that you want the new page
726 * allocated with 1 clear bit. Actually, you are going to
727 * allocate 2 pages from this map. The first is going to be
728 * the map page, the second is the overflow page we were
729 * looking for. The init_bitmap routine automatically, sets
730 * the first bit of itself to indicate that the bitmap itself
731 * is in use. We would explicitly set the second bit, but
732 * don't have to if we tell init_bitmap not to leave it clear
733 * in the first place.
736 (int)OADDR_OF(splitnum, offset), 1, free_page))
738 hashp->SPARES[splitnum]++;
743 if (offset > SPLITMASK) {
744 if (++splitnum >= NCACHED) {
745 (void)_write(STDERR_FILENO, OVMSG,
750 hashp->OVFL_POINT = splitnum;
751 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
752 hashp->SPARES[splitnum-1]--;
757 * Free_bit addresses the last used bit. Bump it to address
758 * the first available bit.
761 SETBIT(freep, free_bit);
764 /* Calculate address of the new overflow page */
765 addr = OADDR_OF(splitnum, offset);
767 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
768 addr, free_bit, free_page);
773 bit = bit + first_free(freep[j]);
780 * Bits are addressed starting with 0, but overflow pages are addressed
781 * beginning at 1. Bit is a bit addressnumber, so we need to increment
782 * it to convert it to a page number.
784 bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT));
785 if (bit >= hashp->LAST_FREED)
786 hashp->LAST_FREED = bit - 1;
788 /* Calculate the split number for this page */
789 for (i = 0; (i < splitnum) && (bit > hashp->SPARES[i]); i++);
790 offset = (i ? bit - hashp->SPARES[i - 1] : bit);
791 if (offset >= SPLITMASK) {
792 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
794 return (0); /* Out of overflow pages */
796 addr = OADDR_OF(i, offset);
798 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
802 /* Allocate and return the overflow page */
807 * Mark this overflow page as free.
810 __free_ovflpage(HTAB *hashp, BUFHEAD *obufp)
814 int bit_address, free_page, free_bit;
819 (void)fprintf(stderr, "Freeing %d\n", addr);
821 ndx = (((u_int16_t)addr) >> SPLITSHIFT);
823 (ndx ? hashp->SPARES[ndx - 1] : 0) + (addr & SPLITMASK) - 1;
824 if (bit_address < hashp->LAST_FREED)
825 hashp->LAST_FREED = bit_address;
826 free_page = (bit_address >> (hashp->BSHIFT + BYTE_SHIFT));
827 free_bit = bit_address & ((hashp->BSIZE << BYTE_SHIFT) - 1);
829 if (!(freep = hashp->mapp[free_page]))
830 freep = fetch_bitmap(hashp, free_page);
833 * This had better never happen. It means we tried to read a bitmap
834 * that has already had overflow pages allocated off it, and we
835 * failed to read it from the file.
840 CLRBIT(freep, free_bit);
842 (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
843 obufp->addr, free_bit, free_page);
845 __reclaim_buf(hashp, obufp);
854 open_temp(HTAB *hashp)
859 char path[MAXPATHLEN];
861 if (issetugid() == 0)
862 envtmp = getenv("TMPDIR");
864 sizeof(path), "%s/_hash.XXXXXX", envtmp ? envtmp : "/tmp");
865 if (len < 0 || len >= (int)sizeof(path)) {
866 errno = ENAMETOOLONG;
870 /* Block signals; make sure file goes away at process exit. */
871 (void)sigfillset(&set);
872 (void)__libc_sigprocmask(SIG_BLOCK, &set, &oset);
873 if ((hashp->fp = mkostemp(path, O_CLOEXEC)) != -1)
875 (void)__libc_sigprocmask(SIG_SETMASK, &oset, (sigset_t *)NULL);
876 return (hashp->fp != -1 ? 0 : -1);
880 * We have to know that the key will fit, but the last entry on the page is
881 * an overflow pair, so we need to shift things.
884 squeeze_key(u_int16_t *sp, const DBT *key, const DBT *val)
887 u_int16_t free_space, n, off, pageno;
891 free_space = FREESPACE(sp);
897 memmove(p + off, key->data, key->size);
900 memmove(p + off, val->data, val->size);
903 sp[n + 2] = OVFLPAGE;
904 FREESPACE(sp) = free_space - PAIRSIZE(key, val);
909 fetch_bitmap(HTAB *hashp, int ndx)
911 if (ndx >= hashp->nmaps)
913 if ((hashp->mapp[ndx] = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
915 if (__get_page(hashp,
916 (char *)hashp->mapp[ndx], hashp->BITMAPS[ndx], 0, 1, 1)) {
917 free(hashp->mapp[ndx]);
920 return (hashp->mapp[ndx]);
925 print_chain(int addr)
930 (void)fprintf(stderr, "%d ", addr);
931 bufp = __get_buf(hashp, addr, NULL, 0);
932 bp = (short *)bufp->page;
933 while (bp[0] && ((bp[bp[0]] == OVFLPAGE) ||
934 ((bp[0] > 2) && bp[2] < REAL_KEY))) {
935 oaddr = bp[bp[0] - 1];
936 (void)fprintf(stderr, "%d ", (int)oaddr);
937 bufp = __get_buf(hashp, (int)oaddr, bufp, 0);
938 bp = (short *)bufp->page;
940 (void)fprintf(stderr, "\n");