xref: /netbsd-src/lib/libc/db/btree/bt_open.c (revision 8b0f9554ff8762542c4defc4f70e1eb76fb508fa)
1 /*	$NetBSD: bt_open.c,v 1.21 2007/02/03 23:46:09 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 1990, 1993, 1994
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Mike Olson.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 #include <sys/cdefs.h>
36 #if defined(LIBC_SCCS) && !defined(lint)
37 #if 0
38 static char sccsid[] = "@(#)bt_open.c	8.10 (Berkeley) 8/17/94";
39 #else
40 __RCSID("$NetBSD: bt_open.c,v 1.21 2007/02/03 23:46:09 christos Exp $");
41 #endif
42 #endif /* LIBC_SCCS and not lint */
43 
44 /*
45  * Implementation of btree access method for 4.4BSD.
46  *
47  * The design here was originally based on that of the btree access method
48  * used in the Postgres database system at UC Berkeley.  This implementation
49  * is wholly independent of the Postgres code.
50  */
51 
52 #include "namespace.h"
53 #include <sys/stat.h>
54 
55 #include <assert.h>
56 #include <errno.h>
57 #include <fcntl.h>
58 #include <limits.h>
59 #include <signal.h>
60 #include <stdio.h>
61 #include <stdlib.h>
62 #include <string.h>
63 #include <unistd.h>
64 #include <paths.h>
65 
66 #include <db.h>
67 #include "btree.h"
68 
69 #ifdef DEBUG
70 #undef	MINPSIZE
71 #define	MINPSIZE	128
72 #endif
73 
74 static int byteorder(void);
75 static int nroot(BTREE *);
76 static int tmp(void);
77 
78 /*
79  * __BT_OPEN -- Open a btree.
80  *
81  * Creates and fills a DB struct, and calls the routine that actually
82  * opens the btree.
83  *
84  * Parameters:
85  *	fname:	filename (NULL for in-memory trees)
86  *	flags:	open flag bits
87  *	mode:	open permission bits
88  *	b:	BTREEINFO pointer
89  *
90  * Returns:
91  *	NULL on failure, pointer to DB on success.
92  *
93  */
94 DB *
95 __bt_open(const char *fname, int flags, mode_t mode, const BTREEINFO *openinfo,
96     int dflags)
97 {
98 	struct stat sb;
99 	BTMETA m;
100 	BTREE *t;
101 	BTREEINFO b;
102 	DB *dbp;
103 	pgno_t ncache;
104 	ssize_t nr;
105 	size_t temp;
106 	int machine_lorder;
107 
108 	t = NULL;
109 
110 	/*
111 	 * Intention is to make sure all of the user's selections are okay
112 	 * here and then use them without checking.  Can't be complete, since
113 	 * we don't know the right page size, lorder or flags until the backing
114 	 * file is opened.  Also, the file's page size can cause the cachesize
115 	 * to change.
116 	 */
117 	machine_lorder = byteorder();
118 	if (openinfo) {
119 		b = *openinfo;
120 
121 		/* Flags: R_DUP. */
122 		if (b.flags & ~(R_DUP))
123 			goto einval;
124 
125 		/*
126 		 * Page size must be indx_t aligned and >= MINPSIZE.  Default
127 		 * page size is set farther on, based on the underlying file
128 		 * transfer size.
129 		 */
130 		if (b.psize &&
131 		    (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 ||
132 		    b.psize & (sizeof(indx_t) - 1)))
133 			goto einval;
134 
135 		/* Minimum number of keys per page; absolute minimum is 2. */
136 		if (b.minkeypage) {
137 			if (b.minkeypage < 2)
138 				goto einval;
139 		} else
140 			b.minkeypage = DEFMINKEYPAGE;
141 
142 		/* If no comparison, use default comparison and prefix. */
143 		if (b.compare == NULL) {
144 			b.compare = __bt_defcmp;
145 			if (b.prefix == NULL)
146 				b.prefix = __bt_defpfx;
147 		}
148 
149 		if (b.lorder == 0)
150 			b.lorder = machine_lorder;
151 	} else {
152 		b.compare = __bt_defcmp;
153 		b.cachesize = 0;
154 		b.flags = 0;
155 		b.lorder = machine_lorder;
156 		b.minkeypage = DEFMINKEYPAGE;
157 		b.prefix = __bt_defpfx;
158 		b.psize = 0;
159 	}
160 
161 	/* Check for the ubiquitous PDP-11. */
162 	if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN)
163 		goto einval;
164 
165 	/* Allocate and initialize DB and BTREE structures. */
166 	if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL)
167 		goto err;
168 	memset(t, 0, sizeof(BTREE));
169 	t->bt_fd = -1;			/* Don't close unopened fd on error. */
170 	t->bt_lorder = b.lorder;
171 	t->bt_order = NOT;
172 	t->bt_cmp = b.compare;
173 	t->bt_pfx = b.prefix;
174 	t->bt_rfd = -1;
175 
176 	if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL)
177 		goto err;
178 	memset(t->bt_dbp, 0, sizeof(DB));
179 	if (t->bt_lorder != machine_lorder)
180 		F_SET(t, B_NEEDSWAP);
181 
182 	dbp->type = DB_BTREE;
183 	dbp->internal = t;
184 	dbp->close = __bt_close;
185 	dbp->del = __bt_delete;
186 	dbp->fd = __bt_fd;
187 	dbp->get = __bt_get;
188 	dbp->put = __bt_put;
189 	dbp->seq = __bt_seq;
190 	dbp->sync = __bt_sync;
191 
192 	/*
193 	 * If no file name was supplied, this is an in-memory btree and we
194 	 * open a backing temporary file.  Otherwise, it's a disk-based tree.
195 	 */
196 	if (fname) {
197 		switch (flags & O_ACCMODE) {
198 		case O_RDONLY:
199 			F_SET(t, B_RDONLY);
200 			break;
201 		case O_RDWR:
202 			break;
203 		case O_WRONLY:
204 		default:
205 			goto einval;
206 		}
207 
208 		if ((t->bt_fd = open(fname, flags, mode)) == -1)
209 			goto err;
210 		if (fcntl(t->bt_fd, F_SETFD, FD_CLOEXEC) == -1)
211 			goto err;
212 	} else {
213 		if ((flags & O_ACCMODE) != O_RDWR)
214 			goto einval;
215 		if ((t->bt_fd = tmp()) == -1)
216 			goto err;
217 		F_SET(t, B_INMEM);
218 	}
219 
220 	if (fcntl(t->bt_fd, F_SETFD, FD_CLOEXEC) == -1)
221 		goto err;
222 
223 	if (fstat(t->bt_fd, &sb))
224 		goto err;
225 	if (sb.st_size) {
226 		if ((nr = read(t->bt_fd, &m, sizeof(BTMETA))) < 0)
227 			goto err;
228 		if (nr != sizeof(BTMETA))
229 			goto eftype;
230 
231 		/*
232 		 * Read in the meta-data.  This can change the notion of what
233 		 * the lorder, page size and flags are, and, when the page size
234 		 * changes, the cachesize value can change too.  If the user
235 		 * specified the wrong byte order for an existing database, we
236 		 * don't bother to return an error, we just clear the NEEDSWAP
237 		 * bit.
238 		 */
239 		if (m.magic == BTREEMAGIC)
240 			F_CLR(t, B_NEEDSWAP);
241 		else {
242 			F_SET(t, B_NEEDSWAP);
243 			M_32_SWAP(m.magic);
244 			M_32_SWAP(m.version);
245 			M_32_SWAP(m.psize);
246 			M_32_SWAP(m.free);
247 			M_32_SWAP(m.nrecs);
248 			M_32_SWAP(m.flags);
249 		}
250 		if (m.magic != BTREEMAGIC || m.version != BTREEVERSION)
251 			goto eftype;
252 		if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 ||
253 		    m.psize & (sizeof(indx_t) - 1))
254 			goto eftype;
255 		if (m.flags & ~SAVEMETA)
256 			goto eftype;
257 		b.psize = m.psize;
258 		F_SET(t, m.flags);
259 		t->bt_free = m.free;
260 		t->bt_nrecs = m.nrecs;
261 	} else {
262 		/*
263 		 * Set the page size to the best value for I/O to this file.
264 		 * Don't overflow the page offset type.
265 		 */
266 		if (b.psize == 0) {
267 			b.psize = sb.st_blksize;
268 			if (b.psize < MINPSIZE)
269 				b.psize = MINPSIZE;
270 			if (b.psize > MAX_PAGE_OFFSET + 1)
271 				b.psize = MAX_PAGE_OFFSET + 1;
272 		}
273 
274 		/* Set flag if duplicates permitted. */
275 		if (!(b.flags & R_DUP))
276 			F_SET(t, B_NODUPS);
277 
278 		t->bt_free = P_INVALID;
279 		t->bt_nrecs = 0;
280 		F_SET(t, B_METADIRTY);
281 	}
282 
283 	t->bt_psize = b.psize;
284 
285 	/* Set the cache size; must be a multiple of the page size. */
286 	if (b.cachesize && b.cachesize & (b.psize - 1))
287 		b.cachesize += (~b.cachesize & (b.psize - 1)) + 1;
288 	if (b.cachesize < b.psize * MINCACHE)
289 		b.cachesize = b.psize * MINCACHE;
290 
291 	/* Calculate number of pages to cache. */
292 	ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize;
293 
294 	/*
295 	 * The btree data structure requires that at least two keys can fit on
296 	 * a page, but other than that there's no fixed requirement.  The user
297 	 * specified a minimum number per page, and we translated that into the
298 	 * number of bytes a key/data pair can use before being placed on an
299 	 * overflow page.  This calculation includes the page header, the size
300 	 * of the index referencing the leaf item and the size of the leaf item
301 	 * structure.  Also, don't let the user specify a minkeypage such that
302 	 * a key/data pair won't fit even if both key and data are on overflow
303 	 * pages.
304 	 */
305 	temp = (t->bt_psize - BTDATAOFF) / b.minkeypage -
306 	    (sizeof(indx_t) + NBLEAFDBT(0, 0));
307 	_DBFIT(temp, indx_t);
308 	t->bt_ovflsize = (indx_t)temp;
309 	if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t))
310 		t->bt_ovflsize =
311 		    NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t);
312 
313 	/* Initialize the buffer pool. */
314 	if ((t->bt_mp =
315 	    mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL)
316 		goto err;
317 	if (!F_ISSET(t, B_INMEM))
318 		mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t);
319 
320 	/* Create a root page if new tree. */
321 	if (nroot(t) == RET_ERROR)
322 		goto err;
323 
324 	/* Global flags. */
325 	if (dflags & DB_LOCK)
326 		F_SET(t, B_DB_LOCK);
327 	if (dflags & DB_SHMEM)
328 		F_SET(t, B_DB_SHMEM);
329 	if (dflags & DB_TXN)
330 		F_SET(t, B_DB_TXN);
331 
332 	return (dbp);
333 
334 einval:	errno = EINVAL;
335 	goto err;
336 
337 eftype:	errno = EFTYPE;
338 	goto err;
339 
340 err:	if (t) {
341 		if (t->bt_dbp)
342 			free(t->bt_dbp);
343 		if (t->bt_fd != -1)
344 			(void)close(t->bt_fd);
345 		free(t);
346 	}
347 	return (NULL);
348 }
349 
350 /*
351  * NROOT -- Create the root of a new tree.
352  *
353  * Parameters:
354  *	t:	tree
355  *
356  * Returns:
357  *	RET_ERROR, RET_SUCCESS
358  */
359 static int
360 nroot(BTREE *t)
361 {
362 	PAGE *meta, *root;
363 	pgno_t npg;
364 
365 	if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) {
366 		mpool_put(t->bt_mp, meta, 0);
367 		return (RET_SUCCESS);
368 	}
369 	if (errno != EINVAL)		/* It's OK to not exist. */
370 		return (RET_ERROR);
371 	errno = 0;
372 
373 	if ((meta = mpool_new(t->bt_mp, &npg)) == NULL)
374 		return (RET_ERROR);
375 
376 	if ((root = mpool_new(t->bt_mp, &npg)) == NULL)
377 		return (RET_ERROR);
378 
379 	if (npg != P_ROOT)
380 		return (RET_ERROR);
381 	root->pgno = npg;
382 	root->prevpg = root->nextpg = P_INVALID;
383 	root->lower = BTDATAOFF;
384 	root->upper = t->bt_psize;
385 	root->flags = P_BLEAF;
386 	memset(meta, 0, t->bt_psize);
387 	mpool_put(t->bt_mp, meta, MPOOL_DIRTY);
388 	mpool_put(t->bt_mp, root, MPOOL_DIRTY);
389 	return (RET_SUCCESS);
390 }
391 
392 static int
393 tmp(void)
394 {
395 	sigset_t set, oset;
396 	size_t len;
397 	int fd;
398 	char *envtmp;
399 	char path[PATH_MAX];
400 
401 	if (issetugid())
402 		envtmp = NULL;
403 	else
404 		envtmp = getenv("TMPDIR");
405 
406 	len = snprintf(path,
407 	    sizeof(path), "%s/bt.XXXXXX", envtmp ? envtmp : _PATH_TMP);
408 	if (len >= sizeof(path))
409 		return -1;
410 
411 	(void)sigfillset(&set);
412 	(void)sigprocmask(SIG_BLOCK, &set, &oset);
413 	if ((fd = mkstemp(path)) != -1) {
414 		(void)unlink(path);
415 		(void)fcntl(fd, F_SETFD, FD_CLOEXEC);
416 	}
417 	(void)sigprocmask(SIG_SETMASK, &oset, NULL);
418 	return(fd);
419 }
420 
421 static int
422 byteorder(void)
423 {
424 	u_int32_t x;
425 	u_char *p;
426 
427 	x = 0x01020304;
428 	p = (u_char *)(void *)&x;
429 	switch (*p) {
430 	case 1:
431 		return (BIG_ENDIAN);
432 	case 4:
433 		return (LITTLE_ENDIAN);
434 	default:
435 		return (0);
436 	}
437 }
438 
439 int
440 __bt_fd(const DB *dbp)
441 {
442 	BTREE *t;
443 
444 	t = dbp->internal;
445 
446 	/* Toss any page pinned across calls. */
447 	if (t->bt_pinned != NULL) {
448 		mpool_put(t->bt_mp, t->bt_pinned, 0);
449 		t->bt_pinned = NULL;
450 	}
451 
452 	/* In-memory database can't have a file descriptor. */
453 	if (F_ISSET(t, B_INMEM)) {
454 		errno = ENOENT;
455 		return (-1);
456 	}
457 	return (t->bt_fd);
458 }
459