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 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 #if defined(LIBC_SCCS) && !defined(lint)
34 static char sccsid[] = "@(#)bt_open.c 8.10 (Berkeley) 8/17/94";
35 #endif /* LIBC_SCCS and not lint */
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
40 * Implementation of btree access method for 4.4BSD.
42 * The design here was originally based on that of the btree access method
43 * used in the Postgres database system at UC Berkeley. This implementation
44 * is wholly independent of the Postgres code.
47 #include "namespace.h"
48 #include <sys/param.h>
59 #include "un-namespace.h"
69 static int byteorder(void);
70 static int nroot(BTREE *);
74 * __BT_OPEN -- Open a btree.
76 * Creates and fills a DB struct, and calls the routine that actually
80 * fname: filename (NULL for in-memory trees)
81 * flags: open flag bits
82 * mode: open permission bits
83 * b: BTREEINFO pointer
86 * NULL on failure, pointer to DB on success.
90 __bt_open(const char *fname, int flags, int mode, const BTREEINFO *openinfo, int dflags)
99 int machine_lorder, saved_errno;
104 * Intention is to make sure all of the user's selections are okay
105 * here and then use them without checking. Can't be complete, since
106 * we don't know the right page size, lorder or flags until the backing
107 * file is opened. Also, the file's page size can cause the cachesize
110 machine_lorder = byteorder();
115 if (b.flags & ~(R_DUP))
119 * Page size must be indx_t aligned and >= MINPSIZE. Default
120 * page size is set farther on, based on the underlying file
124 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 ||
125 b.psize & (sizeof(indx_t) - 1) ))
128 /* Minimum number of keys per page; absolute minimum is 2. */
130 if (b.minkeypage < 2)
133 b.minkeypage = DEFMINKEYPAGE;
135 /* If no comparison, use default comparison and prefix. */
136 if (b.compare == NULL) {
137 b.compare = __bt_defcmp;
138 if (b.prefix == NULL)
139 b.prefix = __bt_defpfx;
143 b.lorder = machine_lorder;
145 b.compare = __bt_defcmp;
148 b.lorder = machine_lorder;
149 b.minkeypage = DEFMINKEYPAGE;
150 b.prefix = __bt_defpfx;
154 /* Check for the ubiquitous PDP-11. */
155 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN)
158 /* Allocate and initialize DB and BTREE structures. */
159 if ((t = (BTREE *)calloc(1, sizeof(BTREE))) == NULL)
161 t->bt_fd = -1; /* Don't close unopened fd on error. */
162 t->bt_lorder = b.lorder;
164 t->bt_cmp = b.compare;
165 t->bt_pfx = b.prefix;
168 if ((t->bt_dbp = dbp = (DB *)calloc(1, sizeof(DB))) == NULL)
170 if (t->bt_lorder != machine_lorder)
171 F_SET(t, B_NEEDSWAP);
173 dbp->type = DB_BTREE;
175 dbp->close = __bt_close;
176 dbp->del = __bt_delete;
181 dbp->sync = __bt_sync;
184 * If no file name was supplied, this is an in-memory btree and we
185 * open a backing temporary file. Otherwise, it's a disk-based tree.
188 switch (flags & O_ACCMODE) {
199 if ((t->bt_fd = _open(fname, flags | O_CLOEXEC, mode)) < 0)
203 if ((flags & O_ACCMODE) != O_RDWR)
205 if ((t->bt_fd = tmp()) == -1)
210 if (_fstat(t->bt_fd, &sb))
213 if ((nr = _read(t->bt_fd, &m, sizeof(BTMETA))) < 0)
215 if (nr != sizeof(BTMETA))
219 * Read in the meta-data. This can change the notion of what
220 * the lorder, page size and flags are, and, when the page size
221 * changes, the cachesize value can change too. If the user
222 * specified the wrong byte order for an existing database, we
223 * don't bother to return an error, we just clear the NEEDSWAP
226 if (m.magic == BTREEMAGIC)
227 F_CLR(t, B_NEEDSWAP);
229 F_SET(t, B_NEEDSWAP);
231 M_32_SWAP(m.version);
237 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION)
239 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 ||
240 m.psize & (sizeof(indx_t) - 1) )
242 if (m.flags & ~SAVEMETA)
247 t->bt_nrecs = m.nrecs;
250 * Set the page size to the best value for I/O to this file.
251 * Don't overflow the page offset type.
254 b.psize = sb.st_blksize;
255 if (b.psize < MINPSIZE)
257 if (b.psize > MAX_PAGE_OFFSET + 1)
258 b.psize = MAX_PAGE_OFFSET + 1;
261 /* Set flag if duplicates permitted. */
262 if (!(b.flags & R_DUP))
265 t->bt_free = P_INVALID;
267 F_SET(t, B_METADIRTY);
270 t->bt_psize = b.psize;
272 /* Set the cache size; must be a multiple of the page size. */
273 if (b.cachesize && b.cachesize & (b.psize - 1) )
274 b.cachesize += (~b.cachesize & (b.psize - 1) ) + 1;
275 if (b.cachesize < b.psize * MINCACHE)
276 b.cachesize = b.psize * MINCACHE;
278 /* Calculate number of pages to cache. */
279 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize;
282 * The btree data structure requires that at least two keys can fit on
283 * a page, but other than that there's no fixed requirement. The user
284 * specified a minimum number per page, and we translated that into the
285 * number of bytes a key/data pair can use before being placed on an
286 * overflow page. This calculation includes the page header, the size
287 * of the index referencing the leaf item and the size of the leaf item
288 * structure. Also, don't let the user specify a minkeypage such that
289 * a key/data pair won't fit even if both key and data are on overflow
292 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage -
293 (sizeof(indx_t) + NBLEAFDBT(0, 0));
294 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t))
296 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t);
298 /* Initialize the buffer pool. */
300 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL)
302 if (!F_ISSET(t, B_INMEM))
303 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t);
305 /* Create a root page if new tree. */
306 if (nroot(t) == RET_ERROR)
310 if (dflags & DB_LOCK)
312 if (dflags & DB_SHMEM)
313 F_SET(t, B_DB_SHMEM);
319 einval: errno = EINVAL;
322 eftype: errno = EFTYPE;
325 err: saved_errno = errno;
330 (void)_close(t->bt_fd);
338 * NROOT -- Create the root of a new tree.
344 * RET_ERROR, RET_SUCCESS
352 if ((root = mpool_get(t->bt_mp, 1, 0)) != NULL) {
353 if (root->lower == 0 &&
355 root->linp[0] == 0) {
356 mpool_delete(t->bt_mp, root);
359 mpool_put(t->bt_mp, root, 0);
360 return (RET_SUCCESS);
363 if (errno != EINVAL) /* It's OK to not exist. */
367 if ((meta = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
370 if ((root = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
376 root->prevpg = root->nextpg = P_INVALID;
377 root->lower = BTDATAOFF;
378 root->upper = t->bt_psize;
379 root->flags = P_BLEAF;
380 memset(meta, 0, t->bt_psize);
381 mpool_put(t->bt_mp, meta, MPOOL_DIRTY);
382 mpool_put(t->bt_mp, root, MPOOL_DIRTY);
383 return (RET_SUCCESS);
392 char path[MAXPATHLEN];
394 if (issetugid() == 0)
395 envtmp = getenv("TMPDIR");
397 sizeof(path), "%s/bt.XXXXXXXXXX", envtmp ? envtmp : "/tmp");
398 if (len < 0 || len >= (int)sizeof(path)) {
399 errno = ENAMETOOLONG;
403 (void)sigfillset(&set);
404 (void)_sigprocmask(SIG_BLOCK, &set, &oset);
405 if ((fd = mkostemp(path, O_CLOEXEC)) != -1)
407 (void)_sigprocmask(SIG_SETMASK, &oset, NULL);
423 return (LITTLE_ENDIAN);
430 __bt_fd(const DB *dbp)
436 /* Toss any page pinned across calls. */
437 if (t->bt_pinned != NULL) {
438 mpool_put(t->bt_mp, t->bt_pinned, 0);
442 /* In-memory database can't have a file descriptor. */
443 if (F_ISSET(t, B_INMEM)) {