2 * Copyright (c) 1990, 1993, 1994
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
37 #if defined(LIBC_SCCS) && !defined(lint)
38 static char sccsid[] = "@(#)bt_open.c 8.11 (Berkeley) 11/2/95";
39 #endif /* LIBC_SCCS and not lint */
42 * Implementation of btree access method for 4.4BSD.
44 * The design here was originally based on that of the btree access method
45 * used in the Postgres database system at UC Berkeley. This implementation
46 * is wholly independent of the Postgres code.
49 #include <sys/param.h>
61 #include "k5-platform.h" /* mkstemp? */
71 static int byteorder __P((void));
72 static int nroot __P((BTREE *));
73 static int tmp __P((void));
76 * __BT_OPEN -- Open a btree.
78 * Creates and fills a DB struct, and calls the routine that actually
82 * fname: filename (NULL for in-memory trees)
83 * flags: open flag bits
84 * mode: open permission bits
85 * b: BTREEINFO pointer
88 * NULL on failure, pointer to DB on success.
92 __bt_open(fname, flags, mode, openinfo, dflags)
94 int flags, mode, dflags;
95 const BTREEINFO *openinfo;
109 * Intention is to make sure all of the user's selections are okay
110 * here and then use them without checking. Can't be complete, since
111 * we don't know the right page size, lorder or flags until the backing
112 * file is opened. Also, the file's page size can cause the cachesize
115 machine_lorder = byteorder();
120 if (b.flags & ~(R_DUP))
124 * Page size must be indx_t aligned and >= MINPSIZE. Default
125 * page size is set farther on, based on the underlying file
129 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 ||
130 b.psize & (sizeof(indx_t) - 1)))
133 /* Minimum number of keys per page; absolute minimum is 2. */
135 if (b.minkeypage < 2)
138 b.minkeypage = DEFMINKEYPAGE;
140 /* If no comparison, use default comparison and prefix. */
141 if (b.compare == NULL) {
142 b.compare = __bt_defcmp;
143 if (b.prefix == NULL)
144 b.prefix = __bt_defpfx;
148 b.lorder = machine_lorder;
150 b.compare = __bt_defcmp;
153 b.lorder = machine_lorder;
154 b.minkeypage = DEFMINKEYPAGE;
155 b.prefix = __bt_defpfx;
159 /* Check for the ubiquitous PDP-11. */
160 if (b.lorder != DB_BIG_ENDIAN && b.lorder != DB_LITTLE_ENDIAN)
163 /* Allocate and initialize DB and BTREE structures. */
164 if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL)
166 memset(t, 0, sizeof(BTREE));
167 t->bt_fd = -1; /* Don't close unopened fd on error. */
168 t->bt_lorder = b.lorder;
170 t->bt_cmp = b.compare;
171 t->bt_pfx = b.prefix;
174 if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL)
176 memset(t->bt_dbp, 0, sizeof(DB));
177 if (t->bt_lorder != machine_lorder)
178 F_SET(t, B_NEEDSWAP);
180 dbp->type = DB_BTREE;
182 dbp->close = __bt_close;
183 dbp->del = __bt_delete;
188 dbp->sync = __bt_sync;
191 * If no file name was supplied, this is an in-memory btree and we
192 * open a backing temporary file. Otherwise, it's a disk-based tree.
195 switch (flags & O_ACCMODE) {
206 if ((t->bt_fd = open(fname, flags | O_BINARY, mode)) < 0)
210 if ((flags & O_ACCMODE) != O_RDWR)
212 if ((t->bt_fd = tmp()) == -1)
217 if (fcntl(t->bt_fd, F_SETFD, 1) == -1)
220 if (fstat(t->bt_fd, &sb))
223 if ((nr = read(t->bt_fd, &m, sizeof(BTMETA))) < 0)
225 if (nr != sizeof(BTMETA))
229 * Read in the meta-data. This can change the notion of what
230 * the lorder, page size and flags are, and, when the page size
231 * changes, the cachesize value can change too. If the user
232 * specified the wrong byte order for an existing database, we
233 * don't bother to return an error, we just clear the NEEDSWAP
236 if (m.magic == BTREEMAGIC)
237 F_CLR(t, B_NEEDSWAP);
239 F_SET(t, B_NEEDSWAP);
241 M_32_SWAP(m.version);
247 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION)
249 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 ||
250 m.psize & (sizeof(indx_t) - 1))
252 if (m.flags & ~SAVEMETA)
257 t->bt_nrecs = m.nrecs;
260 * Set the page size to the best value for I/O to this file.
261 * Don't overflow the page offset type.
264 b.psize = sb.st_blksize;
265 if (b.psize < MINPSIZE)
267 if (b.psize > MAX_PAGE_OFFSET + 1)
268 b.psize = MAX_PAGE_OFFSET + 1;
271 /* Set flag if duplicates permitted. */
272 if (!(b.flags & R_DUP))
275 t->bt_free = P_INVALID;
277 F_SET(t, B_METADIRTY);
280 t->bt_psize = b.psize;
282 /* Set the cache size; must be a multiple of the page size. */
283 if (b.cachesize && b.cachesize & (b.psize - 1))
284 b.cachesize += (~b.cachesize & (b.psize - 1)) + 1;
285 if (b.cachesize < b.psize * MINCACHE)
286 b.cachesize = b.psize * MINCACHE;
288 /* Calculate number of pages to cache. */
289 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize;
292 * The btree data structure requires that at least two keys can fit on
293 * a page, but other than that there's no fixed requirement. The user
294 * specified a minimum number per page, and we translated that into the
295 * number of bytes a key/data pair can use before being placed on an
296 * overflow page. This calculation includes the page header, the size
297 * of the index referencing the leaf item and the size of the leaf item
298 * structure. Also, don't let the user specify a minkeypage such that
299 * a key/data pair won't fit even if both key and data are on overflow
302 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage -
303 (sizeof(indx_t) + NBLEAFDBT(0, 0));
304 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t))
306 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t);
308 /* Initialize the buffer pool. */
310 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL)
312 if (!F_ISSET(t, B_INMEM))
313 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t);
315 /* Create a root page if new tree. */
316 if (nroot(t) == RET_ERROR)
320 if (dflags & DB_LOCK)
322 if (dflags & DB_SHMEM)
323 F_SET(t, B_DB_SHMEM);
329 einval: errno = EINVAL;
332 eftype: errno = EFTYPE;
339 (void)close(t->bt_fd);
346 * NROOT -- Create the root of a new tree.
352 * RET_ERROR, RET_SUCCESS
361 if ((root = mpool_get(t->bt_mp, 1, 0)) != NULL) {
362 if (root->lower == 0 &&
364 root->linp[0] == 0) {
365 mpool_delete(t->bt_mp, root);
368 mpool_put(t->bt_mp, root, 0);
369 return (RET_SUCCESS);
372 if (errno != EINVAL) /* It's OK to not exist. */
376 if ((meta = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
379 if ((root = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
385 root->prevpg = root->nextpg = P_INVALID;
386 root->lower = BTDATAOFF;
387 root->upper = t->bt_psize;
388 root->flags = P_BLEAF;
389 memset(meta, 0, t->bt_psize);
390 mpool_put(t->bt_mp, meta, MPOOL_DIRTY);
391 mpool_put(t->bt_mp, root, MPOOL_DIRTY);
392 return (RET_SUCCESS);
405 char path[MAXPATHLEN];
406 static char fn[] = "/bt.XXXXXX";
408 envtmp = getenv("TMPDIR");
410 /* this used to be done with snprintf(), but since snprintf
411 isn't in most operating systems, and overflow checking in
412 this case is easy, this is what is done */
414 if (envtmp && ((strlen(envtmp)+sizeof(fn)+1) > sizeof(path)))
417 (void)snprintf(path, sizeof(path),
418 "%s%s", (envtmp ? envtmp : "/tmp"), fn);
421 (void)sigfillset(&set);
422 (void)sigprocmask(SIG_BLOCK, &set, &oset);
426 if ((fd = mkstemp(path)) != -1)
430 (void)sigprocmask(SIG_SETMASK, &oset, NULL);
435 /* Ensure the fd is in binary mode. */
436 setmode(fd, O_BINARY);
437 #endif /* __CYGWIN32__ */
452 return (DB_BIG_ENDIAN);
454 return (DB_LITTLE_ENDIAN);
468 /* Toss any page pinned across calls. */
469 if (t->bt_pinned != NULL) {
470 mpool_put(t->bt_mp, t->bt_pinned, 0);
474 /* In-memory database can't have a file descriptor. */
475 if (F_ISSET(t, B_INMEM)) {