17c478bd9Sstevel@tonic-gate /*-
27c478bd9Sstevel@tonic-gate * Copyright (c) 1990, 1993, 1994
37c478bd9Sstevel@tonic-gate * The Regents of the University of California. All rights reserved.
47c478bd9Sstevel@tonic-gate *
57c478bd9Sstevel@tonic-gate * This code is derived from software contributed to Berkeley by
67c478bd9Sstevel@tonic-gate * Mike Olson.
77c478bd9Sstevel@tonic-gate *
87c478bd9Sstevel@tonic-gate * Redistribution and use in source and binary forms, with or without
97c478bd9Sstevel@tonic-gate * modification, are permitted provided that the following conditions
107c478bd9Sstevel@tonic-gate * are met:
117c478bd9Sstevel@tonic-gate * 1. Redistributions of source code must retain the above copyright
127c478bd9Sstevel@tonic-gate * notice, this list of conditions and the following disclaimer.
137c478bd9Sstevel@tonic-gate * 2. Redistributions in binary form must reproduce the above copyright
147c478bd9Sstevel@tonic-gate * notice, this list of conditions and the following disclaimer in the
157c478bd9Sstevel@tonic-gate * documentation and/or other materials provided with the distribution.
167c478bd9Sstevel@tonic-gate * 3. All advertising materials mentioning features or use of this software
177c478bd9Sstevel@tonic-gate * must display the following acknowledgement:
187c478bd9Sstevel@tonic-gate * This product includes software developed by the University of
197c478bd9Sstevel@tonic-gate * California, Berkeley and its contributors.
207c478bd9Sstevel@tonic-gate * 4. Neither the name of the University nor the names of its contributors
217c478bd9Sstevel@tonic-gate * may be used to endorse or promote products derived from this software
227c478bd9Sstevel@tonic-gate * without specific prior written permission.
237c478bd9Sstevel@tonic-gate *
247c478bd9Sstevel@tonic-gate * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
257c478bd9Sstevel@tonic-gate * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
267c478bd9Sstevel@tonic-gate * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
277c478bd9Sstevel@tonic-gate * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
287c478bd9Sstevel@tonic-gate * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
297c478bd9Sstevel@tonic-gate * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
307c478bd9Sstevel@tonic-gate * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
317c478bd9Sstevel@tonic-gate * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
327c478bd9Sstevel@tonic-gate * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
337c478bd9Sstevel@tonic-gate * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
347c478bd9Sstevel@tonic-gate * SUCH DAMAGE.
357c478bd9Sstevel@tonic-gate */
367c478bd9Sstevel@tonic-gate
377c478bd9Sstevel@tonic-gate #if defined(LIBC_SCCS) && !defined(lint)
387c478bd9Sstevel@tonic-gate static char sccsid[] = "@(#)bt_open.c 8.11 (Berkeley) 11/2/95";
397c478bd9Sstevel@tonic-gate #endif /* LIBC_SCCS and not lint */
407c478bd9Sstevel@tonic-gate
417c478bd9Sstevel@tonic-gate /*
427c478bd9Sstevel@tonic-gate * Implementation of btree access method for 4.4BSD.
437c478bd9Sstevel@tonic-gate *
447c478bd9Sstevel@tonic-gate * The design here was originally based on that of the btree access method
457c478bd9Sstevel@tonic-gate * used in the Postgres database system at UC Berkeley. This implementation
467c478bd9Sstevel@tonic-gate * is wholly independent of the Postgres code.
477c478bd9Sstevel@tonic-gate */
487c478bd9Sstevel@tonic-gate
497c478bd9Sstevel@tonic-gate #include <sys/param.h>
507c478bd9Sstevel@tonic-gate #include <sys/stat.h>
517c478bd9Sstevel@tonic-gate
527c478bd9Sstevel@tonic-gate #include <errno.h>
537c478bd9Sstevel@tonic-gate #include <fcntl.h>
547c478bd9Sstevel@tonic-gate #include <limits.h>
557c478bd9Sstevel@tonic-gate #include <signal.h>
567c478bd9Sstevel@tonic-gate #include <stdio.h>
577c478bd9Sstevel@tonic-gate #include <stdlib.h>
587c478bd9Sstevel@tonic-gate #include <string.h>
597c478bd9Sstevel@tonic-gate #include <unistd.h>
607c478bd9Sstevel@tonic-gate
617c478bd9Sstevel@tonic-gate #include "db-int.h"
627c478bd9Sstevel@tonic-gate #include "btree.h"
637c478bd9Sstevel@tonic-gate
6456a424ccSmp #ifdef DEBUG
657c478bd9Sstevel@tonic-gate #undef MINPSIZE
667c478bd9Sstevel@tonic-gate #define MINPSIZE 128
677c478bd9Sstevel@tonic-gate #endif
687c478bd9Sstevel@tonic-gate
697c478bd9Sstevel@tonic-gate static int byteorder __P((void));
707c478bd9Sstevel@tonic-gate static int nroot __P((BTREE *));
717c478bd9Sstevel@tonic-gate static int tmp __P((void));
727c478bd9Sstevel@tonic-gate
737c478bd9Sstevel@tonic-gate /*
747c478bd9Sstevel@tonic-gate * __BT_OPEN -- Open a btree.
757c478bd9Sstevel@tonic-gate *
767c478bd9Sstevel@tonic-gate * Creates and fills a DB struct, and calls the routine that actually
777c478bd9Sstevel@tonic-gate * opens the btree.
787c478bd9Sstevel@tonic-gate *
797c478bd9Sstevel@tonic-gate * Parameters:
807c478bd9Sstevel@tonic-gate * fname: filename (NULL for in-memory trees)
817c478bd9Sstevel@tonic-gate * flags: open flag bits
827c478bd9Sstevel@tonic-gate * mode: open permission bits
837c478bd9Sstevel@tonic-gate * b: BTREEINFO pointer
847c478bd9Sstevel@tonic-gate *
857c478bd9Sstevel@tonic-gate * Returns:
867c478bd9Sstevel@tonic-gate * NULL on failure, pointer to DB on success.
877c478bd9Sstevel@tonic-gate *
887c478bd9Sstevel@tonic-gate */
897c478bd9Sstevel@tonic-gate DB *
__bt_open(fname,flags,mode,openinfo,dflags)907c478bd9Sstevel@tonic-gate __bt_open(fname, flags, mode, openinfo, dflags)
917c478bd9Sstevel@tonic-gate const char *fname;
927c478bd9Sstevel@tonic-gate int flags, mode, dflags;
937c478bd9Sstevel@tonic-gate const BTREEINFO *openinfo;
947c478bd9Sstevel@tonic-gate {
957c478bd9Sstevel@tonic-gate struct stat sb;
967c478bd9Sstevel@tonic-gate BTMETA m;
977c478bd9Sstevel@tonic-gate BTREE *t;
987c478bd9Sstevel@tonic-gate BTREEINFO b;
997c478bd9Sstevel@tonic-gate DB *dbp;
1007c478bd9Sstevel@tonic-gate db_pgno_t ncache;
1017c478bd9Sstevel@tonic-gate ssize_t nr;
1027c478bd9Sstevel@tonic-gate int machine_lorder;
1037c478bd9Sstevel@tonic-gate
1047c478bd9Sstevel@tonic-gate t = NULL;
1057c478bd9Sstevel@tonic-gate
1067c478bd9Sstevel@tonic-gate /*
1077c478bd9Sstevel@tonic-gate * Intention is to make sure all of the user's selections are okay
1087c478bd9Sstevel@tonic-gate * here and then use them without checking. Can't be complete, since
1097c478bd9Sstevel@tonic-gate * we don't know the right page size, lorder or flags until the backing
1107c478bd9Sstevel@tonic-gate * file is opened. Also, the file's page size can cause the cachesize
1117c478bd9Sstevel@tonic-gate * to change.
1127c478bd9Sstevel@tonic-gate */
1137c478bd9Sstevel@tonic-gate machine_lorder = byteorder();
1147c478bd9Sstevel@tonic-gate if (openinfo) {
1157c478bd9Sstevel@tonic-gate b = *openinfo;
1167c478bd9Sstevel@tonic-gate
1177c478bd9Sstevel@tonic-gate /* Flags: R_DUP. */
1187c478bd9Sstevel@tonic-gate if (b.flags & ~(R_DUP))
1197c478bd9Sstevel@tonic-gate goto einval;
1207c478bd9Sstevel@tonic-gate
1217c478bd9Sstevel@tonic-gate /*
1227c478bd9Sstevel@tonic-gate * Page size must be indx_t aligned and >= MINPSIZE. Default
1237c478bd9Sstevel@tonic-gate * page size is set farther on, based on the underlying file
1247c478bd9Sstevel@tonic-gate * transfer size.
1257c478bd9Sstevel@tonic-gate */
1267c478bd9Sstevel@tonic-gate if (b.psize &&
1277c478bd9Sstevel@tonic-gate (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 ||
12856a424ccSmp b.psize & (sizeof(indx_t) - 1)))
1297c478bd9Sstevel@tonic-gate goto einval;
1307c478bd9Sstevel@tonic-gate
1317c478bd9Sstevel@tonic-gate /* Minimum number of keys per page; absolute minimum is 2. */
1327c478bd9Sstevel@tonic-gate if (b.minkeypage) {
1337c478bd9Sstevel@tonic-gate if (b.minkeypage < 2)
1347c478bd9Sstevel@tonic-gate goto einval;
1357c478bd9Sstevel@tonic-gate } else
1367c478bd9Sstevel@tonic-gate b.minkeypage = DEFMINKEYPAGE;
1377c478bd9Sstevel@tonic-gate
1387c478bd9Sstevel@tonic-gate /* If no comparison, use default comparison and prefix. */
1397c478bd9Sstevel@tonic-gate if (b.compare == NULL) {
1407c478bd9Sstevel@tonic-gate b.compare = __bt_defcmp;
1417c478bd9Sstevel@tonic-gate if (b.prefix == NULL)
1427c478bd9Sstevel@tonic-gate b.prefix = __bt_defpfx;
1437c478bd9Sstevel@tonic-gate }
1447c478bd9Sstevel@tonic-gate
1457c478bd9Sstevel@tonic-gate if (b.lorder == 0)
1467c478bd9Sstevel@tonic-gate b.lorder = machine_lorder;
1477c478bd9Sstevel@tonic-gate } else {
1487c478bd9Sstevel@tonic-gate b.compare = __bt_defcmp;
1497c478bd9Sstevel@tonic-gate b.cachesize = 0;
1507c478bd9Sstevel@tonic-gate b.flags = 0;
1517c478bd9Sstevel@tonic-gate b.lorder = machine_lorder;
1527c478bd9Sstevel@tonic-gate b.minkeypage = DEFMINKEYPAGE;
1537c478bd9Sstevel@tonic-gate b.prefix = __bt_defpfx;
1547c478bd9Sstevel@tonic-gate b.psize = 0;
1557c478bd9Sstevel@tonic-gate }
1567c478bd9Sstevel@tonic-gate
1577c478bd9Sstevel@tonic-gate /* Check for the ubiquitous PDP-11. */
1587c478bd9Sstevel@tonic-gate if (b.lorder != DB_BIG_ENDIAN && b.lorder != DB_LITTLE_ENDIAN)
1597c478bd9Sstevel@tonic-gate goto einval;
1607c478bd9Sstevel@tonic-gate
1617c478bd9Sstevel@tonic-gate /* Allocate and initialize DB and BTREE structures. */
1627c478bd9Sstevel@tonic-gate if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL)
1637c478bd9Sstevel@tonic-gate goto err;
1647c478bd9Sstevel@tonic-gate memset(t, 0, sizeof(BTREE));
1657c478bd9Sstevel@tonic-gate t->bt_fd = -1; /* Don't close unopened fd on error. */
1667c478bd9Sstevel@tonic-gate t->bt_lorder = b.lorder;
1677c478bd9Sstevel@tonic-gate t->bt_order = NOT;
1687c478bd9Sstevel@tonic-gate t->bt_cmp = b.compare;
1697c478bd9Sstevel@tonic-gate t->bt_pfx = b.prefix;
1707c478bd9Sstevel@tonic-gate t->bt_rfd = -1;
1717c478bd9Sstevel@tonic-gate
1727c478bd9Sstevel@tonic-gate if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL)
1737c478bd9Sstevel@tonic-gate goto err;
1747c478bd9Sstevel@tonic-gate memset(t->bt_dbp, 0, sizeof(DB));
1757c478bd9Sstevel@tonic-gate if (t->bt_lorder != machine_lorder)
1767c478bd9Sstevel@tonic-gate F_SET(t, B_NEEDSWAP);
1777c478bd9Sstevel@tonic-gate
1787c478bd9Sstevel@tonic-gate dbp->type = DB_BTREE;
1797c478bd9Sstevel@tonic-gate dbp->internal = t;
1807c478bd9Sstevel@tonic-gate dbp->close = __bt_close;
1817c478bd9Sstevel@tonic-gate dbp->del = __bt_delete;
1827c478bd9Sstevel@tonic-gate dbp->fd = __bt_fd;
1837c478bd9Sstevel@tonic-gate dbp->get = __bt_get;
1847c478bd9Sstevel@tonic-gate dbp->put = __bt_put;
1857c478bd9Sstevel@tonic-gate dbp->seq = __bt_seq;
1867c478bd9Sstevel@tonic-gate dbp->sync = __bt_sync;
1877c478bd9Sstevel@tonic-gate
1887c478bd9Sstevel@tonic-gate /*
1897c478bd9Sstevel@tonic-gate * If no file name was supplied, this is an in-memory btree and we
1907c478bd9Sstevel@tonic-gate * open a backing temporary file. Otherwise, it's a disk-based tree.
1917c478bd9Sstevel@tonic-gate */
1927c478bd9Sstevel@tonic-gate if (fname) {
1937c478bd9Sstevel@tonic-gate switch (flags & O_ACCMODE) {
1947c478bd9Sstevel@tonic-gate case O_RDONLY:
1957c478bd9Sstevel@tonic-gate F_SET(t, B_RDONLY);
1967c478bd9Sstevel@tonic-gate break;
1977c478bd9Sstevel@tonic-gate case O_RDWR:
1987c478bd9Sstevel@tonic-gate break;
1997c478bd9Sstevel@tonic-gate case O_WRONLY:
2007c478bd9Sstevel@tonic-gate default:
2017c478bd9Sstevel@tonic-gate goto einval;
2027c478bd9Sstevel@tonic-gate }
203*1da57d55SToomas Soome
2047c478bd9Sstevel@tonic-gate if ((t->bt_fd = open(fname, flags | O_BINARY, mode)) < 0)
2057c478bd9Sstevel@tonic-gate goto err;
2067c478bd9Sstevel@tonic-gate
2077c478bd9Sstevel@tonic-gate } else {
2087c478bd9Sstevel@tonic-gate if ((flags & O_ACCMODE) != O_RDWR)
2097c478bd9Sstevel@tonic-gate goto einval;
2107c478bd9Sstevel@tonic-gate if ((t->bt_fd = tmp()) == -1)
2117c478bd9Sstevel@tonic-gate goto err;
2127c478bd9Sstevel@tonic-gate F_SET(t, B_INMEM);
2137c478bd9Sstevel@tonic-gate }
2147c478bd9Sstevel@tonic-gate
2157c478bd9Sstevel@tonic-gate if (fcntl(t->bt_fd, F_SETFD, 1) == -1)
2167c478bd9Sstevel@tonic-gate goto err;
2177c478bd9Sstevel@tonic-gate
2187c478bd9Sstevel@tonic-gate if (fstat(t->bt_fd, &sb))
2197c478bd9Sstevel@tonic-gate goto err;
2207c478bd9Sstevel@tonic-gate if (sb.st_size) {
2217c478bd9Sstevel@tonic-gate if ((nr = read(t->bt_fd, &m, sizeof(BTMETA))) < 0)
2227c478bd9Sstevel@tonic-gate goto err;
2237c478bd9Sstevel@tonic-gate if (nr != sizeof(BTMETA))
2247c478bd9Sstevel@tonic-gate goto eftype;
2257c478bd9Sstevel@tonic-gate
2267c478bd9Sstevel@tonic-gate /*
2277c478bd9Sstevel@tonic-gate * Read in the meta-data. This can change the notion of what
2287c478bd9Sstevel@tonic-gate * the lorder, page size and flags are, and, when the page size
2297c478bd9Sstevel@tonic-gate * changes, the cachesize value can change too. If the user
2307c478bd9Sstevel@tonic-gate * specified the wrong byte order for an existing database, we
2317c478bd9Sstevel@tonic-gate * don't bother to return an error, we just clear the NEEDSWAP
2327c478bd9Sstevel@tonic-gate * bit.
2337c478bd9Sstevel@tonic-gate */
2347c478bd9Sstevel@tonic-gate if (m.magic == BTREEMAGIC)
2357c478bd9Sstevel@tonic-gate F_CLR(t, B_NEEDSWAP);
2367c478bd9Sstevel@tonic-gate else {
2377c478bd9Sstevel@tonic-gate F_SET(t, B_NEEDSWAP);
2387c478bd9Sstevel@tonic-gate M_32_SWAP(m.magic);
2397c478bd9Sstevel@tonic-gate M_32_SWAP(m.version);
2407c478bd9Sstevel@tonic-gate M_32_SWAP(m.psize);
2417c478bd9Sstevel@tonic-gate M_32_SWAP(m.free);
2427c478bd9Sstevel@tonic-gate M_32_SWAP(m.nrecs);
2437c478bd9Sstevel@tonic-gate M_32_SWAP(m.flags);
2447c478bd9Sstevel@tonic-gate }
2457c478bd9Sstevel@tonic-gate if (m.magic != BTREEMAGIC || m.version != BTREEVERSION)
2467c478bd9Sstevel@tonic-gate goto eftype;
2477c478bd9Sstevel@tonic-gate if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 ||
24856a424ccSmp m.psize & (sizeof(indx_t) - 1))
2497c478bd9Sstevel@tonic-gate goto eftype;
2507c478bd9Sstevel@tonic-gate if (m.flags & ~SAVEMETA)
2517c478bd9Sstevel@tonic-gate goto eftype;
2527c478bd9Sstevel@tonic-gate b.psize = m.psize;
2537c478bd9Sstevel@tonic-gate F_SET(t, m.flags);
2547c478bd9Sstevel@tonic-gate t->bt_free = m.free;
2557c478bd9Sstevel@tonic-gate t->bt_nrecs = m.nrecs;
2567c478bd9Sstevel@tonic-gate } else {
2577c478bd9Sstevel@tonic-gate /*
2587c478bd9Sstevel@tonic-gate * Set the page size to the best value for I/O to this file.
2597c478bd9Sstevel@tonic-gate * Don't overflow the page offset type.
2607c478bd9Sstevel@tonic-gate */
2617c478bd9Sstevel@tonic-gate if (b.psize == 0) {
2627c478bd9Sstevel@tonic-gate b.psize = sb.st_blksize;
2637c478bd9Sstevel@tonic-gate if (b.psize < MINPSIZE)
2647c478bd9Sstevel@tonic-gate b.psize = MINPSIZE;
2657c478bd9Sstevel@tonic-gate if (b.psize > MAX_PAGE_OFFSET + 1)
2667c478bd9Sstevel@tonic-gate b.psize = MAX_PAGE_OFFSET + 1;
2677c478bd9Sstevel@tonic-gate }
2687c478bd9Sstevel@tonic-gate
2697c478bd9Sstevel@tonic-gate /* Set flag if duplicates permitted. */
2707c478bd9Sstevel@tonic-gate if (!(b.flags & R_DUP))
2717c478bd9Sstevel@tonic-gate F_SET(t, B_NODUPS);
2727c478bd9Sstevel@tonic-gate
2737c478bd9Sstevel@tonic-gate t->bt_free = P_INVALID;
2747c478bd9Sstevel@tonic-gate t->bt_nrecs = 0;
2757c478bd9Sstevel@tonic-gate F_SET(t, B_METADIRTY);
2767c478bd9Sstevel@tonic-gate }
2777c478bd9Sstevel@tonic-gate
2787c478bd9Sstevel@tonic-gate t->bt_psize = b.psize;
2797c478bd9Sstevel@tonic-gate
2807c478bd9Sstevel@tonic-gate /* Set the cache size; must be a multiple of the page size. */
28156a424ccSmp if (b.cachesize && b.cachesize & (b.psize - 1))
28256a424ccSmp b.cachesize += (~b.cachesize & (b.psize - 1)) + 1;
2837c478bd9Sstevel@tonic-gate if (b.cachesize < b.psize * MINCACHE)
2847c478bd9Sstevel@tonic-gate b.cachesize = b.psize * MINCACHE;
2857c478bd9Sstevel@tonic-gate
2867c478bd9Sstevel@tonic-gate /* Calculate number of pages to cache. */
2877c478bd9Sstevel@tonic-gate ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize;
2887c478bd9Sstevel@tonic-gate
2897c478bd9Sstevel@tonic-gate /*
2907c478bd9Sstevel@tonic-gate * The btree data structure requires that at least two keys can fit on
2917c478bd9Sstevel@tonic-gate * a page, but other than that there's no fixed requirement. The user
2927c478bd9Sstevel@tonic-gate * specified a minimum number per page, and we translated that into the
2937c478bd9Sstevel@tonic-gate * number of bytes a key/data pair can use before being placed on an
2947c478bd9Sstevel@tonic-gate * overflow page. This calculation includes the page header, the size
2957c478bd9Sstevel@tonic-gate * of the index referencing the leaf item and the size of the leaf item
2967c478bd9Sstevel@tonic-gate * structure. Also, don't let the user specify a minkeypage such that
2977c478bd9Sstevel@tonic-gate * a key/data pair won't fit even if both key and data are on overflow
2987c478bd9Sstevel@tonic-gate * pages.
2997c478bd9Sstevel@tonic-gate */
3007c478bd9Sstevel@tonic-gate t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage -
3017c478bd9Sstevel@tonic-gate (sizeof(indx_t) + NBLEAFDBT(0, 0));
3027c478bd9Sstevel@tonic-gate if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t))
3037c478bd9Sstevel@tonic-gate t->bt_ovflsize =
3047c478bd9Sstevel@tonic-gate NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t);
3057c478bd9Sstevel@tonic-gate
3067c478bd9Sstevel@tonic-gate /* Initialize the buffer pool. */
3077c478bd9Sstevel@tonic-gate if ((t->bt_mp =
3087c478bd9Sstevel@tonic-gate mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL)
3097c478bd9Sstevel@tonic-gate goto err;
3107c478bd9Sstevel@tonic-gate if (!F_ISSET(t, B_INMEM))
3117c478bd9Sstevel@tonic-gate mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t);
3127c478bd9Sstevel@tonic-gate
3137c478bd9Sstevel@tonic-gate /* Create a root page if new tree. */
3147c478bd9Sstevel@tonic-gate if (nroot(t) == RET_ERROR)
3157c478bd9Sstevel@tonic-gate goto err;
3167c478bd9Sstevel@tonic-gate
3177c478bd9Sstevel@tonic-gate /* Global flags. */
3187c478bd9Sstevel@tonic-gate if (dflags & DB_LOCK)
3197c478bd9Sstevel@tonic-gate F_SET(t, B_DB_LOCK);
3207c478bd9Sstevel@tonic-gate if (dflags & DB_SHMEM)
3217c478bd9Sstevel@tonic-gate F_SET(t, B_DB_SHMEM);
3227c478bd9Sstevel@tonic-gate if (dflags & DB_TXN)
3237c478bd9Sstevel@tonic-gate F_SET(t, B_DB_TXN);
3247c478bd9Sstevel@tonic-gate
3257c478bd9Sstevel@tonic-gate return (dbp);
3267c478bd9Sstevel@tonic-gate
3277c478bd9Sstevel@tonic-gate einval: errno = EINVAL;
3287c478bd9Sstevel@tonic-gate goto err;
3297c478bd9Sstevel@tonic-gate
3307c478bd9Sstevel@tonic-gate eftype: errno = EFTYPE;
3317c478bd9Sstevel@tonic-gate goto err;
3327c478bd9Sstevel@tonic-gate
3337c478bd9Sstevel@tonic-gate err: if (t) {
3347c478bd9Sstevel@tonic-gate if (t->bt_dbp)
3357c478bd9Sstevel@tonic-gate free(t->bt_dbp);
3367c478bd9Sstevel@tonic-gate if (t->bt_fd != -1)
3377c478bd9Sstevel@tonic-gate (void)close(t->bt_fd);
3387c478bd9Sstevel@tonic-gate free(t);
3397c478bd9Sstevel@tonic-gate }
3407c478bd9Sstevel@tonic-gate return (NULL);
3417c478bd9Sstevel@tonic-gate }
3427c478bd9Sstevel@tonic-gate
3437c478bd9Sstevel@tonic-gate /*
3447c478bd9Sstevel@tonic-gate * NROOT -- Create the root of a new tree.
3457c478bd9Sstevel@tonic-gate *
3467c478bd9Sstevel@tonic-gate * Parameters:
3477c478bd9Sstevel@tonic-gate * t: tree
3487c478bd9Sstevel@tonic-gate *
3497c478bd9Sstevel@tonic-gate * Returns:
3507c478bd9Sstevel@tonic-gate * RET_ERROR, RET_SUCCESS
3517c478bd9Sstevel@tonic-gate */
3527c478bd9Sstevel@tonic-gate static int
nroot(t)3537c478bd9Sstevel@tonic-gate nroot(t)
3547c478bd9Sstevel@tonic-gate BTREE *t;
3557c478bd9Sstevel@tonic-gate {
3567c478bd9Sstevel@tonic-gate PAGE *meta, *root;
3577c478bd9Sstevel@tonic-gate db_pgno_t npg;
3587c478bd9Sstevel@tonic-gate
3597c478bd9Sstevel@tonic-gate if ((root = mpool_get(t->bt_mp, 1, 0)) != NULL) {
3607c478bd9Sstevel@tonic-gate if (root->lower == 0 &&
3617c478bd9Sstevel@tonic-gate root->pgno == 0 &&
3627c478bd9Sstevel@tonic-gate root->linp[0] == 0) {
3637c478bd9Sstevel@tonic-gate mpool_delete(t->bt_mp, root);
3647c478bd9Sstevel@tonic-gate errno = EINVAL;
3657c478bd9Sstevel@tonic-gate } else {
3667c478bd9Sstevel@tonic-gate mpool_put(t->bt_mp, root, 0);
3677c478bd9Sstevel@tonic-gate return (RET_SUCCESS);
3687c478bd9Sstevel@tonic-gate }
3697c478bd9Sstevel@tonic-gate }
3707c478bd9Sstevel@tonic-gate if (errno != EINVAL) /* It's OK to not exist. */
3717c478bd9Sstevel@tonic-gate return (RET_ERROR);
3727c478bd9Sstevel@tonic-gate errno = 0;
3737c478bd9Sstevel@tonic-gate
3747c478bd9Sstevel@tonic-gate if ((meta = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
3757c478bd9Sstevel@tonic-gate return (RET_ERROR);
3767c478bd9Sstevel@tonic-gate
3777c478bd9Sstevel@tonic-gate if ((root = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
3787c478bd9Sstevel@tonic-gate return (RET_ERROR);
3797c478bd9Sstevel@tonic-gate
3807c478bd9Sstevel@tonic-gate if (npg != P_ROOT)
3817c478bd9Sstevel@tonic-gate return (RET_ERROR);
3827c478bd9Sstevel@tonic-gate root->pgno = npg;
3837c478bd9Sstevel@tonic-gate root->prevpg = root->nextpg = P_INVALID;
3847c478bd9Sstevel@tonic-gate root->lower = BTDATAOFF;
3857c478bd9Sstevel@tonic-gate root->upper = t->bt_psize;
3867c478bd9Sstevel@tonic-gate root->flags = P_BLEAF;
3877c478bd9Sstevel@tonic-gate memset(meta, 0, t->bt_psize);
3887c478bd9Sstevel@tonic-gate mpool_put(t->bt_mp, meta, MPOOL_DIRTY);
3897c478bd9Sstevel@tonic-gate mpool_put(t->bt_mp, root, MPOOL_DIRTY);
3907c478bd9Sstevel@tonic-gate return (RET_SUCCESS);
3917c478bd9Sstevel@tonic-gate }
3927c478bd9Sstevel@tonic-gate
3937c478bd9Sstevel@tonic-gate static int
tmp()3947c478bd9Sstevel@tonic-gate tmp()
3957c478bd9Sstevel@tonic-gate {
3967c478bd9Sstevel@tonic-gate #ifdef SIG_BLOCK
3977c478bd9Sstevel@tonic-gate sigset_t set, oset;
3987c478bd9Sstevel@tonic-gate #else
3997c478bd9Sstevel@tonic-gate int oset;
4007c478bd9Sstevel@tonic-gate #endif
4017c478bd9Sstevel@tonic-gate int fd;
4027c478bd9Sstevel@tonic-gate char *envtmp;
4037c478bd9Sstevel@tonic-gate char path[MAXPATHLEN];
4047c478bd9Sstevel@tonic-gate static char fn[] = "/bt.XXXXXX";
4057c478bd9Sstevel@tonic-gate
4067c478bd9Sstevel@tonic-gate envtmp = getenv("TMPDIR");
4077c478bd9Sstevel@tonic-gate
4087c478bd9Sstevel@tonic-gate /* this used to be done with snprintf(), but since snprintf
4097c478bd9Sstevel@tonic-gate isn't in most operating systems, and overflow checking in
4107c478bd9Sstevel@tonic-gate this case is easy, this is what is done */
4117c478bd9Sstevel@tonic-gate
4127c478bd9Sstevel@tonic-gate if (envtmp && ((strlen(envtmp)+sizeof(fn)+1) > sizeof(path)))
4137c478bd9Sstevel@tonic-gate return(-1);
4147c478bd9Sstevel@tonic-gate
4157c478bd9Sstevel@tonic-gate (void)sprintf(path, "%s%s", (envtmp ? envtmp : "/tmp"), fn);
4167c478bd9Sstevel@tonic-gate
4177c478bd9Sstevel@tonic-gate #ifdef SIG_BLOCK
4187c478bd9Sstevel@tonic-gate (void)sigfillset(&set);
4197c478bd9Sstevel@tonic-gate (void)sigprocmask(SIG_BLOCK, &set, &oset);
4207c478bd9Sstevel@tonic-gate #else
4217c478bd9Sstevel@tonic-gate oset = sigblock(~0);
4227c478bd9Sstevel@tonic-gate #endif
4237c478bd9Sstevel@tonic-gate if ((fd = mkstemp(path)) != -1)
4247c478bd9Sstevel@tonic-gate (void)unlink(path);
4257c478bd9Sstevel@tonic-gate #ifdef SIG_BLOCK
4267c478bd9Sstevel@tonic-gate (void)sigprocmask(SIG_SETMASK, &oset, NULL);
4277c478bd9Sstevel@tonic-gate #else
4287c478bd9Sstevel@tonic-gate sigsetmask(oset);
4297c478bd9Sstevel@tonic-gate #endif
4307c478bd9Sstevel@tonic-gate #ifdef __CYGWIN32__
4317c478bd9Sstevel@tonic-gate /* Ensure the fd is in binary mode. */
4327c478bd9Sstevel@tonic-gate setmode(fd, O_BINARY);
4337c478bd9Sstevel@tonic-gate #endif /* __CYGWIN32__ */
4347c478bd9Sstevel@tonic-gate
4357c478bd9Sstevel@tonic-gate return(fd);
4367c478bd9Sstevel@tonic-gate }
4377c478bd9Sstevel@tonic-gate
4387c478bd9Sstevel@tonic-gate static int
byteorder()4397c478bd9Sstevel@tonic-gate byteorder()
4407c478bd9Sstevel@tonic-gate {
4417c478bd9Sstevel@tonic-gate u_int32_t x;
4427c478bd9Sstevel@tonic-gate u_char *p;
4437c478bd9Sstevel@tonic-gate
4447c478bd9Sstevel@tonic-gate x = 0x01020304;
4457c478bd9Sstevel@tonic-gate p = (u_char *)&x;
4467c478bd9Sstevel@tonic-gate switch (*p) {
4477c478bd9Sstevel@tonic-gate case 1:
4487c478bd9Sstevel@tonic-gate return (DB_BIG_ENDIAN);
4497c478bd9Sstevel@tonic-gate case 4:
4507c478bd9Sstevel@tonic-gate return (DB_LITTLE_ENDIAN);
4517c478bd9Sstevel@tonic-gate default:
4527c478bd9Sstevel@tonic-gate return (0);
4537c478bd9Sstevel@tonic-gate }
4547c478bd9Sstevel@tonic-gate }
4557c478bd9Sstevel@tonic-gate
4567c478bd9Sstevel@tonic-gate int
__bt_fd(dbp)4577c478bd9Sstevel@tonic-gate __bt_fd(dbp)
4587c478bd9Sstevel@tonic-gate const DB *dbp;
4597c478bd9Sstevel@tonic-gate {
4607c478bd9Sstevel@tonic-gate BTREE *t;
4617c478bd9Sstevel@tonic-gate
4627c478bd9Sstevel@tonic-gate t = dbp->internal;
4637c478bd9Sstevel@tonic-gate
4647c478bd9Sstevel@tonic-gate /* Toss any page pinned across calls. */
4657c478bd9Sstevel@tonic-gate if (t->bt_pinned != NULL) {
4667c478bd9Sstevel@tonic-gate mpool_put(t->bt_mp, t->bt_pinned, 0);
4677c478bd9Sstevel@tonic-gate t->bt_pinned = NULL;
4687c478bd9Sstevel@tonic-gate }
4697c478bd9Sstevel@tonic-gate
4707c478bd9Sstevel@tonic-gate /* In-memory database can't have a file descriptor. */
4717c478bd9Sstevel@tonic-gate if (F_ISSET(t, B_INMEM)) {
4727c478bd9Sstevel@tonic-gate errno = ENOENT;
4737c478bd9Sstevel@tonic-gate return (-1);
4747c478bd9Sstevel@tonic-gate }
4757c478bd9Sstevel@tonic-gate return (t->bt_fd);
4767c478bd9Sstevel@tonic-gate }
477