Home | History | Annotate | Line # | Download | only in btree
bt_put.c revision 1.22
      1  1.22  christos /*	$NetBSD: bt_put.c,v 1.22 2016/09/24 21:31:25 christos Exp $	*/
      2   1.7       cgd 
      3   1.1       cgd /*-
      4   1.6       cgd  * Copyright (c) 1990, 1993, 1994
      5   1.1       cgd  *	The Regents of the University of California.  All rights reserved.
      6   1.1       cgd  *
      7   1.1       cgd  * This code is derived from software contributed to Berkeley by
      8   1.1       cgd  * Mike Olson.
      9   1.1       cgd  *
     10   1.1       cgd  * Redistribution and use in source and binary forms, with or without
     11   1.1       cgd  * modification, are permitted provided that the following conditions
     12   1.1       cgd  * are met:
     13   1.1       cgd  * 1. Redistributions of source code must retain the above copyright
     14   1.1       cgd  *    notice, this list of conditions and the following disclaimer.
     15   1.1       cgd  * 2. Redistributions in binary form must reproduce the above copyright
     16   1.1       cgd  *    notice, this list of conditions and the following disclaimer in the
     17   1.1       cgd  *    documentation and/or other materials provided with the distribution.
     18  1.13       agc  * 3. Neither the name of the University nor the names of its contributors
     19   1.1       cgd  *    may be used to endorse or promote products derived from this software
     20   1.1       cgd  *    without specific prior written permission.
     21   1.1       cgd  *
     22   1.1       cgd  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     23   1.1       cgd  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     24   1.1       cgd  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     25   1.1       cgd  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     26   1.1       cgd  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     27   1.1       cgd  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     28   1.1       cgd  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     29   1.1       cgd  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     30   1.1       cgd  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     31   1.1       cgd  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     32   1.1       cgd  * SUCH DAMAGE.
     33   1.1       cgd  */
     34   1.1       cgd 
     35  1.18     joerg #if HAVE_NBTOOL_CONFIG_H
     36  1.18     joerg #include "nbtool_config.h"
     37  1.18     joerg #endif
     38  1.18     joerg 
     39   1.9  christos #include <sys/cdefs.h>
     40  1.22  christos __RCSID("$NetBSD: bt_put.c,v 1.22 2016/09/24 21:31:25 christos Exp $");
     41   1.1       cgd 
     42  1.10       jtc #include "namespace.h"
     43   1.1       cgd #include <sys/types.h>
     44   1.1       cgd 
     45  1.15  christos #include <assert.h>
     46   1.1       cgd #include <errno.h>
     47   1.1       cgd #include <stdio.h>
     48   1.1       cgd #include <stdlib.h>
     49   1.1       cgd #include <string.h>
     50   1.1       cgd 
     51   1.1       cgd #include <db.h>
     52   1.1       cgd #include "btree.h"
     53   1.1       cgd 
     54  1.15  christos static EPG *bt_fast(BTREE *, const DBT *, const DBT *, int *);
     55   1.1       cgd 
     56   1.1       cgd /*
     57   1.1       cgd  * __BT_PUT -- Add a btree item to the tree.
     58   1.1       cgd  *
     59   1.1       cgd  * Parameters:
     60   1.1       cgd  *	dbp:	pointer to access method
     61   1.1       cgd  *	key:	key
     62   1.1       cgd  *	data:	data
     63   1.1       cgd  *	flag:	R_NOOVERWRITE
     64   1.1       cgd  *
     65   1.1       cgd  * Returns:
     66   1.1       cgd  *	RET_ERROR, RET_SUCCESS and RET_SPECIAL if the key is already in the
     67   1.1       cgd  *	tree and R_NOOVERWRITE specified.
     68   1.1       cgd  */
     69   1.1       cgd int
     70  1.15  christos __bt_put(const DB *dbp, DBT *key, const DBT *data, u_int flags)
     71   1.1       cgd {
     72   1.1       cgd 	BTREE *t;
     73   1.1       cgd 	DBT tkey, tdata;
     74   1.9  christos 	EPG *e = NULL; /* pacify gcc */
     75   1.1       cgd 	PAGE *h;
     76  1.12  christos 	indx_t idx, nxtindex;
     77   1.1       cgd 	pgno_t pg;
     78  1.16     joerg 	uint32_t nbytes, temp;
     79   1.1       cgd 	int dflags, exact, status;
     80   1.1       cgd 	char *dest, db[NOVFLSIZE], kb[NOVFLSIZE];
     81   1.1       cgd 
     82   1.1       cgd 	t = dbp->internal;
     83   1.4       cgd 
     84   1.4       cgd 	/* Toss any page pinned across calls. */
     85   1.4       cgd 	if (t->bt_pinned != NULL) {
     86   1.4       cgd 		mpool_put(t->bt_mp, t->bt_pinned, 0);
     87   1.4       cgd 		t->bt_pinned = NULL;
     88   1.4       cgd 	}
     89   1.1       cgd 
     90   1.8       cgd 	/* Check for change to a read-only tree. */
     91   1.8       cgd 	if (F_ISSET(t, B_RDONLY)) {
     92   1.8       cgd 		errno = EPERM;
     93   1.8       cgd 		return (RET_ERROR);
     94   1.8       cgd 	}
     95   1.8       cgd 
     96   1.1       cgd 	switch (flags) {
     97   1.1       cgd 	case 0:
     98   1.1       cgd 	case R_NOOVERWRITE:
     99   1.1       cgd 		break;
    100   1.8       cgd 	case R_CURSOR:
    101   1.8       cgd 		/*
    102   1.8       cgd 		 * If flags is R_CURSOR, put the cursor.  Must already
    103   1.8       cgd 		 * have started a scan and not have already deleted it.
    104   1.8       cgd 		 */
    105   1.8       cgd 		if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
    106   1.8       cgd 		    !F_ISSET(&t->bt_cursor,
    107   1.8       cgd 		        CURS_ACQUIRE | CURS_AFTER | CURS_BEFORE))
    108   1.8       cgd 			break;
    109   1.8       cgd 		/* FALLTHROUGH */
    110   1.1       cgd 	default:
    111   1.8       cgd 		errno = EINVAL;
    112   1.1       cgd 		return (RET_ERROR);
    113   1.1       cgd 	}
    114   1.1       cgd 
    115   1.1       cgd 	/*
    116   1.8       cgd 	 * If the key/data pair won't fit on a page, store it on overflow
    117   1.8       cgd 	 * pages.  Only put the key on the overflow page if the pair are
    118   1.8       cgd 	 * still too big after moving the data to an overflow page.
    119   1.1       cgd 	 *
    120   1.1       cgd 	 * XXX
    121   1.8       cgd 	 * If the insert fails later on, the overflow pages aren't recovered.
    122   1.1       cgd 	 */
    123   1.1       cgd 	dflags = 0;
    124   1.1       cgd 	if (key->size + data->size > t->bt_ovflsize) {
    125   1.1       cgd 		if (key->size > t->bt_ovflsize) {
    126   1.1       cgd storekey:		if (__ovfl_put(t, key, &pg) == RET_ERROR)
    127   1.1       cgd 				return (RET_ERROR);
    128   1.1       cgd 			tkey.data = kb;
    129   1.1       cgd 			tkey.size = NOVFLSIZE;
    130  1.20  christos 			memmove(kb, &pg, sizeof(pg));
    131   1.1       cgd 			memmove(kb + sizeof(pgno_t),
    132  1.16     joerg 			    &key->size, sizeof(uint32_t));
    133   1.1       cgd 			dflags |= P_BIGKEY;
    134   1.1       cgd 			key = &tkey;
    135   1.1       cgd 		}
    136   1.1       cgd 		if (key->size + data->size > t->bt_ovflsize) {
    137   1.1       cgd 			if (__ovfl_put(t, data, &pg) == RET_ERROR)
    138   1.1       cgd 				return (RET_ERROR);
    139   1.1       cgd 			tdata.data = db;
    140   1.1       cgd 			tdata.size = NOVFLSIZE;
    141  1.20  christos 			memmove(db, &pg, sizeof(pg));
    142  1.16     joerg 			_DBFIT(data->size, uint32_t);
    143  1.16     joerg 			temp = (uint32_t)data->size;
    144  1.15  christos 			(void)memmove(db + sizeof(pgno_t),
    145  1.16     joerg 			    &temp, sizeof(uint32_t));
    146   1.1       cgd 			dflags |= P_BIGDATA;
    147   1.1       cgd 			data = &tdata;
    148   1.1       cgd 		}
    149   1.1       cgd 		if (key->size + data->size > t->bt_ovflsize)
    150   1.1       cgd 			goto storekey;
    151   1.1       cgd 	}
    152   1.1       cgd 
    153   1.1       cgd 	/* Replace the cursor. */
    154   1.1       cgd 	if (flags == R_CURSOR) {
    155  1.22  christos 		if ((h = mpool_get(t->bt_mp, t->bt_cursor.pg.pgno, 0)) == NULL)
    156   1.1       cgd 			return (RET_ERROR);
    157  1.12  christos 		idx = t->bt_cursor.pg.index;
    158   1.1       cgd 		goto delete;
    159   1.1       cgd 	}
    160   1.1       cgd 
    161   1.1       cgd 	/*
    162   1.8       cgd 	 * Find the key to delete, or, the location at which to insert.
    163   1.8       cgd 	 * Bt_fast and __bt_search both pin the returned page.
    164   1.1       cgd 	 */
    165   1.1       cgd 	if (t->bt_order == NOT || (e = bt_fast(t, key, data, &exact)) == NULL)
    166   1.1       cgd 		if ((e = __bt_search(t, key, &exact)) == NULL)
    167   1.1       cgd 			return (RET_ERROR);
    168   1.1       cgd 	h = e->page;
    169  1.12  christos 	idx = e->index;
    170   1.1       cgd 
    171   1.1       cgd 	/*
    172   1.8       cgd 	 * Add the key/data pair to the tree.  If an identical key is already
    173   1.8       cgd 	 * in the tree, and R_NOOVERWRITE is set, an error is returned.  If
    174   1.8       cgd 	 * R_NOOVERWRITE is not set, the key is either added (if duplicates are
    175   1.8       cgd 	 * permitted) or an error is returned.
    176   1.1       cgd 	 */
    177   1.1       cgd 	switch (flags) {
    178   1.1       cgd 	case R_NOOVERWRITE:
    179   1.1       cgd 		if (!exact)
    180   1.1       cgd 			break;
    181   1.1       cgd 		mpool_put(t->bt_mp, h, 0);
    182   1.1       cgd 		return (RET_SPECIAL);
    183   1.1       cgd 	default:
    184   1.8       cgd 		if (!exact || !F_ISSET(t, B_NODUPS))
    185   1.1       cgd 			break;
    186   1.8       cgd 		/*
    187   1.8       cgd 		 * !!!
    188   1.8       cgd 		 * Note, the delete may empty the page, so we need to put a
    189   1.8       cgd 		 * new entry into the page immediately.
    190   1.8       cgd 		 */
    191  1.12  christos delete:		if (__bt_dleaf(t, key, h, (u_int)idx) == RET_ERROR) {
    192   1.1       cgd 			mpool_put(t->bt_mp, h, 0);
    193   1.1       cgd 			return (RET_ERROR);
    194   1.1       cgd 		}
    195   1.1       cgd 		break;
    196   1.1       cgd 	}
    197   1.1       cgd 
    198   1.1       cgd 	/*
    199   1.1       cgd 	 * If not enough room, or the user has put a ceiling on the number of
    200   1.1       cgd 	 * keys permitted in the page, split the page.  The split code will
    201   1.1       cgd 	 * insert the key and data and unpin the current page.  If inserting
    202   1.1       cgd 	 * into the offset array, shift the pointers up.
    203   1.1       cgd 	 */
    204   1.1       cgd 	nbytes = NBLEAFDBT(key->size, data->size);
    205  1.19     lukem 	if ((uint32_t)h->upper - (uint32_t)h->lower < nbytes + sizeof(indx_t)) {
    206   1.1       cgd 		if ((status = __bt_split(t, h, key,
    207  1.12  christos 		    data, dflags, nbytes, (u_int)idx)) != RET_SUCCESS)
    208   1.1       cgd 			return (status);
    209   1.1       cgd 		goto success;
    210   1.1       cgd 	}
    211   1.1       cgd 
    212  1.12  christos 	if (idx < (nxtindex = NEXTINDEX(h)))
    213  1.12  christos 		memmove(h->linp + idx + 1, h->linp + idx,
    214  1.12  christos 		    (nxtindex - idx) * sizeof(indx_t));
    215   1.1       cgd 	h->lower += sizeof(indx_t);
    216   1.1       cgd 
    217  1.12  christos 	h->linp[idx] = h->upper -= nbytes;
    218  1.12  christos 	dest = (char *)(void *)h + h->upper;
    219   1.1       cgd 	WR_BLEAF(dest, key, data, dflags);
    220   1.1       cgd 
    221   1.8       cgd 	/* If the cursor is on this page, adjust it as necessary. */
    222   1.8       cgd 	if (F_ISSET(&t->bt_cursor, CURS_INIT) &&
    223   1.8       cgd 	    !F_ISSET(&t->bt_cursor, CURS_ACQUIRE) &&
    224  1.12  christos 	    t->bt_cursor.pg.pgno == h->pgno && t->bt_cursor.pg.index >= idx)
    225   1.8       cgd 		++t->bt_cursor.pg.index;
    226   1.8       cgd 
    227  1.11   thorpej 	if (t->bt_order == NOT) {
    228   1.1       cgd 		if (h->nextpg == P_INVALID) {
    229  1.12  christos 			if (idx == NEXTINDEX(h) - 1) {
    230   1.1       cgd 				t->bt_order = FORWARD;
    231  1.12  christos 				t->bt_last.index = idx;
    232   1.1       cgd 				t->bt_last.pgno = h->pgno;
    233   1.1       cgd 			}
    234   1.1       cgd 		} else if (h->prevpg == P_INVALID) {
    235  1.12  christos 			if (idx == 0) {
    236   1.1       cgd 				t->bt_order = BACK;
    237   1.1       cgd 				t->bt_last.index = 0;
    238   1.1       cgd 				t->bt_last.pgno = h->pgno;
    239   1.1       cgd 			}
    240   1.1       cgd 		}
    241  1.11   thorpej 	}
    242   1.1       cgd 
    243   1.1       cgd 	mpool_put(t->bt_mp, h, MPOOL_DIRTY);
    244   1.1       cgd 
    245   1.1       cgd success:
    246   1.8       cgd 	if (flags == R_SETCURSOR)
    247  1.12  christos 		__bt_setcur(t, e->page->pgno, (u_int)e->index);
    248   1.8       cgd 
    249   1.8       cgd 	F_SET(t, B_MODIFIED);
    250   1.1       cgd 	return (RET_SUCCESS);
    251   1.1       cgd }
    252   1.1       cgd 
    253   1.1       cgd #ifdef STATISTICS
    254  1.16     joerg unsigned long bt_cache_hit, bt_cache_miss;
    255   1.1       cgd #endif
    256   1.1       cgd 
    257   1.1       cgd /*
    258   1.1       cgd  * BT_FAST -- Do a quick check for sorted data.
    259   1.1       cgd  *
    260   1.1       cgd  * Parameters:
    261   1.1       cgd  *	t:	tree
    262   1.1       cgd  *	key:	key to insert
    263   1.1       cgd  *
    264   1.1       cgd  * Returns:
    265   1.1       cgd  * 	EPG for new record or NULL if not found.
    266   1.1       cgd  */
    267   1.1       cgd static EPG *
    268  1.15  christos bt_fast(BTREE *t, const DBT *key, const DBT *data, int *exactp)
    269   1.1       cgd {
    270   1.1       cgd 	PAGE *h;
    271  1.16     joerg 	uint32_t nbytes;
    272   1.1       cgd 	int cmp;
    273   1.1       cgd 
    274  1.22  christos 	if ((h = mpool_get(t->bt_mp, t->bt_last.pgno, 0)) == NULL) {
    275   1.1       cgd 		t->bt_order = NOT;
    276   1.1       cgd 		return (NULL);
    277   1.1       cgd 	}
    278   1.5       cgd 	t->bt_cur.page = h;
    279   1.5       cgd 	t->bt_cur.index = t->bt_last.index;
    280   1.1       cgd 
    281   1.1       cgd 	/*
    282   1.8       cgd 	 * If won't fit in this page or have too many keys in this page,
    283   1.8       cgd 	 * have to search to get split stack.
    284   1.1       cgd 	 */
    285   1.1       cgd 	nbytes = NBLEAFDBT(key->size, data->size);
    286  1.19     lukem 	if ((uint32_t)h->upper - (uint32_t)h->lower < nbytes + sizeof(indx_t))
    287   1.1       cgd 		goto miss;
    288   1.1       cgd 
    289   1.1       cgd 	if (t->bt_order == FORWARD) {
    290   1.5       cgd 		if (t->bt_cur.page->nextpg != P_INVALID)
    291   1.1       cgd 			goto miss;
    292   1.5       cgd 		if (t->bt_cur.index != NEXTINDEX(h) - 1)
    293   1.1       cgd 			goto miss;
    294   1.5       cgd 		if ((cmp = __bt_cmp(t, key, &t->bt_cur)) < 0)
    295   1.1       cgd 			goto miss;
    296   1.5       cgd 		t->bt_last.index = cmp ? ++t->bt_cur.index : t->bt_cur.index;
    297   1.1       cgd 	} else {
    298   1.5       cgd 		if (t->bt_cur.page->prevpg != P_INVALID)
    299   1.1       cgd 			goto miss;
    300   1.5       cgd 		if (t->bt_cur.index != 0)
    301   1.1       cgd 			goto miss;
    302   1.5       cgd 		if ((cmp = __bt_cmp(t, key, &t->bt_cur)) > 0)
    303   1.1       cgd 			goto miss;
    304   1.1       cgd 		t->bt_last.index = 0;
    305   1.1       cgd 	}
    306   1.1       cgd 	*exactp = cmp == 0;
    307   1.1       cgd #ifdef STATISTICS
    308   1.1       cgd 	++bt_cache_hit;
    309   1.1       cgd #endif
    310   1.5       cgd 	return (&t->bt_cur);
    311   1.1       cgd 
    312   1.1       cgd miss:
    313   1.1       cgd #ifdef STATISTICS
    314   1.1       cgd 	++bt_cache_miss;
    315   1.1       cgd #endif
    316   1.1       cgd 	t->bt_order = NOT;
    317   1.1       cgd 	mpool_put(t->bt_mp, h, 0);
    318   1.1       cgd 	return (NULL);
    319   1.1       cgd }
    320