Home | History | Annotate | Line # | Download | only in btree
bt_put.c revision 1.4
      1 /*-
      2  * Copyright (c) 1990, 1993
      3  *	The Regents of the University of California.  All rights reserved.
      4  *
      5  * This code is derived from software contributed to Berkeley by
      6  * Mike Olson.
      7  *
      8  * Redistribution and use in source and binary forms, with or without
      9  * modification, are permitted provided that the following conditions
     10  * are met:
     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  * 3. All advertising materials mentioning features or use of this software
     17  *    must display the following acknowledgement:
     18  *	This product includes software developed by the University of
     19  *	California, Berkeley and its contributors.
     20  * 4. Neither the name of the University nor the names of its contributors
     21  *    may be used to endorse or promote products derived from this software
     22  *    without specific prior written permission.
     23  *
     24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     34  * SUCH DAMAGE.
     35  */
     36 
     37 #if defined(LIBC_SCCS) && !defined(lint)
     38 /* from: static char sccsid[] = "@(#)bt_put.c	8.2 (Berkeley) 9/7/93"; */
     39 static char *rcsid = "$Id: bt_put.c,v 1.4 1993/09/09 02:41:28 cgd Exp $";
     40 #endif /* LIBC_SCCS and not lint */
     41 
     42 #include <sys/types.h>
     43 
     44 #include <errno.h>
     45 #include <stdio.h>
     46 #include <stdlib.h>
     47 #include <string.h>
     48 
     49 #include <db.h>
     50 #include "btree.h"
     51 
     52 static EPG *bt_fast __P((BTREE *, const DBT *, const DBT *, int *));
     53 
     54 /*
     55  * __BT_PUT -- Add a btree item to the tree.
     56  *
     57  * Parameters:
     58  *	dbp:	pointer to access method
     59  *	key:	key
     60  *	data:	data
     61  *	flag:	R_NOOVERWRITE
     62  *
     63  * Returns:
     64  *	RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key is already in the
     65  *	tree and R_NOOVERWRITE specified.
     66  */
     67 int
     68 __bt_put(dbp, key, data, flags)
     69 	const DB *dbp;
     70 	DBT *key;
     71 	const DBT *data;
     72 	u_int flags;
     73 {
     74 	BTREE *t;
     75 	DBT tkey, tdata;
     76 	EPG *e;
     77 	PAGE *h;
     78 	indx_t index, nxtindex;
     79 	pgno_t pg;
     80 	size_t nbytes;
     81 	int dflags, exact, status;
     82 	char *dest, db[NOVFLSIZE], kb[NOVFLSIZE];
     83 
     84 	t = dbp->internal;
     85 
     86 	/* Toss any page pinned across calls. */
     87 	if (t->bt_pinned != NULL) {
     88 		mpool_put(t->bt_mp, t->bt_pinned, 0);
     89 		t->bt_pinned = NULL;
     90 	}
     91 
     92 	switch (flags) {
     93 	case R_CURSOR:
     94 		if (!ISSET(t, B_SEQINIT))
     95 			goto einval;
     96 		if (ISSET(t, B_DELCRSR))
     97 			goto einval;
     98 		break;
     99 	case 0:
    100 	case R_NOOVERWRITE:
    101 		break;
    102 	default:
    103 einval:		errno = EINVAL;
    104 		return (RET_ERROR);
    105 	}
    106 
    107 	if (ISSET(t, B_RDONLY)) {
    108 		errno = EPERM;
    109 		return (RET_ERROR);
    110 	}
    111 
    112 	/*
    113 	 * If the key/data won't fit on a page, store it on indirect pages.
    114 	 * Only store the key on the overflow page if it's too big after the
    115 	 * data is on an overflow page.
    116 	 *
    117 	 * XXX
    118 	 * If the insert fails later on, these pages aren't recovered.
    119 	 */
    120 	dflags = 0;
    121 	if (key->size + data->size > t->bt_ovflsize) {
    122 		if (key->size > t->bt_ovflsize) {
    123 storekey:		if (__ovfl_put(t, key, &pg) == RET_ERROR)
    124 				return (RET_ERROR);
    125 			tkey.data = kb;
    126 			tkey.size = NOVFLSIZE;
    127 			memmove(kb, &pg, sizeof(pgno_t));
    128 			memmove(kb + sizeof(pgno_t),
    129 			    &key->size, sizeof(size_t));
    130 			dflags |= P_BIGKEY;
    131 			key = &tkey;
    132 		}
    133 		if (key->size + data->size > t->bt_ovflsize) {
    134 			if (__ovfl_put(t, data, &pg) == RET_ERROR)
    135 				return (RET_ERROR);
    136 			tdata.data = db;
    137 			tdata.size = NOVFLSIZE;
    138 			memmove(db, &pg, sizeof(pgno_t));
    139 			memmove(db + sizeof(pgno_t),
    140 			    &data->size, sizeof(size_t));
    141 			dflags |= P_BIGDATA;
    142 			data = &tdata;
    143 		}
    144 		if (key->size + data->size > t->bt_ovflsize)
    145 			goto storekey;
    146 	}
    147 
    148 	/* Replace the cursor. */
    149 	if (flags == R_CURSOR) {
    150 		if ((h = mpool_get(t->bt_mp, t->bt_bcursor.pgno, 0)) == NULL)
    151 			return (RET_ERROR);
    152 		index = t->bt_bcursor.index;
    153 		goto delete;
    154 	}
    155 
    156 	/*
    157 	 * Find the key to delete, or, the location at which to insert.  Bt_fast
    158 	 * and __bt_search pin the returned page.
    159 	 */
    160 	if (t->bt_order == NOT || (e = bt_fast(t, key, data, &exact)) == NULL)
    161 		if ((e = __bt_search(t, key, &exact)) == NULL)
    162 			return (RET_ERROR);
    163 	h = e->page;
    164 	index = e->index;
    165 
    166 	/*
    167 	 * Add the specified key/data pair to the tree.  If an identical key
    168 	 * is already in the tree, and R_NOOVERWRITE is set, an error is
    169 	 * returned.  If R_NOOVERWRITE is not set, the key is either added (if
    170 	 * duplicates are permitted) or an error is returned.
    171 	 *
    172 	 * Pages are split as required.
    173 	 */
    174 	switch (flags) {
    175 	case R_NOOVERWRITE:
    176 		if (!exact)
    177 			break;
    178 		/*
    179 		 * One special case is if the cursor references the record and
    180 		 * it's been flagged for deletion.  Then, we delete the record,
    181 		 * leaving the cursor there -- this means that the inserted
    182 		 * record will not be seen in a cursor scan.
    183 		 */
    184 		if (ISSET(t, B_DELCRSR) && t->bt_bcursor.pgno == h->pgno &&
    185 		    t->bt_bcursor.index == index) {
    186 			CLR(t, B_DELCRSR);
    187 			goto delete;
    188 		}
    189 		mpool_put(t->bt_mp, h, 0);
    190 		return (RET_SPECIAL);
    191 	default:
    192 		if (!exact || !ISSET(t, B_NODUPS))
    193 			break;
    194 delete:		if (__bt_dleaf(t, h, index) == RET_ERROR) {
    195 			mpool_put(t->bt_mp, h, 0);
    196 			return (RET_ERROR);
    197 		}
    198 		break;
    199 	}
    200 
    201 	/*
    202 	 * If not enough room, or the user has put a ceiling on the number of
    203 	 * keys permitted in the page, split the page.  The split code will
    204 	 * insert the key and data and unpin the current page.  If inserting
    205 	 * into the offset array, shift the pointers up.
    206 	 */
    207 	nbytes = NBLEAFDBT(key->size, data->size);
    208 	if (h->upper - h->lower < nbytes + sizeof(indx_t)) {
    209 		if ((status = __bt_split(t, h, key,
    210 		    data, dflags, nbytes, index)) != RET_SUCCESS)
    211 			return (status);
    212 		goto success;
    213 	}
    214 
    215 	if (index < (nxtindex = NEXTINDEX(h)))
    216 		memmove(h->linp + index + 1, h->linp + index,
    217 		    (nxtindex - index) * sizeof(indx_t));
    218 	h->lower += sizeof(indx_t);
    219 
    220 	h->linp[index] = h->upper -= nbytes;
    221 	dest = (char *)h + h->upper;
    222 	WR_BLEAF(dest, key, data, dflags);
    223 
    224 	if (t->bt_order == NOT)
    225 		if (h->nextpg == P_INVALID) {
    226 			if (index == NEXTINDEX(h) - 1) {
    227 				t->bt_order = FORWARD;
    228 				t->bt_last.index = index;
    229 				t->bt_last.pgno = h->pgno;
    230 			}
    231 		} else if (h->prevpg == P_INVALID) {
    232 			if (index == 0) {
    233 				t->bt_order = BACK;
    234 				t->bt_last.index = 0;
    235 				t->bt_last.pgno = h->pgno;
    236 			}
    237 		}
    238 
    239 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
    240 
    241 success:
    242 	if (flags == R_SETCURSOR) {
    243 		t->bt_bcursor.pgno = e->page->pgno;
    244 		t->bt_bcursor.index = e->index;
    245 	}
    246 	SET(t, B_MODIFIED);
    247 	return (RET_SUCCESS);
    248 }
    249 
    250 #ifdef STATISTICS
    251 u_long bt_cache_hit, bt_cache_miss;
    252 #endif
    253 
    254 /*
    255  * BT_FAST -- Do a quick check for sorted data.
    256  *
    257  * Parameters:
    258  *	t:	tree
    259  *	key:	key to insert
    260  *
    261  * Returns:
    262  * 	EPG for new record or NULL if not found.
    263  */
    264 static EPG *
    265 bt_fast(t, key, data, exactp)
    266 	BTREE *t;
    267 	const DBT *key, *data;
    268 	int *exactp;
    269 {
    270 	EPG e;
    271 	PAGE *h;
    272 	size_t nbytes;
    273 	int cmp;
    274 
    275 	if ((h = mpool_get(t->bt_mp, t->bt_last.pgno, 0)) == NULL) {
    276 		t->bt_order = NOT;
    277 		return (NULL);
    278 	}
    279 	e.page = h;
    280 	e.index = t->bt_last.index;
    281 
    282 	/*
    283 	 * If won't fit in this page or have too many keys in this page, have
    284 	 * to search to get split stack.
    285 	 */
    286 	nbytes = NBLEAFDBT(key->size, data->size);
    287 	if (h->upper - h->lower < nbytes + sizeof(indx_t))
    288 		goto miss;
    289 
    290 	if (t->bt_order == FORWARD) {
    291 		if (e.page->nextpg != P_INVALID)
    292 			goto miss;
    293 		if (e.index != NEXTINDEX(h) - 1)
    294 			goto miss;
    295 		if ((cmp = __bt_cmp(t, key, &e)) < 0)
    296 			goto miss;
    297 		t->bt_last.index = cmp ? ++e.index : e.index;
    298 	} else {
    299 		if (e.page->prevpg != P_INVALID)
    300 			goto miss;
    301 		if (e.index != 0)
    302 			goto miss;
    303 		if ((cmp = __bt_cmp(t, key, &e)) > 0)
    304 			goto miss;
    305 		t->bt_last.index = 0;
    306 	}
    307 	*exactp = cmp == 0;
    308 #ifdef STATISTICS
    309 	++bt_cache_hit;
    310 #endif
    311 	return (&e);
    312 
    313 miss:
    314 #ifdef STATISTICS
    315 	++bt_cache_miss;
    316 #endif
    317 	t->bt_order = NOT;
    318 	mpool_put(t->bt_mp, h, 0);
    319 	return (NULL);
    320 }
    321