69
70#include <db.h>
71#include "hash.h"
72#include "page.h"
73#include "extern.h"
74
75static u_int32_t *fetch_bitmap(HTAB *, int);
76static u_int32_t first_free(u_int32_t);
77static int open_temp(HTAB *);
78static u_int16_t overflow_page(HTAB *);
79static void putpair(char *, const DBT *, const DBT *);
80static void squeeze_key(u_int16_t *, const DBT *, const DBT *);
81static int ugly_split(HTAB *, u_int32_t, BUFHEAD *, BUFHEAD *, int, int);
82
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; \
87}
88
89/*
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
92 * stuff on.
93 */
94static void
95putpair(char *p, const DBT *key, const DBT *val)
96{
97 u_int16_t *bp, n, off;
98
99 bp = (u_int16_t *)p;
100
101 /* Enter the key first. */
102 n = bp[0];
103
104 off = OFFSET(bp) - key->size;
105 memmove(p + off, key->data, key->size);
106 bp[++n] = off;
107
108 /* Now the data. */
109 off -= val->size;
110 memmove(p + off, val->data, val->size);
111 bp[++n] = off;
112
113 /* Adjust page info. */
114 bp[0] = n;
115 bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t));
116 bp[n + 2] = off;
117}
118
119/*
120 * Returns:
121 * 0 OK
122 * -1 error
123 */
124int
125__delpair(HTAB *hashp, BUFHEAD *bufp, int ndx)
126{
127 u_int16_t *bp, newoff, pairlen;
128 int n;
129
130 bp = (u_int16_t *)bufp->page;
131 n = bp[0];
132
133 if (bp[ndx + 1] < REAL_KEY)
134 return (__big_delete(hashp, bufp));
135 if (ndx != 1)
136 newoff = bp[ndx - 1];
137 else
138 newoff = hashp->BSIZE;
139 pairlen = newoff - bp[ndx + 1];
140
141 if (ndx != (n - 1)) {
142 /* Hard Case -- need to shuffle keys */
143 int i;
144 char *src = bufp->page + (int)OFFSET(bp);
145 char *dst = src + (int)pairlen;
146 memmove(dst, src, bp[ndx + 1] - OFFSET(bp));
147
148 /* Now adjust the pointers */
149 for (i = ndx + 2; i <= n; i += 2) {
150 if (bp[i + 1] == OVFLPAGE) {
151 bp[i - 2] = bp[i];
152 bp[i - 1] = bp[i + 1];
153 } else {
154 bp[i - 2] = bp[i] + pairlen;
155 bp[i - 1] = bp[i + 1] + pairlen;
156 }
157 }
158 if (ndx == hashp->cndx) {
159 /*
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.
163 */
164 hashp->cndx -= 2;
165 }
166 }
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);
170 bp[0] = n - 2;
171 hashp->NKEYS--;
172
173 bufp->flags |= BUF_MOD;
174 return (0);
175}
176/*
177 * Returns:
178 * 0 ==> OK
179 * -1 ==> Error
180 */
181int
182__split_page(HTAB *hashp, u_int32_t obucket, u_int32_t nbucket)
183{
184 BUFHEAD *new_bufp, *old_bufp;
185 u_int16_t *ino;
186 char *np;
187 DBT key, val;
188 int n, ndx, retval;
189 u_int16_t copyto, diff, off, moved;
190 char *op;
191
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)
196 return (-1);
197 new_bufp = __get_buf(hashp, nbucket, NULL, 0);
198 if (new_bufp == NULL)
199 return (-1);
200
201 old_bufp->flags |= (BUF_MOD | BUF_PIN);
202 new_bufp->flags |= (BUF_MOD | BUF_PIN);
203
204 ino = (u_int16_t *)(op = old_bufp->page);
205 np = new_bufp->page;
206
207 moved = 0;
208
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;
215 return (retval);
216
217 }
218 key.data = (u_char *)op + ino[n];
219 key.size = off - ino[n];
220
221 if (__call_hash(hashp, key.data, key.size) == obucket) {
222 /* Don't switch page */
223 diff = copyto - off;
224 if (diff) {
225 copyto = ino[n + 1] + diff;
226 memmove(op + copyto, op + ino[n + 1],
227 off - ino[n + 1]);
228 ino[ndx] = copyto + ino[n] - ino[n + 1];
229 ino[ndx + 1] = copyto;
230 } else
231 copyto = ino[n + 1];
232 ndx += 2;
233 } else {
234 /* Switch page */
235 val.data = (u_char *)op + ino[n + 1];
236 val.size = ino[n] - ino[n + 1];
237 putpair(np, &key, &val);
238 moved += 2;
239 }
240
241 off = ino[n + 1];
242 }
243
244 /* Now clean up the page */
245 ino[0] -= moved;
246 FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3);
247 OFFSET(ino) = copyto;
248
249#ifdef DEBUG3
250 (void)fprintf(stderr, "split %d/%d\n",
251 ((u_int16_t *)np)[0] / 2,
252 ((u_int16_t *)op)[0] / 2);
253#endif
254 /* unpin both pages */
255 old_bufp->flags &= ~BUF_PIN;
256 new_bufp->flags &= ~BUF_PIN;
257 return (0);
258}
259
260/*
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.
265 *
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
269 * big key/data pair.
270 *
271 * Returns:
272 * 0 ==> success
273 * -1 ==> failure
274 */
275static int
276ugly_split(HTAB *hashp,
277 u_int32_t obucket, /* Same as __split_page. */
278 BUFHEAD *old_bufp,
279 BUFHEAD *new_bufp,
280 int copyto, /* First byte on page which contains key/data values. */
281 int moved) /* Number of pairs moved to new page. */
282{
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 */
287
288 BUFHEAD *last_bfp; /* Last buf header OVFL needing to be freed */
289 DBT key, val;
290 SPLIT_RETURN ret;
291 u_int16_t n, off, ov_addr, scopyto;
292 char *cino; /* Character value of ino */
293
294 bufp = old_bufp;
295 ino = (u_int16_t *)old_bufp->page;
296 np = (u_int16_t *)new_bufp->page;
297 op = (u_int16_t *)old_bufp->page;
298 last_bfp = NULL;
299 scopyto = (u_int16_t)copyto; /* ANSI */
300
301 n = ino[0] - 1;
302 while (n < ino[0]) {
303 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) {
304 if (__big_split(hashp, old_bufp,
305 new_bufp, bufp, bufp->addr, obucket, &ret))
306 return (-1);
307 old_bufp = ret.oldp;
308 if (!old_bufp)
309 return (-1);
310 op = (u_int16_t *)old_bufp->page;
311 new_bufp = ret.newp;
312 if (!new_bufp)
313 return (-1);
314 np = (u_int16_t *)new_bufp->page;
315 bufp = ret.nextp;
316 if (!bufp)
317 return (0);
318 cino = (char *)bufp->page;
319 ino = (u_int16_t *)cino;
320 last_bfp = ret.nextp;
321 } else if (ino[n + 1] == OVFLPAGE) {
322 ov_addr = ino[n];
323 /*
324 * Fix up the old page -- the extra 2 are the fields
325 * which contained the overflow information.
326 */
327 ino[0] -= (moved + 2);
328 FREESPACE(ino) =
329 scopyto - sizeof(u_int16_t) * (ino[0] + 3);
330 OFFSET(ino) = scopyto;
331
332 bufp = __get_buf(hashp, ov_addr, bufp, 0);
333 if (!bufp)
334 return (-1);
335
336 ino = (u_int16_t *)bufp->page;
337 n = 1;
338 scopyto = hashp->BSIZE;
339 moved = 0;
340
341 if (last_bfp)
342 __free_ovflpage(hashp, last_bfp);
343 last_bfp = bufp;
344 }
345 /* Move regular sized pairs of there are any */
346 off = hashp->BSIZE;
347 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) {
348 cino = (char *)ino;
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];
353 off = ino[n + 1];
354
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);
359 else {
360 old_bufp =
361 __add_ovflpage(hashp, old_bufp);
362 if (!old_bufp)
363 return (-1);
364 op = (u_int16_t *)old_bufp->page;
365 putpair((char *)op, &key, &val);
366 }
367 old_bufp->flags |= BUF_MOD;
368 } else {
369 /* Move to new page */
370 if (PAIRFITS(np, (&key), (&val)))
371 putpair((char *)np, &key, &val);
372 else {
373 new_bufp =
374 __add_ovflpage(hashp, new_bufp);
375 if (!new_bufp)
376 return (-1);
377 np = (u_int16_t *)new_bufp->page;
378 putpair((char *)np, &key, &val);
379 }
380 new_bufp->flags |= BUF_MOD;
381 }
382 }
383 }
384 if (last_bfp)
385 __free_ovflpage(hashp, last_bfp);
386 return (0);
387}
388
389/*
390 * Add the given pair to the page
391 *
392 * Returns:
393 * 0 ==> OK
394 * 1 ==> failure
395 */
396int
397__addel(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val)
398{
399 u_int16_t *bp, *sop;
400 int do_expand;
401
402 bp = (u_int16_t *)bufp->page;
403 do_expand = 0;
404 while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
405 /* Exception case */
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 */
409 break;
410 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
411 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
412 if (!bufp)
413 return (-1);
414 bp = (u_int16_t *)bufp->page;
415 } else if (bp[bp[0]] != OVFLPAGE) {
416 /* Short key/data pairs, no more pages */
417 break;
418 } else {
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);
423 goto stats;
424 } else {
425 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
426 if (!bufp)
427 return (-1);
428 bp = (u_int16_t *)bufp->page;
429 }
430 }
431
432 if (PAIRFITS(bp, key, val))
433 putpair(bufp->page, key, val);
434 else {
435 do_expand = 1;
436 bufp = __add_ovflpage(hashp, bufp);
437 if (!bufp)
438 return (-1);
439 sop = (u_int16_t *)bufp->page;
440
441 if (PAIRFITS(sop, key, val))
442 putpair((char *)sop, key, val);
443 else
444 if (__big_insert(hashp, bufp, key, val))
445 return (-1);
446 }
447stats:
448 bufp->flags |= BUF_MOD;
449 /*
450 * If the average number of keys per bucket exceeds the fill factor,
451 * expand the table.
452 */
453 hashp->NKEYS++;
454 if (do_expand ||
455 (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
456 return (__expand_table(hashp));
457 return (0);
458}
459
460/*
461 *
462 * Returns:
463 * pointer on success
464 * NULL on error
465 */
466BUFHEAD *
467__add_ovflpage(HTAB *hashp, BUFHEAD *bufp)
468{
469 u_int16_t *sp, ndx, ovfl_num;
470#ifdef DEBUG1
471 int tmp1, tmp2;
472#endif
473 sp = (u_int16_t *)bufp->page;
474
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;
480 }
481 bufp->flags |= BUF_MOD;
482 ovfl_num = overflow_page(hashp);
483#ifdef DEBUG1
484 tmp1 = bufp->addr;
485 tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0;
486#endif
487 if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1)))
488 return (NULL);
489 bufp->ovfl->flags |= BUF_MOD;
490#ifdef DEBUG1
491 (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n",
492 tmp1, tmp2, bufp->ovfl->addr);
493#endif
494 ndx = sp[0];
495 /*
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
498 * the page.
499 */
500 sp[ndx + 4] = OFFSET(sp);
501 sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE;
502 sp[ndx + 1] = ovfl_num;
503 sp[ndx + 2] = OVFLPAGE;
504 sp[0] = ndx + 2;
505#ifdef HASH_STATISTICS
506 hash_overflows++;
507#endif
508 return (bufp->ovfl);
509}
510
511/*
512 * Returns:
513 * 0 indicates SUCCESS
514 * -1 indicates FAILURE
515 */
516int
517__get_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_disk,
518 int is_bitmap)
519{
520 int fd, page, size, rsize;
521 u_int16_t *bp;
522
523 fd = hashp->fp;
524 size = hashp->BSIZE;
525
526 if ((fd == -1) || !is_disk) {
527 PAGE_INIT(p);
528 return (0);
529 }
530 if (is_bucket)
531 page = BUCKET_TO_PAGE(bucket);
532 else
533 page = OADDR_TO_PAGE(bucket);
534 if ((rsize = pread(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
535 return (-1);
536 bp = (u_int16_t *)p;
537 if (!rsize)
538 bp[0] = 0; /* We hit the EOF, so initialize a new page */
539 else
540 if (rsize != size) {
541 errno = EFTYPE;
542 return (-1);
543 }
544 if (!is_bitmap && !bp[0]) {
545 PAGE_INIT(p);
546 } else
547 if (hashp->LORDER != BYTE_ORDER) {
548 int i, max;
549
550 if (is_bitmap) {
551 max = hashp->BSIZE >> 2; /* divide by 4 */
552 for (i = 0; i < max; i++)
553 M_32_SWAP(((int *)p)[i]);
554 } else {
555 M_16_SWAP(bp[0]);
556 max = bp[0] + 2;
557 for (i = 1; i <= max; i++)
558 M_16_SWAP(bp[i]);
559 }
560 }
561 return (0);
562}
563
564/*
565 * Write page p to disk
566 *
567 * Returns:
568 * 0 ==> OK
569 * -1 ==>failure
570 */
571int
572__put_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_bitmap)
573{
574 int fd, page, size, wsize;
575
576 size = hashp->BSIZE;
577 if ((hashp->fp == -1) && open_temp(hashp))
578 return (-1);
579 fd = hashp->fp;
580
581 if (hashp->LORDER != BYTE_ORDER) {
582 int i, max;
583
584 if (is_bitmap) {
585 max = hashp->BSIZE >> 2; /* divide by 4 */
586 for (i = 0; i < max; i++)
587 M_32_SWAP(((int *)p)[i]);
588 } else {
589 max = ((u_int16_t *)p)[0] + 2;
590 for (i = 0; i <= max; i++)
591 M_16_SWAP(((u_int16_t *)p)[i]);
592 }
593 }
594 if (is_bucket)
595 page = BUCKET_TO_PAGE(bucket);
596 else
597 page = OADDR_TO_PAGE(bucket);
598 if ((wsize = pwrite(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
599 /* Errno is set */
600 return (-1);
601 if (wsize != size) {
602 errno = EFTYPE;
603 return (-1);
604 }
605 return (0);
606}
607
608#define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1)
609/*
610 * Initialize a new bitmap page. Bitmap pages are left in memory
611 * once they are read in.
612 */
613int
614__ibitmap(HTAB *hashp, int pnum, int nbits, int ndx)
615{
616 u_int32_t *ip;
617 int clearbytes, clearints;
618
619 if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
620 return (1);
621 hashp->nmaps++;
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);
628 SETBIT(ip, 0);
629 hashp->BITMAPS[ndx] = (u_int16_t)pnum;
630 hashp->mapp[ndx] = ip;
631 return (0);
632}
633
634static u_int32_t
635first_free(u_int32_t map)
636{
637 u_int32_t i, mask;
638
639 mask = 0x1;
640 for (i = 0; i < BITS_PER_MAP; i++) {
641 if (!(mask & map))
642 return (i);
643 mask = mask << 1;
644 }
645 return (i);
646}
647
648static u_int16_t
649overflow_page(HTAB *hashp)
650{
651 u_int32_t *freep;
652 int max_free, offset, splitnum;
653 u_int16_t addr;
654 int bit, first_page, free_bit, free_page, i, in_use_bits, j;
655#ifdef DEBUG2
656 int tmp1, tmp2;
657#endif
658 splitnum = hashp->OVFL_POINT;
659 max_free = hashp->SPARES[splitnum];
660
661 free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
662 free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);
663
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)))
669 return (0);
670 if (i == free_page)
671 in_use_bits = free_bit;
672 else
673 in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;
674
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);
680 } else {
681 bit = 0;
682 j = 0;
683 }
684 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
685 if (freep[j] != ALL_SET)
686 goto found;
687 }
688
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);
694
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);
699 errno = EFBIG;
700 return (0);
701 }
702 hashp->OVFL_POINT = splitnum;
703 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
704 hashp->SPARES[splitnum-1]--;
705 offset = 1;
706 }
707
708 /* Check if we need to allocate a new bitmap page */
709 if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
710 free_page++;
711 if (free_page >= NCACHED) {
712 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
713 errno = EFBIG;
714 return (0);
715 }
716 /*
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.
726 */
727 if (__ibitmap(hashp,
728 (int)OADDR_OF(splitnum, offset), 1, free_page))
729 return (0);
730 hashp->SPARES[splitnum]++;
731#ifdef DEBUG2
732 free_bit = 2;
733#endif
734 offset++;
735 if (offset > SPLITMASK) {
736 if (++splitnum >= NCACHED) {
737 (void)_write(STDERR_FILENO, OVMSG,
738 sizeof(OVMSG) - 1);
739 errno = EFBIG;
740 return (0);
741 }
742 hashp->OVFL_POINT = splitnum;
743 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
744 hashp->SPARES[splitnum-1]--;
745 offset = 0;
746 }
747 } else {
748 /*
749 * Free_bit addresses the last used bit. Bump it to address
750 * the first available bit.
751 */
752 free_bit++;
753 SETBIT(freep, free_bit);
754 }
755
756 /* Calculate address of the new overflow page */
757 addr = OADDR_OF(splitnum, offset);
758#ifdef DEBUG2
759 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
760 addr, free_bit, free_page);
761#endif
762 return (addr);
763
764found:
765 bit = bit + first_free(freep[j]);
766 SETBIT(freep, bit);
767#ifdef DEBUG2
768 tmp1 = bit;
769 tmp2 = i;
770#endif
771 /*
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.
775 */
776 bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT));
777 if (bit >= hashp->LAST_FREED)
778 hashp->LAST_FREED = bit - 1;
779
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);
785 errno = EFBIG;
786 return (0); /* Out of overflow pages */
787 }
788 addr = OADDR_OF(i, offset);
789#ifdef DEBUG2
790 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
791 addr, tmp1, tmp2);
792#endif
793
794 /* Allocate and return the overflow page */
795 return (addr);
796}
797
798/*
799 * Mark this overflow page as free.
800 */
801void
802__free_ovflpage(HTAB *hashp, BUFHEAD *obufp)
803{
804 u_int16_t addr;
805 u_int32_t *freep;
806 int bit_address, free_page, free_bit;
807 u_int16_t ndx;
808
809 addr = obufp->addr;
810#ifdef DEBUG1
811 (void)fprintf(stderr, "Freeing %d\n", addr);
812#endif
813 ndx = (((u_int16_t)addr) >> SPLITSHIFT);
814 bit_address =
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);
820
821 if (!(freep = hashp->mapp[free_page]))
822 freep = fetch_bitmap(hashp, free_page);
823#ifdef DEBUG
824 /*
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.
828 */
829 if (!freep)
830 assert(0);
831#endif
832 CLRBIT(freep, free_bit);
833#ifdef DEBUG2
834 (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
835 obufp->addr, free_bit, free_page);
836#endif
837 __reclaim_buf(hashp, obufp);
838}
839
840/*
841 * Returns:
842 * 0 success
843 * -1 failure
844 */
845static int
846open_temp(HTAB *hashp)
847{
848 sigset_t set, oset;
849 int len;
850 char *envtmp = NULL;
851 char path[MAXPATHLEN];
852
853 if (issetugid() == 0)
854 envtmp = getenv("TMPDIR");
855 len = snprintf(path,
856 sizeof(path), "%s/_hash.XXXXXX", envtmp ? envtmp : "/tmp");
857 if (len < 0 || len >= (int)sizeof(path)) {
858 errno = ENAMETOOLONG;
859 return (-1);
860 }
861
862 /* Block signals; make sure file goes away at process exit. */
863 (void)sigfillset(&set);
| 70
71#include <db.h>
72#include "hash.h"
73#include "page.h"
74#include "extern.h"
75
76static u_int32_t *fetch_bitmap(HTAB *, int);
77static u_int32_t first_free(u_int32_t);
78static int open_temp(HTAB *);
79static u_int16_t overflow_page(HTAB *);
80static void putpair(char *, const DBT *, const DBT *);
81static void squeeze_key(u_int16_t *, const DBT *, const DBT *);
82static int ugly_split(HTAB *, u_int32_t, BUFHEAD *, BUFHEAD *, int, int);
83
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; \
88}
89
90/*
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
93 * stuff on.
94 */
95static void
96putpair(char *p, const DBT *key, const DBT *val)
97{
98 u_int16_t *bp, n, off;
99
100 bp = (u_int16_t *)p;
101
102 /* Enter the key first. */
103 n = bp[0];
104
105 off = OFFSET(bp) - key->size;
106 memmove(p + off, key->data, key->size);
107 bp[++n] = off;
108
109 /* Now the data. */
110 off -= val->size;
111 memmove(p + off, val->data, val->size);
112 bp[++n] = off;
113
114 /* Adjust page info. */
115 bp[0] = n;
116 bp[n + 1] = off - ((n + 3) * sizeof(u_int16_t));
117 bp[n + 2] = off;
118}
119
120/*
121 * Returns:
122 * 0 OK
123 * -1 error
124 */
125int
126__delpair(HTAB *hashp, BUFHEAD *bufp, int ndx)
127{
128 u_int16_t *bp, newoff, pairlen;
129 int n;
130
131 bp = (u_int16_t *)bufp->page;
132 n = bp[0];
133
134 if (bp[ndx + 1] < REAL_KEY)
135 return (__big_delete(hashp, bufp));
136 if (ndx != 1)
137 newoff = bp[ndx - 1];
138 else
139 newoff = hashp->BSIZE;
140 pairlen = newoff - bp[ndx + 1];
141
142 if (ndx != (n - 1)) {
143 /* Hard Case -- need to shuffle keys */
144 int i;
145 char *src = bufp->page + (int)OFFSET(bp);
146 char *dst = src + (int)pairlen;
147 memmove(dst, src, bp[ndx + 1] - OFFSET(bp));
148
149 /* Now adjust the pointers */
150 for (i = ndx + 2; i <= n; i += 2) {
151 if (bp[i + 1] == OVFLPAGE) {
152 bp[i - 2] = bp[i];
153 bp[i - 1] = bp[i + 1];
154 } else {
155 bp[i - 2] = bp[i] + pairlen;
156 bp[i - 1] = bp[i + 1] + pairlen;
157 }
158 }
159 if (ndx == hashp->cndx) {
160 /*
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.
164 */
165 hashp->cndx -= 2;
166 }
167 }
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);
171 bp[0] = n - 2;
172 hashp->NKEYS--;
173
174 bufp->flags |= BUF_MOD;
175 return (0);
176}
177/*
178 * Returns:
179 * 0 ==> OK
180 * -1 ==> Error
181 */
182int
183__split_page(HTAB *hashp, u_int32_t obucket, u_int32_t nbucket)
184{
185 BUFHEAD *new_bufp, *old_bufp;
186 u_int16_t *ino;
187 char *np;
188 DBT key, val;
189 int n, ndx, retval;
190 u_int16_t copyto, diff, off, moved;
191 char *op;
192
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)
197 return (-1);
198 new_bufp = __get_buf(hashp, nbucket, NULL, 0);
199 if (new_bufp == NULL)
200 return (-1);
201
202 old_bufp->flags |= (BUF_MOD | BUF_PIN);
203 new_bufp->flags |= (BUF_MOD | BUF_PIN);
204
205 ino = (u_int16_t *)(op = old_bufp->page);
206 np = new_bufp->page;
207
208 moved = 0;
209
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;
216 return (retval);
217
218 }
219 key.data = (u_char *)op + ino[n];
220 key.size = off - ino[n];
221
222 if (__call_hash(hashp, key.data, key.size) == obucket) {
223 /* Don't switch page */
224 diff = copyto - off;
225 if (diff) {
226 copyto = ino[n + 1] + diff;
227 memmove(op + copyto, op + ino[n + 1],
228 off - ino[n + 1]);
229 ino[ndx] = copyto + ino[n] - ino[n + 1];
230 ino[ndx + 1] = copyto;
231 } else
232 copyto = ino[n + 1];
233 ndx += 2;
234 } else {
235 /* Switch page */
236 val.data = (u_char *)op + ino[n + 1];
237 val.size = ino[n] - ino[n + 1];
238 putpair(np, &key, &val);
239 moved += 2;
240 }
241
242 off = ino[n + 1];
243 }
244
245 /* Now clean up the page */
246 ino[0] -= moved;
247 FREESPACE(ino) = copyto - sizeof(u_int16_t) * (ino[0] + 3);
248 OFFSET(ino) = copyto;
249
250#ifdef DEBUG3
251 (void)fprintf(stderr, "split %d/%d\n",
252 ((u_int16_t *)np)[0] / 2,
253 ((u_int16_t *)op)[0] / 2);
254#endif
255 /* unpin both pages */
256 old_bufp->flags &= ~BUF_PIN;
257 new_bufp->flags &= ~BUF_PIN;
258 return (0);
259}
260
261/*
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.
266 *
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
270 * big key/data pair.
271 *
272 * Returns:
273 * 0 ==> success
274 * -1 ==> failure
275 */
276static int
277ugly_split(HTAB *hashp,
278 u_int32_t obucket, /* Same as __split_page. */
279 BUFHEAD *old_bufp,
280 BUFHEAD *new_bufp,
281 int copyto, /* First byte on page which contains key/data values. */
282 int moved) /* Number of pairs moved to new page. */
283{
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 */
288
289 BUFHEAD *last_bfp; /* Last buf header OVFL needing to be freed */
290 DBT key, val;
291 SPLIT_RETURN ret;
292 u_int16_t n, off, ov_addr, scopyto;
293 char *cino; /* Character value of ino */
294
295 bufp = old_bufp;
296 ino = (u_int16_t *)old_bufp->page;
297 np = (u_int16_t *)new_bufp->page;
298 op = (u_int16_t *)old_bufp->page;
299 last_bfp = NULL;
300 scopyto = (u_int16_t)copyto; /* ANSI */
301
302 n = ino[0] - 1;
303 while (n < ino[0]) {
304 if (ino[2] < REAL_KEY && ino[2] != OVFLPAGE) {
305 if (__big_split(hashp, old_bufp,
306 new_bufp, bufp, bufp->addr, obucket, &ret))
307 return (-1);
308 old_bufp = ret.oldp;
309 if (!old_bufp)
310 return (-1);
311 op = (u_int16_t *)old_bufp->page;
312 new_bufp = ret.newp;
313 if (!new_bufp)
314 return (-1);
315 np = (u_int16_t *)new_bufp->page;
316 bufp = ret.nextp;
317 if (!bufp)
318 return (0);
319 cino = (char *)bufp->page;
320 ino = (u_int16_t *)cino;
321 last_bfp = ret.nextp;
322 } else if (ino[n + 1] == OVFLPAGE) {
323 ov_addr = ino[n];
324 /*
325 * Fix up the old page -- the extra 2 are the fields
326 * which contained the overflow information.
327 */
328 ino[0] -= (moved + 2);
329 FREESPACE(ino) =
330 scopyto - sizeof(u_int16_t) * (ino[0] + 3);
331 OFFSET(ino) = scopyto;
332
333 bufp = __get_buf(hashp, ov_addr, bufp, 0);
334 if (!bufp)
335 return (-1);
336
337 ino = (u_int16_t *)bufp->page;
338 n = 1;
339 scopyto = hashp->BSIZE;
340 moved = 0;
341
342 if (last_bfp)
343 __free_ovflpage(hashp, last_bfp);
344 last_bfp = bufp;
345 }
346 /* Move regular sized pairs of there are any */
347 off = hashp->BSIZE;
348 for (n = 1; (n < ino[0]) && (ino[n + 1] >= REAL_KEY); n += 2) {
349 cino = (char *)ino;
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];
354 off = ino[n + 1];
355
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);
360 else {
361 old_bufp =
362 __add_ovflpage(hashp, old_bufp);
363 if (!old_bufp)
364 return (-1);
365 op = (u_int16_t *)old_bufp->page;
366 putpair((char *)op, &key, &val);
367 }
368 old_bufp->flags |= BUF_MOD;
369 } else {
370 /* Move to new page */
371 if (PAIRFITS(np, (&key), (&val)))
372 putpair((char *)np, &key, &val);
373 else {
374 new_bufp =
375 __add_ovflpage(hashp, new_bufp);
376 if (!new_bufp)
377 return (-1);
378 np = (u_int16_t *)new_bufp->page;
379 putpair((char *)np, &key, &val);
380 }
381 new_bufp->flags |= BUF_MOD;
382 }
383 }
384 }
385 if (last_bfp)
386 __free_ovflpage(hashp, last_bfp);
387 return (0);
388}
389
390/*
391 * Add the given pair to the page
392 *
393 * Returns:
394 * 0 ==> OK
395 * 1 ==> failure
396 */
397int
398__addel(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val)
399{
400 u_int16_t *bp, *sop;
401 int do_expand;
402
403 bp = (u_int16_t *)bufp->page;
404 do_expand = 0;
405 while (bp[0] && (bp[2] < REAL_KEY || bp[bp[0]] < REAL_KEY))
406 /* Exception case */
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 */
410 break;
411 else if (bp[2] < REAL_KEY && bp[bp[0]] != OVFLPAGE) {
412 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
413 if (!bufp)
414 return (-1);
415 bp = (u_int16_t *)bufp->page;
416 } else if (bp[bp[0]] != OVFLPAGE) {
417 /* Short key/data pairs, no more pages */
418 break;
419 } else {
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);
424 goto stats;
425 } else {
426 bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
427 if (!bufp)
428 return (-1);
429 bp = (u_int16_t *)bufp->page;
430 }
431 }
432
433 if (PAIRFITS(bp, key, val))
434 putpair(bufp->page, key, val);
435 else {
436 do_expand = 1;
437 bufp = __add_ovflpage(hashp, bufp);
438 if (!bufp)
439 return (-1);
440 sop = (u_int16_t *)bufp->page;
441
442 if (PAIRFITS(sop, key, val))
443 putpair((char *)sop, key, val);
444 else
445 if (__big_insert(hashp, bufp, key, val))
446 return (-1);
447 }
448stats:
449 bufp->flags |= BUF_MOD;
450 /*
451 * If the average number of keys per bucket exceeds the fill factor,
452 * expand the table.
453 */
454 hashp->NKEYS++;
455 if (do_expand ||
456 (hashp->NKEYS / (hashp->MAX_BUCKET + 1) > hashp->FFACTOR))
457 return (__expand_table(hashp));
458 return (0);
459}
460
461/*
462 *
463 * Returns:
464 * pointer on success
465 * NULL on error
466 */
467BUFHEAD *
468__add_ovflpage(HTAB *hashp, BUFHEAD *bufp)
469{
470 u_int16_t *sp, ndx, ovfl_num;
471#ifdef DEBUG1
472 int tmp1, tmp2;
473#endif
474 sp = (u_int16_t *)bufp->page;
475
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;
481 }
482 bufp->flags |= BUF_MOD;
483 ovfl_num = overflow_page(hashp);
484#ifdef DEBUG1
485 tmp1 = bufp->addr;
486 tmp2 = bufp->ovfl ? bufp->ovfl->addr : 0;
487#endif
488 if (!ovfl_num || !(bufp->ovfl = __get_buf(hashp, ovfl_num, bufp, 1)))
489 return (NULL);
490 bufp->ovfl->flags |= BUF_MOD;
491#ifdef DEBUG1
492 (void)fprintf(stderr, "ADDOVFLPAGE: %d->ovfl was %d is now %d\n",
493 tmp1, tmp2, bufp->ovfl->addr);
494#endif
495 ndx = sp[0];
496 /*
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
499 * the page.
500 */
501 sp[ndx + 4] = OFFSET(sp);
502 sp[ndx + 3] = FREESPACE(sp) - OVFLSIZE;
503 sp[ndx + 1] = ovfl_num;
504 sp[ndx + 2] = OVFLPAGE;
505 sp[0] = ndx + 2;
506#ifdef HASH_STATISTICS
507 hash_overflows++;
508#endif
509 return (bufp->ovfl);
510}
511
512/*
513 * Returns:
514 * 0 indicates SUCCESS
515 * -1 indicates FAILURE
516 */
517int
518__get_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_disk,
519 int is_bitmap)
520{
521 int fd, page, size, rsize;
522 u_int16_t *bp;
523
524 fd = hashp->fp;
525 size = hashp->BSIZE;
526
527 if ((fd == -1) || !is_disk) {
528 PAGE_INIT(p);
529 return (0);
530 }
531 if (is_bucket)
532 page = BUCKET_TO_PAGE(bucket);
533 else
534 page = OADDR_TO_PAGE(bucket);
535 if ((rsize = pread(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
536 return (-1);
537 bp = (u_int16_t *)p;
538 if (!rsize)
539 bp[0] = 0; /* We hit the EOF, so initialize a new page */
540 else
541 if (rsize != size) {
542 errno = EFTYPE;
543 return (-1);
544 }
545 if (!is_bitmap && !bp[0]) {
546 PAGE_INIT(p);
547 } else
548 if (hashp->LORDER != BYTE_ORDER) {
549 int i, max;
550
551 if (is_bitmap) {
552 max = hashp->BSIZE >> 2; /* divide by 4 */
553 for (i = 0; i < max; i++)
554 M_32_SWAP(((int *)p)[i]);
555 } else {
556 M_16_SWAP(bp[0]);
557 max = bp[0] + 2;
558 for (i = 1; i <= max; i++)
559 M_16_SWAP(bp[i]);
560 }
561 }
562 return (0);
563}
564
565/*
566 * Write page p to disk
567 *
568 * Returns:
569 * 0 ==> OK
570 * -1 ==>failure
571 */
572int
573__put_page(HTAB *hashp, char *p, u_int32_t bucket, int is_bucket, int is_bitmap)
574{
575 int fd, page, size, wsize;
576
577 size = hashp->BSIZE;
578 if ((hashp->fp == -1) && open_temp(hashp))
579 return (-1);
580 fd = hashp->fp;
581
582 if (hashp->LORDER != BYTE_ORDER) {
583 int i, max;
584
585 if (is_bitmap) {
586 max = hashp->BSIZE >> 2; /* divide by 4 */
587 for (i = 0; i < max; i++)
588 M_32_SWAP(((int *)p)[i]);
589 } else {
590 max = ((u_int16_t *)p)[0] + 2;
591 for (i = 0; i <= max; i++)
592 M_16_SWAP(((u_int16_t *)p)[i]);
593 }
594 }
595 if (is_bucket)
596 page = BUCKET_TO_PAGE(bucket);
597 else
598 page = OADDR_TO_PAGE(bucket);
599 if ((wsize = pwrite(fd, p, size, (off_t)page << hashp->BSHIFT)) == -1)
600 /* Errno is set */
601 return (-1);
602 if (wsize != size) {
603 errno = EFTYPE;
604 return (-1);
605 }
606 return (0);
607}
608
609#define BYTE_MASK ((1 << INT_BYTE_SHIFT) -1)
610/*
611 * Initialize a new bitmap page. Bitmap pages are left in memory
612 * once they are read in.
613 */
614int
615__ibitmap(HTAB *hashp, int pnum, int nbits, int ndx)
616{
617 u_int32_t *ip;
618 int clearbytes, clearints;
619
620 if ((ip = (u_int32_t *)malloc(hashp->BSIZE)) == NULL)
621 return (1);
622 hashp->nmaps++;
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);
629 SETBIT(ip, 0);
630 hashp->BITMAPS[ndx] = (u_int16_t)pnum;
631 hashp->mapp[ndx] = ip;
632 return (0);
633}
634
635static u_int32_t
636first_free(u_int32_t map)
637{
638 u_int32_t i, mask;
639
640 mask = 0x1;
641 for (i = 0; i < BITS_PER_MAP; i++) {
642 if (!(mask & map))
643 return (i);
644 mask = mask << 1;
645 }
646 return (i);
647}
648
649static u_int16_t
650overflow_page(HTAB *hashp)
651{
652 u_int32_t *freep;
653 int max_free, offset, splitnum;
654 u_int16_t addr;
655 int bit, first_page, free_bit, free_page, i, in_use_bits, j;
656#ifdef DEBUG2
657 int tmp1, tmp2;
658#endif
659 splitnum = hashp->OVFL_POINT;
660 max_free = hashp->SPARES[splitnum];
661
662 free_page = (max_free - 1) >> (hashp->BSHIFT + BYTE_SHIFT);
663 free_bit = (max_free - 1) & ((hashp->BSIZE << BYTE_SHIFT) - 1);
664
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)))
670 return (0);
671 if (i == free_page)
672 in_use_bits = free_bit;
673 else
674 in_use_bits = (hashp->BSIZE << BYTE_SHIFT) - 1;
675
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);
681 } else {
682 bit = 0;
683 j = 0;
684 }
685 for (; bit <= in_use_bits; j++, bit += BITS_PER_MAP)
686 if (freep[j] != ALL_SET)
687 goto found;
688 }
689
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);
695
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);
700 errno = EFBIG;
701 return (0);
702 }
703 hashp->OVFL_POINT = splitnum;
704 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
705 hashp->SPARES[splitnum-1]--;
706 offset = 1;
707 }
708
709 /* Check if we need to allocate a new bitmap page */
710 if (free_bit == (hashp->BSIZE << BYTE_SHIFT) - 1) {
711 free_page++;
712 if (free_page >= NCACHED) {
713 (void)_write(STDERR_FILENO, OVMSG, sizeof(OVMSG) - 1);
714 errno = EFBIG;
715 return (0);
716 }
717 /*
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.
727 */
728 if (__ibitmap(hashp,
729 (int)OADDR_OF(splitnum, offset), 1, free_page))
730 return (0);
731 hashp->SPARES[splitnum]++;
732#ifdef DEBUG2
733 free_bit = 2;
734#endif
735 offset++;
736 if (offset > SPLITMASK) {
737 if (++splitnum >= NCACHED) {
738 (void)_write(STDERR_FILENO, OVMSG,
739 sizeof(OVMSG) - 1);
740 errno = EFBIG;
741 return (0);
742 }
743 hashp->OVFL_POINT = splitnum;
744 hashp->SPARES[splitnum] = hashp->SPARES[splitnum-1];
745 hashp->SPARES[splitnum-1]--;
746 offset = 0;
747 }
748 } else {
749 /*
750 * Free_bit addresses the last used bit. Bump it to address
751 * the first available bit.
752 */
753 free_bit++;
754 SETBIT(freep, free_bit);
755 }
756
757 /* Calculate address of the new overflow page */
758 addr = OADDR_OF(splitnum, offset);
759#ifdef DEBUG2
760 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
761 addr, free_bit, free_page);
762#endif
763 return (addr);
764
765found:
766 bit = bit + first_free(freep[j]);
767 SETBIT(freep, bit);
768#ifdef DEBUG2
769 tmp1 = bit;
770 tmp2 = i;
771#endif
772 /*
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.
776 */
777 bit = 1 + bit + (i * (hashp->BSIZE << BYTE_SHIFT));
778 if (bit >= hashp->LAST_FREED)
779 hashp->LAST_FREED = bit - 1;
780
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);
786 errno = EFBIG;
787 return (0); /* Out of overflow pages */
788 }
789 addr = OADDR_OF(i, offset);
790#ifdef DEBUG2
791 (void)fprintf(stderr, "OVERFLOW_PAGE: ADDR: %d BIT: %d PAGE %d\n",
792 addr, tmp1, tmp2);
793#endif
794
795 /* Allocate and return the overflow page */
796 return (addr);
797}
798
799/*
800 * Mark this overflow page as free.
801 */
802void
803__free_ovflpage(HTAB *hashp, BUFHEAD *obufp)
804{
805 u_int16_t addr;
806 u_int32_t *freep;
807 int bit_address, free_page, free_bit;
808 u_int16_t ndx;
809
810 addr = obufp->addr;
811#ifdef DEBUG1
812 (void)fprintf(stderr, "Freeing %d\n", addr);
813#endif
814 ndx = (((u_int16_t)addr) >> SPLITSHIFT);
815 bit_address =
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);
821
822 if (!(freep = hashp->mapp[free_page]))
823 freep = fetch_bitmap(hashp, free_page);
824#ifdef DEBUG
825 /*
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.
829 */
830 if (!freep)
831 assert(0);
832#endif
833 CLRBIT(freep, free_bit);
834#ifdef DEBUG2
835 (void)fprintf(stderr, "FREE_OVFLPAGE: ADDR: %d BIT: %d PAGE %d\n",
836 obufp->addr, free_bit, free_page);
837#endif
838 __reclaim_buf(hashp, obufp);
839}
840
841/*
842 * Returns:
843 * 0 success
844 * -1 failure
845 */
846static int
847open_temp(HTAB *hashp)
848{
849 sigset_t set, oset;
850 int len;
851 char *envtmp = NULL;
852 char path[MAXPATHLEN];
853
854 if (issetugid() == 0)
855 envtmp = getenv("TMPDIR");
856 len = snprintf(path,
857 sizeof(path), "%s/_hash.XXXXXX", envtmp ? envtmp : "/tmp");
858 if (len < 0 || len >= (int)sizeof(path)) {
859 errno = ENAMETOOLONG;
860 return (-1);
861 }
862
863 /* Block signals; make sure file goes away at process exit. */
864 (void)sigfillset(&set);
|