xref: /freebsd/lib/libc/db/btree/bt_open.c (revision f677a9e2672665f4eb3dd4111c07ee8f1f954262)
1 /*-
2  * Copyright (c) 1990, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Mike Olson.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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.
19  *
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
30  * SUCH DAMAGE.
31  */
32 
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("$FreeBSD$");
38 
39 /*
40  * Implementation of btree access method for 4.4BSD.
41  *
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.
45  */
46 
47 #include "namespace.h"
48 #include <sys/param.h>
49 #include <sys/stat.h>
50 
51 #include <errno.h>
52 #include <fcntl.h>
53 #include <limits.h>
54 #include <signal.h>
55 #include <stdio.h>
56 #include <stdlib.h>
57 #include <string.h>
58 #include <unistd.h>
59 #include "un-namespace.h"
60 
61 #include <db.h>
62 #include "btree.h"
63 
64 #ifdef DEBUG
65 #undef	MINPSIZE
66 #define	MINPSIZE	128
67 #endif
68 
69 static int byteorder(void);
70 static int nroot(BTREE *);
71 static int tmp(void);
72 
73 /*
74  * __BT_OPEN -- Open a btree.
75  *
76  * Creates and fills a DB struct, and calls the routine that actually
77  * opens the btree.
78  *
79  * Parameters:
80  *	fname:	filename (NULL for in-memory trees)
81  *	flags:	open flag bits
82  *	mode:	open permission bits
83  *	b:	BTREEINFO pointer
84  *
85  * Returns:
86  *	NULL on failure, pointer to DB on success.
87  *
88  */
89 DB *
90 __bt_open(const char *fname, int flags, int mode, const BTREEINFO *openinfo, int dflags)
91 {
92 	struct stat sb;
93 	BTMETA m;
94 	BTREE *t;
95 	BTREEINFO b;
96 	DB *dbp;
97 	pgno_t ncache;
98 	ssize_t nr;
99 	int machine_lorder, saved_errno;
100 
101 	t = NULL;
102 
103 	/*
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
108 	 * to change.
109 	 */
110 	machine_lorder = byteorder();
111 	if (openinfo) {
112 		b = *openinfo;
113 
114 		/* Flags: R_DUP. */
115 		if (b.flags & ~(R_DUP))
116 			goto einval;
117 
118 		/*
119 		 * Page size must be indx_t aligned and >= MINPSIZE.  Default
120 		 * page size is set farther on, based on the underlying file
121 		 * transfer size.
122 		 */
123 		if (b.psize &&
124 		    (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 ||
125 		    b.psize & (sizeof(indx_t) - 1) ))
126 			goto einval;
127 
128 		/* Minimum number of keys per page; absolute minimum is 2. */
129 		if (b.minkeypage) {
130 			if (b.minkeypage < 2)
131 				goto einval;
132 		} else
133 			b.minkeypage = DEFMINKEYPAGE;
134 
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;
140 		}
141 
142 		if (b.lorder == 0)
143 			b.lorder = machine_lorder;
144 	} else {
145 		b.compare = __bt_defcmp;
146 		b.cachesize = 0;
147 		b.flags = 0;
148 		b.lorder = machine_lorder;
149 		b.minkeypage = DEFMINKEYPAGE;
150 		b.prefix = __bt_defpfx;
151 		b.psize = 0;
152 	}
153 
154 	/* Check for the ubiquitous PDP-11. */
155 	if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN)
156 		goto einval;
157 
158 	/* Allocate and initialize DB and BTREE structures. */
159 	if ((t = (BTREE *)calloc(1, sizeof(BTREE))) == NULL)
160 		goto err;
161 	t->bt_fd = -1;			/* Don't close unopened fd on error. */
162 	t->bt_lorder = b.lorder;
163 	t->bt_order = NOT;
164 	t->bt_cmp = b.compare;
165 	t->bt_pfx = b.prefix;
166 	t->bt_rfd = -1;
167 
168 	if ((t->bt_dbp = dbp = (DB *)calloc(1, sizeof(DB))) == NULL)
169 		goto err;
170 	if (t->bt_lorder != machine_lorder)
171 		F_SET(t, B_NEEDSWAP);
172 
173 	dbp->type = DB_BTREE;
174 	dbp->internal = t;
175 	dbp->close = __bt_close;
176 	dbp->del = __bt_delete;
177 	dbp->fd = __bt_fd;
178 	dbp->get = __bt_get;
179 	dbp->put = __bt_put;
180 	dbp->seq = __bt_seq;
181 	dbp->sync = __bt_sync;
182 
183 	/*
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.
186 	 */
187 	if (fname) {
188 		switch (flags & O_ACCMODE) {
189 		case O_RDONLY:
190 			F_SET(t, B_RDONLY);
191 			break;
192 		case O_RDWR:
193 			break;
194 		case O_WRONLY:
195 		default:
196 			goto einval;
197 		}
198 
199 		if ((t->bt_fd = _open(fname, flags | O_CLOEXEC, mode)) < 0)
200 			goto err;
201 
202 	} else {
203 		if ((flags & O_ACCMODE) != O_RDWR)
204 			goto einval;
205 		if ((t->bt_fd = tmp()) == -1)
206 			goto err;
207 		F_SET(t, B_INMEM);
208 	}
209 
210 	if (_fstat(t->bt_fd, &sb))
211 		goto err;
212 	if (sb.st_size) {
213 		if ((nr = _read(t->bt_fd, &m, sizeof(BTMETA))) < 0)
214 			goto err;
215 		if (nr != sizeof(BTMETA))
216 			goto eftype;
217 
218 		/*
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
224 		 * bit.
225 		 */
226 		if (m.magic == BTREEMAGIC)
227 			F_CLR(t, B_NEEDSWAP);
228 		else {
229 			F_SET(t, B_NEEDSWAP);
230 			M_32_SWAP(m.magic);
231 			M_32_SWAP(m.version);
232 			M_32_SWAP(m.psize);
233 			M_32_SWAP(m.free);
234 			M_32_SWAP(m.nrecs);
235 			M_32_SWAP(m.flags);
236 		}
237 		if (m.magic != BTREEMAGIC || m.version != BTREEVERSION)
238 			goto eftype;
239 		if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 ||
240 		    m.psize & (sizeof(indx_t) - 1) )
241 			goto eftype;
242 		if (m.flags & ~SAVEMETA)
243 			goto eftype;
244 		b.psize = m.psize;
245 		F_SET(t, m.flags);
246 		t->bt_free = m.free;
247 		t->bt_nrecs = m.nrecs;
248 	} else {
249 		/*
250 		 * Set the page size to the best value for I/O to this file.
251 		 * Don't overflow the page offset type.
252 		 */
253 		if (b.psize == 0) {
254 			b.psize = sb.st_blksize;
255 			if (b.psize < MINPSIZE)
256 				b.psize = MINPSIZE;
257 			if (b.psize > MAX_PAGE_OFFSET + 1)
258 				b.psize = MAX_PAGE_OFFSET + 1;
259 		}
260 
261 		/* Set flag if duplicates permitted. */
262 		if (!(b.flags & R_DUP))
263 			F_SET(t, B_NODUPS);
264 
265 		t->bt_free = P_INVALID;
266 		t->bt_nrecs = 0;
267 		F_SET(t, B_METADIRTY);
268 	}
269 
270 	t->bt_psize = b.psize;
271 
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;
277 
278 	/* Calculate number of pages to cache. */
279 	ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize;
280 
281 	/*
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
290 	 * pages.
291 	 */
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))
295 		t->bt_ovflsize =
296 		    NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t);
297 
298 	/* Initialize the buffer pool. */
299 	if ((t->bt_mp =
300 	    mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL)
301 		goto err;
302 	if (!F_ISSET(t, B_INMEM))
303 		mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t);
304 
305 	/* Create a root page if new tree. */
306 	if (nroot(t) == RET_ERROR)
307 		goto err;
308 
309 	/* Global flags. */
310 	if (dflags & DB_LOCK)
311 		F_SET(t, B_DB_LOCK);
312 	if (dflags & DB_SHMEM)
313 		F_SET(t, B_DB_SHMEM);
314 	if (dflags & DB_TXN)
315 		F_SET(t, B_DB_TXN);
316 
317 	return (dbp);
318 
319 einval:	errno = EINVAL;
320 	goto err;
321 
322 eftype:	errno = EFTYPE;
323 	goto err;
324 
325 err:	saved_errno = errno;
326 	if (t) {
327 		if (t->bt_dbp)
328 			free(t->bt_dbp);
329 		if (t->bt_fd != -1)
330 			(void)_close(t->bt_fd);
331 		free(t);
332 	}
333 	errno = saved_errno;
334 	return (NULL);
335 }
336 
337 /*
338  * NROOT -- Create the root of a new tree.
339  *
340  * Parameters:
341  *	t:	tree
342  *
343  * Returns:
344  *	RET_ERROR, RET_SUCCESS
345  */
346 static int
347 nroot(BTREE *t)
348 {
349 	PAGE *meta, *root;
350 	pgno_t npg;
351 
352 	if ((root = mpool_get(t->bt_mp, 1, 0)) != NULL) {
353 		if (root->lower == 0 &&
354 		    root->pgno == 0 &&
355 		    root->linp[0] == 0) {
356 			mpool_delete(t->bt_mp, root);
357 			errno = EINVAL;
358 		} else {
359 			mpool_put(t->bt_mp, root, 0);
360 			return (RET_SUCCESS);
361 		}
362 	}
363 	if (errno != EINVAL)		/* It's OK to not exist. */
364 		return (RET_ERROR);
365 	errno = 0;
366 
367 	if ((meta = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
368 		return (RET_ERROR);
369 
370 	if ((root = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
371 		return (RET_ERROR);
372 
373 	if (npg != P_ROOT)
374 		return (RET_ERROR);
375 	root->pgno = npg;
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);
384 }
385 
386 static int
387 tmp(void)
388 {
389 	sigset_t set, oset;
390 	int fd, len;
391 	char *envtmp = NULL;
392 	char path[MAXPATHLEN];
393 
394 	if (issetugid() == 0)
395 		envtmp = getenv("TMPDIR");
396 	len = snprintf(path,
397 	    sizeof(path), "%s/bt.XXXXXXXXXX", envtmp ? envtmp : "/tmp");
398 	if (len < 0 || len >= (int)sizeof(path)) {
399 		errno = ENAMETOOLONG;
400 		return(-1);
401 	}
402 
403 	(void)sigfillset(&set);
404 	(void)_sigprocmask(SIG_BLOCK, &set, &oset);
405 	if ((fd = mkostemp(path, O_CLOEXEC)) != -1)
406 		(void)unlink(path);
407 	(void)_sigprocmask(SIG_SETMASK, &oset, NULL);
408 	return(fd);
409 }
410 
411 static int
412 byteorder(void)
413 {
414 	u_int32_t x;
415 	u_char *p;
416 
417 	x = 0x01020304;
418 	p = (u_char *)&x;
419 	switch (*p) {
420 	case 1:
421 		return (BIG_ENDIAN);
422 	case 4:
423 		return (LITTLE_ENDIAN);
424 	default:
425 		return (0);
426 	}
427 }
428 
429 int
430 __bt_fd(const DB *dbp)
431 {
432 	BTREE *t;
433 
434 	t = dbp->internal;
435 
436 	/* Toss any page pinned across calls. */
437 	if (t->bt_pinned != NULL) {
438 		mpool_put(t->bt_mp, t->bt_pinned, 0);
439 		t->bt_pinned = NULL;
440 	}
441 
442 	/* In-memory database can't have a file descriptor. */
443 	if (F_ISSET(t, B_INMEM)) {
444 		errno = ENOENT;
445 		return (-1);
446 	}
447 	return (t->bt_fd);
448 }
449