1 #pragma ident	"%Z%%M%	%I%	%E% SMI"
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. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 #if defined(LIBC_SCCS) && !defined(lint)
40 static char sccsid[] = "@(#)bt_open.c	8.11 (Berkeley) 11/2/95";
41 #endif /* LIBC_SCCS and not lint */
42 
43 /*
44  * Implementation of btree access method for 4.4BSD.
45  *
46  * The design here was originally based on that of the btree access method
47  * used in the Postgres database system at UC Berkeley.  This implementation
48  * is wholly independent of the Postgres code.
49  */
50 
51 #include <sys/param.h>
52 #include <sys/stat.h>
53 
54 #include <errno.h>
55 #include <fcntl.h>
56 #include <limits.h>
57 #include <signal.h>
58 #include <stdio.h>
59 #include <stdlib.h>
60 #include <string.h>
61 #include <unistd.h>
62 
63 #include "db-int.h"
64 #include "btree.h"
65 
66 #ifdef DEBUG
67 #undef	MINPSIZE
68 #define	MINPSIZE	128
69 #endif
70 
71 static int byteorder __P((void));
72 static int nroot __P((BTREE *));
73 static int tmp __P((void));
74 
75 /*
76  * __BT_OPEN -- Open a btree.
77  *
78  * Creates and fills a DB struct, and calls the routine that actually
79  * opens the btree.
80  *
81  * Parameters:
82  *	fname:	filename (NULL for in-memory trees)
83  *	flags:	open flag bits
84  *	mode:	open permission bits
85  *	b:	BTREEINFO pointer
86  *
87  * Returns:
88  *	NULL on failure, pointer to DB on success.
89  *
90  */
91 DB *
92 __bt_open(fname, flags, mode, openinfo, dflags)
93 	const char *fname;
94 	int flags, mode, dflags;
95 	const BTREEINFO *openinfo;
96 {
97 	struct stat sb;
98 	BTMETA m;
99 	BTREE *t;
100 	BTREEINFO b;
101 	DB *dbp;
102 	db_pgno_t ncache;
103 	ssize_t nr;
104 	int machine_lorder;
105 
106 	t = NULL;
107 
108 	/*
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
113 	 * to change.
114 	 */
115 	machine_lorder = byteorder();
116 	if (openinfo) {
117 		b = *openinfo;
118 
119 		/* Flags: R_DUP. */
120 		if (b.flags & ~(R_DUP))
121 			goto einval;
122 
123 		/*
124 		 * Page size must be indx_t aligned and >= MINPSIZE.  Default
125 		 * page size is set farther on, based on the underlying file
126 		 * transfer size.
127 		 */
128 		if (b.psize &&
129 		    (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 ||
130 		    b.psize & (sizeof(indx_t) - 1)))
131 			goto einval;
132 
133 		/* Minimum number of keys per page; absolute minimum is 2. */
134 		if (b.minkeypage) {
135 			if (b.minkeypage < 2)
136 				goto einval;
137 		} else
138 			b.minkeypage = DEFMINKEYPAGE;
139 
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;
145 		}
146 
147 		if (b.lorder == 0)
148 			b.lorder = machine_lorder;
149 	} else {
150 		b.compare = __bt_defcmp;
151 		b.cachesize = 0;
152 		b.flags = 0;
153 		b.lorder = machine_lorder;
154 		b.minkeypage = DEFMINKEYPAGE;
155 		b.prefix = __bt_defpfx;
156 		b.psize = 0;
157 	}
158 
159 	/* Check for the ubiquitous PDP-11. */
160 	if (b.lorder != DB_BIG_ENDIAN && b.lorder != DB_LITTLE_ENDIAN)
161 		goto einval;
162 
163 	/* Allocate and initialize DB and BTREE structures. */
164 	if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL)
165 		goto err;
166 	memset(t, 0, sizeof(BTREE));
167 	t->bt_fd = -1;			/* Don't close unopened fd on error. */
168 	t->bt_lorder = b.lorder;
169 	t->bt_order = NOT;
170 	t->bt_cmp = b.compare;
171 	t->bt_pfx = b.prefix;
172 	t->bt_rfd = -1;
173 
174 	if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL)
175 		goto err;
176 	memset(t->bt_dbp, 0, sizeof(DB));
177 	if (t->bt_lorder != machine_lorder)
178 		F_SET(t, B_NEEDSWAP);
179 
180 	dbp->type = DB_BTREE;
181 	dbp->internal = t;
182 	dbp->close = __bt_close;
183 	dbp->del = __bt_delete;
184 	dbp->fd = __bt_fd;
185 	dbp->get = __bt_get;
186 	dbp->put = __bt_put;
187 	dbp->seq = __bt_seq;
188 	dbp->sync = __bt_sync;
189 
190 	/*
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.
193 	 */
194 	if (fname) {
195 		switch (flags & O_ACCMODE) {
196 		case O_RDONLY:
197 			F_SET(t, B_RDONLY);
198 			break;
199 		case O_RDWR:
200 			break;
201 		case O_WRONLY:
202 		default:
203 			goto einval;
204 		}
205 
206 		if ((t->bt_fd = open(fname, flags | O_BINARY, mode)) < 0)
207 			goto err;
208 
209 	} else {
210 		if ((flags & O_ACCMODE) != O_RDWR)
211 			goto einval;
212 		if ((t->bt_fd = tmp()) == -1)
213 			goto err;
214 		F_SET(t, B_INMEM);
215 	}
216 
217 	if (fcntl(t->bt_fd, F_SETFD, 1) == -1)
218 		goto err;
219 
220 	if (fstat(t->bt_fd, &sb))
221 		goto err;
222 	if (sb.st_size) {
223 		if ((nr = read(t->bt_fd, &m, sizeof(BTMETA))) < 0)
224 			goto err;
225 		if (nr != sizeof(BTMETA))
226 			goto eftype;
227 
228 		/*
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
234 		 * bit.
235 		 */
236 		if (m.magic == BTREEMAGIC)
237 			F_CLR(t, B_NEEDSWAP);
238 		else {
239 			F_SET(t, B_NEEDSWAP);
240 			M_32_SWAP(m.magic);
241 			M_32_SWAP(m.version);
242 			M_32_SWAP(m.psize);
243 			M_32_SWAP(m.free);
244 			M_32_SWAP(m.nrecs);
245 			M_32_SWAP(m.flags);
246 		}
247 		if (m.magic != BTREEMAGIC || m.version != BTREEVERSION)
248 			goto eftype;
249 		if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 ||
250 		    m.psize & (sizeof(indx_t) - 1))
251 			goto eftype;
252 		if (m.flags & ~SAVEMETA)
253 			goto eftype;
254 		b.psize = m.psize;
255 		F_SET(t, m.flags);
256 		t->bt_free = m.free;
257 		t->bt_nrecs = m.nrecs;
258 	} else {
259 		/*
260 		 * Set the page size to the best value for I/O to this file.
261 		 * Don't overflow the page offset type.
262 		 */
263 		if (b.psize == 0) {
264 			b.psize = sb.st_blksize;
265 			if (b.psize < MINPSIZE)
266 				b.psize = MINPSIZE;
267 			if (b.psize > MAX_PAGE_OFFSET + 1)
268 				b.psize = MAX_PAGE_OFFSET + 1;
269 		}
270 
271 		/* Set flag if duplicates permitted. */
272 		if (!(b.flags & R_DUP))
273 			F_SET(t, B_NODUPS);
274 
275 		t->bt_free = P_INVALID;
276 		t->bt_nrecs = 0;
277 		F_SET(t, B_METADIRTY);
278 	}
279 
280 	t->bt_psize = b.psize;
281 
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;
287 
288 	/* Calculate number of pages to cache. */
289 	ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize;
290 
291 	/*
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
300 	 * pages.
301 	 */
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))
305 		t->bt_ovflsize =
306 		    NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t);
307 
308 	/* Initialize the buffer pool. */
309 	if ((t->bt_mp =
310 	    mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL)
311 		goto err;
312 	if (!F_ISSET(t, B_INMEM))
313 		mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t);
314 
315 	/* Create a root page if new tree. */
316 	if (nroot(t) == RET_ERROR)
317 		goto err;
318 
319 	/* Global flags. */
320 	if (dflags & DB_LOCK)
321 		F_SET(t, B_DB_LOCK);
322 	if (dflags & DB_SHMEM)
323 		F_SET(t, B_DB_SHMEM);
324 	if (dflags & DB_TXN)
325 		F_SET(t, B_DB_TXN);
326 
327 	return (dbp);
328 
329 einval:	errno = EINVAL;
330 	goto err;
331 
332 eftype:	errno = EFTYPE;
333 	goto err;
334 
335 err:	if (t) {
336 		if (t->bt_dbp)
337 			free(t->bt_dbp);
338 		if (t->bt_fd != -1)
339 			(void)close(t->bt_fd);
340 		free(t);
341 	}
342 	return (NULL);
343 }
344 
345 /*
346  * NROOT -- Create the root of a new tree.
347  *
348  * Parameters:
349  *	t:	tree
350  *
351  * Returns:
352  *	RET_ERROR, RET_SUCCESS
353  */
354 static int
355 nroot(t)
356 	BTREE *t;
357 {
358 	PAGE *meta, *root;
359 	db_pgno_t npg;
360 
361 	if ((root = mpool_get(t->bt_mp, 1, 0)) != NULL) {
362 		if (root->lower == 0 &&
363 		    root->pgno == 0 &&
364 		    root->linp[0] == 0) {
365 			mpool_delete(t->bt_mp, root);
366 			errno = EINVAL;
367 		} else {
368 			mpool_put(t->bt_mp, root, 0);
369 			return (RET_SUCCESS);
370 		}
371 	}
372 	if (errno != EINVAL)		/* It's OK to not exist. */
373 		return (RET_ERROR);
374 	errno = 0;
375 
376 	if ((meta = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
377 		return (RET_ERROR);
378 
379 	if ((root = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
380 		return (RET_ERROR);
381 
382 	if (npg != P_ROOT)
383 		return (RET_ERROR);
384 	root->pgno = npg;
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);
393 }
394 
395 static int
396 tmp()
397 {
398 #ifdef SIG_BLOCK
399 	sigset_t set, oset;
400 #else
401 	int oset;
402 #endif
403 	int fd;
404 	char *envtmp;
405 	char path[MAXPATHLEN];
406 	static char fn[] = "/bt.XXXXXX";
407 
408 	envtmp = getenv("TMPDIR");
409 
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 */
413 
414 	if (envtmp && ((strlen(envtmp)+sizeof(fn)+1) > sizeof(path)))
415 	    return(-1);
416 
417 	(void)sprintf(path, "%s%s", (envtmp ? envtmp : "/tmp"), fn);
418 
419 #ifdef SIG_BLOCK
420 	(void)sigfillset(&set);
421 	(void)sigprocmask(SIG_BLOCK, &set, &oset);
422 #else
423 	oset = sigblock(~0);
424 #endif
425 	if ((fd = mkstemp(path)) != -1)
426 		(void)unlink(path);
427 #ifdef SIG_BLOCK
428 	(void)sigprocmask(SIG_SETMASK, &oset, NULL);
429 #else
430 	sigsetmask(oset);
431 #endif
432 #ifdef __CYGWIN32__
433       /* Ensure the fd is in binary mode. */
434       setmode(fd, O_BINARY);
435 #endif /* __CYGWIN32__ */
436 
437 	return(fd);
438 }
439 
440 static int
441 byteorder()
442 {
443 	u_int32_t x;
444 	u_char *p;
445 
446 	x = 0x01020304;
447 	p = (u_char *)&x;
448 	switch (*p) {
449 	case 1:
450 		return (DB_BIG_ENDIAN);
451 	case 4:
452 		return (DB_LITTLE_ENDIAN);
453 	default:
454 		return (0);
455 	}
456 }
457 
458 int
459 __bt_fd(dbp)
460         const DB *dbp;
461 {
462 	BTREE *t;
463 
464 	t = dbp->internal;
465 
466 	/* Toss any page pinned across calls. */
467 	if (t->bt_pinned != NULL) {
468 		mpool_put(t->bt_mp, t->bt_pinned, 0);
469 		t->bt_pinned = NULL;
470 	}
471 
472 	/* In-memory database can't have a file descriptor. */
473 	if (F_ISSET(t, B_INMEM)) {
474 		errno = ENOENT;
475 		return (-1);
476 	}
477 	return (t->bt_fd);
478 }
479