xref: /freebsd/lib/libc/db/hash/hash_bigkey.c (revision 0b57cec536236d46e3dba9bd041533462f33dbb7)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1990, 1993, 1994
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Margo Seltzer.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
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.
21  *
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
32  * SUCH DAMAGE.
33  */
34 
35 #if defined(LIBC_SCCS) && !defined(lint)
36 static char sccsid[] = "@(#)hash_bigkey.c	8.3 (Berkeley) 5/31/94";
37 #endif /* LIBC_SCCS and not lint */
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 /*
42  * PACKAGE: hash
43  * DESCRIPTION:
44  *	Big key/data handling for the hashing package.
45  *
46  * ROUTINES:
47  * External
48  *	__big_keydata
49  *	__big_split
50  *	__big_insert
51  *	__big_return
52  *	__big_delete
53  *	__find_last_page
54  * Internal
55  *	collect_key
56  *	collect_data
57  */
58 
59 #include <sys/param.h>
60 
61 #include <errno.h>
62 #include <stdio.h>
63 #include <stdlib.h>
64 #include <string.h>
65 
66 #ifdef DEBUG
67 #include <assert.h>
68 #endif
69 
70 #include <db.h>
71 #include "hash.h"
72 #include "page.h"
73 #include "extern.h"
74 
75 static int collect_key(HTAB *, BUFHEAD *, int, DBT *, int);
76 static int collect_data(HTAB *, BUFHEAD *, int, int);
77 
78 /*
79  * Big_insert
80  *
81  * You need to do an insert and the key/data pair is too big
82  *
83  * Returns:
84  * 0 ==> OK
85  *-1 ==> ERROR
86  */
87 int
88 __big_insert(HTAB *hashp, BUFHEAD *bufp, const DBT *key, const DBT *val)
89 {
90 	u_int16_t *p;
91 	int key_size, n;
92 	unsigned int val_size;
93 	u_int16_t space, move_bytes, off;
94 	char *cp, *key_data, *val_data;
95 
96 	cp = bufp->page;		/* Character pointer of p. */
97 	p = (u_int16_t *)cp;
98 
99 	key_data = (char *)key->data;
100 	key_size = key->size;
101 	val_data = (char *)val->data;
102 	val_size = val->size;
103 
104 	/* First move the Key */
105 	for (space = FREESPACE(p) - BIGOVERHEAD; key_size;
106 	    space = FREESPACE(p) - BIGOVERHEAD) {
107 		move_bytes = MIN(space, key_size);
108 		off = OFFSET(p) - move_bytes;
109 		memmove(cp + off, key_data, move_bytes);
110 		key_size -= move_bytes;
111 		key_data += move_bytes;
112 		n = p[0];
113 		p[++n] = off;
114 		p[0] = ++n;
115 		FREESPACE(p) = off - PAGE_META(n);
116 		OFFSET(p) = off;
117 		p[n] = PARTIAL_KEY;
118 		bufp = __add_ovflpage(hashp, bufp);
119 		if (!bufp)
120 			return (-1);
121 		n = p[0];
122 		if (!key_size) {
123 			space = FREESPACE(p);
124 			if (space) {
125 				move_bytes = MIN(space, val_size);
126 				/*
127 				 * If the data would fit exactly in the
128 				 * remaining space, we must overflow it to the
129 				 * next page; otherwise the invariant that the
130 				 * data must end on a page with FREESPACE
131 				 * non-zero would fail.
132 				 */
133 				if (space == val_size && val_size == val->size)
134 					goto toolarge;
135 				off = OFFSET(p) - move_bytes;
136 				memmove(cp + off, val_data, move_bytes);
137 				val_data += move_bytes;
138 				val_size -= move_bytes;
139 				p[n] = off;
140 				p[n - 2] = FULL_KEY_DATA;
141 				FREESPACE(p) = FREESPACE(p) - move_bytes;
142 				OFFSET(p) = off;
143 			} else {
144 			toolarge:
145 				p[n - 2] = FULL_KEY;
146 			}
147 		}
148 		p = (u_int16_t *)bufp->page;
149 		cp = bufp->page;
150 		bufp->flags |= BUF_MOD;
151 	}
152 
153 	/* Now move the data */
154 	for (space = FREESPACE(p) - BIGOVERHEAD; val_size;
155 	    space = FREESPACE(p) - BIGOVERHEAD) {
156 		move_bytes = MIN(space, val_size);
157 		/*
158 		 * Here's the hack to make sure that if the data ends on the
159 		 * same page as the key ends, FREESPACE is at least one.
160 		 */
161 		if (space == val_size && val_size == val->size)
162 			move_bytes--;
163 		off = OFFSET(p) - move_bytes;
164 		memmove(cp + off, val_data, move_bytes);
165 		val_size -= move_bytes;
166 		val_data += move_bytes;
167 		n = p[0];
168 		p[++n] = off;
169 		p[0] = ++n;
170 		FREESPACE(p) = off - PAGE_META(n);
171 		OFFSET(p) = off;
172 		if (val_size) {
173 			p[n] = FULL_KEY;
174 			bufp = __add_ovflpage(hashp, bufp);
175 			if (!bufp)
176 				return (-1);
177 			cp = bufp->page;
178 			p = (u_int16_t *)cp;
179 		} else
180 			p[n] = FULL_KEY_DATA;
181 		bufp->flags |= BUF_MOD;
182 	}
183 	return (0);
184 }
185 
186 /*
187  * Called when bufp's page  contains a partial key (index should be 1)
188  *
189  * All pages in the big key/data pair except bufp are freed.  We cannot
190  * free bufp because the page pointing to it is lost and we can't get rid
191  * of its pointer.
192  *
193  * Returns:
194  * 0 => OK
195  *-1 => ERROR
196  */
197 int
198 __big_delete(HTAB *hashp, BUFHEAD *bufp)
199 {
200 	BUFHEAD *last_bfp, *rbufp;
201 	u_int16_t *bp, pageno;
202 	int key_done, n;
203 
204 	rbufp = bufp;
205 	last_bfp = NULL;
206 	bp = (u_int16_t *)bufp->page;
207 	pageno = 0;
208 	key_done = 0;
209 
210 	while (!key_done || (bp[2] != FULL_KEY_DATA)) {
211 		if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA)
212 			key_done = 1;
213 
214 		/*
215 		 * If there is freespace left on a FULL_KEY_DATA page, then
216 		 * the data is short and fits entirely on this page, and this
217 		 * is the last page.
218 		 */
219 		if (bp[2] == FULL_KEY_DATA && FREESPACE(bp))
220 			break;
221 		pageno = bp[bp[0] - 1];
222 		rbufp->flags |= BUF_MOD;
223 		rbufp = __get_buf(hashp, pageno, rbufp, 0);
224 		if (last_bfp)
225 			__free_ovflpage(hashp, last_bfp);
226 		last_bfp = rbufp;
227 		if (!rbufp)
228 			return (-1);		/* Error. */
229 		bp = (u_int16_t *)rbufp->page;
230 	}
231 
232 	/*
233 	 * If we get here then rbufp points to the last page of the big
234 	 * key/data pair.  Bufp points to the first one -- it should now be
235 	 * empty pointing to the next page after this pair.  Can't free it
236 	 * because we don't have the page pointing to it.
237 	 */
238 
239 	/* This is information from the last page of the pair. */
240 	n = bp[0];
241 	pageno = bp[n - 1];
242 
243 	/* Now, bp is the first page of the pair. */
244 	bp = (u_int16_t *)bufp->page;
245 	if (n > 2) {
246 		/* There is an overflow page. */
247 		bp[1] = pageno;
248 		bp[2] = OVFLPAGE;
249 		bufp->ovfl = rbufp->ovfl;
250 	} else
251 		/* This is the last page. */
252 		bufp->ovfl = NULL;
253 	n -= 2;
254 	bp[0] = n;
255 	FREESPACE(bp) = hashp->BSIZE - PAGE_META(n);
256 	OFFSET(bp) = hashp->BSIZE;
257 
258 	bufp->flags |= BUF_MOD;
259 	if (rbufp)
260 		__free_ovflpage(hashp, rbufp);
261 	if (last_bfp && last_bfp != rbufp)
262 		__free_ovflpage(hashp, last_bfp);
263 
264 	hashp->NKEYS--;
265 	return (0);
266 }
267 /*
268  * Returns:
269  *  0 = key not found
270  * -1 = get next overflow page
271  * -2 means key not found and this is big key/data
272  * -3 error
273  */
274 int
275 __find_bigpair(HTAB *hashp, BUFHEAD *bufp, int ndx, char *key, int size)
276 {
277 	u_int16_t *bp;
278 	char *p;
279 	int ksize;
280 	u_int16_t bytes;
281 	char *kkey;
282 
283 	bp = (u_int16_t *)bufp->page;
284 	p = bufp->page;
285 	ksize = size;
286 	kkey = key;
287 
288 	for (bytes = hashp->BSIZE - bp[ndx];
289 	    bytes <= size && bp[ndx + 1] == PARTIAL_KEY;
290 	    bytes = hashp->BSIZE - bp[ndx]) {
291 		if (memcmp(p + bp[ndx], kkey, bytes))
292 			return (-2);
293 		kkey += bytes;
294 		ksize -= bytes;
295 		bufp = __get_buf(hashp, bp[ndx + 2], bufp, 0);
296 		if (!bufp)
297 			return (-3);
298 		p = bufp->page;
299 		bp = (u_int16_t *)p;
300 		ndx = 1;
301 	}
302 
303 	if (bytes != ksize || memcmp(p + bp[ndx], kkey, bytes)) {
304 #ifdef HASH_STATISTICS
305 		++hash_collisions;
306 #endif
307 		return (-2);
308 	} else
309 		return (ndx);
310 }
311 
312 /*
313  * Given the buffer pointer of the first overflow page of a big pair,
314  * find the end of the big pair
315  *
316  * This will set bpp to the buffer header of the last page of the big pair.
317  * It will return the pageno of the overflow page following the last page
318  * of the pair; 0 if there isn't any (i.e. big pair is the last key in the
319  * bucket)
320  */
321 u_int16_t
322 __find_last_page(HTAB *hashp, BUFHEAD **bpp)
323 {
324 	BUFHEAD *bufp;
325 	u_int16_t *bp, pageno;
326 	int n;
327 
328 	bufp = *bpp;
329 	bp = (u_int16_t *)bufp->page;
330 	for (;;) {
331 		n = bp[0];
332 
333 		/*
334 		 * This is the last page if: the tag is FULL_KEY_DATA and
335 		 * either only 2 entries OVFLPAGE marker is explicit there
336 		 * is freespace on the page.
337 		 */
338 		if (bp[2] == FULL_KEY_DATA &&
339 		    ((n == 2) || (bp[n] == OVFLPAGE) || (FREESPACE(bp))))
340 			break;
341 
342 		pageno = bp[n - 1];
343 		bufp = __get_buf(hashp, pageno, bufp, 0);
344 		if (!bufp)
345 			return (0);	/* Need to indicate an error! */
346 		bp = (u_int16_t *)bufp->page;
347 	}
348 
349 	*bpp = bufp;
350 	if (bp[0] > 2)
351 		return (bp[3]);
352 	else
353 		return (0);
354 }
355 
356 /*
357  * Return the data for the key/data pair that begins on this page at this
358  * index (index should always be 1).
359  */
360 int
361 __big_return(HTAB *hashp, BUFHEAD *bufp, int ndx, DBT *val, int set_current)
362 {
363 	BUFHEAD *save_p;
364 	u_int16_t *bp, len, off, save_addr;
365 	char *tp;
366 
367 	bp = (u_int16_t *)bufp->page;
368 	while (bp[ndx + 1] == PARTIAL_KEY) {
369 		bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
370 		if (!bufp)
371 			return (-1);
372 		bp = (u_int16_t *)bufp->page;
373 		ndx = 1;
374 	}
375 
376 	if (bp[ndx + 1] == FULL_KEY) {
377 		bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
378 		if (!bufp)
379 			return (-1);
380 		bp = (u_int16_t *)bufp->page;
381 		save_p = bufp;
382 		save_addr = save_p->addr;
383 		off = bp[1];
384 		len = 0;
385 	} else
386 		if (!FREESPACE(bp)) {
387 			/*
388 			 * This is a hack.  We can't distinguish between
389 			 * FULL_KEY_DATA that contains complete data or
390 			 * incomplete data, so we require that if the data
391 			 * is complete, there is at least 1 byte of free
392 			 * space left.
393 			 */
394 			off = bp[bp[0]];
395 			len = bp[1] - off;
396 			save_p = bufp;
397 			save_addr = bufp->addr;
398 			bufp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
399 			if (!bufp)
400 				return (-1);
401 			bp = (u_int16_t *)bufp->page;
402 		} else {
403 			/* The data is all on one page. */
404 			tp = (char *)bp;
405 			off = bp[bp[0]];
406 			val->data = (u_char *)tp + off;
407 			val->size = bp[1] - off;
408 			if (set_current) {
409 				if (bp[0] == 2) {	/* No more buckets in
410 							 * chain */
411 					hashp->cpage = NULL;
412 					hashp->cbucket++;
413 					hashp->cndx = 1;
414 				} else {
415 					hashp->cpage = __get_buf(hashp,
416 					    bp[bp[0] - 1], bufp, 0);
417 					if (!hashp->cpage)
418 						return (-1);
419 					hashp->cndx = 1;
420 					if (!((u_int16_t *)
421 					    hashp->cpage->page)[0]) {
422 						hashp->cbucket++;
423 						hashp->cpage = NULL;
424 					}
425 				}
426 			}
427 			return (0);
428 		}
429 
430 	val->size = (size_t)collect_data(hashp, bufp, (int)len, set_current);
431 	if (val->size == (size_t)-1)
432 		return (-1);
433 	if (save_p->addr != save_addr) {
434 		/* We are pretty short on buffers. */
435 		errno = EINVAL;			/* OUT OF BUFFERS */
436 		return (-1);
437 	}
438 	memmove(hashp->tmp_buf, (save_p->page) + off, len);
439 	val->data = (u_char *)hashp->tmp_buf;
440 	return (0);
441 }
442 /*
443  * Count how big the total datasize is by recursing through the pages.  Then
444  * allocate a buffer and copy the data as you recurse up.
445  */
446 static int
447 collect_data(HTAB *hashp, BUFHEAD *bufp, int len, int set)
448 {
449 	u_int16_t *bp;
450 	char *p;
451 	BUFHEAD *xbp;
452 	u_int16_t save_addr;
453 	int mylen, totlen;
454 
455 	p = bufp->page;
456 	bp = (u_int16_t *)p;
457 	mylen = hashp->BSIZE - bp[1];
458 	save_addr = bufp->addr;
459 
460 	if (bp[2] == FULL_KEY_DATA) {		/* End of Data */
461 		totlen = len + mylen;
462 		if (hashp->tmp_buf)
463 			free(hashp->tmp_buf);
464 		if ((hashp->tmp_buf = (char *)malloc(totlen)) == NULL)
465 			return (-1);
466 		if (set) {
467 			hashp->cndx = 1;
468 			if (bp[0] == 2) {	/* No more buckets in chain */
469 				hashp->cpage = NULL;
470 				hashp->cbucket++;
471 			} else {
472 				hashp->cpage =
473 				    __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
474 				if (!hashp->cpage)
475 					return (-1);
476 				else if (!((u_int16_t *)hashp->cpage->page)[0]) {
477 					hashp->cbucket++;
478 					hashp->cpage = NULL;
479 				}
480 			}
481 		}
482 	} else {
483 		xbp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
484 		if (!xbp || ((totlen =
485 		    collect_data(hashp, xbp, len + mylen, set)) < 1))
486 			return (-1);
487 	}
488 	if (bufp->addr != save_addr) {
489 		errno = EINVAL;			/* Out of buffers. */
490 		return (-1);
491 	}
492 	memmove(&hashp->tmp_buf[len], (bufp->page) + bp[1], mylen);
493 	return (totlen);
494 }
495 
496 /*
497  * Fill in the key and data for this big pair.
498  */
499 int
500 __big_keydata(HTAB *hashp, BUFHEAD *bufp, DBT *key, DBT *val, int set)
501 {
502 	key->size = (size_t)collect_key(hashp, bufp, 0, val, set);
503 	if (key->size == (size_t)-1)
504 		return (-1);
505 	key->data = (u_char *)hashp->tmp_key;
506 	return (0);
507 }
508 
509 /*
510  * Count how big the total key size is by recursing through the pages.  Then
511  * collect the data, allocate a buffer and copy the key as you recurse up.
512  */
513 static int
514 collect_key(HTAB *hashp, BUFHEAD *bufp, int len, DBT *val, int set)
515 {
516 	BUFHEAD *xbp;
517 	char *p;
518 	int mylen, totlen;
519 	u_int16_t *bp, save_addr;
520 
521 	p = bufp->page;
522 	bp = (u_int16_t *)p;
523 	mylen = hashp->BSIZE - bp[1];
524 
525 	save_addr = bufp->addr;
526 	totlen = len + mylen;
527 	if (bp[2] == FULL_KEY || bp[2] == FULL_KEY_DATA) {    /* End of Key. */
528 		if (hashp->tmp_key != NULL)
529 			free(hashp->tmp_key);
530 		if ((hashp->tmp_key = (char *)malloc(totlen)) == NULL)
531 			return (-1);
532 		if (__big_return(hashp, bufp, 1, val, set))
533 			return (-1);
534 	} else {
535 		xbp = __get_buf(hashp, bp[bp[0] - 1], bufp, 0);
536 		if (!xbp || ((totlen =
537 		    collect_key(hashp, xbp, totlen, val, set)) < 1))
538 			return (-1);
539 	}
540 	if (bufp->addr != save_addr) {
541 		errno = EINVAL;		/* MIS -- OUT OF BUFFERS */
542 		return (-1);
543 	}
544 	memmove(&hashp->tmp_key[len], (bufp->page) + bp[1], mylen);
545 	return (totlen);
546 }
547 
548 /*
549  * Returns:
550  *  0 => OK
551  * -1 => error
552  */
553 int
554 __big_split(HTAB *hashp,
555     BUFHEAD *op,	/* Pointer to where to put keys that go in old bucket */
556     BUFHEAD *np,	/* Pointer to new bucket page */
557     BUFHEAD *big_keyp,	/* Pointer to first page containing the big key/data */
558     int addr,		/* Address of big_keyp */
559     u_int32_t obucket,	/* Old Bucket */
560     SPLIT_RETURN *ret)
561 {
562 	BUFHEAD *bp, *tmpp;
563 	DBT key, val;
564 	u_int32_t change;
565 	u_int16_t free_space, n, off, *tp;
566 
567 	bp = big_keyp;
568 
569 	/* Now figure out where the big key/data goes */
570 	if (__big_keydata(hashp, big_keyp, &key, &val, 0))
571 		return (-1);
572 	change = (__call_hash(hashp, key.data, key.size) != obucket);
573 
574 	if ( (ret->next_addr = __find_last_page(hashp, &big_keyp)) ) {
575 		if (!(ret->nextp =
576 		    __get_buf(hashp, ret->next_addr, big_keyp, 0)))
577 			return (-1);
578 	} else
579 		ret->nextp = NULL;
580 
581 	/* Now make one of np/op point to the big key/data pair */
582 #ifdef DEBUG
583 	assert(np->ovfl == NULL);
584 #endif
585 	if (change)
586 		tmpp = np;
587 	else
588 		tmpp = op;
589 
590 	tmpp->flags |= BUF_MOD;
591 #ifdef DEBUG1
592 	(void)fprintf(stderr,
593 	    "BIG_SPLIT: %d->ovfl was %d is now %d\n", tmpp->addr,
594 	    (tmpp->ovfl ? tmpp->ovfl->addr : 0), (bp ? bp->addr : 0));
595 #endif
596 	tmpp->ovfl = bp;	/* one of op/np point to big_keyp */
597 	tp = (u_int16_t *)tmpp->page;
598 #ifdef DEBUG
599 	assert(FREESPACE(tp) >= OVFLSIZE);
600 #endif
601 	n = tp[0];
602 	off = OFFSET(tp);
603 	free_space = FREESPACE(tp);
604 	tp[++n] = (u_int16_t)addr;
605 	tp[++n] = OVFLPAGE;
606 	tp[0] = n;
607 	OFFSET(tp) = off;
608 	FREESPACE(tp) = free_space - OVFLSIZE;
609 
610 	/*
611 	 * Finally, set the new and old return values. BIG_KEYP contains a
612 	 * pointer to the last page of the big key_data pair. Make sure that
613 	 * big_keyp has no following page (2 elements) or create an empty
614 	 * following page.
615 	 */
616 
617 	ret->newp = np;
618 	ret->oldp = op;
619 
620 	tp = (u_int16_t *)big_keyp->page;
621 	big_keyp->flags |= BUF_MOD;
622 	if (tp[0] > 2) {
623 		/*
624 		 * There may be either one or two offsets on this page.  If
625 		 * there is one, then the overflow page is linked on normally
626 		 * and tp[4] is OVFLPAGE.  If there are two, tp[4] contains
627 		 * the second offset and needs to get stuffed in after the
628 		 * next overflow page is added.
629 		 */
630 		n = tp[4];
631 		free_space = FREESPACE(tp);
632 		off = OFFSET(tp);
633 		tp[0] -= 2;
634 		FREESPACE(tp) = free_space + OVFLSIZE;
635 		OFFSET(tp) = off;
636 		tmpp = __add_ovflpage(hashp, big_keyp);
637 		if (!tmpp)
638 			return (-1);
639 		tp[4] = n;
640 	} else
641 		tmpp = big_keyp;
642 
643 	if (change)
644 		ret->newp = tmpp;
645 	else
646 		ret->oldp = tmpp;
647 	return (0);
648 }
649