Home | History | Annotate | Line # | Download | only in btree
bt_put.c revision 1.9
      1 /*	$NetBSD: bt_put.c,v 1.9 1997/07/13 18:51:55 christos Exp $	*/
      2 
      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  * Mike Olson.
      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. All advertising materials mentioning features or use of this software
     19  *    must display the following acknowledgement:
     20  *	This product includes software developed by the University of
     21  *	California, Berkeley and its contributors.
     22  * 4. Neither the name of the University nor the names of its contributors
     23  *    may be used to endorse or promote products derived from this software
     24  *    without specific prior written permission.
     25  *
     26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     36  * SUCH DAMAGE.
     37  */
     38 
     39 #include <sys/cdefs.h>
     40 #if defined(LIBC_SCCS) && !defined(lint)
     41 #if 0
     42 static char sccsid[] = "@(#)bt_put.c	8.8 (Berkeley) 7/26/94";
     43 #else
     44 __RCSID("$NetBSD: bt_put.c,v 1.9 1997/07/13 18:51:55 christos Exp $");
     45 #endif
     46 #endif /* LIBC_SCCS and not lint */
     47 
     48 #include <sys/types.h>
     49 
     50 #include <errno.h>
     51 #include <stdio.h>
     52 #include <stdlib.h>
     53 #include <string.h>
     54 
     55 #include <db.h>
     56 #include "btree.h"
     57 
     58 static EPG *bt_fast __P((BTREE *, const DBT *, const DBT *, int *));
     59 
     60 /*
     61  * __BT_PUT -- Add a btree item to the tree.
     62  *
     63  * Parameters:
     64  *	dbp:	pointer to access method
     65  *	key:	key
     66  *	data:	data
     67  *	flag:	R_NOOVERWRITE
     68  *
     69  * Returns:
     70  *	RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key is already in the
     71  *	tree and R_NOOVERWRITE specified.
     72  */
     73 int
     74 __bt_put(dbp, key, data, flags)
     75 	const DB *dbp;
     76 	DBT *key;
     77 	const DBT *data;
     78 	u_int flags;
     79 {
     80 	BTREE *t;
     81 	DBT tkey, tdata;
     82 	EPG *e = NULL; /* pacify gcc */
     83 	PAGE *h;
     84 	indx_t index, nxtindex;
     85 	pgno_t pg;
     86 	u_int32_t nbytes;
     87 	int dflags, exact, status;
     88 	char *dest, db[NOVFLSIZE], kb[NOVFLSIZE];
     89 
     90 	t = dbp->internal;
     91 
     92 	/* Toss any page pinned across calls. */
     93 	if (t->bt_pinned != NULL) {
     94 		mpool_put(t->bt_mp, t->bt_pinned, 0);
     95 		t->bt_pinned = NULL;
     96 	}
     97 
     98 	/* Check for change to a read-only tree. */
     99 	if (F_ISSET(t, B_RDONLY)) {
    100 		errno = EPERM;
    101 		return (RET_ERROR);
    102 	}
    103 
    104 	switch (flags) {
    105 	case 0:
    106 	case R_NOOVERWRITE:
    107 		break;
    108 	case R_CURSOR:
    109 		/*
    110 		 * If flags is R_CURSOR, put the cursor.  Must already
    111 		 * have started a scan and not have already deleted it.
    112 		 */
    113 		if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
    114 		    !F_ISSET(&t->bt_cursor,
    115 		        CURS_ACQUIRE | CURS_AFTER | CURS_BEFORE))
    116 			break;
    117 		/* FALLTHROUGH */
    118 	default:
    119 		errno = EINVAL;
    120 		return (RET_ERROR);
    121 	}
    122 
    123 	/*
    124 	 * If the key/data pair won't fit on a page, store it on overflow
    125 	 * pages.  Only put the key on the overflow page if the pair are
    126 	 * still too big after moving the data to an overflow page.
    127 	 *
    128 	 * XXX
    129 	 * If the insert fails later on, the overflow pages aren't recovered.
    130 	 */
    131 	dflags = 0;
    132 	if (key->size + data->size > t->bt_ovflsize) {
    133 		if (key->size > t->bt_ovflsize) {
    134 storekey:		if (__ovfl_put(t, key, &pg) == RET_ERROR)
    135 				return (RET_ERROR);
    136 			tkey.data = kb;
    137 			tkey.size = NOVFLSIZE;
    138 			memmove(kb, &pg, sizeof(pgno_t));
    139 			memmove(kb + sizeof(pgno_t),
    140 			    &key->size, sizeof(u_int32_t));
    141 			dflags |= P_BIGKEY;
    142 			key = &tkey;
    143 		}
    144 		if (key->size + data->size > t->bt_ovflsize) {
    145 			if (__ovfl_put(t, data, &pg) == RET_ERROR)
    146 				return (RET_ERROR);
    147 			tdata.data = db;
    148 			tdata.size = NOVFLSIZE;
    149 			memmove(db, &pg, sizeof(pgno_t));
    150 			memmove(db + sizeof(pgno_t),
    151 			    &data->size, sizeof(u_int32_t));
    152 			dflags |= P_BIGDATA;
    153 			data = &tdata;
    154 		}
    155 		if (key->size + data->size > t->bt_ovflsize)
    156 			goto storekey;
    157 	}
    158 
    159 	/* Replace the cursor. */
    160 	if (flags == R_CURSOR) {
    161 		if ((h = mpool_get(t->bt_mp, t->bt_cursor.pg.pgno, 0)) == NULL)
    162 			return (RET_ERROR);
    163 		index = t->bt_cursor.pg.index;
    164 		goto delete;
    165 	}
    166 
    167 	/*
    168 	 * Find the key to delete, or, the location at which to insert.
    169 	 * Bt_fast and __bt_search both pin the returned page.
    170 	 */
    171 	if (t->bt_order == NOT || (e = bt_fast(t, key, data, &exact)) == NULL)
    172 		if ((e = __bt_search(t, key, &exact)) == NULL)
    173 			return (RET_ERROR);
    174 	h = e->page;
    175 	index = e->index;
    176 
    177 	/*
    178 	 * Add the key/data pair to the tree.  If an identical key is already
    179 	 * in the tree, and R_NOOVERWRITE is set, an error is returned.  If
    180 	 * R_NOOVERWRITE is not set, the key is either added (if duplicates are
    181 	 * permitted) or an error is returned.
    182 	 */
    183 	switch (flags) {
    184 	case R_NOOVERWRITE:
    185 		if (!exact)
    186 			break;
    187 		mpool_put(t->bt_mp, h, 0);
    188 		return (RET_SPECIAL);
    189 	default:
    190 		if (!exact || !F_ISSET(t, B_NODUPS))
    191 			break;
    192 		/*
    193 		 * !!!
    194 		 * Note, the delete may empty the page, so we need to put a
    195 		 * new entry into the page immediately.
    196 		 */
    197 delete:		if (__bt_dleaf(t, key, h, index) == RET_ERROR) {
    198 			mpool_put(t->bt_mp, h, 0);
    199 			return (RET_ERROR);
    200 		}
    201 		break;
    202 	}
    203 
    204 	/*
    205 	 * If not enough room, or the user has put a ceiling on the number of
    206 	 * keys permitted in the page, split the page.  The split code will
    207 	 * insert the key and data and unpin the current page.  If inserting
    208 	 * into the offset array, shift the pointers up.
    209 	 */
    210 	nbytes = NBLEAFDBT(key->size, data->size);
    211 	if (h->upper - h->lower < nbytes + sizeof(indx_t)) {
    212 		if ((status = __bt_split(t, h, key,
    213 		    data, dflags, nbytes, index)) != RET_SUCCESS)
    214 			return (status);
    215 		goto success;
    216 	}
    217 
    218 	if (index < (nxtindex = NEXTINDEX(h)))
    219 		memmove(h->linp + index + 1, h->linp + index,
    220 		    (nxtindex - index) * sizeof(indx_t));
    221 	h->lower += sizeof(indx_t);
    222 
    223 	h->linp[index] = h->upper -= nbytes;
    224 	dest = (char *)h + h->upper;
    225 	WR_BLEAF(dest, key, data, dflags);
    226 
    227 	/* If the cursor is on this page, adjust it as necessary. */
    228 	if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
    229 	    !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
    230 	    t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index >= index)
    231 		++t->bt_cursor.pg.index;
    232 
    233 	if (t->bt_order == NOT)
    234 		if (h->nextpg == P_INVALID) {
    235 			if (index == NEXTINDEX(h) - 1) {
    236 				t->bt_order = FORWARD;
    237 				t->bt_last.index = index;
    238 				t->bt_last.pgno = h->pgno;
    239 			}
    240 		} else if (h->prevpg == P_INVALID) {
    241 			if (index == 0) {
    242 				t->bt_order = BACK;
    243 				t->bt_last.index = 0;
    244 				t->bt_last.pgno = h->pgno;
    245 			}
    246 		}
    247 
    248 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
    249 
    250 success:
    251 	if (flags == R_SETCURSOR)
    252 		__bt_setcur(t, e->page->pgno, e->index);
    253 
    254 	F_SET(t, B_MODIFIED);
    255 	return (RET_SUCCESS);
    256 }
    257 
    258 #ifdef STATISTICS
    259 u_long bt_cache_hit, bt_cache_miss;
    260 #endif
    261 
    262 /*
    263  * BT_FAST -- Do a quick check for sorted data.
    264  *
    265  * Parameters:
    266  *	t:	tree
    267  *	key:	key to insert
    268  *
    269  * Returns:
    270  * 	EPG for new record or NULL if not found.
    271  */
    272 static EPG *
    273 bt_fast(t, key, data, exactp)
    274 	BTREE *t;
    275 	const DBT *key, *data;
    276 	int *exactp;
    277 {
    278 	PAGE *h;
    279 	u_int32_t nbytes;
    280 	int cmp;
    281 
    282 	if ((h = mpool_get(t->bt_mp, t->bt_last.pgno, 0)) == NULL) {
    283 		t->bt_order = NOT;
    284 		return (NULL);
    285 	}
    286 	t->bt_cur.page = h;
    287 	t->bt_cur.index = t->bt_last.index;
    288 
    289 	/*
    290 	 * If won't fit in this page or have too many keys in this page,
    291 	 * have to search to get split stack.
    292 	 */
    293 	nbytes = NBLEAFDBT(key->size, data->size);
    294 	if (h->upper - h->lower < nbytes + sizeof(indx_t))
    295 		goto miss;
    296 
    297 	if (t->bt_order == FORWARD) {
    298 		if (t->bt_cur.page->nextpg != P_INVALID)
    299 			goto miss;
    300 		if (t->bt_cur.index != NEXTINDEX(h) - 1)
    301 			goto miss;
    302 		if ((cmp = __bt_cmp(t, key, &t->bt_cur)) < 0)
    303 			goto miss;
    304 		t->bt_last.index = cmp ? ++t->bt_cur.index : t->bt_cur.index;
    305 	} else {
    306 		if (t->bt_cur.page->prevpg != P_INVALID)
    307 			goto miss;
    308 		if (t->bt_cur.index != 0)
    309 			goto miss;
    310 		if ((cmp = __bt_cmp(t, key, &t->bt_cur)) > 0)
    311 			goto miss;
    312 		t->bt_last.index = 0;
    313 	}
    314 	*exactp = cmp == 0;
    315 #ifdef STATISTICS
    316 	++bt_cache_hit;
    317 #endif
    318 	return (&t->bt_cur);
    319 
    320 miss:
    321 #ifdef STATISTICS
    322 	++bt_cache_miss;
    323 #endif
    324 	t->bt_order = NOT;
    325 	mpool_put(t->bt_mp, h, 0);
    326 	return (NULL);
    327 }
    328