1 /* $OpenBSD: bt_open.c,v 1.13 2005/08/05 13:02:59 espie 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 /*
36 * Implementation of btree access method for 4.4BSD.
37 *
38 * The design here was originally based on that of the btree access method
39 * used in the Postgres database system at UC Berkeley. This implementation
40 * is wholly independent of the Postgres code.
41 */
42
43 #include <sys/param.h>
44 #include <sys/stat.h>
45
46 #include <errno.h>
47 #include <fcntl.h>
48 #include <limits.h>
49 #include <signal.h>
50 #include <stdio.h>
51 #include <stdlib.h>
52 #include <string.h>
53 #include <unistd.h>
54
55 #include <db.h>
56 #include "btree.h"
57
58 #ifdef DEBUG
59 #undef MINPSIZE
60 #define MINPSIZE 128
61 #endif
62
63 static int byteorder(void);
64 static int nroot(BTREE *);
65 static int tmp(void);
66
67 /*
68 * __BT_OPEN -- Open a btree.
69 *
70 * Creates and fills a DB struct, and calls the routine that actually
71 * opens the btree.
72 *
73 * Parameters:
74 * fname: filename (NULL for in-memory trees)
75 * flags: open flag bits
76 * mode: open permission bits
77 * b: BTREEINFO pointer
78 *
79 * Returns:
80 * NULL on failure, pointer to DB on success.
81 *
82 */
83 DB *
__bt_open(const char * fname,int flags,int mode,const BTREEINFO * openinfo,int dflags)84 __bt_open(const char *fname, int flags, int mode, const BTREEINFO *openinfo,
85 int dflags)
86 {
87 struct stat sb;
88 BTMETA m;
89 BTREE *t;
90 BTREEINFO b;
91 DB *dbp;
92 pgno_t ncache;
93 ssize_t nr;
94 int machine_lorder;
95
96 t = NULL;
97
98 /*
99 * Intention is to make sure all of the user's selections are okay
100 * here and then use them without checking. Can't be complete, since
101 * we don't know the right page size, lorder or flags until the backing
102 * file is opened. Also, the file's page size can cause the cachesize
103 * to change.
104 */
105 machine_lorder = byteorder();
106 if (openinfo) {
107 b = *openinfo;
108
109 /* Flags: R_DUP. */
110 if (b.flags & ~(R_DUP))
111 goto einval;
112
113 /*
114 * Page size must be indx_t aligned and >= MINPSIZE. Default
115 * page size is set farther on, based on the underlying file
116 * transfer size.
117 */
118 if (b.psize &&
119 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 ||
120 b.psize & (sizeof(indx_t) - 1)))
121 goto einval;
122
123 /* Minimum number of keys per page; absolute minimum is 2. */
124 if (b.minkeypage) {
125 if (b.minkeypage < 2)
126 goto einval;
127 } else
128 b.minkeypage = DEFMINKEYPAGE;
129
130 /* If no comparison, use default comparison and prefix. */
131 if (b.compare == NULL) {
132 b.compare = __bt_defcmp;
133 if (b.prefix == NULL)
134 b.prefix = __bt_defpfx;
135 }
136
137 if (b.lorder == 0)
138 b.lorder = machine_lorder;
139 } else {
140 b.compare = __bt_defcmp;
141 b.cachesize = 0;
142 b.flags = 0;
143 b.lorder = machine_lorder;
144 b.minkeypage = DEFMINKEYPAGE;
145 b.prefix = __bt_defpfx;
146 b.psize = 0;
147 }
148
149 /* Check for the ubiquitous PDP-11. */
150 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN)
151 goto einval;
152
153 /* Allocate and initialize DB and BTREE structures. */
154 if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL)
155 goto err;
156 memset(t, 0, sizeof(BTREE));
157 t->bt_fd = -1; /* Don't close unopened fd on error. */
158 t->bt_lorder = b.lorder;
159 t->bt_order = NOT;
160 t->bt_cmp = b.compare;
161 t->bt_pfx = b.prefix;
162 t->bt_rfd = -1;
163
164 if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL)
165 goto err;
166 memset(t->bt_dbp, 0, sizeof(DB));
167 if (t->bt_lorder != machine_lorder)
168 F_SET(t, B_NEEDSWAP);
169
170 dbp->type = DB_BTREE;
171 dbp->internal = t;
172 dbp->close = __bt_close;
173 dbp->del = __bt_delete;
174 dbp->fd = __bt_fd;
175 dbp->get = __bt_get;
176 dbp->put = __bt_put;
177 dbp->seq = __bt_seq;
178 dbp->sync = __bt_sync;
179
180 /*
181 * If no file name was supplied, this is an in-memory btree and we
182 * open a backing temporary file. Otherwise, it's a disk-based tree.
183 */
184 if (fname) {
185 switch (flags & O_ACCMODE) {
186 case O_RDONLY:
187 F_SET(t, B_RDONLY);
188 break;
189 case O_RDWR:
190 break;
191 case O_WRONLY:
192 default:
193 goto einval;
194 }
195
196 if ((t->bt_fd = open(fname, flags, mode)) < 0)
197 goto err;
198
199 } else {
200 if ((flags & O_ACCMODE) != O_RDWR)
201 goto einval;
202 if ((t->bt_fd = tmp()) == -1)
203 goto err;
204 F_SET(t, B_INMEM);
205 }
206
207 if (fcntl(t->bt_fd, F_SETFD, 1) == -1)
208 goto err;
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: if (t) {
326 if (t->bt_dbp)
327 free(t->bt_dbp);
328 if (t->bt_fd != -1)
329 (void)close(t->bt_fd);
330 free(t);
331 }
332 return (NULL);
333 }
334
335 /*
336 * NROOT -- Create the root of a new tree.
337 *
338 * Parameters:
339 * t: tree
340 *
341 * Returns:
342 * RET_ERROR, RET_SUCCESS
343 */
344 static int
nroot(BTREE * t)345 nroot(BTREE *t)
346 {
347 PAGE *meta, *root;
348 pgno_t npg;
349
350 if ((root = mpool_get(t->bt_mp, 1, 0)) != NULL) {
351 if (root->lower == 0 &&
352 root->pgno == 0 &&
353 root->linp[0] == 0) {
354 mpool_delete(t->bt_mp, root);
355 errno = EINVAL;
356 } else {
357 mpool_put(t->bt_mp, root, 0);
358 return (RET_SUCCESS);
359 }
360 }
361 if (errno != EINVAL) /* It's OK to not exist. */
362 return (RET_ERROR);
363 errno = 0;
364
365 if ((meta = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
366 return (RET_ERROR);
367
368 if ((root = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL)
369 return (RET_ERROR);
370
371 if (npg != P_ROOT)
372 return (RET_ERROR);
373 root->pgno = npg;
374 root->prevpg = root->nextpg = P_INVALID;
375 root->lower = BTDATAOFF;
376 root->upper = t->bt_psize;
377 root->flags = P_BLEAF;
378 memset(meta, 0, t->bt_psize);
379 mpool_put(t->bt_mp, meta, MPOOL_DIRTY);
380 mpool_put(t->bt_mp, root, MPOOL_DIRTY);
381 return (RET_SUCCESS);
382 }
383
384 static int
tmp(void)385 tmp(void)
386 {
387 sigset_t set, oset;
388 int fd;
389 char *envtmp = NULL;
390 char path[MAXPATHLEN];
391
392 if (issetugid() == 0)
393 envtmp = getenv("TMPDIR");
394 (void)snprintf(path,
395 sizeof(path), "%s/bt.XXXXXX", envtmp ? envtmp : "/tmp");
396
397 (void)sigfillset(&set);
398 (void)sigprocmask(SIG_BLOCK, &set, &oset);
399 if ((fd = mkstemp(path)) != -1)
400 (void)unlink(path);
401 (void)sigprocmask(SIG_SETMASK, &oset, NULL);
402 return(fd);
403 }
404
405 static int
byteorder(void)406 byteorder(void)
407 {
408 u_int32_t x;
409 u_char *p;
410
411 x = 0x01020304;
412 p = (u_char *)&x;
413 switch (*p) {
414 case 1:
415 return (BIG_ENDIAN);
416 case 4:
417 return (LITTLE_ENDIAN);
418 default:
419 return (0);
420 }
421 }
422
423 int
__bt_fd(const DB * dbp)424 __bt_fd(const DB *dbp)
425 {
426 BTREE *t;
427
428 t = dbp->internal;
429
430 /* Toss any page pinned across calls. */
431 if (t->bt_pinned != NULL) {
432 mpool_put(t->bt_mp, t->bt_pinned, 0);
433 t->bt_pinned = NULL;
434 }
435
436 /* In-memory database can't have a file descriptor. */
437 if (F_ISSET(t, B_INMEM)) {
438 errno = ENOENT;
439 return (-1);
440 }
441 return (t->bt_fd);
442 }
443